Posts Tagged ‘Neuron’

Schizophrenia, broken-links

Larry H. Bernstein, MD, FCAP, Curator



Runs in the Family

 New findings about schizophrenia rekindle old questions about genes and identity.
BY Annals of Science MARCH 28, 2016 ISSUE      http://www.newyorker.com/magazine/2016/03/28/the-genetics-of-schizophrenia


The author and his father have seen several relatives succumb to mental illness.CREDIT PHOTOGRAPH BY DAYANITA SINGH FOR THE NEW YORKER

In the winter of 2012, I travelled from New Delhi, where I grew up, to Calcutta to visit my cousin Moni. My father accompanied me as a guide and companion, but he was a sullen and brooding presence, lost in a private anguish. He is the youngest of five brothers, and Moni is his firstborn nephew—the eldest brother’s son. Since 2004, Moni, now fifty-two, has been confined to an institution for the mentally ill (a “lunatic home,” as my father calls it), with a diagnosis of schizophrenia. He is kept awash in antipsychotics and sedatives, and an attendant watches, bathes, and feeds him through the day.

My father has never accepted Moni’s diagnosis. Over the years, he has waged a lonely campaign against the psychiatrists charged with his nephew’s care, hoping to convince them that their diagnosis was a colossal error, or that Moni’s broken psyche would somehow mend itself. He has visited the institution in Calcutta twice—once without warning, hoping to see a transformed Moni, living a secretly normal life behind the barred gates. But there was more than just avuncular love at stake for him in these visits. Moni is not the only member of the family with mental illness. Two of my father’s four brothers suffered from various unravellings of the mind. Madness has been among the Mukherjees for generations, and at least part of my father’s reluctance to accept Moni’s diagnosis lies in a grim suspicion that something of the illness may be buried, like toxic waste, in himself.

Rajesh, my father’s third-born brother, had once been the most promising of the Mukherjee boys—the nimblest, the most charismatic, the most admired. But in the summer of 1946, at the age of twenty-two, he began to behave oddly, as if a wire had been tripped in his brain. The most obvious change in his personality was a volatility: good news triggered uncontained outbursts of joy; bad news plunged him into inconsolable desolation. By that winter, the sine curve of Rajesh’s psyche had tightened in its frequency and gained in its amplitude. My father recalls an altered brother: fearful at times, reckless at others, descending and ascending steep slopes of mood, irritable one morning and overjoyed the next. When Rajesh received news of a successful performance on his college exams, he vanished, elated, on a two-night excursion, supposedly “exercising” at a wrestling camp. He was feverish and hallucinating when he returned, and died of pneumonia soon afterward. Only years later, in medical school, did I realize that Rajesh was likely in the throes of an acute manic phase. His mental breakdown was the result of a near-textbook case of bipolar disorder.

Jagu, the fourth-born of my father’s siblings, came to live with us in Delhi in 1975, when I was five years old and he was forty-five. His mind, too, was failing. Tall and rail thin, with a slightly feral look in his eyes and a shock of matted, overgrown hair, he resembled a Bengali Jim Morrison. Unlike Rajesh, whose illness had surfaced in his twenties, Jagu had been troubled from his adolescence. Socially awkward, withdrawn from everyone except my grandmother, he was unable to hold a job or live by himself. By 1975, he had visions, phantasms, and voices in his head that told him what to do. He was still capable of extraordinary bursts of tenderness—when I accidentally smashed a beloved Venetian vase at home, he hid me in his bedclothes and informed my mother that he had “mounds of cash” stashed away, enough to buy “a thousand” replacement vases. But this episode was symptomatic: even his love for me extended the fabric of his psychosis and confabulation.

Unlike Rajesh, Jagu was formally diagnosed. In the late nineteen-seventies, a physician in Delhi examined him and determined that he had schizophrenia. But no medicines were prescribed. Instead, Jagu continued to live at home, half hidden away in my grandmother’s room. (As in many families in India, my grandmother lived with us.) For nearly a decade, she and my father maintained a fragile truce, with Jagu living under her care, eating meals in her room and wearing clothes that she stitched for him. At night, when Jagu was consumed by his fears and fantasies, she put him to bed like a child, with her hand on his forehead. She was his nurse, his housekeeper, his only friend, and, more important, his public defender. When my grandmother died, in 1985, Jagu joined a religious sect in Delhi and disappeared, until his death, a dozen years later.


at schizophrenia runs in families was evident even to the person who first defined the illness. In 1911, Eugen Bleuler, a Swiss-German psychiatrist, published a book describing a series of cases of men and women, typically in their teens and early twenties, whose thoughts had begun to tangle and degenerate. “In this malady, the associations lose their continuity,” Bleuler wrote. “The threads between thoughts are torn.” Psychotic visions and paranoid thoughts flashed out of nowhere. Some patients “feel themselves weak, their spirit escapes, they will never survive the day. There is a growth in their heads. Their bones have turned liquid; their hearts have turned into stone. . . . The patient’s wife must not use eggs in cooking, otherwise he will grow feathers.” His patients were often trapped between flickering emotional states, unable to choose between two radically opposed visions, Bleuler noted. “You devil, you angel, you devil, you angel,” one woman said to her lover.

Bleuler tried to find an explanation for the mysterious symptoms, but there was only one seemingly common element: schizophrenic patients tended to have first-degree relatives who were also schizophrenic. He had no tools to understand the mechanism behind the heredity. The word “gene” had been coined just two years before Bleuler published his book. The notion that a mental illness could be carried across generations by unitary, indivisible factors—corpuscles of information threading through families—would have struck most of Bleuler’s contemporaries as mad in its own right. Still, Bleuler was astonishingly prescient about the complex nature of inheritance. “If one is looking for ‘theheredity,’ one can nearly always find it,” he wrote. “We will not be able to do anything about it even later on, unless the single factor of heredity can be broken down into many hereditary factors along specific lines.”

In the nineteen-sixties, Bleuler’s hunch was confirmed by twin studies. Psychiatrists determined that if an identical twin was schizophrenic the other twin had a forty-to-fifty-per-cent chance of developing the disease—fiftyfold higher than the risk in the general population. By the early two-thousands, large population studies had revealed a strong genetic link between schizophrenia and bipolar disorder. Some of the families described in these studies had a crisscrossing history that was achingly similar to my own: one sibling affected with schizophrenia, another with bipolar disorder, and a nephew or niece also schizophrenic.

“The twin studies clarified two important features of schizophrenia and bipolar disorder,” Jeffrey Lieberman, a Columbia University psychiatrist who has studied schizophrenia for thirty years, told me. “First, it was clear that there wasn’t a single gene, but dozens of genes involved in causing schizophrenia—each perhaps exerting a small effect. And, second, even if you inherited the entire set of risk genes, as identical twins do, you still might not develop the disease. Obviously, there were other triggers or instigators involved in releasing the illness.” But while these studies established that schizophrenia had a genetic basis, they revealed nothing about the nature of the genes involved. “For doctors, patients, and families in the schizophrenia community, genetics became the ultimate mystery,” Lieberman said. “If we knew the identity of the genes, we would find the causes, and if we found the causes we could find medicines.”

In 2006, an international consortium of psychiatric geneticists launched a genomic survey of schizophrenia, hoping to advance the search for the implicated genes. With 3,322 patients and 3,587 controls, this was one of the largest and most rigorous such studies in the history of the disease. Researchers scanned through the nearly seven thousand genomes to find variations in gene segments that were correlated with schizophrenia. This strategy, termed an “association study,” does not pinpoint a gene, but it provides a general location where a disease-linked gene may be found, like a treasure map with a large “X” scratched in a corner of the genome.

The results, reported in 2009 (and updated in 2014) in the journal Nature, were a dispiriting validation of Bleuler’s hunch about multiple hereditary factors: more than a hundred independent segments of the genome were associated with schizophrenia. “There are lots of small, common genetic effects, scattered across the genome,” one researcher said. “There are many different biological processes involved.” Some of the putative culprits made biological sense—if dimly. There were genes linked to transmitters that relay messages between neurons, and genes for molecular channels that move electrical signals up and down nerve cells. But by far the most surprising association involved a gene segment on chromosome 6. This region of the genome—termed the MHC region—carries hundreds of genes typically associated with the immune system.

“The MHC-segment finding was so strange and striking that you had to sit up and take notice,” Lieberman told me. “Here was the most definitive evidence that something in the immune system might have something to do with schizophrenia. There had been hints about an immunological association before, but this was impossible to argue with. It raised an endlessly fascinating question: what was the link between immune-response genes and schizophrenia?”

The Rogue Immune Cells That Wreck the Brain

Beth Stevens thinks she has solved a mystery behind brain disorders such as Alzheimer’s and schizophrenia.

by Adam Piore   April 4, 2016            https://www.technologyreview.com/s/601137/the-rogue-immune-cells-that-wreck-the-brain/

In the first years of her career in brain research, Beth Stevens thought of microglia with annoyance if she thought of them at all. When she gazed into a microscope and saw these ubiquitous cells with their spidery tentacles, she did what most neuroscientists had been doing for generations: she looked right past them and focused on the rest of the brain tissue, just as you might look through specks of dirt on a windshield.

“What are they doing there?” she thought. “They’re in the way.’”

Stevens never would have guessed that just a few years later, she would be running a laboratory at Harvard and Boston’s Children’s Hospital devoted to the study of these obscure little clumps. Or that she would be arguing in the world’s top scientific journals that microglia might hold the key to understanding not just normal brain development but also what causes Alzheimer’s, Huntington’s, autism, schizophrenia, and other intractable brain disorders.

Microglia are part of a larger class of cells—known collectively as glia—that carry out an array of functions in the brain, guiding its development and serving as its immune system by gobbling up diseased or damaged cells and carting away debris. Along with her frequent collaborator and mentor, Stanford biologist Ben Barres, and a growing cadre of other scientists, Stevens, 45, is showing that these long-overlooked cells are more than mere support workers for the neurons they surround. Her work has raised a provocative suggestion: that brain disorders could somehow be triggered by our own bodily defenses gone bad.

A type of glial cell known as an oligodendrocyte

In one groundbreaking paper, in January, Stevens and researchers at the Broad Institute of MIT and Harvard showed that aberrant microglia might play a role in schizophrenia—causing or at least contributing to the massive cell loss that can leave people with devastating cognitive defects. Crucially, the researchers pointed to a chemical pathway that might be targeted to slow or stop the disease. Last week, Stevens and other researchers published a similar finding for Alzheimer’s.

This might be just the beginning. Stevens is also exploring the connection between these tiny structures and other neurological diseases—work that earned her a $625,000 MacArthur Foundation “genius” grant last September.

All of this raises intriguing questions. Is it possible that many common brain disorders, despite their wide-ranging symptoms, are caused or at least worsened by the same culprit, a component of the immune system? If so, could many of these disorders be treated in a similar way—by stopping these rogue cells?

Nature. 2016 Feb 11;530(7589):177-83. http://dx.doi.org:/10.1038/nature16549. Epub 2016 Jan 27.   Schizophrenia risk from complex variation of complement component 4.

Schizophrenia is a heritable brain illness with unknown pathogenic mechanisms. Schizophrenia’s strongest genetic association at a population level involves variation in the major histocompatibility complex (MHC) locus, but the genes and molecular mechanisms accounting for this have been challenging to identify. Here we show that this association arises in part from many structurally diverse alleles of the complement component 4 (C4) genes. We found that these alleles generated widely varying levels of C4A and C4B expression in the brain, with each common C4 allele associating with schizophrenia in proportion to its tendency to generate greater expression of C4A. Human C4 protein localized to neuronal synapses, dendrites, axons, and cell bodies. In mice, C4 mediated synapse elimination during postnatal development. These results implicate excessive complement activity in the development of schizophrenia and may help explain the reduced numbers of synapses in the brains of individuals with schizophrenia.


Science  31 Mar 2016;        http://dx.doi.org:/10.1126/science.aad8373      Complement and microglia mediate early synapse loss in Alzheimer mouse models.
Soyon Hong1Victoria F. Beja-Glasser1,*Bianca M. Nfonoyim1,*,…., Ben A. Barres6Cynthia A. Lemere,2Dennis J. Selkoe2,7Beth Stevens1,8,

 Synapse loss in Alzheimer’s disease (AD) correlates with cognitive decline. Involvement of microglia and complement in AD has been attributed to neuroinflammation, prominent late in disease. Here we show in mouse models that complement and microglia mediate synaptic loss early in AD. C1q, the initiating protein of the classical complement cascade, is increased and associated with synapses before overt plaque deposition. Inhibition of C1q, C3 or the microglial complement receptor CR3, reduces the number of phagocytic microglia as well as the extent of early synapse loss. C1q is necessary for the toxic effects of soluble β-amyloid (Aβ) oligomers on synapses and hippocampal long-term potentiation (LTP). Finally, microglia in adult brains engulf synaptic material in a CR3-dependent process when exposed to soluble Aβ oligomers. Together, these findings suggest that the complement-dependent pathway and microglia that prune excess synapses in development are inappropriately activated and mediate synapse loss in AD.
Genome-wide association studies (GWAS) implicate microglia and complement-related pathways in AD (1). Previous research has demonstrated both beneficial and detrimental roles of complement and microglia in plaque-related neuropathology (2, 3); however, their roles in synapse loss, a major pathological correlate of cognitive decline in AD (4), remain to be identified. Emerging research implicates microglia and immune-related mechanisms in brain wiring in the healthy brain (1). During development, C1q and C3 localize to synapses and mediate synapse elimination by phagocytic microglia (57). We hypothesized that this normal developmental synaptic pruning pathway is activated early in the AD brain and mediates synapse loss.


Complex machinery

It’s not surprising that scientists for years have ignored microglia and other glial cells in favor of neurons. Neurons that fire together allow us to think, breathe, and move. We see, hear, and feel using neurons, and we form memories and associations when the connections between different neurons strengthen at the junctions between them, known as synapses. Many neuroscientists argue that neurons create our very consciousness.

Glia, on the other hand, have always been considered less important and interesting. They have pedestrian duties such as supplying nutrients and oxygen to neurons, as well as mopping up stray chemicals and carting away the garbage.

Scientists have known about glia for some time. In the 1800s, the pathologist Rudolf Virchow noted the presence of small round cells packing the spaces between neurons and named them “nervenkitt” or “neuroglia,” which can be translated as nerve putty or glue. One variety of these cells, known as astrocytes, was defined in 1893. And then in the 1920s, the Spanish scientist Pio del Río Hortega developed novel ways of staining cells taken from the brain. This led him to identify and name two more types of glial cells, including microglia, which are far smaller than the others and are characterized by their spidery shape and multiple branches. It is only when the brain is damaged in adulthood, he suggested, that microglia spring to life—rushing to the injury, where it was thought they helped clean up the area by eating damaged and dead cells. Astrocytes often appeared on the scene as well; it was thought that they created scar tissue.

This emergency convergence of microglia and astrocytes was dubbed “gliosis,” and by the time Ben Barres entered medical school in the late 1970s, it was well established as a hallmark of neurodegenerative diseases, infection, and a wide array of other medical conditions. But no one seemed to understand why it occurred. That intrigued Barres, then a neurologist in training, who saw it every time he looked under a microscope at neural tissue in distress. “It was just really fascinating,” he says. “The great mystery was: what is the point of this gliosis? Is it good? Is it bad? Is it driving the disease process, or is it trying to repair the injured brain?”


Barres began looking for the answer. He learned how to grow glial cells in a dish and apply a new recording technique to them. He could measure their electrical qualities, which determine the biochemical signaling that all brain cells use to communicate and coördinate activity.

“From the second I started recording the glial cells, I thought ‘Oh, my God!’” Barres recalls. The electrical activity was more dynamic and complex than anyone had thought. These strange electrical properties could be explained only if the glial cells were attuned to the conditions around them, and to the signals released from nearby neurons. Barres’s glial cells, in other words, had all the machinery necessary to engage in a complex dialogue with neurons, and presumably to respond to different kinds of conditions in the brain.

Why would they need this machinery, though, if they were simply involved in cleaning up dead cells? What could they possibly be doing? It turns out that in the absence of chemicals released by glia, the neurons committed the biochemical version of suicide. Barres also showed that the astrocytes appeared to play a crucial role in forming synapses, the microscopic connections between neurons that encode memory. In isolation, neurons were capable of forming the spiny appendages necessary to reach the synapses. But without astrocytes, they were incapable of connecting to one another.

Hardly anyone believed him. When he was a young faculty member at Stanford in the 1990s, one of his grant applications to the National Institutes of Health was rejected seven times. “Reviewers kept saying, ‘Nah, there’s no way glia could be doing this,’” Barres recalls. “And even after we published two papers in Science showing that [astrocytes] had profound, almost all-or-nothing effects in controlling synapses’ formation or synapse activity, I still couldn’t get funded! I think it’s still hard to get people to think about glia as doing anything active in the nervous system.”

Marked for elimination

Beth Stevens came to study glia by accident. After graduating from Northeastern University in 1993, she followed her future husband to Washington, D.C., where he had gotten work in the U.S. Senate. Stevens had been pre-med in college and hoped to work in a lab at the National Institutes of Health. But with no previous research experience, she was soundly rebuffed. So she took a job waiting tables at a Chili’s restaurant in nearby Rockville, Maryland, and showed up at NIH with her résumé every week.

After a few months, Stevens received a call from a researcher named Doug Fields, who needed help in his lab. Fields was studying the intricacies of the process in which neurons become insulated in a coating called myelin. That insulation is essential for the transmission of electrical impulses.

As Stevens spent the following years pursuing a PhD at the University of Maryland, she was intrigued by the role that glial cells played in insulating neurons. Along the way, she became familiar with other insights into glial cells that were beginning to emerge, especially from the lab of Ben Barres. Which is why, soon after completing her PhD in 2003, Stevens found herself a postdoc in Barres’s lab at Stanford, about to make a crucial discovery.

Barres’s group had begun to identify the specific compounds astrocytes secreted that seemed to cause neurons to grow synapses. And eventually, they noticed that these compounds also stimulated production of a protein called C1q.

Conventional wisdom held that C1q was activated only in sick cells—the protein marked them to be eaten up by immune cells—and only outside the brain. But Barres had found it in the brain. And it was in healthy neurons that were arguably at their most robust stage: in early development. What was the C1q protein doing there?


A stained astrocyte.

The answer lies in the fact that marking cells for elimination is not something that happens only in diseased brains; it is also essential for development. As brains develop, their neurons form far more synaptic connections than they will eventually need. Only the ones that are used are allowed to remain. This pruning allows for the most efficient flow of neural transmissions in the brain, removing noise that might muddy the signal.

But it was unknown how exactly the process worked. Was it possible that C1q helped signal the brain to prune unused synapses? Stevens focused her postdoctoral research on finding out. “We could have been completely wrong,” she recalls. “But we went for it.”

It paid off. In a 2007 paper, Barres and Stevens showed that C1q indeed plays a role in eliminating unneeded neurons in the developing brain. And they found that the protein is virtually absent in healthy adult neurons.

Now the scientists faced a new puzzle. Does C1q show up in brain diseases because the same mechanism involved in pruning a developing brain later goes awry? Indeed, evidence was already growing that one of the earliest events in neurodegenerative diseases such as Alzheimer’s, Parkinson’s, and Huntington’s was significant loss of synapses.

When Stevens and Barres examined mice bred to develop glaucoma, a neurodegenerative disease that kills neurons in the optic system, they found that C1q appeared long before any other detectable sign that the disease was taking hold. It cropped up even before the cells started dying.

This suggested the immune cells might in fact cause the disease, or at the very least accelerate it. And that offered an intriguing possibility: that something could be made to halt the process. Barres founded a company, Annexon Biosciences, to develop drugs that could block C1q. Last week’s paper published by Barres, Stevens, and other researchers shows that a compound being tested by Annexon appears to be able to prevent the onset of Alzheimer’s in mice bred to develop the disease. Now the company hopes to test it in humans in the next two years.

Paths to treatments

To better understand the process that C1q helps trigger, Stevens and Barres wanted to figure out what actually plays the role of Pac-Man, eating up the synapses marked for death. It was well known that white blood cells known as macrophages gobbled up diseased cells and foreign invaders in the rest of the body. But macrophages are not usually present in the brain. For their theory to work, there had to be some other mechanism. And further research has shown that the cells doing the eating even in healthy brains are those mysterious clusters of material that Beth Stevens, for years, had been gazing right past in the microscope—the microglia that Río Hortega identified almost 100 years ago.

Now Stevens’s lab at Harvard, which she opened in 2008, devotes half its efforts to figuring out what microglia are doing and what causes them to do it. These cells, it turns out, appear in the mouse embryo at day eight, before any other brain cell, which suggests they might help guide the rest of brain development—and could contribute to any number of neurodevelopmental diseases when they go wrong.

Meanwhile, she is also expanding her study of the way different substances determine what happens in the brain. C1q is actually just the first in a series of proteins that accumulate on synapses marked for elimination. Stevens has begun to uncover evidence that there is a wide array of protective “don’t eat me” molecules too. It’s the balance between all these cues that regulates whether microglia are summoned to destroy synapses. Problems in any one could, conceivably, mess up the system.

Evidence is now growing that microglia are involved in several neurodevelopmental and psychiatric problems. The potential link to schizophrenia that was revealed in January emerged after researchers at the Broad Institute, led by Steven McCarroll and a graduate student named Aswin Sekar, followed a trail of genetic clues that led them directly to Stevens’s work. In 2009, three consortia from around the globe had published papers comparing DNA in people with and without schizophrenia. It was Sekar who identified a possible pattern: the more a specific type of protein was present in synapses, the higher the risk of developing the disease. The protein, C4, was closely related to C1q, the one first identified in the brain by Stevens and Barres.

McCarroll knew that schizophrenia strikes in late adolescence and early adulthood, a time when brain circuits in the prefrontal cortex undergo extensive pruning. Others had found that areas of the prefrontal cortex are among those most ravaged by the disease, which leads to massive synapse loss. Could it be that over-pruning by rogue microglia is part of what causes schizophrenia?

To find out, Sekar and McCarroll got in touch with Stevens, and the two labs began to hold joint weekly meetings. They soon demonstrated that C4 also had a role in pruning synapses in the brains of young mice, suggesting that excessive levels of the protein could indeed lead to over-pruning—and to the thinning out of brain tissue that appears to occur as symptoms such as psychotic episodes grow worse.

If the brain damage seen in Parkinson’s and Alzheimer’s stems from over-pruning that might begin early in life, why don’t symptoms of those diseases show up until later? Barres thinks he knows. He notes that the brain can normally compensate for injury by rewiring itself and generating new synapses. It also contains a lot of redundancy. That would explain why patients with Parkinson’s disease don’t show discernible symptoms until they have lost 90 percent of the neurons that produce dopamine.

It also might mean that subtle symptoms could in fact be detected much earlier. Barres points to a study of nuns published in 2000. When researchers analyzed essays the nuns had written upon entering their convents decades before, they found that women who went on to develop Alzheimer’s had shown less “idea density” even in their 20s. “I think the implication of that is they could be lifelong diseases,” Barres says. “The disease process could be going on for decades and the brain is just compensating, rewiring, making new synapses.” At some point, the microglia are triggered to remove too many cells, Barres argues, and the symptoms of the disease begin to manifest fully.

Turning this insight into a treatment is far from straightforward, because much remains unclear. Perhaps an overly aggressive response from microglia is determined by some combination of genetic variants not shared by everyone. Stevens also notes that diseases like schizophrenia are not caused by one mutation; rather, a wide array of mutations with small effects cause problems when they act in concert. The genes that control the production of C4 and other immune-system proteins may be only part of the story. That may explain why not everyone who has a C4 mutation will go on to develop schizophrenia.

Nonetheless, if Barres and Stevens are right that the immune system is a common mechanism behind devastating brain disorders, that in itself is a fundamental breakthrough. Because we have not known the mechanisms that trigger such diseases, medical researchers have been able only to alleviate the symptoms rather than attack the causes. There are no drugs available to halt or even slow neurodegeneration in diseases like Alzheimer’s. Some drugs elevate neurotransmitters in ways that briefly make it easier for individuals with dementia to form new synaptic connections, but they don’t reduce the rate at which existing synapses are destroyed. Similarly, there are no treatments that tackle the causes of autism or schizophrenia. Even slowing the progress of these disorders would be a major advance. We might finally go after diseases that have run unchecked for generations.

“We’re a ways away from a cure,” Stevens says. “But we definitely have a path forward.”

Adam Piore is a freelance writer who wrote “A Shocking Way to Fix the Brain”  in November/December 2015.


Int Immunopharmacol. 2001 Mar;1(3):365-92.

Genetic, structural and functional diversities of human complement components C4A and C4B and their mouse homologues, Slp and C4.

Blanchong CA1Chung EKRupert KLYang YYang ZZhou BMoulds JMYu CY.

Author information


The complement protein C4 is a non-enzymatic component of the C3 and C5 convertases and thus essential for the propagation of the classical complement pathway. The covalent binding of C4 to immunoglobulins and immune complexes (IC) also enhances the solubilization of immune aggregates, and the clearance of IC through complement receptor one (CR1) on erythrocytes. Human C4 is the most polymorphic protein of the complement system. In this review, we summarize the current concepts on the 1-2-3 loci model of C4A and C4B genes in the population, factors affecting the expression levels of C4 transcripts and proteins, and the structural, functional and serological diversities of the C4A and C4B proteins. The diversities and polymorphisms of the mouse homologues Slp and C4 proteins are described and contrasted with their human homologues. The human C4 genes are located in the MHC class III region on chromosome 6. Each human C4 gene consists of 41 exons coding for a 5.4-kb transcript. The long gene is 20.6 kb and the short gene is 14.2 kb. In the Caucasian population 55% of the MHC haplotypes have the 2-locus, C4A-C4B configurations and 45% have an unequal number of C4A and C4B genes. Moreover, three-quarters of C4 genes harbor the 6.4 kb endogenous retrovirus HERV-K(C4) in the intron 9 of the long genes. Duplication of a C4 gene always concurs with its adjacent genes RP, CYP21 and TNX, which together form a genetic unit termed an RCCX module. Monomodular, bimodular and trimodular RCCX structures with 1, 2 and 3 complement C4 genes have frequencies of 17%, 69% and 14%, respectively. Partial deficiencies of C4A and C4B, primarily due to the presence of monomodular haplotypes and homo-expression of C4A proteins from bimodular structures, have a combined frequency of 31.6%. Multiple structural isoforms of each C4A and C4B allotype exist in the circulation because of the imperfect and incomplete proteolytic processing of the precursor protein to form the beta-alpha-gamma structures. Immunofixation experiments of C4A and C4B demonstrate > 41 allotypes in the two classes of proteins. A compilation of polymorphic sites from limited C4 sequences revealed the presence of 24 polymophic residues, mostly clustered C-terminal to the thioester bond within the C4d region of the alpha-chain. The covalent binding affinities of the thioester carbonyl group of C4A and C4B appear to be modulated by four isotypic residues at positions 1101, 1102, 1105 and 1106. Site directed mutagenesis experiments revealed that D1106 is responsible for the effective binding of C4A to form amide bonds with immune aggregates or protein antigens, and H1106 of C4B catalyzes the transacylation of the thioester carbonyl group to form ester bonds with carbohydrate antigens. The expression of C4 is inducible or enhanced by gamma-interferon. The liver is the main organ that synthesizes and secretes C4A and C4B to the circulation but there are many extra-hepatic sites producing moderate quantities of C4 for local defense. The plasma protein levels of C4A and C4B are mainly determined by the corresponding gene dosage. However, C4B proteins encoded by monomodular short genes may have relatively higher concentrations than those from long C4A genes. The 5′ regulatory sequence of a C4 gene contains a Spl site, three E-boxes but no TATA box. The sequences beyond–1524 nt may be completely different as the C4 genes at RCCX module I have RPI-specific sequences, while those at Modules II, III and IV have TNXA-specific sequences. The remarkable genetic diversity of human C4A and C4B probably promotes the exchange of genetic information to create and maintain the quantitative and qualitative variations of C4A and C4B proteins in the population, as driven by the selection pressure against a great variety of microbes. An undesirable accompanying byproduct of this phenomenon is the inherent deleterious recombinations among the RCCX constituents leading to autoimmune and genetic disorders.


C4A isotype is responsible for effective binding to form amide bonds with immune aggregates or protein antigens, while C4B isotype catalyzes the transacylation of the thioester carbonyl group to form ester bonds with carbohydrate antigens.

Derived from proteolytic degradation of complement C4, C4a anaphylatoxin is a mediator of local inflammatory process.


Schizophrenia and the Synapse

Genetic evidence suggests that overactive synaptic pruning drives development of schizophrenia.

By Ruth Williams | January 27, 2016


Compared to the brains of healthy individuals, those of people with schizophrenia have higher expression of a gene called C4, according to a paper published inNature today (January 27). The gene encodes an immune protein that moonlights in the brain as an eradicator of unwanted neural connections (synapses). The findings, which suggest increased synaptic pruning is a feature of the disease, are a direct extension of genome-wide association studies (GWASs) that pointed to the major histocompatibility (MHC) locus as a key region associated with schizophrenia risk.

“The MHC [locus] is the first and the strongest genetic association for schizophrenia, but many people have said this finding is not useful,” said psychiatric geneticist Patrick Sullivan of the University of North Carolina School of Medicine who was not involved in the study. “The value of [the present study is] to show that not only is it useful, but it opens up new and extremely interesting ideas about the biology and therapeutics of schizophrenia.”

Schizophrenia has a strong genetic component—it runs in families—yet, because of the complex nature of the condition, no specific genes or mutations have been identified. The pathological processes driving the disease remain a mystery.

Researchers have turned to GWASs in the hope of finding specific genetic variations associated with schizophrenia, but even these have not provided clear candidates.

“There are some instances where genome-wide association will literally hit one base [in the DNA],” explained Sullivan. While a 2014 schizophrenia GWAS highlighted the MHC locus on chromosome 6 as a strong risk area, the association spanned hundreds of possible genes and did not reveal specific nucleotide changes. In short, any hope of pinpointing the MHC association was going to be “really challenging,” said geneticist Steve McCarroll of Harvard who led the new study.

Nevertheless, McCarroll and colleagues zeroed in on the particular region of the MHC with the highest GWAS score—the C4 gene—and set about examining how the area’s structural architecture varied in patients and healthy people.

The C4 gene can exist in multiple copies (from one to four) on each copy of chromosome 6, and has four different forms: C4A-short, C4B-short, C4A-long, and C4B-long. The researchers first examined the “structural alleles” of the C4 locus—that is, the combinations and copy numbers of the different C4 forms—in healthy individuals. They then examined how these structural alleles related to expression of both C4Aand C4B messenger RNAs (mRNAs) in postmortem brain tissues.


Schizophrenia risk from complex variation of complement component 4

Aswin Sekar, Allison R. Bialas, Heather de Rivera, …, Schizophrenia Working Group of the Psychiatric Genomics Consortium, Mark J. Daly, Michael C. Carroll, Beth Stevens & Steven A. McCarroll

Nature (11 Feb 2016); 530: 177–183 http://dx.doi.org:/10.1038/nature16549

Schizophrenia is a heritable brain illness with unknown pathogenic mechanisms. Schizophrenia’s strongest genetic association at a population level involves variation in the major histocompatibility complex (MHC) locus, but the genes and molecular mechanisms accounting for this have been challenging to identify. Here we show that this association arises in part from many structurally diverse alleles of the complement component 4 (C4) genes. We found that these alleles generated widely varying levels of C4A and C4B expression in the brain, with each common C4 allele associating with schizophrenia in proportion to its tendency to generate greater expression of C4A. Human C4 protein localized to neuronal synapses, dendrites, axons, and cell bodies. In mice, C4 mediated synapse elimination during postnatal development. These results implicate excessive complement activity in the development of schizophrenia and may help explain the reduced numbers of synapses in the brains of individuals with schizophrenia.

  1. Cannon, T. D. et al. Cortex mapping reveals regionally specific patterns of genetic and disease-specific gray-matter deficits in twins discordant for schizophrenia. Proc. Natl Acad. Sci. USA 99, 3228–3233 (2002)
  1. Cannon, T. D. et al. Progressive reduction in cortical thickness as psychosis develops: a multisite longitudinal neuroimaging study of youth at elevated clinical risk. Biol. Psychiatry 77,147–157 (2015)
  1. Garey, L. J. et al. Reduced dendritic spine density on cerebral cortical pyramidal neurons in schizophrenia. J. Neurol. Neurosurg. Psychiatry 65, 446–453 (1998)
  1. Glantz, L. A. & Lewis, D. A. Decreased dendritic spine density on prefrontal cortical pyramidal neurons in schizophrenia. Arch. Gen. Psychiatry 57, 65–73 (2000)
  1. Glausier, J. R. & Lewis, D. A. Dendritic spine pathology in schizophrenia. Neuroscience 251,90–107 (2013)
  1. Schizophrenia Working Group of the Psychiatric Genomics Consortium. Biological insights from 108 schizophrenia-associated genetic loci. Nature 511, 421–427 (2014)
  1. Shi, J. et al. Common variants on chromosome 6p22.1 are associated with schizophrenia. Nature 460, 753–757 (2009)
  1. Stefansson, H. et al. Common variants conferring risk of schizophrenia. Nature 460,744–747 (2009)
  1. International Schizophrenia Consortium et al. Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature 460, 748–752 (2009)
  1. Schizophrenia Psychiatric Genome-Wide Association Study Consortium. Genome-wide association study identifies five new schizophrenia loci. Nature Genet . 43, 969–976 (2011)


The strongest genetic association found in schizophrenia is its association to genetic markers across the major histocompatibility complex (MHC) locus, first described in three Nature papers in 2009. …


Schizophrenia: From genetics to physiology at last

Ryan S. DhindsaDavid B. Goldstein
Nature  (11 Feb 2016); 530:162–163   http://dx.doi.org:/10.1038/nature16874

  1. Schizophrenia Working Group of the Psychiatric Genomics Consortium. Nature511,421–427 (2014).
  2. Stevens, B. et alCell131, 1164–1178 (2007).
  3. Cannon, T. D. et al Psychiatry77, 147–157 (2015).
  4. Glausier, J. R. & Lewis, D. A. Neuroscience251, 90–107 (2013).
  5. Glantz, L. A. & Lewis, D. A.  Gen. Psychiatry57, 65–73 (2000).


 Jianxin Shi1, et al.   Common variants on chromosome 6p22.1 are associated with schizophrenia.  Nature 460, 753-757 (6 August 2009) | doi:10.1038/nature08192; Received 29 May 2009; Accepted 10 June 2009; Published online 1 July 2009; Corrected 6 August 2009

Schizophrenia, a devastating psychiatric disorder, has a prevalence of 0.5–1%, with high heritability (80–85%) and complex transmission1. Recent studies implicate rare, large, high-penetrance copy number variants in some cases2, but the genes or biological mechanisms that underlie susceptibility are not known. Here we show that schizophrenia is significantly associated with single nucleotide polymorphisms (SNPs) in the extended major histocompatibility complex region on chromosome 6. We carried out a genome-wide association study of common SNPs in the Molecular Genetics of Schizophrenia (MGS) case-control sample, and then a meta-analysis of data from the MGS, International Schizophrenia Consortium and SGENE data sets. No MGS finding achieved genome-wide statistical significance. In the meta-analysis of European-ancestry subjects (8,008 cases, 19,077 controls), significant association with schizophrenia was observed in a region of linkage disequilibrium on chromosome 6p22.1 (P = 9.54 × 10-9). This region includes a histone gene cluster and several immunity-related genes—possibly implicating aetiological mechanisms involving chromatin modification, transcriptional regulation, autoimmunity and/or infection. These results demonstrate that common schizophrenia susceptibility alleles can be detected. The characterization of these signals will suggest important directions for research on susceptibility mechanisms.

Editor’s Summary   6 August 2009
Schizophrenia risk: link to chromosome 6p22.1

A genome-wide association study using the Molecular Genetics of Schizophrenia case-control data set, followed by a meta-analysis that included over 8,000 cases and 19,000 controls, revealed that while common genetic variation that underlies risk to schizophrenia can be identified, there probably are few or no single common loci with large effects. The common variants identified here lie on chromosome 6p22.1 in a region that includes a histone gene cluster and several genes implicated in immunity.


Hreinn Stefansson1,48, et al. Common variants conferring risk of schizophrenia.
Nature 460, 744-747 (6 August 2009) | doi:10.1038/nature08186; Received 16 March 2009; Accepted 5 June 2009; Published online 1 July 2009

Schizophrenia is a complex disorder, caused by both genetic and environmental factors and their interactions. Research on pathogenesis has traditionally focused on neurotransmitter systems in the brain, particularly those involving dopamine. Schizophrenia has been considered a separate disease for over a century, but in the absence of clear biological markers, diagnosis has historically been based on signs and symptoms. A fundamental message emerging from genome-wide association studies of copy number variations (CNVs) associated with the disease is that its genetic basis does not necessarily conform to classical nosological disease boundaries. Certain CNVs confer not only high relative risk of schizophrenia but also of other psychiatric disorders1, 2, 3. The structural variations associated with schizophrenia can involve several genes and the phenotypic syndromes, or the ‘genomic disorders’, have not yet been characterized4. Single nucleotide polymorphism (SNP)-based genome-wide association studies with the potential to implicate individual genes in complex diseases may reveal underlying biological pathways. Here we combined SNP data from several large genome-wide scans and followed up the most significant association signals. We found significant association with several markers spanning the major histocompatibility complex (MHC) region on chromosome 6p21.3-22.1, a marker located upstream of the neurogranin gene (NRGN) on 11q24.2 and a marker in intron four of transcription factor 4 (TCF4) on 18q21.2. Our findings implicating the MHC region are consistent with an immune component to schizophrenia risk, whereas the association with NRGN and TCF4 points to perturbation of pathways involved in brain development, memory and cognition.



The International Schizophrenia Consortium. Common polygenic variation contributes to risk of schizophrenia and bipolar disorder.  Nature 460, 748-752 (6 August 2009) | doi:10.1038/nature08185; Received 11 February 2009; Accepted 8 June 2009; Published online 1 July 2009; Corrected 6 August 2009

Schizophrenia is a severe mental disorder with a lifetime risk of about 1%, characterized by hallucinations, delusions and cognitive deficits, with heritability estimated at up to 80%1, 2. We performed a genome-wide association study of 3,322 European individuals with schizophrenia and 3,587 controls. Here we show, using two analytic approaches, the extent to which common genetic variation underlies the risk of schizophrenia. First, we implicate the major histocompatibility complex. Second, we provide molecular genetic evidence for a substantial polygenic component to the risk of schizophrenia involving thousands of common alleles of very small effect. We show that this component also contributes to the risk of bipolar disorder, but not to several non-psychiatric diseases.


The Psychiatric GWAS Consortium Steering Committee. A framework for interpreting genome-wide association studies of psychiatric disorders.  Molecular Psychiatry (2009) 14, 10–17; doi:10.1038/mp.2008.126; published online 11 November 2008

Genome-wide association studies (GWAS) have yielded a plethora of new findings in the past 3 years. By early 2009, GWAS on 47 samples of subjects with attention-deficit hyperactivity disorder, autism, bipolar disorder, major depressive disorder and schizophrenia will be completed. Taken together, these GWAS constitute the largest biological experiment ever conducted in psychiatry (59 000 independent cases and controls, 7700 family trios and >40 billion genotypes). We know that GWAS can work, and the question now is whether it will work for psychiatric disorders. In this review, we describe these studies, the Psychiatric GWAS Consortium for meta-analyses of these data, and provide a logical framework for interpretation of some of the conceivable outcomes.

Keywords: genome-wide association, attention-deficit hyperactivity disorder, autism, bipolar disorder, major depressive disorder, schizophrenia

The purpose of this article is to consider the ‘big picture’ and to provide a logical framework for the possible outcomes of these studies. This is not a review of GWAS per se as many excellent reviews of this technically and statistically intricate methodological approach are available.789101112 This is also not a review of the advantages and disadvantages of different study designs and sampling strategies for the dissection of complex psychiatric traits. We would like to consider how the dozens of GWAS papers that will soon be in the literature can be synthesized: what can integrated mega-analyses (meta-analysis is based on summary data (for example, odds ratios) from all available studies whereas ‘mega-analysis’ uses individual-level genotype and phenotype data) of all available GWAS data tell us about the etiology of these psychiatric disorders? This is an exceptional opportunity as positive or negative results will enable us to learn hard facts about these critically important psychiatric disorders. We suggest that it is not a matter of ‘success versus failure’ or ‘optimism versus pessimism’ but rather an opportunity for systematic and logical approaches to empirical data whereby both positive and appropriately qualified negative findings are informative.

The studies that comprise the Psychiatric GWAS Consortium (PGC; http://pgc.unc.edu) are shown in Table 1. GWAS data for ADHD, autism, bipolar disorder, major depressive disorder and schizophrenia from 42 samples of European subjects should be available for mega-analyses by early 2009 (>59 000 independent cases and controls and >7700 family trios). To our knowledge, the PGC will have access to the largest set of GWAS data available.

A major change in human genetics in the past 5 years has been in the growth of controlled-access data repositories, and individual phenotype and genotype data are now available for many of the studies in Table 1. When the PGC mega-analyses are completed, most data will be available to researchers via the NIMH Human Genetics Initiative (http://nimhgenetics.org). Although the ready availability of GWAS data is a benefit to the field by allowing rapid application of a wide range of analytic strategies to GWAS data, there are potential disadvantages. GWAS mega-analysis is complex and requires considerable care and expertise to be done validly. For psychiatric phenotypes, there is the additional challenge of working with disease entities based largely on clinical description, with unknown biological validity and having both substantial clinical variation within diagnostic categories as well as overlaps across categories.13 Given the urgent need to know if there are replicable genotype–phenotype associations, a new type of collaboration was required.

The purpose of the PGC is to conduct rigorous and comprehensive within- and cross-disorder GWAS mega-analyses. The PGC began in early 2007 with the principal investigators of the four GAIN GWAS,14 and within six months had grown to 110 participating scientists from 54 institutions in 11 countries. The PGC has a coordinating committee, five disease-working groups, a cross-disorder group, a statistical analysis and computational group, and a cluster computer for statistical analysis. It is remarkable that almost all investigators approached agreed to participate and that no one has left the PGC. Most effort is donated but we have obtained funding from the NIMH, the Netherlands Scientific Organization, Hersenstichting Nederland and NARSAD.

The PGC has two major specific aims. (1) Within-disorder mega-analyses: conduct separate mega-analyses of all available GWAS data for ADHD, autism, bipolar disorder, major depressive disorder, and schizophrenia to attempt to identify genetic variation convincingly associated with any one of these five disorders. (2) Cross-disorder mega-analyses: the clinically-derived DSM-IV and ICD-10 definitions may not directly reflect the fundamental genetic architecture.15 There are two subaims. (2a) Conduct mega-analysis to identify genetic variation convincingly associated with conventional definitions of two or more disorders. This nosological aim could assist in delineating the boundaries of this set of disorders. (2b) An expert working group will convert epidemiological and genetic epidemiological evidence into explicit hypotheses about overlap among these disorders, and then conduct mega-analyses based on these definitions (for example, to examine the lifetime presence of idiopathic psychotic features without regard to diagnostic context).

The goal of the PGC is to identify convincing genetic variation-disease associations. A convincing association would be extremely unlikely to result from chance, show consistent effect sizes across all or almost all samples and be impervious to vigorous attempts to disprove the finding (for example, by investigating sources of bias, confirmatory genotyping, and so on). Careful attention will be paid to the impact of potential sources of heterogeneity17 with the goal of assessing its impact without minimizing its presence.

Biological plausibility is not an initial requirement for a convincing statistical association, as there are many examples in human genetics of previously unsuspected candidate genes nonetheless showing highly compelling associations. For example, multiple SNPs in intron 1 of the FTO gene were associated with body mass index in 13 cohorts with 38 759 participants18 and yet ‘FTO’ does not appear in an exhaustive 116 page compilation of genetic studies of obesity.19 Some strong associations are in gene deserts: multiple studies have found convincing association between prostate cancer and a region on 8q24 that is ~250 kb from the nearest annotated gene.20 Both of these examples are being intensively investigated and we suspect that a compelling mechanistic ‘story’ will emerge in the near future. The presence of a compelling association without an obvious biological mechanism establishes a priority research area for molecular biology and neuroscience of a psychiatric disorder.

The PGC will use mega-analysis as the main analytic tool as individual-level data will be available from almost all samples. To wield this tool appropriately, a number of preconditions must be met. First, genotype data from different GWAS platforms must be made comparable as the direct overlap between platforms is often modest. This requires meticulous quality control for the inclusion of both SNPs and subjects and attention to the factors that can cause bias (for example, population stratification, cryptic relatedness or genotyping batch effects). Genotype harmonization can be accomplished using imputation (2122, for example) so that the same set of ~2 million2324 directly or imputed SNP genotypes are available for all subjects. Second, phenotypes need to be harmonized across studies. This is one of the most crucial components of the PGC and we are fortunate to have world experts directing the work. Third, the mega-analyses will assess potential heterogeneity of associations across samples.

A decision-tree schematic of the potential outcomes of the PGC mega-analyses is shown in Figure 1. Note that many of the possibilities in Figure 1 are not mutually exclusive and different disorders may take different paths through this framework. It is possible that there eventually will be dozens or hundreds of sequence variants strictly associated with these disorders with frequencies ranging from very rare to common.



GWAS has the potential to yield considerable insights but it is no panacea and may well perform differently for psychiatric disorders. Even if these psychiatric GWAS efforts are successful, the outcomes will be complex. GWAS may help us learn that clinical syndromes are actually many different things—for example, proportions of individuals with schizophrenia might evidence associations with rare CNVs of major effect,56 with more common genetic variation in dozens (perhaps hundreds) of genomic regions, between genetic variation strongly modified by environmental risk factors, and some proportion may be genetically indistinguishable from the general population. Moreover, as fuel to long-standing ‘lumper versus splitter’ debates in psychiatric nosology, empirical data might show that some clinical disorders or identifiable subsets of subjects might overlap considerably.

The critical advantage of GWAS is the search of a ‘closed’ hypothesis space. If the large amount of GWAS data being generated are analyzed within a strict and coherent framework, it should be possible to establish hard facts about the fundamental genetic architecture of a set of important psychiatric disorders—which might include positive evidence of what these disorders are or exclusionary evidence of what they are not. Whatever the results, these historically large efforts should yield hard facts about ADHD, autism, bipolar disorder, major depressive disorder and schizophrenia that may help guide the next era of psychiatric research.

  1. Pe’er I, Yelensky R, Altshuler D, Daly MJ. Estimation of the multiple testing burden for genomewide association studies of nearly all common variants. Genet Epidemiol 2008; 32: 381–385. | Article | PubMed |
  2. Weiss LA, Shen Y, Korn JM, Arking DE, Miller DT, Fossdal R et al. Association between microdeletion and microduplication at 16p11.2 and autism. N Engl J Med 2008; 358: 667–675. | Article | PubMed | ChemPort |



Hreinn Stefansson1,36, et al. Large recurrent microdeletions associated with schizophrenia. Nature 455, 232-236 (11 September 2008) | doi:10.1038/nature07229; Received 17 April 2008; Accepted 8 July 2008; Corrected 11 September 2008

Reduced fecundity, associated with severe mental disorders1, places negative selection pressure on risk alleles and may explain, in part, why common variants have not been found that confer risk of disorders such as autism2, schizophrenia3 and mental retardation4. Thus, rare variants may account for a larger fraction of the overall genetic risk than previously assumed. In contrast to rare single nucleotide mutations, rare copy number variations (CNVs) can be detected using genome-wide single nucleotide polymorphism arrays. This has led to the identification of CNVs associated with mental retardation4, 5 and autism2. In a genome-wide search for CNVs associating with schizophrenia, we used a population-based sample to identify de novoCNVs by analysing 9,878 transmissions from parents to offspring. The 66 de novo CNVs identified were tested for association in a sample of 1,433 schizophrenia cases and 33,250 controls. Three deletions at 1q21.1, 15q11.2 and 15q13.3 showing nominal association with schizophrenia in the first sample (phase I) were followed up in a second sample of 3,285 cases and 7,951 controls (phase II). All three deletions significantly associate with schizophrenia and related psychoses in the combined sample. The identification of these rare, recurrent risk variants, having occurred independently in multiple founders and being subject to negative selection, is important in itself. CNV analysis may also point the way to the identification of additional and more prevalent risk variants in genes and pathways involved in schizophrenia.


The C4 gene can exist in multiple copies (from one to four) on each copy of chromosome 6, and has four different forms: C4A-short, C4B-short, C4A-long, and C4B-long. The researchers first examined the “structural alleles” of the C4 locus—that is, the combinations and copy numbers of the different C4 forms—in healthy individuals. They then examined how these structural alleles related to expression of both C4Aand C4B messenger RNAs (mRNAs) in postmortem brain tissues.

From this the researchers had a clear picture of how the architecture of the C4 locus affected expression ofC4A and C4B. Next, they compared DNA from roughly 30,000 schizophrenia patients with that from 35,000 healthy controls, and a correlation emerged: the alleles most strongly associated with schizophrenia were also those that were associated with the highest C4A expression. Measuring C4A mRNA levels in the brains of 35 schizophrenia patients and 70 controls then revealed that, on average, C4A levels in the patients’ brains were 1.4-fold higher.

C4 is an immune system “complement” factor—a small secreted protein that assists immune cells in the targeting and removal of pathogens. The discovery of C4’s association to schizophrenia, said McCarroll, “would have seemed random and puzzling if it wasn’t for work . . . showing that other complement components regulate brain wiring.” Indeed, complement protein C3 locates at synapses that are going to be eliminated in the brain, explained McCarroll, “and C4 was known to interact with C3 . . . so we thought well, actually, this might make sense.”

McCarroll’s team went on to perform studies in mice that revealed C4 is necessary for C3 to be deposited at synapses. They also showed that the more copies of the C4 gene present in a mouse, the more the animal’s neurons were pruned.

Synaptic pruning is a normal part of development and is thought to reflect the process of learning, where the brain strengthens some connections and eradicates others. Interestingly, the brains of deceased schizophrenia patients exhibit reduced neuron density. The new results, therefore, “make a lot of sense,” said Cardiff University’s Andrew Pocklington who did not participate in the work. They also make sense “in terms of the time period when synaptic pruning is occurring, which sort of overlaps with the period of onset for schizophrenia: around adolescence and early adulthood,” he added.

“[C4] has not been on anybody’s radar for having anything to do with schizophrenia, and now it is and there’s a whole bunch of really neat stuff that could happen,” said Sullivan. For one, he suggested, “this molecule could be something that is amenable to therapeutics.”




Derived from proteolytic degradation of complement C4, C4a anaphylatoxin is a mediator of local inflammatory process. It induces the contraction of smooth muscle, increases vascular permeability and causes histamine release from mast cells and basophilic leukocytes.

Non-enzymatic component of C3 and C5 convertases and thus essential for the propagation of the classical complement pathway. Covalently binds to immunoglobulins and immune complexes and enhances the solubilization of immune aggregates and the clearance of IC through CR1 on erythrocytes. C4A isotype is responsible for effective binding to form amide bonds with immune aggregates or protein antigens, while C4B isotype catalyzes the transacylation of the thioester carbonyl group to form ester bonds with carbohydrate antigens.


Related Articles in Pharmaceutical Intelligence 

Schizophrenia genomics

Identical Twin Brother Develops Schizophrenia

Imaging Schizophrenia Brain

Outstanding Achievement in Schizophrenia Research

Combining Genetic has Identified Schizophrenia Subtypes

A new relationship identified in preterm stress and development of autism or schizophrenia

Genomic Promise for Neurodegenerative Diseases, Dementias, Autism Spectrum, Schizophrenia, and Serious Depression

Brain Biobank

Protein binding to RNAs in brain

Brain Matters from iBiology

Sleep science

Dopamine-β-Hydroxylase Functional Variants

An inconvenient truth about dreams

Synapse activity in neurotransmission

Mindful Discoveries

Schizophrenic hallucinations

Neurons Reprogrammed

Role of Neurotransmitters and other such Neurosignaling Molecules

microglia and brain maintenance

sequential memory

To reduce symptoms of mental illness and retrain the brain

To understand what happens in the brain to cause mental illness

Icelandic Population Genomic Study Results by deCODE Genetics come to Fruition: Curation of Current genomic studies

The Colors of Life Function

3:15PM 11/12/2014 – Discussion Complex Disorders @10th Annual Personalized Medicine Conference at the Harvard Medical School, Boston

Searchable Database of almost 100,000 functional Brain Scans related to Behavior on patients from 111 countries will change Psychiatry practices set in 170 years

Social Behavior Traits Embedded in Gene Expression

News in Exploration of the Biological Causes of Mental Illness: Potential for New Treatments

Nobel Prize in Physiology or Medicine 2013 for Cell Transport: James E. Rothman of Yale University; Randy W. Schekman of the University of California, Berkeley; and Dr. Thomas C. Südhof of Stanford University

Translational Research on the Mechanism of Water and Electrolyte Movements into the Cell

Synaptotagmin functions as a Calcium Sensor: How Calcium Ions Regulate the fusion of vesicles with cell membranes during Neurotransmission

Signaling Pathway that Makes Young Neurons Connect was discovered @ Scripps Research Institute

No dishonour in depression

MGH’s Largest-ever Genetic Study of Five Psychiatric Disorders: Variation in SNPs in Two Genes involved in Calcium-Channel Signaling

Modern Society, Risk and Mentally Disordered Offender









Read Full Post »

Protein binding to RNAs in brain

Larry H. Bernstein, MD, FCAP, Curator



Regulatory consequences of neuronal ELAV-like protein binding to coding and non-coding RNAs in human brain

 Claudia Scheckel, 

Neuronal ELAV-like (nELAVL) RNA binding proteins have been linked to numerous neurological disorders. We performed crosslinking-immunoprecipitation and RNAseq on human brain, and identified nELAVL binding sites on 8681 transcripts. Using knockout mice and RNAi in human neuroblastoma cells, we showed that nELAVL intronic and 3′ UTR binding regulates human RNA splicing and abundance. We validated hundreds of nELAVL targets among which were important neuronal and disease-associated transcripts, including Alzheimer’s disease (AD) transcripts. We therefore investigated RNA regulation in AD brain, and observed differential splicing of 150 transcripts, which in some cases correlated with differential nELAVL binding. Unexpectedly, the most significant change of nELAVL binding was evident on non-coding Y RNAs. nELAVL/Y RNA complexes were specifically remodeled in AD and after acute UV stress in neuroblastoma cells. We propose that the increased nELAVL/Y RNA association during stress may lead to nELAVL sequestration, redistribution of nELAVL target binding, and altered neuronal RNA splicing.

DOI: http://dx.doi.org/10.7554/eLife.10421.001


eLife digest

When a gene is active, its DNA is copied into a molecule of RNA. This molecule then undergoes a process called splicing which removes certain segments, and the resulting ‘messenger RNA’ molecule is then translated into protein. Many messenger RNAs go through alternative splicing, whereby different segments can be included or excluded from the final molecule. This allows more than one type of protein to be produced from a single gene.

Specialized RNA binding proteins associate with messenger RNAs and regulate not only their splicing, but also their abundance and location within the cell. These activities are crucially important in the brain where forming memories and learning new skills requires thousands of proteins to be made rapidly. Many members of a family of RNA binding proteins called ELAV-like proteins are unique to neurons. These proteins have also been associated with conditions such as Alzheimer’s disease, but it was not known which messenger RNAs were the targets of these proteins in the human brain.

Scheckel, Drapeau et al. have now addressed this question and used a method termed ‘CLIP’ to identify thousands of messenger RNAs that directly bind to neuronal ELAV-like proteins in the human brain. Many of these messenger RNAs coded for proteins that are important for the health of neurons, and neuronal ELAV-like proteins were shown to regulate both the alternative splicing and the abundance of these messenger RNAs.

The regulation of RNA molecules in post-mortem brain samples of people with or without Alzheimer’s disease was then compared. Scheckel, Drapeau et al. unexpectedly observed that, in the Alzheimer’s disease patients, the neuronal ELAV-like proteins were very often associated with a class of RNA molecules known as Y RNAs. These RNA molecules do not code for proteins, and are therefore classified as non-coding RNA. Moreover, massive shifts in the binding of ELAV-like proteins onto Y RNAs were observed in neurons grown in the laboratory that had been briefly stressed by exposure to ultraviolet radiation.

Scheckel, Drapeau et al. suggest that the strong tendency of neuronal ELAV-like proteins to bind to Y RNAs in conditions of short- or long-term stress, including Alzheimer’s disease, might prevent these proteins from associating with their normal messenger RNA targets. This was supported by finding that some messenger RNAs targeted by neuronal ELAV-like proteins showed altered regulation after stress. Such changes to the normal regulation of these messenger RNAs could have a large impact on the proteins that are produced from them.

Together, these findings link Y RNAs to both neuronal stress and Alzheimer’s disease, and suggest a new way that a cell can alter which messenger RNAs are expressed in response to changes in its environment. The next step is to explore what causes the shift in neuronal ELAV-like protein binding from messenger RNAs to Y RNAs and how it might contribute to disease.



RNA binding proteins (RBPs) associate with RNAs throughout their life cycle, regulating all aspects of RNA metabolism and function. More than 800 RBPs have been described in human cells (Castello et al., 2012). The unique structure and function of neurons, and the need to rapidly adapt RNA regulation in the brain both within and at sites distant from the nucleus, are consistent with specialized roles for RBPs in the brain. Indeed, mammalian neurons have developed their own system of RNA regulation (Darnell, 2013), and RBP:mRNA interactions are thought to regulate local protein translation at synapses, perhaps underlying learning and long-term memory (McKee et al., 2005).

Numerous RBPs have been linked to human neurological disorders (reviewed in Richter and Klann (2009)). For example, FUS, TDP-43 and ATXN2 mutations have been found in familial amyotrophic lateral sclerosis patients (Elden et al., 2010; Vance et al., 2009; Sreedharan et al., 2008), TDP-43has additionally been associated with frontotemporal lobar degeneration, Alzheimer’s Disease (AD) and Parkinson’s Disease (PD) (Baloh, 2011), STEX has been linked to amyotrophic lateral sclerosis 4 (Chen et al., 2004), and spinal muscular atrophy can be caused by mutations in SMN (Clermont et al., 1995).

The neuronal ELAV-like (ELAVL) and NOVA RBPs are targeted by the immune system in paraneoplastic neurodegenerative disorders (Buckanovich et al., 1996; Szabo et al., 1991). Mammalian ELAVL proteins include the ubiquitously expressed paralog ELAVL1 (also termed HUA or HUR) and the three neuron-specific paralogs, ELAVL2, 3 and 4 (also termed HUB, C, and D, and collectively referred to as nELAVL; Ince-Dunn et al., 2012). nELAVL proteins are expressed exclusively in neurons in mice (Okano and Darnell, 1997), and they are important for neuronal differentiation and neurite outgrowth in cultured neurons (Akamatsu et al., 1999; Kasashima et al., 1999; Mobarak et al., 2000; Anderson et al., 2000; Antic et al., 1999; Aranda-Abreu et al., 1999). Redundancy between the three nELAVL isoforms complicates in vivo studies of their individual functions. Nevertheless, even haploinsuffiency of Elavl3 is sufficient to trigger cortical hypersynchronization, and Elavl3 and Elavl4 null mice display defects in motor function and neuronal maturation, respectively (Akamatsu et al., 2005; Ince-Dunn et al., 2012).

ELAVL proteins have been shown to regulate several aspects of RNA metabolism. In vitro and in tissue culture cells, nELAVL proteins have been implicated in the regulation of stabilization and/or translation of specific mRNAs, as well as in the regulation of splicing and polyadenylation of select transcripts [reviewed in Pascale et al. (2004)]. A more comprehensive approach was taken by immunoprecipitating an overexpressed isoform of ELAVL4 in mice, although such RNA immunoprecipitation experiments cannot distinguish between direct and indirect targets (Bolognani et al., 2010). Recently, direct binding of nELAVL to target RNAs in mouse brain was demonstrated by high-throughput sequencing of RNA isolated by crosslinking immunoprecipitation (HITS-CLIP; Ince-Dunn et al., 2012); these data, coupled with transcriptome profiling of Elavl3/4 KO mice, demonstrated that nELAVL directly regulates neuronal mRNA abundance and alternative splicing by binding to U-rich elements with interspersed purine residues in 3’UTRs and introns in mouse brain (Ince-Dunn et al., 2012).

While genome-wide approaches have been applied to studying nELAVL proteins in mice, the targets of nELAVL in the human brain remain largely unknown. This is of particular importance, as nELAVL proteins have been implicated in neurological disorders such as AD (Amadio et al., 2009; Kang et al., 2014) and PD (DeStefano et al., 2008; Noureddine et al., 2005). Hence, to advance our understanding of the function of nELAVL in humans and its link to human disease, we set out to investigate nELAVL:RNA interactions in the human brain.

To globally identify transcripts directly bound by nELAVL in human neurons, we generated a genome wide RNA binding map of nELAVL in human brain using CLIP. CLIP allows the identification of functional RNA-protein interactions in vivo by using UV-irradiation of intact tissues to covalently crosslink and then purify RNA-protein complexes present in vivo (Licatalosi and Darnell, 2010; Ule et al., 2003). This method has been adopted for a variety of RBPs (Darnell, 2010; 2013; Moore et al., 2014). Here, we systemically identified tens of thousands of reproducible nELAVL binding sites in human brain and showed that nELAVL binds transcripts that are important for neurological function and that have been linked to neurological diseases such as AD. We validated the functional consequences of nELAVL binding in mice and cultured human neuroblastoma cells and showed that the loss of nELAVL affected mRNA abundance and alternative splicing of hundreds of transcripts. We further investigated RNA regulation in AD brains, and found that numerous transcripts were differentially spliced in AD, which correlated with differential nELAVL binding in some cases. Remarkably, we observed the most significant increase in nELAVL binding in AD on a class of non-coding RNAs, Y RNAs. We recapitulated these findings in human neuroblastoma cells, showing that nELAVL binding is linked to Y ribonucleoprotein (RNP) remodeling acutely during UV-induced stress, and chronically in AD.



Figure 1.

Figure 1.Identification of nELAVL targets in human brain.

(A) Illustration depicting the brain area analyzed by CLIP and RNAseq. The image was generated using BodyParts3D/Anatomography service by DBCLS, Japan. (B) SDS-PAGE separation of radiolabeled nELAVL-RNP complexes. nELAVL-RNP complexes from 40 mg of human brain were specifically immunoprecipitated with Hu-antiserum, compared to control serum (compare lane #4 to #1), which is dependent on UV irradiation (compare lane #4 to #2). Wide-range nELAVL-RNP complexes collapse to a single band in the presence of high RNAse concentration (lane #3). RNAse dilutions: + 19.23 Units/µl; +++ 3846 Units/µl. As in studies of mouse nELAVL (Ince-Dunn et al., 2012), higher molecular weight bands were present in nELAVL CLIP autoradiograms, which correspond at least in part to nELAVL multimers. (C) Shown is the most enriched motif in the top 500 nELAVL peaks, determined with MEME-ChiP. (D) Pie chart of the genomic peak distribution of 75,592 nELAVL peaks (p < 0.01; present in at least 5 individuals). (E) nELAVL binding correlates with mRNA abundance. nELAVL binding (CLIP tags within binding sites per transcript) was compared to mRNA abundance (RNAseq tags per transcript). Only expressed genes with peaks are shown and the correlation coefficient is indicated. The top 1000 targets were identified as genes with highest normalized nELAVL binding (binding sites were normalized for mRNA abundance and summarized per gene). (F) Subnetwork of direct protein-protein interactions of top nELAVL targets. The 1000 top nELAVL target genes and six additional genes highly associated with AD (APP, BACE1, MAPT, PICALM, PSEN1 and PSEN2) were clustered using the organic layout algorithm in yEd. Genes with no direct interactions with other target genes were excluded, leaving 172 nodes from the top nELAVL target list (green) and 5 AD associated genes (blue) in this subnetwork. The size of the nodes is proportional to the connectivity degree. Six clusters (gray circles) containing at least 10 nodes were identified, and subjected to enrichment analysis (see Supplementary file 1F).

DOI: http://dx.doi.org/10.7554/eLife.10421.003


Figure 2.

Figure 2.nELAVL mediated regulation is conserved in mouse and human.

(A) Overlap of nELAVL targets in human and mouse. Human nELAVL targets (n = 8681) were intersected with mouse targets identified by RIP (Bolognani et al., 2010) or HITS-CLIP (Ince-Dunn et al., 2012). 538 genes were identified as nELAVL targets by RIP and were expressed in human brain. 1978 expressed genes had HITS-CLIP nELAVL clusters that were present in at least 3 samples (biological complexity (BC) ≥ 3). Both overlaps (n = 500 and n = 1835) were highly significant (p = 6.5e-74 and p = 2.3e-287; hypergeometric test), compared to expressed transcripts (n = 14,737). (B) Only few nELAVL binding sites are conserved between mice and human, which are predominantly present within 3’UTRs. The genomic distribution of all human nELAVL binding sites (total) and nELAVL binding sites conserved in mouse is shown. The number of nELAVL binding sites (n) within each category is indicated. (C) UCSC Genome Browser images illustrating the 3’UTRs of RAB6B, HCN3, and KCNMB2 and their normalized nELAVL binding profile in human brain. The maximum PeakHeight is indicated by numbers in the right corner. (D) The mRNA levels of transcripts with nELAVL 3’UTR binding decrease in Elavl3/4 knockout (KO) mice. Shown are the mRNA expression fold changes (knockout/wildtype) of RAB6B, HCN3, and KCNMB2. *p< 0.01 (two-tailed t test; Ince-Dunn et al., 2012). (E) UCSC Genome Browser images showing pink cassette exons in the DST, NRXN1, and CELF2 genes and their normalized nELAVL binding profiles in human brain. The maximum PeakHeight is indicated by numbers in the right corner. (F) nELAVL binding adjacent to a cassette exon in the DST gene prevents exon inclusion. Downstream nELAVL binding promotes the inclusion of cassette exons in the NRXN1 and CELF2 genes. The change in alternative exon inclusion (delta inclusion (ΔI): wildtype – Elavl3/4 KO) is shown. * significantly changing (analyzed by Aspire2;Ince-Dunn et al., 2012).

DOI: http://dx.doi.org/10.7554/eLife.10421.010


Figure 3.

Figure 3.nELAVL proteins regulate mRNA abundance of human brain targets.

(A) nELAVL depletion causes mRNA level changes in IMR-32 neuroblastoma cells. The mRNA abundance change was plotted against average mRNA abundance. Significantly changing transcripts (FDR < 0.05; n = 784) are colored in blue. Shown are only expressed genes (n = 12,743), and ELAVL1/2/3/4 transcripts are indicated. (B) nELAVL with exclusively 3’UTR binding decrease upon nELAVL RNAi depletion. Box plots represent the distribution of mRNA level differences between mock and nELAVL RNAi. We compared genes with exclusively 3’UTR (n = 2346) or intronic (n = 1693) binding that were expressed in IMR-32 cells. nELAVL binding was defined as CLIP tags within binding sites per transcript. Transcripts with exclusively 3’UTR binding were less abundant upon nELAVL RNAi compared to remaining transcripts (p = 3.8e-15; two-tailed t-test). In contrast, mRNA levels of transcripts with exclusively intron binding were even slightly increased compared to remaining transcripts (p = 1.7e-4; two-tailed t-test). (C) Transcripts with nELAVL 3’UTR binding decrease upon nELAVL RNAi. Cumulative fraction curves for genes with no 3’UTR nELAVL binding in human brain, 3’UTR binding, and top 3’UTR targets. Top targets were identified as 1000 genes with highest normalized nELAVL 3’UTR binding (binding sites were normalized for mRNA abundance before summarized per gene). 952 of the top 1000 targets were expressed in IMR-32 cells. A curve displacement to the left indicates a downregulation of mRNA abundance upon nELAVL RNAi. p values were calculated with a one-sided KS test, comparing (top) targets to non-targets. (D) Many transcripts that are decreasing upon nELAVL depletion are top nELAVL 3’UTR targets. The mRNA abundance change (nELAVL/mock RNAi) of transcripts expressed in IMR-32 cells and in human brain (n = 12,242) was plotted against average mRNA abundance. Significantly changing transcripts (FDR<0.05; n = 743) are colored in blue and additionally boxed if they are top nELAVL 3’UTR targets. Transcripts shown in E/F are indicated. (E) UCSC Genome Browser images illustrating the 3’UTRs of APPBP2, ATXN3, andSHANK2 and their normalized nELAVL binding profile in human brain. The maximum PeakHeight is indicated by numbers in the right corner. (F) The mRNA abundance of top nELAVL 3’UTR targets decreases upon nELAVL RNAi. Shown are the mRNA level changes (nELAVL/mock RNAi) of APPBP2, ATXN3, and SHANK2. * FDR<0.05 (derived from edgeR).

DOI: http://dx.doi.org/10.7554/eLife.10421.012


Figure 4.

Figure 4.nELAVL regulates splicing of human brain targets.

(A) Analysis of splicing changes upon nELAVL RNAi. Shown is the exon inclusion fraction of cassette exons that are expressed in IMR-32 cells and in human brain (n = 7903). Significantly changing exons (FDR<0.05 and ΔI>0.1) are colored in light blue (n = 473), and additionally boxed in dark blue if adjacent (+/- 2.5 kb) to intronic nELAVL binding sites (n = 155). Significantly changing exons shown in (B/C) are boxed in pink. The two alternative events withinPICALM correspond to the same alternative exon with two different 3’ splice sites. (B) UCSC Genome Browser images depicting cassette exons in pink in the BIN1, PICALM, and APP genes and their normalized nELAVL binding profiles in human brain. The maximum PeakHeight is indicated by numbers in the right corner. (C) nELAVL binding downstream of cassette exons in BIN1 and PICALM promotes exon inclusion, whereas intronic nELAVL binding ofAPP prevents exon inclusion downstream and upstream. The change in alternative exon inclusion (ΔI: mock –nELAVL RNAi) is shown. *FDR< 0.0005; **FDR< 1e-4; ***FDR<1e-16 (GLM likelihood ratio test). (D) Normalized nELAVL binding map of nELAVL regulated exons. Only exons that changed significantly upon nELAVL RNAi (FDR<0.05 and ΔI>0.1) and that are adjacent (+/- 2.5 kb) to intronic nELAVL binding sites (n = 155) were included. Red and blue peaks represent binding associated with nELAVL-dependent exon inclusion and exclusion, respectively.


RNA regulation changes in AD

nELAVL has previously been linked to neurological diseases and we observed that nELAVL regulated the mRNA abundance and splicing of multiple disease-associated genes. We examined nELAVL binding in a set of genes with disease associated 3’UTR single nucleotide polymorphisms (SNPs) (Bruno et al., 2012). We found that these genes were enriched among nELAVL 3’UTR targets (n = 200; p = 0.001; hypergeometric test), and that nELAVL binding sites directly overlapped with 45 disease associated SNPs, including SNPs associated with autism, schizophrenia, depression, AD, and PD (Figure 5—figure supplement 1, Supplementary file 3A).

nELAVL proteins have been implicated in AD (Amadio et al., 2009; Kang et al., 2014), and among the validated nELAVL regulated RNAs were also several AD-related transcripts, which led us to investigate additional AD-linked genes (hereafter termed AD genes; n = 96; Supplementary file 3B). Indeed, we found that the top nELAVL targets were enriched among AD genes (n = 11; p = 0.03; hypergeometric test; contained in Supplementary file 3B) as well as among AD risk loci identified in a genome-wide association study (GWAS) in AD (Naj et al., 2011) (n = 77; p = 1.7e-14; hypergeometric test; Supplementary file 3C). To investigate if nELAVL mediated regulation of AD related and other transcripts might be affected in AD, we performed nELAVL CLIP and RNAseq on AD subject brains, age-matched to control subjects (Figure 5—figure supplement 2, Supplementary file 1A/B and 3D). Importantly, ELAVL3/4 mRNA levels were similar between control and AD samples and ELAVL2 showed only a slight decrease in transcript abundance in AD brains (Supplementary file 1B), which allowed us to compare nELAVL binding profiles between control and AD brains. We did not detect many significant changes in nELAVL binding nor mRNA abundance (Figure 5A/B, Supplementary file 1B and 3D), probably due to the variation between human samples, the small sample size, and the potential heterogeneity of AD. We did however observe that 150 transcripts were differentially spliced in the 9 AD subjects (FDR<0.05 and ΔI>0.1; Figure 5C, Supplementary file 3E). Two of these transcripts, BIN1 and PTPRD, have previously been linked to AD (Tan et al., 2013; Ghani et al., 2012), suggesting that the differential splicing of these two transcripts as well as other RNAs might be linked to AD.

Figure 5.RNA regulation changes in AD.

(A) nELAVL binding changes in AD. The nELAVL peak binding change (AD/Control) was plotted against average nELAVL peak binding. Significantly changing peaks (FDR<0.05; n = 52) are colored in blue, and peaks within AD genes are colored in pink (1811 peaks within 69 genes). Shown are only peaks that are bound in control or AD brain (n = 115,393). (B) mRNA abundance changes in AD. The mRNA abundance change (AD/Control) was plotted against average mRNA abundance. Significantly changing transcripts (FDR<0.05; n = 3) are colored in blue, and AD transcripts are colored in pink (n = 89). Shown are only transcripts that are expressed in control or AD brain (n = 14,875). (C) Analysis of splicing changes in AD. Shown is the inclusion fraction of expressed cassette exons in control and AD subjects (n = 8163). Exons within AD genes are colored in pink (n = 79). Significantly changing exons (FDR<0.05 and ΔI>0.1) are colored in light blue (n = 170), and additionally boxed in pink if within AD genes (n = 2). (D) BIN1 is alternatively spliced in AD. UCSC Genome Browser image illustrating a cassette exon in the BIN1 gene and normalized nELAVL binding profiles in control and AD brain. The maximum PeakHeight is indicated by numbers in the right corner. Bar graphs depict the difference in alternative exon inclusion (ΔI: Control – AD) and nELAVL peak binding (AD/Control) in control and AD brain. Corresponding FDR values derived from edgeR are shown. The inclusion of the exon is promoted by nELAVL (see Figure 4), and exon inclusion as well as nELAVL peak binding are reduced in AD subjects.

DOI: http://dx.doi.org/10.7554/eLife.10421.015


As shown above (Figure 4), nELAVL depletion in IMR-32 cells was associated with the reduced inclusion of an alternative exon of BIN1, suggesting that nELAVL binding promotes the inclusion of this exon. Precisely this exon was differentially spliced in AD subjects, with AD subjects showing a reduced exon inclusion rate compared to control subjects (Figure 5D). Along with the differential exon inclusion, we observed that nELAVL peak binding was fourfold decreased in AD subjects (log2 fold change = -2.35; p = 0.16; Figure 5D). These results are consistent with nELAVL-mediated dysregulation of this exon in AD subjects, with decreased binding leading to decreased exon inclusion. In conclusion, while we did not detect global nELAVL binding and mRNA abundance changes in AD subjects, we observed that splicing of 150 transcripts was affected, which in some cases might be linked to nELAVL dysregulation.

Non-coding Y RNAs are bound by nELAVL in AD

The largest fold changes in nELAVL binding in AD (relative to the age-matched control population) occurred on a specific class of non-coding RNAs, Y RNAs (Wolin et al., 2013). Y RNAs are 100 nt long structured RNAs usually found in complex with RO60 (also known as TROVE2; Figure 6A; modified from Chen and Wolin, 2004). RO60 is believed to act as a sensor of RNA quality, targeting defective RNAs for degradation (Sim and Wolin, 2011). RO60 was initially identified as an autoantigen targeted in systemic lupus (Lerner et al., 1981) and some subjects with the paraneoplastic encephalopathy syndrome harbor both anti-RO and anti-nELAVL (Hu) autoantibodies (Manley et al., 1994). Four canonical Y RNAs, Y1/3/4/5, have been characterized in humans, but numerous slightly divergent copies of these Y RNAs, especially Y1 and Y3, are distributed throughout the human genome (Perreault et al., 2005).

Figure 6.

Figure 6.Non-coding Y RNAs are bound by nELAVL in AD.

(A) Secondary structures of Y1 and Y3. Binding sites of nELAVL and Ro are indicated. Modified from (Chen and Wolin, 2004). (B) The nELAVL binding motif (UUUUUU, allowing a G at any position) is enriched in nELAVL-bound Y RNAs compared to non-bound Y RNAs (p = 1.1e-7; Fisher’s exact test). Y RNAs were scanned for (T)6, allowing a G at any position. nELAVL-bound Y RNAs: nELAVL CLIP tags in at least two samples; n = 320. (C) nELAVL binding of Y RNAs increases in AD compared to control samples (p = 4.47e-51; paired one-sided Wilcoxon rank sum test). The axes depict nELAVL Y RNA binding (nELAVL CLIP tags per Y RNA) in control and AD subjects. Y RNAs with nELAVL binding motif are colored in green. (D) Y RNA levels do not change in AD. Y RNA abundance (RNAseq tags per Y RNA) in AD subjects was plotted against Y RNA abundance in control subjects.

DOI: http://dx.doi.org/10.7554/eLife.10421.018


Surprisingly, we observed nELAVL binding to a total of 320 Y RNAs, although Y RNA copies other than the canonical four Y1/3/4/5 genes had previously been considered to be non-functional and were labeled ‘pseudogenes’ (Supplementary file 3F). We found that 237 of the 320 nELAVL bound Y RNAs were Y3-like RNAs (Supplementary file 3F), and that nELAVL bound Y RNAs showed an enrichment of the nELAVL binding motif (202 Y RNAs contained UUUUUU, allowing a G at any one position), which is also present in the canonical hY3 RNA (Figure 6A/B). We examined the 118 nELAVL bound Y RNAs that did not fit this consensus in more detail. 91 of these Y RNAs (77%) contained either a 5mer version of the motif or the motif with an A or C instead of a G, and we found U/G rich stretches in the remaining 27 Y RNAs (Supplementary file 3F). In addition, some Y RNAs with a strong binding motif did not show any evidence of nELAVL binding. In general, these Y RNAs showed a lower expression compared to nELAVL bound Y RNA, which may explain the absence of detectable nELAVL binding (Figure 6—figure supplement 1).

We next explored nELAVL/Y RNA binding in AD brain. We observed a drastic increase in nELAVL/Y RNA association in AD subjects (Figure 6C), while Y RNA levels remained largely unchanged (Figure 6D). This suggests that Y RNPs undergo nELAVL-dependent remodeling in AD. Interestingly, we did observe a high variability in nELAVL/Y RNA association between AD samples (Figure 6—figure supplement 2), with three of them showing a very strong nELAVL/Y RNA association. Efforts to relate this difference to the expression of stress-related genes, post-mortem interval, age, extent of disease and cause of death were not conclusive, and the cause for the variation in nELAVL binding to Y RNAs among AD subjects remains elusive.

Y RNPs are remodeled during UV stress

The observation of increased nELAVL/Y RNA association in AD raised the possibility that Y RNP remodeling is associated with neuronal stress. Y RNP remodeling has previously been linked to UV-induced stress (Sim et al., 2009), and both bacterial (Chen et al., 2000; Wurtmann and Wolin, 2010) and mouse cells (Chen et al., 2003) show an increased sensitivity to UV stress in the absence of RO60. ELAVL binding can be modulated in response to stress in cultured cells (Bhattacharyya et al., 2006), and ELAVL proteins, which shuttle between nucleus and cytoplasm in response to environmental cues, preferentially accumulate in cytoplasmic stress granules upon stress (Gallouzi et al., 2000; Fan and Steitz, 1998b). We therefore examined the effect of acute UV stress on Y RNP remodeling in IMR-32 cells. IMR-32 cells were exposed to a low dose of UV stress (not sufficient to induce RNA:protein crosslinking) and allowed to recover for 24 h before being analyzed by nELAVL CLIP. We found that nELAVL bound 132 Y RNAs in neuroblastoma cells (Supplementary file 3F), that Y RNAs showed an enrichment of the nELAVL binding motif (Figure 7A) or at least contained a degenerate version of it (Supplementary file 3F), and that non-bound Y RNAs with a motif show a very low expression (Figure 7—figure supplement 1). Moreover, nELAVL binding on Y RNAs was dynamic and increased in UV stressed cells compared to non-stressed cells (Figure 7B and Figure 7—figure supplement 2), while their abundance did not change upon UV irradiation (Figure 7C). To assess whether Y RNA levels were affected by nELAVL, we depleted nELAVL by RNAi three days prior to the UV exposure, and analyzed Y RNA levels by RNAseq. Y RNA abundance was not affected by nELAVL depletion in UV stressed IMR-32 cells (Figures 7D). These results indicate that increased nELAVL binding to Y RNAs is not a function of Y RNA levels, and that nELAVL binding during stress is not required for Y RNA stability.


Figure 7.

Figure 7.Y RNPs are remodeled during UV stress.

(A) The nELAVL binding motif (UUUUUU, allowing a G at any position) is enriched in nELAVL-bound Y RNAs compared to non-bound Y RNAs (p = 6.2e-6; Fisher’s exact test). Y RNAs were scanned for (T)6, allowing a G at any position. nELAVL-bound Y RNAs: nELAVL CLIP tags in at least two samples; n = 132. (B) nELAVL binding of Y RNAs increases during UV stress compared to non-stressed cells (p = 8.23e-29; paired one-sided Wilcoxon rank sum test). The axes depict nELAVL Y RNA binding (nELAVL CLIP tags per Y RNA) in control and UV stressed cells. Y RNAs with nELAVL binding motif are colored in green. (C) Y RNA levels do not change upon UV stress. Y RNA abundance (RNAseq tags per Y RNA) in UV stressed cells was plotted against Y RNA abundance in non-stressed control cells. (D) nELAVL is binding is not required for Y RNA stability. Comparison of Y RNA abundance between mock andnELAVL RNAi treated UV stressed cells.

DOI: http://dx.doi.org/10.7554/eLife.10421.021


Figure 8.nELAVL/Y RNA correlates with loss of nELAVL-mediated splicing.

(A) Samples with high nELAVL/Y RNA association show decreased nELAVL binding on mRNA targets. Columns represent significantly changing nELAVL binding sites. Shown are changes in AD subjects with and without Y RNA association (AD_Y and AD_nY) and changes upon UV treatment. The number of nELAVL binding sites (n) within each category is indicated. (B) Identification of nELAVL-dependent UV-induced splicing changes. Comparison of the differential inclusion rate of expressed cassette exons upon UV stress between mock and nELAVL RNAi treated IMR-32 cells (n = 9397). Significant UV-induced splicing changes that do not change upon UV stress in nELAVL RNA treated cells are boxed in dark blue (FDR<0.05 and ΔI>0.1; n = 260). (C) Many exons that are alternatively spliced upon nELAVL RNAi treatment also change during UV stress in an nELAVL-dependent manner. Shown is the inclusion rate of expressed cassette exons in IMR-32 cells that were subjected to mock or nELAVL RNAi (n = 9397). nELAVL RNAi induced splicing changes are colored in light blue (n = 553), and are additionally boxed in dark blue if they are UV-induced in an nELAVL-dependent manner (n = 68). The plot is related to Figure 4A but contains additional cassette exons expressed in UV stressed cells. (D) nELAVL binding adjacent to exons that are alternatively spliced upon nELAVL RNAi and UV treatment decreases only in AD subjects with an increased Y RNA association. Displayed is the change in nELAVL peak binding. nELAVL peak binding changes were not significant except for CBFA2T2(boxed in pink). * FDR<0.05 (derived from edgeR). (E) UCSC Genome Browser images depicting an overview and an enlarged view of a cassette exon within the CBFA2T2 gene that is alternatively spliced in nELAVL RNAi and UV-treated IMR-32 cells. The nELAVL binding track in human brain and RNAseq tracks in mock and nELAVL RNAi treated non-stressed and UV-stressed IMR-32 cells are shown.

DOI: http://dx.doi.org/10.7554/eLife.10421.026


Figure 9.Y RNA overexpression is linked to nELAVL sequestration from mRNA targets.

(A) Validation of Y RNA overexpression. Shown are RNA expression fold changes of Y3wt or Y3mut infected IMR-32 cells compared to non-infected IMR-32 cells assessed by qPCR. Y RNAs expression increased while control mRNAs (ACTB, GAPDH, ELAVL4) were not affected. Error bars represent SEM. p values were calculated with a two-tailed t-test (ns: not significant; * p<0.05). (B) The expression of endogenous Y3-like Y RNAs increases upon Y3wt but not Y3mut infection. Box plots represent the distribution of endogenous Y3-like and non-Y3-like Y RNA expression fold changes upon Y3wt or Y3mut infection. Y3-like Y RNAs show a slight increase in abundance upon Y3wt compared to non-Y3-like Y RNAs (p = 0.057; one-tailed t-test). In contrast, the mRNA abundance of Y3-like Y RNAs does not change upon Y3mut infection, when compared to non-Y3 like Y RNAs (p = 0.602; one-tailed t-test). (C) Identification of Y3 dependent splicing changes. Shown is the exon inclusion fraction of cassette exons that are expressed in IMR-32 cells subjected to Y3wt or Y3mut infection (n = 10,189). Exons changing significantly between Y3wt and Y3mut infection (FDR<0.05 and ΔI>0.1) are colored in light blue (n = 191). (D) Exons that are alternatively spliced upon Y3wt infection are enriched for nELAVL bound exons. Bar graph representing total expressed exons (n = 10,189), exons that change in either Y3wt (n = 240; blue points in the left panel of Figure 9—figure supplement 4) or Y3mut (n = 151; blue points in the right panel of Figure 9—figure supplement 4) infected cells compared to non-infected cells, and exons that change in Y3wt compared to Y3mut infected cells (n = 191; blue points in Figure 9C). Exons that are alternatively spliced upon Y3wt infection compared to either non-infected (p = 0.037; hypergeometric test) or Y3mut infected cells (p = 0.069; hypergeometric test) are enriched for nELAVL bound exons.

DOI: http://dx.doi.org/10.7554/eLife.10421.029


In contrast to the mRNA abundance changes, only few splicing changes overlapped between Y3wt and Y3mut infection when compared to non-infected cells (17% of Y3wt induced changes overlapped with Y3mut induced changes). Most of the observed splicing changes are therefore likely to be specific to Y RNA overexpression. Importantly, we observed an enrichment of nELAVL bound exons and of nELAVL RNAi dependent exons among the exons that changed upon Y3wt but not Y3mut overexpression (Figure 9C/D and Figure 9—figure supplement 4, 5). The relatively small enrichment is consistent with the modest increase in total Y3-like Y RNAs. These results suggest that Y RNA overexpression results in nELAVL sequestration from some of its intronic targets and consequent splicing changes, and partially recapitulates the stress induced nELAVL sequestration due to increased nELAVL/Y RNA association seen in AD patients and UV treated IMR-32 cells.


nELAVL proteins are abundant neuron-specific RNA binding proteins which have been suggested to regulate various neurological processes and have been linked to neurodegenerative disorders including AD and PD. Yet the RNA targets of nELAVL in human brain were completely unknown. Here, we generated a comprehensive genome-wide RNA binding map of nELAVL in human brain, identifying 75,592 significant binding events within 8681 transcripts. We observed a significant overlap between these binding sites and disease-associated 3’UTR SNPs, and the potential disruption of nELAVL-mediated RNA regulation at these sites might contribute to disease manifestation. Most deleterious variants to date have been identified by exome sequencing while as many as 50% of disease-causing mutations are thought to affect splicing (Ward and Cooper, 2009). With whole genome sequencing being increasingly available, non-coding variants are also increasingly detected, some of which may be linked to disease. As the majority of nELAVL binding occurs in introns and 3’UTRs, we expect that many binding sites will overlap with prospective disease-associated non-coding variants. The overlap between deleterious variants and nELAVL binding sites, and the observation that nELAVL binding at individual sites diverged between mice and human, underscores the importance of this study and illustrates the caveat of relying solely on mouse models when studying human disease. Considering the widespread nature of nELAVL binding in human brain and that RNA dysregulation has been linked to numerous neurological disorders, we believe that this binding map will be a valuable resource for the scientific community.

To analyze the functional consequences of nELAVL binding, we used two different loss-of-function models: Elavl3/4 KO mice and nELAVL RNAi depletion in neuroblastoma cells. Due to the incomplete RNAi depletion of nELAVL in neuroblastoma cells, and potential differences in mRNA abundance and therefore nELAVL binding between the different samples, it is likely that we validated only a fraction of nELAVL-regulated transcripts. Despite these technical limitations we demonstrated that nELAVL impacts mRNA abundance and/or splicing of hundreds of targets. Among the nELAVL regulated transcripts were many transcripts implicated in human disease, including AD, which led us to investigate RNA regulation in AD subjects. Due to the relatively small sample size and the heterogeneity between these samples, likely due to both differences between individuals and sample preservation during postmortem collection, we did not detect many reproducible changes in mRNA abundance or nELAVL binding between AD and non-AD subjects. However, we found that 150 transcripts were differentially spliced in AD subjects, which in some cases coincided with differential nELAVL binding. Unexpectedly, the most significant binding change in AD was a dramatic increase in nELAVL binding to a class of non-coding RNAs, termed Y RNAs. This change was evident on a specific subset of Y RNAs harboring the nELAVL binding site. nELAVL/Y RNA binding also increased during UV stress in human neuroblastoma cells, while the abundance of Y RNAs remained constant in AD subjects and upon UV exposure. The increased nELAVL/Y RNA association correlated with decreased nELAVL binding at a subset of intronic binding sites, and was associated with similar splicing changes as induced by nELAVL depletion, suggesting that nELAVL/Y RNP remodeling during acute and chronic stress sequesters nELAVL from its mRNA targets. We provided further evidence for a Y RNA dependent nELAVL sequestration by overexpressing Y3 RNAs harboring either a wild type or mutated nELAVL binding site. Exons that were differentially spliced upon Y RNA overexpression were enriched for nELAVL bound exons, indicating nELAVL sequestration, which was dependent on an intact nELAVL binding site in the Y RNA.

nELAVL 3’UTR binding has been implicated in increasing mRNA abundance in vivo (Ince-Dunn et al., 2012). We described numerous nELAVL 3’UTR targets in brain, and were able to validate many of these targets, including disease-associated transcripts, indicating that nELAVL 3’UTR binding is important for the regulation of mRNA abundance in human brain. While ELAVL binding is frequently reported to result in an increase in mRNA abundance, we found several cases where nELAVL binding seemed to have an opposing effect. ELAVL proteins can compete or collaborate with miRNAs as well as RBPs like AUF1, CUGBP1 and TIA1 to regulate its targets (Bhattacharyya et al., 2006;Kawai et al., 2006; Lal et al., 2004; Young et al., 2009; Yu et al., 2013; Kim et al., 2009). The ultimate outcome of nELAVL 3’UTR binding might therefore vary between individual transcripts.

nELAVL has also been shown to regulate splicing in mouse brain by binding to intronic sequences (Ince-Dunn et al., 2012). We observed many instances of intronic nELAVL binding events adjacent to alternative exons in brain, and confirmed that nELAVL regulates many of these exons in mice and neuroblastoma cells. In contrast to the position-dependent splicing observed for other RBPs (Licatalosi and Darnell, 2010), we observed that upstream nELAVL binding was associated with both exon skipping and inclusion. While nELAVL binding was observed within 25-50 nucleotides upstream of skipped exons, coinciding with the branch point sequence, nELAVL binding peaked within the proximal 25 nucleotides upstream of included exons, overlapping the polypyrimidine tract. Binding of auxiliary splicing factors, including nELAVL, to the branch point sequence usually interferes with spliceosome assembly and thus leads to exon skipping (Licatalosi and Darnell, 2010). Polypyrimidine tract binding however can lead to both exon inclusion and skipping (Licatalosi et al., 2012; Wei et al., 2012), presumably depending on the recruitment of splicing enhancers or silencers. Our data indicates that upstream nELAVL binding can both interfere with the assembly of the spliceosome as well as promote splicing, most likely by recruiting splicing enhancers.

Splicing defects have been associated with many neurological diseases (Licatalosi and Darnell, 2006), and among the nELAVL-regulated transcripts we describe here are numerous transcripts related to disease, including AD. For example, intronic nELAVL binding of the gene encoding the amyloid precursor protein, APP, was associated with skipping of exons 7 and 8. Both exons have previously been shown to be alternatively spliced and encode for the Kunitz protease inhibitory (KPI) motif, a domain that has been linked to APP processing (Ben Khalifa et al., 2012). Remarkably, KPI domain containing isoforms of APP have been shown to be increased in AD (Zhang et al., 2012), indicating that APP splicing might contribute to AD pathogenesis, and that nELAVL binding in human brain might be important to regulate the inclusion of the KPI domain. nELAVL regulates the splicing of two more AD-related transcripts, PICALM and BIN1, by promoting the inclusion of alternative exons 13 and 6a, respectively. Both proteins have been implicated in APP trafficking and both exons lie within domains mediating protein-protein interactions (Tan et al., 2013; Treusch et al., 2011). Moreover, inclusion of the alternative exon 13 in PICALM has been linked to an AD-associated SNP (Parikh et al., 2014), and we observed in this study that exon 6a of BIN1 shows a higher inclusion rate in controls compared to AD subjects. Since nELAVL binding promotes the inclusion of this exon, and control subjects show higher nELAVL binding, we propose that the altered splicing of BIN1 in AD subjects might be due to differential nELAVL binding. In fact, several nELAVL-regulated exons have been shown to be differentially spliced in AD subjects, further strengthening the link between nELAVL dysregulation and AD.

While Y RNAs have not been linked to AD before, they have been implicated in various types of stress responses. The RNA binding protein RO60 usually associates with Y RNAs and is required for their stabilization (Chen et al., 2000; 2003; Labbé et al., 1999; Wolin et al., 2013; Xue et al., 2003). Besides RO60, Y RNPs contain several other RBPs such as ZBP1, MOV10, and Y-box proteins, and have been found to be remodeled upon stress (Sim et al., 2012). Our data suggests that nELAVL becomes increasingly associated with specific Y RNAs during both UV-induced stress and AD. ELAVL proteins can shuttle between nucleus and cytoplasm in response to environmental cues and preferentially accumulate in cytoplasmic stress granules upon cellular stress (Fan and Steitz, 1998a;Gallouzi et al., 2000), and ELAVL binding to the CAT-1 transcript is modulated in response to stress in cultured cells (Bhattacharyya et al., 2006). Interestingly, while we found that nELAVL specifically associates with Y RNAs during AD and acute UV stress, the nucleocytoplasmic distribution of nELAVL, RO60, and Y RNAs was not affected by UV stress. Because Y RNA levels remained constant, we propose that Y RNP complexes are specifically remodeled during AD and acute stress, which is not likely due to a change in nucleocytoplasmic protein/RNA distribution. These results are consistent with previous observations that stress induced shuttling might be limited to ELAVL1 (Burry and Smith, 2006). Our observation of Y RNP remodeling in two very different systems of neuronal stress suggests that differential nELAVL/Y RNA association may be a widespread phenomenon and a focus of future studies.

In addition to the four canonical human Y RNAs, hY1/3/4/5, hundreds of additional Y RNA genes are distributed throughout the human genome (Perreault et al., 2005). The apparent lack of promoters upstream led to a premature designation of these Y RNAs as pseudogenes. Surprisingly, we found that hundreds of these Y RNA copies are expressed in human brain and neuroblastoma cells, although it remains unclear if these Y RNAs can still associate with RO60, because the RO60 binding site in many Y RNA copies is mutated (Perreault et al., 2005). We observed that numerous Y RNA copies were more strongly associated with nELAVL in AD brain and acutely stressed cells, yet nELAVL binding did not affect their levels, indicating a function for this interaction other than Y RNA stabilization. While the outcome of nELAVL/Y RNA remains to be elucidated, our work revealed an aspect of nELAVL/Y RNA association related to stoichiometry. Hundreds of Y RNAs are bound by nELAVL in AD and UV-stress, which corresponds to up to 5% of all nELAVL CLIP tags. This shift of nELAVL binding may distort the normal stoichiometry of nELAVL interactions with its mRNA targets. Indeed, non-coding RNAs have previously been shown to affect RBP-RNA stoichiometry and therefore the biological function of other RNAs or RBPs (Borah et al., 2011; Cazalla et al., 2010;Hansen et al., 2013). Our data indicate that the binding of nELAVL to Y RNAs during stress may lead to a redistribution of nELAVL binding and/or competition of nELAVL from other RNAs. Consistently, we found that high nELAVL/Y RNA association was associated with a general decrease in nELAVL binding at a subset of binding sites, especially within introns, and consequential splicing changes were reminiscent of splicing changes provoked by nELAVL depletion. Consistently, splicing changes induced by Y RNA overexpression showed an enrichment of nELAVL binding that was dependent on the presence of the ELAVL binding motif in Y RNAs. Hence we propose that the increased association of nELAVL and Y RNAs during stress causes sequestration of nELAVL from its mRNA targets.

Taken together, our data indicate that nELAVL becomes strongly associated with Y RNAs in some AD subjects as well as in cells subjected to UV stress, and this is linked to a sequestration of nELAVL from some of its intronic targets, partially recapitulating splicing changes induced by nELAVL depletion. Our results are consistent with a hypothesis that a relatively subtle and perhaps long-term effect of Y RNA binding on normal nELAVL stoichiometry may underlie subtle and long-term changes in nELAVL biology. Perhaps analogously, the sequestration of the RBP, TDP-43, has previously been linked to neurodegenerative disorders (Lee et al., 2012). While the underlying mechanisms of TDP-43 and nELAVL sequestration are distinct, relatively subtle and long-term rearrangement of RNA:protein stoichiometry and interactions might be a recurrent theme of neurodegeneration.




Read Full Post »

Synapse activity in neurotransmission

Larry H. Bernstein, MD, FCAP, Curator



The case of the silent synapses: Why are only 20% of synapses active during neurotransmission?

Unknown information coding in the brain?
February 26, 2016   http://www.kurzweilai.net/study-finds-only-a-small-portion-of-synapses-may-be-active-during-neurotransmission

Using a fluorescent molecule to track neurotransmission of dopamine in mouse synapses, scientists made a puzzling discovery. … (credit: Sulzer Lab/Columbia University Medical Center)

Columbia University scientists recently tested a new optical technique to study how information is transmitted in the brains of mice and made a surprising discovery: When stimulated electrically to release dopamine (a neurotransmitteror chemical released by neurons, or nerve cells, to send signals to other nerve cells), only about 20 percent of synapses — the connections between cells that control brain activity — were active at any given time.

The effect had never been noticed. “Older techniques only revealed what was going on in large groups of synapses,” explained David Sulzer, PhD, professor of neurobiology in Psychiatry, Neurology, and Pharmacology at Columbia University Medical Center (CUMC). “We needed a way to observe the neurotransmitter activity of individual synapses, to help us better understand their intricate behavior.”

So Sulzer’s team turned to Dalibor Sames, PhD, associate professor of chemistry at Columbia, to develop a novel compound called “fluorescent false neurotransmitter 200″ (FFN200). When added to brain tissue or nerve cells from mice, FFN200 mimicked the brain’s natural neurotransmitters, allowing the researchers to spy on chemical messaging in action, focusing on complex tasks such as learning and memory.


Only 20% of synapses (red) were observed to transmit dopamine. The rest (green) were found to be silent. (credit: Sulzer Lab/Columbia University Medical Center)

Silent synapses: unknown information coding?

Using a fluorescence microscope, the researchers were able for the first time to view the release and re-uptake of dopamine — a neurotransmitter involved in motor learning, habit formation, and reward-seeking behavior — in individual synapses.

When all the neurons were electrically stimulated in a sample of brain tissue, the researchers expected all the synapses to release dopamine. Instead, they found that less than 20 percent of dopaminergic synapses were active following a pulse of electricity.

One possibility: these silent synapses hint at a mechanism of information coding in the brain that’s yet to be revealed, the researchers hypothesize.

The study’s authors plan to pursue that hypothesis in future experiments and examine how other neurotransmitters behave. “If we can work this out, we may learn a lot more about how alterations in dopamine levels are involved in brain disorders such as Parkinson’s disease, addiction, and schizophrenia,” Sulzer said.

The study was published in the latest issue of Nature Neuroscience.

The authors note in the paper that “the state of silent vesicle clusters may be important in disorders such as schizophrenia, which show striatal hyperdopaminergia [excessive release of dopamine in the brain’s reward center] and cortical hypodopaminergia [low amounts of dopamine in the cortex] and processes of  ‘unsilencing’ may have clinical applications for diseases such as Parkinson’s disease.”

Columbia Medical | Study Finds Only a Small Portion of Synapses May Be Active During Neurotransmission

Abstract of Fluorescent false neurotransmitter reveals functionally silent dopamine vesicle clusters in the striatum

Neurotransmission at dopaminergic synapses has been studied with techniques that provide high temporal resolution, but cannot resolve individual synapses. To elucidate the spatial dynamics and heterogeneity of individual dopamine boutons, we developed fluorescent false neurotransmitter 200 (FFN200), a vesicular monoamine transporter 2 (VMAT2) substrate that selectively traces monoamine exocytosis in both neuronal cell culture and brain tissue. By monitoring electrically evoked Ca2+ transients with GCaMP3 and FFN200 release simultaneously, we found that only a small fraction of dopamine boutons that exhibited Ca2+ influx engaged in exocytosis, a result confirmed with activity-dependent loading of the endocytic probe FM1-43. Thus, only a low fraction of striatal dopamine axonal sites with uptake-competent VMAT2 vesicles are capable of transmitter release. This is consistent with the presence of functionally ‘silent’ dopamine vesicle clusters and represents, to the best of our knowledge, the first report suggestive of presynaptically silent neuromodulatory synapses.

Read Full Post »

Neuron welding

Neuron welding

Larry H. Bernstein, MD, FCAP, Curator



How to ‘weld’ neurons with a laser

February 9, 2016



An illustration of how a femtosecond laser pulse is delivered to the target point between an axon and a neuronal soma (cell body) (credit: the authors)

University of Alberta researchers have developed a method of connecting neurons using ultrashort laser pulses. The technique gives researchers complete control over the cell connection process and could lead to new research and treatment methods, including physical reattachment of severed neurons right after injury, the researchers say.

The team’s findings are published in the open-access Nature journal Scientific Reports.


After putting two neurons in a special solution that prevents them from sticking together, the researchers brought them into contact with each other and delivered femtosecond (10-15 seconds) laser pulses to the meeting point of the two cells, causing them to establish solid bonds and form a common membrane at the targeted area.


(Left): An illustration of the phospholipid bilayers of the neuron soma and axon (the attachment region is designated with a circular spot — not the laser focal point). (Center): The laser pulse’s high intensity causes a reversible destabilization of both phospholipid layers. The generated free ions (red) and free electrons (orange) cross the center nonpolar region and break bonds between the fatty acid hydrophobic tails. (Right): The relaxation process results in the formation of new stable bonds and formation of singular, hemifused, cell membrane only at the targeted connection point. (credit: Nir Katchinskiy et al./ Scientific Reports)


The cells remained viable and the connection strong. It took the neurons just 15 milliseconds to stick to each other; the process would have taken hours to occur naturally.

“The preservation of the viability of the neural network will allow researchers to study new complex pathophysiological processes, such as neurogenesis, Wallerian degeneration, segmental demyelination, and axonal degeneration,” the authors note.


Abstract of Novel Method  for Neuronal Nanosurgical Connection

Neuronal injury may cause an irreversible damage to cellular, organ and organism function. While preventing neural injury is ideal, it is not always possible. There are multiple etiologies for neuronal injury including trauma, infection, inflammation, immune mediated disorders, toxins and hereditary conditions. We describe a novel laser application, utilizing femtosecond laser pulses, in order to connect neuronal axon to neuronal soma. We were able to maintain cellular viability, and demonstrate that this technique is universal as it is applicable to multiple cell types and media.

Read Full Post »

Beyond tau and amyloid

Larry H. Bernstein, MD, FCAP, Curator






Neurovascular pathways to neurodegeneration in Alzheimer’s disease and other disorders.

Berislav V. Zlokovic

Nature Reviews Neuroscience 12, 723-738 (December 2011) |   http:dx.doi.org:/10.1038/nrn3114

The neurovascular unit (NVU) comprises brain endothelial cells, pericytes or vascular smooth muscle cells, glia and neurons. The NVU controls blood–brain barrier (BBB) permeability and cerebral blood flow, and maintains the chemical composition of the neuronal ‘milieu’, which is required for proper functioning of neuronal circuits. Recent evidence indicates that BBB dysfunction is associated with the accumulation of several vasculotoxic and neurotoxic molecules within brain parenchyma, a reduction in cerebral blood flow, and hypoxia. Together, these vascular-derived insults might initiate and/or contribute to neuronal degeneration. This article examines mechanisms of BBB dysfunction in neurodegenerative disorders, notably Alzheimer’s disease, and highlights therapeutic opportunities relating to these neurovascular deficits.



The neurovascular unit comprises vascular cells (endothelial cells, pericytes and vascular smooth muscle cells (VSMCs)), glial cells (astrocytes, microglia and oliogodendroglia) and neurons.
Neurodegenerative disorders such as Alzheimer’s disease and amyotrophic lateral sclerosis (ALS) are associated with microvascular dysfunction and/or degeneration in the brain, neurovascular disintegration, defective blood–brain barrier (BBB) function and/or vascular factors.
The interactions between endothelial cells and pericytes are crucial for the formation and maintenance of the BBB. Indeed, pericyte deficiency leads to BBB breakdown and extravasation of multiple vasculotoxic and neurotoxic circulating macromolecules, which can contribute to neuronal dysfunction, cognitive decline and neurodegenerative changes.
Alterations in cerebrovascular metabolic functions can also lead to the secretion of multiple neurotoxic and inflammatory factors.
BBB dysfunction and/or breakdown and cerebral blood flow (CBF) reductions and/or dysregulation may occur in sporadic Alzheimer’s disease and experimental models of this disease before cognitive decline, amyloid-β deposition and brain atrophy. In patients with ALS and in some experimental models of ALS, CBF dysregulation, blood–spinal cord barrier breakdown and spinal cord hypoperfusion have been reported prior to motor neuron cell death.
Several studies in animal models of Alzheimer’s disease and, more recently, in patients with this disorder have shown diminished amyloid-β clearance from brain tissue. The recognition of amyloid-β clearance pathways opens exciting new therapeutic opportunities for this disease.
‘Multiple-target, multiple-action’ agents will stand a better chance of controlling the complex disease mechanisms that mediate neurodegeneration in disorders such as Alzheimer’s disease than will agents that have only one target. According to the vasculo-neuronal-inflammatory triad model of neurodegenerative disorders, in addition to neurons, brain endothelium, VSMCs, pericytes, astrocytes and activated microglia all represent important therapeutic targets.


Neurons depend on blood vessels for their oxygen and nutrient supplies, and for the removal of carbon dioxide and other potentially toxic metabolites from the brain’s interstitial fluid (ISF). The importance of the circulatory system to the human brain is highlighted by the fact that although the brain comprises ~2% of total body mass, it receives up to 20% of cardiac output and is responsible for ~20% and ~25% of the body’s oxygen consumption and glucose consumption, respectively1. To underline this point, when cerebral blood flow (CBF) stops, brain functions end within seconds and damage to neurons occurs within minutes2.

Neurodegenerative disorders such as Alzheimer’s disease and amyotrophic lateral sclerosis (ALS) are associated with microvascular dysfunction and/or degeneration in the brain, neurovascular disintegration, defective blood–brain barrier (BBB) function and/or vascular factors1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12. Microvascular deficits diminish CBF and, consequently, the brain’s supply of oxygen, energy substrates and nutrients. Moreover, such deficits impair the clearance of neurotoxic molecules that accumulate and/or are deposited in the ISF, non-neuronal cells and neurons. Recent evidence suggests that vascular dysfunction leads to neuronal dysfunction and neurodegeneration, and that it might contribute to the development of proteinaceous brain and cerebrovascular ‘storage’ disorders. Such disorders include cerebral β-amyloidosis and cerebral amyloid angiopathy (CAA), which are caused by accumulation of the peptide amyloid-β in the brain and the vessel wall, respectively, and are features of Alzheimer’s disease1.

In this Review, I will discuss neurovascular pathways to neurodegeneration, placing a focus on Alzheimer’s disease because more is known about neurovascular dysfunction in this disease than in other neurodegenerative disorders. The article first examines transport mechanisms for molecules to cross the BBB, before exploring the processes that are involved in BBB breakdown at the molecular and cellular levels, and the consequences of BBB breakdown, hypoperfusion, and hypoxia and endothelial metabolic dysfunction for neuronal function. Next, the article reviews evidence for neurovascular changes during normal ageing and neurovascular BBB dysfunction in various neurodegenerative diseases, including evidence suggesting that vascular defects precede neuronal changes. Finally, the article considers specific mechanisms that are associated with BBB dysfunction in Alzheimer’s disease and ALS, and therapeutic opportunities relating to these neurovascular deficits.

The neurovascular unit

The neurovascular unit (NVU) comprises vascular cells (that is, endothelium, pericytes and vascular smooth muscle cells (VSMCs)), glial cells (that is, astrocytes, microglia and oliogodendroglia) and neurons1,2, 13 (Fig. 1). In the NVU, the endothelial cells together form a highly specialized membrane around blood vessels. This membrane underlies the BBB and limits the entry of plasma components, red blood cells (RBCs) and leukocytes into the brain. The BBB also regulates the delivery into the CNS of circulating energy metabolites and essential nutrients that are required for proper neuronal and synaptic function. Non-neuronal cells and neurons act in concert to control BBB permeability and CBF. Vascular cells and glia are primarily responsible for maintenance of the constant ‘chemical’ composition of the ISF, and the BBB and the blood–spinal cord barrier (BSCB) work together with pericytes to prevent various potentially neurotoxic and vasculotoxic macromolecules in the blood from entering the CNS, and to promote clearance of these substances from the CNS1.

In the brain, pial arteries run through the subarachnoid space (SAS), which contains the cerebrospinal fluid (CSF). These vessels give rise to intracerebral arteries, which penetrate into brain parenchyma. Intracerebral arteries are separated from brain parenchyma by a single, interrupted layer of elongated fibroblast-like cells of the pia and the astrocyte-derived glia limitans membrane that forms the outer wall of the perivascular Virchow–Robin space. These arteries branch into smaller arteries and subsequently arterioles, which lose support from the glia limitans and give rise to pre-capillary arterioles and brain capillaries. In an intracerebral artery, the vascular smooth muscle cell (VSMC) layer occupies most of the vessel wall. At the brain capillary level, vascular endothelial cells and pericytes are attached to the basement membrane. Pericyte processes encase most of the capillary wall, and they communicate with endothelial cells directly through synapse-like contacts containing connexins and N-cadherin. Astrocyte end-foot processes encase the capillary wall, which is composed of endothelium and pericytes. Resting microglia have a ‘ramified’ shape and can sense neuronal injury.

Figure 2 | Blood–brain barrier transport mechanisms.

Small lipophilic drugs, oxygen and carbon dioxide diffuse across the blood–brain barrier (BBB), whereas ions require ATP-dependent transporters such as the (Na++K+)ATPase. Transporters for nutrients include the glucose transporter 1 (GLUT1; also known as solute carrier family 2, facilitated glucose transporter member 1 (SLC2A1)), the lactate transporter monocarboxylate transporter 1 (MCT1) and the L1 and y+ transporters for large neutral and cationic essential amino acids, respectively. These four transporters are expressed at both the luminal and albuminal membranes. Non-essential amino acid transporters (the alanine, serine and cysteine preferring system (ASC), and the alanine preferring system (A)) and excitatory amino acid transporter 1 (EAAT1), EAAT2 and EAAT3 are located at the abluminal side. The ATP-binding cassette (ABC) efflux transporters that are found in the endothelial cells include multidrug resistance protein 1 (ABCB1; also known as ATP-binding cassette subfamily B member 1) and solute carrier organic anion transporter family member 1C1 (OATP1C1). Finally, transporters for peptides or proteins include the endothelial protein C receptor (EPCR) for activated protein C (APC); the insulin receptors (IRs) and the transferrin receptors (TFRs), which are associated with caveolin 1 (CAV1); low-density lipoprotein receptor-related protein 1 (LRP1) for amyloid-β, peptide transport system 1 (PTS1) for encephalins; and the PTS2 and PTS4–vasopressin V1a receptor (V1AR) for arginine vasopressin.


Transport across the blood–brain barrier. The endothelial cells that form the BBB are connected by tight and adherens junctions, and it is the tight junctions that confer the low paracellular permeability of the BBB1. Small lipophilic molecules, oxygen and carbon dioxide diffuse freely across the endothelial cells, and hence the BBB, but normal brain endothelium lacks fenestrae and has limited vesicular transport.

The high number of mitochondria in endothelial cells reflects a high energy demand for active ATP-dependent transport, conferred by transporters such as the sodium pump ((Na++K+)ATPase) and the ATP-binding cassette (ABC) efflux transporters. Sodium influx and potassium efflux across the abluminal side of the BBB is controlled by (Na++K+)ATPase (Fig. 2). Changes in sodium and potassium levels in the ISF influence the generation of action potentials in neurons and thus directly affect neuronal and synaptic functions1, 12.

Brain endothelial cells express transporters that facilitate the transport of nutrients down their concentration gradients, as described in detail elsewhere1, 14 (Fig. 2). Glucose transporter 1 (GLUT1; also known as solute carrier family 2, facilitated glucose transporter member 1 (SLC2A1)) — the BBB-specific glucose transporter — is of special importance because glucose is a key energy source for the brain.

Monocarboxylate transporter 1 (MCT1), which transports lactate, and the L1 and y+ amino acid transporters are expressed at the luminal and abluminal membranes12, 14. Sodium-dependent excitatory amino acid transporter 1 (EAAT1), EAAT2 and EAAT3 are expressed at the abluminal side of the BBB15 and enable removal of glutamate, an excitatory neurotransmitter, from the brain (Fig. 2). Glutamate clearance at the BBB is essential for protecting neurons from overstimulation of glutaminergic receptors, which is neurotoxic16.

ABC transporters limit the penetration of many drugs into the brain17. For example, multidrug resistance protein 1 (ABCB1; also known as ATP-binding cassette subfamily B member 1) controls the rapid removal of ingested toxic lipophilic metabolites17 (Fig. 2). Some ABC transporters also mediate the efflux of nutrients from the endothelium into the ISF. For example, solute carrier organic anion transporter family member 1C1 (OATP1C1) transports thyroid hormones into the brain. MCT8 mediates influx of thyroid hormones from blood into the endothelium18 (Fig. 2).

The transport of circulating peptides across the BBB into the brain is restricted or slow compared with the transport of nutrients19. Carrier-mediated transport of neuroactive peptides controls their low levels in the ISF20, 21, 22, 23, 24 (Fig. 2). Some proteins, including transferrin, insulin, insulin-like growth factor 1 (IGF1), leptin25, 26, 27 and activatedprotein C (APC)28, cross the BBB by receptor-mediated transcytosis (Fig. 2).

Circumventricular organs. Several small neuronal structures that surround brain ventricles lack the BBB and sense chemical changes in blood or the cerebrospinal fluid (CSF) directly. These brain areas are known as circumventricular organs (CVOs). CVOs have important roles in multiple endocrine and autonomic functions, including the control of feeding behaviour as well as regulation of water and salt metabolism29. For example, the subfornical organ is one of the CVOs that are capable of sensing extracellular sodium using astrocyte-derived lactate as a signal for local neurons to initiate neural, hormonal and behavioural responses underlying sodium homeostasis30. Excessive sodium accumulation is detrimental, and increases in plasma sodium above a narrow range are incompatible with life, leading to cerebral oedema (swelling), seizures and death29.

Vascular-mediated pathophysiology

The key pathways of vascular dysfunction that are linked to neurodegenerative diseases include BBB breakdown, hypoperfusion–hypoxia and endothelial metabolic dysfunction (Fig. 3). This section examines processes that are involved in BBB breakdown at the molecular and cellular levels, and explores the consequences of all three pathways for neuronal function and viability.

Figure 3 | Vascular-mediated neuronal damage and neurodegeneration.

a | Blood–brain barrier (BBB) breakdown that is caused by pericyte detachment leads to leakage of serum proteins and focal microhaemorrhages, with extravasation of red blood cells (RBCs). RBCs release haemoglobin, which is a source of iron. In turn, this metal catalyses the formation of toxic reactive oxygen species (ROS) that mediate neuronal injury. Albumin promotes the development of vasogenic oedema, contributing to hypoperfusion and hypoxia of the nervous tissue, which aggravates neuronal injury. A defective BBB allows several potentially vasculotoxic and neurotoxic proteins (for example, thrombin, fibrin and plasmin) to enter the brain. b | Progressive reductions in cerebral blood flow (CBF) lead to increasing neuronal dysfunction. Mild hypoperfusion, oligaemia, leads to a decrease in protein synthesis, whereas more-severe reductions in CBF, leading to hypoxia, cause an array of detrimental effects.

Blood–brain barrier breakdown. Disruption to tight and adherens junctions, an increase in bulk-flow fluid transcytosis, and/or enzymatic degradation of the capillary basement membrane cause physical breakdown of the BBB.

The levels of many tight junction proteins, their adaptor molecules and adherens junction proteins decrease in Alzheimer’s disease and other diseases that cause dementia1, 9, ALS31, multiple sclerosis32 and various animal models of neurological disease8, 33. These decreases might be partly explained by the fact that vascular-associated matrix metalloproteinase (MMP) activity rises in many neurodegenerative disorders and after ischaemic CNS injury34, 35; tight junction proteins and basement membrane extracellular matrix proteins are substrates for these enzymes34. Lowered expression of messenger RNAs that encode several key tight junction proteins, however, has also been reported in some neurodegenerative disorders, such as ALS31.

Endothelial cell–pericyte interactions are crucial for the formation36, 37and maintenance of the BBB33, 38. Pericyte deficiency can lead to a reduction in expression of certain tight junction proteins, including occludin, claudin 5 and ZO1 (Ref. 33), and to an increase in bulk-flow transcytosis across the BBB, causing BBB breakdown38. Both processes can lead to extravasation of multiple small and large circulating macromolecules (up to 500 kDa) into the brain parenchyma33, 38. Moreover, in mice, an age-dependent progressive loss of pericytes can lead to BBB disruption and microvasular degeneration and, subsequently, neuronal dysfunction, cognitive decline and neurodegenerative changes33. In their lysosomes, pericytes concentrate and degrade multiple circulating exogenous39 and endogenous proteins, including serum immunoglobulins and fibrin33, which amplify BBB breakdown in cases of pericyte deficiency.

BBB breakdown typically leads to an accumulation of various molecules in the brain. The build up of serum proteins such as immunoglobulins and albumin can cause brain oedema and suppression of capillary blood flow8, 33, whereas high concentrations of thrombin lead to neurotoxicity and memory impairment40, and accelerate vascular damage and BBB disruption41. The accumulation of plasmin (derived from circulating plasminogen) can catalyse the degradation of neuronal laminin and, hence, promote neuronal injury42, and high fibrin levels accelerate neurovascular damage6. Finally, an increase in the number of RBCs causes deposition of haemoglobin-derived neurotoxic products including iron, which generates neurotoxic reactive oxygen species (ROS)8, 43(Fig. 3a). In addition to protein-mediated vasogenic oedema, local tissue ischaemia–hypoxia depletes ATP stores, causing (Na++K+)ATPase pumps and Na+-dependent ion channels to stop working and, consequently, the endothelium and astrocytes to swell (known as cytotoxic oedema)44. Upregulation of aquaporin 4 water channels in response to ischaemia facilitates the development of cytotoxic oedema in astrocytes45.

Hypoperfusion and hypoxia. CBF is regulated by local neuronal activity and metabolism, known as neurovascular coupling46. The pial and intracerebral arteries control the local increase in CBF that occurs during brain activation, which is termed ‘functional hyperaemia’. Neurovascular coupling requires intact pial circulation, and for VSMCs and pericytes to respond normally to vasoactive stimuli33, 46, 47. In addition to VSMC-mediated constriction and vasodilation of cerebral arteries, recent studies have shown that pericytes modulate brain capillary diameter through constriction of the vessel wall47, which obstructs capillary flow during ischaemia48. Astrocytes regulate the contractility of intracerebral arteries49, 50.

Progressive CBF reductions have increasingly serious consequences for neurons (Fig. 3b). Briefly, mild hypoperfusion — termed oligaemia — affects protein synthesis, which is required for the synaptic plasticity mediating learning and memory46. Moderate to severe CBF reductions and hypoxia affect ATP synthesis, diminishing (Na++K+)ATPase activity and the ability of neurons to generate action potentials9. In addition, such reductions can lower or increase pH, and alter electrolyte balances and water gradients, leading to the development of oedema and white matter lesions, and the accumulation of glutamate and proteinaceous toxins (for example, amyloid-β and hyperphopshorylated tau) in the brain. A reduction of greater than 80% in CBF results in neuronal death2.

The effect of CBF reductions has been extensively studied at the molecular and cellular levels in relation to Alzheimer’s disease. Reduced CBF and/or CBF dysregulation occurs in elderly individuals at high risk of Alzheimer’s disease before cognitive decline, brain atrophy and amyloid-β accumulation10, 46, 51, 52, 53, 54. In animal models, hypoperfusion can induce or amplify Alzheimer’s disease-like neuronal dysfunction and/or neuropathological changes. For example, bilateral carotid occlusion in rats causes memory impairment, neuronal dysfunction, synaptic changes and amyloid-β oligomerization55, leading to accumulation of neurotoxic amyloid-β oligomers56. In a mouse model of Alzheimer’s disease, oligaemia increases neuronal amyloid-β levels and neuronal tau phosphophorylation at an epitope that is associated with Alzheimer’s disease-type paired helical filaments57. In rodents, ischaemia leads to the accumulation of hyperphosphorylated tau in neurons and the formation of filaments that resemble those present in human neurodegenerative tauopathies and Alzheimer’s disease58. Mice expressing amyloid-β precursor protein (APP) and transforming growth factor β1 (TGFβ1) develop deficient neurovascular coupling, cholinergic denervation, enhanced cerebral and cerebrovascular amyloid-β deposition, and age-dependent cognitive decline59.

Recent studies have shown that ischaemia–hypoxia influences amyloidogenic APP processing through mechanisms that increase the activity of two key enzymes that are necessary for amyloid-β production; that is, β-secretase and γ-secretase60, 61, 62, 63. Hypoxia-inducible factor 1α (HIF1α) mediates transcriptional increase in β-secretase expression61. Hypoxia also promotes phosphorylation of tau through the mitogen-activated protein kinase (MAPK; also known as extracellular signal-regulated kinase (ERK)) pathway64, downregulates neprilysin — an amyloid-β-degrading enzyme65 — and leads to alterations in the expression of vascular-specific genes, including a reduction in the expression of the homeobox protein MOX2 gene mesenchyme homeobox 2 (MEOX2) in brain endothelial cells5 and an increase in the expression of the myocardin gene (MYOCD) in VSMCs66. In patients with Alzheimer’s disease and in models of this disorder, these changes cause vessel regression, hypoperfusion and amyloid-β accumulation resulting from the loss of the key amyloid-β clearance lipoprotein receptor (see below). In addition, hypoxia facilitates alternative splicing of Eaat2 mRNA in Alzheimer’s disease transgenic mice before amyloid-β deposition67 and suppresses glutamate reuptake by astrocytes independently of amyloid formation68, resulting in glutamate-mediated neuronal injury that is independent of amyloid-β.

In response to hypoxia, mitochondria release ROS that mediate oxidative damage to the vascular endothelium and to the selective population of neurons that has high metabolic activity. Such damage has been suggested to occur before neuronal degeneration and amyloid-β deposition in Alzheimer’s disease69, 70. Although the exact triggers of hypoxia-mediated neurodegeneration and the role of HIF1α in neurodegeneration versus preconditioning-mediated neuroprotection remain topics of debate, mitochondria-generated ROS seem to have a primary role in the regulation of the HIF1α-mediated transcriptional switch that can activate an array of responses, ranging from mechanisms that increase cell survival and adaptation to mechanisms inducing cell cycle arrest and death71. Whether inhibition of hypoxia-mediated pathogenic pathways will delay onset and/or control progression in neurodegenerative conditions such as Alzheimer’s disease remains to be determined.

When comparing the contributions of BBB breakdown and hypoperfusion to neuronal injury, it is interesting to consider Meox2+/− mice. Such animals have normal pericyte coverage and an intact BBB but a substantial perfusion deficit5 that is comparable to that found in pericyte-deficient mice that develop BBB breakdown33 Notably, however, Meox2+/− mice show less pronounced neurodegenerative changes than pericyte-deficient mice, indicating that chronic hypoperfusion–hypoxia alone can cause neuronal injury, but not to the same extent as hypoperfusion–hypoxia combined with BBB breakdown.

Endothelial neurotoxic and inflammatory factors. Alterations in cerebrovascular metabolic functions can lead to the secretion of multiple neurotoxic and inflammatory factors72, 73. For example, brain microvessels that have been isolated from individuals with Alzheimer’s disease (but not from neurologically normal age-matched and young individuals) and brain microvessels that have been treated with inflammatory proteins release neurotoxic factors that kill neurons74, 75. These factors include thrombin, the levels of which increase with the onset of Alzheimer’s disease76. Thrombin can injure neurons directly40and indirectly by activating microglia and astrocytes73. Compared with those from age-matched controls, brain microvessels from individuals with Alzheimer’s disease secrete increased levels of multiple inflammatory mediators, such as nitric oxide, cytokines (for example, tumour necrosis factor (TNF), TGFβ1, interleukin-1β (IL-1β) and IL-6), chemokines (for example, CC-chemokine ligand 2 (CCL2; also known as monocyte chemoattractant protein 1 (MCP1)) and IL-8), prostaglandins, MMPs and leukocyte adhesion molecules73. Endothelium-derived neurotoxic and inflammatory factors together provide a molecular link between vascular metabolic dysfunction, neuronal injury and inflammation in Alzheimer’s disease and, possibly, in other neurodegenerative disorders.

Neurovascular changes

This section examines evidence for neurovascular changes during normal ageing and for neurovascular and/or BBB dysfunction in various neurodegenerative diseases, as well as the possibility that vascular defects can precede neuronal changes.

Age-associated neurovascular changes. Normal ageing diminishes brain circulatory functions, including a detectable decay of CBF in the limbic and association cortices that has been suggested to underlie age-related cognitive changes77. Alterations in the cerebral microvasculature, but not changes in neural activity, have been shown to lead to age-dependent reductions in functional hyperaemia in the visual system in cats78 and in the sensorimotor cortex in pericyte-deficient mice33. Importantly, a recent longitudinal CBF study in neurologically normal individuals revealed that people bearing the apolipoprotein E (APOE) ɛ4allele — the major genetic risk factor for late-onset Alzheimer’s disease79, 80, 81 — showed greater regional CBF decline in brain regions that are particularly vulnerable to pathological changes in Alzheimer’s disease than did people without this allele82.

A meta-analysis of BBB permeability in 1,953 individuals showed that neurologically healthy humans had an age-dependent increase in vascular permeability83. Moreover, patients with vascular or Alzheimer’s disease-type dementia and leucoaraiosis — a small-vessel disease of the cerebral white matter — had an even greater age-dependent increase in vascular permeability83. Interestingly, an increase in BBB permeability in brain areas with normal white matter in patients with leukoaraiosis has been suggested to play a causal part in disease and the development of lacunar strokes84. Age-related changes in the permeability of the blood–CSF barrier and the choroid plexus have been reported in sheep85.

Vascular pathology. Patients with Alzheimer’s disease or other dementia-causing diseases frequently show focal changes in brain microcirculation. These changes include the appearance of string vessels (collapsed and acellular membrane tubes), a reduction in capillary density, a rise in endothelial pinocytosis, a decrease in mitochondrial content, accumulation of collagen and perlecans in the basement membrane, loss of tight junctions and/or adherens junctions3, 4, 5, 6, 9,46, 86, and BBB breakdown with leakage of blood-borne molecules4, 6,7, 9. The time course of these vascular alterations and how they relate to dementia and Alzheimer’s disease pathology remain unclear, as no protocol that allows the development of the diverse brain vascular pathology to be scored, and hence to be tracked with ageing, has so far been developed and widely validated87. Interestingly, a recent study involving 500 individuals who died between the ages of 69 and 103 years showed that small-vessel disease, infarcts and the presence of more than one vascular pathological change were associated with Alzheimer’s disease-type pathological lesions and dementia in people aged 75 years of age87. These associations were, however, less pronounced in individuals aged 95 years of age, mainly because of a marked ageing-related reduction in Alzheimer’s disease neuropathology relative to a moderate but insignificant ageing-related reduction in vascular pathology87.

Accumulation of amyloid-β and amyloid deposition in pial and intracerebral arteries results in CAA, which is present in over 80% of Alzheimer’s disease cases88. In patients who have Alzheimer’s disease with established CAA in small arteries and arterioles, the VSMC layer frequently shows atrophy, which causes a rupture of the vessel wall and intracerebral bleeding in about 30% of these patients89, 90. These intracerebral bleedings contribute to, and aggravate, dementia. Patients with hereditary cerebral β-amyloidosis and CAA of the Dutch, Iowa, Arctic, Flemish, Italian or Piedmont L34V type have accelerated VSMC degeneration resulting in haemorrhagic strokes and dementia91. Duplication of the gene encoding APP causes early-onset Alzheimer’s disease dementia with CAA and intracerebral haemorrhage92.

Early studies of serum immunoglobulin leakage reported that patients with ALS had BSCB breakdown and BBB breakdown in the motor cortex93. Microhaemorrhages and BSCB breakdown have been shown in the spinal cord of transgenic mice expressing mutant variants of human superoxide dismutase 1 (SOD1), which in mice cause an ALS-like disease8, 94, 95. In mice with ALS-like disease and in patients with ALS, BSCB breakdown has been shown to occur before motor neuron degeneration or brain atrophy8, 11, 95.

BBB breakdown in the substantia nigra and the striatum has been detected in murine models of Parkinson’s disease that are induced by administration of 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP)96, 97, 98. However, the temporal relationship between BBB breakdown and neurodegeneration in Parkinson’s disease is currently unknown. Notably, the prevalence of CAA and vascular lesions increases in Parkinson’s disease99, 100. Vascular lesions in the striatum and lacunar infarcts can cause vascular parkinsonism syndrome101. A recent study reported BBB breakdown in a rat model of Huntington’s disease that is induced with the toxin 3-nitropropionic acid102.

Several studies have established disruption of BBB with a loss of tight junction proteins during neuroinflammatory conditions such as multiple sclerosis and its murine model, experimental allergic encephalitis. Such disruption facilitates leukocyte infiltration, leading to oliogodendrocyte death, axonal damage, demyelination and lesion development32.

Functional changes in the vasculature. In individuals with Alzheimer’s disease, GLUT1 expression at the BBB decreases103, suggesting a shortage in necessary metabolic substrates. Studies using18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) have identified reductions in glucose uptake in asymptomatic individuals with a high risk of dementia104, 105. Several studies have suggested that reduced glucose uptake across the BBB, as seen by FDG PET, precedes brain atrophy104, 105, 106, 107, 108.

Amyloid-β constricts cerebral arteries109. In a mouse model of Alzheimer’s disease, impairment of endothelium-dependent regulation of neocortical microcirculation110, 111 occurs before amyloid-β accumulation. Recent studies have shown that CD36, a scavenger receptor that binds amyloid-β, is essential for the vascular oxidative stress and diminished functional hyperaemia that occurs in response to amyloid-β exposure112. Neuroimaging studies in patients with Alzheimer’s disease have shown that neurovascular uncoupling occurs before neurodegenerative changes10, 51, 52, 53. Moreover, cognitively normal APOE ɛ4 carriers at risk of Alzheimer’s disease show impaired CBF responses to brain activation in the absence of neurodegenerative changes or amyloid-β accumulation54. Recently, patients with Alzheimer’s disease as well as mouse models of this disease with high cerebrovascular levels of serum response factor (SRF) and MYOCD, the two transcription factors that control VSMC differentiation, have been shown to develop a hypercontractile arterial phenotype resulting in brain hypoperfusion, diminished functional hyperaemia and CAA66, 113. More work is needed to establish the exact role of SRF and MYOCD in the vascular dysfunction that results in the Alzheimer’s disease phenotype and CAA.

PET studies with 11C-verapamil, an ABCB1 substrate, have indicated that the function of ABCB1, which removes multiple drugs and toxins from the brain, decreases with ageing114 and is particularly compromised in the midbrain of patients with Parkinson’s disease, progressive supranuclear palsy or multiple system atrophy115. More work is needed to establish the exact roles of ABC BBB transporters in neurodegeneration and whether their failure precedes the loss of dopaminergic neurons that occurs in Parkinson’s disease.

In mice with ALS-like disease and in patients with ALS, hypoperfusion and/or dysregulated CBF have been shown to occur before motor neuron degeneration or brain atrophy8, 116. Reduced regional CBF in basal ganglia and reduced blood volume have been reported in pre-symptomatic gene-tested individuals at risk for Huntington’s disease117. Patients with Huntington’s disease display a reduction in vasomotor activity in the cerebral anterior artery during motor activation118.

Vascular and neuronal common growth factors. Blood vessels and neurons share common growth factors and molecular pathways that regulate their development and maintenance119, 120. Angioneurins are growth factors that exert both vasculotrophic and neurotrophic activities121. The best studied angioneurin is vascular endothelial growth factor (VEGF). VEGF regulates vessel formation, axonal growth and neuronal survival120. Ephrins, semaphorins, slits and netrins are axon guidance factors that also regulate the development of the vascular system121. During embryonic development of the neural tube, blood vessels and choroid plexus secrete IGF2 into the CSF, which regulates the proliferation of neuronal progenitor cells122. Genetic and pharmacological manipulations of angioneurin activity yielded various vascular and cerebral phenotypes121. Given the dual nature of angioneurin action, these studies have not been able to address whether neuronal dysfunction results from a primary insult to neurons and/or whether it is secondary to vascular dysfunction.

Increased levels of VEGF, a hypoxia-inducible angiogenic factor, were found in the walls of intraparenchymal vessels, perivascular deposits, astrocytes and intrathecal space of patients with Alzheimer’s disease, and were consistent with the chronic cerebral hypoperfusion and hypoxia that were observed in these individuals73. In addition to VEGF, brain microvessels in Alzheimer’s disease release several molecules that can influence angiogenesis, including IL-1β, IL-6, IL-8, TNF, TGFβ, MCP1, thrombin, angiopoietin 2, αVβ3 and αVβ5 integrins, and HIF1α73. However, evidence for increased vascularity in Alzheimer’s disease is lacking. On the contrary, several studies have reported that focal vascular regression and diminished microvascular density occur in Alzheimer’s disease4, 5, 73 and in Alzheimer’s disease transgenic mice123. The reason for this discrepancy is not clear. The anti-angiogenic activity of amyloid-β, which accumulates in the brains of individuals with Alzheimer’s disease and Alzheimer’s disease models, may contribute to hypovascularity123. Conversely, genome-wide transcriptional profiling of brain endothelial cells from patients with Alzheimer’s disease revealed that extremely low expression of vascular-restricted MEOX2 mediates aberrant angiogenic responses to VEGF and hypoxia, leading to capillary death5. This finding raises the interesting question of whether capillary degeneration in Alzheimer’s disease results from unsuccessful vascular repair and/or remodelling. Moreover, mice that lack one Meox2 allele have been shown to develop a primary cerebral endothelial hypoplasia with chronic brain hypoperfusion5, resulting in secondary neurodegenerative changes33.

Does vascular dysfunction cause neuronal dysfunction? In summary, the evidence that is discussed above clearly indicates that vascular dysfunction is tightly linked to neuronal dysfunction. There are many examples to illustrate that primary vascular deficits lead to secondary neurodegeneration, including CADASIL (cerebral autosomal dominant arteriopathy with subcortical infarcts), an hereditary small-vessel brain disease resulting in multiple small ischaemic lesions, neurodegeneration and dementia124; mutations in SLC2A1 that cause dysfunction of the BBB-specific GLUT1 transporter in humans resulting in seizures; cognitive impairment and microcephaly125; microcephaly and epileptiform discharges in mice with genetic deletion of a single Slc2a1allele126; and neurodegeneration mediated by a single Meox2 homebox gene deletion restricted to the vascular system33. Patients with hereditary cerebral β-amyloidosis and CAA of the Dutch, Iowa, Arctic, Flemish, Italian or Piedmont L34V type provide another example showing that primary vascular dysfunction — which in this case is caused by deposition of vasculotropic amyloid-β mutants in the arterial vessel wall — leads to dementia and intracerebral bleeding. Moreover, as reviewed in the previous sections, recent evidence suggests that BBB dysfunction and/or breakdown, and CBF reductions and/or dysregulation may occur in sporadic Alzheimer’s disease and experimental models of this disease before cognitive decline, amyloid-β deposition and brain atrophy. In patients with ALS and in some experimental models of ALS, CBF dysregulation, BSCB breakdown and spinal cord hypoperfusion have been reported to occur before motor neuron cell death. Whether neurological changes follow or precede vascular dysfunction in Parkinson’s disease, Huntington’s disease and multiple sclerosis remains less clear. However, there is little doubt that vascular injury mediates, amplifies and/or lowers the threshold for neuronal dysfunction and loss in several neurological disorders.

Disease-specific considerations

This section examines how amyloid-β levels are kept low in the brain, a process in which the BBB has a central role, and how faulty BBB-mediated clearance mechanisms go awry in Alzheimer’s disease. On the basis of this evidence and the findings discussed elsewhere in the Review, a new hypothesis for the pathogenesis of Alzheimer’s disease that incorporates the vascular evidence is presented. ALS-specific disease mechanisms relating to the BBB are then examined.

Alzheimer’s disease. Amyloid-β clearance from the brain by the BBB is the best studied example of clearance of a proteinaceous toxin from the CNS. Multiple pathways regulate brain amyloid-β levels, including its production and clearance (Fig. 4). Recent studies127, 128, 129 have confirmed earlier findings in multiple rodent and non-human primate models demonstrating that peripheral amyloid-β is an important precursor of brain amyloid-β130, 131, 132, 133, 134, 135, 136. Moreover, peripheral amyloid-β sequestering agents such as soluble LRP1 (ref.137), anti-amyloid-β antibodies138, 139, 140, gelsolin and the ganglioside GM1 (Ref. 141), or systemic expression of neprilysin142, 143have been shown to reduce the amyloid burden in Alzheimer’s disease mice by eliminating contributions of the peripheral amyloid-β pool to the total brain pool of this peptide.

Figure 4 | The role of blood–brain barrier transport in brain homeostasis of amyloid-β.

Amyloid-β (Aβ) is produced from the amyloid-β precursor protein (APP), both in the brain and in peripheral tissues. Clearance of amyloid-β from the brain normally maintains its low levels in the brain. This peptide is cleared across the blood–brain barrier (BBB) by the low-density lipoprotein receptor-related protein 1 (LRP1). LRP1 mediates rapid efflux of a free, unbound form of amyloid-β and of amyloid-β bound to apolipoprotein E2 (APOE2), APOE3 or α2-macroglobulin (not shown) from the brain’s interstitial fluid into the blood, and APOE4 inhibits such transport. LRP2 eliminates amyloid-β that is bound to clusterin (CLU; also known as apolipoprotein J (APOJ)) by transport across the BBB, and shows a preference for the 42-amino-acid form of this peptide. ATP-binding cassette subfamily A member 1 (ABCA1; also known as cholesterol efflux regulatory protein) mediates amyloid-β efflux from the brain endothelium to blood across the luminal side of the BBB (not shown). Cerebral endothelial cells, pericytes, vascular smooth muscle cells, astrocytes, microglia and neurons express different amyloid-β-degrading enzymes, including neprilysin (NEP), insulin-degrading enzyme (IDE), tissue plasminogen activator (tPA) and matrix metalloproteinases (MMPs), which contribute to amyloid-β clearance. In the circulation, amyloid-β is bound mainly to soluble LRP1 (sLRP1), which normally prevents its entry into the brain. Systemic clearance of amyloid-β is mediated by its removal by the liver and kidneys. The receptor for advanced glycation end products (RAGE) provides the key mechanism for influx of peripheral amyloid-β into the brain across the BBB either as a free, unbound plasma-derived peptide and/or by amyloid-β-laden monocytes. Faulty vascular clearance of amyloid-β from the brain and/or an increased re-entry of peripheral amyloid-β across the blood vessels into the brain can elevate amyloid-β levels in the brain parenchyma and around cerebral blood vessels. At pathophysiological concentrations, amyloid-β forms neurotoxic oligomers and also self-aggregates, which leads to the development of cerebral β-amyloidosis and cerebral amyloid angiopathy.

The receptor for advanced glycation end products (RAGE) mediates amyloid-β transport in brain and the propagation of its toxicity. RAGE expression in brain endothelium provides a mechanism for influx of amyloid-β144, 145 and amyloid-β-laden monocytes146 across the BBB, as shown in Alzheimer’s disease models (Fig. 4). The amyloid-β-rich environment in Alzheimer’s disease and models of this disorder increases RAGE expression at the BBB and in neurons147, 148, amplifying amyloid-β-mediated pathogenic responses. Blockade of amyloid-β–RAGE signalling in Alzheimer’s disease is a promising strategy to control self-propagation of amyloid-β-mediated injury.

Several studies in animal models of Alzheimer’s disease and, more recently, in patients with this disorder149 have shown that diminished amyloid-β clearance occurs in brain tissue in this disease. LRP1 plays an important part in the three-step serial clearance of this peptide from brain and the rest of the body150 (Fig. 4). In step one, LRP1 in brain endothelium binds brain-derived amyloid-β at the abluminal side of the BBB, initiating its clearance to blood, as shown in many animal models151, 152, 153, 154, 155, 156 and BBB models in vitro151, 157,158. The vasculotropic mutants of amyloid-β that have low binding affinity for LRP1 are poorly cleared from the brain or CSF151, 159, 160. APOE4, but not APOE3 or APOE2, blocks LRP1-mediated amyloid-β clearance from the brain and, hence, promotes its retention161, whereas clusterin (also known as apolipoprotein J (APOJ)) mediates amyloid-β clearance across the BBB via LRP2 (Ref. 153). APOE and clusterin influence amyloid-β aggregation162, 163. Reduced LRP1 levels in brain microvessels, perhaps in addition to altered levels of ABCB1, are associated with amyloid-β cerebrovascular and brain accumulation during ageing in rodents, non-human primates, humans, Alzheimer’s disease mice and patients with Alzheimer’s disease66, 151, 152, 164, 165, 166. Moreover, recent work has shown that brain LRP1 is oxidized in Alzheimer’s disease167, and may contribute to amyloid-β retention in brain because the oxidized form cannot bind and/or transport amyloid-β137. LRP1 also mediates the removal of amyloid-β from the choroid plexus168.

In step two, circulating soluble LRP1 binds more than 70% of plasma amyloid-β in neurologically normal humans137. In patients with Alzheimer’s disease or mild cognitive impairment (MCI), and in Alzheimer’s disease mice, amyloid-β binding to soluble LRP1 is compromised due to oxidative changes137, 169, resulting in elevated plasma levels of free amyloid-β isoforms comprising 40 or 42 amino acids (amyloid-β1–40 and amyloid-β1–42). These peptides can then re-enter the brain, as has been shown in a mouse model of Alzheimer’s disease137. Rapid systemic removal of amyloid-β by the liver is also mediated by LRP1 and comprises step three of the clearance process170.

In brain, amyloid-β is enzymatically degraded by neprilysin171, insulin-degrading enzyme172, tissue plasminogen activator173 and MMPs173,174 in various cell types, including endothelial cells, pericytes, astrocytes, neurons and microglia. Cellular clearance of this peptide by astrocytes and VSMCs is mediated by LRP1 and/or another lipoprotein receptor66, 175. Clearance of amyloid-β aggregates by microglia has an important role in amyloid-β-directed immunotherapy176 and reduction of the amyloid load in brain177. Passive ISF–CSF bulk flow and subsequent clearance through the CSF might contribute to 10–15% of total amyloid-β removal152, 153, 178. In the injured human brain, increasing soluble amyloid-β concentrations in the ISF correlated with improvements in neurological status, suggesting that neuronal activity might regulate extracellular amyloid-β levels179.

The role of BBB dysfunction in amyloid-β accumulation, as discussed above, underlies the contribution of vascular dysfunction to Alzheimer’s disease (see Fig. 5 for a model of vascular damage in Alzheimer’s disease). The amyloid hypothesis for the pathogenesis of Alzheimer’s disease maintains that this peptide initiates a cascade of events leading to neuronal injury and loss and, eventually, dementia180, 181. Here, I present an alternative hypothesis — the two-hit vascular hypothesis of Alzheimer’s disease — that incorporates the vascular contribution to this disease, as discussed in this Review (Box 1). This hypothesis states that primary damage to brain microcirculation (hit one) initiates a non-amyloidogenic pathway of vascular-mediated neuronal dysfunction and injury, which is mediated by BBB dysfunction and is associated with leakage and secretion of multiple neurotoxic molecules and/or diminished brain capillary flow that causes multiple focal ischaemic or hypoxic microinjuries. BBB dysfunction also leads to impairment of amyloid-β clearance, and oligaemia leads to increased amyloid-β generation. Both processes contribute to accumulation of amyloid-β species in the brain (hit two), where these peptides exert vasculotoxic and neurotoxic effects. According to the two-hit vascular hypothesis of Alzheimer’s disease, tau pathology develops secondary to vascular and/or amyloid-β injury.

Figure 5 | A model of vascular damage in Alzheimer’s disease.

a | In the early stages of Alzheimer’s disease, small pial and intracerebral arteries develop a hypercontractile phenotype that underlies dysregulated cerebral blood flow (CBF). This phenotype is accompanied by diminished amyloid-β clearance by the vascular smooth muscle cells (VSMCs). In the later phases of Alzheimer’s disease, amyloid deposition in the walls of intracerebral arteries leads to cerebral amyloid angiopathy (CAA), pronounced reductions in CBF, atrophy of the VSMC layer and rupture of the vessels causing microbleeds. b | At the level of capillaries in the early stages of Alzheimer’s disease, blood–brain barrier (BBB) dysfunction leads to a faulty amyloid-β clearance and accumulation of neurotoxic amyloid-β oligomers in the interstitial fluid (ISF), microhaemorrhages and accumulation of toxic blood-derived molecules (that is, thrombin and fibrin), which affect synaptic and neuronal function. Hyperphosphorylated tau (p-tau) accumulates in neurons in response to hypoperfusion and/or rising amyloid-β levels. At this point, microglia begin to sense neuronal injury. In the later stages of the disease in brain capillaries, microvascular degeneration leads to increased deposition of basement membrane proteins and perivascular amyloid. The deposited proteins and amyloid obstruct capillary blood flow, resulting in failure of the efflux pumps, accumulation of metabolic waste products, changes in pH and electrolyte composition and, subsequently, synaptic and neuronal dysfunction. Neurofibrillary tangles (NFTs) accumulate in response to ischaemic injury and rising amyloid-β levels. Activation of microglia and astrocytes is associated with a pronounced inflammatory response. ROS, reactive oxygen species.

Amyotrophic lateral sclerosis. The cause of sporadic ALS, a fatal adult-onset motor neuron neurodegenerative disease, is not known182. In a relatively small number of patients with inherited SOD1 mutations, the disease is caused by toxic properties of mutant SOD1 (Ref. 183). Mutations in the genes encoding ataxin 2 and TAR DNA-binding protein 43 (TDP43) that cause these proteins to aggregate have been associated with ALS182, 184. Some studies have suggested that abnormal SOD1 species accumulate in sporadic ALS185. Interestingly, studies in ALS transgenic mice expressing a mutant version of human SOD1 in neurons, and in non-neuronal cells neighbouring these neurons, have shown that deletion of this gene from neurons does not influence disease progression186, whereas deletion of this gene from microglia186 or astrocytes187 substantially increases an animal’s lifespan. According to an emerging hypothesis of ALS that is based on studies in SOD1 mutant mice, the toxicity that is derived from non-neuronal neighbouring cells, particularly microglia and astrocytes, contributes to disease progression and motor neuron degeneration186, 187, 188, 189, 190, whereas BBB dysfunction might be critical for disease initiation8, 43, 94, 95. More work is needed to determine whether this concept of disease initiation and progression may also apply to cases of sporadic ALS.

Human data support a role for angiogenic factors and vessels in the pathogenesis of ALS. For example, the presence of VEGF variations (which were identified in large meta-analysis studies) has been linked to ALS191. Angiogenin is another pro-angiogenic gene that is implicated in ALS because heterozygous missense mutations in angiogenin cause familial and sporadic ALS192. Moreover, mice with a mutation that eliminates hypoxia-responsive induction of the Vegf gene (Vegfδ/δ mice) develop late-onset motor neuron degeneration193. Spinal cord ischaemia worsens motor neuron degeneration and functional outcome in Vegfδ/δmice, whereas the absence of hypoxic induction of VEGF in mice that develop motor neuron disease from expression of ALS-linked mutant SOD1G93A results in substantially reduced survival191.

Therapeutic opportunities

Many investigators believe that primary neuronal dysfunction resulting from an intrinsic neuronal disorder is the key underlying event in human neurodegenerative diseases. Thus, most therapeutic efforts for neurodegenerative diseases have so far been directed at the development of so-called ‘single-target, single-action’ agents to target neuronal cells directly and reverse neuronal dysfunction and/or protect neurons from injurious insults. However, most preclinical and clinical studies have shown that such drugs are unable to cure or control human neurological disorders2, 181, 183, 194, 195. For example, although pathological overstimulation of glutaminergic NMDA receptors (NMDARs) has been shown to lead to neuronal injury and death in several disorders, including stroke, Alzheimer’s disease, ALS and Huntington’s disease16, NMDAR antagonists have failed to show a therapeutic benefit in the above-mentioned human neurological disorders.

Recently, my colleagues and I coined the term vasculo-neuronal-inflammatory triad195 to indicate that vascular damage, neuronal injury and/or neurodegeneration, and neuroinflammation comprise a common pathological triad that occurs in multiple neurological disorders. In line with this idea, it is conceivable that ‘multiple-target, multiple-action’ agents (that is, drugs that have more than one target and thus have more than one action) will have a better chance of controlling the complex disease mechanisms that mediate neurodegeneration than agents that have only one target (for example, neurons). According to the vasculo-neuronal-inflammatory triad model, in addition to neurons, brain endothelium, VSMCs, pericytes, astrocytes and activated microglia are all important therapeutic targets.

Here, I will briefly discuss a few therapeutic strategies based on the vasculo-neuronal-inflammatory triad model. VEGF and other angioneurins may have multiple targets, and thus multiple actions, in the CNS120. For example, preclinical studies have shown that treatment of SOD1G93A rats with intracerebroventricular VEGF196 or intramuscular administration of a VEGF-expressing lentiviral vector that is transported retrogradely to motor neurons in SOD1G93A mice197 reduced pathology and extended survival, probably by promoting angiogenesis and increasing the blood flow through the spinal cord as well as through direct neuronal protective effects of VEGF on motor neurons. On the basis of these and other studies, a phase I–II clinical trial has been initiated to evaluate the safety of intracerebroventricular infusion of VEGF in patients with ALS198. Treatment with angiogenin also slowed down disease progression in a mouse model of ALS199.

IGF1 delivery has been shown to promote amyloid-β vascular clearance and to improve learning and memory in a mouse model of Alzheimer’s disease200. Local intracerebral implantation of VEGF-secreting cells in a mouse model of Alzheimer’s disease has been shown to enhance vascular repair, reduce amyloid burden and improve learning and memory201. In contrast to VEGF, which can increase BBB permeability, TGFβ, hepatocyte growth factor and fibroblast growth factor 2 promote BBB integrity by upregulating the expression of endothelial junction proteins121 in a similar way to APC43. However, VEGF and most growth factors do not cross the BBB, so the development of delivery strategies such as Trojan horses is required for their systemic use25.

A recent experimental approach with APC provides an example of a neurovascular medicine that has been shown to favourably regulate multiple pathways in non-neuronal cells and neurons, resulting in vasculoprotection, stabilization of the BBB, neuroprotection and anti-inflammation in several acute and chronic models of the CNS disorders195 (Box 2).

The recognition of amyloid-β clearance pathways (Fig. 4), as discussed above, opens exciting new therapeutic opportunities for Alzheimer’s disease. Amyloid-β clearance pathways are promising therapeutic targets for the future development of neurovascular medicines because it has been shown both in animal models of Alzheimer’s disease1 and in patients with sporadic Alzheimer’s disease149 that faulty clearance from brain and across the BBB primarily determines amyloid-β retention in brain, causing the formation of neurotoxic amyloid-β oligomers56 and the promotion of brain and cerebrovascular amyloidosis3. The targeting of clearance mechanisms might also be beneficial in other diseases; for example, the clearance of extracellular mutant SOD1 in familial ALS, the prion protein in prion disorders and α-synuclein in Parkinson’s disease might all prove beneficial. However, the clearance mechanisms for these proteins in these diseases are not yet understood.

Conclusions and perspectives

Currently, no effective disease-modifying drugs are available to treat the major neurodegenerative disorders202, 203, 204. This fact leads to a question: are we close to solving the mystery of neurodegeneration? The probable answer is yes, because the field has recently begun to recognize that, first, damage to neuronal cells is not the sole contributor to disease initiation and progression, and that, second, correcting disease pathways in vascular and glial cells may offer an array of new approaches to control neuronal degeneration that do not involve targeting neurons directly. These realizations constitute an important shift in paradigm that should bring us closer to a cure for neurodegenerative diseases. Here, I raise some issues concerning the existing models of neurodegeneration and the new neurovascular paradigm.

The discovery of genetic abnormalities and associations by linkage analysis or DNA sequencing has broadened our understanding of neurodegeneration204. However, insufficient effort has been made to link genetic findings with disease biology. Another concern for neurodegenerative research is how we should interpret findings from animal models202. Genetically engineered models of human neurodegenerative disorders in Drosophila melanogaster andCaenorhabditis elegans have been useful for dissecting basic disease mechanisms and screening compounds. However, in addition to having much simpler nervous systems, insects and avascular species do not have cerebrovascular and immune systems that are comparable to humans and, therefore, are unlikely to replicate the complex disease pathology that is found in people.

For most neurodegenerative disorders, early steps in the disease processes remain unclear, and biomarkers for these stages have yet to be identified. Thus, it is difficult to predict whether mammalian models expressing human genes and proteins that we know are implicated in the intermediate or later stages of disease pathophysiology, such as amyloid-β or tau in Alzheimer’s disease7, 181, will help us to discover therapies for the early stages of disease and for disease prevention, because the exact role of these pathological accumulations during disease onset remains uncertain. According to the two-hit vascular hypothesis of Alzheimer’s disease, incorporating vascular factors that are associated with Alzheimer’s disease into current models of this disease may more faithfully replicate dementia events in humans. Alternatively, by focusing on the comorbidities and the initial cellular and molecular mechanisms underlying early neurovascular dysfunction that are associated with Alzheimer’s disease, new models of dementia and neurodegeneration may be developed that do not require the genetic manipulation of amyloid-β or tau expression.

The proposed neurovascular triad model of neurodegenerative diseases challenges the traditional neurocentric view of such disorders. At the same time, this model raises a set of new important issues that require further study. For example, the molecular basis of the neurovascular link with neurodegenerative disorders is poorly understood, in terms of the adhesion molecules that keep the physical association of various cell types together, the molecular crosstalk between different cell types (including endothelial cells, pericytes and astrocytes) and how these cellular interactions influence neuronal activity. Addressing these issues promises to create new opportunities not only to better understand the molecular basis of the neurovascular link with neurodegeneration but also to develop novel neurovascular-based medicines.

The construction of a human BBB molecular atlas will be an important step towards understanding the role of the BBB and neurovascular interactions in health and disease. Achievement of this goal will require identifying new BBB transporters by using genomic and proteomic tools, and by cloning some of the transporters that are already known. Better knowledge of transporters at the human BBB will help us to better understand their potential as therapeutic targets for disease.

Development of higher-resolution imaging methods to evaluate BBB integrity, key transporters’ functions and CBF responses in the microregions of interest (for example, in the entorhinal region of the hippocampus) will help us to understand how BBB dysfunction correlates with cognitive outcomes and neurodegenerative processes in MCI, Alzheimer’s disease and related disorders.

The question persists: are we missing important therapeutic targets by studying the nervous system in isolation from the influence of the vascular system? The probable answer is yes. However, the current exciting and novel research that is based on the neurovascular model has already begun to change the way that we think about neurodegeneration, and will continue to provide further insights in the future, leading to the development of new neurovascular therapies.


  1. Zlokovic, B. V. The blood–brain barrier in health and chronic neurodegenerative disorders. Neuron 57, 178–201 (2008).

  2. Moskowitz, M. A., Lo, E. H. & Iadecola, C. The science of stroke: mechanisms in search of treatments. Neuron 67, 181–198 (2010).
    A comprehensive review describing mechanisms of ischaemic injury to the neurovascular unit.

  3. Zlokovic, B. V. Neurovascular mechanisms of Alzheimer’s neurodegeneration. Trends Neurosci. 28, 202–208 (2005).

  4. Brown, W. R. & Thore, C. R. Review: cerebral microvascular pathology in ageing and neurodegeneration. Neuropathol. Appl. Neurobiol. 37, 56–74 (2011).

  5. Wu, Z. et al. Role of the MEOX2 homeobox gene in neurovascular dysfunction in Alzheimer disease. Nature Med. 11, 959–965 (2005).
    A study demonstrating that low expression of MEOX2 in brain endothelium leads to aberrant angiogenesis and vascular regression in Alzheimer’s disease.

  6. Paul, J., Strickland, S. & Melchor, J. P. Fibrin deposition accelerates neurovascular damage and neuroinflammation in mouse models of Alzheimer’s disease. J. Exp. Med. 204, 1999–2008 (2007).
    A study showing BBB breakdown in models of Alzheimer’s disease.

  7. Zipser, B. D. et al. Microvascular injury and blood–brain barrier leakage in Alzheimer’s disease. Neurobiol. Aging 28, 977–986 (2007).

  8. Zhong, Z. et al. ALS-causing SOD1 mutants generate vascular changes prior to motor neuron degeneration. Nature Neurosci. 11, 420–422 (2008).
    A study demonstrating that BSCB defects precede motor neuron degeneration in mice that develop an ALS-like disease.

  9. Kalaria, R. N. Vascular basis for brain degeneration: faltering controls and risk factors for dementia. Nutr. Rev. 68, S74–S87 (2010).

  10. Knopman, D. S. & Roberts, R. Vascular risk factors: imaging and neuropathologic correlates. J. Alzheimers Dis. 20, 699–709 (2010).

  11. Miyazaki, K. et al. Disruption of neurovascular unit prior to motor neuron degeneration in amyotrophic lateral sclerosis. J. Neurosci. Res. 89, 718–728 (2011).

  12. Neuwelt, E. A. et al. Engaging neuroscience to advance translational research in brain barrier biology. Nature Rev. Neurosci. 12, 169–182 (2011).

  13. Guo, S. & Lo, E. H. Dysfunctional cell–cell signaling in the neurovascular unit as a paradigm for central nervous system disease.Stroke 40, S4–S7 (2009).

  14. Redzic, Z. Molecular biology of the blood–brain and the blood–cerebrospinal fluid barriers: similarities and differences. Fluids Barriers CNS 8, 3 (2011).

  15. O’Kane, R. L., Martinez-Lopez, I., DeJoseph, M. R., Vina, J. R. & Hawkins, R. A. Na+-dependent glutamate transporters (EAAT1, EAAT2, and EAAT3) of the blood–brain barrier. A mechanism for glutamate removal. J. Biol. Chem. 274, 31891–31895 (1999).

………   212

Author affiliations

  1. Department of Physiology and Biophysics, and Center for Neurodegeneration and Regeneration at the Zilkha Neurogenetic Institute, University of Southern California, Keck School of Medicine, 1501 San Pablo Street, Los Angeles, California 90089, USA.
    Email: bzlokovi@usc.edu


Retromer in Alzheimer disease, Parkinson disease and other neurological disorders.

Scott A. Small and Gregory A. Petsko

Nature Reviews Neuroscience  2015; 16:126-132.   http://dx.doi.org:/10.1038/nrn3896


Retromer is a protein assembly that has a central role in endosomal trafficking, and retromer dysfunction has been linked to a growing number of neurological disorders. First linked to Alzheimer disease, retromer dysfunction causes a range of pathophysiological consequences that have been shown to contribute to the core pathological features of the disease. Genetic studies have established that retromer dysfunction is also pathogenically linked to Parkinson disease, although the biological mechanisms that mediate this link are only now being elucidated. Most recently, studies have shown that retromer is a tractable target in drug discovery for these and other disorders of the nervous system.

Yeast has proved to be an informative model organism in cell biology and has provided early insight into much of the molecular machinery that mediates the intracellular transport of proteins1,2. Indeed, the term ‘retromer’ was first introduced in a yeast study in 1998 (Ref. 3). In this study, retromer was referred to as a complex of proteins that was dedicated to transporting cargo in a retrograde direction, from the yeast endosome back to the Golgi.

By 2004, a handful of studies had identified the molecular4 and the functional5, 6 homologies of the mammalian retromer, and in 2005 retromer was linked to its first human disorder, Alzheimer disease (AD)7. At the time, the available evidence suggested that the mammalian retromer might match the simplicity of its yeast homologue. Since then, a dramatic and exponential rise in research focusing on retromer has led to more than 300 publications. These studies have revealed the complexity of the mammalian retromer and its functional diversity in endosomal transport, and have implicated retromer in a growing number of neurological disorders.

New evidence indicates that retromer is a ‘master conductor’ of endosomal sorting and trafficking8. Synaptic function heavily depends on endosomal trafficking, as it contributes to the presynaptic release of neurotransmitters and regulates receptor density in the postsynaptic membrane, a process that is crucial for neuronal plasticity9. Therefore, it is not surprising that a growing number of studies are showing that retromer has an important role in synaptic biology10, 11, 12, 13. These observations may account for why the nervous system seems particularly sensitive to genetic and other defects in retromer. In this Progress article, we briefly review the molecular organization and the functional role of retromer, before discussing studies that have linked retromer dysfunction to several neurological diseases — notably, AD and Parkinson disease (PD).

Function and organization

The endosome is considered a hub for intracellular transport. From the endosome, transmembrane proteins can be actively sorted and trafficked to various intracellular sites via distinct transport routes (Fig. 1a). Studies have shown that the mammalian retromer mediates two of the three transport routes out of endosomes. First, retromer is involved in the retrieval of cargos from endosomes and in their delivery, in a retrograde direction, to the trans-Golgi network (TGN)5,6. Retrograde transport has many cellular functions but, as we describe, it is particularly important for the normal delivery of hydrolases and proteases to the endosomal–lysosomal system. The second transport route in which retromer functions is the recycling of cargos from endosomes back to the cell surface14, 15 (Fig. 1a). It is this transport route that is particularly important for neurons, as it mediates the normal delivery of glutamate and other receptors to the plasma membrane during synaptic remodelling and plasticity10, 11, 12, 13.

Figure 1: Retromer’s endosomal transport function and molecular organization.
Retromer's endosomal transport function and molecular organization.

a | Retromer mediates two transport routes out of endosomes via tubules that extend out of endosomal membranes. The first is the retrograde pathway in which cargo is retrieved from the endosome and trafficked to the trans-Golgi network (TGN). The second is the recycling pathway in which cargo is trafficked back from the endosome to the cell surface. The degradation pathway, which is not mediated by retromer, involves the trafficking of cargo from endosomes to lysosomes for degradation. b | The retromer assembly of proteins can be organized into distinct functional modules, all of which work together as part of retromer’s transport role. The ‘cargo-recognition core’ is the central module of the retromer assembly and comprises a trimer of proteins, in which vacuolar protein sorting-associated protein 26 (VPS26) and VPS29 bind VPS35. The ‘tubulation’ module includes protein complexes that bind the cargo-recognition core and aid in the formation and stabilization of tubules that extend out of endosomes, directing the transport of cargos towards their final destinations. The ‘membrane-recruiting’ proteins recruit the cargo-recognition core to the endosomal membrane. The WAS protein family homologue (WASH) complex of proteins also binds the cargo-recognition core and is involved in endosomal ‘actin remodelling’ to form actin patches, which are important for directing cargos towards retromer’s transport pathways. Retromer cargos includes a range of receptors — which bind the cargo-recognition core — and their ligands. PtdIns3P, phosphatidylinositol-3-phosphate.

As well as extending the endosomal transport routes, recent studies have considerably expanded the number of molecular constituents and what is known about the functional organization of the mammalian retromer. Following this expansion in knowledge of the molecular diversity and organizational complexity, retromer might be best described as a multimodular protein assembly. The protein or group of proteins that make up each module can vary, but each module is defined by its distinct function, and the modules work in unison in support of retromer’s transport role.

Two modules are considered central to the retromer assembly. First and foremost is a trimeric complex that functions as a ‘cargo-recognition core’, which selects and binds to the transmembrane proteins that need to be transported and that reside in endosomal membranes5, 6. This trimeric core comprises vacuolar protein sorting-associated protein 26 (VPS26), VPS29 and VPS35; VPS35 functions as the core’s backbone to which the other two proteins bind16. VPS26 is the only member of the core that has been found to have two paralogues, VPS26a and VPS26b17,18, and studies suggest that VPS26b might be differentially expressed in the brain19, 20. Some studies suggest that VPS26a and VPS26b are functionally redundant21, whereas others suggest that they might form distinct cargo-recognition cores20, 22.

The second central module of the retromer assembly is the ‘tubulation’ module, which is made up of proteins that work together in the formation and the stabilization of tubules that extend out of endosomes and that direct the transport of cargo towards its final destination (Fig. 1b). The proteins in this module, which directly binds the cargo-recognition core, are members of the subgroup of the sorting nexin (SNX) family that are characterized by the inclusion of a carboxy-terminal BIN–amphiphysin–RVS (BAR) domain23. These members include SNX1, SNX2, SNX5 and SNX6 (Refs 24,25). As part of the tubulation module, these SNX-BAR proteins exist in different dimeric combinations, but typically SNX1 interacts with SNX5 or SNX6, and SNX2 interacts with SNX5 or SNX6 (Refs 26,27). The EPS15-homology domain 1 (EHD1) protein can be included in this module, as it is involved in stabilizing the tubules formed by the SNX-BAR proteins28.

A third module of the retromer assembly functions to recruit the cargo-recognition core to endosomal membranes and to stabilize the core once it is there (Fig. 1b). Proteins that are part of this ‘membrane-recruiting’ module include SNX3 (Ref. 29), the RAS-related protein RAB7A30, 31,32 and TBC1 domain family member 5 (TBC1D5), which is a member of the TRE2–BUB2–CDC16 (TBC) family of RAB GTPase-activating proteins (GAPs)28. In addition, the lipid phosphatidylinositol-3-phosphate (PtdIns3P), which is found on endosomal membranes, contributes to recruiting most of the retromer-related SNXs through their phox homology domains33. Interestingly, another SNX with a phox homology domain, SNX27, was recently linked to retromer and its function15, 34. SNX27 functions as an adaptor for binding to PDZ ligand-containing cargos that are destined for transport to the cell surface via the recycling pathway. Thus, according to the functional organization of the retromer assembly, SNX27 belongs to the module that engages in cargo recognition and selection.

Recent studies have identified a fourth module of the retromer assembly. The five proteins in this module — WAS protein family homologue 1 (WASH1), FAM21, strumpellin, coiled-coil domain-containing protein 53 (CCDC53) and KIAA1033 (also known as WASH complex subunit 7) — form the WASH complex and function as an ‘actin-remodelling’ module28, 35, 36 (Fig. 1b). Specifically, the WASH complex functions in the rapid polymerization of actin to create patches of actin filaments on endosomal membranes. The complex is recruited to endosomal membranes by binding VPS35 (Ref. 28), and together they divert cargo towards retromer transport pathways and away from the degradation pathway.

The cargos that are transported by retromer include the receptors that directly bind the cargo-recognition core and the ligands of these receptors that are co-transported with the receptors. The receptors that are transported by retromer that have so far been identified to be the most relevant to neurological diseases are the family of VPS10 domain-containing receptors (including sortilin-related receptor 1 (SORL1; also known as SORLA), sortilin, and SORCS1, SORCS2 and SORCS3)7; the cation-independent mannose-6-phosphate receptor (CIM6PR)6, 5; glutamate receptors10; and phagocytic receptors that mediate the clearing function of microglia37. The most disease-relevant ligand to be identified that is trafficked as retromer cargo is the β-amyloid precursor protein (APP)7, 38, 39, 40, 41, which binds SORL1 and perhaps other VPS10 domain-containing receptors42 at the endosomal membrane.

Retromer dysfunction

Guided by retromer’s established function, and on the basis of empirical evidence, there are three well-defined pathophysiological consequences of retromer dysfunction that have proven to be relevant to AD and nervous system disorders. First, retromer dysfunction can cause cargos that typically transit rapidly through the endosome to reside in the endosome for longer than normal durations, such that they can be pathogenically processed into neurotoxic fragments (for example, APP, when stalled in the endosome, is more likely to be processed into amyloid-β, which is implicated in AD43 (Fig. 2a)). Second, by reducing endosomal outflow via impairment of the recycling pathway, retromer dysfunction can lead to a reduction in the number of cell surface receptors that are important for brain health (for example, microglia phagocytic receptors37 (Fig. 2b)).

Figure 2: The pathophysiology of retromer dysfunction.
The pathophysiology of retromer dysfunction.

Retromer dysfunction has three established pathophysiological consequences. In the examples shown, the left graphic represents a cell with normal retromer function and the right graphic represents a cell with a deficit in retromer function. a | Retromer dysfunction causes increased levels of cargo to reside in endosomes. For example, in primary neurons, retromer transports the β-amyloid precursor protein (APP) out of endosomes. Accordingly, retromer dysfunction increases APP levels in endosomes, leading to accelerated APP processing, resulting in an accumulation of neurotoxic fragments of APP (namely, β-carboxy-terminal fragment (βCTF) and amyloid-β) that are pathogenic in Alzheimer disease. b | Retromer dysfunction causes decreased cargo levels at the cell surface. For example, in microglia, retromer mediates the transport of phagocytic receptors to the cell surface and retromer dysfunction results in a decrease in the delivery of these receptors. Studies suggest that this cellular phenotype might have a pathogenic role in Alzheimer disease. c | Retromer dysfunction causes decreased delivery of proteases to the endosome. Retromer is required for the normal retrograde transport of the cation-independent mannose-6-phosphate receptor (CIM6PR) from the endosome back to the trans-Golgi network (TGN). It is in the TGN that this receptor binds cathepsin D and other proteases, and transports them to the endosome, to support the normal function of the endosomal–lysosomal system. By impairing the retrograde transport of the receptor, retromer dysfunction ultimately leads to reduced delivery of cathepsin D to this system. Cathepsin D deficiency has been shown to disrupt the endosomal–lysosomal system and to trigger tau pathology either within endosomes or secondarily in the cytosol.

The third consequence (Fig. 2c) is a result of the established role that retromer has in the retrograde transport of receptors, such as CIM6PR5, 6 or sortilin44, after these receptors transport proteases from the TGN to the endosome. Once at the endosome, the proteases disengage from the receptors, are released into endosomes and migrate to lysosomes. These proteases function in the endosomal–lysosomal system to degrade proteins, protein oligomers and aggregates45. Retromer functions to transfer the ‘naked’ receptor from the endosome back to the TGN via the retrograde pathway5, 6, allowing the receptors to continue in additional rounds of protease delivery. Accordingly, by reducing the normal retrograde transport of these receptors, retromer dysfunction has been shown to reduce the proper delivery of proteases to the endosomal–lysosomal system5,6, which, as discussed below, is a pathophysiological state linked to several brain disorders.

Although requiring further validation, recent studies suggest that retromer dysfunction might be involved in two other mechanisms that have a role in neurological disease. One study suggested that retromer might be involved in trafficking the transmembrane protein autophagy-related protein 9A (ATG9A) to recycling endosomes, from where it can then be trafficked to autophagosome precursors — a trafficking step that is crucial in the formation and the function of autophagosomes46. Autophagy is an important mechanism by which neurons clear neurotoxic aggregates that accumulate in numerous neurodegenerative diseases47. A second study has suggested that retromer dysfunction might enhance the seeding and the cell-to-cell spread of intracellular neurotoxic aggregates48, which have emerged as novel pathophysiological mechanisms that are relevant to AD49, PD50 and other neurodegenerative diseases.

Alzheimer disease

Retromer was first implicated in AD in a molecular profiling study that relied on functional imaging observations in patients and animal models to guide its molecular analysis7. Collectively, neuroimaging studies confirmed that the entorhinal cortex is the region of the hippocampal circuit that is affected first in AD, even in preclinical stages, and suggested that this effect was independent of ageing (as reviewed in Ref. 51). At the same time, neuroimaging studies identified a neighbouring hippocampal region, the dentate gyrus, that is relatively unaffected in AD52. Guided by this information, a study was carried out in which the two regions of the brain were harvested post mortem from patients with AD and from healthy individuals, intentionally covering a broad range of ages. A statistical analysis was applied to the determined molecular profiles of the regions that was designed to address the following question: among the thousands of profiled molecules, which are the ones that are differentially affected in the entorhinal cortex versus the dentate gyrus, in patients versus controls, but that are not affected by age? The final results led to the determination that the brains of patients with AD are deficient in two core retromer proteins — VPS26 and VPS35 (Ref. 7).

Little was known about the receptors of the neuronal retromer, so to understand how retromer deficiency might be mechanistically linked to AD, an analysis was carried out on the molecular data set that looked for transmembrane molecules for which expression levels correlated with VPS35 expression. The top ‘hit’ was the transcript encoding the transmembrane protein SORL1 (Ref. 43). As SORL1 belongs to the family of VPS10-containing receptors and as VPS10 is the main retromer receptor in yeast3, it was postulated that SORL1 and the family of other VPS10-containing proteins (sortillin, SORCS1, SORCS2 and SORCS3) might function as retromer receptors in neurons7. In addition, SORL1 had recently been reported to bind APP53, so if SORL1 was assumed to be a receptor that is trafficked by retromer, then APP might be the cargo that is co-trafficked by retromer. This led to a model in which retromer traffics APP out of endosomes7, which are the organelles in which APP is most likely to be cleaved by βAPP-cleaving enzyme 1 (BACE1; also known as β-secretase 1)43; this is the initial enzymatic step in the pathogenic processing of APP.

Subsequent studies were required to further establish the pathogenic link between retromer and AD, and to test the proposed model. The pathogenic link was further supported by human genetic studies. First, a genetic study investigating the association between AD, the genes encoding the components of the retromer cargo-recognition core and the family of VPS10-containing receptors found that variants of SORL1 increase the risk of developing AD38. This finding was confirmed by numerous studies, including a recent large-scale AD genome-wide association study54. Other genetic studies identified AD-associated variants in genes encoding proteins that are linked to nearly all modules of the retromer assembly55, including genes encoding proteins of the retromer tubulation module (SNX1), genes encoding proteins of the retromer membrane-recruiting module (SNX3 and RAB7A) and genes encoding proteins of the retromer actin-remodelling module (KIAA1033). In addition, nearly all of the genes encoding the family of VPS10-containing retromer receptors have been found to have variants that associate with AD56. Finally, a study found that brain regions that are differentially affected in AD are deficient in PtdIns3P, which is the phospholipid required for recruiting many sorting nexins to endosomal membranes57. Thus, together with the observation that the brains of patients with AD are deficient in VPS26a and VPS35 (Refs 7,37), all modules in the retromer assembly are implicated in AD.

Studies in mice39, 58, 59, flies39 and cells in culture34, 40, 41, 60, 61 have investigated how retromer dysfunction leads to the pathogenic processing of APP. Although rare discrepancies have been observed among these studies62, when viewed in total, the most consistent findings are that retromer dysfunction causes increased pathogenic processing of APP by increasing the time that APP resides in endosomes. Moreover, these studies have confirmed that SORL1 and other VPS10-containing proteins function as APP receptors that mediate APP trafficking out of endosomes.

Retromer has unexpectedly been linked to microglial abnormalities37 — another core feature of AD — which, on the basis of recent genetic findings, seem to have an upstream role in disease pathogenesis54, 63. A recent study found that microglia harvested from the brains of individuals with AD are deficient in VPS35 and provided evidence suggesting that retromer’s recycling pathway regulates the normal delivery of various phagocytic receptors to the cell surface of microglia37, including the phagocytic receptor triggering receptor expressed on myeloid cells 2 (TREM2) (Fig. 2b). Mutations in TREM2 have been linked to AD63, and a recent study indicates that these mutations cause a reduction in its cell surface delivery and accelerate TREM2 degradation, which suggests that the mutations are linked to a recycling defect64. While they are located at the microglial cell surface, these phagocytic receptors function in the clearance of extracellular proteins and other molecules from the extracellular space65. Taken together, these recent studies suggest that defects in the retromer’s recycling pathway can, at least in part, account for the microglial defects observed in the disease.

The microtubule-associated protein tau is the key element of neurofibrillary tangles, which are the other hallmark histological features of AD. Although a firm link between retromer dysfunction and tau toxicity remains to be established, recent insight into tau biology suggests several plausible mechanisms that are worth considering. Tau is a cytosolic protein, but nonetheless, through mechanisms that are still undetermined, it is released into the extracellular space from where it gains access to neuronal endosomes via endocytosis66, 67. In fact, recent studies suggest that the pathogenic processing of tau is triggered after it is endocytosed into neurons and while it resides in endosomes67. Of note, it still remains unknown which specific tau processing step — its phosphorylation, cleavage or aggregation — is an obligate step towards tau-related neurotoxicity. Accordingly, if defects in microglia or in other phagocytic cells reduce their capacity to clear extracellular tau, this would accelerate tau endocytosis in neurons and its pathogenic processing.

A second possibility comes from the established role retromer has in the proper delivery of cathepsin D and other proteases to the endosomal–lysosomal system via CIM6PR or sortilin (Fig. 2c). Studies in sheep, mice and flies68 have shown that cathepsin D deficiency can enhance tau toxicity and that this is mediated by a defective endosomal–lysosomal system68. Whether this mechanism leads to abnormal processing of tau within endosomes or in the cytosol via caspase activation68 remains unclear. As discussed above, retromer dysfunction will lead to a decrease in the normal delivery of cathepsin D to the endosome and will result in endosomal–lysosomal system defects. Retromer dysfunction can therefore be considered as a functional phenocopy of cathepsin D deficiency, which suggests a plausible link between retromer dysfunction and tau toxicity. Nevertheless, although these recent insights establish plausibility and support further investigation into the link between retromer and tau toxicity, whether this link exists and how it may be mediated remain open and outstanding questions.

Parkinson disease

The pathogenic link between retromer and PD is singular and straightforward: exome sequencing has identified autosomal-dominant mutations in VPS35 that cause late-onset PD69, 70, one of a handful of genetic causes of late-onset disease. However, the precise mechanism by which these mutations cause the disease is less clear.

Among a group of recent studies, all46, 48, 71, 72, 73, 74, 75, 76 but one77 strongly suggest that these mutations cause a loss of retromer function. At the molecular level, the mutations do not seem to disrupt mutant VPS35 from interacting normally with VPS26 and VPS29, and from forming the cargo-recognition core. Rather, two studies suggest that the mutations have a restricted effect on the retromer assembly but reduce the ability of VPS35 to associate with the WASH complex46, 75. Studies disagree about the pathophysiological consequences of the mutations. Four studies suggest that the mutations affect the normal retrograde transport of CIM6PR71, 73, 75, 76 from the endosome back to the TGN (Fig. 2c). In this scenario, the normal delivery of cathepsin D to the endosomal–lysosomal system should be reduced and this has been empirically shown73. Cathepsin D has been shown to be the dominant endosomal–lysosomal protease for the normal processing of α-synuclein76, and mutations could therefore lead to abnormal α-synuclein processing and to the formation of α-synuclein aggregates, which are thought to have a key pathogenic role in PD.

A separate study suggested that the mutation might cause a mistrafficking of ATG9, and thereby, as discussed above, reduce the formation and the function of autophagosomes46. Autophagosomes have also been implicated as an intracellular site in which α-synuclein aggregates are cleared. Thus, although future studies are needed to resolve these discrepant findings (which may in fact not be mutually exclusive), these studies are generally in agreement that retromer defects will probably increase the neurotoxic levels of α-synuclein aggregates48.

Several studies in flies71, 74 and in rat neuronal cultures71 provide strong evidence that increasing retromer function by overexpressing VPS35 rescues the neurotoxic effects of the most common PD-causing mutations in leucine-rich repeat kinase 2 (LRRK2). Moreover, a separate study has shown that increasing retromer levels rescues the neurotoxic effect of α-synuclein aggregates in a mouse model48. These findings have immediate therapeutic implications for drugs that increase VPS35 and retromer function, as discussed in the next section, but they also offer mechanistic insight. LRRK2 mutations were found to phenocopy the transport defects caused either by theVPS35 mutations or by knocking down VPS35 (Ref. 71). Together, this and other studies78suggest that LRRK2 might have a role in retromer-dependent transport, but future studies are required to clarify this role.

Other neurological disorders

Besides AD and PD, in which a convergence of findings has established a strong pathogenic link, retromer is being implicated in an increasing number of other neurological disorders. Below, we briefly review three disorders for which the evidence of the involvement of retromer in their pathophysiology is currently the most compelling.

The first of these disorders is Down syndrome (DS), which is caused by an additional copy of chromosome 21. Given the hundreds of genes that are duplicated in DS, it has been difficult to identify which ones drive the intellectual impairments that characterize this condition. A recent elegant study provides strong evidence that a deficiency in the retromer cargo-selection protein SNX27 might be a primary driver for some of these impairments79. This study found that the brains of individuals with DS were deficient in SNX27 and that this deficiency may be caused by an extra copy of a microRNA (miRNA) encoded by human chromosome 21 (the miRNA is produced at elevated levels and thereby decreases SNX27 expression). Consistent with the known role of SNX27 in retromer function, decreased expression of this protein in mice disrupted glutamate receptor recycling in the hippocampus and led to dendritic dysfunction. Importantly, overexpression of SNX27 rescued cognitive and other defects in animal models79, which not only strengthens the causal link between retromer dysfunction and cognitive impairment in DS but also has important therapeutic implications.

Hereditary spastic paraplegia (HSP) is another disorder linked to retromer. HSP is caused by genetic mutations that affect upper motor neurons and is characterized by progressive lower limb spasticity and weakness. Although there are numerous mutations that cause HSP, most are unified by their effects on intracellular transport80. One HSP-associated gene in particular encodes strumpellin81, which is a member of the WASH complex.

The third disorder linked to retromer is neuronal ceroid lipofuscinosis (NCL). NCL is a young-onset neurodegenerative disorder that is part of a larger family of lysosomal storage diseases and is caused by mutations in one of ten identified genes — nine neuronal ceroid lipofuscinosis (CLN) genes and the gene encoding cathepsin D82. Besides cathepsin D, for which the link to retromer has been discussed above, CLN3 seems to function in the normal trafficking of CIM6PR83. However, the most direct link to retromer has been recently described for CLN5, which seems to function, at least in part, as a retromer membrane-recruiting protein84.

Retromer as a therapeutic target

As suggested by the first study implicating retromer in AD7, and in several subsequent studies71,85, increasing the levels of retromer’s cargo-recognition core enhances retromer’s transport function. Motivated by this observation and after a decade-long search86, we identified a novel class of ‘retromer pharmacological chaperones’ that can bind and stabilize retromer’s cargo-recognition core and increase retromer levels in neurons61.

Validating the motivating hypothesis, the chaperones were found to enhance retromer function, as shown by the increased transport of APP out of endosomes and a reduction in the accumulation of APP-derived neurotoxic fragments61. Although there are numerous other pharmacological approaches for enhancing retromer function, this success provides the proof-of-principle that retromer is a tractable therapeutic target.

As retromer functions in all cells, a general concern is whether enhancing its function will have toxic adverse effects. However, studies have found that in stark contrast to even mild retromer deficiencies, increasing retromer levels has no obvious negative consequences in yeast, neuronal cultures, flies or mice40, 48, 61, 71. This might make sense because unlike drugs that, for example, function as inhibitors, simply increasing the normal flow of transport through the endosome might not be cytotoxic.

If retromer drugs are safe and can effectively enhance retromer function in the nervous system — which are still outstanding issues — there are two general indications for considering their clinical application. One rests on the idea that these agents will only be efficacious in patients who have predetermined evidence of retromer dysfunction. The most immediate example is that of individuals with PD that is caused by LRRK2 mutations. As discussed above, several ‘preclinical’ studies in flies and neuronal cultures have already established that increasing retromer levels71, 74can reverse the neurotoxic effects of such mutations and, thus, if this approach is proven to be safe, LRRK2-linked PD might be an appropriate indication for clinical trials.

Alternatively, the pathophysiology of a disease might be such that retromer-enhancing drugs would be efficacious regardless of whether there is documented evidence of retromer dysfunction. AD illustrates this point. As reviewed above, current evidence suggests that retromer-enhancing drugs will, at the very least, decrease pathogenic processing of APP in neurons and enhance microglial function, even if there are no pre-existing defects in retromer.

More generally, histological studies comparing the entorhinal cortex of patients with sporadic AD to age-matched controls have documented that enlarged endosomes are a defining cellular abnormality in AD87, 88. Importantly, enlarged endosomes are uniformly observed in a broad range of patients with sporadic AD, which suggests that enlarged endosomes reflect an intracellular site at which molecular aetiologies converge87. In addition, because they are observed in early stages of the disease in regions of the brain without evidence of amyloid pathology87, enlarged endosomes are thought to be an upstream event. Mechanistically, the most likely cause of enlarged endosomes is either too much cargo flowing into endosomes — as occurs, for example, with apolipoprotein E4 (APOE4), which has been shown to accelerate endocytosis89, 90 — or too little cargo flowing out, as observed in retromer dysfunction40, 61 and related transport defects57. By any mechanism, retromer-enhancing drugs might correct this unifying cellular defect and might be expected to be beneficial regardless of the specific aetiology.


The fact that retromer defects, including those derived from bona fide genetic mutations, seem to differentially target the nervous system suggests that the nervous system is differentially dependent on retromer for its normal function. We think that this reflects the unique cellular properties of neurons and how synaptic biology heavily depends on endosomal transport and trafficking. Although plausible, future studies are required to confirm and to test the details of this hypothesis.

However, currently, it is the clinical rather than the basic neuroscience of retromer that is much better understood, with the established pathophysiological consequences of retromer dysfunction providing a mechanistic link to the disorders in which retromer has been implicated. Nevertheless, many questions remain. The two most interesting questions, which are in fact inversions of each other, relate to regional vulnerability in the nervous system. First, why does retromer dysfunction target specific neuronal populations? Second, how can retromer dysfunction cause diseases that target different regions of the nervous system? Recent evidence hints at answers to both questions, which must somehow be rooted in the functional and molecular diversity of retromer.

The type and the extent of retromer defects linked to different disorders might provide pathophysiological clues as well as reasons for differential vulnerability. As discussed, in AD there seem to be across-the-board defects in retromer, such that each module of the retromer assembly as well as multiple retromer cargos have been pathogenically implicated. By contrast, the profile of retromer defects in PD seems to be more circumscribed, involving selective disruption of the interaction between VPS35 and the WASH complex. These insights might agree with histological87, 88 and large-scale genetic studies54 that suggest that endosomal dysfunction is a unifying focal point in the cellular pathogenesis of AD. In contrast, genetics and other studies91suggest that the cellular pathobiology of PD is more distributed, implicating the endosome but other organelles as well, in particular the mitochondria.

Interestingly, studies suggest that the entorhinal cortex — a region that is differentially vulnerable to AD — has unique dendritic structure and function92, which are highly dependent on endosomal transport. We speculate that it is the unique synaptic biology of the entorhinal cortex that can account for why it might be particularly sensitive to defects in endosomal transport in general and retromer dysfunction in particular, and for why this region is the early site of disease. Future studies are required to investigate this hypothesis, as well as to understand why the substantia nigra or other regions that are differentially vulnerable to PD would be particularly sensitive to the more circumscribed defect in retromer.

Perhaps the most important observation for clinical neuroscience is the now well-established fact that increasing levels of retromer proteins enhances retromer function and has already proved capable of reversing defects associated with AD, PD and DS in either cell culture or in animal models. The relationships between protein levels and function are not always simple, but emerging pharmaceutical technologies that selectively and safely increase protein levels are now a tractable goal in drug discovery93. With the evidence mounting that retromer has a pathogenic role in two of the most common neurodegenerative diseases, we think that targeting retromer to increase its functional activity is an important goal that has strong therapeutic promise.


  • Schekman, R. Charting the secretory pathway in a simple eukaryote. Mol. Biol. Cell 21,37813784 (2010).
  • Henne, W. M., Buchkovich, N. J. & Emr, S. D. The ESCRT pathway. Dev. Cell 21, 7791(2011).
  • Seaman, M. N., McCaffery, J. M. & Emr, S. D. A membrane coat complex essential for endosome-to-Golgi retrograde transport in yeast. J. Cell Biol. 142, 665681 (1998).
  • Haft, C. R.et al. Human orthologs of yeast vacuolar protein sorting proteins Vps26, 29, and 35: assembly into multimeric complexes. Mol. Biol. Cell 11, 41054116 (2000).
  • Seaman, M. N. Cargo-selective endosomal sorting for retrieval to the Golgi requires retromer. J. Cell Biol. 165, 111122 (2004).
  • Arighi, C. N., Hartnell, L. M., Aguilar, R. C., Haft, C. R. & Bonifacino, J. S. Role of the mammalian retromer in sorting of the cation-independent mannose 6-phosphate receptor. J. Cell Biol. 165, 123133 (2004).
  • Small, S. A.et al. Model-guided microarray implicates the retromer complex in Alzheimer’s disease. Ann. Neurol. 58, 909919 (2005).
  • Burd, C. & Cullen, P. J. Retromer: a master conductor of endosome sorting. Cold Spring Harb. Perspect. Biol. 6, a016774 (2014).
  • Carroll, R. C., Beattie, E. C., von Zastrow, M. & Malenka, R. C. Role of AMPA receptor endocytosis in synaptic plasticity. Nature Rev. Neurosci. 2, 315324 (2001).
  • Choy, R. W.et al. Retromer mediates a discrete route of local membrane delivery to dendrites. Neuron 82, 5562 (2014).
  • Zhang, D.et al. RAB-6.2 and the retromer regulate glutamate receptor recycling through a retrograde pathway. J. Cell Biol. 196, 85101 (2012).
  • Hussain, N. K., Diering, G. H., Sole, J., Anggono, V. & Huganir, R. L. Sorting nexin 27 regulates basal and activity-dependent trafficking of AMPARs. Proc. Natl Acad. Sci. USA111, 1184011845 (2014).
  • Loo, L. S., Tang, N., Al-Haddawi, M., Dawe, G. S. & Hong, W. A role for sorting nexin 27 in AMPA receptor trafficking. Nature Commun. 5, 3176 (2014).
  • Feinstein, T. N.et al. Retromer terminates the generation of cAMP by internalized PTH receptors. Nature Chem. Biol. 7, 278284 (2011).
  • Temkin, P.et al. SNX27 mediates retromer tubule entry and endosome-to-plasma membrane trafficking of signalling receptors. Nature Cell Biol. 13, 715721 (2011).
  • Seaman, M. N. Recycle your receptors with retromer. Trends Cell Biol. 15, 6875 (2005).
  • Kerr, M. C.et al. A novel mammalian retromer component, Vps26B. Traffic 6, 9911001(2005).
  • Collins, B. M.et al. Structure of Vps26B and mapping of its interaction with the retromer protein complex. Traffic 9, 366379 (2008).
  • Kim, E.et al. Identification of novel retromer complexes in the mouse testis. Biochem. Biophys. Res. Commun. 375, 1621 (2008).
  • Bugarcic, A.et al. Vps26A and Vps26B subunits define distinct retromer complexes. Traffic12, 17591773 (2011).

……. 93


Taub Institute for Research on Alzheimer’s Disease and the Ageing Brain, Departments of Neurology, Radiology, and Psychiatry, Columbia University College of Physicians and Surgeons, New York, New York 10032, USA.

Scott A. Small

Helen and Robert Appel Alzheimer’s Disease Research Institute, Department of Neurology and Feil Family Brain and Mind Research Institute, Weill Cornell Medical College, New York, New York 10065, USA.

Gregory A. Petsko


See also:

Neurobiol Aging. 2011 Nov;32(11):2109.e1-14. doi: 10.1016/j.neurobiolaging.2011.05.025.
Altered intrinsic neuronal excitability and reduced Na+ currents in a mouse model of Alzheimer’s disease.
Brown JT, Chin J, Leiser SC, Pangalos MN, Randall AD.

Trends Neurosci. 2013 Jun;36(6):325-35. doi: 10.1016/j.tins.2013.03.002.
Why size matters – balancing mitochondrial dynamics in Alzheimer’s disease.
DuBoff B, Feany M, Götz J.

Neuron. 2014 Dec 3;84(5):1023-33. doi: 10.1016/j.neuron.2014.10.024.
Dendritic structural degeneration is functionally linked to cellular hyperexcitability in a mouse model of Alzheimer’s disease.
Šišková Z, Justus D, Kaneko H, Friedrichs D, Henneberg N, Beutel T, Pitsch J, Schoch S, Becker A, von der Kammer H, Remy S.



Video: How can we tease out the role of other toxic insults in AD pathogenesis?




Read Full Post »

Visualizing metal-impregnated neurons with spectral confocal microscopy

Larry H. Bernstein, MD, FCAP, Curator



Novel Imaging Captures Beauty of Metal-Labeled Neurons in 3D

These are silver-impregnated dendrites from an insect motor neuron, captured through spectral confocal microscopy. (Credit: Grant M. Barthel, Karen A. Mesce, Karen J. Thompson)

Researchers have discovered a dazzling new method of visualizing neurons that promises to benefit neuroscientists and cell biologists alike: by using spectral confocal microscopy to image tissues impregnated with silver or gold.

Rather than relying on the amount of light reflecting off metal particles, this novel process, to be presented in the journal eLife, involves delivering light energy to silver or gold nanoparticles deposited on neurons and imaging the higher energy levels resulting from their vibrations, known as surface plasmons.

This technique is particularly effective as the light emitted from metal particles is resistant to fading, meaning that decades-old tissue samples achieved through other processes, such as the Golgi stain method from the late 1880s, can be imaged repeatedly.

The new process was achieved by using spectral detection on a Laser Scanning Confocal Microscope (LSCM), first made available in the late 1980s and, until now, used most extensively for fluorescent imaging.

Paired with such methods, silver- and gold-based cell labeling is poised to unlock new information in a myriad of archived specimens. Furthermore, silver-impregnated preparations should retain their high image quality for a century or more, allowing for archivability that could aid in clinical research and disease-related diagnostic techniques for cancer and neurological disorders.

“For the purposes of medical diagnostics, older and newer specimens could be compared with the knowledge that signal intensity would remain fairly uniform regardless of sample age or repeated light exposure,” says contributing author Karen Mesce from the University of Minnesota.

“With the prediction that superior resolution microscopic techniques will continue to evolve, older archived samples could be reimagined with newer technologies and with the confidence that the signal in question was preserved. The progression or stability of a cancer or other disease could therefore be charted with accuracy over long periods of time.”

To appreciate the enhanced image quality produced by the new technique, the team first examined a conventional brightfield image of a metal-labelled neuron within a grasshopper’s abdominal ganglion, a type of mini-brain which, even at that size, presented out-of-focus structures.

They then imaged the same ganglion with the spectral LSCM adjusted to the manufacturer’s traditional fluorescence settings, resulting only in strong natural fluorescence and a collective dark blur in place of the silver-labelled neurons.

However, after collecting the light energy emitted from vibrating surface plasmons in the spectral LSCM, the team obtained spectacular three-dimensional computer images of silver and gold-impregnated neurons. This holds enormous potential for stimulating a re-examination of archived preparations, including Golgi-stained and cobalt/silver-labelled nervous systems.

Additionally, by using a number of different metal-based cell-labeling techniques in combination with the new LSCM protocols, tissue and cell specimens can be generated and imaged with ease and in great three-dimensional detail. Changes in even small structural details of neurons can be identified, which are often important indicators of neurological disease, learning and memory, and brain development.

“Both new and archived preparations are essentially permanent and the information gathered from them increases the data available for characterizing neurons as individuals or as members of classes for comparative studies, adding to emerging neuronal banks,” says co-first author Karen Thompson from Agnes Scott College.

“Just as plasmon resonance can explain the continued intensity of the red (caused by silver nanoparticles) and yellow (gold nanoparticles) colors in centuries-old medieval stained glass and other works of art, metal-impregnated neurons are also likely never to fade, neither in the information they provide nor in their intrinsic beauty,” adds Mesce.



This chapter is partly modified from:

Towards a 3D View of Cellular Architecture: Correlative Light Microscopy and Electron Tomography.

Valentijn J.A.; van Driel L.F.; Jansen K.A.; Valentijn K.M.; Koster A.J.
Book chapter in: Reiner Salzer. Biomedical Imaging: Principles and Applications. Wiley, 2010.

The present thesis reports on newly developed tools and strategies for correlative light and electron microscopy, and their application to cell biology research aimed at furthering our understanding of the structure-functional mechanisms of varying current biological applications. Accordingly, this Introduction consists of two main parts: the first one will discuss past and present strategies for correlative light and electron microscopy (CLEM) as an introduction to the subsequent chapters in this thesis, all of which will describe new developments and applications in the field; the second one will elaborate on the rationale behind the biological applications that were undertaken.

The term ‘correlative microscopy’ is employed in the biomedical literature to designate any combination of two or more microscopic techniques applied to the same region in a biological specimen. The purpose of correlative microscopy is to obtain complementary data, each imaging modality providing different information, on the specimen that is under investigation. Correlative light and electron microscopy (CLEM) is by far the most widespread form of correlative microscopy. CLEM makes use of the fact that imaging with photons on the one hand, and electrons on the other hand, each offers specific advantages over one another. For instance, the low-magnification range inherent to light microscopy (LM) is particularly well-suited for the rapid scanning of large and heterogeneous sample areas, while the high magnification and -resolution that can be achieved by electron microscopy (EM) allows for the subsequent zooming in on selected areas of interest to obtain ultrastructural detail. A further advantage of LM is that it can be used to study dynamic processes, up to the molecular level, in living cells and tissues. The recent surge in live cell imaging research has catalyzed a renewed interest in CLEM methodologies, as the interpretation of the dynamic processes observed by LM often requires high-resolution information from EM data. CLEM is also gaining in momentum in the field of cryo-electron microscopy where the low contrast conditions and low electron dose requirements put a constraint on the detection efficacy.

Current CLEM procedures face a number of challenges. Firstly, sample preparation methods for LM and EM can be quite divergent due to different requirements for preservation, embedding, sectioning, and counterstaining. Therefore, alternative sample preparation protocols need to be devised that are suitable for both LM and EM. Secondly, CLEM often requires the correlated localization of specific molecules in cells or tissues, for which specialized detection systems need to be developed. Standard detection methods are based on tagging of molecules either with fluorochromes for LM, or with gold particles for EM, whereas some CLEM applications require a tag to be visible in both modalities. Thirdly, the transition from imaging by LM to EM may involve handling and additional processing of samples, which can lead to changes in orientation and morphology of the sample. This in turn can hamper the finding back of, and correlation with previously established areas of interest.

In the present post-genomics climate, EM is coming back with a vengeance. Despite the dip in EM-based research during the previous decade, the development of novel EM technologies moved forward at a steady pace, resulting in several breakthrough applications. Among them are electron tomography and cryo-electron tomography, which are techniques for high-resolution 3D visualization and which are gradually becoming mainstream tools in structural molecular biology. As will be discussed in more detail below, (cryo-)electron tomography is often hampered by the lack of landmarks in the 2D views used to select areas of interest.

The diversity of goals to be achieved by CLEM constrains the development of universally applicable protocols. For instance, correlating live cell imaging data of a fluorescent protein with an ultrastructural endpoint requires a different CLEM approach than if the main goal is to pinpoint a rarely occurring structure of interest for EM investigation. CLEM can also be used as an alternative for immuno-EM, or to locate structures for cryo-EM if high-resolution structural details are required. As a consequence of the diversity of applications, there are to date numerous methods to correlate LM and EM data, and more developments, improvements, and applications, are likely to follow. Depending on the purpose of the application, three groups of CLEM methodologies can be distinguished:

  • Those that combine live cell imaging with ultrastructural information (see figure 1),
  • Those that combine LM of fixed or immobilized samples with ultrastructural information (figure 2),
  • Those that combine LM and EM data from the same sections (figure 3)



Using Laser Scanning Confocal Microscopy as a Guide for Electron Microscopic Study: A Simple Method for Correlation of Light and Electron Microscopy’

J Histochem and Cytochcem 1995; 43(3):329-335


Anatomic study of synaptic connections in the nervous system is laborious and difficult, especially when neurons are large or have fine branches embedded among many other processes. Although electron microscopy provides a powerful tool for such study, the correlation of light microscopic appearance and electron microscopic detail is very time consuming. We report here a simple method combining laser scanning confocal microscopy and electron microscopy for study of the synaptic relationships of the neurons in the antennal lobe, the first central neuropil in the olfactory pathway, of the moth Manduca sexta. Neurons were labeled intracellularly with neurobiotin or biocytin, two widely used stains. The tissue was then sectioned on a vibratome and processed with both streptavidin-nanogold (for electron microscopic study) and streptavidin-Cy3 (for confocal microscopic study) and embedded in epon/araldite. Interesting areas of the labeled neuron were imaged in the epon/araldite sectioned at the indicated depth for electron microscopic study. This method provides an easy, reliable way to correlate three-dimensional light miaoscopic information with electron microscopic detail, and can be very useful in studies of synaptic connections.

Introduction Study of synaptic connections is fundamental to an understanding of nervous system function. Physiological recording in combination with cell staining with different dyes yields especially useful information about the functional properties and morphology of neurons in various nervous systems. Light microscopic (LM) analysis of neurons, however, often raises the question of whether synaptic connections occur at specialized regions of the neurites. Knowledge of whether synaptic connections between two neurons exist and where on the neuritic tree they occur provides valuable information about how signals are integrated by the neurons and thus about the role the neurons play in a given pathway. Electron microscopy (EM) provides a powerful tool for this type of study. Unfortunately, correlation of three-dimensional LM data with EM detail is often difficult, as neurons are often large (ranging up to hundreds of micrometers in length) with complex branching patterns. Moreover, EM data usually lose most large-scale three- dimensional information. Much work has been done to attempt to overcome these problems (3,4,8,10,14,19,21,25,26,31,34,35). To date, the most successful methods for correlating EM and threedimensional data have been

(a) three-dimensional reconstruction of thin sections,

(b) semi-thin sectioning of tissue, serial reconstruction of the neuron at the light microscopic level, re-embedment, and finally thin-sectioning of selected semi-thin sections for electron microscopic study, and

(c) high-voltage EM observation of relatively thick sections and reconstruction of high-voltage EM data.

Although combination of these techniques or computer-assisted three-dimensional reconstruction and image processing techniques greatly facilitate the task, it is still very time-consuming to pursue this type of study.

Recently, laser scanning confocal microscopy (LSCM) has provided a convenient way to obtain three-dimensional morphology of neurons, as optical sections of relatively thick tissue can be obtained easily and rapidly (5-7,20,24,27,28,30). Equally importantly, information is gathered as digitized optical sections and is readily displayed as three-dimensional stacks. Combination of LSCM and EM appears to be a promising way to study synaptic connections.


Several different techniques have been used to bridge the gap between LM and EM (3,4,8,10,14,19,21,25,26,31,34,35). One common way is to section the block into semi-thin sections, make a threedimensional reconstruction of the neuron at the LM level, and then thin-section areas of interest (4,8,10,19,21). However, three-dimensional reconstruction from sections is tedious, and valuable tissue may be lost during reembedding and resectioning. High-voltage EM is a reasonable alternative, as relatively thicker sections can be observed, thus reducing the number of sections needed to span a neuron. This method, however, requires access to a high-voltage EM. We have developed a technique using LSCM as a guide for EM study. A biotin-labeled neuron is rendered detectable by both fluorescence (for LSCM) and immunogold (for EM). After such double labeling, interesting areas of physiologically identified and intracellularly labeled neurons can be investigated at the EM level to see where synaptic connections are formed. Although with our method three-dimensional reconstruction might also be needed, the number of vibratome sections needed to span a neuron is small (two to four in our case with 80-pm sections). This greatly reduces the task.

The major challenge in correlation of LSCM and EM is the compatibility of the labeling methods. Visualization methods using diaminobenzidine (either in a peroxidase reaction or in a reaction catalyzed by photo-oxidation of fluorescent dyes) and Golgi staining have been used widely for both LM and EM observations (see, e.g., 8,26,35). The light- and electron-dense label, however, is not suitable for LSCM imaging, which is best performed on fluorescent  labels. One possibility is to use fluorescent labels, perform LSCM, and, after LSCM imaging, convert the fluorescent dyes into electron-dense materials by using either photo-oxidation of fluorescent dyes [such as Lucifer Yellow (16) or DiI as suggested by Vischer and Durrenberger (32)] in the presence of diaminobenzidine or an antibody against the dye injected into the cell. It is possible to determine the position of an interesting area of a neuron by LSCM with this method, but relocating the same area after plastic embedding is much more difficult than the presently reported method because of shrinkage of the tissue during tissue processing for EM and the inability to monitor the exact position by LSCM during sectioning. The reflection mode of LSCM has been applied successfully to detect electron-dense label in several systems (6,7,22,33). With this mode of imaging, Deitch et al. (6) have proposed another way of combining LSCM and EM. Reflection signals in the LSCM are often weak, however, and do not take full advantage of the excellent resolving capabilities of the confocal microscope.


The unending fascination with the Golgi method

Y Koyama*
Department of Anatomy and Neuroscience, Graduate School of Medicine, Osaka University, Suita, Osaka, Japan

In the second half of the 19th century, Camillo Golgi provided a breakthrough staining technique for visualizing whole neurons, which are seen as black bodies due to intracellular staining with microcrystalline silver chromate. The high contrast, selective staining properties enabled identification of complete neuronal morphology. This staining technique, termed the Golgi method, was later improved by Ramón y Cajal and popularised through his tireless experiments. Morphological analysis, using the Golgi method, led to the discovery of neuronal microstructures such as dendritic spines and growth cones and helped give rise to the ‘neuron doctrine’. Many post-mortem human brains as well as brains of experimental animals have since been stained using this method. In combination with other morphological techniques (e.g. electron microscopy and immunohistochemistry), the Golgi method has allowed us to glean more information regarding the neuronal networks present in various brain regions. However, the Golgi method is a difficult first choice for morphological analysis since it is capricious, complicated and time-consuming and has poor reproducibility.

Recent increases in the number of in vivo animal experiments and of post-mortem brains collected following neurological disorders heighten the need for the Golgi method to be viewed as a crucial morphological tool for assessing abnormalities in single neurons, as well as in neuronal networks. Fortunately, over 100 years of neuroanatomical diligence has seen significant contributions to overcoming the shortfalls of this method. The advent of modified Golgi methods with potential use as routine techniques, together with the development of the kit-based Golgi–Cox method, has made the Golgi method more accessible to neuroanatomists.

This review surveys the technical fundamentals, history and evolution of Golgi methods and intends to spark an interest in the Golgi method within every neuroscientist, novice and old pro alike and to allow them to appreciate this useful technique.


Many neuroanatomists, including us, feel a strong attraction to the Golgi method as a powerful morphological tool. Our researchers have identified unwanted issues of the various Golgi methods and then have been working to remedy these problems. We encourage the reader to adopt staining using the Golgi method as its utility continues to evolve.


While studying the brain function, it is extremely important to investigate the precise shape and morphological changes of individual neurons. A neuron is a specialised cell that emits an electrical signal to allow information exchange, a characteristic not found in the cells of other organs. Both the short ‘dendrite’, which is intricately branched like a tree branch and the long ‘axon’ emanate from the nerve cell body. In addition, neuronal spines located on dendrites receive electric signals from other neurons and are involved in neuronal plasticity. Thus, together with neural cells such as neuroglia, neurons form complicated networks known as ‘neural circuits’.

To appreciate the complexity of such intricate neural networks, staining methods that allow the visualisation of neuronal cells in thinly sliced brain sections are used. In particular, Nissl stains and silver impregnation are commonly used. Nissl staining, which is based on a mechanism combining a basic aniline dye with Nissl granules, can stain both the cell nucleoli and rough endoplasmic reticula in neurons. In contrast, silver impregnation using neuronal argent affinity can stain an entire neuron, but not the myelin sheath. Neural circuitry refers to the combination of many interacting neural cells and is immensely complex morphologically, with many neurons intertwined with one another within a restricted space. Unfortunately, because the above techniques stain all neural cells with equal probability, it is difficult to identify and appreciate the morphology of a single cell amongst the mass of other stained cells.

In contrast, the Golgi method, focused in this review, has allowed for the visualisation of entire neurons and glia in high detail and with good contrast. Moreover, compared to Nissl staining and silver impregnation, the Golgi method has the beneficial feature of characteristically selective staining. Because neurons are stained only sparsely with the Golgi method, it is a powerful staining technique for providing a complete, detailed representation of a single neuron. The aim of this review was to discuss the history and evolution of the Golgi method.


Dendritic vulnerability in neurodegenerative disease: insights from analyses of cortical pyramidal neurons in transgenic mouse models

In neurodegenerative disorders, such as Alzheimer’s disease, neuronal dendrites and dendritic spines undergo significant pathological changes. Because of the determinant role of these highly dynamic structures in signaling by individual neurons and ultimately in the functionality of neuronal networks that mediate cognitive functions, a detailed understanding of these changes is of paramount importance. Mutant murine models, such as the Tg2576 APP mutant mouse and the rTg4510 tau mutant mouse have been developed to provide insight into pathogenesis involving the abnormal production and aggregation of amyloid and tau proteins, because of the key role that these proteins play in neurodegenerative disease. This review showcases the multidimensional approach taken by our collaborative group to increase understanding of pathological mechanisms in neurodegenerative disease using these mouse models. This approach includes analyses of empirical 3D morphological and electrophysiological data acquired from frontal cortical pyramidal neurons using confocal laser scanning microscopy and whole-cell patch-clamp recording techniques, combined with computational modeling methodologies. These collaborative studies are designed to shed insight on the repercussions of dystrophic changes in neocortical neurons, define the cellular phenotype of differential neuronal vulnerability in relevant models of neurodegenerative disease, and provide a basis upon which to develop meaningful therapeutic strategies aimed at preventing, reversing, or compensating for neurodegenerative changes in dementia.

Keywords: Alzheimer’s disease, Amyloid, Computational modeling, Dendritic spine, Tau, Whole-cell patch-clamp

Neocortical pyramidal neurons possess extensive apical and basilar dendritic trees, which integrate information from thousands of excitatory and inhibitory synaptic inputs. Dendrites respond to inputs with postsynaptic potentials, which are relayed to the soma where they are summed; if a threshold potential is exceeded, an action potential is generated. Voltage attenuation in space and time along dendrites is fundamental to summation and is influenced by a number of interacting morphological properties (such as diameter and length) and active properties (such as distribution of ion channels) of the dendritic shaft (Hausser et al. 2000; Kampa et al. 2007; Stuart and Spruston 2007; Henry et al. 2008). Further complexity arises from the presence of thousands of biophysically active dendritic spines, the principal receptive site for excitatory glutamatergic inputs to a neuron.

Dendrites and dendritic spines in particular are dynamic structures, which undergo significant changes across the life span. Under non-pathological conditions, the number of spines on pyramidal neuron dendrites increases substantially over the course of development, is reduced during maturation to adulthood, and remains relatively stable during adulthood (for review see Bhatt et al. 2009). Then, during normal aging, significant changes in spine number, distribution, and morphology occur (Duan et al. 2003). Spines also undergo significant structural modifications under conditions where synaptic strength is experimentally modified, usually with protocols designed to evoke longterm potentiation or long-term depression (for review see Alvarez and Sabatini 2007; Bhatt et al. 2009; Holtmaat and Svoboda 2009). In many neurodegenerative diseases, significant alterations in the dendritic arbor occur, together with substantial spine loss and alterations in spine morphology (reviewed in Halpain et al. 2005). Gaining an understanding of these sublethal changes to neurons, which detrimentally impact neuronal signaling, and ultimately cognitive function, is an important goal.

Transgenic mouse models have been useful for elucidating mechanisms of amyloid-and tau-induced pathology, although no single model fully recapitulates human disease (for review see Duff and Suleman 2004; Spires et al. 2005). Two of the most commonly employed mouse models of neurodegenerative disease are the Tg2576 amyloid precursor protein (APP) mutant mouse and the rTg4510 tau mutant mouse, which develop significant pathological aggregations of amyloid-beta (Aβ) and tau proteins, respectively. Tg2576 transgenic mice overexpress the Swedish double mutation of the human APP gene, which leads to progressive formation of soluble Aβ peptides and fibrillar Aβ deposits in the form of amyloid plaques. Increased Aβ levels in these mice are associated with progressive structural changes to neurons (although not with neuron death), and cognitive impairment (Hsiao et al. 1996). In the rTg4510 mouse model, expression of the P301L mutant human tau variant leads to progressive development of neurofibrillary tangles (NFTs), neuronal death, and memory impairment reminiscent of the pathology observed in human tauopathies (Santacruz et al. 2005).

In this review, we discuss changes in the structure and function of dendrites and spines of pyramidal neurons that are associated with pathological aggregation of Aβ and tau in these and other mouse models of neurodegenerative disease. We also discuss, as a point of comparison, the impact of normal brain aging on the morphofunctional properties of pyramidal neurons in aged macaque monkeys. This is not a comprehensive review of the literature on such changes in human neurodegenerative diseases or of the many studies of these changes in mouse models (for reviews see Duff and Suleman 2004; Spires and Hyman 2004;Lewis and Hutton 2005; Duyckaerts et al. 2008; Giannakopoulos et al. 2009). Rather, the focus here is specifically on our multidimensional collaborative efforts to understand the functional consequences of pathological changes in the structure of individual layer 3 frontal cortical pyramidal neurons in commonly employed mouse models of neurodegeneration. These studies focus on layer 3 pyramidal neurons in the neocortex because they are the principal neurons involved in corticocortical circuits that mediate many cognitive functions of the frontal cortex and may be selectively targeted in neurodegenerative disease (Morrison and Hof 1997, 2002; Hof and Morrison 2004).

An external file that holds a picture, illustration, etc.
Object name is nihms273899f1.jpg


Structural properties of mouse neocortical pyramidal neurons. a 40 × CLSM image of a layer 3 pyramidal cell from the frontal cortex of a wild-type mouse. The neuron was filled with biocytin during patch-clamp recording and subsequently labeled with Alexastreptavidin 488. b100 × CLSM image of the spiny dendrite shown within the red box in (a). c Electron micrograph showing the ultrastructure of a pyramidal neuron spine, with a prominent postsynaptic density and spine apparatus. Courtesy of Dr. Alan Peters. Scale bars a 40, b 5, c 0.5 μm

The structural properties of dendrites underlie their passive membrane (cable) properties, which are membrane capacitance Cm, specific membrane resistance Rm and axial resistivity Ra. These cable properties determine, at a fundamental level, the degree of summation of synaptic inputs and the spatial distribution of electrical signals. Because summation of synaptic inputs by dendrites is determined in large part by their structure, dendritic morphology plays a critical role in action potential generation (Mainen and Sejnowski 1996; Koch and Segev 2000; Euler and Denk 2001; Vetter et al. 2001; Krichmar et al. 2002; Ascoli 2003). Importantly, both branching topology and surface irregularities including dendritic varicosities (Surkis et al. 1998) contribute to the excitable properties of neurons. For example, recent simulation studies have demonstrated that marked differences in the efficacy of action potential back propagation in different neural types are attributable in large part to variations in dendritic morphology (for review see Stuart et al. 1997;Waters et al. 2005).

Integration of synaptic inputs by dendrites is mediated not only by basic passive cable properties, but also by active properties, which include the number and distribution of voltage, ligand, as well as second messenger-gated transmembrane ionic channels (reviewed in Migliore and Shepherd 2002; Magee and Johnston 2005;Johnston and Narayanan 2008; Nusser 2009). Dendrites possess a rich array of sodium, calcium, and potassium channels that are distributed either uniformly or non-uniformly across a given dendrite (for reviews, see Johnston et al. 1996; Migliore and Shepherd 2002; Magee and Johnston 2005; Johnston and Narayanan 2008; Nusser 2009). For example, layer 5 cortical pyramidal neuron dendrites possess a gradient of the hyperpolarization-activated mixed cationic HCN channels that increase in density from the soma to the apical tuft (Berger et al. 2001; Lorincz et al. 2002). The interaction of intrinsic ionic and synaptic conductances with passive properties determined by dendritic morphology can effectively alter the cable properties of the dendritic tree (Bernander et al. 1991; Segev and London 2000; Bekkers and Hausser 2007) adding a further layer of complexity to signaling by individual neurons. Computational modeling approaches have been extensively employed to shed light on dendritic structure–function relationships and into potential interactions between a multitude of dendritic active and passive properties. These approaches, as discussed at the end of this review, have been useful for gaining insight on dendrites that are too thin to be studied with electrophysiological approaches, and for providing empirical researchers with testable hypotheses relevant to the functional consequences of dendritic dystrophy in neurodegenerative diseases.

Dendritic spines

Dendrites of glutamatergic pyramidal neurons are studded by thousands of dendritic spines, which are the site of most of the glutamatergic synapses in the brain, that confer further functional complexity to these processes. Spine density, shape, and distribution are all important contributors to neuronal excitability that are superimposed on the electrophysiological properties of dendritic shafts (Wilson 1988; Stratford et al. 1989;Baer and Rinzel 1991; Tsay and Yuste 2002). Spines are small appendages that extend from approximately 0.5–3 lm from dendritic shafts and are usually <1 μm in diameter (Fig. 1b, c). Mouse layer 3 frontal cortical pyramidal neurons typically possess approximately 6,000 dendritic spines (Rocher et al. 2008). Spines can be broadly categorized as falling into one of several different morphological types, namely “thin”, “stubby”, “mushroom”, and “filopodia” which are normally seen in large numbers only during development (for review see Yuste and Bonhoeffer 2004; Bourne and Harris 2007). Although most spines do fall into one of these broad categories, serial electron microscopy reveals that there is also a continuum between the different morphological subtypes (Bourne and Harris 2007). Spine morphology determines the strength, stability and function of excitatory synaptic connections that subserve the neuronal networks underlying cognitive function. Smaller spines, such as the thin and filopodia types are less stable and more motile (Trachtenberg et al. 2002; Kasai et al. 2003; Holtmaat et al. 2005) and as a result, are more plastic than large spines such as the mushroom and stubby types (Grutzendler et al. 2002). The size and morphology of the spine head is correlated with the number of docked presynaptic vesicles (Schikorski and Stevens 1999) and the number of postsynaptic receptors (Nusser et al. 1998), and hence with the size of synaptic currents and synapse strength. In addition, a small head size permits fast diffusion of calcium within the spine, while the neck length shapes the time constant for calcium compartmentalization (for review see Yuste and Bonhoeffer 2001; Nimchinsky et al. 2002), modulating postsynaptic mechanisms that play an important role in synaptic plasticity linked to functions, such as learning and memory (Holthoff et al. 2002; Alvarez and Sabatini 2007; Bloodgood and Sabatini 2007). Spine neck length and diameter also affect diffusional coupling between dendrite and spine (Svoboda et al. 1996; Yuste et al. 2000; Bloodgood and Sabatini 2005), and spine density and shape regulate the degree of anomalous diffusion of chemical signals within the dendrite (Santamaria et al. 2006). Spine morphology also varies dynamically in response to synaptic activity (Lendvai et al. 2000; Hering and Sheng 2001; Zuo et al. 2005).

Over the past few decades, it has become increasingly evident that local dendritic spine structure and distribution (Matus and Shepherd 2000) play a key role in the electrical and biochemical signaling of dendrites (Nimchinsky et al. 2002; Matus 2005; Bourne and Harris 2007). However, spines present challenges to the standard cable model of dendrites. The common way to model the effects of spines in a passive cable equation model is to reduce the membrane resistance and increase the membrane capacitance by a factor proportional to the increased membrane surface area due to spines (Jaslove 1992). This modification predicts that voltage should attenuate more drastically in space along spiny dendrites, relative to their smooth counterparts. Because of their capacity for plasticity and because they are the location of most excitatory synapses in the cerebral cortex, accurately characterizing the structure of dendritic spines is essential for understanding their contributions to neuronal, and ultimately to cognitive function.

An external file that holds a picture, illustration, etc.
Object name is nihms273899f2.jpg


Analytical methods to quantify morphological structure. a xy projection of the montage of CLSM tiled image stacks from a wildtype mouse layer 3 frontal cortical pyramidal neuron. Scale bar 40 μm. b Tree structure from the data shown in (a), extracted using NeuronStudio. cAutomatic spine detection and visualization in NeuronStudio. Left automatically detected spines overlaid on a maximum projection of a typical dataset; right the same data and spines volume-rendered in 3D. d Left 2D Rayburst sample used for diameter estimation. Rays cast using the sampling core is shown in orange with the one chosen as the diameter shown in blue. The green line indicates the surface detected by the Rayburst samples. Right spine diameter estimation using a 2D Rayburst run at the center of mass (small green squares) of a single layer. Theblue line indicates the resulting width of the structure as calculated by Rayburst, and provides an approximate length of 0.7 μm. e Optimized fits of scaling exponents (black fitted lines) and optimal scaling regions (gray shaded bands) for the apical dendrites of a typical layer 3 pyramidal cell projecting from superior temporal cortex to area 46 (see Kabaso et al. (2009) for details). f Contributions of branching and tapering exponents to global mass scaling (measured by total area exponent dA) in spine-corrected apical dendrites of long projection neurons of young rhesus monkeys, in the proximal and medial scaling regions (I, II in e). The total branching and tapering vary across the two scaling regions, yet the total area in each region is conserved [modified from Wearne et al. (2005) and Kabaso et al. (2009)]

An external file that holds a picture, illustration, etc.
Object name is nihms273899f4.jpg


Dendritic spine loss in pyramidal neurons from Tg2576 and rTg4510 mice. CLSM images of dendritic segments with spines of neurons from wild-type (top) and transgenic (bottom) mice. Scale bars 10 μm [modified from Rocher et al. (2008, 2009)]

3D reconstruction and analytical approaches

Early methods for digitizing 3D neuronal structures relied on interactive manual tracing from a computer screen (Capowski 1985). These methods were time-consuming, subjective, and lacked precision. In recent years, automated methods have been developed that use image analysis algorithms to extract neuronal morphology directly from 3D microscopy and overcome the limitations of manual techniques (Koh et al. 2002; He et al. 2003; Wearne et al. 2005). Newer methods use pattern recognition routines to track or detect a structure locally without the need for global image operations (Al-Kofahi et al. 2002; Streekstra and van Pelt 2002; Schmitt et al. 2004; Myatt et al. 2006; Santamaria-Pang et al. 2007). Most are designed to work on a broad range of signal-to-noise ratios and even on multiple imaging modalities. This results in increased computational complexity, which makes the use of these methods as interactive reconstruction tools for high-resolution data less than optimal.

In our studies of neuronal structure, morphological reconstruction is performed using NeuronStudio, a neuron morphology reconstruction software tool (http://www.mssm.edu/cnic/tools.html), developed by Wearne et al. (2005) and Rodriguez et al. (2006, 2008, 2009). Neuron- Studio has been designed for low computational complexity to allow interactive semi-automated extraction of neuronal morphology from medium to high-quality de-convolved 3D CLSM and MPLSM image stacks of fluorescently labeled neurons. It features automated extraction of dendrites and dendritic spines as well as a rich set of manual editing and visualization modalities. Figure 2a shows an xy projection of CLSM data from a wild-type mouse layer 3 cortical pyramidal neuron and Fig. 2b shows the automated reconstruction of the neuron’s dendritic arbor obtained using NeuronStudio. Because fluorescence intensity can vary with adequacy of filling, imaging depth and xyspatial extent in CLSM and MPLSM image stacks, data segmentation within NeuronStudio adapts the iterative self-organizing data analysis (ISODATA) method (Ridler and Calvard 1978) to compute local thresholds dynamically. This method is appropriate for datasets exhibiting a bimodal distribution of intensity values, such as the grayscale images characteristic of de-convolved LSM image stacks.

Automated 3D reconstruction of dendritic spines

The assessment of spine numbers and distribution and their classification into subtypes has historically been a labor intensive and relatively inaccurate process. Spine numbers could only be estimated because spines extending primarily in the z plane relative to the dendritic shaft could not be counted. With the advent of CLSM, accurate 3D spine assessment came a step closer, but was still a highly timeconsuming and labor-intensive undertaking. Improving upon previous spine detection algorithms (Koh et al. 2002; Cheng et al. 2007), Rodriguez et al. (2008) devised an efficient and robust method for automated spine detection, available in NeuronStudio.

3D measures of spatial complexity

Traditionally, Sholl analysis (1953) has been used in two dimensions to quantify the spatial complexity of dendritic branching patterns with increasing distance from the soma. Fractal analyses have also been used (Smith et al. 1989; Caserta et al. 1995; Jelinek and Elston 2001; Henry et al. 2002) to quantify spatial complexity as a power law scaling exponent, describing the rate of change of the number of branches over a large portion of the dendrites, and best visualized as the slope of a log–log plot of these two quantities. We have recently extended this work (Rothnie et al. 2006; Kabaso et al. 2009) to include three power law exponents describing global spatial complexity: rates of change of dendritic mass, branching, and taper.


In conclusion, these analytical tools provide rapid, objective analyses of the high-resolution data that we collect from wild-type and transgenic mouse neurons. With NeuronStudio, we are able to perform analyses of differences in dendrite diameter and spine shapes that were not available previously, and these kinds of studies are currently ongoing in our laboratories. Moving forward, we will evaluate whether the global patterns in spatial complexity observed in rhesus monkey pyramidal neurons are similarly present in mouse neurons. We will also compare the spatial complexity of wild-type and transgenic neurons to determine whether global mass homeostasis is conserved in neurodegeneration as it seems to be in aging.

Alterations in the structure of dendrites and spines of cortical pyramidal neurons in Tg2576 and rTg4510 mutant mice

Dendritic changes in neurodegenerative disease ….

Clearly, there is a need for many more studies on the functional electrophysiological consequences to individual neurons of the significant structural changes in neurodegenerative disease. Given the lack of such studies, and also technical considerations such as space clamp limitations and the impossibility of recording from distal dendrites or spines, the use of modeling methods to understand potential functional consequences of structural changes is very important.

Insights from modeling

For nearly 60 years, mathematical models have been used to investigate neuronal function. Hodgkin and Huxley’s (1952) mathematical model of action potential generation predicted the existence of ion channels, decades before ion channels were observed experimentally (Neher and Sakmann 1976). Other models were groundbreaking in describing how dendrites filter signals as passive electrical cables (Rall 1959; Goldstein and Rall 1974), but were limited in their ability to apply directly to realistic morphologic data. Since then, applications of mathematical techniques (Fitzhugh 1961; Nagumo et al. 1962; Rinzel and Ermentrout 1989), advances in computational software (Bower and Beeman 1998; Carnevale and Hines 2006), model reduction (Clements and Redman 1989; Pinsky and Rinzel 1994), and computing power have resulted in models that have great potential for yielding insights into neuronal function. In particular, models have shown that morphology is a critical determinant of neuronal firing properties (Zador et al. 1995; Mainen and Sejnowski 1996; Vetter et al. 2001; Schaefer et al. 2003; Stiefel and Sejnowski 2007). The effects of morphology are further amplified by the actions of ion channels distributed throughout the dendrites (for review see Johnston and Narayanan 2008), both of which shape patterns of synaptic input. Our ability to understand neuronal function depends largely on analysis of the nonlinear interactions between morphology, electrical membrane properties, and synaptic input.


Our modeling results make several predictions that may apply to transgenic mouse models of neurodegeneration. First, changes in dendrite length, diameter, and spine densities/numbers in transgenic neurons compared to wildtype neurons may significantly impact the attenuation of signals to and from the soma. This may happen over an entire neuron, or in limited regions, such as dendrites passing through or near fibrillar amyloid deposits, or in apical tufts that undergo atrophy. Long projection neurons are particularly vulnerable in AD (Hof et al. 1990; Bussiére et al. 2003; Hof and Morrison 2004), giving greater weight to the significant differences in voltage attenuation observed in long projection neurons (Kabaso et al. 2009). Second, changes in spine density may affect the diffusion rate of intracellular messengers and ions. Elongated dendrites may result in less trapping of electrical and chemical signals within spines; this phenomenon may be functionally significant. We must also consider how spine loss might impact the amount of excitatory input that a neuron receives. Finally, it is likely that interactions between morphology and active parameters vary between wild-type and transgenic neurons. To study this more fully we must measure both detailed morphological properties and ionic currents in wild-type and transgenic neurons. To evaluate the degree to which these predictions truly apply to the transgenic mouse models, we will apply the mathematical methods described here directly to the transgenic data that we have collected. These computational studies will likely lead to explicit predictions of which parameters to change, and by how much, to counteract morphological changes that affect physiological function in neurodegenerative disease.

Read Full Post »

Remyelination of axon requires Gli1 inhibition

Larry H. Bernstein, MD, FCAP, Curator



Inhibition of Gli1 Enhances Remyelination Abilities of Endogenous Stem Cell Populations





If myelin is damaged, the speed of nerve impulse transmission slows substantially. Multiple sclerosis is one example of a disease that causes systematic loss of the myelin sheath. Inflammatory demyelinating diseases also cause progressive damage and loss of the myelin sheath. Regenerating the myelin sheath in these patients is one of the goals of regenerative medicine.

A good deal of data tells us that endogenous remyelination does occur. Unfortunately, this process is overwhelmed by the degree of demyelination in these diseases. A stem cell population called the parenchymal oligodendrocyte progenitor cells and endogenous adult neural stem cells in the brain are known to remyelinate demyelinated axons.

The Salzer laboratory at the New York Neuroscience Institute examined the ability of a specific adult neural stem cell population to remyelinate axons. These stem cells expressed the transcription factor Gli1.

Salzer and his team showed that this subventricular zone-specific group of neural stem cells were efficiently recruited to demyelinated portions of the brain. This same neural stem cell population was never observed entering healthy axon tracts. This finding shows that these cells seem to specialize in making new myelin sheaths for damaged axon tracts.

Since these neural stem cells expressed Gli1, and since there are drugs that can inhibit Gli1 activity, Salzer’s group wanted to show that Gli1 was a necessary factor for neural stem cell activity. Surprisingly, differentiation of these neural stem cells into oligodendrocytes (which make myelin and remyelinate axons) is significantly enhanced by inhibition of Gli1.

A specific signaling pathway called the hedgehog pathway is known to activate Gli1 and other members of the Gli gene family. However, when the hedgehog pathway in these neural stem cells was completely inhibited, it did not have the same effect and Gli1 inhibition. This suggests that Gli1 is doing more than responding to the hedgehog pathway in these neural stem cells.

Salzer and his colleagues showed that Gli1 inhibition improved myelin deposition in an animal model of experimental autoimmune encephalomyelitis; an inflammatory demyelination disease. Thus, inhibition of Gli1 activity in this preclinical model system increase regeneration of the myelin sheath in demyelinated neurons.

This work elegantly showed that endogenous neural stem cells that can remyelinate axons are present and can be activated by inhibiting Gli1. Furthermore, this activation will nicely enhance the therapeutic capacity of these endogenous cells. This potentially identifies a new therapeutic avenue for the treatment of demyelinating disorders.

This work was published in Nature. 2015 Oct 15;526(7573):448-52. doi: 10.1038/nature14957.


Inhibition of Gli1 mobilizes endogenous neural stem cells for remyelination
Inhibition of Gli1 mobilizes endogenous neural stem cells for remyelination.

 Nature. 2015 Oct 15;526(7573):448-52.   http://dx.doi.org:/10.1038/nature14957   Epub 2015 Sep 30.

Enhancing repair of myelin is an important but still elusive therapeutic goal in many neurological disorders. In multiple sclerosis, an inflammatory demyelinating disease, endogenous remyelination does occur but is frequently insufficient to restore function. Both parenchymal oligodendrocyte progenitor cells and endogenous adult neural stem cells resident within the subventricular zone are known sources of remyelinating cells. Here we characterize the contribution to remyelination of a subset of adult neural stem cells, identified by their expression of Gli1, a transcriptional effector of the sonic hedgehog pathway. We show that these cells are recruited from the subventricular zone to populate demyelinated lesions in the forebrain but never enter healthy, white matter tracts. Unexpectedly, recruitment of this pool of neural stem cells, and their differentiation into oligodendrocytes, is significantly enhanced by genetic or pharmacological inhibition of Gli1. Importantly, complete inhibition of canonical hedgehog signalling was ineffective, indicating that the role of Gli1 both in augmenting hedgehog signalling and in retarding myelination is specialized. Indeed, inhibition of Gli1 improves the functional outcome in a relapsing/remitting model of experimental autoimmune encephalomyelitis and is neuroprotective. Thus, endogenous neural stem cells can be mobilized for the repair of demyelinated lesions by inhibiting Gli1, identifying a new therapeutic avenue for the treatment of demyelinating disorders.


Read Full Post »

Older Posts »