Funding, Deals & Partnerships: BIOLOGICS & MEDICAL DEVICES; BioMed e-Series; Medicine and Life Sciences Scientific Journal – http://PharmaceuticalIntelligence.com
Science Policy Forum: Should we trust healthcare explanations from AI predictive systems?
Some in industry voice their concerns
Curator: Stephen J. Williams, PhD
Post on AI healthcare and explainable AI
In a Policy Forum article in Science “Beware explanations from AI in health care”, Boris Babic, Sara Gerke, Theodoros Evgeniou, and Glenn Cohen discuss the caveats on relying on explainable versus interpretable artificial intelligence (AI) and Machine Learning (ML) algorithms to make complex health decisions. The FDA has already approved some AI/ML algorithms for analysis of medical images for diagnostic purposes. These have been discussed in prior posts on this site, as well as issues arising from multi-center trials. The authors of this perspective article argue that choice of type of algorithm (explainable versus interpretable) algorithms may have far reaching consequences in health care.
Summary
Artificial intelligence and machine learning (AI/ML) algorithms are increasingly developed in health care for diagnosis and treatment of a variety of medical conditions (1). However, despite the technical prowess of such systems, their adoption has been challenging, and whether and how much they will actually improve health care remains to be seen. A central reason for this is that the effectiveness of AI/ML-based medical devices depends largely on the behavioral characteristics of its users, who, for example, are often vulnerable to well-documented biases or algorithmic aversion (2). Many stakeholders increasingly identify the so-called black-box nature of predictive algorithms as the core source of users’ skepticism, lack of trust, and slow uptake (3, 4). As a result, lawmakers have been moving in the direction of requiring the availability of explanations for black-box algorithmic decisions (5). Indeed, a near-consensus is emerging in favor of explainable AI/ML among academics, governments, and civil society groups. Many are drawn to this approach to harness the accuracy benefits of noninterpretable AI/ML such as deep learning or neural nets while also supporting transparency, trust, and adoption. We argue that this consensus, at least as applied to health care, both overstates the benefits and undercounts the drawbacks of requiring black-box algorithms to be explainable.
Types of AI/ML Algorithms: Explainable and Interpretable algorithms
Interpretable AI: A typical AI/ML task requires constructing algorithms from vector inputs and generating an output related to an outcome (like diagnosing a cardiac event from an image). Generally the algorithm has to be trained on past data with known parameters. When an algorithm is called interpretable, this means that the algorithm uses a transparent or “white box” function which is easily understandable. Such example might be a linear function to determine relationships where parameters are simple and not complex. Although they may not be as accurate as the more complex explainable AI/ML algorithms, they are open, transparent, and easily understood by the operators.
Explainable AI/ML: This type of algorithm depends upon multiple complex parameters and takes a first round of predictions from a “black box” model then uses a second algorithm from an interpretable function to better approximate outputs of the first model. The first algorithm is trained not with original data but based on predictions resembling multiple iterations of computing. Therefore this method is more accurate or deemed more reliable in prediction however is very complex and is not easily understandable. Many medical devices that use an AI/ML algorithm use this type. An example is deep learning and neural networks.
The purpose of both these methodologies is to deal with problems of opacity, or that AI predictions based from a black box undermines trust in the AI.
For a deeper understanding of these two types of algorithms see here:
How interpretability is different from explainability
Why a model might need to be interpretable and/or explainable
Who is working to solve the black box problem—and how
What is interpretability?
Does Chipotle make your stomach hurt? Does loud noise accelerate hearing loss? Are women less aggressive than men? If a machine learning model can create a definition around these relationships, it is interpretable.
All models must start with a hypothesis. Human curiosity propels a being to intuit that one thing relates to another. “Hmm…multiple black people shot by policemen…seemingly out of proportion to other races…something might be systemic?” Explore.
People create internal models to interpret their surroundings. In the field of machine learning, these models can be tested and verified as either accurate or inaccurate representations of the world.
Interpretability means that the cause and effect can be determined.
What is explainability?
ML models are often called black-box models because they allow a pre-set number of empty parameters, or nodes, to be assigned values by the machine learning algorithm. Specifically, the back-propagation step is responsible for updating the weights based on its error function.
To predict when a person might die—the fun gamble one might play when calculating a life insurance premium, and the strange bet a person makes against their own life when purchasing a life insurance package—a model will take in its inputs, and output a percent chance the given person has at living to age 80.
Below is an image of a neural network. The inputs are the yellow; the outputs are the orange. Like a rubric to an overall grade, explainability shows how significant each of the parameters, all the blue nodes, contribute to the final decision.
In this neural network, the hidden layers (the two columns of blue dots) would be the black box.
For example, we have these data inputs:
Age
BMI score
Number of years spent smoking
Career category
If this model had high explainability, we’d be able to say, for instance:
The career category is about 40% important
The number of years spent smoking weighs in at 35% important
The age is 15% important
The BMI score is 10% important
Explainability: important, not always necessary
Explainability becomes significant in the field of machine learning because, often, it is not apparent. Explainability is often unnecessary. A machine learning engineer can build a model without ever having considered the model’s explainability. It is an extra step in the building process—like wearing a seat belt while driving a car. It is unnecessary for the car to perform, but offers insurance when things crash.
The benefit a deep neural net offers to engineers is it creates a black box of parameters, like fake additional data points, that allow a model to base its decisions against. These fake data points go unknown to the engineer. The black box, or hidden layers, allow a model to make associations among the given data points to predict better results. For example, if we are deciding how long someone might have to live, and we use career data as an input, it is possible the model sorts the careers into high- and low-risk career options all on its own.
Perhaps we inspect a node and see it relates oil rig workers, underwater welders, and boat cooks to each other. It is possible the neural net makes connections between the lifespan of these individuals and puts a placeholder in the deep net to associate these. If we were to examine the individual nodes in the black box, we could note this clustering interprets water careers to be a high-risk job.
In the previous chart, each one of the lines connecting from the yellow dot to the blue dot can represent a signal, weighing the importance of that node in determining the overall score of the output.
If that signal is high, that node is significant to the model’s overall performance.
If that signal is low, the node is insignificant.
With this understanding, we can define explainability as:
Knowledge of what one node represents and how important it is to the model’s performance.
So how does choice of these two different algorithms make a difference with respect to health care and medical decision making?
The authors argue:
“Regulators like the FDA should focus on those aspects of the AI/ML system that directly bear on its safety and effectiveness – in particular, how does it perform in the hands of its intended users?”
A suggestion for
Enhanced more involved clinical trials
Provide individuals added flexibility when interacting with a model, for example inputting their own test data
More interaction between user and model generators
Determining in which situations call for interpretable AI versus explainable (for instance predicting which patients will require dialysis after kidney damage)
Other articles on AI/ML in medicine and healthcare on this Open Access Journal include
For the first time, treatment with genetically engineered T-cells has used CD4 T-cells instead of the CD8 T-cells, which are used in the chimeric antigen receptor (CAR) T-cell approach. Early data suggest that this CD4 T-cell approach has activity against solid tumors, whereas the CAR T-cell approach so far has achieved dramatic success in hematologic malignancies.
In the new approach, CD4 T-cells were genetically engineered to target MAGE-A3, a protein found on many tumor cells. The treatment was found to be safe in patients with metastatic cancers, according to data from a phase 1 clinical study presented here at the American Association for Cancer Research (AACR) 2016 Annual Meeting.
“This is the first trial testing an immunotherapy using genetically engineered CD4 T-cells,” senior author Steven A. Rosenberg, MD, PhD, chief of the Surgery Branch at the National Cancer Institute (NCI), told Medscape Medical News.
Most approaches use CD8 T-cells. Although CD8 T-cells are known be cytotoxic and CD4 T-cells are normally considered helper cells, CD4 T-cells can induce tumor regression, he said.
Louis M. Weiner, MD, director of the Lombardi Comprehensive Cancer Center at Georgetown University, in Washington, DC, indicated that in contrast with CAR T-cells, these CD4 T-cells target proteins on solid tumors. “CAR T-cells are not tumor specific and do not target solid tumors,” he said.
Engineering CD4 Cells
Immunotherapy with engineered CD4 T-cells was personalized for each patient whose tumors had not responded to or had recurred following treatment with least one standard therapy. The immunotherapy was specific for patients in whom a specific human leukocyte antigen (HLA) — HLA-DPB1*0401 — was found to be expressed on their cells and whose tumors expressed MAGE-A3.
MAGE-A3 belongs to a class of proteins expressed during fetal development. The expression is lost in normal adult tissue but is reexpressed on tumor cells, explained presenter Yong-Chen William Lu, PhD, a research fellow in the Surgery Branch of the NCI.
Targeting MAGE-A3 is relevant, because it is frequently expressed in a variety of cancers, such as melanoma and urothelial, esophageal, and cervical cancers, he pointed out.
Researchers purified CD4 T-cells from the peripheral blood of patients. Next, the CD4 T-cells were genetically engineered with a retrovirus carrying the T-cell receptor (TCR) gene that recognizes MAGE-A3. The modified cells were grown ex vivo and were transferred back into the patient.
Clinical Results
Dr Lu presented data for 14 patients enrolled into the study: eight patients received cell doses from 10 million to 30 billion cells, and six patients received up to 100 billion cells.
This was similar to a phase 1 dose-finding study, except the researchers were seeking to determine the maximum number of genetically engineered CD4 T-cells that a patient could safely receive.
One patient with metastatic cervical cancer, another with metastatic esophageal cancer, and a third with metastatic urothelial cancer experienced partial objective responses. At 15 months, the response is ongoing in the patient with cervical cancer; after 7 months of treatment, the response was durable in the patient with urothelial cancer; and a response lasting 4 months was reported for the patient with esophageal cancer.
Dr Lu said that a phase 2 trial has been initiated to study the clinical responses of this T-cell receptor therapy in different types of metastatic cancers.
In his discussion of the paper, Michel Sadelain, MD, of the Memorial Sloan Kettering Cancer Center, New York City, said, “Although therapy with CD4 cells has been evaluated using endogenous receptor, this is the first study using genetically engineered CD4 T-cells.”
Although the study showed that therapy with genetically engineered T-cells is safe and efficacious at least in three patients, the mechanism of cytotoxicity remains unclear, Dr Sadelain indicated.
Comparison With CAR T-cells
CAR T-cells act in much the same way. CARs are chimeric antigen receptors that have an antigen-recognition domain of an antibody (the V region) and a “business end,” which activates T-cells. In this case, CD8 T-cells from the patients are used to genetically engineer T-cells ex vivo. In the majority of cases, dramatic responses have been seen in hematologic malignancies.
CARs, directed against self-proteins, result in on-target, off-tumor effects, Gregory L. Beatty, MD, PhD, assistant professor of medicine at the University of Pennsylvania, in Philadelphia, indicated when he reported the first success story of CAR T-cells in a solid pancreatic cancer tumor.
Side effects of therapy with CD4 T-cells targeting MAGE-A3 were different and similar to side effects of chemotherapy, because patients received a lymphodepleting regimen of cyclophosphamide and fludabarine. Toxicities included high fever, which was experienced by the majority of patients (12/14). The fever lasted 1 to 2 weeks and was easily manageable.
High levels of the cytokine interleukin-6 (IL-6) were detected in the serum of all patients after treatment. However, the elevation in IL-6 levels was not considered to be a cytokine release syndrome, because no side effects occurred that correlated with the syndrome, Dr Liu indicated.
He also indicated that future studies are planned that will employ genetically engineered CD4 T-cells in combination with programmed cell death protein 1–blocking antibodies.
This study was funded by Intramural Research Program of the National Institutes of Health. The NCI’s research and development of T-cell receptor therapy targeting MAGE-A3 are supported in part under a cooperative research and development agreement between the NCI and Kite Pharma, Inc. Kite has an exclusive, worldwide license with the NIH for intellectual property relating to retrovirally transduced HLA-DPB1*0401 and HLA A1 T-cell receptor therapy targeting MAGE-A3 antigen. Dr Lu and Dr Rosenberg have disclosed no relevant financial relationships.
American Association for Cancer Research (AACR) 2016 Annual Meeting: Abstract CT003, presented April 17, 2016.
Searches Related to immunotherapy using genetically engineered CD4 T-cells
To be effective for the treatment of cancer and infectious diseases, T cell adoptive immunotherapy requires large numbers of cells with abundant proliferative reserves and intact effector functions. We are achieving these goals using a gene therapy strategy wherein the desired characteristics are introduced into a starting cell population, primarily by high efficiency lentiviral vector-mediated transduction. Modified cells are then expanded using ex vivo expansion protocols designed to minimally alter the desired cellular phenotype. In this article, we focus on strategies to (1) dissect the signals controlling T cell proliferation; (2) render CD4 T cells resistant to HIV-1 infection; and (3) redirect CD8 T cell antigen specificity.
Adoptive T cell therapy is a form of transfusion therapy involving the infusion of large numbers of T cells with the aim of eliminating, or at least controlling, malignancies or infectious diseases. Successful applications of this technique include the infusion of CMV-or EBVspecific CTLs to protect immunosuppressed patients from these transplantation-associated diseases [1,2]. Furthermore, donor lymphocyte infusions of ex vivo-expanded allogeneic T cells have been used to successfully treat hematological malignancies in patients with relapsed disease following allogeneic hematopoietic stem cell transplant [3]. However, in many other malignancies and chronic viral infections such as HIV-1, adoptive T cell therapy has achieved inconsistent and/or marginal successes. Nevertheless, there are compelling reasons for optimism on this strategy. For example, the existence of HIV-positive elite non-progressors [4], as well as the correlation between the presence of intratumoral T cells and a favorable prognosis in malignancies such as ovarian [5,6] and colon carcinoma [7,8], provides in vivo evidence for the critical role of the immune system in controlling both HIV and cancer.
The key to successful adoptive immunotherapy strategies appears to consist of (1) using the “right” T cell type(s) and (2) obtaining therapeutically effective numbers of these cells without compromising their effector functions or their ability to engraft within the host. This article is focused on strategies employed in our laboratory to generate the “right” cell through genetic engineering approaches, with an emphasis on redirecting the antigen specificity of CD8 T cells, and rendering CD4 T cells resistant to HIV-1 infection. The article by Paulos et al. describes the evolving process of how to best obtain therapeutically effective numbers of the “right” cells by optimizing ex vivo cell expansion strategies.
Our laboratory’s overall strategy and flow plan for development and evaluation of engineered T cells is depicted in Fig. 1. We work almost exclusively with primary human T cells; little or no work is performed with conventional established cell lines. Thus, we benefit substantially from our close association with the UPenn Human Immunology Core. The Core performs leukaphereses on healthy donors 2–3 times a week, and provides purified peripheral blood mononuclear cell subsets, ensuring a constant influx of fresh human T cells into our laboratory. We have extensive experience in developing both bead- and cell-based artificial antigen presenting cells (aAPCs), as described in detail in the article by Paulos et al. The ability to genetically modify T cells at high efficiency is critical for virtually every project within the laboratory. We have adapted the lentiviral vector system described by Dull [15] for most, but not all, of the engineering applications in our laboratory.
CD4 T cells are the primary target of HIV-1, and decreasing CD4 T cell numbers is a hallmark of advancing HIV-1 disease [34]. Thus, strategies that protect CD4 T cells from HIV-1 infection in vivo would conceivably provide sufficient immunological help to control HIV-1 infection. Our early observations that CD3/CD28 costimulation resulted in improved ex vivo expansion of CD4 T cells from both healthy and HIV-infected donors, as well as enhanced resistance to HIV-1 infection [35,36], ultimately led to the first-in-human trial of lentiviral vector-modified CD4 T cells [37]. In this trial, CD4 T cells from HIV-positive subjects who had failed antiretroviral therapy were transduced with a lentiviral vector encoding an antisense RNA that targeted a 937 bp region in the HIV-1 envelope gene. Preclinical studies demonstrated that this antisense region, directed against the HIV-1NL4-3 envelope, provided robust protection from a broad range of both R5-and X4-tropic HIV-1 isolates [38]. One year after administration of a single dose of the gene-modified cells, four of the five enrolled patients had increased peripheral blood CD4 T cell counts, and in one subject, a 1.7 log decrease in viral load was observed. Finally, in two of the five patients, persistence of the gene-modified cells was detected one year post-infusion.
Since its identification as the primary co-receptor involved in HIV transmission, CCR5 has attracted considerable attention as a target for HIV therapy [42,43]. Indeed, “experiments of nature” have shown that individuals with a homozygous CCR5 Δ32 deletion are highly resistant to HIV-1 infection. Thus, we hypothesized that knocking out the CCR5 locus would generate CD4 T cells permanently resistant to infection by R5 isolates of HIV-1. To test this hypothesis we took advantage of zinc-finger nuclease (ZFN) technology [44]. ZFNs introduce sequencespecific double-strand DNA breakage, which is imperfectly repaired by non-homologous endjoining. This results in the permanent disruption of the genomic target, a process termed genome editing (Fig. 3).
Genetic modification of T cells to redirect antigen specificity is an attractive strategy compared to the lengthy process of growing T cell lines or CTL clones for adoptive transfer. Genetically modified, adoptively transferred T cells are capable of long-term persistence in humans [37, 46,47], demonstrating the feasibility of this approach. When compared to the months it can take to generate an infusion dose of antigen-specific CTL lines or clones from a patient, a homogeneous population of redirected antigen-specific cells can be expanded to therapeutically relevant numbers in about two weeks [3]. Several strategies are being explored to bypass the need to expand antigen-specific T cells for adoptive T cell therapy. The approaches currently studied in our laboratory involve the genetic transfer of chimeric antigen receptors and supraphysiologic T cell receptors.
Chimeric antigen receptors (CARs or T-bodies) are artificial T cell receptors that combine the extracellular single-chain variable fragment (scFv) of an antibody with intracellular signaling domains, such as CD3ζ or Fc(ε)RIγ [48–50]. When expressed on T cells, the receptor bypasses the need for antigen presentation on MHC since the scFv binds directly to cell surface antigens. This is an important feature, since many tumors and virus-infected cells downregulate MHCI, rendering them invisible to the adaptive immune system. The high-affinity nature of the scFv domain makes these engineered T cells highly sensitive to low antigen densities. In addition, new chimeric antigen receptors are relatively easy to produce from hybridomas. The key to this approach is the identification of antigens with high surface expression on tumor cells, but reduced or absent expression on normal tissues. Since one can redirect both CD4 and CD8 T cells, the T-body approach to immunotherapy represents a near universal “off the shelf” method to generate large numbers of antigen-specific helper and cytotoxic T cells.
Many T-bodies targeting diverse tumors have been developed [51], and four have been evaluated clinically [52–55]. Three of the four studies were characterized by poor transgene expression and limited T-body engraftment. However, in a study of metastatic renal cell carcinoma using a T-body directed against carbonic anhydrase IX [55], T-body-expressing cells were detectable in the peripheral blood for nearly 2 months post-administration.
The major goals in the T-body field currently are to optimize their engraftment and maximize their effector functions. Our laboratory is addressing both problems simultaneously through an in-depth study of the requirements for T-body activation. We hypothesize that their limited persistence is due to incomplete cell activation due to the lack of costimulation. While naïve T cells depend on costimulation through CD28 ligation to avoid anergy and undergo full activation in response to antigen, it is recognized that effector cells also require costimulation to properly proliferate and produce cytokines [56]. Previous studies have shown that providing CD28 costimulation is crucial for the antitumoral function of adoptively transferred T cells and T-bodies [57–59]. Unlike conventional T cell activation, which requires two discrete signals, T-bodies can be engineered to provide both costimulation and CD3 signaling through one binding event.
A different approach for redirecting specificity to T cells for adoptive immunotherapy involves the genetic transfer of full-length TCR genes. A T cell’s specificity for its cognate antigen is solely determined by its TCR. Genes encoding the α and β chains of a T cell receptor (TCR) can be isolated from a T cell specific for the antigen of interest and restricted to a defined HLA allele, inserted into a vector, and then introduced into large numbers of T cells of individual patients that share the restricting HLA allele as well as the targeted antigen. In 1999, Clay and colleagues from Rosenberg’s group at the National Cancer Institute were the first to report the transfer of TCR genes via a retroviral vector into human lymphocytes and to show that T cells gained stable reactivity to MART-1 [67]. To date, many others have shown that the same approach can be used to transfer specificity for multiple viral and tumor associated antigens in mice and human systems. These T cells gain effector functions against the transferred TCR’s cognate antigen, as defined by proliferation, cytokine production, lysis of targets presenting the antigen, trafficking to tumor sites in vivo, and clearance of tumors and viral infection.
In 2006, Rosenberg’s group redirected patients’ PBLs with the naturally occurring, MART-1- specific TCR reported in 1999 by Clay. In the first clinical trial to test TCR-transfer immunotherapy, these modified T cells were infused into melanoma patients [68]. While the transduced T cells persisted in vivo, only two of the 17 patients had an objective response to this therapy. One issue revealed by the study was the poor expression of the transgenic TCRs by the transferred T cells. Nonetheless, the results from this trial showed the potential of TCR transfer immunotherapy as a safe form of therapy for cancer and highlighted the need to optimize such therapy to attain maximum potency.
The adoptive immunotherapy field is advancing by a tried-and-true method: learning from disappointments and moving forward. Our ability to fully realize the therapeutic potential of adoptive T cell therapy is tied to a more complete understanding of how human T cells receive signals, kill targets, and modulate effective immune responses. Our goal is to perform labbased experiments that provide insight into how primary T cells function in a manner that will facilitate and enable adoptive T cell therapy clinical trials. Our ability to efficiently modify (and expand) T cells ex vivo provides the opportunity to deliver sufficient immune firepower where it has heretofore been lacking. Sustained transgene expression, coupled with enhanced in vivo engraftment capability, will move adoptive immunotherapy into a realm where longterm therapeutic benefits are the norm rather than the exception.
Genetic Modification of T Lymphocytes for Adoptive Immunotherapy
Adoptive transfer of T lymphocytes is a promising therapy for malignancies—particularly of the hemopoietic system—and for otherwise intractable viral diseases. Efforts to broaden the approach have been limited by the physiology of the T cells themselves and by a range of immune evasion mechanisms developed by tumor cells. In this review we show how genetic modification of T cells is being used preclinically and in patients to overcome these limitations, by incorporation of novel receptors, resistance mechanisms, and control genes. We also discuss how the increasing safety and effectiveness of gene transfer technologies will lead to an increase in the use of gene-modified T cells for the treatment of a wider range of disorders.
That gene transfer could be used to improve the effectiveness of T lymphocytes was apparent from the beginning of clinical studies in the field. T cells were the very first targets for genetic modification in human gene transfer experiments. Rosenberg’s group marked tumor-infiltrating lymphocytes ex vivo with a Moloney retroviral vector encoding neomycin phosphotransferase before reinfusing them and attempting to demonstrate selective accumulation at tumor sites. Shortly thereafter, Blaese and Anderson led a group that infused corrected T cells into two children with severe combined immunodeficiency due to ADA deficiency. While neither study was completely successful in terms of outcome, both showed the feasibility of ex vivo gene transfer into human cells and set the stage for many of the studies that followed. More recently, a second wave of interest in adoptive T cell therapies has developed, based on their success in the prevention and treatment of viral infections such as EBV and cytomegalovirus (CMV) and on their apparent ability to eradicate hematologic and perhaps solid malignancies1,2,3,4,5,6. There has been a corresponding increase in studies directed toward enhancing the antineoplastic and antiviral properties of the T cells. In this article we will review how gene transfer may be used to produce the desired improvements focusing on vectors and genes that have had clinical application.
Currently available viral and nonviral vector systems lack a pattern of biodistribution that would favor T cell transduction in vivo—as occurs, for example, with adenovectors and the liver or liposomal vectors and the lung. This lack of favorable biodistribution cannot yet be compensated for by the introduction of specific T-cell-targeting ligands into vectors. Hence, all T cell gene transfer studies conducted to date have used ex vivo transduction followed by adoptive transfer of gene-modified cells. This approach is inherently less attractive for commercial development than directin vivo gene transfer and has probably restricted interest in developing clinical applications using these cells. On the other hand, ex vivo transduction may be more readily controlled, characterized, and standardized than in vivo efforts and may ultimately produce a better defined final product (the transduced cell).
The gene products of suicide and coexpressed resistance genes are highly immunogenic and may induce immune-mediated rejection of the transduced cells. In one study, the persistence of adoptively transferred autologous CD8+ HIV-specific CTL clones modified to express the hygromycin phosphotransferase (Hy) gene and the herpesvirus thymidine kinase gene as a fusion gene was limited by the induction of a potent CD8+ class I MHC-restricted CTL response specific for epitopes derived from the Hy-tk protein126. Less immunogenic suicide and selection marker genes, preferably of human origin, may reduce the immunological inactivation of genetically modified donor lymphocytes. Human-derived prodrug-activating systems include the human folylpolyglutamate synthetase/methotrexate127, the deoxycytidine/cytosine arabinoside128, or the carboxylesterase/irinotecan129 systems. These systems do not activate nontoxic prodrugs but are based on enhancement of already potent chemotherapeutic agents. The administration of methotrexate to treat severe GVHD may not only kill transduced donor lymphocytes but may also have additional inhibitory activity on nontransduced but activated T cells.
Finally, endogenous proapoptotic molecules have been proposed as nonimmunogenic suicide genes. A chimeric protein that contains the FK506-binding protein FKBP12 linked to the intracellular domain of human Fas130 was recently introduced. Addition of the dimerizing prodrug induces Fas crosslinking with subsequent triggering of an apoptotic death signal.
Genetic engineering of T lymphocytes should help deliver on the promise of immunotherapies for cancer, infection, and autoimmune disease. Improvements in transduction, selection, and expansion techniques and the development of new viral vectors incapable of insertional mutagenesis will reduce the risks and further enhance the integration of T cell and gene therapies. Nonetheless, successful application of the proposed modifications to the clinical setting still requires many iterative studies to allow investigators to optimize the individual components of the approach.
Genetically modified T cells in cancer therapy: opportunities and challenges
Tumours use many strategies to evade the host immune response, including downregulation or weak immunogenicity of target antigens and creation of an immune-suppressive tumour environment. T cells play a key role in cell-mediated immunity and, recently, strategies to genetically modify T cells either through altering the specificity of the T cell receptor (TCR) or through introducing antibody-like recognition in chimeric antigen receptors (CARs) have made substantial advances. The potential of these approaches has been demonstrated in particular by the successful use of genetically modified T cells to treat B cell haematological malignancies in clinical trials. This clinical success is reflected in the growing number of strategic partnerships in this area that have attracted a high level of investment and involve large pharmaceutical organisations. Although our understanding of the factors that influence the safety and efficacy of these therapies has increased, challenges for bringing genetically modified T-cell immunotherapy to many patients with different tumour types remain. These challenges range from the selection of antigen targets and dealing with regulatory and safety issues to successfully navigating the routes to commercial development. However, the encouraging clinical data, the progress in the scientific understanding of tumour immunology and the improvements in the manufacture of cell products are all advancing the clinical translation of these important cellular immunotherapies.
The immune system is split into two arms: the innate and adaptive immune systems (Fig. 1). Through immune surveillance, any molecules that are identified as non-self are eliminated. Targets include not only virally infected cells but also transformed (tumour) cells, which can acquire antigenicity (see Box 1for a glossary of terms) and hence immunogenicity through the expression of neo-antigens that can be recognised as non-self. However, cancer cells have developed strategies to escape and suppress the immune system (Pinzon-Charry et al., 2005; Blankenstein et al., 2012), which results in a failure to initiate and maintain adequate antitumour immunity, and consequently facilitates tumour survival and progression. Strategies include tumour antigens being only weakly immunogenic. Alternatively, the tumour might downregulate or modulate the expression of antigens, thus evading immune-cell detection. In addition, tumours can suppress the immune response through the synthesis of various immune suppressants that have roles in maintaining self-tolerance, or that inhibit effector immune cell function. Tumour immune suppression affects all branches of the immune system and can result in tumour escape from the immune system.
Fig. 1. Cells of the innate and adaptive immune systems. The innate immune system provides an immediate response to foreign targets, with responses typically within minutes to hours. It consists of a number of soluble factors and proteins as well as a diverse set of cells, including granulocytes, macrophages, dendritic cells and natural killer cells. The second branch of the immune system is the adaptive or acquired immune system, which provides specific, long-lasting immune responses. The adaptive and innate immune systems are linked; for example dendritic cells are important adaptive immune system cell activators. The adaptive immune system consists of antibodies, B cells, and CD4+ and CD8+ T cells, and these enable a highly specific response against a particular target. Natural killer T cells and γδ T cells are cytotoxic lymphocytes that overlap both innate and adaptive immunity. Cells from both arms of the immune system are in development as potential cellular immunotherapies.
The concept of transferring T cells to patients (adoptive T-cell transfer) to treat disease has been established over many years through the ex vivo manipulation, expansion and reinfusion of T cells that target specific viruses, for example to treat viral infections, such as cytomegalovirus or Epstein Barr virus infections following haematopoietic stem cell transplantation (Walter et al., 1995; Heslop et al., 2010; Rooney and Leen, 2012). As described above, rare populations of tumour-antigen-specific T cells do exist and specifically can be isolated at the site of the tumour, and these are known as tumour infiltrating lymphocytes (TILs) (Kawakami et al., 1994; Robbins et al., 2013). TILs can be isolated from excised tumour tissue, cultivated, activated and expanded ex vivo, and, on reinfusion, have shown promising efficacy in the clinic, particularly in the treatment of melanoma (Rosenberg et al., 1988,Besser et al., 2013; Kvistborg et al., 2012; Dudley et al., 2013), supporting the therapeutic potential of tumour-specific T cells.
An alternative option to these approaches that is now starting to generate compelling clinical data is based on the premise that the antigen specificity of T cells can be manipulated by genetic modification and redirected to successfully target antigens that are expressed by tumours. In particular, T cells can be engineered to express modified TCRs (so-called TCR therapies) or protein-fusion-derived chimeric antigen receptors (CARs) that have enhanced antigen specificity (Fig. 3). These approaches could overcome the fundamental limitations associated with central and peripheral tolerance, and generate T cells that will be more efficient at targeting tumours without the requirement for de novo T-cell activation in the patient.
Fig. 3. Genetically modified TCRs for cancer immunotherapy. (A) T-cell response can be manipulated and redirected against cancer, with improved specificity and affinity for tumour antigens, via genetic engineering of the endogenous TCR. (B) Genetically modified TCR: gene sequences are transferred to the T cell to encode new TCR α and β chains with different peptide specificity. In addition, there can also be transmembrane changes (red bars). To minimise interchain mispairing with the endogenous TCR, modifications such as the addition of a disulphide bridge (ss) are made. (C) Alternatively, a fusion receptor can be generated, a chimeric antigen receptor (CAR). Typically, these consist of three parts: a recognition sequence [represented here by an antibody-derived single-chain variable fragment (scFv)], a transmembrane element and an intracellular bespoke signalling domain (CD3ζ), which also contains co-stimulatory molecules, such as CD28 and tumour necrosis factor receptors (TNFr) such as OX-40.
The feasibility of T-cell adoptive transfer was first reported nearly 20 years ago (Walter et al., 1995) and the field of T-cell therapies is now poised for significant clinical advances. Recent clinical trial successes have been achieved through multiple small advances, improved understanding of immunology and emerging technologies. As the key challenges of T-cell avidity, persistence and ability to exert the desired anti-tumour effects as well as the identification of new target antigens are addressed, a broader clinical application of these therapies could be achieved. As the clinical data emerges, the challenge of making these therapies available to patients shifts to implementing robust, scalable and cost-effective manufacture and to the further evolution of the regulatory requirements to ensure an appropriate but proportionate system that is adapted to the characteristics of these innovative new medicines.
2012pharmaceutical
Amandeep Kaur
Aashir Awan, Phd
Abhisar Anand
Adina Hazan
Alan F. Kaul, PharmD., MS, MBA, FCCP
alexcrystal6
anamikasarkar
apreconasia
aviralvatsa
David Orchard-Webb, PhD
danutdaagmailcom
Demet Sag, Ph.D., CRA, GCP
Dror Nir
dsmolyar
Ethan Coomber
evelinacohn
Gail S Thornton
Irina Robu
jayzmit48
jdpmd
jshertok
kellyperlman
Ed Kislauskis
larryhbern
Madison Davis
marzankhan
megbaker58
ofermar2020
Dr. Pati
pkandala
ritusaxena
Rick Mandahl
sjwilliamspa
Srinivas Sriram
stuartlpbi
Dr. Sudipta Saha
tildabarliya
vaishnavee24
zraviv06
zs22
