Metabolomic analysis of two leukemia cell lines. II.
Larry H. Bernstein, MD, FCAP, Reviewer and Curator
Leaders in Pharmaceutical Intelligence
In Part I of metabolomics of two leukemia cell lines, we have established a major premise for the study, an insight into the use of an experimental model, and some insight into questions raised.
I here return to examine these before pursuing more detail in the study.
Q1. What strong metabolic pathways come into focus in this study?
Answer – The aerobic and anaerobic glycolytic pathways, with a difference measured in the extent of participation of mitochondrial oxidative phosphorylation.
Q2. Would we expect to also gain insight into the effect, on balance, played by a suppressed ubiquitin pathway?
Answer – lets look into this in Part II.
Q3. Would the synthesis of phospholipid and the maintenance of membrane structures requires availability of NADPH, which would be a reversal of the TCA cycle at the cost of delta G in catabolic energy, be consistent with increased dependence of anaerobic glycolysis with unchecked replication?
Answer: Part II might show this, as the direction and the difference between the cell lines is consistent with a Warburg (Pasteur) effect.
Recall the observation that the model is based on experimental results from lymphocytic leukemia cell lines in cell culture. The internal metabolic state is inferred from measurement of external metabolites.
The classification of the lymphocytic leukemias in humans is based on T-cell and B-cell lineages, but actually uses cell differentiation (CD) markers on the cytoskeleton for recognition. It is only a conjecture that if the cells line were highly anaplastic, they might not be sustainable in cell culture in perpetuity.
The analogue of these cells to what I would expect to see in humans is the SLL having the characteristic marking: CD5, see http://www.pathologyoutlines.com/topic/lymphomaSLL.html
Micro description
=======================================================
● Effacement of nodal architecture by pale staining pseudofollicles or proliferation centers with ill-defined borders, containing small round mature lymphocytes, prolymphocytes (larger than small lymphocytes, abundant basophilic cytoplasm, prominent nucleoli), paraimmunoblasts (larger cells with distinct nucleoli) and many smudge cells
● Pseudofollicular centers are highlighted by decreasing light through the condenser at low power; cells have pale cytoplasm but resemble soccer balls or smudge cells on peripheral smear (cytoplasm is bubbly in mantle cell lymphoma); may have plasmacytoid features
● May have marginal zone, perifollicular or interfollicular patterns, but these cases also have proliferation centers (Mod Pathol 2000;13:1161)
● Interfollicular pattern: large, reactive germinal centers; resembles follicular lymphoma but germinal centers are bcl2 negative and tumor cells resemble SLL by morphology and immunostains
(Am J Clin Path 2000;114:41)
● Paraimmunoblastic variant: diffuse proliferation of paraimmunoblasts (normally just in pseudoproliferation centers); rare, <30 reported cases; usually multiple lymphadenopathies and rapid disease progression; case report in 69 year old man (Hum Pathol 2002;33:1145); consider as mantile cell lymphoma if t(11;14)(q13;q32) is present; may also represent CD5+ diffuse large B cell lymphoma
● Bone marrow: small focal aggregates of variable size with irregular, poorly circumscribed outlines; lymphocytes are well differentiated, small, round with minimal atypia; may have foci of transformation; rarely has granulomas (J Clin Pathol 2005;58:815)
● Marrow infiltrative patterns are also described as diffuse (unmutated IgH genes, ZAP-70+, more aggressive), nodular (associated with IgH hypermutation, ZAP-70 negative) or mixed (variable mutation of IgH, variable ZAP-70, Hum Pathol 2006;37:1153)
Positive stains
=======================================================
● CD5, CD19, CD20 (dim), CD23, surface Ig light chain, surface IgM (dim)
● Also CD43, CD79a, CD79b (dim in 20%, Arch Pathol Lab Med 2003;127:561), bcl2
● Variable CD11c, FMC7 (42%)
Negative stains
=======================================================
● CD10, cyclin D1
Molecular
=======================================================
● Trisomy 12 (30%, associated with atypical CLL and CD79b), deletion 13q14 (25-50%),
deletion of 11q23 (worse prognosis, 10-20%)
Results
We set up a pipeline that could be used to
- infer intracellular metabolic states from semi-quantitative data
- regarding metabolites exchanged between cells and their environment.
Our pipeline combined the following four steps:
- data acquisition,
- data analysis,
- metabolic modeling and
- experimental validation of
- the model predictions (Fig. 1A).
We demonstrated the pipeline and the predictive potential
- to predict metabolic alternations in diseases such as cancer
- based on two lymphoblastic leukemia cell lines.
The resulting Molt-4 and CCRF-CEM condition-specific cell line models were able
- to explain metabolite uptake and secretion
- by predicting the distinct utilization of central metabolic pathways by the two cell lines.
Whereas the CCRF-CEM model
- resembled more a glycolytic, commonly referred to as ‘Warburg’ phenotype,
- our predictions suggested a more respiratory phenotype for the Molt-4 model.
We found these predictions to be in agreement with measured gene expression differences
- at key regulatory steps in the central metabolic pathways, and
- they were also consistent with data regarding the energy and redox states of the cells.
After a brief discussion of the data generation and analysis steps, the results
- derived from model generation and analysis will be described in detail.
2.1 Pipeline for generation of condition-specific metabolic cell line models
2.1.1 Generation of experimental data
We monitored the growth and viability of lymphoblastic leukemia cell lines in
serum- free medium (File S2, Fig. S1). Multiple omics data sets were derived from these cells.
Extracellular metabolomics (exo-metabolomic) data,
- comprising measurements of the metabolites in the spent medium of the cell cultures
(Paglia et al. 2012a), - were collected along with transcriptomic data, and
- these data sets were used to construct the models.
2.1.4 Condition-specific models for CCRF-CEM and Molt-4 cells
To determine whether we had obtained two distinct models,
- we evaluated the reactions, metabolites, and genes of the two models.
Both the Molt-4 and CCRF-CEM models contained approximately
- half of the reactions and metabolites present in the global model (Fig. 1C).
They were very similar to each other in terms of their
- reactions,
- metabolites, and
- genes (File S1, Table S5A–C).
The Molt– 4 model contained
- seven reactions that were not present in the CCRF-CEM model
(Co-A biosynthesis pathway and exchange reactions).
In contrast, the CCRF-CEM contained
31 unique reactions
- arginine and proline metabolism,
- vitamin B6 metabolism,
- fatty acid activation,
- transport, and exchange reaction.
- There were 2 and 15 unique metabolites in the Molt-4 and CCRF-CEM models, respectively
(File S1, Table S5B).
Approximately three quarters of the global model genesremained in the condition-specific cell line models (Fig. 1C).
The Molt-4 model contained
- 15 unique genes, and
the CCRF-CEM model had
- 4 unique genes (File S1, Table S5C).
Both models lacked NADH dehydrogenase
(complex I of the electron transport chain—ETC),
- determined by the absence of expression of a mandatory subunit
(NDUFB3, Entrez gene ID 4709).
The ETC was fueled by FADH2 originating from
- succinate dehydrogenase and
- from fatty acid oxidation, which
- through flavoprotein electron transfer
- could contribute to the same ubiquinone pool as
- complex I and complex II (succinate dehydrogenase).
Despite their different in vitro growth rates
(which differed by 11 %, see File S2, Fig. S1) and
- differences in exo-metabolomic data (Fig. 1B) and
- transcriptomic data,
- the internal networks were largely conserved
- in the two condition-specific cell line models.
2.1.5 Condition-specific cell line models predict distinct metabolic strategies
Despite the overall similarity of the metabolic models,
- differences in their cellular uptake and secretion patterns suggested
- distinct metabolic states in the two cell lines
(Fig. 1B and see “Materials and methods” section for more detail).
To interrogate the metabolic differences, we sampled the solution space
- of each model using an Artificial Centering Hit-and-Run (ACHR) sampler (Thiele et al. 2005).
For this analysis, additional constraints were applied, emphasizing
- the quantitative differences in commonly uptaken and secreted metabolites.
The maximum possible uptake and maximum possible secretion flux rates were
- reduced according to the measured relative differences between the cell lines
(Fig. 1D, see “Materials and methods” section).
We plotted the number of sample points containing a particular flux rate for each reaction. The resulting
- binned histograms can be understood as representing the probability that
- a particular reaction can have a certain flux value.
A comparison of the sample points obtained for the Molt-4 and CCRF-CEM models revealed
- a considerable shift in the distributions, suggesting
- a higher utilization of glycolysis by the CCRF-CEM model (File S2, Fig. S2).
This result was further supported by differences
- in medians calculated from sampling points (File S1, Table S6).
The shift persisted throughout all reactions of the pathway and
- was induced by the higher glucose uptake (35 %) from
- the extracellular medium in CCRF-CEM cells.
The sampling median for glucose uptake was 34 % higher
- in the CCRF-CEM model than in Molt-4 model (File S2, Fig. S2).
The usage of the TCA cycle was also distinct in the two condition-specific cell-line models (Fig. 2).
The Molt-4 model utilized an associated reaction to generate FADH2, whereas
- in the CCRF-CEM model, the histogram was shifted in the opposite direction,
- toward the generation of succinate.
Additionally, there was a higher efflux of citrate toward
- amino acid and lipid metabolism in the CCRF-CEM model (Fig. 2).
There was higher flux through anaplerotic and cataplerotic reactions
- in the CCRF-CEM model than in the Molt-4 model (Fig. 2);
- these reactions include the efflux of citrate through
- ATP-citrate lyase,
- uptake of glutamine,
- generation of glutamate from glutamine,
- transamination of pyruvate and
- glutamate to alanine and to 2-oxoglutarate,
- secretion of nitrogen, and
- secretion of alanine.
The Molt-4 model showed higher utilization of oxidative phosphorylation (Fig. 3),
- supported by elevated median flux through ATP synthase (36 %) and other enzymes,
- which contributed to higher oxidative metabolism.
The sampling analysis therefore revealed different usage of
- central metabolic pathways by the condition-specific models.
Fig. 2
Differences in the use of the TCA cycle by the CCRF-CEM model (red) and the Molt-4 model (blue).
The table provides the median values of the sampling results. Negative values in histograms and Table
- describe reversible reactions with flux in the reverse direction.
There are multiple reversible reactions for the transformation of
- isocitrate and α-ketoglutarate,
- malate and fumarate, and
- succinyl-CoA and succinate.
These reactions are unbounded, and therefore histograms are not shown.
The details of participating cofactors have been removed.
Atp ATP, cit citrate, adp ADP, pi phosphate, oaa oxaloacetate, accoa acetyl-CoA, coa coenzyme-A,
icit isocitrate, αkg α-ketoglutarate, succ–coa succinyl-CoA, succ succinate, fumfumarate, mal malate,
oxa oxaloacetate, pyr pyruvate, lac lactate, ala alanine, gln glutamine, ETC electron transport chain.
Electronic supplementary material The online version of this article
http://dx.doi.org:/10.1007/s11306-014-0721-3
contains supplementary material, which is available to authorized users.
- K. Aurich _ G. Paglia _ O ´ . Rolfsson _ S. Hrafnsdo´ ttir _
- Magnu´sdo´ ttir _ B. Ø. Palsson _ R. M. T. Fleming _ I. Thiele. Center for Systems Biology,
University of Iceland, Reykjavik, Iceland - K. Aurich _ R. M. T. Fleming _ I. Thiele (&). Luxembourg Centre for Systems Biomedicine,
University of Luxembourg, Campus Belval, Esch-Sur-Alzette, Luxembourg
e-mail: ines.thiele@uni.lu - M. Stefaniak. School of Health Science, Faculty of Food Science and Nutrition,
University of Iceland, Reykjavik, Iceland - Ø. Palsson. Department of Bioengineering, University of California San Diego, La Jolla, CA, USA
http://link.springer.com/static-content/images/404/art%253A10.1007%252
Fs11306-014-0721-3/MediaObjects/11306_2014_721_Fig3_HTML.gif
Fig. 3
Sampling reveals different utilization of oxidative phosphorylation by the
- generated models.
Different distributions are observed for the CCRF-CEM model (red) and the Molt-4 model (blue).
- Molt-4 has higher median flux through ETC reactions II–IV.
The table provides the median values of the sampling results. Negative values in the histograms and in the table describe
- reversible reactions with flux in the reverse direction.
Both models lack Complex I of the ETC because of constraints
- arising from the mapping of transcriptomic data.
Electron transfer flavoprotein and
- electron transfer flavoprotein–ubiquinone oxidoreductase
- both also carry higher flux in the Molt-4 model
2.1.6 Experimental validation of energy and redox status of CCRF-CEM and Molt-4 cells
Cancer cells have to balance their needs
- for energy and biosynthetic precursors, and they have
- to maintain redox homeostasis to proliferate (Cairns et al. 2011).
We conducted enzymatic assays of cell lysates to measure levels and/or ratios of
- ATP,
- NADPH + NADP,
- NADH + NAD, and
- glutathione.
These measurements were used to provide support for
- the in silico predicted metabolic differences (Fig. 4).
Additionally, an Oxygen Radical Absorbance Capacity (ORAC) assay was used
- to evaluate the cellular antioxidant status (Fig. 4B).
Total concentrations of NADH + NAD, GSH + GSSG, NADPH + NADP and ATP, were higher in Molt-4 cells (Fig. 4A).
The higher ATP concentration in Molt-4 cells could either result from
- high production rates, or intracellular accumulation connected to high or
- low reactions fluxes (Fig. 4A).
Our simplified view that oxidative Molt-4 produces less ATP and was contradicted by
- the higher ATP concentrations measured (Fig. 4L).
Yet we want to emphasize that concentrations
- cannot be compared to flux values,
- since we are modeling at steady-state.
NADH/NAD+ ratios for both cell lines were shifted toward NADH (Fig. 4D, E), but
- the shift toward NADH was more pronounced in CCRF-CEM (Fig. 4E),
- which matched our expectation based on the higher utilization of
- glycolysis and 2-oxoglutarate dehydrogenase in the CCRF-CEM model (Fig. 4L).
Fig. 4 (not shown)
A–K Experimentally determined ATP, NADH + NAD, NADPH + NADP, and GSH + GSSG concentrations, and ROS detoxification in the CCRF-CEM and Molt-4 cells.
L Expectations for cellular energy and redox states. Expectations are based on predicted metabolic differences of the Molt-4 and CCRF-CEM models
2.1.7 Comparison of network utilization and alteration in gene expression
With the assumption that
- differential expression of particular genes would cause reaction flux changes,
we determined how the differences in gene expression (between CCRF-CEM and Molt-4)
- compared to the flux differences observed in the models.
Specifically, we checked whether the reactions associated with genes upregulated
(significantly more expressed in CCRF-CEM cells compared to Molt-4 cells)
- were indeed more utilized by the CCRF-CEM model,
and we checked whether downregulated genes
- were associated with reactions more utilized by the Molt-4 model.
The set of downregulated genes was associated with 15 reactions, and
- the set of 49 upregulated genes was associated with 113 reactions in the models.
Reactions were defined as differently utilized
- if the difference in flux exceeded 10 % (considering only non-loop reactions).
Of the reactions associated with upregulated genes,
- 72.57 % were more utilized by the CCRF-CEM model, and
- 2.65 % were more utilized by the Molt-4 model (File S1, Table S7).
In contrast, all 15 reactions associated with the 12 downregulated genes
- were more utilized in the CCRF-CEM model (File S1, Table S8).
After this initial analysis, we approached the question from a different angle, asking
- whether the majority of the reactions associated with each individual gene
- upregulated in CCRF-CEM were more utilized by the CCRF-CEM model.
- this was the case for 77.55 % of the upregulated genes.
The majority of reactions associated with two (16.67 %) downregulated genes
- were more utilized by the Molt-4 model.
Taken together, our comparisons of the
- direction of gene expression with the fluxes of the two cancer cell-line models
- confirmed that reactions associated with upregulated genes in the CCRF-CEM
cells were generally more utilized by the CCRF-CEM model.
2.1.8 Accumulation of DEGs and AS genes at key metabolic steps
After we confirmed that most reactions associated with upregulated genes
- were more utilized by the CCRF-CEM model,
we checked the locations of DEGs within the network. In this analysis, we paid special attention to
- the central metabolic pathways that we had found
- to be distinctively utilized by the two models.
Several DEGs and AS events were associated with
- glycolysis,
- the ETC,
- pyruvate metabolism, and
- the PPP (Table 1).
Table 1
DEGs and AS events of central metabolic and cancer-related pathways
Full lists of DEGs and AS are provided in the supplementary material.
Upregulated significantly more expressed in CCRF-CEM compared to Molt-4 cells
PPP pentose phosphate pathway, OxPhos oxidative phosphorylation, Glycolysis/gluconglycolysis/gluconeogenesis, Pyruvate met. pyruvate metabolism
Moreover, in glycolysis, the DEGs and/or AS genes
- were associated with all three rate-limiting steps, i.e., the steps mediated by
- hexokinase,
- pyruvate kinase, and
- phosphofructokinase.
Of these key enzymes,
- hexokinase 1 (Entrez Gene ID: 3098) was alternatively spliced,
- pyruvate kinase (PKM, Entrez gene ID: 5315) was significantly more
expressed in the CCRF-CEM cells (Table 1),
in agreement with the higher in silico predicted flux.
However, in contrast to the observed
- higher utilization of glycolysis in the CCRF-CEM model,
- the gene associated with the rate-limiting glycolysis step, phosphofructokinase (Entrez Gene ID: 5213),
- was significantly upregulated in Molt-4 cells relative to CCRF-CEM cells.
This higher expression was detected for only a single isozyme, however. Two of
the three genes associated with phosphofructokinase were also subject to
alternative splicing (Table 1). In addition to the key enzymes, fructose
bisphosphate aldolase (Entrez Gene ID: 230) was also significantly
- upregulated in Molt-4 cells relative to CCRF-CEM cells,
- in contrast to the predicted higher utilization of glycolysis in the CCRF-CEM model.
Additionally, glucose-6P-dehydrogenase (G6PD), which catalyzes
- the first reaction and committed step of the PPP,
- was an AS gene (Table 1).
A second AS gene associated with
- the PPP reaction of the deoxyribokinase
- was RBKS (Entrez Gene ID: 64080).
This gene is also associated with ribokinase, but ribokinase was removed
- because of the lack of ribose uptake or secretion.
Single AS genes were associated with different complexes of the ETC (Table 1).
Literature query revealed that at least 13 genes associated with alternative
- splicing events were mentioned previously in connection with both alternative
splicing and cancer (File S1, Table S14), and - 37 genes were associated with cancer, e.g., upregulated, downregulated at the
level of mRNA or protein, or otherwise - connected to cancer metabolism and signaling.
One general observation was that there was a surprising
- accumulation of metabolite transporters among the AS.
Overall, the high incidence of
- differential gene expression events at metabolic control points
- increases the plausibility of the in silico predictions.
2.1.9 Single gene deletion
Analyses of essential genes in metabolic models have been used
- to predict candidate drug targets for cancer cells (Folger et al. 2011).
Here, we conducted an in silico gene deletion study for all model genes to identify
- a unique set of knock-out (KO) genes
- for each condition-specific cell line model.
The analysis yielded 63 shared lethal KO genes and
- distinct sets of KO genes for the CCRF-CEM model (11 genes) and the Molt-4 model (3 genes).
For three of the unique CCRF-CEM KO genes,
- the genes were only present in the CCRF-CEM model (File S1, Table S9).
The essential genes for both models were then
- related to the cell-line-specific differences in metabolite uptake and secretion (Fig. 1B).
The CCRF-CEM model
- needed to generate putrescine from ornithine
(ORNDC, Entrez Gene ID: 4953) - to subsequently produce 5-methylthioadenosine for secretion (Fig. 1B).
- S-adenosylmethioninamine produced by adenosylmethionine decarboxylase
(arginine and proline metabolism, associated with Entrez Gene ID: 262)
- is a substrate required for generation of 5-methylthioadenosine.
Another example of a KO gene connected to an enforced exchange reaction was
- glutamic-oxaloacetic transaminase 1 (GOT1, Entrez Gene ID: 2805).
Without GOT1, the CCRF-CEM model was forced to secrete
- 4-hydroxyphenylpyruvate (Fig. 1B),
- the second product of tyrosine transaminase,
- which is produced only by that enzyme.
One KO gene in the Molt-4 model (Entrez Gene ID: 26227) was associated with
- phosphoglycerate dehydrogenase (PGDH),
- which catalyzes the conversion of 3-phospho-d-glycerate to 3-phosphohydroxypyruvate
- while generating NADH from NAD+.
This KO gene is particularly interesting, given
- the involvement of this reaction in a novel pathway for ATP generation in rapidly proliferating cells
(Locasale et al. 2011; Vander Heiden 2011; Vazquez et al. 2011).
Reactions associated with unique KO genes were in many cases utilized more by the model, in which
- the gene KO was lethal,
- underlining the potential importance of these reactions for the models.
Thus, single gene deletion provided unique sets of lethal genes that could be
- specifically targeted to kill these cells.
3 Discussion
In the current study, we explored the possibility of
- semi-quantitatively integrating metabolomic data with
- the human genome-scale reconstruction to facilitate analysis.
By constructing condition-specific cell line models
- to provide a structured framework,
- we derived insights that could not have been obtained from data analysis alone.
We derived condition-specific cell line models
- for CCRF-CEM and
- Molt-4 cells
that were able to explain the observed exo-metabolomic differences (Fig. 1B).
Despite the overall similarities between the models, the analysis revealed
- distinct usage of central metabolic pathways (Figs. 2, 3, 4),
- which we validated based on experimental data and
- differential gene expression.
The additional data sufficiently supported
- metabolic differences in the cell lines,
- providing confidence in the generated models and the model-based predictions.
We used the validated models
- to predict unique sets of lethal genes
- to identify weak links in each model.
These weak links may represent potential drug targets.
Integrating omics data with the human genome-scale reconstruction
- provides a structured framework (i.e., pathways)
- that is based on careful consideration of the available biochemical literature
(Thiele and Palsson2010).
This network context can simplify omics data analysis, and
- it allows even non-biochemical experts
- to gain fast and comprehensive insights
- into the metabolic aspects of omics data sets.
Compared to transcriptomic data,
- methods for the integration and analysis of metabolomic data
- in the context of metabolic models are less well established,
although it is an active field of research (Li et al. 2013; Paglia et al. 2012b).
In contrast to other studies, our approach emphasizes
- the representation of experimental conditions rather than
- the reconstruction of a generic, cell-line-specific network,
- which would require the combination of data sets from
- many experimental conditions and extensive manual curation.
Rather, our way of model construction allowed us to efficiently
- assess the metabolic characteristics of cells.
Despite the fact, that only a limited number of exchanged metabolites can be
- measured by available metabolomics platforms and
- at reasonable time-scale,
and that pathways of measured metabolites might still be unknown to date
(File S1, Tables S2–S3), our methods have the potential
- to reveal metabolic characteristics of cells
- which could be useful for biomedicine and personalized health.
The reasons why some cancers respond to certain treatments and not others
remain unclear, and choosing a treatment for a specific patient is often difficult
(Vander Heiden 2011). One potential application of our approach could be the
characterization of cancer phenotypes to explore how cancer cells or other cell
types
- with particular metabolic characteristics respond to drugs.
The generation of our condition-specific cell line models involved
- only limited manual curation,
- making this approach a fast way to place metabolomic data
- into a network context.
Model building mainly involves
- the rigid reduction of metabolite exchanges
- to match the observed metabolite exchange pattern
- with as few additional metabolite exchanges as possible.
It should be noted that this reduction determines,
- which pathways can be utilized by the model.
Our approach mostly conserved the internal network redundancy. However, a
- more significant reduction may be achieved using different data.
Generally, a trade-off exists between the reduction of the internal network and
- the increasing number of network gaps that need to be curated
- by using additional omics data, such as transcriptomics and proteomics.
One way to prevent the emergence of network gaps would be
- to use mapping algorithms that conserve network functionality,
such as GIMME (Becker and Palsson 2008).
However, several additional methods exist for the integration of
transcriptomic data (Blazier and Papin 2012), and
- which model-building method is best depends on the available data.
Interestingly, the lack of a significant contribution of our
- gene expression data to the reduction of network size
- suggests that the use of transcriptomic data is not necessary
- to identify distinct metabolic strategies;
- rather, the integration of exo-metabolomic data alone
may provide sufficient insight.
However, sampling of the cell line models constrained
- according to the exo-metabolomic profiles only, or
- increasing the cutoff for the generation of absent and present calls (p < 0.01),
- did not yield the same insights as presented herein (File S1, Table S18).
Only recently Gene Inactivation Moderated by Metabolism, Metabolomics and
Expression (GIM(3)E) became available, which
- enforces minimum turnover of detected metabolites
- based on intracellular metabolomics data as well as
- gene expression microarray data (Schmidt et al. 2013).
In contrast to this approach, we emphasized our analysis on the
- relative differences in the exo-metabolomic data of two cell lines.
GIM(3)E constitutes another integration method when the analysis should be
- emphasized on intracellular metabolomics data (Schmidt et al. 2013).
The metabolic differences predicted by the models are generally plausible.
Cancers are known to be heterogeneous (Cairns et al. 2011), and
- the contribution of oxidative phosphorylation to cellular ATP production
may vary (Zu and Guppy 2004).
Moreover, leukemia cell lines have been shown
- to depend on glucose, glutamine, and fatty acids to varying extents
- to support proliferation.
Such dependence may cause the cells to adapt their metabolism
- to the environmental conditions (Suganuma et al. 2010).
In addition to identifying supporting data in the literature, we performed
- several analyses to validate the models and model predictions.
Our expectations regarding the levels and ratios of metabolites
- relevant to energy and redox state were largely met (Fig. 4L).
The more pronounced shift of the NADH/NAD+ ratio
- toward NADH in the CCRF-CEM cells
- was in agreement with the predicted Warburg phenotype (Fig. 4),
- and the higher lactate secretion in the CCRF-CEM cells (File S2, Fig. S2)
- implies an increase in NADH relative to NAD+
(Chiarugi et al. 2012; Nikiforov et al. 2011), again - matching the known Warburg phenotype.
ROS production is enhanced in certain types of cancer (Droge 2002; Ha et al. 2000), and
- the generation of ROS is thought to contribute to
However, decreased mitochondrial glucose oxidation and
- a transition to aerobic glycolysis
- protect cells against ROS damage during biosynthesis and cell division
(Brand and Hermfisse1997).
The higher ROS detoxification capability in Molt-4 cells, in combination with
- higher spermidine dismutase utilization by the Molt-4 model (Fig. 4),
- provided a consistent picture of the predicted respiratory phenotype (Fig. 4L).
Control of NADPH maintains the redox potential through GSH and
- protects against oxidative stress, yet
- changes in the NADPH ratio in response to oxidative damage
- are not well understood (Ogasawara et al.2009).
Under stress conditions, as assumed for Molt-4 cells,
- the NADPH/NADP+ ratio is expected to decrease because of
- the continuous reduction of GSSG (Fig. 4L), and
- this was confirmed in the Molt-4 cells (Fig. 4).
The higher amounts of GSH found in Molt-4 cells in vitro may demonstrate
- an additional need for ROS scavengers because of
- a greater reliance on oxidative metabolism.
Cancer is related to metabolic reprogramming, which results from
- alterations of gene expression and
- the expression of specific isoforms or
- splice forms to support proliferation
(Cortes-Cros et al. 2013; Marin-Hernandez et al. 2009).
The gene expression differences detected between the two cell lines in this study
supported the existence of
- metabolic differences in these cell lines, particularly because
- key steps of the metabolic pathways central to cancer metabolism
- seemed to be differentially regulated (Table 1).
The detailed analysis of the respective
- differences on the pathway fluxes exceeds the scope of this study, which was to
- demonstrate the potential of the integration of exo-metabolomic data into the network context.
We found discrepancies between differential gene regulation and
- the flux differences between the two models as well as
- the utilization AS gene-associated reaction.
This is not surprising, since analysis of the detailed system is required
- to make any further assumptions on the impact that
- the differential regulation or splicing might have on the reaction flux,
- given that for many of the concerned enzymes isozymes exist, or
- only one of multiple subunits of a protein complex was concerned.
Additionally, reaction fluxes are regulated by numerous post-translational factors, e.g.,
- protein modification,
- inhibition through proteins or metabolites,
- alter reaction fluxes (Lenzen 2014),
which are out of the scope of constraint-based steady-state modeling.
Rather, the results of the presented approach
- demonstrate how the models can be used to generate
- informed hypothesis that can guide experimental work.
The combination of our tailored metabolic models and
- differential gene expression analysis seems well-suited
- to determine the potential drivers
- involved in metabolic differences between cells.
Such information could be valuable for drug discovery, especially when more
- peripheral metabolic pathways are considered.
Statistical comparisons of gene expression data with sampling-derived flux data
- could be useful in future studies (Mardinoglu et al. 2013).
A single-gene-deletion analysis revealed that PGDH was
- a lethal KO gene for the Molt-4 model only.
Differences in PGDH protein levels
- correspond to the amount of glycolytic carbon
- diverted into glycine biosynthesis.
Rapidly proliferating cells may use an
- alternative glycolytic pathway for ATP generation,
- which may provide an advantage in the case of
- extensive oxidative phosphorylation and proliferation
(Locasale et al.2011; Vander Heiden 2011; Vazquez et al. 2011).
For breast cancer cell lines, variable dependency on
- the expression of PGDH has already been demonstrated
(Locasale et al. 2011).
This example of a unique KO gene demonstrates how
- in silico gene deletion in metabolomics-driven models
- can identify the metabolic pathways used by cancer cells.
This approach can provide valuable information for drug discovery.
In conclusion, our contextualization method produced
- metabolic models that agreed in many ways with the validation data sets.
The analyses described in this study have great potential to reveal
- the mechanisms of metabolic reprogramming,
- not only in cancer cells but also in other cells affected by diseases, and
- for drug discovery in general.
…
4.3 Analysis of the extracellular metabolome
Mass spectrometry analysis of the exo-metabolome was performed by
Metabolon®, Inc. (Durham, NC, USA) using a standardized analytical platform.
In total, 75 extracellular metabolites were detected in the initial data set for at
least 1 of the 2 cell lines (Paglia et al. 2012a). Of these metabolites, 15 were not
part of our global model and were discarded. Apart from being absent in our
global model, an independent search in HMDB (Wishart et al. 2013) revealed no
pathway information was available for most of these metabolites (File S1, Tables S2–S3).
It should be noted that metabolites e.g.,
- N-acetylisoleucine,
- N-acetyl-methionine or pseudouridine,
constitute protein and RNA degradation products, which were out of the scope
of the metabolic network.
Thiamin (Vitamin B1) was part of the minimal medium of essential compounds
supplied to both models.Riboflavin (Vitamin B2) and Trehalose were excluded
since these compounds cannot be produced by human cells. Erythrose and
fructose were also excluded. In contrast 46 metabolites that were part of the
global model. The data set included two different time points, which allowed us
to treat the increase/decrease of a metabolite signal between time points as
- evidence for uptake or secretion when the change was greater than 5 %
from what was observed in the control (File S1, Tables S2–S3).
We found 12 metabolites that were taken up by both cell lines and
10 metabolites that were commonly secreted by both cell lines over
the course of the experiment.
Molt-4 cells took up three metabolites not taken up by CCRF-CEM cells, and
secreted one metabolite not secreted by CCRF-CEM cells. Two of the three
uniquely uptaken metabolites were essential amino acids:
- valine and
- methionine.
It is unlikely that these metabolites were not taken up by the CCRF-CEM cells,
and the CCRF-CEM model was allowed to take up this metabolite. Therefore,
no quantitative constraints were applied for the sampling analysis either.
CCRF-CEM cells had
- four unique uptaken
- and seven unique secreted metabolites
(exchange not detected in Molt-4 cells).
4.4 Network refinement based on exo-metabolic data
Despite its comprehensiveness, the human metabolic reconstruction is
- not complete with respect to extracellular metabolite transporters
(Sahoo et al. 2014; Thiele et al. 2013).
Accordingly, we identified metabolite transport systems
- from the literature for metabolites that were already part of the global model,
- but whose extracellular transport was not yet accounted for.
Diffusion reactions were included whenever a respective transporter could not be identified.
In total, 34 reactions [11 exchange reactions, 16 transport reactions and 7 demand reactions
(File S1, Table S11)] were added to Recon 2 (Thiele et al. 2013), and 2 additional reactions
were added to the global model (File S1, Table S10).
4.5 Expression profiling
Molt-4 and CCRF-CEM cells were grown in advanced RPMI 1640 and 2 mM
GlutaMax, and the cells were resuspended in medium containing DMSO
(0.67 %) at a concentration of 5 × 105 cells/mL. The cell suspension (2 mL)
was seeded in 12-well plates in triplicate. After 48 h of growth, the cells
were collected by centrifugation at 201×g for 5 min. Cell pellets were snap-frozen
in liquid N2 and kept frozen until RNA extraction and analysis by Aros
(Aarhus, Denmark).
4.6 Analysis of transcriptomic data
We used the Affymetrix GeneChip Human Exon 1.0 ST Array to measure whole
genome exon expression. We generated detection above background (DABG) calls
using ROOT (version 22) and the XPS package for R (version 11.1), with Robust
Multi-array Analysis summarization. Calls for data mapping were assigned based
on p < 0.05 as the cutoff probability to distinguish presence versus absence for
the 1,278 model genes (File S1, Table S12).
Differential gene expression and alternative splicing analyses were performed by
using AltAnalyse software (v2.02beta) with default options on the raw data files
(CEL files). The Homo sapiens Ensemble 65 database was used, probe set filtering
was kept as DABG p < 0.05, and non-log expression < 70 was used for
constitutive probe sets to determine gene expression levels. For the comparison,
CCRF-CEM was the experimental group and Molt-4 was the baseline group. The
set of DEGs between cell lines was identified based on a p < 0.05 FDR cutoff
(File S1, Table S13A–B). Alternative splicing analysis was performed on core probe sets
with a minimum alternative exon score of 2 and a maximum absolute gene
expression change of 3 because alternative splicing is a less critical factor among
highly DEGs (File S1, Table S14). Gene expression data, complete lists of DABG p-values,
DEGs and alternative splicing events have been deposited in the Gene
Expression Omnibus (GEO) database (Accession number: GSE53123).
4.7 Deriving cell-type-specific subnetworks
Transcriptomic data were mapped to the model in a manual fashion (COBRA
function: deleteModelGenes). Specifically, reactions dependent on gene products
that were called as “absent” were constrained to zero, such that fluxes through
these reactions were disabled. Submodels were extracted based on the set of
reactions carrying flux (network pruning) by running fastFVA
(Gudmundsson and Thiele 2010) after mapping the metabolomic and
transcriptomic data using the COBRA toolbox (Schellenberger et al. 2011).
…..
Electronic supplementary material
Below is the link to the electronic supplementary material.
File S1. Supplementary material 1 (XLSX 915 kb)
File S2. Supplementary material 2 (DOCX 448 kb)
References
Antonucci, R., Pilloni, M. D., Atzori, L., & Fanos, V. (2012). Pharmaceutical research and metabolomics in the newborn. Journal of Maternal-Fetal and Neonatal Medicine, 25, 22–26.PubMedCrossRef
Barrett, T., Troup, D. B., Wilhite, S. E., Ledoux, P., Evangelista, C., Kim, I. F., et al. (2011). NCBI GEO: archive for functional genomics data sets—10 years on. Nucleic Acids Research, 39, D1005–D1010.PubMedCentralPubMedCrossRef
Beck, M., Schmidt, A., Malmstroem, J., Claassen, M., Ori, A., Szymborska, A., et al. (2011). The quantitative proteome of a human cell line.Molecular Systems Biology, 7, 549.PubMedCentralPubMedCrossRef
Becker, S. A., & Palsson, B. O. (2008). Context-specific metabolic networks are consistent with experiments. PLoS Computational Biology, 4, e1000082.PubMedCentralPubMedCrossRef
Blazier, A. S., & Papin, J. A. (2012). Integration of expression data in genome-scale metabolic network reconstructions. Frontiers in Physiology, 3, 299.PubMedCentralPubMedCrossRef
Bordbar, A., Lewis, N. E., Schellenberger, J., Palsson, B. O., & Jamshidi, N. (2010). Insight into human alveolar macrophage and M. tuberculosisinteractions via metabolic reconstructions. Molecular Systems Biology, 6, 422.PubMedCentralPubMedCrossRef
Bordbar, A., & Palsson, B. O. (2012). Using the reconstructed genome-scale human metabolic network to study physiology and pathology. Journal of Internal Medicine, 271, 131–141.PubMedCentralPubMedCrossRef
Brand, K. A., & Hermfisse, U. (1997). Aerobic glycolysis by proliferating cells: a protective strategy against reactive oxygen species. FASEB Journal, 11, 388–395.PubMed
Cairns, R. A., Harris, I. S., & Mak, T. W. (2011). Regulation of cancer cell metabolism. Nature Reviews Cancer, 11, 85–95.PubMedCrossRef
Chance, B., Sies, H., & Boveris, A. (1979). Hydroperoxide metabolism in mammalian organs. Physiological Reviews, 59, 527–605.PubMed
Chapman, E. H., Kurec, A. S., & Davey, F. R. (1981). Cell volumes of normal and malignant mononuclear cells. Journal of Clinical Pathology, 34, 1083–1090.PubMedCentralPubMedCrossRef
Chiarugi, A., Dolle, C., Felici, R., & Ziegler, M. (2012). The NAD metabolome—a key determinant of cancer cell biology. Nature Reviews Cancer, 12, 741–752.PubMedCrossRef
Cortes-Cros, M., Hemmerlin, C., Ferretti, S., Zhang, J., Gounarides, J. S., Yin, H., et al. (2013). M2 isoform of pyruvate kinase is dispensable for tumor maintenance and growth. Proceedings of the National Academy of Sciences of the United States of America, 110, 489–494.PubMedCentralPubMedCrossRef
Dreher, D., & Junod, A. F. (1996). Role of oxygen free radicals in cancer development. European Journal of Cancer, 32a, 30–38.PubMedCrossRef
Droge, W. (2002). Free radicals in the physiological control of cell function. Physiological Reviews, 82, 47–95.PubMed
Duarte, N. C., Becker, S. A., Jamshidi, N., Thiele, I., Mo, M. L., Vo, T. D., et al. (2007). Global reconstruction of the human metabolic network based on genomic and bibliomic data. Proceedings of the National Academy of Sciences of the United States of America, 104, 1777–1782.PubMedCentralPubMedCrossRef
Durot, M., Bourguignon, P. Y., & Schachter, V. (2009). Genome-scale models of bacterial metabolism: Reconstruction and applications. FEMS Microbiology Reviews, 33, 164–190.PubMedCentralPubMedCrossRef
Fleming, R. M., Thiele, I., & Nasheuer, H. P. (2009). Quantitative assignment of reaction directionality in constraint-based models of metabolism: Application to Escherichia coli. Biophysical Chemistry, 145, 47–56.PubMedCentralPubMedCrossRef
Folger, O., Jerby, L., Frezza, C., Gottlieb, E., Ruppin, E., & Shlomi, T. (2011). Predicting selective drug targets in cancer through metabolic networks. Molecular Systems Biology, 7, 501.PubMedCentralPubMedCrossRef
Frezza, C., Zheng, L., Folger, O., Rajagopalan, K. N., MacKenzie, E. D., Jerby, L., et al. (2011). Haem oxygenase is synthetically lethal with the tumour suppressor fumarate hydratase. Nature, 477, 225–228.PubMedCrossRef
Ganske, F., & Dell, E. J. (2006). ORAC assay on the FLUOstar OPTIMA to determine antioxidant capacity. BMG LABTECH.
Gudmundsson, S., & Thiele, I. (2010). Computationally efficient flux variability analysis. BMC Bioinformatics, 11, 489.PubMedCentralPubMedCrossRef
Ha, H. C., Thiagalingam, A., Nelkin, B. D., & Casero, R. A, Jr. (2000). Reactive oxygen species are critical for the growth and differentiation of medullary thyroid carcinoma cells. Clinical Cancer Research, 6, 3783–3787.PubMed
Hyduke, D. R., Lewis, N. E., & Palsson, B. O. (2013). Analysis of omics data with genome-scale models of metabolism. Molecular BioSystems, 9, 167–174.PubMedCentralPubMedCrossRef
Jerby, L., & Ruppin, E. (2012). Predicting drug targets and biomarkers of cancer via genome-scale metabolic modeling. Clinical Cancer Research,18, 5572–5584.PubMedCrossRef
Jerby, L., Shlomi, T., & Ruppin, E. (2010). Computational reconstruction of tissue-specific metabolic models: Application to human liver metabolism.Molecular Systems Biology, 6, 401.PubMedCentralPubMedCrossRef
Jerby, L., Wolf, L., Denkert, C., Stein, G. Y., Hilvo, M., Oresic, M., et al. (2012). Metabolic associations of reduced proliferation and oxidative stress in advanced breast cancer. Cancer Research, 72, 5712–5720.PubMedCrossRef
Lenzen, S. (2014). A fresh view of glycolysis and glucokinase regulation: History and current status. Journal of Biological Chemistry, 289, 12189–12194.PubMedCrossRef
Lewis, N. E., Nagarajan, H., & Palsson, B. O. (2012). Constraining the metabolic genotype–phenotype relationship using a phylogeny of in silico methods. Nature Reviews Microbiology, 10, 291–305.PubMedCentralPubMed
Lewis, N. E., Schramm, G., Bordbar, A., Schellenberger, J., Andersen, M. P., Cheng, J. K., et al. (2010). Large-scale in silico modeling of metabolic interactions between cell types in the human brain. Nature Biotechnology, 28, 1279–1285.PubMedCentralPubMedCrossRef
Li, S., Park, Y., Duraisingham, S., Strobel, F. H., Khan, N., Soltow, Q. A., et al. (2013). Predicting network activity from high throughput metabolomics. PLoS Computational Biology, 9, e1003123.PubMedCentralPubMedCrossRef
Locasale, J. W., Grassian, A. R., Melman, T., Lyssiotis, C. A., Mattaini, K. R., Bass, A. J., et al. (2011). Phosphoglycerate dehydrogenase diverts glycolytic flux and contributes to oncogenesis. Nature Genetics, 43, 869–874.PubMedCentralPubMedCrossRef
Mardinoglu, A., Agren, R., Kampf, C., Asplund, A., Nookaew, I., Jacobson, P., et al. (2013). Integration of clinical data with a genome-scale metabolic model of the human adipocyte. Molecular Systems Biology, 9, 649.PubMedCentralPubMedCrossRef
Marin-Hernandez, A., Gallardo-Perez, J. C., Ralph, S. J., Rodriguez-Enriquez, S., & Moreno-Sanchez, R. (2009). HIF-1alpha modulates energy metabolism in cancer cells by inducing over-expression of specific glycolytic isoforms. Mini Reviews in Medicinal Chemistry, 9, 1084–1101.PubMedCrossRef
Mir, M., Wang, Z., Shen, Z., Bednarz, M., Bashir, R., Golding, I., et al. (2011). Optical measurement of cycle-dependent cell growth. Proceedings of the National Academy of Sciences of the United States of America, 108, 13124–13129.PubMedCentralPubMedCrossRef
Mo, M. L., Palsson, B. O., & Herrgard, M. J. (2009). Connecting extracellular metabolomic measurements to intracellular flux states in yeast. BMC Systems Biology, 3, 37.PubMedCentralPubMedCrossRef
Nikiforov, A., Dolle, C., Niere, M., & Ziegler, M. (2011). Pathways and subcellular compartmentation of NAD biosynthesis in human cells: From entry of extracellular precursors to mitochondrial NAD generation. The Journal of biological chemistry, 286, 21767–21778.PubMedCentralPubMedCrossRef
Ogasawara, Y., Funakoshi, M., & Ishii, K. (2009). Determination of reduced nicotinamide adenine dinucleotide phosphate concentration using high-performance liquid chromatography with fluorescence detection: Ratio of the reduced form as a biomarker of oxidative stress. Biological & Pharmaceutical Bulletin, 32, 1819–1823.CrossRef
Paglia, G., Hrafnsdottir, S., Magnusdottir, M., Fleming, R. M., Thorlacius, S., Palsson, B. O., et al. (2012a). Monitoring metabolites consumption and secretion in cultured cells using ultra-performance liquid chromatography quadrupole-time of flight mass spectrometry (UPLC-Q-ToF-MS).Analytical and Bioanalytical Chemistry, 402, 1183–1198.PubMedCrossRef
Paglia, G., Palsson, B. O., & Sigurjonsson, O. E. (2012b). Systems biology of stored blood cells: Can it help to extend the expiration date? Journal of Proteomics, 76, 163–167.PubMedCrossRef
Price, N. D., Schellenberger, J., & Palsson, B. O. (2004). Uniform sampling of steady-state flux spaces: Means to design experiments and to interpret enzymopathies. Biophysical Journal, 87, 2172–2186.PubMedCentralPubMedCrossRef
Reed, J. L., Famili, I., Thiele, I., & Palsson, B. O. (2006). Towards multidimensional genome annotation. Nature Reviews Genetics, 7, 130–141.PubMedCrossRef
Sahoo, S., Aurich, M. K., Jonsson, J. J., & Thiele, I. (2014). Membrane transporters in a human genome-scale metabolic knowledgebase and their implications for disease. Frontiers in Physiology, 5, 91.PubMedCentralPubMedCrossRef
Sahoo, S., & Thiele, I. (2013). Predicting the impact of diet and enzymopathies on human small intestinal epithelial cells. Human Molecular Genetics, 22, 2705–2722.PubMedCentralPubMedCrossRef
Schellenberger, J., & Palsson, B. O. (2009). Use of randomized sampling for analysis of metabolic networks. The Journal of biological chemistry,284, 5457–5461.PubMedCrossRef
Schellenberger, J., Que, R., Fleming, R. M., Thiele, I., Orth, J. D., Feist, A. M., et al. (2011). Quantitative prediction of cellular metabolism with constraint-based models: The COBRA Toolbox v2.0. Nature Protocols, 6, 1290–1307.PubMedCentralPubMedCrossRef
Schmidt, B. J., Ebrahim, A., Metz, T. O., Adkins, J. N., Palsson, B. O., & Hyduke, D. R. (2013). GIM3E: Condition-specific models of cellular metabolism developed from metabolomics and expression data. Bioinformatics (Oxford, England), 29, 2900–2908.CrossRef
Suganuma, K., Miwa, H., Imai, N., Shikami, M., Gotou, M., Goto, M., et al. (2010). Energy metabolism of leukemia cells: Glycolysis versus oxidative phosphorylation. Leukemia & Lymphoma, 51, 2112–2119.CrossRef
Thiele, I., & Palsson, B. O. (2010). A protocol for generating a high-quality genome-scale metabolic reconstruction. Nature Protocols, 5, 93–121.PubMedCentralPubMedCrossRef
Thiele, I., Price, N. D., Vo, T. D., & Palsson, B. O. (2005). Candidate metabolic network states in human mitochondria. Impact of diabetes, ischemia, and diet. The Journal of biological chemistry, 280, 11683–11695.PubMedCrossRef
Thiele, I., Swainston, N., Fleming, R. M., Hoppe, A., Sahoo, S., Aurich, M. K., et al. (2013). A community-driven global reconstruction of human metabolism. Nature Biotechnology, 31, 419–425.PubMedCrossRef
Uhlen, M., Oksvold, P., Fagerberg, L., Lundberg, E., Jonasson, K., Forsberg, M., et al. (2010). Towards a knowledge-based human protein Atlas.Nature Biotechnology, 28, 1248–1250.PubMedCrossRef
Vander Heiden, M. G. (2011). Targeting cancer metabolism: A therapeutic window opens. Nature Reviews Drug Discovery, 10, 671–684.PubMedCrossRef
Vazquez, A., Markert, E. K., & Oltvai, Z. N. (2011). Serine biosynthesis with one carbon catabolism and the glycine cleavage system represents a novel pathway for ATP generation. PLoS ONE, 6, e25881.PubMedCentralPubMedCrossRef
Wishart, D. S., Jewison, T., Guo, A. C., Wilson, M., Knox, C., Liu, Y., et al. (2013). HMDB 3.0—The human metabolome database in 2013. Nucleic Acids Research, 41, D801–D807.PubMedCentralPubMedCrossRef
Zu, X. L., & Guppy, M. (2004). Cancer metabolism: Facts, fantasy, and fiction. Biochemical and Biophysical Research Communications, 313, 459–465.PubMedCrossRef
Leave a Reply