Conceived and designed the experiments: JLYC MW JTC. Performed the experiments: JLYC TS JW JTC. Analyzed the data: JLYC JEL SM JN MD MW JTC. Contributed reagents/materials/analysis tools: JEL JTC. Wrote the paper: JLYC JEL JN MW JTC.
The authors have declared that no competing interests exist.
The tumor microenvironment has a significant impact on tumor development. Two important determinants in this environment are hypoxia and lactic acidosis. Although lactic acidosis has long been recognized as an important factor in cancer, relatively little is known about how cells respond to lactic acidosis and how that response relates to cancer phenotypes. We develop genome-scale gene expression studies to dissect transcriptional responses of primary human mammary epithelial cells to lactic acidosis and hypoxia
It is well recognized that tumor microenvironments play an important role in modulating tumor progression in human cancers. Although previous studies have highlighted the importance of hypoxia, there is limited knowledge on the effects of other components in tumor microenvironments. Therefore, we use gene expression to compare and analyze how cells respond to lactate, acidity, and hypoxia, as well as how these responses can be utilized to predict the clinical outcomes of patients with breast cancers. We uncover an unexpected association with better clinical outcome of the strong lactic acidosis and acidosis response in breast cancers as a result of their abilities to inhibit glycolysis and favor oxidative phosphorylation for energy generation. This effect is caused by not only the repression of the gene expression of glycolysis genes but also the inhibition of Akt activation of cells exposed to lactic acidosis and acidosis. In conclusion, we propose that lactic acidosis and acidosis to be considered as independent prognostic factors for human cancers.
The tumor microenvironment is characterized by oxygen depletion (hypoxia), high lactate and extracellular acidosis (lactic acidosis) as well as glucose and energy deprivation
The accumulation of lactic acid in solid tumors is often thought to be caused by tumor hypoxia – a by-product of glycolysis as the tumor cells shift to an anaerobic mode of energy production under hypoxia or due to the altered metabolic profiles of cancer cells. In spite of this apparent mechanistic link, these two factors exhibit significant disparities in their spatial and temporal distribution in tumors
Research on tumor microenvironments relies on our ability to manipulate
To characterize the gene expression program generated in response to lactic acidosis, hypoxia, and combined lactic acidosis and hypoxia, we made use of human mammary epithelial cells (HMEC) brought to replicative arrest by growth factor/serum withdrawal for 24 hours. Since the HMEC represent normal epithelial cells with intact signaling components, the response elicited in HMECs is likely to reflect the cellular response not biased by genetic mutations present in cancer cell lines. We exposed HMECs to four different culture environments for 24 hours in triplicate samples: 1) control – ambient oxygen level (∼21%O2) with neutral pH; 2) lactic acidosis – 25 mM lactic acidosis with pH 6.7; 3) hypoxia (2% O2) with neutral pH; 4) combined lactic acidosis and hypoxia. We did not find a significant change in media pH at the end of the 24 hour culture. The gene expression of these HMEC samples were interrogated with Affymetrix GeneChip U133 plus 2.0 arrays to measure the expression of more than 54,000 probe sets and at least 47,000 transcripts and variants. Gene expression profiles of cellular responses to hypoxia, lactic acidosis and combined stresses were first normalized by RMA, mean centered and filtered with the criteria of at least 2 (out of 3 samples in each experimental condition) observations with at least 1.75 fold changes to select 4722 probes sets. A clustering analysis on these genes revealed that hypoxia and lactic acidosis induced distinct sets of genes (
(A,B) The gene expression response of HEMC is shown when exposed to control, hypoxia, lactic acidosis and combined hypoxia and lactic acidosis conditions. 4722 genes with expression variations of at least 1.75 fold in two samples were selected and hierarchically clustered. Genes induced by hypoxia (vertical blue bar), lactic acidosis (pink), and repressed by lactic acidosis (orange) are marked and further expanded in (B). (C) The expression of three lactic acidosis-induced genes ERBB3, CD55 and PLAUR normalized by actin-beta were confirmed by real time RT-PCR. Similar results were observed when normalized by another control gene, B2M (D) Venn diagram showing the number of genes changed by lactic acidosis (1585 genes), hypoxia (217 genes) and overlap (54 genes) for whom the probability of an expression change exceeded 0.99. (E, F) The expression of genes comprising hypoxia (E) and genes comprising lactic acidosis (F) gene signatures was shown in respective heat maps in all four indicated conditions.
Compared with hypoxia, the lactic acidosis response is more dramatic with alterations of many more genes. Genes induced by lactic acidosis included: PLAUR, ERBB3, CD55, interleukin 15, CXCL16, angiogenin and MHC class I genes (
A supervised analysis of the full set of data on all 47,000 probe sets was performed using Bayesian multivariate regression analysis (BFRM) that has been utilized in a number of prior studies
To survey the molecular pathways triggered by lactic acidosis and hypoxia, we analyzed the Gene Ontology (GO) enrichment in the genes induced and repressed by hypoxia and lactic acidosis using GATHER
Previous studies have shown that lowering the extracellular pH from 7.4 to ∼6.7 will lead to a slight lowering of intracellular pH (pHi) from 7.4 to 6.9–7.0, which is likely to be mediated by monocarboxylate transporter (MCT) proteins
To further assess the extent to which the hypoxia and lactic acidosis response overlaps, we used binary logistic regression to estimate the probability of activation for the overall hypoxia and lactic acidosis pathways revealed in gene expression under individual conditions. We first used the control vs. hypoxia (for hypoxia probability) and control vs. lactic acidosis (for lactic acidosis probability) groups to provide the training sets to generate respectively hypoxia (
High lactate (lactosis) and low pH (acidosis) often co-exist in tumor lactic acidosis, but these two factors are not necessarily present simultaneously. To determine the respective contributions of lactosis and acidosis in the lactic acidosis response, we created culture conditions to separate the lactic acidosis condition (pH 6.7 created by 25 mM lactic acid) into lactosis (25 mM sodium lactate with neutral pH) and acidosis (pH 6.7 created by HCl) conditions. We analyzed the transcriptional responses of HMECs under lactosis and acidosis via microarrays. Gene expression profiles of the cellular responses to lactosis and acidosis were normalized by RMA, mean centered and filtered with the criteria of at least 4 (out of 6 sample in each experimental condition) observations with at least 1.75 fold change. 213 probes were identified and clustered in
(A) HMECs were exposed to three indicated environments: control, lactosis, and acidosis. 213 genes with expression varied from the mean at least 1.75 fold in 4 samples were selected and hierarchically clustered. A cluster of genes strongly induced by acidosis is shown (yellow vertical bar). (B) The expression of genes comprising acidosis gene signatures under indicated conditions was shown in heat maps. The expression of genes selected by statistical analysis (see
We previously showed that the hypoxia response elicited in cultured epithelial cells provides a molecular gauge of hypoxia response for cancerous human tissues
The gene signatures in hypoxia (A), lactic acidosis (B), and acidosis (C) response were assessed in the four indicated breast cancer expression datasets. The tumors stratified by the degrees of these responses were used to generate the Kaplan-Meier survival curves for the clinical outcomes exhibiting high and low indicated responses are shown. (D) In the Miller dataset, the lactic acidosis response score is significantly higher among the tumors with wild type p53 than mutant p53 (p = 3.288×10−11). (E) Acidosis response were also estimated for a group of breast cancer cell lines with different metastatic abilities and found to be negatively correlated with tumor aggressiveness determined in xenografted mice
Given the perceived relation of lactic acidosis with hypoxia in tumors, we evaluated the prognostic value of the lactic acidosis signature using the same statistical approach. We found that tumors with high lactic acidosis response signatures have significantly improved overall survival, in contrast to that for hypoxia response signature. This association with favorable clinical outcomes is consistent across all four breast cancer datasets with their different target populations and stages of cancers (
Among the breast cancer patients included in the Sotiriou dataset, 64 patients were treated with tamoxifen while the remaining 125 patients were untreated. Both hypoxia and lactic acidosis pathway signatures are predictive of poor and favorable outcomes respectively in the patients treated with tamoxifen (
To understand whether the prognostic value of lactic acidosis is evident in a cell autonomous manner, we tested the prognostic value of the lactic acidosis signature in various breast cancer cell lines grown in a controlled culture milieu. We used binary logistic regression to determine the probability of acidosis response in each of a set of breast cancer cell lines that vary in their potential for distant metastasis
We also tested the prognostic value of gene signatures reflecting lactic acidosis response and hypoxia response in a multivariate survival analysis using the Sotiriou data set. When both gene signatures are included in the Cox survival model on all samples for which we have survival, ER status, tumor size, and data on node involvement, the p-values for lactic acidosis and hypoxia are 0.0379 and 0.0069 respectively. When we include these two pathway variables along with clinical variables indicating ER status, tumor size >2 cm and node involvement, the p-values are .07, .008, .85, .0032 and .77 respectively. Dropping ER status and node involvement (poor predictors for this dataset) gives p-values of .06, .008, and .002.
We further evaluated parametric Weibull survival models involving different combinations of clinical variables and expression signatures.
Since lactic acid production and accumulation in solid tumors is likely to relate to the shift to glycolysis under hypoxia, a high degree of correlation is expected between hypoxia and lactic acid in solid tumors. What is the relationship between the degree of hypoxia and lactic acidosis responses in the same tumors? With our quantitative, probabilistic assessment of the pathway activities based on tumor gene expression, we can directly investigate the relationship between these two factors in individual tumors. Unexpectedly, the lactic acidosis response score correlated significantly in a negative fashion with the hypoxia score in all four expression studies, with R ranging from −0.35 to −0.45 and
(A) Scatter plots showing the relationship between the probability of hypoxia response (Y-axis) and lactic acidosis response (X-axis). Each point in the scatter plots represents a single tumor from the indicated breast cancer data sets. The overall correlation (R) and probability (p) between hypoxia and lactic acidosis signatures across all samples is shown for the indicated data set. (B) The tumors in the indicated breast cancer data set are separated into four groups based on hypoxia and lactic acidosis responses. Kaplan-Meier curves for the clinical outcomes of these four groups of tumors are shown with indicated colors. (C) The top ten genesets based on normalized enrichment score (NES) from GSEA analysis for the difference in pathway composition between the tumors with high vs. low lactic acidosis responses in the Pawitan data. (D) The breast samples in the Pawitan data was arranged from left to right by descending lactic acidosis score (top row). The expression of genes in TCA cycles in these samples is shown in heat map (orange means higher expression whereas blue means lower expression) together with lactic acidosis score.
To understand the unexpected association between lactic acidosis response and favorable clinical outcomes, we compared the pathway composition between tumors with high vs. low lactic acidosis responses in all four expression datasets using Gene Set Enrichment Analysis (GSEA)
There are two major pathways for ATP-generation in mammalian cells – glycolysis or aerobic respiration. One of the fundamental properties of cancer cells is their preferential utilization of glycolysis over aerobic respiration to produce ATP. The glycolytic phenotype of cancer cells is thought to offer selective advantages since the disruption of glycolysis phenotype (e.g. silencing of LDH-A) results in stimulation of mitochondrial respiration and significantly compromises their tumorigenicity and the proliferation under hypoxia
To investigate the possibility that lactic acidosis directly modulates the balance of energy production, we measured its influence on ATP production in cultured cells when aerobic respiration is inhibited by rotenone. In control conditions without hypoxia or lactic acidosis, we found that approximately 35 %of ATP production in Siha cells is sensitive to rotenone at 48 hours (
The contribution of aerobic respiration and glycolysis to ATP generation under control, lactic acidosis and hypoxia is measured by the degree of inhibition of ATP generation after treatment of rotenone (A) and 2-DG (B) at the indicated time after treatment. (C) The amount of ATP generation at different time points under the indicated conditions. (D) The genes in the glycolysis pathways were up-regulated by hypoxia and down-regulated by lactic acidosis. (E) The expression of genes listed as “glycolysis pathway” was extracted and clustered. (F) The mean expression values of the 53 glycolysis genes for each HMEC under hypoxia, lactic acidosis and hypoxia/lactic acidosis are calculated and shown. (G)(H) The expression of genes in the glycolytic pathways under hypoxia and lactic acidosis were used to predict the pathways activity and stratified the indicated breast cancer samples. This small set of genes recapitulated the result using the whole lactic acidosis and hypoxia gene signatures. (I) Scatter plots showing the relationship between the probability of hypoxia response (Y-axis) and lactic acidosis response (X-axis) for genes in the glycolysis pathways. Each point in the scatter plots represents a single tumor from the indicated breast cancer data sets. The probability (p) between hypoxia and lactic acidosis signatures across all samples in the indicated data set is shown.
These data reveal the distinct manner by which lactic acidosis and hypoxia redirect energy utilization – hypoxia favors the glycolytic pathways while lactic acidosis favors the aerobic respiration. This suggests these two stresses may impact cellular metabolism in distinct manner and have synergic effects on ATP inhibition when cells are exposed to simultaneous hypoxia and lactic acidosis. We measured ATP production under control, lactic acidosis, hypoxia and combined hypoxia and lactic acidosis. We found that lactic acidosis and hypoxia reduced the ATP production to ∼50% and 63% respectively in 48 and 96 hours (
To further understand how lactic acidosis and hypoxia modulate the balance between the aerobic respiration and glycolysis, we mapped their respective effects on gene expression onto the framework of metabolic pathways in energy metabolism (
In contrast, expression levels of these glycolytic genes were consistently repressed by lactic acidosis (
Under hypoxia, there was also a noticeable reduction in the expression of the genes in the TCA cycles and other mitochondria genes essential for aerobic respiration, consistent with previous studies
To further explore the signal transduction pathways of lactic acidosis through genomic analysis, we compared the lactic acidosis gene signatures with the database of “connectivity map”
(A) PI3K inhibitors are highly ranked with lactic acidosis signature in the connectivity map analysis. The “barview” is constructed from 453 horizontal lines, each representing an individual treatment instance, ordered by their corresponding connectivity scores calculated with lactic acidosis signature (+1, top; −1, bottom) with the instances corresponding to wormannin and LY-294002 were shown as black bars. Colors applied to the remaining instances reflect the sign of their scores (green, positive; gray, null; red, negative). (B), (C) The relationship between the predicted Akt and Acidosis pathway activities in the gene expression pattern of a breast cancer expression studies (B) and prostate tissue is shown between wild (WT) and Akt transgenic mouse (AKT-Tg) treated with placebo or mTOR inhibitor RAD001. (D) The effect of lactic acidosis on Akt activation in DU145 cells during serum exposure. Indicated amount of serum are added to the DU145 which have been placed in 0.2% serum conditions for 24 hours without (−) or with (+) 25 mM lactic acid. The same amount of cell lysates of DU145 cultured under indicated conditions were separated, transferred to blot and probed with indicated antibodies. (E), (F) The amount (mM) of lactate production (orange) and glucose consumption (blue) in 48 hour per million of WiDr (E) and SiHa (F) cells under hypoxia (left) or normoxia (right) with the following media conditions (1) control, (2) 25 mM lactic acidosis (pH 6.7), (3) 25 mM sodium lactate, (4) pH 6.7, (5) pH 6.5, (6) pH 6.0. (G) The model of the differentially modulated the balance of glycolysis and aerobic respiration as means of energy generation under control, hypoxia and lactic acidosis.
Gene signatures representing different pathways can be evaluated with lactic acidosis response signature in the gene expression data sets of breast cancer to further elucidate the molecular mechanisms of lactic acidosis
This inverse correlation between the lactic acidosis and Akt pathway activity in the tumor expression data lead us to hypothesize that lactic acidosis can inhibit the Akt pathway in the tumor cells. This possibility is also consistent with the correlation between lactic acidosis response and PI3K inhibition noted in the connectivity map analysis. Since this observed pathway activity seen in the gene expression may be caused by the corresponding change in the Akt enzymatic activity, we formally tested the effect of lactic acidosis on activation of Akt enzymatic activity in prostate cancer cell line DU145 during serum exposure. Growth of DU145 cells in a serum-free condition resulted in the inhibition of the Akt enzymatic activity, as seen by the absence of phosphorylation of Ser473 (
Given the important role of the PI3K/Akt pathway in the tumor glycolytic phenotypes
Taken together, we propose a model (
Lactic acidosis and hypoxia are two well recognized features in human cancers. Although tumor lactic acidosis is often thought to co-exist with hypoxia, relatively little is known about its cellular response, relationship with hypoxia or its role in tumor progression. Our current study presents, to our best knowledge, the first genomic analysis of lactic acidosis activities in human breast cancers
A number of studies have now described the power in utilizing large scale gene expression data to develop signatures representing important biological states – in this context, the signature becomes a surrogate phenotype that can be used to explore the biological relevance in the diverse space of
One of the principal metabolic properties of cancer cells is their preferential use of glycolysis for energy generation even in the presence of oxygen (aerobic glycolysis). Warburg has speculated that this is caused by defective or functionally impaired mitochondria. Recent studies suggest that mutations affecting mitochondrial DNA or enzymes of the TCA cycle might contribute to tumor formation, tumor progression and the Warburg effect. Two enzymes in the TCA cycle enzymes, SDH and FH, are found to be tumor suppressor genes
In addition to generation of ATP, glycolysis is responsible for the generation of acetyl-CoA, which feeds into the TCA cycle for aerobic respiration. Thus, the reduction in glycolysis under lactic acidosis may imply a reduction in the amount of acetyl-CoA derived from the glycolytic process. To maintain cellular energy, cells may increase the use of other energy sources, such as the β-oxidation of fatty acids as source of acetyl-CoA. Since Akt pathway activity is known to suppress β-oxidation and thus energy generation from fatty acids
These data suggest that the lactic acidosis-induced gene expression program may have a direct causal role in impacting tumor biology to affect clinical outcomes. It will be important to understand what specific lactic acidosis-driven biological processes underlie the phenotypic differences between groups of tumors separated by lactic acidosis pathway activity. The mechanisms underlying the variation in lactic acidosis responses in breast are still unknown. There are three reasonable possibilities; variations in the lactic acidosis-response program could reflect: 1) actual variations in lactate and/or acidity in the tumors 2) cell-type-specific variations in the magnitude of, or threshold for, the response to
It is widely believed that the intimate relationship between tumor progression, tumor microenvironment and the response of the tumor to that environment, when combined with other genetic and clinical factors, offers promise for improving our understanding of the heterogeneity of the disease. Progress in this direction will require a substantial advance in our – currently limited – ability to dissect the roles played by multiple characteristics of the tumor microenvironment. We aim to develop this approach to further dissect other various microenvironmental factors in cancers, such as glucose starvation, reoxygenation and ATP depletion
Human mammalian epithelial cells (HMEC) were cultured in MEGM (Cambrex) and growth factors were withdrawn for 24 hrs before being placed under different environmental stresses. DU145 cells were cultured in RPMI1640 with 10% FBS, 1% sodium pyruvate, 1% L-glutamine, 1% Hepes and 1% antibiotics (penicillin, 10000 UI/ml; streptomycin, 10000 UI/ml). WiDr and SiHa cells were cultured in DMEM with 10%FBS. Lactic acidosis conditions were created with the addition of 25 mM lactic acid (Sigma) to pH 6.7. Hypoxia was created by lowering the oxygen level to 2%. Similarly, the lactosis condition was created with the addition of 25 mM sodium lactate, while acidosis conditions were created by titrating media to pH of 6.7 with HCl.
RNAs were extracted by miRVana kits (Ambion) and hybridized to Affymetrix Hu133 plus 2 genechips with standard protocol. All microarray data are available on GEO (GEO accession number GSE9649). Hierarchical clustering with weighted average linkage clustering was performed after indicated data filtering based on spot quality and variations in signal intensity as described
RNAs were reverse-transcribed to cDNAs with SuperScript II reverse transcription kit following the manufacturer's protocol (Invitrogen). cDNAs were then used as the substrate for gene expression level measurements by qPCR with Power SYBRGreen PCR Mix (Applied Biosystems) and primers specific for ErbB3(Forward:
DU145 cells were serum-starved (0.2%FBS) for 24 hrs, followed by 24 hr of continuous incubation in serum starved (0.2%FBS) media (as the control) and media with 25 mM lactic acid. 20%, 10%, and 5% FBS were applied for 30 mins to induce Akt activation. Proteins were extracted with PARIS kit (Ambion) and equal amount of protein samples were loaded to SDS-PAGE gels and blotted with pSer473 Akt antibody (Cell signaling) and other indicated antibodies.
WiDr and SiHa cells were plated in six-well dishes (800,000 cells per well). The next day fresh media of respective conditions, including control, 25 mM lactic acidosis, 25 mM sodium lactate, acidosis of pH 6.7, pH 6.5 and pH 6.0 were applied to cells with the continuous incubation of 48 hrs under either normoxia or hypoxia (0.5% O2). After 48hr incubation, media were collected for glucose (ACCU-CHECK, Roche) and lactate (ARKRAY) measurements and normalized against cell number to obtain the glucose consumption/lactate production per million cells.
SiHa cells were plated at the density of 2×104 cells/ml. On the next day, respective media of control and 25 mM lactic acidosis, as well as media containing drug inhibitors, 2-DG and rotenone (Sigma), would be applied. They will then be incubated under normoxia and hypoxia (1% oxygen) respectively. ATP was measured by ATPlite 1 step luminescence ATP detection assay system kit with the protocol provided by the manufacturer after 48 and 72 hours (Perkin Elmer). To prevent the interference caused by different colors of control versus lactic acidosis media, we replaced culture media with PBS right before the addition of substrate solution.
The pathway activities of hypoxia (A) and lactic acidosis (B) signatures were estimated as probability (Y-axis) by binary regression models in the HMECs exposed to indicated treatments (control, lactic acidosis, hypoxia and combined lactic acidosis and hypoxia, X-axis).
(1.02 MB EPS)
The change in gene expression in the indicated conditions for a set of genes whose expression is induced (A) or repressed (B) only in the presence of both hypoxia and lactic acidosis.
(3.02 MB EPS)
The confirmation of the induction of ERBB3 and SOD2 expression by acidosis using real time RT-PCR.
(1.09 MB EPS)
The prognostic values of gene signatures reflecting hypoxia (A, C) and lactic acidosis (B, D) response among the patients treated (A, B) and untreated (C, D) with tamoxifen in the Sotiriou datasets. The graphs show Kaplan-Meier curves for two patient subsets simply split on the median signature level in each case, for visual presentation. The p-values are for regression coefficients of the signature in the survival model analysis.
(0.93 MB EPS)
Prognostic values of predictive model based on different combination of gene signatures and clinical variables. The breast cancers are separated in (A) based on hypoxia, lactic acidosis response and nodal status; in (B) based on hypoxia, lactic acidosis response, tumor size, ER status and nodal status; in (C) tumor size, ER status and nodal status. The graphs show Kaplan-Meier curves for two patient subsets simply split on the median signature level in each case, for visual presentation. The p-values are for regression coefficients of the signature in the survival model analysis.
(1.31 MB EPS)
Genes in the TCA cycle gene set are highly enriched in the tumors with high lactic acidosis using GSEA.
(0.02 MB PNG)
Breast samples in the Pawitan data set arranged from left to right by descending hypoxia score (top row). The expression of genes in TCA cycles in these samples is shown in heat map, orange color being higher expression while blue color being lower expression, together with hypoxia score.
(0.57 MB EPS)
Higher hypoxia response of TCA cycle genes was associated with better prognosis, in contrast to the association with poor prognosis for hypoxia genelists (A). On the other hand, TCA cycle genes of lactic acidosis response were not associated with differences in clinical outcome (B).
(1.87 MB EPS)
Estimated posterior probability and corresponding Bayesian p-value (1-probability) for gene probe sets showing responses to hypoxia, lactic acidosis and acidosis.
(10.86 MB XLS)
GO terms enrichments (with p values) for genes which are upregulated and downregulated in HMECs under lactic acidosis.
(0.06 MB XLS)
GO terms enrichments (with p values) for genes which are upregulated and downregulated in HMECs under hypoxia.
(0.03 MB XLS)
The comparison between GO term enrichment under hypoxia and lactic acidosis.
(0.02 MB XLS)
The expression mean-centered values of indicated MCT family proteins under hypoxia and lactic acidosis.
(0.02 MB XLS)
The lists of genes which are strongly induced or repressed (top 1% probability) under lactic acidosis and hypoxia.
(0.18 MB XLS)
Gene sets enriched in the four expression datasets of breast cancers with high vs. low lactic acidosis pathways from the GSEA analysis.
(0.05 MB XLS)
The GO terms enriched for genes used for the analysis of glycolysis pathways.
(0.02 MB TXT)
p-values for the changes in gene expression under lactic acidosis and hypoxia for the listed glycolysis genes.
(0.02 MB XLS)
Connectivity map analysis of the lactic acidosis gene signatures.
(0.09 MB XLS)
Relationship between lactic acidosis signature and Akt pathway activities in different tumor expression datasets.
(0.04 MB XLS)
Statistical supplement. Details of Bayesian factor regression models and survival analyses.
(0.14 MB PDF)
We appreciate the technical assistance from Vivian Lentchitsky, the critical comments on the manuscripts from Gregory LaMonte and materials/reagents from Bala Blakumaran.