Research on samples from human acute myeloid leukemia (AML) patients has suggested that the a leukemia stem cell (LSC) gene expression profile can accurately be used to predict patient survival, even among AML patients with cytogenetically normal disease. An international team led by researchers at Ontario’s University Health Network and the University of Toronto’s Department of Molecular Genetics claim their findings support those from xenograft models suggesting that the disease is propagated by a population of rare LSCs.
Describing their findings in Nature Medicine, John E. Dick, M.D., Kolja Eppert, M.D., and colleagues report that the LSC gene signature is similar to that of hematopoietic stem cells (HSCs), and that both gene signatures were significant and independent predictors of patient survival. Moreover, the LSC gene signature could accurately predict poor survival in about 50% of AML patients with cytogenetically normal disease who would otherwise be classified as low risk on the basis of mutations in specific individual genes. Importantly, the LSC signature could be identified even in unsorted AML cell populations in which LSCs were very rare, suggesting that LSC-derived AML-blasts retain at least part of their progenitors’ expression profiles.
The researchers finally highlight the need to develop LSC biomarkers for use in patient evaluation, and claim that therapies targeting LSCs would improve survival outcomes. Their paper is titled “Stem cell gene expression programs influence clinical outcome in human leukemia.”
Xenograft studies indicate that some solid tumors and leukemias are organized as cellular hierarchies sustained by cancer stem cells (CSCs), the authors write. Although the clinical relevance of the CSC model in humans has, to date, remained uncertain, recent evidence from leukemia patients does support the notion that LSC properties may be prognostic: correlative studies have linked outcome with either the capacity for a sample to be xenografted or surface expression of LSC-linked markers.
Moreover, they continue, if CSCs are more prognostic than non-CSCs, the molecular machinery underpinning the properties of such stem cells is likely to likely to influence clinical outcome. “Indeed, LSCs have the core set of biological functions common to all stem cells, including self-renewal and the ability to produce differentiated, non-stem cell progeny.”
The University Health Network team set out to investigate the specific properties of CSCs in leukemia further by defining the gene expression signatures of both LSCs and hematopoietic stem cells (HSCs) from functionally validated, sorted fractions of primary human AML samples. Sixteen primary human AML samples were sorted into four cell populations on the basis of surface expression of the markers CD34 and CD38: previous studies have suggested LSCs are primarily CD34+CD38-. Each cell fraction was then assayed using an optimized xenotransplantation assay to identify LSC-enriched and LSC-depleted populations.
As expected, LSCs were found in the CD34+CD38− fraction in all but one of the informative cases. However, the cells were also found in at least one other fraction in the majority of AML samples, and in 50% of cases the majority of LSCs were found in the CD34+CD38+ fraction, “establishing heterogeneity between cell surface marker expression and LSC activity among individual samples,” the authors state. Overall, LSCs ranged in frequency from 1 in 1.6 × 103 cells, to 1 in 1.1 × 106, but were generally found at the highest frequency in the CD34+CD38− fraction. Importantly, the researchers note, “LSC-containing fractions that initiated leukemia generated a xenograft that acquired the marker phenotype of non-LSC fractions, confirming earlier reports that AML is hierarchically organized.”
Each functionally validated fraction was then subjected to global gene expression analysis to identify LSC-related (LSC-R) gene profiles. Bioinformatics analyses compared global gene expression patterns of 25 LSC-enriched fractions to 29 fractions without LSCs. The results threw up an initial LSC-R signature including 42 genes, which was validated by RT-PCR. This same gene signature was identified in AML samples with a variety of karyotypic alterations and French-American-British subtype.
Then, because LSCs and HSCs both have canonical stem cell functions including self-renewal and the ability to make non-stem cell, mature progeny, the team also generated HSC gene-expression profiles to see whether human LSCs share molecular mechanisms and gene-expression programs with HSCs.
A comparison of the resulting 121-gene HSC-related (HSC-R) profile with the LSC-R gene profile identified 44 leading-edge genes, termed the core enriched HSC-LSC genes (CE-HSC-LSC), which appeared to represent HSC genes that are also differentially expressed in LSCs. Of these, 18 are implicated in stem cell regulation, oncogenesis, or both, the authors note. However, protein interaction network analyses highlighted the enrichment of multiple pathways distinct from the progenitor network, including Notch and Jak-STAT signaling, which are implicated in stem cell regulation, “thereby supporting the stem cell nature of the HSC- and LSC-related gene profiles.”
Significantly, when the researchers compared their data with previously generated gene datasets from stem, progenitor, and mature cell populations, and from embryonic stem cells, they found that LSC gene expression correlated positively with primitive cell gene sets and negatively with gene sets derived from more differentiated cells and from ES cells. “Collectively, our data establishes that an HSC expression program and not a common lineage-committed progenitor or ES cell-expression pattern is preferentially expressed in LSCs compared with more mature non-LSC leukemic cells,” they conclude.
To test the clinical relevance of LSCs, the team then looked for a link between LSC-R and HSC-R gene signatures and clinical outcomes, in three large clinically annotated gene expression datasets derived from unsorted AML cells. They found that the LSC-R and HSC-R profiles were very similar among a cohort of 285 AML samples, and that the profiles were either positively or negatively correlated with cluster gene sets characterized by molecular markers indicative of poor prognosis or good prognosis.
They then analyzed the LSC-R, HSC-R, and CE-HSC-LSC signatures in new gene expression data generated on about a third of the AML samples that had been stratified in terms of poor or good prognostic risk groups on the basis of cytogenetic alterations. In this case higher expression of all three signatures distinguished poor prognostic risk subjects. “These findings demonstrate that AML samples associated with worse prognosis express stem cell-related genes more highly than less aggressive AML samples,” they note.
As a final test of clinical relevance, the researchers looked at a cohort of 160 cytogenetically normal AML subjects for whom gene expression and survival data were available. For this analysis they used the LSC-R or HSC-R gene signature to divide the subjects into two equal groups, based on the median expression of the respective signatures in bulk AML bone marrow cells.
Both signatures negatively correlated with overall survival and event-free survival with a high degree of significance. Moreover, there was a significant negative correlation between the occurrence of complete remission and high expression of the LSC-R signature, and an almost significant negative correlation between complete remission occurrence and the HSC-R signature. “Our data demonstrate that high expression of stem cell expression signatures directly predicts patient survival in cytogenetically normal AML and, therefore, that variation in stem cell expression programs among subject samples is highly correlated with heterogeneity in disease outcome,” the team writes.
Of particular interest, multivariate analyses showed that the LSC-R and HSC-R signatures predicted overall and event-free survival independently of known molecular prognostic factors in cytogenetically normal patients, such as molecular risk status and CEBPA mutation. In fact, after subdividing the AML samples that were informative for molecular risk status, they found that each stem cell signature identified subsets of subjects with poor survival in both high molecular risk (HMR) group and low molecular risk (LMR) groups. “Thus, the LSC-R and HSC-R signatures can be used to separate patients currently identified as low risk into groups who respond well to standard therapy and those who may benefit from more intensive therapy, including stem cell transplant.”
The researchers state their studies provide evidence supporting the hierarchical organization of AML according to the CSC model, and support the notion that LSCs are not just artifacts of experimental xenograft models. Moreover, they suggest a similar investigative approach could be used to assess both the identity and clinical relevance of LSCs and CSCs from other leukemias and solid tumors.
“Our findings warrant validation in a large cohort and a clinical trial to test the LSC-R signature in the LMR subgroup,” they continue. “If our results are confirmed, poor-risk patients might benefit from more aggressive therapy such as allogeneic transplant or alternative therapy.”
The authors admit that they found it rather counterintuitive that a stem cell signature could be detected in unsorted samples containing a very small population of LSCs, and in which the vast majority of cells are differentiated non-LSC blasts. This, they remark, suggests that stem cell expression programs persist in these blasts at a population level, even though no individual blast retains the full repertoire or expression level of stem cell genes.