P53-INDEPENDENT, NORMAL STEM CELL SPARING EPIGENETIC-DIFFERENTIATION THERAPY FOR MYELOID AND OTHER MALIGNANCIES

Summary

Cytotoxic chemotherapy for acute myeloid leukemia (AML) usually produces only temporary remissions, at the cost of significant toxicity and risk for death. One fundamental reason for treatment failure is that it is designed to activate apoptosis genes (eg., TP53) that may be unavailable because of mutation or deletion. Unlike deletion of apoptosis genes, genes that mediate cell cycle exit by differentiation are present in myelodysplastic syndrome (MDS) and AML cells but are epigenetically repressed: MDS/AML cells express high levels of key lineage-specifying transcription factors (TF). Mutation in these TF (eg., CEBPA) or their cofactors (eg., RUNX1) affect transactivation function and produce epigenetic repression of late-differentiation genes that antagonize MYC. Importantly, this aberrant epigenetic repression can be redressed clinically by depleting DNA methyltransferase 1 (DNMT1, a central component of the epigenetic network that mediates transcription repression) using the deoxycytidine analogue decitabine (DAC) at non-cytotoxic concentrations. The DNMT1 depletion is sufficient to trigger upregulation of late-differentiation genes and irreversible cell cycle exit by p53-independent differentiation mechanisms. Fortuitously, the same treatment maintains or increases self-renewal of normal hematopoietic stem cells (HSC), which do not express high levels of lineage-specifying TF. The biological rationale for this approach to therapy appears to apply to cancers other than MDS/AML also. DAC or 5-azacytidine dose and schedule can be rationalized to emphasize this mechanism of action, as an alternative or complement to conventional apoptosis-based oncotherapy.

The lineage and maturation context of cancer cells

A major goal of cancer research is to find differences between normal stem cells and cancer cells that can be used to selectively destroy cancer cells. One difference is frequent mutation or deletion of key apoptosis genes (e.g., TP53, p16/CDKN2A) in cancer cells^1-7. However, this feature of cancer cells works against the objectives of conventional apoptosis-based chemo- or radiation therapy, contributing to treatment resistance and toxicity.

Instead of increasing cancer cell apoptosis/death as the primary treatment objective, an alternative is to identify and target pathways of cancer cell proliferation⁸. Since these pathways may be differentiation-context dependent, it is useful to understand the lineage and maturation stage of cancer cells. Examination of cancer cell morphology and surface phenotype usually reveals lineage-commitment. Indeed, lineage-markers (morphologic and immunohistochemical) underpin classification of malignancy, and lineage-commitment is an implicit component of some oncotherapy, for example, hormonal blockade to treat breast and prostate cancer.

The expression pattern of key DNA binding transcription factors that drive lineage-commitment and progressive maturation provide further valuable insight into differentiation context: lineage-commitment and progressive maturation absolutely requires and is driven by coordinated expression of key transcription factors^9-11. Missense mutations in genes for these factors, for example CEBPA and GATA1, are known initiating events in acute myeloid leukemia (AML) pathogenesis^12;13. These genes are expressed at high levels only with lineage-commitment. Indeed, AML cells express high levels of these factors (Figure 1)¹⁴. Despite the high expression of CEBPA, AML cells express relatively low levels of the key late-differentiation driver transcription factor CEBPE (Figure 1)^15;16 (expression levels of CEBPE increase during the transition from proliferating pro-myelocytes to non-proliferating myelocytes¹¹, and CEBPE terminates proliferation in myeloid and AML cells^10;17-22).

Figure 1.

The pattern of expression of key lineage-specifying and late differentiation factors in AML cells suggests impaired differentiation in lineage-committed cells. A) CEBPA (lineage-specifying factor) and CEBPE (late differentiation factor) expression in AML myeloblasts (n=318) compared with normal CD34+ cells, bone marrow and peripheral blood (n=38), and normal myeloblasts (n=3) (p-values Wilcoxon Two-sample test. Raw data extracted from GEO Datasets ^148;149, gene expression measured by microarray). B) CD34+CD38- cells from AML patient bone marrow (n=9) express higher levels of CEBPA, and a higher CEBPA/HOXB4 ratio, than CD34+CD38- cells from normal bone marrow (n=4). p-values Median Two-Sample test. Raw data extracted from GEO Datasets ¹⁵⁰. C) Similar findings in CD34+ AML cells (n=10) and normal CD34+ cells (n=11) analyzed by gene-expression microarray. Expression levels represented by heat-map. p-values Wilcoxon Two-sample test. Raw data extracted from GEO Datasets ^151;152D) CD34+CD38- cells from AML patient bone marrow (n=9) express higher levels of CEBPα, and a higher CEBPα/HOXB4 ratio, than CD34+CD38- cells from normal bone marrow (n=4) ¹⁵⁰. p-values Median Two-Sample test. Raw data extracted from GEO Datasets ¹⁵⁰.

The identity and gene expression profiles of key transcription factors that drive progressive maturation of solid tissues are not as well characterized as for hematopoiesis²³. Nonetheless, where the identity of these transcription factors is known, the solid tumors that arise from these tissues express high levels of lineage-commitment transcription factors, and low levels of key late-differentiation genes: malignant melanoma cells express high levels of the melanocyte commitment factor MITF, and point mutations in melanoma target MITF and another key driver of melanocyte commitment and early differentiation, SOX10^24;25. However, melanocyte late-differentiation driver genes, eg., SOX9, are epigenetically repressed^26;27. Medulloblastoma cells express high levels of genes that are expressed early in cerebellar development/differentiation, but relatively low levels of late cerebellar differentiation genes²⁸ (the medulloblastoma gene expression profile corresponds to the normal maturation stage with the highest rate of proliferation and migration²⁸). Squamous cell lung carcinoma cells demonstrate a gene expression profile of an intermediate stage of normal lung development/differentiation²⁸. Rhabdomyosarcomas express high levels of the lineage-specifying transcription factor MYOD²⁹, however, disruptions to the usual interactions between MYOD and E-proteins results in repression rather than activation of late-differentiation target genes²⁹. Chromosome translocations target the master regulator of differentiation ETV6 in breast cancer and sarcoma^30;31, PAX8 in follicular thyroid cancer³², PAX3 and PAX7 in rhabdomyosarcoma^33;34 and TFE3 in papillary renal cell cancer³⁵. Evidence for lineage-commitment and lineage-dependency in solid tumors has also been reviewed elsewhere³⁶.

Is lineage-commitment a feature of cancer ‘stem cells’ or cancer initiating cells?

Surface phenotype can be used to sort cancer cell populations into subsets. These subsets can then be xeno-transplanted into immunocompromised mice for evaluation of cancer initiating efficiency (as a measurement of self-renewal capacity). The earliest studies suggested that AML cells with leukemia-initiating capacity had a surface phenotype resembling that of normal hematopoietic stem cells (CD34+38-)^37;38. This suggested that AML cell populations might recapitulate the hierarchical structure of normal hematopoiesis, with only cells with a stem cell phenotype having the self-renewal capacity to sustain the bulk AML cell population³⁸. Recently, it has been reported that technical factors may have biased results from the earliest studies³⁹. Accordingly, in numerous recent studies, AML initiating cells had a surface phenotype suggesting lineage-commitment (progenitor phenotype) (CD34+38+, CLL-1+, CD71+, CD90 -, c-Kit -)^39-46. Also, with use of more immunocompromised mice as recipients, AML initiating cell surface-phenotypes are not stem cell restricted^47-50. Even AML cells with a stem cell surface phenotype (CD34+CD38-) express high levels of the lineage-specifying transcription factor CEBPA, low levels of the late-differentiation driver CEBPE, and low levels of stem cell genes such as HOXB4, when compared to normal CD34+38- cells (Figure 1)¹⁵.

In solid cancers, surface markers that identify cancer cell subsets with the highest cancer initiating efficiency, such as CD133, are surface markers of both progenitors and stem cells^51-54, and therefore, provide limited information regarding stem versus progenitor context. Similar to the experience with AML cells, use of more immunocompromised mice in these assays suggests cancer-initiating capacity is less restricted than early estimates, with a similar frequency of cancer initiating capacity in the CD133 positive and negative compartments⁵⁵.

Gene expression profiles of cancer cells and embryonic stem cells can overlap⁴². Recently, it has been shown that the overlapping gene expression signatures can be attributed, in large part, to the activity of MYC, a driver of cell proliferation⁵⁶. Importantly, MYC upregulation is a feature of, and required for, the active proliferation that occurs with lineage-commitment by stem cells^57;58. In other words, the MYC module that is associated with cancer is a normal feature of early progenitors.

Progressive epigenetic repression of late-differentiation genes during neoplastic evolution

Coactivator protein complexes recruited by DNA binding transcription factors contain chromatin modifying enzymes that create activation marks on histones, and also assist in recruitment of the basal transcription factor complex. Conversely, corepressor protein complexes that can be alternatively recruited by transcription factors contain chromatin-modifying enzymes that create repression marks on histones and DNA. As discussed earlier, mutations or translocations in lineage-commitment/early-differentiation transcription factors (or their cofactors) affect corepressor/coactivator recruitment decisions⁵⁹ and produce epigenetic repression of late-differentiation genes that would otherwise terminate active MYC-driven proliferation^{10;16-22;59;60}.

Typically, evolution and progression of cancer is accompanied by increasing impairment of maturation, illustrated for example during myelodysplastic syndrome (MDS) progression into AML (Figure 2). Conceivably, this progressive impairment of maturation is caused by further imbalance in corepressor/coactivator recruitment by transcription factors that regulate expression of late-differentiation genes. Supporting this possibility, genetic abnormalities that target chromatin modifying enzymes (e.g., ASXL1) accompany MDS progression to AML^61-63 (Figure 2). A high rate of mutation or amplification in chromatin modifying enzymes is also observed in solid tumors⁶⁴ (comprehensively reviewed elsewhere^65-69). One net consequence of these abnormalities is an increase in repression marks on histones and in gene promoters that have correlated with disease aggression in multi-variate analyses of multiple malignancies including AML, ALL, and multiple solid tumors^70-76. Indeed, repressive chromosome marks can be observed over large chromosome regions in cancer cells (reviewed in⁷⁷).

Figure 2.

Progression of myelodysplastic syndrome (MDS) into acute myeloid leukemia (AML). A) Normal hematopoiesis: Hematopoietic stem cells (HSC) can self-renew or give rise to lineage-committed daughter cells (progenitors). Progenitors actively divide (transit-amplification) until mature cells are formed. B) The initial abnormality (first hit) in multi-hit neoplastic evolution may occur in an HSC (the cell of origin is an HSC). However, the growth advantage is conferred to progenitors, by epigenetic repression of late differentiation genes. Early in the disease process, the differentiation impairment produced by the initiating abnormality may not be severe enough to decrease mature cell numbers. However, a left-shift in the marrow compartment may be noted, as differentiation impaired precursors accumulate. C) Mature cell numbers decrease, and there is a progressive left-shift, with additional hits that cause progressive epigenetic repression of late differentiation genes in the progenitor compartment, conferring the property of self-renewal to lineage-committed cells (producing leukemia-initiating cells, LIC, and evolution of MDS into AML).

The model suggested by these observations

The preceding observations suggest the following model of MDS/AML and possibly other cancers:

Adult stem cells are quiescent (reviewed in⁷⁸). However, daughter cells that lineage-commit proliferate actively (a MYC-driven process^57;58). This proliferation is usually self-limited by the activation of late-differentiation genes that antagonize MYC^16;22;60;79. However, mutation or translocation in early-differentiation driving transcription factors produces aberrant epigenetic repression of these late-differentiation genes. In other words, the MYC activation may be physiologic, having occurred as a consequence of lineage-commitment by stem cells; the pathologic event is failure to activate late-differentiation genes that antagonize MYC. Since the repression of late-differentiation genes is epigenetic and not genetic, it is potentially reversible. Furthermore, the key DNA-binding factors that usually drive expression of the late-differentiation genes are expressed at high levels in the lineage-committed malignant cells. The fundamental problem is an imbalance in corepressor versus coactivator recruitment at late-differentiation genes⁵⁹.

This model is readily tested: treating MDS/AML or other cancer cells with conditions or drugs that antagonize corepressor function should restore late-differentiation gene expression and terminate proliferation. This has indeed been observed, with a number of different strategies to relax chromatin, and in a spectrum of cancer histologies and genotypes: aggressive, differentiation-impaired melanoma and breast cancer cells resumed differentiation and exited cell cycle when exposed to an embryonic cell micro-environment that opens chromatin^80;81. Similarly, oocyte extracts, another micro-environment that induces DNA hypomethylation and removes repressive histone marks, terminated tumorigenicity of breast cancer cells⁸². Drugs that inhibit histone deacetylases (chromatin modifying enzymes that create repressive histone marks) induce terminal differentiation in a spectrum of leukemia and cancer primary cells and cell lines^83-89. Similarly, the deoxycytidine analogue decitabine, which depletes DNA methyl-transferase 1 (DNMT1) (DNMT1 creates the methyl-CpG DNA repression mark) and relaxes chromatin^90;91 terminates proliferation of various AML and cancer primary cells and cell lines^85;88;92-95.

Irreversible cell cycle exit by epigenetic-differentiation does not require functional p53 or p16/CDKN2A

p53 and p16/CDKN2A-null mice, although cancer-prone, demonstrate essentially normal development^96;97, suggesting that cell cycle exit by differentiation may not require these master regulators of apoptosis. Evaluating this possibility using clinically available drugs is particularly relevant from a translational perspective. One confounding factor in the interpretation of such studies is that drug therapy can have non-epigenetic effects, including antimetabolite or DNA damaging effects that cause apoptosis, that may contribute to cell cycle exit⁹⁸. To address this issue, we have conducted experiments focused on the drug decitabine^14-16;27: unlike the cytidine analogues cytarabine or gemcitabine, the sugar moiety of decitabine is unmodified. Therefore, at low concentrations, DNA-incorporated decitabine does not terminate DNA chain elongation^99;100. Accordingly, decitabine can be administered at doses that deplete DNMT1 without causing significant DNA damage or cytotoxicity, both in vitro and in vivo^92;99-103.

Treatment of AML, renal cell cancer and melanoma cells with these concentrations of decitabine allowed one or two cell divisions, upregulated key drivers of late myeloid (CEBPE), epithelial (HNF4A) and melanocyte (SOX9) differentiation respectively, and induced cell cycle exit accompanied by upregulation of p27/CDKN1B, the cyclin dependent kinase inhibitor that mediates cell cycle exit by differentiation^104-107. Further underlining the p53-independence of the differentiation-mediated cell cycle exit, many of the cells used in these experiments were p16/CDKN2A and/or p53 null^15;16;27;108. In vitro observations were readily recapitulated in murine xenotransplantation models of AML, renal cell cancer and melanoma, by using a dose and schedule of decitabine that depleted DNMT1 without in vivo myelotoxicity^15;16;27;108. Fortuitously, p27/CDKN1B is rarely deleted from cancer cells, unlike p16/CDKN2A and p53: in 770 cell lines analyzed by the Cancer Genome Project (Wellcome Trust Sanger Institute), there was homozygous deletion of p16/CDKN2A in 218 cell lines, loss of heterozygosity in 278 cell lines, and mutation in 276 cell lines. There was homozygous deletion of TP53 in 5 cell lines, loss of hetrozygosity in 482 cell lines, and mutation in 482 cell lines. In contrast, there was p27/CDKN1B loss of heterozygosity in 161 cell lines but no mutations or homozygous deletions in any of the 770 cell lines analyzed.

Why self-renewal of normal stem cells is maintained with this treatment approach

Hematopoietic stem cell genes, such as HOXB4 and c-Kit, are rapidly repressed during the process of hematopoietic lineage commitment and differentiation¹⁰⁹. Since DNMT1 is a central component of the epigenetic network that mediates transcription repression¹¹⁰, DNMT1 depletion by shRNA or decitabine prevented stem cell gene repression by differentiation stimuli and maintained stem cell phenotype¹⁰⁹. These observations also explain why other drugs that antagonize transcription repression, such as histone deacetylase inhibitors and 5-azacytidine, also increase hematopoietic stem cell self-renewal^111-118. However, decitabine treatment after the stem cell gene repression phase of the differentiation process augmented differentiation¹⁰⁹. Therefore, the cell fate consequences of depleting DNMT1 with decitabine depend on baseline maturation stage (Figure 3). In other words, differences in maturation stage/lineage-commitment underlie the contrasting effects of chromatin-relaxing drugs on self-renewal and differentiation of malignant and normal stem cells.

Figure 3.

Maturation context explains why non-cytotoxic, DNA methyl-transferase 1 (DNMT1) depleting concentrations of decitabine increase normal hematopoietic stem cell self-renewal but induce terminal differentiation of AML cells. A) DNMT1 plays a central role in the network of chromatin-modifying enzymes that are implicated in transcription repression. B) In normal hematopoietic stem cells (HSC), decitabine (DAC) to deplete DNMT1 and antagonize transcription repression prevents a necessary first step in lineage-commitment, which is repression of stem cell gene expression. Therefore, DAC treatment maintains HSC self-renewal, even in differentiation promoting conditions ^109;111-118. If DAC is added shortly after the differentiation-inducing stimulus (after the phase of stem cell gene repression), it does not prevent and may even increase differentiation ¹⁰⁹. C) Leukemia-initiating events, such as RUNX1 mutation, can originate in the germ-line or in hematopoietic stem cells, however, RUNX1 deficient stem cells can lineage-commit in response to a differentiation stimulus, with intact repression of stem cell genes. Instead, RUNX1 cooperation with lineage-specifying transcription factor to active late-differentiation genes is impaired, by coactivator/corepressor imbalance at late-differentiation genes, producing aberrant self-renewal (proliferation at the same level of differentiation) in lineage-committed cells. In these cells, primed to differentiate with high levels of lineage-specifying factors, and in which repression of late differentiation genes is by epigenetic means, DAC can resume differentiation and differentiation-mediated cell cycle exit.

Towards effective clinical translation

The idea of using differentiation to terminate malignant cell proliferation (differentiation therapy) was mooted more than 50 years ago^119-121. However, as a primary objective of clinical therapy, it is currently limited to all-trans retinoic acid (ATRA) treatment of acute promyelocytic leukemia (APL)¹²².

One reason has been that the biological model or pathway basis for pursuing epigenetic-differentiation therapy has not been clear. Indeed, the model of cancer as being sustained by self-renewing cancer ‘stem cells’, does not provide a rationale for this mode of therapy, since in the stem cell model, differentiation-impairment is presumably a consequence rather than a cause of malignant self-renewal^36;38;123. However, recent data, as outlined above, challenge the stem cell model, and provide a biologic and mechanistic rationale for epigenetic-differentiation therapy.

Another important reason for limited translation is pharmacologic and mechanism of action complexity and limitations of compounds that differentiate AML or cancer cells in vitro: ATRA targets leukemia fusion proteins containing RARA, which are present only in APL. Low dose cytosine arabinoside (cytarabine, AraC) may induce differentiation but is predominantly cytotoxic¹²⁰. Histone deacetylase inhibitor drugs may induce differentiation but also have cytotoxic effects, and it is difficult to separate epigenetic effects of these drugs from DNA damage and apoptosis induction^124-126. Similarly, decitabine that depletes DNMT1 can induce both apoptosis and epigenetic/differentiation effects. Indeed, decitabine was originally developed as a DNA-damaging agent for cytotoxic therapy¹²⁷. Therefore, doses to treat AML were escalated to maximum tolerated levels in traditional phase 1 studies¹²⁸. Although the regimen in common use to treat MDS has de-escalated doses with an epigenetic mechanism of action in mind^129;130, therapy continues to resemble pulse-cycled cytotoxic therapy, and the potential or actual cytotoxicity of current regimens has resulted in controversy regarding the relative importance of differentiation to the clinical mechanism of action⁹⁸. Although cytotoxicity can contribute to tumor kill in vitro and in vivo, cytotoxicity also impairs treatment eligibility, tolerance and feasible exposure, and destroys normal hematopoietic stem cells required for relief of cytopenia and durable remission of myeloid cancers. Furthermore, mutagenicity and micro-environmental insult from anti-metabolite actions can potentially accelerate malignant evolution and resistance^131;132.

There is ample pre-clinical evidence that decitabine, at non-cytotoxic but DNMT1 depleting concentrations, can induce cancer cell cycle exit by differentiation pathways^14-16;27. Indeed, in the earliest cell biology studies the in vitro differentiation modifying effects of decitabine were most potent at low, non-cytotoxic concentrations⁹². Therefore, a DNMT1 depleting, but not necessarily cytotoxic dose, administered frequently, should be safer and more effective than a higher, cytotoxic dose administered infrequently, since exposure timings and distribution are critical considerations for S-phase specific depletion of DNMT1¹⁰¹. Furthermore, maximizing cell cycle exit by epigenetic-differentiation would offer a true p53/p16-independent alternative or complement to conventional apoptosis-based therapy. We are currently testing this approach in a National Institutes of Health sponsored clinical trial in MDS, administering decitabine 0.1-0.2 mg/kg (3.5-7 mg/m²) (a dose lower than in previous studies^130;133-135, since even these low doses are sufficient to deplete DNMT1), by the subcutaneous route (to avoid high peak drug levels that can cause apoptosis), administered from 1-3X/week (to produce greater exposure than with previous clinical trials, and to distribute exposure and capture MDS cells entering cell cycle asynchronously at different points in time). This type of decitabine dose and schedule has been used to treat non-malignant disease^103;136, demonstrating its clinical safety and noncytotoxic epigenetic and differentiation modifying actions^103;136.

In parallel, we are developing an approach to oral therapy that combines decitabine with tetrahydrouridine, an inhibitor of cytidine deaminase, the enzyme which rapidly metabolizes decitabine in vivo (because of cytidine deaminase activity, the in vivo half-life of decitabine is <20 minutes¹³⁷ in contrast to an in vitro half-life at 37°C of approximately 9 hours). Oral administration of tetrahydrouridine-decitabine is more likely to produce the desired pharmacologic profile of low peak drug levels (~0.005-0.2 μM) to avoid cytotoxicity, but extended half-life (hours rather than minutes) to increase depletion of DNMT1, and could decrease inter-individual variability in decitabine pharmacokinetics that arises from pharmacogenetic variation in cytidine deamianse (Lavelle et al., submitted manuscript). In addition, an oral formulation will provide major cost, logistical and accessibility advantages for long-term outpatient therapy. Finally, inhibition of cytidine deaminase may address a mechanism of cancer cell resistance, and sanctuary from the effects of decitabine that can occur in organs that express high levels of cytidine deaminase, for example, the liver and intestines^138-146.

Currently, decitabine (~10% of the related compound 5-azacytidine is converted to decitabine by ribonucleotide reductase in vivo) is the only drug that can be repurposed clinically for non-cytotoxic, p53/p16-independent epigenetic-differentiation therapy (since it is difficult to separate the chromatin-modifying effects of histone deacetylase inhibitor drugs from cytotoxicity). Other compounds that can inhibit components of the chromatin modifying network without inducing apoptosis have been identified or will be identified^69;147. The maturation and epigenetic context of cancer cells suggests that these agents can also be developed for the purpose of p53/p16-independent, normal stem cell sparing epigenetic-differentiation oncotherapy.

Conclusion

Conventional therapy focuses on inducing irreversible cell cycle exit in cancer cells by DNA damage or metabolic insult that activates apoptosis pathways (cytotoxicity). This approach to therapy has a major limitation: malignant cells frequently mutate or delete key apoptosis genes. Hence, the goal of activating apoptosis genes may be futile, since these genes may not be present, yet treatment destroys normal stem cells which have intact apoptosis pathways. This manifests clinically as short-term improvement but frequent relapse with more aggressive, apoptosis-resistant disease. Unlike apoptosis genes, genes that mediate cell cycle exit by differentiation are typically present, but are aberrantly repressed by epigenetic means. Furthermore, cancer cells express high levels of lineage-specifying transcription factors. Because of this epigenetic and maturation context of cancer cells, non-cytotoxic antagonism of chromatin-modifying enzymes that mediate transcription repression terminates proliferation by differentiation^104-107. Normal stem cells, which do not express high levels of lineage-specifying transcription factors, are spared^109-118. Using differentiation to terminate cancer cell proliferation was first described more than 50 years ago^119-121, but is not a major component of current clinical therapy. The biological and translational insights that have accrued in the intervening five decades renew the importance and promise of this approach, not just for the myeloid malignancies, but for cancer in general.