Enhancing the Cytotoxic Effects of PARP Inhibitors with DNA Demethylating Agents – A Potential Therapy for Cancer

SUMMARY

Poly (ADP-ribose) polymerase inhibitors (PARPis) are clinically effective predominantly for BRCA-mutant tumors. We introduce a mechanism-based strategy to enhance PARPi efficacy based on DNA damage-related binding between DNA methyltransferases (DNMTs) and PARP1. In AML and breast cancer cells, DNMT inhibitors (DNMTis) alone covalently bind DNMTs into DNA and increase PARP1 tightly bound into chromatin. Low doses of DNMTis plus PARPis, versus each drug alone, increase PARPi efficacy, increasing amplitude and retention of PARP1 directly at laser-induced DNA damage sites. This correlates with increased DNA damage, synergistic tumor cytotoxicity, blunting of self-renewal and strong anti-tumor responses in unfavorable AML subtypes and BRCA wild-type breast cancer cells. Our combinatorial approach introduces a strategy to enhance efficacy of PARPis in treating cancer.

Graphical Abstract

eTOC Blurb

Using a mechanism-based strategy to enhance PARP1 inhibitor efficacy based on DNA damage-induced interactions, Muvarak et al. show that combining DNMT and PARP1 inhibitors results in significant increases in anti-tumor effects in vivo, suggesting an approach for cancers not responsive to either inhibitor alone.

INTRODUCTION

PARPis are a very exciting spectrum of drugs currently used in the clinic in cancer management. Overall, the therapy works through interfering with how PARP functions in allowing cancer cells to survive ongoing DNA damage. In this regard, PARP1 is an abundant nuclear protein that senses and contributes to repair of DNA single-strand breaks (SSBs) (De Vos et al., 2012). PARP1 is also active in repair of DNA double-strand breaks (DSBs) (Audebert et al., 2004), working through catalyzing poly-ADP-ribosylation of itself, histones and other target proteins (Gibson and Kraus, 2012). In particular, PARP1 is involved in a highly error-prone form of DSB repair, alternative non-homologous end-joining (ALT NHEJ) (Nussenzweig and Nussenzweig, 2007; Rassool and Tomkinson, 2010). Both expression of PARP1 and ALT NHEJ activity are increased in breast cancer and leukemia cells, compared with non-tumorigenic counterparts (Ha et al., 2014; Tobin et al., 2012a; Tobin et al., 2012b). Blocking the catalytic activity of PARP1 has been shown to inhibit BER repair, resulting in accumulation of SSBs, as well as DSBs, during replication (Mariano et al., 2015), and this damage in turn activates homologous recombination (HR) (Chevanne et al., 2010). Recent studies have shown that disruptions of any HR-related pathway (Mateo et al., 2015), such as by BRCA mutations, and disruption of Fanconi Anemia (FA) (D’Andrea, 2010) and ATM genes (Murai et al., 2012), can predict sensitivity and tumor cytotoxicity to PARP1 inhibition by small molecule inhibitors. Additionally, blocking PARP1 in combination with another ALT NHEJ protein, DNA ligase IIIα, in multiple cancers results in significant reduction of ALT NHEJ activity, leading to increased cytotoxic DSBs and cell death (Ceccaldi et al., 2015; Ha et al., 2014; Tobin et al., 2012a; Tobin et al., 2012b).

Particularly important with respect to the future of PARPis in cancer therapy are the recent advances in understanding how and where, at a molecular level, these agents best work as cytotoxic agents, and recent progress in developing the best reagents. Substantial efficacy has been shown with clinically available PARPis, especially for treatment of breast and ovarian cancers in patients with hereditary deletions of the HR BRCA1/2 genes. Cancers presenting with such mutations represent 5–10% of all triple-negative breast cancers (estrogen, progesterone and HER2 receptor negative breast cancers ;TNBCs) (Bryant et al., 2005; Farmer et al., 2005; Guastafierro et al., 2008; Pedersen-Bjergaard et al., 2006). However, responses to PARPi therapy, even in BRCA-mutant breast cancers, have not been highly durable. Furthermore, PARPis have failed to show impressive clinical benefit for patients with sporadic TNBCs (Guha, 2011) and/or other cancers, suggesting the necessity for developing new strategies to maximize the efficacy for using these agents, which is the focus of the present paper.

PARP-DNA complexing by PARPi is proposed to be a direct interaction between DNA and PARP1 via the DNA-binding site of the latter (Horton and Wilson, 2013; Murai et al., 2014). A key for the above need for improving PARPi therapy is the recent development of new PARPis with much increased potency, such as BMN 673 (talazoparib) (Shen et al., 2015). The primary cytotoxic effect of PARPis has been correlated with trapping of cytotoxic DNA-PARP1 complexes at sites of DNA damage (Murai et al., 2012). Biochemically, PARP1/2 are trapped at 5’-dRP lesions generated during BER steps under PARPi treatment (Murai et al., 2012). Moreover, and with particular importance to our present work, increases in the amplitude and duration of this trapping appear to be key parameters for efficacy of PARPis. This is well reflected in the fact that up to 100-fold greater inhibitory activity is associated with the increased ability of the new and strongest PARPi, talazoparib, to trap DNA-PARP1 complexes, compared to weaker PARPis such as veliparib (ABT888) (Shen et al., 2015).

DNA methyltransferase inhibitors (DNMTis) are approved by the Food and drug administration (FDA) for treatment of myelodysplastic syndromes (MDS) (Kantarjian et al., 2006) and are used for treatment of AML as well (Carbone et al., 2006; Issa et al., 2004; Zampieri et al., 2012). These agents, including 5-azacytidine (AZA) and decitabine (DAC), are potent inhibitors of DNA methylation and DNMTs [reviewed in (Tsai et al., 2012)]. Reversing the gene expression changes associated with DNA methylation abnormalities in cancer is one proposed mechanism for the clinical efficacy of DNMTis (Baylin and Jones, 2011; Issa, 2007). Moreover, DNMTis become incorporated into replicating DNA as an altered cytosine base and covalently bind DNMTs, creating a cytotoxic DNMT-DNA complex while simultaneously triggering the degradation of soluble DNMT’s (Patel et al., 2010). In the current study, we explore the use of these two classes of drugs together as a potential therapy for TNBCs and AML.

RESULTS

PARP1 interacts with DNMTs in a damage-induced, enlarging protein complex

PARP1 interacts non-covalently with DNMT1 (Caiafa et al., 2009; Reale et al., 2005), suggesting that epigenetic and DNA repair pathways are linked mechanistically. Our present study was suggested by an initial extension of this possibility. Using human embryonic and colon carcinoma cancer cells, we previously reported that within 30 min exposure to hydrogen peroxide (H₂O₂), an enlarging protein complex is induced which includes DNMTs 1 and 3B, histone deacetylases, and polycomb silencing proteins (O’Hagan et al., 2011). This damage also induces tightening of the complex components to chromatin, especially DNMT1 localization to sites of DNA damage. In a prelude to the present work, using mass spectrometry analysis and gel filtration studies of proteins associating with immunoprecipitated DNMTs 1 and 3B, we found that PARP1 is a constituent of this above complex, including when it enlarges during DNA damage (Figure S1A and S1B).

Consistent with the above studies, using co-immunoprecipitation (co-IP) analyses of chromatin extracted from untreated TNBC (MDA-MB-231) cells and AML cells (MV411), we found that PARP1 also interacts with DNMT1 before and after DNA damage (Figures 1A and 1B). Moreover, following treatment with the DNA damaging drug camptothecin (1 µM CPT) or the alkylating agent methyl methanesulfonate (0.01% MMS), larger amounts of DNMT1 co-IP with PARP1 (Figure 1B). These data suggest that PARP1-DNMT1-containing complexes are recruited to sites of DNA damage. Importantly, in response to DNA damage [1 µM CPT or 10 Gy irradiation (IR)], PARP1 co-localizes with the DSB marker γH2AX in immunostaining studies of chromatin fibers from MV411 treated cells (Figure S1C).

Figure 1. PARP1 interacts with DNMT1 and DNMTi-PARPi combination binds PARP1 tightly in chromatin.

Co-immunoprecipitation of DNMT1 or PARP1 in chromatin in (A) MDA-MB-231 or (B) MV411 cells untreated or treated with camptothecin (CPT, 1 µM) or methyl methanesulphate (MMS, 0.01%). Left panels: Western blotting for input proteins; right panels: co-IPs. IgG served as negative control. Note: in (A), lower panel (PARP1 IP), the DNMT1 band was cropped and moved next to IgG lane from the same gel. (C–F) PARP1, DNMT1 and histone H3 (loading control) in chromatin (nuclear insoluble) fractions from untreated and drug-treated cells. Upper panels, representative blots; middle/lower panels, quantitation of PARP1 and/or DNMT1 trapping. MOLM14 cells were collected post 72 hr treatment with (C) BMN 673- 1, 2.5, 5 nM or (D) DAC- 1, 2.5, 5 nM. (E) Combination treatments in MOLM14 with DAC and BMN 673 with indicated doses for 72 hr. (F) Combination treatment in MDA-MB-231 with AZA and BMN 673 with indicated doses for 72 hr followed by 4 Gy IR. Cells were collected 4 hr post IR for Western blotting. Experiments in triplicates are represented, mean ± SD (represented by error bars), ^# p<0.05 by t-test, single treatments vs. control, * p<0.05, by t-test, combination treatments vs. control and single treatments. See also Figures S1 and, S2.

Combined DNMTis and PARPis increase tight binding of PARP1 in chromatin

PARPis, especially talazoparib, have been shown to not only catalytically inhibit PARP1 activity but also trap PARP1 in chromatin, forming cytotoxic DNA-PARP1 complexes (Murai et al., 2012; Shen et al., 2015). In our present study, talazoparib at very low nM concentrations (1, 2.5 or 5 nM) also traps PARP1 in chromatin extracts (Figure 1C) to the same extent as the weaker PARPi ABT888 at much higher concentrations (500 nM) (Figure S2A). As previously introduced, DNMTis by their obligatory mechanisms of action are incorporated into replicating DNA where they covalently bind DNMTs (Lyko and Brown, 2005). In agreement with previous studies showing that DNMTi-induced damage is repaired by BER (Covey et al., 1986), and that BER repair intermediates (abasic sites) are covalently bound by PARP1 (Prasad et al., 2014), we also find that the DNMTis retain PARP1 in chromatin fractions (Figures 1D and S2A). Consequently, in AML cells, combinations of low-dose DNMTis and the PARPi talazoparib, (given concurrently) lead to enhanced PARP1 bound in chromatin, as compared to either drug alone (Figure 1E). Combinations of low-dose DNMTis and talazoparib given sequentially also lead to enhanced tight binding of PARP1 in chromatin (Figure S2A). In TNBC cells, significant tight binding of PARP1 is seen with combination treatment (given concurrently), versus each agent alone, following 72 hr daily drug treatments and 4 hr post IR (4 Gy) (Figures 1F, S2B, and S2C). To ensure that the increased binding of PARP1 in chromatin post combination treatments is not due to global increase in PARP1 protein expression, we examined whole cell extracts from MOLM14 and MV411 cells, and observed no change in PARP1 protein expression in combination-treated cells relative to untreated or single drug-treated cells (Figure S2D). Lastly, the purity of subcellular fractionation was confirmed by immunoblotting for β-tubulin (cytoplasmic), Lamin B (nuclear, soluble), and H3 (chromatin, insoluble) (Figure S2E).

Combining DNMTis and PARPis increases retention of PARP1 and DNMT1 at laser-induced DNA damage sites

A key question with regard to our mechanistic hypothesis is whether the above chromatin data translate into increased tight binding of DNMT1 and PARP1 by the DNMTis and PARPis directly at DNA damage sites. We find by multiple criteria that the low-dose combination increases PARP directly at induced DNA damage sites.

We performed these studies using laser micro-irradiation (Rogakou et al., 1999) at 365 nm wavelength known to create both SSBs and DSBs in the DNA at the cut sites (Lan et al., 2004; Rulten et al., 2008), followed by immunofluorescence to detect recruitment of DNMT1 or PARP1 (Ha et al., 2011; Haince et al., 2008). These studies require adherent cells and so we used TNBC MDA-MB-231 cells.

First, given that laser-induced DNA damage causes both SSBs and DSBs, we used DSB marker γH2AX to delineate DNA damage sites in nuclei and show that DNMT1 and PARP1 can be seen localizing at 1, 5, 15 and 30 min after induction, in the absence of DNMTi or PARPi treatment, as determined by the mean intensity of immunofluorescence for co-localizations (Figure S3A, S3B, S3D, and S3E). Similar results are seen by counting the number of cells in which these proteins are present at the induced DNA damage sites (Figure S3C and S3F). The localization of each protein to DNA damage sites increases significantly with either AZA or BMN treatment alone, starting at 1 min and persisting at all time points.

Second, the above results are paralleled by the finding that DNMT1 and PARP1 co-localize with one another at induced DNA damage sites (Figure 2A). In untreated cells, this co-localization is seen in just over 60% of cells at 30 min post laser induction, but the signals for both proteins are diminished in all cells after 30 min (Figure 2A). While treatment with low doses of AZA and BMN alone only slightly increases the percentage of cells with co-localization to ~ 70 % at 30 min, 10–15% of cells show co-localization at 3 and 6 hr after laser cuts for each drug treatment that is significant with respect to untreated controls (Figures 2A, right panel, S4A and S4B). When both drugs are combined, co-localization of DNMT1 and PARP1 increases to 80% of cells at 30 min and remains in over 60% at 3 and 6 hr post laser-induced DNA damage (Figures 2A–E). These results thus indicate an increase in retention of both proteins at DNA damage sites, as compared to untreated cells and cells treated with each drug alone, and again the most dramatic point is the high retention at 3 and 6 hr post the induced laser damage. When the image intensity for each protein is quantitated, this retention is most apparent for DNMT1 (Figure 2D). For PARP1, combined treatment results in a particularly great increase at 30 min with continued significant retention at 3 and 6 hr (Figure 2E). Finally, the above laser DNA damage site binding dynamics, especially retention at the late time points (3 and 6 hr post laser damage), rely on an interdependency between levels of DNMT1 and PARP1 in the cells. Thus, shRNA depletion of PARP1 (Figure S4C) sharply reduces DNMT1 at the induced laser cut sites at all studied time points with the combination treatment (Fig 2A–E). Similarly, with lentiviral shRNA knockdown of DNMT1 (Figure S4D), PARP1 levels are reduced back to basal levels at laser cut sites, which is especially noted with the combination treatment at 3 and 6 hr post laser induction (Fig 2A–E).

Figure 2. Combination of AZA and BMN 673 increases the retention of DNMT1 and PARP1 at laser-induced DNA damage sites and increases cytotoxic DSBs.

(A) MDA-MB-231 cells were treated with AZA (150 nM) and BMN 673 (10 nM) alone or in combination for 72 hr, followed by laser-induced DNA damage, then fixed at the indicated time points. Localization of DNMT1 and PARP1 to DNA damage sites was examined by immunofluorescence staining. Left panel, representative images at indicated time points. Right panel, graph of percentage cells with co-localization of PARP1 and DNMT1 at damage sites with single and combination drug treatments. Experiments in triplicates are represented. 30 cells per time point and treatment were analyzed for each experiment. * p<0.05, by t-test. Scale bar (upper left white bar) =10µm. (B,C) Graphs represent percentages of cells with γH2AX micro-irradiation tracts that have visible accumulation of DNMT1 (B) or PARP1 (C) co-localizing with γH2AX. Experiments in triplicates are represented. (D,E) Quantitation of the mean intensities of DNMT1 or PARP1 co-localized to γH2AX tracts that were quantified in single cells at the indicated time points after laser treatment. Experiments in triplicates are represented. 30 cells per time point and treatment were analyzed for each experiment * p<0.05, by t-test.

(F,G) Graphs of γH2AX foci examined by immunofluorescence in untreated, AZA, BMN 673 and combination treated MDA-MB-231 cells (F) at indicated time points after IR (4 Gy). MOLM14 AML cells (G) untreated or treated with DAC, BMN 673 or the drug combination. Scale bars (lower left white bar) = 5 µm. Experiments in triplicate are represented; percent mean ± SD (represented by error bars), * p<0.05, ** p<0.01. See also Figures S3 and S4.

Combination drug administration induces increased frequency of DSBs and synergistic cytotoxicity

The above increased amplitude and retention of PARP1 and DNMT1 co-localization at laser-induced sites of DNA damage in TNBC MDA-MB-231 cells also correlate with significantly more cytotoxic DSBs, compared with single talazoparib or AZA treatments, respectively. Increased γH2AX foci, as quantitated by immunostaining, are seen with the combination treatments, compared with single treatments and controls, at 4 hr and 24 hr after DSB induction by IR (4 Gy) (Figure 2F). Also, in MOLM14 AML cells, combination dosing with DAC (5 nM) and BMN 673 (5 nM) induces increases in γH2AX foci, as compared with single drug treatments (Figure 2G). In a key association with the above data, both MOLM14 AML and MDA-MB-231 TNBC cells treated with the combination of the two drugs exhibit synergistic cytotoxicity, as assessed by the CalcuSyn model (Chou, 1991) (Figures 3A–D, top panels). Significant decrease in survival was also observed with combination drug treatments (AZA and BMN 673 or DAC and BMN 673), compared with either drug alone (Figures 3A–D, middle and lower panels).

Figure 3. DNMTi and PARPi act synergistically to produce cytotoxicity.

(A, B) MDA-MB-231 cells treated with (A) AZA (50–400 nM) and BMN 673 (2.5–20 nM), or (B) DAC (2.5–20 nM) and BMN 673 (2.5–20 nM). (C, D) MOLM14 cells treated with (C) AZA (50–400 nM) and BMN 673 (2.5–20 nM), or (D) DAC (2.5–20 nM) and BMN 673 (2.5–20 nM). Cells were treated daily for 7 days with the indicated drug combinations, followed by MTS assay to determine cytotoxicity. Upper panel, x-axis; Fa = fraction of cells affected; y-axis = combination index (CI). Combinations below red line are synergistic. Middle and bottom panels, survival of cells treated with DNMTi (DAC or AZA) or BMN 673 alone or in combination. Data are from 3 independent experiments, mean ± SD (represented by error bars). * p<0.05, ** p<0.01, by t-test, combination treatments vs single treatments.

DNMTis combined with PARPis decrease clonogenicity in cultured TNBC and AML cells

The data in the section immediately above translate to our drug combination inducing significant decreases in clonogenicity of TNBC cells (Figure 4A). As expected, TNBC cells mutant for BRCA1 (SUM149PT) are sensitive to talazoparib (10 nM) alone, but these cells are even more sensitive to a combination of talazoparib (1 and 10 nM) and AZA (150 nM). Interestingly, another BRCA mutant cell line, HCC1937, exhibits significant decreases in colony formation (p<0.05) only when treated with the drug combination (Figure 4A). Finally, and importantly, MDA-MB-231 and MDA-MB-468 breast cancer cells with intact BRCA genes are significantly (p<0.05) sensitive to the combination of AZA (150nM) plus talazoparib (10nM), versus either drug alone. In contrast to all of the above results, non-tumorigenic MCF10A cells were unaffected by the above drug treatments (Figure 4A). This is not surprising, since we find that MCF10A cells, as compared to breast cancer cell lines, have lower levels of PARP1 (Figure S5A), DNMT1 (Figure S5B), and PARylation (Figure S5C). The increase in PARP levels and PARylation are reported to correlate with potency of PARPis (Benafif and Hall, 2015). The relatively higher levels of DNMT1 in breast cancer cells are likely due to increased protein stability (Agoston et al., 2005).

Figure 4. DNMTis in combination with PARPis decreases clonogenicity.

(A) Colony formation of non-tumorigenic MCF10A and TNBCs (SUM149PT, HCC1937, MDA-MB-231, MDA-MB-468). Cells were treated daily for 7 days with the indicated drug treatments. Results expressed as % survival relative to mock (Ctrl) treated groups. (B) Colony formation of AML (MOLM-14, MV411, KASUMI, MOLM13) cell lines and (C) primary AML cells untreated or treated with DAC (10–20 nM), AZA (50–100 nM), BMN 673 (1–5 nM) or combination. Cell lines were treated daily for 72 hr with indicated doses of DNMTis, followed by a 24 hr recovery period without DNMTis, then plated in methylcellulose with or without the indicated doses of BMN 673 and incubated for 10–14 days. Primary AML samples were treated as AML cell lines, except they were plated in Methocult for 14 days. Samples #081, #090, #107, #110 were treated with AZA, and samples #29, #34, #086, #092, #109 were treated with DAC. Experiments performed in triplicates; mean ± SD (represented by error bars). * p<0.05 by t-test, combination vs control and single treatments. See also Figures S5 and S6.

Because the methylcellulose colony formation assay for AML cells (non-adherent) precluded continuous drug treatments as for adherent TNBCs, AML cells were pretreated with the DNMTi DAC and plated in medium containing talazoparib. In a similar fashion to the above breast cancer cell lines, the AML cell lines MOLM14 and especially KASUMI-1 showed profound reductions in colony formation following treatment with the drug combinations (Figure 4B). MV411 and MOLM13 cells are also sensitive to the combination treatments but to a lesser degree (Figure 4B). KASUMI-1 cells treated with 2.5 nM DAC and subsequently plated in methylcellulose in the presence of 2.5 nM talazoparib produced 60% fewer colonies (Figure 4B), while FLT3/ITD-positive MOLM14 cells showed a similar decrease at slightly higher drug concentrations (Figure 4B). When treated with higher doses of the weaker PARPi ABT888 (500 nM) alone, as previously described (Ha et al., 2014), FLT3/ITD-positive MOLM14 and MV411 cells exhibited a significant reduction in colony numbers, compared to control-treated cells (Figures S6B and S6C). However, treatment of all AML cell lines with DAC followed by ABT888 led to significantly (p<0.05) fewer colonies (Figures S6A–C). Notably, the enhanced PARP binding to chromatin was also seen with the higher doses of the weaker PARPi ABT888 in these studies (Figure S2A).

The above combinatorial effects are not limited to cell lines, and can also be seen in primary AML cells from patients (Table 1). In 8 of 13 patient samples (61.5%) (#15, #29, #34, #081, #086, #092, #109, #110), colony formation in methylcellulose is significantly reduced when 10–20 nM DAC is used prior to talazoparib (1–10 nM) treatment, as compared to treatment with each drug alone (Figures 4C and S6D–G). Studies with 50–100 nM AZA followed by talazoparib gave similar results in 2 of 5 (40%) patient samples (#081, #086, #090, #092, #107) (Figures 4C and S6H). Similarly, 3 of 6 (50%) patient samples (#9, #15, #29, #34, #16, #30) treated with DAC followed by higher doses of the ABT888 compound (500 nM) showed significant (p<0.05) reductions in colonies (Figure S6I).

Table 1.

Cytogenetic and Molecular features of AML samples treated with DNMTis in combination with PARPis

PT#	Samplea	Sexb	BM Blastsc	PB Blastsc	Cytogenetic Findingsd	Categorye	Response to DAC/AZA +ABT/BMNf
9	BM	F	-	-	49,XX,11,+21,+3mar[11]/49,XX,i(11)(q10),+3mar[3}	Complex	NO
15	BM	M	-	-	ND	FLT3/ITD	YES
16	BM	F	35	21	46,XX	Unknown	NO
29	PB	F	-	95	46,XX	FLT3/ITD	YES
30	BM	F	-	5	46,XX	Neither	NO
34	PB	-	-	-	ND	FLT3/ITD	YES
081	BM	M	58	0	45–48,XY,-4,+11,+14,-17,-18,+2–4mar	Complex	YES
086	BM	M	86	63	46XY	FLT3/ITD	YES
090	PB	M	-	79	46XY,+8	Neither	NO
092	BM/PB	M	78	94	46–51,XY,+4,+8,+ 3mar	Complex	YES
107	BM	F	92	89	46,XX	FLT3/ITD	NO
109	BM	F	25	5	41–44,XX,-3,del(3)(p21),-5,del(7)(q22q36),-12,-15,- 17,-20,+1–3mar	Complex	YES
110	BM	F	88	91	47,XX,+11	FLT3/ITD	YES

^aBM= Bone Marrow; PB= Peripheral Blood

^bF= female; M= male; hyphen = information unavailable

^cNumbers represent percentages

^dND= Not determined

^eComplex refers to complex karyotype

^fResponse to combination (yes/no) is relative to single drug treatment

TNBC and AML xenografts are sensitive to AZA and BMN 673 combination treatment

As further evidence for the important translational implications of the present studies, in in vivo therapy models in immune-deficient mice (Figures S7A and S7B), our low-dose combinations of DNMTis and PARPis yielded potent anti-tumor responses. First, for BRCA mutant SUM149PT xenografts (Figures 5A–C), as expected, mice treated with talazoparib, (0.3 mg/kg BMN 673) alone had a significant reduction in tumor burden after day 28 (denoted by asterisk and arrow) versus those treated with vehicle or AZA (0.5 mg/kg) alone (Figure 5A). However, the combination treatment yielded a further tumor burden reduction and a significant (p<0.05) survival advantage compared to Ctrl (vehicle), AZA (0.5 mg/kg) or talazoparib, (0.3 mg/kg) alone (Figure 5B). Importantly, the treatment was well tolerated, as no significant weight loss was observed throughout the study (Figure 5C). Most importantly, combined treatment of MDA-MB-231 TNBC cells (wild-type BRCA1) with AZA (0.5 mg/kg) and talazoparib at 0.2 mg/kg (data not shown) or 0.3 mg/kg (Figures 5D–F) resulted in marked reduction in tumor volumes from day 25 onward (denoted by asterisk and arrow), compared with either drug alone (Figure 5D). The combination group also showed a significant (p<0.05) increase in survival (Figure 5E). This survival result is actually conservative, in that tumors in mice treated with both agents developed severe ulcerations before reaching 1500 mm³ and mice were thus euthanized even though they were doing well. Ulcerations were seen only for a few mice in the other groups and at larger tumor volumes (Figure 5E, see triangles). Drug doses were well tolerated as evidenced by stable body weights in all groups (Figure 5F). Interestingly we could still detect tight binding of PARP1 to chromatin extracted from the 1500 mm3 - size tumors from euthanized mice (from all treatment groups) at the end of the study (n=2 per group) (Figure 5G). This suggests a measure of drug effect even at these increased tumor burdens.

Figure 5. Anti-tumorigenic effect in xenograft models of TNBCs.

(A–C) SUM149PT xenograft model. (A) Tumor volume (mm³) measurements (mean ± SD) in vehicle and drug treated groups with time for the duration of the experiment and euthanasia. Significant difference in tumor volume from Day 28 post treatment to the end of study is denoted by the asterisk and arrow. (B) Graphs of % survival with time (until tumor volume reached 1500 mm³) for the duration of the experiment. (C) Graph of mean ± SD % body weight change vs time for the duration of the experiment and euthanasia.

(D–G) MDA-MB-231 xenograft model. (D) Graph of tumor volume (mm³) measurements (mean ± SD) in vehicle and drug treated groups with time for the duration of the experiment and euthanasia. Significant difference in tumor volume from Day 24 post treatment to the end of study is denoted by the asterisk and arrow. (E) Graph of % survival with time (until tumor volume reached 1500 mm³ or showed necrosis) for the duration of the experiment. Triangles denote mice removed from study due to necrosis of tumor. (F) Graph of mean ± SD % body weight change vs time for the duration of the experiment and euthanasia. (G) Tight binding of PARP1 into chromatin in tumors from euthanized MDA-MB-231 xenograft mice from control and treatment groups. Upper panel, PARP1 levels in chromatin. H3: loading control. Lower panel, quantitation of PARP1 levels. In all of the above, error bars represent the SD. See also Figure S7.

Our second major translational findings are the very robust responses of AML MV411 and MOLM14 cells in an in vivo therapy model using cells with luciferase labeling. Drug administration began 3 days after intravenous cell injection, when leukemia cells engrafted in the bone marrow, as measured by photon intensity. All mice in all treatment groups had equal tumor burdens at the start of therapy (Figures 6A and 7A). For MV411 cells, the combination of low-dose talazoparib and AZA resulted in a significant (p<0.05) decrease in tumor burden up to day 30 following start of treatment, compared with either drug alone (Figures 6A and 6B). Long-term survival determination was compromised due to a characteristic of the model, hind limb paralysis related to leukemia burden, which requires euthanasia (O’Farrell et al., 2003). Using paralysis as an endpoint (14/20 mice), the mice treated with the combination had marked delay in onset of paralysis compared to mice treated with vehicle or either drug alone. While leukemia burden was still present in all groups on day 30 post treatment, at the time of sacrifice, spleen weight was significantly lower in the combination treatment group, compared to control (Figure 6C), and blast cell numbers in the peripheral blood were significantly diminished with the treatment combination compared with either drug treatment alone (Figure 6D). As in the breast cancer studies described earlier, treatments were well tolerated, as gauged by body weight (Figure 6E).

Figure 6. Anti-leukemic effect of BMN 673 and AZA in a systemic IV leukemia model MV411.

(A) Bioluminescence measurements (mean ± SD) of photon intensity showing relative leukemia burden. (B) Graph of photon intensity with time post drug treatment up to 47 days and euthanasia. (C) Graph of spleen size (mm³) in different treatment groups post euthanasia. Measurements represent mean ± SD. Representative pictures of spleens are placed above graph for each group. (D) Graph of % blasts in the peripheral blood of mice from different treatment groups post euthanasia. (E) % body weight change vs time. Measurements represent mean ± SD. In all of the above, error bars represent the SD. See also Figure S7.

FIGURE 7. Anti-leukemic effect of BMN 673 and AZA in a systemic IV leukemia model MOLM14.

(A) Bioluminescence measurements of photon intensity showing leukemia burden for duration of the experiment. Measurements represent mean ± SD. (B) Graph of photon intensity with time post drug treatment up to 61 days and euthanasia. Measurements represent mean ± SD (C) Graph of % survival with time post therapy. (D) % body weight change vs time. Measurements represent mean ± SD. In all of the above, error bars represent the SD. See also Figures S7 and S8.

An even better treatment outcome is observed with the AML cell line MOLM14 (Figures 7A–D and S8). Of note, in methylcellulose clonogenic assays (Figure 4B) MV411 was less sensitive to the two drugs combined compared with MOLM14, and this is also reflected in vivo. Indeed, the combination groups showed not only significantly reduced tumor burden as measured by photon intensity, compared with vehicle-treated mice and mice treated with AZA alone, but also increased survival for both 0.3 mg/kg (data not shown) and 0.1 mg/kg doses of talazoparib paired with 0.5 mg/kg of AZA (Figures 7A–C). In this model, the leukemia cells home preferentially to the bone marrow, with little evidence of disease in peripheral blood and spleen, as previously described (Damjanac et al., 2009). While mice treated with 0.3 mg/kg and 0.5 mg/kg (data not shown) BMN 673 in the drug combination lost weight, mice treated with 0.1 mg/kg in the combination gained weight during the study, demonstrating that this dosing regimen is again well tolerated (Figure 7D). With this combination dose at 61 days there was no hind limb paralysis, while mock-treated and mice treated with BMN 673 alone were all euthanized for this effect by 26 and 34 days, respectively, and all mice treated with AZA alone started developing paralysis by day 45, with the last mouse euthanized at day 61 (Figure 7C).

DISCUSSION

While substantial efficacy has been shown with clinically available PARPis in treatment of breast and/or ovarian cancers in patients with hereditary deletions of BRCA1/2 (Farmer et al., 2005), the high promise of these drugs has not yet been realized for non-BRCA1/2-defective cancers (Westman et al., 2013). This has spurred the development of a generation of potent PARPis, such as talazoparib, that are in late-stage clinical trials (Shen et al., 2013). As previously noted, these most potent PARPis induce cytotoxicity proportional to the amplitude and duration of PARP1 trapping in chromatin (Murai et al., 2012; Shen et al., 2015). We now take advantage of the ability of DNMTis to increase tight binding of PARP1 into chromatin, based on the following data: 1) DNMTs and PARP1 interact (Reale et al., 2005); 2) DNMTis covalently bind DNMTs into DNA (Ghoshal and Bai, 2007) and this can occur at sites of DNA damage (O’Hagan et al., 2011) and thus can enhance PARP1 recruitment and tight binding to chromatin; and 3) These interactions allow DNMTis to be combined with potent PARPis such as talazoparib to increase tight binding of PARP1 in chromatin, increase the presence and retention of PARP1 and DNMT1 at DNA damage sites, and enhance cytotoxic effects in TNBC and AML cells in vitro and in vivo. These dynamics allow design of a combinatorial treatment strategy with the potential for clinical efficacy in treating patients with TNBCs with intact BRCA genes and poor-prognosis AML patients.

With special reference to AML therapy, there have been over three decades of clinical efforts to improve outcomes in this disease by increasing the intensity of chemotherapy (Pachauri et al., 2012). However, patients with unfavorable AML subtypes or with refractory or relapsed AML continue to do poorly, and less than 10% achieve long-term survival (Kumar, 2011). DNMTis are frequently used to treat AML patients unlikely to respond to conventional chemotherapy (Fenaux, 2005) following their FDA approval for the treatment of MDS (Ghoshal and Bai, 2007). Most recently, SGI-110, a pro-drug for DAC, has been developed as a novel potent DNMTi, with a response rate of 55% in untreated AML patients unfit for intensive chemotherapy (Paccosi et al., 2012), but only 16% in refractory and relapsed AML patients (Issa et al., 2015). Thus, the efficacy of DNMTis needs to be augmented by combination with other therapies, and PARPis appear to be good candidates. Importantly, PARPis alone have not shown any efficacy in AML, and our data suggest that combining them with DNMTis could give PARPis a place in management of AML and other cancers. In this regard, synergistic tumor cytotoxicity of both drugs at very low doses appears to provide good tolerability in our in vivo pre-clinical models, even though each drug alone can induce bone marrow toxicity (Issa et al., 2004; Kantarjian et al., 2013; Underhill et al., 2011). This is emphasized by the fact that the best survival results for both TNBC and AML occurred when we used the lowest doses of talazoparib and AZA.

In terms of mechanisms underlying the efficacy of our combined treatment paradigm, there are important insights from the present study, and the findings will certainly be expanded in future work. All of our present data are consistent with the key hypothesis underlying our model that the DNMTis and PARPis combine to increase the amplitude and duration of DNMT1 and PARP1 at sites of laser-induced DNA damage that include both SSB’s and DSB’s. It is apparent that these dynamics may well underlie our observed synergistic effects on tumor cell cytotoxicity.

PARP1 is known to be recruited to DNA damage sites and to be covalently trapped by PARPis specifically at SSB’s, and most potently by talazoparib (Murai et al., 2012; Shen et al., 2015). Importantly for our model, DNMTis covalently trap for prolonged periods of time DNMT’s at CpG sites in DNA (Oz et al., 2014). Thus, following treatment with these agents, it would be expected that some of the DNA at our laser induced damage sites, which induces both SSB’s and DSB’s, would include DNA with incorporated DNMTi, and thus, covalently-bound DNMT’s. It is to this DNA bound DNMT’s that we propose interacting PARP1 is bound and would be increased in chromatin and possibly also provide for increased amounts of this latter protein to be formally trapped by PARPis at SSB’s. In fact, as shown by our shRNA studies of both DNMT1 and PARP1 in combination inhibitor treatment with laser damage induction there is an interdependency between DNMT1 and PARP1 required for retention of PARP1 at DNA damage sites at later time points after DNA damage induction. With respect to the above dynamics, several points are key to the interactions between DNMT1 and PARP1 for chromatin binding of each and potential for formal trapping of the latter by PARPis at DNA damage sites. First, we have previously shown that DNMT1 is recruited directly to both sites of induced DSBs (O’Hagan et al., 2008) and sites involved with 8-oxo-guanosine adducts induced by increased cellular ROS (O’Hagan et al., 2011). The latter is particularly key for our present findings in that the DNMT1 recruitment is associated, as part of a large protein complex which enlarges after damage, and with a tightening to chromatin, especially for DNMT1, which is highly salt resistant (O’Hagan et al., 2011). Upstream of this binding are DNA damage proteins including mismatch repair proteins (Ding et al., 2015) and the BER protein OGG1 (O’Hagan et al., 2011). Thus, it is well possible that some of the DNMTi-enhanced binding of PARP1 into chromatin could well be formally trapped at DNA damage sites, including SSB’s, when the DNMTi and PARPi are combined.

With respect to the above possibilities, other mechanistic considerations may apply which may be contributing without being mutually exclusive. Possibly, combination treatment leads, as shown by the timing in the laser induction studies, to a marked delay in repair of DNA damage that could be independent of the above discussed DNMT1 covalent binding to DNA induced by DNMT is and subsequent direct interactions with PARP1. This delayed, impaired, and ongoing damage repair could in and of itself contribute to the increased amplitude and duration of localization of both DNMT1 and PARP at the damage sites. We previously reported that sporadic TNBCs with intact BRCA genes are dependent on an alternative and highly error-prone form of non-homologous end-joining, ALT NHEJ (Tobin et al., 2012a). Retention of PARP1 and DNMT1 at DSBs may prevent access of ALT NHEJ factors to DSB sites, leading to decreased ALT NHEJ repair. In addition to activation of HR (Chevanne et al., 2010), PARPis administered singly also increase classical non-homologous end joining (c-NHEJ) activity through inhibition of Ku protein PARylation and increased DNA-PK phosphorylation (Patel et al., 2011). In contrast, DNMTi treatment alone does not affect c-NHEJ (Moscariello and Iliakis, 2013). Recent reports do suggest that treatment with AZA at much higher doses than employed in our present study induces DNA damage lesions that are recognized by the BER machinery (Orta et al., 2014). Thus, some of the enhanced cytotoxicity induced by our drug combination could stem from disrupting BER of incorporated DNMTis. As we have previously shown (Tsai et al., 2012), and also found here, during the time of administration, DNMTis at the low concentrations employed do not appear over 72 hr to generate significant DNA damage.

As mentioned, none of the mechanisms above may be mutually exclusive, and they may have summation effects. It is clear from our overall results that our combination approach enhances the parameters that facilitate the efficacy of PARPis and the high potential as a therapeutic approach paradigm. It is important to remember that while all of our present studies pinpoint actions of our dual treatment strategy directly at DSB’s, low dose DNMTi treatment also effects DNA methylation changes (Tsai et al, 2012). Downstream responses to these triggering events may also link to DNA damage repair changes and augment the pathways and events studied above to explain our combinatorial drug effects (Tsai et al., 2012). Thus, DNMTi treatment may induce a BRCAness phenotype that contributes to synthetic lethality with PARPis in AML and TNBC cells (Wiegmans et al., 2015).

In summary, our preclinical data in AML cell lines, primary cells, and mouse xenografts, as outlined in this study, suggest the potential for improving the clinical efficacies of both DNMTis and PARPis by combining low doses of both drugs for patients with AML. Moreover, our initial data suggest that the poor-prognosis subgroups of AML, including AML with FLT3/ITD, constituting at least a third of AML cases, are likely to be sensitive to our therapeutic approach. We are already translating the combination paradigm into an upcoming clinical trial for AML. This is designed to test whether low doses of DNMTis (Issa and Kantarjian, 2005a; Issa and Kantarjian, 2005b) and the PARPi talazoparib can be safely combined, and then to test whether this combination therapy shows efficacy for contributing to the management of newly diagnosed AML patients unfit for intensive chemotherapy and patients with relapsed/refractory AML. Our present results in breast cancer cells may also lay the groundwork for similar combination trials in this and other solid tumors.

EXPERIMENTAL PROCEDURES

Cell culture and drugs

Human cell lines were cultured as described in Supplemental Experimental Procedures. Decitabine (DAC, Sigma-Aldrich) was prepared at 10 mM in DMSO. 5-azacytidine (AZA, Sigma-Aldrich) was prepared at 500 µM in PBS. PARPis Veliparib (ABT888; 200 mM in water) was obtained from Enzo Life Sciences, and BMN 673 (5 mM in DMSO) was obtained from Abmole BioScience, (Kowloon, Hong Kong).

In vitro culture of patient samples

Mononuclear cells (MNCs) from AML primary samples were isolated using Histopaque-1077 (Sigma-Aldrich) according to manufacturer’s instructions. MNCs were incubated overnight in Hematopoietic Progenitor Growth Media (Lonza, Walkersville, MD) supplemented with 50 ng/ml thrombopoietin and FLT3L, 25 ng/ml stem cell factor, 10 ng/ml IL-3, IL-6, GM-CSF and G-CSF (Gemini Bio). Primary AML patient samples were obtained with informed consent on a protocol approved by the institutional review board of the University of Maryland School of Medicine (IRB # H25314).

Colony forming assay

AML cells were treated with vehicle or DNMTis daily for 72 hr, followed by a 24 hr recovery. Viable cells were then plated in triplicates in methylcellulose (cell lines) or MethoCult (MNCs) with or without PARPis, and incubated for 10–14 days. Colonies were stained with 2-(4-iodophenyl)-3-(4-nitrophenyl)-5-phenlytetrazolium chloride (1 mg/mL) overnight at 37°C and counted using a colony (Synbiosis, Frederick, MD). Breast cancer cells were treated with both drugs combined for 7 days, and colonies were stained with 0.05% crystal violet solution for 30 min then counted.

Irradiation

For IR studies, cells were exposed to 4 Gy X-Ray radiation using a Pantak HF320 X-Ray machine (250 kV peak, 13 mA; half-value layer, 1.65 mm copper) at a dose rate of 2.4 Gy/min.

Subcellular fractionation and PARP trapping assay

Chromatin extraction was performed using Subcellular Fractionation kit (Thermo Fisher Scientific) according to manufacturer’s instructions. See Supplemental Experimental Procedures for more details. PARP-binding in chromatin was assessed by immunoblotting as described below.

Immunoblotting

Refer to Supplemental Experimental Procedures for details.

Immunofluorescence

Refer to Supplemental Experimental Procedures for details.

Chromatin fiber analysis

Method was modified from (Sullivan, 2010). Refer to Supplemental Experimental Procedures for details.

Co-Immunoprecipitation

Refer to Supplemental Experimental Procedures for more details.

Determination of synergism

Cells were plated on 96-well plates and treated with various concentrations of drugs alone or in combination at a constant ratio (1:20 BMN:AZA and 1:1 BMN:DAC). Following daily treatments for 7 days, the assays were terminated using CellTiter96 MTS reagent (Promega) and absorbance values were used to determine the fraction of cells affected (Fa) in each treatment and determine combination indices according to the Chou-Talalay method using CompuSyn software.

Laser irradiation and confocal microscopy

Refer to Supplemental Experimental Procedures for details.

Xenografts models

Female nude and NOD scid gamma (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, NSG) mice (6–8 weeks old) were used for MDA-MB-231 and SUM149PT xenograft studies, respectively. Exponentially growing cells (3×10⁶) were mixed with 33% matrigel and injected subcutaneously above right hind leg of the mice. One week post injection, volumes were externally measured in 2D using electronic calipers [(length X width²)/2]. When tumors reached 80–100 mm³, mice were sorted into 6 groups (n= 7 or 8) so the mean tumor volume between groups was similar. Mice were observed daily, weighed 5 days per week and tumor volume was measured twice per week. Mice were treated with 0.2 and 0.3 mg/kg BMN 673, 0.5 mg/kg AZA, the combination or vehicle. AZA was prepared in PBS at 0.05 mg/ml and stored in −80°C. Talazoparib was prepared weekly in 10% DMAc, 6% solutol and 84% PBS at 0.02 mg/ml and 0.03 mg/ml and stored in 4 °C in the dark. Talazoparib was administered by oral gavage and AZA was administered by subcutaneous injection daily, 5 days per week for the duration of the study. Mice were euthanized when tumors reached 1500 mm³ or showed necrosis in case of MDA-MB-231 xenograft.

For AML xenografts, female NSG mice (6–8 weeks old) were and. Exponentially growing MOLM14-luc (0.5×10⁶), a gift from Dr. Xiaochun Chen, and MV411-luc cells (1×10⁶), a gift from Dr. Sharyn Baker, (Ohio State University) were injected intravenously into the lateral tail vein of restrained mice. Three days later, cell engraftment was assessed after injection of D-luciferin (150 mg/kg IP) on a Xenogen IVIS-2-Imaging System (Alameda, CA). Mice were sorted into 4 treatment groups so the mean intensity of signal was equal. Mice were observed daily, weighed 5 days per week and leukemic burden was assessed weekly by non-invasive luciferin imaging. Mice were treated with 0.1 mg/kg BMN 673, 0.5 mg/kg AZA, the combination or vehicle. Drug administration was performed as described above. All mice were housed in a 12 hr light/dark cycle with access to food and water ad libitum. Studies were performed with IACUC approval (protocol# 1113007).

Statistical analysis

Statistical analyses for biological assays were performed using 2-tailed unpaired t-test. For chromatin fiber analysis, Z-test for two population proportions was used.

Box 1. Highlights.

• Low doses of DNMTis in combination with PARPis increase PARP1 binding in chromatin.

• DNMTis and PARPis increase PARP1 and DNMT1 retention at DSBs inducing cytotoxicity.

• The DNMTi-PARPi combination shows strong anti-tumor effects in vivo.

• This paradigm provides for PARPis to be a therapeutic option for multiple cancers.

SIGNIFICANCE.

We introduce a mechanism-based therapeutic strategy, which combines DNMTis with potent PARPis. Efficacy is seen in both wild-type BRCA breast cancer cells and unfavorable AML sub-types, which are not clinically responsive to PARPis. The mechanism is based on DNMT’s and PARP1 interacting during DNA damage and DNMTis increasing tight binding of both DNMTs and PARP1 into chromatin. The inhibitor combination, versus each agent alone, increases the amplitude and duration of PARP1 at laser-induced DNA damage sites and is dependent on DNMT1 and PARP1. The significant increases in anti-tumor effects with the drug combination in vivo suggest a potentially potent therapeutic strategy for cancers not responsive to PARPis or DNMTis alone and which is advancing to clinical trials in AML.