Triptolide sensitizes cancer cells to nucleoside DNA methyltransferase inhibitors through inhibition of DCTPP1-mediated cell-intrinsic resistance

Abstract

Abnormal DNA hypermethylation mediated by DNA methyltransferases (DNMT) is a nearly universal hallmark of human cancers. However, while DNA methyltransferase inhibitors (DNMTi) such as decitabine and azacitidine are effective in treating myelodysplatic syndrome/leukemia, they have had limited utility for the majority of other cancers. Through a chemical library screen, we identify that triptolide, a diterpenoid epoxide from Tripterygium wilfordii, and multiple of its analogs, significantly augment the epigenetic and anti-cancer effects of decitabine in vitro and in vivo. These effects are attributable to inhibition of DCTPP1-mediated cleavage of 5-aza-deoxycytidine triphosphate, the convergent activated metabolite of nucleoside DNMTi, leading to enhanced drug incorporation into genomic DNA, increased DNMT degradation, enhanced DNA demethylation and associated transcriptional reprogramming. We show that high DCTPP1 expression mediates cell-intrinsic resistance to nucleoside DNMTi, and that triptolide and its analogs could overcome this resistance. These findings nominate combining DNMTi with triptolide or its analogs as a rational cancer therapeutic strategy.

Abnormal DNA hypermethylation is common in cancer, but DNA methyltransferase inhibitors show limited activity in many tumors. Here, the authors employ a chemical library screen to identify triptolide as an active agent that enhances these drugs by blocking DCTPP1, increasing drug incorporation into DNA, promoting demethylation and epigenetic reprogramming, and improving anti-cancer effects.

Introduction

In cancer cells, somatic alterations in DNA methylation appear to occur consistently, to arise early, and to be potentially reversible^1–5. Increased CpG dinucleotide methylation at transcriptional regulatory regions of many genes typically facilitates gene repression, leading to silencing of critical cancer genes^2,6,7. Suggestive of a driver role in human carcinogenesis and disease progression, such DNA hypermethylation alterations can be strongly maintained and subject to selection through disease progression and metastatic dissemination^5,8–10. For this reason, therapeutic reversal of epigenetic gene silencing, with nucleoside analog DNA methyltransferase (DNMT) inhibitors such as decitabine (DAC, Dacogen®), azacitidine (Vidaza®), and newly developed oral formulations of these drugs (ASTX-727 (Inqovi®) and Onureg®), have been developed as rational anti-cancer treatment strategies. The mechanism of action of these drugs involves metabolic conversion to 5-aza-2′-deoxycytidine triphosphate (5-aza-dCTP), incorporation into genomic DNA, trapping of DNMT enzymes and their subsequent proteolytic degradation, and ultimately DNA demethylation through passive loss of 5-methylcytosine^11–13. Thus far, the DNMTi have shown promise when used for myelodysplastic syndrome (MDS), for certain acute myeloid leukemias (AMLs), and a limited number of other cancers^14–18.

However, DNMTi have not been effective as single agents for the majority of cancer types despite the nearly universal prevalence of DNA hypermethylation alterations across cancer types^19,20. Additionally, the clinical use of nucleoside analog DNMTi has been generally limited by poor bioavailability, instability in blood, and substantial toxicity when administered at high doses^21,22. The difficulties in administering higher nucleoside DNMTi doses has even obfuscated whether DNMT inhibition is responsible per se for the responses seen in hematologic disorders²³. Some of the poor pharmacologic properties of the DNMTi have been attributed to rapid metabolism of the nucleoside analogs by cytidine deaminase (CDA)^24–26. Strategies aimed at inhibiting CDA in combination with decitabine or azacitidine to enhance the bioavailability and stability of the DNMTi have been translated to clinical use^27–31. Lower doses of decitabine or azacitidine, given alone or in combination with histone deacetylase inhibitors, have exhibited improved safety with preserved efficacy in clinical trials^32–34.

In this work, we hypothesize that there may be cancer cell intrinsic mechanisms that could confer primary or acquired resistance to DNMT inhibitors, potentially limiting their utility in reversing DNA methylation and anti-cancer efficacy. To ascertain whether available drugs could inhibit such resistance mechanisms, we screen 3800 compounds from the Johns Hopkins Drug Library collection^35,36 alone or in combination with decitabine, and identify that triptolide (TPL), a diterpenoid epoxide found in the Thunder God Vine (Tripterygium wilfordii), exhibits marked synergy with decitabine in cancer cell growth inhibition and cytotoxicity. Using pharmacological/biochemical, molecular genetic, and biophysical approaches, we identify that inhibition of DCTPP1 by triptolide and its analogs accounted for the synergy between triptolide and decitabine. By inhibiting DCTPP1-mediated pyrophosphate cleavage from the active decitabine metabolite 5-aza-dCTP, triptolide/analogs facilitate greater drug incorporation into the genome, triggering increased DNMT degradation, enhanced demethylation and modulation of gene expression programs, including interferon response pathways. Furthermore, the degree of DCTPP1 expression is associated with sensitivity to nucleoside DNMTi in correlative and mechanistic studies. Collectively, these results implicate DCTPP1 is cancer cell intrinsic resistance to nucleoside analog DNMTi. The combination of triptolide or its DCTPP1-inhibiting analogs with the existing nucleoside analog DNMTi is therefore ripe for further preclinical and clinical testing in human cancers.

Results

A chemical library screen for synergistic combinations with decitabine

We first screened the Johns Hopkins Drug Library, comprised of 3800 of the estimated 10,000 drugs used in medical practice^35–37. We identified those agents that showed >50% growth inhibition (measured by tritiated thymidine incorporation³⁶) of the prostate cancer cell line DU145 either alone (5 μM of each drug) or in combination with a low dose of decitabine (100 nM) that is well below circulating concentrations of decitabine achieved by clinically used doses (Fig. 1A). This narrowed the initial set of compounds to 131 compounds that could have any potential for interaction with decitabine to inhibit DU145 cell growth. After filtering out those compounds that are used only topically, we selected 66 of the remaining compounds to represent a diverse array of compound classes for further analysis in combination studies across a wide dose range of each hit and of decitabine (0 nM to 10 μM each drug). Among these 66 compounds, 24 compounds showed formal synergy with decitabine for DU145 growth inhibition (Chou-Talalay combination index (CI) < 1 at a minimum of 2 decitabine doses and sum of Bliss independence index across the full dose response > 0)^38,39 (Fig. 1B and Supplementary Fig. 1). Among these 24 synergistic hits, there are 6 anti-neoplastic drugs, 6 protein kinase inhibitors, 3 cardiotonic agents, 3 anti-rheumatics and immunosuppressants, 2 antibiotics, 2 mitotic poisons and 2 nucleoside analogs. Interestingly, two of the top 10 synergistic compounds, triptolide and triptonide, are diterpene triepoxides derived from the Thunder God Vine (Tripterygium wilfordii) (Fig. 1B). Triptolide is reported to be the principal active ingredient from extracts of T. wilfordii^40,41, and has been tested in multiple pre-clinical and clinical settings, including as an anti-inflammatory, immunosuppressive, anti-fertility, and anticancer agent^41,42. Here, we investigate the role of triptolide in sensitization to nucleoside analog DNMTi.

Fig. 1. Identification of compounds in the Johns Hopkins Drug Library demonstrating synergy with decitabine for cancer cell growth inhibition.

A Schematic outline of Johns Hopkins Drug Library (JHDL) screen. DU145 cells were treated with 100 nM decitabine or vehicle control for 3 days and then seeded into 96-well plates for overnight attachment followed by exposure to JHDL compounds (5 μM) for 72 h. The degree of ³H-thymidine incorporation (added at day 6 for 18 h) was measured to evaluate the degree of growth inhibition. 131 compounds showed more than 50% growth inhibition, when given alone or in combination with decitabine, of which 66 compounds were selected for a wide dose-response titration for synergy analysis. Among these, 24 compounds showed synergy with decitabine in inhibiting DU145 cell growth. B Scatter plot of 24 synergistic compounds with decitabine, ranked by Combination Index (CI) at a decitabine concentration of 100 nM (red circles) or 250 nM (black diamonds). Triptolide and triptonide are shown in red. Source data are provided as a Source Data file.

Triptolide synergistically sensitizes cancer cells to nucleoside analog DNMTi in vitro

We examined whether the synergy between triptolide and decitabine was more generalizable. When tested alone and in combination using a broad dose response of each agent, triptolide and decitabine exhibited marked synergistic interaction in inhibiting cancer cell growth and/or viability on multiple cancer cell lines including: 6 prostate cancer cell lines (DU145, LNCaP, C4-2B, LAPC4, CWR22Rv1, PC3) (Fig. 2A and Supplementary Fig. 2A), 3 leukemia cell lines (K562, CCRF-CEM, MOLT4) (Fig. 2B), 3 lung cancer cell lines (A549, HOP62, HOP92) (Fig. 2C) and 3 breast cancer cell lines (MCF7, T47D, MDAMB231) (Fig. 2D). The combination of triptolide and decitabine also significantly reduced clonogenic survival compared to either drug alone in multiple cancer cell models (Supplementary Fig. 2B). Interestingly, non-malignant RWPE1 prostate epithelial cells and normal hepatocytes in primary culture were largely resistant to the growth inhibitory effects of triptolide and decitabine, both alone and in combination (Fig. 2E). This suggests that the combination of triptolide and decitabine may exhibit a favorable therapeutic index.

Fig. 2. Triptolide and decitabine synergistically and selectively inhibit cancer cell growth, viability, and survival.

A Inhibition of growth and viability of human prostate cancer cell lines DU145 (left) and LNCaP (right) treated with decitabine and triptolide. Growth curves were measured as the percent of total confluence in each well over time, shown are mean of percent confluence ± SEM for 16 fields in each condition. Viability measurements of cells treated with decitabine and triptolide alone and in combinations across a dose series as indicated are represented in a heatmap scaled as shown in the color legend, with each measurement representing the percent viability with respect to the control treatment. The degree of synergy of combinations of decitabine and triptolide was calculated as the Bliss synergy score across the full dose ranges (values > 0 indicate greater than additive effect). B Inhibition of cell viability and synergy of decitabine and triptolide in human leukemia cell lines. CCRF-CEM, K562 and MOLT4 cells were treated in a dose series of decitabine and triptolide for 5 days. Viability measurements of triplicate wells treated with decitabine and triptolide alone and in combinations across a dose series as indicated are represented in a heatmap scaled as shown in the color legend, with each measurement representing the percent viability with respect to the control treatment. C Inhibition of cell viability and synergy of decitabine and triptolide in human lung cancer cell lines. A549, HOP62 and HOP92 cells were treated in a dose series of decitabine and triptolide as in (B). D Inhibition of cell viability and synergy of decitabine and triptolide in human breast cancer cell lines. MCF7, T47D and MDAMB231 cells were treated in a dose series of decitabine and triptolide as in (B). In (A–D) for all the tested cell lines, multiple dose combinations exhibited greater than additive effects. E Growth curve analysis as in (A) for non-malignant RWPE-1 prostate epithelial cells (left) and normal hepatocytes in primary culture (right). At doses that showed synergistic growth inhibition of cancer cell lines (A-D), decitabine, triptolide and their combination showed minimal growth inhibition of these non-malignant cells. Source data are provided as a Source Data file.

Combination of triptolide and decitabine showed enhanced efficacy at well-tolerated doses in vivo

To further explore the efficacy and safety of the combination of triptolide and decitabine, we tested each drug alone and in combination in a DU145 prostate cancer xenograft model in athymic nude mice. The combination of decitabine (1 mg/kg doses given i.p. on week 1 on days 1/3/5) and triptolide (0.2 mg/kg doses given each week on days 1–5, 8–12, 15–19) for two 3-week cycles significantly diminished xenograft growth and improved survival compared to triptolide alone, decitabine alone or vehicle control (Fig. 3A–C). Interestingly, at these doses in which the combination of triptolide and decitabine showed significantly enhanced efficacy in reducing xenograft growth, we did not observe increased systemic toxicity measured by mouse body weight loss, hematological parameters (WBC, RBC, hemoglobin), liver function (AST, ALT, total protein), and renal function (BUN, creatinine) throughout the treatment course (Fig. 3D–G). No necrosis or other morphological anomalies were observed in spleen, kidney and liver tissues in any treatment group (n = 3 animals in each group) after thorough histological examination by a veterinary pathologist of hematoxylin and eosin (H&E) stained formalin-fixed paraffin-embedded tissue sections (Supplementary Fig. 3). We also did not observe any major changes in feeding or other behavioral patterns, or overt gastrointestinal bleeding in any of the treatment groups throughout the treatment course. Taken together, these data suggest that the combination of triptolide and decitabine exhibited favorable efficacy and safety in vivo.

Fig. 3. Combination of triptolide and decitabine shows favorable efficacy and safety in vivo.

A Xenograft study design: NOD/SCID mice bearing DU145 human prostate cancer cell xenografts were divided randomly into 4 groups (starting with n = 10 mice in each group) and then treated with normal saline(Control, Arm 1), 1 mg/kg decitabine (Arm 2, three 1 mg/kg doses given i.p. on week 1 of two 3-week cycles), 0.2 mg/kg Triptolide (Arm 3, five 0.2 mg/kg doses given i.p. each week of two 3-week cycles) or in combination (Arm 4) upon xenograft tumors reaching an average size of 200 mm³. B DU145 xenograft tumor growth curve. The tumor volumes were measured every week. The tumor volume in each animal was estimated according to the formula: tumor volume (mm³) = 1/2 × Length ×  Width². Combination treatment (Arm 4) significantly inhibited the growth of DU145 xenografts compared to single treatment (Arm 2 and Arm 3) and vehicle control (Arm 1). Shown are the mean ± SEM of all xenografted tumors (n = 10 in each Arm at the start of treatment). P-values were assessed by two-way ANOVA with Tukey’s test. C Kaplan–Meier survival curves of DU145 xenograft mice. Combination of decitabine and triptolide significantly improved progression-free survival (time to >5× growth of DU145 xenograft in nude mice) compared to vehicle control (p = 0.003), decitabine alone (p = 0.003), or triptolide alone (p = 0.035). For Kaplan–Meier analysis, the two-sided Mantel–Cox log-rank test was used. D Mouse body weight curve. Mean mouse body weight ± SEM from these experiments are shown (n = 10 in each Arm at start of treatment). P-values were assessed by two-way ANOVA with Tukey’s test. There was no significant difference between any pairs of treatment arms. E–G Changes of hematological (E) parameters (RBC and WBC counting, hemoglobin) and biochemical parameters (Total protein, AST and ALT, glucose, BUN and creatinine) for liver (F) and kidney (G) toxicity monitoring. Each value represents the mean ± SEM of three mice from each treatment arm during treatment (at week 0, 2, and 6). Dashed lines indicate the normal range. P-values were assessed by two-way ANOVA with Tukey’s test. There was no significant difference between any pairs of treatment arms. Source data are provided as a Source Data file.

The synergistic interaction of triptolide and decitabine is mediated through inhibition of DCTPP1

We next sought to understand the mechanistic basis for the observed synergy between triptolide and the nucleoside analog DNMTi. To date, a number of direct cellular binding targets for triptolide have been reported, including Xeroderma Pigmentosum B (XPB)/ERCC3 subunit of general transcription factor TFIIH, the dCTP pyrophosphatase DCTPP1, the calcium channel polycystin-2, the membrane protease ADAM10, and the kinase TAK1 partner protein TAB1^40,43–47. Ensuing mechanistic studies revealed that only XPB among those putative targets is likely the physiological target mediating the antiproliferative activity of triptolide^44,48. Among the remaining targets, DCTPP1 appeared to be relevant due to its ability to hydrolyze nucleotide triphosphate in addition to dCTP^43,49,50. To understand which of these two direct targets (XPB and DCTPP1) are most responsible for the observed synergy between triptolide and decitabine, we first undertook a chemical biology approach using a series of 6 triptolide analogs that exhibited different potencies for inhibiting DCTPP1 and XPB in vitro (Table 1 and Supplementary Fig. 4A). For each triptolide analog, we determined the IC₅₀ for DU145 cancer cell growth inhibition (defined as the dose of the compound needed to reduce cell growth at 150 h to 50% of vehicle control) with and without combined exposure to decitabine (50 nM). The growth inhibitory potency of each compound alone was significantly correlated to its potency in inhibiting the target XPB (R² = 0.882, P = 0.006), but not to its potency for DCTPP1 inhibition (R² = 0.463, P = 0.137; Fig. 4A). In a multivariate analysis, the correlation with XPB was preserved (coefficient = 1.11, p = 0.04), while the low trend of correlation with DCTPP1 potency was largely eliminated (coefficient = 0.12, p = 0.76); this suggests that although treatment with single agent triptolide and its analogs can inhibit both XPB and DCTPP1 to varying degrees, the effect on growth inhibition is largely attributable to XPB inhibitory activity, and not DCTPP1 inhibitory activity, in agreement with previous findings^40,44,48.

Table 1 .

Bioactivities of triptolide (TPL) and its analogs

Entry	Name	Structure	IC50 (μM), XPB activity	IC50 (μM), DCTPP1 activity	IC50 (nM), DU145 growth	IC50 with 50 nM DAC (nM), DU145 growth
1	TPL		0.145	10.0	0.448	0.219
2	TPL-2004		1.44	3.10	6.05	1.60
3	TPL-2005		0.156	22.4	0.766	0.675
4	TPL-2015		2.90	>1000	8.71	9.78
5	TPL-2021		10.1	60.0	8.55	2.07
6	TPL-2024		100	>1000	3330	864

Fig. 4. Triptolide analogs with stronger potency for inhibition of DCTPP1 relative to XPB show greater synergy with decitabine.

A Correlation between potency (log₁₀[IC₅₀]) of inhibition of DU145 cell growth and potency (log₁₀[IC₅₀]) of inhibition of XPB (left panel) and DCTPP1 (right panel) activity in vitro for a series of triptolide analogs. B Correlation between the log-ratio of [potency of inhibition of DU145 cell growth of each triptolide analog with and without 50 nM decitabine] and the log-ratio of [potency of inhibition of DCTPP1 activity to that of XPB activity]. A, B Goodness-of-fit was evaluated by linear regression (R²), and the strength of the linear relationship was assessed using Pearson correlation analysis (P-values indicated on graphs). C Potency of inhibition of DCTPP1 (left) and XPB (right) in vitro activity by the triptolide analogs TPL-2004 and TPL-2015. Shown are the mean ± SEM of in vitro assays (n = 3 each). D DU145 growth curve analysis as in Fig. 2A with vehicle control, decitabine alone, TPL-2004 or TPL-2015 alone, or each in combination with 50 nM decitabine. Left, TPL-2004; Right, TPL-2015. The mean percent confluence ± SEM for 16 fields in each condition are shown. E (Left) The triptolide binding site overlaps with the nucleotide binding pocket of DCTPP1. The structure of DCTPP1 is shown colored in green, blue, and white by monomer subunit of DCTPP1. Triptolide is depicted as yellow sticks, with the dCTP-binding pocket as present in the mouse DCTPP1 homolog (PDB ID 6SQZ) outlined in magenta. A magnesium ion is represented as a green sphere. (Center) The triptolide binding pocket is predominantly formed by a hydrophobic sandwich of tryptophan residues W73 and W47, with H51 hydrogen bonding to the lactone carbonyl of the compound. (Right) Triptolide derivatives TPL-2015 and TPL-2004, represented as green sticks, manually overlayed onto the structure of DCTPP1 bound to triptolide. The additional moiety on TPL-2015 helps to close off the triptolide binding pocket. A steric clash appears capable of forming between the indole nitrogen of W73 and the hydroxyl modification present on TPL-2004 (as represented by transparent spheres colored by atom type). F Clonogenic survival and quantification in DU145 (left panel) and PC3 (right panel) cells. Shown are representative images of triplicate samples, quantification data are presented as mean ± SEM (n = 3 per group). Data were analyzed with ordinary one-way ANOVA with Tukey’s post-hoc test for multiple-testing correction. Source data are provided as a Source Data file.

Interestingly, the log ratio of the growth inhibitory IC₅₀ for each compound with decitabine to that without decitabine was significantly correlated to the log ratio of IC₅₀ for inhibition of DCTPP1 activity vs. XPB activity (R² = 0.901, P = 0.004; Fig. 4B). Thus, the extent to which each compound enhanced growth inhibition with decitabine co-treatment could be attributed to their relative potency for DCTPP1 with respect to XPB. In support of this notion, the degree of synergy between triptolide analogs and decitabine in inhibiting cell growth (measured by the average Bliss synergy score) was correlated with the potency of triptolide analogs in inhibiting DCTPP1 activity more so than that for inhibiting XPB activity (Supplementary Fig. 4B).

To explore this further, we closely examined the behavior of two triptolide analogs, TPL-2004 and TPL-2015 (Table 1 and Fig. 4C). Each of these compounds had comparable IC₅₀ for inhibition of XPB in activity assays. However, while TPL-2004 had strong potency for inhibition of DCTPP1 activity, comparable to that of the parent compound triptolide, TPL-2015 did not effectively inhibit DCTPP1 (Fig. 4C). Likewise, while both TPL-2004 and TPL-2015 showed similar single agent potency for inhibition of DU145 cell growth (6.05 and 8.71 nM IC₅₀ respectively; Table 1), likely attributable to their comparable potency of inhibiting XPB (Fig. 4C), only TPL-2004 and not TPL-2015 showed significantly enhanced growth inhibition with decitabine (Fig. 4D). This selective enhancement of growth inhibition by the combination of TPL-2004 with decitabine compared to the combination of TPL-2015 with decitabine was further confirmed by clonogenic survival assays in both DU145 and PC3 cells (Fig. 4F).

To define the structural basis for the inhibition of DCTPP1 by triptolide and to understand the differences between triptolide and its analogs, we determined the structure of the catalytic core of human DCTPP1 (residues 21–130) bound to triptolide using X-ray crystallography (Supplementary Table 1). A structure of mouse DCTPP1 has been previously solved in the presence of non-hydrolyzable 5-methyl dCTP (indicated as *-dCTP or Me-dCPNPP in Fig. 4E and Supplementary Fig. 5, respectively)⁵¹, as well as several structures bound to dCTP analogs⁵². As seen with the mouse enzyme⁵¹, human DCTPP1 adopts a tetramer in which each active site is formed by contributions from three different monomer chains. The dCTP binding pocket is composed of a hydrophobic pyrimidine-binding site and a triphosphate-binding locus lined with charged residues capable of coordinating a catalytic magnesium ion. The pyrimidine-binding pocket comprises Trp47, Phe50 from one subunit and Trp73 from an adjacent monomer (human DCTPP1 numbering). The human protein was highly similar structurally to its mouse homolog (Cα rmsd = 0.242 Å) (Supplementary Fig. 5A), as expected given the sequence similarity between the two proteins (88% identity, 93% similarity for the catalytic core regions). Inspection of the human DCTPP1•triptolide complex revealed that the inhibitor bound in part to the pyrimidine-binding locus (Fig. 4E and Supplementary Fig. 5B). Trp47 and Trp73 sandwiched the lactone ring and a portion of the diterpine ring of triptolide, while the lactone carbonyl of the compound formed hydrogen bonds to His51, a residue also seen to engage the 4-amino group of dCTP. Phe50 also made close contact to the diterpene ring of triptolide, while residues Tyr102 and Tyr129 formed an additional adjacent monomer to further define the hydrophobic binding pocket.

Triptolide contains three epoxides. The inhibition of XPB by triptolide has been attributed to covalent adduct formation between a cysteine in the protein and one of the three epoxides⁴⁰. In the DCTPP1•triptolide complex, the binding pocket contained no cysteine or serine and no covalent adduct were observed (Fig. 4E and Supplementary Fig. 5B). The structure of the complex also explained the disparate activities of analogs TPL-2004 and TPL-2015. The Hydroxyl substitution of TPL-2015 likely disrupts the hydrophobic sandwich formed by W47 and W73, accounting for the >100-fold difference in IC₅₀ of this derivative for DCTPP1 compared to triptolide (Fig. 4E and Table 1). By comparison, TPL-2004 was functionalized with an imidazole group linked through a thio-enol ester to a hydroxyl group on triptolide distal to its lactone ring. This imidazole substituent likely closes off the hydrophobic binding pocket exploited by triptolide (Fig. 4E), accounting for its observed ~threefold improvement in IC₅₀ (Fig. 4E and Table 1).

To further determine the mechanism of synergy between decitabine and triptolide, we carried out additional genetic experiments to complement the chemical biology approaches described above. CRISPR/CAS9-mediated genetic disruption of DCTPP1 had little effect on baseline cell growth (Supplementary Fig. 6), but led to sensitization of DU145 and PC3 cells to single agent decitabine (Fig. 5A), and greatly abolished the synergy across a wide dose response matrix of decitabine and triptolide combinations (P < 0.001, Fig. 5B, D). Moreover, forced expression of a DCTTP1 rescue construct harboring a synonymous mutation that protects from CAS9/sgRNA targeting (DCTPP1-Rescue) in DCTPP1-KO DU145 cells re-established overall synergy between triptolide and decitabine (Fig. 5F and Supplementary Fig. 7). However, knocking out XPB in DU145 and PC3 did not significantly impact the sensitivity to decitabine or the synergy between decitabine and triptolide (Supplementary Fig. 8A–C).

Fig. 5. The synergy of triptolide and decitabine is dependent on DCTPP1.

A Cellular sensitivity to decitabine is dependent on DCTPP1 levels. Human prostate cancer cell lines PC3 and DU145 were transduced with CRISPR-Cas9 system with DCTPP1 guide-RNA to knockout DCTPP1 (KO-DCTPP1), or non-targeting guide-RNA (KO-Control), or DCTPP1 overexpression lentivirus (OE-DCTPP1), or empty lentivirus (OE-Control). Potency of inhibition of growth and viability of the transduced isogenic cell lines was measured across a 8-point dose range of decitabine (n = 16 fields for each dose for each cell line). Shown are log₁₀[IC₅₀] with 95% confidence intervals (Left panel) with corresponding tabulated IC₅₀ values and 95% confidence intervals in linear scale (Right panel). P-values are determined using an extra sum-of-squares F-test. B–E The synergy between decitabine and triptolide is dependent on DCTPP1 expression levels in isogenic DU145 and PC3 cells. Viability measurements of DU145 and PC3 cells treated with decitabine and triptolide alone and in combinations across the indicated dose series are represented in heatmaps scaled as shown in the color legend, with each measurement representing the percent viability with respect to the control treatment. The degree of synergy of combinations of decitabine and triptolide was calculated as the Bliss synergy score across the indicated dose range. Knocking out DCTPP1 in DU145 (B) and PC3 (D) cells showed no overall synergy across serial dose combinations of decitabine and triptolide (Bliss values < 0 indicate less than additive effect). Both DU145 (C) and PC3 (E) cells with DCTPP1 overexpression retained slightly better overall synergy across serial dose combinations of decitabine and triptolide (Bliss values > 0 indicate greater than additive effect). F Restoring DCTPP1 expression in DCTPP1 knockout DU145 cells re-established synergy between decitabine and triptolide. Viability measurements and the degree of synergy are analyzed as in (B–E). There was strong synergy across serial dose combinations of decitabine and triptolide in control cells with intact endogenous DCTPP1 expression (KO-Control + Empty Vector cells, Bliss Sum = 0.89). The cells with DCTPP1-KO eliminated this strong synergy (KO-DCTPP1 + EV cells, Bliss Sum = −0.26). Rescuing DCTPP1 expression in these KO cells strongly re-established synergy across serial dose combinations of decitabine and triptolide (KO-DCTPP1 + Rescue cells, Bliss Sum = 2.58). B–F P-values are determined using the Wilcoxon Rank-Sum Test. Source data are provided as a Source Data file.

On the other hand, while forced expression of DCTPP1 in DU145 and PC3 cells had no impact on baseline growth (Supplementary Fig. 6), it led to significant desensitization to single agent decitabine (Fig. 5A) and enhanced the synergy between decitabine and triptolide in inhibiting cancer cell proliferation across a dose response matrix (Fig. 5C, E). In contrast, forced expression of XPB in DU145 and PC3 cells had little impact on sensitivity to single agent decitabine (Supplementary Fig. 8D).

Taken together, these data show that: (i) TPL and its analogs would be expected to inhibit both XPB and DCTPP1 to varying extents when used alone or in combination with decitabine; (ii) the inhibition of cell growth by single agent triptolide or its analogs is likely due to their ability to inhibit the target XPB and unrelated to their inhibitory activity for DCTPP1; and (iii) the observed synergy between triptolide and its analogs with decitabine for inhibition of cancer cell growth and survival is attributable to inhibition of DCTPP1 and not related to inhibitory activity against XPB.

Defining the mechanism of synergy of DNMT and DCTPP1 inhibition

Since DCTPP1 was the major target for triptolide and its analogs for synergy with decitabine in inhibiting cancer cell growth and survival, we next explored the underlying mechanisms. DCTPP1 encodes the deoxycytidine triphosphate pyrophosphatase, which is involved in balancing levels of deoxycytidine triphosphate (dCTP) and deoxycytidine monophosphate (dCMP) in cells. DCTPP1 has also been shown to hydrolyze pyrophosphate from non-natural and modified dCTP, including 5-halogenated, 5-formyl, 5-methyl, 5-hydroxymethyl, more so than dCTP itself, potentially acting to more selectively prevent the incorporation of these modified dCTP nucleotides into genomic DNA^43,49,50.

The convergent mechanism of action of the nucleoside analog DNMTi, including decitabine, azacitidine, and guadecitabine, involves cellular conversion to 5-aza-dCTP and incorporation into the genome during DNA synthesis, triggering trapping and degradation of the DNMT enzymes and passive loss of methylation across DNA replication^11–13. We hypothesized that DCTPP1 may have pyrophosphatase activity on 5-aza-dCTP, thereby preventing its incorporation into genomic DNA, presenting a cell-intrinsic resistance mechanism to nucleoside analog DNMTi.

To begin testing this hypothesis, we examined the enzymatic activity of DCTPP1 for hydrolysis of pyrophosphate from 5-aza-dCTP and dCTP. These analyses revealed that DCTPP1 had slightly increased catalytic efficiency (Kcat/Km) for 5-aza-dCTP compared to dCTP (Supplementary Fig. 9A). We also observed that triptolide can inhibit DCTPP1 pyrophosphatase activity for 5-aza-dCTP with an IC₅₀ similar to that for dCTP (14.7 and 8.49 μM respectively) (Supplementary Fig. 9B). Additionally, we showed that decitabine did not inhibit DCTPP1 pyrophosphatase activity (Supplementary Fig. 9C). These data suggest that DCTPP1 may reduce 5-aza-dCTP availability for incorporation into genomic DNA, and that triptolide can increase the availability of 5-aza-dCTP incorporation by inhibiting DCTPP1 pyrophosphatase activity.

Consistent with this notion, co-treatment with decitabine and triptolide led to enhanced 5-aza-dC incorporation into genomic DNA compared to treatment with decitabine alone as measured by a liquid chromatography/tandem mass spectrometry assay (Fig. 6A, B)⁵³. Similarly, compared to that in control DU145 cells, the level of 5-aza-dC incorporation into genomic DNA after decitabine treatment was significantly reduced in DCTPP1-overexpressing cells (Supplementary Fig. 10A), and was significantly increased in DCTPP1-knockout cells (Supplementary Fig. 10B). Intriguingly, this enhanced genomic incorporation of 5-aza-dC by combination decitabine and triptolide treatment translated to significantly enhanced trapping and degradation of all three DNMTs (DNMT1, DNMT3A, and DNMT3B) in the DU145 cells (Fig. 6C and Supplementary Fig. 11A) as well as a greater degree of global DNA demethylation (Fig. 6D, E), compared to decitabine alone, which could only lead to reduction of DNMT1 levels. Furthermore, the enhanced trapping and loss of DNMT1 is a common result of combination treatment with decitabine and triptolide in multiple cell lines, including PC3, LNCAP, C4-2B, LAPC4, and CWR22RV1 (Supplementary Fig. 11A–D). This enhanced demethylation was also observed when decitabine was combined with the DCTPP1-inhibiting triptolide analog TPL-2004, but not with TPL-2015 (Fig. 6F).

Fig. 6. DCTPP1 inhibition by triptolide enhances the action of nucleoside analog DNMT inhibitors.

A, B 5-aza-dC incorporation into genomic DNA, determined by LC-MS/MS, in DU-145 (A) and PC3 (B) cells treated with vehicle control, triptolide or decitabine alone or in combination. The mean of the amount of 5-aza-dC per 1000 2-dC in genomic DNA ± SEM of triplicate treatments is shown. Triptolide treatment significantly enhanced the decitabine incorporation into genomic DNA in DU145 and PC3 cells. C The combination treatment of triptolide and decitabine enhanced DNMT loss. Western Blot of DNMT1, DNMT3A, DNMT3B in DU145 cells treated with vehicle control, decitabine (250 nM) or triptolide (1 nM) alone or in combination for 24 h. β-actin blotting is shown as protein loading control. Data shown are representative of triplicate samples. D Genomic 5-methyl-deoxycytidine content by LC-MS/MS in DU145 cells treated in vitro with vehicle control, decitabine or triptolide or both at indicated doses. Each value represents the mean ± SEM of 5-methyl-deoxycytidine content from triplicate measurements at 5 days from the start of treatment. The combination treatment of triptolide and all tested doses of decitabine significantly lowered the levels of 5-methyl-deoxycytidine content in genomic DNA compared to each dose of decitabine alone. E Genomic 5-methyl-deoxycytidine content by LC-MS/MS in DU145 xenograft tumors treated in vivo with vehicle control, decitabine or triptolide or both. Each value represents the mean ± SEM of 5-methyl-deoxycytidine content from three xenografted tumors as a percentage of that measured in xenografted tumors from vehicle control treated animals. F Genomic 5-methyl-deoxycytidine content by LC-MS/MS in DU145 cells treated in vitro with vehicle control, triptolide analogs (TPL-2004 or TPL-2015), decitabine or triptolide or both. Each value represents the mean ± SEM of 5-methyl-deoxycytidine content from triplicate measurements at 5 days from the start of treatment. P-values were assessed by ordinary one-way ANOVA with Tukey’s post-hoc test for multiple-testing correction (A, B, D–F). Source data are provided as a Source Data file.

Combination of DNMTi with triptolide enhances the epigenetic effects of decitabine in surviving cells

Prior work has suggested that high doses of nucleoside DNMTi lead to significant cytotoxicity without mediating significant epigenetic effects in the surviving cells; in contrast, low doses of nucleoside DNMTi mediated significant DNA demethylation and long term epigenetic reprogramming of gene expression but without significant cytotoxicity^9,54–58. We hypothesized that the combination of triptolide (or its DCTPP1-inhibiting analogs) with decitabine could allow significant cancer cell cytotoxicity, while also mediating DNA demethylation and epigenetic reprogramming in any surviving cells. To test this hypothesis, we exposed DU145 cells to vehicle controls, triptolide (0.5 nM) or decitabine (33 nM) alone at low doses, or the combination, for 3 days, replaced with drug free media, and assessed the surviving cells for genome-wide DNA methylation (using the Infinium EPIC microarray platform) and gene expression (using RNA-seq) patterns over a time course at 5 days and 20 days post initial exposure. As expected, the combination of triptolide and decitabine showed significant inhibition of growth/survival at 5 days while each drug alone had minimal effect. Also as expected, low dose decitabine alone, but not triptolide alone, led to marked genome-wide DNA demethylation (Supplementary Fig. 12A). Interestingly, the combination of low dose triptolide and decitabine also showed marked genome-wide DNA demethylation, and enhanced the degree of DNA demethylation compared to low-dose decitabine alone to some extent. Moreover, this pattern of genome-wide DNA demethylation was retained in the surviving cells even at 20 days post-exposure (Supplementary Fig. 12B). Consistent with these in vitro findings, in the long-term in vivo experiment, mice treated with the combination of decitabine and triptolide showed significantly greater loss of 5-methyl-deoxycytidine (5mdC) in the residual tumor genomic DNA compared to that in mice treated with decitabine alone (Fig. 6E).

Remarkably, in terms of gene expression, we also observed that the genes significantly induced by decitabine alone (Benjamini–Hochberg adjusted p (padj) <0.001, log₂(fold change) > 0.5), across the board, were subtly but consistently further upregulated by the combination of decitabine and triptolide (p < 0.001; Fig. 7A and Supplementary Fig. 13A, B). As an example, the long noncoding RNA H19, which is known to exhibit loss of imprinting and DNA methylation mediated silencing in DU145 cells, was highly induced by decitabine alone, and was further upregulated by the combination of triptolide and decitabine, which was confirmed by qRT-PCR (Fig. 7B). Deep bisulfite sequencing of the previously characterized imprint control region of H19^59,60 showed that the combination of decitabine and triptolide led to a greater fraction of completely demethylated alleles compared to decitabine alone (Fig. 7C). Consistent with prior reports^9,61, in our experimental system, decitabine strongly induced expression of endogenous retroviruses, and interferon response genes (Fig. 7D–F). The combination of triptolide and decitabine further enhanced induction of these genes compared to decitabine alone (Fig. 7D–F).

Fig. 7. The combination treatment of triptolide and decitabine enhances the epigenetic effects of decitabine.

A The set of genes induced by decitabine (n = 720 genes, DAC vs Control, log₂[Fold Change] > 0, P < 0.001) are further upregulated by treatment with the combination of triptolide and decitabine. (Upper panel) Heatmap representing day-20 gene expression profiles of triplicate samples of DU145 cells treated with vehicle control (Control), 2 nM triptolide alone (TPL), 33 nM decitabine alone (DAC), or in combination (DAC + TPL). (Lower panel) Barcode distribution plot of decitabine induced genes in the comparison between DAC + TPL vs DAC alone. Each vertical line represents a decitabine upregulated gene as defined in (A), ranked by log₂[fold change] of DAC + TPL vs DAC alone. Red vertical line represents the median for decitabine induced genes; Blue vertical line represents median of all genes measured. 525 of 720 decitabine induced genes were further upregulated by triptolide and decitabine combination treatment (p < 0.001 by Wilcoxon rank sum test). B qRT-PCR analysis of H19 expression showed significantly higher expression in DAC + TPL compared to decitabine alone, triptolide alone or vehicle control. Shown are mean ± SEM of triplicate samples per group (n = 3 per group). C Bisulfite sequencing analysis of H19 imprint control region in the representative samples at day-20 after the indicated treatment. Data shown are the numbers and patterns of methylated and unmethylated clones (Left), percentages of complete demethylation clones with 95% confidence interval (Right). p-values were assessed by Two-sided Fisher’s Exact test. Color scale shows methylation fraction at each CpG site. D, E DU145 cells treated with decitabine alone (D) and combination of triptolide and decitabine (E) were enriched for increased expression of interferon-α response genes (upper panel, p < 0.001, GSEA) and interferon-γ response genes (lower panel, p < 0.001, GSEA). F qRT-PCR validation of endogenous retroviral gene HERV-FC1 (upper panel) and interferon response gene IFI27 (lower panel). Data shown are represented as mean ± SEM of three biological replicates at day-15 after indicated treatment; y-axis is represented as fold change over vehicle control treatment (n = 3 per group). B, F P-values were assessed by ordinary one-way ANOVA with Tukey’s post-hoc test for multiple-testing correction. **, p-value < 0.0001. Source data are provided as a Source Data file.

In addition to the decitabine induced gene set described above, we also analyzed the RNA-seq data to assess whether there were genes that were uniquely re-expressed after combined treatment with decitabine and triptolide relative to that with decitabine alone or any other condition (control and TPL alone). A collection of 478 genes were robustly upregulated by the combination, but only inconsistently, if at all, by decitabine alone, and largely downregulated by TPL alone compared to the control conditions (Supplementary Fig. 13A, B). Intriguingly, 56 of these 478 genes (~ 11.7%) were annotated tumor suppressor genes by the TSGene knowledge base⁶². Additionally, 101 of these 478 genes harbored promoter hypermethylation in DU145 cells (DNA methylation beta value in the upper quartile (75%ile or above) among all NCI-60 cell lines). Finally, of the 56 tumor suppressor genes, multiple were hypermethylated in DU145 cells at baseline, and showed re-expression in the DAC + TPL treatment. These included notable tumor suppressor genes such as CEACAM1, GLIPR1, SPARC (Supplementary Fig. 14).

As previously reported, DCTPP1 plays essential roles in maintaining dCTP levels, preventing the incorporation of non-canonical deoxycytidine analogues into genomic DNA, maintaining genomic stability and regulating transcription^63,64. Here, we further tested the role of DCTPP1 in modulating gene expression programing by overexpressing or knocking out DCTPP1 in DU145 cells carrying out RNA-seq transcriptome analyses. Many differentially expressed genes (DEGs) displayed reciprocal regulation between DCTPP1 knockout and overexpression conditions (Supplementary Figs. 15A, B and 16A, B). Likewise, this reciprocal modulation by DCTPP1 knockout and overexpression was also evident at a pathway level (Supplementary Fig. 15C). Notably, DCTPP1 knockout resulted in higher immune response pathways, including TNF-α signaling, and inflammatory and interferon-γ response; while DCTPP1 overexpression resulted in upregulation of hedgehog and Wnt signaling pathway genes. Interestingly, TPL treatment induced pathway alterations that were highly analogous to those mediated by DCTPP1-knockout, indicating that triptolide treatment induced a DCTPP1 knockout-like transcriptional state (Supplementary Fig. 15C).

To statistically examine whether TPL treatment closely aligned with DCTPP1 inactivation at a gene level, we developed an upstream regulator approach to assess whether TPL treatment was inducing a DCTPP1 inactivation state. For this approach, we used the DCTPP1 knockout and overexpression RNA-seq data to develop a pathway activation z-score that would indicate DCTPP1 pathway activation vs. pathway inactivation in the DU-145 cells treated with TPL compared to control. An empirical, non-parametric, one-tailed p-value was determined by creating a null distribution from randomly permuted (n = 10,000) gene expression values to assess what fraction of the random datasets showed a z-score that surpassed the observed z-score. For this pathway inactivation z-score, a strong negative value would be consistent with a DCTPP1 inactive state, while a strong positive value would indicate a DCTPP1 active state. TPL treatment induced a strong DCTPP1 inactive state, which was statistically significant (z = −2.68, p = 0.0037) (Supplementary Fig. 15D), indicating that TPL treatment induced a transcriptional state that was consistent with DCTPP1 loss of function.

Thus, compared to decitabine alone, the combination of triptolide and low dose decitabine captures the most desired effects of both low-dose and high-dose nucleoside DNMTi, in that it leads to both selective cancer cell cytotoxicity while also mediating, and even enhancing, the long-lived epigenetic effects of DNA demethylation and gene re-expression in surviving cells.

DCTPP1 inhibition by triptolide overcomes a cell intrinsic resistance mechanism to nucleoside analog DNMTi

These data implicate DCTPP1 as a cell-intrinsic factor in mediating primary resistance to nucleoside analog DNMTi by preventing their incorporation into genomic DNA. To further investigate this notion, we assessed whether the degree of incorporation of decitabine into genomic DNA in a panel of 16 human cancer cell lines (5 lung cancer, 4 breast cancer, 4 melanoma, 2 prostate cancer, and 1 ovarian cancer) was associated with the level of expression of DCTPP1 or with any other genes from the “pyrimidine metabolism” pathway curated in the Kyoto Encyclopedia of Genes and Genomes (KEGG; Fig. 8A)¹³. Interestingly, there was a significant inverse correlation with 5-aza-dC incorporation and DCTPP1 expression levels (R² = 0.493, p = 0.003, Fig. 8A–C)^65,66, and this correlation was preserved even after normalizing to differences in the rate of proliferation between cell lines as measured by tritiated thymidine incorporation (Supplementary Fig. 17A). This correlation with DCTPP1 expression was the top-ranked correlation among all genes in the “pyrimidine metabolism” KEGG pathway, including those that are known to be directly involved in the cellular uptake and metabolism of decitabine (Fig. 8A, B). Conversely, there was no significant correlation between 5-aza-dC incorporation and the level of expression of XPB, the other known major direct target of triptolide (Supplementary Fig. 17B). Furthermore, the growth inhibitory potencies of decitabine were also significantly correlated with the DCTPP1 expression levels among a panel of 18 cancer cell lines (R² = 0.238, p = 0.0399, Fig. 8B). In cell lines with intermediate to high DCTPP1 expression (DU145, PC3, A549 and MCF7), CRISPR/CAS9 mediated disruption of DCTPP1 significantly sensitized (reduced IC₅₀) the cells to single agent decitabine compared to those of the isogenic control cells (Figs. 5A and 8E and Supplementary Fig. 18A and C). On the other hand, in cell lines having low to intermediate DCTPP1 expression (MDA-MB-231, DU145 and PC3), overexpressing DCTPP1 conferred resistance (increased IC₅₀) to single agent decitabine compared isogenic control cells (Figs. 5A and 8F and Supplementary Fig. 18A, B).

Fig. 8. DCTPP1 expression leads to reduced 5-aza-dC incorporation and decitabine potency.

A A plot of the correlation coefficient versus the [–log₁₀(p-value)] for the correlation between 5-aza-dC incorporation into genomic DNA and the level of expression for each gene from the Kyoto Encyclopedia of Genes and Genomes (KEGG) “pyrimidine metabolism” pathway. The Human Genome Organization (HUGO) gene symbols for the top-ranked genes (with p < 0.1) are indicated. B The sensitivity to decitabine (log₁₀(IC₅₀)) is significantly correlated with the level of DCTPP1 expression in a panel of 16 cancer cell lines. C The level of incorporation of 5-aza-dC into genomic DNA after two days of exposure to 250 nM decitabine is significantly inversely correlated with the level of expression of DCTPP1 in the panel of 16 cancer cell lines. The cells selected for genetic manipulation are indicated with arrows. A–C Goodness-of-fit was evaluated by linear regression (R²), and the strength of the linear relationship was assessed using Pearson correlation analysis (P-values indicated on graphs). D A schematic representation of the enzymes (with HUGO gene symbols of encoding genes in parentheses) curated in the “pyrimidine metabolism” KEGG pathway that are directly involved in decitabine uptake, metabolism and incorporation. Our proposed model of synergy between decitabine and triptolide is shown with bolded arrows. E, F Genetic disruption of DCTPP1 significantly sensitized A549 and MCF7cells to decitabine (E), while overexpressing DCTPP1 in MDA-MB-231 cells conferred significant resistance to decitabine (F). Triplicate data were collected from each treatment condition to calculate IC₅₀ (n = 3). Data represent log₁₀ [IC₅₀, DAC (nM)] and 95% confidence intervals. P-values are determined using an Extra Sum-of-Squares F-test. G DCTPP1 expression levels in acute myeloid leukemia are significantly lower than those in most solid tumors in The Cancer Genome Atlas (TCGA) datasets. The box of the box-and-whisker plots showed the median and interquartile range, and the whiskers represent the minimum and maximum log₂(RPKM) value (reads per kilobase per million). P-Values are determined with ordinary one-way ANOVA with Tukey’s post-hoc test for multiple-testing correction. **, p < 0.001. Source data are provided as a Source Data file.

Nucleoside analog DNMTi, including decitabine, azacitidine, and guadecitabine, have shown clinical efficacy in MDS and AML^{13,20,29,67,68}, but have been disappointing as single agents in many solid organ cancers. Interestingly, an analysis of DCTPP1 mRNA expression levels in multiple cancer types from the TCGA showed that the level of DCTPP1 expression in AML (n = 175) is significantly lower than those in the majority of major solid organ cancers (Fig. 8G). Furthermore normal bone marrow and liver tissues had relatively low DCTPP1 expression (Supplementary Fig. 19), suggesting that there may not be an enhanced hematopoietic or liver toxicity of combination DNMT and DCTPP1 inhibition compared to DNMT inhibition alone. We can speculate that the relatively low level of DCTPP1 expression in bone marrow and hematological malignancies compared to most normal tissues and solid tumors respectively may be partially responsible for the observed hematological side effects and clinical sensitivity to nucleoside analog DNMTi in hematological malignancies like MDS and AML.

We further tested whether inhibition of DCTPP1 by triptolide may have more broad implications for sensitization to several classes of cytidine nucleoside analog drugs whose mechanism of action involves incorporation into DNA. The other FDA approved nucleoside DNMTi azacitidine can be metabolized and incorporated into both RNA (~ 80–90%) and DNA (~ 10–20%); however, the major mechanism of action in MDS/AML and other cancers is thought to arise from the conversion (through ribonucleotide reductase and other enzymes) to 5-aza-dCTP and incorporation into DNA, thereby exerting both epigenetic demethylation and DNA damage responses^22,69,70. We found that, similar to decitabine, co-treatment of DU145 cells with azacitidine and triptolide or its DCTPP1-inhibiting analog TPL-2004 showed synergistic cytotoxicity, enhanced 5-aza-dC incorporation into genomic DNA and global demethylation compared to treatment with azacitidine alone (Supplementary Figs. 20A, B and 21A, B). Since DCTPP1 does not have enzymatic activity on ribonucleosides, these data provide further confirmation that the major mechanism of action of azacitidine is through incorporation in DNA and not RNA. In further support of this notion, we found that across a panel of 16 cell lines, the degree of 5-aza-2′dC incorporation into genomic DNA and the potency of decitabine in inhibiting cancer cell growth was highly correlated to those of azacitidine (Supplementary Figs. 20A and 21A). Since decitabine cannot be incorporated into RNA, this high correlation provides further support that azacitidine exerts growth inhibition through the fraction that gets incorporated into DNA. In contrast to 5-aza-2′dCTP, gemcitabine or cytarabine triphosphate are not substrates for DCTPP1^50,71. Consistent with this substrate selectivity, there was no appreciable synergy for the combination of triptolide and gemcitabine or cytarabine in multiple cell lines (Supplementary Fig. 22A–C).

Discussion

Although nucleoside analog DNMTi have been effective for treatment of MDS and leukemia, they have shown limited single-agent clinical efficacy for the majority of human cancer types. Using a chemical screen of known drugs with subsequent validation and mechanistic studies, we identified that triptolide, a major active ingredient of Tripterygium wilfordii, and several of its analogs showed strong synergy with nucleoside analog DNMTi in suppressing growth and survival of cancer cells in vitro and enhanced efficacy of inhibiting xenograft tumor growth at well-tolerated doses in vivo. We used chemical biology, genetic approaches (knockout and overexpression of XPB or DCTPP1) and structural analysis to determine that the mechanism of synergy between triptolide and decitabine was attributable to targeting DCTPP1. By inhibiting DCTPP1-mediated hydrolysis of 5-aza-dCTP, triptolide and its analogs (i) enhanced the genomic incorporation of nucleoside analog DNMTi, (ii) led to trapping and degradation of DNMT1, and to some extent DNMT3A and DNMT3B, and (iii) maintained or even enhanced DNA demethylation and decitabine induced gene re-expression programs despite the synergistic cancer selective cytotoxicity of the combination.

Consistent with this notion, decitabine incorporation into genomic DNA and its efficacy were significantly associated with the level of expression of DCTPP1, but not XPB or any other enzymes involved in metabolism of deoxycytidine or decitabine en route to incorporation into genomic DNA in a series of cancer cell lines (Fig. 8A–C). Knocking out DCTPP1 in cancer cells with intermediate to high DCTPP1 expression sensitized these cancer cells to single agent decitabine, while overexpressing DCTPP1 in cancer cells with low to intermediate DCTPP1 expression conferred resistance to single agent decitabine. These data suggest that DCTPP1 may represent a major cancer cell intrinsic mechanism of primary resistance to nucleoside analog DNMTi. The synergy between triptolide and decitabine was evident in multiple cancer types tested in vitro including prostate cancer cell lines (AR-positive: LNCaP, CWR22RV1, LAPC4; and AR-negative: DU145, PC3; Fig. 2A and Supplementary Fig. 2A), leukemia cell lines (CCRF-CEM, K562, MOLT4; Fig. 2B), lung cancer cell lines (A549, HOP62, HOP92; Fig. 2C) and breast cancer cell lines (MCF7, T47D, MDAMB231; Fig. 2D). Interestingly, a recent study has shown that DCTPP1 nuclear expression is often significantly enhanced in the majority of solid organ cancers studied compared to normal counterparts⁷², consistent with the poor activity of nucleoside analog DNMTi as single agents in solid organ cancers. This notion is further supported by our analysis of TCGA RNA-seq data showing that the majority of major solid organ cancers exhibited significantly higher DCTPP1 expression compared to that in hematological malignancies (Fig. 8G). Other enzymes, such as CDA, deoxycytidine kinase, and the nucleotide transporters hENT1/2, may also be involved in cell intrinsic and extrinsic resistance mechanisms as has been suggested by prior studies^{24–26,73–76}, but their level of mRNA expression does not appear to be rate-limiting in determining the degree to which decitabine can get incorporated into genomic DNA.

We show here that treatment with TPL and its DCTPP1-inhibiting analogs can inhibit primary resistance to decitabine by inhibiting DCTPP1-mediated inactivation of the active triphosphate metabolite of decitabine. It is likely that combined triptolide and decitabine treatment may lead to additional secondary or acquired resistance mechanisms. Future studies, using powerful forward genetic screening approaches with inducible mutagenesis screens may prove useful in anticipating such secondary resistance mechanisms to combined decitabine and triptolide⁷⁷.

Inhibition of DCTPP1 by triptolide also sensitized to the nucleoside DNMTi azacitidine, but did not sensitize to other cytidine analog drugs such as gemcitabine and cytarabine. This is likely due to the lack of enzymatic activity of DCTPP1 for the triphosphate metabolites of gemcitabine and cytarabine^43,50. Targeting DCTPP1 may also have some utility as a single agent by inhibiting the sanitation of damaged deoxycytidine nucleoside pools and by promoting random salvage incorporation of various naturally occurring modified cytidine nucleosides, such as 5-methylcytosine, hydroxymethylcytosine, formylcytosine, thereby interfering with their tightly regulated signaling functions in cancer cells^{26,43,49,50,78}. DCTPP1 may thus play a role in safe-guarding cancer cells against genomic and epigenomic instability.

The findings here provide strong rationale for advancing a combination of triptolide or its analogs with existing nucleoside analog DNMTi for testing in clinical trials. There is mounting interest in using nucleoside analog DNMTi alone or in combination with other epigenetic drugs and to use these combinations to prime sensitivity to subsequent or concomitant administration of other agents, including targeted drugs, chemotherapeutic agents, and immunotherapies. In our in vitro treated cell model, there were significant enhancements of global demethylation and gene expression profiles, especially in interferon-α response genes and interferon-γ response genes (Fig. 7D–F). Given the strong efficacy and safety and enhanced epigenetic effects seen in our in vitro and in vivo models for the combination of decitabine with triptolide and its DCTPP1-inhibiting analogs, these combinations may be more effective than DNMTi monotherapy when deployed in such trials.

In the future, development and testing of selective DCTPP1 inhibitors, such as compounds with diverse heterocyclic scaffolds that have been recently reported, in combination with nucleoside analog DNMTi would also warrant clinical testing^79–81. Based on our mechanistic studies, we can evaluate the utility of DCTPP1 gene and protein expression and the degree of decitabine genomic incorporation as predictive and pharmacodynamic biomarkers to inform such clinical trials. Thus, the combination of triptolide and its DCTPP1-inhibiting analogs with nucleoside analog DNMTi is poised for further testing in other solid tumors. Additionally, in future clinical studies, we can evaluate the potential of DCTPP1 expression and/or activity levels as a predictive biomarker for sensitivity to nucleoside-analog DNMTi monotherapy in MDS/leukemia. Finally, recent studies have reported that dysregulated DNMT function in prostate cancer can be a driver of lineage plasticity as a mechanism of resistance to androgen receptor targeted therapies in prostate cancer; and that inhibition of DNMT and other epigenetic drivers may be of utility in treating neuroendocrine and other treatment-resistant prostate cancers^82–89. In future efforts, it will be interesting to assess the utility of combined DNMTi and DCTPP1i with and without androgen receptor blockade for treatment for various states of prostate cancer, including AR-positive, AR-negative, neuroendocrine, and double-negative states.

Methods

Materials and cell culture

All the cell lines, unless otherwise stated, were obtained from American Type Culture Collection and grew in the recommended medium. Human prostate epithelial cell line RWPE1 was purchased from Lonza (Basel, Switzerland) and were cultured in PrEGM medium (Lonza). All cell lines were maintained in a humidified incubator with 5% CO₂ at 37 °C. All chemicals, unless otherwise stated, were purchased from Sigma Aldrich (St. Louis, MO) and had purity greater than 98%. [α -³²P]-dCTP was purchased from Perkin Elmer (Waltham, MS); 5-methyl-2′-deoxycytidine (5mC), 5-azacitidine-¹⁵N₄ (5AC-¹⁵N₄), 2′-deoxycytidine-¹³C¹⁵N₂ (2dC-¹³C¹⁵N₂), and 5-methyl-2′-deoxycytidine-d3 (5mC-d3) were purchased from Toronto Research Chemical (Toronto, ON).

Clinical drug library screen

3800 clinical compounds were included in Johns Hopkins Drug Library (JHDL)^35–37. Human prostate cancer DU145 cells were treated with 100 nM Decitabine (DAC) or DMSO for 4 days in T75 flasks. Cells were then split and seeded into 96-well plates at 5000 cells/well in complete media and allowed to adhere overnight. On day 5, JHDL drugs were added to the plates in duplicate wells at a final concentration of 5 μM for a total of 72 h, with 16 wells left untreated as no-drug controls. After the first 48 h in JHDL compounds, 1 μCi [³H]-thymidine was added, and incubated for an additional 24 h (in both drug and [³H]-thymidine). Cells were then trypsinized and transferred onto FilterMat-A glass fiber filters (Wallac, Turku, Finland) using the Harvester-96 cell harvester (Tomtec, Hamden, CT). After washing with distilled water five times to remove free [³H]-thymidine, the incorporated [³H]-thymidine was read on a MicroBeta plate reader (Perkin Elmer, MA, USA), and the mean percentage of inhibition was calculated relative to no-drug control wells. In each group (DAC or DMSO pretreated cells), JHDL compounds demonstrating greater than 50% inhibition of [³H]-thymidine incorporation considered primary hits. After filtering out those compounds that are used only topically, 66 compounds representing a diverse array of classes were selected for combination titration studies across a wide dose range of each hit and of decitabine (0 nM to 10 μM each drug). For synthetic lethality and synergy analysis, the Combination Index (CI) was calculated with CompuSyn software using Chou-Talalay method (CI < 1 represents synergy) ³⁸and sum of Bliss independence index was calculated with Graphpad Prism 10 (Bliss independence index > 0 represents synergy)³⁹.

In vitro DCTPP1 pyrophosphatase activity assays

[α -³²P]-dCTP was used to detect dCTP hydrolysis by DCTPP1 with a thin-layer chromatography assay. Briefly, a 20 μL reaction mixture contained 50 mM HEPES (pH = 8.0), 100 mM NaCl, 5 mM MgCl₂, 1 mM Tris (2-carboxyethyl) phosphine hydrochloride (TCEP), indicated concentration of triptolide analogs, 10 μM dCTP, 1 μCi [α -³²P]dCTP, 100 μg/ml BSA, 10 nM His-DCTPP1. The reactions were started by addition of DCTPP1 for 5 min, and were stopped by addition of 5 μL of 0.5 M EDTA and dilution up to 100 μL with TE buffer. An aliquot of 1 μL reaction mixture was spotted on PEI-cellulose and the chromatogram was developed with 0.9 M guanidine hydrochloride (PH = 6.0). The percent of dCTP hydrolysis to dCMP was quantified using a PhosphorImager (Molecular Dynamics, Sunnyvale, CA).

To enable measurement of DCTPP1 pyrophosphatase activity on multiple deoxycytidine triphosphate analog substrates, we measured the rate of DCTPP1 pyrophosphate production through a coupled ATP regeneration reaction followed by detection of luciferase-mediated luminescence. Briefly, the pyrophosphate generated by DCTPP1 after hydrolysis of dCTP or 5-aza-dCTP was converted by ATP sulfurylase to ATP, which was in turn used to produce luminescence by a luciferase assay. A 45 μL reaction mixture contained 50 mM HEPES (pH = 8.0), 100 mM NaCl, 5 mM MgCl₂, 1 mM TCEP, various concentrations of triptolide, 100 μg/ml BSA, 0.015U ATP sulfurylase, 10 nM His-DCTPP1. The reactions were started by addition of various concentrations of 5-Aza-dCTP or dCTP, and incubated at room temperature at the indicated times followed by quenching via addition of 5 μL of 0.1 M EDTA. The luminescence produced by addition of 50 μL Buffer A (50 mM HEPES(PH = 8.0), 15 mM MgCl₂, 5 μM Adenosine 5′-phosphosulfate (APS), 0.015 U luciferase, and 100 μg/ml D-luciferin, using injector mode of Microbeta 1450-023(Perkins Elmer). The amount of DCTPP1-mediated production of pyrophosphate from 5-aza-dCTP or dCTP was quantified comparing to a pyrophosphate standard curve. The Kcat and Km for each substrate was calculated by Prism Graphpad 10.

In vitro XPB activity assays

The TFIIH complex containing XPB was purified and its DNA-dependent ATPase assay was performed as previously described in refs. ^40,44. Briefly, a 10-μl reaction mixture contained 20 mM Tris (pH 7.9), 4 mM MgCl₂, 1 μM of ATP, 1 μCi [γ-³²P]ATP (3000 Ci/mmol), 100 μg/ml BSA, 100 nM AdMLP, 5 nM TFIIH and indicated concentrations of triptolide or its analogs. The reactions were started by either addition of TFIIH for 2 h and stopped by addition of 2 μl of 0.5 M EDTA and dilution up to 100 μl with TE buffer. An aliquot of 1 μl reaction mixture was spotted on PEI-cellulose and the chromatogram was developed with 0.5 M LiCl and 1 M HCOOH. The percent of ATP hydrolysis was quantified using a PhosphorImager.

Quantitation of genomic 5-aza-dC and 5-methyl-dC content by high-pressure liquid chromatography-tandem mass spectrometry (LC-MS/MS)

The 5-aza-deoxycytidine (5-aza-dC) and 5-methyl-deoxycytidine (5-methyl-dC) content in genomic DNA was determined by a high-pressure liquid chromatography/tandem mass spectrometry (LC-MS/MS) procedure as described previously, by the Johns Hopkins Analytical Pharmacology Core facility^10,53. Briefly, ~2 to 5 μg genomic DNA in 50 μL of HPLC grade water was digested with 4 Units of Nuclease P1 (Sigma) at 65 °C for 10 min in a digestion buffer containing 0.04 mM desferrioxamine mesylate (DFAM), 3.25 mM ammonium acetate pH 5.0, 0.5 mM zinc chloride in a final volume of 100 μL. Subsequently, 20 μL of 100 mM Trizma base, pH 8.5 was added and this reaction was treated with 4 Units of Alkaline Phosphatase (Roche Life Science) at 37 °C for 1 h. Following incubation, 20 μL of 300 mM ammonium acetate, pH 5.0 and 6 μL of 0.25 mM DFAM in 50 nM EDTA was added to stop digestion. Standards and quality control (QCs) samples were prepared by adding known concentrations of decitabine, 2′-deoxycytidine (2dC), and 5-methyl-2′-deoxycytidine (5mC) into blank digest matrix (all solvents used during DNA digest except for the enzymes). All samples, standards and QCs were prepared for analysis by adding 20 µL internal standard (5-azacitidine (5AC)-¹⁵N₄ 2dC-¹³C¹⁵N₂, and 5mC-d3 in water) to 100 µL of sample and vortex-mix briefly. The samples were injected onto a LC-MS/MS system consisting of a Waters Acquity UPLC (Milford, MA) interfaced with an AB Sciex 5500 triple quadruple mass spectrometer (Foster City, CA). Chromatographic separation was achieved using Thermo Hypercarb porous graphite analytical column (100 × 2.1 mm, 5 µm, Waltham, MA) running isocratic elution with a mobile phase consisting of 10 mM ammonium acetate:acetonitrile with 0.1% formic acid (70:30, v/v) at a flow rate of 0.3 mL/min. The mass spectrometer was run in positive electrospray ionization mode monitoring for the following MRM transitions: 5-aza-dC (DAC): 228.9 → 113.0, 2dC: 228.0 →112.0, 5-methyl-dC: 242.0 → 126.0, 5AC-¹⁵N₄ (internal standard): 249.0 → 117.0, 2dC-¹³C¹⁵N₂ (internal standard): 230.8 → 115.0, and 5mC-d3 (internal standard): 245.8 →129.0. The calibration range was 2–400 ng/mL for 5-aza-dC, 5–1000 ng/mL for 5-methyl-dC, and 50-10,000 ng/mL for 2dC. All analytes used quadratic regression with 1/x² weighing. Results were reported as 5-aza-dC (DAC) and 5mC content per thousand 2dC.

Western blot detection of DNMT, DCTPP1and XPB proteins

Cells were collected by trypsinization, washed with PBS, and lysed in RIPA buffer (ThermoFisher) supplemented with protease and proteosomal inhibitors. Lysates were denatured in NuPage LDS Sample buffer 4× (Invitrogen, US) at 100 °C and 15 μg of protein were separated with 4–12% Bis-Tris gels (Invitrogen, US) and transferred to polyvinylidene fluoride membrane using NUPAGE transfer buffer (Invitrogen, US). Membranes were incubated with Odyssey blocking buffer (Li-Cor, Lincoln, NE) prior to incubation with appropriate antibodies, anti-DNMT-1(#PLA0011), DNMT-3A(#SAB5701326), DNMT-3B (#SAB4500664) antibody, 1:1000, Sigma; anti-DCTPP1 antibody, 1:500, Abcam (#AB224051); anti-XPB antibody, 1:500, Abcam(AB190698); anti-β-actin antibody, 1:5000 Cell Signaling (#3700) overnight at 4 °C. After three 5-min washes with PBS-Tween20 (0.05%), goat anti-rabbit (#926-32211) or goat anti-mouse (#926-32210) secondary antibody (1:5000, LI-COR) was applied for 1 h at room temperature, followed by three washes with PBS-Tween20 (0.05%). Visualization and quantification were carried out with the LI-COR Odyssey scanner and software (Li-Cor, Lincoln, NE).

Clonogenic survival assay

Clonogenic assays were performed to assess long-term cell survival. Human prostate cancer cell lines, DU145 and PC3 cells were plated in triplicates in a density of 2000 cells/well in 6-well plates for adhering overnight. Decitabine was freshly dissolved in DMSO and diluted with PBS to indicated concentration and added to culture plates within 15 min. Cells were cultured for 10–14 days to allow formation of colonies. After 30 min fixation with 20% methanol, the clones were stained with crystal violet (Sigma, US) and the colony areas were calculated with ImageJ⁹⁰. All dishes from one cell line were stained at the same time point. The average colony area for each condition was normalized to that in DMSO-treated controls for each cell line and each separate experiment. Statistical significance of differences was assessed by Student’s t-tests.

Alamar Blue cell viability assay

Cells were seeded into 96-well plates one day before compound treatments. After incubating with appropriate drugs for 96 h, 20 µL Alamar Blue reagent (Thermo Fisher, USA) was added into each well and incubated for 2 h. The fluorescence intensity was read on a microplate reader at 590 nm wavelength.

Growth curve assay

500 cancer cells were seeded per well in 48-well plates one day before compound treatments. Decitabine, triptolide and other drugs were freshly prepared and added to culture plates at indicated concentrations. For growth curve experiments, growth data were determined by imaging cell confluence at 6 h intervals at each of 16 fields per well using the IncuCyte™ ZOOM System (Essen Bioscience, MA) in the Johns Hopkins SKCCC Imaging Core Facility. The percent confluence was calculated with Incucyte built-in software. Growth curves were analyzed and constructed with Graphpad Prism software 10 (La Jolla, CA).

Growth inhibition of subcutaneous xenografted tumors

Protocols for all animal experiments conducted at Johns Hopkins University were approved by the University Animal Care and Use Committee. Human prostate cancer DU145 xenografts were generated by subcutaneous injection of 2 × 10⁶ cells into 8–10 week old athymic nude mice (Harlan, Indianapolis, IN). Tumor volumes (calculated as length × (width)²/2) were measured every week. When the tumors reached approximately 200 mm³ in size, 10 mice were randomly assigned into each treatment arm and received intraperitoneal injection of 100 µl of appropriate drug as outlined in Fig. 3A: control (normal saline, Arm 1), decitabine (three 1 mg/kg doses given i.p. every other day on week 1 of two 3-week cycles, Arm 2), triptolide (0.2 mg/kg doses given each day for 5 days with two days’ rest each week for two 3-week cycles, Arm 3) or both (Arm 4).

For analysis of systemic toxicity, mouse body weight was measured every week; whole blood samples were collected from three randomly selected mice in each group at baseline, at week 2, and at completion of treatment (week 6) by submandibular bleeding. Mouse blood samples were processed within four hours for measurement of complete blood count/hematology and clinical chemistry parameters by the Johns Hopkins Phenotyping Core facility. Animals were removed from the study and euthanized if a tumor is greater than 2 cm in any dimension, or any signs of distress, bleeding or ulceration were detected. No mice exceeded the maximal allowable tumor size throughout the entire study.

Lentiviral-mediated stable gene overexpression

pLenti-6 containing DCTPP1 or XPB ORF were shared from Dr. Jun O Liu’s lab in Johns Hopkins University. The reference sequence for DCTPP1 is available at https://www.ncbi.nlm.nih.gov/nuccore/BC001344.1. Packaged viral particles were used to infect selected cancer cell lines using polybrene media (at a final concentration of 8 μg/ml). Transduced cells were selected by blasticidin (5–10 μg/ml) (Sigma). Expression efficacy was determined by western blotting and cells were used for further studies as described.

Lentiviral-mediated stable gene knockout and rescue

pLenti-CRISPR.V2 plasmids containing DCTPP1 CRISPR guide RNA GCTTACATGTCCTCGAGCG), or XPB CRISPR guide RNA (AAACGGGCGTGCACGTTCGG), or non-targeting control were obtained from Genscript (Piscataway, NJ). Synonymous point mutations (changing from 5′-CCCACGCTCGAGGACATC to 5′-CCTACTCTTGAGGACATC) in the protospacer adjacent motif and coding frame was introduced into the DCTPP1 overexpression clone in order to create a DCTPP1-Rescue construct, using the Gibson Assembly HiFi Cloning Kit (Invitrogen). Packaged viral particles were used to infect selected cancer cell lines using polybrene media (at a final concentration of 8 μg/ml). Transduced cells were selected by puromycin (1–2.5 μg/ml) (Sigma). Knockdown efficacy was determined by western blotting and cells were used for further studies as described.

X-ray crystallography

Constructs and cloning

The gene for full-length human DCTPP1 was codon optimized for bacterial expression and synthesized by Twist Bioscience (www.twistbioscience.com). A slightly truncated crystallization construct comprising residues 21–130 (DCTPP1^(21–130)) was designed using sequence alignments with existing crystal structures of mouse DCTPP1 (2A3Q, 2OIG). DCTPP1^(21–130) was generated by PCR and subcloned into a modified pET vector containing an N-terminal hexahistidine SUMO tag using ligation independent cloning⁹¹.

Expression and purification

Expression of DCTPP1^(21–130) was conducted in BL21[DE]-RIL E. coli (www.agilent.com). Cells were grown in 2XTY at 37 °C to OD₆₀₀ of 0.5 before lowering the temperature to 18 °C for 30 min, followed by induction with 300 μM IPTG. Cells were harvested after 16 h at 18 °C and resuspended in Buffer A (50 mM Tris-HCl (pH 7.8); 800 mM NaCl; 10% glycerol; 10 mM imidazole; 0.5 mM TCEP-HCl; 1 mM PMSF; 1 μg/mL Leupeptin; 1 μg/mL Pepstatin). Cells were lysed by sonication and clarified by centrifugation before passing the lysate over a 5 mL HisTrap-HP column (www.cytivalifesciences.com). The column was washed with 200 mL of Buffer A and eluted with 30 mL Buffer B (50 mM Tris-HCl (pH 7.8); 800 mM NaCl; 10% glycerol; 500 mM imidazole; 0.5 mM TCEP-HCl; 1 mM PMSF; 1 μg/mL Leupeptin; 1 μg/mL Pepstatin). SUMO-tagged DCTPP1^(21–130) was cleaved with SENP protease and dialyzed overnight against Buffer A. Cleaved DCTPP1^(21–130) was again applied to a 5 mL HisTrap-HP column and flow-through containing protein was collected and concentrated for application onto a size exclusion column. Concentrated protein was applied to a Sephacryl S-300 column (www.cytivalifesciences.com) pre-equilibrated in Buffer C (50 mM Tris 7.8; 500 mM NaCl; 10% glycerol; 0.5 mM TCEP). Peak fractions were pooled and concentrated to 20 mg/mL. Final glycerol concentration was brought to 30% for flash freezing and storage at −80 °C.

Crystallization, data collection and model building

For crystallization, DCTPP1^(21–130) was dialyzed overnight against Buffer D (20 mM Tris 7.8; 500 mM NaCl; 0.5 M TCEP). After dialysis, 96 μL of DCTPP1 (10 mg/mL) was complexed with 4 μL triptolide (50 mM in 100% DMSO) to a final concentration of 2 mM Triptolide. The DCTPP1^(21–130)•triptolide complex was crystallized by hanging drop vapor diffusion. Protein and compound was mixed in a 1:1 ratio with crystallization buffer containing 100 mM Tris-HCl (pH 7.5), 8% PEG3350, and 300 mM proline and allowed to equilibrate by vapor diffusion at 20 °C. Crystals were cryoprotected in crystallization buffer supplemented with 25% glycerol and 2 mM triptolide before plunge freezing in liquid nitrogen. Diffraction data were collected at NSLS-II beamline 17-ID-1 (AMX) and auto-processed using FASTDP⁹². A molecular replacement solution was obtained by using 2OIG as a search model, stripped of ligands and waters using PHASER⁹³ in the PHENIX software suite⁹⁴. Model building was conducted using COOT⁹⁵. Refinement was carried out in PHENIX⁹⁴ and figures were generated using PYMOL (www.pymol.org). Coordinates have been deposited in the PDB (accession number 7MU5).

RNA-seq experiments

DU145 human prostate cancer cells were treated with 2 nM triptolide (TPL) alone, 33 nM decitabine (DAC) alone, or the combination (2 nM TPL + 33 nM DAC) once daily for 3 consecutive days, followed by 17 days of drug-free growth. Total RNA was extracted on day 20 using the Qiagen AllPrep kit (Qiagen, Germantown, MD). Sample-barcoded, strand-specific, RNA-seq libraries were prepared using the Illumina TruSeq Stranded Total RNA Sample Preparation kit (Illumina, San Diego, CA) and sequenced on an Illumina HiSeq platform. Reads were aligned to the hg19 reference and quantified using RSEM v1.2.09. Gene level counts were normalized to transcripts per million (TPM). Differential expression analysis was performed using the R package DESeq2⁹⁶. Heatmaps were generated using the CRAN package gplots. The Wilcoxon rank-sum test was applied to log2 fold change values (DAC + TPL versus DAC) to assess whether genes upregulated by DAC alone showed further induction with combination treatment. Gene set enrichment analysis (GSEA)⁹⁷ was performed for genes upregulated in DAC + TPL versus DAC using HALLMARK gene sets (15–500 genes) downloaded from the Molecular Signatures Database (MsigDB)⁹⁸.

Isogenic DU145 cell lines with genetic manipulation of DCTPP1 were generated by lentiviral transduction as described above. These included CRISPR/Cas9-mediated DCTPP1 knockout cells (DU145_KO_DCTPP1), non-targeting sgRNA controls (DU145_KO_CTRL), DCTPP1 overexpression cells (DU145_OE_DCTPP1), empty vector controls (DU145_OE_CTRL), and parental DU145 cells. Three independent biological replicates were collected per condition. Total RNA was extracted using the Qiagen AllPrep kit. Sample-barcoded, strand-specific RNA-seq libraries were prepared using the Illumina TruSeq Stranded Total RNA Ribo-Zero Human/Mouse/Rat kit (Illumina, San Diego, CA) and sequenced on an Illumina NovaSeq X platform to generate 150-bp paired-end reads. BCL files were converted to FASTQ using bcl2fastq v2.20.0. Read quality was assessed using FastQC v0.11.7 and trimming was performed using fastp v0.24.0. Reads were aligned to the hg38 human genome and quantified using RSEM v1.3.3 to obtain gene- and transcript-level expression, which was normalized to TPM. Differential expression analysis was performed using DESeq2 v1.44.0⁹⁶. Heatmaps were generated using ggplot2⁹⁹ and ComplexHeatmap¹⁰⁰. GSEA was performed using the fgsea¹⁰¹ R package for differentially expressed genes in DCTPP1 knockout, DCTPP1 overexpression, and triptolide-treated DU145 cells, using HALLMARK gene sets from MSigDB.

DCTPP1 pathway activation score

For DCTPP1 knockout, differentially expressed genes (DEGs) were defined by comparing DU145_KO_DCTPP1 with DU145_KO_CTRL using thresholds of |log₂ fold change| > 1 and p_adj < 0.05. For DCTPP1 overexpression, DEGs were defined by comparing DU145_OE_DCTPP1 with DU145_OE_CTRL using thresholds of |log₂ fold change| > 2 and p_adj < 0.0001. Core DCTPP1 pathway genes were then defined as those exhibiting reciprocal regulation between DCTPP1 knockout and DCTPP1 overexpression. Specifically, DCTPP1-upregulated genes were defined as those upregulated with DCTPP1 overexpression and downregulated with DCTPP1 knockout, whereas DCTPP1-downregulated genes were defined as those downregulated with overexpression and upregulated with knockout. These gene sets were identified by intersecting the corresponding DEG lists from each comparison, resulting in 120 DCTPP1-upregulated genes and 214 DCTPP1-downregulated genes, for a total of 334 core DCTPP1 pathway genes (Fig. S16).

To assess DCTPP1 pathway activity in response to triptolide treatment, a DCTPP1 pathway activation score was calculated to determine whether gene expression changes in triptolide-treated versus vehicle-treated DU145 cells were consistent with DCTPP1 activation or inactivation. For each of the 334 core DCTPP1 pathway genes, a score of +1 was assigned if the observed log₂ fold change was concordant with DCTPP1 activation (upregulation of DCTPP1-upregulated genes or downregulation of DCTPP1-downregulated genes) and −1 if discordant. The overall pathway activation score was defined as the sum across all 334 genes.

Statistical significance was assessed using permutation testing. Gene labels were randomly permuted 10,000 times to generate a null distribution of pathway scores, and an empirical one-tailed p-value was calculated as the fraction of permuted scores that equaled or exceeded the observed score in the direction of interest. The observed score was converted to a Z-score using the mean and standard deviation of the null distribution. Negative Z-scores indicate DCTPP1 pathway inactivation, whereas positive Z-scores indicate pathway activation.

Global genomic DNA methylation assays

Cells were treated with vehicle control, decitabine, triptolide or in combination for 5 days. Genomic DNA was extracted with DNeasy kit (Qiangen, Germantown, MD) according manufacturer’s protocol. The global DNA methylation were analyzed by direct Pyrosequencing after bisulfite conversion of DNA in EpigenDX (Hopkinton, MA).

Global gene expression and methylation analysis

Gene expression profiles for queried cells were analyzed using Agilent Human 44 K expression arrays (Agilent Techonologies, Santa Clara, CA). Data was loess-normalized to correct for dye and ratio bias at different signal intensities. Global methylation analysis was performed using the Illumina Infinium Human Methylation27 BeadChip (Illumina, Inc., San Diego, CA). Data quality was verified using in vitro methylated DNA (IVD) and DNMT1 and 3b double knock-out HCT-116 cells (DKO), which is highly depleted of methylation. Data from the above two platforms were analyzed with R and BioConductor.

Statistics

Statistical analyses were performed using Prism GraphPad 10 software (San Diego, CA), unless described separately above. The priori level of significance was set at 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.L., J.O.L., W.G.N., and S.Y. designed the study, analyzed data, and wrote the article. J.L., A.R., P.N., and J.S.S. performed Johns Hopkins Drug Library screen and synergy analysis. J.L., W.G.N., and S.Y. designed and analyzed animal studies for human prostate cancers. N.C., A.M.V., and J.L. conducted all animal studies. J.L., J(Jianya) Z., N.M.A., T.W., H.G., M.A.R., and S.Y. designed and conducted all the experiments related to LC-MS/MS. J.L., D.E., J(Jianying) Z. and A.M.V. conducted all the in vitro cell viability assay, cell growth assay and clonogenic survival assays. Q.L.H., J.L., J.O.L., W.G.N., and S.Y. designed and conducted all the in vitro DCTPP1 and XPB assays. J.L., R.C.S., R.V.C., A.M.V., W.G.N., and S.Y. designed and conducted all the global gene expression and methylation experiments. G.H. and J.M.B. designed and conducted all the X-ray crystallography. K.G., A.D.M., and J.Y.K. designed and conducted all the histology of animal studies. J.L., A.M.V., J.M.G., Y.Y., A.M., T.A., R.E., P.L., A.C.O., and J.K. designed and conducted DCTPP1 rescue experiments, RNA-seq experiments and analysis of DCTPP1 isogenic cells. All authors contributed to the writing, editing, and approval of the final manuscript.

Peer review

Peer review information

Nature Communications thanks Shusuke Akamatsu, Yong Chen and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by NIH/NCI grants P50CA058236, R01CA183965, R01CA070196, P30CA006973, S10RR026824, UL1TR003098, Prostate Cancer Foundation Challenge Awards, Patrick C. Walsh Koch Research Award, the Ellen B. Masenheimer Fellowship (S.Y.), the Flight Attendant Medical Research Institute (J.O.L.) and the Commonwealth Foundation. This research used beamlines AMX(17ID-1) and FMX(17ID-2) of the National Synchrotron Light Source II (NSLS-II), a U.S. Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Brookhaven National Laboratory under Contract No. DE-SC0012704. The Center for BioMolecular Structure (CBMS) at NSLS-II is primarily supported by the National Institutes of Health, National Institute of General Medical Sciences (NIGMS) through a Center Core P30 Grant (P30GM133893), and by the DOE Office of Biological and Environmental Research (KP1605010).

Data availability

The crystallization data and coordinates related to Fig. 4E and Supplementary Fig. S5 have been deposited in Worldwide Protein Data Bank (wwPDB, accession number 7MU5), www.wwpdb.org/pdb?id=pdb_00007mu5, or https://www.rcsb.org/structure/7MU5. And the “Full wwPDB X-Ray Structure Validation Report” for this research is also available on the PDB website, www.wwpdb.org/pdb?id=pdb_00007mu5. The methylation profiling data by the Infinium EPIC arrays (Illumina) related to Fig. 7C have been deposited in the Gene Expression Omnibus database (GEO, Accession code GSE185594). It is available on www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185594. The expression profiling data by RNA-seq in this study (related to Fig. 7A, D, and E and Supplementary Figs. S13–S17) have been deposited in the Gene Expression Omnibus (GEO, Accession codes GSE185651, GSE185650, GSE329432). It is available on https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185651. DCTPP1 mRNA expression data in The Cancer Genome Atlas (TCGA) is publicly available from http://www.cbioportal.org. The remaining data that support the findings of this study are available within the article, supplementary information, or Source Data file. Source data are provided with this paper.

Competing interests

J.L., J.O.L., W.G.N., and S.Y. are inventors on a provisional patent related to this work.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76852-2.