Selective targeting of NRF2-high pancreatic ductal adenocarcinoma with an NQO1-activatable prodrug

Significance

NRF2 activation is associated with resistance to oxidant-producing chemo- and radiotherapies, as well as enhanced pancreatic ductal adenocarcinoma (PDAC) growth and metastatic spread, and poor patient survival. Here, we show that NRF2^high PDAC can be selectively controlled by a prodrug, C29h, that is specifically activated by the NRF2-inducible enzyme NAD(P)H:quinone oxidoreductase-1 (NQO1). In addition to targeted therapy of this PDAC subset, C29h and related NQO1-activatable prodrugs may augment responses to standard-of-care drugs while reducing the systemic toxicity of chemotherapy. Given its greater efficacy in immunocompetent hosts, we propose that C29h may also augment the response to immunotherapy.

Abstract

Activation of transcription factor NRF2 in pancreatic ductal adenocarcinoma (PDAC) promotes aggressive tumor phenotype and protection from therapy-induced oxidative stress. We postulated that NRF2^high PDAC can be selectively targeted by C29h, a prodrug that is activated by the NRF2-induced enzyme NAD(P)H:quinone oxidoreductase-1 (NQO1), which is elevated in human pancreatic tumors. Initial evaluations of C29h alone or together with the standard-of-care chemotherapeutic drug gemcitabine were conducted on NQO1^high human and mouse PDAC cell lines and patient-derived organoids. As PDAC is enriched in collagen-containing extracellular matrix (ECM) that activates NRF2 and induces NQO1 expression, we examined the ECM effect on the response to C29h, as well as in vivo tumor control in IKKα-deficient Kras^G12D/Ikkα^ΔPEC mice in which NRF2 is strongly activated, immunocompromised Nu/Nu mice orthotopically transplanted with human PDAC cells and C57BL/6n and NOD/SCID mice transplanted with mouse PDAC. C29h led to NQO1-dependent killing of human and mouse PDAC cell lines and organoids and acted additively with gemcitabine. Furthermore, ECM-plated PDAC cells were more susceptible to C29h cytotoxicity than cells grown on plastic. Importantly, C29h treatment induced tumor regression and increased the survival of PDAC-bearing mice and optimal C29h-induced tumor regression was dependent on CD8⁺ T lymphocytes whose tumoral recruitment was enhanced by drug treatment. This study supports the use of C29h alone or as part of a drug combination as an effective and promising strategy for selective eradication of NRF2^high PDAC.

Pancreatic ductal adenocarcinoma (PDAC), the most common pancreatic cancer, is clinically challenging due to advanced stage at diagnosis and resistance to conventional chemo- and immunotherapies (1). Despite recent advances, PDAC’s 1-y and 5-y survival rates remain at 38% and 12%, respectively (2), rendering it a leading cause of cancer-related deaths in the United States and Europe. With only a minority of patients eligible for surgical resection (3), the standard of care for PDAC consists of cytotoxic combination chemotherapy, typically FOLFIRINOX or gemcitabine plus nanoparticle albumin–bound (nab)-paclitaxel (4). However, the success of chemotherapy is limited by acquired drug resistance and toxicity to healthy tissues and cells. An important contributor to oxidant stress inducing therapeutics is NRF2 (5), the master activator of the antioxidant response (6). NRF2 is genetically and posttranscriptionally activated in numerous cancers (7), including PDAC (8) and was recognized as an activator of metabolic genes that support rapid tumor growth (9). In PDAC, a highly desmoplastic cancer (10), NRF2 is activated on engagement of the collagen receptor DDR1 by matrix metalloprotease (MMP)-cleaved collagen I, resulting in upregulation of mitochondrial biogenesis and respiration genes (11). Moreover, we recently established NRF2 to be a critical component of an epigenetic switch that converts KRAS^G12D-induced premalignant lesions to invasive PDAC (12). Congruently, genetic and pharmacological manipulations that inhibit NRF2 activation curtail growth of mouse and human PDAC allografts (8). Although NRF2 is an attractive drug target, its systemic inhibition may increase susceptibility of normal cells to oxidative damage, especially when combined with chemo- or radiotherapy. We, therefore, sought ways for targeting PDAC with prodrugs that are selectively activated in cells with high NRF2 activity.

Cancer prodrugs are therapeutically inert and devoid of systemic toxicity and are converted to cytotoxic DNA-damaging entities within the proper target cell (13). A standard approach to prodrug design is masking of the cytotoxic functionality of an active chemical moiety using a cleavable linker, such that the inert compound is selectively activated by an enzyme that is highly expressed by the target cell (14). One commonly used enzyme is NAD(P)H:quinone oxidoreductase-1 (NQO1), an NRF2-inducible flavoprotein that catalyzes the reduction of quinones (15). NQO1 is weakly expressed in adipocytes, vascular endothelium, small intestine parasympathetic ganglia, and nerve fibers, and is barely detectable in healthy human hepatocytes, bile duct epithelium, and myocardium (16). NQO1 is also weakly expressed in normal tissues adjacent to renal cell and hepatocellular carcinomas (17), while being highly expressed in these cancers, and those of the colon, breast, pancreas, and lung (18). NQO1 was linked to tumor progression, aggressiveness, and drug resistance (19). Several classes of bioreductive prodrugs that undergo NQO1-dependent enzymatic reduction to release their toxic moieties were described (20, 21). Among these, Compound-29h (C29h) prepared by functionalization of 3-(hydroxymethyl) indolequinone, an NQO1 substrate that is subject to nonreversible two-electron reduction, and linkage to the cytotoxic diterpenoid (oridonin) leaving group (22), caught our attention. The quinone nucleus is present in many bioreductive compounds, including mitomycin C (MMC), which is used to treat several human cancers, as well as the structurally similar EO9 (apaziquinone), and is often linked to reductive activation (23). The cytotoxic diterpenoid oridonin was first extracted from Radosia rubescens, and identified in 1967 to have tumoricidal activity (24). C29h itself exhibits high NQO1-dependent toxicity in colon and lung carcinomas (22), and has the ability to target ROR^high/NQO1^high chronic lymphocytic leukemia cells, enhancing their susceptibility to venetoclax therapy (25). Here, we examine the anti-PDAC activity of C29h ex vivo, in the absence or presence of collagen-enriched ECM, and in vivo, using an autochthonous model and mice transplanted with human and mouse PDAC cells.

Results

NQO1 Upregulation in Human PDAC.

Moderate to strong staining for NQO1 was detected in 92% of resected PDACs, and 100% of endoscopic ultrasound-guided fine needle aspirations (EUS-FNA) with malignant PDAC diagnoses, independently of tumor grade/stage (26). To extend these findings, we analyzed PDAC GSE1547135 and GSE165153647, and GSE19650 of microdissected preneoplastic tissue and confirmed NQO1 mRNA to be highly elevated in PDAC compared to normal pancreas (average >10-fold; n = 50 PDAC tumors and n = 24 adjacent normal tissues; Fig. 1 A and B and SI Appendix, Fig. S1 A–C). TCGA analysis (n = 150) showed NQO1 mRNA to be more highly expressed in Classical (27, 28), and Progenitor (29) subtypes vs. other subtypes (Fig. 1C). These analyses were performed on treatment-naïve patient samples collected prior to chemotherapy exposure. No association with patient outcome, expressed as progression-free survival (PFS) or disease-specific survival (DSS) was found (SI Appendix, Fig. S1D). High NQO1 protein was observed in six specimens from patients with PDAC at the UCSD Cancer Center (Fig. 1D and SI Appendix, Table S1). Of these, four samples were collected from treatment-naïve patients at the time of surgery, and two samples were obtained following neoadjuvant therapy [FOLFIRINOX or Gemcitabine (GEM) plus Abraxane]. Given the small size and heterogeneity of this cohort, no statistical correlation could be established between NQO1 amounts and treatment history. Elevated NQO1 was also confirmed in 65.5% of specimens in a different human PDAC cohort (UCLA-TMA, SI Appendix, Fig. S1E and Table S2). We found no significant association between NQO1 amounts and other clinicopathologic variables (SI Appendix, Table S2). These data indicate that NQO1 expression is not merely a consequence of treatment but may serve as a biomarker for pretreatment patient stratification.

Fig. 1.

NQO1 is upregulated in human pancreatic cancer. (A) NQO1 mRNA in human PDAC (n = 50) and adjacent nonmalignant tissues (n = 24) from public datasets (GSE15471, GSE16515). **** Student’s t test, P < 0.0001. (B) NQO1 mRNA across preneoplastic stages (GSE19650): normal ducts, IPMA, IPMN, and IPMC. Kruskal–Wallis, P = 0.0006. (C) NQO1 mRNA in TCGA PDAC (n = 150) showing high expression in Classical subtypes (Moffitt, Collisson) and Progenitor (Bailey); one-way ANOVA, P < 0.0001. (D) Representative NQO1 IHC of PDAC with adjacent nontumor (A, green) and tumor (T, red) regions (dotted line). (Scale bar, 50 µm.) (SI Appendix, Table S1).

C29h Exhibits Potent Cytotoxicity against NQO^high PDAC Cells.

Given elevated NQO1 in most human PDACs, we tested whether the NQO1-activateable prodrug C29h selectively kills NRF2^high NQO1-overexpressing cancer cells. Relative NQO1 amounts were assessed in human PDAC and other cancer cell lines and compared to the C29h-sensitive colon cancer cell line HT29 (22). MIA-PaCa-2 and Capan-2 cells exhibited high NQO1 expression, whereas NQO1 was low or barely detectable in SW620 and Caco-2 colorectal adenocarcinoma cells (Fig. 2A and SI Appendix, Fig. S2A). In HT29 cells, C29h induces apoptosis and activates apoptotic caspases (22). To examine C29h induced apoptosis in NQO1^high PDAC cells we incubated MIA-PaCa-2, Capan-2, and Caco-2 cells with increasing C29h concentrations and monitored caspase-3 and PARP-1 cleavage. As little as 0.1 to 0.5 µM of C29h induced caspase-3 cleavage (cCasp-3) in the PDAC cell lines (Fig. 2B). No caspase-3, nor PARP-1 cleavage were detected in NQO1^neg Caco-2 cells (Fig. 2B). Congruently, C29h diminished cell proliferation/viability in NQO1^high MIA-PaCa-2 and Capan-2 cells, but not in NQO1^neg Caco-2 cells (Fig. 2C), with a similar IC₅₀ value to HT29 cells (SI Appendix, Fig. S2B). To confirm NQO1 dependence of C29h cytotoxicity, we ablated NQO1 in the PDAC cell lines using CRISPR-Cas9 and established polyclonal NQO1KO cell populations (SI Appendix, Fig. S2C). In MIA-PaCa-2 cells, residual NQO1 expression was minimal and detectable only by high-sensitivity immunoblotting (IB). Accordingly, NQO1KO cells barely showed caspase-3 and PARP-1 cleavage and were refractory to C29h-induced cytotoxicity (SI Appendix, Fig. S2 D–F). Because cytotoxic drugs target highly dividing cells, we compared the growth of parental (CTRL) and NQO1KO cells and ruled out that C29h resistance in NQO1KO cells was due to a proliferation defect (SI Appendix, Fig. S2G). We next investigated the benefit of combining GEM, which provides a marginal survival benefit (30), with C29h, and found that C29h additively increased GEM cytotoxicity in CTRL NQO1^high cells, but not in NQO1KO cells, which were as sensitive to GEM as their NQO1^high counterparts (Fig. 2D and SI Appendix, Fig. S2 H and I).

Fig. 2.

C29h is cytotoxic to PDAC cells with high NQO1 expression. (A) NQO1 protein in human cancer cell lines; ERK1/2, loading control. (B) Cleaved/total caspase-3 and PARP-1 in NQO1⁺ and NQO1⁻ cancer lines treated for 48 h with the indicated C29h doses; actin, loading control; 0 = DMSO control. (C) C29h dose–response; IC₅₀ from 72 h MTT assays. (D) Viability of parental and NQO1KO Capan-2 cells after 48 h treatment; data shown as % of DMSO controls (100%). Mean ± SD. Mean ± SD (n = 3 biological replicates). Two-way ANOVA, Dunnett’s multiple comparisons test; *P < 0.05, **P < 0.01, ****P < 0.0001, ns-not significant. (E–G) NQO1 protein (E), densitometry (F), and mRNA (G) in PDAC PDXs; MIA-PaCa-2 = 1.0 A.U.; actin, loading control. Mean ± SEM (n = 2 to 10 technical replicates). Kruskal–Wallis, Dunn’s multiple comparisons test; *P < 0.05, ***P < 0.001, ****P < 0.0001. (H) Clonogenic survival of AA1444 PDX cells ± 150 mM C29h for 8 d. (I) AA1444 viability after 48 h treatment; % MTT vs. NT control. One-way ANOVA, Dunnett’s multiple comparisons test; **P < 0.01, ***P < 0.001, ns-not significant.

Approximately 50% of PDAC PDXs (31, 32) showed elevated NQO1 protein and mRNA (Fig. 2 E–G and SI Appendix, Fig. S2J). We prepared 2D cultures from two NQO1^high PDXs, 1444 and 1305, which also showed elevated NRF2 (33). Both PDXs were sensitive to C29h ± GEM (Fig. 2 H and I and SI Appendix, Fig. S2 K and L). In our cohort, 4 of 12 (33.3%) PDX-derived lines were NQO1^low, reflecting tumor heterogeneity influenced by genetic background, cell origin, differentiation, and prior treatments (SI Appendix, Fig. S2J). This proportion is consistent with the UCLA Stage I/II pancreatic cancer cohort, where ~34.5% of cases were NQO1^low (SI Appendix, Table S2).

ECM Potentiates C29h Cytotoxicity via NQO1 Induction.

PDAC chemoresistance has been attributed to cancer intrinsic and tumor microenvironmental (TME) factors (34). The PDAC TME is mainly composed of fibroblasts that secrete an ECM enriched in fibrillar collagens, mainly consisting of Col I (35). While a small fraction of PDACs exhibit so-called inert stroma with little collagen cleavage, most PDACs contain reactive stroma in which Col I and other fibrillar collagens were cleaved by MMPs (11, 36). MMP–cleaved collagen I (cCol I), but not intact Col I (iCol I), activates the collagen receptor DDR1 and its downstream MAPK and NF-κB–p62–NRF2 signaling pathways, thereby stimulating mitochondrial biogenesis, energy production, and PDAC growth (11). To determine whether DDR1 engagement increases C29h sensitivity we cultured PDAC cells on uncoated and ECM-coated plates (Fig. 3 A, Top and SI Appendix, Fig. S3 A, Top) and monitored NRF2 activation with an antioxidant response element (ARE) driven reporter. Elevated reporter expression was observed 24 and 72 h after seeding PDAC cells on ECM-coated plates, correlating with DDR1 and ERK activation and NQO1 induction (Fig. 3 A, Bottom and SI Appendix, Fig. S3 A and B). This was accompanied by increased C29h, but not GEM, cytotoxicity in MIA-PaCa-2 and PDX-1444 cells (Fig. 3C and SI Appendix, Fig. S3C). CRISPR-Cas9 targeting of NQO1 in MIA-PaCa-2 cells reduced ECM-enhanced C29h cytotoxicity with no effect on DDR1 and MAPK activation (Fig. 3C and SI Appendix, Fig. S3D). The residual cytotoxic effect of C29h on ECM-plated NQO1KO cells is likely due to heterogeneous gene editing, generating a mixed population of cells, some of which retained functional NQO1, as described above (Fig. 3C and SI Appendix, Fig. S2D).

Fig. 3.

Culture on ECM coated plates increases C29h cytotoxicity. (A) Experimental schematic (Top) and IB of AA1444 PDAC cells cultured on plastic (none) or ECM for 72 h (Bottom Left) with densitometric quantification (Bottom Right; A.U. vs. HSP90). HSP90 and ERK1/2, loading controls. Representative of two biological and three technical replicates. Mean ± SD (n = 6). Two-tailed unpaired t test; P < 0.05. (B) Experimental schematic and assessment of C29h and GEM cytotoxicity (SI Appendix, Fig. S3C). (C) Parental (CTRL) and NQO1KO MIA-PaCa-2 cells on plastic or ECM treated 48 h with 60 nM C29h. Viability (CCK-8) expressed as % of DMSO controls (100%). Mean ± SD (n = biological replicates). One-way ANOVA, Šídák’s test; P < 0.05, *P < 0.01, ***P < 0.0001, ns.

C29h Induces Human PDAC Regression and Decreases Organoid Viability.

Athymic (nu/nu) mice were orthotopically transplanted with MIA-PaCa-2 cells and one wk later were randomly assigned into two groups receiving intraperitoneal (i.p.) C29h (20 mg/Kg) or vehicle (CTRL) injections, 3×/wk for 4 wk (Fig. 4A). C29h reduced pancreas weight (a surrogate for tumor burden), and tumor volume, while substantially increasing overall survival (Fig. 4 B–D). C29h-treated tumors showed decreased Ki-67 and CD44 (marking highly plastic PDAC cells), fewer F4/80⁺ macrophages and NQO1⁺ cells, and increased cCasp-3, confirming C29h’s ability to drive NQO1^high cancer cell apoptosis in vivo, with no effect on total Col I expression (Fig. 4E). General C29h toxicity was excluded by monitoring mouse weight during the treatment’s cycle (SI Appendix, Fig. S4A) along with weight, gross morphology, and internal organ (liver, kidney, heart) histology at the experimental endpoint (SI Appendix, Fig. S4 B–E). Notably, C29h did not affect liver NQO1 or cCasp-3 expression (SI Appendix, Fig. S4F).

Fig. 4.

C29h induces orthotopic PDAC regression and cytotoxicity. (A) Experimental schematic. (B) Gross morphology of pancreata bearing MIA-PaCa-2 tumors after indicated treatments. (C) Pancreas weight and tumor volume at endpoint. Mean ± SEM (n = 5 mice). Paired two-tailed t test (Right) and Mann–Whitney test (Left); P < 0.05, *P < 0.01. (D) Kaplan–Meier survival of mice with MIA-PaCa-2 tumors ± C29h (n = 6 to 8). Log-rank, P = 0.0339; Gehan–Breslow–Wilcoxon, P = 0.0368. (E) H&E and IHC of pancreata at endpoint with quantification. (Scale bars, 50 µm and 20 µm.) (NQO1, Ki-67). Mean ± SEM (n = 3 to 5 tumors). Each point represents one tumor (per-mouse average). Paired two-tailed t test and Mann–Whitney test (F4/80, cCasp-3); P < 0.05, *P < 0.01. (F) Clinicopathological features of patient tumors used for organoid cultures and relative NQO1 mRNA (Top) and protein (Bottom); MIA-PaCa-2, Capan-2, Caco-2 reference high/low NQO1 and C29h response; HSP90, loading control. (G) Viability of patient-derived organoids (PDOs) treated with C29h for 72 h; IC₅₀ from CellTiter-Glo assay. Mean ± SD (n = 3 biological replicates).

We validated these findings in patient-derived organoids (PDO). Analyzing the transcriptomic profiles of 44 PDAC-confirmed PDOs (37), we selected cultures with high (hM1A), intermediate (hF23), and low (hF3) NQO1 mRNA amounts (Fig. 4 F, Top). NQO1 protein was IB validated and compared to the C29h-sensitive MIA-PaCa-2 and Capan-2 cells, as well as C29h-resistant Caco-2 cells (Fig. 4 F, Bottom). Consistently, C29h treatment reduced the viability of NQO1^high PDO cultures (hM1A and hF23), while its effect on the NQO1^low hF3 culture was less pronounced, exhibiting a 10-fold higher IC₅₀ value (Fig. 4G).

C29h Blunts PanIN to PDAC Progression While Modulating the TME.

IKKα deficiency in pancreatic epithelial cells (PEC) causes chronic pancreatitis that accelerates PanIN to PDAC progression when introduced into Pdx1-Cre;LSL-Kras^G12D mice with PEC-restricted oncogenic Kras^G12D (hereafter Kras^G12D and Kras^G12D; Ikkα^ΔPEC) (8). Whereas only 10% of Kras^G12D mice progress to PDAC after a long latency, Kras^G12D; Ikkα^ΔPEC mice exhibit multiple PanIN2/3 lesions at 5 wk-of-age and full-blown invasive PDAC with elevated NRF2 and high NQO1 expression at 2 mo-of-age (8). We took advantage of the NRF2^high nature of this model to examine C29h’s effect on tumor development, regression, and host survival. Six-week-old female and male mice were i.p. injected with C29h (20 mg/kg) or vehicle control (Fig. 5A). After four cycles pancreata were isolated and analyzed, revealing that C29h substantially reduced the number of NQO1⁺, Ki-67⁺, KRT19⁺, and CD44⁺ cells, while increasing amylase⁺ acinar cell mass (Fig. 5 A and B). Moreover, C29h blunted PDAC progression as indicated by loss of advanced PanIN2/3 lesions in favor of acinar-ductal metaplastic (ADM) events (Fig. 5 B and C). Congruently, C29h-treated pancreata showed higher amounts of acinar identity mRNAs (Nr5a2, Mist1, and Amy2a5), whereas ductal (Krt19), stem cell (Cd44, Aldh1), and inflammatory (Adgre1) mRNAs were reduced (Fig. 5 B and D–F). C29h-treated pancreata contained more CD8⁺ T cells, and cCasp-3, marking apoptotic cell death (Fig. 5B). Consistent with the lower PanIN3 count and blunted tumor progression, C29h-treated pancreata showed reduced cCol I, although total Col I was barely affected. No significant changes were observed in DDR1, Acta2 mRNA (marking stellate cell activation), or Sirius red–positive desmoplastic areas. The reduced cCol I (3/4 fragment) content correlated with lower Mmp2 and Mmp13 mRNA levels (SI Appendix, Fig. S5 A and B). C29h-treated Kras^G12D;Ikkα^ΔPEC pancreata also showed reduced mRNA markers of cancer-associated fibroblasts (CAF), including Fap, whose ablation blunts tumor growth and potentiates antitumor immunity (38), Cxcl12, a homeostatic CXC chemokine secreted by FAP⁺ CAFs that was linked to T cell exclusion (39), and Saa3, a protumorigenic member of the serum amyloid A family (40) (Fig. 5F).

Fig. 5.

C29h inhibits PDAC development in Kras^G12D/PEC/Ikkα^ΔPEC mice. (A) C29h treatment scheme. (B) H&E and IHC of untreated and C29h-treated pancreata with quantification. (Scale bars, 50 µm and 100 µm.) (Amylase). CD8⁺ T cell density normalized to tissue area (cells/mm²) and averaged per tumor. Mean ± SD (n = 3 to 6 tumors). Unpaired two-tailed t tests and Mann–Whitney test (NQO1); P < 0.05, *P < 0.01, **P < 0.001. (C and D) Quantification of ADM and PanINs (C) and acinar-identity genes mRNAs (D) in Kras^G12D/PEC/Ikkα^ΔPEC pancreata ± C29h. Mean ± SEM (lesion counts, n = 56 to 77 fields) or mean ± SD (mRNA, n = 4 to 6). Kruskal–Wallis with Dunn’s test (C); multiple unpaired t tests, Holm–Šídák method (D); *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; ns-not significant. (E) NQO1 mRNA (Left) and protein (Right) in the same pancreata; ERK1/2, loading control. Mean ± SD (n = 4 to 6). Unpaired two-tailed t tests; P < 0.05. (F) Indicated mRNAs in above pancreata. Mean ± SD or ± SEM (Cxcl12) (n = 4 to 7). Unpaired two-tailed t tests; P < 0.05, *P < 0.01.

We further evaluated C29h’s effect on PanIN1-to-PDAC progression using our ex vivo organoid transformation system, which exhibits strong NRF2 dependence (12). As anticipated, C29h inhibited organoid transformation induced by the NRF2 activator sulforaphane, reducing organoid yield and size and triggering organoid death (SI Appendix, Fig. S5 C and D). The absence of an effect on NRF2KO organoids from Kras^G12D;Nrf2^ΔPEC pancreata confirmed the selective cytotoxicity of C29h toward NRF2^high/NQO1^high Kras^G12D PDAC progenitors (SI Appendix, Fig. S5 E–G). Notably, NRF2 ablation did not alter GEM-induced cytotoxicity (SI Appendix, Fig. S5 E and G).

Given increased CD8⁺ T cell infiltration, we investigated whether these cells potentiate the response to C29h. We orthotopically transplanted UN-KC6141 cells (41) into C57BL/6n and NOD/SCID mice (Fig. 6A) and gavemice a single C29h cycle when tumors became palpable. We note that tumor burden at treatment initiation may have contributed to early mortality in some animals, reducing group sizes at the endpoint. Compared to tumors in immunocompetent mice, tumors in T and B cell–deficient NOD/SCID mice were minimally affected by C29h, evidenced by unchanged pancreatic weight, histopathological features, and the amounts of NQO1, Ki-67, CD44, F4/80, cCasp-3, and Fap mRNA (Fig. 6 B–D). Consistent with these findings, C29h treatment increased tumoral Cd8 mRNA content (Fig. 6 E, Top), suggesting enhanced tumor infiltration with CD8⁺ T cells in C57BL/6n mice, which was confirmed by IHC (SI Appendix, Fig. S6A) and flow cytometry (Fig. 6 E, Bottom). Further analysis revealed that Fas, Fasl, and Il15 mRNAs were increased, while Kras^G12D mRNA was decreased following C29h treatment of C57BL/6n, but not in NOD/SCID, mice (SI Appendix, Fig. S6B). IHC analysis confirmed more FAS-positive cells in C29h treated C57BL/6n-grown tumors, but not in NOD/SCID-grown tumors (SI Appendix, Fig. S6A). A similar expression pattern was found for cCasp-8, suggesting that CD8⁺ T cell–mediated cytotoxicity could be mediated via FAS–FASL interaction and Casp-8 (SI Appendix, Fig. S6 A and C). FACS analysis confirmed that C29h led to CD8⁺ T cell upregulation within the tumor. In parallel, C29h significantly reshaped the TME, increasing CD8⁺ T cells, and reducing the proportion of CD3e⁺CD4⁺FoxP3⁺ regulatory T cells (Fig. 6 E, Bottom). Tumor-associated macrophages (TAMs) contribute to tumor progression via secretion of growth-promoting inflammatory cytokines and suppression of T cell function, and drive resistance to conventional and immune therapies (42). We analyzed expression of TAM-associated transcripts, including Nos2 (inducible nitric oxide synthase) and Ym1/Chil3 (Chitinase-like 3, SI Appendix, Fig. S6D). C29h substantially increased Nos2 expression (SI Appendix, Fig. S6D), which may influence tumor cell survival by enhancing DNA damage repair, reducing chemotherapy efficacy, and modulating immune responses (43). Conversely, C29h reduced Ym1/Chil3 mRNA amounts, suggesting a shift from M2-like TAM polarization, which is typically associated with immunosuppressive functions and tumor progression. Flow cytometry further revealed that C29h downregulated CD45⁺CD11b⁺F4/80⁺PD-L1⁺ myeloid cells (Fig. 6 E, Bottom), supporting a reduction in immunosuppressive TAMs (44). Collectively, these results support the notion that C29h reshapes the TME, potentially enhancing antitumor immune responses.

Fig. 6.

C29h induced PDAC regression is lymphocyte dependent. (A) Experimental design and treatment scheme. (B) Gross morphology of tumor-bearing pancreata, treated as indicated (Left), and corresponding weights (Right). Mean ± SD (n = 3 to 8 mice), one-way ANOVA test, *P < 0.05; ***P < 0.001, ns-not significant. (C) H&E and IHC of pancreata with quantification. (Scale bars, 50 µm and 100 µm.) (H&E). Values from multiple fields were averaged per tumor; each point represents one tumor. Mean ± SD (n = 3 to 5 tumors). Kruskal–Wallis, Dunn’s multiple comparison test, *P < 0.05; **P < 0.01; ****P < 0.0001, ns-not significant. (D) Nqo1, Fap, and Cd44 mRNAs in pancreata from C57BL/6n and NOD/SCID mice under indicated treatments. Mean ± SD (n = 4 to 7). ANOVA, Holm–Šídák’s multiple comparisons test. *P < 0.05; ***P < 0.001, ns-not significant. (E) Cd8 mRNA (Top) and flow cytometry of immune cells from C57BL/6n pancreata (Bottom). CD45⁺ cells analyzed for CD8⁺, CD44⁺CD8⁺PD-1⁺ Tex, CD3e⁺CD4⁺Foxp3⁺ Treg, and CD11b⁺F4/80⁺PD-L1⁺ TAMs. Mean ±SEM (n/pancreata = 4 to 8, Top; 3 to 6, Bottom). One-way ANOVA, Holm–Šídák’s multiple comparisons test (Top), and multiple unpaired t tests (Bottom). *P < 0.05; ***P < 0.001; ****P < 0.000; ns-not significant.

To assess the role of T cell subsets in C29h antitumor activity, CD4⁺ and CD8⁺ T cells were depleted in orthotopic PDAC-bearing C57BL/6 mice (45) and efficient depletion of each subset was confirmed in spleen and pancreatic tumors (Fig. 7A and SI Appendix, Fig. S7A). Consistent with the above results, C29h efficacy was completely lost upon CD8⁺ T cell depletion, as shown by unchanged tumor weight, absence of cCasp-3, and persistence of Ki-67⁺ cancer cells (Fig. 7 B–D). In contrast, CD4⁺ T cell depletion did not affect the antitumor activity of C29h, indicating that CD8⁺, but not CD4⁺, T cells are essential for tumor regression. Notably, removal of CD8⁺ T cells prevented the C29h-mediated reduction of Nqo1 and Cd44 expression and abolished Fasl induction, whereas loss of CD4⁺ T cells altered stromal and myeloid-associated transcripts, leading to Fap upregulation and loss of Ym1/Chil3 suppression. Moreover, Nos2 induction by C29h was abolished following either CD4⁺ or CD8⁺ T cell depletion (SI Appendix, Fig. S7B). These findings demonstrate that C29h requires adaptive immunity for in vivo efficacy.

Fig. 7.

Impact of CD4⁺ and CD8⁺ T cell depletion on C29h cytotoxicity. (A) Schematic of experimental design and treatments. Orthotopic pancreatic tumors were generated in C57BL/6 N mice and treated with C29h, CD4⁺ or CD8⁺ T cell–depleting antibodies, or isotype controls. (B) Gross morphology of tumor-bearing pancreata (Left) and corresponding tumor weights (Right). Each point = one tumor. Mean ± SD (n = 4 to 8 mice per group). One-way ANOVA with Šídák’s multiple comparisons test; *P < 0.05, **P < 0.01, ns-not significant. (C) Representative H&E and IHC for cCasp-3, CD8, and Ki-67 in pancreata from above-treated mice. (Scale bars, 50 µm.) (D) Quantification of IHC in the same pancreata; CD8⁺ T cell density normalized to tissue area (cells/mm²) and averaged per tumor. Data are shown as mean ± SEM (n = 4 to 8 mice per group). One-way ANOVA with Šídák’s multiple comparisons test; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns-not significant.

Discussion

Despite the scarcity of NRF2 activating mutations, PDAC development depends on NFE2L2 mRNA induction downstream to RAS-MAPK signaling (46) and subsequent posttranslational NRF2 activation by exogenous factors (8). NRF2 stimulates PDAC metabolism (33), and induces detoxifying enzymes that protect cells from oxidants and electrophiles, including chemotherapeutic drugs (5). One of the major NRF2-inducible protective enzymes is NQO1 (47), which is highly expressed in advanced PDAC, suggesting that highly aggressive and fast growing NRF2^high PDAC (12), could be targeted by NQO1-activateable bioreductive prodrugs (22). The results described above confirm this assumption and show that C29h effectively induces regression of NRF2^high pancreatic tumors generated by orthotopic transplantation of human and mouse PDAC cells and significantly extends host survival. In vitro studies demonstrated that C29h-induced cytotoxicity was NQO1 dependent. A unique feature of PDAC that contributes to resistance to conventional anticancer therapies is desmoplasia, characterized by excessive accumulation of fibroblastic stroma and deposition of a collagen-rich ECM (48). Consistent with NRF2 activation on DDR1 engagement by cCol I (11), PDAC cells cultured on ECM show increased susceptibility to C29h-induced cytotoxicity compared with those grown on uncoated plastic. Moreover, C29h reduced expression of PDAC-associated MMPs through a mechanism that remains to be elucidated, thereby reducing Col I cleavage. These findings suggest that although the desmoplastic stroma limits penetration of large therapeutic molecules, it may represent an “Achilles Heel” for C29h-based therapy. To further explore this interaction between the TME and drug response, we used fibroblast-derived ECM to investigate whether collagen remodeling enhances NQO1 expression and NRF2 signaling, thereby modulating sensitivity to the NQO1-specific prodrug C29h. Indeed, cCol I activated a DDR1–NF-κB–p62–NRF2 signaling cascade that significantly increased NQO1 amounts, sensitizing NQO1^high PDAC cells to C29h. We suggest that studies performed on ECM-cultured cancer cells are needed for more complete evaluation of cytotoxic anticancer drugs. Consistent with its ability to selectively eliminate NRF2^high cells, which may serve as PDAC progenitors (12), C29h treatment of Kras^G12D;Ikkα^ΔPEC mice reduced progenitor and stemness markers and blocked ADM to advanced PanIN progression, compromising tumor growth. In established tumors, however, our studies revealed that the full therapeutic efficacy of C29h extends beyond direct cytotoxicity and requires adaptive immunity. Consistent with other chemotherapies, including GEM (49–51) and KRAS inhibitors (52), C29h-induced PDAC regression increases CD8⁺ T cell infiltration and depends on their tumoricidal activity. An unanswered question is how C29h, which is potently cytotoxic in vitro, requires adaptive immunity for its early in vivo effects. We suggest that C29h-mediated cancer cell death and antigen release rapidly engage CD8⁺ T cells to amplify antitumor activity. Our observations are in line with prior reports that chemotherapy enhances tumor immunogenicity through multiple pathways, including immunogenic cell death, release of tumor antigens, upregulation of death receptors, and depletion of immunosuppressive populations. Enhanced CD8⁺ T cell recruitment correlated with decreased expression of CXCL12, a CAF-produced chemokine previously shown to induce T cell exclusion (39, 53). Fas–FasL signaling, a well-established mechanism by which CD8⁺ cytotoxic T lymphocytes eliminate cancer cells (54), could also be involved, as suggested by Casp-8 activation. Curiously, CD4⁺ T cell depletion did not impair tumor control but altered the expression of stromal and myeloid-associated genes, suggesting that CD4⁺ T cells provide contextual cues that shape the tumor microenvironment but are not required for early tumor clearance. C29h treatment also reduced intratumoral Tregs, key contributors to the immunosuppressive PDAC TME. Collectively, these results provide a rationale for combining C29h-like drugs with immune checkpoint inhibitors, a strategy that will require further studies and optimization of treatment regimens.

In conclusion, our results suggest that C29h-like prodrugs with improved potency and selectivity should be considered for further development and addition to the anti-PDAC armamentarium, both as stand-alone chemotherapeutics and as adjuvants for immune checkpoint inhibitors.

Materials and Methods

Genetically engineered mouse models of pancreatic ductal adenocarcinoma (PDAC), including Pdx1-Cre;LSL-Kras^G12D, Pdx1-Cre;Ikkα^F/F;LSL-Kras^G12D, Kras^G12D/PEC;Nrf2^ΔPEC mice, were generated on a C57BL/6 background and maintained under approved institutional and NIH animal care guidelines. Age- and sex-matched littermates were used for all comparative analyses. Human pancreatic and colorectal cancer cell lines, syngeneic mouse PDAC cells, and patient-derived xenograft (PDX) lines were cultured under standard conditions.

The NQO1-activatable prodrug C29h was synthesized as previously described and administered intraperitoneally at 20 mg/kg using defined dosing schedules in autochthonous and orthotopic tumor models. Orthotopic pancreatic tumor implantation was performed using established surgical procedures in immunocompetent and immunodeficient mice. Tumor growth, therapeutic response, and survival were evaluated by tumor weight, histopathology, and molecular analyses. Immune contributions to treatment efficacy were examined using CD4⁺ and CD8⁺ T cell depletion strategies combined with flow-cytometric profiling of tumor-infiltrating immune populations.

In vitro and ex vivo studies included cell viability, clonogenicity, and drug-sensitivity assays performed in two-dimensional cultures, extracellular matrix-based systems, and pancreatic organoids. NRF2 pathway activation and NQO1 expression were assessed using ARE-luciferase reporter assays, quantitative RT-PCR, immunoblotting, and immunohistochemistry. Transcriptomic analyses of human PDAC datasets were conducted using publicly available TCGA and GEO resources. Statistical analyses were performed using appropriate parametric or nonparametric tests, with data presented as mean ± SD or SEM, as indicated. Detailed experimental procedures are provided in the SI Appendix, SI Materials and Methods.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

No custom code was generated during the course of this study. All relevant package and software information is provided in Materials and Methods. Reagents developed for this study are not subjected to any availability restriction, as long as they have not been exhausted.