The extracellular matrix drives guanylate production and protects pancreatic cancer cells from oxaliplatin-induced DNA damage Authors: Georgios Efthymiou, Eugénie Lohmann, Claudio Montenegro, Paraskevi Kousteridou, Pierre Bertrand, Julia Pereira, Heather S. Carr, Issam Ben-Sahra, Coralie Compare, Ivan Nemazanyy, Ghislain Bidaut, Pascal Finetti, François Bertucci, Christophe Lachaud, Fabienne Guillaumond, Richard Tomasini, Sophie Vasseur Published: 2025 ABSTRACT The excessive production of extracellular matrix (ECM) and the metabolic adaptations in pancreatic ductal adenocarcinoma (PDAC) contribute individually to enhanced chemoresistance, marked tumor progression, and dismal patient survival. However, ECM-driven metabolic alterations that promote chemoresistance in PDAC are so far unexplored. Here, we use in vitro–generated ECM bioscaffolds that recapitulate cell-ECM interactions and induce broad metabolic alterations in PDAC cells. High-throughput integration of multiomics datasets coupled with metabolic tracing showed that the ECM enhances the generation of guanylates in PDAC cells, the accumulation of which alleviates oxaliplatin-induced DNA damage, and boosts PDAC cell proliferation. These events are guided by the guanosine monophosphate (GMP)–producing enzymes inosine monophosphate dehydrogenase (IMPDH) and guanosine monophosphate synthase (GMPS), the expression of which correlated with that of matrisomal and DNA repair genes in samples from patients with PDAC. We propose that targeting ECM-driven metabolic processes, such as the enhanced IMPDH activity, may be an effective therapeutic approach for patients with PDAC to bypass the negative side effects of direct targeting of the ECM itself. INTRODUCTION Pancreatic ductal adenocarcinoma (PDAC) is an inherently aggressive disease characterized by strong resistance to current therapeutic schemes. The extracellular matrix (ECM) greatly contributes to drug resistance and PDAC progression, resulting to a dismal 5-year relative survival of 13.3% (1). Upon activation by tumor-cell–derived signals, myofibroblastic-like cancer-associated fibroblasts (myCAFs) become the prominent ECM producers within the tumor, generating excessive amounts of collagen, fibronectin (FN), laminin, and hyaluronic acid (2–5). This process, known as desmoplasia (2, 6), plays a central role in the formation of a dense and rigid “activated” stromal compartment around the tumor cells (6) that accounts for up to 90% of the tumor mass (7) and is associated with poor prognosis (8). The rigidified ECM forms a barrier around the tumor cells, causes blood vessel collapse, and results in a gradient of oxygen concentration and nutrient availability to the tumor cells (9, 10). Under these unfavorable conditions, PDAC cells deploy metabolic reprogramming that maintains the levels of metabolic intermediates needed for proliferation. Nucleotides are essential components for anabolic processes that support the aggressive behavior of rapidly proliferating PDAC cells (11, 12). Low-differentiated quasimesenchymal PDAC cells preferentially promote aerobic glycolysis to provide the pentose phosphate pathway with the necessary carbon for nucleotide synthesis (13, 14). Similarly, addiction to glutamine provides ample nitrogen for the de novo generation of nucleotides and is correlated with poor prognosis (13). Over the past decade, few studies have focused on the ECM as a driver of cancer cell metabolism. Extracellular protein catabolism has been demonstrated to provide PDAC cells with numerous nutrients under nutrient-limited conditions (15, 16). Our team has shown that PDAC cells degrade collagen from the ECM and use collagen-derived proline to feed the tricarboxylic acid (TCA) cycle, thus promoting tumor cell survival and PDAC growth (17). Similarly, hyaluronic acid, an abundant component of PDAC ECM (18), has been described as a fuel for the hexosamine biosynthetic pathway, supporting PDAC growth (19). Apart from a nutrient pool, the ECM presents a biochemical scaffold, the remodeling of which may drive multiple cellular processes. Remodeling of the hyaluronic acid network was shown to enable the interaction between cell receptors and ECM-bound factors, inducing an acute glycolytic response in PDAC cell lines (20). Whether this resulted in enhanced PDAC tumorigenicity is not known. Cell-ECM interactions and ECM-related cues are numerous, highly complex and depend on the spatial properties of the ECM topology and the residing cells (21–23). Exploration of these interactions will help us understand their effect in PDAC physiology, progression, and chemoresistance. Oxaliplatin (OX) is a major antineoplastic drug used routinely for the treatment of patients with unresectable PDAC in combination with fluorouracil and irinotecan [in a drug cocktail known as FOLFIRINOX (FFX)]. Tumor resistance to OX is a major challenge in patients with PDAC who often develop rapid tumor recurrence with extensive ECM remodeling, in association with pronounced decrease in survival (24–28). This is attributed, at least in part, to the chemotherapy-induced changes in the production and remodeling of the local ECM (24). Neoadjuvant OX-containing chemotherapy induces a marked accumulation of ECM-producing myofibroblasts in colorectal cancer, resulting in a pronounced fibrotic response (29). Similar results were observed in pancreatic cancer postchemotherapy where tumor cells were surrounded by thin, spindle-shaped fibroblasts and mature ECM fibers with chemoprotective properties (30). Despite the established role of the ECM as a conductor of tumorigenicity, and of metabolism as a cancer hallmark (31, 32), ECM-driven metabolism and its effect in PDAC chemoresistance are uncharted territories. Standard cell culture studies often use nonphysiologically relevant or variable ECM formulations (coatings and soluble) that may mask the full range of ECM-relayed cues [reviewed in (33)]. To address this issue, we implemented a straightforward, highly robust in vitro model consisting of the culture of PDAC cells on CAF-derived ECM scaffolds that mimic the ECM compartment of PDAC tumors. Using this biologically relevant system, we performed a high-throughput bioinformatic integration of RNA sequencing (RNA-seq) and metabolomics, combined with steady-state metabolic tracing in PDAC cells. We show that the ECM favors the production and accumulation of guanylates in PDAC cells, and this is closely linked with their enhanced capacity to repair OX-induced DNA lesions and maintain a high rate of proliferation. We identified Impdh and Gmps as the enzyme(s) responsible for the accumulation of guanylates and for the capacity of cancer cells to repair OX-induced DNA lesions. This study puts forth the presence of an ECM–metabolism–DNA repair axis, which may be specifically targeted in combination with standard chemotherapy regimens, paving the way for the development of tailored therapeutic strategies against PDAC. RESULTS CAF-derived ECM induces purine imbalance in PDAC cells To understand ECM-modulated metabolism in PDAC, we implemented a cell-on-matrix in vitro system where primary murine tumor cells (hereafter PK4A), isolated from the PDAC of a KIC (Pdx-Cre;LSL-KrasG12D/+;Ink4a/Arfflox/flox) mouse (34), were cultivated on the ECM produced by CAFs derived from patients with PDAC (Fig. 1A). This CAF-derived matrix (CDM) represents a bioscaffold with structural and topological features that mimic PDAC ECM (35) and recapitulates cell-ECM interactions [(36, 37) and references there]. Within this setting, we used human plasma–like medium (HPLM), which unlike synthetic formulations, contains physiologically relevant concentrations of nutrients (38), allowing us to achieve a more accurate metabolite makeup of PDAC. For the generation of the CDM, first we characterized six immortalized patient-derived CAF populations (hereafter CAFi). All CAFi express similar levels of the inflammatory CAF (iCAF) markers integrin b1 (ITGB1) (39) and platelet-derived growth factor receptor b (PDGFRb) (40, 41), and the myCAF marker alpha smooth muscle actin (aSMA) (40), suggesting that these cells have a mixture of iCAF and myCAF features (fig. S1A). However, CAFi #1, #2, and #3 express considerably more PDPN (podoplanin), a protumoral myCAF marker (40), and generate rich, homogeneous FN networks of similar thickness, unlike CAFi #4, #5, and #6 that barely express any PDPN and demonstrate irregular FN fibrilogenesis with too short, aggregated, or fragmented fibrils (fig. S1, A and B). These results suggest that CAFi #1, #2, and #3 correspond to a protumoral myCAF phenotype (42), the main contributor of ECM deposition and assembly in PDAC (6). Among the three myCAFi populations, CAFi #1 and #2 produce and secrete less FN (fig. S1C) and present a high proliferative capacity (fig. S1D) compared to CAFi #3, in line with the thin and dense FN fibrils they assemble (fig. S1B). In contrast, CAFi #3 displays strong FN expression and secretion (fig. S1C) and moderate proliferation (fig. S1D). This results in an elaborate FN network with both thin long fibrils and coarse short ones, which correspond to FN deposition and fibril maturation, respectively. Thus, CAFi #3 better recapitulates the physiological features of myCAF in tissues in terms of proliferation and ECM production (43, 44) and was therefore selected for the generation of the CDM throughout the study. CAF-derived ECM drives purine imbalance in PDAC cells. (A) CAFs from biopsies of patients with PDAC were immortalized (CAFi) and cultivated for the generation of CDMs. Murine PDAC cells (PK4A) were seeded on decellularized CDM or on CDM-free tissue culture plates (TCP) for 48 hours in HPLM for RNA-seq and semi-targeted metabolomics. (B) RNA-seq and metabolomics integration in PK4A_1 cells. Dots represent Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, and dot size reflects the number of identified features (genes/metabolites) within that pathway. (C) Heatmap of purine metabolism compounds in PK4A cells on TCP and CDM for 48 hours in HPLM. Data from two independent experiments in two batches of PK4A cells. Compounds are ranked by significance with the –log10p value shown on the right. The dotted line delineates the threshold of significance (−log10p ≥ 1.3). ATP, adenosine 5′-triphosphate; GDP, guanosine 5′-triphosphate; ADP, adenosine 5′-diphosphate; GTP, guanosine 5′-triphosphate. (D) Top: Normalized abundance of guanosine (Guo) and guanine (Gua) in PK4A cells cultivated on CDM or TCP. Data from two independent experiments in two batches of PK4A cells. Unpaired parametric t test. *P ≤ 0.05. Bottom: Purine metabolism. Brown: de novo purine biosynthesis; blue: salvage pathway; pink: degradation pathway. PRPP, phosphoribosylpyrophosphate; AdSA, adenylosuccininc acid. (E) Heatmap of significantly deregulated purine metabolism genes (−log10p ≥ 1.3) identified and extracted from (D) and divided into groups [associated with guanylates only (purple), adenylates only (green), or both (yellow)] and ranked by decreasing –log10p value shown on the right. G, guanylate; A, adenylate. To explore the ECM-driven metabolic adaptations that occur in PDAC cells, we cultivated two biologically independent PK4A (PK4A_1 and PK4A_2) cell populations, each derived from the PDAC of two separate KIC mice, on CDM from CAFi #3 or on CDM-free tissue culture plastic plates (TCP). Using these PDAC-derived primary cells, we set out to avoid the accumulation of genetic and phenotypical alterations frequently occurring in cell lines with continuous passaging. We also limit metabolic reprogramming due to long-term culture in nutrient-rich media. Both PK4A cell populations positively respond to the presence of the matrix by increasing their proliferative capacity when grown on CDM compared to TCP (fig. S1, E to G). Next, we performed RNA-seq and semi-targeted metabolomics in PK4A cells cultivated on CDM and TCP, and we integrated the comparative datasets (CDM versus TCP) into a joint pathway analysis (Fig. 1A). Among the most significantly deregulated metabolic pathways observed in both PK4A cell batches (Fig. 1B, fig. S1H, and tables S1 and S2), purine metabolism had the highest cumulative impact (an index of the positional importance of the matched metabolites within the pathway) and P value (fig. S1I). Guanine (Gua) and guanosine (Guo) were significantly up-regulated in CDM compared to TCP (Fig. 1, C and D, top), while no difference was observed for adenine (Ade) or adenosine (Ado) (Fig. 1C and table S3). Inosine monophosphate (IMP) and inosine (Ino) were also up-regulated in PK4A cells on CDM (Fig. 1C). IMP is converted to guanosine monophosphate (GMP) and adenosine monophosphate (AMP) via the de novo purine biosynthesis, while Ino may give rise to AMP or GMP via multiple steps that include the salvage pathway (Fig. 1D, bottom). This simultaneous increase of IMP, Ino, Guo, and Gua in cancer cells grown on CDM with no concomitant rise of Ado or Ade indicates an imbalance in cellular purine pools, suggesting a shift toward guanylate production. Half of the significantly deregulated purine metabolism genes (table S4) are involved in the production and turnover of guanylates, while only a few are associated with adenylates or both (Fig. 1E). Among these, guanine deaminase (Gda) down-regulation may impede guanylate degradation in CDM-grown PK4A cells, while the up-regulation of inosine monophosphate dehydrogenase 1 (Impdh1) with the simultaneous down-regulation of guanosine monophosphate reductase (Gmpr) may promote GMP production [Fig. 1, D (bottom), and E]. Furthermore, adenine phosphoribosyltransferase (Aprt) and adenosine monophosphate deaminase (Ampd3) overexpression in cells cultivated on CDM may deplete adenylate pools for IMP regeneration [Fig. 1, D (bottom) and E]. Notably, the expression of genes involved in pyrimidine metabolism remained unaltered between cells grown on CDM and TCP (fig. S1J and table S5), and no concentration imbalance was observed between uracil- and cytosine-containing compounds in cells grown on CDM (fig. S1K and table S6). Last, the substrate rigidity perceived by the cells grown on TCP and CDM was similar, as indicated by the nuclear localization of the mechanosensor Yes-associated protein 1 under both conditions (fig. S1L). Together, these data suggest that the CDM modulates a mechanosensing-independent metabolic alteration in PDAC cells, where the cellular concentration of Gua-containing compounds is increased. CDM drives guanylate accumulation and OX resistance in PDAC cells Guanylate accumulation has previously been associated with the repair of radiation-induced DNA damages in cancer cells, thus promoting chemoresistance and cell survival (45). Therefore, we next addressed whether the CDM potentiates the metabolic flux of the de novo purine biosynthetic pathway toward an enhanced production of guanylates to promote chemoprotection of PDAC cells against OX treatment. To that end, we traced the incorporation of glutamine-amide-15N (15N-Gln) in the purines of OX-treated PK4A cells grown on CDM and TCP (Fig. 2A). As expected, the CDM significantly enhanced the incorporation of Gln-derived 15N in GMP (GMP M + 3) upon OX treatment in PDAC cells compared to TCP (Fig. 2B, left), while the OX-mediated decrease in Gln-derived nitrogen incorporation into AMP (AMP M + 2) was similar between CDM and TCP (Fig. 2B, right). This indicates that in OX-challenged PK4A cells, the CDM enhances the overall contribution of the de novo purine biosynthetic pathway toward the production of GMP but not of AMP. Similarly, we determined the contribution of the purine salvage pathway to the production of GMP and AMP using U-13C-hypoxanthine (U-13C-Hxn) (Fig. 2A). We observed that Hxn-derived carbon incorporation in IMP (IMP M + 5) was considerably lower in OX-treated PDAC cells grown on CDM compared to OX-treated cells on TCP (Fig. 2C, left). In addition, OX addition decreased the incorporation of Hxn-derived carbon into AMP (AMP M + 5) into cells grown on either substrate (Fig. 2C, center). The M + 5 fraction of GMP remained high in CDM-grown OX-treated cells (Fig. 2C, left), indicating that despite the reduced Hprt (hypoxanthine-guanine phosphoribosyltransferase)–mediated purine salvage, GMP production remains intact in these cells. Together, these results suggest that the CDM enhances the rate of IMP-to-GMP conversion in OX-treated PDAC cells, a process mediated by Impdh and Gmps (GMP synthase). Consistent with the tracing results, both isoforms of Impdh (Impdh1 and Impdh2), the enzyme that catalyzes the conversion of IMP to xanthosine monophosphate (XMP) (Fig. 2A), were significantly up-regulated in OX-treated PK4A cells in a CDM-dependent manner (Fig. 2D). Notably, this increase was lost in OX-treated cells cultivated with a homogenized CDM (hCDM) preparation (Fig. 2D, dotted bars). This highlights the role of the intact topological properties of the ECM in the transcriptional regulation of genes involved in purine metabolism. On the other hand, the expression of adenylosuccinate synthase (Adss) and adenylosuccinate lyase (Adsl), involved in the AMP-producing arm of the de novo purine biosynthesis, remained unchanged under all conditions tested (fig. S2A). These data demonstrate that the ECM strongly favors the activity of the GMP-generating arm of the de novo purine biosynthetic pathway in OX-challenged PK4A cells grown on CDM, thus contributing to the overrepresentation of guanylates in these cells. While the addition of OX resulted, as expected, in a significant 40% decrease in the number of viable cells grown on TCP, this decrease was blunted to 20% by the presence of the CDM in PK4A cells (Fig. 2, E and F). Similar results were observed upon validation of the data in a second batch of PK4A cells (fig. S2, B and C). This protective feature of the CDM against OX was lost when PK4A cells were cultured with hCDM (Fig. 2E and fig. S2B, dotted bars). These results suggest that the spatial distribution and high-order assembly of the ECM components trigger protumoral cellular processes, including the observed shift in purine production and chemoprotection. CDM drives GMP production and OX resistance in PDAC cells. (A) Gln-derived 15N, Hxn-derived 13C incorporation into purines via the de novo (brown), and the salvage pathway (blue), respectively. (B) Bar graphs demonstrating the GMP M + 3 and AMP M + 2 fractions in PK4A cells cultivated on CDM and TCP for 3 hours in HPLM containing OX (0.5 μM). Mean ± SD from two batches of PK4A cells. (C) Bar graph demonstrating the (M + 5) fractions of IMP, GMP, and AMP in PK4A cells cultivated on CDM and TCP for 3 hours in HPLM containing OX (0.5 μM). Mean ± SD from two batches of PK4A cells. (D) Impdh1 and Impdh2 mRNA expression in PK4A cells grown on CDM, hCDM, and TCP in HPLM with OX (0.5 μM) for 48 hours. Data represented as mean fold change (FC) relative to TCP (TCP + OX versus TCP; CDM + OX versus TCP + OX; hCDM + OX versus TCP + OX). (E) Left: Bar graph of the mean ± SD number of viable PK4A_1 cells grown on TCP, CDM, and hCDM after 48 hours in HPLM containing OX (0.5 μM). Right: Percent reduction of the number of viable PK4A_1 cells on TCP or CDM due to the presence of OX compared to the respective nontreated (NT) control. (F) Phase contrast images of PK4A_1 cells cultivated on TCP, CDM, and hCDM for 6 hours (top) or 48 hours (bottom) in HPLM with OX (0.5 μM). Scale bar, 50 μm. h, hours. Data from ≥3 independent experiments. Statistics: repeated-measures one-way analysis of variance (ANOVA) with Tukey’s multiple-comparison test, with a single pooled variance in [(B) and (C)]; ordinary one-way ANOVA in (D); ordinary one-way ANOVA, with Tukey’s multiple-comparison test, with a single pooled variance in (E) (left); unpaired parametric t test in (E) (right). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, and n.s. (not significant). To further confirm the impact of CDM-driven production and accumulation of guanylates on PK4A cell proliferation, we used mizoribine, a potent inhibitor of Impdh and Gmps (Fig. 1D, bottom). As expected, the addition of mizoribine to PK4A cells upon seeding markedly decreased the number of viable cells grown either on CDM or TCP (fivefold and threefold decrease compared to corresponding nontreated cells, respectively) (Fig. 3A). The addition of either GMP or Guo reversed the effect of mizoribine independently of the substrate, with CDM-grown cells displaying a stronger rescue effect (Fig. 3A). Next, we cultivated PK4A cells on CDM and TCP for 24 hours to allow for guanylates to accumulate in cells grown on CDM, followed by mizoribine addition to inhibit the GMP-generating branch of the de novo purine biosynthetic pathway for another 24 hours (Fig. 3B). As expected, the addition of mizoribine resulted in 51% decrease in the number of viable PK4A cells grown on TCP, while PK4A cells grown on CDM were less sensitive to Impdh/Gmps inhibition (~28%) and reached a number comparable to that of control TCP (Fig. 3B). These data suggest that during the first 24 hours, the CDM supported the accumulation of guanylates, which subsequently significantly counteracted the inhibitory effect of mizoribine and maintained the proliferative capacity of these cells. The CDM-mediated guanylate accumulation is further supported by the significantly lower levels of Gda (Fig. 3C) that converts Gua to xanthine and eventually to allantoin (Alla), the final product of purine degradation in the mouse (Fig. 2A). Hxn-derived 13C incorporation in Alla was lower in CDM- compared to TCP-grown cells (Fig. 3D). CDM promotes guanylate production and accumulation in PDAC cells. (A) Experimental design and bar graph of the mean ± SD of the number of PK4A cells grown on TCP or CDM after 48 hours in HPLM with mizoribine (Mizo; 2.5 μM), GMP, and Guo (20 μM each), added upon cell seeding. (B) Left: Experimental design and bar graph of the mean ± SD of the number of PK4A cells grown on TCP and CDM for 24 hours in HPLM with Mizo (2.5 μM) after a 24-hour drug-free incubation. Right: Percent reduction of the number of viable cells grown on TCP or CDM due to Mizo compared to the respective NT control. (C) Western blot showing the decreased expression of Gda in PK4A cells grown on CDM compared to TCP for 48 hours in HPLM. b-Tubulin as loading control. The numbers beneath the blots express the mean FC ± SD relative to TCP. (D) Ratio of the 13C-containing fractions of Alla in PK4A cells cultivated on CDM and TCP for 3 hours in HPLM. Mean ± SD from two batches of PK4A cells. (E) Forest plot showing the strong correlation between the matrisome metagene (214 matrisomal genes) and IMPDH1 expression in 938 samples from patients with PDAC. Odds ratio (box) indicated with confidence interval (line). P value of the logistic regression was estimated by specifying a binomial family for model with a logit function. CI, confidence interval. Data from ≥3 independent experiments. Statistics: ordinary two-way ANOVA, with Tukey’s multiple-comparison test, with individual variances computed for each comparison in [(A) and (B)] (left); and unpaired parametric t test in [(B) right], (C), and (D). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, and ****P ≤ 0.0001; not significant comparisons were omitted for clarity. Together, these results demonstrate that the increased cellular guanylate pools in PDAC cells are due, in part, to the ECM-controlled enhancement of the GMP-generating arm of the de novo purine biosynthetic pathway, through the activity of Impdh and/or Gmps, together with the decrease of Gua degradation. Analysis of gene expression data from 938 primary PDAC samples showed that a high expression of a set of 214 matrisomal genes was strongly associated with high IMPDH1 expression in patients with PDAC (Fig. 3E). These data show a strong correlation between the expression of a key guanylate-producing enzyme and ECM-related genes, suggesting that IMPDH-mediated guanylate production may be associated with specific features of PDAC ECM. CDM promotes the repair of OX-induced DNA lesions in an Impdh/Gmps-dependent manner We showed that the CDM promotes the production and accumulation of Gua -containing compounds in PDAC cells while protecting them from OX treatment. OX is a platinum-based compound that binds and forms adducts between contiguous Gua residues in the DNA of rapidly proliferating cells, causing extensive DNA damage, which, in turn, results in polymerase stalling, replication stress, and, ultimately, to apoptotic signaling and cell death [reviewed in (25, 46)]. To assess the capacity of PK4A cells to repair OX-induced DNA damage when cultivated on CDM, we performed cytometry-based quantification of gH2AX intensity, a phosphorylated form of H2A histone family member X, which is an indicator of DNA double strand breaks (47). As expected, addition of OX in TCP-grown PK4A cells resulted in a significant increase in gH2AX intensity compared to nontreated cells, which was significantly alleviated in OX-treated PK4A cells grown on CDM (Fig. 4A). To assess whether the ECM controls the expression of genes involved in the repair of DNA damages, we performed expression analysis of the hallmark DNA repair gene set (48) in PK4A cells grown on CDM versus TCP. A total of 42% of the gene set (22 of 52 genes) displayed significantly up-regulated expression in PK4A cells grown on CDM compared to those cultivated on TCP, while the rest remained unchanged (Fig. 4B and table S7). To associate these 22 deregulated genes with cellular functions, we performed enrichment analysis, which revealed that they are involved in nucleotide excision repair (NER), the main pathway of the identified Gene Ontology term “UV damage removal” (Fig. 4C and table S8) (49). NER is a mechanism implemented by cancer cells to eliminate DNA lesions induced by chemotherapeutic agents, including platinum salts (50, 51). Key genes involved in NER were significantly up-regulated in OX-treated PK4A cells specifically grown on CDM, including the endonuclease component Ercc1, the transcription factor Gtf2h1, the lesion recognition component Rad23, and the polymerases Pold4 and Polb (Fig. 4D). In addition, the NER negative regulator Parp9 was down-regulated in these cells (Fig. 4D). These results are in line with previous works reporting that cancer cells acquire resistance to OX by promoting NER [reviewed in (26)]. Because OX alone did not alter the expression of any of these genes in cells grown on TCP (Fig. 4D), our findings suggest the critical role of the ECM as a driver of OX resistance in PDAC. CDM drives repair of OX-induced DNA lesions in an IMPDH/GMPS-dependent manner. (A) Mean ± SD of gH2AX intensity in PK4A cells grown for 48 hours on TCP and CDM in HPLM with OX (0.5 μM) added at cell seeding. (B) Heatmap showing 22 of 52 DNA repair genes in PK4A cells grown on CDM and TCP ranked by their –log10p value. (C) Pathway score determined by enrichment analysis using the genes in (B). UV, ultraviolet; ER, excision repair. (D) mRNA expression of NER genes in PK4A cells grown on CDM and TCP as in (A). Mean FC ± SD relative to TCP (TCP + OX versus TCP; CDM + OX versus TCP + OX). (E) Experimental design and mean ± SD of gH2AX intensity in PK4A cells on TCP and CDM in HPLM containing Mizo (2.5 μM) and OX (2 μM) for 24 hours, after a 24-hour drug-free incubation. Rescue with GMP, Guo, or Gua. (F) Left: Mean ± SD of the number of PK4A cells on TCP and CDM as in (E). Right: Percent reduction of the number of viable cells on TCP and CDM due to OX ± Mizo compared to respective NT control. (G) mRNA expression of NER genes in MIA PaCa-2 grown for 48 hours on CDM and TCP with OX (5 μM) after a 24-hour drug-free incubation. Mean FC ± SD relative to TCP. (H) Experimental design and mean ± SD of gH2AX intensity in MIA PaCa-2 on TCP and CDM as in (G). Rescue with GMP. (I) Mean ± SD number of viable MIA PaCa-2 on TCP and CDM as in (H). Data from ≥3 independent experiments. Statistics: ordinary two-way ANOVA, with Fisher’s least significant difference (LSD) test, with individual variances computed for each comparison in [(A), (E), (F) (left), (H), and (I)]; ordinary one-way ANOVA in (D); and ordinary two-way ANOVA with Tukey’s comparison test, with individual variances computed for each comparison in [(F) right] and (G). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, and ****P ≤ 0.0001; not significant comparisons are omitted for clarity unless specified: n.s. (not significant). We next addressed the association between Impdh/Gmps activity (and the subsequent guanylate production) and the capacity of PDAC cells to repair OX-induced DNA lesions by measuring gH2AX intensity in PK4A cells cultivated on TCP or CDM in the presence of both OX and mizoribine over the course of 48 hours. Simultaneous addition of the drugs upon PK4A seeding resulted in a similar and marked increase in gH2AX intensity 24 hours later (fig. S3A), highlighting the crucial role of a functional Impdh/Gmps duo and of sufficient guanylate pools for inherent DNA repair in these cells. To evaluate whether the CDM-driven accumulation of guanylates can alleviate OX-induced DNA damage, we cultivated PK4A cells for 24 hours on TCP or CDM and then simultaneously treated with OX and mizoribine (Fig. 4E). gH2AX intensity was significantly increased in cells grown on TCP upon treatment with mizoribine/OX compared to OX alone, and the addition of any Gua -containing compound fully restored basal gH2AX levels, underlining the prominent role of guanylates in DNA repair. In contrast, gH2AX intensity was considerably lower in OX-treated PK4A cells cultivated on CDM compared to TCP irrespective of the presence of mizoribine or the exogenous addition of guanylates (Fig. 4E). Consistently, the levels of Ercc1, a key element in NER, were unchanged in PK4A cells grown on TCP but remained constitutively up-regulated in PK4A cells grown on CDM even in presence of mizoribine (fig. S3B). In accordance with the above, the number of viable PK4A cells grown on CDM was substantially less affected by the combined treatment with mizoribine and OX, contrary to those cultivated on TCP, which displayed a markedly decreased proliferative capacity after OX alone or combined with mizoribine (Fig. 4F). Similar results were obtained when PK4A cells were cultivated on the CDM generated by CAFi #2 (fig. S3, C and D), validating that myCAF produce an ECM that confers enhanced DNA repair capacity and proliferation to PK4A cells under the challenging conditions of OX by favoring guanylate accumulation. This is in accordance with the increased phosphorylation levels of p70 S6 kinase and of Akt in cells grown on CDM compared to TCP despite the presence of OX (fig. S3E), suggesting active proliferative signaling. Last, when we challenged the human pancreatic cancer cell line MIA PaCa-2 with OX, we confirmed that the chemotherapy strongly exacerbated the transcriptional activation of the DNA-repair–implicated genes GTF2H1, POLD4, and POLB in the cells grown on CDM (Fig. 4G). Although the OX-induced DNA damage content was similar in MIA PaCa-2 cells grown on TCP and CDM (Fig. 4H), the addition of mycophenolic acid (MPA; an alternative to mizoribine) resulted in further generation of DNA damages in MIA PaCa-2 cells grown on TCP (Fig. 4H), leading to lower cell viability (Fig. 4I), which was subsequently reversed with GMP supplementation (Fig. 4, H and I, and fig. S3F). In contrast, the CDM-driven accumulation of guanylates before OX treatment and IMPDH inhibition prevented further DNA damage in MIA PaCa-2 cells grown on CDM (Fig. 4H) and resulted in maintained cell proliferation (Fig. 4, H and I, and fig. S3F). Together, these results demonstrate that the CDM-induced accumulation of guanylates in PDAC cells limits OX-induced DNA damage, counteracts the inhibition of GMP production, and supports their proliferative potential. They further indicate that depletion of tumor cells from guanylates may be an attractive strategy to sensitize PDAC to chemotherapy. Guanylate metabolism couples ECM remodeling with DNA repair in PDAC To address the influence of the accumulating desmoplasia on purine metabolism and DNA repair throughout cancer progression, we performed differential gene expression analysis on whole PDAC from 4-, 6-, and 9-week-old KIC mice, which correspond to early, intermediate, and late stages of tumor development, respectively, compared to the pancreases from age-matched healthy littermates (Fig. 5A). We found that a considerable number of matrisomal genes were significantly deregulated in KIC pancreases compared to healthy ones as early as 4 weeks of age (91 genes up and 4 down; Fig. 5B and table S9), suggesting that the alterations of the pancreatic tissue on the molecular level occur before the appearance of anatomopathological structures associated with tumoral features (52). At intermediate-stage tumors (6 weeks), similar number of genes were significantly deregulated as at 4 weeks. In advanced disease stage (9 weeks), however, the number and level of deregulation of differentially expressed matrisomal genes in PDAC compared to healthy pancreas increased considerably (Fig. 5B and table S9), in line with the prominent ECM deposition at that stage (fig. S4A) (17). Histological examination showed initial signs of tissue transformation in 4-week-old mice, followed by the appearance of multiple tumoral glands at 6 weeks, and pronounced desmoplasia surrounding the tumor cells in 9-week-old mice (fig. S4A). Genes involved in purine metabolism and in DNA repair displayed the highest deregulation in late-stage PDAC compared to the tumor at early and intermediate stages (Fig. 5, C and D, and tables S10 and S11). Analysis of the deferentially expressed DNA repair genes in advanced PDAC compared to healthy tissue revealed strong enrichment of NER-related genes (Fig. 5E and table S12), including Ddb1, Ercc1, and Pold4 that were also modulated by the desmoplasia-mimicking CDM (Fig. 4, B and D). To confirm the clinical significance of our findings, we performed a compartment-specific assessment of key players identified in our study. Analysis of single-cell transcriptomics datasets from patients with PDAC [(53) and table S18] revealed that myCAFs are indeed the main producers of desmoplasia in PDAC, as indicated by the strong expression of FN1 and COL1A1 (Fig. 5F) by this CAF subtype. Cancer cells, however, annotated as “epithelial cells” in these patient samples, present high levels of the guanylate producing IMPDH1, IMPDH2, and GMPS (Fig. 5G, top, and fig. S4B, left). In addition, these cells also express high amounts of transcripts involved in NER, such as ERCC1, RAD23A, POLB, and POLD4 compared to the other cell types (Fig. 5G, bottom, and fig. S4B, right). High mRNA expression of the guanylate producing enzymes IMPDH1, IMPDH2, and GMPS in PDAC strongly correlates with high expression of DNA repair genes (Fig. 5H). Last, patients with PDAC and high expression of both COL1A1 and IMPDH1 present a very poor survival compared to those with low expression of either gene individually or combined (Fig. 5I). These results suggest that strong desmoplasia in conjunction with enhanced guanylate production are markers of bad prognosis for patients with PDAC. Collectively, our data show that concomitantly with ECM production by CAF in the PDAC tumor microenvironment (TME), tumor cells activate the expression of genes involved in DNA repair and guanylate production to enhance NER via adaptations in purine metabolism. This promotes PDAC aggressiveness and is associated with poor survival of patients with PDAC. Guanylate metabolism couples ECM remodeling with DNA repair in PDAC. (A) Experimental plan followed to assess the transcriptomic alterations in KIC mice during PDAC progression (at 4, 6, and 9 weeks of age) compared to age-matched healthy animals. w, week. (B) Graph showing the logFC of differentially expressed matrisomal genes in 4-, 6-, and 9-week-old KIC versus healthy mice. Each dot represents a differentially expressed gene. Red: up-regulated genes; blue: down-regulated genes. (C) Graph showing the logFC of differentially expressed purine metabolism genes in 4-, 6-, and 9-week-old KIC versus healthy mice as in (B). (D) Graph showing the logFC of differentially expressed DNA repair genes in 4-, 6-, and 9-week-old KIC versus healthy mice as in (B). (E) Pathway score determined by enrichment analysis using the genes extracted from week 9 shown in (D). Relevant pathways: red; full list in table S12; UA, uric acid; CUL4/CRL, cullin 4/RING E3 ubiquitin ligase; SS, single-strand; BER, base excission repair. (F) Dot plot demonstrating the high expression of FN1 and COL1A1 (collagen type I alpha 1 chain) mRNA particularly by myCAF in patients with PDAC. Dot size represents the percentage of cells expressing the gene, and color intensity indicates average expression. NK, natural killer. (G) Heatmap of IMPDH1 and ERCC1 mRNA expression levels across cell types in 11 patients with PDAC determined by single-cell transcriptomics. Both genes are mainly expressed by tumor cells, annotated here as epithelial cells. (H) Forest plot representation of the correlation between the DNA repair metagene and IMPDH1, IMPDH2, and GMPS expression classes in 938 samples from patients with PDAC as in Fig. 3E. (I) Kaplan-Meier survival curve representing the disease-free survival (DFS) in patients with PDAC stratified by combined IMPDH1 and COL1A1 expression classes: COL1A1high + IMPDH1high (purple), COL1A1high + IMPDH1low (green), COL1A1low + IMPDH1high (blue), and COL1A1low + IMPDH1low (orange). Time is shown in months after diagnosis, and the 2-year survival rate is indicated. P values were obtained using the log-rank test. y, year. DISCUSSION In this study, we established a robust and highly customizable in vitro model using the ECM produced by CAF derived from PDAC patients. This approach involves the devitalization of CAF monolayers and the use of unabridged ECM substrates that maintain their three-dimensional (3D) structure, fiber topology, and secreted factor bioavailability. This system highlights the role of the ECM composition/structure on cellular responses that may go undetected in experimental settings that overlook the topological features of native ECM (e.g., soluble ECM components, gels, and coatings) (33). For the generation of the CDM, we used CAFi that express markers of activated myofibroblasts to simulate the ECM generated by myCAF, the major matrix producers and/or remodelers in PDAC (2–5). Although a comprehensive study of the ECM-induced cellular responses as a function of the CAF subtype is beyond the scope of this work, using two separate PDAC-derived myCAFi populations, we demonstrate that myCAF-derived matrices induce similar biological effects in PDAC cells in terms of proliferative potential and DNA repair (Fig. 4, E and F, and fig. S3, C and D). A recent study revealed that CAF-subtype–dependent ECM heterogeneity influences patient prognosis in gastric cancer (54). Given the wide range of components present in the tumor ECM (55), we consider that the effects we observed are the combined outcome of a multitude of signaling and metabolic processes rather than individual responses induced by a single component (e.g., collagen). These responses are further modulated by the topological structure and distribution of the ECM fibers, as the presentation of disorganized ECM components (hCDM) resulted in loss of cellular responses. The importance of the ECM in PDAC aggressiveness has previously been emphasized by the chemoprotection the desmoplasia confers to tumor cells by forming a protective shield around them, preventing drug penetrance (56, 57). However, whether and how the ECM biochemically contributes to genotoxic stress resistance remain poorly understood. Tumor cell adhesion to FN, a major ECM component with multiple isoforms in solid tumors (44), alleviated radiation-induced genotoxic injury in squamous cancer cells (58), but the underlying mechanism is unknown. Collagen, the most abundant component of the noncellular compartment of the PDAC TME has a dual role. On the one hand, PDAC cells enhance their proliferative capacity by feeding the TCA cycle with collagen-derived proline acquired from the TME under nutrient stress (17). On the other hand, collagen has been shown to confer chemoresistance to 2D- and 3D-cultured PDAC cells by reducing apoptosis (59, 60) and by enhancing the DNA damage response (61). The biomechanical and biochemical action of PDAC ECM in chemoresistance was recently highlighted when PDAC cell lines were less sensitive to FFX treatment when cultivated in decellularized tissue derived from patients with PDAC (62). Although the underlying mechanisms are not yet elucidated, studies similar to this stress the functional role of the ECM as a driver of protumoral processes (e.g., chemoresistance) that, in turn, contribute to PDAC aggressiveness. Here, we demonstrate that ECM-modulated metabolism blunts drug sensitivity in PDAC. More specifically, we demonstrate that PDAC ECM induces an adaptation in purine metabolism in PDAC cells, which fosters PDAC cell survival despite the presence of OX. This is mediated via two routes: (i) a metabolic shift in purine metabolism favoring the GMP-generating arm of the de novo purine biosynthetic pathway at the expense of AMP production and (ii) the down-regulation of guanylate degradation. Together, these routes result in the accumulation of Guo and Gua. Our transcriptomics and metabolomics data have revealed large-scale alterations in gene expression and the metabolic profile of PDAC cells induced by CDM. Because the hCDM did not influence Impdh1 and Impdh22 expression, nor did it protect cells from OX, we expect that the observed responses were not due to the use of the CDM as a mere nutrient pool by the tumor cells. Rather, we consider that specific and highly complex interactions between topologically structured ECM components (e.g., collagens, FN, and secreted factors) and cell receptors induce signal transduction that modulates the observed gene expression profiles and metabolic adaptations. However, to our knowledge, no specific extracellular cues are known to induce alterations in purine metabolism. Selective inhibition of key cell-ECM interactions (e.g., blocking of the cell-binding domain of the a5b1 integrin to inhibit FN-mediated integrin signaling) coupled with metabolomics could provide mechanistic links between ECM-relayed cues and metabolic adaptations, as well as novel therapeutic targets. Our study demonstrates that the CDM potentiates the removal of OX-induced DNA lesions via the simultaneous up-regulation of NER gene expression and the maintenance of cellular guanylate pools. This reveals a strong dependency of PDAC cells on guanylate production to counteract the deleterious effect of chemotherapy. The sensitization of tumor cells to genotoxic stress under guanylate-limited conditions has so far only been described in glioblastoma (GBM). In that setting, GBM cells acquired resistance to DNA-damage–inducing radiotherapy by driving the generation of guanylates that, in turn, promoted DNA repair mechanisms and proliferative signaling (45, 63). In the same study, the authors found a decrease in the survival probability in patients with GBM and high expression of IMPDH1 (45), as we observed in patients with a highly desmoplastic PDAC together with high expression of guanylate production enzymes. The practical significance of our study lies in the discovery that an ECM–metabolism–DNA repair axis is involved in PDAC cell resistance to OX, one of the three components of FFX. Despite its strong cytotoxic effects, a major factor that contributes to the failure of FFX is the acquired resistance of PDAC cells via the up-regulation of DNA damage response mechanisms (64, 65). Complementary to our in vitro data in murine and human PDAC cells, our in vivo transcriptomics demonstrate a synchronous deregulation of matrisomal, purine metabolism, and DNA repair genes, as the tumor progresses toward an advanced desmoplastic stage, which strengthens the association of ECM deposition with purine metabolism and DNA repair. Our hypothesis is also corroborated by the single-cell–scale analysis, demonstrating the strong expression of genes involved in purine metabolism and NER by tumor cells and of matrisomal genes by myCAF in patients with PDAC. It is also supported by the strong correlation between the expression of IMPDH1, IMPDH2, and GMPS with that of DNA repair genes in patients with PDAC. A study using the GMP-depleting compound MPA, or its derivative mycophenolate mofetil, showed a moderate decrease in the size of xenografts derived from patients with PDAC (PDX), but no significant antitumoral effects were observed when administered in a small cohort of six patients with PDAC (66). Although clinically approved for the prevention of rejection in renal transplantation (67), mizoribine has not yet been tested in PDAC mouse models or PDAC clinical trials. Because desmoplasia promotes the accumulation of guanylates via the enhanced action of Impdh/Gmps, we propose that a suitably timed pretreatment of patients with PDAC and mizoribine to deplete PDAC cells from the previously accumulated guanylates will sensitize them before the administration of platinum-based chemotherapy. In this frame of administration, mizoribine repurposing may be an appealing approach worthy of further investigation. MATERIALS AND METHODS Materials and reagents All materials and reagents were from Sigma-Merck (St. Louis, MO) and Euromedex (France) unless otherwise specified. Cell culture, live cell imaging, proliferation, and cell counting CAFi were generated with human telomerase reverse transcriptase overexpression as described before (68) and cultivated in Dulbecco’s modified Eagle’s medium F12 (DMEM F12) (Thermo Fisher Scientific, Waltham, MA), supplemented with 10% fetal bovine serum (Biosera, France), 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B (Thermo Fisher Scientific). CDMs were generated as described before (69). Briefly, 72,000 cells/cm2 were seeded in tissue culture–treated vessels and cultivated in culture medium supplemented with 50 μg/ml of l-ascorbic acid. Culture medium was replaced every 2 days for a total of 8 days. Matrices were subsequently devitalized (decellularized) with 20 mM NH4OH and 0.1% Triton X-100 in phosphate-buffered saline (PBS), residual DNA was digested with 15 μg/ml of deoxyribonuclease I (Roche, Switzerland) in PBS containing Ca2+Mg2+ 1 mM, and they were subsequently equilibrated in tumor cell medium (see below). Matrix homogenization was performed using a cell scraper to detach the devitalized, DNA-free matrix, which was subsequently passed through a 31-gauge insulin syringe. CAFi #3 was used throughout the study for the CDM generation unless otherwise stated. For CAF proliferation curves, CAFi #1, #2, and #3 were seeded at an initial density of 50,000 cells per well in 24-well plates. The medium was replaced the day after and every 2 days. Cells were enumerated with a Vi-CELL XR Cell Viability Analyzer (Beckman Coulter, Brea, CA) every day for 5 days. Two batches of tumor cells (hereafter PK4A; RRID: CVCL_WB21) were isolated from two male animals of a KIC mouse model as described before (10). Cells were routinely cultivated in Gln-free DMEM (Thermo Fisher Scientific), supplemented with 10% fetal bovine serum, 2 mM Gln (Thermo Fisher Scientific), 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B. For PK4A proliferation curves, cells were seeded at an initial density of 50,000 cells per well in 24-well plates and were enumerated every 24 hours with a ViCell XR Cell Viabilitty Analyzer. MIA PaCa-2 cells were obtained from the American Type Culture Collection (catalog no. CRL-1420, RRID: CVCL_0428) and cultivated in Gln-free DMEM (Thermo Fisher Scientific), supplemented with 10% fetal bovine serum, 2 mM Gln (Thermo Fisher Scientific), 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B. All cells were routinely tested for mycoplasma contamination with the MycoAlert Mycoplasma Detection Kit (Lonza, Switzerland) according to the manufacturer’s instructions. Before all experiments, murine tumor cells were cultivated in HPLM (Thermo Fisher Scientific), supplemented with 10% fetal bovine serum, 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B, for 2 to 5 days. Then, cells were seeded on CDM-coated or CDM-free 12-well plates (105 cells per well) in HPLM supplemented with 10% fetal bovine serum, 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B for 48 hours, unless otherwise stated. Live cell imaging and video microscopy were performed with an Incucyte S3 Live-Cell Imaging and Analysis System (Sartorius AG, Germany), while endpoint cell counting was performed with Vi-CELL XR Cell Viability Analyzer, and the number of viable cells is indicated in the corresponding panels. To assess the ratio (percentage) of the treatment-induced reduction of the number of viable cells relative to the respective untreated control per substrate (TCP and CDM separately), we used the following function: % reduction of viable cells in X condition = (number of viable cells in nontreated control – number of viable cells in condition X)/number of viable cells in untreated control × 100. RNA extraction for RNA-seq and transcriptomics analysis from in vitro cultures For the transcriptomic analysis, PK4A cells were cultivated on CDM-coated or noncoated 12-well plates (105 cells per well) in HPLM for 48 hours. RNA from three independent experiments was subsequently isolated using an RNeasy Plus Kit (QIAGEN, Netherlands), and the RNA integrity number was determined with an RNA analysis kit (Agilent, Santa Clara, CA). mRNA library preparation was performed with 50 ng of RNA per sample using an Illumina Stranded mRNA Prep (Illumina, San Diego, CA) following the manufacturer’s instructions. Samples were sequenced on a NovaSeq 6000 Sequencing System (Illumina) using an S1 Reagent Kit v1.5 (200 cycles) corresponding to 2 × 30 million reads per sample after demultiplexing. For data analysis, we applied a pipeline developed by the Cibi (CRCM Integrative Bioinformatics) platform. A FASTQ data preprocessing tool was used (Fastp version 0.23.2) for first nucleotide elimination (cutoff quality > 21) and adapter removal. Reads were then mapped to the mm10 reference genome using Subread v1.6.4 with an average mapping score > 90% determined with the featureCounts function. Differential analysis was performed with DESeq2 v1.32.0. Only transcripts with more than 20 counts within the same condition were kept. Gene set enrichment analysis was performed using clusterProfiler v4.0.5, and results were visualized using Ridgeplot and DotPlot. Heatmaps were generated with NMF v0.24.0. Other graphs were created with ggplot2 v3.3.5. For the identification of enriched cellular processes based on differentially expressed gene sets or gene subgroups, we used the online tool enrichR (70). The combined score is calculated for each process by multiplying the P value with the z-score. The P value is computed using Fisher’s exact test, while the z-score is computed by assessing the deviation from the expected rank. In vivo studies and PDAC progression transcriptome analysis Male PDAC-bearing mice Pdx1-Cre;LSL-KrasG12D;Ink4a/Arffl/fl (KIC) and control LSL-KrasG12D;Ink4a/Arffl/fl (KI) littermates were obtained and maintained as previously described (10, 34). The experimental study design was performed as described in (52). Mice were euthanized at 4 weeks of age (n = 3 KI mice and n = 3 KIC mice), and pieces of tumor or control pancreas were either fixed in 4% (wt/vol) formaldehyde, snapfrozen in cold isopentane for further analysis, or directly homogenized in 4 M guanidinium isothiocyanate lysis buffer for efficient pancreatic RNA extraction as described before (10, 52). All animal care and experimental procedures were performed in agreement with the Animal Ethics Committee of Marseille under reference (APAFIS#191832019021414576818). For expression data in 6- and 9-week-old KIC mice, we used the previously established datasets GSE127891 and GSE61412, respectively (52). Significantly deregulated genes were extracted from datasets by applying a fold change (FC) cutoff of 1.5 and q ≤ 0.5 in PDAC versus control in each time point (4, 6, and 9 weeks). Histological examination of murine tissue sections Formalin-fixed paraffin-embedded murine PDAC sections (5 μm) were deparaffinized in xylene and rehydrated with serial ethanol dilutions. Hematoxylin and eosin staining was performed using an automated Leica ST5010 Autostainer (Leica Biosystems, Nussloch, Germany). Semi-targeted metabolomics in cell lysates For metabolomics analysis, PK4A cells were cultivated on CDM-coated or noncoated 12-well plates as described in the previous section in three independent experiments in technical duplicates: one for metabolite detection, and the second to determine the cell number per condition and for subsequent normalization. Metabolites were extracted with a 5:3:1 solution of methanol, acetonitrile, and water by adjusting the volume to the cell number (1 ml per 107 cells). The samples were vigorously mixed for 5 min at 4°C and then centrifuged at 16,000g for 15 min at 4°C. The supernatants were subsequently analyzed by liquid chromatography–mass spectrometry (LC-MS) with a Q Exactive Plus Orbitrap mass spectrometer equipped with an Ion Max source and a HESI-II probe coupled to a Dionex UltiMate 3000 uHPLC system (Thermo Fisher Scientific). External mass calibration was performed using a standard calibration mixture every 7 days, as recommended by the manufacturer. The 5-μl samples were injected onto a ZIC-pHILIC column [150 mm by 2.1 mm; inner diameter (i.d.) of 5 μm] with a guard column (20 mm by 2.1 mm; i.d. of 5 μm) (Millipore) for LC separation. Buffer A was 20 mM ammonium carbonate and 0.1% ammonium hydroxide (pH 9.2), and buffer B was acetonitrile. The chromatographic gradient was run at a flow rate of 0.2 μl min−1 as follows: 0 to 20 min—linear gradient from 80 to 20% of buffer B; 20 to 20.5 min—linear gradient from 20 to 80% of buffer B; 20.5 to 28 min—80% buffer B. The mass spectrometer was operated in full scan, polarity switching mode with the spray voltage was set to 2.5 kV, and the heated capillary held at 320°C. The sheath gas flow was set to 20 U, the auxiliary gas flow to 5 U, and the sweep gas flow to 0 U. The metabolites were detected across a mass range of 75 to 1000 mass/charge ratio (m/z) at a resolution of 35,000 (at 200 m/z) with the automatic gain control target at 106 and the maximum injection time at 250 ms. Lock masses were used to ensure mass accuracy below 5 parts per million. Data were acquired with Xcalibur Data Acquisition and Intepretation Software (Thermo Fisher Scientific). The peak areas of metabolites were determined using TraceFinder Software (Thermo Fisher Scientific), identified by the exact mass of each singly charged ion and by the known retention time on the high-performance liquid chromatography (HPLC) column. Metabolomics and joint pathway analysis For metabolomics analysis, the peak areas previously identified were uploaded in csv file format in MetaboAnalyst 5.0 in two groups (TCP and CDM). Values were normalized by sum and range scaled. Univariate analysis including FC analysis and t tests was performed within the software. Joint pathway analysis was performed with MetaboAnalyst 5.0 (https://metaboanalyst.ca/MetaboAnalyst/home.xhtml; accessed on 7 November 2025) using the previously generated transcriptomics and metabolomics datasets. For joint pathway analysis, metabolites with FC ≥ 1.5 and P ≤0.05 and genes with P ≤ 0.05 (0.1 for PK4A batch 2) in CDM versus TCP were selected. The integration method was performed with the application of a hypergeometric test for enrichment analysis using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database on both queries. The pathway impact was computed as the aggregate of the importance measures of the matched metabolites (relative betweenness, centrality, and node degree), normalized by the importance measures of all metabolites within each pathway (71). For pathway selection, a threshold of 1.3 was applied for –log10p val and 1.0 for pathway impact. Targeted metabolic flux analysis For metabolic flux analysis of Gln and Hxn, a “custom-made” HPLM (cm-HPLM) was prepared by combining the method published by Cantor et al. (38) with the official HPLM composition from Thermo Fisher Scientific. PK4A cells were seeded on CDM-coated or noncoated 10-cm dishes (3 × 106 cells per dish) in HPLM supplemented with 10% fetal bovine serum, 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B. OX (Hospira, Lake Forest, IL) was added to the cells at the time of seeding at a final concentration of 0.32 μM to condition the cells without significantly reducing their viability. Cells were then incubated overnight at 37°C in a humidified incubator supplemented with 5% CO2. For amido-15N-Gln flux studies, cells were washed once with Gln-free cm-HPLM and then incubated for 3 hours in cm-HPLM supplemented with 10% dialyzed fetal bovine serum (Thermo Fisher Scientific), 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B, containing 4 mM amido-15N-Gln (Cambridge Isotope Laboratories, Tewksbury, MA) and 0.5 μM OX. For 13C-Hxn flux studies, cells were washed once with Hxn-free cm-HPLM and then incubated for 3 hours in cm-HPLM supplemented with 10% fetal bovine serum, 100 U/ml of penicillin, 100 μg/ml of streptomycin, and 0.25 μg/ml of amphotericin B, containing 100 μM 13C-Hxn (Cambridge Isotope Laboratories) and 0.5 μM OX. The medium was then removed, and cells were incubated with 4 ml of 80% methanol for 20 min at –80°C and were subsequently scraped off the culture dishes and transferred into 15-ml conical tubes. Insoluble material was pelleted by centrifugation at 3000g for 5 min at –4°C, followed by two subsequent centrifugations of the insoluble pellet with an additional 0.5 ml of 80% methanol and centrifugation at 21,000g for 5 min at 4°C. The pooled supernatants were dried under nitrogen gas using an N-EVAP (Organomation Associates, Berlin, MA). Dried pellets were resuspended in 20 μl of HPLC-grade water for mass spectrometry, and 5 μl were injected using a 5500 QTRAP hybrid triple quadrupole mass spectrometer (ABI/SCIEX, Framingham, MA) coupled to a Prominence UFLC HPLC system (Shimadzu Corporation, Japan) with Amide XBridge HILIC chromatography (Waters Corporation, Milford, MA) via selected reaction monitoring (SRM) (72) and polarity switching between positive and negative modes. The parameters for instrumental and software analysis were as described before (73). Peak areas from the total ion current for each metabolite SRM transition were integrated using MultiQuant v2.0 software (SCIEX, Framingham, MA). Custom SRMs were created for expected 15N or 13C incorporation in various forms for targeted LC–tandem MS. RNA extraction and quantitative polymerase chain reaction amplification Cells were seeded in 12-well plates as described above for 48 hours and then lysed, and RNA was isolated with TRIzol (Thermo Fisher Scientific) according to the manufacturer’s guidelines. Reconstituted RNA was quantified with a FLUOstar Omega spectrophotometer (BMG Labtech, Germany). cDNA was generated with a GoScript cDNA generation kit (Promega, France) using random primers starting from 1 μg of RNA. Each quantitative polymerase chain reaction (qPCR) reaction was performed with 5 ng of cDNA in triplicates using a GoTaq SYBR Green qPCR Mix (Promega) in an AriaMx qPCR System (Agilent) using the gene specific primers listed in table S13. Raw fluorescence values were baseline corrected with a passive reference dye (ROX), normalized with the murine TATA box–binding protein (Tbp) or the human ribosomal protein L27a (RPL27A) using the Pfaffle method for primer efficiency correction (74), and represented as FC relative to the indicated control. Western blot in whole-cell extracts and conditioned media Cells were seeded in 12-well plates for 48 hours as described above, then rinsed with ice-cold PBS, and lysed with 3× Laemmli lysis buffer [6% SDS, 30% glycerol, 187.5 mM, and tris-HCl (pH 6.8)]. Lysates were subsequently sonicated with a Pulse 150 ultrasonic homogenizer (Benchmark Scientific, Sayreville, NJ), heated at 95°C for 5 min, and quantified with a BCA Quantification Assay Kit (Thermo Fisher Scientific). Conditioned media were centrifuged (800g for 5 min at 4°C), then mixed with the appropriate volume of 3× Laemmli buffer, and then boiled at 95°C for 5 min. Polyacrylamide gels were cast and 20 to 40 μg of whole-cell lysates, or 30 μl of conditioned medium mix, were loaded per well for electrophoresis and transferred on polyvinylidene difluoride membranes (Merck Millipore, Burlington, MA). Membranes were subsequently blocked with a 5% skim milk in tris-buffered saline (TBS) solution, followed by primary antibody incubation in 3% bovine serum albumin (BSA) in TBS–Tween 0.1% (see antibody list in table S14). A secondary antibody coupled with horseradish peroxidase (Southern Biotech, Birmingham, AL) was used against the primary antibody isotype, followed by an enhanced chemiluminescence reaction (Merck Millipore). Protein bands were visualized in a PXi Multi-Application Gel Imagin System (Syngene, UK), and band densitometry analysis was performed with FIJI (75). Immunofluorescence and imaging For immunofluorescence-based protein detection and microscopy, PK4A cells were seeded on glass coverslips (Paul Marienfeld, Germany) as described above. Cells were then incubated with a fixing solution (4% paraformaldehyde and 3% sucrose in PBS) at 37°C for 20 min, washed with PBS, permeabilized with 0.2% Triton X-100 in PBS for 30 min, blocked with 3% BSA in PBS, and incubated with the respective primary antibodies, followed by secondary antibodies coupled to Alexa Fluor fluorophores (see table S14). Cells were counterstained with Hoechst 33342 (Thermo Fisher Scientific), and coverslips were lastly mounted on glass slides with ProLong Gold anti-fade reagent (Thermo Fisher Scientific). Cell visualization and image acquisition were performed with a Zeiss Axio Imager microscopy system (Zeiss, Germany). Images were treated and analyzed with FIJI. Cytometry For CAF phenotyping, cells were cultivated as described above. Cells were detached using StemPro Accutase cell dissociation reagent (Thermo Fisher Scientific) and washed with PBS. Samples were resuspended in flow cytometry buffer (0.5% BSA and 2 mM EDTA in PBS) and stained with extracellular fluorochrome-coupled antibodies (see table S14) for 20 min at 4°C. Samples were then washed with PBS, and one drop of viability marker SYTOX green (Thermo Fisher Scientific) was added to each sample 5 min before analysis. Single-stained samples for compensation were made by staining UltraComp eBeads Compensation Beads (Thermo Fisher Scientific) with the same protocol as above. Samples were analyzed using a BD LSRFortessa X-20 Cell Analyzer (Becton Dickinson Biosciences, Franklin Lakes, NJ), and data were analyzed using FlowJo software (Becton Dickinson Biosciences). For gH2AX staining, PK4A or MIA PaCa-2 cells were seeded in 12-well plates as described before, with or without OX and/or mizoribine (MPA for MIA PaCa-2) and purine compounds (all from Sigma-Merck) at cell seeding unless otherwise specified for the indicated duration. Cells were then trypsinized, fixed with 3.7% paraformaldehyde in PBS, and incubated with a 1/10 dilution of a BD Pharmingen Alexa Fluor 647 Mouse anti-H2AX, pS139 (Becton Dickinson Biosciences). Cytometry was performed in a BD LSRFortessa X-20 Cell Analyzer, and data were analyzed with FlowJo v.10.10 and expressed as the mean cellular fluorescence intensity. For cell cycle analysis, cells were fixed with 70% ethanol for 30 min at 4°C, centrifuged (850g for 5 min at 4°C), and incubated in a mixture containing 1 mg/ml of ribonuclease I and 1 μg/ml of propidium iodide for 30 min at 37°C. Samples were analyzed using a BD LSRFortessa X-20 Cell Analyzer, and the data were curated using the built-in Cell Cycle module in FlowJo v.10.10. Dataset analysis of patients with PDAC Clinicopathological and gene expression data were collected from 17 publicly available mRNA expression datasets (table S15 and references there) comprising a total of 938 primary PDAC samples. Patients’ informed consent to participate was obtained by the authors of each study, while ethical considerations were overseen by the respective institutional review board. Data preanalytic processing was performed as described before (69). Batch effect correction on genes of interest was performed using z-score normalization with the mean and the SD measured on primary cancer samples of each dataset. Tumor expression of purine genes was measured as discrete values after comparison with median expression in the 938 samples: High expression was defined by value > median and low expression by value ≤ median. Matrisomal and DNA repair genes (tables S16 and S17) were similarly normalized, and both metagenes were computed as the mean expression of all standardized genes from the respective list. A total of 185 primary PDA samples was informative for both disease-free survival (DFS) and delivery or not of adjuvant chemotherapy treatment: A total of 106 had received FFX regimen and 79 had not received any chemotherapy. DFS was calculated from the date of diagnosis to the date of disease recurrence or death. Follow-up was measured from the date of diagnosis to the date of last news for event-free patients. Survivals, calculated using the Kaplan-Meier method, were compared with the log-rank test. All statistical tests were two sided at the 5% level of significance. Statistical analysis was performed using the survival package (version 3.5-5) in the R software (version 4.1.2; www.cran.r-project.org/). Single-cell transcriptomics analysis derived from patients with PDAC Single-cell RNA-seq (scRNA-seq) data from 11 patients with PDAC without any treatment were acquired from a previously published study (53) available through Gene Expression Omnibus under accession number GSE205013. Raw data counts were imported and processed using Seurat (v5) in R (76). Only cells with a gene count between 200 and 10,000 or with a mitochondrial gene content ≤ 15% were included in the analysis, and the SCTransform method was applied to normalize the expression data and correct for technical variation. Raw count matrices were normalized using the NormalizeData() function with the LogNormalize method and a scale factor of 10,000. Data were then scaled to standardize gene expression values across all cells. Principal components analysis was performed with the first 50 principal components computed to capture most transcriptional variation. To identify cellular neighborhoods, a shared nearest neighbor graph was constructed using the FindNeighbors() function, with the first 30 principal components (dims = 1:30) as input. Clustering was performed at a resolution of 1.2 to delineate transcriptionally distinct cell populations. For visualization of the high-dimensional data in two dimensions, uniform manifold approximation and projection was conducted with dims = 1:30, a neighborhood size of 30 (n.neighbors = 30), and a minimum distance parameter (min.dist) set to 0.3 to balance cluster separation and continuity in the projection. For cell type annotation, normalized expression matrices were extracted from the SCT assay, and annotation was performed using the SingleR package (77) with the Blueprint+Encode reference dataset. The most frequent predicted cell type in each cluster was assigned as the representative annotation, which was then incorporated into the Seurat object metadata for downstream visualization and analysis. To discriminate CAF subtypes, ACTA2 (actin alpha 2), matrix metalloproteinase 11 (MMP11), and collagen type X alpha 1 chain (COL10A1) were used to identify myCAFs, while component of complement 7 (C7), complement factor D (CFD), and prostaglandin D2 synthase (PTGDS) were used to assess iCAFs. For pseudobulking analysis, the raw counts were extracted and gathered into a single matrix with the values transformed by log1p. Heatmap plots of gene expression were generated using ComplexHeatmap package (78, 79). Statistics Statistical analyses were performed with GraphPad Prism (GraphPad Software, Boston, MA). For comparisons involving multiple conditions within two groups with similar variance, we used ordinary one-way analysis of variance (ANOVA) with Tukey’s multiple-comparison test, with individual variances computed for each comparison. For more than two groups, we used ordinary two-way ANOVA with Tukey’s multiple-comparison test, with individual variances computed for each comparison. For comparisons with different variance among groups, we used ordinary one- or two-way ANOVA with Fisher’s least significant difference (LSD) test with individual variances computed for each comparison. For all other comparisons, appropriate statistical tests tailored to the specific conditions/analyses are described in detail in the figure legends. Data points represent biological replicates from at least three independent experiments, unless otherwise specified. Experimental versus control group differences were considered significant when P ≤ 0.05. Graphs and charts were created with GraphPad Prism or Excel 365 (Microsoft Corporation, Redmond, WA). For the FC analysis in the murine PDAC progression dataset, statistical differences were assessed with pairwise t test in log-transformed FC data coupled with Bonferroni correction to adjust P values (70).