Induction of a mismatch repair deficient genotype by tailored chemical mutagenesis in experimental models of cancer

SUMMARY

Mismatch repair deficient (MMRd) tumors harbor thousands of somatic mutations enriched for insertion–deletion (indels) conferring high sensitivity to immunotherapy. We sought to reproduce this phenotype using mutagenic agents to engineer an MMRd genotype in immunoresistant cells. The combination of temozolomide (TMZ) and cisplatin led to a rapid accumulation of a high mutational load enriched for indels in murine cell lines resulting from the epigenetic loss of Msh2. Pretreated cells showed sensitivity to PD-1 blockade. Systemic treatment with TMZ, cisplatin, and anti-PD-1 bearing immunoresistant tumor cells led to increased survival, intratumoral T cell infiltration, and downregulation of Msh2 expression without affecting healthy tissues. In a clinical trial with 18 patients with refractory mismatch repair proficient colorectal cancer, no responses were seen, but MMRd signatures emerged in cell-free DNA. These findings show that recapitulating an MMRd genotype through chemical mutagenesis can generate an immunogenic phenotype.

Graphical Abstract

In brief

Rousseau et al. show that tailored mutagenic chemical treatment induces a mismatch repair deficient genotype by epigenetic Msh2 silencing leading to potent immune response in immunoresistant tumor models and sensitizing to PD-1 blockade.

INTRODUCTION

Cancers that respond to immune checkpoint blockade (ICB) are often characterized by a high tumor mutation burden (TMB).^1,2 However, this phenotype alone is typically insufficient to generate a clinically relevant immune response. Instead, the clonal dominance, quality, and etiology of neoantigens have been proposed as important determinants of immunotherapy sensitivity.^3–9

Mismatch repair deficient (MMRd) tumors are highly immunogenic and often exhibit durable response to ICB.^10–12 MMRd tumors have exceptionally high TMB, typically higher than 50 mutations per megabase (mt/Mb). Their mutation profile is enriched for frameshift (FS) mutations,^13,14 which alter the open reading frame downstream of the mutation and lead to a novel amino acid sequence that persists until the first occurrence of a stop codon.^7,10 This can, in turn, lead to expression of non-self-peptides that appear foreign to the host immune system.¹⁰

Most solid tumors (>95%) are mismatch repair proficient (MMRp), non-immunogenic, and unresponsive to ICB.^2,15 Interestingly, genetically inactivating MMR proteins in experimental models produces immunogenic tumors that have a genomic profile consistent with MMRd. However, this only occurred after the deficient tumor cells were passaged for several months, which allowed for the gradual accumulation of immunogenic mutations.^7,8 We were interested in using a pharmacologic approach to directly and more rapidly generate an immunogenic tumor whose genome harbored a high overall mutation burden with a high proportion of frameshift mutations and mutational signatures overall reflective of the MMRd genotype.

There have been several previous attempts at employing pharmacologic approaches to induce anti-tumor immunity by modulating mutation load or neoantigen expression. Although temozolomide (TMZ) monotherapy often leads to substantial increases in tumor mutation load, it has failed to yield meaningful immunogenicity to date.^16–19 Another recent approach demonstrated that pharmacologic modulation of splicing with RBM39 degradation or type I PRMT inhibition induced neoantigen expression rendering tumors immunogenic in preclinical models.²⁰

We hypothesized that mimicking the mutational profile of MMRd tumors would recapitulate a MMRd genotype and phenotype. The MMRd genotype is characterized by T to C, C to T, and C to A single base substitutions (SBS). Our screening strategy was based on prior data from cell lines exposed to chemicals¹⁶ to identify agents inducing a high mutation burden with a MMRd genotype enriched for frameshift mutations. The combination of two agents quickly emerged as our lead strategy. TMZ is known to induce a high mutation burden in cancer with T to C substitutions but no frameshift mutations,^16,21 while cisplatin (CDDP) generates C to T and C to A substitutions and a high number of frameshift mutations. Based on these observations, we expected that combining TMZ and CDDP would lead to a mutational profile characteristic of MMRd tumors and potentially induce an immunogenic phenotype.

RESULTS

The combination of temozolomide and cisplatin induces a mismatch repair deficient genotype

We first evaluated a panel of compounds and combinations for mutagenic potential in immunoresistant murine cancer cells (Figures 1A and 1B). We focused on the combination of TMZ and CDDP. As a control mutagenic agent, we used DMBA (7,12-Dimethylbenz[a]anthracene), a polycyclic aromatic hydrocarbon (PAH), a component of tobacco smoke²² known to induce insertion-deletion mutations (indels), including C deletions,¹⁶ and to be tumorigenic in murine models.

Figure 1. The combination of temozolomide and cisplatin induces a mismatch repair deficient (MMRd) genotype.

(A) Strategy to identify mutagenic drugs that induce immunogenicity. 8W: 8 weeks; 1 W: 1 week.

(B) Selection of compounds aimed at inducing an MMRd genotype, increasing TMB and frameshift mutations. 7,12-Dimethyl benz[a]anthracene (DMBA) was used as a mutagenic control.

(C) Viability (MTS assay) of CT26 and B16-F10 cells treated with increasing doses of temozolomide, cisplatin, and DMBA. Data shown as mean viability (percentage) and standard error of the mean for each concentration of drug.

(D) Whole exome sequencing (WES, ≥250X mean coverage) of CT26 cells after 8 weeks exposure to vehicle, mutagenic compounds, or combination: temozolomide (TMZ) 20 μM, cisplatin (CDDP) 0.5 μM, temozolomide 20 μM + cisplatin 0.5 μM (TMZ+CDDP), DMBA 25 μM. Mutations are classified as shared (blue) or gained (red) compared to vehicle. TMB changes assessed via non-synonymous single nucleotide variants (NSSNV, red) and frameshift mutations (FS, blue). The microsatellite instability (MSI) score corresponds to absolute value after treatment.

(E) Single base substitution (SBS) by trinucleotide context and related COSMIC V3 SBS signatures of gained mutations assessed by WES after exposure of CT26 cells for 8 weeks to vehicle, TMZ 20 μM, CDDP 0.5 μM, TMZ 20 μM + CDDP 0.5 μM or DMBA 25 μM. Gained mutations were obtained by subtracting shared mutations with the CT26 parental cell line and deconvolved for mutational SBS. Total number of mutations and cosine similarity are displayed. Abbreviations: Ultraviolet (UV) light, Reactive Oxygen Species (ROS), Polymerase Epsilon proofreading defect (POLE), Chemotherapy (Chemo). See also Figures S1 and S2.

Single agent dose-response curves were generated to identify sub-lethal doses (Figure 1C) to prioritize mutagenesis over cytotoxicity. Since TMZ had no inhibitory effect on cell viability up to 200 μM, and given the possibility of a potential complementary effect, we tested the combination of TMZ+CDDP (Figures S1A and S1B). Doses achieving inhibitory concentration of 10–25% (IC10 to IC25) were selected allowing cell lines to proliferate as mutations were acquired during replication. After treatment with mutagenic agents, the proliferation rate remained unchanged in all models (Figures S1C and S1D).

We evaluated sequence alterations in the non-immunogenic murine colon cancer cell line CT26 following 8 weeks of exposure to the mutagenic compounds and a brief washout period. We performed whole exome sequencing (WES) with mean target coverage of ≥250X. Compared with baseline parental CT26 cells, cells treated with the vehicle control, TMZ, CDDP or DMBA exhibited a minimal increase in missense, frameshift mutations, and MSIsensor score (Figure 1D). In contrast, treatment with the combination of TMZ+CDDP resulted in a TMB gain of 125 mt/Mb and a significant gain in missense mutations, frameshift mutations, and MSIsensor score.²³

Mutational signature analyses of SBS demonstrated that nearly 70% of gained mutations induced by TMZ+CDDP were consistent with signatures associated with MMRd comprised of T to C transitions (Figure 1E). The MMRd signature was minimally observed or not observed at all with the other agents. When measuring mutational signatures associated with indels, only those indels induced by TMZ+CDDP were consistent with MMRd (S1E).

We next explored the mutational profiles following exposure to vehicle, TMZ, CDDP, TMZ+CDDP, or DMBA in the murine melanoma cell line B16-F10. As with CT26 cells, exposure to TMZ+CDDP in B16-F10 led to a mutational signature that was dominant for MMRd and characterized by a robust increase in total mutations, missense mutations, frameshift mutations, and MSIsensor score consistent with microsatellite instability (Figures S2A–S2D).

Temozolomide and cisplatin inhibit mismatch repair via epigenetic silencing

To investigate a possible mechanism for TMZ+CDDP leading to a MMRd genotype, we first looked for inactivation of the MMR protein family. While inactivating mutations have been reported in MMR genes after exposure to TMZ,^17,19,24 no deleterious genomic alteration in Mlh1, Pms2, Msh2, or Msh6 were found in our cohort. We then evaluated if the treatments modified the expression of MMR proteins and found a complete loss of the MSH2/MSH6 complex in the TMZ+CDDP treated CT26 cell lines (Figure 2A) not observed with TMZ or CDDP monotherapies. Protein quantification confirmed a significant decrease of MSH2 expression by 95% and MSH6 by 70% in CT26 cells treated with TMZ+CDDP (Figure 2B).

Figure 2. The combination of temozolomide and cisplatin induces MSH2 loss through epigenetic silencing of Msh2.

Treatments applied to CT26 cells include vehicle, TMZ 20 μM, CDDP 0.5 μM, and TMZ (20 μM) + CDDP (0.5 μM).

(A) Protein expression of MLH1, MSH2 and MSH6 assessed by western blot for CT26 cell line pretreated for 8W and CT26 Mlh1^−/− and CT26 Msh2^−/−.

(B) Relative protein expression of MLH1, MSH2 and MSH6 blot for CT26 cell line pretreated for 8W. One-way ANOVA.

(C) Assessment of microsatellite instability induced by 72h treatment in CT26 cell line. CT26 cells transfected with reporter plasmid containing out of frame luciferase due to microsatellite (CA17 repeats). The luciferase expression is restored by Indels. The boxplot represents the median (middle line), upper quartile (top line), and lower quartile (bottow line). Confidence interval at 90% is reported as whiskers. Please note that there were no statistical differences between groups.

(D) Assessment of Msh2 mRNA expression by RT-qPCR after 2, 4 or 8 weeks of treatment. One-way ANOVA.

(E) Promoter methylation assessment by MSP-PCR of Mlh1, Msh2 in the CT26 cell line pretreated 8 weeks.

(F) Quantification of Msh2 promoter methylation pretreated 4 weeks or 8 weeks. One-way ANOVA.

(G) Assessment of microsatellite instability in CT26 parental, CT26 Msh2^−/− and CT26 pretreated 8 weeks by TMZ 20 μM + CDDP 0.5 μM using the same reporter plasmid as in Figure 2C. The boxplot represents the median (middle line), upper quartile (top line), and lower quartile (bottow line). Confidence interval at 90% is reported. One-way ANOVA.

(H) Comparison of WES results for the CT26 cell line pretreated 2, 4 and 8 weeks. TMB gain assessment was assessed compared to the baseline TMB of parental CT26. Total frameshift mutations and MSI score per condition are displayed. ****: p ≤ 0.0001, ***: p ≤ 0.001, **: p ≤ 0.01; *: p > 0.01 and ≤0.05. Data are means +/− SEM. “ns” dictates not significant. See also Figure S3.

To understand if TMZ+CDDP had a direct impact on the generation of mutations or if the resulting genotype was the consequence of acquired MSH2 loss-of-function, we designed experiments investigating early exposure to the drugs and the kinetics of Msh2 downregulation. First, we developed a plasmid reporter in which luciferase expression is downstream of an out-of-frame microsatellite comprising CA17 repeats. This reporter when transfected, measures microsatellite instability proportional to indels occurring in the microsatellite as the in-frame expression of luciferase is restored. Short-term exposure (72 h) of CT26 cells to TMZ alone, CDDP alone, or TMZ+CDDP was not able to drive microsatellite instability via their sole alkylating/DNA damage properties (Figure 2C).

Then, we assessed Msh2 and Msh6 mRNA expression over the 8 weeks of treatment with TMZ+CDDP (Figures 2D and S3A). We found that Msh2 mRNA downregulation occurred as early as 4 weeks into treatment and was persistent at 8 weeks. This downregulation in mRNA expression was proportional to an increase in CpG island methylation of Msh2 promoter (Figures 2E and 2F). After 8 weeks’ exposure to TMZ+CDDP, CT26 cells developed a significant level of microsatellite instability that corresponded with the timing of Msh2 loss, even without continued exposure to the alkylating agents (Figure 2G).

To examine the association between Msh2 epigenetic loss and mutation burden, we performed WES in samples treated 2, 4, or 8 weeks with vehicle, TMZ, CDDP, or combination (Figure 2H). Before 4 weeks, we observed no differences between vehicle and any treatment for TMB, frameshift, and microsatellite instability. After 4 weeks, only the TMZ+CDDP combination showed an increase in TMB, frameshift, and microsatellite instability. This corresponded to the time frame in which Msh2 expression was lost. The TMB and frameshift burden further increased during weeks 4–8, indicating that the accumulation of non-synonymous mutations is proportional to time only after the loss of MSH2. Altogether, these results indicate that long term exposure ≥4 weeks to these alkylating agents is necessary to induce the downregulation of Msh2 through epigenetic reprogramming and an MMRd genotype.

Shorter exposure for 2 weeks resulted in upregulation of Msh2 and Msh6 mRNA (Figures 2D and S3A), suggesting that the MSH2-MSH6 complex is initially activated upon exposure to DNA alkylating agents and subsequently lost as a possible mechanism of resistance. CT26 cells were found to express O⁶-methylguanine methyltransferase (MGMT) and excision repair cross-complementation group 1 (ERCC1), other mechanisms of resistance to TMZ and CDDP, respectively,^25,26 in all conditions while MSH2 was lost only after TMZ+CDDP exposure (Figure S3B). We then investigated if exposure to TMZ+CDDP modified the overall methylation landscape of CT26 cells during treatment and found that methylation remained globally low in all conditions (<0.1%, Figure S3C), suggesting a specific epigenetic loss of Msh2. Finally, to assess functional resistance to the alkylating agents, we measured drug-induced apoptosis levels in CT26 parental, Msh2^−/− and CT26 pretreated 8 weeks by TMZ+CDDP (Figure S3D). We found that CT26 Msh2^−/− and CT26 pretreated by TMZ+CDDP presented significantly less apoptosis compared to the parental CT26 cell line, suggesting a more resistant phenotype to these agents induced by MSH2 loss.

In vitro immunogenicity of cell lines exposed to temozolomide and cisplatin

To assess the immunogenicity from exposure to these mutagenic agents, we performed syngeneic in vitro co-culture experiments of pretreated CT26 cells with splenocytes isolated from naive BALB/c mice and activated with anti-CD3 antibody. CT26 Msh2^−/− was used as a positive control. While no significant proliferation differences were observed for pretreated cells in the absence of syngeneic immune cells, growth was delayed in the presence of syngeneic immune cells for the CT26 Msh2^−/− cell line, and the TMZ+CDDP pretreated cell lines (Figure S4A). Furthermore, the CT26 Msh2^−/− cell line and cells pretreated with TMZ+CDDP led to increased immune cell-mediated apoptosis (Figure S4B).

We then assessed the impact of PD-1 blockade in coculture experiments. TMZ+CDDP pretreated cell line and the Msh2^−/− positive control showed a significant growth delay compared to other conditions (Figure S4C) and higher degree of immune cell-induced apoptosis (Figure S4D).

In vivo immunogenicity of cell lines exposed to temozolomide and cisplatin

We then performed immunogenicity assessment in mice. Cells pretreated with 8 weeks of mutagenic agents or the CT26 Msh2^−/− control cell line were implanted into immunocompromised mice. No growth difference was found compared to vehicle control (Figure 3A).

Figure 3. In vivo immunogenicity of cell lines exposed to temozolomide and cisplatin.

Treatments applied to CT26 cells include vehicle, TMZ 20 μM, CDDP 0.5 μM, and TMZ (20 μM) + CDDP (0.5 μM) or DMBA (25 μM).

(A) Tumor growth assessment in immunodeficient NSG mice engrafted with 1 million CT26 cells pretreated for 8W or CT26 Msh2^−/− (n = 5 mice per group). One-way ANOVA at last time point. (B) Follow up of tumor volume (mm³) of CT26 cells pretreated for 8W engrafted in syngeneic BALB/c mice (n = 6 per group). CT26 Msh2^−/− is included as a control. Comparison is reported at 20 days (D20).

(C) Tumor growth assessment in syngeneic mice engrafted with 1 million CT26 B2m^−/− cells pretreated 8W or CT26 B2m^WT (wild type) pre-treated 8W by Vehicle or TMZ+CDDP (n = 5 mice per group). One-way ANOVA at last time point.

(D) Randomized preclinical sensitivity assessment to anti PD-1 (aPD-1) versus isotype control (IgG) of tumors derived from CT26 cell line pre-treated for 8W. engrafted with 1 million cells in syngeneic BALB/c mice (N = 5 per group). CT26 Msh2^−/− cell line is used as a positive control. No statistical comparison was performed due to high immunorejection rate in the IgG group for the TMZ+CDDP condition.

(E and F) Exome sequencing comparison of missense (E) and frameshift (F) mutations in tumor compared to baseline cells. One-way ANOVA. Only significant comparisons are displayed.

(G) Absolute loss of TMB in mice tumors samples derived from CT26 cell lines pretreated 8W and engrafted in syngeneic mice compared to cell line from matched condition before engraftment. TMB was assessed by WES. One-way ANOVA.

(H) SBS signatures of lost mutations assessed by WES in CT26 derived tumors pretreated 8W before engraftment. Lost mutations were determined by comparing tumoral mutations with the baseline cell line. The total number of lost mutations and cosine similarity are reported. Abbreviations: Ultra Violet (UV) light, Mismatch Repair Deficiency (MMRd), Reactive Oxygen Species (ROS), Polymerase Epsilon proofreading defect (POLE), Chemotherapy (Chemo).

(I) MLH1, MSH2, and MSH6 protein expression assessment by immunohistochemistry of syngeneic orthotopic mice tumors derived from CT26 pretreated 8W or CT26 Msh2^−/− at time of sacrifice. Representative staining based on two tumor samples and multiple fields assessment are displayed. *: p ≤ 0.05; **: p ≤ 0.01; ***: p ≤ 0.001; ****: p ≤ 0.0001, Data are means +/− SEM. “NS” dictates not significant. See also Figures S4 and S5.

This was followed by studies in syngeneic and immunocompetent BALB/c mice (Figure 3B, n = 6 per group). Growth delay was strongest in CT26 cells pretreated with TMZ+CDDP with a magnitude comparable to Msh2^−/− tumors, while intermediate growth delay was observed with single agents. In the TMZ+CDDP group, some tumors were spontaneously rejected (33%). Tumors derived from CT26 cells pretreated by TMZ+CDDP and DMBA presented a significant increase in CD3⁺ and CD8⁺ T cells and increased intensity of PD-L1 staining (Figure 3C).

To assess the role of major histocompatibility complex (MHC) class I, we exposed a CT26 parental cell line and one deficient for the beta-2-microglobulin (CT26 B2m^−/−) with 8 weeks of vehicle, TMZ, CDDP, or TMZ+CDDP. The CT26 B2m^−/− was previously confirmed to have loss of MHC I expression.²⁰ Tumor growth was delayed in mice injected with CT26 parental cell line exposed to TMZ+CDDP but not in the CT26 B2m^−/− exposed to TMZ+CDDP (Figure S4E).

We then conducted a randomized preclinical study in mice to assess the efficacy of ICB on tumors derived from CT26 cells that had been pretreated with vehicle, TMZ, CDDP, or TMZ+CDDP. One million cells were injected subcutaneously into the flank of BALB/c mice. Once the tumor reached 100 mm³ mice were exposed to anti-PD-1 or an isotype control (IgG). The CT26 Msh2^−/− cell line served as a positive control (n = 5 per group, Figure 3D). No difference in tumor growth was observed for vehicle, TMZ, or CDDP with control IgG or anti-PD-1. Tumors from the CT26 Msh2^−/− cell line showed delay of growth in mice treated with PD-1 blockade, but not mice treated with control IgG. However, in the cells pretreated with TMZ+CDDP, no tumors arose in the anti-PD-1 treated group and 1/5 mice developed tumors when treated with IgG.

To circumvent the spontaneous rejection of tumors, we injected 5 million CT26 cells pretreated with TMZ+CDDP subcutaneously into the flank of BALB/c mice that were then randomized to receive anti-PD-1 or IgG (Figure S4F). In the anti-PD-1 group, we observed a complete response in all mice; all tumors were unpalpable by day 25. After discontinuation of anti-PD-1 at day 30, only one mouse experienced disease relapse. In the IgG group, 5 of 7 mice showed persistent tumor growth and the remaining 2 mice experienced spontaneous rejection of tumor (Figure S4G). The median overall survival (OS) was superior in the anti-PD-1 treatment group compared to IgG (Figure S4H).

Exome sequences of tumors derived from pretreated CT26 cell lines were compared to the exome sequences from the pre-implantation cell line to assess immunoediting. All tumors showed loss of mutations, with most significant loss in the CT26 cells pretreated by TMZ+CDDP displaying a disproportionate loss in frameshift mutations and decrease in mutation burden compared to baseline (Figures 3E–3G). Mutational signature assessment of lost mutations suggested predominant immunoediting of mutations originating from MMR deficiency (Figures 3H and S4I) while no significant difference in microsatellite score loss was observed compared to vehicle (Figure S4J).

MMR protein expression from tumors developed in syngeneic mice showed MSH2/MSH6 loss in CT26 cells pre-treated with TMZ+CDDP (Figure 3I). Loss of MSH2/MSH6 expression was persistent after cell line implantation and observed more than 8 weeks after last exposure to TMZ+CDDP, this sustained DNA repair defect being reflected by a higher onset of novel mutations in derived tumor samples (Figure 3E and 3F).

To confirm these findings we tested a second cell line. We evaluated the immunogenicity of B16-F10 cell line following 8 weeks of exposure to mutagenic agents in C57BL/6 mice. The melanoma model pretreated by TMZ+CDDP showed a significant growth delay or rejection of tumors (Figures S5A and S5B). No differences in tumor growth were observed with cells cultured with single agent TMZ, CDDP, or DMBA compared to vehicle (Figure S5A). Tumors derived from melanoma cells pretreated by TMZ+CDDP showed significant tumor growth delay when exposed to anti-PD-1 compared to isotype control (Figure S5C). Loss of MSH2 with decreased expression of MSH6 was observed in B16-F10 cells treated for 8 weeks with TMZ or TMZ+CDDP in cell lines (Figure S5D) and tumors grown in syngeneic mice (Figure S5E). Msh2 promoter hypermethylation was noted in both conditions at 8 weeks (Figure S5F).

Reprogramming of the tumor immune microenvironment after cancer cells exposure to temozolomide and cisplatin

We performed single-cell RNA sequencing (scRNA-seq) on tumor derived from CT26 cells pre-treated with 8 weeks of vehicle, TMZ, CDDP, or combination of TMZ+CDDP (Figures 4A, 4B, S6A, and S6B). Tumors were collected 2 weeks after engraftment to allow immune cell migration and activation when reaching a size of 500 mm³ scRNA-seq analyses identified that the quantity of T and NK cells infiltrating the tumor was doubled after TMZ+CDDP pretreatment compared to vehicle. We also identified the emergence of clusters of effector CD8⁺ T cells and M1-like macrophages in tumors originating from tumor cells pre-exposed to TMZ+CDDP.

Figure 4. Reprogramming of the tumor immune microenvironment after cancer cells exposure to temozolomide and cisplatin.

(A and B) Uniform Manifold Approximation and Projection (UMAP) representation of single cell clusters from tumor derived from CT26 pretreated by vehicle for 8W (A) and CT26 pretreated by TMZ+CDDP for 8W (B). ISG+: Interferon-Stimulated Gene positive cells; CAFs: Cancer-Associated Fibroblasts. NK: Natural Killers.

(C) Gene ontology analysis of bulk scRNA-seq and in specific clusters of cells displaying enriched pathways compared to vehicle. *: p ≤ 0.05; **: p ≤ 0.01; ***: p ≤ 0.001; ****: p ≤ 0.0001, ANOVA corrected for multiplicity. Only significant comparisons are displayed.

(D and E) VDJ TCR sequencing from T cells from tumors derived from CT26 cells pretreated 8 weeks by vehicle, TMZ 20 μM, CDDP 0.5 μM or combination of TMZ 20 μM and CDDP 0.5 μM (TMZ+CDDP) engrafted in syngeneic BALB/c mice (D) and TCR diversity score (E). See also Figure S6.

When comparing the mRNA profiles of tumors originating from cells pre-exposed to TMZ+CDDP as compared to control groups (Figure 4C), we found a profile consistent with immune cell activation including higher production of interferon-gamma (IFN-γ) and activation of JAK-STAT pathway, activation of innate immunity, higher class I and class II presentation, and activation and cytotoxicity of T cell and B cell immunity. TCR-seq analyses also highlighted the high degree of T cell activation with a high number of unique VDJ TCR+ cells (Figure 4D) and TCR diversity increased by 6-fold compared to control groups (Figure 4E). Altogether, these results demonstrate activation of both innate and adaptative immunity after exposure to TMZ+CDDP.

Systemic administration of temozolomide and cisplatin to mice bearing immunoresistant tumors

We next investigated whether we could convert immunoresistant tumors in vivo with systemic administration of TMZ and CDDP and explore any potential toxicity. We implanted 10,000 CT26 cells intraperitoneally into BALB/c mice and began treatment three days after implantation for 4 weeks with vehicle or TMZ+CDDP with either anti-PD-1 or IgG (Figure 5A). Most mice treated with vehicle or TMZ+CDDP and a control IgG antibody or vehicle+anti-PD-1 developed obvious signs of tumor growth and succumbed between 4 and 6 weeks of treatment, whereas mice treated with TMZ+CDDP+anti-PD-1 showed evidence of delayed tumor growth (Figure 5B) and 5 of 15 (30%) remained tumor-free (follow-up period >12 weeks). We evaluated immune infiltrate in tumor exposed at least 4 weeks with vehicle or TMZ+CDDP with either anti-PD-1 or IgG. We found a significant increase in CD3⁺ and CD8⁺ T cells in mice treated with TMZ+CDDP and anti-PD-1 compared to other groups (Figure 5C). In tumors of mice treated with TMZ+CDDP, we found downregulation of Msh2 mRNA expression compared to those mice treated with vehicle alone (Figure 5D).

Figure 5. Systemic administration of temozolomide and cisplatin with anti-PD-1 to mice bearing immunoresistant tumors.

(A) BALB/c mice were engrafted intraperitoneally (IP) with CT26 parental cells and treated with vehicle TMZ + CDDP ± anti-PD-1 or IgG starting 4 days post-engraftment. Mice received daily IP injections of DMSO (0.08%) or TMZ (40 μg, 5 days on, 2 days off), weekly CDDP (1 μg IP), and anti-PD-1 or IgG (200 μg IP, three times weekly) for up to 4 weeks or until sacrifice. The table displays final tumor rejection rates.

(B) Kaplan-Meier cancer-free survival analysis with relevant group comparisons (log rank test).

(C) Immunofluorescence assessment of CD3⁺ and CD8⁺ intratumoral cells. Quantification was performed on 10 independent fields of 0.2 mm² for each tumor. One-way ANOVA. Data are represented as means +/− SEM.

(D) Normalized mRNA expression in tumors originating from CT26 cell line and challenged intraperitoneally by vehicle, TMZ+CDDP and TMZ+CDDP + anti-PD-1. RT-qPCR was performed on 10 tumors samples of 1 mm³ per condition. The violin plot represents the median (middle dashed line), upper quartile (top dashed line), and lower quartile (bottom dashed line). The violin plot extends to minimum and maximum observed values. One-way ANOVA.

(E) Exome sequencing of healthy tissues (brain, liver, bone marrow, GI tract, lung) from mice after 8W exposure to vehicle control (mouse 1: M1) or TMZ+CDDP +anti PD-1 (mouse 4: M4, mouse 5: M5). Number of mutations are displayed as mutation per megabase by class. BM: Bone Marrow. One-way ANOVA. Data are means +/− SEM.

(F) Comparison of TMB and MSI scores in healthy tissues of mice treated by vehicle (5 organs from 1 mouse) or intraperitoneal combination of TMZ+CDDP + anti PD-1 (9 organs from 2 mice). Mice received 8 weeks treatment. The violin plot represents the median (middle dashed line), upper quartile (top dashed line), and lower quartile (bottow dashed line). The violin plot extends to minimum and maximum observed values. Student’s t test.

(G) SBS signatures identified by WES in healthy tissues of mice treated by vehicle (5 organs/mouse) or intraperitoneal combination of TMZ+CDDP+ anti PD-1 (9 organs from 2 mice). Lost mutations were determined by comparing tumoral mutations with the baseline cell line. The total number of lost mutations and cosine similarities are reported. Abbreviations: cytosine deaminases (APOBEC).

(H) Immunohistochemistry assessment of MLH1, MSH2 and MSH6 expression in tumors derived from mice challenged intraperitoneally by TMZ+CDDP. Tumors were collected 8W after engraftment at time of sacrifice. Regions with focal losses are highlighted and not observed with vehicle treatment. T: Tumor deposit; GI: gastrointestinal region. ****: p ≤ 0.0001; ***: p ≤ 0.001, **: p ≤ 0.01, *: p ≤ 0.05, ns: p > 0.05. See also Figure S6 and Data S1.

To check if MMR gene expression could be affected in healthy tissue by long-term exposure to TMZ and CDDP, we performed MLH1, MSH2, and MSH6 immunohistochemistry and qPCR of Msh2 and Msh6, in healthy tissue of mice exposed (6–8 weeks) to vehicle or TMZ+CDDP. We found that MLH1, MSH2 and MSH6 expression was conserved in all healthy tissues in all conditions tested (data not shown). Interestingly, we found that upon long-term exposure to TMZ+CDDP, healthy tissues were significantly upregulating Msh2 by 10- to 50-fold (Figure S6C). These results suggest that healthy tissues upregulate MSH2 to counteract DNA damage induced by the combination of alkylating agents. Contrary to cancer cells downregulating MSH2 expression upon long-term exposure, healthy tissue displays sustained MSH2 overexpression.

In a separate cohort of BALB/c mice, we treated with TMZ+CDDP for 8 weeks. We then assessed for evidence of toxicity at the end of the treatment or after 12 weeks of drug wash out (n = 4). After acute exposure, mouse weight was stable and in the normal range. The only histopathologic findings interpreted as TMZ+CDDP-related were mild inflammation in the liver, uterus, and pancreas. After 12 weeks of wash out these findings were still observed. Other histopathologic findings were interpreted as naturally occurring background lesions unrelated to treatment, as known to occur spontaneously in this strain.^27,28 No significant blood chemistry abnormalities were observed (Data S1).

We then investigated the systemic genotoxicity induced by long-term exposure to TMZ+CDDP+anti-PD-1 by sequencing multiple healthy organs of BALB/C mice from the experiment depicted in Figure 5A, and without evidence of cancer at the time of sacrifice and compared to healthy organ of mice exposed to vehicle control (Figure 5E). Compared to vehicle, a slight increase of non-synonymous and synonymous mutations was observed in the healthy organs of mice exposed to TMZ+CDDP+anti-PD-1. The effect was mainly driven by the higher onset of missense and synonymous mutations in one mouse (Mouse 5), observed in the bone marrow and lung tissues. The mean TMB gain between vehicle and the combination was respectively 0.2 and 0.3 mutation/megabase (Figure 5F). Then, we looked at evidence of MMR defects and investigated MSI score and SBS mutational signatures (Figures 5F and 5G). Compared to vehicle, we found lower MSI score in tissues submitted to TMZ+CDDP + anti-PD-1. (Figure 5G), while the main gained signatures in the treated condition were consistent with TMZ and CDDP exposure. MSH2 overexpression in healthy tissue after exposure to TMZ+CDDP (Figure S6C) may explain the apparent reduction of microsatellite instability compared to vehicle control and the absence of MMRd signatures in healthy tissues.

To further understand the potential differential effect on MMR-induced TMZ+CDDP treatment between cancer and healthy cells, we performed MMR staining in BABL/C mice engrafted intraperitoneally by CT26 and challenged by TMZ+CDDP (Figure 5H). Looking at gastrointestinal tract regions with tumor deposits, we found a downregulation of MSH2 and MSH6 in cancer cell regions while the healthy epithelium was spared.

Treatment of patients with refractory metastatic colorectal cancers with temozolomide and cisplatin and PD-1 blockade

Based on these preclinical findings, we designed an exploratory clinical trial to investigate the combination of TMZ, cisplatin, and PD-1 blockade in patients with chemo-refractory MMRp metastatic colorectal (mCRC) cancer, a cancer typically unresponsive to ICB.^9,29 Over the course of a 28-day treatment cycle, patients received oral TMZ 150 mg/m² on days 1–5 during cycle 1 and 200 mg/m² on days 1–5 during cycle 2. CDDP 40 mg/m² was administered intravenously on days 1 and 15, and nivolumab 480 mg delivered intravenously on day 1. Treatment was continued until progression, unacceptable toxicity, or patient withdrawal (Figure 6A). The co-primary endpoints were the overall response rate (ORR) and the 16-week PFS rate. These hypotheses were tested using a two-stage Simon’s optimal design.

Figure 6. Treatment of patients with refractory metastatic colorectal cancers with temozolomide, cisplatin and PD-1 blockade and resulting genomic profiles assessed by ctDNA.

(A) Study design. Patients with refractory, mismatch repair proficient colorectal cancer were treated with temozolomide, cisplatin and nivolumab. Baseline status for low mutational burden and microsatellite score was confirmed by tumoral sequencing using the MSK-IMPACT panel and blood TMB and blood MSI score dynamics were studied using cell-free DNA analyses.

(B and C) Cell-free DNA measurements of (B) blood tumor mutational burden (bTMB), (C) blood microsatellite instability (bMSI) score at baseline and at last available sample. One-way ANOVA mixed-effects analysis.

(D) Number of mutations from baseline, to cycle 3 to cycle 5. Lines indicate samples from the same patients. Paired t-test.

(E) Microsatellite score from baseline, to cycle 3 to cycle 5. Lines indicate samples coming from the same patients. Paired t-test.

(F) Tumor fraction from baseline, to cycle 3 to cycle 5. Lines represent samples coming from the same patients. Paired t-test.

(G) Relative contribution of SBS signatures between mutations in baseline samples vs. gained in subsequent samples.

(H) Relative contribution of SBS signatures between potentially immunoedited and non-immunoedited mutations. HRD, Homologous recombination. The violin plots (B–F) represent the median (middle dashed line), upper quartile (top dashed line), and lower quartile (bottow dashed line). Each violin plot extends to minimum and maximum observed values. ***: p ≤ 0.001; **: p ≤ 0.01; *: p ≤ 0.05; ns: p > 0.05. See also Figure S7.

A safety run-in was performed. In the first 6 patients, no dose limiting toxicity was observed after 4 weeks of treatment. In general, the toxicity was moderate and limited to grade 2 hematologic toxicities and only one patient had a delay of treatment because of thrombopenia. No immune related adverse event was observed during the trial.

In the first 18 patients, no response was observed and the 16-week PFS rate was below the predefined threshold. We stopped patient recruitment to focus on genomic analyses and assess if a specific subset of patients may have benefited from the strategy and potential markers associated with resistance to chemotherapy-induced mutagenesis. While the overall clinical study based on the primary endpoint was negative, we were interested if any translational findings could be identified aiding the clinical development of this concept moving forward. We measured circulating tumor DNA (ctDNA) to follow the genetic evolution of patients’ tumors while on therapy. Serially collected plasma was available for ctDNA analyses for 15 of 18 patients. Of these, 10 (66%) showed a gain in the absolute number of circulating mutations and 2 (13%) experienced a gain of greater than 100% (Figure 6B) in blood TMB (bTMB). Twelve (80%) patients showed an increase in blood microsatellite instability (bMSI) scores following therapy (Figure 6C). This bTMB and bMSI score increase was gradual and proportional to time (Figures 6D and 6E). We did not observe a significant change in tumor fraction on treatment (Figure 6F). Analysis of gained mutations pooled from patients (Figure S7A) demonstrated a single nucleotide variant mutational signature that corresponded to MMRd as evidenced by the emergence of T to C transitions (Figure S7B) and an increase in SBS MMRd mutational signature up to 23% (Figure 6G). Since acquired and baseline mutations show different trajectories on treatment over time (Figure S7C), we investigated the impact of immunoediting during treatment, defined as mutations in the last treatment cycle with variant allelic frequency (VAF) at least lower by 50% than its peak VAF or if subsequently undetectable (VAF = 0). We observed that immunoedited mutations on treatment were disproportionally presenting a MMRd SBS mutational signature (27%) compared to non-immunoedited mutations (9%) (Figure 6H).

While no patients achieved an objective radiographic response, five patients experienced stable disease at 8 weeks (Figure 7A and Table S1). Four patients presented MGMT promoter hypermethylation at baseline. MGMT methylated patients had a higher rate of stabilization (50%, n = 2/4) than non-methylated patients (27%, n = 3/11, Figure 7A). Compared with patients who showed progressive disease, those with stable disease showed an early increase (≤3 months of initiating therapy) in mutation burden and MSI scores as measured by ctDNA, with proportional increase during treatment (Figures 6D and 6E). One patient with a significant increase in bTMB but unchanged bMSI score also showed stable disease. All patients eventually progressed, but those who did not experience an increase in bTMB or bMSI score showed progression earlier than 8 weeks.

Figure 7. Survival outcomes of patients with refractory metastatic colorectal cancers treated with temozolomide and cisplatin and PD-1 blockade based on acquired genotypes.

(A) Swimmer plot summarizing the genotype and clinical outcomes of each patient. Gain in bTMB and bMSI score are defined based on the median increases observed in the clinical trial.

(B) Overall survival (OS) according to acquired bTMB and bMSI score on treatment. Gain in bTMB and in bMSI score are defined based on the median increases observed in the clinical trial as explained above. We report the median OS per group. Log rank test.

(C) Assessment of OS according to acquired bTMB and bMSI score from Cycle 3 (C3). Log rank test.

(D) Oncoprint displaying genomic features of patients at baseline and at the end of treatment with temozolomide, cisplatin, and nivolumab. Copy number gains in mismatch repair (MMR) genes are reported at baseline and at the end of treatment. See also Figure S7 and Table S1.

Progression-free survival (PFS) and OS showed a wide variability across the cohort (Figure 7A). A higher baseline bTMB or bMSI score did not impact PFS (Figures S7D and S7E). Based on our preclinical results, in which immunogenicity was observed when both TMB and MSI score increased, we investigated if different survival outcomes in the clinical trial were observed based on the relative increase in bTMB and bMSI score on treatment. Improved OS was only observed in patients showing an early gain in both bTMB and bMSI scores, even after discontinuation of therapy (Figures 7A–7C and S7F).

To understand why some patients were more prone to increase both their bTMB and bMSI score on treatment, we assessed genomic alterations in the mismatch repair pathway from baseline to end of treatment (Figure 7D). MGMT methylated patients at baseline were more likely to gain both bTMB and bMSI score in our cohort. Patients with bTMB and bMSI gain had lower tumor fraction at baseline. Patients not increasing their bTMB and/or bMSI scores were more likely to present aneuploid amplifications of one or more MMR genes at baseline (64%, n = 7/11). On treatment, patients not increasing their bTMB and/or bMSI scores presented de novo copy number gain of MSH2 or MSH6, with any gain in MMR genes observed in 82% (n = 9/11) at time of progression. In patients increasing both bTMB and bMSI scores, only one patient with MGMT methylation had a baseline PMS2 gain and acquired MSH2 gain, and 75% (n = 3/4) of patients retained intact MMR gene copies at the end of treatment.

DISCUSSION

In this study, we tailor chemical mutagenesis to induce a MMRd genotype and anti-tumor immunity in cancers that were previously non-immunogenic. After exposure to mutagenic agent combination, cell lines acquired a high mutation burden with a preponderance of frameshift mutations and an immunogenic signature consistent with the MMRd state. Tumors derived from these mutagenized cell lines exhibited significant infiltration with tumor infiltrating lymphocytes and were consistently immunorejected. Interestingly, the sole gain of a high mutation burden alone, as was observed with single agent TMZ, did not result in anti-tumor immunity and adequate reprogramming of the immune tumor microenvironment. These results are consistent with clinical data in glioblastoma where TMZ monotherapy leads to massive somatic mutation gains but fails to sensitize tumors to ICB.¹⁹

We found that the combination of TMZ with CDDP rendered cells MMRd with high mutation burden following only 4–8 weeks of treatment. This differs from prior data showing that Mlh1 or Msh2 deletions in cell lines required up to 1 year to develop a high mutation burden and immunogenicity.^7,8 In patients with advanced colorectal cancers treated with TMZ, the time to develop a high mutation burden required a minimum of 4–6 months of treatment.¹⁷ In contrast, we noted the development of MMRd as early as 4 weeks into treatment and noted a substantial accumulation of mutations by 4–8 weeks. Importantly, we uncovered that the accumulation of mutations was contingent on Msh2 loss which is acquired on treatment after 2–4 weeks, as Msh2 loss drove resistance to TMZ+CDDP combination. Unlike the other scenarios described above, the combination of TMZ+CDDP rapidly inactivates mismatch repair whereby any new mutations would be retained.

TMZ has been reported to enhance genome-wide methylation of cytosines³⁰ and occurrence of an MMRd phenotype with MSH6 loss has been noted with long-term exposure to TMZ.^31,32 Several studies have identified MSH2 promoter hypermethylation after CDDP exposure.³³ These data suggest that the combination of TMZ+CDDP might act synergistically to increase MSH2 promoter methylation, consistent with our own observations that MSH2 loss was a mechanism of resistance to the combination and necessary to accumulate mutations.

One important concern is that chemical mutagenesis would not be limited to cancer cells and that healthy tissue would accumulate detrimental mutations. Long-term exposure to TMZ+CDDP was well tolerated by mice, and mild systemic toxicity was observed. Sequencing of healthy tissues exposed to the combination revealed only the accumulation of few mutations and mutational signatures consistent with TMZ or CDDP exposure, without inducing an MMRd genotype. Of note, tumors exposed to TMZ+CDDP showed downregulation of MSH2 and MSH6, while the adjacent normal epithelium was spared. In contrast to cancer cells, healthy tissues presented a sustained MSH2 upregulation upon long-term exposure to TMZ+CDDP. These data support the concept that chemical mutagenesis has a greater effect on cancer cells than normal cells and suggest that cancer cells have an intrinsic level of genomic instability leading to MSH2 loss to become resistant to the alkylating drugs combination, explaining the higher accumulation of mutations and the occurrence of a MMRd genotype.

Our pilot clinical trial combining TMZ, cisplatin and nivolumab did not meet any of the co-primary endpoints. Nevertheless, translational analyses of clinical specimens identified some signals that were consistent with the pre-clinical analyses. Systemic administration of TMZ+CDDP in a subset of cases (1) induced the acquisition of mutational signatures consistent with MMRd, (2) showed immunoediting of mutations related to the MMRd signature, and (3) showed preliminary evidence of initial clinical benefit in those who had an increase in their circulating mutation number and microsatellite instability. The acquisition of an MMRd-like genotype was more often observed in patients with baseline MGMT methylation, low tumor fraction, and in the absence of copy number gain in MMR genes. As in our preclinical experiments, identifying that TMZ+CDDP first induced MSH2 overexpression before MSH2 epigenetic silencing, we uncovered that patients not developing the MMRd genotype acquired de novo copy number gains in MSH2 and/or MSH6, potentially counteracting the DNA damage induced by the combination and impairing MMR loss in cancer cells.

While the clinical trial was negative, these genomic and correlative results are provocative. Indeed, longer exposure to TMZ+CDDP could result in improved genomic and immunologic outcomes as duration of treatment correlated with higher gain in bTMB and bMSI scores. Another important observation is the possible rapid immunoediting of MMRd-related mutations as assessed by mutational signature on treatment. Contrary to other trials in which a priming phase with single agent TMZ preceded immunotherapy exposure, our trial incorporated nivolumab from the start of mutagenic chemotherapy. The impact of timing of immunotherapy introduction relative to mutagenic chemotherapy remains an open question especially when comparing the discrepancy between the pre-clinical and clinical data. In addition, while there is extensive clinical experience with these agents, and despite our reassuring preclinical results, the potential induction of mutations in non-cancerous tissues remains of concern and requires further judicious evaluation in the clinical setting.

In conclusion, we find that chemical mutagens can be used to induce a MMRd genotype in cancer cells and, in turn, convert poorly immunogenic tumors into tumors with high immunogenic potential in mice, but without meaningful clinical impact in an unselected population. From a biologic perspective, our findings further emphasize the importance of genotype in defining the immunogenicity of a tumor.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Luis A Diaz Jr (ldiaz@mskcc.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Mouse Whole Exome Sequencing data have been deposited at the Sequencing Read Archive (SRA) and are publicly available as of the date of publication. DOIs are listed in the key resources table.

• Mouse single cell sequencing Cell Ranger output has been deposited at Zenodo and is publicly available as of the date of publication.

• Patients’ sequencing data reported in this study cannot be deposited in a public repository in compliance with patient’s consent agreement and to protect patients’ privacy. Data would be available upon request if access is granted through a data transfer agreement. To request access, contact Dr Luis A. Diaz, the lead contact. Summary genomic results are available in Table S1 as of the date of publication.

• All original code has been deposited at Zenodo and is publicly available as of the date of publication. DOIs are listed in the key resources table.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Armenian Hamster Anti-mouse CD3	Biolegend	Cat# 100301; RRID: AB_312666
Rat AntiP-Mouse PD-1	Thermofischer/Ebioscience	Cat# 14-9982-82; RRID: AB_468664
Rat IgG2a Isotype control	Biolegend	Cat# 400563; RRID: AB_326423
Rabbit Anti-Mouse CD3	Dako	Cat# A0452; RRID: AB_2335690
Rabbit Anti-Mouse CD4	Abcam	Cat# 183685; RRID: AB_2686917
Rabbit anti-Mouse CD8	Cell Signaling	Cat# 98941; RRID: AB_2756376
Rabbit Anti-Mouse Ki67	Abcam	Cat# 16667; RRID: AB_302459
Goat Anti-Mouse PD-L1	R&D	Cat# AF1019; RRID: AB_354685
Biotynylated Goat Anti-Rabbit IgG	Vector Labs	Cat# PK6101; RRID: AB_2336819
Rabbit Anti-mouse β-Actin	Abcam	Cat# Ab115777; RRID: AB_10711040
Rabbit Anti-mouse MLH1	Abcam	Cat# ab92312; RRID: AB_10562149
Rabbit Anti-mouse MSH2	Abcam	Cat# ab70270; RRID: AB_1209509
Rabbit Anti-mouse MSH6	Abcam	Cat# ab92471; RRID: AB_10563502
Rabbit Anti-MGMT	Life technology	Cat# PA5-120050; RRID: AB_2915054
Rabbit Anti-ERCC1	Proteintech	Cat# 14586-1-AP; RRID: AB_10693616
Bacterial and virus strains
Not applicable
Biological samples
Not applicable
Chemicals, peptides, and recombinant proteins
Dimethyl sulfoxide	Fischer scientific	BP231-10
Temozolomide	Sigma Aldrich	T2577
Cisplatin	Sigma Aldrich	1134357
QTracker® 655 cell labeling kit	Life Technology	Q25029
Phosphatase inhibitor cocktail	Thermoscientific	78425
DNAse	Roche	1010159001
7,12-Dimethylbenz(A)anthracene	Sigma Aldrich	D3254
Streptavidin HRP and DAB Map kit	Ventana Medical Systems	760-124
Green color Fluorophore TSA A488	Life Tech	B40932
Red color Fluorophore TSA CF594	Biotium	92174
Red Blood Cell Lysis Buffer	Biolegend	420301
Critical commercial assays
CellTiter 96® AQueous One Solution Reagent	Promega	G3582
Caspase 3/7 green apoptosis assay reagent	IncuCyte	4440
Qubit™ RNA Assay Kit	Invitrogen	Q32852
iSCript Reverse Transcription Supermix	Bio-Rad	1708841
PowerUp™ SYBR™ Green Master Mix	Thermofischer scientific	100029283
Qubit dsDNA HS Assay kit	Thermofischer scientific	Q32851
NEBNext Enzymatic Methyl-seq Conversion Module	New England Biolab	E7125L
Methylated Mouse DNA Standard	Zymo	D5012
EpiTect MethyLight PCR + ROX Vial Kit	EpiTect	59496
D1000 ScreenTapes	Agilent	5067-5582
D1000 Reagents	Agilent	5067-5583
KAPA Hyper Prep Kit	Kapa Biosystem	KK8504
SureSelectXT Mouse All Exon	Agilent	5190-4641
SinglePlex Mouse Exome	Twist	102036
HiSeq 3000/4000 SBS kit	Illumina	FC-410-1002
NovaSeq 6000 S4 Reagent Kit	Illumina	20028312
BCA assay	Thermoscientific	23225
4–20% tris-glycine gels	Thermoscientific	23225
PVDF membrane	Invitrogen	IB24001
PNL2.1[nluc/Hygro]	Promega	N1061
DH5α competent cells	Thermoscientific	EC0112
Plasmid plus midi kit	Qiagen	12943
SE Cell Line 4D-Nucleofector X Kit	Lonza	V4XC-1032
Nano-Glo Luciferase assay	Promega	N1110
Annexin Binding Buffer	Thermoscientific	V13246
FITC Annexin V	BD Bioscience	556420
Propidium Iodide staining solution	BD Bioscience	556463
Tumor Dissociation Kit, Mouse	Miltenyi	130-096-730
Chromium Next GEM Chip K Single Cell Kit	10x Genomics	PN1000286
Chromium Next GEM Single Cell 5’ Kit v2	10x Genomics	PN1000263
NovaSeq X 25B Reagent Kit	Illumina	20125967
Human TCR Amplification Kit	10x Genomics	PN1000252
Global Methylation assay kit	Abcam	ab233486
Background Blocking reagent	Innovex	NB306
Cell Conditioning 1	Ventana	950-500
DNeasy Blood and Tissue kit	QIAGEN	69504
Deposited data
Code for whole exome sequencing analysis	This study	https://doi.org/10.5281/zenodo.10551140
Mouse Whole Exome Sequencing data (FASTQ)	This study	PRJNA1071536
Code and data for scRNAseq analysis	This study	https://doi.org/10.5281/zenodo.14549539
Patient summary of sequencing data (Excel)	This study	Table S1
Experimental models: Cell lines
CT26	ATCC	CRL2638
B16-F10	ATCC	CRL-6475
CT26 Msh2−/−	Mandal et al.7	Not applicable
B16-F10 Msh2−/−	Mandal et al.7	Not applicable
CT26 Mlh1−/−	Germano et al.8	Not Applicable
CT26 B2m−/−	Lu et al.20	Not Applicable
Experimental models: Organisms/strains
C57/Bl6	The Jackson laboratory	strain #000664
BALB/c	The Jackson l aboratory	strain# 000651
NSG	The Jackson l aboratory	strain#005557
Oligonucleotides
See Table S2
Recombinant DNA
Not Applicable
Software and algorithms
R V4.4.0	www.R-project.org
GraphPad Prism V9	Dotmatics	–
Gen5 V3	BioTek	–
MSIsensor program	Middha et al.34	–
R package MutationalPatterns V3.4.1	Manders et al.35	–
Cell Ranger v8.0	10x Genomics (https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/latest)	–
Seurat v5.1.0	Hao et al.36	–
DoubletFinder v2.0.4	McGinnis et al.37	–
SingleR v2.6.0	Aran et al.38	–
celldex v1.14.0 (MouseRNAseqData)	Aran et al.38	–
scType v2021	Ianevski et al.39	–
clusterProfiler v4.12.6	Wu et al.40	–
GRCm39 mouse reference genome (10x refdata-gex-GRCm39-2024-A)	10x Genomics (https://support.10xgenomics.com/single-cell-gene-expression/software/release-notes/references)	–
Other
Cytation 1 multi-mode reader microscope	BioTek	–
BioTek BioSpa 8 Automated Incubator	BioTek	–
4200 TapeStation System	Agilent	G2991BA
Guardant Omni and Guardant infinity assays	Guardant Health Inc	–

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines

CT26 (female) and B16-F10 (male) cell lines were obtained from American Type Culture Collection (ATCC). All the CRISPR-Cas9 cell lines used in the present study have been described in previous works.^7,8,20 CT26 Msh2 Knock Out, B16F10 Msh2 Knock out were obtained from Tim Chan’s lab (MSKCC, USA),⁷ CT26 Mlh1 Knock out from Alberto Bardelli’s Lab (Candiolo Cancer Institute, Italy)⁸ and CT26 B2m Knock Out from Omar Abdel-Wahad’s Lab (MSKCC, USA).²⁰ All cell lines were cultured in RPMI 1640 supplemented with 10% fetal bovine serum and 10 UI/mL penicillin 10 μg/mL streptomycin. All cell lines were cultivated in 5% CO₂ incubator at 37° C.

Animal studies

C57/Bl6(strain #000664), BALB/c (strain# 000651) and NSG (strain#005557) mice (The Jackson laboratory) were used for in vivo studies and cared for in accordance with guidelines approved by the Memorial Sloan Kettering Institutional Animal Care and Use Committee and Research Animal Resource Center. 6–8 weeks old mice (male or female) were injected subcutaneously with 0.1–5 million CT26, Pan02 or B16-F10 cells together with matrigel (Corning) at 1:1 ratio. Tumors were measured 2–3 times/week using calipers, and volume was calculated using the formula: length × width2 × 0.52. Bodyweight was also assessed twice weekly. Animal protocol was approved by the Institutional Animal Care and Use Committee.

Clinical trial (NCT04457284)

We designed a single arm, Simon two-stage design, monocenter (MSKCC) phase II trial investigating the combination of Temozolomide, Cisplatin and Nivolumab in patients with chemorefractory advanced Mismatch Repair proficient colorectal cancer. This clinical trial was approved by Memorial Sloan Kettering Cancer Center’s Institutional Review Board.

Main inclusion criteria were the following: adult patient with written informed consent for the trial, histologically-confirmed locally advanced unresectable or metastatic colorectal adenocarcinoma, tested for MSI/dMMR and determined to be MSS or MMR proficient, tested for BRAF and POLE mutations and determined to be wild type, subjects refractory to, or intolerant of, at least 2 lines of standard chemotherapy, and at a minimum prior exposed to oxaliplatin, irinotecan and a fluoropyrimidine, at least one index lesion which is measurable based on RECIST 1.1, ECOG performance status of 0 or 1, consent for use of archival tissue and blood draws for research purposes, adequate organ function, no prior exposure to anti-PD-1, anti-PDL-1, anti-PDL-2, anti-Cytotoxic T-lymphocyte-associated antigen-4 (CTLA-4) antibody or any other antibody or drug specifically targeting T cell co-stimulation or checkpoint pathways, no history of immunodeficiency, of active autoimmune disease or systemic steroid therapy (>10 mg/day prednisone, or equivalent) or any other form of immunosuppressive therapy within 7 days prior to the first dose of trial treatment.

Subjects received oral Temozolomide at 150–200 mg/m² day 1–5 every 4 weeks, cisplatin via IV infusion at 40 mg/m² every two weeks (Q2W), and nivolumab via IV infusion at 480 mg every four weeks (Q4W). The primary objective was to determine the Disease Control Rate (DCR) defined by Complete, Partial Responses and Stable Disease (SD) according by RECIST 1.1. Secondary objectives included safety and tolerability according to Common Terminology Criteria for Adverse Events (CTCAE) V5, Progression-Free Survival, Overall Survival, and evaluate the effects of TMZ, Cisplatin and nivolumab on occurrence of new mutations (InDels/frameshift/Missenses), tumor mutational burden in plasma and radiologic response.

The primary objective of the study was to determine both the 16 weeks PFS and the objective response rate (ORR) (complete response plus partial response) according to RECIST 1.1 in patients with CRC treated with TMZ, cisplatin and nivolumab. Both factors will be considered for purposes of sample size determination as follows: A 2-stage design was employed; An initial cohort of 18 patients will be treated; If 5 or more patients are alive and progression free at the 16 weeks follow-up or if at least 2 objective responses are observed, an additional 17 patients will be accrued for a total of 35 patients. Secondary objectives included safety and tolerability, overall survival. Exploratory objectives were to evaluate the effects of temozolomide (TMZ), cisplatin and nivolumab on occurrence of new immunogenic mutations (insertions and deletions [InDels]/frameshift/missenses) and tumor mutational burden; and the effect on the immune response.

METHOD DETAILS

Study design

The main objective of the study was to investigate pharmacological modalities to convert non-immunological tumors to neoantigen comprising immunological tumor. To do so, we first created invitro models using CT26 via exposing to various drugs and then used B16-F10 and Pan02 mice cancer models for further validation. We assessed the immunogenicity of newly engineered cell lines via co-culture assays in presence of syngeneic splenocytes for the CT26 model. The cell lines were then sequenced to analyze changes in mutation burden and mutational signatures. We then tested the pre-treated cell lines in in-vivo model in absence and presence of immune check point blockade by anti PD1 treatment. Each in vitro experiment was performed with technical triplicates unless otherwise mentioned. All in-vivo experiments were performed with 3–5 biological replicates unless otherwise mentioned.

Based on best drug candidates in preclinical models, available drugs for clinical use and established clinical data for safety of drug combination of temozolomide and cisplatin,^41–43 we then performed a clinical trial with translational studies investigating a drug induced mutagenesis strategy combined with PD-1 blockade in patients with chemorefractory metastatic colorectal cancer (NCT04457284). Patients were not selected based on MGMT promotor methylation status as it has been reported that the clinical impact of temozolomide and cisplatin combination did not correlate with MGMT activity.⁴³

MTS assays

1,000 to 5,000 cells were plated in 96 wells plate overnight at 37 ° C 5% CO₂. On Day 2, serial dilutions of temozolomide, Cisplatin, DMBA or vehicle control (Dimethyl sulfoxide, DMSO) were prepared and used to treat the cells in triplicate for 3 consecutive days at 37 ° C 5% CO₂. On Day 5, treated cells were incubated with CellTiter 96 AQ_ueous One Solution Reagent (Promega, G3582) for 2 to 4 h and then absorbance was recorded at 490nm using a 96-well plate reader. IC50 was assessed using a log(inhibitor) vs. response model using GraphPad Prism v9. For combination treatment, fixed doses of drugs were used as single agents or in combination to treat the cells in triplicate, using the same revelation method.

Long-term drug treatment assays

CT26, B16-F10 and Pan02 parental cell lines were treated for 8 weeksby vehicle control (Dimethyl sulfoxide, DMSO at 0.014%, Fischer, scientific, BP231–10), temozolomide 20 μM (Sigma Aldrich, T2577), Cisplatin 0.5 μM (Sigma Aldrich, 1134357), combination of temozolomide 20 μM + Cisplatin 0.5 μM, and 7,12-Dimethylbenz(A)anthracene (DMBA) 25 μM (Sigma Aldrich, D3254) in RPMI 1640 supplemented with 10% fetal bovine serum and 10 UI/mL penicillin 10 μg/mL streptomycin at 37° C 5% CO₂. The dose of each compound or combination was determined according to MTS assay to reach a IC10 to IC25 dose to allow the cancers cells to proliferate enough as mutations were expected to occur during replication. In short, each week, on day 1 the cells were trypsinized and plated between 250,000 and 500,000 cells in a 150 mm dish in RPMI 1640 supplemented with 10% fetal bovine serum and 10 UI/mL penicillin 10 μg/mL streptomycin at 37° C 5% CO₂. On day 2, the media was removed and fresh media containing the drug or combination of interest was added. On day 4, media was removed and fresh media with the drugs was added again. On day 7, media was removed and fresh media without the drugs was added to allow the cells to rest before trypsinization the next day, then repeating the forementioned weekly cycle up to 8 weeks. Two million cells were collected, pelleted or stored in 5% DMSO at 1W, 2W, 4W and after 8W + 1 week of wash out for subsequent experiments. The proliferation time of each cell line in each condition after 8W of treatment was determined by taking bright field pictures every three to 5 h for 5 days using the Cytation-1TM 5 multi-mode reader microscope (BioTek).

Splenocytes and coculture experiments

On Day 1, CT26 cell line pretreated 8 weeks by vehicle control, Temozolomide, Cisplatin, Temozolomide+Cisplatin combination, or DMBA, were plated in 96 well plates at 1,500 cells per well and let it grow overnight at 37° C 5% CO₂.

On Day 2, Spleens were harvested from female BALB/c mice in PBS at 4° C. Spleens were crushed in PBS and resulting solution was poured through a 100 μM filter. After a 5min 250g centrifugation, the cells were resuspended in 6 mL of Red Blood Cell Lysis Buffer (Biolegend, 420301) 4 min at room temperature and reaction was stopped using 20 mL of PBS. After another 5min 250g centrifugation, cells were resuspended in 10 mL PBS at room temperature and counted. Immune cells were then prepared at 1 million cells per mL and stained using Qtracker 655 cell labeling kit (Life technologies, Q25029). Immune cells were then activated for 1 h in RPMI-1640 + 2mM Glutamine +10% fetal bovine serum and 10 UI/mL penicillin 10 μg/mL streptomycin with anti-mouse CD3 antibodies at 2 μg/mL (Biolegend, clone 9145–2C11) with or without anti mouse PD-1 antibody at 0.5 μg/mL (Thermofisher/Ebioscience, clone RMPI1–14). 1X Caspase-3/7 green apoptosis assay reagent (IncuCyte,4440) was then added to the media with or without immune cells to allow the follow up of apoptosis. The co-culture experiment itself was performed. 200 μL of the desired immune cell solution (with/without anti-PD-1) or solution without immune cells were added in each well in quadruplicate. A ratio of 3 immune cells for 1 cancer cells was used for conditions without anti PD-1 while a 1 to 1 ratio was used for conditions with anti PD1 antibody. 96-well plates were protected from light and pictures were automatically acquired by the Cytation-1 multi-mode reader microscope (BioTek) every three to 5 h for 5 days. Live apoptosis and cancer cells confluence were then analyzed. To differentiate cancer and immune cells, a strategy of gating based on cell size (>14 μm) was applied on each picture, and apoptotic caspase3/7 analyses were also performed using the same gating by assessing green fluorescence. Results are reported using ΔConfluence corresponding to each time point confluence minus the baseline mean confluence of the condition, and immune induced apoptosis using a normalized ratio of caspase3/7 positive fluorescent cells at each time point corresponding to the number of fluorescent cells in each well exposed to immune cells divided by the number of caspase3/7 positive fluorescent cells in conditions not exposed to immune cells. Baseline ratio before adding the immune cells was considered as 0 as measured before adding the caspase3/7 reagent to the cancer cells.

RNA Extraction and MMR gene expression study

RNA was extracted from cell pellets of 5 million cells or 1–10 mm3 of bulk tumors from syngeneic mice experiments (5 tumoral pieces per mice) using RNeasy Plus Mini Kit (Qiagen, 74134)) using the recommended manufacturer protocol. The extracted RNA was quantified using a Qubit Fluorometer with the Qubit RNA HS Assay Kit (Invitrogen,REF: Q32852). For RT-qPCR, 1μg of RNA was converted into cDNA using the iScript Reverse Transcription Supermix for RT-qPCR from Bio-Rad Laboratories (Bio-Rad,Cat. #1708841) using the recommended manufacturer protocol. For each qPCR, 10 μL PowerUp SYBR Green Master Mix from (ThermoFisher, Scientific (ref. 100029283), 1 μL 5μM forward primer, 1 μL 5μM reverse primer, 7 μL ultrapure water, and 1 μL of cDNA template from the previously described iScript reaction were mixed to form 20 μL reactions and were cycled at 95 ° C for 2 min, 40 cycles of 95 ° C for 15 s, 60 ° C for 15 s, and 72 ° C for 1 min. Data from RT-qPCR was analyzed using the delta-delta Ct method with RPL13A serving as the housekeeping gene and MSH2 and MSH6 as the genes of interest. The list of forward and reverse primers for each gene are shown in Table S2.

DNA methylation study of MMR genes

DNA was extracted from cell pellets of 5 million cells using DNeasy Blood & Tissue Kit from QIAGEN (Cat. No. 69504). The extracted DNA was quantified using a Qubit Fluorometer with the Qubit dsDNA HS Assay Kit (Cat. No. Q32851). Next, 500ng of extracted DNA was enzymatically converted for downstream use in MSP-PCR using the NEBNext Enzymatic Methyl-seq Conversion Module (Cat. No. E7125L) according to manufacturer protocol. Universal Methylated Moue DNA Standard from ZYMO (Cat. No. D5012) was used as a positive control for the MSP-PCRs. For MSP-PCRs, 10μL EpiTect MethyLight PCR MasterMix (w/o ROX) from the EpiTect MethyLight PCR + ROX Vial Kit (Cat. No. 59496), 1 μL 1μM forward primer, 1μL 1μM reverse primer, 6 μL ultrapure water, and 2 μL enzymatically converted DNA template were used in each 20 μL reaction. To aid with primer sensitivity and specificity, each 20 μL reaction was put through a touchdown PCR thermal cycling protocol of 95 ° C for 2 min, 3 cycles of 95 ° C for 30 s, 67 ° C for 30 s, 3 cycles of 95 ° C for 30 s, 64 ° C for 30 s, 3 cycles of 95 ° C for 30 s, 61 ° C for 30 s, and 40 cycles of 95° C for 30 s, 60 ° C for 30 s. The MSP-PCR products were visualized using D1000 ScreenTapes (Part No. 5067–5582) and D1000 Reagents (Part No. 5067–5583) on a 4200 TapeStation System (Part No. G2991BA) from Agilent Technologies. Both forward and reverse primers for MSP-PCR were designed to contain at least 2 CpG sites each. The CpG sites of interest are found within the first CpG island directly upstream of the TSS (transcription start site) for each gene of interest. The 1000 bps directly upstream of the TSS were retrieved using (https://epd.epfl.ch//index.php), and appropriate primers were designed using (http://www.urogene.org/cgi-bin/methprimer/methprimer.cgi) for all 1000-bp sequences upstream of TSS. CpG islands were defined by (http://www.urogene.org/cgi-bin/methprimer/methprimer.cgi) as at least a 100-bp region of DNA with a GC content higher than 50% and an observed CpG versus expected CpG ratio greater or equal to 0.6. The list of forward and reverse primers used for all MSP-PCRs for the promoter region of each region of interest are available in Table S2.

DNA extraction and sequencing

DNA was extracted from either cell pellets (1 million) or FFPE mice tumor tissue (microdissected to reach at least of volume 1 mm3) using DNeasy Qiagen blood and tissue kit (ref. 69556) by following manufactured protocol. After PicoGreen quantification and quality control by Agilent BioAnalyzer, 100 ng of DNA were used to prepare libraries using the KAPA Hyper Prep Kit (Kapa Biosystems KK8504) with 8 cycles of PCR. After sample barcoding, 500 ng of library were captured by hybridization using either SureSelectXT Mouse All Exon (Agilent catalog # 5190–4641) or SinglePlex Mouse Exome (Twist catalog # 102036) according to the manufacturer’s protocol. PCR amplification of the post-capture libraries was carried out for 10 cycles. Samples were run on a HiSeq 4000 in a PE100 run, using the HiSeq 3000/4000 SBS Kit (Illumina) or a NovaSeq 6000 in a PE100 run, using the NovaSeq 6000 S4 Reagent Kit (200 Cycles) (Illumina). Samples were covered to an average of 250X.

Whole exome sequencing bioinformatical analyses

Somatic variant calling in mouse

To call somatic variants from tumor samples, we built an analysis pipeline based on GATK best practices (GATK4 and Picard v2.16.0). FASTQ files of tumor and normal samples were subjected to FASTQC (v0.11.4) to check raw sequencing quality and TrimGalore (v0.6.0) to remove low-quality bases, adapters and short reads. Trimmed FASTQ files were converted to uBAM files using FastqToSam (Picard) and aligned to the mouse reference genome (GRCm38) using BWA MEM (v0.7.15). Duplicated reads were marked in aligned BAM files using MarkDuplicates (Picard). Base quality scores were recalibrated using BaseRecalibrator and ApplyBQSR (Picard). We used Mutect2 (GATK4) in tumor/normal or healthy tissue/normal mode to call SNVs and indels from recalibrated BAM files. Low quality variants calls were removed with FilterMutectCalls (GATK4). Germline variants were excluded by removing all common SNPs recorded in mouse dbSNP142 and indels recorded in the MGP5 database. To further limit false positive variants, we manually filtered the call set to keep only variants with alternative allele depth (ALT_AD) ≥ 4, mutation allele fraction (MAF) > 1%, mean base quality (MBQ) > 20; mean mapping quality (MMQ) > 50; 5), and mean position to reads end (MPOS) > 5 bp. The final variant call set was functionally annotated using Variant Effect Predictor (VEP v98) by Ensembl and classified by variant functional types.

Description of the total, gained and lost mutations analysis strategy

To identify mutations that were gained or lost compared to the parental cell line at baseline, we used SelectVariants (GATK4). Gained mutations were determined by excluding mutations identified in the parental cell line at baseline from drug-treated cell lines. Lost mutations were determined by excluding mutations identified in drug-treated cell lines from the parental cell line at baseline. Lost mutations were also assessed comparing mice tumors for each condition compared to matched cell line before engraftment treated 8W by the same agent for immunoediting analyses.

Single Base substitution and insertion/deletions signatures

Deconvolution of mutational signatures was performed using the R package MutationalPatterns (v3.4.1).³⁵ We report the strict fit results for Single Base Substitution (SBS) or Insertion/Deletions (ID). Cosine similarity result for each mutational profile is displayed to assess the profile quality.

For combined SBS signatures: Clock-like (aging): SBS1 + SBS5 + SBS40; Ultra Violet (UV) light: SBS7a + SBS7b+ SBS7c + SBS38; Tobacco: SBS4 + SBS29 + SBS92; Mismatch Repair Deficiency (MMRd): SBS3 + SBS6 + SBS14 + SBS15 + SBS20 + SBS21 + SBS26 + SBS44; Polymerase Epsilon proofreading defect (POLE): SBS10a + SBS10b + SBS28; Reactive Oxygen Species (ROS): SBS17a + SBS17b + SBS18; Temozolomide (TMZ): SBS11; Platinum: SBS31 + SBS35; Chemotherapy (Chemo)/chemical: SBS22 + SBS24 + SBS25 +SBS32 +SBS42 +SBS86 + SBS87 +SBS88 +SBS90; APOBEC: SBS2 + SBS13; Other/Unknown: sum of all other SBS signatures.

For combined ID signatures: Slippage during replication: ID1 + ID2; Tobacco: ID3; MMRd: ID7 + ID11; Non-Homologous DNA End-Joining (NHEJ): ID6 + ID8; UV light: ID13; Other/Unknown: sum of all other ID signatures.

MSKCC IMPACT panel sequencing

Pretreatment Tumoral genetic data from 16 patients included in the clinical trial were obtained from MSKCC internal database. Tumor samples were submitted to the next generation sequencing targeted “MSKCC IMPACT” panel of somatic mutations (341, 410, 468 and 505 genes successive panels). Tumoral Tumor mutational burden was assessed by MSKCC IMPACT as previously described.⁴⁴ All patients in the clinical cohort had previously signed a consent form (IRB 12–245) for genomic analysis of their tumors.

Determination of the tumoral MSI status

Pretreatment tumoral samples in the clinical trial and tumor cell lines microsatellite instability (MSI) status was confirmed using by analyzing a clinically-validated MSI impact score using the MSIsensor program.³⁴ An MSI impact score of ≥10 was sufficient to classify the sample as MSI-high. Samples with an MSI impact score <10 were classified as MSS.

Guardant Omni and Guardant Infinity assays

Plasma samples for 15 patients were obtained at baseline, before cycle 2, cycle 3, cycle 5 and at the end of treatment. Blood was collected in two 10-mL Streck tubes (Streck) from each patient. All cell-free DNA isolation and sequencing were performed at Guardant Health Inc. (Redwood City, CA), a Clinical Laboratory Improvement Amendments–certified, CAP-accredited facility. Samples from the 15 patients were analyzed with the GuardantOMNI assay. Samples were analyzed using the Guardant OMNI assay as described previously.^45,46 All samples sequenced were processed with the Guardant bioinformatics pipeline.

Samples from the 15 patients were also analyzed using Guardant Infinity assay. Isolated plasma samples were analyzed using the Reveal assay, which uses proprietary next-generation sequencing technology and a bioinformatics classifier to evaluate thousands of differentially methylated regions for tumor fraction (TF) detection. Sequencing libraries are generated using cell-free DNA that has been partitioned based on its methylation state. TF is estimated by normalizing cancer-specific differentially methylated regions with appropriately matched control regions within each sample. Comprehensive genomic profiling is accomplished using the Guardant360 genomic panel in samples with detectable ctDNA and includes genotyping of >700 genes. Promoter methylation regions are quantified relative to constitutively hypermethylated control regions, measuring gene promoter methylation relative to total cfDNA.^47,48

Identification of immunoedited mutations

To identify mutations that might undergo immunoediting, we assessed the variant allele frequency (VAF) dynamics of mutations for each patient across treatment cycles. A mutation was considered potentially immunoedited if its VAF in the last treatment cycle was at least 50% lower than its peak VAF or if it was undetectable (VAF = 0).

Preclinical treatment by anti PD1 or isotype control

For the cohorts of the mice treated with anti PD1 and Isotype control, once tumors reached an average volume of 100 mm3, mice were randomized and divided into treatment or control vehicle groups. Dosing schedule was as follows: anti PD1 antibody (Biolegend Inc., 114115) or Isotype control (Biolegend Inc.400563) 200 μg/mouse i.p. three times/week. Mice were observed daily throughout the treatment period for signs of morbidity/mortality. At the end of the study tumor samples were collected for histology and biochemistry analysis.

Intraperitoneal tumor engraftment and preclinical drug challenge

For the in vivo preclinical challenge, mice were engrafted intraperitoneally with 10,000 CT26 cancer cells (N = 5 per condition). After D3, mice received intraperitoneally, vehicle control, or a combination of daily Temozolomide at 20 mg/kg 5 times a weekand Cisplatin at 10 mg/kg once a week. Additionally, a group of mice received anti PD-1 intraperitoneally at 200 μg three times a week. Each cycle was repeated each week up to 4 weeks or until mice presented extended abdomen or unacceptable toxicities. Tumors were collected at the time of sacrifice or death for further analyses. Mice apparently cured of cancer were submitted to necropsies to assess the long-term toxicity of the combination of temozolomide+cisplatin with or without anti PD-1. An independent experiment in mice free of tumors was performed with similar regimen during 8 weeks and necropsies were performed at the end of the drug challenge or after a 12 weeks drug wash out.

Necropsy and histopathology

Mice were euthanized with carbon dioxide (CO2) inhalation. Following gross examination all organs were fixed in 10% neutral buffered formalin, followed by decalcification of bone in a formic acid solution (Surgipath Decalcifier I, Leica Biosystems). Tissues were then processed in ethanol and xylene and embedded in paraffin in a Leica ASP6025 tissue processor. Paraffin blocks were sectioned at 5 microns, stained with hematoxylin and eosin (H&E), and examined by a board-certified veterinary pathologist. The following tissues were processed and examined: heart, thymus, lungs, liver, gallbladder, kidneys, pancreas, stomach, duodenum, jejunum, ileum, cecum, colon, lymph nodes (submandibular, mesenteric), salivary glands, skin (trunk and head), urinary bladder, uterus, cervix, vagina, ovaries, oviducts, adrenal glands, spleen, thyroid gland, esophagus, trachea, spinal cord, vertebrae, sternum, femur, tibia, stifle join, skeletal muscle, nerves, skull, nasal cavity, oral cavity, teeth, ears, eyes, pituitary gland, brain. A subset of healthy tissues was further sequenced including lung, liver, brain, gastrointestinal tract and bone marrow.

Mismatch repair protein, ERRC1 and MGMT western blots and quantification

Cell pellet (1 million cells) or tumor tissue were harvested with RIPA buffer (thermoscientific-89900) containing protease and phosphatase inhibitor cocktail (thermoscientific-78425) and DNAse (Roche-10104159001). The resulting solution was spun down at 1500 g × 5 min to remove the cell debris. The supernatant was collected, and total protein was quantified using BCA assay (thermoscientific-23225). Equal amount of total proteins were loaded onto 4–20% tris-glycine gels (Thermoscientific-XP04205BOX) followed by electrophoresis at 100 V. The proteins were then transferred to PVDF membrane (InvitrogenIB24001) using immunoblot2. After blocking with 3% BSA solution for 1 h, the membranes were probed with primary antibodies overnight at 4° C (Rabbit anti MLH1, Abcam, ab92312, 1:1000 dilution; Rabbit Anti MSH2, Abcam, ab70270, 1:2000 dilution, Rabbit anti MSH6, Abcam, ab92471, 1:1000 dilution, Rabbit anti β-Actin, Abcam, ab115777, 1:2000 dilution, Rabbit Anti-MGMT, Life technology PA5–120050, 1:1000 dilution Rabbit Anti-ERCC1, Proteintech, 14586–1-AP, dilution 1:2500). The next day, the membranes were incubated with HRP conjugated secondary antibodies at room temperature for 1 h. The membranes were then visualized by chemiluminescence (Thermoscientific 34580). Quantification of the MMR proteins level was performed using ImageJ Image analysis software and MMR protein level were normalized using β-Actin protein expression.

Immunohistochemistry, immunofluorescence and quantification

Tumors obtained from mice were fixated using 4% PFA overnight and embedded in paraffin. 10 μm slices were produced for further analyses. IHC and IF experiments for CD3, CD4, CD8, PDL-1, Iba1 and Ki67 were performed at Molecular Cytology Core Facility of Memorial Sloan Kettering Cancer Center using Discovery XT processor or Ultra processor (Ventana Medical Systems-Roche).

After 32 min of heat and CC1 (Cell Conditioning 1, Ventana cat#950–500) retrieval, the tissue sections were blocked first for 30 min in Background Blocking reagent (Innovex, catalog#: NB306). A rabbit polyclonal anti-CD3 antibody (Dako, cat. #A0452) was used in 2.4 μg/mL concentrations and incubated 6 h). A rabbit monoclonal anti-CD4 antibody (abcam, cat. #183685) was used in 1.4 μg/mL concentrations and incubated for 4 h. A rabbit polyclonal anti-CD8 antibody (Cell Signaling, cat#98941) was used in 4.8 μg/mL concentrations and incubated for 6 h. A rabbit polyclonal Iba-1 antibody (Wako, cat#019–19741) was used in 0.1 μg/mL concentrations and incubated for 5 h. A goat anti-mouse PDL-1 antibody (R&D, cat#AF1019) was used in 2 μg/ml concentration and incubated for 3 h. A rabbit polyclonal Ki67 (abcam, cat#16667) was used in 2.5μg/ml concentrations and incubated for 6 h. Primary incubations were followed by 60 min incubation with biotinylated goat anti-rabbit IgG (Vector labs, cat#:PK6101) in in 5.75 μg/mL. Blocker D, Streptavidin-HRP and DAB Map kit (Cat#760–124, Ventana Medical Systems). The slides were counterstained with hematoxylin and coverslipped with Permount (Fisher Scientific).

For Immunofluorescence protocols, all IF protocols are same as above IHC protocols until Streptavidin-HRP step; after Streptavidin- HRP; we used the green color fluorophore TSA A488 (Life Tech, cat#B40932) prepared according to manufacturer instruction at 1:100 and applied for 16 min or the red color fluorophore TSA CF594 (Biotium, cat.#92174) prepared according to manufacturer instruction at 1:2000 for 16 min.

Automated quantification of immune cells and PD-L1 expressing cells was performed on 20 representative tumor area (100,000 μm² for each area) on duplicate samples. Cell nuclei were segmented using the DAPI signal. Segmented nuclei were expanded five pixels outward to encompass the cytoplasm. For each marker, if the cell contained an intensity greater than a certain threshold, that cell was marked as positive.

Immunohistochemistry images of MLH1, MSH2, and MSH6 staining in formalin fixed, paraffin embedded mouse tissue sections, were performed on a Leica Bond system using the modified protocol F in the Department of Pathology and Laboratory Medicine of Weill Cornell Medicine, New York, NY, USA. The section was pre-treated using heat mediated antigen retrieval with sodium citrate buffer (pH6, epitope retrieval solution 1, Leica) for 30 min (MLH1) or EDTA based buffer (pH 9, epitope retrieval solution 2, Leica) for 20 min (MSH2 and MSH6). The section was then incubated with MLH1/MSH2/MSH6 antibody (Abcam, ab92312, 1:300 dilution; ab70270, 1:300 dilution, and ab92471, 1:500 dilution) for 15 min at room temperature and detected using an HRP conjugated compact polymer system. DAB was used as the chromogen. The section was then counterstained with hematoxylin and mounted with micromount.

Vector plasmid

Vector plasmid (pNL2.1[Nluc/Hygro]) was purchased from Promega (cat# N1061). CNV promoter was cloned upstream of Nluc enzyme sequence. A short microsatellite of (CA)17 was cloned in an N-terminal of Nluc enzyme in such way that the enzyme sequence goes out of frame.⁴⁹ The cloning of CNV promoter and (CA)17was performed by GenScript. The resulting plasmid (M-pNL2.1) was then amplified using DH5α competent cells (Thermo Fisher Cat #EC0112) and purified using Qiagen plasmid plus midi kit (Qiagen cat# 12943).

Nanoluc assay

M-pNL2.1 was nucleaofected using SE Cell Line 4D-Nucleofector X Kit S (Lonza cat# V4XC-1032) in CT26-WT and CT26-combo cell lines. The cells were then allowed to recover for 2 weeks in presence of hygromycin (5 μg/mL). Recovered cells were then plated at 5000 cells per well in a 96 well plate and allowed to grow overnight. The next day the cells were treated with Temozolomide, Cisplatin or Temozolomide + Cisplatin for 48 h. Luminescence was detected using Nano-Glo Luciferase assay (Promega cat# N1110).

Apoptosis assessment with flow cytometry

Cells were plated at a seeding density of 1 million cells. The next day cells were treated with Temozolomide, Cisplatin, Temozolomide + Cisplatin for 48 h. 5-Flurouracil was used at positive control for apoptosis. The cells were then trypsinized and resuspended in 100 μL of apoptosis binding buffer (Thermo Scietific cat# V13246). The cells were stained with 5 μL of BD Pharmingen FITC Annexin V (BD bioscience cat#556420) per reaction for 15 min. The cells were then washed twice with 2 mL of Annexin staining buffer. 5 μL of Propidium Iodide staining solution (BD bioscience cat#556463) was then added. The stained cells were then subjected to flow cytometry analysis.

Global methylation assay

The Global Methylation Assay Kit (Abcam cat#ab233486) was used, and the manufacturer’s protocol was followed to assess global methylation. A total of 100 ng of DNA, harvested from either cells or tumors, with 100 μL of binding solution were added to the well. After gently mixing the plate, the DNA was allowed to bind at 37° C for 60 min. The wells were washed three times, followed by the addition of 50 μL of the 5-mC detection solution, and incubated for 50 min. The plate was then washed five times with wash solution. Subsequently, 100 μL of developer solution was added, and the color was allowed to develop for up to 15 min. Finally, the plate was read at 450 nm after adding the stop solution to each well.

Single-cell transcriptome sequencing

The tumor was harvested from mice and subjected to tissue digestion using tissue dissociation kit (Miltenyi cat# 130-096-730) according to manufacturer’s protocol. Single cell suspensions were stained with Trypan blue and Countess II Automated Cell Counter (ThermoFisher) was used to assess both cell number and viability. Following QC, the single cell suspension was loaded onto Chromium Next GEM Chip K (10X Genomics PN 1000286) and GEM generation, cDNA synthesis, cDNA amplification, and library preparation of ~10,000 cells proceeded using the Chromium Next GEM Single Cell 5′ Kit v2 (10X Genomics PN 1000263) according to the manufacturer’s protocol. cDNA amplification included 13 cycles and 50ng of the material was used to prepare sequencing libraries with 14 cycles of PCR. Indexed libraries were pooled equimolar and sequenced on a NovaSeq X in a PE28/88 run using the NovaSeq X 25B Reagent Kit (100 cycles) (Illumina). An average of 40 thousand paired reads was generated per cell.

Single-cell V(D)J analysis from RNA

An aliquot of cDNA generated using the methods described above was used to enrich for V(D)J regions using the Human TCR Amplification Kit (10X Genomics PN 1000252) according to the manufacturer’s protocol with 10 cycles of PCR during enrichment and 8 cycles during library preparation. Indexed libraries were pooled equimolar and sequenced on a NovaSeq X in a PE28/88 run using the NovaSeq X 25B Reagent Kit (100 cycles) (Illumina) (Illumina). An average of 780 thousand paired reads was generated per T cell.

Single cell RNA bioinformatical analysis

Raw 10X scRNA-Seq fastq file processing

For each sequenced scRNA-Seq pool, Cell Ranger (v8.0) software from 10x Genomics (https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/latest) was used to process the generated FASTQ files. The processing pipeline included alignment and quantification of unique molecular identifier (UMI) counts against the GRCm39 mouse reference genome (10x reference: refdata-gex-GRCm39–2024-A). A preliminary round of filtering was performed on the Cell Ranger platform to generate high-quality matrix files.

Cell Ranger employs a series of filtering steps to identify high-confidence cell-associated barcodes while excluding background noise. Initially, barcodes are filtered based on a whitelist of expected sequences. A barcode rank plot is then generated to differentiate high-confidence barcodes corresponding to cells from those representing empty droplets or ambient RNA. Barcodes with low total UMI counts, typically falling below the inflection point of the rank plot, are excluded from further analysis.

Next, UMIs are deduplicated by collapsing identical UMIs mapped to the same gene and barcode, effectively reducing PCR amplification biases. Only reads that map confidently to the transcriptome, such as exonic regions of annotated genes, are retained in the UMI count matrix. Reads that are multimapped or mapped to non-coding regions are excluded. Additionally, Cell Ranger estimates and corrects ambient RNA contamination originating from empty droplets, ensuring the adjustment of gene expression levels.

Finally, the filtered output matrices include only high-confidence barcodes with associated gene expression data, summarized in the filtered_feature_bc_matrix output. This ensures the downstream analysis focuses on high-quality, cell-associated data.

Downstream 10x scRNA-Seq data analysis

Filtered UMI count matrices were imported into R (v4.4.0) for further filtering and downstream processing using the Seurat R package (v5.1.0).³⁶ Mitochondrial and ribosomal RNA content were calculated for each droplet barcode using the PercentageFeatureSet function in Seurat. Droplets with a high percentage of total UMIs derived from mitochondrial RNA (>15%), low gene counts (<200), or excessively high gene counts (>2500) were filtered out. Additionally, droplets with fewer than 500 UMIs were excluded. Data normalization, variable feature identification, scaling, and dimensionality reduction were performed using Seurat functions. Principal component analysis (PCA) was employed to reduce the dimensionality of the data, and the top 10 principal components were used to construct a k-nearest neighbors (kNN) graph. Clustering was performed using the Louvain algorithm at a resolution of 0.5, and cell clusters were visualized with the UMAP algorithm (parameters: dims = 1:10, n.neighbors = 30, spread = 1, min.dist = 0.1). DoubletFinder (v2.0.4) was applied to identify and exclude potential doublets (droplets containing more than one cell) from the dataset.³⁷

Cell annotation

Cell type annotation was performed on a semi-automatic manner for higher accuracy and better cell type discrimination. Briefly the SingleR package (2.6.0) was used for automatic annotation using the MouseRNAseqData reference dataset (main and fine annotations) from the celldex (1.14.0) package.³⁸ Using this information along with the 30 overexpressed and 10 underexpressed markers per cluster we performed semi-automatic annotation using the package scType (v2021).³⁹

Differential Gene Expression

The annotated Seurat objects were saved and used for Differential Gene Expression (DGE) analysis between the respective cell types under different conditions. Prior to DGE, datasets were integrated using Seurat’s anchor-based integration workflow to account for batch effects and harmonize data across samples. Specifically, datasets were normalized using the NormalizeData function, followed by selection of the top 2000 highly variable features with SelectIntegrationFeatures. Integration anchors were identified with FindIntegrationAnchors, and the datasets were integrated with IntegrateData (k.weight = 30) to generate a combined expression matrix. Following integration, the data were scaled using ScaleData to center the gene expression values and remove technical variability. Dimensionality reduction was performed via PCA using the top 30 principal components.

Differential Gene Expression analysis was conducted using Seurat’s FindMarkers function to identify genes significantly upregulated or downregulated between the conditions. A log-fold change threshold of 0.25 and a minimum expression percentage (min.pct) of 0.1 were applied to focus on biologically relevant changes.

Gene Ontology analysis

Pathway enrichment analysis was performed to identify biological processes and pathways associated with differentially expressed genes (DEGs). Gene Set Enrichment Analysis (GSEA) was conducted using the clusterProfiler package (v. 4.12.6) to examine Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways.^40,50,51

The analysis began with the mapping of gene identifiers from the DEG results. Symbols for significantly upregulated and downregulated genes were converted into ENSEMBL and ENTREZID formats using the bitr function from clusterProfiler, ensuring compatibility with downstream enrichment tools. Genes were ranked by their log-fold change values, with the ranking lists sorted in descending order. Separate ranked gene lists were prepared for GO and KEGG enrichment analyses.

GSEA for GO biological processes was performed using the gseGO function, which identifies enriched GO terms within the ranked gene list. Parameters were set to focus on gene sets of size 10–500 genes, with a p-value cutoff of 0.05 to ensure significance. Similarly, KEGG pathway enrichment was conducted using the gseKEGG function, which assesses pathways relevant to the ranked gene list.

QUANTIFICATION AND STATISTICAL ANALYSIS

Experiments were powered with at least three or more samples for statistical analysis. According to parameters, Categorical variables are expressed in the form of frequencies or and percentages and are compared where appropriate using two-sided Fisher’s exact testing or Pearson’s Chi-Squared test. Continuous variables are calculated as mean and median with variation expressed as the (qualitative variables), mean and standard deviation (SEM) and interquartile range, respectively. Continuous variables are compared where appropriate with Kruskal Wallis rank-sum test and student t testing. For multiple comparisons, we used oneway ordinary ANOVA tests, corrected for multiplicity using the Dunnet test. Tumor growth curves are assessed with generalized multivariate analysis of variance modeling. median and interquartile range (quantitative variables). Survival curve estimates calculated with the Kaplan Meier method and compared using log rank test. Confidence intervals are constructed based on normal-based 95% confidence intervals. In all cases, a p-value less than or equal to 0.05 was considered significant.

Statistical tests included Fisher’s exact test, Kruskal-Wallis rank-sum test and Pearson’s Chi-squared test. Data has been analyzed with GraphPad Prism V9 software (Dotmatics, MA, USA) and R (v4.0.0; https://www.R-project.org/) software package. The level of significance displayed in all the figures across the article is the following: ns: non significant, p > 0.05; *: p ≤ 0.05 and p > 0.01; **: p ≤ 0.01 and >0.001, ***: p ≤ 0.001 and p > 0.0001; ****: p ≤ 0.0001. https://www.r-project.org/.

ADDITIONAL RESOURCES

Not applicable.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.ccell.2025.05.010.

Highlights.

• Tailored mutagenic chemical treatment leads to epigenetic loss of Msh2

• Temozolomide plus cisplatin induces high TMB and microsatellite instability

• Chemically induced MMRd genotype generates high immunogenicity in preclinical models

• In treated patients, MMRd signatures emerged but no responses were observed