Acquired High Tumor Mutational Burden and Activity of Immunotherapy after Targeted Therapy in Microsatellite Stable Colorectal Cancer

Abstract

Purpose:

Microsatellite stable (MSS) colorectal cancers, in contrast to microsatellite instability–high colorectal cancers, have few mutations and are insensitive to immune checkpoint blockade (ICB). Colorectal cancers treated with targeted agents often acquire a high number of genomic alterations at progression. We asked whether targeted therapy could be used to generate a high tumor mutational burden (TMB) in MSS colorectal cancer and sensitize these tumors to ICB.

Experimental Design:

In patients with MSS metastatic colorectal cancer treated with targeted therapy, we evaluated baseline and progression TMB and response to ICB for patients whose tumors developed high TMB. We determined types of alterations, mutational signatures, neoantigenicity, and clonality associated with emergent genomic alterations in cases of acquired high TMB.

Results:

Among 26 cases, nine acquired high TMB at progression. Three of these patients received ICB but none had a response. In the TMB-high cases, we found no induction of tumor-infiltrating lymphocytes or PD-L1 expression. Acquired genomic alterations consisted predominantly of single-nucleotide variants, were enriched for single base substitution 17a/b mutational signature, and did not enhance predicted MHC class I binding. TMB was higher in plasma, driven by highly subclonal acquired alterations, compared with tissue samples, which harbored few resistance alterations.

Conclusions:

A substantial number of MSS colorectal cancers acquire high TMB following targeted therapy. However, this change is not associated with sensitization to ICB. The high TMB is due to subclonal alterations unique to individual disease sites that are inadequate to elicit a robust antitumor immune response.

See related commentary by Parseghian and Eluri, p. 999

Translational Relevance.

The large majority of metastatic colorectal cancers are microsatellite stable and thus do not respond to immunotherapy. Treatment with targeted therapy, however, can result in the acquisition of a high number of genomic alterations and acquired high tumor mutation burden (TMB) in a substantial portion of patients with microsatellite stable colorectal cancer. We investigated whether this effect could be leveraged to sensitize colorectal cancers to immunotherapy. Patients whose tumors acquired high TMB had rapid progression on immunotherapy without clinical benefit. Analyses of mutational signatures, immunogenicity, clonality, and single-cell sequencing indicated that acquired resistance to targeted therapies gives rise to many “islands” (subclones) of resistant tumor, each with TMB that remains low after treatment and consists of less immunogenic variants inadequate to elicit a robust antitumor immune response.

Introduction

Immune checkpoint blockade (ICB) has transformed clinical outcomes and is now standard of care for patients with advanced, mismatch repair–deficient (MMRd)/microsatellite instability–high (MSI-H) colorectal cancer. Mismatch repair–deficient/MSI-H tumors are characterized by genetic or epigenetic inactivation of MMR genes leading to a high number of mutations and enrichment in neoantigens, conferring responsiveness to ICB (1). In contrast, with the exception of DNA polymerase epsilon (POLE)–mutated colorectal cancer (2), mismatch repair–proficient (MMRp)/microsatellite stable (MSS) tumors are generally characterized by chromosomal instability but low tumor mutational burden (TMB), resulting in low levels of tumor-infiltrating lymphocytes and minimal sensitivity to ICB (1). The MMRp/MSS subtype makes up ∼95% of metastatic colorectal cancer (mCRC) cases, and thus, there remains an urgent clinical need to develop novel treatments for these immunologically “cold” tumors.

Therapies targeting common oncogenic drivers in colorectal cancer, including EGFR and, more recently, BRAF^V600E, KRAS^G12C, and HER2 overexpression/amplification, have expanded treatment options and prolonged survival for molecularly selected patients with mCRC. However, relative to other tumor types, the success of targeted therapies in colorectal cancer has been limited by short duration of response and rapid onset of acquired resistance. Studies of acquired resistance to targeted therapy in colorectal cancer suggest that cancer cells evade the effects of treatment by inducing a program of enhanced mutagenesis (3), and analyses of patient samples at the time of disease progression commonly identify the emergence of multiple resistance alterations within the same patient (4–7). We therefore asked whether selective pressure from targeted therapy could generate a high TMB in colorectal cancer and whether this effect could be used to prime response to ICB.

Materials and Methods

Patient population

Patients studied had mCRC treated at Memorial Sloan Kettering Cancer Center (MSK) with a genomically matched targeted therapy and had progression tumor tissue and/or plasma samples genomically sequenced with panels sufficient to assess TMB. Blinding, randomization of groups, and power analysis were not relevant to this study. Patients provided written informed consent for analysis of their samples (IRB #06-107 and #12-245), and this study was performed under protocols approved by MSK’s institutional review board. The study was conducted in accordance with recognized ethical guidelines (e.g., Declaration of Helsinki, Council for International Organizations of Medical Sciences, Belmont Report, and U.S. Common Rule). Clinical and demographic details were extracted from the electronic medical record. About half of patients (14 of 26, 54%) were male. The median age was 51.3 years (range, 30.7–70.9 years).

DNA sequencing

Tumor tissue was profiled using MSK-Integrated Profiling of Actionable Cancer Targets (MSK-IMPACT), a hybridization capture–based next-generation sequencing (NGS) panel of 340 to 505 genes (8). The MSIsensor algorithm was used to determine microsatellite status as previously described (RRID: SCR_006418; ref. 9). To determine tissue TMB (tTMB), the total number of somatic, nonsynonymous mutations, including driver mutations in oncogenes, was divided by the exonic coverage of the respective MSK-IMPACT panel in megabases (Mb).

Cell-free DNA (cfDNA) was profiled using the GuardantOMNI (Guardant Health, Inc.) assay. GuardantOMNI is a targeted NGS panel that includes detection of single-nucleotide variants (SNV) and small insertions/deletions (indel) in 496 genes, copy-number variants in 106 genes, fusions in 21 genes, microsatellite instability, and plasma TMB (pTMB; refs. 10, 11). Methods for MSI-H detection (12) and pTMB (13) analysis have been described previously. Molecular pathway annotation was performed as previously described (14).

Mutational signature analysis

Mutational signature analysis was performed using the MutationalPatterns R package (v3.14.0, RRID: SCR_024247; ref. 15), which deconvolutes somatic mutations detected in cfDNA into established Catalogue of Somatic Mutations in Cancer single-base substitution (SBS) signatures (v3.2, RRID: SCR_002260; ref. 16). Cosine similarity between sample-derived and reference signatures was calculated using the cos_sim() function in MutationalPatterns, which computes the cosine similarity between vectors of equal length to assess signature similarity.

MHC-I binding prediction

Somatic variants called by the GuardantOMNI pipeline were converted to VCF files using maf2vcf (17) and annotated with SNPEff (v5.2, RRID: SCR_005191; ref. 18). We extracted mutated protein sequences and generated custom FASTA files. Patient human leukocyte antigen (HLA) class I alleles were obtained from MSK-IMPACT sequencing. MHC-binding prediction was performed using netMHCpan-4.1 (RRID: SCR_018182; ref. 19) for patients with available HLA class I genotyping. Peptides 8 to 11 amino acids in length were analyzed. Peptides with a percentile rank (EL rank) < 0.5 were deemed “strong binders,” and those with a percentile range of < 2 were deemed “weak binders.”

Immune infiltrate

Hematoxylin and eosin stains were performed on available paired (baseline and progression) tumor biopsies. The expression of CD3, CD8, and PD-L1 was evaluated by IHC staining and analyzed by dedicated subspecialty pathologists. Antibodies used were CD3 (clone LN10, dilution 1:250, Leica Biosystems, RRID: AB_3073619), CD8 (clone C8/144B, diluted 1:1,000, Dako, RRID: AB_2075537), and PD-L1 (clone E1L3N, diluted 1:400, Cell Signaling Technology, RRID: AB_2799389).

Single-cell DNA sequencing

Single nuclei from frozen core biopsies were isolated and processed as previously described (20). Briefly, nuclei were suspended in Mission Bio cell buffer at a maximum concentration of 4,000 nuclei/μL, encapsulated in Tapestri microfluidics cartridge, lysed, and barcoded (RRID: SCR_025736). Barcoded samples were then put through targeted PCR amplification with patient-specific panels, which covered mutations detected by bulk sequencing. Downstream bioinformatics analyses were performed following a previously described pipeline (21).

Results

Identification of cases with high TMB following exposure to targeted therapies

We first determined whether targeted therapy treatment converted tumors to a TMB-high phenotype. We reviewed patients with MMRp/MSS mCRC who (i) received targeted therapy at our institution; (ii) discontinued targeted therapy because of disease progression; and (iii) had plasma and/or tumor tissue biopsies collected at baseline (prior to targeted therapy) and at progression, with samples sequenced before July 16, 2024, using panels sufficient to assess TMB. We identified 26 such patients who received genomically matched therapies targeting EGFR (n = 8), KRAS^G12C (n = 11), and BRAF^V600E (n = 7) as standard of care (n = 11), off-label use (n = 2), or on a clinical trial (n = 13; Supplementary Table S1). We evaluated TMB at baseline and at progression (Fig. 1A). pTMB typically exceeded tTMB in samples collected at the same time point (Supplementary Tables S1 and S2; Supplementary Fig. S1A) likely because of plasma capturing DNA shedding across tumor sites. To account for this discordance in sample type, we set a conservative definition of acquired high TMB as follows: (i) in cases with different sample types available at baseline and progression (e.g., baseline tissue and progression plasma), at least five-fold increase in TMB from baseline to progression and in cases with the same sample type (e.g., baseline plasma and progression plasma), at least two-fold increase in TMB from baseline to progression and (ii) tTMB > 10 mut/Mb (22) or pTMB > 20 mut/Mb (13) at progression. Of the 26 patients in our series, nine had mCRC that fulfilled the above criteria for acquisition of high TMB following targeted therapy (Supplementary Table S1; Fig. 1A and B).

Figure 1.

Change in tumor mutational burden (TMB) after targeted therapy. A, Cohort of 26 patients who received targeted therapy for metastatic colorectal cancer and underwent baseline and progression tissue and/or plasma sampling. For each case, baseline TMB (top), progression TMB (middle), and fold change in TMB from baseline to progression (bottom) are shown. Sample types are shown with blue (tissue) and red (plasma) bars. Samples used to determine fold change in TMB in each case are specified in Supplementary Table S1. B, OncoPrint showing baseline genomic alterations and putative resistance alterations, categorized by molecular pathway, detected in baseline and progression samples in cases with acquired high TMB. Amp., amplification; mut/Mb, mutations per megabase; PI3K, phosphatidylinositol-3 kinase; RTK, receptor tyrosine kinase.

Molecular determinants of acquired high TMB

We sought to identify potential predictors of acquired high TMB after targeted therapy. Sequencing of baseline tumor tissue revealed no significant differences in tTMB (Supplementary Fig. S1B), MSI score (Supplementary Fig. S1C), or distribution of variant types (Supplementary Fig. S1D) between tumors that acquired high TMB after targeted therapy and those that did not. Baseline mutational signatures associated with acquired high TMB were apolipoprotein B mRNA-editing enzyme catalytic polypeptide–like (APOBEC; SBS2; P = 0.012) and UV exposure (SBS7a/c/d and SBS38; P = 0.023; Supplementary Fig. S1E). These analyses are limited (median cosine similarity per sample = 0.73; SD = 0.1) by small patient numbers and the low number of mutations detected in baseline, single-site biopsies.

Clinical response to ICB

Three patients whose tumors acquired high TMB after targeted therapy were subsequently treated with ICB (patients #21, #25, and #26; Fig. 2A). Patient #21 with RAS/BRAF wild-type (WT) metastatic rectal cancer involving the liver, lung, and bone received irinotecan plus the anti-EGFR antibody panitumumab (Supplementary Table S1). A baseline liver metastasis had a tTMB of 6.9 mut/Mb. A progression bone metastasis had a tTMB of 11.5 mut/Mb, whereas pTMB was 40.19 mut/Mb. Notably, the progression plasma sample was classified as MSI-H, possibly related to an acquired somatic pathogenic MLH1 mutation (E37*; Fig. 1B) occurring at 3% variant allele frequency (VAF). The progression bone metastasis remained MSS and MMRp by IHC. The patient then received the anti–PD-1 antibody pembrolizumab but had no tumor response with significant clinical and radiographic progression on ICB at 1 month (Fig. 2A). Patient #25 with KRAS^G12C-mutated mCRC involving the liver, lung, and peritoneum received the KRAS^G12C inhibitor sotorasib plus panitumumab (Supplementary Table S1). Progression pTMB was 145 mut/Mb (MSS). She subsequently received regorafenib with the addition of pembrolizumab upon discovery of acquired high pTMB but had clinical and radiographic progression after 1.4 months (Fig. 2A). Patient #26 with RAS/BRAF WT metastatic rectal cancer involving the liver and lymph nodes received 5-fluorouracil and irinotecan (FOLFIRI) plus panitumumab. Baseline tTMB of a rectal biopsy was 4.9 mut/Mb and pTMB was 6.7 mut/Mb. tTMB of a progressing liver metastasis was 11.5 mut/Mb. The patient was subsequently treated on a clinical trial with an anti–PD-1 antibody plus a bispecific antibody and had significant clinical and radiographic progression of disease after 1.5 months. After progression on ICB, pTMB was 124 mut/Mb (MSS; Fig. 2A).

Figure 2.

Clinical response to immune checkpoint blockade (ICB) in cases of acquired high tumor mutational burden (TMB) after targeted therapy. A, Timeline of sample collection and systemic and locoregional therapies administered to patients whose tumors acquired high TMB and subsequently received ICB (patients #21, #25, and #26). CT images before starting ICB vs. after progression on ICB are included for each patient. Patient #21, left to right: liver, lung, lymph node (arrowheads), and bone metastases. Patient #25, left to right: lung, liver, and soft-tissue metastases. Patient #26, left to right: liver and retroperitoneal lymph node (arrowheads) metastases. B, Representative images of CD3 (left), CD8 (middle), and PD-L1 (right) expression by IHC in the progression sample of patient #26.

We evaluated whether acquisition of high TMB was associated with immune infiltration. In the tumor biopsies of patients #14, #21, and #26, we quantitatively assessed the intratumoral immune infiltrate and detected very few or no tumor-infiltrating lymphocytes both before and after treatment with targeted therapy. In addition, PD-L1 expression was not detected in any sample (Fig. 2B; Supplementary Table S3).

Mutational profile and signature analysis

To better understand the lack of response to ICB, we evaluated the acquired alterations detected after targeted therapy and their potential immunogenicity. As all nine patients whose tumors acquired high TMB had progression plasma samples collected, we focused the following analyses on alterations detected in cfDNA for consistency.

First, we characterized the types of alterations that emerged on targeted therapy. Of 664 total acquired alterations across progression samples of patients whose tumors acquired high TMB, 61% were SNVs (59% missense, 2% nonsense, and 1% other), 5% were indels (5% frameshift), and 33% were copy-number variations (26% amplifications and 7% deep deletions; Fig. 3A). The increase in TMB from baseline to progression was due predominantly to an increase in SNVs (Fig. 3B). A roughly consistent proportion (median, 21%; range, 17%–39%) of acquired alterations occurred in genes belonging to the RTK/RAS/RAF/MAPK pathway (Supplementary Fig. S2A).

Figure 3.

Mutational profiles, signature analyses, and MHC-I binding prediction of genomic alterations detected in cfDNA of patients whose tumors acquired high tumor mutational burden (TMB) after targeted therapy. A, Distribution of acquired alteration types detected across progression cfDNA samples of the nine patients whose tumors acquired high TMB after targeted therapy. n, % = total number (n) and proportion (%) of variants detected across all samples. B, Comparison of mean number of variants between sequenced baseline and progression samples. n = number of patients with cfDNA samples. C, Relative exposures of mutational signatures (contribution of each mutational signature to the total number of mutations) extracted from pooled baseline (left) and acquired (right) mutations detected in all sequenced cfDNA samples (baseline and progression) of the nine patients whose tumors acquired high TMB. D, Relative exposures of mutational signatures extracted from all mutations detected in sequenced baseline (left, n = 3) and progression (right, n = 9) cfDNA samples of the patients whose tumors acquired high TMB. E, Distribution of patient harmonic-mean best rank scores of predicted neoantigens derived from baseline (left) and acquired (right) mutations detected across cfDNA samples of patients whose tumors acquired high TMB and for whom HLA class I genotyping was available through MSK-IMPACT. n = number of mutations pooled across the seven cases. P value was calculated using the Wilcoxon rank-sum test. Box plots indicate median (center line), IQR (edges), and minimum and maximum values not considered outliers (whiskers). APOBEC, apolipoprotein B mRNA-editing enzyme catalytic polypeptide–like; CNV, copy-number variant; HRD, homologous recombination deficiency; LGR, large genomic rearrangement; MMRd, mismatch repair deficiency; POLE, DNA polymerase epsilon.

To understand the mutational processes underlying acquired high TMB, we extracted SBS mutational signatures (Catalogue of Somatic Mutations in Cancer; ref. 16) from mutations detected in the cfDNA of the nine patients whose tumors acquired high TMB, pooling data by alteration status (baseline vs. acquired; Fig. 3C) and also per patient sample (Fig. 3D). The dominant signature that emerged at progression was SBS17a/b, which has been associated with DNA damage from reactive oxygen species (ROS; ref. 23).

We also compared mutational signatures extracted from the progression cfDNA of patients with and without acquired high TMB, focusing only on acquired mutations. Signatures associated with ROS (SBS 17a/b) were more commonly observed in cases of acquired high TMB (P = 0.011; Supplementary Fig. S2B and S2C).

Immunogenicity analysis

We hypothesized based on the predominance of missense mutations at progression (Fig. 3A) that acquired alterations may not generate neoantigens that are effectively presented on MHC-I and thus will be less immunogenic. Using the software netMHCpan 4.1 (19), we predicted the MHC-I binding affinity of all possible peptides originating from mutations detected in cfDNA across seven cases of acquired high TMB in which HLA class I genotyping was available. Using a binding rank < 2% threshold, we found that 47% of acquired mutations that pooled across all nine cases (median, 41%; range, 33%–69%) resulted in the generation of new putative peptide binders. Although many more peptides were generated from alterations acquired at progression, on average, peptide binding affinity did not vary significantly between acquired and baseline mutations (P = 0.46; Fig. 3E; Supplementary Fig. S2D).

Clonality of acquired alterations

We hypothesized that the acquired high mutational burden did not translate to immune responsiveness because of the subclonal nature of acquired mutations and resulting neoantigens. We found that 98% of acquired alterations pooled across the nine cases were subclonal (VAF < 0.5 maximum VAF in the sample), 98% of which were ultrasubclonal (VAF < 0.2 maximum VAF; Fig. 4A). Finally, we assessed the contribution of clonal and subclonal alterations to total pTMB. The increase in total pTMB from baseline to progression (P = 0.0003) was predominantly ultrasubclonal (P = 0.0004), whereas the change in clonal pTMB from baseline to progression was not statistically significant (P = 0.2032; Fig. 4B).

Figure 4.

Clonality analyses of genomic alterations detected in patient tumors with acquired high tumor mutational burden (TMB). A, Histogram showing distribution of variant allelic frequencies (VAF) of mutations detected in cfDNA samples (baseline and progression) of the nine patients whose tumors acquired high TMB. The VAF of each mutation was normalized against the maximum VAF detected in the same sample (VAF/maximum VAF, %). The right panel is a higher magnification of the histogram representing mutations with VAF in the ≤ 0.02 VAF/maximum VAF range. B, Comparison of baseline and progression pTMB (plasma TMB) in the three patients with cfDNA sequenced at both time points. P values were calculated using two-way ANOVA corrected for multiplicity. Box plots indicate median (center line), IQR (edges), and minimum and maximum values not considered outliers (whiskers). C, Timeline of sample collection and systemic and locoregional therapies administered to patient #14. D, Single-cell mutation heatmap of single-nucleotide DNA sequencing libraries generated from two frozen samples of a metastatic liver lesion biopsied after progression from patient #14. Each column represents a bulk data–validated variant in each sample. Each row represents a single nucleus in the library. Heat indicates VAF of each mutation in each single nucleus. E, Single-cell per-amplicon–normalized read count heatmap. Each column represents one amplicon targeting KRAS. Each row represents a matched single nucleus from D. Heat indicates normalized read count of each amplicon in each single nucleus. maxVAF, maximum VAF.

Heterogeneity of resistance mechanisms

Our data suggest that plasma profiling may indicate a high TMB by pooling together alterations from many sites of disease, whereas there remains a relatively low TMB in each individual tumor site. In progression tissue samples, we found very few (zero to one) putative resistance alterations (Fig. 1B). Their VAFs, when normalized to those of a truncal driver (e.g., KRAS^G12C, TP53, or APC mutations), were > 50% in all cases. For one patient (patient #14), we had access to frozen tissue at resistance and performed single-cell DNA (scDNA) sequencing to better understand the prevalence and co-occurrence of acquired alterations. The patient had KRAS^G12C-mutated mCRC involving the liver and lung that had progressed on standard chemotherapy (Supplementary Table S1) and received the KRAS^G12C inhibitor adagrasib in combination with the anti-EGFR antibody cetuximab (Fig. 4C). The progression plasma sample revealed a KRAS^G12C mutation at 45.9% VAF, an emergent high-level KRAS amplification, 78 acquired mutations with VAFs ranging 0.01% to 55.6% [including secondary KRAS mutations (G12A, G12V, G13D, G12D, G12S, Q61H, and Y96S with VAFs ranging 0.01%–1.2%) and NRAS mutation (G12D at 0.02% VAF)], and pTMB of 62.2 mut/Mb (Fig. 1B; Supplementary Table S2). Bulk DNA sequencing (MSK-IMPACT) of the progression liver biopsy identified none of the secondary KRAS/NRAS mutations seen in the cfDNA sample, KRAS^G12C at 75.5% VAF, and the emergent KRAS amplification (4.1-fold change) consistent with amplification of the mutant allele. tTMB was 8.2 mut/Mb. We performed scDNA sequencing on two separate cores collected at the same time as the tissue sample subjected to bulk sequencing. In the first core, KRAS^G12C was identified in the same cells with the other truncal mutations (PIK3CA^E545K and TP53^K305fs) but at a much lower allele frequency (Fig. 4D). Single-cell copy number analysis identified high copy number of KRAS amplicons in these cells (Fig. 4E). These results suggest a relative decrease in KRAS^G12C dosage likely through amplification of the WT allele. In the second core, KRAS^G12C remained clonal but with allelic imbalance (median VAF, 61.7%; median copy number, 1.46), suggesting relative increase in KRAS^G12C dosage (Fig. 4D and E; Supplementary Fig. S3).

Discussion

In the present study, we investigate a subset of MSS colorectal cancers that, following exposure to targeted therapies, develops significant increases in TMB, often detectable only in plasma. We show that despite very high pTMB, these tumors do not become sensitized to ICB. Analyses of mutational signatures, immunogenicity, and clonality provide mechanistic insights into the lack of immune responsiveness in these cases.

Among the nine cases of acquired high TMB, cfDNA contained up to 132 (median, 35; range, 6–132) acquired mutations per case, enriched for SNVs, which are less immunogenic than frameshift indels, and nearly all subclonal/ultrasubclonal. In cases with both tissue and plasma samples at progression, none of the acquired SNVs or indels identified in cfDNA were detected on corresponding single-site biopsies, suggesting that plasma profiling captures acquired alterations from many sites. At the single-cell level, we identified one likely resistance mechanism to KRAS^G12C inhibition that varied spatially—one clone harboring amplification of the WT KRAS allele and, thus, decrease in KRAS^G12C dosage, and a separate clone harboring increase in KRAS^G12C dosage. Taken together, these findings suggest the presence of profound intratumor heterogeneity and support a model in which acquired resistance to targeted therapies gives rise to many “islands” (subclones) of resistant tumor, each with TMB that remains low after treatment. Pooling together these subclones, the measured pTMB is significantly increased. In our dataset, two patients had plasma samples sequenced after progression on ICB with persistence of very high TMB, further suggesting lack of immune recognition and/or clearance of subclones contributing to acquired high total TMB. Consistent with our working model, McGranahan and colleagues (24) previously showed that high clonal neoantigen burden and low neoantigen intratumor heterogeneity are necessary to elicit T-cell immunoreactivity and robust response to ICB. This is conceivably because tumor heterogeneity dilutes down the neoantigen dosage, resulting in decreased T-cell receptor recognition, priming, and activation.

Although the mutational landscape of resistance to targeted therapies in colorectal cancer is well characterized, the underlying mechanisms that give rise to these emergent mutations remain uncertain. Russo and colleagues (3) previously demonstrated in colorectal cancer cell lines that exposure to EGFR ± BRAF–targeted agents leads to transcriptional downregulation of genes involved in DNA MMR and homologous recombination with simultaneous induction of DNA damage, increased production of ROS, and a shift toward repair of DNA damage by error-prone polymerases. Their data lend support to a model in which the therapeutic pressure associated with targeted therapies induces cellular stress, triggering adaptive mutability as a stress response. Among cases of acquired high TMB in our dataset, the dominant mutational signature that emerged at progression was SBS17a/b, a signature that has been associated with DNA damage by ROS and acquired resistance to EGFR- (25) and BRAF-targeted (5) therapies. Finally, one patient in our dataset had cfDNA that converted from MSS to MSI-H at progression, with an acquired somatic pathogenic MLH1 mutation also identified. Although limited by small patient numbers, our data add to the existing body of real-world, clinical evidence in support of adaptive mutability as an underlying mutational process in colorectal cancers that develop resistance to targeted therapies.

Our scDNA analysis is the first, to our knowledge, to provide evidence for loss of KRAS^G12C and amplification of the WT allele as putative resistance mechanisms to KRAS^G12C inhibition in colorectal cancer. Through sequencing of cfDNA, simultaneous scDNA sequencing of multiple cores from the same lesion, as well as bulk sequencing of another core from the same metastatic lesion, we show that this patient’s tumor also harbored amplification of the KRAS^G12C allele in some cells. These data indicate that high chromosomal instability likely contributes to intratumor heterogeneity under the pressure of targeted therapy.

Our group previously found that, in patients with colorectal cancer treated with targeted therapy, a higher number of acquired alterations were associated with shorter time on treatment (26). Recent data in breast and lung cancer demonstrate that APOBEC-mediated mutagenesis drives resistance to targeted therapy (27, 28), but little is known about the role of APOBEC in colorectal cancer. Our data nominate APOBEC as a baseline mutational signature that could potentially predict acquisition of high TMB and, thus, more rapid onset of acquired resistance to targeted therapy in colorectal cancer. This could serve as a future therapeutic target to delay or counteract resistance in such cases.

Because of the limited efficacy of ICB in MSS colorectal cancer (1), there are ongoing efforts to convert these “cold” tumors to an immunologically “hot” phenotype. Multiple clinical trials (29–31) have been designed to test the strategy of deliberately increasing TMB by administering mutagenic chemotherapy. The ARETHUSA study (31) is an ongoing phase II trial in which patients with MSS mCRC are primed with temozolomide, an alkylating agent known to induce inactivation of MMR genes in glioblastoma (32, 33). Patients whose tumors develop high TMB (≥ 20 mut/Mb) after temozolomide priming are subsequently treated with pembrolizumab. Paralleling our experience with targeted therapies, Crisafulli and colleagues (34) observed that high TMB following temozolomide priming largely comprised SNVs. The increase in TMB was predominantly subclonal, whereas clonal TMB remained low. In addition, post-priming tTMB varied across tumor sites even within the same metastatic lesion. Consistent with our findings, preliminary data from the ARETHUSA trial demonstrated no objective responses.

Our study has several limitations. First, this is a small retrospective study, and only three patients whose tumors acquired high TMB were ultimately treated with ICB. Second, the cohort studied was heterogeneous, with many patients receiving chemotherapy in addition to targeted therapy. Third, the assays used to calculate TMB varied between time points, with the majority of patients undergoing tissue sampling at baseline and plasma sampling at progression. Finally, other factors, such as metastatic sites involved and host immune response, may have affected tumor sensitivity to ICB. All patients whose tumors acquired high TMB in our dataset had liver involvement, a site that can confer resistance to ICB (35). Patients in our dataset had also received multiple lines of chemotherapy. However, chemotherapy for colorectal cancer does not affect patients’ ability to mount effective immune responses to infection (36) and has been shown to prime host antitumor immune responses.

Overall, our data do not support the strategy of administering ICB to patients with colorectal cancers that acquire high TMB after targeted therapy. They suggest that pTMB does not reflect a true increase in intratumoral TMB at all sites of disease but, instead, represents the sum total of very subclonal acquired alterations emerging from multiple sites of disease that individually harbor low TMB insufficient to generate a robust antitumor immune response. In order for a mutagenic process or agent to convert an immunologically “cold” tumor into a “hot” tumor, it must uniformly increase TMB in all sites of disease. Our data provide real-world clinical evidence that the TMB of individual tumor subclones remains low despite a high total pTMB after targeted therapy, precluding the generation of a robust antitumor immune response.

Supplementary Material

Supplementary Figure S1

Comparisons of genomic characteristics in patient tumors that acquired versus did not acquire high TMB after targeted therapy

Supplementary Figure S2

Mutational analyses of cases with and without acquired high TMB

Supplementary Figure S3

Distribution of single-cell KRAS G12C VAF in clinical samples

Supplementary Table S1

Disease characteristics and treatment details

Supplementary Table S2

Genomic alterations detected in plasma cfDNA and tumor tissue

Supplementary Table S3

Quantitation of tumor-infiltrating lymphocytes

Data Availability

Clinical data and NGS results are included within the article and Supplementary Data. Raw sequencing data from this study cannot be deposited in a public repository in compliance with patients’ consent agreement and to protect patients’ privacy. Original code and test data with environment setup and script running instructions are available at Github (https://github.com/oliverartz/20250521_TMBH_acquisition). Downstream analyses to reproduce figures relevant to Tapestri scDNA sequencing analysis are available at Github (https://github.com/haochenz96/tapestri_crc_krasi.git). Any additional inquiries should be addressed to the corresponding author.

Authors’ Disclosures

H. Zhang reports other support from Valar Labs and Revolution Medicines outside the submitted work. P. Ntiamoah reports other support from Bridge Dermpath Laboratory (consultant). C. Weipert reports employment with Guardant Health, Inc and ownership of Guardant Health, Inc stock. F. Keane reports other support from Takeda (consulting/honoraria) and Novartis (consulting/honoraria). N.H. Segal reports personal fees from 3T Biosciences, Pfizer, Astellas, Agenus, Regeneron, Puretech, Novartis, and Numab and other support from Roche/Genentech, Pfizer, Merck, Bristol Myers Squibb, AstraZeneca, Puretech, Immunocore, Regeneron, and Agenus outside the submitted work. A. Cercek reports personal fees from Amgen, AbbVie, Agenus, Daiichi Sankyo, Merck, Roche, Janssen, Summit, 3T Biosciences, Urogen, and Regeneron and grants and personal fees from GSK and Pfizer outside the submitted work, as well as a patent for Neoadjuvant PD-1 blockade in mismatch repair–deficient rectal cancer pending to MSK. S.B. Maron reports personal fees from Elevation Oncology, Purple Biotech, Pinetree Therapeutics, Calcium, Bolt Biotherapeutics, Amgen, and 1Cell.AI; nonfinancial support from AstraZeneca; and grants from Paige.AI, Guardant Health, and Conquer Cancer Foundation outside the submitted work. L.A. Diaz reports other support from Quest Diagnostics (board member), Epitope (board member), Function Health (consultant), Grove (consultant), Innovatus CP (consultant), Seer (consultant), Absci (consultant), GSK (consultant), Delfi (consultant), and Neophore (consultant), as well as multiple licensed patents related to technology for circulating tumor DNA analyses and mismatch repair deficiency for diagnosis and therapy (inventor). C.A. Iacobuzio-Donahue reports other support from Bristol Myers Squibb outside the submitted work. J. Shia reports other support from Tempus AI outside the submitted work. B. Rousseau reports grants and personal fees from Neophore LTD during the conduct of the study as well as personal fees from Artios pharma outside the submitted work; in addition, B. Rousseau reports a patent for PCT/US 2021/013212 issued. R. Yaeger reports grants and personal fees from Mirati Therapeutics, Revolution Medicines, Eli Lilly and Company, and Parabilis Medicines; personal fees from Erasca, Merck, and Bayer; and grants from Daiichi Sankyo, Boundless Bio, Pfizer, and Boehringer Ingelheim outside the submitted work, as well as other support from CytoDyn (Data and Safety Monitoring Board member). No disclosures were reported by the other authors.

Authors’ Contributions

C. Yeh: Conceptualization, data curation, investigation, writing–original draft, writing–review and editing. O. Artz: Formal analysis, methodology, writing–original draft, writing–review and editing. H. Zhang: Formal analysis, visualization, methodology, writing–original draft, writing–review and editing. E.-R. Karnoub: Formal analysis, investigation, writing–review and editing. P. Ntiamoah: Investigation, writing–review and editing. C. Weipert: Data curation, writing–review and editing. H. Walch: Visualization, writing–review and editing. E. Harrold: Data curation, writing–review and editing. F. Keane: Data curation, writing–review and editing. S. Chalasani: Investigation, writing–review and editing. N.H. Segal: Investigation, writing–review and editing. M.B. Foote: Investigation, writing–review and editing. A. Cercek: Investigation, writing–review and editing. A. Pagano: Investigation, visualization, writing–review and editing. S.B. Maron: Data curation, writing–review and editing. L.A. Diaz: Investigation, writing–review and editing. C.A. Iacobuzio-Donahue: Formal analysis, writing–review and editing. J. Shia: Investigation, writing–review and editing. B. Rousseau: Formal analysis, investigation, writing–original draft, writing–review and editing. R. Yaeger: Conceptualization, data curation, investigation, writing–original draft, writing–review and editing.