Introduction

In the United States, approximately 4% of people will develop colorectal cancer (CRC) at some point in their lifetime [1, 2]. In 2023, it is predicted that CRC will account for 153,000 new cancer cases and cause over 52,500 deaths in the US [3]. Patients with CRC that have metastasized eventually become refractory to traditional systemic therapeutic approaches and succumb to the disease [4]. Recently, the epidemiology of the disease has shifted, with an increased incidence observed in individuals younger than 50 years. The average age of diagnosis was 72 in the early 2000s but has decreased to age 66 in the 2020s [5]. As such, there are significant efforts to identify novel therapeutic targets and implement immunotherapies in CRC.

Although there is a relatively high mortality rate associated with CRC, improvements in molecular characterization of tumors have permitted the development of novel treatment options for a subset of patients. These improvements have extended the median overall survival (OS) of patients with CRC from approximately 12 months to 25–30 months over the past 5 decades [6, 7]. Of note, molecular targets with therapeutic implications include epithelial growth factor receptor (EGFR) and V-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog (KRAS), among others [8]. Targeting known drivers of cancer progression is critical to expanding the fraction of patients who may benefit from precision treatment of their CRC [8]. Most recently, other genetic abnormalities, including the high frequency of FBXW7 mutations, a finding that portends a poor prognosis, are under active investigation as additional potential molecular targets in CRC [9]. However, a majority of patients with CRC have tumors that are microsatellite stable (MSS), a genotype that globally tends to be resistant to immunotherapies [10]. With continued advancements in molecular characterization of MSS CRC tumors, it is hoped that methods of immunological targeting will be developed to improve upon current immunotherapy success levels in patients presenting with an MSS genotype.

A promising cellular therapeutic target under investigation is mesothelin (MSLN). Mesothelin, a protein with a suspected cell adhesion function in normal tissues, is overexpressed in significant subsets of patients with numerous cancer types, including mesothelioma, lung, pancreas, and CRC [11,12,13,14,15,16]. Both preclinical and clinical studies examining MSLN expression in mesothelioma, pancreatic cancer, and lung adenocarcinoma show evidence that enhanced MSLN expression portends a reduction in survival in animals and humans [13, 14, 17, 18]. Specifically, in CRC, increased MSLN expression correlated with the development of metastatic disease, a reduction in patient survival, and an upregulation of CRC cell proliferation [11]. Numerous efforts to develop MSLN-targeted therapies have been underway based on these preclinical data, specifically in mesothelioma, lung, and pancreatic cancers [11,12,13,15,17,18,19]. However, work is still needed to exploit the therapeutic potential of high levels of MSLN expression in CRC.

This study seeks to correlate MSLN expression with known CRC prognostic signatures, including tumor-sidedness, metastatic sites, and the consensus molecular subtypes (CMS). Additionally, we seek to uncover associations of MSLN expression with specific CRC and immune-related molecular markers through examination of these relationships in all CRC patients and in the subset of MSS CRC patients. These interactions could identify new therapeutic options for MSS CRC patients with high levels of MSLN expression and lead to further investigations into therapeutics with clinical relevance to MSLN expression patterns.

Methods

Next generation sequencing (NGS)

Next generation sequencing was performed on genomic DNA isolated from 14,892 formalin-fixed paraffin-embedded (FFPE) CRC tumor samples by a commercial CAP/CLIA lab (Caris Life Sciences, Phoenix, AZ) using the NextSeq or NovaSeq 6000 platforms (Illumina, Inc., San Diego, CA). The Caris platform uses a 592-gene panel and a 700-gene panel at high depth and coverage, as previously described [20]. Tumor enrichment was performed using manual microdissection techniques. Genetic variants identified were interpreted by board-certified molecular geneticists and categorized as ‘pathogenic,’ ‘likely pathogenic,’ ‘variant of unknown significance,’ ‘likely benign,’ or ‘benign,’ according to the American College of Medical Genetics and Genomics (ACMG) standards. When assessing mutation frequencies of individual genes, ’pathogenic,’ and ‘likely pathogenic’ were counted as mutations [21]. The copy number alteration (CNA) of each exon was determined by the average depth of the sample along with the sequencing depth of each exon and compared to a pre-calibrated value.

Tumor mutation burden (TMB)

Tumor mutation burden was measured by counting all mutations found per tumor that had not been previously denoted as germline alterations in dbSNP151, Genome Aggregation Database (gnomAD) databases, or benign variants identified by Caris geneticists as previously described [22]. A cutoff level of ≥10 mutations per MB was used to characterize tumors as TMB high based on evidence from the KEYNOTE-158 pembrolizumab trial, which showed that patients with a TMB of ≥10 mt/MB across several tumor types had higher response rates than patients with a TMB of <10 mt/MB [23].

Whole transcriptome sequencing (WTS)

A Qiagen RNA FFPE tissue extraction kit (Germantown, MD) was used for extraction, and the RNA quality and quantity were determined using the Agilent TapeStation (Santa Clara, CA) prior to WTS on the Illumina NovaSeq platform (Illumina, Inc., San Diego, CA) as previously described [24]. For transcript counting, transcripts per million numbers were generated using the Salmon expression pipeline [25]. RNA-deconvolution was performed via quanTIseq to assess immune cell infiltration within the tumor microenvironment (TME) [26]. Tumors were characterized as MSLN-high(H) and MSLN-low(L) based on the top and bottom quartile of transcripts per million (TPM) expression, respectively. Gene set enrichment analysis (GSEA) was conducted in order to examine the enrichment or depletion of groups of genes associated with different biological pathways. This analysis was performed using Broad Institute software [27].

Immunohistochemistry (IHC)

Immunohistochemistry of PD-L1 (SP142 clone), MLH1 (M1 clone), MSH2 (G2191129 clone), MSH6 (44 clone), and PMS2 (EPR3947 clone) was completed on FFPE tissue slides. Slides were stained as per the manufacturer’s instructions (Ventana Medical Systems, Inc. Tucson, AZ), and were optimized and validated per CLIA/CAP and ISO requirements. Staining was scored for intensity (0 = no staining; 1+ = weak staining; 2+ = moderate staining; 3+ = strong staining) and staining percentage (0–100%). The complete absence of protein expression of any of the 4 proteins tested (0+ in 100% of cells) was considered deficient MMR. A board-certified pathologist evaluated all IHC results independently.

Mesothelin IHC

Mesothelin expression (MN-1 clone, 1:100–1:2000 dilution) by IHC was evaluated by a pathologist blinded to the clinicopathological characteristics using percentage staining and staining intensity. MSLN IHC positivity was determined to be positive if the sum of percentage staining and staining intensity was ≥3. <3 (1 or 2) was considered negative. Additional MSLN IHC details have been reported previously [28].

Microsatellite instability or mismatch repair (MSI/MMR) status

Multiple test platforms were used to determine the MSI or MMR status of the tumors profiled, including fragment analysis (FA, Promega, Madison, WI), IHC (see IHC method), and NGS (7000 target microsatellite loci were examined and compared to the reference genome hg19 from the University of California). The three platforms generated highly concordant results, as previously reported. In the rare cases of discordant results, the MSI or MMR status of the tumor was determined in the order of IHC, FA, and NGS [29].

Mitogen activated protein kinase activation score (MPAS)

MAPK activation score (MPAS) was calculated based on the TPM values of RNA expression of CCND1, DUSP4, DUSP6, EPHA2, EPHA4, ETV4, ETV5, PHLDA1, SPRY2, and SPRY4 using a previously reported algorithm as a transcriptomic indicator of MAPK pathway activation [30].

CODEai™

Insurance claims data were used to calculate real-world overall survival (rwOS) via ‘first of treatment’ to ‘last contact’ patient records. Patients were monitored from the time of first treatment through the last clinical contact for these calculations, and any patients without contact/claims data for a period of at least 100 days were presumed deceased. Kaplan–Meier estimates were calculated for molecularly defined patient cohorts.

Statistical analysis

Statistical significance was determined using Chi-square and Mann–Whitney U tests, with p values adjusted for multiple comparisons (q ≤ 0.05). rwOS significance was determined with p ≤ 0.05.

Results

Mesothelin cohort demographics and expression distribution in primary tumors, metastatic site, tumor side, and CMS subtypes

The study population comprised 7446 patients representing the low and high MSLN expressing quartiles of the 14,892 total CRC patient samples in the Caris Life Sciences database, of which 6847 patients were assigned to the “MSS cohort” according to MSI status. Both the entire patient cohort and MSS cohort were dichotomized as MSLN Low or MSLN High. Shown in Table 1A, the MSLN low group included 3723 patients with 54.9% being male and 45.1% female. The MSLN high group included 3723 patients, with 52.9% being male and 47.1% female. The median age was 63 years and 62 years respectively. In Table 1B, the MSLN low cohort included 3377 patients with a median age of 62 years, with 56.5% being male and 43.5% female. The MSLN high group was comprised of 3470 patients, 53.7% of whom were male and 46.3% female. The median age was 62 years for both MSLN expression levels in the MSS cohort.

Table 1 Demographic patient data by mesothelin (MSLN) expression level in the entire cohort of patients (A) and MSS cohort of patients (B). Patients were categorized as either MSLN low or MSLN high by RNA TPM quartiles.
Full size table

MSLN expression patterns were compared across primary and metastatic sites, metastatic location, CRC side of origin, and CMS subtypes (Fig. 1). Figure 1A, B shows MSLN expression in the primary tumor versus metastatic sites in all patients. Overall, tumor samples from metastatic sites expressed MSLN at significantly higher levels compared to the primary tumor site (q ≤ 0.01) (Fig. 1A). Samples from metastases to the skin (39.4 transcripts per million (TPM)), connective/soft tissue (11.9 TPM), and the peritoneum/retroperitoneum (11.7 TPM) exhibit the highest MSLN expression levels amongst metastatic sites (Fig. 1B). Shown in Fig. 1C, MSLN expression was highest in right-sided CRC tumors in both the entire cohort (6.1 TPM) and MSS cohorts (6.2 TPM) (q ≤ 0.01). Additionally, MSLN expression was high in transverse tumor locations in both patient cohorts. Left-sided tumors exhibited the lowest median MSLN expression levels at 4.8 TPM. Similar expression patterns between the entire and MSS cohorts were observed when comparing CMS subtypes (Fig. 1D). Both cohorts exhibited the highest MSLN expression in CMS1 (8.3 and 10.4 TPM, respectively, q ≤ 0.001) and CMS4 (10.1 and 10.0 TPM respectively, q ≤ 0.001). Moreover, we analyzed MSLN IHC protein expression in CRC normal paired/matched and CRC tumor tissues. Our data suggests a higher proportion of MSLN positivity in paired/matched tumor tissue (Supplementary Fig 1B, D, F) compared to normal tissue (Supplementary Fig 1A, C, E).

Fig. 1: Mesothelin expression in CRC is enhanced in metastatic tissue, right-sided CRC, and CMS1 and CMS4 molecular subtypes.
figure 1

Median MSLN TPM expression was analyzed in primary and metastatic CRC (A), various sites of CRC metastasis (B), primary tumor sidedness across patient cohorts (C), and CMS subtypes across patient cohorts (D). Analysis was completed via WTS and subpopulations were tested for statistical significance using Chi-square and Mann–Whitney U tests, with p values adjusted for multiple comparisons. *q < 0.05; **q < 0.01; ***q < 0.001, ****q < 0.0001.

Full size image

Mesothelin expression is associated with increased mutations in several cancer-associated genes

Numerous oncogenic driver mutation rates and their relation to MSLN gene expression were explored. Within the entire and MSS patient cohorts, comparisons were made between MSLN high-expressing tissues and MSLN low-expressing tissues (Fig. 2). High and low expression quartiles were determined as described in the methods above. The entire cohort of patients (Fig. 2A) displayed significant differences for numerous gene mutations between the MSLN high and low expression cohorts. APC (p = 1.7e-20), BCL9 (p = 0.0006), CREBBP (p = 8.5e−5), FLCN (p = 1.39e−7), NSD2 (p = 0.002), and EP300 (p = 1.42e−6) all had significantly higher mutation rates in MSLN low expressing tumor samples compared to MSLN high expressing tumor samples (Fig. 2A). Conversely, KRAS (p = 2.67e−90), FBXW7 (p = 1.29e−14), BRAF (p = 9.32e−8), GNAS (p = 2.54e−29), and SMAD2 (p = 3.93e−5) mutated tumors were significantly more common in MSLN high expressing tumor samples versus MSLN low expressing tumor samples (Fig. 2A). These mutation patterns were quite similar in the MSS cohort. APC (p = 8.35e−26), TSC2 (p = 0.001), FLCN (p = 3.77e−7), and MAP2K4 (p = 0.002) were significantly more likely to be mutated in MSLN low tumor samples whereas KRAS (p = 2.58e−89), FBXW7 (p = 8.38e−16), RNF43 (p = 4.06e−10), GNAS (p = 4.36e−29), SMAD2 (p = 4.7e−6), BMPR1A (p = 1e−4), and STK11 (p = 8.61e−6) were significantly more likely to be mutated in MSLN high tumor samples (Fig. 2B). Of note, APC, KRAS, and TP53, regardless of patient cohort, exhibited high mutation rates of at least 30% in each MSLN expression quartile. Full mutation data are presented in Supplementary Table 1.

Fig. 2: Mesothelin expression yields high rates of oncogenic driver mutations.
figure 2

Mutation analysis was performed in CRC samples expressing low and high MSLN across the entire cohort (A) and MSS cohort (B). Mutation frequencies were calculated via NGS and statistical significance was determined using Chi-square and Mann–Whitney U tests, where *p < 0.05 is considered significant.

Full size image

Mesothelin high tumors exhibit greater expression of PD-L1 and higher T-cell inflamed score, immune cell infiltration, and expression of immunosuppressive genes

Figure 3 illustrates the prevalence of immune markers across MSLN high and low tissue expression split into the two cohorts as mentioned previously. Across the entire cohort (Fig. 3A), MSLN low expression had a significant association with higher tumor mutation burden (TMB) and DNA mismatch repair (dMMR)/microsatellite instability-high (MSI-H) positivity in comparison to MSLN high expression. However, MSLN high expression exhibited an association with high PD-L1 expression via IHC (IHC-PD-L1) compared to MSLN low expression. In the MSS cohort, MSLN high tumor samples yielded significantly higher IHC-PD-L1 positivity compared to MSLN low expression (Fig. 3B). Figure 3C, D shows similar results regarding T-cell inflammation quantification and IFN-γ scores. Across both cohorts, it was found that MSLN high expression significantly correlated with both markers of T-cell inflammation and low IFN-γ scores.

Fig. 3: Microsatellite stable tumors with high MSLN expression yield high PD-L1 staining and T-cell inflammation.
figure 3

Immune landscape characterization was quantified using tumor mutation burden (TMB), dMMR/MSI-high status, and PD-L1 percent positivity (A) T-cell inflamed score (B) and IFN-y score (C). Analyses and statistical comparisons were performed in MSLN low and high CRC, with adjustments made for multiple comparisons. *q < 0.05; **q < 0.01; ***q < 0.001.

Full size image

Specific immune cell infiltration into the TME (Fig. 4A) and immune gene expression (Fig. 4B) were examined within each cohort in addition to the immune marker results presented in Fig. 3. Several immune cell types including B cells, M1 and M2 macrophages, neutrophils, natural killer (NK) cells, and regulatory T-cells (Tregs) were more prevalent in MSLN high expressing tumors across both the entire cohort and MSS cohort. Specifically, M1 and M2 macrophages were most significant within these groups. Macrophage infiltration-associated cytokine and growth factor expression was also increased in MSLN high CRC across both patient cohorts (Supp. Fig. 2A, B). Only dendritic cells (DC) were more prevalent in MSLN low tumors. Full immune cell fraction data values are presented in supplementary table 2. Immune marker gene expression levels, shown in Fig. 4B, were all higher in MSLN high tumors across both cohorts except for IL12A. Particularly, HAVCR2 (TIM-3), CD80, CD86, and IL1B expression were of the highest magnitude change between MSLN high and low-expressing tumors. In the entire cohort of patients, comparing MSLN high and MSLN low groups, HAVCR2 (15.33 TPM vs. 8.39 TPM, q ≤ 0.05), CD80 (4.14 TPM vs. 2.73 TPM, q ≤ 0.05), CD86 (7.07 TPM vs. 4.12 TPM, q ≤ 0.05), and IL1B (9.00 TPM vs. 8.76 TPM, q ≤ 0.05) were all significantly higher in the MSLN high group. A relationship in HAVCR2 expression was seen in the MSS cohort between MSLN high and low tissue (15.00 TPM vs 8.03 TPM, q ≤ 0.05). In the MSS cohort, MSLN high tissue again yielded increased gene expression in HAVCR2 (15.00 TPM vs. 8.03 TPM, q ≤ 0.05), CD80 (4.08 TPM vs. 2.66 TPM, q ≤ 0.05), CD86 (8.76 TPM vs. 3.97 TPM, q ≤ 0.05), and IL1B (8.76 TPM vs. 6.71 TPM, q ≤ 0.05).

Fig. 4: Significant immune cell fractions in MSLN high CRC are found in B cells macrophages, neutrophils, NK cells, and T-regs regardless of patient cohort.
figure 4

RNA-seq using Quantiseq and subsequent deconvolution to estimate immune cell fraction (A) further characterized the tumor immune microenvironment. Immune-related gene expression (B) was analyzed in MSLN low and high CRC in the entire and MSS cohorts via WTS, where Chi-square and Mann–Whitney U tests were implemented for statistical analyses, with p values adjusted for multiple comparisons. *q < 0.05; **q < 0.01; ***q < 0.001.

Full size image

Gene set enrichment analysis of MSLN high tumors

The association of immune cell recruitment, tumor microenvironment, and corresponding gene sets related to immune response can support the possibility of a gene target being a marker for tumor susceptibility to antigen-specific immunotherapy. Complementing the immune marker and microenvironment analyses, a gene set enrichment analysis (GSEA) was performed to evaluate the differences in MSLN-associated gene expression between patients with MSLN low and MSLN high tumors within each cohort. For all gene sets shown (Fig. 5A, B), a positive normalized enrichment score (NES) was observed, indicating higher gene enrichment for patients with MSLN high tumors. Each cohort of patients exhibited significantly high NES for gene sets related to immune response related to MSLN high expression, shown using red bars. Specifically, TNFα signaling via NFKβ (NES = 1.33, false discovery rate (FDR) = 0.09), IL2 STAT5 signaling (NES = 1.32, FDR = 0.08), IFN-γ response (NES = 1.33, FDR = 0.09), and IL6 JAK STAT3 signaling (NES = 1.29, FDR = 0.12) were three such pathway enrichments related to immune response that were significantly enriched in patients with MSLN high tumors in contract with those who had MSLN low tumors. In the MSS cohort, it must first be noted that one of the most highly enriched pathways was that of the inflammatory response (NES = 1.44, FDR = 0.01). Additionally, immune-related pathways including IFN-γ Response (NES = 1.44, FDR = 0.02), IL2 STAT 5 Signaling (NES = 1.37, FDR = 0.06), IFNα Response (NES = 1.37, FDR = 0.07), IL6 JAK STAT3 Signaling (NES = 1.36, FDR = 0.07), and TNFα Signaling via NFKβ (NES = 1.39, FDR = 0.09) were all highly significant in the MSS cohort. These pathways were more significant based on FDR in the MSS cohort than in the entire cohort, possibly suggesting a higher role of immune modulation in patients with MSS tumors.

Fig. 5: Microsatellite stable CRC with high MSLN exhibits significant gene enrichment for inflammation and immune response.
figure 5

Gene set enrichment analysis (GSEA) was performed in MSLN low and high CRC entire cohort (A) and MSS cohort (B). Positive values for normalized enrichment score (NES) indicate enhanced pathway signaling for the indicated gene set. Where FDR ≤ 0.25 is considered significant between MSS and the entire cohort. Significantly different gene sets are indicated in red.

Full size image

Patient outcomes in association with MSLN expression

Survival outcomes relative to MSLN expression levels were analyzed using Caris CODEai™. Patient outcomes in each cohort of patients (MSS cohort and entire cohort) have been compared across various treatment modalities based on chemotherapy (5-fluorouracil (FU) based) or immune checkpoint inhibitory (ICI) therapy (nivolumab (nivo), ipilimumab (ipi), and pembrolizumab (pembro). Both the entire cohort (Fig. 6A, Supp. Fig. 3A) and the MSS cohort of patients (Fig. 6B, Supp. Fig. 3B) displayed similar patterns in that MSLN high expression had an associated lower survival than patients whose tumors were MSLN low in the 5-FU based chemotherapy-treated group. In the entire cohort, nivolumab (18.1 months vs 11.4 months, HR = 1.433, 95% CI 0.854–2.403, p = 0.172), ipilimumab (Inf vs 21.1 months, HR = 2.596, 95% CI 0.798 – 8.444, p = 0.1), and the combination Ipi-Nivo (18.1 months vs 11.4 months, HR = 1.447, 95% CI 0.863–2.426, p = 0.16) were the exceptions in that patients with MSLN high tumors trended towards a longer median survival time compared to those with MSLN low tumors (Fig. 6A, Supp. Fig. 3A). In the MSS cohort, patients with MSLN high tumors treated with immunotherapy exhibited trends toward better survival. Patients with tumors exhibiting MSLN high expression survived a median of 13.0 months vs 4.6 months in those with MSLN low tumors when treated with nivolumab (HR = 1.731, 95% CI 0.957–3.13, p = 0.068, Supp Fig. 3B). Patients treated with pembrolizumab whose tumors exhibited MSLN high expression had an improved survival trend at a median of 10.6 months while patients with MSLN low tumors survived a median of 8.0 months (HR = 1.044, 95% CI 0.651–1.675, p = 0.86, Fig. 6B). The Ipi-Nivo combination therapy also resulted in an improved median survival trend for patients with MSLN high tumors of 13.0 months versus those with MSLN low tumors who had a median survival time of 4.6 months (HR = 1.762, 95% CI 0.975–3.186, p = 0.059, Supp. Fig. 3B). Finally, ICI treatment in MSLN high tumor patients yielded a median survival time of 12.2 months, while MSLN low tumor patients survived a median of 8.0 months (HR = 1.221, 95% CI 0.855–1.743, p = 0.272, Fig. 6B).

Fig. 6: Patient survival outcomes improve for MSLN high MSS CRC when treated with pembrolizumab and ICI.
figure 6

Patient insurance claims data provided to Caris Life Sciences were used to generate Kaplan–Meier curves via the CODEai™ data portal. Curves depict survival from the time of tissue collection or first treatment to last contact for the entire cohort (A) and MSS cohort (B) of CRC patients with MSLN low (blue/cohort 1) compared to MSLN high (red/cohort 2) across several therapeutic interventions.

Full size image

Survival outcomes in patients treated with immune checkpoint inhibitors, analyzed based on MSLN expression quartiles

Survival outcomes were again analyzed via Caris CODEai™ using the highest and lowest 25% MSLN expression quartiles to further evaluate the correlation of MSLN expression with treatment efficacy. Each cohort, the entire cohort (Fig. 7A, Supp. Fig. 4A) and the MSS cohort (Fig. 7B, Supp. Fig. 4B), were categorized by treatment intervention, all being immunotherapeutic in nature: nivolumab, pembrolizumab, Ipi-Nivo combination, and all ICI combined. For the entire cohort (Fig. 7A, Supp. Fig. 4A), patients with tumors in the lowest 25% MSLN expression (blue) had a trend toward longer median survival time than patients with tumors exhibiting the highest 25% MSLN expression (red) in pembrolizumab (20.6 months vs 18.8 months; HR 0.94, 95% CI 0.574–1.55, p = 0.816), nivolumab (21.1 months vs 15.6 months; HR 1.18, 95% CI 0.528–2.495, p = 0.73), Ipi-Nivo combination (21.1 months vs 15.6 months; HR 1.172, 95% CI 0.54–2.547), and the combination of each ICI (20.6 months vs 18.4 months; HR 1.08, 95% CI 0.724–1.625, p = 0.694) treatment groups. In the MSS patient cohort (Fig. 7B, Supp. Fig. 4B), the opposite effect of ICI appeared to be present. Across pembrolizumab, nivolumab, Ipi-Nivo, and all ICI, the patients with the highest 25% MSLN expression in their tumors (red) survived longer than the patients with tumors exhibiting the lowest 25% MSLN expression (blue). Median survival times for pembrolizumab were 8.0 months and 11.1 months for the bottom and top 25% MSLN expression groupings, respectively (HR = 1.164, 95% CI = 0.629–2.151, p = 0.629). In the nivolumab-treated group, median survival times were 3.9 months and 12.2 months for the bottom and top 25% MSLN expression cohorts, respectively (HR = 4.87, 95% CI 1.583–14.983, p = 0.003). The combined Ipi-Nivo treatment yielded survival times of 3.9 months and 12.2 months for the bottom 25% of the MSLN-expressing cohort and the top 25% of the MSLN-expressing cohort, respectively (HR = 5.128, 95% CI 0.67–15.747, p = 0.002). In the combined ICI curve, patients with tumors exhibiting the lowest 25% MSLN expression survived a median of 7.5 months. Those whose tumors had the highest 25% MSLN expression survived a median of 12.7 months (HR = 1.423, 95% CI 0.871–2.326, p = 0.157). Both the cohorts with high MSLN expression who were treated with nivolumab had significantly longer survival times at p = 0.003.

Fig. 7: Patient survival outcomes improve in MSLN high tumors with MSS status across all immunotherapeutic treatment strategies.
figure 7

Patient insurance claims data provided to Caris Life Sciences were used to generate Kaplan–Meier curves via the CODEai™ data portal. Curves depict survival from the time of tissue collection or first treatment to last contact for the entire cohort (A) and MSS cohort (B) of CRC patients by the lowest 25% MSLN expression quartile (blue/cohort 1) and highest 25% MSLN expression quartile (red/cohort 2) across several therapeutic interventions.

Full size image

Discussion

Colorectal cancer is the third most common cause of cancer and the second highest cause of cancer-related mortality in the United States [31]. There is significant heterogeneity among patients with CRC in terms of genomic abnormalities that drive tumor progression and these differences correlate with prognosis. By establishing connections between novel biomarkers and prognostic signatures of CRC, including tumor sidedness, metastatic sites, CMS category, and immune cell recruitment, we may be able to preferentially select treatments for patients with enhanced expression patterns of these biomarkers, possibly leading to an improvement in the survival outcomes. Mesothelin, shown to play a role in promoting cancer cell survival, proliferation, and invasion of cancer cell lines enabling cancer cells to thrive in an inflammatory milieu, is significantly overexpressed in CRC tissue compared to healthy colorectal tissue and is a candidate biomarker for targeted therapy [18, 32, 33].

Our work begins to uncover the relevance of MSLN overexpression in CRC patients as well as the implications of MSLN expression levels on pivotal CRC prognostic information, including CRC tumor sidedness, CMS subtype, genetic mutation landscape, and immune response modulation. We have shown that increased MSLN expression is positively correlated with right-sided CRC, CMS1, and CMS4 molecular subtypes, possibly playing a role in the poor prognosis associated with right-sided CRC and CMS1/4 molecular subtypes [34, 35]. Further, we have shown high positive protein IHC staining of MSLN in CRC compared to normal tissue (Supp. Figure 1), consistent with transcript data from The Cancer Genome Atlas and The Protein Atlas, as well as two previous studies conducted in CRC and pancreatic cancer [36,37,38]. Moreover, MSLN expression results discussed here specifically increases in metastatic distributions, are also reflected in previous works [28].

It is understood that CMS1 is characterized by increased immune-related gene expression, including activation of immune evasion that is commonly observed in MSI CRC, with high rates of right-sided tumors and poor survival rates with disease relapse [39]. CMS2 is typically categorized by upregulation of WNT and MYC, leading to carcinogenesis, and enhanced epithelial cell differentiation [39]. CMS3 is often seen with mixed MSI status but high rates of KRAS mutation and CMS4 is associated with high rates of epithelial-to-mesenchymal transition (EMT) gene expression in addition to activation of TGF-β and inflammatory pathways, contributing to the poorest overall survival among CMS subtypes [39]. Regarding tumor sidedness, previous work has shown that right-sided tumors are typically associated with poorer 5-year survival (55% vs 59%) and an increased risk of CRC recurrence versus left-sided tumors [40, 41].

As stated, our results show an increased association of CMS1, CMS4, and right-sided CRC with high MSLN expression. Due to this relationship, it can be posited that this increase in MSLN gene expression also plays a role in the associated features of these molecular subtypes and tumor location. The immune and inflammatory enhancements seen with CMS1 and CMS4 CRC are mirrored in the increased immune cell recruitment, immune-related gene expression, and enhancement of immune and inflammatory pathways shown with high MSLN expression (Figs. 4 and 5). Further, CMS4 has specifically been linked to EMT enhancements creating desirable conditions for rapid tumor growth and metastatic activity [42]. It is plausible that with increased MSLN expression, specifically in CMS4, there may be a link between MSLN expression and EMT, though further studies should be conducted to fully support this relationship.

Bringing focused attention to the MSS cohort of patients and the relationship between MSLN and EMT activation, we have shown that high MSLN expression in this subset of patients exhibits increased TGF-β signaling (Fig. 5B), KRAS signaling (Fig. 5B), MAPK activation (Supp. Fig. 4), and significant enrichment of the EMT gene set (Fig. 5B) [43]. Each of these factors, in addition to associated increased CA125/MUC16 expression (Supp. Figure 6), an MSLN binding domain, which has also been shown to play a role in tumorigenesis and STAT3 phosphorylation [44], supports the consideration of MSLN as a prospective biomarker in CRC.

Directly related to the potential clinical relevance of MSLN as a biomarker is the difference in survival time for patients with tumors exhibiting high MSLN expression versus low MSLN expression across multiple therapeutic regimens. Illustrated in Fig. 6 and Supplementary Fig. 3, CODEai™ analysis revealed some significant differences across these patient cohorts, specifically in patients with tumors classified as MSS. MSS tumors that had high MSLN expression had a worse associated median survival time when treated with 5-FU-chemotherapy and ipilimumab alone (Supp. Figure 3B). Notably, there was a shift in this pattern when patients with MSLN high tumors that were also MSS exhibited a higher median survival time than the MSLN low cohort when treated with nivolumab, pembrolizumab, Ipi-Nivo combination, or the ICI combination. By determining MSLN expression, oncologists may choose to recommend treatment with ICI, and that strategy may enhance survival outcomes. The patients with the highest and lowest quartiles of MSLN expression were also investigated using CODEai™ analysis (Fig. 7, Supp. Figure 5). The lowest 25% of MSLN-expressing tumors in the entire cohort exhibited longer survival times than the tumors with the highest 25% of MSLN expression when treated with different forms of ICI (Fig. 7A, Supp. Fig. 5). What is most noteworthy, is that when we analyzed the cohort of patients with tumors classified as MSS, once again, the survival patterns shifted, with the MSLN high cohorts surviving longer than the MSLN low cohort across all ICI treatment groups (consisting of pembrolizumab, nivolumab, Ipi-Nivo, and combination ICI treatment) (Fig. 7B, Supp. Fig. 5). It is this combination of high MSLN expression and MSS status that yields an improved survival with ICI treatment that implicates MSLN as a possible biomarker in MSS CRCs. Moreover, it may be suggested that this improved survival outcome with ICI in patients with MSLN high MSS CRC is due to the immunosuppressive TME induced by increased MSLN expression.

Our study has demonstrated a strong association between MSLN expression and a differentially modulated immune microenvironment, highlighted by an increase in M1/M2 macrophage recruitment and macrophage-associated cytokine and growth factor expression, but a lowered IFN-γ score and general immunosuppressive effect, though pathway activation is shown in GSEA analysis. It should also be emphasized that high MSLN expression, especially in MSS patients, was associated with significantly increased IHC-PD-L1 staining (Fig. 3B) and enhanced gene expression of PD-1, CTLA4, FOXP3, LAG3, and TIM3 (Fig. 4B). Further, total gene set enrichment analysis revealed significant activation of gene sets associated with TGF-β signaling, IL6/JAK2/STAT pathway signaling, IL2/STAT5 pathway signaling, and PIK3/AKT/mTOR signaling (Fig. 5B). These findings are supported by other studies that have described the effects of MSLN-related immunosuppression and tumor immune escape in other cancers via upregulation of immunosuppressive genes and cytokine production [16, 45].

Immune cell response, relative ‘heat’ of the tumor microenvironment, immune/inflammatory pathway activations, and genetic mutation profiles may predict the likelihood of benefit from immunotherapy in relation to MSLN expression, specifically in patients with MSS tumors. Regarding gene mutation patterns, recent studies have determined that increased mutation in KRAS and FBXW7 have been implicated as immunosuppressive biomarkers that play a role in poor immunotherapeutic responses in CRC [46, 47]. Additionally, high co-expression of HAVCR2 and MSLN have been shown to be prevalent in several cancers and may also be vulnerable to the use of CAR-T therapies. Pre-clinical studies have shown the utility of CAR-T therapy in an MSLN high ovarian cancer model [48]. Taking this deeper into the HAVCR2-MSLN relationship, RNA-interference therapies have also been used in combination with CAR-T approaches to first knock down HAVCR2/TIM3 expression and enhance the cytotoxic functioning and proliferation capacity of therapeutic CAR-T cells in numerous in vitro cancer models [49]. TIM-3 has been shown to be a candidate for gene knockdown to enhance cell-based immunotherapy previously [50]. Taking these studies together, in combination with our conclusions of MSLN high implications on patient response to immunotherapies, genes, and immunotherapy, may yield a prospective combination therapy approach.

In MSLN high-expressing tumors that are MSS, we have reported significantly increased mutation rates in KRAS and FBXW7 as opposed to our findings in MSLN low-expressing patient cohorts, suggesting that MSLN status may also be relevant to the modulation of the TME and immune response in CRC. Taking this one step further, the top quartile of MSLN-expressing tumor cohorts exhibited significantly longer survival times, again supporting MSLN as a potential biomarker. We have also demonstrated (Figs. 6, 7) that a subset of patients, particularly the MSS subset, show an MSLN expression-dependent response to ICI therapies, highlighting the prospect of MSLN as a clinically relevant biomarker for therapeutic selection. However, it is recommended that additional pre-clinical testing using genetic strategies to modulate levels of MSLN be completed to determine if any causative relationship exists between MSLN expression and modulation of the TME towards immune suppression in MSS CRC. Improved survival in MSLN high CRCs post-ICI could be attributed to increased PD-L1 protein, CD274 (PD-L1), PDCD1 (PD-1), and CTLA4 gene expression, and enrichment of inflammatory response pathways in these tumors. However, the small size of the ICI therapy cohorts for survival analysis is one of the limitations of this study, which will require further validation in larger studies. In addition, data on tumor stage and grade were available in a limited number of patients included in our analysis and, therefore, we were also unable to determine the relationship of these prognostic variables with treatment outcomes. Several details not found in the insurance claims data, including prior treatments and clinical staging, do provide limitations to this study, and addressing these shortcomings would be pertinent to improving future works. We also acknowledge that there is a limit to the data availability within the Caris dataset, most notably the lack of MSLN IHC data within the larger cohort that further details expression information in noncancerous tissue as well as complete staging and diagnostic information. Future prospective analysis will be required to answer such important questions in addition to uncovering the similarities between CMS4 and MSLN high-expression tumor profiles. Further studies should be performed to fully uncover the implication of MSLN expression in CRC patients, specifically MSS patients, to improve the efficacy and availability of therapeutic options to a patient population that currently sees limited benefit beyond progression on the current standard of care treatment options. Large scale clinical trials should also be considered to fully capture patient data to provide a more complete MSLN profile and to fully validate the use of immunotherapies in MSLN high MSS CRC.