Genomic Characterization and Clinical Outcomes of Patients with Peritoneal Metastases from the AACR GENIE Biopharma Collaborative Colorectal Cancer Registry

Metastatic colorectal cancer (mCRC) is the third most frequent cause of cancer death in North America (1). Peritoneal metastases (PM) are a common site of metastatic spread, occurring in approximately 13% of patients with mCRC (2). However, the incidence of PM may be underestimated because of limitations of radiological imaging, as autopsy studies indicate that up to 80% of patients who die from mCRC have PM (3). Palliative systemic chemotherapy is widely used to improve survival (4). Patients with PM as the only site of metastasis may benefit from local therapies such as cytoreductive surgery and hyperthermic intraperitoneal chemotherapy (HIPEC; ref. 5). Nevertheless, the outcomes for patients with PM are variable. To refine treatment strategies, there is critical need for prognostic and predictive biomarkers that can help identify individuals who are most likely to derive benefits from aggressive local treatment interventions. Currently, the peritoneal cancer index, based on intraoperative assessment, is the only criterion suggested for the selection of patients with PM who may benefit from aggressive surgical treatment (6).

We hypothesized that patients with PM may have distinct characteristics that lead to different clinical outcomes compared with those without PM. Therefore, molecular characterization could offer insights into the clinical outcomes for patients with PM and allow for identification of unique biological drivers of peritoneal dissemination that can be targeted. Retrospective studies using single gene analysis or small targeted panel next-generation sequencing (NGS) have been performed in mCRC with PM (7–9). There is evidence that BRAF V600 mutations (10, 11) and microsatellite instability (MSI; ref. 12) are enriched in patients with PM. However, these studies often feature small sample sizes, lack a comparator group (non-PM mCRC) and include only limited genomic characterization. Project Genomics Evidence Neoplasia Information Exchange (GENIE) is an international genomics registry and data-sharing consortium launched in 2015 by the American Association of Cancer Research (AACR; ref. 13). Within GENIE, the BioPharma Collaborative (GENIE BPC) project augments the genomic data with comprehensive clinical data for cohorts of patients with selected cancers, including colorectal cancer. Our primary objective in this study was to characterize the genomic differences between patients with mCRC with and without PM. As secondary objectives, we examined the association between clinical characteristics and the presence of PM. In addition, we analysed the impact of PM and selected genomic alterations on overall survival (OS) from advanced disease and progression-free survival (PFS) from the initiation of first-line systemic therapy in the advanced setting.

Patients from the GENIE BPC colorectal cancer (CRC) v2.0-public data release were analyzed. These patients were selected for Project GENIE BPC based on the following eligibility criteria: underwent somatic genomic sequencing at one of three institutions (Dana-Farber Cancer Institute, Memorial Sloan Kettering Cancer Center, Vanderbilt-Ingram Cancer Center) between January 1, 2015 and April 30, 2018; aged 18 or older at the time of genomic sequencing; a minimum of 2 years of potential follow-up after sequencing; and one of the following OncoTree diagnoses: colorectal adenocarcinoma (COADREAD), colon adenocarcinoma (COAD), mucinous adenocarcinoma of the colon and rectum (MACR), signet ring cell adenocarcinoma of the colon and rectum (SRCCR), or rectal adenocarcinoma (READ). The PRISSMM (Pathology; Radiology; Imaging; Signs and Symptoms; tumor Markers; Medical oncology assessments) framework was used to curate structured clinical, demographic, treatment, pathology and imaging reports, and medical oncologist notes (14). The GENIE BPC project was approved by the research ethics board at each participating institution. Written informed consent was obtained for each patient to be included in this registry. Study was conducted in accordance to Declaration of Helsinki ethical guidelines.

Patients diagnosed with stage IV disease or stage I–III disease who later developed distant metastasis were evaluated. PM were defined as the presence of metastasis documented under any of the following ICD-O-3 (topography) codes on a cancer diagnosis form, radiology report, or pathology report at or after initial colorectal cancer diagnosis: C48.1 (specified parts of peritoneum), C48.2 (peritoneum not otherwise specified), C56.9 (ovary), C57.4 (uterine adnexa), and F20 (peritoneal fluid/ascites). Patients with PM at advanced diagnosis were further categorized into: (i) PM only, defined as those with PM as the only distant site disease present at the time of advanced diagnosis; and (ii) PM and other sites, defined as those with PM in addition to other locations of distant metastasis noted at the time of advanced disease diagnosis.

Each institution performed NGS analysis on archival formalin-fixed paraffin-embedded tissue from the primary tumor and/or metastases to detect single-nucleotide variants, small indels, copy-number alterations, and/or structural variants. Details regarding sequencing panels are provided in Supplementary Table S1 and in the GENIE data guide (www.aacr.org/wp-content/uploads/2022/02/GENIE_data_guide_11.0-public-1.pdf). In cases where a patient had multiple NGS reports available, only the first report was considered for analysis. A gene was considered altered in the presence of any mutation, fusion, amplification, or deletion. Genomic data were analyzed both with and without annotation for the OncoKB (precision Oncology Knowledge Base), such that the annotated genomic alterations were restricted to oncogenic and likely oncogenic variants.

Mismatch repair (MMR) status was determined on the basis of the pathology reports that indicated the expression of MLH1, MSH2, MSH6, or PMS2. Deficient MMR (dMMR) was defined as a loss of nuclear expression of at least one of these proteins. Patients were considered as dMMR if this status was ever documented on any available pathology report, while those with proficient MMR (pMMR) were identified if pMMR was recorded on any pathology report available. MMR non-concordance was assigned when both dMMR and pMMR were noted across pathology reports. For those tested for microsatellite instability (MSI) using PCR for microsatellites, the designation MSI-high (MSI-H) was assigned if MSI-H was recorded on any pathology report, and MSI-low/microsatellite stable (MSI-L/MSS) if either MSI-L or MSS was recorded on any pathology report. Patients were defined as MSI non-concordant if both MSI-H and MSI-L/MSS were recorded across pathology reports. Further details on the derivation of MMR and MSI statuses are shown in Supplementary Fig. S1.

The study cohort was summarized using descriptive statistics including frequencies, medians, and ranges. Associations between baseline clinical and pathologic characteristics with presence of PM were calculated using logistic regression. Multivariable analysis (MVA) for presence of PM included the variables that were significant in univariable analyses at a threshold of P < 0.05 and was adjusted for stage at diagnosis (IV vs. I–III), sex, primary tumor location, primary tumor histology, histologic grade of primary tumor, presence of liver metastasis at diagnosis of advanced disease (yes or no), and presence of lung metastasis at diagnosis of advanced disease (yes or no). Associations between genomic variables with PM were evaluated using the χ² test or Fisher exact test. To control the false discovery rate (FDR) among comparisons of genomic alterations, adjustments were made using the Benjamini–Hochberg method in comparisons between genomic alterations and PM. A q-value < 0.05 was considered statistically significant.

We evaluated the associations between clinical and genomic variables of interest and time-to-event endpoints. OS was defined from diagnosis of advanced disease (either at diagnosis for stage IV or from date of distant metastasis among stage I–III) until death or last follow-up. OS calculations were restricted to patients known to be alive at the time of NGS report.

PFS was defined from initiation of the first combination drug therapy regimen received following an advanced diagnosis. First-received combination therapies included in the PFS cohort consisted of those recommended by the National Comprehensive Cancer Network: 5-fluorouracil, oxaliplatin and leucovorin (FOLFOX), 5-fluorouracil, irinotecan and leucovorin (FOLFIRI), 5-fluorouracil, oxaliplatin, irinotecan and leucovorin (FOLFOXIRI), or capecitabine and oxaliplatin (XELOX) ± Bevacizumab/Cetuximab. The PRISSMM framework defines four real-world PFS endpoints: (i) PFS-I (time to disease progression according to an imaging report, or death); (ii) PFS-M (time to disease progression according to a medical oncologist assessment, or death); (iii) PFS-I-or-M (time to disease progression according to the earlier of an imaging report or a medical oncologist assessment, or death); (iv) PFS-I-and-M (time to disease progression according to the later documentation of disease worsening that was recorded in both an imaging report and medical oncologist assessment, or death). All PFS endpoints censor patients without progression at the initiation of a subsequent drug regimen, if applicable, or last follow-up. In alignment with a previous analysis demonstrating that PFS-I-and-M correlates most strongly with OS (14), analyses of PFS-I-and-M are presented as the primary results, while the remaining three PFS analyses are provided in Supplementary Materials. All PFS calculations were restricted to patients who were alive and progression-free at the time of NGS report.

OS and PFS were estimated using Kaplan–Meier methodology. HRs along with 95% confidence intervals (CI) were calculated using the Cox proportional hazards models. All survival analyses account for the left truncation bias inherent to the data by using risk set adjustment methods (i.e., entering patients into the risk set at the time of NGS report; ref. 15). Because of the delayed entry, risk tables below Kaplan–Meier curves may show the number of patients at risk increasing over time. All Cox models were adjusted for time from diagnosis of advanced disease to NGS report to account for possible dependent left truncation. All covariates included in Cox models were assessed for proportional hazards using tests of weighted residuals (16). MVA for both OS and PFS included the variables that were significant in univariable setting at a threshold of P ≤ 0.05. MVA for OS was adjusted for the presence of PM at time of advanced disease (yes or no), stage at diagnosis (IV vs. I–III), any RAS/BRAF mutation (yes or no), any APC mutation (yes or no), presence of liver metastasis at diagnosis of advanced disease (yes or no), presence of lung metastasis at diagnosis of advanced disease (yes or no), and months from diagnosis of advanced disease to NGS report. MVA for PFS-I-and-M was adjusted for any APC mutation, presence of liver metastasis at advanced disease, and months from initiation of first-line drug regimen to NGS report. All P values were from two-sided tests and results were deemed statistically significant at P < 0.05.

Analyses were performed using R version 4.2.1. All genomic data were extracted from cBioPortal for cancer genomics using the cbioportalR R package, while all clinical data were extracted from Synapse using the genieBPC R package. Note that both cbioportalR and genieBPC are available on CRAN (17). This study has been performed and reported according to STROBE and REMARK guidelines from the Equator network.

All of the clinical and genomic data are publicly available on https://www.synapse.org/#!Synapse:syn30991637. To access the publicly available GENIE BPC data:

• Register for a Synapse account (https://www.synapse.org/#)

• Review and accept the Synapse account terms of use

• Navigate to the CRC v2.0-public data release (https://www.synapse.org/#!Synapse:syn30991637)

• Click “Request Access,” review the terms of data use, and select “Accept”

• All the R source codes are publicly available in the following link: https://github.com/slb2240/patients_with_peritoneal_metastasis_from_colorectal_cancer

A total of 1,485 patients were included in the GENIE BPC colorectal cancer cohort. The study cohort was comprised of the 1,281 patients with mCRC, including 700 patients diagnosed with stage IV colorectal cancer and 581 patients diagnosed with stage I–III colorectal cancer who later developed distant metastasis. PM were present in 244 patients, representing 16% of the entire GENIE colorectal cancer cohort and 19% of the study cohort (Fig. 1). Among these 244 patients with PM, 135 (55%) had PM only.

Clinical and demographic characteristics of the study population stratified by the presence of PM at advanced diagnosis are summarized in Table 1 and according to stage at diagnosis in Supplementary Table S2. The median age at diagnosis in the study cohort was 54 years [interquartile range (IQR): 47–64] and 55% were male. 77% (N = 982) of the patients in the study cohort were non-Hispanic White, while 7% (N = 91) of patients were non-Hispanic Black and 5% (N = 71) were Asian, Asian American, and Pacific Islander. Most cases were adenocarcinoma (N = 996, 91%) and 58% of patients had liver metastases present at time of advanced diagnosis (N = 723).

Associations between clinical and pathologic characteristics with PM are shown in Table 2. In a multivariable model for presence of PM at diagnosis of advanced disease, PM were associated with female sex (OR: 1.67; 95% CI: 1.11–2.54; P = 0.014) and high histologic grade (OR: 1.72; 95% CI: 1.08–2.71; P = 0.002) after adjustment for other independently significant clinical variables. Tumor location (left colon vs. right colon vs. rectum) was associated with presence of PM (rectal vs. left colon OR: 0.51, 95% CI: 0.29–0.88; right colon vs. left colon OR: 1.41, 95% CI: 0.86–2.32; P < 0.001). Presence of liver metastasis and lung metastasis were less frequent in patients with PM compared with non-PM (liver OR: 0.04, 95% CI: 0.03–0.08, P < 0.001; lung OR: 0.33, 95% CI: 0.18–0.58, P < 0.001). Histology was associated with presence of PM (P = 0.001), with mucinous adenocarcinoma occurring more frequently than adenocarcinoma in patients with PM (OR: 3.55; 95% CI: 1.72–7.42).

A total of 1,345 NGS reports were available for the 1,281 patients included in the study cohort. 62% (N = 792) of patients’ NGS samples were taken from the primary tumor, with a median time interval from advanced diagnosis to NGS report of 7 months (IQR: 2–20). The site of NGS sample (primary tumor vs. metastasis) and type of NGS panel are summarized in Supplementary Table S3. Genes of particular clinical interest, NRAS and MED12, as well as frequencies of all genes altered in at least 7% of the samples and tested in at least 20% of the samples, are presented in Table 3 and are visualized by presence of PM in Supplementary Fig. S2 and S3. The five most frequently altered genes were APC (N = 151, 64% vs. N = 788, 79%), TP53 (N = 167, 68% vs. N = 779, 75%), KRAS (N = 122, 50% vs. N = 457, 44%), PIK3CA (N = 50, 20% vs. N = 458, 21%), and SMAD4 (N = 50, 21% vs. N = 176, 18%) in PM and non-PM patients, respectively. These frequencies were similar when restricting to only oncogenic and likely oncogenic variants according to OncoKB annotation.

Three genes with direct impact in upfront therapy selection for mCRC (KRAS, NRAS, and BRAF) were sequenced on all NGS panels. There were no differences in frequency of mutations in these three genes between PM and without PM (Table 3), nor was there any difference in the presence of BRAF V600E mutation between PM and non-PM (9.8% vs. 6.6%, P = 0.10). After controlling for the FDR, the only significant differences in genomic alterations between PM only versus PM and other sites versus no PM were alterations in APC and MED12. APC was less frequently altered in patients with PM (60%) compared with patients with PM and other sites (68%) and non-PM (79%; q < 0.01). MED12 was more frequently altered among patients with PM only (13%) compared with patients with PM and other sites of distant metastasis (7.4%) and patients without PM (4%; q = 0.02). These differences were also observed when comparing MED12 and APC alterations between patients with PM versus no PM (Table 3). However, when restricting the analysis to oncogenic and likely oncogenic variants based on OncoKB annotation, there were no significant differences observed in the presence of MED12 alterations when comparing patients with PM only versus PM and other sites versus no PM or between patients with PM versus without PM. NOTCH1 alterations were observed more frequently in patients with PM versus without PM (11% vs. 6.4%) but these differences were not statistically significant after adjusting by multiple comparisons (q-value 0.06).

At least one of MSI status or MMR status was known in 82% of the patients (N = 1,049). MSI results were available for 22% of the study cohort (N = 279). Among patients with MSI results available, 7.5% were MSI-H (N = 21) and 4.3% were MSI-L/MSS (N = 258). MMR results were available in 958 patients (75%), and 5% (N = 62) of these were dMMR. There were no differences in MMR status between patients with and without PM (dMMR 6% vs. 4%, P = 0.7) in the entire study cohort, although among patients diagnosed at stage IV disease, dMMR was more frequent in patients with PM compared with non-PM (9% vs. 2%, P < 0.01; Supplementary Table S2).

The cohort evaluable for OS included 1,204 patients after excluding 77 patients from the study cohort whose NGS report was returned after the last follow-up visit or death (Fig. 1). Baseline characteristics of this population are summarized in Supplementary Table S4. OS estimates according to PM and stage at diagnosis are summarized in Fig. 2. The median OS from the diagnosis of advanced disease was 31.4 months (95% CI: 28.9–34.2). OS was worse in patients with PM compared with those without PM [median OS: 26.8 months (95% CI: 22.2–31.3) vs. 32.9 months (95% CI: 30.4–37.7), P < 0.01] (Fig. 2A). After stratifying by stage at diagnosis, this difference was still observed among patients diagnosed with stage IV disease (P = 0.02), but not among patients diagnosed with stage I–III disease who later developed metastasis (P = 0.42; Fig. 2C and D).

OS from advanced disease was significantly shorter for patients with PM and other sites of metastases (median OS: 19.6 months, 95% CI: 10.6–24.8) compared with those with PM only (median OS: 34.3 months, 95% CI: 29.9–44.0) and those without PM but with other sites of metastases (median OS: 32.9 months, 95% CI: 30.4–37.7; P < 0.01; Fig. 2B). Stratification by stage at diagnosis demonstrated that this difference was maintained for patients diagnosed with stage IV disease (P < 0.01), but not for patients diagnosed with stage I–III colorectal cancer who later developed metastasis (P = 0.17).

We sought to explore whether there was an impact on OS based on the presence of PM and the status of KRAS, BRAF, NRAS, any of RAS/BRAF, or three known driver mutations in colorectal cancer: TP53, APC, and PIK3CA (Supplementary Table S5). OS was significantly different for all four analyses of presence of PM by KRAS (P < 0.01), BRAF (P = 0.04), NRAS (P = 0.04), and RAS/BRAF (P < 0.01) mutant (mt) versus wild-type (wt) patients, with the longest median OS in the non-PM wt subgroups compared with the non-PM mt, PM wt, and PM mt subgroups. In contrast, in the analyses of presence of PM by TP53, APC, and PIK3CA alteration statuses, OS differed significantly among the subgroups with non-PM mt patients having the longest estimated median OS (TP53: P = 0.03; APC: P < 0.01; PIK3CA: P = 0.01) compared with the non-PM wt, PM mt, and PM wt subgroups. These associations remained statistically significant when limited to oncogenic and likely oncogenic variants according to OncoKB.

Univariable analyses were conducted to examine associations between clinical and genomic factors of interest with OS (Table 4). Among the clinical factors, univariable analyses showed that stage at diagnosis (stage IV vs. stage I–III) and the presence of PM were associated with worse OS [stage HR: 1.80 (95% CI: 1.54–2.12), P < 0.001; PM HR: 1.30 (95% CI: 1.07–1.57), P < 0.01]. In addition, the presence of liver metastasis and lung metastasis at advanced diagnosis were associated with worse OS [liver metastasis HR: 1.43 (95% CI: 1.21–1.68), P < 0.01; lung metastasis HR: 1.24 (95% CI: 1.03–1.49), P = 0.02]. No significant OS differences were observed between location of primary tumor (right colon vs. left colon vs. rectum; P = 0.3), MSI status (MSI-H vs. MSI-L/MSS; P = 0.4), or MMR status (pMMR vs. dMMR vs. MMR non-concordant; P = 0.065). Beyond the clinical variables, we explored associations between OS and RAF/BRAF, TP53, APC, and PIK3CA alteration status. In the univariable setting, the presence of any RAS/BRAF alteration was associated with worse OS [HR: 1.53 (95% CI: 1.31–1.80), P < 0.01]. There were no differences in OS by the presence of TP53 or PIK3CA alterations, but the presence of APC alterations was associated with longer OS under univariable analysis [HR: 0.75 (95% CI: 0.63–0.90), P < 0.01]. Similar associations with OS were observed when restricting to oncogenic and likely oncogenic variants.

MVA demonstrated that the presence of PM at advanced diagnosis was an independent prognostic factor for worse OS [HR: 1.45 (95% CI: 1.16–1.81), P < 0.01] after adjustment for stage at diagnosis, RAS/BRAF and APC alteration status, presence of liver metastasis and presence of lung metastasis at diagnosis of advanced disease, and months from diagnosis of advanced disease to NGS report (Table 4).

There were 473 patients who received first-line approved combination drug therapies and were included in the PFS-I-and-M cohort (101 patients with PM, 21.4%; Fig. 1). Most patients received a combination of fluoropyrimidines and oxaliplatin as their first-line systemic therapy (69%, Supplementary Table S6). PFS-I-and-M according to PM and stage are summarized in Fig. 3. There was no significant difference in PFS-I-and-M from initiation of first-line therapy when comparing patients with PM versus without PM [median PFS: 13.8 months (95% CI: 10.6–26.5) vs. 14.9 months (95% CI: 13.3–17.7), respectively; P value = 0.46] (Fig. 3A). Stratification by stage at diagnosis revealed no significant differences in comparisons of PM versus non-PM and PFS-I-and-M (Fig. 3C and D). Similarly, there was no difference in PFS-I-and-M when comparing among patients with PM only versus PM and other sites versus non-PM [median PFS-I-and-M: 13.9 months (95% CI: 11.7–not reached) vs. 9.5 months (95% CI: 4.4–not reached) vs. 14.9 months (95% CI: 13.3–17.7); P = 0.40] (Fig. 3B). Secondary analyses estimating the other PRISSMM-defined PFS endpoints (PFS-I, PFS-M, and PFS-I-or-M) by presence of PM are presented in Supplementary Table S7.

We examined whether there was an impact on PFS-I-and-M based in the presence of PM and KRAS, BRAF, RAS/BRAF, PIK3CA, APC, and TP53 mutation status (Supplementary Table S8). None of the comparisons revealed any significant differences in PFS-I-and-M based in the presence of PM and alterations. Note that the impact of NRAS by the presence of PM on PFS-I-and-M could not be explored because of limited subgroup sample sizes.

Univariable analyses of the associations between clinical and genomic variables and PFS-I-and-M were examined (Supplementary Table S9). Presence of liver metastasis at advanced diagnosis was the only clinical variable marginally associated with worse PFS-I-and-M [HR: 1.37, (95% CI: 1.00–1.88), P = 0.05]. Notably, neither the presence of PM (PM vs. non-PM) nor PM only versus PM and other sites of distant metastasis versus non-PM were significantly associated with PFS-I-and-M (P = 0.5 and P = 0.4, respectively). With respect to the genomic variables, the presence of APC alteration was significantly associated with better PFS-I-and-M [HR: 0.69 (95% CI: 0.48–0.98), P = 0.04]. However, this association dissipated after limiting to only oncogenic and likely oncogenic variants as per OncoKB annotation (P = 0.07).

MVA for PFS-I-and-M demonstrated that APC was significantly associated with better PFS-I-and-M after adjusting for the presence of liver metastasis at advanced diagnosis and time from advanced diagnosis to NGS report [HR: 0.61 (95% CI: 0.42–0.89), P = 0.01]. In addition, liver metastasis at diagnosis of advanced disease was significantly associated with worse PFS-I-and-M after adjusting for APC mutation status and time to NGS report.

The aim of this study was to characterize and compare the clinical and genomic characteristics of patients with mCRC with versus without PM. Consistent with prior large studies, PM were present in 19% of patients at diagnosis of advanced disease (18). The presence of PM was associated with several clinical factors that have been identified in prior series, including female sex (2), high histologic grade (19, 20), and mucinous or signet cell histology (21, 22). Most tumors (95%) were sequenced with large (326–468 genes) targeted tumor sequencing panels.

We identified two important genomic findings in patients with mCRC with PM. First, the presence of MED12 alterations was associated with PM. Previously described as a tumor suppressor in colorectal cancer, MED12 encodes a component of the mediator transcription regulation complex that is necessary for the initiation of the transcription (23, 24). When MED12 is mutated or its expression is lost, it may induce an epithelial/mesenchymal-like phenotype and activation of the TGF receptor pathway that confer drug resistance in colorectal cancer models (25). This provides a therapeutic rationale to test TGFβ pathway targeted drugs in patients with MED12 alterations or loss of expression. However, this association was not observed when variants were annotated by OncoKB, as most MED12 alterations were non-oncogenic or of unknown significance because variants in this gene have not been well biologically characterized. In contrast, APC alterations were less frequent in patients with PM in the study cohort, even after restricting for oncogenic and likely oncogenic variants, supporting a key finding from a prior study that demonstrated that APC mutations were more frequent in primary colorectal cancer tumors compared with unmatched PM samples (26). APC is a key negative regulator of the Wnt pathway, which controls the cell proliferation and dedifferentiation of the gastrointestinal tract (27). In addition, APC alterations can contribute to the loss of cell adhesion and errors in cell cycle control or in DNA repair (28). Gene expression analysis has also shown that PM are associated with activation of the WNT/β-catenin pathway that can be activated because of other genomic alterations beyond APC (29). Therefore, WNT/β-catenin pathway may be more frequently activated in mCRC with PM. Moreover, certain MED12 variants induce WNT/β-catenin pathway activation in myometrial cells (30, 31). This provides a rationale for targeting the WNT/β-catenin pathway in patients with PM.

In comparison with a recent systematic review, we observed a higher prevalence of mutations in TP53 (68% vs. 54%) and APC (63% vs. 44%) and a similar prevalence of KRAS alterations (50% vs. 44%) in patients with mCRC with PM (8). In contrast to prior reports, BRAF was not associated with PM (10, 11, 32). Notably, this study cohort captures BRAF mutations beyond the canonical V600 hotspot analyzed in previous studies. When considering BRAF V600E canonical mutations only, such alterations did not show an increased frequency in patients with PM compared with no PM (9.8% vs. 6.6%).

Patients with PM had shorter OS from advanced disease compared with patients without PM as described previously (2). Notably, patients with PM and other sites of distant metastasis present at the time of advanced diagnosis had worse OS than patients with PM only and patients without PM. After adjustment for stage at diagnosis, RAS/BRAF and APC alteration statuses, presence of liver and lung metastases, and time from diagnosis of advanced disease to NGS report, patients with any PM had significantly shorter OS than those without PM, indicating that PM are an independent negative prognostic factor for patients with mCRC. KRAS and BRAF mutations are associated with shorter survival in mCRC irrespective of the presence of PM (33, 34). We observed in our cohort that patients with PM and KRAS mutation showed the worst prognosis. There were no differences in PFS-I-and-M from initiation of combination first-line cancer-directed drug regimens between patients with versus without PM or among patients with PM only versus PM and other sites of metastasis versus no PM. Sensitivity analyses of PFS-I, PFS-M, and PFS-I-or-M demonstrated similar patterns to the PFS-I-and-M results, with patients with PM generally having shorter median PFS estimates than those without PM, and patients with PM and other sites of metastasis with shorter median PFS compared with those with PM only and non-PM, though the differences were not statistically significant.

There are several limitations of this retrospective study. Comprised of data from three large academic medical institutions in the United States, this study may not accurately reflect the outcomes of patients with PM treated outside such centers specializing in management of mCRC. MSI status was largely missing within the study cohort but many other samples were tested instead for MMR protein status, that is usually performed as an alternative to MSI analysis. In addition, PM were identified using different ICD-O-3 codes that were abstracted by the curators in the GENIE BPC project. Therefore, some patients with PM at diagnosis may not have been identified if the sites of metastatic disease were not fully captured in the clinical notes, imaging or pathology reports during the initial diagnosis. Furthermore, clinical data on peritoneal surgery were not available, which could potentially influence the survival results. Although all patients had NGS testing, there were differences in the genes included in these assays. Our analysis of genomic alterations associated with PM accounts for these differences in gene coverage by restricting analyses of genomic alterations to tumors with the gene(s) of interest included on the sequencing panel. Finally, analysis was performed with the first available NGS testing when more than one sample was available, and it is possible that genomic alterations were acquired over the course of treatment.

In conclusion, this study demonstrates that patients with mCRC with PM have distinct clinical and molecular characteristics compared with those without PM, including differences in histologic grade, tumor location, and presence of MED12 and APC alterations. To our knowledge, this is the largest multicenter, clinico-genomic study to evaluate the impact of PM on patients with colorectal cancer. Further research is needed to understand these biological differences and develop therapeutic strategies to prevent and treat mCRC with PM.

E. Sanz Garcia reports grants from GSK and other from Novartis outside the submitted work. S. Brown reports grants from AACR Project GENIE Biopharma Collaborative during the conduct of the study. J.A. Lavery reports grants from NCI and other from American Association for Cancer Research Project Genomics Evidence Neoplasia Information Exchange Biopharma Collaborative during the conduct of the study. H.E. Fuchs reports grants from AACR during the conduct of the study. J.L. Warner reports grants from AACR during the conduct of the study; grants from NIH, Brown Physicians Inc; personal fees from Westat, Flatiron Health, Melax Tech; and other from HemOnc.org LLC outside the submitted work. M.L. LeNoue-Newton reports other from American Association for Cancer Research during the conduct of the study; other from GE Healthcare outside the submitted work; in addition, M.L. LeNoue-Newton has a patent to 63/335,215 pending. S.M. Sweeney reports grants from Amgen, AstraZeneca, Bristol Meyers Squibb, Boehringer Ingelheim, Bayer Pharmaceuticals, Janssen, Genentech, Novartis, Pfizer, and Merck during the conduct of the study. C. Nichols reports grants from American Association for Cancer Research during the conduct of the study. R. Kundra reports AACR GENIE funds cBioPortal. G.J. Riely reports grants from AACR during the conduct of the study; grants from Pfizer, Novartis, Takeda, Roche, Merck, and Lilly outside the submitted work. D. Schrag reports grants from AACR during the conduct of the study; personal fees from JAMA and grants from Grail outside the submitted work. K.S. Panageas reports other from AACR Project GENIE Biopharmaceutical Consortium and grants from NIH/NCI Cancer Center grant during the conduct of the study. P.L. Bedard reports grants from BMS, Pfizer, Seagen, Lilly, Amgen, Merck, Gilead, Zymeworks, Bicara Therapeutics, LegoChem, GSK, Roche/Genentech, Medicenna, AstraZeneca, Bayer, and Takeda outside the submitted work; and Uncompensated advisory for Janssen, Repare, Seagen, Zymeworks, Gilead, Merck, Lilly, Pfizer, BMS. No disclosures were reported by the other authors.

E. Sanz-Garcia: Conceptualization, data curation, formal analysis, investigation, visualization, writing-original draft, writing-review and editing. S. Brown: Formal analysis, investigation, visualization, methodology, writing-original draft, writing-review and editing. J.A. Lavery: Formal analysis, visualization, writing-review and editing. J. Weiss: Formal analysis, writing-review and editing. H.E. Fuchs: Data curation, formal analysis, writing-review and editing. A. Newcomb: Data curation, formal analysis, writing-review and editing. A. Postle: Data curation, formal analysis. J.L. Warner: Resources, supervision, methodology, writing-review and editing. M.L. LeNoue-Newton: Resources, supervision, methodology, writing-review and editing. S.M. Sweeney: Resources, supervision, funding acquisition, project administration, writing-review and editing. S. Pillai: Formal analysis, visualization, writing-review and editing. C. Yu: Data curation, project administration, writing-review and editing. C. Nichols: Resources, formal analysis, writing-review and editing. B. Mastrogiacomo: Formal analysis, visualization, methodology, writing-review and editing. R. Kundra: Formal analysis, visualization, writing-review and editing. N. Schultz: Data curation, formal analysis, writing-review and editing. K.L. Kehl: Resources, supervision, writing-review and editing. G.J. Riely: Resources, supervision, writing-review and editing. D. Schrag: Resources, supervision, writing-review and editing. A. Govindarajan: Supervision, writing-review and editing. K.S. Panageas: Resources, formal analysis, supervision, investigation, writing-original draft, writing-review and editing. P.L. Bedard: Conceptualization, resources, visualization, methodology, writing-original draft, writing-review and editing.

The authors would like to acknowledge the AACR and its financial and material support in the development of the AACR Project GENIE registry as well as members of the AACR Project GENIE consortium for their commitment to data sharing; GENIE Coordinating Center, Sage Bionetworks, and/or cBioPortal staff who contributed substantively to the development and analysis of data, as well as writing the article. This study was supported in part by the NIH and NCI Cancer Center Support Grant P30 CA008748.