. Author manuscript; available in PMC: 2026 Sep 22.

Published in final edited form as: Cancer Cell. 2026 Sep 1;44(9):1746–1760.e4. doi: 10.1016/j.ccell.2026.08.004

Integrated clinicogenomic analysis reveals the evolution and metastatic tropisms of advanced colorectal cancer

Paolo Manca ^1,^2,^*, Henry Walch ^3,^4,^5,^*, Jierui Xu ^3,^4,⁵, Natalie Small ⁴, Ayush Kris ⁴, Evan Seffar ^3,^4,⁵, Allison Richards ³, Tiffany Xue ⁴, Adam Price ³, Mark T A Donoghue ³, Christopher Fong ^3,^4,⁵, Michele Waters ^3,^4,⁵, Justin Jee ^1,^4,⁵, Sachin Kumar Deshmukh ⁶, George W Sledge Jr ⁶, Midhun Malla ⁷, Andrea Cercek ¹, Leonard Saltz ¹, Neil H Segal ¹, Zsofia K Stadler ¹, Maliha Nusrat ¹, Michael B Foote ¹, Avni Desai ¹, Sree Chalasani ¹, Devika Rao ¹, Louise Connell ¹, Richard K G Do ⁸, Karuna Ganesh ¹, Martin R Weiser ⁴, Michael F Berger ^3,⁹, David B Solit ^1,³, Jinru Shia ⁹, Efsevia Vakiani ⁹, Julio Garcia-Aguilar ¹⁰, Nikolaus Schultz ^3,^4,⁵, Francisco Sanchez-Vega ^3,^4,^5,¹⁰, Walid K Chatila ^3,^4,^5,⁺, Rona Yaeger ^1,^+,⁺⁺

PMCID: PMC13593342 NIHMSID: NIHMS2208821 PMID: 42679811

SUMMARY

We performed an integrated clinical and genomic analysis of over 7,000 consecutively sequenced colorectal cancer (CRC) samples to comprehensively characterize genetic drivers and metastatic tropisms of CRC. We find that genomic evolutionary changes, such as clonal mutations and oncogenic mutant allele imbalance, selectively enhance the impact of recurrent oncogenic alterations. We identify the relative timing of organ-specific metastasis, showing sequential metastatic progression in microsatellite stable CRC with brain and adrenal metastases as late events; metastatic sites that cluster together, such as lung, bone, and brain metastases; and genomic events that enhance or decrease risk for each metastatic site with WNT pathway activation as overall protective while RAS pathway activation increased risk for spread to all metastatic sites. Our data suggest that despite the heterogeneity in CRC, genomic evolution increases the impact of recurrent alterations, and integrating information about tumor primary location and genomics can be used to predict organ-specific metastasis risk.

eTOC Blurb:

Manca et al. analyze > 7000 colorectal cancers, integrating next-generation sequencing and clinical data. They define tumor genomics, clonality and dosage of mutations, and metastatic tropisms and their effects on patient outcomes. Incorporating metastatic sites, primary tumor location, and genomics, they develop an organ-specific risk model for metastasis.

Graphical Abstract

graphic file with name nihms-2208821-f0007.webp

INTRODUCTION

Colorectal cancer (CRC) is one of the deadliest cancers: fewer than a fifth of patients with metastatic CRC are alive 5 years after diagnosis and rising incidence has made it the leading cause of cancer death in people under 50 years old¹. However, the timing and extent of metastatic involvement varies greatly across patients. An understanding of metastatic tropisms, the timing and prognosis of each metastatic site involved, and the contribution of CRC molecular subtypes, primary tumor location, and tumor genomics would allow more tailored surveillance strategies and identify potential targets to try to intercept metastatic extension.

CRC is also a model for the multi-step acquisition of genomic alterations in the progression of cancer^2,3. Three to six driver gene alterations have been hypothesized as needed for the development of invasive CRC⁴. These recurrent, initiating events are well defined^5,6, and selective inhibitors targeting key driver alterations, including BRAF V600E, KRAS G12C, and TP53 Y220C, have demonstrated clinical utility in patients^7–10. Nevertheless, chemotherapy remains the cornerstone of metastatic CRC treatment even in the presence of genetic drivers^11,12, and early experience with all of these selective inhibitors suggests lower activity in CRC compared to other cancer types. A more complete genomic understanding of these targets, including whether these alterations are clonal, present in all metastatic sites, and enhanced during tumor evolution through gain of the mutant or loss of the wild-type allele (allelic imbalance) and the role, frequency, and clonality of secondary mutations, is needed to refine therapeutic approaches for CRC.

To address these knowledge gaps, we queried a dataset of over 7,000 CRCs prospectively sequenced at high depth in the clinical setting with full clinical data and molecular profiles. With this large series, we set out to comprehensively characterize genomic alterations in CRC. We leveraged analysis of clonality and allele imbalance to define drivers of CRC oncogenesis and progression and their association with global genomic features, including aneuploidy and whole genome doubling (WGD). We also identified patterns of metastatic progression, associations between genomic alterations and metastatic sites, and their impact on patient survival. In light of the rise in early-onset CRC¹³, we characterized genomic patterns and metastatic tropisms by patient age group to identify distinct features of early-onset CRC.

RESULTS

Cohort definition

The full series of CRC cases sequenced with MSK-IMPACT (Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets) in the clinical setting consisted of 7,238 colorectal adenocarcinoma samples from 6,789 patients, including 5,174 primary and 2,063 metastatic samples (Table 1). Compared to the 2018 TCGA (The Cancer Genome Atlas) CRC¹⁴, the current series included a higher proportion of metastatic samples, rectal primary tumors, patients diagnosed under the age of 50 years, and patients with stage IV disease at diagnosis and a lower proportion of MSI-H (microsatellite instability-high) tumors (Figure 1A). Expanding on the 2018 MSK CRC cohort⁶, overall, the current cohort had more patients with earlier stage disease at diagnosis (reflecting a shift to earlier sequencing in patients’ course) and a higher proportion of MSI-H tumors and rectal primary tumors (reflecting areas of research at MSK) (Figure 1A). Consistent with prior studies, early-onset CRC cases had predominantly left-sided or rectal primary location (Figure S1A). We defined two subgroups for analysis: the FACETS Cohort (n=5,520) with full allelic copy number information from FACETS¹⁵ (Fraction and Allele-Specific Copy Number Estimates from Tumor Sequencing) and the Clinical Cohort (n=4,517) with full clinical data (Figures 1B and S1B). Breakdown by molecular subtypes was 5,946 microsatellite stable (MSS), 775 MSI-H, and 68 ultra-mutated (pathogenic POLE/POLD1, pPOLE/POLD1) (Figure 1C).

Table 1.

Patient Characteristics

Characteristic	Genomics Cohort (n=6,789) N (%)

Sex
Male	3,689 (54.3%)
Female	3,100 (45.7%)

Age at diagnosis
<35 years	391 (5.8%)
35-50 years	1871 (27.5%)
≥50 years	4404 (64.9%)
N/A	123 (1.8%)

Primary tumor location
Right colon	2059 (30.3%)
Left colon	2577 (38.0%)
Rectum	2153 (31.7%)

Stage at diagnosis
Stages I-III	3578 (52.7%)
Stage IV	2607 (38.4%)
N/A	604 (8.9%)

Tumor molecular subtype
Microsatellite stable	5946 (87.6%)
Microsatellite instability-high	775 (11.4%)
Ultramutated (Pathogenic POLE/POLD1)	68 (1.0%)

Open in a new tab

Abbreviations: N/A: not available

Figure 1. — A) Comparison of TCGA PanCancer Atlas colorectal cancer (CRC) cohort (TCGA CRC), the 2018 CRC MSK-IMPACT series (MSK 2018), and the current MSK CRC cohort (MSK 2025).

B) Breakdown of MSK CRC 2025 cohort into subgroups for analyses. C) Comparison of microsatellite stable (MSS) (n=5,946), microsatellite instability-high (MSI-H) (n=775), and ultra-mutated (pathogenic POLE/POLD1, pPOLE/POLD1) (n=68) cases. D) Treatment history by molecular subtype and, for MSS cases, by primary tumor location.

See also Figure S1.

Treatment patterns were evaluated in patients treated at MSK who had data available for at least 5 years or to date of death (Figures 1D and S1C). Chemotherapy was the dominant treatment for metastatic CRC; over 90% of patients with metastatic MSS tumors received chemotherapy, and, as expected, very few of these patients received immune checkpoint blockade (ICB). For MSI-H metastatic CRC, about half of patients received chemotherapy during their course and about half received ICB. For patients with ultra-mutated metastatic tumors, nearly all patients still received chemotherapy, and a large proportion also received ICB. Only about 3% of all patients with metastatic CRC received BRAF targeted therapy, with this frequency increasing to 4% in patients with right-sided primary tumors. The use of EGFR (epidermal growth factor receptor) inhibitors for MSI-H metastatic CRC was also lower than expected (5%). For non-metastatic CRC, most patients with MSS and ultramutated tumors received chemotherapy, with much less common chemotherapy use for MSI-H tumors. Among MSS tumors, the proportion getting chemotherapy was similar for right- and left-sided colon primaries and higher for rectal primaries. Treatment patterns did not substantially vary by patient age with potentially higher adjuvant chemotherapy administration in younger patients (Figure S1C). For cases with pre-treatment and post-treatment samples (n=131), no clear differences were seen in comparison of tumor genomics, fraction genome altered (FGA), mutation spectrum, and tumor mutation burden (TMB) (Figure S1D).

Leveraging the power of this large series, we assessed recurrent genomic alterations in CRC (Figures 2A and S2A). The most frequently altered genes in MSS CRC were TP53, APC, and KRAS, followed by SMAD4, PIK3CA, FBXW7, SOX9, TCF7L2, and BRAF. Recurrent amplifications involved the q arm of chromosome 20 (BCL2L1, DNMT3B, SRC), MYC, ERBB2, EGFR, and ERBB3 (Figure 2A). For pPOLE/D1 cases, alterations in genes within the phosphatidylinositol 3-kinases (PI3K) pathway (PIK3CA, PIK3R1, PTEN, FBXW7) were particularly common. Recurrent genomic alterations were similar across age groups, once accounting for differences in primary tumor location, with a significant but modest increase in TP53 and decrease in APC mutations in early-onset CRC (Figure 2B).

Figure 2. — A) Frequently altered genes. Bars are colored for alteration type. B) Recurrent alterations in MSS CRC, subdivided by patient age group and primary tumor location. C) Relationship between mutation prevalence (x-axis) and relative fraction of clonal mutations (y-axis). Each dot represents a gene. Oncogenic/likely oncogenic mutations per oncoKB are in purple, others are in grey. Dashed lines indicate Spearman correlation fits, and the corresponding Spearman correlation coefficients (ρ) and p-values are shown.

See also Figures S2 and S3.

The three molecular subtypes had distinct genomic features. pPOLE/D1 cases had the highest TMB (median: 158.0 mut/Mb); MSI-H cases had the highest proportion of insertion/deletion (InDel) mutations (median: 34.3%); and MSS cases had the highest FGA (58.0%) (Figure S2B). In the MSS cohort, oncogenic mutations were more often clonal compared to non-oncogenic mutations (Figure S2C). Allelic imbalance, where the proportion of mutant and wild-type alleles differ from 50%, was most frequently observed in the MSS tumors and with gain-of-function (GOF) and loss-of-function (LOF) mutations (35.8% and 58.1%, respectively) (Figure S2D). Frequencies of allelic imbalance in the most recurrently mutated genes were overall similar across patient age groups (Figure S2E).

Clonality and mutation dosage in MSS CRC

Recent genomic studies indicate clonality and allelic imbalance may affect cancer outcomes¹⁶. In MSS CRC, oncogenic mutations in the most recurrently mutated genes were usually clonal and were significantly more likely to be clonal compared to oncogenic mutations in genes less frequently altered (ρ=0.23, p=0.0037) (Figures 2C, S3A, and S3B). For mutations of unknown significance, the proportion of clonal mutations was not correlated with gene alteration frequency (Figure 2C) and was similar to that for oncogenic mutations in rarely altered genes (Figure S3B).

Clonal mutations in TP53, KRAS, PIK3CA, SOX9, and BRAF, but not APC, were associated with positive selection of the mutant allele (Figure 3A). We hypothesized that the presence of co-existing APC mutations in the same sample may have influenced the clonality and pressure for allelic imbalance. Consistent with this, the percentage of subclonal mutations was higher in samples with ≥2 APC mutations (21.4% versus 12.1%; p<0.001) and mutation dosage was strikingly higher in tumors with only one APC mutation (79.3% vs 16.7%; p<0.001) (Figure S3C). Thus, samples with only one APC oncogenic mutation maintained the same relationship between clonality and dosage as for other genomic drivers of CRC (Figure 3B). Single APC mutation with allelic imbalance and multiple mutations in APC without underlying copy number changes thus represent two mechanisms that enhance the effect of these alterations.

We investigated whether mutation clonality and allelic imbalance varied for alterations that were private versus shared in 139 paired primary-metastasis samples. Oncogenic mutations were highly concordant between matched primary-metastasis samples, with 475 out of 542 (87.6%) primary tumor oncogenic mutations shared with the respective metastatic sample (Figure S3D). Among primary tumor oncogenic mutations, shared mutations more often displayed positive allelic selection (57.1% vs 39.0%, p=0.012) and were more often clonal (86.1% vs 51.7%, p<0.001) compared to private primary mutations (Figure S3E). These data indicate that the alterations selected in the primary tumor as clonal and enhanced by allelic imbalance were more likely to be present in metastases.

Lastly, we assessed how chromosomal instability and WGD influence mutation dosage (Figures 3C, S3F–S3J). Overall, we found a higher fraction of mutations with positive mutant allele selection in WGD+ due to a higher frequency of wild-type allele loss and not increased copies of the mutant allele. However, for samples with clonal KRAS mutation, gain of the mutant KRAS allele was relatively more prevalent than loss of the wild-type allele (Figure S3K).

To understand the clinical implications of mutation dosage, we evaluated the effect of mutation dosage in each of the genes most recurrently mutated in CRC on patient survival (Figures 3D, S3L, and S3M). As expected, the presence of a mutation in KRAS or BRAF was associated with worse survival. However, enhanced mutant dosage of either gene did not affect survival, except in metastatic left-sided CRC, where mutant KRAS gain was associated with shorter survival (HR=1.58, 95%CI:1.17-2.12, p=0.003) (Figure 3E). APC mutation was associated with improved survival, without enhancement of the effect by mutation dosage. The only other gene associated with survival was SMAD4, an established poor prognostic marker¹⁷, where interestingly the presence of a heterozygous mutation without allelic imbalance led to worse outcomes and higher mutant dosage seemed to provide some protection with survival similar to wild-type SMAD4 CRC.

Impact of TP53 status on copy number changes

To understand how genomic alterations affect tumor development, we looked more closely at the relationship of p53 pathway alterations (TP53, ATM, or CHEK2 mutation, MDM2 or MDM4 amplification) with gene and chromosome-level copy number alterations. There was a progressive increase in FGA from MSI-H to MSS WGD− to MSS WGD+ cases (Figure S4A). In parallel, there was an increasing selection for p53 pathway alterations across these groups, as seen in the absolute p53 pathway alterations frequency (41.8%, 76.2%, and 89.1%), relative fraction of subclonal mutations in this pathway (21.6%, 9.3%, and 2.7%), and relative fraction of mutations in this pathway with positive selection for the mutant allele (6.3%, 81.6%, and 93.9%) (Figures S4B and S4C), and this translated to higher FGA (Figure S4D). These data suggest that p53 pathway alterations may underlie differences in FGA between MSI-H and MSS cases and within MSS cases.

We next investigated the effect of p53 pathway status on chromosome-level copy numbers in the MSS/WGD-tumors. While chromosome losses and LOH (loss of heterozygosity) were enriched in p53 pathway altered samples, gains were evenly distributed between p53 mutant and wild-type cases (Figures 4A, 4B, and S4E); intriguingly, we observed an increased frequency of gains in chromosome 12 in the p53 pathway wild-type samples: 12p (OR=4.54, q<0.001) and 12q (OR=7.35, q<0.001), with these arms carrying the KRAS and MDM2 genes, respectively.

Figure 4. — A) FACETS-based total copy number (TCN) of chromosome arms (rows) depending on samples’ (columns) p53 pathway clonality status (clonal: at least one clonal mutation; subclonal: only subclonal mutations). B) Differential enrichment of chromosome arm CNAs, including gains (TCN>2), losses (TCN<2), and losses of heterozygosity (LOH, lower copy number or LCN=0), according to p53 pathway clonal alteration status in MSS WGD- samples. The x-axis shows the odds ratio (OR) for CNA enrichment; the y-axis shows unadjusted p-values (−log10). Odds ratios and p-values were calculated using Fisher’s exact test, and q-values were obtained by FDR correction. Each chromosome arm is represented three times: losses, gains, LOHs. Colored dots represent significant CNA-chromosome arm pairs (gain: red; loss: blue; LOH: green; q<0.10), while grey dots represent q-value≥0.10. CNA-chromosome arm pairs with |log₂(OR)|>2 are labeled. The dashed line indicates OR=1. C) Baseline and progression clinical samples for sotorasib/panitumumab treatment showing acquired *KRAS* amplification and *MDM2* gain. D) Proportion of samples with chromosome 12p TCN>2 by *KRAS* mutation and p53 pathway clonality status. E) *KRAS* and *MDM2* mRNA expression in TCGA PanCan cohort. F) *MDM2* mRNA expression by *TP53* mutation status. G) Fraction of genome altered (FGA) by TCN of *MDM2* and p53 pathway mutation status. H) *MDM2* mRNA expression by *TP53/KRAS* genotype in TCGA and Caris Life Sciences cohorts. F, G, H) Horizontal lines represent medians, boxes represent the interquartile range (25th–75th percentiles), whiskers extend to the most extreme observations within 1.5 × the interquartile range, and points represent outliers. Abbreviations: Wt: wild-type, Mut: Mutant.

See also Figure S4.

These data raise the possibility that chromosome 12 gain may be due to KRAS mutant allele selection and that MDM2 may be secondarily gained after this event, potentially affecting selective need for p53 pathway alteration. Pre-treatment and progression samples from a patient with KRAS G12C CRC treated with the KRAS G12C inhibitor sotorasib plus EGFR inhibitor panitumumab illustrate a gain in MDM2 co-occurring with the acquisition of increased copies of KRAS (Figure 4C). In fact, the highest rate of chromosome 12p gain was seen in the KRAS mutant, p53 unaltered cases (Figure 4D), and gains in the 12p and 12q arms were highly concordant. In a separate cohort of 386 MSS CRC samples with mRNA expression available¹⁴, we validated the significant co-expression of MDM2 and KRAS (Spearman ρ=0.567, q-value=5×10⁻³¹) (Figure 4E) and that MDM2 expression was particularly high in the TP53 wild-type samples (FC=1.75, q-value=1.7×10⁻¹⁷) (Figure 4F). Moreover, in the p53 pathway unaltered cases, MDM2 gain was associated with a higher FGA, suggesting that this gain has a functional effect (Figure 4G). In TCGA and the independent Caris Life Sciences MSS CRC cohort (n=24,225), MDM2 expression was highest in CRC with wild-type TP53 and mutant KRAS compared to the other TP53/KRAS genotypes (Figure 4H). Aligned with this, drug response data from Genomics of Drug Sensitivity in Cancer (GDSC) showed a trend for increased sensitivity to the MDM2 inhibitor Nutlin-3a in TP53 wild-type CRC cell lines with MDM2 gain compared those with MDM2 neutral copy number, with median natural log IC50 of 1.52 compared to 2.38 (p=0.060) (Figure S4F).

Mutations in some key genomic drivers of CRC were enriched in the CRC cases without p53 pathway alterations, including KRAS (75.8% vs 42.9%, q<0.001), PIK3CA (33.3% vs 14.3%, q<0.001) and SOX9 (21.1% vs 7.2%, q<0.001) (Figure S4G).

Finally, we used matched primary tumor and metastatic samples to investigate the putative “pre-WGD” state and found that, among MSS cases, WGD− primary with WGD+ metastasis had a higher frequency of p53 pathway alterations and higher FGA, compared to WGD− primary with WGD− metastasis (Figure S4H–J).

Patterns of metastatic spread

We used data automatically collected from radiology reports¹⁸ to investigate the patterns of metastasis to seven sites: liver, lung, peritoneum, reproductive system, bones, brain, and adrenal glands in the Clinical Cohort. Median follow-up was 44.6 months. A median of 11 radiology reports (MR, CT or PET scans (interquartile range [IQR] 6-18)) performed after CRC diagnosis were available per patient. Presence of metastatic disease was recorded in 72.2% of those with MSS CRC, 36.9% of those with MSI-H CRC, and 40.4% of those with pPOLE/D1.

At all three landmark time points (time of metastasis diagnosis, 1 year from metastasis diagnosis, and 3 years from metastasis diagnosis), a significantly larger portion of patients with MSS CRC had metastasis to two or more different organ sites compared to patients with MSI-H CRC (Figure 5A). At time of metastasis diagnosis, different metastatic patterns were already apparent (Figure S5A). Liver metastases were less frequent among patients with MSI-H CRC compared to those with MSS CRC (38.6% vs 65.4%; p<0.001, q<0.001) and those with pPOLE/D1 CRC (38.6% vs 63.2%; p=0.049, q=0.244). Patients with MSS CRC also displayed the highest relative frequency of lung metastases (37.3% vs 25.2% for MSI-H; p<0.001, q=0.005), whereas peritoneal metastases were more frequent in patients with MSI-H CRC (29.9% in MSS vs 40.1% in MSI-H; p=0.003, q=0.024).

Figure 5. — A) Number of metastatic sites (liver, lung, peritoneal, reproductive (repr.) system, bone, brain, and adrenal glands) by molecular subgroup at time of metastasis diagnosis (yr 0) and 1 and 3 years (yr) later. B) Cumulative incidence of metastasis to each site for patients with MSS CRC who develop metastasis to that site. Median time to metastases (mTTMs) are reported. C) Proportion of metastatic CRC patients who developed metastasis to a specific organ by primary tumor location and molecular classification. pPOLE/D1 CRC was not included. Significant relationships in pair-wise comparisons by Fisher’s exact test are marked with an asterisk. D) Coefficients from multivariate, time dependent Cox regressions to investigate interdependency of different metastatic sites. Metastatic sites already involved are on the y-axis and are each used as dependent variables; hazard ratios (HR) are on the x-axes. Each dot corresponds to a coefficient of the model with its 95% confidence interval (CI). Colors reflect q-values. Vertical, dashed lines indicate HR=1. E) Time to brain metastases depending on presence of bone and lung metastases at diagnosis in MSS cases. F) Multivariate models for associations of metastatic sites with overall survival (OS): presence of metastases—separately for the seven organs (top panels) and any organ (bottom panels)—are modelled as time dependent covariates for MSS (left panels) and MSI-H (right panels) cases. Age at diagnosis was also used as a covariate. Each dot represents the HR for death in the presence of metastases to that specific site and the respective 95%CI. G) OS depending on molecular subgroup and presence of metastatic disease at diagnosis. H) Median OS from the date of diagnosis of a metastasis to a specific organ. Lines represent 95%CI. E, G) Hazard ratios (HRs), 95% confidence intervals, and p-values were estimated using univariate Cox proportional hazards regression, with the indicated reference group for each comparison. P-values were calculated with Wilcoxon signed-rank sum test.

See also Figure S5 and Tables S1 and S2.

In metastatic MSS cases, median time to metastasis (TTM) ranged from 1.2 months (IQR: 0.2-5.6 months) for liver metastasis to 23.2 months (IQR: 10.3-38.4 months) for brain metastasis (Figures 5B and S5B). Conversely, in the MSI-H cases, an order of metastatic site involvement was less evident (Figures S5B and S5C). We separately examined the impact of primary tumor location on metastatic tropisms in the MSS and MSI-H cases. Metastasis patterns varied by primary site, including lower frequency of liver metastases and higher frequency of lung metastases in MSS rectal cancer, higher frequency of peritoneal metastases in MSS right-sided colon cancer and MSI-H left-sided colon cancer, and higher frequency of bone metastases in rectal cancer (Figure 5C; Table S1). Within MSS cases, patient age largely did not impact the number and identity of metastatic sites involved at the time of diagnosis of metastasis (Figures S5D and S5E). In the patients diagnosed before 35 years with right-sided primaries, there was the suggestion of increased metastatic load, potentially due to later diagnosis from right-sided tumors, which are unlikely to present with gross rectal bleeding.

We then used a multivariate Cox regression with time dependent covariate and correction for primary tumor location to measure the conditional risk of metastasis to each designated site in the presence of metastases to other sites in MSS CRC (Figures 5D and S5F). Notably, among the different co-occurrence patterns, we uncovered a much higher brain metastasis risk in the presence of bone or lung metastasis (Figures 5D and 5E). Reproductive system metastasis most commonly occurred in the presence of peritoneal metastasis; while liver, reproductive system, and adrenal metastases, all abdominal organs, were associated with increased risk for peritoneal involvement. Bone metastasis generally did not occur as a single, first-site of metastasis and was associated with prior involvement of any of the other metastatic sites, with the strongest relationship with brain and adrenal gland involvement.

The effect of metastatic disease on OS varied depending on the affected organ. In the MSS cohort, the highest risk of death was observed in the presence of liver or peritoneal metastases, whereas metastases to the reproductive system and lung were the least detrimental for OS (Figure 5F; Table S2). The effect on OS of having metastasis to any site was greater in the MSS cases (HR=20.91; 95%CI: 15.01-29.14) compared to the MSI-H cases (HR=8.24; 95%CI: 4.49-15.14) (Figures 5F and 5G), likely driven by the different metastatic load of the two groups (Figure S5G). The presence of metastatic disease to only one organ was associated with lower risk of death compared to multi-organ spread (Figure S5H); specifically, in the MSS cases, using patients without metastases as a reference, the longest OS were observed, in order, in patients with lung-only (3-year OS: 92.3%, 95%CI: 87.9-96.9), reproductive system-only (3-year OS: 87.6%, 95%CI: 79.3.8-96.7), liver only (3-year OS: 75.9%, 95%CI: 70.8-81.3) and peritoneal-only (3-year OS: 73.3%, 95%CI: 65.2-82.5) metastasis (Figure S5H).

To combine the diverse prognostic effect and the different timing of metastatic sites, we calculated, for each organ, the median OS from the diagnosis of metastasis to that site (Figure 5H). In the MSS cohort, the median OS after diagnosis of liver metastasis was 33.3 months, while brain and adrenal metastases were associated with a median OS of less than a year. These data identify brain and adrenal involvement as generally late events in the metastatic spread of CRC.

Role of primary tumor genomics and primary location in metastatic tropisms

We investigated associations between primary tumor genomic profile and metastatic tropisms of MSS CRC. First, using a multivariate logistic regression with correction for primary tumor location, we looked at the effect of alterations in key oncogenic pathways. We observed a higher frequency of metastatic disease in the presence of oncogenic alterations activating the MYC (OR=1.80, q=0.031), RAS (OR=1.51, q<0.001), or TGFβ pathways (OR=1.40, q=0.012). Conversely, metastasis less commonly developed in patients whose tumors harbored inactivating alterations in DDR (DNA Damage Response) (OR=0.68, q=0.012) or alterations activating NOTCH (OR=0.71, q=0.012) or WNT (OR=0.72, q=0.043) pathways (Figure 6A). We observed a positive correlation between the number of metastatic sites and the frequency of alterations in SMAD4, KRAS, and BRAF (Figure 6B), suggesting a possible role for these genes in metastasis. However, increased mutant dosage of the recurrently altered genes did not appear associated with increased metastatic penetrance (Figure S6A). We also checked for associations between recurrent copy number alterations (20q gain, 12p gain, 12q gain, 13q gain) and metastatic tropisms. Among these, chromosome 20q gain was associated with a decreased risk of metastasis to any sites, with a significant protective effect seen for metastasis to lung, bone, peritoneum, and reproductive system (Figure S6B).

Pathways influencing metastasis risk differed across anatomic sites (Figure 6C). We observed notable similarities between lung, brain, and adrenal gland metastases, which were more frequent in the presence of RAS and p53 pathways alterations, and between peritoneal and reproductive system metastases, which were more frequent in the presence of RAS and TGFβ pathways alterations and less frequent in the presence of WNT pathway alterations. RAS pathway activation was associated with increased risk of metastasis to all seven metastatic sites, suggesting higher metastatic competence from activation of this pathway. In contrast, WNT pathway activation was only associated with increased risk of liver metastasis, a more readily accessible landing site for CRC due to drainage of the colon into the portal vein, suggesting limited metastatic competence from this pathway.

We next aimed to stratify the metastatic risk for specific organs based on primary tumor location and tumor genomic profile. For each metastatic site, we first built a multivariate logistic regression using pathway alterations as predictors along with primary tumor location (Figure S7A). To confirm the robustness of pathways selected by logistic regression, we employed a range of complementary machine learning approaches—including LASSO, elastic net, gradient boosting—which showed largely consistent results (Figure S7B). We subsequently calculated for each patient seven organ-specific “risk scores”, based on the presence of prognostic genomic alterations and primary tumor location (Table S3); the factors for the brain and adrenal gland organ-specific models were the same. The risk scores were significantly associated with the risk of metastasis to the respective organs and could stratify across different timepoints the proportion of patients with metastatic disease (Figure S7A; Tables S4–S5). As such, combining primary location and pathway alterations effectively stratified organ-specific time to metastases (Figure 6D). We validated these scores using an independent cohort¹⁹ of patients with MSS CRC with available time to metastasis data and primary tumor comprehensive genomic profiling. In this separate cohort, we observed that the liver, lung, and peritoneal scores—but not the bone or brain scores—were significantly associated with the respective risk of metastases (Figure S7C; Table S6). Finally, we identified a profile highly associated with risk of brain metastasis, consisting of lung and bone metastases, which was further enhanced by the presence of alterations in the TGFβ pathway (Figure 6E).

DISCUSSION

Here, we present a deep integrated genomic and clinical analysis of over 7,000 CRCs. Our data suggest a convergence in the development of CRC with a key group of recurrent oncogenic alterations that appears to be enhanced and conserved through tumor evolution and metastasis. Alterations in oncogenic drivers appear to be selected to become clonal within the tumor, while non-oncogenic alterations or alterations in genes infrequently mutated in CRC may be less important to the tumor’s survival and thus more likely to remain subclonal. Tumors with the highest FGA were associated with higher frequency of p53 pathway alterations, clonal mutations in this pathway, and allelic imbalance affecting this pathway, indicating the functional effect of these genomic modulations. Further, our analysis of paired primary and metastatic samples suggests that mutations in the primary tumor that are subclonal or lack mutant allele selection are less likely to be conserved in the metastasis and thus may be less impactful for metastatic seeding. It is possible that a change in the allelic dosage of important oncogenes might facilitate the clonal expansion of CRC and its progression. Some key oncogenic drivers, such as KRAS, BRAF, and SMAD4 were associated with increased risk for metastasis with progressively higher alteration frequencies of these genes in patients with more sites of metastases, further suggesting their importance to CRC metastatic invasion.

While our analysis of clonality and mutation dosage highlights a central set of genomic drivers in CRC, it reveals a complicated network of signaling. For example, KRAS gain in KRAS mutant CRC was associated with worse survival only among left-sided primaries and not in right-sided primaries. We speculate that the KRAS gain in the left-sided colon cancers uniquely amplifies EGFR signaling, which is higher in the left-colon, leading to higher mitogenic signaling and worse outcomes. In WGD+ CRC, we found that the increased incidence of allelic imbalance for most oncogenes resulted from loss of the wild-type allele; however, in KRAS mutant CRC, there is gain of the mutant allele, suggesting a potential role for the wild-type allele in tumor fitness. Consistent with this, a recent study showed that deletion of wild-type Kras in a mouse model of metastatic Kras G12D/Trp53 loss intestinal cancer suppressed metastatic progression through modulation of the immune microenvironment and decreased tumor invasiveness and aggression²⁰. The transcriptional data suggest that KRAS expression is not increased in advanced CRC despite allelic imbalance. We speculate that it is possible that KRAS expression is higher at the adenoma stage, when increased KRAS activation has been associated with more effective tumor initiation²⁰, but, when the tumor becomes metastatic, there may be some modulation to bring down KRAS expression to support tumor invasion, migration, and metastatic progression^21,22. Our identification of an association of KRAS mutation and allelic imbalance with MDM2 gain in TP53 wild-type CRC suggests that KRAS allelic imbalance is an early event, prior to the adenoma-carcinoma transition when TP53 alterations are often acquired^2,3.

In the WGD− CRC with KRAS mutation, the allelic imbalance in chromosome 12 likely results in a gain in MDM2, which appears functional as it is strongly enriched in otherwise p53 pathway wild-type tumors, associated with increased FGA, and confers increased sensitivity to the MDM2 inhibitor nutulin-3a. Consistent with this, whole genome sequencing of about 2,000 MSS CRC cases revealed an association of WGD with both TP53 mutation and KRAS wild-type status²³. Even so, the gain in MDM2 likely has a weaker effect on the p53 pathway compared to mutations in TP53, as it isn’t as commonly selected in metastatic CRC.

The presence of mutations in key genomic drivers, such as KRAS, PIK3CA, and SOX9, may decrease selective pressure for p53 pathway alterations. Thus, the mutational profile of CRC and early evolution of oncogenic mutant allele imbalance may shape the tumor’s subsequent progression towards a higher chromosome instability or mitogenic pathway activation dependency. Given the frequency of allelic imbalance within key metastatic CRC drivers, particularly those in the mitogenic pathway, future trials of targeted treatments may need to consider the role of the target’s allelic configuration on drug sensitivity²⁴.

In our analyses, we specifically focused on the effect of patient age on tumor characteristics because of the concerning rise in early-onset CRC. Consistent with prior studies²⁵, early-onset CRC was largely similar to average-onset disease, after accounting for primary tumor location. We found a modest increase in TP53 mutation and decrease in APC mutation, possibly reflecting a component of bowel inflammation in some early-onset cases^26,27.

Using natural language processing, we were able to collect treatment exposure history for our large clinical series. Several interesting patterns emerged. A lower percentage of patients received BRAF targeted therapy than would be expected for the frequency of BRAF V600E mutation in CRC, possibly due to the poor clinical status associated with this mutation limiting the ability of patients to receive treatment²⁸. There was also a low frequency of EGFR inhibitor use for MSI-H CRC. We speculate that this may be due to the combined factors of relatively frequent resistance mutations, right-sided primary location, and efficacy of ICB precluding need for this treatment.

Finally, this study provides a comprehensive characterization of the clinical and genomic determinants underlying organ-specific metastatic risk in CRC. We identify marked differences between MSS and MSI-H disease, with MSS CRC presenting with a higher metastatic burden at diagnosis, a greater incidence of synchronous metastases, and more frequent involvement of the liver, lungs, and bones. In contrast, MSI-H CRC exhibits a compressed pattern of metastatic dissemination without clearly defined early or late metastatic sites. In MSS CRC, differential metastatic tropism results in an organ-specific order of metastatic progression and reproducible patterns of site co-occurrence, with three major clusters: (1) reproductive system and peritoneum, (2) adrenal glands, brain, and bone, and (3) liver and lung. These metastatic sites retain varied prognostic effects, emphasizing that metastatic location—not only metastatic burden—should be considered in routine clinical care and in the interpretation of clinical trials.

Several genomic pathways significantly affect metastatic tropisms. WNT pathway activation was associated with better prognosis, consistent with prior studies^6,29,30, and with decreased risk of peritoneal metastasis. This likely reflects the enrichment of WNT pathway wild-type tumors in the poor prognosis subgroups of BRAF V600E CRC and colitis-associated cancers^26,31–34, both of which are associated with peritoneal spread³⁵. WNT activation may also limit metastatic competence as CRC metastatic progression has been associated with a shift from intestinal cell identities and reprogramming into non-canonical differentiation³⁶. Consistent with prior studies^37,38, we also found that PI3K pathway activation in CRC was not associated with metastasis, raising questions about the role of this pathway in CRC invasiveness. We acknowledge that the pathway analysis may be an oversimplification as different nodes within the same pathway may have different effects; for example, we found that BRAF mutation had a stronger predilection for peritoneal metastases than KRAS mutation, while the latter was more associated with lung metastases.

By integrating primary tumor location, patterns of metastatic site co-occurrence, and primary tumor genomic features, metastatic tropism to individual organ sites can be inferred. Following prospective validation, organ-specific metastasis risk stratification may enable a shift from reactive to proactive diagnostic strategies, particularly for sites that are poorly assessed by standard imaging, such as brain metastases, where routine chest and abdominal CT scans are uninformative, or peritoneal carcinomatosis, where conventional abdominal CT imaging lacks sensitivity. For example, our data suggest that patients with lung and bone metastases whose tumors harbor TGFβ pathway alterations have a very high chance of developing brain metastasis and should be considered for surveillance brain imaging.

We acknowledge some limitations of our cohort. This cohort was assembled from patients treated at a single, major cancer referral center with the potential inherent biases of increased fitness and social economic status of this population. This study primarily focused on genomic features and further integration of transcriptomic and proteomic data is needed to fully characterize the downstream effect of the genomic changes identified.

Our large dataset provides an important resource, and we have made the annotated clinical and genomic data available through the cBioPortal for Cancer Genomics website. We believe this public database will accelerate the testing and validation of new hypotheses about metastatic CRC.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources should be directed to the lead contact, Rona Yaeger, (yaegerr@mskcc.org).

Materials availability

This study did not generate any new reagents.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS MSK-IMPACT cohort

A total of 7,238 tumor specimens from 6,789 patients with colorectal cancer (CRC) were collected and sequenced as part of routine clinical care at Memorial Sloan Kettering Cancer Center (MSK) before the data cut-off of November 17, 2023. All cases had a confirmed diagnosis of invasive adenocarcinoma and known primary tumor location within the colon or rectum. The cohort included 3,689 (54.3%) male and 3,100 (45.7%) female patients. Four incremental versions of the MSK-IMPACT panel were used: IM3 (341 genes), IM5 (410 genes), IM6 (468 genes), and IM7 (505 genes). Patient demographics and clinical characteristics, including age at diagnosis, sex, primary tumor location, stage at diagnosis, and molecular subtype, are summarized in Figure 1A and Table 1. Additional clinical characteristics, including smoking history, prior anti-cancer therapy, baseline Eastern Cooperative Oncology Group (ECOG) performance status, and baseline CEA levels, are summarized in Figure 1C. Genetic ancestry was inferred from matched normal blood sequencing as previously described⁴² and is reported in Table 1. Socioeconomic status data were not available. Included in this series are 1,062 patients and 1,054 samples previously reported⁶ with new data now available on mutation clonality, allelic imbalance, and metastatic sites. The study was approved by the MSK Institutional Review Board (IRB), and all patients provided written informed consent for tumor sequencing (MSK IRB 12-245) or had consented to institutional tumor banking protocols (MSK IRB 06-107, 16-1261, WA0096-12). All analyses were performed using one sample per patient unless otherwise specified; when multiple samples were available, the sample with a positive FACETS (Fraction and Allele-Specific Copy Number Estimates from Tumor Sequencing) quality check and the highest tumor purity estimate by the pathologist was selected.

Subcohorts and validation

Three overlapping subcohorts were defined for different analyses (Figure 1B): the Genomics Cohort (n=6,789 patients, 7,238 samples), the FACETS Cohort (n=5,520 patients, 5,807 samples), and the Clinical Cohort (n=4,517 patients, 4,784 samples). The Clinical Cohort was restricted to fully clinically annotated patients; sequential exclusion criteria are outlined in Figure S1B. The Clinical Cohort was further stratified by molecular subtype: microsatellite stable (MSS) (n=3,923 patients, 4,177 samples), microsatellite instability-high (MSI-H) (n=547 patients, 559 samples), and pathogenic POLE/POLD (pPOLE/POLD1) (n=47 patients, 48 samples). Several separate cohorts were included for validation and comparative purposes. The TCGA PanCancer Atlas cohort was used for demographic comparison and analysis of mRNA expression data. The GENIE BPC CRC V2.0 cohort, excluding MSK cases, was used to validate the association between primary tumor genomics and patterns of metastasis. DNA and RNA molecular profiling data from Caris Life Sciences were used as an independent validation cohort to evaluate MDM2 expression across TP53 and KRAS genotypes. The influence of sex and genetic ancestry on the genomic and clinical findings was not independently analyzed in this study, as the primary focus was on the association between molecular subtypes, genomic alterations, and organ-specific patterns of metastasis. The absence of these analyses may limit the generalizability of the findings across demographic subgroups.

Cell lines

No cell lines were directly cultured in this study. Drug sensitivity and copy number data for CRC cell lines with wild-type TP53 were obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) database (https://cellmodelpassports.sanger.ac.uk/downloads).

METHOD DETAILS

Clinical data abstraction

Clinical data elements were abstracted automatically from the electronic health record, partly using natural language processing (NLP) methods, as previously described,¹⁸ and included: smoking status at time of diagnosis, ECOG performance status (ECOG PS) at time of diagnosis, whether prior anti-tumor medications were administered before establishing care at MSK, anti-tumor medications administered at MSK, and presence of metastasis in specific organ sites at each available tumor scan. Presence of lymph nodes metastasis was not included in this analysis as the model does not differentiate between regional and distant, metastatic nodal involvement. International classification of disease version 9 (ICD-9) billing codes were used to inform staging at the time of diagnosis—comprising the diagnosis date and primary tumor location. Missing data were manually retrieved. Staging at diagnosis was dichotomized as stage I-III or IV. Patients with clinical or pathological evidence of metastases within 6 months from the time of diagnosis were categorized as stage IV. Overall survival (OS) was calculated from the date of diagnosis to date of death for any cause or last follow-up; except for the analyses of metastatic sites, where OS was calculated from the date of first diagnosis of metastases. Patients not experiencing an event were censored.

The analysis of organ-specific patterns of metastasis was restricted to the Clinical Cohort, and the patient flow diagram for the subgroup of fully clinically annotated patients (n=4,517 patients, 4,784 samples) is outlined in Figure S1A. From this cohort, we excluded the following: (1) patients with concomitant non-CRC metastatic malignancies, which could confound prognostic assessment; (2) patients with a primary lung or liver tumor, to prevent misclassification of lung or liver metastases by the NLP algorithm; (3) patients with multiple primary CRC diagnoses, as it was not possible to determine which primary tumor gave rise to the metastases; (4) patients without any CT or PET imaging performed at MSK; (5) patients whose primary CRC diagnosis occurred before January 1, 2015, prior to the routine implementation of MSK-IMPACT; and (6) patients whose first imaging used for NLP analysis was performed more than 6 months after diagnosis and showed metastases, as it was not possible to determine whether these represented synchronous or metachronous disease.

DNA extraction and sample sequencing

All tumor and matched blood normals underwent sequencing with MSK-IMPACT, a capture-based next-generation sequencing platform that can detect mutations, copy number alterations and selected rearrangements in 341–505 cancer-associated genes, depending on the version of the panel. Sequencing was performed in a Clinical Laboratory Improvement Amendments (CLIA)-certified molecular laboratory, as previously described⁴³. The average depth of sequencing was 622x.

Genomic analyses

All somatic mutations, copy number alteration calls, and fusions involving select genes were detected using a clinical validated pipeline⁴⁴. Gene panel differences (genes not present on earlier panels) were accounted for when comparing the frequency of alterations across cohorts by subsetting for samples sequenced with the earliest panel that did not include the gene being tested. Tumor mutational burden (TMB) was defined as the number of non-synonymous mutations per megabase sequenced. OncoKB, a precision oncology knowledgebase⁴¹, was used to annotate genomic alterations for their relevance in cancer development and potential clinical actionability. Variants were classified by OncoKB as oncogenic, likely oncogenic, or of unknown significance, and only oncogenic and likely oncogenic alterations were included in the genomic alterations analysis. Microsatellite instability (MSI) status was determined using MSIsensor⁴⁰. Tumors with an MSIsensor score ≥10 were classified as MSI-H, consistent with previous studies⁴⁵.

The FACETS algorithm^15,46 was used to infer allele-specific copy number profiles and estimate tumor purity and ploidy. FACETS outputs were subsequently processed using the facets-suite R package (https://github.com/mskcc/facets-suite/tree/master) to estimate cancer cell fraction (CCF), total and minor copy number, and the number of mutant allele copies, as previously described^47,48.

The MSK-IMPACT assay was designed to capture 2,025 common polymorphic sites distributed across the genome in addition to >5,000 common single nucleotide polymorphisms (SNPs) located within targeted exons and introns, resulting in an average of ~3,100 heterozygous SNPs observed per sample. FACETS integrates tumor and matched-normal sequencing coverage and allelic fractions at heterozygous SNPs to estimate tumor purity and ploidy, as well as integer major and minor allele copy number across the genome, enabling detection of allelic imbalance and loss of heterozygosity (LOH). Tumors were considered to have undergone whole-genome doubling (WGD) if more than 50% of the autosomal genome had a major copy number (MCN) ≥₂49_.

Mutation dosage was defined as the ratio of mutant allele copies to the total copy number (TCN) at the corresponding locus, using allele-specific copy number estimates inferred by FACETS and processed with facets-suite. A mutation dosage different from 0.5 was considered indicative of allelic imbalance, and mutation dosage >0.5 was interpreted as preferential selection of the mutant allele. Gene-level copy number gain was defined according to facets-suite criteria, which integrates the observed total copy number with the expected copy number given the tumor’s WGD status⁵⁰.

LOH was defined as the loss of one parental allele at a given locus and was identified when the minor (lesser) copy number (LCN) was 0 while the TCN remained greater than 0. Mutation clonality was inferred using FACETS-derived CCF estimates; mutations were classified as subclonal when the upper bound of the 95% confidence interval for the CCF was <1 and the posterior probability that CCF <0.5 exceeded 0.95⁴⁸. FACETS was also used to generate purity-corrected copy-number segmentation profiles, which were used to calculate the fraction of genome altered (FGA), defined as the proportion of the sequenced genome that deviated from a balanced diploid state. All samples included in the FACETS cohort passed quality control criteria as implemented in the facets-suite quality control pipeline⁵¹.

Caris Life Sciences cohort

Next Generation Sequencing (NGS)/Whole Exome Sequencing (WES)

NGS was performed on genomic DNA isolated from formalin-fixed paraffin-embedded (FFPE) tumor samples by a commercial CAP/CLIA laboratory (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 700 gene panel at high depth and coverage as previously described⁵². 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. Mutation assessment of individual genes stratified mutations as ‘pathogenic,’ and ‘likely pathogenic’⁵³.

Whole Transcriptome Sequencing (WTS)

Tumor-specific RNA was extracted from microdissected FFPE specimens with at least 10% tumor content using a Qiagen RNeasy FFPE Kit (Germantown, MD). RNA quality and quantity were determined using the Agilent TapeStation (Santa Clara, CA). Biotinylated RNA baits were hybridized to the synthesized and purified cDNA targets, and the bait-target complexes were amplified in a post-capture PCR reaction. Whole transcriptome sequencing (WTS) on Illumina NovaSeq (Illumina, Inc., San Diego, CA) generated an average of 60M reads. Raw data was demultiplexed by Illumina Dragen BioIT accelerator, trimmed, counted, PCR-duplicates removed, and aligned to human reference genome hg19 by STAR aligner (v.2.7.8a). Transcripts per million (TPMs) were calculated with Salmon expression pipeline⁵⁴.

Immunohistochemistry

Immunohistochemistry (IHC) of 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 (mismatch mutation repair). A board-certified pathologist evaluated all IHC results independently.

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 NGS⁵⁵.

The Caris dataset was used as an independent validation cohort to evaluate MDM2 expression across TP53 and KRAS genotypes.

Cell line analysis

We utilized the GDSC (Genomics of Drug Sensitivity in Cancer) database (https://cellmodelpassports.sanger.ac.uk/downloads) to assess response to the MDM2 inhibitor Nutlin-3a (-) in cell lines with wild-type TP53. We limited our analysis to cell lines with cancer type annotation of Colon adenocarcinoma (COAD) and Rectal adenocarcinoma (READ). The natural log of half-maximal inhibitory concentration (IC50) was used as a measurement for drug responsiveness. We derived copy number status directly from the CNA file downloaded, which categorizes genes for gain, amplification, loss, deletion, and neutral status.

QUANTIFICATION, STATISTICAL ANALYSIS, AND MACHINE LEARNING

Fisher’s exact test was used to compare categorical data. Spearman rank correlation was used to compute ρ coefficients and p values for numeric variables. Right-censored data were modelled using Cox univariate and multivariate regressions. Within multivariate Cox regressions, the presence of metastasis to a specific organ was considered as a time dependent covariate; that is, the patient was considered metastasis-free in the interval from the diagnosis until the occurrence of metastasis, the event of interest, or the end of the follow-up. When using time-to-metastasis to a specific organ as the dependent variable in cox regressions, death was considered as a competing event. The frequency of metastasis to a specific organ by molecular subtype (MSS, MSI-H or pPOLE/POLD1) or primary tumor location was calculated at three landmark timepoints—time of metastasis diagnosis, one year after metastasis diagnosis, and three years after metastasis diagnosis. Metastases clinically detected within 6 months of diagnosis were considered present at diagnosis, consistent with clinical staging.

We developed organ-specific risk scores to predict risk of metastasis to each of seven designated organs (liver, lung, peritoneum, reproductive system, bone, brain, and adrenal glands). For each designated metastasis site, we first built a multivariate Cox regression using the time-to-metastasis to the designated organ as the dependent variable. Covariates for each model included primary tumor location and presence or absence of at least one oncogenic alteration in any of the genes belonging to canonical signaling pathways, as defined in Sanchez-Vega et al⁵⁶. Next, for each designated organ, we identified which primary sites (Figure 5C) and pathways (Figure 6C) were significantly associated with risk of metastasis to that site. Finally, we tallied up points to calculate the seven organ-specific metastasis risk scores for each patient: the presence of at least one pathogenic alteration in a pathway significantly and positively (HR>1.00) associated with metastasis to the designated organ scored 1 point; the absence of alterations in a pathway negatively (HR<1.00) associated with metastasis to the designated organ scored 1 point; the presence of a primary tumor location positively associated with organ-specific metastasis scored 1 point. By design, each organ-specific risk score is intended to maximize metastasis risk stratification to a single organ; it is not meant to be informative for other metastatic sites, as the genomic alterations and primary tumor locations associated with metastasis to each specific organ are not necessarily linked to an increased risk of metastasis to a different site.

Machine learning-based survival modeling approaches were used to complement traditional statistical analyses and to evaluate robustness of associations with time-to-metastasis outcomes. In order to perform a cross validation of time to metastases analysis, penalized Cox regression (LASSO and Elastic Net) and gradient boosting survival analyses. Model hyperparameters for regression, including the strength of regularization and the relative weighting, were tuned by grid search with 5-fold cross-validation to maximize the concordance index. For gradient boosting survival analyses, hyperparameters (learning rate, number of boosting stages, and tree depth) were similarly grid-searched with cross-validation. After determining the optimal hyperparameters, we assessed feature significance using a permutation-based approach: each candidate feature was individually shuffled (while keeping all other features and the model parameters fixed), and the drop in predictive performance—quantified by the concordance index—was measured across multiple permutations (1000 repeats). A permutation-based p-value was then calculated for each feature as the fraction of permutations whose concordance index matched or exceeded the baseline index from the unpermuted data.

A threshold of 0.05 was used for uncorrected p-values significance. The false discovery rate (FDR) method was used to correct p-values to account for multiple comparisons. FDR-corrected p-values (q-values) were deemed significant below 0.10.

The majority of analyses were performed in R version 4.3.2⁵⁷. Cross validation of the results for the time-to-metastasis multivariate Cox regressions were performed in Python 3.10.14 using the scikit-learn (v1.5.2) and scikit-survival (0.23.0) libraries.

Supplementary Material

NIHMS2208821-supplement-1.pdf^{(19.1MB, pdf)}

Highlights:

Prospective targeted sequencing of >7000 colorectal cancers in the clinical setting
Recurrent alterations in CRC are often clonal and enhanced by allelic imbalance
Mutant KRAS allelic imbalance is likely an early event and may lead to MDM2 gain
Organ-specific metastasis risk based on prior metastases, tumor location, genomics

ACKNOWLEDGEMENTS

This work was supported by NIH/NCI P30 CA00878 to MSK and R01 CA233736 (to R.Y.). We acknowledge the Colorectal Cancer Research Center, the Marie-Josée and Henry R. Kravis Center for Molecular Oncology, and the Molecular Diagnostics Service at MSK. F.S.-V. and K.G. acknowledge research support from the Fund for Innovation in Cancer Informatics (ICI). F.S.-V. was also supported by a Chris4Life Colorectal Cancer Alliance Early Career Investigator Award Grant and a Corning MSKCC Health Equity Scholarship. We thank the patients included in this study.

DECLARATION OF INTERESTS

S.K.M. and G.W.S. are employees of Caris Life Sciences.

M.M. is a consultant with Caris Life Sciences.

A.C. has been a member of the following advisory boards: Abbvie, Amgen, Agenus, Daiichi Sankyo, GSK, Merck, Regeneron, Janssen, Pfizer, Roche, Summit, 3T Biosciences, and Urogen. She has also received research funding from GSK and Pfizer.

N.H.S. reports Consulting/ Advisory board: 3T Biosciences, Regeneron, Pfizer, Agenus, Astellas, Pfizer, Puretech, Novartis and Research funding: Roche/Genentech, Pfizer, Merck, BMS, AstraZeneca, Puretech, Immunocore, Regeneron, Agenus.

Z.K.S. has intellectual property rights in SOPHiA Genetics and serves as an Associate Editor for JCO Precision Oncology and as a Section Editor for UpToDate. Z.K.S.’s immediate family member serves on the Board of Directors for Adverum Biotechnologies, is Co-Founder, CMO and President for Blue Gen Therapeutics Foundation, and serves as a consultant in Ophthalmology for Beacon Therapeutics and Regeneron Pharmaceuticals, Inc. outside the submitted work.

M.N. has received research funding paid to her institution from Taiho Oncology, Inc. and Syndax Pharmaceuticals, Inc. She has provided consultation to Takeda Pharmaceuticals (uncompensated, other than travel arrangements) and has served as CME faculty for Haymarket Media (uncompensated).

M.B.F. has served on advisory boards for Bristol Meyers Squibb, Inter, Genzyme, and Abbott Laboratories.

M.F.B, reports Personal fees from AstraZeneca and Paige.AI, research support from BoundlessBio, and intellectual property rights from SOPHiA Genetics.

R.Y. has served as an advisor for Mirati Therapeutics, Revolution Medicines, Loxo@Lilly, Erasca, Merck, Parabilis Medicines, and Bayer; served as a Data and Safety Monitoring Board member for CytoDyn; and received research support to her institution from Pfizer, Boehringer Ingelheim, Mirati Therapeutics, Daiichi Sankyo, Boundless Bio, Revolution Medicines, Parabilis Medicines, and Eli Lilly.

All other authors report no competing interests.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Data and code availability

Our complete dataset including genomic and clinical data is publicly available through the cBioPortal for Cancer Genomics: https://www.cbioportal.org/study/summary?id=crc_msk_2026. All accompanying genomic and clinical data are included in this manuscript as a compressed ZIP file that is also deposited in the public Zenodo digital library at https://zenodo.org/records/20707132. MSK-IMPACT clinical sequencing data are considered protected patient information and access to raw sequencing data is contingent on additional institutional approvals. Accession links for all publicly available datasets analyzed in this study (TCGA, GENIE BPC, GDSC) are listed in the Key Resources Table. For the Caris Life Sciences’ data, contact Caris Life Sciences or go through their letter of intent process. This paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Human tumor and matched normal (blood) samples	This study	N/A
Deposited data
Genomic and clinical data at patient and tumor level is deposited for visualization	This study	cBioPortal.org: https://www.cbioportal.org/study?id=crc_msk_2026
Genomic and clinical data at patient and tumor level is deposited for download	This study	Zenodo: https://zenodo.org/records/20707132
TCGA RNA-seq data (COAD/READ)	NCI Genomic Data Commons	https://portal.gdc.cancer.gov/ ; RRID: SCR_014514
GENIE BPC CRC V2.0 cohort (excluding MSK cases)	AACR Project GENIE	https://genie.cbioportal.org/study/summary?id=crc_public_genie_bpc
GDSC drug sensitivity, mutations, and copy number data	Genomics of Drug Sensitivity in Cancer (GDSC1_fitted_dose_response_27Oct23.xlsx, mutations_summary_20230202.zip, cnv_summary_20230303.zip)	https://cellmodelpassports.sanger.ac.uk/downloads
Caris Life Sciences colorectal cancer DNA (targeted panel and/or WES) and RNA molecular profiling data	Caris Life Sciences	N/A (proprietary; not available, contact Caris Life Sciences)
Software and algorithms
R (v4.3.2)	R Project	https://www.r-project.org/ ; RRID: SCR_001905
survival R package	https://cran.r-project.org/web/packages/survival/	https://doi.org/10.32614/CRAN.package.survival
Python (v3.10.14)	Python Software Foundation	https://www.python.org/ ; RRID: SCR_008394
scikit-learn (v1.5.2)	Pedregosa et al., 2011	https://scikit-learn.org/ ; RRID: SCR_002577
scikit-survival (v0.23.0)	Pölsterl, 2020³⁹	https://scikit-survival.readthedocs.io/
FACETS	Shen & Seshan, 2016¹⁵	https://github.com/mskcc/facets ; RRID: SCR_026264
MSIsensor	Niu et al., 2014⁴⁰	https://github.com/ding-lab/msisensor; RRID: SCR_006418
OncoKB	Chakravarty et al., 2017⁴¹	https://www.oncokb.org/ ; RRID: SCR_014782
Other
cBioPortal		http://www.cbioportal.org/ ; RRID:SCR_014555

Open in a new tab

REFERENCES

1.Siegel RL, Wagle NS, and Jemal A. (2026). Leading cancer deaths in people younger than 50 years. JAMA 335, 632–634. [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Fearon ER, and Vogelstein B. (1990). A genetic model for colorectal tumorigenesis. Cell 61, 759–767. [DOI] [PubMed] [Google Scholar]
3.Vogelstein B, Fearon ER, Hamilton SR, Kern SE, Preisinger AC, Leppert M, Nakamura Y, White R, Smits AM, and Bos JL. (1988). Genetic alterations during colorectal-tumor development. N. Engl. J. Med. 319, 525–532. [DOI] [PubMed] [Google Scholar]
4.Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, Diaz LA Jr, and Kinzler KW. (2013). Cancer genome landscapes. Science 339, 1546–1558. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Cancer Genome Atlas Network (2012). Comprehensive molecular characterization of human colon and rectal cancer. Nature 487, 330–337. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Yaeger R, Chatila WK, Lipsyc MD, Hechtman JF, Cercek A, Sanchez-Vega F, Jayakumaran G, Middha S, Zehir A, Donoghue MTA, et al. (2018). Clinical sequencing defines the genomic landscape of metastatic colorectal cancer. Cancer Cell 33, 125–136.e3. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Kopetz S, Desai J, Chan E, Hecht JR, O’Dwyer PJ, Maru D, Morris V, Janku F, Dasari A, Chung W, et al. (2015). Phase II pilot study of vemurafenib in patients with metastatic BRAF-mutated colorectal cancer. J. Clin. Oncol. 33, 4032–4038. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Hong DS, Fakih MG, Strickler JH, Desai J, Durm GA, Shapiro GI, Falchook GS, Price TJ, Sacher A, Denlinger CS, et al. (2020). KRASG12C inhibition with sotorasib in advanced Solid Tumors. N. Engl. J. Med. 383, 1207–1217. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Jiang J, Jiang L, Maldonato BJ, Wang Y, Holderfield M, Aronchik I, Winters IP, Salman Z, Blaj C, Menard M, et al. (2024). Translational and therapeutic evaluation of RAS-GTP inhibition by RMC-6236 in RAS-driven cancers. Cancer Discov. 14, 994–1017. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Dumbrava EE, Shapiro GI, Parikh AR, Johnson ML, Tolcher AW, Thompson JA, El-Khoueiry AB, Vandross AL, Kummar S, Shepard DR, et al. (2026). Phase 1 study of rezatapopt, a p53 reactivator, in TP53 Y220C-mutated tumors. N. Engl. J. Med. 394, 872–883. [DOI] [PubMed] [Google Scholar]
11.Kopetz S, Yoshino T, Van Cutsem E, Eng C, Kim TW, Wasan HS, Desai J, Ciardiello F, Yaeger R, Maughan TS, et al. (2025). Encorafenib, cetuximab and chemotherapy in BRAF-mutant colorectal cancer: a randomized phase 3 trial. Nat. Med. 10.1038/s41591-024-03443-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Siena S, Yamaguchi K, Ruffinelli Rodriguez JC, Corral de la Fuente E, Kuboki Y, Cremolini C, Victoria Ruiz I, Elez Fernandez ME, Strickler JH, Furqan M, et al. (2024). 505O Sotorasib (soto), panitumumab (pani) and FOLFIRI in the first-line (1L) setting for KRAS G12C–mutated metastatic colorectal cancer (mCRC): Safety and efficacy analysis from the phase Ib CodeBreaK 101 study. Ann. Oncol. 35, S429–S430. [Google Scholar]
13.Siegel RL, Kratzer TB, Wagle NS, Sung H, and Jemal A. (2026). Cancer statistics, 2026. CA Cancer J. Clin. 76, e70043. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Liu J, Lichtenberg T, Hoadley KA, Poisson LM, Lazar AJ, Cherniack AD, Kovatich AJ, Benz CC, Levine DA, Lee AV, et al. (2018). An integrated TCGA Pan-Cancer Clinical Data Resource to drive high-quality survival outcome analytics. Cell 173, 400–416.e11. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Shen R, and Seshan VE (2016). FACETS: allele-specific copy number and clonal heterogeneity analysis tool for high-throughput DNA sequencing. Nucleic Acids Res. 44, e131. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Varghese AM, Perry MA, Chou JF, Nandakumar S, Muldoon D, Erakky A, Zucker A, Fong C, Mehine M, Nguyen B, et al. (2025). Clinicogenomic landscape of pancreatic adenocarcinoma identifies KRAS mutant dosage as prognostic of overall survival. Nat. Med. 10.1038/s41591-024-03362-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Yan P, Klingbiel D, Saridaki Z, Ceppa P, Curto M, McKee TA, Roth A, Tejpar S, Delorenzi M, Bosman FT, et al. (2016). Reduced expression of SMAD4 is associated with poor survival in colon cancer. Clin. Cancer Res. 22, 3037–3047. [DOI] [PubMed] [Google Scholar]
18.Jee J, Fong C, Pichotta K, Tran TN, Luthra A, Waters M, Fu C, Altoe M, Liu S-Y, Maron SB, et al. (2024). Automated real-world data integration improves cancer outcome prediction. Nature 636, 728–736. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.AACR project GENIE Biopharma Collaborative releases colorectal cancer dataset (2023). Cancer Discov. 13, OF6. [DOI] [PubMed] [Google Scholar]
20.Najumudeen AK, Fey SK, Millett LM, Ford CA, Gilroy K, Gunduz N, Ridgway RA, Anderson E, Strathdee D, Clark W, et al. (2024). KRAS allelic imbalance drives tumour initiation yet suppresses metastasis in colorectal cancer in vivo. Nat. Commun. 15, 100. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Gadal S, Boyer JA, Roy SF, Outmezguine NA, Sharma M, Li H, Fan N, Chan E, Romin Y, Barlas A, et al. (2025). Tumorigenesis driven by BRAFV600E requires secondary mutations that overcome its feedback inhibition of RAC1 and migration. Cancer Res. 85, 1611–1627. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Ferguson J, Arozarena I, Ehrhardt M, and Wellbrock C (2013). Combination of MEK and SRC inhibition suppresses melanoma cell growth and invasion. Oncogene 32, 86–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Cornish AJ, Gruber AJ, Kinnersley B, Chubb D, Frangou A, Caravagna G, Noyvert B, Lakatos E, Wood HM, Thorn S, et al. (2024). The genomic landscape of 2,023 colorectal cancers. Nature 633, 127–136. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Hyman DM, Smyth LM, Donoghue MTA, Westin SN, Bedard PL, Dean EJ, Bando H, El-Khoueiry AB, Pérez-Fidalgo JA, Mita A, et al. (2017). AKT inhibition in Solid Tumors with AKT1 mutations. J. Clin. Oncol. 35, 2251–2259. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Cercek A, Chatila WK, Yaeger R, Walch H, Fernandes GDS, Krishnan A, Palmaira L, Maio A, Kemel Y, Srinivasan P, et al. (2021). A comprehensive comparison of early-onset and average-onset colorectal cancers. J. Natl. Cancer Inst. 113, 1683–1692. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Chatila WK, Walch H, Hechtman JF, Moyer SM, Sgambati V, Faleck DM, Srivastava A, Tang L, Benhamida J, Ismailgeci D, et al. (2023). Integrated clinical and genomic analysis identifies driver events and molecular evolution of colitis-associated cancers. Nat. Commun. 14, 110. [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Willauer AN, Liu Y, Pereira AAL, Lam M, Morris JS, Raghav KPS, Morris VK, Menter D, Broaddus R, Meric-Bernstam F, et al. (2019). Clinical and molecular characterization of early-onset colorectal cancer. Cancer 125, 2002–2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Chu JE, Johnson B, Kugathasan L, Morris VK, Raghav K, Swanson L, Lim HJ, Renouf DJ, Gill S, Wolber R, et al. (2020). Population-based screening for BRAF V600E in metastatic colorectal cancer reveals increased prevalence and poor prognosis. Clin. Cancer Res. 26, 4599–4605. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Jorissen RN, Christie M, Mouradov D, Sakthianandeswaren A, Li S, Love C, Xu Z-Z, Molloy PL, Jones IT, McLaughlin S, et al. (2015). Wild-type APC predicts poor prognosis in microsatellite-stable proximal colon cancer. Br. J. Cancer 113, 979–988. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Mondaca S, Walch H, Nandakumar S, Chatila WK, Schultz N, and Yaeger R (2020). Specific mutations in APC, but not alterations in DNA damage response, associate with outcomes of patients with metastatic colorectal cancer. Gastroenterology 159, 1975–1978.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Robles AI, Traverso G, Zhang M, Roberts NJ, Khan MA, Joseph C, Lauwers GY, Selaru FM, Popoli M, Pittman ME, et al. (2016). Whole-exome sequencing analyses of inflammatory bowel disease-associated colorectal cancers. Gastroenterology 150, 931–943. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Yaeger R, Shah MA, Miller VA, Kelsen JR, Wang K, Heins ZJ, Ross JS, He Y, Sanford E, Yantiss RK, et al. (2016). Genomic alterations observed in colitis-associated cancers are distinct from those found in sporadic colorectal cancers and vary by type of inflammatory bowel disease. Gastroenterology 151, 278–287.e6. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Yaeger R, Paroder V, Bates DDB, Capanu M, Chou J, Tang L, Chatila W, Schultz N, Hersch J, and Kelsen D (2020). Systemic chemotherapy for metastatic colitis-associated cancer has a worse outcome than sporadic colorectal cancer: Matched case cohort analysis. Clin. Colorectal Cancer 19, e151–e156. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Adams SV, Ahnen DJ, Baron JA, Campbell PT, Gallinger S, Grady WM, LeMarchand L, Lindor NM, Potter JD, and Newcomb PA (2013). Survival after inflammatory bowel disease-associated colorectal cancer in the Colon Cancer Family Registry. World J. Gastroenterol. 19, 3241–3248. [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Tran B, Kopetz S, Tie J, Gibbs P, Jiang Z-Q, Lieu CH, Agarwal A, Maru DM, Sieber O, and Desai J (2011). Impact of BRAF mutation and microsatellite instability on the pattern of metastatic spread and prognosis in metastatic colorectal cancer. Cancer 117, 4623–4632. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Moorman A, Benitez EK, Cambulli F, Jiang Q, Mahmoud A, Lumish M, Hartner S, Balkaran S, Bermeo J, Asawa S, et al. (2025). Progressive plasticity during colorectal cancer metastasis. Nature 637, 947–954. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Yaeger R, Cowell E, Chou JF, Gewirtz AN, Borsu L, Vakiani E, Solit DB, Rosen N, Capanu M, Ladanyi M, et al. (2015). RAS mutations affect pattern of metastatic spread and increase propensity for brain metastasis in colorectal cancer. Cancer 121, 1195–1203. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Lan Y-T, Jen-Kou L, Lin C-H, Yang S-H, Lin C-C, Wang H-S, Chen W-S, Lin T-C, Jiang J-K, and Chang S-C (2015). Mutations in the RAS and PI3K pathways are associated with metastatic location in colorectal cancers. J. Surg. Oncol. 111, 905–910. [DOI] [PubMed] [Google Scholar]
39.Pölsterl S (2020). scikit-survival: A Library for Time-to-Event Analysis Built on Top of scikit-learn. J. Mach. Learn. Res. 21, 212:1–212:6. [Google Scholar]
40.Niu B, Ye K, Zhang Q, Lu C, Xie M, McLellan MD, Wendl MC, and Ding L (2014). MSIsensor: microsatellite instability detection using paired tumor-normal sequence data. Bioinformatics 30, 1015–1016. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Chakravarty D, Gao J, Phillips SM, Kundra R, Zhang H, Wang J, Rudolph JE, Yaeger R, Soumerai T, Nissan MH, et al. (2017). OncoKB: A precision oncology knowledge base. JCO Precis. Oncol. 2017. 10.1200/PO.17.00011. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Arora K, Tran TN, Kemel Y, Mehine M, Liu YL, Nandakumar S, Smith SA, Brannon AR, Ostrovnaya I, Stopsack KH, et al. (2022). Genetic ancestry correlates with somatic differences in a real-world clinical cancer sequencing cohort. Cancer Discov. 12, 2552–2565. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Cheng DT, Mitchell TN, Zehir A, Shah RH, Benayed R, Syed A, Chandramohan R, Liu ZY, Won HH, Scott SN, et al. (2015). Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT): A hybridization capture-based next-generation sequencing clinical assay for solid tumor molecular oncology. J. Mol. Diagn. 17, 251–264. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Zehir A, Benayed R, Shah RH, Syed A, Middha S, Kim HR, Srinivasan P, Gao J, Chakravarty D, Devlin SM, et al. (2017). Mutational landscape of metastatic cancer revealed from prospective clinical sequencing of 10,000 patients. Nat. Med. 23, 703–713. [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Middha S, Zhang L, Nafa K, Jayakumaran G, Wong D, Kim HR, Sadowska J, Berger MF, Delair DF, Shia J, et al. (2017). Reliable pan-cancer microsatellite instability assessment by using targeted next-generation sequencing data. JCO Precis. Oncol. 2017. 10.1200/PO.17.00084. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.mskcc/facets-suite: Functions to run and parse output from FACETS (2022). https://rdrr.io/github/mskcc/facets-suite.
47.Dentro SC, Wedge DC, and Van Loo P (2017). Principles of reconstructing the subclonal architecture of cancers. Cold Spring Harb. Perspect. Med. 7, a026625. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Bielski CM, Donoghue MTA, Gadiya M, Hanrahan AJ, Won HH, Chang MT, Jonsson P, Penson AV, Gorelick A, Harris C, et al. (2018). Widespread selection for oncogenic mutant allele imbalance in cancer. Cancer Cell 34, 852–862.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Bielski CM, Zehir A, Penson AV, Donoghue MTA, Chatila W, Armenia J, Chang MT, Schram AM, Jonsson P, Bandlamudi C, et al. (2018). Genome doubling shapes the evolution and prognosis of advanced cancers. Nat. Genet. 50, 1189–1195. [DOI] [PMC free article] [PubMed] [Google Scholar]
50.mskcc/facets-suite source: data-raw/sysdata.R https://rdrr.io/github/mskcc/facets-suite/src/data-raw/sysdata.R.
51.check_fit: Sample QC in mskcc/facets-suite: Functions to run and parse output from FACETS (2022). https://rdrr.io/github/mskcc/facets-suite/man/check_fit.html.
52.Nazha B, Zhuang T, Wu S, Brown JT, Magee D, Carthon BC, Kucuk O, Nabhan C, Barata PC, Heath EI, et al. (2023). Comprehensive genomic profiling of penile squamous cell carcinoma and the impact of human papillomavirus status on immune-checkpoint inhibitor-related biomarkers. Cancer 129, 3884–3893. [DOI] [PubMed] [Google Scholar]
53.Richards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, Grody WW, Hegde M, Lyon E, Spector E, et al. (2015). Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet. Med. 17, 405–424. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Patro R, Duggal G, Love MI, Irizarry RA, and Kingsford C (2017). Salmon provides fast and bias-aware quantification of transcript expression. Nat. Methods 14, 417–419. [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Vanderwalde A, Spetzler D, Xiao N, Gatalica Z, and Marshall J (2018). Microsatellite instability status determined by next-generation sequencing and compared with PD-L1 and tumor mutational burden in 11,348 patients. Cancer Med. 7, 746–756. [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Sanchez-Vega F, Mina M, Armenia J, Chatila WK, Luna A, La KC, Dimitriadoy S, Liu DL, Kantheti HS, Saghafinia S, et al. (2018). Oncogenic signaling pathways in The Cancer Genome Atlas. Cell 173, 321–337.e10. [DOI] [PMC free article] [PubMed] [Google Scholar]
57.The R Project for Statistical Computing https://www.r-project.org/.

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

NIHMS2208821-supplement-1.pdf^{(19.1MB, pdf)}

Data Availability Statement

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Human tumor and matched normal (blood) samples	This study	N/A
Deposited data
Genomic and clinical data at patient and tumor level is deposited for visualization	This study	cBioPortal.org: https://www.cbioportal.org/study?id=crc_msk_2026
Genomic and clinical data at patient and tumor level is deposited for download	This study	Zenodo: https://zenodo.org/records/20707132
TCGA RNA-seq data (COAD/READ)	NCI Genomic Data Commons	https://portal.gdc.cancer.gov/ ; RRID: SCR_014514
GENIE BPC CRC V2.0 cohort (excluding MSK cases)	AACR Project GENIE	https://genie.cbioportal.org/study/summary?id=crc_public_genie_bpc
GDSC drug sensitivity, mutations, and copy number data	Genomics of Drug Sensitivity in Cancer (GDSC1_fitted_dose_response_27Oct23.xlsx, mutations_summary_20230202.zip, cnv_summary_20230303.zip)	https://cellmodelpassports.sanger.ac.uk/downloads
Caris Life Sciences colorectal cancer DNA (targeted panel and/or WES) and RNA molecular profiling data	Caris Life Sciences	N/A (proprietary; not available, contact Caris Life Sciences)
Software and algorithms
R (v4.3.2)	R Project	https://www.r-project.org/ ; RRID: SCR_001905
survival R package	https://cran.r-project.org/web/packages/survival/	https://doi.org/10.32614/CRAN.package.survival
Python (v3.10.14)	Python Software Foundation	https://www.python.org/ ; RRID: SCR_008394
scikit-learn (v1.5.2)	Pedregosa et al., 2011	https://scikit-learn.org/ ; RRID: SCR_002577
scikit-survival (v0.23.0)	Pölsterl, 2020³⁹	https://scikit-survival.readthedocs.io/
FACETS	Shen & Seshan, 2016¹⁵	https://github.com/mskcc/facets ; RRID: SCR_026264
MSIsensor	Niu et al., 2014⁴⁰	https://github.com/ding-lab/msisensor; RRID: SCR_006418
OncoKB	Chakravarty et al., 2017⁴¹	https://www.oncokb.org/ ; RRID: SCR_014782
Other
cBioPortal		http://www.cbioportal.org/ ; RRID:SCR_014555

Open in a new tab

PERMALINK

Integrated clinicogenomic analysis reveals the evolution and metastatic tropisms of advanced colorectal cancer

Paolo Manca

Henry Walch

Jierui Xu

Natalie Small

Ayush Kris

Evan Seffar

Allison Richards

Tiffany Xue

Adam Price

Mark T A Donoghue

Christopher Fong

Michele Waters

Justin Jee

Sachin Kumar Deshmukh

George W Sledge Jr

Midhun Malla

Andrea Cercek

Leonard Saltz

Neil H Segal

Zsofia K Stadler

Maliha Nusrat

Michael B Foote

Avni Desai

Sree Chalasani

Devika Rao

Louise Connell

Richard K G Do

Karuna Ganesh

Martin R Weiser

Michael F Berger

David B Solit

Jinru Shia

Efsevia Vakiani

Julio Garcia-Aguilar

Nikolaus Schultz

Francisco Sanchez-Vega

Walid K Chatila

Rona Yaeger

SUMMARY

eTOC Blurb:

Graphical Abstract

INTRODUCTION

RESULTS

Cohort definition

Table 1.

Figure 1. Study cohort characteristics.

Figure 2. Microsatellite stable (MSS) CRC mutational landscape.

Clonality and mutation dosage in MSS CRC

Figure 3. Allelic imbalance in MSS CRC.

Impact of TP53 status on copy number changes

Figure 4. p53 pathway and copy number alterations (CNAs).

Patterns of metastatic spread

Figure 5. Metastatic patterns.

Role of primary tumor genomics and primary location in metastatic tropisms

Figure 6. Organ-specific metastatic risk scores.

DISCUSSION

RESOURCE AVAILABILITY

Lead contact

Materials availability

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS MSK-IMPACT cohort

Subcohorts and validation

Cell lines

METHOD DETAILS

Clinical data abstraction

DNA extraction and sample sequencing

Genomic analyses

Caris Life Sciences cohort

Next Generation Sequencing (NGS)/Whole Exome Sequencing (WES)

Whole Transcriptome Sequencing (WTS)

Immunohistochemistry

Microsatellite instability or mismatch repair (MSI/MMR) status

Cell line analysis

QUANTIFICATION, STATISTICAL ANALYSIS, AND MACHINE LEARNING

Supplementary Material

Highlights:

ACKNOWLEDGEMENTS

DECLARATION OF INTERESTS