Age-stratified mutation patterns in early-onset colorectal cancer reveal distinct molecular features and therapeutic implications

Abstract

Background

Colorectal cancer (CRC) is increasingly diagnosed in younger adults, with evidence that early-onset cases (age <50 years) differ in the spectrum of prevalent gene mutations compared with older individuals. To evaluate how these age-related differences may inform testing guidelines and therapeutic development, we examined mutation rates of the most prevalent gene mutations across four age-stratified cohorts.

Patients and methods

Clinicogenomic data were obtained from Memorial Sloan Kettering Center for Harmonized Onco-genomic Research Dataset and China Pan-Cancer cohorts available in cBioPortal. A total of 6762 samples were analyzed. Mutation frequencies for a comprehensive panel of the 100 most prevalent CRC genes were compared across four age groups: 18-29 (n = 79), 30-39 (n = 402), 40-49 (n = 1064), and ≥50 (n = 5217) using chi-square analysis. False discovery rate (FDR) correction for multiple comparisons was carried out using Benjamini–Hochberg procedure.

Results

Statistically significant variation in mutation frequency across age groups was seen in 22 key genes. APC mutations increased with age and were seen in 49.4% of patients in the 18-29 group, 69.7% in 30-39, 73.3% in 40-49, and 75.25% of patients ≥50 (P < 0.001, FDR < 0.001). The oldest cohort was more than three times more likely to have an APC mutation than the youngest [odds ratio (OR) = 3.74, 95% confidence interval (CI) 2.44-5.74, P < 0.001]. In contrast, SMAD4 mutations were twice as common in the youngest age group at 31.6% compared with those over 40, with a prevalence of 17.29% in patients 40-49, and 18.84% in patients over 50 (OR = 2.03, 95% CI 1.26-3.27, P < 0.001, FDR < 0.001). POLE mutations peaked in the 30-39 age group with a prevalence of 10.7% compared with 6.3% in patients aged 18-29, 4.9% in patients aged 40-49, and 5.9% in patients aged ≥50 (P < 0.001, FDR < 0.001). Individuals in the 30-39 group were nearly twice as likely to carry a POLE mutation compared with those over 40 (OR = 1.96, 95% CI 1.41-2.74, P < 0.001).

Conclusions

Differences in mutations of key genes including a lower prevalence of APC mutations and increased SMAD4 mutations in younger individuals provides further supporting evidence that early-onset CRC may represent a distinct biological subtype of CRC. Enrichment of POLE mutations in younger patients highlights the importance of expanded molecular profiling in early-onset CRC, which could help identify patients most likely to benefit from immunotherapy and advance personalized treatment strategies in CRC. Together, these findings reinforce the need to approach early-onset CRC as a distinct biological entity and ensure that appropriate molecular assays are incorporated to guide care.

Highlights

• •Age-stratified analysis by decade reveals distinct CRC mutation patterns in early versus average-onset disease.
• •APC mutations increase, whereas SMAD4 is enriched in younger individuals.
• •POLE mutations peak in patients aged 30-39 with implications for favorable responses to immunotherapy.
• •EOCRC shows noncanonical mutation patterns.
• •Findings support expanded molecular profiling in younger patients.

Introduction

Colorectal cancer (CRC) remains a leading cause of cancer-related death worldwide.¹ Previously considered a disease predominantly affecting older adults, recent decades have seen an epidemiological shift with a rise in CRC cases among younger individuals.^2,3 Early-onset colorectal cancer (EOCRC), defined as diagnosis before age 50, now accounts for up to 10% of new cases in high-income countries and is projected to increase in the next decade.^3,4 In contrast, the incidence of CRC in adults over 50 has declined, a trend largely attributed to improvements in screening.3, 4, 5

Although several risk factors have been associated with the development of EOCRC including obesity, red meat consumption, low fiber intake, microbiome disruption, inflammatory bowel disease (IBD), and hereditary cancer syndromes such as Lynch syndrome and familial adenomatous polyposis, the underlying reasons for the rising incidence of EOCRC remain largely unknown.^2,3,5, 6, 7, 8 Although patients with IBD and hereditary syndromes have a substantially increased individual risk, most EOCRC cases remain sporadic, suggesting that additional genetic, environmental, and lifestyle factors likely contribute to disease development. However, definitive causal mechanisms have not been established.^2,3,5,6

EOCRC exhibits unique clinical and genomic characteristics compared with late-onset disease. Younger individuals are more likely to present with advanced disease, left-sided tumors, distal colon or rectal involvement, poor histological differentiation, and lymph node or distant metastasis.^4,5 In addition to distinct clinical features, molecular profiling suggests that EOCRC exhibits a unique mutational landscape compared with later-onset disease. Previous research has found that EOCRC is associated with increased frequencies of tumor protein P53 (TP53) and catenin beta 1 (CTNNB1) mutations and decreased rates of adenomatous polyposis coli (APC), KRAS proto-oncogene, GTPase (KRAS), and B-Raf proto-oncogene, serine/threonine kinase (BRAF) mutations, particularly in microsatellite-stable tumors.⁹ However, comprehensive analyses of age-related molecular differences across large, diverse cohorts remain limited.

A more complete understanding of age-related genomic differences can have direct clinical implications by informing molecular testing; standard testing strategies may miss clinically relevant biology and lead to suboptimal therapies for younger patients.^6,8,9 To address this knowledge gap, we conducted a large-scale analysis of mutation frequencies across age-stratified cohorts using clinicogenomic data from internal cancer databases to identify how mutation patterns in the most prevalent CRC-associated genes vary by age at diagnosis, with a focus on identifying distinguishing molecular features of EOCRC that may inform diagnostic testing and therapeutic decision-making.

Patients and methods

Patients

We conducted a retrospective analysis using publicly available clinicogenomic data from the Memorial Sloan Kettering Center for Harmonized Onco-genomic Research Dataset (MSK-CHORD, Nature 2024) and China Pan-Cancer cohort, accessed via cBioPortal.¹⁰ MSK-CHORD is a harmonized dataset comprising demographic, clinical, and genomic data from over 24 000 patients with cancer, including 5543 pathology-confirmed CRC cases that has been validated for real-world oncological research and outcome prediction.¹¹ Our study used the harmonized clinicogenomic dataset rather than the CHORD homologous recombination deficiency classifier. The China Pan-Cancer cohort contains next-generation sequencing (NGS) data from multiple tertiary-care hospitals across China with pathology-confirmed diagnoses and has been validated in several large-scale genomic studies.¹² We selected these datasets because they provide high-quality, uniformly curated genomic and demographic data with patient-level age information, allowing us to conduct a decade-stratified analysis. Other CRC datasets available through cBioPortal were considered but were either substantially smaller, lacked sufficient numbers of young patients, or lacked necessary demographic data. Inclusion criteria were age 18 and above, a diagnosis of colorectal carcinoma as defined by the originating datasets and availability of genomic profiling data for mutation frequency analysis. Cases were excluded if they lacked somatic sequencing data, had incomplete or ambiguous diagnostic information, or missing demographic data. Cohorts were stratified across four age groups: 18-29, 30-39, 40-49, and ≥50. These age ranges were chosen to better characterize EOCRC by identifying potential decade-specific mutations.

Statistical analysis

Somatic mutation frequency data were obtained for the 100 most prevalent CRC-associated genes represented across the cohorts. For each gene, a contingency table was constructed for the number of mutated and nonmutated cases across the four age groups. Statistical significance of differences in mutation frequency by age was assessed using the chi-square test for independence (chi2_contingency, SciPy, SciPy Developers) in Python 3.11 (Python Software Foundation, Beaverton, OR). Multiple hypothesis correction was applied across all genes using the Benjamini–Hochberg false discovery rate (FDR) procedure (multipletests, statsmodels, statsmodels Developers) in Python 3.11 with an FDR-adjusted P value < 0.05 considered statistically significant. For genes with significant overall differences, post hoc pairwise comparisons between age groups were carried out with FDR correction.

Following peer review, cohort-specific sensitivity analyses of POLE mutation frequencies were carried out separately for the MSK-CHORD and China Pan-Cancer datasets using the same statistical approach as the primary analysis.

Oversight

This analysis was conducted exclusively using deidentified, publicly accessible datasets obtained through cBioPortal. In accordance with institutional policy, the use of such data meets criteria for exemption from Institutional Board Review oversight.

Results

Cohort characteristics

We included 6762 patients with CRC from the MSK-CHORD and China Pan-Cancer cohorts and stratified patients into four age groups: 18-29 years (n = 79, 1.2%), 30-39 years (n = 402, 5.9%), 40-49 years (n = 1064, 15.7%), and ≥50 years (n = 5217, 77.2%). Demographic and clinical features are summarized in Table 1. In the combined cohort, 55.1% of patients were male and 44.9% were female. Most cases (82%) derived from the MSK-CHORD dataset, whereas the remaining 18% came from the China Pan-Cancer cohort. Primary tumor sites were distributed across the right colon (24.2%), left colon (31.5%), and rectum (25.1%) with 19.2% unspecified. Stages I-III at diagnosis represented 57.5% of cases, whereas stage IV represented 41.5%.

Table 1.

Characteristics of patients with CRC stratified by age group

Characteristic	All patients (n = 6762)	18-29 (n = 79)	30-39 (n = 402)	40-49 (n = 1064)	≥50 (n = 5217)
Sex, n (%)
Male	3725 (55.1)	42 (53.2)	216 (53.7)	569 (53.5)	2898 (55.5)
Female	3037 (44.9)	37 (46.8)	186 (46.3)	495 (46.5)	2319 (44.5)
Cohort, n (%)
MSK-CHORD	5542 (82.0)	51 (64.6)	311 (77.4)	860 (80.8)	4318 (82.8)
China Pan-Cancer	1220 (18.0)	28 (35.4)	91 (22.6)	204 (19.2)	899 (17.2)
Primary tumor
Location, n (%)
Right colon	1639 (24.2)	15 (19.0)	58 (14.4)	164 (15.4)	1402 (26.9)
Left colon	2129 (31.5)	19 (24.1)	137 (34.1)	393 (36.9)	1580 (30.3)
Rectum	1694 (25.1)	15 (19.0)	113 (28.1)	294 (27.6)	1272 (24.4)
Unspecified	1301 (19.2)	30 (38.0)	94 (24.4)	214 (20.1)	963 (18.5)
Stage at diagnosis, n (%)
Stage I-III	3887 (57.5)	44 (55.7)	206 (51.2)	580 (54.5)	3057 (58.6)
Stage IV	2807 (41.5)	35 (44.3)	189 (47.0)	479 (45.0)	2104 (40.3)
Unspecified	68 (1.0)	0 (0)	7 (1.7)	5 (0.5)	56 (1.1)

Demographic and clinical characteristics of 6762 patients with CRC stratified by age group including sex distribution, cohort source, primary tumor location, and stage at diagnosis across four age categories (18-29, 30-39, 40-49, and ≥50 years).

CRC, colorectal cancer; MSK-CHORD, Memorial Sloan Kettering Center for Harmonized Onco-genomic Research Dataset.

Age-stratified mutation differences

Across age groups, we observed several distinct genomic differences between early-onset and late-onset CRCs. The prevalence of APC mutations increased progressively with age, occurring in 49.4% of patients aged 18-29, 69.7% in patients aged 30-39, 73.3% in patients aged 40-49, and 75.2% in patients aged ≥50 (P < 0.001, FDR < 0.001). Patients ≥50 years had more than three times higher odds of harboring an APC mutation compared with those aged 18-29 [odds ratio (OR) 3.74, 95% confidence interval (CI) 2.44-5.74, P < 0.001]. The youngest cohort showed the largest difference with a 25.9% lower APC mutation prevalence compared with individuals ≥50, whereas the 30-39 group showed a smaller yet significant reduction (−5.6%).

In contrast, SMAD family member 4 (SMAD4) mutations demonstrated the opposite trend and were twice as common in the youngest cohort, occurring in 31.6% of patients aged 18-29 compared with 18.8% in those aged ≥50 (OR 2.03, 95% CI 1.26-3.27, P < 0.001, FDR < 0.001). The 30-39 group demonstrated an intermediate prevalence (21.4%). We also observed that enrichment of SMAD4 in younger patients was accompanied by increased variation in other transforming growth factor-β (TGF-β) pathway genes, including ACVR2A and TGFBR2, although these did not reach statistical significance after FDR correction (Figure 1).

Figure 1.

Heatmap of differential mutation frequencies in early-onset colorectal cancer compared with patients aged ≥50 years. This plot displays the change (Δ) in mutation rates for significantly altered genes in patients aged 18-29 and 30-39 relative to the ≥50 reference group. Blue shading indicates lower mutation prevalence and red shading indicates higher prevalence compared with older adults. APC mutations were markedly less common in the youngest cohort (−25.9%), whereas SMAD4, POLE, and several chromatin-remodeling genes (e.g. KMT2A, SETD2, and NSD2) were enriched in early-onset cases.

Furthermore, we observed age-specific enrichment of DNA polymerase epsilon, catalytic subunit (POLE) mutations, which peaked in the 30-39 age group (10.7%) and declined in older patients (6.3% in 18-29, 4.9% in 40-49, and 5.9% in ≥50; P < 0.001, FDR < 0.001). Individuals aged 30-39 were nearly twice as likely to harbor a POLE mutation compared with those aged ≥40 (OR 1.96, 95% CI 1.41-2.74, P < 0.001, FDR < 0.001). To evaluate whether this finding was driven by either dataset alone, cohort-specific sensitivity analyses were carried out. In the China Pan-Cancer cohort, POLE mutations remained significantly enriched in patients aged 30-39 years (20.99% versus 7.14%, 8.89%, and 4.14% in the 18-29, 40-49, and ≥50 age groups, respectively; P < 0.001, FDR < 0.001). In the MSK-CHORD cohort, the highest POLE mutation frequency was likewise observed in the 30- to 39-year group (7.95% versus 5.88%, 4.08%, and 6.21%, respectively), although this trend did not reach statistical significance (P = 0.054, FDR = 0.628).

A volcano plot of effect size versus statistical significance (Figure 2) further illustrates these findings. Genes with the largest and most significant differences included decreased APC mutations (Δ −15.8%, FDR < 0.001) and enrichment of SMAD4, POLE, and multiple chromatin-remodeling genes. Together, these analyses demonstrate that early-onset CRC is characterized by a relative reduction in canonical APC-driven carcinogenesis and an increase in disruption of TGF-β signaling and DNA repair pathways.

Figure 2.

Volcano plot of mutation frequency differences between early-onset and late-onset CRC. The x-axis represents the absolute difference in mutation rates (Δ mutation rate, early-onset minus late-onset) and the y-axis shows statistical significance as −log₁₀(FDR). Each point corresponds to a gene; red points indicate genes with significant differences (FDR < 0.05), whereas gray points were not significant. The dashed horizontal line denotes the FDR threshold of 0.05. Notably, APC mutations were significantly depleted in early-onset patients, whereas SMAD4, POLE, and several chromatin-remodeling genes showed enrichment.

CRC, colorectal cancer; FDR, false discovery rate.

In contrast to the age-dependent alterations observed in the Wnt signaling pathway and TGF-β pathway, the mutation frequencies of TP53 and KRAS were not significantly associated with age at diagnosis. The differences in TP53 mutation rates across age groups ranged from 74.2% to 81.0% and were not statistically significant (P = 0.056). Similarly, the prevalence of KRAS mutations in early versus late-onset remained relatively constant at 44.0%-48.8% (P = 0.42).

Discussion

In recent years, rising incidence of younger CRC affecting individuals has prompted investigation into the unique clinical and genetic features and molecular underpinnings of early-onset disease. In our analysis of mutation profiles across two large clinicogenomic cohorts encompassing >6000 patients, we observed several age-associated genomic differences that highlight potential mechanisms of early tumorigenesis. Although the genomic landscape of patients under 40 demonstrated distinct features, the 40-49 age cohort notably did not show significant differences from the late-onset (≥50) group, suggesting that molecular drivers of tumorigenesis may follow a biological gradient.

APC mutations were substantially less frequent in the youngest cohort, indicating that canonical WNT pathway activation may play a less dominant role in tumorigenesis at younger ages, particularly in those under 30 years old.^6,7,9,13, 14, 15, 16 Reduced APC mutation prevalence has been consistently observed in prior EOCRC studies, raising the possibility that other precursor pathways may play a more prominent role in malignant transformation for younger patients.^7,9,13, 14, 15, 16 Therefore, targeted molecular testing for canonical CRC genes such as APC, KRAS, and BRAF is insufficient in this population and comprehensive NGS-based profiling is necessary to capture the full spectrum of driver mutations and pathway alterations.

In contrast, SMAD4 mutations were significantly enriched in younger patients and prior studies have similarly reported higher rates of SMAD4 alterations in EOCRC.^13,15,17,18 SMAD4 loss of function mutations promote dysregulated TGF-β signaling, which has been associated with metastatic potential, chemoresistance, and inferior prognosis.^17,18 This aligns with clinical observations of aggressive disease and advanced stage at diagnosis in younger age groups.^8,16,17,19 Therefore, SMAD4 status may play a role in risk stratification and this finding supports further investigation of TGF-β pathway-targeted strategies in EOCRC.20, 21, 22

Additionally, POLE mutations demonstrated a nonlinear distribution, peaking among patients in the 30-39 age group rather than the youngest cohort. The high mutation burden in tumors with POLE alterations and their robust response to immune checkpoint inhibitors (ICI) suggest that a subset of patients with EOCRC may have heightened immunotherapeutic sensitivity.23, 24, 25, 26 Given that POLE mutations can occur in microsatellite-stable tumors, assessment of POLE exonuclease-domain variants and tumor mutational burden may identify patients who could benefit from ICI beyond standard microsatellite instability (MSI) testing, particularly among younger individuals.23, 24, 25, 26, 27

Sensitivity analyses carried out separately in the two datasets showed that the enrichment observed in the 30-39 age group was statistically significant in the China Pan-Cancer cohort and directionally consistent, although not statistically significant in the MSK-CHORD cohort. These findings suggest that the observed age-associated pattern is reproducible across both datasets, although its magnitude differed and is less likely to be solely an artifact of combining the two datasets. Differences in cohort composition, sequencing platforms, patient population, and the relatively small number of patients aged 18-29 may have contributed to the observed heterogeneity. The nonlinear distribution may also reflect the biological heterogeneity of EOCRC rather than a uniform, age-dependent increase, supporting growing evidence suggesting that EOCRC comprises multiple molecular subtypes with distinct oncogenic mechanisms rather than a single disease entity.^15,28,29 Additional validation in independent cohorts will be important to determine whether this represents a reproducible biological feature of EOCRC.

Furthermore, enrichment of chromatin-remodeling gene alterations, such as Set Domain Containing 2 (SETD2) and nuclear receptor-binding SET domain protein 2 (NSD2), supports epigenetic and transcriptional dysregulation as potential mechanisms of carcinogenesis.^13,14,30, 31, 32, 33 Prior multiomic investigations have highlighted epigenetic shifts, altered DNA methylation trajectories, and stem-cell-related transcriptional programs in subsets of EOCRC, reinforcing the need to expand beyond canonical pathways in disease characterization among younger individuals.30, 31, 32, 33, 34 Collectively, these findings underscore the hypothesis that EOCRC comprises heterogenous molecular subtypes with distinct therapeutic vulnerabilities and that identification of noncanonical mutations may guide prognosis, treatment selection, and future screening strategies.^{3,4,8,13,14,25,31} We found this distinction to be most prominent in patients under 40, whereas patients in their 40s may represent a transitional phenotype more akin to traditional late-onset disease. As incidence continues to rise among younger adults, integration of age-informed genomics into risk-adapted screening and prevention frameworks while accounting for modifiable risk factors will be crucial.2, 3, 4, 5

Notably, we found no significant differences in the mutation rates of TP53 and KRAS across age-stratified cohorts, suggesting that dysregulation of p53 and RAS/MAPK pathways is a common oncogenic events central to CRC tumorigenesis regardless of age. Although these canonical driver mutations appear as shared molecular end states across all age groups, the routes by which tumors arrive at these end states may differ substantially in younger versus older adults. Therefore, EOCRC may be characterized by unique precursor lesions, epigenetics, or microenvironmental cues compared with average-onset colorectal cancer (AOCRC). These upstream differences can lead to distinct therapeutic vulnerabilities, treatment response profiles, and resistance mechanisms, remaining a potential avenue of therapeutic exploration.

Although this analysis benefits from two large clinicogenomic cohorts, limitations that may impact the interpretation of these data do exist. This includes potential selection, referral, and geographic biases as data from tertiary-care centers in two countries may not fully represent the demographic and environmental diversity of the general population. The relatively small size of the 18-29 age group also limits statistical power to detect rare but biologically relevant variants. It is also possible that our age cohort selections by decade mask key transition points between mutational genotypes that may occur in smaller intervals. Because this study relied on publicly available clinicogenomic datasets, comprehensive information regarding IBD and hereditary cancer syndromes was not uniformly available. Consequently, patients with known germline pathogenic variants or IBD-associated CRC could not be systematically excluded. However, hereditary syndromes account for only a minority of EOCRC cases and therefore are unlikely to fully explain the observed age-associated findings. This analysis is one of the first to attempt to analyze the mutational landscape by decade instead of the traditional binary EOCRC versus AOCRC. Prospective, multi-institutional, population-based studies with comprehensive exposure data will be critical to validate these observations, further characterize the oncogenic pathways driving EOCRC, and optimize treatment strategies.

Conclusion

Our decade-stratified analysis shows that while CRCs can ultimately converge on shared core oncogenic pathways, molecular routes to reach these endpoints differ by age and these differences are most distinct in patients under 40. These findings reinforce the emerging view of EOCRC as a heterogenous entity with unique biological drivers and therapeutic opportunities. As incidence continues to rise among younger adults, integrating age-informed genomic insights into screening, prevention, and treatment will be essential for improving outcomes and developing more biologically precise strategies for treating CRC across the age spectrum.