Mutational and Co-Mutational Landscape of Early Onset Colorectal Cancer

Abstract

Introduction:

Colorectal cancer (CRC) incidence and mortality before 50 have been rising alarmingly in the recent decades.

Methods:

Using a cohort of 10,000 patients, this study investigates the clinical, mutational, and co-mutational features of CRC in early-onset (EOCRC, < 50 years) compared to late-onset (LOCRC, ≥ 50 years).

Results:

EOCRC was associated with a higher prevalence of Asian and Hispanic patients, rectal or left-sided tumors (72% vs. 59%), and advanced-stage disease. Molecular analyses revealed differences in mutation patterns, with EOCRC having higher frequencies of TP53 (74% vs. 68%, P < 0.01) and SMAD4 (17% vs. 14%, P = 0.015), while BRAF (5% vs. 11%, P < 0.001) and NOTCH1 (2.7% vs. 4.1%, P = 0.01) mutations were more prevalent in LOCRC. Stratification by tumor site and MSI status highlighted significant location- and age-specific molecular differences, such as increased KRAS and CTNNB1 mutations in right-sided EOCRC and higher BRAF prevalence in MSI-H LOCRC (47% vs. 6.7%, P < 0.001). Additionally, co-occurrence analysis revealed unique mutational networks in EOCRC MSS, including significant co-occurrences of FBXW7 with NOTCH3, RB1, and PIK3R1.

Conclusion:

This study highlights the significance of age-specific molecular profiling, offering insights into the unique biology of EOCRC and potential clinical applications.

INTRODUCTION

Colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second leading cause of cancer-related death worldwide, imposing significant morbidity and mortality on affected individuals ^1,2. Traditionally regarded as a disease of older adults, the incidence of CRC among populations over 50 years of age has been declining, largely due to widespread implementation of screening programs and increased awareness ³. In contrast, early-onset colorectal cancer (EOCRC), defined as CRC diagnosed in individuals under 50 years of age, has been rising at an alarming rate ^4–7. The rising incidence of EOCRC presents a significant public health concern ^8,9. The incidence of new CRC cases in this age demographic nearly doubled from 1994 to 2019, increasing from 11% to 20%; furthermore, EOCRC is now believed to account for 12% of all new CRC diagnoses ^3,8,10,11. Conversely, late-onset colorectal cancer (LOCRC, age ≥ 50 years) has been in decline. For example, CRC diagnosed in individuals over 60 years of age has demonstrated an average annual decline of roughly 4% ^3,12,13. Recent reports indicate that CRC is now the primary cause of cancer death among young males aged 20 to 49 years and the second leading cause for females, following breast cancer ^3,4,7. This pattern is not confined to the United States and has been thoroughly documented in over 20 countries worldwide, lacking a clearly defined epidemiological or etiological cause ^14–16. Although older patients are more prone to severe postoperative complications, there is no consensus that age directly impacts survival outcomes ¹⁷. Prognosis in elderly individuals is often influenced by factors such as stage at diagnosis, tumor location, comorbidities, and treatment modalities. In contrast, EOCRC patients, despite undergoing more aggressive treatment regimens, have been reported to exhibit lower response rates and poorer survival outcomes ^18–23. The U.S. Preventive Services Task Force and the American Cancer Society updated their guidelines to recommend CRC screening beginning at age 45 rather than 50 in response to the rising incidence of EOCRC ²⁴.

Early-onset colorectal cancers exhibit heterogeneous somatic and molecular characteristics ^9,25. Approximately 10–16% of EOCRC cases result from germline mutations, while the majority are sporadic with no identified etiology ^10,25–27. The primary characteristics of EOCRC tumors encompass left-sided origins, an elevated probability of lymph node involvement and metastasis, and a higher prevalence of poor or undifferentiated histology ^25,26,28,29. No molecular mechanism or risk factor for sporadic EOCRC is currently clearly defined ⁹.

Driver mutations and genetic alterations are fundamental to the pathophysiology of CRC, providing tumor cells with a growth advantage crucial for cancer development and dissemination ^30–32. Prior research revealed that EOCRC tumors have a higher frequency of changes in the TP53, PTEN, ATM, and CTNNB1 genes and a lower frequency of APC, BRAF, KRAS, and NRAS mutations ^25,26,28. However, other studies demonstrated a conflicting prevalence of CRC driver genes in EOCRC ^33–36. These variations arise from differences in study design and the inability to exclude patients that skewed the results, such as those with inherited CRC, MSI status or varying age cutoffs for age groups.

Further research is therefore needed to better define these molecular features, as identifying the spectrum of driver mutations in EOCRC could reveal novel biomarkers for early detection, insights into potential etiologies, and therapeutic targets for personalized medicine. Moreover, comparing the mutation landscapes between EOCRC and LOCRC may uncover age-related molecular differences that could inform tailored screening strategies and interventions.

In this study, we conducted a retrospective analysis of CRC patients diagnosed at MD Anderson Cancer Center (MD Anderson) with available genomic data. We identified 10,000 cases between 2003 and 2023 and performed molecular characterization to examine the molecular spectrum between EOCRC and LOCRC.

METHODS

Patient Selection:

In this study, we included patients with a confirmed diagnosis of colorectal adenocarcinoma who underwent somatic mutation testing on tumor tissue at MD Anderson Cancer Center from November 1, 2003, to November 1, 2023. The software system of Palantir Foundry (Palantir Technologies, Denver, CO) has been used to extract patient information from the longitudinal electronic health records (EHR) and tumor registry. The pathology report of each patient was used to gather histology, tumor grade, and pathological stage. We included only patients with demographic, histopathological, and molecular data available. Patients younger than 18 years old at time of diagnosis were excluded from this study (Supplementary Figure 1). Based on age at time of diagnosis, patients were then divided into two groups: those under 50 years were labeled as EOCRC while those 50 years or over as LOCRC. The tumor location, stage, and grade at diagnosis were obtained from clinical notes in the EHR and the MD Anderson Cancer Registry. Tumor sites were reclassified according to ICD-10 diagnostic codes into right, left, rectal and overlapping categories ³⁷. The Institutional Review Board (IRB) of MD Anderson Cancer Center approved the research in compliance with IRB protocol (09–0373). A waiver of informed consent was approved in compliance with US federal regulation 45 CFR § 46.116 (Common Rule) due to the study’s minimal risk to participants.

Molecular Testing:

Targeted clinical genetic testing for CRC was performed at MD Anderson’s molecular diagnostics laboratory, accredited by the College of American Pathologists (CAP) and certified by the Clinical Laboratory Improvement Amendments (CLIA). We used multiple clinically validated institutional multigene next-generation sequencing (NGS) panels and only included somatic gene variations in our analysis ³⁸. Matching tumor-normal sequencing or the publicly accessible dbSNP and COSMIC databases were used to eliminate germline variants ^39,40. Following the classification of all genetic alterations using a five-tier system, only variants classified as pathogenic or likely pathogenic were included ^41,42. Both the total number of genes tested and the number of hotspots for each gene on the NGS panels grew over time ³⁸. The microsatellite instability (MSI) status has been assessed through PCR, mismatch repair (MMR) protein expression via IHC, or NGS testing ^42,43. Patients with identified germline variants in CRC driver genes were excluded from somatic profiling to focus on tumor-specific somatic mutations, as germline alterations may confound the analysis by introducing inherited variants that do not reflect the somatic mutational landscape.

Statistical Analysis:

Frequency was used to summarize the baseline clinicopathological characteristics and molecular characteristics of the patient population. All statistical analyses were performed with R v 4.2.2 (https://www.r-project.org/)⁴⁴. Continuous variables were analyzed using the Wilcoxon rank test or, when applicable, the student’s t test, while categorical data were analyzed using Fisher’s exact test. Unless otherwise specified, all statistical tests were two-sided, and the significance threshold was set at P < 0.05. The P values were adjusted for multiple testing using the Benjamini-Hochberg method, with a q-value cutoff of 0.1 ⁴⁵. To generate data visualizations, the R packages annotate (version 3.5.1), ggrepel (version 0.9.6), ggplot2 (version 3.5.1), and treemap (version 2.4) were utilized.

RESULTS

Overview of the study’s cohort characteristics

A total of 10,000 patients were included in this cohort, with histologically confirmed colorectal carcinomas. The majority were men (57%), and 70% identified as White or Caucasian. The median age at diagnosis was 57 years (IQR: 48–65), and most patients (89%) presented with advanced-stage. All included patients had structured histological data available. Histopathological profiles indicated that 98% of tumors were adenocarcinomas, whereas 8.2% were mucinous tumors as defined by the standard histopathological diagnostic criteria (> 50% extracellular mucin of the tumor). Most of them were moderately differentiated (76%). The majority of patients had left-sided or rectal primary tumors (56%). Microsatellite instability (MSI) status was available for more than 75% of the patients, of whom the vast majority (91%) were microsatellite stable (MSS) or mismatch repair proficient (pMMR). Detailed baseline characteristics of the study population are outlined in Table 1 and the selection of the study cohort in flowchart in Supplementary Figure 1.

Table 1:

Patient characteristics for the entire cohort and by age group:

		Age Group
Characteristic	Overall N = 10,0001	EOCRC1	LOCRC1	p-value2
		< 50 years	≥ 50 years
Patient Age at Diagnosis	57 (48, 65)	44 (38, 47)	62 (56, 68)
Sex				<0.001
Female	4,300 (43%)	1,490 (47%)	2,810 (41%)
Male	5,700 (57%)	1,695 (53%)	4,005 (59%)
Race / Ethnicity				<0.001
White or Caucasian	6,968 (70%)	2,107 (66%)	4,861 (71%)
Asian	499 (5.0%)	212 (6.7%)	287 (4.2%)
Black or African American	1,015 (10%)	323 (10%)	692 (10%)
Hispanic	1,229 (12%)	441 (14%)	788 (12%)
Other	156 (1.6%)	56 (1.8%)	100 (1.5%)
Unknown	133 (1.3%)	46 (1.4%)	87 (1.3%)
Histology				<0.001
Adenocarcinoma	6,431 (64%)	1,994 (63%)	4,437 (65%)
Adenocarcinoma-mucinous	820 (8.2%)	248 (7.8%)	572 (8.4%)
Adenocarcinoma-signet	280 (2.8%)	118 (3.7%)	162 (2.4%)
Carcinoma	124 (1.2%)	15 (0.5%)	109 (1.6%)
Unknown	2,345 (23%)	810 (25%)	1,535 (23%)
Tumor Site				<0.001
Left	3,875 (39%)	1,409 (44%)	2,466 (36%)
Rectal	2,431 (24%)	878 (28%)	1,553 (23%)
Right	3,154 (32%)	755 (24%)	2,399 (35%)
Overlapping	540 (5.4%)	143 (4.5%)	397 (5.8%)
Microsatellite Status				<0.001
MSI-H	654 (6.5%)	181 (5.7%)	473 (6.9%)
MSS	6,939 (69%)	2,538 (80%)	4,401 (65%)
Unknown	2,407 (24%)	466 (15%)	1,941 (28%)
Stage at Evaluation				<0.001
Stage I-II	1,050 (11%)	211 (7.2%)	839 (13%)
Stage III	1,867 (20%)	575 (20%)	1,292 (20%)
Stage IV	6,356 (69%)	2,161 (73%)	4,195 (66%)
Unknown	727	238	489
Grade at Evaluation				0.36
Well Differentiated	84 (0.9%)	23 (0.7%)	61 (0.9%)
Moderately Differentiated	7,272 (76%)	2,378 (77%)	4,894 (75%)
Poorly Differentiated	2,231 (23%)	703 (23%)	1,528 (24%)
Unknown	413	81	332

¹Median (Q1, Q3); n (%)

²Wilcoxon rank sum test; Pearson’s Chi-squared test

Differences in cohort characteristics between the age groups

When we stratified patients by age at time of diagnosis, EOCRC accounted for 32% (N = 3,185) of the cohort, with a median age of 44 years (IQR: 38–47). In contrast, LOCRC had a median age of 62 years (IQR: 56–68) (N = 6,815).

Racial and ethnic distributions also differed significantly between EOCRC and LOCRC groups, with EOCRC having a higher proportion of Asian and Hispanic patients (P < 0.001). Female patients were disproportionately represented in the EOCRC group, while male patients were more likely to have LOCRC (47% vs. 41%; P < 0.001). In addition, the majority of EOCRC were found in the rectum or left colon (72% vs. 59%; P < 0.001), were moderately differentiated, and were often diagnosed at stage IV, which is consistent with the MD Anderson patient population and CRC patients undergoing NGS (73% vs. 66%; P < 0.001). The pathological and clinical differences observed highlight distinct characteristics between EOCRC and LOCRC patients (Table 1).

Differences in the molecular landscape between the age groups

Comparing the frequency of somatic mutations in EOCRC and LOCRC across a number of CRC genes showed that the two subgroups had modestly different patterns of mutation (Figure 1, Figure 2, and Table 2). The overall prevalence of mutations in key CRC driver genes, such as TP53, APC, KRAS, and BRAF, was largely consistent with their known frequency (Figure 1A). The most common CRC driver genes in EOCRC were TP53 (74%), APC (61%), KRAS (48%), PIK3CA (18%) and BRAF (5%). However, there was some variation in the frequency of the same genes found in LOCRC tumors, which were 68%, 65%, 46%, 20%, and 11%, respectively.

Figure 1: Comparison of mutation prevalence between early-onset colorectal cancer (EOCRC) and late-onset colorectal cancer (LOCRC).

The scatter plots display the prevalence of colorectal cancer somatic mutations in EOCRC versus LOCRC. Each point represents a gene. The identity line (y = x) is represented by the dashed blue diagonal line, which indicates that EOCRC and LOCRC are similarly prevalent. Genes above the line are more prevalent in EOCRC, while genes below the line are more prevalent in LOCRC. The x-axis and y-axis represent mutation prevalence percentages for EOCRC and LOCRC, respectively. Legend categories represent mutation group classifications by microsatellite instability (MSI) status in (B) and tumor site in (C).

Figure 2: Difference in mutation prevalence between early-onset colorectal cancer (EOCRC) and late-onset colorectal cancer (LOCRC).

A. The treemap shows the difference in mutation prevalence for various genes, comparing EOCRC to LOCRC. Each rectangle represents a gene, with its size proportional to its overall mutation prevalence across both groups. Colors indicate the direction and magnitude of the prevalence difference: green shades represent higher prevalence in EOCRC, while red shades represent higher prevalence in LOCRC. The intensity of the color corresponds to the magnitude of the difference, as indicated by the color scale below the plot.

B, C and D. The bar chart illustrates the difference in mutation prevalence for selected genes between EOCRC and LOCRC groups. Positive values indicate higher prevalence in EOCRC, while negative values indicate higher prevalence in LOCRC. Error bars represent the standard error (SE) for the differences in prevalence. Arrows on the right highlight the directionality: “Higher in EOCRC” for positive differences and “Higher in LOCRC” for negative differences. Genes with an absolute prevalence difference of more than 1% in the entire cohort are displayed in B. Figure C illustrates the prevalence results of the same genes for MSS tumors, whereas Figure D depicts the results for MSI-H tumors.

Table 2.

Genes prevalence summary.

All Tumors				Microsatellite Stable (MSS) Tumors			Microsatellite Instable (MSI) Tumors
	EOCRC1 N= 2,895	LOCRC1 N = 6,137	p-value2	EOCRC1 N= 2,302	LOCRC1 N = 3,934	p-value2	EOCRC1 N= 167	LOCRC1 N = 448	p-value2
APC			0.52			<0.001			0.003
Mutation	870 (64%)	1,725 (65%)		732 (65%)	1,400 (71%)		36 (69%)	39 (43%)
Wildtype	497 (36%)	942 (35%)		389 (35%)	561 (29%)		16 (31%)	51 (57%)
ARID1A			0.36			0.92			0.94
Mutation	78 (11%)	176 (12%)		57 (8.6%)	106 (8.7%)		18 (42%)	51 (43%)
Wildtype	663 (89%)	1,312 (88%)		609 (91%)	1,114 (91%)		25 (58%)	69 (58%)
ATM			0.24			0.57			0.10
Mutation	126 (7.4%)	284 (8.3%)		101 (7.1%)	191 (7.6%)		13 (15%)	44 (24%)
Wildtype	1,588 (93%)	3,140 (92%)		1,326 (93%)	2,331 (92%)		71 (85%)	136 (76%)
BRAF			<0.001			<0.001			<0.001
Mutation	145 (5.3%)	642 (11%)		114 (5.2%)	308 (8.2%)		11 (6.7%)	206 (47%)
Wildtype	2,600 (95%)	5,040 (89%)		2,085 (95%)	3,443 (92%)		152 (93%)	229 (53%)
CDKN2A			0.061			0.38			0.28
Mutation	18 (1.1%)	58 (1.8%)		15 (1.1%)	34 (1.4%)		1 (1.2%)	7 (4.1%)
Wildtype	1,664 (99%)	3,238 (98%)		1,385 (99%)	2,392 (99%)		82 (99%)	163 (96%)
EGFR			0.29			0.24			0.22
Mutation	25 (1.4%)	64 (1.8%)		16 (1.1%)	39 (1.6%)		8 (9.2%)	9 (5.2%)
Wildtype	1,752 (99%)	3,495 (98%)		1,428 (99%)	2,463 (98%)		79 (91%)	164 (95%)
FBXW7			0.96			0.28			0.58
Mutation	221 (12%)	436 (13%)		171 (12%)	326 (13%)		25 (29%)	59 (32%)
Wildtype	1,548 (88%)	3,039 (87%)		1,297 (88%)	2,218 (87%)		62 (71%)	125 (68%)
FGFR3			0.43			0.22			0.51
Mutation	17 (1.0%)	42 (1.3%)		8 (0.6%)	23 (0.9%)		5 (5.8%)	14 (8.1%)
Wildtype	1,671 (99%)	3,290 (99%)		1,393 (99%)	2,425 (99%)		81 (94%)	159 (92%)
GNAS			0.10			0.057			0.54
Mutation	37 (2.1%)	103 (2.8%)		25 (1.7%)	68 (2.6%)		7 (8.1%)	20 (11%)
Wildtype	1,767 (98%)	3,588 (97%)		1,465 (98%)	2,551 (97%)		79 (92%)	170 (89%)
KIT			>0.99			0.22			0.93
Mutation	62 (3.5%)	126 (3.5%)		44 (3.1%)	61 (2.4%)		7 (8.0%)	15 (8.4%)
Wildtype	1,692 (96%)	3,444 (96%)		1,387 (97%)	2,452 (98%)		80 (92%)	164 (92%)
KRAS			0.11			0.39			<0.001
Mutation	1,297 (48%)	2,600 (46%)		1,051 (48%)	1,856 (49%)		60 (48%)	73 (28%)
Wildtype	1,405 (52%)	3,034 (54%)		1,150 (52%)	1,940 (51%)		64 (52%)	191 (72%)
MET			0.71			0.13			0.35
Mutation	46 (2.5%)	87 (2.4%)		33 (2.2%)	40 (1.5%)		2 (2.3%)	10 (5.5%)
Wildtype	1,771 (97%)	3,587 (98%)		1,462 (98%)	2,543 (98%)		84 (98%)	172 (95%)
MPL			0.36			0.72			0.54
Mutation	2 (0.2%)	9 (0.4%)		2 (0.3%)	6 (0.4%)		0 (0%)	2 (2.4%)
Wildtype	1,006 (100%)	2,030 (100%)		784 (100%)	1,329 (100%)		47 (100%)	82 (98%)
NOTCH1			0.010			0.031			0.068
Mutation	46 (2.7%)	142 (4.1%)		26 (1.8%)	74 (3.0%)		15 (17%)	53 (28%)
Wildtype	1,670 (97%)	3,312 (96%)		1,396 (98%)	2,427 (97%)		71 (83%)	139 (72%)
NRAS			0.068			0.018			0.060
Mutation	95 (4.2%)	241 (5.2%)		76 (4.0%)	185 (5.5%)		5 (5.0%)	3 (1.3%)
Wildtype	2,165 (96%)	4,383 (95%)		1,820 (96%)	3,187 (95%)		96 (95%)	227 (99%)
PIK3CA			0.12			0.012			0.10
Mutation	411 (18%)	893 (20%)		332 (18%)	660 (21%)		38 (39%)	64 (29%)
Wildtype	1,837 (82%)	3,601 (80%)		1,509 (82%)	2,484 (79%)		60 (61%)	154 (71%)
RB1			0.80			0.69			0.22
Mutation	30 (1.8%)	56 (1.7%)		23 (1.6%)	36 (1.5%)		3 (3.6%)	13 (7.6%)
Wildtype	1,662 (98%)	3,284 (98%)		1,385 (98%)	2,417 (99%)		80 (96%)	158 (92%)
RET			0.18			0.25			0.45
Mutation	19 (1.1%)	55 (1.5%)		15 (1.0%)	37 (1.4%)		4 (4.6%)	13 (7.0%)
Wildtype	1,778 (99%)	3,587 (98%)		1,474 (99%)	2,562 (99%)		83 (95%)	174 (93%)
RNF43			0.049			0.58			0.014
Mutation	42 (5.8%)	118 (8.1%)		34 (5.2%)	70 (5.8%)		7 (18%)	45 (39%)
Wildtype	684 (94%)	1,336 (92%)		619 (95%)	1,130 (94%)		33 (83%)	71 (61%)
SMAD4			0.015			0.12			0.92
Mutation	297 (17%)	492 (14%)		260 (18%)	403 (16%)		10 (12%)	21 (12%)
Wildtype	1,474 (83%)	2,965 (86%)		1,224 (82%)	2,170 (84%)		74 (88%)	149 (88%)
SOX9
Mutation	63 (100%)	104 (100%)		56 (100%)	91 (100%)		4 (100%)	11 (100%)
STK11			0.88			0.26			0.047
Mutation	19 (1.1%)	39 (1.2%)		8 (0.6%)	22 (0.9%)		7 (8.3%)	4 (2.4%)
Wildtype	1,670 (99%)	3,291 (99%)		1,396 (99%)	2,427 (99%)		77 (92%)	162 (98%)
TP53			<0.001			0.001			0.094
Mutation	1,568 (75%)	2,848 (69%)		1,376 (78%)	2,291 (74%)		35 (40%)	58 (30%)
Wildtype	524 (25%)	1,289 (31%)		390 (22%)	816 (26%)		52 (60%)	135 (70%)
BRCA2			0.95			0.46			0.41
Mutation	67 (6.4%)	129 (6.3%)		53 (5.8%)	85 (5.1%)		11 (18%)	31 (23%)
Wildtype	987 (94%)	1,920 (94%)		865 (94%)	1,585 (95%)		50 (82%)	102 (77%)
CREBBP			0.31			0.28			0.28
Mutation	58 (7.8%)	101 (6.6%)		37 (5.6%)	55 (4.5%)		16 (36%)	32 (28%)
Wildtype	686 (92%)	1,423 (93%)		624 (94%)	1,176 (96%)		28 (64%)	84 (72%)
CTNNB1			0.78			0.76			0.53
Mutation	47 (2.8%)	97 (2.9%)		35 (2.5%)	65 (2.6%)		10 (12%)	25 (14%)
Wildtype	1,658 (97%)	3,257 (97%)		1,381 (98%)	2,406 (97%)		76 (88%)	148 (86%)
ERBB2			0.31			0.050			0.24
Mutation	37 (2.2%)	89 (2.6%)		21 (1.5%)	60 (2.4%)		12 (14%)	16 (9.2%)
Wildtype	1,677 (98%)	3,306 (97%)		1,404 (99%)	2,442 (98%)		74 (86%)	158 (91%)
ERBB4			0.97			0.72			0.24
Mutation	39 (2.3%)	77 (2.3%)		32 (2.2%)	60 (2.4%)		7 (8.2%)	7 (4.1%)
Wildtype	1,672 (98%)	3,275 (98%)		1,393 (98%)	2,414 (98%)		78 (92%)	162 (96%)
JAK3			0.53			0.77			0.51
Mutation	19 (1.1%)	45 (1.3%)		14 (1.0%)	27 (1.1%)		2 (2.4%)	8 (4.5%)
Wildtype	1,687 (99%)	3,362 (99%)		1,401 (99%)	2,458 (99%)		83 (98%)	171 (96%)
LRP1B			0.44			0.66			>0.99
Mutation	65 (98%)	135 (95%)		54 (98%)	103 (95%)		5 (100%)	25 (93%)
Wildtype	1 (1.5%)	7 (4.9%)		1 (1.8%)	5 (4.6%)		0 (0%)	2 (7.4%)
NF1			0.14			0.38			0.045
Mutation	65 (6.1%)	103 (4.9%)		40 (4.4%)	62 (3.7%)		22 (35%)	30 (22%)
Wildtype	996 (94%)	2,008 (95%)		878 (96%)	1,629 (96%)		41 (65%)	109 (78%)
NOTCH3			0.47			0.68			0.87
Mutation	49 (6.8%)	110 (7.6%)		32 (4.9%)	64 (5.4%)		14 (34%)	36 (33%)
Wildtype	673 (93%)	1,329 (92%)		617 (95%)	1,127 (95%)		27 (66%)	74 (67%)
PIK3R1			0.46			0.32			0.57
Mutation	56 (5.3%)	97 (4.7%)		43 (4.7%)	65 (3.9%)		10 (17%)	17 (13%)
Wildtype	991 (95%)	1,947 (95%)		867 (95%)	1,602 (96%)		50 (83%)	109 (87%)
PTEN			0.57			0.26			0.50
Mutation	83 (4.8%)	179 (5.1%)		64 (4.5%)	133 (5.3%)		15 (17%)	26 (14%)
Wildtype	1,652 (95%)	3,298 (95%)		1,371 (96%)	2,390 (95%)		71 (83%)	156 (86%)
TCF7L2
Mutation	50 (100%)	96 (100%)		44 (100%)	77 (100%)		5 (100%)	16 (100%)

¹Median (IQR) or Frequency (%);

²Pearson’s Chi-squared test; Fisher’s exact test

Among the genes with notable differences in prevalence, TP53 (75% vs. 69%; P < 0.01) and SMAD4 (17% vs. 14%; P = 0.015) had a significantly greater frequency in EOCRC as compared to LOCRC (Figure 1A, Figure 2A, Figure 2B and Table 2). In contrast, mutations in BRAF (5% vs. 11%; P < 0.001) and NOTCH1 (2.7% vs. 4.1%; P = 0.01) were significantly more prevalent in LOCRC (Table 2).

Differences in the molecular landscape between age groups and across different tumor sites

Primary tumor location is associated with specific molecular features in CRC; therefore, we also investigated whether these were emerging across age groups. Indeed, we found that there were significant differences in the mutational landscape according to age of diagnosis and tumor site, highlighting location-specific differences. For instance, in rectal tumors, prevalence of TP53 mutations was significantly higher in early vs. late onset (79% vs. 73%; P = 0.01). On the other hand, early-onset right-sided tumors were characterized by a significantly lower prevalence of BRAF (8% vs. 20%; P < 0.001) mutations and a significantly higher prevalence of KRAS (64.5% vs. 53%; P < 0.001) and CTNNB1 (7% vs. 4%; P < 0.001).

Differences in the molecular landscape across age of onset deciles

Examining these genes’ prevalence across age deciles, further revealed the anticipated variation across them (Figure 3). The Wnt/β-catenin signaling pathway genes, including APC and CTNNB1 exhibited varying mutation prevalence across age deciles. The APC gene showed the highest prevalence, starting at 44% in individuals in their 20s, peaking at 64% in the 40s, and gradually declining to 50% in the 90s. In the 30s decile of EOCRC, prevalence of CTNNB1 mutations was slightly higher at 3.6%, while in the 40s to 60s, the rate remained steady at about 2%. When examining the genes involved in the TP53 signaling pathway, TP53 mutations were most common in younger age groups, with rates exceeding 70% in the 30s to 50s.

Figure 3: Somatic Gene Mutation Prevalence by Age in Patients with Colorectal Cancer.

The line plots represent the prevalence of mutations in APC, TP53, KRAS, BRAF, RNF43, and CTNNB1 genes across different age deciles. The x-axis represents age deciles (20s to 90s), while the y-axis indicates the mutation prevalence percentage. Each line shows the gene prevalence for each decile, with the shaded region representing the 95% confidence interval.

The RTK-RAS pathway, represented by BRAF, KRAS, NRAS, EGFR, ERBB2, ERBB4, FGFR3, and MET, demonstrated distinct age-related mutation prevalence patterns. The highest frequency was found in KRAS mutations, which were consistently between 38% and 48% across all age deciles. The prevalence of BRAF mutations significantly increased with age, peaking at 10% in the 60s after beginning at about 4% in the 20s and 30s and increasing to 6% in the 40s. NRAS mutations are uncommon, with a prevalence consistently ranging from 4% to 5% in patients < 70 years old. Similarly, EGFR, ERBB2, and ERBB4 mutations were infrequent in younger groups (<2%) but showed a modest increase in prevalence to 3–4% in individuals over 60 years. The frequency of FGFR3 and MET mutations was also low, at less than 1% across all deciles. These results indicate that KRAS and BRAF are essential in the RTK-RAS pathway, particularly in CRC among various age-of-onset groups (Figure 3 and Supplementary Table 1).

Differences by microsatellite instability (MSI)

Recognizing that MSI-High (MSI-H) tumors are hypermutated and that this introduces bias into genomic studies, we aimed to further investigate the comparison between EOCRC and LOCRC among MSI-H and MSS tumors separately. Subsequent analysis of genes that differed significantly between EOCRC and LOCRC revealed a correlation with MSI status.

Several genes display strikingly opposite prevalence trends between MSI-H and MSS tumors. Although the overall APC somatic rate did not differ significantly between the two age groups, the comparison of EOCRC and LOCRC across MSS and MSI-H tumors did, with MSS EOCRC having a lower prevalence (65% vs. 71%; P < 0.001) while MSI-H EOCRC (69% vs. 43%; P < 0.001) showed a higher prevalence than MSI-H LOCRC. Similar inverse patterns were notable for the genes KRAS, SMAD4, PIK3CA, and CREBBP (Figure 2C, Figure 2D, Table 2).

The MSI status had a significant impact on BRAF prevalence in LOCRC. While BRAF mutations were significantly more common in LOCRC MSS cases (5.2% EOCRC vs. 8.2% LOCRC; P < 0.001), they were much more common in EOCRC than LOCRC in MSI-H cases (6.7% vs. 47%; P < 0.001) with a 38.8% rise in prevalence in MSI-H LOCRC (P < 0.001), reflecting the epigenetic silencing of MLH1 related to sporadic MSI-H CRC tumors in the LOCRC and the hereditary component for the EOCRC (Figure 2C, Figure 2D, Table 2).

Among other genes, RNF43 alterations were more prevalent in EO compared to LO CRC (5.8% vs 8.1%, P = 0.049) in the entire cohort; however, this difference was confirmed only for MSI-H tumors (18% vs 39%, P = 0.014) and not for MSS. Moreover, NF1 showed a higher prevalence of mutations in EO MSI-H tumors (35% vs 22%, P = 0.045) but the difference did not emerge in the MSS group.

Co-occurrence of mutations by microsatellite instability (MSI)

For MSI-H and MSS tumors, we examined the pattern of mutational co-occurrence separately for each age group.

In MSS EOCRC, several gene pairs showed significant co-occurrence that appeared to be absent in LOCRC, highlighting unique mutational interactions in this subgroup (Figure 4A). Among them, FBXW7 mutations co-occurred more frequently with those of NOTCH3, RB1, and PIK3R1 (FDR-adjusted P < 0.01), suggesting their potential involvement in the carcinogenesis of MSS EOCRC. The distinct mutational networks in MSS EOCRC are further highlighted by the significant co-occurrence of ATM mutations with those of other specific genes (i.e., NOTCH3, EGFR, ERBB4, CREBBP, FGFR3, RNF43, NOTCH1 and RET; FDR-adjusted P < 0.03), further emphasizing distinct mutational networks in MSS EOCRC. KRAS mutations exhibited co-occurrence with SOX9, and NF1 mutations, which are strongly linked with both NOTCH3 and RB1 (Figure 4A).

Figure 4: Co-occurrence of mutations in early-onset colorectal cancer (EOCRC) and late-onset colorectal cancer (LOCRC) by microsatellite instability (MSI) status.

A) is for MSS and B) is for MSI-H tumors. The heatmap compares the co-mutation patterns between EOCRC (upper triangle) and LOCRC (lower triangle). Each cell represents the log2-transformed co-occurrence of mutations between two genes. Darker shades of red indicate higher co-occurrence, while lighter shades of blue indicate lower co-occurrence. Significant co-occurrences, based on Fisher’s exact test, are marked with symbols: “*” (FDR-adjusted P < 0.05) and “.” (FDR-adjusted P < 0.1). Genes are arranged alphabetically along both axes for consistency. The diagonal line separates the separation of plot between the two age groups, highlighting distinct co-mutation patterns.

The following co-occurred significantly in MSS LOCRC: BRAF with PTEN and RNF43 (FDR-adjusted P<0.03), APC with NOTCH1 and ARID1A (FDR-adjusted P<0.05), MET with FBXW7 and STK11 (FDR-adjusted P<0.05) (Figure 4A).

As previously mentioned, BRAF mutations are commonly seen in MSI-H LOCRC; an FDR-adjusted P <0.02 was found for co-occurring of BRAF with several genes, including PIK3CA, ERBB2, CTNNB1, FGFR3, and LRP1B in MSI-H LOCRC tumors. Interestingly, ARID1A mutations co-occurred with those of CREBBP, NOTCH3, CTNNB1, BRACA2, TCF7L2, and LRP1B in MSI-H LOCRC (FDR-adjusted P<0.001). No significant gene mutations’ co-occurrences were found in the MSI-H EOCRC subgroup (Figure 4B).

DISCUSSION

Over the past few decades, the incidence and death rate of EOCRC have been steadily rising, whereas the morbidity and mortality rate for CRC in people over 50 has been declining ^1,2. Although some previous studies have characterized the clinical and molecular profiles of EOCRC using large cohorts of CRC patients, including a study previously published by our group²⁵, the majority of them have included fewer than 500 EOCRC cases ^{10,25,33,46,47}. Here, we analyzed the molecular and clinicopathological characteristics of 10,000 CRC patients, revealing distinct features between 3,185 EOCRCs and 6,815 LOCRCs.

Recent findings suggest that EOCRCs have distinct clinicopathological profiles compared to LOCRCs ^8–10,25,26. The majority of EOCRCs were located in the rectum or left colon, consistent with prior clinical profiling and epidemiological studies of this subgroup ^{2,10,25,26,29,34}. Further consistent with previously reported observations, the prevalence of EOCRC patients was higher at stage IV, which could be attributed to observational bias of delayed diagnosis and lack of screening for this age group ^{2,8–10,25,26}. While some studies suggest a higher rate of poorly differentiated tumors in EOCRC, we observed that moderately differentiated histology was the predominant presentation (77%), with no significant difference compared to LOCRC (75%) ^48,49. Similar findings to our observations were reported by Cercek et al ³⁴.

We observed significant disparities in racial and ethnic prevalence, with a greater incidence of EOCRC among Asian and Hispanic individuals. Our results are consistent with Siegel et al.’s report on CRC statistics showing that the Hispanic population’s incidence of EOCRC was increasing at the highest rate in ³. Nonetheless, these findings may be affected by referral biases and demographic factors, as the risk of EOCRC is impacted by the birth cohort effects, with Hispanic and Asian populations in the United States exhibiting a notably younger median age ^3,50.

In EOCRC cases, germline mutations constitute approximately 14%, while the remaining 86% are sporadic ^51–56. In this study, we excluded patients with identified germline mutations and focused on somatic variants according to applied NGS assays. However, it is important to emphasize that genetic counseling, possibly extended to the entire family, is crucial for identifying potential inherited syndromes, paving the way for families to get enrolled into prevention programs.

Distinct mutational distributions have been observed in EOCRC ^25,26,46,57. Lieu et al. compared very early-onset cases (< 40 years), an intermediate-aged group (40–49 years), and an older-onset group (≥ 50 years) and reported consistent results with our observation of a higher prevalence of TP53 and a lower prevalence of APC, BRAF and KRAS in MSS tumors ²⁶. Our study supports these findings by including a larger number of individuals in the larger early-onset group (< 50 years old). However, other studies have shown a contradictory frequency of mutations, such as a higher prevalence of KRAS and APC ^33,34.

These findings highlight discrepancies in the literature regarding the molecular landscape of EOCRC that are mostly due to differences in study design and included population. The molecular landscape of CRC is heterogeneous and complex, influenced by various factors, such as MSI status and primary tumor site, which were both accounted for in our study ^58–60.

Mutations in the TP53 gene were more prevalent in EOCRC, consistent with previous findings of higher frequencies in very early-onset cases (< 40 years) ²⁶. We found this significant association to be specific with MSS EOCRC in alignment with the trend we previously reported ²⁵. Our data also showed significant differences in primary tumor location, with right-sided tumors exhibiting lower TP53 mutation prevalence in LOCRC compared to EOCRC aligning with previous reports that associate the left-side with TP53 mutations ⁶¹. Furthermore, the significant co-occurrence of TP53 mutations with other driver mutations in EOCRC highlights its potential central role in tumorigenesis in younger patients.

APC mutations displayed distinct patterns across MSI-H and MSS tumors, with a higher prevalence in MSI-H EOCRC but a lower prevalence in MSS EOCRC. These results are consistent with those of Lieu et al. and our earlier finding that APC mutations are more common in very early-onset tumors ^25,26. These two studies also found CTNNB1 and ATM variants to be more prevalent in the very early-onset group; we didn’t find these two to be statistically significant when comparing early- to late-onset CRC using the cutoff of 50 years old. With age decile analysis, however, we did discover that they were more prevalent in younger age groups.

The progression from adenoma to carcinoma, driven primarily by the Wnt/β-catenin signaling pathway, includes APC and CTNNB1. The traditional adenoma-carcinoma-metastasis model of CRC tumorigenesis is observed in 70–90% of sporadic CRC patients ^60,62–64. However, the lower prevalence of APC mutations in MSS EOCRC suggests additional aberrations may contribute to the pathogenesis of early-onset disease. Furthermore, the Wnt/β-catenin pathway interacts with other CRC pathways like PI3K/AKT and RAS. ^58,65–68. Our cohort’s findings that the Wnt/β-catenin signaling pathway genes were less prevalent in MSS EOCRC raise the possibility that the disease may have additional aberrations that contribute to its pathogenesis.

Mutations in the RAS signaling pathways are among the most frequently mutated genes in CRC, with KRAS being one of the key mutations in the chromosomal instability pathway (CIN) of CRC carcinogenesis, as identified by adenoma-carcinoma sequencing ^60,69–74. Different prevalence patterns were observed for KRAS mutations, with MSI-H EOCRC exhibiting a significantly higher frequency than LOCRC. KRAS mutations co-occurred with SOX9 and NF1 mutations, which are strongly linked to NOTCH3 and RB1, indicating the involvement of cell signaling and tumor suppressor pathways in MSS EOCRC. In right-sided EOCRC, KRAS mutations were more prevalent (64.5% vs. 53%), consistent with previous findings ⁷⁵. The prevalence of KRAS mutations in age decile exploration showed a pattern of being less prevalent in younger age groups in alignment with our previous observation and others ^25,26,34. Although Watson et. al. reported contradicting results of a higher prevalence of KRAS in early-onset (≤40 years), we believe that other observations and ours are more accurate considering the low sample size of their study (n=68) ^25,26,33,34.

Consistent with our finding of lower BRAF prevalence in right-sided EOCRC (8% vs. 20%), previous research has indicated a higher frequency of BRAF mutations in right-sided LOCRC ⁷⁵. This finding was particularly significant with MSS EOCRC tumors. In contrast, BRAF mutations were more common in MSI-H tumors, a finding consistent with their association with hypermutation and epigenetic silencing of MLH1 in sporadic MSI-H CRC ^76,77. The early-onset group has a higher rate of MSI-H cancers, even when patients with familial susceptibility are eliminated. ^78–80. Understanding the distinct molecular landscape of MSI status reported here emphasizes the critical importance of thorough germline and MSI status testing in CRC among young patients.

The significant co-occurrence of FBXW7 mutations with NOTCH3, RB1, and PIK3R1 in MSS EOCRC suggests unique mutational interactions distinct from late-onset cases. This finding aligns with previous studies indicating that FBXW7 mutations are associated with EOCRC ⁸¹. Additionally, FBXW7 was reported to interact with the Notch signaling pathway in CRC ^82,83. Further supporting reports that ATM mutations contribute to genomic instability in CRC is our significant finding of co-occurrence of ATM mutations with genes like NOTCH3, EGFR, and ERBB4 in MSS EOCRC, which highlights the involvement of DNA damage response pathways. These distinct mutational networks in EOCRC highlight the necessity for age-specific molecular profiling to inform targeted therapeutic strategies.

Targeted therapies and precision medicine strategies have revolutionized the therapy of CRC by targeting specific driver genes and pathways, including anti-EGFR for KRAS wild-type, BRAF inhibitors for BRAF-mutated tumors, and immune checkpoint inhibitors for MSI-H tumors ^84–89. Our findings highlight EOCRC’s distinct genetic and clinicopathological characteristics, which may guide the development of tailored therapy regimens and precision medicine methods for early-onset patients to improve clinical outcomes.

As previously mentioned, inconsistent results in the molecular profiling of EOCRC tumors have been reported across multiple studies ^74,90–93, likely due to variations in research design. One major source of discrepancy is the differing age cutoffs used to define EOCRC, with some studies classifying early-onset as <40 years at diagnosis, while others using average-onset CRC (>50 years) as the reference for late-onset. Our analysis of mutation prevalence across age deciles highlights the significant impact of these differing age thresholds on reported findings. Additionally, by stratifying tumors based on MSI status, we demonstrated how the inclusion of tumors with distinct molecular characteristics can skew the overall molecular landscape, underscoring the need for careful cohort selection. Another limitation in prior studies is the reliance on small sample sizes of EOCRC patients, which compromises the robustness of conclusions. To address this we analyzed a cohort of 3,185 EOCRC cases, providing a substantially larger and more comprehensive dataset. This expanded cohort enabled more rigorous analyses and offered deeper insights into the unique characteristics of EOCRC, distinguishing our study from previous efforts and paving the way for more precise strategies in the diagnosis and treatment of EOCRC. Although our study profiled the clinicopathological and molecular characteristics of EOCRC from a large cohort of 10,000 CRC cases using data from a large, tertiary center-based cohort, its single-institutional design may limit generalizability to larger populations. Furthermore, it is possible that some mutations that were not covered by the earlier NGS panels were falsely underrepresented. Furthermore, not all tumor stages underwent sequencing as part of the standard of care, which may have influenced the representation of molecular profiles across different disease stages. Lastly, the retrospective design limits causal interpretations, and despite stratification by MSI and MSS status, other molecular or environmental factors may have confounded our results.

While this study provides valuable insights into the clinical and molecular characteristics of EOCRC and LOCRC, several limitations and opportunities for future research should be noted. First, the retrospective design may introduce selection biases and limit the generalizability of the findings. Second, survival data and treatment outcomes were not available, which restricts the ability to evaluate the prognostic and therapeutic implications of the identified molecular differences. Third, the molecular data were derived from clinical targeted panel testing, which, while informative, may be influenced by the timing of testing and may not comprehensively capture all relevant genomic alterations. Additionally, metabolic and immunological profiles, which could offer further biological insights, were beyond the scope of this study. While MSI status and molecular data from existing records provided valuable information, they may not encompass all potential biomarkers.

Given these limitations, further action is necessary due to the clinical importance and rising concerns with EOCRC. Improving preventative programs and awareness campaigns is essential to promote early detection and mitigate healthcare disparities. Although the present findings may not currently justify EOCRC-specific interventional clinical trials, rigorous translational research and correlative studies are imperative to address this gap. Incorporating broader datasets, multi-omics approaches, and prospective designs, as well as insights from metabolic and immunological profiles, could enhance our understanding of the carcinogenesis process in this patient population. Such advancements would provide the foundation for developing targeted interventions and personalized therapies, ultimately improving clinical outcomes and informing public health strategies for EOCRC. Despite these limitations, the large cohort and detailed molecular analysis presented in this study underscore its value and significance in addressing the unique challenges of EOCRC.

In conclusion, this study provides comprehensive insights into the unique clinical and molecular characteristics of EOCRC, leveraging a large cohort and highlighting the need for tailored diagnostic, therapeutic, and preventive strategies to improve outcomes for EOCRC patient population. Future multicenter and prospective studies are needed to validate these findings and ensure broader applicability.

CLINICAL SIGNIFICANCE:

1. The unique mutational and co-mutational patterns identified in early-onset colorectal cancer (EOCRC) highlight the necessity for age-specific molecular profiling. These findings can guide tailored therapeutic approaches and improve precision medicine strategies for younger patients with colorectal cancer.

2. The raised prevalence of advanced-stage disease and unique molecular features in EOCRC, such as higher TP53 and lower APC mutations, highlight the importance of developing age-adapted screening protocols and prognostic tools to detect and manage EOCRC more effectively.

DATA AVAILABILITY:

Because of HIPPA regulations and privacy and proprietary concerns, the data used in this study was retrieved from patients’ electronic health records (EHRs) in a real-world healthcare setting and is subject to restricted access. When feasible, the paper and its Supplementary Figures/Tables have provided derived data that support the study’s conclusions.