Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Sep 29.
Published in final edited form as: JCO Precis Oncol. 2026 Sep 1;10(9):e2600389. doi: 10.1200/PO-26-00389

Comparative analysis of potential clinical actionability of genomic alterations in early-onset versus later-onset gastrointestinal cancers

Lawrence W Wu 1, Jimyung Park 1, Sungjoo Jang 2, Ryan H Moy 1
PMCID: PMC13619329  NIHMSID: NIHMS2208556  PMID: 42679331

Abstract

Purpose:

As biomarker-directed therapy increasingly shapes gastrointestinal (GI) oncology, it remains unclear whether early-onset (EO) and later-onset (LO) GI cancers harbor comparable opportunities for clinically actionable targeting. We compared the landscape of potentially actionable genomic alterations in EO versus LO GI cancers using AACR Project GENIE v19.0.

Methods:

GI tumor samples were assigned to 10 prespecified tumor groups using OncoTree codes. Samples were annotated with OncoKB therapeutic levels and classified as potentially actionable if they harbored at least one level 1–3B alteration. EO and LO disease were defined as age at sequencing <50 years and ≥50 years, respectively. Group-wise comparisons used Wilcoxon rank-sum, chi-square, or Fisher exact testing as appropriate and with false discovery rate correction. Multivariable logistic regression evaluated age group associations overall and within tumor groups.

Results:

Among 53,945 GI tumor samples, 10,573 (19.6%) were EO and 43,372 (80.4%) were LO. EO tumors had lower prevalence of potentially actionable alterations in colorectal (71.0% vs 78.1%, q<0.001), esophagogastric (53.3% vs 57.9%, q=0.0097), gastrointestinal stromal tumor (75.1% vs 90.9%, q<0.001), liver (25.4% vs 35.4%, q=0.0021), and pancreatic tumors (82.9% vs 90.7%, q<0.001). In the overall model, LO status was associated with higher odds of potential actionability (OR 1.39, 95% CI 1.33–1.47, p<0.001). Tumor group-specific associations persisted in colorectal, GIST, liver, and pancreatic tumors after multivariable adjustment.

Conclusion:

Potential clinical actionability differs between EO and LO GI cancers in a tumor lineage-specific manner. Several major EO GI tumor groups appear relatively depleted of potentially actionable alterations, suggesting that the expanding therapeutic reach of precision oncology may not be distributed evenly across age-defined GI cancer populations and underscoring the need for EO-focused biomarker discovery and therapeutic development.

Keywords: Early-onset cancer, gastrointestinal cancers, clinical actionability, precision oncology, AACR Project GENIE, OncoKB

Introduction:

There is an alarming rise in early-onset (EO) gastrointestinal (GI) cancers, defined as GI cancers diagnosed in adults younger than 50 years 1. The most common EO GI cancers include colorectal, esophagogastric, and pancreatic cancers; however, increasing incidence has also been observed in rarer GI malignancies such as appendix, small bowel, biliary tract, and neuroendocrine cancers.1,2

Several modifiable risk factors are shared across common GI cancers, including colorectal, pancreatic, and esophagogastric cancers. These overlapping risk factors include tobacco use, alcohol consumption, obesity, and sedentary behavior.3–12 In addition to these shared exposures, tumor-specific modifiable factors also contribute to disease risk, such as Helicobacter pylori infection for gastric cancer.13 Despite these associations, the rise in EO GI cancers is likely not fully explained by recognized risk factor trends alone, suggesting additional contributions from environmental and host factors.

There is accumulating evidence that at least a subset of EO GI cancers may be molecularly distinct from later onset (LO) counterparts. For example, EO colorectal cancers (CRC) have higher prevalence of germline mutations associated with Lynch syndrome (MLH1, MSH2, MSH6, PMS2), resulting in increased rates of microsatellite instability.14 With respect to somatic tumor genomic alterations, EO CRC is characterized by a higher frequency of alterations in TP53 and CTNBB1, alongside a lower frequency of alterations in APC, KRAS, and BRAF compared to LO CRC.15 Similarly, EO esophagogastric cancers (EGC) harbors distinctive germline and somatic genomic landscapes. Germline mutations in CDH1 are more prevalent in EO EGC.16,17 At the somatic level, EO EGC demonstrates increased frequencies of alterations in CDH1, FGFR2, CCNE1, and MYC, as well as decreased prevalence of high microsatellite instability.16,18

Given the rising incidence of several EO GI cancers and parallel advances in comprehensive tumor genomic profiling, it is critical to develop a precision oncology framework to systematically define molecular differences between EO and LO GI cancers. The American Association for Cancer Research (AACR) Project Genomics Evidence Neoplasia Information Exchange (GENIE) is an international, collaborative pan-cancer registry with real-world clinicogenomic annotation and is the largest such database that is publicly accessible.19 The most recent version (v19.0) of AACR GENIE released in January 2026 has 271,837 sequenced tumor samples from over 227,696 patients. AACR GENIE therefore enables large-scale clinicogenomic analyses across tumor types.

To assess the therapeutic implications of genomic alterations identified within GENIE, we have leveraged OncoKB, a rigorously validated database that provides evidence-based annotation of the clinical actionability of genomic alterations.20 OncoKB classifies alterations according to levels of therapeutic evidence (ranging from levels 1–4), with Level 1 being an FDA-recognized biomarker predictive of response to an FDA-approved drug for a particular tumor type; Level 2 being a standard of care biomarker recommended by professional guidelines predictive of response to an FDA-approved drug for a particular tumor type; Level 3A being compelling clinical evidence supporting the biomarker as predictive of response to a drug for a particular tumor type; Level 3B being a standard of care or investigational biomarker predictive of response to an FDA-approved or investigational drug in another tumor type; and Level 4 being a biomarker with compelling biological evidence as predictive of a response to a drug for a particular tumor type. Prior pan-cancer analyses of AACR GENIE revealed a marked increase in clinically actionable alterations between 2017 to 2022, largely driven by the approval of targeted therapies against alterations such as PIK3CA, FGFR2, IDH1, RET, and KRAS G12C.21

Despite these advances, age-specific analyses of the clinical actionability of molecular alterations remain limited, and the expanded contemporary GENIE dataset has not been systematically examined in the context of early-onset disease. To address this gap, we compared the landscape of potentially actionable alterations between early-onset and later-onset GI cancers in the AACR GENIE database.

Methods:

Data Source

AACR Project GENIE is a multi-institutional, real world clinicogenomic registry composed of tumor sequencing data from 20 participating international cancer centers. The AACR GENIE v19.0 database was used to select GI cancers samples by OncoTree code. GI cancer samples were classified into 10 prespecified tumor groups: anal, appendix, biliary, colorectal, esophagogastric, gastrointestinal stromal tumor (GIST), liver, neuroendocrine, pancreatic, and small bowel (Supplementary Table 1). We utilized OncoKB Annotator (https://github.com/oncokb/oncokb-annotator) to annotate all samples and their somatic mutations, copy number alterations, and structural variants by therapeutic level of evidence (Levels 1–4). For samples with multiple actionable alterations, the alteration associated with the highest level of clinical evidence was utilized. Samples without any annotated therapeutic level were categorized as NA. We defined a tumor sample to have a potentially actionable alteration if there was at least one Level 1–3B alteration present. EO disease was defined as age of sequencing <50 and LO disease as age of sequencing ≥ 50; samples lacking age data were excluded. Clinicodemographic variables included sex, race, ethnicity, site of tumor sample, contributing center, and sequencing assay. Tumor mutation burden (TMB) was extrapolated directly from the GENIE database and was classified as TMB-High if TMB was ≥10 mutations/megabase per the Food and Drug Administration (FDA) tumor-agnostic approval indication.22 TMB-high status is considered Level 1 for all solid tumors by OncoKB.

Statistical Analysis

Chi-square or Fisher’s exact testing compared categorical variables between EO and LO tumor groups. Continuous variables were compared using Wilcoxon rank sum test between EO and LO tumor groups. For sample-level comparisons of highest OncoKB category across tumor groups and OncoTree-defined subgroups, false discovery rate (FDR) was controlled using the Benjamini-Hochberg procedure, with statistical significance defined as q<0.05. The highest OncoKB level across any alteration class, including mutation, copy number alteration, structural variant, or TMB-high, was used for sample-level classification. Separate analyses were performed for specific OncoTree codes within each tumor group. Only OncoTree codes with at least 30 samples in both early onset and later onset samples were included. Alteration-class specific analyses were also performed separately for mutations, copy number alterations, structural variants, and TMB-high status.

Multivariable logistic regression was used to evaluate the association between later-onset age group and the presence of potentially actionable alterations. The overall model adjusted for sex, race, ethnicity, and tumor group, whereas tumor group-specific models adjusted for sex, race, and ethnicity. For tumor group-specific multivariable analyses, family-wise error rate (FWER) was controlled using the Holm method, and FWER-adjusted confidence intervals are reported.

To evaluate whether the findings differed by how OncoKB-annotated potential actionability was defined, we performed two sensitivity analyses: (1) OncoKB Level 1–3B and excluding TMB-High and (2) OncoKB Level 1–2 and excluding TMB-High. We repeated tumor group-level, alteration class-level, and multivariable analyses using these definitions.

All analyses were performed using R version 4.2.1. Key packages included dplyr, tidyr, readr, stringr, purrr, forcats, ggplot2, and scales.

Ethics Statement

In accordance with 45 CFR 46, it was determined that this study was exempt from institutional review board review because it utilized exclusively publicly available deidentified data.

Results:

Cohort characteristics

A total of 53,945 GI tumor samples were analyzed, of which 10,573 (19.6%) were EO and 43,372 (80.4%) were LO samples. The median age of sequencing was 43 years in the EO group and 66 years in the LO group (q<0.001) (Table 1). Compared with LO tumors, EO tumors were more frequently derived from female patients (45.3% vs 41.2%, q<0.001). There were significant differences in tumor group distribution. EO cases were enriched for colorectal (60.3% vs 39.2%), whereas LO cases were enriched for pancreatic (22.8% vs 7.8%), biliary (9.3% vs 5.4%) and esophagogastric tumors (14.5% vs 10.6%) (p<0.001). The distribution of contributing centers and sequencing assays differed between EO and LO cohorts (Supplementary Table 2), highlighting the heterogeneity in the AACR GENIE database.

Table 1: Clinicodemographic characteristics of early-onset and later-onset gastrointestinal cohorts.

Early onset (EO) and later-onset (LO) cohorts were defined by age of sequencing as <50 years and ≥ 50 years, respectively. Data are presented as median (interquartile range, IQR) for continuous variables and number (%) for categorical variables. P-values were calculated using Wilcoxon rank-sum test for continuous variables and chi-square test for categorical variables.

Characteristic Early-onset Later-onset p-value
Samples, Number 10573 43372
Age, Median (IQR) 43 (37–47) 66 (59–73) <0.001
Sex, Number (%) <0.001
 Male 5195 (49.1) 23250 (53.6)
 Female 4793 (45.3) 17885 (41.2)
 Other/Unknown 585 (5.5) 2237 (5.2)
Sample Type, Number (%) <0.001
 Primary 6435 (60.9) 25224 (58.2)
 Metastatic 3306 (31.3) 12941 (29.8)
 Other/Unknown 832 (7.9) 5207 (12.0)
Race, Number (%) <0.001
 White 6540 (61.9) 27624 (63.7)
 Black 721 (6.8) 2591 (6.0)
 Asian/Pacific Islander 776 (7.3) 2309 (5.3)
 Native American 24 (0.2) 72 (0.2)
 Other/Unknown/Not Collected/Not Applicable 2512 (23.8) 10774 (24.8)
Ethnicity, Number (%) <0.001
 Non- Spanish/non- Hispanic 6809 (64.4) 28148 (64.9)
 Spanish/Hispanic 1123 (10.6) 3518 (8.1)
 Unknown/Not Collected 2641 (25.0) 11706 (27.0)
Tumor Group, Number (%) <0.001
 Anal 73 (0.7) 496 (1.1)
 Appendix 295 (2.8) 788 (1.8)
 Biliary 568 (5.4) 4017 (9.3)
 Colorectal 6374 (60.3) 16990 (39.2)
 Esophagogastric 1123 (10.6) 6302 (14.5)
 Gastrointestinal Stromal Tumor 418 (4.0) 1890 (4.4)
 Liver 323 (3.1) 982 (2.3)
 Neuroendocrine 456 (4.3) 1442 (3.3)
 Pancreatic 832 (7.8) 9902 (22.8)
 Small Bowel 120 (1.1) 563 (1.3)
Open in a new tab

Landscape of potential clinical actionability across the study population

Next, we compared the prevalence of potentially actionable genomic alterations between EO and LO disease across GI tumor types to evaluable age-associated differences in therapeutic actionability. Across the study population, there were fewer potentially actionable alterations in EO colorectal (71.0% vs 78.1%, q<0.001), esophagogastric (53.3% vs 57.9%, q=0.0097) gastrointestinal stromal tumor (75.1% vs 90.9%, q<0.001), liver (25.4% vs 35.4%, q=0.0021), and pancreatic tumors (82.9% vs 90.7%, q<0.001) compared to respective LO groups (Figure 1A, Supplementary Table 3). No statistically significant differences were observed between EO and LO disease in the remaining GI tumor groups after false discovery rate correction.

Figure 1: Distribution of OncoKB therapeutic level by age, tumor group, and alteration class.

Figure 1:

Figure 1:

Open in a new tab

(A) Stacked bar plots show, within each tumor group, the proportion of samples whose highest OncoKB level across alterations falls into Level 1, 2, 3A, 3B, 4, or NA. The highest level across any alteration class (mutation, copy number alteration, structural variant, tumor mutation burden high) is used per sample. (B) Faceted bar plots for each gastrointestinal tumor group with the percentage of tumor samples with at least one potentially actionable alteration in each alteration class. Potential actionability was defined by the presence of an OncoKB level 1–3B alteration. Brackets indicate tumor groups with a statistically significant difference in proportion of samples with a potentially actionable level in specific alteration types between EO and LO after Benjamini-Hochberg false discovery rate correction. Significance is denoted by asterisks (* q<0.05, ** q<0.01, *** q<0.001)

Decomposition of potentially actionable alterations by specific alteration type (mutation, copy number alteration, structural variant, or tumor mutation burden) revealed that age-associated differences in molecular landscape were tumor lineage-specific (Figure 1B, Supplementary Tables -7). For example, in colorectal cancer, EO tumors exhibited lower frequencies of potentially actionable mutations (62.7% vs 68.7%, q<0.001), structural variants (1.4% vs 1.9%, q=0.022), and TMB-H status (27.4% vs 35.7%, q<0.001) compared to LO tumors, while copy number alterations were slightly more frequent in EO tumors (4.2% vs 3.2%, q<0.001). In the EO pancreatic tumor group, there were decreased potentially actionable mutations (77.3% vs 88.5%, q<0.001) and higher potentially actionable structural variants (3.8% vs 2.2%, q=0.018) compared to the respective LO group. However, there were no significant differences in copy number alterations or TMB-H. In contrast, EO biliary tumors had higher potentially actionable structural variants compared to LO tumors (13.6% vs 6.0%, q<0.001), but no significant differences in other alteration types. Together, these findings demonstrate that age-associated differences in molecular actionability are not uniform across GI cancers but instead reflect tumor lineage-specific shifts in the types of genomic alterations present in EO tumors.

To evaluate which specific tumor types were driving the observed differences between EO and LO disease, we performed further stratified analyses by specific OncoTree codes within tumor groups that demonstrated significant age-associated differences (Figure 2, Supplementary Tables 8–16). These analyses were intended as secondary descriptive analyses, as the primary tumor group analysis combined related OncoTree codes into prespecified tumor groups. In the colorectal tumor group, several histologic subtypes contributed to the observed differences, including EO colon adenocarcinoma (COAD) (73.4% vs 80.2%, q<0.001), colorectal adenocarcinoma (COADREAD) (73.1% vs 77.5%, q=0.0022), mucinous adenocarcinoma of the colon and rectum (MACR) (76.6% vs 88.2%, q=0.0022), rectal adenocarcinoma (READ) (64.4% vs 70.9%, q<0.001), and signet ring cell adenocarcinoma of the colon and rectum (SRCCR) (43.5% vs 66.7%, q=0.0074) (Figure 2A, Supplemental Table11). In contrast, within the esophagogastric tumor group, the difference was driven solely by EO stomach adenocarcinoma (STAD) (47.7% vs 57.3%, q=0.0065) (Figure 2B, Supplemental Table 12). Similarly, in the pancreatic tumor group, the only subtype demonstrating a significant difference was pancreatic adenocarcinoma (PAAD) (87.1% vs 91.5%, q<0.001) (Figure 2C, Supplemental Table 15). No significant differences were observed among liver tumor subtypes following stratification by OncoTree code (Figure 2D, Supplementary Table 13).

Figure 2: Distribution of OncoKB therapeutic level by OncoTree Code in select tumor groups.

Figure 2:

Figure 2:

Figure 2:

Figure 2:

Open in a new tab

Stacked bar plots show for each OncoTree code, the proportion of samples whose highest OncoKB level across alterations falls into Level 1, 2, 3A, 3B, 4, or NA. The highest level across any alteration class (mutation, copy number alteration, structural variant, tumor mutation burden high) is used per sample. Only OncoTree codes with at least 30 samples in both early onset and later onset samples were included. (A) Colorectal tumor OncoTree codes (COAD – colon adenocarcinoma; COADREAD – colorectal adenocarcinoma; MACR – mucinous adenocarcinoma of the colon and rectum; READ – rectal adenocarcinoma; SRCCR – signet ring cell adenocarcinoma of the colon and rectum; TAC – tubular adenoma of the colon). (B) Esophagogastric tumor OncoTree codes (DSTAD – diffuse type stomach adenocarcinoma; EGC – esophagogastric adenocarcinoma; ESCA – esophageal adenocarcinoma; ESCC – esophageal squamous cell carcinoma; GEJ = adenocarcinoma of the gastroesophageal junction; SPDAC – poorly differentiated carcinoma of the stomach; SSRCC – signet ring cell carcinoma of the stomach; STAD – stomach adenocarcinoma). (C) Pancreatic tumor OncoTree codes (PAAD – pancreatic adenocarcinoma). (D) Liver tumor OncoTree codes (HCC – hepatocellular carcinoma). Brackets indicate tumor groups with a statistically significant difference in proportion of samples with a potentially actionable level between EO and LO after Benjamini-Hochberg false discovery rate correction. Significance is denoted by asterisks (* q<0.05, ** q<0.01, *** q<0.001).

To determine whether these age-associated differences persisted after accounting for possible confounders, we performed multivariable logistic regression analyses. In the overall model, LO status was independently associated with higher odds of potential actionability (OR 1.39, 95% CI: 1.33–1.47, p<0.001) (Supplementary Table 17). Within tumor group-specific models, LO status remained associated with higher odds of potential actionability in GIST (OR 3.57, 95% CI: 2.69–4.73, q<0.001), colorectal (OR 1.44, 95% CI: 1.35–1.54, q<0.001), liver (OR: 1.66, 95% CI: 1.24–2.22, q=0.0046), and pancreatic (OR 2.00, 95% CI: 1.65–2.43, q<0.001) tumor groups compared to respective LO tumor groups (Figure 3, Supplementary Table 18). The point estimate for esophagogastric tumors remained directionally similar (OR 1.17, 95% CI 1.03–1.33) but did not retain significance after Holm correction (q=0.097). These results overall support the effect of age on potential clinical actionability across select tumor groups after adjustment for measured covariates.

Figure 3: Adjusted odds ratios for potential actionability across gastrointestinal tumor groups.

Figure 3:

Open in a new tab

Forest plots display tumor group-specific adjusted odds ratios (ORs) with 95% confidence intervals for the presence of any potentially actionable alteration in EO vs LO tumors. OR>1 indicate higher odds of potential actionability in LO tumors, whereas OR<1 indicate higher odds of potential actionability in EO tumors. Significance is denoted by asterisks after family-wise error rate adjustment across tumor groups using the Holm method correction (*p<0.05, **p<0.01, ***p<0.001).

To evaluate whether the findings differed according to how OncoKB-annotated potential actionability was defined, we performed two sensitivity analyses. In the first sensitivity analysis, we retained OncoKB Level 1–3B and excluded TMB-High. In this analysis, EO colorectal, GIST, and pancreatic tumors continued to show lower prevalence of potential actionability compared to their respective LO counterparts (Supplementary Figures 3–5, Supplementary Tables 21, 23, 25). In the second sensitivity analysis, we further restricted to OncoKB Level 1–2 and excluded TMB-High. With this definition, we found that EO colorectal tumors and GIST continued to show lower prevalence of potential actionability, whereas EO biliary and pancreatic tumors showed higher prevalence of potential actionability compared to respective LO groups (Supplementary Figures 3–5, Supplementary Tables 22, 24, 26). These analyses demonstrated evidence-threshold-dependent differences.

Gene-level potentially actionable landscape across tumor groups

We next examined gene-level patterns of potentially actionable alterations across GI tumor types stratified by age of onset. Across the entire study population, the most frequently altered actionable genes were KRAS, PIK3CA, and ERBB2. Across GI tumor groups, the potentially actionable gene landscape differed between EO and LO tumors. Although the overall spectrum of recurrent alterations within each tumor group was largely similar between EO and LO tumors, several loci showed statistically significant age-associated differences after FDR adjustment (Figure 4A; Supplementary Table 19). EO colorectal tumors had significantly lower frequencies of BRAF alterations compared with LO tumors (3.9 vs 10.7%, q<0.001). In GIST, EO tumors exhibited reduced KIT alterations (59.1% vs 75.6.%, q<0.001) and PDGFRA alterations (5.3% vs 10.5%, q=0.0077). In pancreatic cancer, EO tumors were depleted for KRAS alterations (71.9% vs 85.4%, q<0.001) compared to respective LO tumors.

Figure 4: Gene-level differences in potentially actionable alterations across gastrointestinal tumor groups and select driver effect sizes.

Figure 4:

Figure 4:

Open in a new tab

(A) Bubble plot with the prevalence of potentially actionable alterations by gene across gastrointestinal tumor groups stratified by age group. For each tumor group, genes displayed represent the top five potentially actionable genes in EO and LO strata (ranked by the number of affected samples within each stratum). Bubble size is proportional to the percentage of samples in the corresponding tumor group and age stratum harboring at least one potentially actionable alteration in the indicated gene. When shown, asterisks denote Benjamini-Hochberg false discover rate adjusted significance for EO vs LO differences within each tumor group (*q<0.05; **q<0.01; ***q<0.001). (B) Forest plots display odds ratios with 95% confidence intervals for the presence of selected potentially actionable driver alterations. OR <1 indicate lower odds of potential actionability in EO relative to LO. OR>1 indicate higher odds in EO relative to LO. Significance is denoted by asterisks after FDR correction (*p<0.05, **p<0.01, ***p<0.001).

To complement these gene-level prevalence analyses, we evaluated select potentially actionable alterations using effect-size estimates (Figure 4B, Supplementary Table 20). Consistent with gene level findings, EO colorectal tumors were depleted for BRAF alterations (OR 0.34, 95% CI: 0.30–0.39, q<0.001), EO GIST were depleted for PDGFRA (OR 0.48, 95% CI: 0.31–0.76, q=0.0077), KIT (OR 0.47, 95% CI 0.37–0.58, q<0.001), and MTAP (OR 0.29, 95% CI: 0.11–0.77, q=0.28) alterations, and EO pancreatic tumors were depleted for KRAS alterations (OR 0.44, 95% CI: 0.37–0.51, q<0.001). Together, these results suggest that the lower overall potential clinical actionability observed in several EO GI tumor groups was driven by specific recurrent alterations.

Discussion

In this large real-world clinicogenomic analysis of the AACR Project GENIE v19.0 database annotated by OncoKB, we found that potential clinical actionability of genomic alterations differs between EO and LO GI cancers in a tumor lineage-specific manner. Several major EO GI tumors, including colorectal, GIST, liver, and pancreatic tumors, were relatively depleted for potentially actionable alterations, and these associations persisted after multivariable adjustment.

The lineage-specific nature of the observed differences is notable. In colorectal cancer, EO tumors had reduced prevalence of potentially actionable BRAF alterations. The decreased prevalence of BRAF in EO CRC has been previously reported.15 This difference may in part reflect greater contribution of MLH1 promoter hypermethylation in LO disease. Since methylation profiling and mismatch repair protein/microsatellite instability status was not available in GENIE, this potential mechanism could not be directly assessed. In GIST, EO tumors had markedly reduced potentially actionable alterations, driven by lower KIT and PDGFRA alterations. This is consistent with prior observations where younger patients are more likely to harbor wild type KIT and PDGFRA tumors, which include succinate dehydrogenase deficient subtypes which have distinctive clinical behavior.23,24 Given the integral role of targeted therapies in GIST-directed therapeutics which can effectively target several KIT alterations, this finding may have implications for expected response to standard-of-care therapeutics and motivates further study of the underlying biology of EO GIST, particularly for those with KIT and PDGFRA wild-type tumors. EO pancreatic cancer similarly demonstrated reduced potential actionability, which was driven by reduced prevalence of KRAS alterations. A lower KRAS fraction may imply a larger KRAS wild-type subgroup, which has other distinct oncogenic drivers such as fusions.25 There was comparatively less fusion profiling compared to somatic mutations, and some of these may have been missed by more narrow sequencing panels. This highlights the need for comprehensive molecular profiling that includes fusion detection.

The OncoKB Level 1–3B endpoint was utilized as a broad, hypothesis-generating measure of potential actionability. Levels 3A and 3B include emerging or context-dependent therapeutic evidence, and there is less clinical evidence for potential targeted therapeutics against those alterations in that tumor lineage. TMB-High also requires disease-specific interpretation. While TMB-High is recognized as a tumor-agnostic biomarker and was therefore included in the primary endpoint, its predictive relevance in colorectal cancer depends strongly on mismatch repair protein/microsatellite stability status.26Mismatch repair protein/microsatellite stability status was not available in GENIE, so microsatellite instability high (MSI-H)/TMB-High and microsatellite stable (MSS)/TMB-High colorectal tumors could not be distinguished. Sensitivity analyses excluding TMB-High and restricting to higher evidence of Level 1–2 demonstrated that EO versus LO differences depend on the evidence threshold used.

Importantly, this analysis cannot determine whether observed EO versus LO genomic differences contribute to early-onset tumor development or instead reflect downstream features of established tumors. Potential explanations include differences in environmental exposures, inherited susceptibility, epigenetic states, tumor evolution or other unmeasured factors that cannot be directly assessed in AACR GENIE. The expansion of targeted therapies and biomarker-linked indications has increased the proportion of tumors with potentially actionable alterations across cancer types. However, our results highlight that benefits of these advances may not accrue equally across populations. In several major GI tumor groups, EO tumors were depleted for canonical potentially actionable targets that currently guide standard-of-care treatment, trial enrollment, and ongoing biomarker-selected therapeutic development. At the same time, lower prevalence of a potentially actionable alteration should not automatically be interpreted as clinically unfavorable. For example, BRAFV600E mutations in colorectal cancer may be targetable, but BRAFV600E-mutated, microsatellite-stable colorectal cancer has historically been associated with worse prognosis compared to BRAF-wild-type colorectal cancer.27–29 More recently, BRAF/EGFR-directed regimens have significantly improved survival outcomes in BRAFV600E-mutated metastatic colorectal cancer, illustrating that the clinical implications of an annotated alteration may evolve as more effective matched therapies emerge.30These molecular differences therefore have implications for precision oncology frameworks and motivate further biomarker discovery and consideration of EO-specific therapeutic development.

At the same time, the data do not support a uniform narrative that all EO tumors are less targetable globally. Rather, the results suggest potential actionability shifts according to lineage and alteration class. For example, EO biliary tumors showed higher potential actionability in structural variants. This is justification for more refined age-stratified molecular frameworks rather than broad generalizations across all EO GI cancers.

The clinical relevance of OncoKB annotations will continue to evolve as new therapeutics mature. For example, emerging RAS-directed strategies may alter the therapeutic relevance of KRAS-mutant pancreatic cancer beyond limited allele-specific contexts.31,32 Lower-level OncoKB categories should be interpreted cautiously and in the context of therapeutic evidence available at time of annotation. Age-stratified clinicogenomic analyses may help identify which EO populations align with current biomarker-defined therapeutic strategies and which require further biomarker and therapeutic discovery.

Limitations:

The study has limitations inherent to the AACR GENIE database. Microsatellite instability, methylation profiling, and other potentially relevant immunohistochemistry results could not be directly assessed. There was heterogeneity between the next-generation sequencing panels utilized across contributing cancer centers to AACR GENIE, which may have led to differential detection of certain alterations. In addition, AACR GENIE reflects tumors selected for sequencing at participating cancer centers, and observed differences in demographic and tumor group composition may reflect referral patterns, institutional sequencing practices, and incomplete demographic data capture. Patient-level geographic residence was not available, limiting assessment of patient-level geographic or regional ascertainment patterns. Furthermore, there is a lack of clinical data on prior treatments received, treatment response, and survival outcomes in this database, limiting the ability to link potential actionability to real-world outcomes or assess treatment-related clonal evolution including acquired alterations over time. Age at sequencing may not directly reflect age of diagnosis for all cases.

Conclusions:

To our knowledge, this study represents the largest age-stratified clinicogenomic analysis of therapeutic actionability across GI tumors. EO GI cancers demonstrated a lower prevalence of potential clinical actionable genomic alterations compared to respective LO GI cancers across several major GI tumor groups. Together, these findings suggest that several EO GI cancers may have fewer established precision oncology targets, underscoring the need for EO-focused biomarker discovery and novel therapeutic strategies.

Supplementary Material

Supplementary Files

Context Summary.

Key Objective:

Do early-onset and later-onset gastrointestinal cancers harbor comparable opportunities for current potentially actionable biomarker-directed therapy?

Knowledge Generated:

In a clinicogenomic analysis of 53,945 gastrointestinal tumor samples from AACR Project GENIE v19.0, several major early-onset tumor groups, including colorectal, gastrointestinal stromal tumor, liver, and pancreatic tumors, showed lower prevalence of potentially actionable alterations than later-onset counterparts. These age-associated differences were tumor lineage-specific and included differences in potentially actionable alterations involving BRAF, KRAS, KIT, and PDGFRA.

Relevance:

These findings suggest that current precision-oncology opportunities may not be evenly distributed across age-defined gastrointestinal cancer populations and support further investigation into early-onset biomarker discovery and therapeutic development.

Acknowledgements:

The authors acknowledge the American Association for Cancer Research and its financial and material support in the development of the AACR Project GENIE registry as well as members of the AACR Project GENIE consortium for their commitment to data sharing. Interpretations are the responsibility of the study authors. LWW is supported by the NIH/NCI Molecular Oncology Training Program (5T32CA203703-10). RHM is supported by grants from the NIH (K08CA263304), Gastric Cancer Foundation, the 2023 AACR-Debbie’s Dream Foundation Innovation and Discovery Grant (Grant Number 23-80-41-MOY), and the US Department of Defense (HT9425-24-1-0419). This publication was supported by the NIH/NCI Cancer Center Support Grant P30CA013696.

Footnotes

Conflict of Interest:

LWW, JP, SJ declare no conflict of interest. RHM reports advisory board or consulting fees from Amgen, Bristol Myers Squibb, BeOne Medicines, Gilead Sciences, and Nimbus Therapeutics, and research funding from Nimbus Therapeutics and Repare Therapeutics.

Prior Presentation: Presented in part at the ASCO Gastrointestinal Cancers Symposium, San Francisco, CA, January 8–10, 2026.

Data Availability Statement

The data analyzed in this study is publicly accessible through cBioPortal and the Synapse platform, subject to account registration and AACR Project GENIE terms of access. OncoKB annotation was performed using https://github.com/oncokb/oncokb-annotator. Derived summary data supporting the findings of this study are available within the article and the Supplementary Data. Custom code used for cohort curation, statistical analysis, and figure generation are available from the corresponding author upon reasonable request.

References

  • 1.Jayakrishnan T, Ng K: Early-Onset Gastrointestinal Cancers: A Review. JAMA, 2025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Koh B, Tan DJH, Ng CH, et al. : Patterns in Cancer Incidence Among People Younger Than 50 Years in the US, 2010 to 2019. JAMA Netw Open 6:e2328171, 2023 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Li Q, Weitz J, Li C, et al. : Smoking as a risk factor for colorectal neoplasms in young individuals? A systematic meta-analysis. Int J Colorectal Dis 38:114, 2023 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.O'Sullivan DE, Sutherland RL, Town S, et al. : Risk Factors for Early-Onset Colorectal Cancer: A Systematic Review and Meta-analysis. Clin Gastroenterol Hepatol 20:1229–1240 e5, 2022 [DOI] [PubMed] [Google Scholar]
  • 5.Li H, Boakye D, Chen X, et al. : Association of Body Mass Index With Risk of Early-Onset Colorectal Cancer: Systematic Review and Meta-Analysis. Am J Gastroenterol 116:2173–2183, 2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Yuan C, Kim J, Wang QL, et al. : The age-dependent association of risk factors with pancreatic cancer. Ann Oncol 33:693–701, 2022 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.McWilliams RR, Maisonneuve P, Bamlet WR, et al. : Risk Factors for Early-Onset and Very-Early-Onset Pancreatic Adenocarcinoma: A Pancreatic Cancer Case-Control Consortium (PanC4) Analysis. Pancreas 45:311–6, 2016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Nguyen LH, Liu PH, Zheng X, et al. : Sedentary Behaviors, TV Viewing Time, and Risk of Young-Onset Colorectal Cancer. JNCI Cancer Spectr 2:pky073, 2018 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Collaboration NCDRF: Worldwide trends in underweight and obesity from 1990 to 2022: a pooled analysis of 3663 population-representative studies with 222 million children, adolescents, and adults. Lancet 403:1027–1050, 2024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Pandeya N, Olsen CM, Whiteman DC: Sex differences in the proportion of esophageal squamous cell carcinoma cases attributable to tobacco smoking and alcohol consumption. Cancer Epidemiol 37:579–84, 2013 [DOI] [PubMed] [Google Scholar]
  • 11.Arnold M, Laversanne M, Brown LM, et al. : Predicting the Future Burden of Esophageal Cancer by Histological Subtype: International Trends in Incidence up to 2030. Am J Gastroenterol 112:1247–1255, 2017 [DOI] [PubMed] [Google Scholar]
  • 12.Li Y, Hahn AI, Laszkowska M, et al. : Clinicopathological Characteristics and Risk Factors of Young-Onset Gastric Carcinoma: A Systematic Review and Meta-analysis. Clin Transl Gastroenterol 15:e1, 2024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.You WC, Zhang L, Gail MH, et al. : Gastric dysplasia and gastric cancer: Helicobacter pylori, serum vitamin C, and other risk factors. J Natl Cancer Inst 92:1607–12, 2000 [DOI] [PubMed] [Google Scholar]
  • 14.Stoffel EM, Koeppe E, Everett J, et al. : Germline Genetic Features of Young Individuals With Colorectal Cancer. Gastroenterology 154:897–905 e1, 2018 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Lieu CH, Golemis EA, Serebriiskii IG, et al. : Comprehensive Genomic Landscapes in Early and Later Onset Colorectal Cancer. Clin Cancer Res 25:5852–5858, 2019 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Lumish MA, Walch H, Maron SB, et al. : Clinical and molecular characteristics of early-onset vs average-onset esophagogastric cancer. J Natl Cancer Inst 116:299–308, 2024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Oliveira C, Pinheiro H, Figueiredo J, et al. : Familial gastric cancer: genetic susceptibility, pathology, and implications for management. Lancet Oncol 16:e60–70, 2015 [DOI] [PubMed] [Google Scholar]
  • 18.Wu LW, Deshmukh SK, Wu S, et al. : Multi-omic characterization of early-onset esophagogastric cancer. NPJ Precis Oncol 9:241, 2025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Consortium APG: AACR Project GENIE: Powering Precision Medicine through an International Consortium. Cancer Discov 7:818–831, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Chakravarty D, Gao J, Phillips SM, et al. : OncoKB: A Precision Oncology Knowledge Base. JCO Precis Oncol 2017, 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Suehnholz SP, Nissan MH, Zhang H, et al. : Quantifying the Expanding Landscape of Clinical Actionability for Patients with Cancer. Cancer Discov 14:49–65, 2024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Subbiah V, Solit DB, Chan TA, et al. : The FDA approval of pembrolizumab for adult and pediatric patients with tumor mutational burden (TMB) >/=10: a decision centered on empowering patients and their physicians. Ann Oncol 31:1115–1118, 2020 [DOI] [PubMed] [Google Scholar]
  • 23.Boikos SA, Pappo AS, Killian JK, et al. : Molecular Subtypes of KIT/PDGFRA Wild-Type Gastrointestinal Stromal Tumors: A Report From the National Institutes of Health Gastrointestinal Stromal Tumor Clinic. JAMA Oncol 2:922–8, 2016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Miettinen M, Lasota J: Succinate dehydrogenase deficient gastrointestinal stromal tumors (GISTs) - a review. Int J Biochem Cell Biol 53:514–9, 2014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Ogobuiro I, Baca Y, Ribeiro JR, et al. : Multiomic Characterization Reveals a Distinct Molecular Landscape in Young-Onset Pancreatic Cancer. JCO Precis Oncol 7:e2300152, 2023 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Rousseau B, Foote MB, Maron SB, et al. : The Spectrum of Benefit from Checkpoint Blockade in Hypermutated Tumors. N Engl J Med 384:1168–1170, 2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Safaee Ardekani G, Jafarnejad SM, Tan L, et al. : The prognostic value of BRAF mutation in colorectal cancer and melanoma: a systematic review and meta-analysis. PLoS One 7:e47054, 2012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Taieb J, Zaanan A, Le Malicot K, et al. : Prognostic Effect of BRAF and KRAS Mutations in Patients With Stage III Colon Cancer Treated With Leucovorin, Fluorouracil, and Oxaliplatin With or Without Cetuximab: A Post Hoc Analysis of the PETACC-8 Trial. JAMA Oncol 2:643–653, 2016 [DOI] [PubMed] [Google Scholar]
  • 29.Samowitz WS, Sweeney C, Herrick J, et al. : Poor survival associated with the BRAF V600E mutation in microsatellite-stable colon cancers. Cancer Res 65:6063–9, 2005 [DOI] [PubMed] [Google Scholar]
  • 30.Elez E, Yoshino T, Shen L, et al. : Encorafenib, Cetuximab, and mFOLFOX6 in BRAF-Mutated Colorectal Cancer. N Engl J Med 392:2425–2437, 2025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Reiss KA, Soares KC, Torphy RJ, et al. : Treatment Innovations in Pancreatic Cancer: Putting Patient Priorities First. Am Soc Clin Oncol Educ Book 45:e473204, 2025 [DOI] [PubMed] [Google Scholar]
  • 32.Rodon J, Prenen H, Sacher A, et al. : First-in-human study of AMG 193, an MTA-cooperative PRMT5 inhibitor, in patients with MTAP-deleted solid tumors: results from phase I dose exploration. Ann Oncol 35:1138–1147, 2024 [DOI] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Supplementary Files
NIHMS2208556-supplement-Supplementary_Files.docx (1.2MB, docx)

Data Availability Statement

The data analyzed in this study is publicly accessible through cBioPortal and the Synapse platform, subject to account registration and AACR Project GENIE terms of access. OncoKB annotation was performed using https://github.com/oncokb/oncokb-annotator. Derived summary data supporting the findings of this study are available within the article and the Supplementary Data. Custom code used for cohort curation, statistical analysis, and figure generation are available from the corresponding author upon reasonable request.

RESOURCES