Molecular and clinicopathologic impact of GNAS variants across solid tumors

Abstract

Purpose:

The molecular drivers underlying mucinous tumor pathogenicity are poorly understood. GNAS mutations predict metastatic burden and treatment resistance in mucinous appendiceal adenocarcinoma. We investigated the pan-cancer clinicopathologic relevance of GNAS variants.

Patients and Methods:

We assessed 58,043 patients with MSK-IMPACT sequenced solid tumors to identify oncogenic variants, including GNAS, associated with mucinous tumor phenotype. We then performed comprehensive molecular analyses to compare GNAS-mutant (mut) and wild type (wt) tumors across cancers. Gene expression patterns associated with GNAS-mut tumors were assessed in a TCGA cohort. Associations between GNAS variant status and peritoneal metastasis, first-line systemic therapy response, progression-free survival (PFS) and overall survival (OS) were determined using a propensity-matched sub-cohort of patients with metastatic disease.

Results:

Mucinous tumors were enriched for oncogenic GNAS variants. GNAS was mutated in >1% of small bowel, cervical, colorectal, pancreatic, esophagogastric, hepatobiliary, and GI neuroendocrine cancers. Across these cancers, GNAS-mut tumors exhibited a generally conserved C-to-T mutation-high, aneuploidy-low molecular profile with co-occurring prevalent KRAS variants (65% of GNAS-mut tumors) and fewer TP53 alterations. GNAS-mut tumors exhibited recurrently co-mutated alternative tumor suppressors (RBM10, INPPL1) and upregulation of MAPK and cell surface modulators. GNAS-mut tumors demonstrate an increased prevalence of peritoneal metastases (Odds Ratio [OR] 1.7; 95% Confidence Interval [CI]: 1.1-2.5; P=0.006), worse response to first-line systemic therapy (OR 2.2; 95% CI: 1.3-3.8; P=0.003) and shorter PFS (median 5.6 versus 7.0 months; P=0.047). In a multivariable analysis, GNAS mutated status was independently prognostic of worse OS (Hazard Ratio [HR] 1.25; 95% CI: 1.01-1.56; adjusted-P=0.04).

Conclusion:

Across the assessed cancers, GNAS-mut tumors exhibit a conserved molecular and clinical phenotype defined by mucinous tumor status, increased peritoneal metastasis, poor response to first-line systemic therapy, and worse survival.

Introduction

Tumor mucin secretion is associated with features that facilitate metastatic spread including increased angiogenesis, extracellular membrane remodeling, and immunosuppression.^1-6 The drivers of mucinous tumors remain incompletely defined, yet several potential targets have been identified in rare malignancies. Appendiceal adenocarcinomas (AC) are particularly paradigmatic as these cancers secrete copious mucin and demonstrate peritoneal-trophic metastasis.^7-12

Mucinous ACs are enriched for gain-of-function mutations in the Guanine nucleotide binding protein, alpha stimulating activity polypeptide (GNAS) complex locus.^{7, 13} GNAS mediates downstream protein kinase A (PKA) signaling and may plausibly promote mucin secretion.^14-18 In AC, GNAS variants disproportionately occur with clonal KRAS alterations, suggesting that GNAS and KRAS are co-enabling .^{7, 13} Patients with GNAS-mutant (mut) mucinous ACs exhibit a significantly higher burden of peritoneal disease, increased chemoresistance, and worse survival compared to those with KRAS-mut/GNAS-wild-type(wt) tumors.⁷

GNAS-modulated signaling has been profiled in various entities ranging from intraductal papillary mucinous neoplasms (IPMNs) to rare lung cancers, but the broader functional and clinical significance of these uncommon alterations remain unknown.^{14, 15, 19, 20} In this study, we investigated the enrichment of GNAS alterations among mucinous cancers and further defined the pan-cancer clinicopathologic impact of aberrant GNAS signaling.

Methods

Patient Population

The main study cohort included patients from Memorial Sloan Kettering (MSK) who had malignant solid tumors profiled with the MSK-IMPACT next generation sequencing (NGS) panel between 4/2015 and 9/2023.²¹ Patients consented to retrospective evaluation for this study approved by the MSK Institutional Review Board. Clinicopathological features, including mucinous tumor status, were verified from the electronic medical record (Supplemental Methods). A second cohort was assessed from the publicly available The Cancer Genome Atlas (TCGA) RNA-sequencing (RNA-Seq) dataset. For both cohorts, tumors with mismatch repair deficiency by immunohistochemistry, MSIsensor score ≥3, or oncogenic DNA polymerase epsilon (POLE) or polymerase delta 1 (POLD1) variants were excluded.^22-24

Molecular Analysis

We initially evaluated OncoKB defined oncogenic variants disproportionately enriched in mucinous versus non-mucinous MSK-IMPACT samples after Benjamini-Hochberg (BH) multiplicity adjustment.²⁵ Subsequent analyses were performed in a Molecular Comparison Subcohort of non-AC cancers with a greater than 1% prevalence of oncogenic GNAS variants. AC was separately profiled for comparison.^{7, 26} We investigated associations between GNAS oncogenic variant status and tumor mutation burden (TMB), fraction genome altered (FGA), and co-occurring oncogenic variants. Single-base substitutions (SBS) were analyzed using MutationalPatterns.²⁷ Cancer-cell fraction and allele-specific clonality estimates (clonal versus subclonal) were calculated for KRAS and GNAS oncogenic variants using FACETS.²⁸ We predicted the order of KRAS and GNAS variant acquisition using a Bradley-Terry model that assessed each instance of a clonal dominating a subclonal variant for tumors with co-altered, discordant clonal events (i.e., KRAS clonal and GNAS subclonal).

Differentially expressed genes (DEGs) between GNAS-mut and GNAS-wt tumors were assessed per cancer type from the TCGA cohort (Supplemental Methods). The per-gene log-fold expression change and statistical significance was determined with Benjamini-Hochberg (BH) multiplicity adjustment. Consensus significant DEGs in at least two cancer types were evaluated for gene ontology.^{29, 30}

Clinical Outcomes

We used nearest-neighbor propensity matching to generate a balanced 2 (GNAS-wt) to 1 (GNAS-mut) matched Clinical Subcohort of patients with metastatic cancers from the Molecular Comparison Subcohort (Supplemental Methods). Propensity-matched variables included patient age, sex, primary tumor type, pertinent primary tumor location (i.e., right colon), and oncogenic KRAS status. Rates of radiographic tumor regression, stability, or growth were confirmed in patients with available imaging who underwent at least one full cycle of first-line chemotherapy. Primary analyses were conducted in overall cohorts of GNAS-mut versus GNAS-wt tumors, with exploratory sub-analyses performed per-cancer.

Progression-Free-Survival (PFS) after first-line therapy was defined as the time from treatment start until disease progression or death from cancer; censoring was performed at the time of last follow-up or if the patient discontinued therapy for other reasons such as toxicity. Exploratory analyses included tumor regression rates and PFS to specific anti-neoplastic therapies. Overall Survival (OS) was defined as the time from the diagnosis of metastatic disease until death from any cause with censoring at last follow-up. Peritoneal metastasis was verified based on radiographic and/or intraoperative pathologic findings.

Statistical Analyses

Continuous covariates were evaluated with Wilcoxon-Mann-Whitney testing. Categorical variables were compared using Fisher’s Exact. Odds ratios (OR) and confidence intervals for co-occurrence analyses were derived with pairwise Fisher’s Exact test with multiplicity-adjusted significance as per the BH method (P<0.05). Bradley Terry regression model point estimates and 95% confidence intervals were derived for all tumors and per-cancer. Survival curves were generated with the Kaplan Meier method and significant associations were determined using the log rank test. Multivariable Cox proportional hazards models were adjusted for clinically relevant covariates. All analyses were performed using R (v4.0.0). Statistical tests were two-sided with P<0.05 considered statistically significant.

Results

GNAS mutations are prevalent in mucinous tumors

The MSK Overall Cohort consisted of 58,043 patients with 47 different cancer types (Figure 1). Mucinous tumors were rare (n=534/58,043; 0.9%) and predominantly of GI origin.³¹ Notably, 8.5% (33/388) of cervical tumors were mucinous (Supplemental Table 1). Of all MSK-IMPACT interrogated variants, oncogenic mutations in GNAS exhibited the greatest disproportionate prevalence in mucinous (n=179/534; 33.5%) versus non-mucinous (n=341/57,509; 0.6%) tumors (OR: 84.5; BH-adjusted P<0.0001) (Figure 2A). We observed similar results in a sensitivity analysis without AC (Supplemental Figure 1). Oncogenic GNAS variants were found in 0.9% (n=520/58,043) of tumors and most often impacted the arginine 201 residue (n = 493/530; 93.0%: Figure 1B-1C).

Figure 1: Consort diagram for the MSK patient cohort.

Figure 2: GNAS mutations in solid tumors.

(A) Differential prevalence of molecular alterations in mucinous versus non-mucinous tumors assessed by Fisher’s Exact Method. Frequencies of genes significantly enriched in mucinous tumors (BH adjusted P < 0.05) are shown in an adjacent bar plot. Asterisks indicate statistical significance as follows: BH-adjusted *P < 0.05, **P < 0.01, and ***P < 0.001. (B) Prevalence of oncogenic GNAS variants across solid tumor types. (C) Lollipop plot illustrating the number (y-axis) and amino acid substitution (x-axis and color) of observed GNAS oncogenic variants.

Conserved molecular features of GNAS mutant tumors

We compared the molecular features of GNAS–mut versus GNAS–wt tumors for the seven non-appendiceal cancer types with an oncogenic GNAS variant prevalence >1% (“Molecular Comparison Subcohort”; Figure 1). Overall, GNAS-mut tumors exhibited a significantly lower median FGA (5.3% versus 12.1%; P<0.001) and slightly higher TMB (5.8 versus 4.4; P<0.001) [Supplemental Figure 2]. Aggregated SBS profiles per cancer type showed that GNAS-mut tumors had increased C-to-T transitions [Supplemental Figures 2-3]. C-to-T transitions constituted 39% (n=208/530) of all oncogenic GNAS variants. This overall molecular profile was generally conserved in 6 of the 7 assessed cancers except for small bowel, and recapitulated the profile seen in AC (Supplemental Figure 2B).

We next identified recurrent oncogenic variants that disproportionately co-occur with GNAS (Supplemental Figure 4). Overall, KRAS was mutated in 64.9% (n = 156/244) of GNAS-mut tumors versus 42.9% (n=5,837/13,605) of GNAS-wt tumors (OR 2.4; 95% CI: 1.8-3.1; BH-adjusted P<0.0001: Figure 3A). In contrast, TP53 variants were rarer in GNAS-mut (n=66/244; 27.5%) compared to GNAS-wt (n=8,928/13,605; 65.6%) tumors (OR 0.19; 95% CI: 0.15-0.26; BH-adjusted P < 0.0001). This profile was observed across cancers except for pancreatic and small bowel cancers (for KRAS) and cervical cancer (for TP53). Only 12/244 (4.9%) GNAS-mut tumors harbored concurrent OncoKB Level 1 (FDA-recognized biomarker) or Level 2 (standard of care biomarker) targetable alterations (queried 6/1/2023: Figure 3B). Most KRAS variants were p.G12D (21.3%: 52/244) or p.G12V (14.3%: 35/244); only 1-4% of tumors per cancer type harbored KRAS p.G12C variants (Supplemental Figure 5).

Figure 3: Co-alterations in GNAS-mutant solid tumors.

(A) Prevalence of oncogenic variants in genes disproportionately altered in GNAS-mutant or wild-type tumors in two or more different cancer types. Genes were chosen based on panel-wide analyses shown in Supplemental Figure 4. Alteration frequency is compared with Fisher’s Exact Method with Benjamini-Hochberg (BH) multiplicity adjustment. Asterisks indicate statistical significance adjusted for multiplicity as represented: BH-adjusted *P < 0.05, **P < 0.01, ***P < 0.001. (B) Clinically actionable alterations in GNAS-mutant tumors. OncoKB actionability level (queried 6/1/2023) is variant and cancer-specific for response to an FDA-approved drug in the respective indication (Level 1: FDA-recognized biomarker, Level 2: standard of care biomarker, Level 3A: biomarker with compelling clinical evidence). Remaining variants represent standard-of-care or investigational biomarkers predictive of therapy response in another indication (Level 3B), or variants with compelling pre-clinical biological evidence of predictive capacity (Level 4). The “Others” category reflects combined frequency of level 3B actionable alterations in ALK, BARD1, FANCA, TSCT, BRIP1, and NBIN. (C) Timing of GNAS and KRAS clonal and sub-clonal variants in co-altered tumors. Point estimates were calculated with Bradley Terry models of tumors with discordant (clonal vs sub-clonal) variants: clonal variants were defined as earlier events.

We assessed the clonal patterns of ninety-six tumors with co-altered oncogenic KRAS and GNAS variants. Among these, the average cancer-cell fraction (CCF) for KRAS variants was significantly higher (0.92; standard deviation [SD] 0.13) than for GNAS variants (0.81; SD 0.26; Wilcoxon P=0.006: Figure 3C). This association was conserved in sub-analyses of colorectal (n=44; P=0.005) and pancreatic cancers (n=32; P=0.21). Most co-altered tumors had concordant GNAS and KRAS clonality (both clonal in 66/96; both subclonal in 2/96). In 82% of discordant cases (n=23/28), KRAS variants were clonal while GNAS variants were subclonal. We used a Bradley-Terry model to estimate time to clonal dominance and found that GNAS variants were significantly more likely to occur after KRAS in the overall cohort (point estimate: −1.53; 95% CI: −2.62 to −0.64; P=0.002) as well as in colorectal cancers (n = 19; point estimate: −1.67; 95% CI: −3.13 to −0.57; P=0.008: Figure 3C).

Gene expression patterns in GNAS mutated cancers

Out of 1,149 TCGA patients with colorectal, esophageal, hepatocellular, or pancreatic cancer and RNA-sequencing data, 1131 (98.4%) were GNAS-wt and 18 (1.6%) were GNAS-mut (Supplemental Table 2). In total, 22 genes were significantly (BH-adjusted p<0.05) up-regulated or down-regulated in GNAS-mut versus GNAS-wt tumors in two or more cancer types (Figure 4; Supplemental Table 3). Genes recurrently up-regulated in GNAS-mut tumors were active in membrane cadherin structural binding (S100P and PROM1), protein synthesis (RPS28), and RAS modulation (RASSF6). Recurrently down-regulated genes included membrane-associated tumor suppressors notably associated with GPCR signaling (KCNK2, GAST, ODAM), cell membrane binding and adhesion (CDH12, CLDN10), and negative WNT regulation (WIF1) [Figure 4].

Figure 4: Recurrent differentially expressed genes in GNAS-mutant tumors.

Gene expression RNA sequencing data was obtained from the TCGA cohort. All displayed genes exhibited significant (Benjamini Hochberg adjusted p < 0.05) differential expression in two or more different cancer types with a log2 expression ratio ≥1 (up-regulated, green) or ≤ –1 (down-regulated, red) in GNAS-mut versus GNAS-wild-type tumors. Gene function was assigned based on common intersecting Gene Ontology classifiers.

GNAS mutated tumors exhibit adverse clinical outcomes

We next evaluated the clinical relevance of GNAS mutations in a 2 (GNAS-wt) to 1 (GNAS-mut) propensity matched cohort of patients with stage IV disease derived from the Molecular Subcohort (Table 1; Figure 1). This Clinical Subcohort included 336 GNAS-wt patients and 168 GNAS-mut patients and was balanced for age, sex, cancer type, primary tumor location (if applicable), and KRAS variant frequency (Table 1; Supplemental Table 4; Supplemental Figure 6). Overall, GNAS-mut tumors were disproportionately mucinous (n=71/166; 42.8%) compared to GNAS-wt tumors (n=49/328; 14.9%; OR: 4.2; 95% CI: 2.7-6.7; P<0.0001: Figure 5A; Supplemental Table 5). In per-cancer sub-analyses, this association was seen for each cancer type except hepatobiliary and GI neuroendocrine cancers. GNAS-mut tumors also exhibited an increased prevalence of peritoneal metastasis (n=97/167; 58.1%) compared to GNAS-wt tumors (n=151/336; 44.9%; OR 1.7; 95% CI: 1.1-2.5; P=0.006: Figure 5B). GNAS-mut colorectal cancers were significantly more likely to spread to the peritoneum (60.9% versus 40.2%; OR 2.3; 95% CI: 1.1-5.1; P=0.03). Similar trends were observed in pancreatic, small bowel and hepatobiliary cancer-specific sub-analyses although these did not reach statistical significance (Supplemental Table 5).

Table 1.

Clinicopathologic characteristics of the propensity matched cohort

	GNAS WT (N = 336)	GNAS Mut (N = 168)
Age (years) - Median (Interquartile Range)	67 (60, 73)	67 (58, 73)
Sex – N (%)
Female	178 (53%)	84 (50%)
Male	158 (47%)	84 (50%)
Race – N (%)
Asian	34 (10%)	17 (10%)
Black	29 (8.6%)	2 (1.2%)
White	245 (73%)	141 (84%)
Other	10 (3.0%)	4 (2.4%)
Unknown	18 (5.4%)	4 (2.4%)
Tumor Differentiation- N (%)
Well	17 (5.1%)	5 (3.0%)
Moderately	154 (46%)	72 (43%)
Moderately to poorly	36 (11%)	24 (14%)
Poorly	74 (22%)	44 (26%)
Unknown	55 (16%)	23 (14%)
KRAS Mut – N (%)	202 (60%)	106 (63%)
Cancer type - N (%)
Colorectal	92 (27%)	46 (27%)
Pancreatic	110 (33%)	55 (33%)
Esophagogastric	54 (16%)	27 (16%)
Hepatobiliary	28 (8.3%)	14 (8.3%)
GI Neuroendocrine	14 (4.2%)	7 (4.2%)
Cervical	22 (6.5%)	11 (6.5%)
Small Bowel	16 (4.8%)	8 (4.8%)
Primary tumor site – N (%)
Left colon	41 (12.2%)	20 (11.9%)
Right colon	51 (15.2%)	26 (15.5%)
Esophagus	5 (1.5%)	5 (3.0%)
Gastroesophageal junction	6 (1.8%)	3 (1.8%)
Stomach	43 (12.8%)	19 (11.3%)
Colorectal NET	3 (0.9%)	4 (2.4%)
Gastric NET	1 (0.3%)	2 (1.2%)
Pancreatic NET	6 (1.8%)	1 (0.6%)
Small bowel NET	4 (1.2%)	0 (0.0%)
Cholangiocarcinoma	13 (3.9%)	8 (4.8%)
Gallbladder	6 (1.8%)	2 (1.2%)
Hepatocellular	8 (2.4%)	4 (2.4%)

Figure 5: Clinical characteristics and outcomes of patients with GNAS mutant tumors.

For patients with GNAS-mut and GNAS-wt tumors, the frequency of (A) mucinous tumor differentiation, (B) peritoneal metastasis, and (C) response to first line systemic therapy are shown for the overall analysis and per-cancer sub-analyses. The number with each feature and overall number of patients is presented next to each bar plot. Asterisks indicate statistical significance as follows: *P < 0.05, **P < 0.01, and ***P < 0.001. Kaplan-Meier curves show (D) progression-free survival (PFS) after first-line systemic therapy and (E) overall survival (OS) in eligible patients with metastatic GNAS-mut and GNAS-wt tumors propensity-matched for clinicopathologic characteristics (Table 1). P-values for E are calculated with the log-rank method. (F) Multivariable Cox regression model for OS. Hazard ratios (HR) for death are shown together with adjusted P values.

We evaluated the treatment sensitivity of GNAS-mut versus GNAS-wt tumors for patients who received similar first-line systemic therapies and had evaluable disease response (Supplemental Table 6). Overall, GNAS-mut tumors were significantly less likely to experience tumor regression after first-line systemic therapy than GNAS-wt tumors (35.7% versus 55.1%; OR 2.2; 95% CI: 1.3-3.8; P=0.003: Figure 5C). In cancer-specific sub-analyses, GNAS-mut colorectal cancers were significantly less likely to regress after first-line systemic therapy (28.0% versus 72.7%; P=0.0001). This trend was observed for all cancer types except esophagogastric cancer. Patients with GNAS-mut tumors had significantly worse PFS (median 5.6 months; 95% CI: 4.6 - 7.1 months) after first-line systemic therapy compared to patients with GNAS-wt tumors (median 7.0 months; 95% CI: 6.2-7.9 months; log-rank P=0.047: Figure 5D). In a multivariable sensitivity analysis adjusting for age and tumor type, the effect size of this association was maintained (HR: 1.31; CI: 0.99–1.72; adjusted-P = 0.056) [Supplemental Table 7]. We performed exploratory analyses for specific treatment types and found that this PFS trend was conserved in analyses of cancers treated with fluoropyrimidine-containing or platinum-containing regimens (Supplemental Figure 7A-B). Two patients with RAS-wt/GNAS-mut L-sided CRC exhibited no response and numerically worse PFS after EGFR-inhibition compared to 5 patients with RAS-wt/GNAS-mut tumors (median PFS 2.4 vs 6.4 months; log-rank P=0.08: Supplemental Figure 7D-E).

In the propensity matched cohort, patients with GNAS-mut tumors had significantly worse OS (median 16.6 months; 95% CI: 14.8-19.2 months) than those with GNAS-wt tumors (median 21.6 months; 95% CI: 18.1-25.3 months; log-rank P=0.019: Figure 5E). A multivariable analysis demonstrated that GNAS-mut tumor status was independently associated with worse OS irrespective of cancer type or patient age (HR: 1.25; 95% CI: 1.01–1.56; adjusted-P =0.04: Figure 5F). In a sensitivity analysis that stratified patients by GNAS variant status and mucinous tumor differentiation, patients with GNAS-mut tumors exhibited worse median OS than those with GNAS-wt tumors for both mucinous (20.7 versus 28.3 months) and non-mucinous (15.3 versus 19.3 months) tumor subtypes (log-rank P=0.0005) [Supplemental Figure 8A]. Similarly, patients with GNAS-mut tumors exhibited worse median OS compared to patients with GNAS-wt tumors among those with (15.8 versus 18.2 months) and without (19.0 versus 26.1 months) peritoneal metastasis (log-rank P=0.014: Supplemental Figure 8B).

Discussion

Mucinous tumors are associated with adverse patient outcomes across multiple cancer types, but the molecular drivers underlying their pathogenicity are relatively unknown. Previous studies have shown that GNAS-mutant mucinous AC constitutes a distinct subtype with nearly uniform KRAS-mut/TP53-wt status, increased peritoneal metastases, and chemoresistance.^{7, 13, 32} Our integrated molecular and clinical analysis of the largest sequenced dataset of non-appendiceal GNAS-mut tumors extends this profile to define GNAS as a clinically relevant pan-cancer driver.

Our study supports this conclusion through several main findings. First, tumors with oncogenic GNAS variants exhibit a distinct, recurrent molecular profile. Across diverse gastrointestinal and cervical cancers, GNAS-mut tumors harbor a genomic signature with frequent C-to-T transitions, which may possibly be due to age-related spontaneous deamination or oxidative stress.^{33, 34} GNAS-mut tumors are also characterized by reduced aneuploidy compared to their GNAS-wt counterparts, which is akin to AC. Analyses of dual-altered clones show disproportionately high rates of GNAS variants as secondary events in KRAS-mut clones suggesting co-enabling potentiation. Another defining feature of GNAS-mut tumors is the strikingly low frequency of TP53 alterations, which typically occur late in cancer progression and are associated with permissive tumor aneuploidy. GNAS-mut tumors may commit to a molecular lineage less dependent on TP53 aberration potentially through the inactivation of alternative tumor suppressors such as RMB10 and INPPL1, which we found were enriched in GNAS-mut tumors across multiple cancers. These variants have been associated with TP53 mutual-exclusivity, EGFR modulation, and down-stream activation of PI3K-AKT and MAPK-associated pro-growth signals.^35-40 Our per-cancer analysis strategy defines these recurrent GNAS-mut associated signals for future validation.

Second, GNAS-mut tumors exhibit a mucinous tumor phenotype with a higher incidence of peritoneal metastasis. Across over 50,000 sequenced tumors, GNAS variants exhibited the highest disproportionate enrichment of any variant in mucinous versus non-mucinous cancers. In a propensity-matched cohort of patients with stage IV cancers balanced for KRAS status, tumor type and location, age, and sex, GNAS-mut tumors were more likely to be mucinous and spread to the peritoneum. Peritoneal metastases are associated with high morbidity and mortality as well as notoriously high false-negative radiographic detection rates.^41-45 Patients with GNAS-mut tumors may benefit from additional diagnostic attention to the peritoneal cavity, including potential laparoscopic evaluation, given the risk of peritoneal spread. The implications of GNAS-mediated signaling behind this phenotype remain opaque, yet our study newly contextualizes prior findings. GNAS and KRAS mutations are prevalent in early tumors, such as low-grade appendiceal neoplasms and IPMNs, that are highly mucinous and exhibit progressive, chemo-resistant growth that may mimic the biology of the more advanced adenocarcinomas in our study.^{13, 15, 46, 20, 47} GNAS-mut silencing in IPMNs results in complicated, tissue-specific patterns of mucin expression without consensus mediators.⁴⁸ Importantly, our recurrent gene expression data derived from a small cohort of rare GNAS-mut tumors suggests that across cancers, GNAS-mut tumors upregulate consensus biomolecular synthesis pathways and cadherin-associated factors that may enable enhanced secretory and metastatic functionality.^49-53 These findings resonate with the correlation between GNAS and E-Cadherin activation in a breast cancer microarray model associated with pro-metastatic epithelial-to-mesenchymal transitions.⁵⁴ Our hypothesis-generating results derived from recurrent signals across different tumor histologies point to promising new pathways that should be tested in dedicated experiments.

Finally, we demonstrate in a large, balanced multi-cancer cohort that GNAS is a clinically relevant predictive and prognostic biomarker. In multivariable models of propensity-matched stage IV cancers, GNAS-mut status was significantly prognostic of poor OS. We show in our balanced cohort and sensitivity analyses that this prognostic impact was independent of KRAS mutations, mucinous tumor type, or peritoneal metastasis. GNAS-mut tumor mortality may be a result of increased treatment resistance. Patients with GNAS-mut tumors exhibited significantly worse response to first line systemic therapy and decreased PFS. Our exploratory analyses suggest that resistance may specifically impact fluoropyrimidine, platinum, and/or MAPK-targeted therapies. These preliminary findings in small patient populations require prospective validation.

Overall, our data suggest that new therapies may be needed for GNAS-mut tumors. Although KRAS inhibitors may be effective in highly KRAS-co-altered, GNAS-mut tumors, GNAS-mobilized resistance is possible. This study provides a rationale for therapeutic GNAS–inhibition, which could impact 1% of all solid tumor patients and may be particularly important in the management of GNAS-mut enriched tumors, such as AC. A recent study by Dai et al identified two promising cyclic peptides that inhibit GNAS.¹⁶ However, the in vivo potencies of these compounds were limited, and neither were fully mutant-specific. Hereditary disorders of GNAS inactivation provide insight into the possible adverse effect profiles of GNAS inhibitors. For instance, Albright’s hereditary osteodystrophy is associated with clinically significant endocrinopathies that can cause growth and neurologic defects.⁵⁵ Highly specific GNAS-inhibitors may mitigate the risk of serious toxicity associated with GNAS-deficiency in non-cancerous cells.

This study has several limitations. First, the targeted NGS platform MSK-IMPACT identifies somatic exonic alterations in a predefined panel of cancer-related genes. We therefore cannot exclude the possibility that GNAS-mut tumors harbor additional co-variants of significance. Second, we focused our molecular and clinicopathologic analyses of GNAS mutant tumors on the seven non-AC types with GNAS variant levels >1%, which predominantly included GI tumors. Third, given the overall rarity of GNAS mutations, our sub-analyses were not powered to detect statistically significant associations in less common cancer types such as small bowel cancer and GI neuroendocrine tumors even when their estimated direction and magnitude paralleled the overall cohort. Finally, the primary cohort for this study comes from a single large academic cancer center (MSK) and is therefore subject to the possibility of inherent selection bias. Although we affirm associations in discrete cancer-type analyses as well as in the overall combined MSK population, additional validation in a large external cohort would further support our findings.

In summary, we found that a subset of tumors harboring oncogenic GNAS variants exhibit a conserved molecular and clinicopathologic profile characterized by mucinous differentiation, peritoneal metastasis, potential resistance to systemic therapy, and poor overall survival. GNAS inhibition may represent a promising new therapeutic strategy.

Supplementary Material

PV Data Supplement_1

Supplemental Table 1: Clinicodemographic characteristics of the overall MSK cohort

Supplemental Table 2: Clinicodemographic characteristics of the TCGA cohort

Supplemental Table 3: Genes significantly up-regulated or down-regulated for GNAS mutant versus GNAS wild-type tumors

Supplemental Table 4: Clinicodemographic characteristics of the propensity matched cohort of patients with stage IV cancers

Supplemental Table 5: Odds ratio for mucinous tumor differentiation, peritoneal metastasis, and tumor regression after first-line therapy in the propensity matched cohort of patients with stage IV cancers

Supplemental Table 6: Clinicodemographic characteristics including treatment regimen for patients from the PFS analysis

Supplemental Table 7: Cox multivariable analysis for progression free survival

PV Data Supplement_2

Supplemental Figure 1: Molecular landscape of mucinous versus non-mucinous non-appendiceal solid tumors. Volcano plot illustrating the differential prevalence of molecular alterations in mucinous and non-mucinous non-appendiceal solid tumors. For genes that were significantly enriched in tumors with mucinous differentiation, their prevalence in mucinous and non-mucinous tumors is shown in the adjacent bar plot. Asterisks indicate statistical significance as follows: Benjamini-Hochberg adjusted *P < 0.05, **P < 0.01, and ***P < 0.001.

Supplemental Figure 2: Molecular features of GNAS-mutant solid tumors. (A) Features of tumors from the main MSK Molecular Subcohort in cancer types with a prevalence of GNAS oncogenic variants greater than or equal to 1%. Tumors without estimable FGA (n = 75/13,849; 0.5%) or TMB (n = 303/13,849; 2.2%) were removed from the respective comparisons. (B) Comparative assessment of appendiceal adenocarcinomas derived from the Overall MSK cohort and previously characterized.⁷ All analyses include the FGA, TMB, and nucleotide base substitution patterns of GNAS-mutant (Mut, orange) versus GNAS-wild type (WT, blue) tumors. All tumors were profiled with MSK IMPACT and statistical comparisons were performed with the Mann-Whitney-U test.

Supplemental Figure 3: Single base substitution signatures in GNAS mutant cancers. (A) Single base substitutions (SBS) for aggregated variants from GNAS-mut versus GNAS-wt tumors per cancer. The number of variants considered in each analysis is indicated (n). (B) Relative contribution of C-to-T alterations per cancer type in GNAS-wt and GNAS-mut tumors. Each cancer type is indicated with a point that represents the pooled variants of all tumors of that type (as shown in Supplemental Figure 3A). Box plots represent median, interquartile range (25 - 75%), and outliers.

Supplemental Figure 4: Co-mutated variants in GNAS-mutant versus GNAS-wild-type tumors per cancer type. Prevalence of oncogenic variants in genes disproportionately altered in GNAS-mutant (red) or wild-type (blue) tumors both overall (A) and per-cancer (B-H). Odds ratios (log2 scale) and -log10(adjusted p-values) were calculated Fisher’s Exact Method with Benjamini-Hochberg multiplicity adjustment.

Supplemental Figure 5: Co-occurring KRAS gene variants in GNAS-mutant solid tumors. The relative prevalence of each KRAS amino-acid change is demonstrated by circle diameter.

Supplemental Figure 6: Distribution of estimated propensity scores in the propensity-matched cohort. Propensity scores were determined for a nearest neighbor propensity score matching analysis and were estimated via logistic regression.

Supplemental Figure 7: Response to anti-neoplastic therapies in patients with GNAS-mutant versus GNAS-wild-type tumors. (A) Progression-free survival of patients treated with fluoropyrimidine containing chemotherapies (5-FU, FUDR, capecitabine containing) (B). Progression-free survival of patients treated with platinum containing chemotherapies (carboplatin, cisplatin, oxaliplatin) (C) Progression-free survival of patients treated with taxane containing chemotherapies (paclitaxel or nab-paclitaxel). Details of each therapy are shown in Supplemental Table 6. (D) Regression vs no-regression after anti-EGFR-based therapy for patients with left-sided GNAS-mut (n = 2) and GNAS-wt (n = 5) colorectal tumors. (E) Kaplan-Meier curve of progression free survival (PFS) based on GNAS mutation status in patients who received anti-EGFR-based therapy. P-values are defined using the log-rank test.

Supplemental Figure 8: Overall survival of patients with GNAS oncogenic variants stratified by mucinous tumor status or peritoneal metastatic incidence. Kaplan-Meier curves show overall survival (OS) for patients classified by their GNAS mutation status and stratified by (A) mucinous tumor differentiation and (B) peritoneal metastasis status. Abbreviations are as follows: with (“w/”), without (“w/o”), and metastasis (Mets). P-values are defined using the log-rank test.

Context Summary:

Key Objective:

We aimed to comprehensively define the molecular and clinicopathologic features of GNAS-mutant tumors in a large cohort of patients with diverse cancers.

Knowledge generated:

Across the evaluated cancers, tumors with oncogenic GNAS variants were disproportionately mucinous and exhibited a conserved molecular profile with frequent co-occurring KRAS mutations. Patient with GNAS-mutated tumors were at increased risk of peritoneal metastasis, reduced benefit from first-line systemic treatment, and worse overall survival.

Relevance (written by Dr. Gary K. Schwartz):

By using molecular profiling, this study identifies a unique subset of mucinous tumors which harbor GNAS mutations that carry a poor prognosis and for which new therapies are needed.

Data Sharing:

Data will be made available upon reasonable request.