Comprehensive Molecular Characterization of Gallbladder Carcinoma and Potential Targets for Intervention

Abstract

Purpose:

Gallbladder Carcinoma (GBC) is an uncommon and aggressive disease that remains poorly defined at a molecular level. Here, we aimed to characterize the molecular landscape of GBC and identify markers with potential prognostic and therapeutic implications.

Experimental Design:

GBC samples were analyzed using the MSK-IMPACT platform (targeted NGS assay that analyzes 505 cancer-associated genes.) Variants with therapeutic implications were identified using OncoKB database. The associations between recurrent genetic alterations and clinicopathologic characteristics (Fisher’s exact tests) or overall survival (univariate Cox regression) were evaluated. P-values were adjusted for multiple testing.

Results:

Overall, 244 samples (57% primary tumors and 43% metastases) from 233 patients were studied (85% adenocarcinomas, 10% carcinomas with squamous differentiation, and 5% neuroendocrine carcinomas.) The most common oncogenic molecular alterations appeared in the Cell Cycle (TP53 63% and CDKN2A 21%) and RTK_RAS pathways (ERBB2 15% and KRAS 11%). No recurrent structural variants were identified. There were no differences in the molecular landscape of primary and metastasis samples. Variants in SMAD4 and STK11 independently associated with reduced survival in patients with metastatic disease. Alterations considered clinically actionable in GBC or other solid tumor types (e.g., NTRK1 fusions or oncogenic variants in ERBB2, PIK3CA, or BRCA1/2) were identified in 35% of patients; and 18% of patients with metastatic disease were treated off-label or enrolled in a clinical trial based on molecular findings.

Conclusions:

GBC is a genetically diverse malignancy. This large-scale genomic analysis revealed alterations with potential prognostic and therapeutic implications and provides guidance for the development of targeted therapies.

Introduction

Gallbladder carcinoma (GBC) is an uncommon but highly fatal malignancy. Due to the anatomic location of this organ and the absence of specific symptoms early in the course of the disease, the majority of cases are diagnosed incidentally at advanced clinical stages and the patients frequently relapse after surgery.(1,2) Risk factors for the development of GBC are mainly associated with chronic gallbladder inflammation, including gallstone disease, hyalinizing cholecystitis, chronic infection with gram-negative organisms e.g., Salmonella typhi and Helicobacter bilis, obesity, and congenital developmental pancreatobiliary abnormalities.(3–6) It is believed that heterogeneity in the regional frequency of these risk factors accounts for the pronounced variation in GBC incidence across geographic regions, including high rates in certain South American and Asian countries, as well as Native American populations within the United States.(7–9)

Large-scale studies assessing the molecular mechanisms underlying the pathogenesis of GBC are still limited due to the low incidence of this tumor. Two main molecular pathways have been described: one governed by TP53 alterations and enriched in patients with a history of cholelithiasis, and the other governed by KRAS alterations mostly in patients with pancreatobiliary duct maljunction.(4,7) Previous studies in GBC cohorts have suggested potential genetic alterations associated with poor prognosis in this tumor type, including variants in ERBB2, ERBB3,(10) and ELF3.(11) However, there is still limited information about other molecular alterations in this cancer type, including recurrent structural variants (SV), or microbiome signatures. The identification of susceptibility genes, as well as the elucidation of the role of the microbiome in the pathogenesis of GBC is of relevance to expand treatment options and improve the prognosis of patients with this aggressive neoplasm. In this study, we aim to characterize the molecular landscape of a large cohort of GBC to correlate recurrent genetic alterations with clinicopathologic characteristics and identify markers with potential prognostic and therapeutic implications.

Methods

Specimen selection and clinical information collection

This study was conducted in accordance with the Declaration of Helsinki and was performed following MSK Institutional Review Board approval. This protocol allows for the retrieval of clinical, pathologic, and molecular data from patients who signed an informed written consent or with a waiver of consent. Two hundred and forty-four specimens from 233 unique patients with GBC acquired from 2014–to 2021 were identified in the Diagnosis Molecular Pathology archives and were included in this study. When more than one sample was available for an individual patient, only the primary tumor (if available) or the earliest collected metastasis was included in the analysis. However, if both the primary and a metastatic lesion were available, the latter was also included in the analyses comprising metastatic lesions only.

Clinicopathologic information was retrieved from electronic medical records, and included patient’s age, race/ethnicity, therapies received, clinical stage at the time of diagnosis and sample collection, sample type (i.e., primary vs. metastatic), histologic tumor subtype, tumor grade, T stage, N stage, and presence of cholelithiasis. Response to treatment was determined by the treating physician based on serial radiologic assessment performed following therapy initiation.

Sample preparation

The diagnosis of GBC was confirmed by a board-certified pathologist sub-specialized in gastrointestinal pathology, and a representative formalin-fixed, paraffin-embedded block was chosen from each specimen to perform next-generation sequencing (NGS). Macro-dissection was performed to enrich for tumor when needed and appropriate. Samples with estimated tumor purity <10% based on histopathologic assessment and that were not amenable to enrichment by manual macro-dissection were excluded from the study. A blood sample was also obtained from each patient and was used as a patient specific normal control. DNA was extracted using standard validated protocols.

Genetic analysis

MSK-IMPACT, an in-house hybridization capture-based NGS assay for targeted deep sequencing was used to identify somatic genomic alterations in this cohort.(12) Testing is performed bases on both tumor tissue and a patient specific normal control (blood) to ensure all variant calls are somatic in nature. Briefly, custom DNA probes were designed to capture all exons and selected introns of 341 to 505 (depending on the NGS panel version) oncogenes, tumor suppressor genes, and members of pathways deemed potentially actionable by targeted therapies. The standard DNA input for library construction was 250 ng; for selected cases, a lower input was used (50–250 ng) if the DNA quantity was limited. Captured libraries from both tumor and normal samples were sequenced on Illumina HiSeq 2500 or NovaSeq 6000 instruments. MuTect,(13) Vardict,(14) and SomaticIndelDetector,(15) were used for detecting point mutations and indels. Copy number alterations (CNA) were identified by comparing loess-normalized sequence coverage of targeted regions in a tumor sample relative to a standard diploid non-tumor sample. Structural variants were called by Delly.(16) Variants were annotated using VEP, and germline events were filtered out using a patient-matched (blood) normal.

Genetic alterations were classified according to OncoKB(17) by 1) oncogenic effect, four categories i.e., oncogenic, likely-oncogenic, neutral, or inconclusive; 2) biological effect, nine categories i.e., Gain-of-Function (GOF), Loss-of-Function (LOF), Switch-of-Function (SOF), likely GOF, likely LOF, likely SOF, neutral, likely neutral or inconclusive; and 3) therapeutic actionability, scale of 1–4 i.e., levels 1–2 alterations indicating standard therapeutic interventions and levels 3–4 including investigational therapeutic alterations that may direct a patient toward a clinical trial relevant to that biomarker. The OncoKB levels of evidence map 1:1 to both the AMP/ASCO/CAP consensus recommendation of variant categories(18) and ESMO’s ESCAT Tiers.(19) Specifically, OncoKB levels 1 and 2 map to AMP/ASCO/CAP Level A, OncoKB level 3A maps to AMP/ASCO/CAP Level B, OncoKB level 3B maps to AMP/ASCO/CAP Level C and OncoKB level 4 maps to AMP/ASCO/CAP Level D (https://www.oncokb.org/levels#version=AAC). Additionally, OncoKB levels of evidence 1 and 2 map to ESCAT Tier I, OncoKB level of evidence 3A maps to ESCAT Tier II and OncoKB levels of evidence 3B and 4 maps to ESCAT Tiers III and IV.

Contributions of different mutation signatures were identified for each sample according to the distribution of the 6 substitution classes (C>A, C>G, C>T, T>A, T>C, T>G) and the bases immediately 5′ and 3′ of the mutated base, as described previously.(20) In this manuscript, we focused on 30 previously described signatures (v2 - March 2015).(21) A sample was determined to have a dominant signature if it harbored a Tumor Mutation Burden (TMB) >13.8 and >40% of the observed mutations were attributable to a given signature.(22) We also analyzed the mechanisms and patterns of somatic alterations in ten canonical pathways as defined by Sanchez-Vega et al. i.e., cell cycle, Hippo, Myc, Notch, Nrf2, PI-3-Kinase/Akt, RTK-RAS, TGFβ signaling, p53, and β-catenin/Wnt.(23) We used Oncodrive’s algorithm oncodriveCLUST (24) to identify potential cancer driver genes from a given mutant allele fraction.

Microsatellite stability status [Microsatellite Instability-High (MSI-High) or Microsatellite Stable (MSS)] was determined for each sample using MSIsensor algorithm,(25) MiMSI,(26) or IDYLLA MSI.(27) Tumors were classified as MSI-High when any of these assays delivered a positive result. TMB was calculated by dividing the number of somatic missense mutations by the size of the captured region in our NGS assay.

Microbiome pipeline of MSK-IMPACT

For tumor microbiome characterization, we used an in-house analysis pipeline that has been previously validated.(28,29) Briefly, the sequenced DNA reads that did not map to the targeted regions (8–12% of the typical 8 million reads sequenced) were separated on a binary alignment map (BAM) file with samtools (v1.9) and saved as fasta files. These were analyzed with the basic local alignment search tool (BLAST) algorithm blastn (v2.5.0) to find sequences with high homology across repositories of known DNA sequence databases. The nucleotide database from NCBI <ftp://ftp.ncbi.nlm.nih.gov/blast/db/FASTA/nt.gz> (last accessed June 20, 2020) was utilized to identify and quantify the reads derived from microbial species in the sequenced tissue. The blastn output files were mapped to taxonomy IDs from the National Center for Biotechnology and Information (NCBI) database using ClassifyBlast program from the KronaTools package.(30) An internally developed R (executed with R v3.5.1) script was used to quantify the reads for each genus and species and classify the taxonomies as either fungi, bacteria, or viruses. At least two reads needed to be present to consider a sample positive for a given microorganism. Bacteriophages were excluded from the analysis.

Statistical analysis

Clinical and pathologic variables were included in the analysis. We assessed the correlation of mutated genes and canonical pathways with clinical and pathologic variables using Fisher’s exact test. To test the association of clinicopathologic (i.e., age, stage at time of collection, tumor grade, and tumor histology) and molecular features with overall survival, univariate Cox regression analysis was performed separately on primary and metastatic tumors. Univariate Cox regression analysis were adjusted for clinical variables significantly associated with survival. Overall survival was defined as time from the date of diagnosis (diagnosis of primary disease for primary tumors and diagnosis of metastatic disease for metastatic tumors) to death or last follow-up. For metastatic tumors, samples were collected at the date of diagnosis. The majority of primary tumor samples (90%) were collected within 1 month of diagnosis; to account for survival bias in the remaining 10% of samples, survival estimates were calculated with left truncation.(31) Only molecular events with a frequency higher than 5% were included in this analysis. P-values were adjusted for multiple testing with the False Discovery Rate (FDR.)(32)

For the microbiome analysis, a Shannon alpha diversity was calculated for each sample as previously described.(29,33) T-tests were performed to analyze differences across analyzed groups. Odds ratio plots were created using a univariate chi-square analysis to compare the likelihood of bacteria enrichment according to clinical, pathologic, and molecular characteristics. The p-values were statistically adjusted for multiple testing using the FDR method.

Data availability

The data generated in this study are publicly available in cBioPortal at https://www.cbioportal.org/study/summary?id=gbc_mskcc_2022 and Synapse database at https://doi.org/10.7303/syn38174088. The sample list including a unique sample per patient (only the primary tumor (if available) or the earliest collected metastasis) is indicated in Table S1.

Results

Cohort

Two-hundred and forty-four samples from 233 patients were analyzed, Table 1. Across the unique 233 samples (i.e., primary when available or earliest collected metastasis), 199 (85%) were adenocarcinomas, 23 (10%) were carcinomas with squamous differentiation (20 adenosquamous carcinoma and 3 squamous cell carcinomas), 6 (3%) were small cell neuroendocrine carcinomas, and 5 (2%) were poorly differentiated carcinomas with neuroendocrine differentiation (3 mixed adenocarcinoma and neuroendocrine carcinoma and 2 large cell neuroendocrine carcinomas). Of these, 57% were primary lesions. The most common metastasis site was the liver (49%) followed by lymph nodes (13%) and peritoneum (10%). Most patients (63%) were Caucasian, followed by African American (12%), and Asian (12%). Most patients were diagnosed with stage IV (67%) and stage III (18%) disease at the time of surgery/collection. For the cohort of patients with primary tumors, with a median follow up time from sample collection among survivors of 26 (IQR 13–46) months, the median overall survival was 38 (95% CI 30–46) months. For the metastatic patients, with a median follow-up time among survivors of 27 (IQR 15–42) months, the median overall survival was 18 (95% CI 12–24) months.

Table 1.

Clinicopathologic characteristics of the cohort

Clinical characteristics		Number (%)
Sex	Female	160 (69)
	Male	73 (31)
Ethnicity/race	Caucasian/White	147 (63)
	African American/Black	29 (12)
	Asian	29 (12)
	Hispanic	9 (4)
	Unknown	19 (8)
Age (at collection)	Median (range)	66 (37–90)
Stage (at collection)	I	3 (1)
	II	34 (15)
	III	41 (18)
	IV	155 (67)
Sample analyzed	Primary only	129 (55)
	Metastasis only	95 (41)
	Primary and metastasis	5 (2)
	2 or more metastases (no primary)	4 (2)
Histologic Subtype	Adenocarcinoma	199 (85)
	Carcinoma with squamous differentiation	23 (10)
	Small cell carcinoma	6 (3)
	Other poorly differentiated neuroendocrine carcinomas	5 (2)
Tumor grade	Well-differentiated	8 (3%)
	Moderately differentiated	95 (41%)
	Poorly differentated	121 (52%)
	Unknown (cytology specimens)	9 (4%)
Microsatellite instability	Stable	227 (97%)
	Instable	6 (3%)
Cholelithiasis (primary samples, n=134)	Present	46 (34%)
	Absent	23 (17%)
	Unknown	65 (49%)

Somatic mutations and tumor microbiome

A total of 1567 mutations and 660 CNA were identified among the 233 samples. Of these, 656 mutations and 387 CNA were classified as being ‘Oncogenic’ (mutations n=111, CNA n=241) or ‘Likely Oncogenic’ (mutations n=545, CNA n=146) by OncoKB.(17) Three samples had no somatic genetic alterations identified, despite them having an estimated tumor purity ≥20% based in histomorphology. The median sample coverage was 634X (range 72X to 1150X). Six tumors (3%) were MSI-High.

Oncogenic/likely oncogenic mutations were noted in multiple genes, including TP53 [63%, all LOF], SMAD4 (19%, 16% LOF and 3% unknown), ARID1A (18%, all LOF), PIK3CA [9%, all GOF], ELF3 (9%, all LOF), CDKN2A (9%, 8% LOF and 1% unknown), KRAS (7%, all GOF), ERBB2 (6%, all GOF), ARID2 (6%, all LOF), STK11 (6%, 5% LOF and 1% unknown), CTNNB1 (6%, all GOF), KMT2C (6%, all LOF), TERT promoter mutation (6%, all GOF), and RB1 (3.5%, all LOF), Figure 1A. Based on oncodrive clustering analysis,(24) mutations in CTNNB1 (most frequent alteration p.S45F/P), KRAS (most frequent alteration p.G12A/C/D/R), and ERBB2 (most frequent alteration p.S310F/Y) were identified as potential molecular drivers in GBC, Figure 1B and Figure 1C.

Figure 1. Oncogenic somatic alterations in primary and metastatic GBC.

(A) Oncoprint plot displaying selected genes altered in more than 5% of samples in histologic subtypes across n=233 GBC, arranged by canonical mutational pathways. Alterations are color-coded by mutation type. The frequency of somatic alterations is shown on the left. (B) Disease-associated driver genes identified by Oncodrive in n=233 GBC (tumors with no mutations were excluded.) The number of closely spaced mutational clusters is highlighted within brackets. Only genes with at least 7 mutations were included, and those with an FDR<0.05 are highlighted in red. (C) Lollipop plot displaying the oncogenic/likely oncogenic mutations in 5 potential driver genes in GBC. (D) Fraction of samples with copy number alterations (CNA) according to genomic coordinates. Genes altered in ≥5% of samples are highlighted. (E) Number of patients with structural variants (SV) according to the involved gene, color-coded by possible biological effect (red gain-of-function, blue loss-of-function). Only genes involved in at least two SVs in this cohort are displayed. NEC, neuroendocrine carcinoma.

Next, we determined the mutation signatures in cases with TMB >13.8 (n=14), as previously described,(20) Figure S1. Only seven cases showed a dominant mutation signature, which included APOBEC (altered in 3 cases) characterized primarily by C>T or C>G mutations at TpCpN trinucleotides, and MMR (altered in 3 cases) characterized by high numbers of small (shorter than 3bp) insertions and deletions at mono/polynucleotide repeats. The other case showed a mutation signature SBS5, which has an unknown etiology and exhibits transcriptional strand bias for T>C substitutions at ApTpN context.

Potentially oncogenic CNA were noted in multiple genes, including CDKN2A (14%), CDKN2B (14%), SMAD4 (5%), and RB1 (3%) deletions; and amplifications in MDM2 (11%), ERBB2 (10%), CCNE1 (9%), MYC (7%), ERBB3 (5%), CDK4 (5%), and KRAS (4%), Figure 1D.

MSK-IMPACT sequencing identified 74 SV in 63 out of 233 (27%) samples. Of these, 38 SV were classified as being ‘Oncogenic’ or ‘Likely Oncogenic’ by OncoKB, Table S2. No recurrent SV were found. Genes involved in oncogenic/likely-oncogenic structural alterations in more than one sample included TP53 (n=3), CDK12 (n=3), AKT2 (n=2), ARID2 (n=2), AXIN1 (n=2), RET (n=2), SMAD4 (n=2), and SMARCA4 (n=2), Figure 1E. Two cases showed in-frame fusions, namely FGFR3∷TACC3 and POLRMT∷MAP2K2, the first classified as an oncogenic GOF fusion and the latter as a variant of unknown oncogenic potential by OncoKB.

Analysis of the patterns of somatic mutations/CNA highlighted that the most common altered canonical pathways(23) in GBC include TP53 (altered samples 74%, including genes TP53 and MDM2), Cell Cycle (46%, including genes CDKN2A, CDKN2B, and RB1), RTK-RAS (46%, including genes KRAS, ERBB2, and ERNN3), Epigenetic (40%, including genes ARID1A and ARID2), TGF-Beta (28%, including gene SMAD4), and NOTCH (9%, including gene NOTCH1).

Certain molecular alterations correlated with tumor histopathologic characteristics, Supplemental File 1. Variants in RB1 were enriched in tumors with small cell neuroendocrine carcinoma morphology (q<0.001). The fraction of tumors with alterations in NOTCH pathway was higher in carcinomas with squamous differentiation (26%) (e.g., vs. adenocarcinoma 8%, q=0.07), while alterations in Cell Cycle pathway were more common in non-adenocarcinoma tumors (e.g., adenocarcinoma 42% vs. small cell neuroendocrine carcinoma 83%, q=0.01), Figure S1. There were higher frequencies of alterations in PIK3CA (15% vs 5%), SMAD4 (29% vs 17%), and STK11 (11% vs 2.5%) in well/moderately differentiated tumors compared to poorly differentiated tumors, though results were not significant after multiple testing correction. Variants in ERBB3 were more frequent in T stages 1 and 2 (16%) vs. 3 and 4 (0%) among primary tumors. Interestingly, there were higher frequencies of TERT (17% vs 2.2%) and CTNNB1 (22% vs 4.3%) alterations in primary lesions without cholelithiasis on resection, though missing data on cholelithiasis limited power in the analysis, which was not significant after correction for multiple testing. There were no significant differences in the frequency of molecular alterations according to the patient’s race/ethnicity, clinical stage at the time of collection, N stage (primary tumors only), or between primary and metastasis samples.

Analysis of the microbiome revealed the metastatic lesions had higher Shannon diversity index than primary samples, Figure S2. More specifically, metastases were enriched in aerobic, gram + cocci/rods e.g., Arthrobacter, Nocardia, Mycobacteroides, Rhodococcus, Mycobacterium, Nocardioides. We found no differences in the microbiome diversity according to the presence of the most frequent molecular alterations in this cohort. However, tumors with SMAD4 alterations were enriched in gram-negative aerobic bacteria e.g., Thiomona, Vitreoscilla, Dyella, Acidiphilium, and Azoarcus. Interestingly, none of the cases showed reads mapping to the genus Helicobacter or Salmonella.

Molecular predictors of clinical outcome

Overall survival (OS) data was available for 219 (94%) patients. Therapy data was available for 212 (91%) patients, Table S3. One-hundred and ninety-one (82%) of patients received first-line chemotherapy for advanced disease (143 patients: gemcitabine/platinum). Additional regimens used in the first-line setting included FOLFIRI, FOLFOX, capecitabine monotherapy, gemcitabine monotherapy, and 5-fluorouracil monotherapy.

The presence of certain molecular alterations correlated with the patient’s clinical outcome. We analyzed separately primary (n=124) and metastatic samples (n=100). In the primary setting, none of the analyzed molecular variants correlated with OS, Supplemental File 1. However, increasing age and advanced clinical stages at time of tumor collection correlated with shorted OS in this subgroup of patients. In the metastatic setting, patients with alterations in STK11 (7%) and SMAD4 (23%) had shorter OS than patients with no alterations in those genes (STK11: q=0.05, HR 3.94 [95% CI1.62–9.55]; SMAD4: q=0.10, HR 2.17 [1.21–3.89]), Figure 2A and Figure 2B. In contrast to primary samples, in metastatic patients none of the evaluated clinicopathologic variables (i.e., age, tumor grade, and tumor histology) associated with prognosis. Interestingly, mutations in both genes stayed independently associated with shorter OS in the multivariate Cox model that included them both (STK11: p=0.004, HR 3.76 [95% CI1.54–9.16]; SMAD4: p=0.012, HR 2.11 [1.18–3.80], Supplemental File 1.)

Figure 2. Molecular predictors of clinical outcome in metastatic GBC.

(A) Forest plot displaying the prognosis associated with the most common (≥5% frequency) somatic variants in GBC. Hazard-Ration (HR), 95%-Confidence Interval (CI), p-value, and q-value by cox-univariate modeling for OS are displayed. (B) Kaplan-Meier curves displaying OS according to the presence of STK11 (top panel) or SMAD4 (bottom panel) alterations in metastatic GBC.

Potentially actionable mutations

Eighty-two patients (35.2%) had at least one mutation or structural variant that is considered actionable in GBC or other solid tumor types (OncoKB therapeutic level 1, 3A or 3B),(17) making them strong candidates for clinical trial enrollment, Figure 3A. Several patients had more than one of potentially targetable genetic alteration, with 3 patients harboring 4, 5 patients with 3, and 17 patients with 2.

Figure 3. Oncogenic potentially-targetable alterations in primary and metastatic GBC.

Oncoprint plot displaying the genes with potentially actionable alterations across n=233 GBC, color-coded by therapeutic level of evidence for actionability (scale of 1–4 according to OncoKB) and variant type (i.e., mutation, CNA or SV.) The highest overall level of evidence for each sample is displayed on top. The frequency of somatic alterations is shown on the left. NEC, neuroendocrine carcinoma.

One patient harbored an LMNA∷NTRK1 mid-exon fusion (Exon 2 of LMNA to an intron of NTRK1(+) 113bp before exon 12, including its kinase domain), which is an FDA-recognized biomarker predictive of response to entrectinib and larotrectinib in all solid tumors types, including GBC (therapeutic level 1 by OncoKB.) In addition, the only somatic variant classified as level 3A (compelling clinical evidence supporting the biomarkers as being predictive of response to a drug in GBC) was KRAS p.G12C, which was present in 2 patients (1%) and has shown promising response rates in a phase II clinical trial that included patients with biliary tract tumors other than GBC.(34)

Seventy-eight patients (33%) harbored somatic alterations that were classified as level 3B (standard of care or investigational biomarker predictive of response to and FDA-approved or investigational drug in another indication.) Variants for which there is an FDA-approved targeted therapy in tumors types other than GBC included ERBB2 amplification (n=24), and oncogenic/likely-oncogenic somatic mutations in PIK3CA (i.e., p.E545G, p.E545K, p.E542K, p.H1047L or p.H1047R, n=16), BRCA2 (n=6), ATM (n=5), BRCA1 (n=3), CDK12 (n=3 patients), TSC2 (n=3), TSC1 (n=2), PALB2 (n=2), and BRIP1 (n=1). Variants for which there is a standard-of-care investigational drug used in tumors other than GBC included MET amplification (n=5 patients) and RAD51D deletion (n=1), and mutations in ERBB2 (n=18), PIK3CA (i.e., p.G118D, p.M1043I, p.M1043V, p.N345S, p.R88Q, n=5), AKT1 (n=1), BRAF (n=1), EGFR (n=1), ERCC2 (n=1), MAP2K1 (n=1), and NRAS (n=1). In addition, seven patients harbored structural variants classified as level 3B, which included 5 alterations likely leading to LOF of CDK12 (i.e., one SMARCE1∷CDK12 fusion, one CDK12 intragenic deletion, and one CDK12∷ERBB2 fusion) or RET (i.e., one RET∷LOC100129055 fusion and one UXS1∷RET fusion); one variant leading to GOF of FGFR3 (i.e., FGFR3∷TACC3 in-frame fusion); and one variant likely leading to GOF of ROS1 (i.e., ROS1∷PARK2 fusion.)

Sixteen out of 157 patients (10%) with metastatic disease at time of collection or who developed metastatic lesions after the initial surgical resection received targeted therapy based on the molecular profiling results, including 10 treated with HER2-directed therapy (n=5 trastuzumab, n=2 trastuzumab/pan-HER2 TKI, n=1 trastuzumab/pyrotinib, n=1 pan-HER2 TKI, n=1 other anti-HER2 monoclonal antibodies), 3 with MEK-inhibitors, 1 with olaparib/ATM inhibitors, 1 with ERK-inhibitors, and 1 with NTKR-inhibitors. Three of the 16 patients (19%) treated with targeted therapy had evidence of response to treatment based on the treating physician assessment. Two additional patients with locally advanced disease received neoadjuvant target therapy (n=1 gemcitabine/cisplatin/MEK-inhibitors and n=1 gemcitabine/cisplatin followed by olaparib) and both showed a pathologic response ≥70%.

Furthermore, an additional set of twelve patients received checkpoint inhibitors (8% of the metastatic cases), including pembrolizumab (n=9), atezolizumab (n=1), nivolumab monotherapy (n=1), and ipilimumab/nivolumab dual therapy (n=1). Five of the 12 patients (42%) treated with checkpoint inhibitors had evidence of response to treatment. Among these five patients, 3 harbored MSI-High tumors, 1 had an MSS tumor with high TMB (21.9 mutations/megabase), and 1 had an MSS tumor with low TMB (4.4 mutations/megabase).

Discussion

To date, the main molecular drivers in GBC remain to be defined. Consistent with previous reports,(10,11) we found that oncogenic alterations in TP53, ERBB2/3, CDKN2A/B, and SMAD4 dominate the mutational landscape of this tumor type. Furthermore, the size of this cohort allowed us to build on previous reports and evaluate the oncogenic role of other less commonly mutated genes. Our data highlighted CTNNB1, KRAS, and PIK3CA as novel potential molecular drivers in GBC. Studies assessing the sequential genetic evolution of gallbladder adenomas to invasive carcinomas(35) support this finding, as they established that mutations in CTNNB1 and KRAS are one of the initiating events during carcinogenesis and could play a major role in the evolution of pre-malignant to malignant cell states.

The wide use of NGS technologies has significantly expanded our understanding of the correlation between molecular events and histomorphology in solid tumors. Our data suggest that molecular alterations are differentially distributed across GBC histologic subtypes. Specifically, we found that RB1 LOF mutations are more common in tumors with neuroendocrine differentiation i.e., small cell neuroendocrine carcinoma, and GOF variants in the NOTCH pathway genes are enriched in tumors with squamous morphology. Previous studies support these associations. Indeed, mutations in RB1 are significantly enriched in neoplasms with small cell morphology in several tumor types e.g., non-small cell lung cancer (NSCLC)(36) and esophageal carcinoma.(37) Furthermore, NOTCH1/3 activation has been implicated in the regulation of squamous differentiation of both healthy and neoplastic tissues.(38,39) One of the limitations of this study is the low number of tumors with neuroendocrine histology available for analysis. Future studies assessing a larger number of NEC could shed light on the origin of this tumor type and uncover other molecular alterations enriched in this histology.

In contrast with our findings in histologic subtypes, we did not observe significant differences in the prevalence of oncogenic alterations in tumor subgroups according to other histopathologic characteristics (e.g., tumor grade, T or N stage), clinical stage at the time of sample collection, or even between primary and metastatic tumor samples. However, we found that mutations in SMAD4 and STK11 associate with reduced survival in patients with metastatic disease. Interestingly, STK11 and SMAD4 alterations have been both associated with aggressive tumor behavior in other cancer types such as NSCLC.(40,41) Furthermore, mutations in STK11 have been reported to be associated with resistance to checkpoint inhibitors.(42) In this cohort, two patients with STK11 mutations were treated with checkpoint blockade and both responded to therapy. Finally, in contrast with other reports, ERBB2/ERBB3, TP53, and ELF3 alterations did not associate with clinical outcomes in our cohort.(10,11,43) The absence of an association between ERBB2 alterations and adverse clinical outcome in our cohort could be related to the relatively large proportion of patients treated with HER2-targeted therapies which could have potentially improved patient outcomes.

Previous studies have highlighted the influence of geography, ethnicity, and cultural differences in the molecular landscape of GBC.(11,44) In this cohort we did not find differences in the frequency of molecular variants across four categories of patient race/ethnicity i.e., Asian, Black/African American, Caucasian/White or Hispanic. One limitation of this analysis is that most of the analyzed patients lived within the continental United States, precluding further assessment of the correlation between geography/environmental factors and molecular alterations. Interestingly, we did not find any evidence supporting the presence of commonly reported pathogens associated with GBC in this cohort i.e., Salmonella spp or Helicobacter spp.(45,46) Since most of the previously reported microbiologic assessments in GBC have been performed in endemic areas for this disease, our results suggest that the microbiome in GBC and its relationship with oncogenic pathways vary across regions. This is despite that our first effort along that line are not revealing as such.(9) Further studies are necessary to investigate other historically underrepresented populations and look at specific cohorts originating in different geographies.

This effort sheds light on the remarkable differences in the molecular landscape of GBC and other biliary tree tumors, such as cholangiocarcinoma. In contrast to the latter, only a small fraction of GBC harbors mutations in IDH1 or BAP1, and we did not find fusions encompassing FGFR2.(47,48) These differences support a diverse etiology and cell of origin in both diseases.

In this study, we found that a remarkably large percentage of GBC patients harbor alterations considered clinically actionable in GBC or other solid tumor types e.g., ERBB2 amplification, or PIK3CA and BRCA2 mutations. Furthermore, we provide evidence supporting that a significant fraction of patients treated with targeted therapies or checkpoint blockade had evidence of response to therapy. Altogether, these findings highlight the necessity to expand access to DNA tumor sequencing in GBC patients to assure optimal tailoring of therapies, especially in advanced stages of the disease where treatment options are currently limited. One of the main challenges that the oncology field is currently facing is the numerous barriers for accessing unlicensed targeted therapies, and this study highlights the potential benefit of using off-label agents for the treatment of GBC.

In keeping with previous reports, our findings support that GBC is an immunomodulated tumor type that could potentially benefit from immunotherapy-based treatment approaches(11,49,50). Here we found that approximately 3% of the GBC are MSI-High and about 10% of patients harbor a high TMB. Previous studies in GBC have shown similar fractions of MSI and TMB-high tumors.(11,51) Interestingly, in our cohort, 42% of the patients treated with checkpoint inhibitors had evidence of response to treatment. These included 3 MSI-High/TMB-High tumors, 1 MSS/TMB-high tumor, and one MSS/TMB-low tumor.

In summary, our study shows that GBC carcinoma is a genetically diverse cancer. Furthermore, this large-scale genomic analysis revealed alterations potentially associated with negative prognostic implications in patients with advanced GBC and provides guidance for the development of targeted therapies. Ongoing efforts continue to explore a broader global perspective, which we have already initiated with Chile and following with India.

Translational relevance:

Gallbladder Carcinoma (GBC) is a rare and aggressive disease that remains poorly defined at a molecular level. To identify markers with potential prognostic and therapeutic implications we characterize the molecular landscape of a large cohort of GBC (n=233). We identified oncogenic variants associated with shorted overall survival in patients with metastatic disease. Furthermore, we described that approximately one third of GBC patients harbor alterations considered clinically actionable in this malignancy or other solid tumors. In this cohort of patients with GBC from Memorial Sloan Kettering (MSK), 18% of patients with metastatic disease received biomarker-directed therapy or were enrolled in clinical trials based on their tumor molecular characterization. This study reveals novel recurrent alterations with potential prognostic and therapeutic implications in GBC and supports that DNA tumor sequencing should become standard-of-care to allow tailoring of treatment in this aggressive malignancy.