Molecular and Clinical Determinants of Targeted Therapy Treatment in Biliary Tract Cancer

Abstract

Purpose:

Actionable genomic alterations occur in all anatomic subsets of biliary tract cancer (BTC); however, targeted therapies have not shown a survival advantage over cytotoxics, and resistance mechanisms require further characterization.

Experimental Design:

We analyzed a prospectively maintained cohort of 1254 patients with histologically confirmed BTC who underwent molecular profiling using an FDA-authorized targeted next-generation sequencing assay. We defined actionable alterations across anatomic subsets, compared outcomes with targeted therapy versus cytotoxics, and evaluated genomic correlates of resistance using longitudinal samples.

Results:

Overall, 59% of patients harbored at least one OncoKB alteration, and 32.2% (intrahepatic 40%, extrahepatic 15%, gallbladder 22%) had a level 1/2 alteration. Emerging targets included KRAS alterations (17%), MTAP deletions (12.8%), MDM2 amplification (6.5%), and MET amplification (1.5%). Targeted therapy was associated with improved progression-free survival but not overall survival. Co-occurring TP53/RAS pathway and SMAD4 alterations were associated with inferior outcomes in IDH1/FGFR2- and ERBB2-driven tumors, respectively. Longitudinal profiling demonstrated ERBB2 loss in ERBB2-driven tumors, whereas IDH-, FGFR-, BRAF-, and NTRK-driven tumors retained the primary oncogenic driver. Acquired resistance was associated with alterations in RAS, MEK, MET, MYC, and CDKN2A.

Conclusions:

This comprehensive molecular profiling study illustrates the real-world utility and limitations of targeted next-generation sequencing of BTC and affirms the use of precision medicine in patients with these diseases. Characterization of genomic heterogeneity and therapeutic resistance has the potential to inform ongoing drug development efforts for BTC.

Introduction

Biliary tract cancers (BTC), which include intrahepatic (iCCA) and extrahepatic cholangiocarcinomas (eCCA) and gallbladder carcinomas (GBC), are a clinically and molecularly heterogeneous group of malignancies with a poor prognosis (1,2). Chemotherapy in combination with immunotherapy offers a modest survival benefit as first-line systemic therapy for patients with advanced disease (3,4), whereas cytotoxic chemotherapies in the second line have modest to no antitumor activity (5–7), highlighting the unmet need for novel therapeutic strategies.

Retrospective molecular profiling of studies of patients with BTC has identified potentially actionable molecular alterations, including somatic alterations in IDH1, FGFR2, ERBB2, BRAF, RET, and NTRK, as well as hypermutated or microsatellite instability high (MSI-H) tumors (8–10). These alterations are bona fide therapeutic vulnerabilities, and several groups have established knowledge bases to annotate clinical actionability, facilitating matching of genomic profiles to cytotoxic or targeted therapy (11).

Several targeted therapies have received regulatory approval for the treatment of patients with BTC following disease progression on first-line therapy (12–16) or as disease-agnostic indications (17–19), or have been recommended in treatment guideline recommendations (20,21)(22). A critical problem facing patients and clinicians is that the evidence supporting most targeted therapeutics in BTC is based on single-arm, non-comparative, small prospective (~100 patients or fewer) phase 2 clinical trials, or infrequent cases treated within the context of tumor-agnostic basket studies. The magnitude of survival benefit, if any, of precision medicines in patients with mutationally selected BTC is, therefore, largely unknown. Indeed, comparative phase 3 studies for FGFR2 fusion and rearrangements have all failed to accrue due to the rarity of alterations in this gene (23). Ivosidenib, a selective IDH1 inhibitor, was compared to placebo in the second and third-line setting for IDH1 mutant iCCA and is the only targeted therapy, when accounting for intentional cross-over, to have demonstrated improved patient outcomes (12,13). Thus, analyses of large real-world comparative data and benchmarks for outcomes will be a critical tool for guiding future drug development and treatment selection for patients with this rare but typically fatal cancer type (24).

Precision medicine does not work in all patients with an identifiable actionable genomic alteration, and the majority of responding patients develop acquired treatment resistance. Mutations in the targeted gene and oncogenic pathway bypass or downstream mechanisms of drug resistance have been documented to occur in patients with IDH1 and FGFR2-driven BTCs (25–27); however, such mechanisms have not been clearly established in other mutationally defined cohorts (28). Therefore, both pretreatment and sequential tumor profiling of patients with BTC treated with molecularly guided systemic therapies are required to identify novel mechanisms of resistance that could inform the development of more effective targeted therapies or drug combinations.

Here, we sought to systematically characterize our single-institution experience by leveraging a prospectively maintained database of patients with BTC who underwent real-time, clinical-grade, next-generation sequencing (NGS) as a guide to therapy selection. This resource, coupled with retrospective data extraction, has allowed for a comprehensive description of the mutational landscape and frequency of actionable genomic alterations for each anatomic subsite of BTC, the exploration of the impact of targeted therapy on outcome, and the nomination of genomic mechanisms of resistance via available paired pre- and post-targeted treatment tumor samples.

Materials and Methods

Clinical Cohort.

Patients with BTC who underwent targeted NGS at Memorial Sloan Kettering Cancer Center (MSK) were enrolled on a prospective clinical genomic profiling protocol (NCT01775072, MSK IRB 12–245) from April 2014 to September 2022 (Fig. S1, Table S1). All patients provided written informed consent. The study was approved by the MSK Institutional Review Board (IRB) and was conducted in accordance with the U.S. Common Rule. Tumor samples with low tumor purity (<10%) or with no somatic alterations identified were excluded from the analysis (n = 111). One sample per patient was selected, prioritizing primary tumors and samples of higher quality, for analysis of the genomic landscape of BTC and associations with patient demographic and clinical outcomes. To analyze mechanisms of drug resistance, all samples, including paired tumor samples collected pretreatment and following disease progression, along with plasma circulating tumor DNA (ctDNA), were included (patients: n = 1254, samples n = 1285 (Table S1).

Mutations were classified as oncogenic or clinically actionable using OncoKB (RRID:SCR_014782). To assess the expansion in actionable genomic alterations over time, we utilized the OncoKB knowledge base (11,29) releases from March 2017 (2017v1.8) and July 2023 (2023v4.7). The potentially actionable cohort (n = 514) was defined as those tumors harboring an OncoKB level 1 (United States Food and Drug Administration (FDA) recognized indication), level 2 (standard care based on National Comprehensive Cancer Network [NCCN] or other guidelines), level 3A (compelling clinical evidence in iCCA, eCCA or GBC), or select level 3B (compelling clinical evidence in another indication) alteration in a gene clinically relevant to BTC. The level 3b genes considered potentially actionable were: BRAF non-V600E, BRCA1, BRCA2, FGFR1, FGFR3, IDH2, and MET. Patients with missing survival data (n = 46) and incomplete clinical follow-up (n = 55) were excluded from this cohort to define the group of patients that were clinically eligible for matched-targeted therapy (n = 413). The matched therapy group (43%; n = 176) comprised patients who received a molecularly targeted therapy based on NGS results (RRID:SCR_005182). The ‘unmatched therapy group’ (57%; n = 237) comprised patients who had a potentially actionable alteration identified but did not receive a molecularly targeted therapy based on the NGS results. In analyses comparing matched vs unmatched alteration-specific treatment cohorts, patients with more than one actionable alteration were included in each of the unmatched alteration-specific treatment cohorts (Table S2). To address lead time selection bias in our univariate analyses, we included patients who either received matched-targeted therapy as second-line systemic therapy (treated, n = 100) or patients who received 2 or more lines of systemic therapy (untreated, n = 113).

Patient demographics, including sex, date of birth, and sample type, were extracted from prospectively maintained institutional databases. Additional clinicopathologic features were retrospectively extracted via manual chart review, including date of diagnosis; anatomic subtype of BTC; stage of disease; Eastern Cooperative Oncology Group (ECOG) performance status; surgical history, dates, and types of systemic treatment; radiographic images, date of progression, and date of death or last follow-up. For patients with OncoKB level 1, 2, or 3 alterations who were not treated with the corresponding targeted therapy, the rationale for nontreatment was recorded.

Sample Sequencing and Tumor Profiling.

Tumor and matched normal blood samples were analyzed using MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets), a clinically validated hybridization capture-based targeted NGS assay, that detects mutations, copy number alterations (CNAs), and select structural rearrangements in 341–505 cancer-associated genes, depending on the version of the panel (341 genes, n = 55; 410, n = 200; 468, n = 663; 505, n = 336) (30). In select cases, additional molecular profiling of cfDNA was performed using MSK-ACCESS (31).

Genomic Analysis.

Somatic mutations, CNAs, and structural variants were interpreted for each tumor sample using previously published methods (32). All analyses were adjusted for the differences in profiled genes across the various versions of the MSK-IMPACT panels (Table S3). Tumor mutational burden (TMB) was calculated as the total number of nonsynonymous mutations per megabase sequenced. The FACETS (Fraction and Allele-Specific Copy Number Estimates from Tumor Sequencing) algorithm (33,34) and the FACETS-suite package (https://github.com/mskcc/facets-suite) were used to calculate purity-corrected copy number estimates and cancer cell fractions of detected mutations. Mutations were considered to be subclonal when the upper bound of the 95% CI for the CCF was less than one and the probability of CCF < 0.5 was greater than 0.95, as estimated by FACETS (35,36). Genes were grouped into pathways using curated templates from The Cancer Genome Atlas (TCGA) (37).

A combination of MSIsensor (38) and the MiMSI algorithm (39) was used to evaluate microsatellite instability (MSI) status. Tumors with an MSISensor score ≥10 were classified as MSI (n = 16) (40). All patients with TMB-high (TMB ≥10) (n = 72) were further interrogated using the MiMSI algorithm, leading to the classification of an additional 10 patients as MSI. Hypermutant patients (TMB ≥ 30) with a microsatellite stable (MSS) status as defined by both methods were classified as hypermutant MSS (n = 2).

Ancestry analysis.

Ancestry was assessed using an ADMIXTURE analysis as previously described (41).

Statistical Analysis.

A two-sided Mann-Whitney U test was used to compare continuous variables between 2 groups, while a two-sided Fisher’s exact test was used to compare categorical variables. Overall survival (OS) was calculated from the date of commencement of second-line therapy until death or last follow-up. Progression-free survival (PFS) was determined from the date of treatment initiation to the date of investigator-assessed progression or death, whichever occurred first. Patients alive without evidence of progression and who did not receive additional treatment by the last follow-up were censored at the last available date of radiological disease assessment. The Kaplan-Meier method was used to estimate OS and PFS probabilities, and log-rank tests were used to compare them for different groups in the univariate setting, while Cox proportional hazards were used for multivariate analyses of OS and PFS. All statistical analyses were performed using R Version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). All P-values were two-sided. Multiple hypothesis correction was applied when necessary, using the Benjamini-Hochberg method.

Data Availability

Deidentified patient data are not publicly available due to patient privacy and current regulatory approvals; however, the data are available upon request to James Harding, MD. The summary data are available at https://www.cbioportal.org/study/summary?id=biliary_tract_msk_2026.

Results

Patient characteristics and the genomic landscape of BTC

Patient and sample characteristics are summarized in Table S1 and Figure 1A/1B. Of 1254 patients undergoing tumor genomic profiling, 767 tumors (61.2%) were iCCA, 210 (16.8%) were eCCA, and 277 (22.1%) were GBC (Table 1).

Figure 1: Clinical and genomic landscape of biliary tract cancers (BTC).

A. Comparison of clinical features across BTC subtypes. B. Proportion of primary and metastatic tumor samples, with metastatic sites indicated separately. Outer layers depict the proportion of samples with actionable alterations and patients who received targeted therapies. C. Tumor mutation burden (Mut/Mb) and fraction genome altered across BTC subtypes. D. Proportion of whole genome doubling across BTC subtypes. E. Oncoprint showing top altered and actionable genes stratified by BTC subtypes. F. Percentage of samples harboring OncoKB levels 1–4 and oncogenic alterations in 2017 and 2023, full cohort followed by stratification by subtype. G. Comparison of the number and types of actionable alterations across BTC subtypes. H. Bar plot comparing the types of KRAS mutation broken down by subtype. I. Mutant allele selection status and clonality of different types of KRAS mutations.

Table 1.

Patient and tumor characteristics

	Full Cohort	iCCA	eCCA	GBC
Clinical characteristics	1254 (%)	767 (61.2%)	210 (16.7%)	277 (22.1%)
Sex
Male	603 (48%)	398 (51.9%)	118 (56.2%)	87 (31.4%)
Female	651 (52%)	369 (48.1%)	92 (43.8%)	190 (68.6%)
Anatomic location
Intrahepatic	767 (61%)
Extrahepatic	210 (16.7%)
Gallbladder	277 (22%)
Ethnicity
Caucasian	931 (74.2%)	601 (78.3%)	150 (71.4%)	180 (64.9%)
Asian	129 (10.2%)	67 (8.7%)	31 (14.8%)	31 (11.2%)
African American	79 (6.2%)	33 (4.3%)	12 (5.7%)	34 (12.3%)
Age
Median (range)	65 (21–89)	65 (21–88)	66 (21–89)	66 (35–89)
Sample sequenced
Primary	879 (70%)	592 (77.2%)	140 (66.6%)	147 (53.1%)
Metastatic site	375 (30%)	175 (22.8%)	70 (33.3%)	130 (46.9%)
Highest level of actionability identified
I	348 (27.8%)	294 (38.3%)	19 (9.04%)	35 (12.6%)
II	57 (4.5%)	17 (4.5%)	13 (6.2%)	27 (9.7%)
IIIa	14 (1.1%)	9 (1.1%)	2 (0.95%)	3 (1.1%)
IIIb	219 (17.5%)	132 (17.2%)	34 (16.19%)	53 (19.1%)

Most patients were of European ancestry (60.1%) with a slight female predominance (51.91%) (Fig. 1A). Nearly all tumors were MSS (1226/1254 [97.8%]); MSI-H accounted for 26 of 1254 (2.1%), with similar rates of MSI-H disease across subtypes (iCCA, 12/767 [1.5%]; GBC, 6/277 2.2%]; and eCCA, 8/210 [3.8%]). The median TMB of the full cohort was 3.3 mutations/megabase (mut/Mb) (range, 0–74.6 mut/Mb). Among MSS tumors, the median TMB was 3.0 mut/Mb (range, 0–29 mut/Mb), with 48 tumors classified as TMB-H (≥10 mut/Mb). When limited to MSS tumors, the median TMB (4.1 mut/Mb vs. 2.6 mut/Mb; P =.002) and median fraction genome altered (FGA) (0.495 vs. 0.43; P = .014) was significantly higher in GBC compared to iCCA (Fig. 1C). The rate of WGD was significantly higher in eCCA (34.2%; P = 0.002) and GBC (32%; P = 0.001), as compared to iCCA (20.7%) (Fig. 1D). Among MSS tumors, the most frequently altered genes were TP53 (36%), CDKN2A (21%), ARID1A (18%), KRAS (17%), and IDH1/2 (16%) (Fig. 1E). Age-related changes were noted in FGFR2 and MDM2 altered tumors (Fig. S2A). Alterations in IDH1, FGFR2, BAP1, and PBRM1 were found to be significantly enriched in iCCA tumors compared to both eCCA and GBC tumors, whereas alterations in TP53, SMAD4, and MDM2 were less frequent in iCCA compared to either eCCA or GBC (Fig. S2B). Alterations in KRAS were more frequent in eCCA compared to iCCA and GBC, and ERBB2 alterations were more frequent in GBC compared to iCCA (Fig. 1E, Table S4). The significant trends in alteration frequencies at the pathway level largely reflected what was seen at the individual gene level (Fig. S2C, Table S5). KRAS-G13D alterations were enriched in the GBC subset, whereas KRAS-G12V was only seen in iCCA and eCCA (Fig. 1F). Within the entire cohort, we observed the highest frequency of mutant allele selection in the G13D mutants. The majority of the “Other” KRAS (defined as A146T/V, L19F, V14I) alterations were found to be subclonal (Fig. 1G).

While most driver and passenger mutations as defined by OncoKB across subtypes were clonal, passenger mutations had a slightly higher rate of subclonal mutations compared to the driver mutations (Fig. S2D). ERBB2 mutations in GBC were subclonal in 41.2% of tumors (Fig. S2D), whereas ERBB2 amplifications were predominantly clonal across all subtypes. Comparing samples with clonal and subclonal alterations in the GBC cohort, we observed an enrichment of ERBB3 and NF1 alterations at the gene level and the epigenetic, NRF2, and PI3K pathways in the samples harboring a subclonal ERBB2 mutation (Fig. S2E).

Clinical utility of molecular profiling

Consistent with the increase in the availability of molecularly guided therapeutic options, including both those FDA-approved specifically for BTC and those approved for other cancer types, we observed a notable increase in the percentage of BTC harboring OncoKB level 1 and 2 alterations from 2017 to 2023 (Fig. 1H) and among tumors with any actionable alteration, most harbored only one actionable alteration, typically a missense mutation (Fig. 1I). Emerging druggable targets for which promising clinical results have been observed in patients include KRAS mutations (G12D, 5.4%; G12C, 1.1%), MTAP deletions (12.8% of tumors profiled; n = 336), MDM2 amplification (6.5%), and MET amplification (1.5%) (Fig. S2F).

Clinical outcomes of targeted therapy

Among 413 patients with level 1, 2, and 3A actionable alterations, 176 (42.6%) received genomically matched-targeted therapies (IDH1 mutation 73/150 (48.7%, iCCA = 72/147), FGFR2 fusion 37/72 (51.4%, iCCA = 37/71), ERBB2 amplification 22/54 (40.7%, 27/67 in ERBB2 amplifications and mutations combined), MSI-H 14/21 (66.7%), non-MSI TMB-H 12/31 (38.7%), BRAF mutation 8/25 (32.0%; class I: n = 16, 64%; class II: n = 5, 20%; class III: n = 4, 16%), and NTRK1 fusion 2/5 (40.0%))(Fig. S1). Patients with level 3b alterations who received a matched-targeted therapy included those with BRCA1 mutation 2/4 (50.0%) and MET amplification 1/14 (7.1%) (Table S2). Matched-targeted therapy was predominantly administered in the second-line setting (100; 62.5%).

From 2014 to 2021, the proportion of patients who received targeted therapy increased, mirroring the increase in FDA-approved and guideline-endorsed therapies during this timeframe (Fig. S2G, Table S6). The most common reasons patients did not receive a targeted therapy included no FDA-approved agent available within 6 months of their date of death (83/237; 35.0%), the patient was still responding to first-line therapy at the time of data analysis (29/237; 12.2%), or the patient died prior to eligibility (38/237; 16.0%)

No significant differences were observed in age, sex, anatomic site, or oncogenic alterations between those who received and did not receive molecularly matched therapy. (Table S7, S8).

We observed a significant PFS benefit in patients who received matched-targeted therapy as second-line treatment (n = 99) compared to those who received 5-FU-based chemotherapy (n = 25) in the same line (median, 6.6 months [95% CI, 4.2–8.6 months] vs. 3.3 months [95% CI, 2.1–8.1 months], P =.005) (Fig. 2A, Fig. S3A). Additional results show balanced clinical characteristics across the 2 groups (Table S7).

Figure 2: Outcomes for patients with actional alterations matched to target therapy.

A. Kaplan-Meier progression-free survival curve from the start of second-line therapy comparing patients who received 5-fu based treatment versus targeted therapy. B. Kaplan-Meier overall survival curve from the start of second-line therapy comparing patients treated with a targeted therapy to those who did not. To address the potential lead time bias, the treated group was restricted to patients who received targeted treatment in the second line, and the untreated group was restricted to patients who received 2 or more lines of systemic therapy.

No significant difference in OS was seen in the matched therapy group (median follow-up, 13.8 months) compared to the unmatched therapy group (median follow-up, 9.0 months) (median OS, 17.0 months [95% CI: 12.0–27.0 months] vs. 11.0 months [95% CI, 8.5–14.0 months]; HR 0.8, P = .16) (Fig. 2B). A similar result was observed when excluding patients that received immune checkpoint inhibitors for MSI/TMB-H disease (16.0 months [95% CI, 12.0–25.0 months] vs. 11.0 months [95% CI, 8.1–14.0 months]; P =.28, with median follow-up times of 13.2 months and 9.0 months, respectively) (Fig. S3B). When limiting the analysis to patients with OncoKB level 1 and 2 alterations, no significant difference in OS was observed between the matched and the unmatched therapy cohorts (17.0 months [95% CI, 13–27 months] vs. 13.0 months [95% CI, 8.9–16.0 months], P =.34) (Fig. S3C). In univariate survival analyses, age, sex, anatomic site, FGA, and sample type were not significantly associated with survival (Table S9)

Molecularly defined subgroups

The median PFS and OS for those with IDH1 mutations (Fig. 3A, S3D, Table S7, S8), FGFR2 fusions (Fig. 3B, S3D, Table S7, S8), ERBB2 amplifications (Fig. 3C, S3D, Table S7, S8), BRAF mutations (Fig. S3E), and MSI-H/TMB-H (Fig. 3D, S3F–G), treated with matched-targeted therapy are listed in Table 2. In the BRAF cohort (N=16) the median OS in those patients who received targeted therapy was 37.0 months vs 9.5 months for those patients not treated with targeted therapy (p = 0.013). No other mutationally defined cohort exhibited an association with longer OS with targeted therapy. We then assessed whether any clinical or genomic features were associated with outcomes within each of our molecularly defined subgroups. We found that TP53 and RAS-RAF pathway alterations, TP53 pathway alterations, and SMAD4 alterations conferred a worse OS in those with IDH1 mutations, FGFR2 fusions, and ERBB2 amplifications, respectively. No clinical characteristics or genomic alterations stratified PFS for any subgroup (Table S10).

Figure 3: Survival patterns of the IDH1/2, FGFR2, and ERBB2 cohorts.

A. Swimmers plots showing the progression-free survival in the IDH1/2 treated patients and Kaplan-Meier overall survival curves from the start of second-line therapy comparing IDH1/2 mutated patients treated with a targeted therapy to those who did not. B. Swimmer's plots show the progression-free survival in the FGFR2i-treated patients. Kaplan-Meier overall survival curves from the start of second-line therapy comparing patients harboring FGFR2 fusions treated with a targeted therapy to those who did not. C. Swimmers plot showing the progression-free survival in the ERBB2i-treated patients. Kaplan-Meier overall survival curves from the start of second-line therapy comparing ERBB2 amplified patients treated with a targeted therapy to those who did not. D. Swimmers plot depicting the progression-free survival in patients treated with immunotherapy. Kaplan-Meier overall survival curves from stage IV unresectable disease and progression-free survival from the start of IO therapy, comparing MSI-H vs TMB-H patients.

Table 2.

Molecularly defined subgroups

Alteration (n=PFS, n=OS, proportion treated)	PFS	OS (treated vs untreated, months; 95%CI)	Significant alteration associated with outcome	OS adjusted in MV model
IDH1 (n = 72/165, n = 40/74)	4.2 months (2.9–6.8)	13.0 months (18.0–21.2) vs. 14.0 months (12.0-NR); P =.082	TP53 and RAS-RAF pathways alterations	HR 1.5, 95% CI 0.9–2.7; P =.149
FGFR2 fusion (n = 37/72, n = 21/42)	7.4 months (6.5–9.4)	27.0 months (12.1-NR) vs. 13.0 months (3.3–31.3); P =.6	TP53 pathway alterations	HR 1.1, 95% CI 0.5–2.4; P =.835
BRAF (n = 9/16, n = 4/8)	16.6 months (3.68-NR)	37.0 months (NR-NR) vs. 9.5 months (1.9-NR), HR 0.17, P =.013	Nil
ERBB2 (n = 22/54, n = 13/29)	8.2 months (2.9–16.7)	12.0 months (2.4-NR) vs. 6.2 months (5.1–17.8); P =.42	SMAD4 alterations	HR 0.74, 95% CI 0.31–1.8; P =.502
MSI/TMB-H (n = 26/52; 14 MSI, 12 TMB-H, n = 13/19)	22.5 months (1.97-NR)	35.0 months (2.2-NR) vs. 19.0 months (8.71-NR); P =.55	Nil

Patterns of de novo and acquired resistance to targeted therapy

Twenty-one patients with MSS BTC (6 IDH1/IDH2; 8 ERBB2; 5 FGFR2 fusions; 2 NTRK fusions; and 1 BRAF_V600E) and 2 patients with MSI matched to targeted therapy had sequential samples analyzed (Fig. 4A). Patients treated with BRAF inhibitor therapy exhibited the highest rates of mutation concordance after filtering for oncogenic driver mutations (Fig. S4A). All tumors with oncogenic FGFR2, IDH1, and NTRK retained their oncogenic driver mutations at the time of treatment resistance. Among the ERBB2 group, 2 patients exhibited loss of their driver ERBB2 mutation in the post-treatment samples (S310F and R678Q), and 3 had loss of ERBB2 amplifications. Loss of these alterations was confirmed upon review of the post-sample sequencing BAMs.

Figure 4: Patterns of acquired resistance.

A. Oncoprint showing commonly altered genes in between pre- and post-treatment samples. B. Patient timeline depicting sequential samples collected from an ERBB2 targeted therapy-treated patient with subsequent amplifications of MET and MYC. C. Patient timeline depicting sequential samples collected from an ERBB2 targeted therapy-treated patient with loss of the ERBB2 mutation. D. Patient timeline depicting sequential samples collected from a BRAF targeted therapy-treated patient.

Among the RTK/Ras-driven groups (ERBB2, FGFR2, and BRAF), a review of all altered RTK/Ras genes was performed to identify potential mechanisms of resistance (Fig. S4B).

Among the ERBB2 group, a 69-year-old patient presented with de novo metastatic GBC to the liver and peritoneum (Fig. 4B). Pretreatment tumor profiling demonstrated HER2 overexpression (IHC 3+) with ERBB2 amplification (19.3-fold change over normal) and 6 other alterations. The patient initially received FOLFOX due to hyperbilirubinemia for 3 months while undergoing biliary decompression, followed by gemcitabine and cisplatin for 3 months, with progression of disease as the best response to both treatments. The patient initiated zanidatamab, a HER2-targeted bispecific antibody, achieving a partial response (100% tumor shrinkage by RECIST v 1.1 criteria, Ca19.9 pretreatment 11,016, nadir 19) and remained on treatment for 24 months. A solitary escape lesion in the liver (HER2 positron emission tomography [PET] showing no uptake) was biopsied at progression and was found to have a lower level of HER2 overexpression (IHC 2+) and no evidence of ERBB2 amplification, but MYC (23.2-fold relative normal) and MET (8.2-fold change) amplifications that were not observed in the baseline tumor samples. This patient went on to receive off-label crizotinib with tumor shrinkage.

In a second case, a 41-year-old with metastatic GBC with disease progression after gemcitabine and cisplatin, and HER2 overexpression (IHC 3+), ERBB2 amplification (17.3 fold relative normal) and ERBB2 R678Q mutation in their pretreatment tumor had a partial response (84% RECIST v1.1 reduction) to zanidatamab which lasted for 14 months followed by progression in a common hepatic lymph node (Fig. 4C). HER2 PET demonstrated intense focal tracer avidity (SUV 14.5) suggestive of HER2 expression, and biopsy of a site of disease progression confirmed HER2 IHC3+ and retention of ERBB2 amplification (10-fold relative to normal). Given retention, the patient went on to receive trastuzumab detruxtecan with sustained tumor shrinkage ongoing for 16+ months. In the additional 2 patients with ERBB2 amp or single-nucleotide variant (SNV) loss, novel alterations identified at progression included MET and MYC amplification. In the 4 patients who retained ERBB2 at the time of disease progression on HER2-targeted therapy, novel alterations coincident with progression included KRAS and MYC amplifications (Fig. 4A).

In serial tumor and liquid biopsy of a patient with BTC harboring BRAF V600E, the emergence of oncogenic RAS and MEK SNV was identified over the course of treatment and at disease progression (Fig. 4D, S4C). As reported previously, we observed a second site mutation in FGFR2 (N549K) following treatment with an FGFR2 inhibitor in patients with FGFR rearranged BTC (27) as well as MET amplification at progression in a patient with an NTRK rearrangement treated with a TRK inhibitor (42) (Fig. S4B). Although we previously reported a case of IDH isoform switching, we did not observe this in other patients with IDH1 mutant iCCA treated with ivosidenib (26).

Discussion

This study leverages a large biorepository of patients with BTC with extensive clinical annotation, a uniform assay for molecular profiling with long follow-up time to investigate the molecular profile of BTC, the current state of clinical actionability of tumor molecular profiling, and the identification of clinicopathologic biomarkers of response and resistance to genomically matched therapy. Our findings confirm prior reports of frequent actionable genomic alterations in these diseases but also provide new insights regarding the genomic heterogeneity of BTC across anatomic sites, the variability of clinical outcomes following treatment with molecularly guided targeted therapies, and potential mechanisms of acquired drug resistance.

Notable limitations of our work include the single-center design, a proportion of missing data, potential lead time, and immortality bias in the comparative analysis of outcomes--which we attempted to mitigate by restricting the analysis from the start of second-line treatment, and genomic heterogeneity not captured by our NGS assay. Despite these limitations, we find 15–40% of cases harbored actionable alterations, supporting molecular profiling as a standard of care for all subsets of BTC. We were also able to quantify emerging molecular targets in BTC, including RAS alterations, MTAP deletion, and MDM2 and MET amplification. Ongoing clinical trials will clarify the predictive significance of these emerging genomic biomarkers to matched-targeted therapy.

Importantly, 22% of patients with an actionable alteration by current guidelines did not receive a targeted therapy due to impaired clinical status or death, and 8% of the entire cohort had insufficient sample purity to complete NGS. These data argue for sequencing at an earlier disease stage, re-biopsy, and/or undergoing molecular profiling through analysis of cfDNA.

The current portfolio of available targeted therapies in BTC is based largely on antitumor response (43). Large, comparative, phase 3 studies with the hard endpoint of survival have been largely infeasible (24); thus, the critical need to evaluate real-world evidence. Evaluation of our cohort clinically eligible for matched therapy revealed that second-line PFS was modestly longer in those patients who underwent matched-targeted treatment compared to those who received fluoropyrimidine-based chemotherapy, without a statistically significant longer OS. In comparison to other retrospective series, our data also affirms a PFS advantage with targeted therapy over chemotherapy and indicates a similar absolute OS improvement of 5 months (albeit not statistically significant)(44,45). Although these data support the use of targeted therapy in patients with advanced BTC, the totality of the data suggests that targeted monotherapy has a modest benefit that is largely restricted to a small subset of patients. Our data reinforce the need for novel prospective randomized studies to judge the full impact of targeted therapeutics on outcome and to consider combinatorial approaches when rationale and feasible (46).

Our paired biopsy cohort also provides evidence that at the time of disease progression on targeted therapy in BTC, genes predicted to reactivate the primary oncogenic pathway emerge in a subset of cases (i.e., RAF/MEK alterations in BRAF V600E and ERBB2 driven tumors, MET amplification in ERBB2 (47), and NTRK-driven tumors (42), and FGFR SNV in FGFR2 fused tumors (27,28,48)). To our knowledge, the current report includes the largest cohort of HER2-driven BTCs treated with HER2-targeted therapies with sequential samples for genomic analysis. In contrast to other RTK-driven BTCs, where the primary oncogenic driver was retained following treatment with targeted therapy in all cases, we observed ERBB2 alteration loss in 4 of 8 cases in the cohort at the time of disease progression on first-line HER2-targeted therapy—suggesting that preexisting genomic heterogeneity may impact treatment response and resistance. Although a hypothesis-generating observation, clinical anecdotes reported herein also indicate retention of ERBB2 amplification and overexpression after progression on initial HER2-targeted therapy may be rescued with sequential HER2-targeted therapy, as has been reported in other HER2-dependent tumors (49,50). Recurrent MET amplification at disease progression reported herein, along with recently published preclinical data, suggests a role for MET targeting, as well as combinatorial treatment with HER2-targeted therapy in HER2-driven tumors. Ongoing basket clinical trials utilizing novel MET-targeted therapeutics in BTC and MET amplified tumors will define antitumor activity of such approaches (e.g. NCT06084481). Furthermore, as a relatively high proportion of BTC have RAS alterations, and our paired cohort nominates RAS as a genomic escape mechanism, future studies should investigate RAS mutant allele-specific and pan-RAS inhibitors in these contexts. Notably, our NGS assay evolved over time, and MTAP deletions—which may have prognostic implications and/or may be implicated in acquired resistance—were not defined in the entire cohort, and this will require future study (51).

In sum, this cohort represents one of the largest single-center experiences of genetic profiling in the Western Hemisphere, which details a real-world experience in the implementation of targeted therapy in patients with advanced BTC. Our data support the continued use of precision medicine in BTC; however, they also highlight key limitations of this therapeutic approach. Future prospective studies should incorporate adequately sized control groups and repeat biopsy on clinical trials to define the full spectrum of resistance to inform novel combinatorial strategies.

Translational Relevance:

Comprehensive genomic profiling is increasingly incorporated into the clinical management of biliary tract cancers (BTC), yet the real-world benefit of matched-targeted therapies remains incompletely defined. In this large prospective cohort, we demonstrate that actionable alterations are common across BTC subtypes but that targeted therapy confers improvement in progression-free survival without a corresponding overall survival advantage compared with cytotoxic chemotherapy. Our findings identify recurrent genomic contexts associated with inferior outcomes and resistance, including co-occurring TP53/RAS pathway and SMAD4 alterations, as well as emergent bypass pathway activation involving RAS, MEK, MET, and cell-cycle regulators. These data support the routine use of next-generation sequencing to guide therapeutic selection while highlighting the need for combination strategies and adaptive treatment approaches to overcome resistance. Integration of resistance-informed molecular profiling into clinical trial design and treatment sequencing may improve durable responses and optimize precision oncology strategies for patients with BTC.