Clinical utility and prognostic implications of circulating cell-free DNA in biliary tract cancer

Abstract

BACKGROUND:

An estimated 25% of patients with biliary tract cancer (BTC) do not undergo genotyping, representing a missed opportunity for therapeutic targeting.

METHODS:

Cell-free DNA (cfDNA) and matched tumor sample from patients with BTC were analyzed using targeted next-generation sequencing (NGS) assay and compared. We sought to define the molecular profile of cancer-derived cfDNA, frequency of OncoKB level 1/2 alterations, plasma: tumor genotype concordance, the prognostic impact of cfDNA, and clonal evolution after targeted therapy progression.

RESULTS:

cfDNA-based genotyping was performed on 297 blood samples from 170 patients with BTC. The most frequently altered genes were TP53 (29%), FGFR2 (16%), ARID1A (13%), CDKN2A (11%), and KRAS (11%); 25% of patients had OncoKB level 1/2 alterations and 36.2% of potentially actionable alterations were detected in plasma alone. The cfDNA: tissue concordance accuracy was high (96% IDH1, 98% BRAF, 92% KRAS mutations, 99% ERBB2 amplifications, 96% FGFR2 fusions). Detectable tumor-derived cfDNA after resection did not predict recurrence. In treatment-naïve metastatic BTC, high variant allele fraction was associated with worse progression-free and overall survival. RAS alterations not detected in samples pre- treatment were identified at progression in 24% of patients who received BRAF−, FGFR−, or HER2-directed therapy, identifying RAS alterations as a convergent mechanism of targeted therapy resistance.

CONCLUSIONS:

Molecular profiling of cfDNA from patients with BTC identified OncoKB level 1/2 gene alterations and putative genomic resistance mechanisms to targeted therapy. Concordance analysis suggests that cfDNA-based NGS is complementary to that of tissue-based sequencing in the identification of potentially actionable alterations.

INTRODUCTION

Patients with biliary tract cancer (BTC), comprising intrahepatic cholangiocarcinoma (iCCA), extrahepatic cholangiocarcinoma (eCCA), and gallbladder cancer (GBC), have a poor prognosis. For those with locally advanced or metastatic BTC treated with gemcitabine and cisplatin plus immune checkpoint blockade, median overall survival (OS) is 12.8 months ^1,2. Molecular profiling studies have identified BTC driven by oncogenic IDH1, FGFR2, ERBB2, BRAF, RET, and NTRK alterations, and several precision medicines have received regulatory approval for the treatment of patients with these alterations ^3–11. Within the current landscape of targeted therapeutics for patients with BTC, approximately 40% of iCCA and ~20% of eCCA and GBC harbor OncoKB level 1 or 2 genomic alterations (i.e., a Food and Drug Administration [FDA]–approved indication or National Comprehensive Cancer Network–recognized off-label use in BTC) ^12–16.

One barrier to precision oncology for patients with BTC is the limited tumor tissue available for genomic profiling ¹⁶. Tissue biopsies may be associated with hepatobiliary injury or bleeding; may not be feasible due to tumor location, comorbidities, or cancer-associated complications; or may produce limited or low-purity tumor tissue for molecular profiling. Tissue biopsies are also prone to sampling bias and may not capture tumor heterogeneity ¹⁷. It is estimated that up to 25% of patients with BTC cannot undergo molecular profiling ¹⁸. Orthogonal methods for detecting actionable genomic alterations are required to facilitate optimal therapy selection for all patients with BTC.

Collection and analysis of tumor-derived cell-free DNA (cfDNA) may address many of the limitations associated with tumor tissue genotyping. cfDNA analysis is minimally invasive and rapid, can be repeated serially to allow real-time dynamic monitoring of tumor evolution and treatment response, and may better reflect the entire tumor burden by showing spatial genomic heterogeneity ¹⁹. Furthermore, there is potential to detect minimal residual disease (MRD) prior to clinical or radiographic evidence of disease, aid in prognostication, and match patients to appropriate targeted therapeutics. We sought to determine the feasibility of targeted genotyping of cfDNA from patients with BTC, its concordance with tissue-based genomic profiling, and its prognostic significance in patients with locally advanced and metastatic BTC.

METHODS

Study Design

This was a single-center, prospective genotyping study of patients with BTC who underwent molecular profiling of cfDNA using the MSK-ACCESS (Memorial Sloan Kettering– Analysis of Circulating cfDNA to Examine Somatic Status) assay at Memorial Sloan Kettering Cancer Center (MSK) from April 2014 to September 2022. The primary objective was to determine the landscape of genomic alterations via next-generation sequencing (NGS) of cfDNA extracted from the blood of patients with BTC. Secondary objectives included assessing genotype concordance between sequencing from cfDNA extracted from blood and DNA extracted from tumors, the fraction of patients who had BTC with clinical actionable genomic alterations identified by cfDNA analysis, the potential prognostic significance of cfDNA detection and variant allele frequency (VAF) in patients with early-stage and advanced disease, and the ability of cfDNA to nominate resistance alterations in patients who had responded to and then progressed on treatment with genotype-matched targeted therapies.

Patient demographic data, including sex, date of birth, and sample types used for genomic profiling, were collected. 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; date of progression, and date of death or last follow-up.

Genomic Analysis

All patients underwent at least one peripheral blood collection using cfDNA blood collection tubes (Streck, La Vista, NE). Matched leukocyte DNA was used as normal control to identify germline variants and alterations consistent with clonal hematopoiesis. Extracted cfDNA underwent NGS testing using MSK-ACCESS, a hybridization and deep-sequencing assay approved for clinical use by New York State, which employs duplex unique molecular indexes and collapsing of replicate read pairs to filter polymerase chain reaction and sequencing-based artifacts ²⁰. MSK-ACCESS was designed to detect genomic alterations in 129 cancer-associated genes (Table S1).

Tumor tissue was analyzed using MSK-IMPACT (MSK–Integrated Mutation Profiling of Actionable Cancer Targets), an FDA-authorized hybridization capture-based targeted NGS assay that detects mutations, copy number alterations, and select structural rearrangements in 341–505 cancer-associated genes, depending on the version of the panel ²¹. MSK-IMPACT achieves a high depth of sequencing (~800x) and is performed in a Clinical Laboratory Improvement Amendments–certified molecular laboratory (Table S2). Somatic mutations, copy number alterations, and structural variants were called for each cfDNA and tumor sample using previously published methods. Tumor mutational burden was calculated as total number of nonsynonymous mutations per megabase sequenced. OncoKB, a precision oncology knowledge base of clinically actionable variants, was used to identify potentially actionable genomic alterations ¹². Genes were grouped into pathways using curated templates from The Cancer Genome Atlas ²².

Average VAF was calculated as the sum of the number of reads containing detected variants in MSK-ACCESS divided by the sum of overall coverage at those loci for which variants were identified. Maximum VAF was calculated as the single highest VAF at any loci. VAFhigh was defined as > median VAFmax. All data are available for visualization and analysis via the cBioPortal for Cancer Genomics (http://cbioportal.org/).

Statistical Analysis

Categorical variables were summarized using frequencies and percentages and continuous variables using median and interquartile range (IQR) and range. Odds ratios (ORs) were used to determine differences in detection rates within different timing groups. A generalized estimating equation was used to control for within-subject correlation, since patients were in multiple groups and had multiple samples within groups.

Concordance was calculated using the closest blood sample following the patient’s first tissue sample. Fourteen paired tissue samples were not included as they were acquired post- collection of blood samples for cfDNA analysis. Concordance of mutation calling between tissue and plasma was measured using overall agreement and critical success index (CSI). Overall agreement was calculated as [(true positive + true negatives) / (true positives + false positives + true negatives + false negatives)]. CSI was calculated as [true positives / (true positives + false negatives + false positives). Overall accuracy could be inflated when the proportion of negative/negative calls was high, whereas CSI removed these calls where both tests were negative. If both accuracy and CSI were high, we were more confident the tests were in agreement. If accuracy was high and CSI low, we knew that one test was over or under calling a mutation. A 5% prevalence threshold in either blood or tissue data was used for a mutation to be included in the concordance analysis.

Kaplan-Meier methods and Cox proportional hazard models were used to analyze progression-free survival (PFS) and OS. Patients with a cfDNA sample collected in the perioperative setting were included in the MRD cohort, which used landmark analysis to analyze OS. The landmark was defined as 12 months post-surgery. A patient must have had at least one cfDNA sample prior to the landmark. A patient had positive MRD if any sample between surgery and the landmark was positive, with OS defined as time from surgical landmark to death or last follow-up. Association of PFS and OS with tumor-derived cfDNA VAF was analyzed in a subset of patients who had systemic-treatment naïve stage IV disease with cfDNA samples within 14 days of first chemotherapy (n=60). OS was defined as time from sample collection to death or last follow-up. PFS was defined as time from sample collection to progression, death, or last follow- up.

SAS 9.4 (Cary, NC) or R (version 4.4.1; R Foundation for Statistical Computing, Vienna, Austria) was used for all analyses. All tests were two-sided. P <.05 was considered significant.

Ethics

The study was approved by the MSK Institutional Review Board (IRB) and conducted in accordance with the US Common Rule. All samples were collected following written informed consent on an IRB-approved research protocol (MSK IRB 12–245; NCT01775072).

RESULTS

Patient Characteristics

NGS of cfDNA was performed on 297 samples from 170 unique patients with BTC between 2014 and 2022. Patient characteristics are listed in Table 1. The majority had iCCA (121, 71%) followed by eCCA (29, 17%) and GBC (20, 12%). Most patients (146, 86%) had locally advanced unresectable or metastatic disease.

Table 1:

Cohort Demographics

	Total (N = 170)
Sex
Female	85 (50%)
Male	85 (50%)
Age at diagnosis
Median (IQR)	64.7 (56.5 – 72.4)
Race
Asian	17 (10%)
Black	5 (2.9%)
Hispanic	18 (10.6%)
White	130 (76.5%)
Anatomic site
Extrahepatic	29 (17.1%)
Gallbladder Cancer	20 (11.8%)
Intrahepatic	121 (71.2%)
CA-19-9 at time of cfDNA1
Median (IQR)	78.5 (24.0 – 391.0)

Abbreviations: Interquartile range, IQR; Number of patients, N

Genomic Landscape

From the 170-patient cohort, 297 samples were sequenced to a median duplex coverage depth post read-collapsing of 1360x (range 460x-2730x). Genetic alterations were detected in 75% of plasma samples (222/297) and 92% of all patients (156/170). The detection of a mutation in cfDNA was more likely in patients for whom samples were collected in the setting of treatment- naïve advanced disease (59/69 [85%]; OR, 3.35; 95% confidence interval [CI], 1.62–6.96; P=0.001) or in those progressing on systemic therapy (69/78 [88%]; OR, 2.54; 95% CI, 1.36–4.77; P=0.004), compared to samples collected when patients were responding to systemic therapy (42/62 [68%]). There was no statistically significant difference in the likelihood of detecting tumor-derived cfDNA in samples taken before definitive surgery (17/24 [71%]; OR, 1.81; 95% CI, 0.58–5.66; p=0.309) or after definitive surgery (34/64 [53%]; OR, 0.97; 95% CI, 0.34–2.75; P=>0.95) compared to those responding to systemic therapy. The median average VAF per sample was.0029 (range: 0,.49, IQR: 0,.019) (Table S3 and S4).

The most frequently mutated genes were TP53 (29%), FGFR2 (16%), ARID1A (12%), CDKN2A (11%), and KRAS (11%) (Figure 1A). Levels 1 and 2 actionable genomic alterations were identified in 24.7% of all patients (Figure 1B), most commonly in IDH1 (10.0%: iCCA 14.0%, eCCA 0%, GBC 0%), FGFR2 fusions (9.4%: iCCA 13.2%, eCCA 0%, GBC 0%), ERBB2 amplification (2.9%: iCCA 3.3%, eCCA 0%, GBC 5.0%), BRAF V600E (2.4%: iCCA 3.3%, eCCA 0%, GBC 0%) and KRAS^G12C (1.1%: iCCA 1.7%, eCCA 0%, GBC 0%) (Figure 1C). Emerging therapeutic targets such as KRAS^G12D, KRAS^G12V, and MDM2 and MET amplifications were identified in 2.9%, 4.7%, 2.4%, and 0.6%, respectively. MSI was not identified.

Figure 1A:

Oncoprint of genomic landscape in first cfDNA sample per patient stratified by anatomic subtype, N =170 B: Frequency of OncoKB level alteration detection; 1C: Bar chart frequencies of select targetable alterations in the first sample per patient of treatment-naïve advanced disease cohort. Abbreviation: cell-free DNA, cfDNA

In patients whose tumor NGS analyses were unsuccessful or did not undergo tissue-based sequencing due to sample quality or inadequate tissue, 36.2% (21/58) had a potentially actionable alteration (OncoKB level 1, 2 or 3a/b) identified in plasma, including 5 patients with FGFR2 fusions, 3 patients with IDH1 R132 mutations, and one patient with a BRAF V600E mutation.

Concordance Between Tissue and Plasma Genotype

Patient-matched tumor-based NGS sequencing of 98 patients was included in the concordance analysis. Within this cohort, frequencies of alterations identified were detected at similar rates: KRAS (tissue 12.2% vs. plasma 8.2%) TP53 (32.7% vs. 28.6%) IDH1 (15.3% vs.13.3%), FGFR2 (21.4% vs. 17.3%), and BRAF (7.1% vs. 5.1%), ERBB2 (8.2% vs 6.1%), and CDKN2A (23.5% vs 10.2%) (Figure 2A). The overall rate of potentially actionable alterations detected was numerically higher for tissue-based sequencing compared to plasma (65.3% vs 41.8%) (Figure 2B).

Figure 2:

A. Detection of common alterations in tissue and plasma; B. Frequencies of specific OncoKB level 1–3b alterations detected; C: Concordance of presence or absence of alteration detection in tissue and plasma

Concordance at the individual gene level for all paired samples varied on the single nucleotide variant detected, such as oncogenic mutations in IDH1 (CSI 75%, accuracy 96%), KRAS (CSI 43%, accuracy 92%), TP53 (CSI 69%, accuracy 89%), ARID1A (CSI 59%, accuracy 91%), BRAF (CSI 71%, accuracy 98%) (Figure 2C, Table S5). Structural variants, amplifications, and deletions were detected at lower frequencies between cfDNA and tissue overall but had concordance rates similar to that observed for mutations: FGFR2 fusions (tissue 13.3% vs plasma 10.2%; CSI 71%, accuracy 96%), CDKN2A deletions (tissue 10.2% vs. plasma 0%; CSI 0%, accuracy 90%), and ERBB2 amplifications (tissue 5.1% vs. plasma 4.1%, CSI 80%, accuracy 99%). (Figure 2C)

Minimal Residual Disease

Fifty (29%) patients in the cohort underwent a curative intent resection of their disease (Table S6). Among this subset, 24 had at least one blood sample for cfDNA analysis collected prior to surgery, 23 had at least one sample collected post-resection as part of surveillance, and 21 patients had paired pre-surgical and at least one post-surgical cfDNA plasma sample analyzed (Figure 3a, Table S6). Twenty-two patients had at least one post-operative cfDNA analysis, and 13 patients had 2 or more timepoints sampled after definitive surgery for the monitoring of MRD (61 samples; range 2–8 samples per patient). The median time prior to resection for sample acquisition was 0.3 months (range, 0.1–9.3) and post-surgery was 0.2 months (range, 0.1–10.2). Tumor-associated genomic alterations were detected in 13/23 (57%) blood samples collected pre- surgery and 7/22 (32%) of the first samples collected post-surgery (Table S7).

Figure 3.

a: Swimmer’s plot of cfDNA sample timing and clinical outcomes for patients who underwent definitive curative intent resection; 3b: Overall survival stratified by positive cfDNA detected any time within the first year post-surgery. Abbreviation: cell-free DNA, cfDNA

We did not observe a statistically significant difference in recurrence-free survival or OS following resection in patients based on clearance of tumor-derived cfDNA (Figure 3a). Any positive cfDNA sample postoperatively trended toward a shorter OS compared to those that had no positive post-operative sample (38 [3.1-not reached(NR)] vs NR [27-NR] months, P=0.304) (Figure 3b), a result likely limited by sample size.

cfDNA in Advanced Disease

Sixty patients were treatment-naïve at the time of cfDNA NGS assessment and commenced systemic treatment within 14 days from the time of plasma sampling (Table S8). Tumor-derived cfDNA was detected in 83% of patients with a median average VAF of 0.012 (IQR, 0.001–0.038) and the median VAFmax detected being 0.014 (IQR, 0.002–0.058). VAFhigh defined as above the median VAFmax was set at ≽0.014.

PFS or OS was not statistically different regardless of whether any tumor-derived cfDNA was detected. However, when adopting the threshold set by the median, VAFhigh predicted worse PFS-to-first-line treatment and OS. Those with VAFhigh had a shortened PFS (VAFhigh: median PFS, 4.47 months [2.17–6.48] vs. VAFlow: median PFS, 11.84 months [5.39–18.42]; P=0.001; Figure 4a) and OS (VAFhigh: median OS, 14.05 months [5.39–19.97] vs. VAFlow: median OS, 32.04 months [13.82–39.93]; P=0.002; HR, 3.14; Figure 4b).

Figure 4.

a: Progression-free survival stratified by high/low VAF; 4b Overall survival stratified by high/low VAF. Abbreviation: variant allele frequency, VAF

Resistance Alterations Post Targeted Therapy

Serial cfDNA samples were analyzed post-progression for 28 patients with BTC treated with targeted therapy, including 25 treated with BRAF-, FGFR- or HER2-targeted therapies (Figure 5). RAS alterations not detected in pre-treatment samples were detected in 6 of 25 patients (24%), including 3 of 5 patients who received BRAF/MEK inhibition for BRAF V600E+ BTC (emergent KRAS G12D in 1, NRAS Q61K in 1, polyclonal KRAS G12D, G12V, G13D in 1), 2 of 13 patients who received FGFR-targeted therapy for FGFR2 fusion+ BTC (emergent KRAS Q61H in 1, NRAS Q61R in 1), and one of 7 patients who received HER2-targeted therapy for ERBB2- amplified BTC (emergent KRAS G12C). An additional patient receiving HER2-targeted therapy had KRAS amplification detected on tissue-based sequencing in a post-treatment sample. Serial cfDNA profiling revealed canonical FGFR2 resistance mutations in 3 of 13 patients who received FGFR-targeted therapy (acquired N549K in n = 1, N549T in n = 1, and N549K and N549H in n = 1). Among 7 patients treated with HER2-targeted therapy, cfDNA profiling at resistance revealed loss of index ERBB2 amplification in 4, emergent ERBB2 resistance mutation in one, emergent MYC amplification in one, and emergent MET amplification in one.

Figure 5:

Oncoprint of paired post-progression cfDNA samples

DISCUSSION:

Our analyses indicate that targeted NGS of tumor-derived cfDNA using MSK-ACCESS identifies a spectrum of oncogenic alterations in the peripheral blood of patients with BTC. The ability to detect genomic alterations in cfDNA is dependent on disease stage and status relative to therapy, with a higher likelihood of detection for patients with treatment-naïve, advanced, or metastatic, or progressing disease. Importantly, profiling of cfDNA was able to identify level 1/2/3 alterations in 36.2% of cases that did not have available tissue for genotyping. OncoKB level 1/2 alterations were detected in 24.7% of patients, but concordance to tumor genotype varied based on the underlying alteration. As in other reports, detection of ERBB2 amplifications and, more so, FGFR2 fusions were discordant between tissue and blood, suggesting value for both tumoral and cfDNA targeted NGS ^23,24. In addition to known OncoKB level 1/2 alterations, we identified emerging targets including KRAS isoforms and MDM2 amplification. Higher levels of VAF were associated with worse outcomes for patients with advanced disease on first-line systemic treatment. Finally, in a cohort of patients with BTC amenable to genomic alteration and treated with targeted therapy, we observed evidence of emergence of genomic alteration predicted to reactivate the MAPK pathway.

Several reports have investigated the use of targeted NGS to genotype cfDNA extracted from patients with BTC, though most cohorts included a small number of cases, were geographically restricted to patients from China, and/or lacked clinical annotation ^23,25–29. With a sample size of 170 comprehensively clinico-genomically–annotated patients with BTC, our cohort is complementary to a recent multicenter study of 225 clinically annotated cases ²³. In contrast to other reports, our assay utilizes normal blood control–eliminating incorrect interpretation from clonal hematopoiesis and increases the accuracy of genotyping, improving the sensitivity for identification of true somatic variants. The design of our study, a “real-world” dataset which included all stages of disease, provides useful clinical information on appropriate timepoints for testing to maximize detection and mutational profiling. The ability to serially sequence through targeted treatment allowed for the nomination of several mechanisms of resistance, particularly in ERBB2- and RAF- driven tumors, which to our knowledge have not been reported to date in BTC.

Acknowledging the limitation of a small perioperative cohort with heterogeneous post- operative treatment, we did not observe a statistically significant difference in relapse-free survival or OS in patients with either persistently negative or cleared cfDNA following surgical resection. More sensitive assays such as those informed by tumor or methylation signatures may be required for improved detection of MRD, and ongoing prospective adjuvant studies (e.g., NCT06109779, NCT03079427) will clarify the role of MRD via cfDNA analysis. We observed that VAFhigh in treatment-naïve patients was associated with a poor outcome, though prospective validation accounting for known and unknown confounding factors, not accessible in our analyses, are needed ²⁹.

Given high concordance between tissue and cfDNA targeted NGS for mutations and amplifications, our results demonstrate frequent emergence of acquired RAS mutations among patients with BRAF−, FGFR− or HER2-targeted therapy resistance and frequent loss of index ERBB2 amplifications among patients with HER2-targeted therapy resistance, suggesting that cfDNA profiling may be used to define the biology of resistant disease, identify next-line therapeutic strategies, and nominate combinatorial approaches that might be used to delay resistance.

Our study has several other limitations. It was conducted retrospectively at a single center and has missing data. The limitation of using targeted sequencing of cfDNA to evaluate for loss of an index driver alteration (e.g. ERBB2 amplification) is low sensitivity in the setting of low disease burden/cfDNA levels.

This work adds to the growing body of evidence that targeted sequencing of cfDNA in patients with advanced BTC is independently important and complementary to that of tissue-based NGS. Beyond selection of patients for targeted therapy, MSK-ACCESS has the potential to function as both a prognostic and predictive biomarker and may identify novel mechanisms of resistance to targeted therapy.

Supplementary Material

PV Data Supplement

Supplementary Tables:

Table S1: List of genes included in the MSK-ACCESS panel

Table S2: List of genes included in MSK-IMPACT panel

Table S3: Generalized estimate equation of sample positivity depending on clinical timing

Table S4: Odds ratio for detection of positive cfDNA depending on clinical timing of sample acquisition, with those taken at the time of responding to therapy being the reference value.

Table S5: Accuracy and critical success index for alterations in cfDNA and paired tissue

Table S6: Demographics for minimum residual disease cohort

Table S7: Clinical outcomes associated with perioperative cfDNA samples

Table S8: Demographics for treatment-naïve advanced disease cohort

Context Summary.

Key Objective:

Does circulating cell-free DNA (cfDNA) analysis offer a clinically valuable, non-invasive alternative to tissue biopsy for comprehensive genomic profiling and resistance monitoring in biliary tract cancer (BTC)?

Knowledge Generated:

Targeted cfDNA sequencing identified actionable alterations in BTC with high concordance to tissue. High cfDNA variant allele fraction predicted worse survival, and serial profiling revealed emergent mutations as a resistance mechanism post-targeted therapy.

Relevance:

cfDNA profiling serves as a crucial, minimally invasive tool for optimizing therapeutic selection, monitoring disease progression, and identifying resistance pathways in BTC patients, especially when tissue is limited. This can directly inform treatment decisions and improve patient outcomes.

Data availability statement:

The data underlying this article are available in Cbioportal, pending link.