Concordance in Oncogenic Alterations Between the Primary Tumor and Advanced Disease: Insights Into the Heterogeneity of Intrahepatic Cholangiocarcinoma

Abstract

PURPOSE

Intrahepatic cholangiocarcinoma (ICCA) is characterized by significant phenotypic and clinical heterogeneities and poor response to systemic therapy, potentially related to underlying heterogeneity in oncogenic alterations. We aimed to characterize the genomic heterogeneity between primary tumors and advanced disease in patients with ICCA.

METHODS

Biopsy-proven CCA specimens (primary tumor and paired advanced disease [metastatic disease, progressive disease on systemic therapy, or postoperative recurrence]) from two institutions were subjected to targeted next-generation sequencing. Overall concordance (oncogenic driver mutations, copy number alterations, and fusion events) and mutational concordance (only oncogenic mutations) were compared across paired samples. A subgroup analysis was performed on the basis of exposure to systemic therapy. Patients with extrahepatic CCA (ECCA) were included as a comparison group.

RESULTS

Sample pairs from 65 patients with ICCA (n = 54) and ECCA (n = 11) were analyzed. The median time between sample collection was 19.6 months (range, 2.7-122.9). For the entire cohort, the overall oncogenic concordance was 49% and the mutational concordance was 62% between primary and advanced disease samples. Subgroup analyses of ICCA and ECCA revealed overall/mutational concordance rates of 47%/58% and 60%/84%, respectively. Oncogenic concordance was similarly low for pairs exposed to systemic therapy between sample collections (n = 50, 53% overall, 68% mutational). In patients treated with targeted therapy for IDH1/2 alterations (n = 6) or FGFR2 fusions (n = 3), there was 100% concordance between the primary and advanced disease specimens. In two patients, FGFR2 (n = 1) and IDH1 (n = 1) alterations were detected de novo in the advanced disease specimens.

CONCLUSION

The results reflect a high degree of heterogeneity in ICCA and argue for reassessment of the dominant driver mutations with change in disease status.

The low concordance between primary and recurrent/metastatic IHC reflects a high degree of genomic heterogeneity.

INTRODUCTION

Cholangiocarcinoma (CCA) is a rare and diverse group of malignancies that arise along the biliary tree, from the intrahepatic radicals to the ampulla of Vater.¹ Although surgery remains the only potentially curative therapy, less than one third of patients present with resectable disease, and many who undergo resection recur.^2-4 Systemic chemotherapy plus either durvalumab, a PD-L1 inhibitor, or pembrolizumab, an immune checkpoint inhibitor, remains the first-line treatment for patients with unresectable disease, but conveys only modest survival benefits (median <13 months).^5-7

CONTEXT

• Key Objective

• What is the concordance in oncogenic alterations between primary tumor and advanced disease in cholangiocarcinoma (CCA)? This study aimed to gain insight into the genomic heterogeneity of intrahepatic CCA (ICCA) by comparing the primary and advanced diseases.

• Knowledge Generated

• The genomic profile of the primary CCA differs from that of the advanced disease (concordance = 49%), with this difference exaggerated in ICCA (concordance = 47%) compared with extrahepatic CCA (concordance = 60%). In some patients, targetable alterations (eg, FGFR2, IDH1) were identified in the advanced disease but not in the primary tumor.

• Relevance

• ICCA demonstrates significant heterogeneity across stages of disease, likely reflecting its polyclonal genetic makeup and sampling variance over both the anatomic location and time. Our results advocate for reassessment of dominant driver alterations with change in disease status.

Recent efforts have yielded a better understanding of the genomic landscape of CCA, resulting in effective targeted therapies in a minority of patients harboring certain alterations, particularly in intrahepatic CCA (ICCA).^8-14 However, these studies have also highlighted the marked genomic heterogeneity of ICCA, which likely contributes to its characteristic phenotypical divergencies, biologic distinctiveness, treatment resistance, and propensity for recurrence and/or metastasis.

Common alterations in ICCA include KRAS, TP53, ARID1A, CDKN2A/B, IDH1/2, and FGFR2, among others.^15,16 However, no individual alteration accounts for more than 30% of cases. Some of these alterations carry prognostic value, such as KRAS, TP53, and CDKN2A, which are associated with poor prognosis, and there is significant alteration interaction with described co-occurrences and mutual exclusivity.^15,17 Given this degree of genomic heterogeneity, genomic profiling is used increasingly to identify patients with actionable or prognostic biomarkers.

Genomic concordance, or the degree to which the genomic alteration profile of one tumor reflects that of another, was high in one study across intraindividual multifocal ICCA lesions, suggesting a common monoclonal origin and genetic similarity.⁹ It has been suggested, nonetheless, that founder cell populations likely coexist with clonal subtypes, the phylogeny of which is driven by changes in tumor microenvironment.¹⁶ However, the degree of genomic alteration concordance between the primary tumor and advanced ICCA disease remains unknown, obscuring an understanding of the intertemporal heterogeneity of disease and presenting a challenge for personalized therapy across disease stages.

As different cancers demonstrate varying degrees of genomic concordance across stages,^18-20 which might have therapeutic implications, this study aimed to determine the genomic concordance between ICCA primary tumors and advanced disease and the degree of such concordance compared with tumors from patients with extrahepatic CCA (ECCA).

METHODS

Study Design and Patient Selection

This study included patients with ICCA and ECCA from the Memorial Sloan Kettering Cancer Center and the Duke University Medical Center with histologically confirmed tissue from the primary tumor (initial presentation, regardless of stage) and paired tissue from advanced disease. Advanced disease was defined as metastatic disease (disease identified at a distant site, either on systemic therapy or after resection), progressive disease (local disease progression on systemic therapy), or postoperative recurrence (recurrent disease in the liver after resection). This cohort included both patients who underwent curative-intent resection and those who were managed nonoperatively. Institutional Review Board approval was granted from both centers with a waiver of consent (Protocol No. 16-635).

Data Collection

Clinical data were obtained through prospectively maintained databases and electronic medical record review, including classification as ICCA or ECCA, sex, race, age, therapy before sample collection (including chemotherapy, hepatic artery infusion pump [HAIP] chemotherapy, and targeted therapy [IDH1 inhibitors, FGFR2 inhibitors, PARP inhibitors, BRAF inhibitors, ALK inhibitors, VEGF inhibitors, and other immune checkpoint inhibitors]), whether curative-intent resection had been performed, margin status, multifocality of disease, tumor grade, lymphovascular invasion, perineurial invasion, lymph node status (positive defined as histologically positive lymph nodes or clinically/radiographically positive lymph nodes), biopsy site (liver/bile duct, lymph node, metastasis), and time interval between sample collections.^21-23

Genomic Profiling

Each patient had biopsy samples representative of primary and advanced diseases, along with matched nontumor tissue or blood. All specimens were sequenced using the Memorial Sloan Kettering–Integrated Mutation Profiling of Actionable Cancer Targets (IMPACT) assay, a clinically validated hybridization capture–based targeted next-generation sequencing array capable of detecting mutations, copy number alterations (CNAs), and certain structural rearrangements.²⁴ Specific IMPACT panels used included a 468-gene panel (n = 53, 40.8%), a 505-gene panel (n = 52, 40.8%), a 410-gene panel (n = 20, 15.4%), and a 341-gene panel (n = 5, 3.8%).

Genes with recurrent alterations occurring in ≥5% of the sample were identified. Genetic alterations were filtered for oncogenic variants using OncoKB (version 4.7), a precision oncology database that tracks driver gene alterations and called against the matched normal samples.²⁵ Genomic data are available on cBioPortal.²⁶ In addition, the following pathways were evaluated: cell cycle, DNA damage repair, Hippo, Myc, Notch, nuclear factor erythroid 2–related factor (NRF2), phosphoinositide 3-kinase (PI3K), receptor tyrosine kinase (RTK/Ras), TGF-β, TP53, and Wnt.

Tumor mutational burden (TMB), a surrogate marker of DNA repair pathway integrity, was defined as the total number of somatic protein mutations divided by coding regions captured by IMPACT.²⁷ FACETS (PMID: 27270079) and the facets-suite R package (GitHub²⁸) were used to assess tumor purity and generate estimates of chromosomal instability (fraction of genome altered [FGA]).²⁹

Statistical Analysis

Categorical variables are expressed as number (percentage) and compared using the Fischer’s exact test, and continuous variables are expressed as median (IQR) and compared using the Wilcoxon rank-sum test. For all comparisons, a two-sided significance level was set a priori at P < .05. Concordance was defined as the percentage of all alterations (shared, unique to primary disease and unique to advanced disease) that were shared between primary and advanced samples. Concordance is reported as overall concordance (all alterations, including oncogenic mutations, CNAs, and structural rearrangements) and mutational concordance (only oncogenic mutations). Pathway concordance was recorded if alterations were observed in the same pathways between disease stages. All statistical analyses were performed using R/RStudio (version 3.14).

RESULTS

Patient and Tumor Characteristics

We identified 65 patients diagnosed with CCA between 2002 and 2021 (Table 1); 54 (83%) had ICCA, and 11 (17%) had ECCA. Of the advanced disease samples, 34 of 65 (52%, n = 25 ICCA, n = 9 ECCA) represented metastatic disease, 17 of 65 (26%, n = 17 ICCA) represented progression on systemic therapy, and 14 of 65 (22%, n = 12 ICCA, n = 2 ECCA) represented recurrent disease after resection. Patients with ICCA were less likely to undergo curative-intent resection (56% v 91%) and more likely to have poorly differentiated tumors (33% v 9%). ECCA was associated with a greater median time between sample collections compared with ICCA (37 v 19 months, respectively).

TABLE 1.

Baseline Demographic, Clinical, and Pathologic Characteristics

Characteristic	ICCA	ECCA
No. (%)	54 (83)	11 (17)
Age, years, median (IQR)
At first sample collection	58 (46-65)	59 (54-71)
At second sample collection	60 (48-67)	62 (55-75)
Male sex, No. (%)	25 (46)	7 (64)
Race, No. (%)
White	35 (65)	5 (46)
Black	1 (2)	0
Hispanic	2 (4)	0
Asian	2 (4)	4 (36)
Native American	1 (2)	0
Unknown	13 (24)	2 (18)
Characteristics of primary samples, No. (%)
Chemotherapy exposure before first sample collection	5 (9)	1 (9)
Unknown	1	0
Curative-intent resection	30 (56)	10 (91)
R0 resection	25 (81)	7 (78)
Unknown	0	1
Multifocal disease	6 (21)	0 (0)
Unknown	25	2
Tumor grade
Well differentiated	0	3 (27)
Moderately differentiated	35 (67)	7 (64)
Poorly differentiated	17 (33)	1 (9)
Unknown	2	0
Lymphovascular invasion	17 (57)	6 (67)
Unknown	24	2
Perineural invasion	12 (50)	7 (78)
Unknown	30	2
Positive lymph node	7 (37)	4 (57)
Unknown	35	4
Site
Liver/bile duct	51 (94)	9 (82)
Lymph node	1 (2)	1 (9)
Metastasis	2 (4)	1 (9)
Characteristics of advanced disease samples
Time between sample collections, months, median (IQR)	19 (8-29)	37 (14-63)
Type of advanced disease, No. (%)
Metastasis	25 (46)	9 (82)
Progression	17 (31)	0
Recurrence	12 (22)	2 (18)
Therapy between samples, No. (%)
Chemotherapy	46 (89)	10 (91)
Hepatic artery infusion pump	7 (13)	0
Targeted therapy	14 (27)	1 (9)
FGFR inhibitor	3 (6)	0
IDH1 inhibitor	6 (12)	0
Site, No. (%)
Liver/bile duct/gallbladder	30 (56)	4 (36)
Lymph node	4 (7)	4 (36)
Metastasis
Lung/pleural fluid	12 (22)	1 (9)
Adrenal	1 (2)	0
Other	7 (13)	2 (18)

Abbreviations: ECCA, extrahepatic cholangiocarcinoma; ICCA, intrahepatic cholangiocarcinoma.

Overall Genomic Landscape and Concordance

We identified 849 alterations (367 concordant, 43%; Fig 1A), including both oncogenic driver alterations and bystander alterations. The most common oncogenic driver alterations occurred in ARID1A, TP53, CDKN2A, KRAS, BRAF, IDH1/2, BAP1, ARID2, PBRM1, and FGFR2. The most common bystander alterations occurred in FAT1, ARID1B, KMT2C, ATM, and REL. Of these 849, 320 (38%; 198 mutations, 100 CNAs, 22 fusion events) were oncogenic. Forty-nine percent of these oncogenic alterations were concordant between pairs (62% mutations, 22% CNAs, 59% fusion events). Tumor purities were similar between primary and advanced samples (median 0.36 primary v 0.38 advanced; P = .64; Appendix Fig A1). Of note, cumulative concordance in pairs with at least one sample purity <20% was 60%, whereas it was 49% for pairs with both tumor purities ≥20% (P = .26).

FIG 1.

(A) Percentage of mutations shared and unique to primary or advanced disease between paired samples (oncogenic and variants of unknown significance). (B) TMB for primary versus advanced disease (left) and FGA for primary versus advanced disease (right). (C) Oncoprint for the overall cohort stratified by gene alterations, with top rows indicating primary disease and bottom rows indicating advanced disease. Colors and shapes indicate the type of driver mutation or the type of structural alteration (refer to key in figure). Overall concordance and mutational concordance are reported. Of note, Appendix Figure A2 represents an oncoprint of all patients evaluated. ^aStructural fusion concordance is reported for FGFR2. FGA, fraction of genome altered; TMB, tumor mutational burden.

Median TMB was low and similar between primary and advanced samples (median 3.5 v 4.1 mutations/Mb; P = .62; Fig 1B). Only two patients had a TMB >40 mutations/Mb in the primary tumor, and both retained this feature in advanced disease. No patients in this cohort had microsatellite instability or mismatch repair deficiency. FGA was also similar between primary and advanced diseases (17.7% v 19.2%, respectively; P = .94), with very few outliers.

We noted significant genomic heterogeneity as no single alteration was present in more than a third of patients. The most frequently altered genes across all patients were ARID1A (n = 21 of 65, 32%), CDKN2A (n = 20 of 65, 31%), and TP53 (n = 16 of 65, 25%; Appendix Table A1). Alterations in IDH1 occurred at a rate of 18% (n = 12 of 65), FGFR2 events were seen in 14% (n = 9 of 65), and both occurred exclusively in ICCA. The genes with the highest degree of alteration concordance were IDH1 (overall and mutational = 92%), IDH2 (overall and mutational = 83%), TP53 (overall and mutational = 81%), and BRAF (overall = 78%; mutational = 100%; Fig 1C; Appendix Fig A2). Deep deletion events were heavily represented in patients with CDKN2A alterations (80%), whereas only four patients were found to have nonstructural mutations. The concordance of FGFR2 fusion events was 71%.

Rates of pathway alterations were similar between primary and advanced samples (Fig 2), with RTK-Ras (55%), TP53 (31%), and cell cycle (30%) as the most frequently altered. The WNT (77%), RTK-RAS (71%), TP53 (67%), and NOTCH (60%) pathways demonstrated highest alteration concordance between samples.

FIG 2.

Frequency of pathway alterations and concordance between paired samples for each pathway.

ICCA Versus ECCA

TP53 alterations were more common in ECCA compared with ICCA (6 of 11 [55%] v 10 of 54 [19%], respectively; P = .020). By contrast, IDH1 and FGFR2 were altered, respectively, in 12 of 54 (22%) and 9 of 54 (17%) patients with ICCA and no patients with ECCA (P = .11; P = .33, respectively). Frequencies of other gene alterations were statistically similar between primary ICCA and ECCA. The cumulative oncogenic overall concordance and the mutational concordance between the primary tumor and advanced disease specimens were 47% and 58%, respectively, for ICCA and 60% and 84%, respectively, for ECCA.

Metastasis Versus Progression Versus Recurrence

A breakdown of the overall oncoprint on the basis of the type of advanced disease sample (n = 34 metastasis, n = 17 progression, n = 14 recurrence) is shown in Figure 3. The cumulative oncogenic overall concordance for all patients was 47% for metastases, 50% for progression, and 53% for recurrence samples; the oncogenic mutational concordance was 61% for metastases, 68% for progression, and 60% for recurrence. In patients with metastatic disease, the oncogenic overall concordance and the mutational concordance in patients with ICCA (n = 25) were 42% and 52% compared with 61% and 88% for ECCA (n = 9), respectively. Only two patients with ECCA in this data set had recurrence or progression on systemic therapy, precluding comparison with ICCA in these subgroups.

FIG 3.

Gene alterations and concordance of paired samples separated by advanced disease type (metastatic, progression, recurrence).

Effect of Systemic and Targeted Therapies

Fifty-six patients received systemic therapy between sample collections (n = 46 ICCA, n = 10 ECCA). Of those with ICCA, 22 had metastatic disease, 16 progressed on systemic therapy, and eight experienced recurrence after resection. Of those with ECCA, eight developed metastatic disease, whereas two experienced recurrence after resection.

Tumors from patients who received systemic therapy between sample collections had cumulative oncogenic overall and mutational concordances of 53% and 68%, respectively. Cumulative oncogenic overall and mutational concordances were lower for ICCA (51% and 65%, respectively) than for ECCA (61% and 83%, respectively). Similar to the overall cohort (Fig 1C), the most concordant genes were IDH1 (overall and mutational = 100%), IDH2 (overall and mutational = 83%), TP53 (overall = 85%, mutational = 84%), and BRAF (overall = 78%; mutational = 100%; Appendix Fig A3). Other gene-specific alteration concordance values resembled the overall cohort. In addition, patients who were treated with HAIP therapy between biopsies had cumulative and mutational concordances of 33% and 47%, respectively (Appendix Fig A4).

We identified nine patients who were treated with targeted agents (n = 3 FGFR2 inhibitor, n = 6 IDH inhibitor; Fig 4). For these patients, the alteration targeted in the primary sample persisted in the advanced sample. All patients received targeted therapy before the second sample collection. Interestingly, in the overall cohort, de novo FGFR2 and IDH1 alterations (ie, only detected in the advanced disease specimen) were seen in 1 of 9 and 1 of 12 patients, respectively, who harbored alterations in these genes.

FIG 4.

Concordance of mutations in patients receiving targeted therapy between sample collections. Timelines denote sample collection times and duration of therapy, and individual oncoprints show other oncogenic alterations in these patients. IDH1 inhibitors: ivosidenib and vorasidenib. IDH2 inhibitors: enasidenib.

DISCUSSION

CCA is characterized by a predilection for advanced presentation, a propensity for recurrence after resection, resistance to systemic therapy, and unpredictable biologic behavior that often transcends stage.^1,3,5,10,30 The genetic heterogeneity, particularly in ICCA, recapitulates its phenotypic variation as no single genomic alteration or combination of alterations accounts for more than a third of observed cases, contrary to many other cancers.^15,19,20 However, the degree to which a single patient with CCA demonstrates genetic heterogeneity between multiple tumors or different stages over time is unclear. In a small review of patients with multifocal ICCA, Lee et al provided insight to this question, demonstrating a high degree of genomic concordance between the primary tumor and intrahepatic metastasis or satellite lesions, but heterogeneity between different patients.⁹ The study herein gained further insight into the heterogeneity of ICCA by evaluating genomic alteration concordance between the primary disease and advanced disease.

In the current study, only 49% of oncogenic alterations were concordant between sample pairs, which is lower than the rough 85% concordance described for colorectal cancer primary-advanced pairs.¹⁹ Notably, this is greater than the only 43% concordance including both bystander alterations and oncogenic alterations. The lack of concordance was consistent between metastatic, progressive, and recurrent diseases, and the overall concordance was almost 50% lower for ICCA than for ECCA. TMB and FGA were relatively low and similar between sample pairs, underscoring a lack of significant gained DNA damage repair deficiencies or progressive chromosomal instability.

The most frequently identified altered genes in our cohort were ARID1A, CDKN2A (deep deletions), TP53, and KRAS, none of which occurred in more than 32% of patients. Consistent with the literature, TP53 mutations were more commonly observed in ECCA, whereas IDH1 and FGFR2 were exclusively observed in ICCA.^15,16 In addition, IDH1 mutations (18%) and FGFR2 fusions (14%) occurred in our cohort at previously described rates.^10,14 It should be noted that some genes, including IDH1, TP53, and BRAF, were highly concordant between pairs.¹⁵ Similarly, WNT and RTK-RAS demonstrated pathway concordance over 70%.

Detection of some of these mutations in ICCA has allowed for tailored therapy. In a multicenter phase III randomized controlled trial by Abou-Alfa et al,¹⁰ patients with unresectable/metastatic IDH1-mutated ICCA who progressed on chemotherapy saw modest improvements in overall survival and progression-free survival when treated with ivosidenib (IDH1 inhibitor) versus placebo. Similarly, the FGFR inhibitors infigratinib, pemigatinib, and futibatinib have all shown promise in a subgroup of patients with FGFR2 fusions in phase II trials.^11-14 The use of targeted agents in patients with certain BRAF mutations (V600E) and HER2 mutations is under investigation.^31,32 Furthermore, traction is growing in targeting some of the observed alterations in other cancers, including KRAS pan-inhibition and KRAS subtype inhibition (G12C and G12D).^33-35 Interestingly, although KRAS mutations have been described as truncal mutations in CCA, the concordance for alterations in this gene was 58%.³⁶ To further investigate this, binary alignment map files were individually reviewed for discordant cases. Three of four patients with discordant KRAS alterations had reads below the threshold for calling the mutation, suggesting that KRAS events may not be truncal in all tumors and supporting significant heterogeneity and spatial sampling variance.

While a low degree of concordance was observed in patients receiving first-line chemotherapy between samples, patients with FGFR2 fusions or IDH1 mutations who received targeted therapy between sample collections demonstrated a 100% concordance in these genes. While the focus of this project was on the persistence of driver gene alterations from the primary tumor sample to advanced disease, rather than on mechanisms of resistance, we note that Goyal et al describe secondary FGFR2 kinase domain mutations that confer resistance to particular targeted agents.^13,37 Of particular interest, FGFR2 fusions were detected de novo in one patient with advanced disease and one de novo IDH1 mutation was identified. Notably, all IMPACT panels used were equipped to detect these mutations and tumor purities of the primary samples were not particularly low (0.60 FGFR2, 0.40 IDH1). While these were rare events, the recognition of these alterations presents opportunities for targeted therapy that might have been missed had genetic reassessment been omitted.

There are multiple hypotheses for our findings. First, patients with ICCA might have polyclonal cellular populations characterized by different driver alterations and varying mutational signatures.¹⁶ In this case, biopsy samples submitted from the primary tumor may comprise cells of different clonal origins than those from advanced disease. Second, biopsy samples obtained by methods other than resection might have been less pure, limiting the ability to detect certain alterations, particularly CNAs.³⁸ However, our tumor purity analysis argues against this being the case. Third, ICCA may accumulate additional mutations in advanced disease stages that distinguish it from the primary tumor and potentially contribute to treatment resistance.¹⁶ Ultimately, it is likely that a combination of these factors plays a role in the observed findings. Further investigation of de novo alterations in advanced disease, their prognostic value, and their association with systemic therapy resistance is warranted. In addition, genetic testing on multiple sites of the same ICCA tumor might further elucidate intratumoral heterogeneity. Ultimately, circulating tumor DNA may prove to be more useful in this regard but remains investigational.³⁹ Current limitations of circulating tumor DNA include limited concordance to solid tissue biopsies (60%-90%), inability to detect alterations below a certain threshold (representing a challenge in early-stage cancers), and inability to differentiate between primary and metastatic diseases.^40,41

There are several limitations to our study. This was a retrospective study without an established biopsy collection protocol; as such, patients received biopsies at diverse timepoints in their disease course after exposure to different therapies. Sample sizes were low, especially for the ECCA group, which might have affected ICCA versus ECCA comparisons. In addition, we identified rare driver alterations (ie, IDH2, ARID2, and PBRM1) in which a small number of discordance events could translate to large differences in concordance. However, the use of cumulative concordance analysis overcomes this issue. In addition, samples were only taken from one portion of the tumor, and it is possible that the genetic profile observed reflects only a subset of the clonal groups that comprise the tumor. Finally, not all samples were obtained via surgical biopsy, and the purity of these samples might have affected the ability to call certain alterations, especially CNAs. However, a major strength of this study was the genomic analysis of two different disease stages per patient in a multi-institutional cohort.

In conclusion, CCA, particularly ICCA, is a genetically diverse cancer with relatively low genetic alteration concordance between primary and advanced diseases. This low degree of concordance reflects the genetic heterogeneity of ICCA, offers insight into frequent treatment failures in advanced disease, and argues for reassessment of driver gene alterations as the disease evolves clinically.

APPENDIX

TABLE A1.

Gene-Specific Alterations, Mutations, and CNAs in the Overall Cohort (n = 65 patients)

Gene	Patients With Alterations, No. (%)	Patients With Mutations, No.	Patients With CNAs, No.
ARID1A	21 (32)	21	0
CDKN2A	20 (31)	4	16
TP53	16 (25)	16	0
KRAS	12 (18)	12	0
IDH1	12 (18)	12	0
BAP1	10 (15)	6	4
BRAF	9 (14)	9	0
FGFR2	9 (14)	7a	2
IDH2	6 (9)	6	0
ARID2	6 (9)	6	0
PBRM1	5 (8)	5	0

Abbreviation: CNAs, copy number alterations.

^a FGFR2 mutations include six fusion events, one mutation, and two CNAs.

FIG A1.

Tumor purity in primary disease versus advanced disease. Box notes median and IQR. P value is demonstrated above. ECCA, extrahepatic cholangiocarcinoma; ICCA, intrahepatic cholangiocarcinoma.

FIG A2.

Oncoprint of all patients included in analysis. Each row represents an individual patient, with the top row demonstrating primary disease and the bottom row demonstrating advanced disease.

FIG A3.

Oncoprint of samples from patients who received systemic therapy between sample collections. ^aStructural rearrangement concordance.

FIG A4.

Oncoprint including sample pairs from patients who received hepatic artery infusion pump chemotherapy between sample collections. ^aStructural rearrangement concordance.

AUTHOR CONTRIBUTIONS

Conception and design: Sarah M. McIntyre, William A. Preston, Peter J. Allen, James J. Harding, Ghassan K. Abou-Alfa, Wungki Park, Kevin C. Soares, T. Peter Kingham, Christine Iacobuzio-Donahue, William R. Jarnagin

Financial support: James J. Harding, Andrea Cercek, Christina Iacobuzio-Donahue, William R. Jarnagin

Administrative support: Christine Iacobuzio-Donahue, William R. Jarnagin

Provision of study materials or patients: Jeremy Sharib, Peter J. Allen, Michael A. Morse, Wei Chen, James J. Harding, Ghassan K. Abou-Alfa, Wungki Park, Alice Wei

Collection and assembly of data: Sarah M. McIntyre, William A. Preston, Jeremy Sharib, Ritika Kundra, Carlie Sigel, Michael A. Morse, Wei Chen, Andrea Cercek, James J. Harding, Ghassan K. Abou-Alfa, Kevin C. Soares, Alice Wei, William R. Jarnagin

Data analysis and interpretation: Sarah M. McIntyre, William A. Preston, Henry Walch, Ritika Kundra, Michael E. Lidsky, Peter J. Allen, Michael A. Morse, Andrea Cercek, James J. Harding, Ghassan K. Abou-Alfa, Wungki Park, Vinod P. Balachandran, Jeffrey Drebin, Kevin C. Soares, William R. Jarnagin

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).