Molecular profiling of biliary tract cancers reveals distinct genomic landscapes between circulating and tissue tumor DNA

Abstract

Biliary tract cancers (BTCs) are heterogeneous malignancies with dismal prognosis due to tumor aggressiveness and poor response to limited current therapeutic options. Tumor exome profiling has allowed to successfully establish targeted therapeutic strategies in the clinical management of cholangiocarcinoma (CCA). Still, whether liquid biopsy profiling could inform on BTC biology and patient management is unknown. In order to test this and generate novel insight into BTC biology, we analyzed the molecular landscape of 128 CCA patients, using a 394-gene NGS panel (Foundation Medicine). Among them, 32 patients had matched circulating tumor (ct) DNA and tumor DNA samples, where both samples were profiled. In both tumor and liquid biopsies, we identified an increased frequency of alterations in genes involved in genome integrity or chromatin remodeling, including ARID1A (15%), PBRM1 (9%), and BAP1 (14%), which were validated using an in-house-developed immunohistochemistry panel. ctDNA and tumor DNA showed variable concordance, with a significant correlation in the total number of detected variants, but some heterogeneity in the detection of actionable mutations. FGFR2 mutations were more frequently identified in liquid biopsies, whereas KRAS alterations were mostly found in tumors. All IDH1 mutations detected in tumor DNA were also identified in liquid biopsies. These findings provide novel insights in the concordance between the tumor and liquid biopsies genomic landscape in a large cohort of patients with BTC and highlight the complementarity of both analyses when guiding therapeutic prescription.

To the editor,

Incidence and mortality rates of biliary tract cancers (BTC) are rising [1] and most patients present with advanced disease, where standard therapies bring limited benefit [2]. Genomic profiling has allowed the identification of recurrent molecular alterations, leading to successful precision medicine-based therapies (e.g., FGFR or IDH1 inhibitors). Yet, most patients still do not benefit from targeted therapy, due to the absence of actionable alteration or lack of available molecular profile. Lesion accessibility, limited material and tumor heterogeneity represent hurdles to tumor molecular profiling. By contrast, circulating tumor DNA (ctDNA) analysis is minimally invasive, feasible in all patients and possibly recapitulates the molecular landscape of multiple lesions. However, ensuring that alterations identified in liquid biopsies are representative of the ones present in tumors is necessary.

To explore the concordance between molecular alterations in tumor and ctDNA, we characterized the genomic landscape of 128 BTC and 32 matched ctDNA samples using a targeted 394-gene panel (Fig. 1A, B).

Fig. 1

Cholangiocarcinoma harbor frequent DNA repair and chromatin remodeling alterations. A Flowchart of patients and samples characterization. B Clinico-pathological features of patients with CCA. Prior lines of therapy describe the number of treatment lines received prior to sample collection for molecular characterization; iCCA intrahepatic CCA. eCCA extrahepatic CCA. C Oncoplot of mutations detected in tumor biopsies in 128 patients. Significantly mutated genes are listed vertically in decreasing order and hierarchized by functional category. Colored boxes indicate alteration categories observed in each gene and tumor. D Frequency of targetable mutations for OncoKB level 1–4 alterations with a > 5% prevalence. Level 1 = FDA-approved drugs, Level 3 = Clinical Evidence, Level 4 = Biological Evidence. ARID1A status should be updated to level 1–2 according to latest FDA approval of Tulmimetostat. E Somatic interactions plot for the 20 most frequently altered genes. Colored boxes show mutual exclusivity and co-occurring alterations between two genes (*p < 0.05)

Molecular features

Overall, 1357 genomic alterations were detected in 333 genes of the 128 tumor samples. Most frequently altered genes were TP53, KRAS, KMT2D, ARID1A, IDH1, ATM and BAP1. Among 71 tumors for which microsatellite stability (MS) status was assessed, 69 were MS-Stable and two were MS-Instable (Additional file 1: Fig. S1A). Most frequently altered genes were involved in “Genome integrity” (62%) and “Chromatin remodeling” (54%), notably subunits of the SWI/SNF chromatin remodeling complex (35% of cases, including ARID1A, PBRM1, ARID1B, SMARCA4 and ARID2). Alterations in these pathways were significantly associated with a higher number of variants (Additional file 1: Fig. S1B). Interestingly, we found that, as previously described for FGFR and IDH1 mutations [3], BAP1 alterations were restricted to iCCA (Fig. 1C).

When focusing on clinically actionable alterations, KRAS (G12/V/C, OncoKB level 1), IDH1 (level 1) and ARID1A (level 1–2) [4, 5] were most frequently altered (18%, 15% and 15% of patients, respectively; Fig. 1D). KRAS and TP53 alterations tended to co-occur (p < 0.05), like DNMT3A and MED12. By contrast, TP53 mutations were mutually exclusive with ATM and BAP1 alterations, as KMT2D and KRAS (Fig. 1E).

Genomic landscape of tissue versus liquid biopsies

We next compared molecular alterations in ctDNA versus tumor tissue in a subset of 32 patients with paired samples, including eight with concomitant and 24 with sequential sampling (Additional file 1: Fig. S2A). We identified multiple alterations that were more frequent in liquid biopsies: DNMT3A (44% vs 6%), TP53 (38% vs 25%), ATR (25% vs 19%) and CHEK2 (25% vs 3%). Such discrepancies could result from clonal hematopoiesis of indeterminate potential (CHIP), intra- or inter-tumor heterogeneity, temporal heterogeneity or, low tumor content of the biopsy and should therefore be interpreted with caution. Conversely, KRAS mutations were found in 22% of tumor samples, but only in 9% of liquid biopsies. The somatic interactome also showed different patterns: based on tissue analysis, TP53 alterations co-occurred with POLE, ERBB3, CDKN2A and KRAS mutations; this was not observed in plasma samples, where CHEK2 alterations co-occurred with DNMT3A and ATRX mutations, potentially linked to CHIP (Fig. 2A, B).

Fig. 2

ctDNA and tumor DNA show variable concordance in the detection of molecular alterations. A, B Oncoplot of mutations detected in tumor (A) or liquid (B) biopsies in 32 patients with paired samples available. Left panel: white stars represent mutations that were found both in the tumor and in ctDNA. Right panels show somatic interactions of the top 20 altered genes in each sample type. C Venn Diagram showing the conservation of IDH1, ATM, FGFR2 and KRAS alterations between ctDNA and tissue biopsies; boxes below the Venn diagram depict the proportion of actionable mutations. D Scatter plot showing the correlation between the number of alterations in ctDNA and tissue DNA in all tumor samples (n = 30, after exclusion of the two outlier patients with MSI-H/MMRd tumors; Spearman correlation score r = 0.4148, p-val = 0.0227, two-tailed t-test). E Scatter plot showing the correlation between the number of alterations in ctDNA and tumor samples in patients with concomitant samplings only (n = 8; Spearman correlation score r = 0.9212, p-val = 0.0023, two-tailed t-test)

When focusing on clinically actionable alterations, all IDH1 and ATM mutations were conserved in tumor and liquid biopsies. Interestingly, 8/9 FGFR2 actionable alterations were only detected in liquid biopsies. This may result from the polyclonal secondary mutations following FGFR2-inhibitors therapy [6], since four out of the five patients with actionable alterations received pemigatinib or futibatinib, and progressed at the biopsy timepoint (Additional file 1: Fig. S2B). By contrast, four out of six actionable KRAS mutations were only identified in tissue samples (Fig. 2C).

A higher number of ctDNA alterations was found in heavily pre-treated patients, potentially reflecting increased tumor genetic complexity over time and exposure to treatments (Additional file 1: Fig. S3G). No correlation was identified with any other clinical parameter (histotype, metastatic burden, sex, age, differentiation, grade, stage at diagnosis; Additional file 1: Fig. S3 A–H). Noteworthy, the number of alterations found in ctDNA and tumor DNA was significantly correlated, especially in patients with concomitant tumor and ctDNA sampling (Spearman correlation r = 0.9212, p-val = 0.0023 two-tailed t-test; Fig. 2D, E, Additional file 1: Fig. S4).

Epigenomic alterations

Since deleterious mutations in SWI/SNF or BAP1 detected in > 30% of this 128-patient cohort are potentially actionable [5, 7, 8], it is crucial to reliably identify them. Still, the pathogenicity of SWI/SNF genes alterations remains vastly unknown. We therefore optimized an immunohistochemistry panel to measure ARID1A, PBRM1, SMARCA4 and SMARCB1 protein expression. ARID1A expression was lost in all ARID1A-mutant cases as well as in one ARID1A-wild-type PBRM1-mutant case. PBRM1 expression was decreased in 5/6 PBRM1-mutant samples, and was completely lost in two BAP1-mutant cases. Importantly, alteration in SWI/SNF subunit was not anti-correlated with H3K27me3 or EZH2 expression, highlighting the challenge to detect the SWI/SNF-PCR2 epigenetic antagonism in tumors [9] (Additional file 1: Fig. S5).

Conclusion

In conclusion, molecular profiling of 128 BTC showed frequent alterations in DNA repair and chromatin remodeling factors, notably in SWI/SNF subunits, which increases the proportion of patients with actionable mutations beyond FGFR2 and IDH1. This dataset represents the second largest series of molecular landscape comparison between ctDNA and tumor tissue biopsies in BTC [10, 11]. Limitations of our study include its retrospective and multi-centric nature, the limited sample size, and the heterogeneous patients’ characteristics. Still, we found a significantly positive correlation between the number of alterations detected in matched tumor-liquid biopsy samples, and alterations detected in liquid biopsies were overall concordant with the ones found in tumor tissue. Importantly, discrepancies were also observed notably in actionable alterations, potentially due to spatial or temporal heterogeneity in tumor sampling, or to CHIP-associated false positive [12]. This overall highlights the complementarity of both analyses when guiding therapeutic prescription.