ABSTRACT
-
Background/Aims
Biliary tract cancer (BTC) is a rare malignancy with poor prognosis. We investigated genomic determinants of clinical benefit from gemcitabine, cisplatin, and nab-paclitaxel (GAP) versus gemcitabine and cisplatin (GC) in advanced BTC.
-
Methods
Clinical and genomic data using TruSight Oncology 500 were analyzed from patients treated with GAP (N=198) or GC (N=89) as first-line therapy.
-
Results
With a median follow-up of 33.0 months, GAP modestly improved progression-free survival (PFS) (hazard ratio [HR] 0.764; 95% confidence interval [CI] 0.591–0.989) without significant overall survival (OS) difference compared to GC. Genomic profiling revealed frequent alterations in TP53 (35.2%), KRAS (16.4%), SMAD4 (10.5%), and TNFRSF14 (10.5%), involving RTK/RAS (44.3%), TP53 (41.8%), and PI3K (20.2%) pathways. Single-gene mutations did not predict treatment benefit. However, pathway-level analysis identified PI3K pathway activation as significantly associated with inferior PFS (HR 2.148; 95% CI 1.478–3.124) and OS (HR 2.096; 95% CI 1.413–3.109) in patients receiving GAP, an effect not observed with GC. Importantly, GAP conferred clinical benefit only in patients without PI3K pathway activation, while no survival advantage was seen in those with such alterations (Pinteraction=0.023 for PFS, Pinteraction=0.003 for OS). Similar results were obtained in the independent validation cohort treated with GAP (N=103) or GC (N=64) for BTC.
-
Conclusions
Genomic profiling using next-generation sequencing identified PI3K pathway activation as key molecular determinant that differentiates patient outcomes between GAP and GC treatments in advanced BTC.
-
Keywords: Biliary tract neoplasms; Genomic analysis; PI3K pathway; Systemic treatment
Study Highlights
• Genomic profiling was conducted in patients with advanced BTC treated with GAP or GC.
• Single gene mutations did not predict the benefit of GAP or GC.
• Pathway-level analysis showed that patients without PI3K pathway activation derived significant benefit from GAP.
• Molecular stratification determines distinct benefit of systemic treatment in advanced BTC.
Graphical Abstract
INTRODUCTION
Biliary tract cancer (BTC), including intrahepatic cholangiocarcinoma (IHCC), extrahepatic cholangiocarcinoma (EHCC), or gallbladder cancer (GBC), is often diagnosed at advanced stages, limiting curative options [
1]. Despite numerous efforts evaluating several systemic and targeted therapies, the prognosis of BTC has remained dismal for the past decade [
2]. Recently, TOPAZ (NCT03875235) and KEYNOTE-966 (NCT04003636) studies have shown improvement in overall survival (OS) with durvalumab or pembrolizumab plus gemcitabine and cisplatin (GC) compared to GC only [
3,
4]. The median OS remained at approximately 12 months, highlighting the need for novel therapeutics or biomarker-driven enrichment strategies [
5]. These studies underscore the critical need for biomarker-driven strategies to optimize treatment selection in BTC and improve patient outcomes by identifying those most likely to benefit from targeted therapies.
Nab-paclitaxel is an albumin-bound form of paclitaxel that has been approved in combination with gemcitabine as a first-line treatment for patients with advanced pancreatic adenocarcinoma [
6]. A phase II study with nab-paclitaxel plus GC (NCT02392637) has shown a high response rate and encouraging survival in patients with advanced BTC [
7]. Based on these results, a phase III SWOG 1815 study was conducted to compare GC with or without nab-paclitaxel for patients with advanced BTC (NCT03768414). The median OS was 14.0 months with the addition of nab-paclitaxel and 13.6 months without the addition of nab-paclitaxel (
P=0.41) [
8]. An intriguing trend toward better survival with nab-paclitaxel plus GC has been observed in specific subgroups, including patients with GBC and locally advanced cancer stages; however, there is an unmet need for identifying subsets who could benefit from adding nab-paclitaxel based on genomic analysis.
We hypothesized that molecular alterations determined by genomic analysis may identify specific genomic subgroups of patients with advanced BTC who will derive benefit from adding nab-paclitaxel to GC therapy. To test this hypothesis, we examined 287 patients with BTC profiled by targeted next-generation sequencing (NGS) and treated with first-line systemic treatment with GC with or without nab-paclitaxel. Furthermore, we validated our hypothesis with independent external validation cohort. Based on this approach, the genomic features predictive of nab-paclitaxel plus GC can be identified and incorporated to guide patient selection regarding the addition of nab-paclitaxel to first-line chemotherapy for advanced BTC.
MATERIALS AND METHODS
Patients
We identified 287 patients with available TruSight Oncology 500 data for advanced BTC treated with gemcitabine, cisplatin, and nab-paclitaxel (GAP) or GC between August 2018 and April 2023. Eligible patients were at least 18 years of age; had histologically or cytologically confirmed IHCC, EHCC, or GBC; and had metastatic or locally advanced unresectable disease documented on diagnostic imaging studies. The exclusion criteria included curative surgery, including complete resection of the tumor with no evidence of disease recurrence and cases where the cancer was not primary BTC. An independent external validation cohort of patients with advanced BTC was retrospectively collected from CHA Bundang Medical Center between January 2018 and July 2024. All procedures were conducted in accordance with the Declaration of Helsinki and the International Conference on Harmonization Guidelines for Good Clinical Practice. This retrospective study was approved by the Institutional Review Board of the Yonsei Cancer Center, Seoul, Republic of Korea (IRB No. 4-2023-0485) and CHA University, Seongnam, Republic of Korea (IRB No. 2021-01-010, 2021-03-045, and 2022-11-002). All patients provided written informed consent before genomic analysis and systemic chemotherapy administration. Abbreviations:
Treatment
GAP included gemcitabine (800 mg/m
2), cisplatin (25 mg/m
2), and nab-paclitaxel (100 mg/m
2) administered on days 1 and 8 of a 21-day cycle. GC included standard doses of gemcitabine (1,000 mg/m
2) and cisplatin (25 mg/m
2) administered on days 1 and 8 of a 21-day cycle. Patients were treated until disease progression, defined according to the Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1 [
9], or unacceptable toxicity. Dose modifications, interruptions, and treatment with growth factors were permitted for adverse effect management in accordance with the American Society of Clinical Oncology guidelines.
Tumor tissue blocks obtained prior to initiation of first-line therapy were reviewed by a pathologist, and only specimens with ≥40% tumor cellularity were selected. From each formalin-fixed, paraffin-embedded (FFPE) sample, we co-extracted genomic DNA and total RNA (40 ng each) using the Qiagen AllPrep DNA/RNA FFPE Kit following the manufacturer’s instructions. After hybridization capturebased target enrichment, paired-end sequencing (2 × 150 bp) was performed using an Illumina NextSeq according to the manufacturer’s instructions.
Bioinformatic processing and variant calling
The raw DNA sequencing file was aligned to the human genome reference build, Genome Research Consortium human build 38 (GRCh38), using BWA-MEM algorithm [
10]. The tag containing the unique molecular identifiers’ information was marked using an in-house script. The Mark duplicate process was performed using UmiAwareMarkDuplicatesWithMateCigar (Picard). Somatic variants were called with Mutect2: (i) total depth ≥100x, (ii) variant allelic frequency ≥10%, (iii) combined annotation dependent depletion phred score ≥25, and (iv) minor allele frequency <0.1% in the gnomAD database for the global and East Asian populations [
11]. The pathogenicity of the filtered variants using haplotype caller was determined using the recommendations of the American College of Medical Genetics and Genomics and Association for Molecular Pathology [
12]. Copy-number variations (CNV) were analyzed using CNVkit, focusing on the 523 genes targeted by the TruSight Oncology 500 panel. The filtering criteria were set at a log2 value of <–1.2 for deletions and >2.0 for amplifications. We selected only CNVs that had been previously reported in patients with BTC. Gene fusion candidates were analyzed using Star-fusion [
13]. A fusion expression filter was applied to the initial calls (fusion fragments per million total RNA-seq fragments <0.5). An oncoplot was constructed using MAFtools [
14].
Genomic DNA was extracted using the RecoverAll Multi-Sample RNA/DNA Isolation Workflow (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s protocol. Targeted sequencing was performed with the Oncomine Comprehensive Assay (OCA, Thermo Fisher Scientific, Waltham, MA, USA). Specifically, 10 samples were analyzed using OCA v1, which targets 143 cancer-associated genes; 20 samples were analyzed using OCA v3, which covers 161 genes; and 137 samples were analyzed using the OCA plus, which encompasses a broader range of 501 genes, including 144 genes overlapping with OCA v3. NGS libraries were prepared and sequenced on the Ion Torrent ™ S5™XL platform. Raw sequencing reads were aligned to human reference genome 19 (hg19) using Torrent Suite software (v5.18); variant calling was performed using Ion Reporter™ software (v5.18).
Follow-up and evaluation
Laboratory tests and physical examinations were performed regularly to assess treatment response. Treatment response was evaluated according to the RECIST 1.1 criteria. Tumor imaging was performed at weeks 6 and 12, every 9 weeks until week 48, and every 12 weeks thereafter. Based on the RECIST 1.1 criteria, treatment response was classified as complete response, partial response, stable disease, or progressive disease [
9]. Progression-free survival (PFS) was measured as the time from treatment initiation to either disease progression or death. OS was defined as the time from treatment initiation to death from any cause. The censoring date was July 5, 2024.
Categorical and continuous variables are summarized descriptively using frequencies, percentages and medians. Differences among continuous and categorical variables were examined for significance using the Student’s
t-test and the chi-squared test. Event-time distributions were estimated using the Kaplan–Meier method, and the adjusted and unadjusted hazard ratios (HRs) for PFS and OS with corresponding 95% confidence intervals (CIs) were estimated using Cox hazard models. To reduce potential confounding bias due to non-randomized research settings, we performed an inverse probability of treatment weighting (IPTW) analysis with the estimated propensity scores. Propensity scores were derived from a multivariable logistic regression model including age, sex, stage, tumor location, and histologic differentiation. Stabilized weights were calculated as the inverse probability of receiving the actual treatment (GAP or GC) and cut at the 1st and 99th percentiles. Covariate balance before and after weighting was evaluated using standardized mean differences, with an absolute standardized mean difference <0.1 considered indicative of adequate balance. All
P-values were two-sided, and CIs were at the 95% level, with significance pre-defined to be at the two-sided 0.05 level. All statistical analyses were performed using R version 4.0.4 (
http://www.R-project.org) and GraphPad Prism (version 6.0; GraphPad Software, San Diego, CA, USA).
RESULTS
Clinicopathologic characteristics of the cohort
Table 1 presents a detailed description of the demographic and clinicopathological characteristics of the study cohort. The median age of the patients was 65 years, and the male patients (56.1%) were slightly more than the female patients. According to anatomical location, 45.6% had IHCC, 28.9% had EHCC, and 25.4% had GBC. Furthermore, 198 patients (69.0%) received GAP, and 89 (31.0%) received GC as first-line systemic treatment. The median age of those who received GAP was younger than that of those who received GC (
P<0.001). Sex, stage, tumor location, and histological differentiation were comparable between patients who received GC and those who received GAP.
Clinical outcomes
The median follow-up time for surviving patients was 33.0 months (95% CI 30.1–35.9 months). Tumor response was evaluated in 97.9% of the patients (281/287). The objective response rates were significantly higher in patients treated with GAP (76/195, 39.0%) compared to those treated with GC (22/84, 26.2%) (
Fig. 1A). The disease control rate was numerically higher in patients treated with GAP (147/195, 75.4%) compared to those treated with GC (58/84, 69.0%), but this difference was not statistically significant (
P=0.271) (
Fig. 1B). PFS was slightly longer in patients treated with GAP compared to those treated with GC (
Fig. 1C; median PFS 189 vs. 151 days; HR 0.764; 95% CI 0.591–0.989;
P=0.041). Regarding OS, a trend toward improved outcome was observed in patients treated with GAP compared to those treated with GC (
Fig. 1D; median OS 428 vs. 314 days; HR 0.778; 95% CI 0.592–1.023;
P=0.073), although it was not statistically significant similarly to SWOG S1815 study [
8]. No interaction was observed between tumor location and survival outcomes according to treatment in terms of PFS (P
interaction=0.637) or OS (P
interaction=0.733).
In total, 287 tumors were sequenced to a mean depth of 744.4-fold with targeted deep sequencing for 523 genes (
Fig. 2A). After stringent filtering, we identified 385 altered genes, including 343 single-nucleotide variations, 32 CNVs, and 38 fusions. The samples contained a mean of 4.36 single nucleotide variations (SNVs), 0.55 CNVs, and 0.17 fusions (2.21 SNVs/Mb, 0.28 CNVs/Mb and 0.09 fusions/Mb, total, 2.58 mutations/Mb). Figure 2B shows a summary of the frequently altered genes (N=22; observed in over 3% of patients) encompassing mutations, copy number alterations, and fusions. Of the 39 significantly mutated genes (
Supplementary Table 1),
TP53 was the most commonly mutated gene (35.2%), followed by
KRAS (16.4%),
ARID1A (9.1%),
SMAD4 (8.4%),
ARID2 (6.3%), and
CDKN2A (5.2%). Primary site-specific biases have also been observed in several genes. Mutations in
IDH1 (4.2%) and
IDH2 (1.7%) occurred mutually exclusively, and almost all
IDH1/
IDH2 mutations were observed in IHCC (16/17, 94.1%), except for one case of GBC (1/17, 5.9%). In contrast,
TP53 mutations were negatively enriched in IHCC (24.4%, 32/131) compared to GBC or EHCC (44.2%, 69/156,
P<0.001).
RBM10 mutations were most prevalent in EHCC (8.4%), followed by GBC (5.5%) and IHCC (0.8%) (
P=0.020). GBC was significantly enriched with
STK 11 (9.6% vs. 3.3%,
P=0.030) and
KMT2C mutation (8.2% vs. 2.8%,
P=0.046).
Copy number alterations were detected in 92 patients. Of these, the tumor suppressor gene
TNFRSF14 was the most commonly deleted gene, and it occurred in 30 patients (10.5%). IHCC was enriched with
TNFRSF14 deletion compared to GBC or EHCC (15.3% vs. 6.4%,
P=0.015). The tumor suppressor genes
CDKN2A/
CDKN2B were deleted simultaneously in 12 patients (4.2%), whereas
CCND1 amplification was detected in six patients (2.1%). Amplification of
ERBB2 was enriched in non-IHCC (5.8% vs. 0.8%,
P=0.021). Amplification of
MDM2, the target of several agents currently under development [
15], was observed in 10 patients (3.5%). Amplification of the fibroblast growth factor receptor (FGFR) pathway ligands and/or receptors occurred in six patients (2.1%).
RNA-seq analysis revealed fusion events in 24 patients. Of these, FGFR2 fusion transcripts were most frequently captured (eight events), followed by FGFR3 and MET fusion transcripts (two events).
Integrative pathway analysis for mutation, copy number alteration, and fusion
The analysis of oncogenic signaling pathways in the Cancer Genome Atlas revealed alterations in 10 molecular pathways (
Fig. 3) [
16]. Of these pathways, receptor tyrosine kinase/RAS was the most altered pathway (44.3%), followed by TP53 (41.8%) and phosphoinositide 3-kinase (PI3K) (20.2%) pathways. A distinct pattern of pathway activation according to the anatomical location of BTC was revealed. Alterations in the Nrf2 pathway were positively enriched in GBC (11.0% vs. 2.8%,
P=0.005), whereas alterations in TP53 (31.3% vs. 50.6%,
P<0.001) and WNT (5.3% vs. 15.4%,
P=0.006) pathways were negatively enriched in IHCC.
We used Cox regression to assess whether the relative efficacy of GAP versus GC varied based on given genomic features. To this end, we assessed whether mutations in individual genes were associated with outcomes of GC or GAP therapy. In this analysis, no single genetic alteration observed in over 5% of patients was associated with treatment outcomes in patients treated with either GAP or GC in terms of PFS and OS (
Supplementary Fig. 1). In oncogenic signaling pathway analysis, activation in the PI3K pathway was significantly associated with worse PFS (HR 2.148; 95% CI 1.478–3.124;
P<0.001) and OS (HR 2.096; 95% CI 1.413–3.109;
P<0.001) in patients treated with GAP (
Fig. 4A). However, activation in the PI3K pathway was not significantly associated with PFS (HR 1.067; 95% CI 0.654–1.741;
P=0.795) and OS (HR 0.802; 95% CI 0.472–1.364;
P=0.461) in patients treated with GC (
Fig. 4B). In addition, GAP was substantially associated with favorable PFS (
Fig. 4C; HR 0.692; 95% CI 0.516–0.930;
P=0.015) and OS (
Fig. 4D; HR 0.644; 95% CI 0.471–0.881;
P=0.006) compared to GC in patients without PI3K pathway activation (n=229), whereas PFS (
Fig. 4E; HR 1.524; 95% CI 0.885–2.625;
P=0.129) and OS (
Fig. 4F; HR 1.675; 95% CI 0.915–3.067;
P=0.095) did not differ significantly in patients with PI3K pathway activation (n=58) according to the treatment. Alterations in the MYC pathway were associated with worse outcomes in patients treated with GC, which was not observed in those treated with GAP. However, the extremely low frequency of patients with alteration in the MYC pathway in their tumors (1.0%) significantly limited the clinical relevance of the predictive value of alteration in the MYC pathway.
To accurately define PI3K pathway activation, we analyzed the effects of gene-level activating and inactivating alterations on survival outcomes (
Supplementary Table 2). Gain-of-function alterations included
AKT1,
MTOR,
PIK3CA,
PIK3CB,
PIK3R2,
RICTOR,
RPS6KB1, and
RPTOR. Loss-of-function alterations comprised
INPP4B,
PIK3R1,
PTEN,
STK11,
TSC1, and
TSC2. Mode of PI3K pathway activation (gain-of-function in oncogenes versus loss-of-function in tumor suppressor genes) was not associated with survival outcomes in patients treated with GAP (
Supplementary Table 3) or GC (
Supplementary Table 4).
When multivariate analysis encompassing age, sex, stage, tumor location, histologic differentiation, and activation in the PI3K pathway was performed in patients treated with GAP, we observed that activation in the PI3K pathway was independently associated with worse PFS (HR 2.092; 95% CI 1.397–3.132;
P<0.001) and OS (HR 2.014; 95% CI 1.318–3.078;
P<0.001) in patients treated with GAP (
Supplementary Table 5), which was not observed in those treated with GC (
Supplementary Table 6). After adjustment for multiple comparisons using the Benjamini–Hochberg correction, PI3K pathway activation was consistently associated with inferior PFS and OS in patients treated with GAP (
Supplementary Table 7), which was not observed in those treated with GC (
Supplementary Table 8). Notably, a significant interaction was observed between the presence of activation in PI3K pathway and survival outcomes according to the treatment (P
interaction=0.023 for PFS and P
interaction =0.003 for OS;
Fig. 5). After adjustment using IPTW (
Supplementary Fig. 2), the differential association between PI3K pathway activation and treatment regimen remained consistent (
Supplementary Table 9). Collectively, activation in PI3K pathway could potentially serve as an independent genomic biomarker in predicting benefits from GAP compared to GC.
To validate whether PI3K pathway activation is associated with differential benefit of GAP and GC, we analyzed genomic data and clinical outcomes from 167 patients (
Supplementary Table 10). All patients received first-line systemic therapy with GAP (N=103) or GC (N=64) with available targeted NGS data. The baseline characteristics were comparable between patients treated with GAP or GC. PI3K pathway activation was detected in 31.1% (52/167) of patients (26.2% in GAP vs. 39.1% in GC). In the validation cohort, patients without PI3K pathway activation showed significantly longer PFS (
Fig. 6A; HR 0.481; 95% CI 0.295–0.785;
P=0.003) and OS (
Fig. 6B; HR 0.403; 95% CI 0.229–0.771;
P=0.002) with GAP compared to GC. Conversely, patients with PI3K pathway activation had significantly worse PFS (
Fig. 6C; HR 1.905; 95% CI 1.033–3.513;
P=0.039) and OS (
Fig. 6D; HR 2.271; 95% CI 1.089–4.735;
P=0.029) with GAP compared to GC. Collectively, these findings support the robustness of the PI3K pathway activation in predicting differential benefits from GAP versus GC.
DISCUSSION
We conducted an integrative analysis of molecular profiling from patients with advanced BTC treated with GAP or GC and identified genomic features that distinguished the outcomes between the two therapies. We uncovered genomic determinants that could predict which patients with advanced BTC would benefit from the addition of nab-paclitaxel to standard GC therapy. Collectively, our study highlights the clinical relevance of genomic analysis in guiding treatment decisions in patients with BTC, taking the presence of PI3K pathway activation as an example. To the best of our knowledge, this is the first molecular biomarker study performed in patients with BTC who received two different first-line treatments with GAP or GC.
BTC is rare, accounting for <5% of all gastrointestinal neoplasms [
17]. BTC is difficult to treat, with a median OS of approximately 1 year with standard-of-care GC-based regimens [
18]. We repeatedly observed a poor prognosis of advanced BTC in our study, with a median OS of 10.5 months for patients treated with GC. Advanced BTC is an area of significant unmet need because of its aggressive nature and limited treatment options. Recently, systemic therapy for patients with metastatic or unresectable BTC has improved with the recent success of durvalumab or pembrolizumab combined with GC [
3,
4]. In addition, specific molecularly targeted agents in BTC have also recently received regulatory approval based on genomic profiling [
19]. In the prospective studies with first-line chemotherapy with or without immune checkpoint inhibitors (KEYNOTE-966 and TOPAZ study) [
3,
4], the median OS did not exceed 13 months for the experimental arms with programmed cell death-(ligand) 1 inhibitors, suggesting the presence of a highly unmet need for dismal prognosis in patients with BTC. Beyond immunotherapy-based regimens, several trials have explored triplet or combination chemotherapy backbones in BTC. KHBO1401-MITSUBA (NCT02182778) demonstrated the survival benefit of adding S-1 to GC [
20]. SWOG S1815, a phase III randomized trial of GAP versus GC in patients with newly diagnosed, advanced BTC, is another example of these efforts. However, the SWOG1815 trial failed to meet the primary endpoint of significant improvement in OS. Importantly, significant improvements in response and survival have been documented in a subset of patients, revealing that triplet regimens encompassing nabpaclitaxel can be useful in specific subgroups of patients, warranting biomarker identification to select patients who can benefit from triplet regimens. In our study, we explored the association between genomic profiles and treatment outcomes to identify genomic-defined subgroup of patients who would benefit more from adding nab-paclitaxel to advanced BTC treatment. Specifically, we noticed that activation in the PI3K pathway was associated with a significantly diminished benefit from GAP compared to GC. In addition, we suggest that investigating PI3K pathway activation can be useful in designing biomarker-driven enrichment trials, as improvement in PFS and OS was evident in patients with BTC without PI3K activation.
PI3K/AKT/mTOR pathway is a core regulator of cell metabolism, growth, and survival and is critically involved in initiation and progression of BTC [
21]. In this study, 20.2% of patients harbored genetic activation in the PI3K pathway, and no differences were observed in the tumor location (
P=0.189), revealing that inhibition or modulation of PI3K/AKT/mTOR can be a compelling treatment strategy in BTC in general. Notably, activation in the PI3K pathway was significantly associated with poor prognosis and diminished benefit from GAP compared to GC in this study. Since the proportion of patients with PI3K pathway activation exceeds 30% (12/36) in samples from the Cancer Genome Atlas [
22], the failure of SWOG1815 study was partially attributed to a significant proportion of patients with activation in PI3K pathway. In other words, improved PFS due to GAP compared to GC in the current study may be associated with a relatively lower frequency of samples with PI3K pathway activation (20.2%). Mechanistically, paclitaxel indirectly reduces PI3K/AKT/mTOR signaling in some cancer models, resulting in reduced proliferation and enhanced apoptosis [
23,
24]. In various preclinical studies, activation in the PI3K pathway is frequently associated with multidrug resistance, including microtubule modulators such as paclitaxel [
25]. In line with this preclinical observation, paclitaxel resistance in certain types of cancer can be overcome by PI3K/mTOR pathway inhibition [
26]. Building on these data, biomarker-driven clinical trials in BTC can be conceptually designed. For PI3K pathway-activated tumors, clinical benefit of single-agent or combination strategies with alpelisib (PI3Kα inhibitor), capivasertib (AKT inhibitor), or inavolisib (next-generation PI3Kα inhibitor) could be explored. Conversely, in PI3K-wild-type disease—where GAP showed greater benefit—triplet backbones could be refined and tested against current standards (GC combined with programmed cell death-(ligand) 1 inhibitors), while prospectively validating the absence of PI3K activation as an enrichment biomarker to derive much benefit from contemporarily preferred options. Of note, adaptive or umbrella designs with preplanned interaction testing and early pharmacodynamic readouts (p-AKT/p-mTOR/p-S6) would accelerate signal detection with on-treatment biomarker development [
27-
29].
In our study, targetable genetic alterations were identified in around 20% of patients, including alterations in
IDH1/2 mutation (N=17),
ERBB2 mutation or amplification (N=16),
MDM2 amplification (N=10),
FGFR2 rearrangement (N=8),
KRAS G12C mutation (N=4),
FGFR3 rearrangement (N=2),
RNF43 mutation (N=2) and
BRAF mutation (N=1). However, the presence of these alterations could not predict benefit from GAP or GC. Furthermore, no single genetic alteration observed in over 5% of patients was found to significantly predict response to GAP or GC, prompting us to conduct integrative pathway analysis based on cancer drivers and therapeutic targets [
16]. Realizing genomic biomarker-based precision medicine can be challenging in BTC as we witnessed from the premature closure of the PROOF-301 study with infigratinib (NCT03773302) or FOENIX-CCA3 study with futibatinib (NCT04093362); however, shortening of turnaround time, standardization of platform, and introduction of less-invasive biomarker tests would provoke a new wave toward precision medicine for BTC, which could bring major benefits to patients’ treatment outcomes and quality of life.
Our study has several important limitations that should be acknowledged. First, the non‑randomized, retrospective nature of our analysis introduces potential selection biases. Treatment assignment to GAP versus GC was based on physician discretion and institutional practice patterns, rather than standardized criteria, which may confound the observed differences in outcomes. Although we additionally performed IPTW adjustment analysis, residual confounding can exist. Second, as the phase III SWOG S1815 trial failed to demonstrate an OS benefit for adding nab-paclitaxel to GC, GAP regimen has not been adopted as a first-line standard currently. Given current standards combining GC with an immune checkpoint inhibitor, head-to-head comparisons between GAP and GC alone warrant cautious, historically contextualized interpretation. Third, as the combination of gemcitabine plus cisplatin with an immune checkpoint inhibitor constitutes the current standard treatment regimen, further research is needed to elucidate the association between genomic landscape and the benefit of current standard treatment. Finally, our analysis lacks functional validation of PI3K/AKT/mTOR signaling. Because panel-based genomic profiling provides only an indirect proxy of pathway activity, future studies should prospectively incorporate pharmacodynamic endpoints—such as p-AKT or p-S6 quantification—to confirm pathway activation and its therapeutic implications.
In summary, we explored and validated that PI3K pathway activation is a robust genomic determinant differentiating outcomes between GAP and GC in advanced BTC. To our knowledge, this is the first molecular biomarker study to use comprehensive genomic profiling to identify PI3K pathway activation as an independent determinant for differential treatment outcomes based on large patient cohorts with advanced BTC. Our findings support the clinical implementation of biomarker-guided treatment decision and establish a foundation for future combination strategies integrating molecular stratification in advanced BTC.
FOOTNOTES
-
Authors’ contributions
Conceptualization: D.K., N.S.S., S.W., H.J.Chon, C.G.K., Y.N.P., H.J.Choi; Acquisition of data: All authors; Statistical analysis: D.K., N.S.S. S.W., C.G.K.; Drafting of the manuscript: D.K., N.S.S. S.W., C.G.K.; Critical revision and final approval of the manuscript: All authors; Study supervision: H.J.Chon, C.G.K., Y.N.P., H.J.Choi.
-
Acknowledgements
This research was supported by the National Research Foundation funded by Ministry of Science and ICT of the Republic of Korea (RS-2023-00261820 to S.K., RS-202400348654 to C.G.K., and NRF-2023R1A2C2004339 to H.J.Chon), Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI) grant funded by Ministry of Health & Welfare of the Republic of Korea (RS-2024-00407544 to Y.N.P.), Severance Hospital Research fund for clinical excellence by Yonsei University College of Medicine (C-2023-0040 to C.G.K.), and faculty research grant by Yonsei University College of Medicine (6-2022-0106 to S.N.S.).
-
Conflicts of Interest
Choong-kun Lee reported consulting or advisory role from Eisai, Servier, Roche, and Daiichi Sankyo, honoraria from AstraZeneca, Servier, Dong-A ST, Boryung Pharmaceuticals, and Astellas, and research funding from Ono Pharmaceuticals, Boryung Pharmaceuticals, GC Biopharma Corp, and Lunit Inc. Hong Jae Chon reported consulting or advisory role from Eisai, Roche, Bayer, ONO, MSD, BMS, Celgene, Sanofi, Servier, AstraZeneca, SillaJen, Menarini, and GreenCross Cell, and research funding from Roche, Dong-A ST, and Boryung Pharmaceuticals. Chang Gon Kim reported consulting or advisory role from Amgen, Astellas, BeOne Medicine, Eisai, Johnson & Johnson/Janssen, Novartis, Ono Pharmaceutical, and Pfizer, speakers’ bureau from Astellas, AstraZeneca, Boehringer Ingelheim, Dong-A ST, Hanmi, Eisai, Merck, MSD Oncology, Novartis, and Takeda, and research funding from Exelixis, Handok and Ipsen. Hye Jin Choi reported consulting or advisory role from BeOne Medicine. The remaining authors declare no potential conflicts of interest.
SUPPLEMENTARY MATERIAL
Supplementary material is available at Clinical and Molecular Hepatology website (
http://www.e-cmh.org).
Supplementary Figure 1.
Outcome of treatment according to the genetic alteration. (A) Statistics for survival among patients treated with GAP with genetic alteration. (B) Statistics for survival among patients treated with GC with genetic alteration. CI, confidence interval; GAP, gemcitabine, cisplatin, and nab-paclitaxel; GC, gemcitabine and cisplatin; HR, hazard ratio; OS, overall survival; PFS, progression-free survival.
Supplementary Table 3.
Survival analysis comparing oncogenic gain-of-function versus tumor-suppressor loss-of-function PI3K alterations in patients treated with gemcitabine, cisplatin and nab-paclitaxel
Supplementary Table 4.
Survival analysis comparing oncogenic gain-of-function versus tumor-suppressor loss-of-function PI3K alterations in patients treated with gemcitabine and cisplatin
Supplementary Table 7.
Multiple comparisons of pathway alterations in patients treated with gemcitabine, cisplatin, and nab-paclitaxel using the Benjamini–Hochberg correction
Supplementary Table 8.
Multiple comparisons of pathway alterations in patients treated with gemcitabine and cisplatin using the Benjamini–Hochberg correction
Supplementary Table 9.
Inverse probability of treatment weighting adjusted Cox proportional hazards models evaluating the interaction between PI3K pathway and treatment on survival
Figure 1.Outcomes according to the treatment. (A) Objective response rate. (B) Disease control rate. (C) Kaplan–Meier plot for progression-free survival. (D) Kaplan–Meier plot for overall survival. GAP, gemcitabine, cisplatin, and nab-paclitaxel; GC, gemcitabine and cisplatin.
Figure 2.Landscape of genomic alteration in 287 patients with biliary tract cancer. (A) An Oncoplot depicting the genomic alteration affecting over 3% of patients. (B) Mode of somatic variation for frequently altered genes affecting over 3% of patients.
Figure 3.Landscape of oncogenic signaling pathway alteration in 287 patients with biliary tract cancer. (A) Oncoplot depicting the oncogenic pathway alteration. (B) Curated pathways for genetic alteration. GAP, gemcitabine, cisplatin, and nab-paclitaxel; GC, gemcitabine and cisplatin.
Figure 4.Outcome of treatment according to the oncogenic signaling pathway activation. (A) Statistics for survival among patients treated with GAP with oncogenic signaling pathway activation. (B) Statistics for survival among patients treated with GC with oncogenic signaling pathway activation. (C) Kaplan–Meier plot for progression-free survival (PFS) among patients without PI3K pathway activation. (D) Kaplan–Meier plot for overall survival (OS) among patients without PI3K pathway activation. (E) Kaplan–Meier plot for PFS among patients with PI3K pathway activation. (F) Kaplan–Meier plot for OS among patients with PI3K pathway activation. CI, confidence interval; GAP, gemcitabine, cisplatin, and nab-paclitaxel; GC, gemcitabine and cisplatin; HR, hazard ratio; PI3K, phosphoinositide 3-kinase.
Figure 5.Subgroup analysis with interaction terms. (A) Subgroup analysis of progression-free survival according to the variables. (B) Subgroup analysis of overall survival according to the variables. CI, confidence interval; GAP, gemcitabine, cisplatin, and nab-paclitaxel; GC, gemcitabine and cisplatin; PI3K, phosphoinositide 3-kinase.
Figure 6.External validation of the outcome of treatment according to the oncogenic signaling pathway activation. (A) Kaplan–Meier plot for progression-free survival among patients without PI3K pathway activation. (B) Kaplan–Meier plot for overall survival among patients without PI3K pathway activation. (C) Kaplan–Meier plot for progression-free survival among patients with PI3K pathway activation. (D) Kaplan–Meier plot for overall survival among patients with PI3K pathway activation. GAP, gemcitabine, cisplatin, and nab-paclitaxel; GC, gemcitabine and cisplatin; PI3K, phosphoinositide 3-kinase.
Table 1.Baseline characteristics of the patients
Table 1.
|
Total (N=287) |
GAP (N=198) |
GC (N=89) |
P-value |
|
Age |
65 (25–85) |
64 (35–84) |
68 (25–85) |
<0.001 |
|
Sex |
|
|
|
0.620 |
|
Male |
161 (56.1) |
113 (57.1) |
48 (53.9) |
|
|
Female |
126 (43.9) |
85 (42.9) |
41 (46.1) |
|
|
Stage |
|
|
|
0.709 |
|
Locally advanced |
8 (2.8) |
6 (3.0) |
2 (2.2) |
|
|
Metastatic |
279 (97.2) |
192 (97.0) |
87 (97.8) |
|
|
Tumor location |
|
|
|
0.242 |
|
Intrahepatic |
131 (45.6) |
88 (44.4) |
43 (48.3) |
|
|
Extrahepatic |
83 (28.9) |
54 (27.3) |
29 (32.6) |
|
|
Gallbladder |
73 (25.4) |
56 (28.3) |
17 (19.1) |
|
|
Differentiation |
|
|
|
0.910 |
|
Well differentiated |
15 (5.2) |
11 (5.6) |
4 (4.5) |
|
|
Moderately differentiated |
147 (51.2) |
100 (50.5) |
47 (52.8) |
|
|
Poorly differentiated |
98 (34.1) |
67 (33.8) |
31 (34.8) |
|
|
Not assessed |
27 (9.4) |
20 (10.1) |
7 (7.9) |
|
Abbreviations
extrahepatic cholangiocarcinoma
gemcitabine and cisplatin
intrahepatic cholangiocarcinoma
inverse probability of treatment weighting
next-generation sequencing
progression-free survival
REFERENCES
- 1. Valle JW, Kelley RK, Nervi B, Oh DY, Zhu AX. Biliary tract cancer. Lancet 2021;397:428-444.
- 2. Scott AJ, Sharman R, Shroff RT. Precision medicine in biliary tract cancer. J Clin Oncol 2022;40:2716-2734.
- 3. Oh DY, Ruth He A, Qin S, Chen LT, Okusaka T, Vogel A, et al. Durvalumab plus gemcitabine and cisplatin in advanced biliary tract cancer. NEJM Evid 2022;1:EVIDoa2200015.
- 4. Kelley RK, Ueno M, Yoo C, Finn RS, Furuse J, Ren Z, et al. Pembrolizumab in combination with gemcitabine and cisplatin compared with gemcitabine and cisplatin alone for patients with advanced biliary tract cancer (KEYNOTE-966): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet 2023;401:1853-1865.
- 5. Demir T, Moloney C, Mahalingam D. Emerging targeted therapies and strategies to overcome resistance in biliary tract cancers. Crit Rev Oncol Hematol 2024;199:104388.
- 6. Park W, Chawla A, O’Reilly EM. Pancreatic cancer: a review. JAMA 2021;326:851-862.
- 7. Shroff RT, Javle MM, Xiao L, Kaseb AO, Varadhachary GR, Wolff RA, et al. Gemcitabine, cisplatin, and nab-paclitaxel for the treatment of advanced biliary tract cancers: a phase 2 clinical trial. JAMA Oncol 2019;5:824-830.
- 8. Shroff RT, King G, Colby S, Scott AJ, Borad MJ, Goff L, et al. SWOG S1815: a phase III randomized trial of gemcitabine, cisplatin, and nab-paclitaxel versus gemcitabine and cisplatin in newly diagnosed, advanced biliary tract cancers. J Clin Oncol 2025;43:536-544.
- 9. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45:228-247.
- 10. Schneider VA, Graves-Lindsay T, Howe K, Bouk N, Chen HC, Kitts PA, et al. Evaluation of GRCh38 and de novo haploid genome assemblies demonstrates the enduring quality of the reference assembly. Genome Res 2017;27:849-864.
- 11. Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 2020;581:434-443.
- 12. Richards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med 2015;17:405-424.
- 13. Haas BJ, Dobin A, Li B, Stransky N, Pochet N, Regev A, et al. Accuracy assessment of fusion transcript detection via read-mapping and de novo fusion transcript assembly-based methods. Genome Biol 2019;20:213.
- 14. Mayakonda A, Lin DC, Assenov Y, Plass C, Koef fler HP. Maftools: efficient and comprehensive analysis of somatic variants in cancer. Genome Res 2018;28:1747-1756.
- 15. LoRusso P, Yamamoto N, Patel MR, Laurie SA, Bauer TM, Geng J, et al. The MDM2-p53 antagonist Brigimadlin (BI 907828) in patients with advanced or metastatic solid tumors: results of a phase Ia, first-in-human, dose-escalation study. Cancer Discov 2023;13:1802-1813.
- 16. Sanchez-Vega F, Mina M, Armenia J, Chatila WK, Luna A, La KC, et al. Oncogenic signaling pathways in the cancer genome atlas. Cell 2018;173:321-337.e10.
- 17. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49.
- 18. Valle J, Wasan H, Palmer DH, Cunningham D, Anthoney A, Maraveyas A, et al. Cisplatin plus gemcitabine versus gemcitabine for biliary tract cancer. N Engl J Med 2010;362:1273-1281.
- 19. Harding JJ, Khalil DN, Fabris L, Abou-Alfa GK. Rational development of combination therapies for biliary tract cancers. J Hepatol 2023;78:217-228.
- 20. Ioka T, Kanai M, Kobayashi S, Sakai D, Eguchi H, Baba H, et al. Randomized phase III study of gemcitabine, cisplatin plus S-1 versus gemcitabine, cisplatin for advanced biliary tract cancer (KHBO1401-MITSUBA). J Hepatobiliary Pancreat Sci 2023;30:102-110.
- 21. Corti F, Nichetti F, Raimondi A, Niger M, Prinzi N, Torchio M, et al. Targeting the PI3K/AKT/mTOR pathway in biliary tract cancers: a review of current evidences and future perspectives. Cancer Treat Rev 2019;72:45-55.
- 22. Farshidfar F, Zheng S, Gingras MC, Newton Y, Shih J, Robertson AG, et al. Integrative genomic analysis of cholangiocarcinoma identifies distinct IDH-mutant molecular profiles. Cell Rep 2017;19:2878-2880.
- 23. Li G, Xu D, Sun J, Zhao S, Zheng D. Paclitaxel inhibits proliferation and invasion and promotes apoptosis of breast cancer cells by blocking activation of the PI3K/AKT signaling pathway. Adv Clin Exp Med 2020;29:1337-1345.
- 24. Su Y, Wu M, Zhou B, Bai Z, Pang R, Liu Z, et al. Paclitaxel mediates the PI3K/AKT/mTOR pathway to reduce proliferation of FLT3‑ITD+ AML cells and promote apoptosis. Exp Ther Med 2024;27:161.
- 25. Liu R, Chen Y, Liu G, Li C, Song Y, Cao Z, et al. PI3K/AKT pathway as a key link modulates the multidrug resistance of cancers. Cell Death Dis 2020;11:797.
- 26. Chen D, Lin X, Zhang C, Liu Z, Chen Z, Li Z, et al. Dual PI3K/mTOR inhibitor BEZ235 as a promising therapeutic strategy against paclitaxel-resistant gastric cancer via targeting PI3K/Akt/mTOR pathway. Cell Death Dis 2018;9:123.
- 27. André F, Ciruelos E, Rubovszky G, Campone M, Loibl S, Rugo HS, et al. Alpelisib for PIK3CA-mutated, hormone receptor-positive advanced breast cancer. N Engl J Med 2019;380:1929-1940.
- 28. Turner NC, Oliveira M, Howell SJ, Dalenc F, Cortes J, Gomez Moreno HL, et al. Capivasertib in hormone receptor-positive advanced breast cancer. N Engl J Med 2023;388:2058-2070.
- 29. Turner NC, Im SA, Saura C, Juric D, Loibl S, Kalinsky K, et al. Inavolisib-based therapy in PIK3CA-mutated advanced breast cancer. N Engl J Med 2024;391:1584-1596.
, Chang Gon Kim1
