Extracellular matrix biomarkers of T-cell infiltration and tumor fibrosis predict response to nivolumab ± ipilimumab with SBRT in biliary tract cancer: insights from the CheckPAC trial

Abstract

Background

Biliary tract cancer (BTC) is an uncommon malignancy with limited treatment options and poor prognosis. BTC is typically characterized by a desmoplastic, collagen-rich extracellular matrix (ECM), which has been linked to immune exclusion and therapy resistance. Although immune checkpoint inhibitors (ICI) combined with gemcitabine/cisplatin have become first-line treatment for advanced BTC, durable responses are rare, and predictive biomarkers for immunotherapy are lacking. We investigated the pharmacodynamic and predictive potential of liquid, ECM-derived biomarkers reflecting cytotoxic T-cell activity (granzyme B-degraded type IV collagen [C4G]) and fibrotic activity (pro-peptides of type III [PRO-C3] and VI [PRO-C6] collagens, matrix metalloprotease-degraded type I [reC1M], III [C3M], and IV collagens [C4M]) in patients with metastatic BTC receiving combined immunotherapy and radiotherapy.

Methods

Biomarkers (C4G, PRO-C3, PRO-C6, reC1M, C3M, and C4M) were measured in serum from 61 patients with metastatic BTC enrolled in CheckPAC (NCT02866383), treated with stereotactic body radiotherapy (SBRT) combined with nivolumab (n = 19) or nivolumab/ipilimumab (n = 42). Biomarkers were assessed at baseline and day 60. Associations of baseline levels and on-treatment changes with overall survival (OS) and clinical benefit rate were evaluated; longitudinal analyses used a landmark approach.

Results

Higher baseline PRO-C3 and reC1M were associated with lack of clinical benefit (p < 0.05) and shorter OS (p < 0.05). In multivariable Cox regression adjusting for CA 19 - 9, ECOG performance status, and modified Glasgow Prognostic Score, PRO-C3 remained independently associated with OS. Longitudinally, C4G increased from baseline to day 60 in all patients with clinical benefit (p < 0.001), whereas no consistent changes were observed among patients without clinical benefit. For the landmark analyses, C4G increase was associated with clinical benefit (p = 0.007) and longer OS (p = 0.0045). Patients with low PRO-C3 and increased C4G at day 60 showed the most favorable survival, including a subgroup without RECIST-defined clinical benefit (p < 0.001).

Conclusions

Serological biomarkers reflecting tumor fibrosis (PRO-C3) and cytotoxic T-cell infiltration (C4G) were associated with clinical benefit and OS and showed pharmacodynamic changes during therapy in patients with metastatic BTC treated with SBRT plus ICI. These biomarkers enable tracking of pharmacodynamic response to ICI, while independent validation is necessary to ensure their predictive utility in immunotherapy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04989-4.

Background

Biliary tract cancer (BTC) continues to have a dismal prognosis with a 5-year overall survival (OS) rate below 15% and limited treatment options for recurrent or metastatic disease [1]. While both innate and adaptive immune cells are present in the tumor microenvironment (TME), clinical benefit from immune checkpoint inhibition remains restricted to a subset of patients. Immune checkpoint inhibitors (ICI) combined with gemcitabine/cisplatin have recently become a standard first-line option for unresectable or metastatic BTC [2–5], yet durable responses beyond 12 months remain uncommon [4, 5]. Approaches under investigation to improve outcomes in metastatic (m)BTC include the combination of ICI with either conventional radiotherapy or stereotactic body radiotherapy, with the aim of enhancing tumor immunogenicity [6, 7].

Biomarkers for prediction and monitoring of response to immunotherapy are a major unmet need in mBTC [8]. Established immunotherapy biomarkers (e.g., PD-L1 expression and tumor mutational burden [TMB]) have found widespread use in immunologically active solid tumors [9, 10]. However, most have shown limited and inconsistent predictive value in fibrosis-rich tumors such as BTC, renal cell carcinoma, and small cell lung cancer [4, 11, 12]. High TMB or microsatellite instability occur at low frequencies in BTC, limiting their clinical utility as generally applicable biomarkers [13]. A major contributing factor to the treatment resistance is the immunosuppressive and desmoplastic (fibrotic) TME characterizing BTC [14, 15]. In recent years, extensive remodeling of the extracellular matrix (ECM) has been recognized as an important component of the fibrotic TME in gastrointestinal cancers, such as pancreatic ductal adenocarcinoma and BTC [16, 17]. The fibrotic reaction leads to tumor fibrosis with increased collagen density and stiffness that promotes immunosuppression and tumor growth [18–20]. In BTC, expansion of cancer-associated fibroblasts (CAFs) and subsequent remodeling of the ECM have been associated with lymph node metastasis and limited immune response [21, 22]. We previously reported that a serum biomarker of type III collagen formation (PRO-C3) reflects CAF activity through fibrotic activity and is a prognostic biomarker in BTC and pancreatic cancer [23, 24]. Additional blood-based biomarkers of collagen remodeling: type I, III, and IV collagen degradation (reC1M, C3M, C4M) and type VI collagen formation (PRO-C6) have also been linked to fibrotic activity and poor outcomes in patients with BTC treated with chemotherapy [23, 25, 26].

Beyond fibrosis in the TME, there is an urgent unmet need for biomarkers that capture treatment-induced immune engagement within the TME [27, 28]. Monitoring T-cell activation during treatment can be challenging as ctDNA cannot account for TME changes, while liquid biopsy approaches such as cytokine analysis or T-cell receptor sequencing give information about systemic immune activation, but with limited correlation to intratumoral T-cells [29]. Meanwhile, serial tumor biopsies are invasive and may be limited by sampling bias and tumor heterogeneity [30, 31]. Activation and infiltration of cytotoxic T-cells is a hallmark of immunotherapy response [32]. Activated T-cells infiltrating tumor tissue release granzyme B, which cleaves type IV collagen within vascular basement membranes facilitating immune-cell extravasation [33]. We previously showed that a quantifiable fragment of granzyme B degraded type IV collagen (C4G) predicts ICI response in metastatic melanoma [34].

In this study we investigated whether serum biomarkers of TME fibrosis (PRO-C3 and related collagen remodeling markers) and T-cell activation-associated collagen remodeling (C4G) are associated with clinical benefit and survival in patients with metastatic BTC treated with SBRT plus nivolumab with or without ipilimumab in the CheckPAC trial. We further assessed whether longitudinal changes in C4G provide pharmacodynamic information during combined ICI and radiotherapy.

Methods

Patients

Serum samples were obtained within the CHOCA biomarker study framework (NCT05184400) from the 61 patients with BTC enrolled in the CheckPAC trial (NCT02866383), a randomized phase 2 trial [6]. Briefly, patients were randomized to treatment with stereotactic body radiotherapy (SBRT) plus nivolumab (n = 19, Arm A) or SBRT plus nivolumab/ipilimumab (n = 42, Arm B) for up to 52 weeks or until progressive disease (PD). At interim analysis the SBRT + Nivolumab arm was closed due to limited treatment response, although no significant tests were performed due to the limited number of patients in each arm. Serum samples were collected at baseline, before second treatment cycle at 2 weeks, and every 8 weeks during treatment until PD. Response was evaluated every 8 weeks by computed tomography scan (CT) according to Response Evaluation Criteria in Solid Tumors (RECIST) 1.1.

Collagen biomarker measurements

Biomarkers were measured by competitive enzyme-linked immunosorbent assay (ELISA) or electro-chemiluminescence immunoassay (ECLIA) according to manufacturer’s protocol (Nordic Bioscience, Herlev, Denmark). The following biomarkers were measured: reC1M (NordicC1M™, cat.no. 1000-01), PRO-C3 (NordicPRO-C3™, cat.no 1700AF01), C3M (NordicC3M™, cat.no 1200-02/1200-03), C4M (NordicC4M™, cat.no 1300-01/1300-02), C4G (NordicC4G™, cat.no R1027-00), PRO-C6 (NordicPRO-C6™, cat.no 4000AG01). Description of biomarkers and reference to technical descriptions are provided in Table 1.

Table 1.

Biomarker description

Biomarker	Biological source of biomarker	Reference
reC1M	MMP-degraded type I collagen	[35]
PRO-C3	N-terminal pro-peptide of type III collagen	[36]
C3M	MMP-degraded type III collagen	[37]
C4M	MMP-degraded IV collagen	[38]
C4G	granzyme B-degraded type IV collagen	[34]
PRO-C6	Pro-peptide of type VI collagen	[39]

Briefly, 96-well streptavidin-coated plates were coated with biotinylated peptides specific for the protein of interest, washed, and 20 µL standard peptide or pre-diluted serum was added along with peroxidase-conjugated monoclonal antibodies. Following incubation and washing, tetramethylbenzidine (Kem-En-Tec Diagnostics, Denmark) substrate was added while the substrate peroxidase reaction was stopped using 1% H2SO4 after incubation. Serum samples were run in technical duplicates and quantified from a standard curve of known standard peptide concentrations. Biomarker levels above the upper limit of detection or below the lower detection thresholds were assigned to the lower and upper limits, respectively, and technical duplicates above 20% variance were rerun. Biomarker healthy reference ranges were provided by the biomarker assay catalogues.

Statistical analysis

Clinical benefit was defined as RECIST 1.1 disease control (complete response [CR] + partial response [PR] + stable disease [SD] versus PD). Treatment arm A and B were pooled to obtain statistical power due to the limited number of patients with clinical benefit. Biomarker differences between response groups were assessed using Students’ t-test and biomarker correlations at baseline were evaluated using the Pearson correlation coefficient. Biomarker differences across RECIST 1.1 response groups were evaluated using analysis of variance and compared using an F-test. Biomarkers were log-transformed prior to testing.

Kaplan-Meier analysis with log-rank test used for statistical significance was conducted on biomarkers stratified by median biomarker concentration. PRO-C3 was stratified by the 75th percentile, as this has previously been shown to be superior [40]. Prognostic value of biomarkers was evaluated using univariate Cox regression with OS at outcome and multivariate Cox regression with OS as outcome and carbohydrate antigen 19 − 9 (CA 19-9), Eastern Cooperative Oncology Group Performance Status (ECOG PS), and modified Glasgow Prognostic Score (mGPS) as covariates due to their established prognostic significance in BTC [41–43].

Longitudinal changes were evaluated using landmark analysis, due to a high patient dropout after second treatment cycle and before third blood sample (~ 45–65 days after treatment start and initial CT-evaluation). Patients with samples available after 45 days of treatment were included in the landmark analysis and their most recent sample after inclusion used as landmark time. A total of 37 patients were available for landmark analysis. Two patients were excluded due to third sample after landmark being > 100 days after treatment initiation. Median time from baseline to treatment cycle 2 was 18 days [range 11–32 days] and median time from baseline to landmark serum sample was 58 days [range 46–87 days]. These time points are referred to as day 20 and day 60 throughout the manuscript. Biomarker differences at baseline between patients included and excluded from landmark analysis was evaluated using Wilcoxon rank-sum test.

Log-fold changes in C4G between baseline and landmark were evaluated in patients with clinical benefit and patients with no clinical benefit, respectively, using a paired t-test. C4G changes were stratified based on increase (C4G increase) or decrease (C4G decrease) relative to the patient’s baseline level. C4G change in different clinical benefit categories were evaluated using Fisher’s Exact test, and differences in C4G change between treatment arms A and B were also evaluated using Fisher’s Exact test. Kaplan-Meier and Cox regression analyses were used to evaluate OS from date of landmark time between C4G change groups and treatment type was used as covariate in multivariate Cox regression. Kaplan-Meier and Cox regression analyses were also used to evaluate OS from date of landmark time in patients with C4G increase and low PRO-C3 levels at day 60 but without clinical benefit, compared to patients with high PRO-C3 and/or C4G decrease without clinical benefit.

Statistical analysis was performed using R (v4.4.3) in RStudio (v2025.09.2). A two-sided p-value < 0.05 was considered statistically significant.

Results

Fibrosis biomarkers at baseline predict clinical benefit and survival time

To evaluate biomarker potential at baseline, we measured C4G, PRO-C3, PRO-C6, C3M, C4M, and reC1M in 61 patients with BTC treated in the CheckPAC trial [6]. Baseline patient characteristics are shown in Tabel 2. A total of nine patients were excluded from analysis, either due to missing response evaluation (n = 6) or missing biomarker measurements (n = 3). All patients were deceased within the follow-up period with a median OS of 5.62 months (1.12–39.95 months). Four patients had PR, 8 patients had SD, and 40 patients had PD.

Table 2.

Patient characteristics

Characteristic	CheckPAC (n = 61)
Median age, years (range)	62 (34–81)
Sex (% female)	34/61 (56%)
ECOG PS, n (%)
0	33 (54%)
1	28 (46%)
Weight loss before diagnosis, n (%)
<5%	26 (43%)
≥5%	27 (44%)
Unknown	8 (13%)
Primary tumor location, n (%)
Intrahepatic cholangiocarcinoma	41 (67%)
Perihilar cholangiocarcinoma	2 (3%)
Extrahepatic cholangiocarcinoma	4 (7%)
Unspecified	1 (2%
Gallbladder	13 (21%)
Biliary stent (%)	11/61 (18%)
CA19-9, median (IQR), kU/L	203 (38-4020)
mGPS, n (%)
0	19 (31%)
1	25 (41%)
2	17 (28%)
Treatment, n (%)
SBRT/Nivolumab	19 (31%)
SBRT/Nivolumab/Ipilimumab	42 (69%)
RECIST 1.1 response:
CR	0
PR	5 (8%)
SD	10 (16%)
PD	40 (66%)
Not evaluable*	6 (10%)

ECOG PS Eastern Cooperative Oncology Group Performance Status, CA19-9 Carbohydrate Antigen 19 - 9, mGPS modified Glasgow Prognostic Score, SBRT Stereotactic body radiotherapy, RECIST 1.1 Response Evaluation Criteria In Solid Tumors 1.1, CR Complete response, PR: Partial response, SD Stable disease, PD Progressive disease

*No post-baseline tumor assessment

Initial analysis of biomarker levels and clinical benefit showed significantly elevated levels of PRO-C3 (p = 0.037, Fig. 1B) and reC1M (p = 0.044, Fig. 1F) in patients with no clinical benefit (PD) compared with those with disease control (SD/PR). We observed no differences in biomarkers C4G, C3M, and C4M (Fig. 1A and D, and 1E), and PRO-C6 showed a non-significant trend (Fig. 1C, p = 0.054). Biomarker levels across RECIST 1.1 response groups (PD vs. SD vs. PR) were also evaluated (Additional file 1: Figure S1).

Fig. 1.

Baseline ECM-related biomarker concentrations in biliary tract cancer patients with and without clinical benefit. Boxplots showing serum concentration of (A) C4G, (B) PRO-C3, (C) PRO-C6, (D) C3M, (E) C4M, and (F) C1M in patients with clinical benefit and patients without clinical benefit. Shaded area represents healthy reference range for each biomarker. C4G, granzyme B-degraded type IV collagen; PRO-C3, PRO-C6, pro-peptide of type III and VI collagen; C1M, C3M, C4M, matrix metalloprotease-degraded type I, III, and IV collagen

In Kaplan-Meier analysis, higher baseline PRO-C3 (p < 0.0001; Fig. 2B), PRO-C6 (p = 0.029; Fig. 2C), C3M (p = 0.0097; Fig. 2D), C4M (p = 0.0073; Fig. 2E), and reC1M (p = 0.0025; Fig. 2F) were each associated with shorter OS, whereas baseline C4G was not associated with OS (p = 0.4; Fig. 2A). C3M, C4M, and reC1M were highly correlated, suggesting shared underlying biology (Additional file 1: Figure S2).

Fig. 2.

Association of ECM biomarkers at baseline and overall survival. Kaplan-Meier plots demonstrating the association of ECM biomarkers with OS for patients with BTC. (A) C4G, (C) PRO-C6, (D) C3M, (E) C4M, and (F) C1M were dichotomized on median biomarker level. (B) PRO-C3 was dichotomized at 75th percentile biomarker level. C4G, granzyme B-degraded type IV collagen; PRO-C3, PRO-C6, pro-peptide of type III and VI collagen; C1M, C3M, C4M, matrix metalloprotease-degraded type I, III, and IV collagen

Based on their significant associations to both clinical benefit and longer OS, PRO-C3 and reC1M were further analyzed in univariate and multivariate Cox regression analysis. High PRO-C3 was significantly associated with OS risk in univariate analysis (HR = 3.9, 95% CI 1.9-8, p < 0.001; Table 3) and was independent of CA19-9, ECOG PS, and mGPS score in multivariate analysis (HR = 2.9, 95% CI 1.4–6.2, p = 0.005; Table 3). Although high reC1M was significantly associated with OS in univariate analysis (HR = 2.4, 95% CI 1.3–4.2, p = 0.003; Table 3), reC1M did not retain significance after covariate adjustment (HR = 1.2, 95% CI 0.5–2.7, p = 0.708; Table 3). Hence, baseline levels of the fibrotic activity biomarker PRO-C3 were significantly associated with OS in BTC independent of common risk factors.

Table 3.

Univariate and multivariate Cox regression analysis of survival and PRO-C3, C1M, and prognostic clinical variables

Univariate Cox regression					Multivariate Cox regression					Multivariate Cox regression
Variable		N	HR (95% CI)	p value	Variable		N	HR (95% CI)	p value	Variable		N	HR (95% CI)	p value
PRO-C3	Low	39	-	-	PRO-C3	Low	39	-	-	reC1M	Low	26	-	-
	High	13	3.9 (1.9-8)	<0.001		High	13	2.9 (1.4–6.2)	0.005		High	26	1.2 (0.5–2.7)	0.708
reC1M	Low	26	-	-	CA19-9 (kU/L)	<198	27	-	-	CA19-9 (kU/L)	<198	27	-	-
	High	26	2.4 (1.3–4.2)	0.003		≥198	28	1.0 (0.5–1.8)	0.931		≥198	28	0.9 (0.5–1.6)	0.696
CA19-9 (kU/L)	< 198	27	-	-	ECOG PS	0	31	-	-	ECOG PS	0	31	-	-
	≥ 198	28	1.3 (0.8–2.3)	0.331		1	24	1.4 (0.8–2.5)	0.278		1	24	1.2 (0.6–2.3)	0.604
ECOG PS	0	31	-	-	mGPS	0	18	-	-	mGPS	0	18	-	-
	1	24	1.5 (0.9–2.5)	0.166		1	22	2.8 (1.2–6.6)	0.017		1	22	3.5 (1.4–8.8)	0.008
mGPS	0	18	-	-		2	15	2.3 (1.0-5.1)	0.041		2	15	2.4 (0.9–6.4)	0.071
	1	22	3.5 (1.7-7.0)	<0.001
	2	15	2.7 (1.3–5.6)	0.008

Significant p values (p < 0.05) are highlighted

HR Hazard ratio, CI: Confidence interval, ECOG PS Eastern Cooperative Oncology Group Performance Status, mGPS modified Glasgow Prognostic Score

C4G changes during treatment suggest pharmacodynamic effect and predict clinical benefit

Next, we wanted to investigate biomarker dynamics during treatment, especially T-cell infiltration dynamics during treatment through biomarker C4G, a marker for T-cell infiltration into the tumor. Due to high patient dropout, we utilized landmark analysis (n = 35) to evaluate biomarker changes from baseline to 60 days after treatment initiation. Baseline comparisons between included vs. excluded patients showed no significant difference in baseline C4G, suggesting that C4G-related findings were unlikely driven by landmark selection; PRO-C3 differed between groups (Additional file 1: Figure S3), consistent with its association with poor prognosis.

Biomarkers related to fibrotic activity (PRO-C3, PRO-C6) or collagen degradation (reC1M, C3M, C4M) did not show consistent separation by clinical benefit, with most patients having increased serum levels of PRO-C3, C3M, and C4M at day 60 (Additional file 1: Figure S4).

Interestingly, all patients with clinical benefit had an increase in C4G from baseline to day 60, whereas C4G decreases occurred only among patients without clinical benefit (Fig. 3A). Accordingly, patients with clinical benefit had a significant increase in C4G between baseline and day 60 (p < 0.001), whereas patients without clinical benefit did not (p = 0.9) (Fig. 3B). Specifically, 44% of patients with C4G increase experienced clinical benefit of treatment, whereas no clinical benefit was observed among patients without a C4G increase (p = 0.007, Fig. 3C). Patients with C4G increase also had significantly longer median OS compared to patients with a C4G decrease (8.25 months vs. 3.55 months, p = 0.0045, Fig. 3D).

Fig. 3.

Increased C4G between baseline and day 60 associated with clinical benefit and longer OS. (A) Spaghetti plot of C4G changes relative to baseline between baseline and day 60 in patients with clinical benefit and without clinical benefit. (B) Paired line plot of C4G serum concentration at baseline and day 60 in patient with and without clinical benefit. (C) Stacked bar plot of clinical benefit in patients with C4G increase and C4G decrease at day 60, respectively. (D) Kaplan-Meier plot showing the association of C4G increase and decrease with OS. C4G, granzyme B-degraded type IV collagen

Looking into the individual treatment arms, Arm A (SBRT + Nivolumab) had significantly fewer patients with a C4G increase during treatment compared to Arm B (SBRT + Nivolumab/Ipilimumab) (p = 0.004; Additional file 1: Figure S5A). In a multivariate model including treatment arm as covariate, C4G change remained associated with OS (p = 0.015; Additional file 1: Figure S5B), supporting C4G as a pharmacodynamic marker linked to outcome independent of treatment.

Low fibrosis and increased C4G identifies favorable subgroup, including patients without RECIST benefit

We next investigated the combined effect of tumor fibrosis and immune activation due to the known immunosuppressive effect of tumor fibrosis [20]. Specifically, we looked at tumor fibrosis evaluated by PRO-C3 level at day 60 and immune activation based on change in C4G from baseline to day 60 in patients with clinical benefit and no clinical benefit separately.

All patients with clinical benefit (6 with SD and 4 with PR) had both low PRO-C3 levels (PRO-C3 < 75th percentile) and increased C4G (Fig. 4A). Interestingly, a patient subgroup without clinical benefit (n = 9) also exhibited low PRO-C3 with increased C4G (Fig. 4A) and a trend for longer median OS than the remaining patients without benefit (8.25 months vs. 3.55 months p = 0.12; Fig. 4B). When groups were compared overall, the combined profile of low PRO-C3 and increased C4G was associated with improved OS (p = 0.00032; Fig. 4C). Consistently, 6-months OS was 90% among patients with clinical benefit, 66% among patients without clinical benefit but with low PRO-C3/increased C4G, and 31% among the remaining patients without clinical benefit (Fig. 4D).

Fig. 4.

Low PRO-C3 and C4G increase associated with clinical benefit and longer OS without treatment response. (A) Boxplot showing PRO-C3 serum concentrations on day 60 in patients stratified by C4G increase or decrease as well as by clinical benefit. Shaded area represents PRO-C3 healthy reference range. Light-blue line represents PRO-C3 75th percentile split (High vs. low PRO-C3). (B) Kaplan-Meier plot showing the association of PRO-C3 low + C4G increase or PRO-C3 high and/or C4G decrease groups with OS in patients with no clinical benefit. (C) Kaplan-Meier plot showing the association of OS and patients with treatment response, NR: PRO-C3 low + C4G increase, and NR: PRO-C3 high and/or C4G decrease. (D) 6-months OS for patients with treatment response, NR: PRO-C3 low + C4G increase, and NR: PRO-C3 high and/or C4G decrease. C4G, granzyme B-degraded type IV collagen; PRO-C3, pro-peptide of type III collagen; NR, non-response

Discussion

In this study, we demonstrated how biomarkers of collagen remodeling and T-cell activation, namely PRO-C3 and C4G, could serve as non-invasive biomarkers of the TME, capturing both tumor fibrosis and T-cell engagement in immunosuppressive tumors (Fig. 5). Our results showed C4G increase associated with clinical benefit and was particularly pronounced in patients receiving the more potent dual ICI regimen. Furthermore, patients with a C4G increase and low tumor fibrosis, reflected by low PRO-C3 levels, exhibited improved OS even in the absence of a radiological response.

Fig. 5.

Overview of study findings and biomarker utility. (Top) Illustration of ECM-derived biomarker association with TME phenotypes. Immune excluded tumors with active fibrosis release PRO-C3 while immune inflamed tumors with high T-cell infiltration release C4G. (Bottom) Biomarkers quantified in serum can potentially identify TME phenotypes in patients with BTC and predict response to combined immunotherapy and radiotherapy. Created in BioRender. Rasmussen, M. (2026) https://BioRender.com/6frqbf6

High baseline PRO-C3, a biomarker of type III collagen formation and fibrosis, was significantly associated with poor clinical outcome when assessing both clinical benefit and OS, retaining its significance when adjusting for common prognostic factors. We have previously linked high baseline PRO-C3 levels to poor prognosis in patients with BTC treated with chemotherapy [23, 26]. A link between PRO-C3 and resistance to ICI have been observed in malignant melanoma [25, 40], while a general association between tumor fibrosis and immunotherapy resistance has been characterized across multiple cancer [44–46]. Specifically, type III collagen forms a barrier around tumor clusters, physically preventing T-cell infiltration and reducing T-cell activity through collagen density [18, 47]. Meanwhile type III collagen binding of the leukocyte-associated immunoglobulin-like receptor 1 (LAIR-1) immune regulator can directly regulate immune cell function [48]. In BTC, overexpression of ECM proteins and excess deposition of type I and III collagen have been associated with poor outcome [19, 49]. Patients with high PRO-C3 levels in this cohort likely correspond to the mesenchymal-like subtype of BTC, characterized by high fibrosis, widespread immunosuppression, and poor OS [50].

Various kinase inhibitors approved for treatment of fibrosis-rich cancers have been shown to modulate the crosstalk between ECM-producing CAFs and tumor cells [51, 52]. Nintedanib, an FGFR, PDGFR, and VEGFR inhibitor, directly inhibit collagen production from CAFs, while FGFR inhibitor Erdafitinib have been linked to increased T cell infiltration due to breakage of CAF-derived ECM [52, 53]. FGFR inhibitors are currently being investigated to treat FGFR-altered BTC, and immunologically cold intrahepatic cholangiocarcinoma (a subtype of BTC) have been shown to be significantly enriched for FGFR alterations [54–57] Our results suggest that patients with high tumor fibrosis/high PRO-C3 may be especially resistant to immunotherapy and combination of immunotherapy with a fibrosis-modulating kinase inhibitors could therefore improve treatment efficacy in this patient population.

Landmark analysis demonstrated that an increase in C4G during treatment was associated with clinical benefit and improved OS, supporting its role as a pharmacodynamic marker of T-cell engagement. Furthermore, the single-agent checkpoint blockade arm was closed at interim analysis due to limited efficacy compared with dual ICI, in accordance with our observation of few patients exhibiting a C4G increase in the nivolumab arm A. While C4G showed prognostic value independent of treatment arm, these findings indicate that the addition of Ipilimumab enhances T-cell priming, leading to increased immune infiltration and improved treatment response. We have previously shown that high C4G is predictive of ICI response in patients with malignant melanoma and is prognostic for OS in patients with glioblastoma [34, 58]. Consistent with earlier findings, baseline C4G was not associated with treatment response [26], which is unsurprising given the limited predictive value of other biomarkers for immunotherapy response in BTC [59]. Importantly, all patients experiencing clinical benefit exhibited an increase in C4G during therapy, and C4G increase was significantly associated with both clinical benefit and longer OS.

All patients with clinical benefit were characterized by a combined profile of increased C4G and low PRO-C3 levels. Furthermore, a subset of patients without RECIST-defined clinical benefit also displayed a similar profile (C4G increase, PRO-C3 low) and demonstrated longer OS. Hence, biomarker-defined immune engagement may confer survival benefit even without meeting radiographic criteria for disease control. Potential explanations include slowed tumor growth without categorical response, mixed responses, or limitations of RECIST in capturing immunotherapy effects. However, the lack of paired tumor biopsies or comprehensive longitudinal analyses of cytokines dynamics limit further mechanistic interpretation. In contrast, patients without C4G increase, or those with a C4G increase in combination with high PRO-C3 levels, did not experience a survival benefit compared with patients exhibiting both C4G increase and low PRO-C3 levels. These patterns are consistent with the immune-excluded BTC phenotype, where either insufficient immune priming or stromal barriers may limit effective T-cell infiltration and antitumor activity (Fig. 5) [8, 50].

Certain limitations are present in this study. No matching tumor biopsies were available, precluding direct validation of intratumoral T-cell infiltration, fibrotic activity, collagen architecture, and molecular subtype assignment. However, blood-based biomarkers are used because of the challenging and painful process to re-biopsy patients with metastatic BTC. C4G association to radiological response supports a role for C4G as a marker of treatment-induced T-cell recruitment, whereas PRO-C3 has previously been shown to be associated with active tissue fibrosis [36]. The trial also lacked a radiotherapy-only or non-ICI control arm, limiting conclusions about whether biomarkers are predictive of immunotherapy benefit versus prognostic and reflecting general treatment responsiveness. Nonetheless, the significant association of C4G increase to the combined immunotherapy arm does suggest correlation to efficacy and immune activation. In-depth cytokine and immune cell activity investigations could be used to directly link biomarkers and immune cell activity. Lastly, cohort size and high early mortality limited statistical power and motivated landmark analysis. Importantly, excluding patients with higher baseline PRO-C3 in landmark analysis implies that findings for combined PRO-3/C4G may be conservative. The biomarkers should be validated in a large, independent, multicenter cohort of patients with BTC receiving combined radiotherapy and immunotherapy. However, the low incidence of BTC creates challenges in assembling a sufficiently large cohort. Indeed, the predictive potential of C4G and PRO-C3 could instead be investigated across cancer types known for harboring a fibrotic TME and responding to ICI.

We evaluated longitudinal changes in C4G as a landmark analysis to minimize bias related to early deaths and to avoid immortal time bias, as patients with available measurements beyond 45 days are, by definition, required to survive and remain on treatment until that timepoint. To investigate any difference between included and excluded patients, baseline biomarker levels were compared between the two groups. While C4G levels were similar, PRO-C3 levels were significantly higher in patients excluded from the landmark analysis compared to included patients. This is consistent with the association between high baseline PRO-C3 levels, lack of clinical benefit, and short OS. However, it also suggests that we underestimated the prognostic potential observed when combining C4G change and PRO-C3 level. Analyzing the full cohort without the patients who died early (if this could be done without confounding the data) would show the full predictive potential of assessing both T-cell infiltration and tumor fibrosis.

To the best of our knowledge, this is the first study to utilize liquid biopsies to track immune activation in BTC during treatment and suggests that collagen-derived biomarkers such as C4G and PRO-C3 could be used as non-invasive precision biomarkers to identify patient benefitting from immunotherapy (Fig. 5). It even could be speculated that high PRO-C3 levels are prevalent across tumors with fibrotic TMEs whereby PRO-C3 and C4G could be utilized as a predictive biomarker for immunotherapy across these tumors where predictive biomarkers fall short.

Conclusions

In this study, we identified C4G (T-cells) and PRO-C3 (fibrosis) as biomarkers associated with clinical benefit and overall survival in patients with BTC treated with SBRT plus nivolumab with or without ipilimumab. These biomarkers enable characterization of the fibrotic/immune state of the TME and provide a non-invasive approach for early treatment monitoring. While they exhibit potential predictive value for immunotherapy response, predictive utility requires additional validation in controlled cohorts.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BTC

Biliary tract cancer

C3M

Matrix metalloprotease-degraded type III collagen

C4G

Granzyme B-degraded type IV collagen

C4M

Matrix metalloprotease-degraded type IV collagen

CA19-9

Carbohydrate Antigen 19 − 9

CAF

Cancer-associated fibroblasts

CI

Confidence Interval

CR

Complete response

ECM

Extracellular matrix

ECOG PS

Eastern Cooperative Oncology Group Performance Status

FGFR2

Fibroblast growth factor receptor 2

ICI

Immune checkpoint inhibitors

mGPS

modified Glasgow Prognostic Score

OS

Overall survival

PD

Progressive disease

PR

Partial response

PRO-C3

Pro-peptide of type III collagen

PRO-C6

Pro-peptide of type VI collagen

reC1M

Matrix metalloprotease-degraded type I collagen

SBRT

Stereotactic body radiotherapy

SD

Stable disease

TME

Tumor microenvironment

Author contributions

Conceptualization, M.B.R., M.A.K., N.W.; Writing: M.B.R. and N.W.; data curation and data analysis, M.B.R.; patient inclusion and sample collection, I.M.C., J.S.J., S.T., A.M., and T.D.C.; collection of clinical data, I.M.C., J.S.J., S.T., A.M., and T.D.C; critical revision of the article, all authors. All authors read and approved the final manuscript.

Funding

Nordic Bioscience A/S, Denmark, provided salary for MBR, MAK, and NW and paid for expenses related to laboratory analysis of ECM biomarkers.

Data availability

The CheckPAC trial protocol lacked a provision for data sharing; consequently, trial data will not be disseminated publicly, as this aspect was not part of the initial ethical approval request. Materials and data used in this study can be made available upon reasonable request. Some restrictions may apply due to Material Transfer Agreements (MTAs) or institutional policies. Requests should be directed to corresponding author Martin B. Rasmussen, mras@nordicbio.com.

Declarations

Ethics approval and consent to participate

This study was approved by the Regional Ethics Committee (no. H-16031247) and the Danish Medicine Agency (EudraCT no. 2016-001883-12). All patients provided written informed consent. All patients were also included in the Danish CHOCA study “Identification of New Biomarkers for Patients with Cholangiocarcinoma and Gallbladder Cancer” (NCT05184400), approved by the Danish Ethics Committee (H-3-2014-055) and the Danish Data Protection Agency (j.nr. HEH 2014-097, I-Suite j.nr. 03260, PACTIUS-2020-797).

Consent for publication

All authors have agreed to the submission and publication of this research.

Competing interests

M.B.R., M.A.K., and N.W. are employed at Nordic Bioscience. M.A.K. and N.W. have stocks in Nordic Bioscience. Besides these, the authors declare no conflict of interest.