Clinicopathological and prognostic implications of vessels encapsulate tumor clusters with PD-L1 in intrahepatic cholangiocarcinoma patients

Abstract

Background

Frequently abnormal vascularization and immunologic derangement have been uncovered in malignant tumors. In present research, we evaluated prognostic characteristic and clinicopathological features of vessels encapsulate tumor clusters (VETC) and the immune checkpoint molecule, programmed cell death-ligand 1 (PD-L1) in patients diagnosed as intrahepatic cholangiocarcinoma (ICC).

Methods

VETC and PD-L1 were investigated in two cohort enrolling 412 ICC patients. VETC and PD-L1 was easily detectable in whole slides and tissue microarray (TMA). Prognostic analysis was performed through Kaplan-Meier cures, log-rank tests and nomograms.

Results

VETC+ was significantly associated with aggressive tumor features. VETC+ predicted a significantly unfavorable survival and higher metastasis and recurrence rates. Furthermore, nomograms integrated by the combination of VETC and PD-L1, that heralded better prognostic value compared to previous staging systems.

Conclusions

Heterogeneous patterns of VETC phenotype and PD-L1 status were both illustrated to be an independent prognostic predictor for clinical outcomes. Therapies designed to target both vascularization and autoimmunity may open a novel direction for HCC. HCC should be replaced by ICC.

Introduction

Intrahepatic cholangiocarcinoma (ICC) ranks second and accounts 10–15% in primary liver cancers (1). Although ICC has made great progress in molecular basis, diagnosis and treatment, its morbidity and mortality are still steadily increasing worldwide (2). Currently, most ICC patients are diagnosed with advanced disease, as they are not eligible for complete surgical resection (3). Therefore, new models and biomarkers are urgently needed to stratify ICC patients based on their prognosis for better risk stratification and comprehensive treatment.

Tumor metastasis is a multi-step and complex process that can be divided into local infiltration and intravascular perfusion (4). Intravascular perfusion is a process that relies on tumor cells entering the surrounding blood vessels, as angiogenic tumors are more likely to be perfused intravascularly (5). Therefore, the unique pattern of vascularization by tumor-associated angiogenesis and pathological capillary formation predicts rapid tumor diffusing and high recurrence rates (6). Remarkably, previous researched illustrated a novel pattern of vascularization, characterized by CD34 positive staining completely encapsulating tumor clusters, named VETC, was significant associated with higher metastasis and recurrence rates in hepatocellular carcinoma (HCC) (7). Furthermore, significantly benefit with the treatment of sorafenib was further uncovered in the presence of VETC than those absence in HCC patients (8). This novel pattern of vascularization, which is different from traditional capillaries, forming a spider web network and tumor islands (9). However, it remained unknown how this vascularization is formed in ICC. As the prognostic significance associated with VETC was unknown in ICC, these observations prompted us to consider that whether VETC affected angiogenesis in ICC.

The immune checkpoint index combined programmed death receptor-1 (PD-1) and programmed death ligand 1 (PD-L1) was treated as the main target of immunotargeted therapy in several malignant tumors with aberrant PD-L1 expression (10-12). In addition, elevated expression of PD-1/PD-L1 index was illustrated as survival predictor for HCC (13). Aberrant status of PD-1/PD-L1 discovered on tumor associated lymphocytes, endothelial cells and tumor cells, was defined as a signal of immune suppression (14). Previous researches indicated that elevated PD-L1 was found and further defined as an immune escape mechanism for occupational cholangiocarcinoma (15). However, due to the tumor heterogeneity and complex etiology, previous studies had indicated that PD-L1 was elevated and predicted dismal prognosis in ICC (16). Therefore, we evaluated the PD-L1 status in two independent cohorts enrolling 412 ICC cases from a single institution. As immune checkpoint blockade test with anti-PD-1 inhibitor was performed in several clinical trials, we confirmed that the drug resistance within anti-PD-1 inhibitor would be the majority challenge in ICC patients.

Five-year survival rate of advanced ICC is poorer than 5% due to poor efficacy of non-systemic treatment and chemotherapy drugs (17), and rare randomized trials of chemotherapy were launched in patients with advanced ICC (18). Previous research indicated that gemcitabine and platinum was defined as the first-line treatment for advanced cholangiocarcinoma. In addition, the treatment of gemcitabine combined with oxaliplatin and cetuximab, indicated a positive objective response rate of 63% in three cholangiocarcinoma patients (19). Nonetheless, the systemic therapeutic efficacy in ICC is far from satisfactory.

Whether the combination of immune checkpoint blockade with other types of therapies could improve anti-tumor efficacy in ICC, would be a leading challenge in the near future. Consistent with the previous studies in HCC, ICC also presented an elevated vascularization (20). Investigating this vascularization pattern was crucial, since combination therapy might produce better efficacy than monotherapy (21). Remarkably, previous research indicated that immune checkpoint blockade could enhance intra-tumor blood perfusion through the vascular normalization in both preclinical models of colorectal and breast cancers (22). In addition, the combination of anti-PD-1 and anti-VEGFR-2 inhibitors could evaluate the normal vascularization and enhance the anti-tumor efficacy in various malignancies (23). Nonetheless, the specific role remained unknown in the combination of immune checkpoint blockade with anti-VEGF/R therapy in ICC.

According to this, we assumed that VETC presenting and elevated PD-L1 expression could be defined as survival predictors for ICC. In addition, VETC presenting and elevated PD-L1 expression were significantly correlated with aggressive tumor features and independently associated with dismal clinical results, which could effectively stratify patients. Furthermore, we established an integrated nomogram with VETC/PD-L1 index for a more accurate prognosis prediction for ICC.

Methods

Patients and study design

Four hundred and twelve patients performed partial hepatectomy and diagnosed as ICC from January 2005 to December 2015 at Zhongshan Hospital (Shanghai, China) were selected in present study, and randomly grouped into training cohort (n=214), validation cohort (n=108) and external validation cohort (n=90). The study was approved by research ethics committee board of Zhongshan Hospital, Fudan University (No.: y2017-179). Written informed consent was obtained from the patient for publication of this study and signed consent forms are kept in the medical records library. Overall survival (OS) and time to relapse (TTR) were carried out as described previously based on our established guidelines (24).

IHC staining

Detailed construction protocol of tissue microarray (TMA) and immunohistochemistry (IHC) protocol were summarized and consistent with previous study (25). Briefly, after deparaffinization of paraffin-embedded sections, antigen recovery was operated using buffer citric acid (pH =6.0). Slides were incubated with primary antibody overnight at 4 °C. Followed rewarming for 45 minutes and incubating with secondary antibody for 30 minutes, slides were stained with 3, 3'-diaminobenzidine solution and then visualized by hematoxylin. Detailed information of IHC reagents were summarized: CD34 (CD34, clone QBEnd/10, 1:200; Santa Cruz Bio-technology), PD-L1 (PD-L1, clone SP263, 1:200; Ventana).

Evaluation of immunohistochemical staining signals

For CD34 evaluation, immunoreactivity that is continuously arranged around the tumor cluster, and VETC⁺ was evaluated semi-quantitatively and defined as CD34 positive area ≥55%.

For PD-L1 evaluation, three representative images were obtained through Leica DM IRE2 microscope combined with Leica CCD camera DFC420. The combination of intensity and area of positive PD-L1 staining were calculated as the PD-L1 density.

Statistical analysis

IBM SPSS Statistics (Version 26) and R software were applied in the statistical analysis. Related transcriptome sequencing data of pan-cancer and cholangiocarcinoma (CHOL) cohorts were download from The Cancer Genome Atlas (TCGA) database. Continuous variables were compared though GraphPad Prism 7 software applying the Mann-Whitney U, χ², and Fisher’s exact tests, respectively. Kaplan-Meier and log-rank test were used for OS and TTR evaluation. Cox regression analysis was performed for univariate and multivariate analyzing. Nomogram models were constructed by “Rms” package, and the Harrell’s concordance index (C-index) was used for evaluating the discrimination performance of nomograms. All P values <0.05 were defined as statistical significant.

Results

Clinical features of selected ICC patients

The clinicopathological features of 412 ICC patients enrolled in training cohort, validation cohort and external validation cohorts were summarized in Table 1. Briefly, a strong HBV infection predominance was observed (60.3%, 64.8% and 77.8%). Most of ICC cases were Child-Pugh stage A (95.8%, 98.1% and 72.2%), and single tumor accounted to 78.5%, 70.4% and 76.7%, respectively. Tumor diagnosed as poorly differentiated of ICC (Edmondson grade III-IV) were accounted for 37.9%, 40.7% and 30%, respectively. The cumulative 1-, 3-, and 5-year OS rates were 76%, 46%, and 36% for training cohort. For validation cohort, the cumulative 1-, 3-, and 5-year OS rates were 78%, 44% and 34%, respectively. Additionally, the cumulative 1- and 3-year OS rates were 77%, 43% for external validation cohort. One hundred and one (47.2%), 63 (58.3%) and 38 (42.2%) occurred tumor recurrence within 2 years after surgery (early recurrence), 29 (13.5%), 7 (6.5%) and 38 (42.2%) after 24 months (late recurrence), and 84 (39.3%), 38 (35.2%) and 14 (15.6%) patients without recurrence for training, validation and external validation cohort, respectively.

Table 1. Baseline demographic, clinicopathological characteristics and phenotypical features of the whole ICC series (n=412).

Variables	Training cohort (n=214)	Validation cohort (n=108)	External validation cohort (n=90)
Clinical features
Gender (male vs. female)	123/91	71/37	53/37
Age, median (range), years	58 [31–81]	58 [27–79]	64 [36–93]
HBV infection (negative vs. positive)	85/129	38/70	20/70
AFP (ng/mL) (<20 vs. ≥20)	193/21	93/15	80/10
CA-199 (U/mL) (<37 vs. ≥37)	103/111	60/48	47/43
Lymphonodus metastasis (absent vs. present)	177/37	89/19	62/28
TNM stage (I vs. II+III)	164/50	83/25	67/23
Child-Pugh stage (A vs. B)	205/9	106/2	65/25
General macroscopic
Tumor number (single vs. multiple)	168/46	76/32	69/21
Tumor size (≤5 vs. >5)	91/123	54/54	33/57
Macrovascular invasion (absent vs. present)	181/33	95/13	75/15
General microscopic
Liver cirrhosis (absent vs. present)	154/60	82/26	62/28
Microvascular invasion (absent vs. present)	181/33	95/13	75/15
Tumor encapsulation (complete vs. none)	21/193	18/90	19/71
Tumor differentiation (I+II vs. III+IV)	133/81	64/44	63/27
Follow-up
Survival (no vs. yes)	61/153	40/68	67/23
Recurrence (no vs. yes)	84/130	38/70	14/76
Recurrence (≤2 vs. >2 years)	101/29	63/7	38/38

ICC, intrahepatic cholangiocarcinoma; CA-199, carbohydrate antigen 199; TNM, tumor-nodes-metastases.

VETC pattern and PD-L1 status in ICC patients

To identify VETC pattern and PD-L1 status in ICC, the transcriptomics profiles CD34 and PD-L1 were downloaded from FireBrowse database (26). Our finding indicated that the mRNA level of CD34 and PD-L1 were significant elevated in tumor area than those in non-tumor tissues in ICC (Figure S1A,B). According to the transcriptomics profiles of CD34 and PD-L1 in some tumors, down-regulation were also illustrated in other tumors. Hence, the specific position of VETC and PD-L1 should be evaluated in a specific role. Furthermore, positively and significantly correlation between the mRNA level of CD34 and PD-L1 in CHOL was uncovered by the TCGA database (pan-cancer cohort, R=0.13, P=1.7e-28; CHOL cohort, R=0.54, P=0.00012; Figure S2A,B).

VETC phenotype was evaluated by IHC in total 412 ICC using TMA. Interestingly, two distinct vascular patterns in ICC was observed: tumor associated vessels combined with discrete lumens (defined as classical capillary vessels), and tumor associated vessels which encapsulated tumor cluster (VETC pattern). ICC patients were divided into VETC⁺ and VETC^–, according to the VETC pattern (Figure 1A). Sequence slices further showed that VETC formed a network around a single ICC nodule, while capillaries indicated a discrete and disordered pattern. However, among 412 ICC tissues examined, 61.9% were VETC⁺ cases (255/412). These findings indicated that VETC is a prevalent pattern of vascularization in ICC.

Figure 1.

VETC pattern and PD-L1 expression in ICC. Representative images of VETC (A) pattern and PD-L1 (B) expression in ICC and adjacent non-tumor tissues. Up: adjacent non-tumor tissues. Bottom: different staining intensities in ICC. Magnification: ×200. VETC, vessels encapsulate tumor clusters; ICC, intrahepatic cholangiocarcinoma.

Meanwhile, IHC evaluation of PD-L1 was performed in the same cohorts. The expression and distribution patterns of PD-L1 were found mainly distributed in both tumor cytoplasm and cell membrane (Figure 1B). However, Heterogeneous PD-L1 status within intra-tumor from different cases were also investigated. Consistent with the mRNA expression in TCGA database, the comparison had showed that significantly elevated PD-L1 expression in ICC intra-tumor area was found than those in adjacent normal intrahepatic biliary tissues (Figure 1B). Consistent with previous study, PD-L1 status was classified as hyper-activated in 50% (206 of 412) of intra-tumor areas, but 19.9% (82 of 412) were defined as hyper-activated in paired normal areas.

Relationship of clinicopathological features with VETC and PD-L1

To further investigated the association between clinicopathological features and VETC and PD-L1 in ICC patients, training cohort were sub-grouped into absent (VETC^-) and present (VETC⁺) groups, high (PD-L1^high) and low (PD-L1^low) expression groups, respectively.

Correlation analysis between clinicopathological features and VETC and PD-L1 were performed and summarized in Table 2, respectively. Remarkably, the presence of VETC⁺ in intra-tumor area was significantly associated with more lymph node metastasis (P=0.017), more microvascular invasion (P=0.018), higher preoperative serum CA-199 level (P=0.021) and early postoperative recurrence (P=0.014). Likewise, elevated PD-L1 status in intra-tumor area was positively correlated with malignant characteristics in ICC, including susceptibility to HBV infection (P=0.035), more liver cirrhosis (P=0.017) and more tendentiousness of lymph node metastasis (P=0.024).

Table 2. Correlation between VETC presenting and PD-L1 expression with clinicopathologic characteristics of the whole ICC series (n=412).

Characteristics	Training cohort (n=214)								Validation cohort (n=108)								External validation cohort (n=90)
	VETC				PD-L1				VETC				PD-L1				VETC				PD-L1
	Absent	Present	P		Low	High	P		Absent	Present	P		Low	High	P		Absent	Present	P		Low	High	P
Gender
Male	56	67	0.231		55	68	0.072		27	44	0.767		32	39	0.155		16	37	0.962		25	28	0.533
Female	34	57			52	39			13	24			22	15			11	26			15	22
Age, year
≤58	54	63	0.182		65	52	0.074		19	33	0.917		22	30	0.123		16	38	0.925		25	29	0.665
>58	36	61			42	55			21	35			32	24			11	25			15	21
HBsAg
Negative	50	81	0.147		73	58	0.035		18	20	0.101		32	6	<0.001		8	12	0.268		14	6	0.009
Positive	40	43			34	49			22	48			22	48			19	51			26	44
AFP (ng/mL)
≤20	79	114	0.313		97	96	0.818		33	60	0.405		48	45	0.403		22	58	0.143		35	45	0.746
>20	11	10			10	11			7	8			6	9			5	5			5	5
CA-199 (U/mL)
<37	35	68	0.021		49	54	0.494		16	44	0.012		32	28	0.438		8	34	0.007		24	23	0.186
≥37	55	56			58	53			24	24			22	26			18	29			16	27
Liver cirrhosis
Absent	68	93	0.925		88	73	0.017		28	54	0.269		36	46	0.024		21	41	0.233		28	34	0.838
Present	22	31			19	34			12	14			18	8			6	22			12	16
Tumor size (cm)
≤5	35	56	0.359		43	48	0.489		16	38	0.111		26	28	0.700		10	23	0.961		26	7	<0.001
>5	55	68			64	59			24	30			28	26			17	40			14	43
Tumor number
Single	71	97	0.907		86	82	0.505		26	50	0.348		41	35	0.206		22	47	0.591		32	37	0.387
Multiple	19	27			21	25			14	18			13	19			5	16			18	13
Microvascular invasion
Negative	70	111	0.018		92	89	0.570		30	65	0.003		48	47	0.767		18	57	0.005		36	39	0.161
Positive	20	13			15	18			10	3			6	7			9	6			4	11
Tumor encapsulation
None	84	109	0.187		98	95	0.491		33	57	0.858		44	46	0.605		23	48	0.409		32	39	0.817
Complete	6	15			9	12			7	11			10	8			4	15			8	11
Tumor differentiation
I+II	57	78	0.948		67	68	0.887		23	41	0.775		35	29	0.240		22	59	0.121		35	46	0.503
III+IV	33	46			40	39			17	27			19	25			5	4			5	4
Lymphonodus metastasis
Absent	82	98	0.017		96	84	0.024		29	60	0.038		48	41	0.076		24	42	0.037		20	42	0.0005
Present	8	26			11	23			11	8			6	3			3	21			20	8
TNM stage
I	73	93	0.290		84	82	0.743		29	54	0.411		43	40	0.493		21	46	0.842		31	35	0.424
II+III	17	31			23	25			11	14			11	14			7	17			9	15
Child-Pugh stage
0-A	86	119	0.882		105	100	0.088		39	67	>0.99		54	52	0.495		18	47	0.258		20	45	0.010
B-C	4	5			2	7			1	1			0	2			10	15			20	15
Follow-up
Early recurrence (≤2 years)	38	65	0.014		52	51	0.667		39	24	0.037		31	32	0.996		9	29	0.009		10	28	0.091
Late recurrence (>2 years)	17	10			12	15			2	6			4	4			6	2			5	3

*, P value <0.05 showed statistical significant. CA-199, carbohydrate antigen 199; TNM, tumor-nodes-metastases; VETC, vessels encapsulate tumor clusters; ICC, intrahepatic cholangiocarcinoma.

Similar results were also found in validation cohort and external independent cohorts. the presence of VETC⁺ in intra-tumor area was significantly associated with higher preoperative serum CA-199 level (P=0.012), more microvascular invasion (P=0.003), more tendentiousness of lymph node metastasis (P=0.038) and early postoperative recurrence (P=0.037). Moreover, elevated PD-L1 status was significantly associated with more tendentiousness of lymph node metastasis (P=0.01) and elevated preoperative serum CA-199 level (P=0.028) (Table 2). Our findings presented that the presence of VETC and elevated PD-L1 status in ICC intra-tumor areas may signify the dismal clinical prognosis and malignant characteristics.

Prognostic values of VETC and PD-L1 in ICC

To further investigate the prognostic values, we assessed potential associations of VETC phenotype and PD-L1 status with patients’ OS and TTR. In the training cohort, VETC⁺ phenotype indicated significantly dismal OS (27 versus 72 months, P=0.0149; Figure 2A) and poorer TTR (14 versus 40.5 months, P=0.0022; Figure 2B) than those in VETC⁻ patients. Similarly, in the validation cohort, VETC⁺ patients showed both unfavorable survival and elevated recurrence (OS, P<0.0001; TTR, P=0.0002; Figure S3A). Consistently, in the external validation cohort, VETC⁺ phenotype illustrated unfavorable prognosis than those in VETC^- patients (OS, P=0.0004; TTR, P=0.0036; Figure S3B).

Figure 2.

Clinical implications of VETC and PD-L1 in ICC. (A,B) Kaplan-Meier curves for OS and TTR based on VETC pattern in training cohort (n=214); (C,D) Kaplan-Meier curves for OS and TTR based on PD-L1 expression in training cohort (n=214); (E,F) Kaplan-Meier curves for OS and TTR based on combined VETC pattern and PD-L1 expression in training cohort (n=214). VETC, vessels encapsulate tumor clusters; ICC, intrahepatic cholangiocarcinoma; TTR, time to relapse; OS, overall survival.

Furthermore, in the training cohort, PD-L1^high patients significantly indicated poorer OS and elevated recurrence than those in PD-L1^low (OS, P=0.0023; TTR, P=0.027; Figure 2C,D). Similarly, results were also found in the validation cohort (OS, P=0.0184; TTR, P=0.0306; Figure S3C) and external validation cohort (OS, P<0.001; TTR, P=0.109; Figure S3D).

Since a potential relationship of VETC and PD-L1 in tumor vascularization and a significantly correlation between CD34 and PD-L1 expression were found, we further constructed a VETC/PD-L1 index. According to this index, training cohort were sub-grouped into three distinct groups: (group I) VETC^- and PD-L1^low; (group II) VETC⁺ or PD-L1^high; and (group III) VETC⁺ and PD-L1^high. Note-worthily, significant prognostic differences were illustrated within these three groups (OS, P<0.0001; TTR, P=0.0003; Figure 2E,F). The 5-year OS rates were 61%, 33.6% and 22.6% for group I, II, and III, respectively. Consistently, similar findings were also evaluated in the validation and external validation cohort (validation cohort, OS, P=0.0001, TTR, P<0.0001; Figure S4A,B; external validation cohort, OS, P<0.0001, TTR, P=0.004; Figure S4C,D). Furthermore, we investigated the univariate and multivariate analyses in both training cohort (Table 3) and validation cohort (Table S1), and VETC/PD-L1 index was illustrated as an independent predictor for both OS and early postoperative recurrence.

Table 3. Impact of clinical and pathological features on OS, TTR and early recurrence in training cohort (n=214).

Characteristics	OS (n=214)						TTR (n=214)						Early recurrence (n=214)
	Univariable analysis			Multivariable analysis			Univariable analysis			Multivariable analysis			Univariable analysis			Multivariable analysis
	HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P
Clinical features
Gender (male vs. female)	1.005 (0.726–1.329)	0.974			NA		0.872 (0.615–1.235)	0.441			NA		0.973 (0.654–1.446)	0.891			NA
Age, median (range), years	1.128 (0.821–1.551)	0.457			NA		1.116 (0.789–1.577)	0.535			NA		1.100 (0.746–1.623)	0.631			NA
HBV infection (negative vs. positive)	0.847 (0.610–1.176)	0.321			NA		0.781 (0.545–1.119)	0.178			NA		0.805 (0.538–1.204)	0.290			NA
AFP (ng/mL) (<20 vs. ≥20)	0.956 (0.552–1.657)	0.872			NA		0.928 (0.500–1.722)	0.812			NA		0.942 (0.475–1.867)	0.864			NA
CA–199 (U/mL) (<37 vs. ≥37)	1.434 (1.0421.973)	0.027*		1.462 (1.057–2.022)	0.022*		1.084 (0.767–1.532)	0.647			NA		1.092 (0.742–1.608)	0.656			NA
Lymphonodus metastasis (absent vs. present)	2.220 (1.476–3.339)	<0.001*		1.564 (0.794–3.082)	0.196		1.874 (1.199–2.929)	0.006*		1.129 (0.516–2.470)	0.760		2.024 (1.261–3.248)	0.004*		1.258 (0.555–2.851)	0.582
TNM stage (I vs. II+III)	1.982 (1.383–2.842)	<0.001*		1.212 (0.660–2.225)	0.535		1.681 (1.130–2.501)	0.01*		1.491 (0.740–3.002)	0.264		1.865 (1.218–2.858)	0.004*		1.439 (0.689–3.004)	0.333
Child–Pugh stage (A vs. B)	0.653 (0.268–1.549)	0.350			NA		0.490 (0.130–1.286)	0.126			NA		0.347 (0.086–1.405)	0.138			NA
General macroscopic
Tumor number (single vs. multiple)	1.437 (0.987–2.091)	0.058			NA		1.451 (0.963–2.186)	0.075			NA		1.363 (0.863–2.153)	0.185			NA
Tumor size (≤5 vs. >5)	1.328 (0.960–1.837)	0.087			NA		1.264 (0.889–1.795)	0.192			NA		1.431 (0.956–2.141)	0.081			NA
Macrovascular invasion (absent vs. present)	1.212 (0.783–1.875)	0.389			NA		1.660 (1.079–2.553)	0.021*		1.420 (0.914–2.205)	0.119		1.519 (0.932–2.475)	0.094			NA
General microscopic
Liver cirrhosis (absent vs. present)	1.453 (1.021–2.07)	0.038*		1.209 (0.822–1.776)	0.335		1.123 (0.750–1.682)	0.574			NA		1.212 (0.781–1.881)	0.391			NA
Tumor encapsulation (complete vs. none)	1.578 (0.891–2.795)	0.118			NA		2.487 (1.212–5.104)	0.013*		2.556 (1.205–5.420)	0.014*		2.722 (1.108–6.691)	0.029*		2.747 (1.114–6.772)	0.028*
Differentiation (I/II vs. III/IV)	0.982 (0.704–1.371)	0.917			NA		1.180 (0.827–1.684)	0.360			NA		1.281 (0.863–1.901)	0.219			NA
Follow-up
Recurrence (≤2 vs. >2 years)	1.290 (1.057–1.574)	0.012*		1.216 (0.990–1.494)	0.063		2.232 (1.812–2.749)	<0.001*		2.182 (1.758–2.709)	<0.001*			NA			NA
VETC (absent vs. present)	2.218 (1.506–3.009)	<0.001*			NA		1.745 (1.213–2.510)	0.003*			NA		1.640 (1.090–2.467)	0.018*			NA
PD-L1 (low vs. high)	1.638 (1.188–2.258)	0.003*			NA		1.471 (1.040–2.079)	0.029*			NA		1.593 (1.079–2.353)	0.019*			NA
Intra-tumoral VETC/PD-L1 index		<0.001*			0.001*			0.003*			0.051			0.008*			0.011*
I	Ref.			Ref.			Ref.			Ref.			Ref.			Ref.
II	0.296 (0.180–0.489)			0.356 (0.210–0.602)			0.422 (0.257–0.692)			0.552 (0.333–0.914)			0.409 (0.231–0.722)			0.422 (0.238–0.749)
III	0.738 (0.523–1.042)			0.780 (0.541–1.124)			0.725 (0.493–1.066)			0.711 (0.480–1.053)			0.710 (0.465–1.085)			0.696 (0.453–1.069)

*, P value showed statistical significant. CA-199, carbohydrate antigen 199; TNM, tumor-nodes-metastases; Ref., reference; HR, hazard ratio; CI, confidential interval; NA, not adopted; NS, not significant; TTR, time to relapse; OS, overall survival.

Table S1. Impact of clinical and pathological features on OS, TTR and early recurrence in validation cohort (n=108).

Variables	OS (n=108)						TTR (n=108)						Early recurrence (n=108)
	Univariable analysis			Multivariable analysis			Univariable analysis			Multivariable analysis			Univariable analysis			Multivariable analysis
	HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P
Clinical features
Gender (male vs. female)	1.266 (0.881–1.819)	0.202			NA		1.106 (0.773–1.583)	0.581			NA		1.448 (0.857–2.448)	0.166			NA
Age, median (range), years	1.026 (0.726–1.448)	0.886			NA		0.994 (0.703–1.406)	0.973			NA		0.667 (0.406–1.907)	0.111			NA
HBV infection (negative vs. positive)	0.751 (0.528–1.068)	0.111			NA		0.755 (0.528–1.078)	0.122			NA		1.018 (0.612–1.693)	0.945			NA
AFP (ng/mL) (<20 vs. ≥20)	0.628 (0.347–1.138)	0.125			NA		0.784 (0.450–1.366)	0.390			NA		0.780 (0.371–1.639)	0.513			NA
CA-199 (U/mL) (<37 vs. ≥37)	1.284 (0.910–1.813)	0.155			NA		1.232 (0.871–1.744)	0.238			NA		1.825 (1.112–2.996)	0.017*		1.768 (1.035–3.020)	0.037*
Lymphonodus metastasis (absent vs. present)	3.700 (2.468–5.547)	<0.001*		5.075 (1.383–18.616)	0.014*		3.041 (2.012–4.598)	<0.001*		3.983 (1.208–13.234)	0.023		3.890 (2.211–6.843)	<0.001*		2.893 (0.792–10.569)	0.108
TNM stage (I vs. II+III)	2.819 (1.929–4.120)	<0.001*		0.629 (0.185–2.136)	0.457		2.343 (1.587–3.461)	<0.001*		0.763 (0.262–2.226)	0.621		2.903 (1.710–4.926)	<0.001*		0.899 (0.266–3.040)	0.864
Child-Pugh stage (A vs. B)	0.694 (0.221–2.183)	0.532			NA		0.869 (0.321–2.354)	0.782			NA		5.138 (1.207–21.864)	0.027*		1.394 (0.301–6.445)	0.671
General macroscopic
Tumor number (single vs. multiple)	1.846 (1.266–2.692)	0.001*	0.950 (0.541–1.669)		0.858		2.020 (1.385–2.946)	<0.001*		1.750 (1.002–3.056)	0.049*		1.986 (1.178–3.346)	0.01*		1.760 (0.978–3.167)	0.059
Tumor size (≤5 vs. >5)	1.735 (1.218–2.472)	0.002*	1.712 (1.029–2.847)		0.039*		1.340 (0.944–1.902)	0.101			NA		1.542 (0.937–2.536)	0.088			NA
Macrovascular invasion (absent vs. present)	1.212 (0.744–1.975)	0.440			NA		1.310 (0.813–2.112)	0.268			NA		0.817 (0.372–1.793)	0.614			NA
General microscopic
Liver cirrhosis (absent vs. present)	1.102 (0.750–1.620)	0.620			NA		1.073 (0.725–1.590)	0.724			NA		1.152 (0.682–1.945)	0.597			NA
Tumor encapsulation (complete vs. none)	1.230 (0.727–2.079)	0.441			NA		1.274 (0.743–2.184)	0.379			NA		0.817 (0.443–1.505)	0.516			NA
Follow-up
Recurrence (≤2 vs. >2 years)	1.939 (1.380–2.723)	<0.001*	1.407 (0.914–2.166)		0.121		2.793 (2.044–3.816)	<0.001*		2.513 (1.713–3.686)	<0.001*			NA			NA
VETC (absent vs. present)	2.064 (1.428–2.982)	<0.001*					1.899 (1.318–2.734)	0.001*					2.368 (1.339–4.188)	0.003*
PD-L1 (low vs. high)	2.096 (1.466–2.997)	<0.001*					1.750 (1.229–2.493)	0.002*					1.591 (0.966–2.618)	0.068
Intratumoral VETC/PD-L1 index		<0.001*			0.023*			<0.001*			0.111			0.003*			0.002
I	Ref.		Ref.				Ref.			Ref.			Ref.			Ref.
II	0.201 (0.109–0.372)		0.265 (0.101–0.700)				0.261 (0.145–0.469)			0.431 (0.163–1.142)			0.233 (0.095–0.570)			0.261 (0.106–0.644)
III	0.755 (0.496–1.149)		0.924 (0.556–1.595)				0.782 (0.508–1.204)			1.206 (0.716–2.031)			1.058 (0.627–1.786)			1.370 (0.775–2.420)

*, P value showed statistical significant. CA-199, carbohydrate antigen 199; TNM, tumor-nodes-metastases; Ref., reference; HR, hazard ratio; CI, confidential interval; NA, not adopted; NS, not significant; TTR, time to relapse; OS, overall survival.

The construction and validation of the prognostic nomogram

Multivariable models were constructed by appropriate categories for all variables simultaneously. Based on our findings of both univariate and multivariate analysis, CA-199 level combined with lymph node metastasis and intra-tumoral VETC/PD-L1 index were subsequently used to build a corresponding nomogram for the prediction of OS (Figure 3A,B) at 3, 5 years after surgery. The total score of each nomogram indicates that the hierarchical prediction of patient prognosis is more accurate. For each nomogram, the predicted cumulative incidence of 3 or 5 years was compared to the observed actual incidence of 3 or 5 years, which showed a good calibration (Figure 3C,D). In addition, we found that the corresponding C-index for this specific OS nomogram was 0.718 [95% confidence interval (CI): 0.640–0.797] in the present study, which was better than those in TNM [the AJCC 7th edition Cancer Staging (27), C-index: 0.594, 95% CI: 0.55–0.638] (28), LCSGJ (C-index: 0.605, 95% CI: 0.561–0.649) (27), Nathan (C-index: 0.588, 95% CI: 0.543–0.633) (29) and Okabayashi staging systems (C-index: 0.594, 95% CI: 0.55–0.638) (30).

Figure 3.

ICC OS nomogram and calibration curve analysis. (A,B) The ICC OS and TTR nomogram comprising CA-199, lymph node metastasis and VETC/PD-L1 index in training cohort (n=214); (C,D) The calibration curves for predicting three-year and five-year OS in training cohort (n=214). ICC, intrahepatic cholangiocarcinoma; VETC, vessels encapsulate tumor clusters; OS, overall survival.

To validate the prognostic value in the validation cohort, nomograms constructed with similar features from training set were further used to predict the probability of OS (Figure S5A,B), with a corresponding C-index (0.691, 95% CI: 0.583–0.800) for this specific OS nomogram. The 3- and 5-year survival rates indicated by nomogram suitable well with this predicted model. Our findings indicated that a good concordance between predicted and observed survival probabilities were constructed through a favorable nomogram.

Discussion

ICC is an uncommon malignant tumor with a unfavorable prognosis due to an poor understanding of its molecular pathogenesis, the insufficient benefits of standard chemotherapy, and no optimal biomarkers used clinically to predict prognosis (31). Since the dismal prognosis in ICC, optimal predictors to sub-group ICC patients were significantly indeed.

It remained unknown that the vascularization pattern in intra-tumoral area could predict clinical benefit. VETC phenotype was defined as a novel peculiar vascular pattern with a common feature, that tumor nest were surrounded with dilated sinusoid-like structures (7). Previous studies indicated that VETC was not only tumor-riched in HCC, but also in follicular thyroid cancer and renal cell cancer (32). Recently, a multi-center cohort of HCC cases from different countries illustrated the universality of this vascularization pattern, and VETC⁺ phenotype was defined as an independent predictor for dismal OS and elevated recurrence (7). In addition, VETC pattern may present an effective transfer model through the promotion of tumor clusters releasing (33).

Furthermore, VETC phenotype may act as a novel indication for HCC patients with sorafenib applying (8). Consistently, recent studies suggest that sorafenib monotherapy may show promising anticancer activity in patients with advanced ICC with controllable toxicity (34). According to the results from the case-control study, we had shown that VETC⁺ in ICC was ubiquitous at different clinical stages and accounted to 57.9–70%. To date, our present research was the first attempt to evaluate the clinical association of VETC phenotype in ICC. Consistent with previous studies in HCC, present study further confirmed the prognostic significance of VETC phenotype, as a robust prognostic parameter discriminating aggressive ICC. Our findings remarkably indicated that this VETC phenotype may promote malignant tumor progression of ICC.

Recently, combination therapy has made great achievements in the application of tumor therapies (35). Targeting and immunotherapy (such as PD-1 antibody) have played an essential role in HCC therapy (36). However, only about 5% of ICC patients were microsatellite unstable, which were sensitive to PD-1 antibody (37). Whether the combination therapy, such as PD-1 antibody combined with targeted drugs, can achieve the effect similar to HCC, is unknown. At present, only very preliminary, but not conclusive evidence can be found in some phase II clinical trials, and further exploration is needed.

Notably, we simultaneously evaluated the heterogeneous PD-L1 expression profile in ICC. Previously, a large cohort of epidemiological data indicated that HBV infection (38), which might result in chronic liver inflammation, immune imbalance in ICC tumorigenesis. Our results revealed that hyper-activated PD-L1 expression in intra-tumor area was positively associated with HBV infection. Furthermore, elevated PD-L1 status in intra-tumor area had dismal prognosis than those low. Our findings illustrated that the combination of amplified PD-L1 signals and HBV infection in intra-tumor areas might play an essential role in the malignant tumor progression of ICC.

Furthermore, to illustrate the clinical value of VETC/PD-L1 index, an integrated nomogram combined with VETC and PD-L1 for OS was constructed, indicating a better prognostic performance. Values of nomograms in predicting prognosis are drawing emerging attentions in many malignancies, such as ICC (39) and HCC (40). In present research, the C-index established through our integrated nomogram was significantly better than these traditional systems, and a good concordance between predicted and observed survival probabilities was also found.

Several limitations were presented in the present research. Our findings were only made through IHC evaluations. Hence, more researches are indeed needed to uncover the potential mechanism of VETC/PD-L1 index in promoting ICC malignant progression. Moreover, the prognostic value of VETC/PD-L1 index needs further validation in prospective studies.

Understanding the key mechanisms of tumor metastasis is of great significance for tumor treatment. The comprehensive evaluation of tumor vascularization pattern, micro-environmental profile and its potential mechanisms not only offers novel ideas for the progression of anti-tumor therapy but also provides a specific theoretical basis for ICC patients. Nonetheless, further trials focused on the effects of sequential or combination therapy in ICC are warranted.

In summary, this present research illustrated that VETC⁺ phenotype and elevated PD-L1 expression in ICC significantly associated with malignant characteristics and dismal survival. In addition, an integrated nomogram combined with VETC and PD-L1 for OS showed a better prognostic value for ICC patients than these traditional systems. The clinical significance of VETC/PD-L1 index ensured it a promising indicator of future risk stratification and customized therapy strategies.

Figure S1.

Transcriptome sequencing data of CD34 and PD-L1 in normal and tumor tissues from TCGA database. The mRNA level of CD34 (A) and PD-L1 (B) in pan-cancer cohort and cholangiocarcinoma (CHOL) cohort (Red) from TCGA database. BLCA, bladder urothelial Carcinoma; BRCA, breast invasive carcinoma; CHOL, cholangiocarcinoma; COAD, colon adenocarcinoma; ESCA, esophageal carcinoma; HNSC, head and neck squamous cell carcinoma; KICH, kidney chromophobe; KIRC, kidney renal clear cell carcinoma; KIRP, kidney renal papillary cell carcinoma; LIHC, liver hepatocellular carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; PAAD, pancreatic adenocarcinoma; PRAD, prostate adenocarcinoma; READ, rectum adenocarcinoma; STAD, stomach adenocarcinoma; THCA, thyroid carcinoma.

Figure S2.

Correlation between CD34 and PD-L1 mRNA expression in pan-cancer cohort (A) and cholangiocarcinoma (CHOL) cohort (B).

Figure S3.

Clinical implications of VETC and PD-L1 in ICC in validation and external validation cohorts. (A) Kaplan-Meier curves for OS and TTR based on VETC pattern invalidation cohort (n=108); (B) Kaplan-Meier curves for OS and TTR based on VETC pattern in external validation cohort (n=90); (C) Kaplan-Meier curves for OS and TTR based on PD-L1 expression invalidation cohort (n=108); (D) Kaplan-Meier curves for OS and TTR based on PD-L1 expression in external validation cohort (n=90). VETC, vessels encapsulate tumor clusters; ICC, intrahepatic cholangiocarcinoma; TTR, time to relapse; OS, overall survival.

Figure S4.

Clinical implications of VETC and PD-L1 in ICC. Kaplan-Meier curves for OS and TTR based on combined VETC pattern and PD-L1 expression in validation (n=108) and external validation cohorts (n=90). VETC, vessels encapsulate tumor clusters; ICC, intrahepatic cholangiocarcinoma; TTR, time to relapse; OS, overall survival.

Figure S5.

ICC OS nomogram and calibration curve analysis. The calibration curves for predicting 3-year (A) and 5-year (B) OS in validation cohort (n=108). ICC, intrahepatic cholangiocarcinoma; OS, overall survival.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Ethical approval was obtained from the research ethics committee of Zhongshan Hospital, Fudan University, and written informed consent was obtained from each patient.