Intrahepatic cholangiocarcinoma (ICC) is the second most common primary liver malignancy [1]. The survival rates of ICC remain stagnant, despite great progresses have been made on the molecular basis, diagnosis and treatment modalities [1, 2]. Surgical resection offers the only chance to cure, but most ICC patients are diagnosed at advanced clinical stages when only palliative treatments can be performed [1]. Some of these palliative treatments are proved to be effective, but their prolongations of survival are still unsatisfactory [1, 3, 4]. Consequently, therapeutic targets that can significantly improve the survival of ICC are urgently needed.

Cancer cells can express immune-inhibitory molecules innately or adaptively to evade immune attacks from the hosts [5]. Recently, immunotherapies that normalize immune responses in the tumor microenvironment (TME), particularly through targeting the program cell death (PD) pathway, have been proved to achieve high objective response rates in several refractory malignancies [5–7]. To date, anti-PD therapy has been approved by FDA with more than 10 cancer indications, and PD-L1 expression level in tumor samples is an important biomarker to predict treatment responses of anti-PD therapy [8].

Although the effect of anti-PD therapy in biliary tract cancers remains scarcely reported, a recent case report showed that PD-1 inhibitor pembrolizumab brought strong and durable control to an advanced cholangiocarcinoma case [9]. Moreover, previous studies reported that PD-L1 expression rates ranged from 17.7 to 72.2% in different ICC cohorts and T cell infiltrates were found in majority of ICC samples [10–12]. These results altogether suggest that ICC is very likely to benefit from immunotherapies that normalize the TME. However, previous studies evaluating PD-L1 expression levels in ICC were different in materials, sample sizes and scoring systems. Gani reported that 39 out of 54 (72.2%) ICC cases were PD-L1 positive on cells within tumor front, whereas other studies mainly evaluated PD-L1 expression within tumor area and reported much lower expression rates ranging from 17.7 to 29.8% [10, 12]. Biomarkers that predict treatment responses towards anti-PD therapy are not limited to PD-L1 expression levels in tumor samples [8]. A PD-L1 negative cholangiocarcinoma with high-level microsatellite instability (MSI-H) were proved to be sensitive to anti-PD-1 therapy, which indicated that MSI status may also serve as a predictive biomarker for anti-PD therapy in ICC [9]. To our disappointment, the incidence of MSI-H in cholangiocarcinoma is also extremely low (1.3%, 4/308). Therefore, expression levels of other immune checkpoints are worth exploration to offer more immunotherapeutic choices for ICC patients who have been excluded from anti-PD therapy [9, 13].

HHLA2, short for HERV-HLTR-associating 2, is a newly identified immune checkpoint belonging to the B7 family [14]. Compared to other members of the B7 family, HHLA2 possesses some unique features: as a type I transmembrane molecule, it has three extracellular Ig domains [15]; it is constitutively expressed on peripheral monocytes and is inducible on B cells rather than on T cells [14, 15]. HHLA2 has been evidenced to inhibit TCR mediated proliferation and cytokine production of CD4+ and CD8+ T cells [14, 16]. Receptors of HHLA2 exist on a wide range of immune cells, such as T cells, monocytes and B cells, as well as on endothelial cells [15]. Most of these receptors remain unidentified, except for TMIGD2, a CD28 family member which is expressed on endothelial cells and may be a participant in angiogenesis [17, 18].

The expression pattern and clinical relevance of HHLA2 have so far been studied in detail in triple negative breast cancer (TNBC), osteosarcoma and non-small-cell lung cancer (NSCLC) [16, 17, 19, 20]. In TNBC and osteosarcoma, the high expression rates of HHLA2 were 56% (28/50) and 68% (42/62), respectively. Moreover, patients with high HHLA2 expression were more likely to have advanced and metastatic diseases [17, 19]. In NSCLC, 66% cases were HHLA2 positive (413/625). HHLA2 expression was associated with EGFR mutation status and high tumor infiltrating lymphocytes (TILs) density [20]. Moreover, HHLA2 was widely expressed in PD-L1 negative NSCLC samples [16]. These literature altogether indicated that HHLA2 was potentially involved in cancer progression through immune inhibition and could be a promising target next to PD-L1 for cancer immunotherapy [17, 19].

So far as we know, the expression pattern of HHLA2 and its association with PD-L1 expression in ICC remain unclear. In the present study, we performed a comprehensive comparison between HHLA2 and PD-L1 in terms of the expression pattern, clinical relevance and their associations with tumor infiltrating CD3+, CD8+, CD4 + Foxp3+, CD68+, CD163+ and CD20+ immune cells in a ICC cohort after curative resection.

In this study, we observed that HHLA2, a newly identified B7 family member, was more prevalent than PD-L1 in ICC and HHLA2 overexpression was common in PD-L1 negative ICC. Moreover, HHLA2 was identified as an independent prognostic indicator for OS in two independent cohorts whereas PD-L1 showed no significant prognostic value. The presence of PD-L1 and HHLA2 was associated with differential infiltration of immune cells: PD-L1 positive tumors were observed to have higher densities of CD3 + TILs, CD68+ TAMs and CD163+ TAMs, whereas HHLA2 overexpression was significantly correlated with sparser CD3+ TILs, CD8+ TILs and a higher CD4 + Foxp3+/CD8+ TIL ratio.

HHLA2 was previously observed with expression rates ranging from zero to 70% in a variety of cancer types [17]. The high expression rates of HHLA2 in the two independent ICC cohorts were 49.0 and 67.7% respectively, which was on a medium to high end of the expression spectrum, higher than previous data on gall bladder cancers (0/10) and liver cancers (4/10) [17]. We found that HHLA2 expression was positively associated with serum CEA and CA19–9 levels in the training cohort as well as the presence of LN metastasis, serum CEA and clinical stage in the validation cohort. In addition, in the validation cohort which was featured by more LN metastases, HHLA2 expression was more frequently observed. These data were in line with previous findings in TNBC and osteosarcoma, where HHLA2 expression was found to be more frequent in patients with LN metastasis and metastatic lesions [17, 19]. Moreover, we found that HHLA2 expression strongly predicted poor OS in the two independent ICC cohorts. The basis for these results may, on one hand, attributed to the known co-inhibitory roles of HHLA2 on T cells [14]. On the other hand, the positive correlations between HHLA2 expression and serum CEA, CA19–9 levels, two tumor biomarkers that generally reflect tumor burden both at primary sites and in the circulation for ICC, indicating that patients with HHLA2 overexpression are more likely to suffer from recurrence and metastasis [29]. Along with previous findings, in ICC, HHLA2 may not only work as an inhibitory checkpoint, but also potentially contribute to tumor progression through binding to TMIGD2, a recently identified ligand which is also involved in angiogenesis, or through other unknown mechanisms [18]. Therefore, targeting HHLA2 may inhibit cancer dissemination through immune-independent pathways and the underlying molecular basis requires further investigation.

Our study reported a PD-L1 positive rate of 28.1 and 17.1% on TC and IC, respectively. These results were similar to a previous report from another large Asian ICC cohort (n = 192), in which 17.7% of the cases were found to be PD-L1 positive [12]. Intriguingly, the prognostic significances of PD-L1 were inconsistent among different studies [10–12]. Studies of Gani et al. and Sabbatino et al. observed that PD-L1 positive was associated with unfavorable survival, but Zhu et al. reported opposite results [10–12]. In our study, PD-L1 expression had no significant correlation with survival. Moreover, we identified that PD-L1 positive cases had prominent T cell infiltration, which was similar to findings from Zhu et al. in ICC and the report from Schalper et al. concerning lung cancer [30]. The infiltration of T cells, especially CD8+ T cells, is a known factor that indicates favorable survival and this finding was also confirmed in our cohort (Additional file 6: Table S6). Although PD-L1 expression, based on the molecular basis, is speculated to have a negative impact on survival, whereas its positive correlation with T cell infiltration mainly exert an opposing impact and consequently leads to the uncertainty of prognostic significance [24]. However, the discrepant prognostic significances of PD-L1 may not be a limitation for its possible role as a therapeutic target and a predictive biomarker for treatment response [30].

Previous studies reported discrepant results among different cancer types on the correlations between immune cell infiltration and HHLA2 expression. In NSCLC, HHLA2 expression was independently correlated with high TIL infiltration [20]. Whereas for osteosarcoma, no significant correlation was found between the presence of TILs and HHLA2 expression [19]. In this study, we evaluated the immune cells in a more detailed manner where the exact counts of stained immune cells of 5 high power field were calculated [24]. We observed that HHLA2 overexpression was conversely associated with T cell and CTL infiltration in ICC. Although no significant correlation was found between HHLA2 expression and the counts of Tregs, patients with high HHLA2 expression were found to have higher ratio of CTLs to Tregs within tumor area. Collectively speaking, the expression of HHLA2 was correlated with an inhibitory TME featured by decreased T cells, CTLs and imbalance between Tregs and CTLs, which indicated the possible role of HHLA2 as an immunotherapeutic target.

HHLA2-high patients and PD-L1 positive cases showed significant difference in the infiltrating patterns of immune cells. HHLA2 was associated with fewer CTLs and higher ratio of Treg to CTLs, whereas the presence of PD-L1 was accompanied by prominent T cells and CD163+ TAMs. This phenomenon is intricate but explainable. The positive correlation between T cell infiltration and PD-L1 expression on TC may indicate that, in ICC, PD-L1 is mainly adaptively expressed in tumor immune response, perhaps through the well-established IFN-γdependent manner [31]. CD163+ TAMs mainly represent M2-like TAMs which contribute to immune suppression [32]. TAMs have been observed to express PD-L1 and induce immune evasion of cancer through autocrine CXCL8 [33]. Moreover, targeting PD-L1 can polarize macrophages to a more pro-inflammatory phenotype [34]. Therefore, the significant infiltration of CD163+ TAMs in PD-L1 IC- and TC- positive ICC may also be a consequence of immune reactions to tumor. On the other hand, concerning the high expression rate of HHLA2 and its correlation with sparser T cell infiltration, HHLA2 is more likely to be innately expressed in ICC and thus limiting further infiltration of active T cells. Although HHLA2 can be induced on B cells and monocytes by IFN-γ and LPS, the mechanism of how tumor cells enhance HHLA2 expression remains to be further investigated [14, 17].

Several limitations concerning this study were underlined as follows. Firstly, both training and validation cohort were derived from a single institution in China, therefore the expression pattern of HHLA2 of ICC patients in other ethnic groups has yet to be investigated. Secondly, limited to the number of TMAs, PD-L1 expression as well as the correlations between typical subsets of tumor infiltrating immune cells and PD-L1, HHLA2 expression were only studied in our training cohort. Therefore, validation on the immune infiltration features of ICC with PD-L1, HHLA2 expression in another independent ICC cohort as well as further investigation on its molecular basis will provide more solid evidence.

In summary, HHLA2 is more commonly expressed and possesses more significant prognostic value compared with PD-L1 in ICC cases. High HHLA2 is frequent in PD-L1 negative cases and is associated with fewer infiltrating CTLs as well as more Tregs in proportion, suggesting that HHLA2 may represent an ideal target next to PD-L1 for immunotherapy of ICC.