Fresh paired HCC tissues and peripheral blood samples were obtained from 56 patients receiving hepatectomy at Zhongshan Hospital of Fudan University (Shanghai, China) from Jan. to Sep. 2015. TMAs were constructed using specimens from two independent cohorts of HCC patients (n = 358 and n = 254) who underwent primary resection in 2006 and 2007 respectively. Tissue cores of 1 mm diameter were selected from high immune cells infiltrated region in paraffin-embedded tumor and peri-tumor tissue based on the HE staining. Sections (4 μm) were cut and applied to APES- coated slides as previously described [4]. Patient informed consent was obtained and the study was approved by the institutional ethics committee. Tumor differentiation was graded by the Edmondson grading system [4]. None of the patients received antitumor or immunosuppressive treatments before surgery.

Multiplex immunohistochemistry (mIHC) was performed according to manufacturer’s instruction (PerkinElmer, Opal® Kit). The process was performed on the following antibodies and fluorescent dyes in the flowing order: CD3/Opal570, CD8/Opal690, PD1/Opal520, TIM3/Opal650. Detailed procedures of mIHC and quantitative analysis are presented in Additional file 1. Slides were scanned and imaged using the PerkinElmer Vectra3® platform and were analyzed in batches using PerkinElmer inform and R script for quantification of positively stained cells.

Peripheral mononuclear leukocytes were isolated by Lymphoprep (STEMCELL Technologies) density gradient centrifugation, and fresh tissue-infiltrating mononuclear leukocytes were obtained as described previously [16]. Detailed procedures of isolation of mononuclear cells are presented in Additional file 1.

Peripheral blood leukocytes as well as tumor- and adjacent non-tumor liver tissue-infiltrating leukocytes were stained with fluorochrome-conjugated antibodies against CD3, CD4, CD8, PD1, and TIM3, or control antibodies to identify the exhausted T cells in a flow cytometry analyzer (BD LSR Fortessa, BD, CA). Data were analyzed by FlowJo software (v9.3.2; TreeStar, USA). PD1-positivity among CD8⁺ T cells was defined based on isotype antibody control and the separation of PD1-high (PD1^Hi) from PD1-intermediate (PD1^int) among PD1⁺ CD8⁺T cells was based on mean fluorescence intensity (MFI) and the expression of TIM 3 [17, 18]. Details of the fluorochrome-conjugated antibodies and isotype controls are listed in Additional file 2: Table S1.

Statistical analysis was performed with the R software, SPSS (v22, IBM, Armonk, NY) and Prism 6.0 (GraphPad Soft Inc., SanDiego, CA) softwares. Comparisons were performed using Student’s t-test, ANOVA test or Chi-square test as appropriate. Kaplan–Meier curves were used and estimated by the log-rank test. Patients were classified as ‘low’ and ‘high’ groups according to the Youden index to achieve the optimal cutoffs. The number of nearest neighbor was calculated on the assigned coordinates of each cell and performed by R software using the spatstat package. Multivariate analysis was performed by Cox regression analysis. Two-sided P <  0.05 considered as statistically significant.

Previously, PD1 and TIM3 expression were found to be upregulated in tumor-infiltrating lymphocytes (TILs) of tumor-bearing mice and HCC patients [15]. Co-expression of PD1 and TIM3 on CD8⁺ T cells was also demonstrated to be associated with T cell exhaustion in melanoma and non–small cell lung carcinoma [19, 20]. Herein, to assess the potential role of PD1 and TIM3 in HCC immunopathology, we first examined the expression of PD1 and TIM3 on CD8⁺ T cells from 30 HCC patients including paired peripheral blood, peri-tumor and tumor tissues by flow cytometry. As shown in Fig. 1a, PD1⁺ CD8⁺T cells were readily identified in blood and liver tissues. In addition, PD1⁺ cells can be further subdivided into PD1-high (PD1^Hi) and PD1-intermediate (PD1^int) and TIM3 expression was limited to PD1^Hi CD8⁺T cells. Consequently, based on the TIM3 expression, PD1^Hi CD8⁺ TILs could be also subdivided into TIM3⁻PD1^Hi and TIM3⁺PD1^Hi CD8⁺ TILs.Fig1Fig. 1

PD1 and TIM3 expression on HCC infiltrating CD8⁺ cytotoxic T cells. a, Representative flow cytometric plots to show the expression of PD1 and TIM3 on CD8⁺ cytotoxic T cells from paired blood, peri-tumor and tumor tissues of HCC patients. b-d, Comparison of the frequencies of PD1⁺ among CD8⁺ T cells (b), PD1^Hi among CD8⁺ T cells (c) and PD1^Hi among CD8⁺PD1⁺ T cells (d) across paired blood, peri-tumor and tumor tissues of HCC patients (n = 30). e-f, Comparison of the frequencies of PD1^Hi among CD8⁺PD1⁺ T cells in relation to tumor stages (E, 11 stage I and 19 stage II and III) and tumor size (F, 8 tumor size≤5 cm and 19 tumor size> 5 cm) across paired blood, peri-tumor and tumor tissues of 11 stage I and 19 stage II and III HCC patients; and paired blood, peri-tumor tissue and tumor tissue from 8 tumor size≤5 cm and 19 tumor size> 5 cm HCC patients. Error bars indicated median with interquartile range. Significance was assessed by Wilcoxon matched-pairs signed rank test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001

To better understand the phenotypic characteristics and functional status of PD1^Hi, PD1^Int and PD1⁻ CD8⁺ TILs in HCC, we performed a comprehensive characterization of these cells by flow cytometry, including 64 cluster of differentiation (CD) surface markers, 24 transcription factors, 18 chemokine receptors and 10 cytokine receptors. The differentially expressed makers among PD1^Hi, PD1^Int, and PD1⁻ CD8⁺ TILs were displayed in the heat map (Fig. 2a) and selected markers were presented as overlaid histograms (Fig. 2b-f and Additional file 4: Figure S3 and Additional file 5: Figure S4).Fig2Fig. 2

Phenotypic characteristics of tumor infiltrating PD1^Hi CD8⁺ T cells in HCC. a, A heatmap to show the global phenotypic characteristics of PD1^Hi, PD1^Int and PD1⁻ CD8⁺T cells from HCC tumor tissues detected by the flow cytometry. Data represent three individual patients. b-f, Representative flow cytometric overlays of different markers expressed by tumor infiltrating PD1^Hi (red line), PD1^Int (blue line) and PD1⁻ (black line) CD8⁺T cells, including co-inhibitory receptors (b), exhaustion associated transcription factors (c), cytotoxic molecules (d), differentiation and activation markers (e-f)

Next, we investigated the cytokine-producing capacity of CD8⁺ TILs based on PD1 expression. CD8⁺ TILs from 9 HCC patients were stimulated by PMA and ionomycin in the presence of Brefeldin A, followed by cytokine measurement. As is known, IL-2 is the first compromised cytokine after T cell exhaustion [27]. We observed that the frequency of IL-2-producing PD1^Hi CD8⁺ T cells (median = 2.89, 1.11–5.88%) was 10–15 times fewer than those of PD1^Int (median = 44.56, 36.54–62.20%; P <  0.0001) and PD1⁻ CD8⁺ T cells (median = 30.21, 21.24–43.27%; P < 0.0001) (Fig. 3a and b). Furthermore, PD1^Hi CD8⁺ T cells exhibited defective production of IFN-γ and TNF-α (median = 35.09, 7.44–67.41%; median = 13.47, 2.12–30.58%), which are typical Th1 cytokines essential for effective anti-tumor responses, in comparison with the PD1^Int (median = 77.01, 61.25–82.34%; P = 0.0059; median = 70.67, 60.70–80.60%; P = 0.002) and PD1⁻ CD8⁺ T cell (median = 71.74, 49.70–90.09%; P = 0.002; median = 58.74, 54.47–80.20%; P = 0.002). Meanwhile, we also found that PD1^Hi CD8⁺ TILs produced much less IL-4, IL-17A, IL-22 and GM-CSF than PD1^Int and PD1⁻ CD8⁺ TILs (Fig. 3a and b). IL-10 is a potent immune-suppressive cytokine and contributes to the induction of B7-H1 (PDL1) on monocytes [12, 13]. Notably, the frequency of IL-10-producing cells was elevated to 2.57% (0.69–4.52%) in PD1^Hi CD8⁺ T cells, which was significantly higher than PD1^Int (median = 0.7, 0.26–1.55%; P = 0.002) and PD1⁻ CD8⁺ TILs (median = 0.1, 0–0.21%; P = 0.002) (Fig. 3a and b).Fig3Fig. 3

Assessment of pro/anti-inflammatory cytokines profiling and cytotoxic potential of tumor infiltrating PD1^Hi CD8⁺ T cells. a-b, Representative flow cytometric plots (a) and accumulated data (b) to show the pro-inflammatory cytokines IFN-γ, IL-2, TNF-α, GM-CSF, IL-4, IL-22, IL-17a and anti-inflammatory IL-10 secreting profile of tumor infiltrating PD1^Hi, PD1^Int and PD1⁻ CD8⁺ T cells following the stimulation of PMA, ionomycin and BFA for 5 h (n = 9). c, Representative flow cytometric overlays of intracellular Granzyme B, perforin and CD107a expression of tumor infiltrating PD1^Hi, PD1^Int, and PD1⁻ CD8⁺ T cells. Granzyme B and perforin were detected from fresh samples (n = 11) and CD107a expression was measured after overnight stimulation of coated anti-CD3 (10 μg/mL) and soluble anti-CD28 mAb (1 μg/mL) (n = 8). d, Apoptosis quantification of HCCLM3 tumor cell line after co-culture with anti-CD3 (10 μg/mL)/CD28 mAb (1 μg/mL) stimulated CD8⁺ T cells subsets for 18 h and pooled statistics of viable HCCLM3 tumor cells shown in bar graph (n = 6). Significance was assessed by Wilcoxon matched-pairs signed rank test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001

Subsequently, we applied multispectral staining to depict CD8⁺ TILs within tumor and peri-tumor of HCC patients (Additional file 8: Figure S6A). Analogous to the flow cytometry, we could classify CD8⁺ TILs into 3 distinct subsets as PD1^Hi, PD1^Int, and PD1⁻ based on the PD1 expression. Furthermore, PD1^Hi CD8⁺ TILs can be subdivided into TIM3⁺PD1^Hi and TIM3⁻PD1^Hi when TIM3 was added (Fig. 4a). As expected, we demonstrated the distinct classification of these four CD8⁺ T cells subpopulation by t-SNE (Fig. 4b). Then, we applied this panel on TMA comprising tumor and peri-tumor from two independent cohorts of 358 and 254 HCC patients as training and validation sets, respectively (Additional file 8: Figure S6B). PD1 ‘high’ and TIM3 ‘positive’ thresholds were identified with mean pixel intensity plotting, gated on CD3⁺CD8⁺ T cells (Additional file 8: Figure S6C). The clinicopathologic characteristics of two cohorts were detailed in Additional file 9: Table S2. No significant differences in clinicopathologic features were observed between the two cohorts. The 1, 3, 5-year Overall Survival (OS) rates were 93%, 64%, 40% and 83%, 59%, 51% for the training and validation cohorts, respectively.Fig4Fig. 4

Increased infiltration of exhausted PD1^Hi CD8⁺ T cells in HCC tumor tissues revealed by multiplex immunohistochemistry. a, Representative multiplex immunofluorescence images to show the distribution of CD8⁺ T cell subsets in peri-tumor and tumor: red arrows (CD8⁺TIM3⁺PD1^Hi), purple arrow (CD8⁺TIM3⁻PD1^Hi), green arrow (CD8⁺PD1^Int) and cyan arrow (CD8⁺PD1⁻). Tissue slides were stained by TSA method and scanned at 20x by Vectra3.0 Automated Imaging System. Scale bar, 50 μm. b, t-SNE analysis of CD8⁺ T cells from paired peri-tumor and tumor tissues validated the distinct classification of four CD8⁺ T cell subsets. c, Relative distribution analysis of total CD3⁺ T cells in paired peri-tumor and tumor tissues by dividing the total CD3⁺ T cells into CD8⁺ (cytotoxic T cells) and other cells (upper panel). CD8⁺ T cells were further split into CD8⁺PD1⁻, CD8⁺PD1^Int and CD8⁺PD1^Hi subpopulations (middle panel); and finally CD8⁺PD1⁺ T cells were separated into CD8⁺TIM3⁺PD1^Hi and CD8⁺TIM3⁻PD1^Int basing on the TIM3 expression (lower panel). d-h, Comparisons of the proportions of CD8⁺ among CD3⁺ T cells (d) and CD8⁺PD1^Hi (e), CD8⁺PD1^Int (f), CD8⁺TIM3⁺PD1^Hi (g) and CD8⁺TIM3⁻PD1^Hi (h) among CD8⁺PD1⁺ T cells between paired peri-tumor and tumor tissues in training cohort (n = 358). Error bars indicated median with interquartile range. Significance was assessed by Wilcoxon matched-pairs signed rank test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001

We then sought to define whether the infiltration of PD1^Hi CD8⁺ T cells in HCC was related to patient survival. We calculated the percentages of defined subpopulations within total CD3⁺ T cells, CD8⁺ T cells, and CD8⁺PD1⁺ T cells successively for each patient. Then, patients were stratified into high and low groups according to the highest Youden index to achieve the optimal cutoffs of each T cell subpopulation. We found that a high proportion of CD8⁺ T cells within CD3⁺ T cells in tumor was associated with prolonged OS (P = 0.019) but not relapse-free survival (RFS, P = 0.129) (Fig. 5a and b). However, high proportions of PD1⁺ cells among CD8⁺ T cells neither correlated with patients’ OS (P = 0.067) nor RFS (P = 0.693) (Fig. 5a and b). However, patients with a high proportion of CD8⁺PD1^Hi within CD8⁺PD1⁺ correlated with a significantly poor OS (P = 0.004) and RFS (P = 0.007). Moreover, a higher proportion of CD8⁺TIM3⁺PD1^Hi within CD8⁺PD1⁺ showed more striking tendency on OS (P = 0.002) and RFS (P < 0.0001) than the proportion of CD8⁺PD1^Hi within CD8⁺PD1⁺ TILs. In contrast, we observed that a high proportion of CD8⁺PD1^Int within CD8⁺PD1⁺ were more likely to have better OS (P = 0.004) and RFS (P = 0.007) (Fig. 5c-f). Prognostic significance of these TIL subgroups in the validation cohort was similar with that in the training cohort (Additional file 10: Figure S7A-D).Fig5Fig. 5

Prognostic significance of the subsets of CD8⁺ TILs in the training cohort. a-f, Kaplan-Meier analysis of overall survival (OS) (a, c, e) and relapse free survival (RFS) (b, d, f) in HCC tumors according to the proportion of CD8⁺ among CD3⁺ TILs and CD8⁺PD1⁺ among CD8⁺ TILs (a and b), CD8⁺PD1^Int and CD8⁺PD1^Hi among CD8⁺PD1⁺ TILs (c and d) and CD8⁺TIM3⁻PD1^Hi and CD8⁺TIM3⁺PD1^Hi among CD8⁺PD1⁺ TILs (e and f) in the TMA training cohort (n = 358)

Previously, it has been shown that PD-L1 and Galentin9, the ligands of PD1 and TIM3 respectively, were primarily expressed on tumor cells and CD68⁺ tumor associated macrophages (TAMs) in HCC that promoted immune escape [14, 28]. Strikingly, we found that the proportions of TIM3⁺PD1^Hi CD8⁺ TILs were positively correlated with the frequency of PD-L1⁺ TAMs (r = 0.4121; P < 0.0001) (Fig. 6a). By contrast, the proportions of both TIM3⁻PD1^Hi and PD1^Int CD8⁺ TILs showed negative correlations with PD-L1⁺ TAMs (r = − 0.1792; P = 0.0007; r = − 0.2551; P < 0.0001; respectively) (Fig. 6b and c). Moreover, PD-L1⁺ tumor cells (PDL1⁺CD68⁻) were weakly or not correlated with these three T cell subsets (Additional file 10: Figure S7E). We thus speculated that PD-L1⁺ TAMs, but not PDL1⁺ tumor cells, were positioned near to specific CD8⁺ T cell subsets in order to exert an inhibitory effect. Then we performed spatial analysis and calculated the relative number of PD-L1⁺ TAMs from each PD1^Int and TIM3⁺PD1^Hi CD8⁺ TIL within a series of distances from 20 to 50 μm (Fig. 6d). We found that in all distances studied, significantly higher numbers of PD-L1⁺ TAMs were around TIM3⁺PD1^Hi than those around PD1^Int CD8⁺ TILs (Fig. 6e). Moreover, the densities of PD-L1⁺ TAMs were significantly higher within 20 μm of TIM3⁺PD1^Hi TILs in comparison with those above 20 μm (Fig. 6e). Together, our results suggested that PD-L1⁺ TAMs may intimately interact with TIM3⁺PD1^Hi CD8⁺ TILs in situ which could jointly dampen the effective anti-tumor immune responses.Fig6Fig. 6

Association of tumor-associated macrophages infiltration with PD-1^Int and TIM3⁺PD1^Hi CD8⁺ TILs. a-c, Correlation analysis between the proportion of tumor infiltrating CD8⁺TIM3⁺PD1^Hi (a), CD8⁺TIM3⁻PD1^Hi (b) and CD8⁺PD1^Int (c) among CD8⁺PD1⁺ TILs and the proportion of PDL1⁺ TAMs within CD68⁺ macrophages per core respectively. Correlation is evaluated by the Spearman correlation coefficient. d, Representative multiplex immune-fluorescence image to show the staining for CD8 (yellow), PD1 (green), TIM3 (red), CD68 (cyan), PDL1 (magenta) in the HCC tumor tissue. Cellular phenotype of the fluorescence image depicted the spatial location of CD68⁺PDL1⁺ (blue dots), CD8⁺TIM3⁺PD1^Hi (red dots), CD8⁺PD1^Int (green dots) in the situ tumor tissue. Solid plots and dash line connected the nearest cells within 20 μm from the CD8⁺TIM3⁺PD1^Hi and CD8⁺PD1^Int to the CD68⁺PDL1⁺ respectively. Scale bar, 200 μm. (e) The infiltrating density of PDL1⁺ TAMs within the indicated hierarchy distances of CD8⁺TIM3⁺PD1^Hi and CD8⁺PD1^Int in the HCC tumor tissues, respectively. Error bars indicated median with interquartile range. Significance was assessed by Wilcoxon matched-pairs signed rank test. ****, P < 0.0001. TAMs: tumor associated macrophages