Composition of T cells in HCC

To generate a comprehensive view of the infiltrated T-cell ecosystem of HCC, we applied scRNA on 8047 sorted CD3⁺CD45RO⁺ cells from the HCC adjacent non-cancerous tissue (ANT) and tumor core (TM). CD8A and IL-7R were used as markers to classify CD8⁺ or CD4⁺ T cells. To reveal the intrinsic structure and potential functional subtypes of the overall T-cell populations, we performed unsupervised clustering of all cells using spectral clustering. For the integrated data, a total of eight stable clusters emerged, including three clusters for CD8⁺ cells and three clusters for CD4⁺ cells (figure 1A). We named each cluster and compared gene expression in figure 1A–1C. The percentage of each cluster is listed in figure 1B. We compared the expression of presentative genes coding co-inhibitory receptors, effector molecules, and other markers in figure 1C, online supplementary figure S1 and table S3.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Clustering HCC-infiltrating T cells based on single-cell gene expression. (A) 2D visualization of single-cell clusters by t-SNE, showing the formation of 8 main clusters in different colors from integrated cell pool of HCC ANT and TM. (B) Pie chart showing the percentage of each cluster. (C) Violin plots showing the expression profile of multiple inhibitory receptors, cell markers, effector molecules in each cluster. (D) The branched trajectory of CD8⁺ T-cell state transition in a two-dimensional state-space inferred by Monocle (V.2). Each dot corresponds to one single cell, colored according to its cluster label. Arrows show the increasing directions of certain T-cell properties. ANT, adjacent non-cancerous tissue; HCC, hepatocellular carcinoma; TM, tumor core.

Notably, C0-CD8-GZMA cluster is the largest cell population, accounting for 36.00% of the total cell. Cells of C0 cluster expressed multiple known exhaustion marker genes such as CTLA4, TIGIT, PDCD1, and LAG3. Meanwhile, they overexpressed the effector genes IFNG and GZMA as well. The transcription signatures indicate that it is a cluster of partially exhausted effector CD8⁺ T cells. C1-CD4-CD44 and C2-CD4-CD40LG, the two clusters, shared some mRNA signatures including IL7R, CD44, and CD69 genes. When compared with C1, C2 has higher expression level of killer cell lectin-like receptor G1 KLRG1 (figure 1C). KLRG1 is a lymphocyte co-inhibitory, or immune checkpoint, receptor expressed predominantly on late-differentiated effector and effector memory CD8⁺ T and NK cells.22 Therefore, the C2 is likely effector CD4⁺ T cell at a terminally differentiated state. Regulatory T cells (Tregs) were detected in the C3-CD4-FOXP3 and accounted for 8.48% of total cells, highlighting the immunosuppressive nature of the tumor microenvironment of HCC. C4-CD8-CXCL13 also expressed high levels of checkpoint genes PDCD1, CTLA4, TIGIT, and CXCL13. This leads us to conclude that both C0 and C4 clusters are exhausted and effector CD8⁺ T cells, and that the differentially expressed genes between the two clusters reflect heterogeneity and gradual development of CD8⁺ T-cell exhaustion. At the same time, the sixth cluster, C5-CD8-GNLY, highly expresses GNLY, GZMB, and PRF1 which are typically natural killer T-cell markers. The seventh cluster, C6-PCNA, accounted for 2.06% of total cells, suggesting a small portion of proliferating T cells existed in HCC. Cells in the C7-CD68 cluster expressed high levels of CD68, which is typically antigen expressed by tissue macrophages and cells of myeloid/mononuclear lineage. However, previous studies reported that CD68 antigen is constitutively expressed in NK cells and T cells and that its expression is strongly induced in activated CD4⁺ and CD8⁺ T lymphocytes.23 However, the character of C7-CD68 needs further study. The heatmap of top genes in each cluster is shown in online supplementary figure S1.

Next, we analyzed CD4⁺ T and CD8⁺ T cells to determine the differentiation trajectory based on the pseudotime result (figure 1D). Although the major C0-CD8-GZMA cells were located in different directions with C4-CD8-CXCL13, C5-CD8-GNLY cells in the pseudotime trajectory plot, some C0-CD8-GZMA cells were mixed with them. Therefore, we inferred that C0-CD8-GZMA cells were more closely linked to intermediate populations between the effector populations (C5-CD8-GNLY) and the advanced exhausted population (C4-CD8-CXCL13). It remains to be seen if therapeutic strategies targeting these intermediate populations could prevent their transition to exhausted cells and instead promote them to effector cells. C3-CD4-FOXP3, C1-CD4-CD44, and C2-CD4-CD40LG were located at different branches, indicating functional divergence of these CD4⁺ T cells in HCC.

Metabolic reprogramming of Tex is characterized by enhanced lipolysis and fatty acid oxidation

Although C0 and C4 clusters are both exhausted CD8⁺ T cells, we observed that they differentially expressed multiple important genes including PRF1, IFNG, PCNA, and so on. The difference reflects the heterogeneity of CD8⁺ T cells’ exhaustion in HCC. To further explore the detail mechanism, we compared the expression of metabolic genes in the two clusters (figure 2A). Although most genes are non-differentially expressed between them, we found that FABP5 was specifically upregulated in cluster C0 which has been identified as partially exhausted CD8⁺ T cells (figure 2A) and validated its expression on HCC-infiltrating CD8⁺ T cells by multiplexed immunohistochemistry (figure 2B–2C) and FACS analysis (figure 2D). Increased CD8⁺FABP⁺ T cells are infiltrated in TM instead of ANT (figure 2B–2C). Furthermore, the frequency of FABP5^high cells of CD3⁺CD45RO⁺CD8⁺ T cells was also higher in fresh TM (p<0.05, figure 2D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Increased expression of FABP5 in one Tex cluster. (A) Heatmap showing differentially expressed metabolic genes between two CD8⁺ Tex clusters. C0 exhibited increased mRNA expression of FABP5. (B) Opal multicolor immunohistochemistry (IHC) staining with anti-CD8, FABP5 antibodies, and DAPI. Image on the left shows staining of ANT, while staining of TM is shown on the right. CD8, green; FABP5, white; DAPI, blue. (C) Positivity of CD8⁺FABP5⁺ cells in ANT and TM tissue array of 118 patients with HCC. Scatter diagram was obtained from positive rates of IHC staining. (D) The frequency of FABP5⁺ cells in CD8⁺ T cells of TM. (E) The frequency of FABP5⁺ CD8⁺ T cells in PD1⁺ or PD1⁻ cells of fresh HCC tumor. (F, G) MFI of lipid uptake and intracellular neutral lipid content (measured by BODIPY FL C16 and Bodipy), and (H) comparison of CPT1a protein level between FABP5^high and FABP5^low cells of CD3⁺CD8⁺CD45RO⁺ T cells. (I) Mitochondrial membrane potential measured by Mito Red and TMRM staining. (J) The frequency of CPT1a⁺CD8⁺ T cells in fresh HCC tumor and (K) the levels of ketone in supernatant of cultured CD3⁺CD8⁺CD45RO⁺ T cells isolated from ANT and TM. Data represent mean±SEM of reads from triplicate samples from one experiment and are representative of data from three experiments. p<0.05 was considered statistically significant. *p<0.05; **p<0.01. CPT1a, carnitine palmitoyltransferase IA; FABP5, fatty acid-binding protein 5; MFI, mean florescence intensity; TMRM, tetramethylrhodamine, methyl ester.

Most of the FABP5⁺CD8⁺ T cells are PD1 positive (figure 2E). Given that FABP5 functions as a fatty acid-binding protein, we hypothesized that fatty acid metabolism might be different in CD3⁺CD8⁺CD45RO⁺FABP^high T cells of TM. Consistent with increased FABP5 levels, we determined that lipid uptake, measured by BODIPY FL C16 incorporation, was increased in CD3⁺CD8⁺CD45RO⁺FABP^high T cells of TM (figure 2F), although cellular neutral lipid content remained unchanged (figure 2G). Then carnitine palmitoyltransferase 1a (CPT1a), a rate-limiting enzyme of mitochondrial FAO which plays an important role in the utilization of fatty acids, was expressed at higher levels in CD3⁺CD8⁺CD45RO⁺FABP^high cells (figure 2H). Meanwhile, by using mitochondria probes (Mito Red and TMRM), we found that CD3⁺CD8⁺CD45RO⁺FABP^high cells have enhanced mitochondrial membrane potential (figure 2I). Also, the frequency of CPT1a⁺CD8⁺ T cells are accumulated in TM (p<0.05, figure 2J). The gating strategy of cells that were identified in measuring FABP5 and CPT1a is provided in online supplementary figure S2A. Consistent with the increased levels of CPT1a, a significant elevation of the ketone body 3-hydroxybutyrate, which is produced during FAO, was detected in culture supernatants from CD3⁺CD8⁺CD45RO⁺ cells of TM (figure 2K). Thus, the upregulated FABP5 in exhausted T cells is indicative of an increase in FAO, lipolysis, and utilization of exogenous fatty acids for β-oxidation. To verify that FABP5^highCD8⁺ T cells are exhausted, we tested the expression of additional exhaustion markers (TIM3, CTLA4, and LAG3) in cells from tumor sample by flow cytometry. There is no significant difference of LAG3 protein level between FABP5^high and FABP5^low cells (online supplementary figure 2B). The protein level of TIM3 or CTLA4 is below the detection level (data not shown).

Level of FABP5 expression defines distinct exhausted CD8⁺ T cell infiltrated in HCC

We compared function genes of two CD8⁺ T-cell clusters within TM in figure 3. Notably, the FABP5^high CD8⁺ T cells increasingly expressed IFNG and CD27, which is required for generation and long-term maintenance of T-cell immunity24 25 (figure 3A–3B). Regarding transcription factors, the FABP5^highCD8⁺ T cells expressed upregulated T-bet, IRF4, NFATC1, and Blimp-1 that are typically associated with effector and exhausted CD8⁺ T cells, highlighting the distinct transcriptional fates of these two CD8⁺ T-cell subsets (figure 3A–3B). Besides, the protein level of anti-apoptosis and proliferation genes (Bcl2, Bcl-xL, Ki67) increased in FABP5^highCD8⁺ T cells as well.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

FABP5 expression defines two distinct Tex subpopulations of HCC. Flow cytometry was performed for patients with HCC to validate whether FABP5 expression defines different Tex of HCC. Exhausted CD8⁺ T cells of HCC tumor core were divided into two groups based on high/low FABP5 expression, then expression of CD27, CD127, IFNγ, Bcl2, Ki67, T-bet, TCF7, NFATC1, IRF4, Blimp1, Bcl-xL, and Eomes were compared. The representative flow cytometry figures and statistical data are shown in (A) and (B). Data are mean±SEM of reads from triplicate samples from one experiment, representative of 3–5. All differences shown are statistically significant (at least p<0.05 by Student’s t-test); p<0.05 was considered statistically significant. *p<0.05; **p<0.01; ***p<0.001.

Naturally upregulated FABP5 enriched in CD137-stimulated T cells

We found the mRNA level of TNFRSFs was distinct in the two CD8⁺ T-cell clusters (figure 4A). The total 85.33%±7.669% of FABP5⁺CD8⁺ T cells were TNFRSF9⁺ (CD137) cells, and we confirmed it on protein level (figure 4B). To verify whether the FABP5^highCD8⁺ T cells and the features discussed previously were specific in HCC, we performed similar tests in dissected tumor or adjacent tissue of non-small cell lung cancer (NSCLC) in online supplementary figure S3. FABP5 expression presented in both adjacent tissue and tumor core of NSCLC. Similarly, FABP5 ^highCD8⁺ T cells have higher amount of CD137, Bcl2, IFNγ, CPT1a, and increase lipid uptake (C16) in comparison with FABP5^lowCD8⁺ T cells in online supplementary figure S3. However, PD1 and Ki67 remain unchanged between two groups (data not shown). We observed similar changes between FABP5^high and FABP5^low cells from distinct locations. The enhanced expression of CD137, CPT1a, IFNγ, and PPARγ (except Bcl-2) were also found in FABP5^highCD8⁺ T cells from peritumor tissue of NSCLC. Therefore, FABP5, as immunometabolic marker, has the potential to apply to multiple cancer types although we have no data regarding the clinical relevance of FABP5⁺ cells in NSCLC.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Enhanced FABP5 expression of CD137-enriched Tex. (A) Heatmap of scRNA showing differentially expressed genes of tumor necrosis factor (TNF) and TNF-receptor superfamily (CD8⁺FABP5^highT cells vs CD8⁺FABP5^lowT cells). (B) Correlation of FABP5 and CD137 of CD8⁺ T cell in tumor core was analyzed through flow cytometry for patients with HCC. The left plot is isotype control of CD137 antibody. (C) Representative flow cytometry plots and MFI of CD137 in CD127⁺KLRG1⁻CD8⁺ T cells or CD127⁻KLRG1⁺CD8⁺ T cells of tumor core. (D) Volcano plot showing differentially expressed genes between the CD137⁺ and CD137⁻ populations. The plot was generated from scRNA data. Data are mean±SEM of reads from triplicate samples from one experiment, representative of 3–5. All differences shown are statistically significant; p<0.05 was considered statistically significant. *p<0.05; **p<0.01.

CD127⁻KLRG1⁺ and CD127⁺KLRG1⁻ are respectively used to define short-lived effector cells (SLECs) and memory precursor cells (MPECs).26 27 To evaluate the state of CD137⁺CD8⁺ T cells from HCC, CD127 and KLRG1 surface expression were detected by flow cytometry. CD137 expression was significantly increased in CD127⁺KLRG1⁻ T cell (figure 4C).

Volcano plot showing differentially expressed genes between the CD137⁺ and CD137⁻ populations. Compared with the CD137⁻ population, CD137⁺CD8⁺ T cells highly expressed genes associated with metabolic functions such as FABP5 and Mitochondrially Encoded Cytochrome C Oxidase I (MT-CO1), further supporting the finding that CD137⁺ cells were antigen-experienced and also induced metabolic reprogramming of CD8⁺ T cells (figure 4D).

Effects of inhibiting FABP5 expression on T cells

To confirm that these metabolic alterations were due to the signaling domains of the receptor, similar experiments were performed with mesothelin-specific CART cells. We generated CD28 CART and CD137 CART. CD28 CART has higher basal OCR than CD137 CART cells (figure 5A). However, there was an increase in spare respiratory capacity (SRC) that was specific to the CD137 CART cells (figure 5A), following decoupling of the mitochondrial membrane using carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP). Increased SRC likely supports T-cell function in a hostile tumor environment.6 7

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

CD137 signaling instructs PPARγ-FABP5 expression. (A) Oxygen consumption rate (OCR) spectra of CD28 CART and CD137 CART using Seahorse XF96 analyzer. Basal OCR, maximal OCR, and spare respiratory capacity (SRC) were compared. (B) The expression of multiple metabolic genes in activated CD137 CART using real-time PCR. (C) The expression level of PPARγ of tumor-infiltrating CD8⁺ T cells was compared between FABP5^high and FABP5^low cells. (D) Cell suspension of HCC tumor tissue was treated with TOFA (10 µM), UK-5099 (50 µM), and PPARγ inhibitor GW9662 (10 µM) for 24 hours. Then expression of FABP5 and lipid uptake (measured by BODIPY FL C16) of CD8⁺ T cell was detected by cell flow cytometry. (E) After treatment of GW9662 for 24 hours, MFI of IFNγ, Ki67, and BCL2 in CD137 CART cells was detected by flow cytometry. (F, G) CFSE proliferation assays, Annexin V, and PI staining were also performed in CART cells treated with GW9662. (H, I) Survival of CD137 CART cells affected by metabolic intervention. We treated the activated CD137 CART cells with metabolic inhibitors for 24 hours and then performed live/dead staining. (J) The effects of energy deprivation on in vitro tumor-killing capability of CD137 CART cells. CD137 CART cells respectively treated with metabolic inhibitors and co-cultured with K562-mesothelin cell line at a ratio of 10:1. After 24 hours, cytotoxicity was evaluated through LDH cytotoxicity assay. Data represent mean±SEM of reads from triplicate samples from one experiment representative of 3. p<0.05 was considered statistically significant. *p<0.05; **p<0.01. PPARγ, peroxisome proliferator-activated receptor γ.

To gain insight into the mechanism leading to the metabolic differences conferred by CD137 signaling domain, we measured the expression of candidate genes that are implicated in glycolytic and lipid metabolism (figure 5B). Primer sequences of genes implicated in glycolytic and lipid metabolism mentioned were provided in online supplementary table S2. Several genes involved in glycolysis are targets of hypoxia-inducible factor 1-alpha (HIF1α) including glucose transporter 1 (Glut1), phosphofructokinase (PFKP), and pyruvate kinase isozymes M2 (PKM2) were downregulated in stimulated-CD137 CART cells (figure 5B). Meanwhile, we found that genes associated with mitochondrial FAO were upregulated (figure 5B). In addition, mRNA levels of FABP5, which plays a critical role in long-chain fatty acid uptake, transport, and metabolism, were significantly upregulated in CD137 CART cells. These findings suggest that CD137 programs T cells to use fatty acids as the predominant energy source, then supports T-cell proliferation, survival, and function in a nutrient-limited tumor environment. We noticed that peroxisome proliferator-activated receptor gamma (PPARγ) was extraordinarily upregulated in CD137 CART cells after specific antigen stimulation. Meanwhile, the mRNA expression level of PPARα was increased, which is consistent with enhanced expression of lipid oxidative genes. Then we confirmed that the expression level of PPARγ of tumor-infiltrating CD8⁺ T cells was higher in FABP5^high cells (figure 5C). PPARγ has been proven as a major regulator of lipid metabolism–related gene expression which includes FABPs. So we thought it was possible that PPARγ instructed the upregulated FABP5 expression of CD137-activated Tex. Then we use PPARg inhibitor GW9662 to test whether PPARg is responsible for FABP5 expression of CD137 signaling. In comparison with TOFA, de novo fatty acid synthesis inhibitor, or UK-5099, the inhibitor of the mitochondrial pyruvate carrier (MPC), GW9662 treatment specifically inhibited the protein amount of FABP5 and lipid uptake (figure 5D). Then we tested whether blocking PPARγ has effects on important biological functions of CD137 CART. The results showed that the decreased Ki67 and Bcl-2 expression presented in GW9662 treatment (figure 5E). However, there was no difference between ±GW9662-treated cells in IFNγ production of CD137⁺CD8⁺ T cells from HCC (figure 5E). Consistent with the Bcl-2 and Ki67 change, Annexin V-PI and CFSE proliferation assays support that the inhibition of FABP5 expression leads to impaired anti-apoptosis and proliferation (figure 5F–5G). To test the effects of metabolic deprivation on CART cell survival and cytotoxic function, we treated the activated CD137 CART cells with metabolic inhibitors TOFA, UK, and GW9662 for 24 hours, then measured the percentage of live/dead cells. TOFA and GW9662 treatments cause more dead cells although no significant statistical change was observed in live cells (figure 5H–5I). Therefore, fatty acid uptake and transport and fatty acid synthesis are both important for CD137 CART cell survival. In contrast, UK-5099, a specific MPC inhibitor, inhibited in vitro tumor killing of CART cells instead of cell survival as shown in figure 5H–5J. Thus, the effect of CD137 signaling–induced metabolic reprogramming contributes to long persistence of tumor-infiltrating lymphocytes (TILs) in tumor microenvironment rather than killing capacity.

FABP5^high T cells of HCC are highly clinically relevant

To evaluate clinical relevance of metabolic markers of the infiltrated T cells in HCC, we collected resected tumor and adjacent tissue from 118 patients and developed TMA. Clinical information of patients is summarized in figure 6A and online supplementary table S1. By the end of 5 years’ follow-up, 48.3% (57/118) of patients suffered a recurrence and 33.1% (39/118) died. The 1-year, 3-year, and 5-year OS rates were 81%, 70%, and 63% and the cumulative recurrence rates were 33%, 41%, and 56%, respectively. Tumor size, Edmondson-Steiner grade (III–IV vs I–II), γ-glutamyl transferase (γ-GT), BCLC stage system (B+C vs A), and vascular invasion were predictors of OS and TTR for univariate analysis. The individual clinicopathological features that presented significance in the univariate analysis were adopted as covariates in a multivariate Cox proportional-hazards model for further analysis. γ-GT was only one predictor for both OS and TTR in univariate/multivariate analysis (online supplementary table S1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

OS and TTR curves from ANT and TM tissue microarray. Based on cut points of OS status, data of OS and TTR status were divided into two groups (low vs high) judged by X-tile Kaplan-Meier analysis. p<0.05 was considered statistically significant. OS, overall survival; TTR, time to recurrence. CD8⁺FABP5⁺_TM^low, low double positive expression of CD8 and FABP5 in TM; CD8⁺FABP5⁺_TM^high, high double positive expression of CD8 and FABP5 in TM; CD8⁺FABP5⁺_ANT+TM^low, low double positive expression of CD8 and FABP5 in both ANT and TM; CD8⁺FABP5⁺_ANT+TM^high, high double positive expression of CD8 and FABP5 in both ANT and TM.

Except FABP5 discussed in this study, we added LDHA which is one of the top expressed metabolic genes among our scRNA data. LDHA has been reported that it is induced in activated T cells to support aerobic glycolysis and promotes IFNγ expression.28 We analyzed the correlation between the percentage of CD8⁺FABP5⁺ or CD8⁺LDHA⁺ T cells, overall survival rate, and cumulative recurrence rate by using Kaplan-Meier curve (figure 6B and Supplemental Figure S4). CD8⁺FABP5^highT cells of patients with HCC seems to link with improved survival rate (figure 6B). However, p values are respectively 0.0995 and 0.1842 for TM and ANT+TM. Next, we found that CD8⁺FABP5⁺_TM^high predicts longer recurrence-free survival. Meanwhile, the cumulative recurrence rates of CD8⁺FABP5⁺_ANT+TM^high patients were also lower than that of CD8⁺FABP5⁺_ANT+TM^low (1 year: 34% vs 37%; 3 years: 30% vs 45%; 5 years: 36% vs 63%) (figure 6B). The data suggest that FABP5 has the potential of predicting clinical outcomes of HCC. Although LDHA plays an important role of affecting T-cell effector function, in our study, the protein level of LDHA in HCC tissue array has no correlation with patient survival and recurrence rate (online supplementary figure 4).