If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Immunology Programme, Life Sciences Institute, Immunology Translational Research Program and Department of Microbiology and Immunology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore 117456, Singapore
Department of Oncology, Asan Medical Center (AMC), University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea
Translational Immunology Institute (TII), SingHealth-DukeNUS Academic Medical Centre, Singapore 169856, SingaporeDuke-NUS Medical School, Singapore 169857, Singapore
Genome Institute of Singapore (GIS), Agency for Science, Technology and Research (A∗STAR), Singapore 138672, SingaporeSchool of Biological Sciences, Nanyang Technological University, Singapore 637551, Singapore
Department of Oncology, Asan Medical Center (AMC), University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea
Duke-NUS Medical School, Singapore 169857, SingaporeDepartment of Anatomical Pathology, Singapore General Hospital (SGH), Singapore 169856, SingaporeInstitute of Molecular and Cell Biology (IMCB), Agency for Science, Technology and Research (A∗STAR), Singapore 138673, Singapore
Translational Immunology Institute (TII), SingHealth-DukeNUS Academic Medical Centre, Singapore 169856, SingaporeDuke-NUS Medical School, Singapore 169857, Singapore
Genome Institute of Singapore (GIS), Agency for Science, Technology and Research (A∗STAR), Singapore 138672, SingaporeKey Laboratory of Zoological Systematics and Evolution, Institute of Zoology, Chinese Academy of Sciences, Beijing 100107, ChinaCenter for Excellence in Animal Evolution and Genetics, Chinese Academy of Sciences, Kunming 650223, China
Department of Oncology, Asan Medical Center (AMC), University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea
Immunology Programme, Life Sciences Institute, Immunology Translational Research Program and Department of Microbiology and Immunology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore 117456, Singapore
Division of Medical Oncology, National Cancer Centre Singapore, Singapore 169610, SingaporeCurie Oncology, Mount Elizabeth Novena Specialist Centre, Singapore 329563, Singapore
Translational Immunology Institute (TII), SingHealth-DukeNUS Academic Medical Centre, Singapore 169856, SingaporeDuke-NUS Medical School, Singapore 169857, Singapore
Distinct immune subsets are linked to response and toxicity in immunotherapy.
•
Antigen-presenting cells and CD8 TEM cell function act as an interface between response and toxicity.
•
Diverging immune trajectories of immune subsets contribute to either response or toxicity.
•
TNFR2 inhibition uncouples response and toxicity in anti-PD-1 immunotherapy.
Background & Aims
While immune checkpoint blockade (ICB) has shown promise in patients with hepatocellular carcinoma (HCC), it is associated with modest response rates and immune-related adverse events (irAEs) are common. In this study, we aimed to decipher immune trajectories and mechanisms of response and/or irAEs in patients with HCC receiving anti-programmed cell death 1 (anti-PD-1) therapy.
Methods
Pre- and on-treatment peripheral blood samples (n = 60) obtained from 32 patients with HCC (Singapore cohort) were analysed by cytometry by time-of-flight and single-cell RNA sequencing, with flow cytometric validation in an independent Korean cohort (n = 29). Mechanistic validation was conducted by bulk RNA sequencing of 20 pre- and on-treatment tumour biopsies and using a murine HCC model treated with different immunotherapeutic combinations.
Results
Single-cell analyses identified CXCR3+CD8+ effector memory T (TEM) cells and CD11c+ antigen-presenting cells (APC) as associated with response (p = 0.0004 and 0.0255, respectively), progression-free survival (p = 0.00079 and 0.0015, respectively), and irAEs (p = 0.0034 and 0.0125, respectively) in anti-PD-1-treated patients with HCC. Type-1 conventional dendritic cells were identified as the specific APC associated with response, while 2 immunosuppressive CD14+ myeloid clusters were linked to reduced irAEs. Further analyses of CXCR3+CD8+ TEM cells showed cell-cell interactions specific to response vs. irAEs, from which the anti-PD-1 and anti-TNFR2 combination was harnessed to uncouple these effects, resulting in enhanced response without increased irAEs in a murine HCC model.
Conclusions
This study identifies early predictors of clinical response to anti-PD-1 ICB in patients with HCC and offers mechanistic insights into the immune trajectories of these immune subsets at the interface between response and toxicity. We also propose a new combination immunotherapy for HCC to enhance response without exacerbating irAEs.
Clinical trial number
NCT03695952.
Lay summary
Response rates to immune checkpoint blockade (ICB) treatment in hepatocellular carcinoma (HCC) remain modest and adverse events are common. Herein, we identified early predictors of response and gained an in-depth understanding of the immunological mechanisms behind response and adverse events in patients with HCC treated with ICB. We also proposed a new combination immunotherapy for HCC that enhances response without exacerbating adverse events.
We read with great interest the recent study by Chuah et al.1 aiming to find immune predictors of treatment response and mechanisms of response versus immune-related adverse events (irAE) in hepatocellular carcinoma (HCC) patients treated with anti-programmed cell death 1 (PD-1). The authors noticed the recent successful combinations of immunotherapies achieved increased response rates compared to monotherapy, but with an accompanying increase in irAEs. The authors were inspired by this and conducted an innovative study to investigate how to increase response rates and reduce irAEs simultaneously.
We thank Zhou1 for their interest and comments on our work.2 Zhou commented that the two binary endpoints of efficacy and safety should be analyzed separately or reorganized into four groups. Indeed, the relationship between response and toxicity to immunotherapy is widely recognised, which we also acknowledged by citing the clinical trials of CheckMate040 and IMBRAVE150, where immune related adverse events (irAEs) increased in tandem with greater objective response rates (ORRs) in combination immunotherapy.
Immune checkpoint blockade (ICB) by antibodies targeting molecules such as programmed cell death-1 (PD-1) and cytotoxic T lymphocyte-associated protein-4 (CTLA4) are among the most widely used cancer immunotherapies. ICB has achieved promising outcomes in various malignancies, including hepatocellular carcinoma (HCC), which remains the sixth-most common cancer and fourth leading cause of cancer mortality worldwide.
The use of anti-PD-1 ICB monotherapy in patients with advanced HCC produced modest objective response rates (ORRs) of 15% or 18.3% in phase III trials for nivolumab
While recently reported combination immunotherapies for HCC conferred greater ORRs, irAEs increased in tandem. For instance, anti-PD-1 combined with anti-CTLA4 for patients with advanced HCC resulted in ORRs of 31% and grade 3/4 irAEs of 37% (CheckMate040),
and anti-programmed death-ligand 1 (PD-L1) combined with anti-vascular endothelial growth factor-A resulted in ORRs of 27.3% and grade 3/4 AEs of 56.5% (IMBRAVE150).
Therefore, a better understanding of the mechanisms of response and/or irAEs is critical for improving outcomes associated with ICB.
The development of single-cell, multi-parametric technologies has provided the means to extract valuable data from limited samples, enabling in-depth characterisation of the immune landscape for mechanistic and biomarker discovery.
Response to immunotherapy requires re-activation of the immunosuppressive tumour microenvironment (TME). Nonetheless, the systemic immune landscape plays an important role in the anti-tumour immune response
and provides a practical and minimally invasive source of biomarkers in the clinical setting. Moreover, irAEs that could be fatal and delay or disrupt treatment
commonly manifest as systemic autoimmune conditions.
In this study, we conducted deep single-cell immunoprofiling of patients with HCC to identify immune signatures predictive of response to anti-PD-1 therapy and to elucidate the mechanisms behind response vs. irAEs.
Materials and methods
Patient samples
Patients with HCC receiving anti-PD-1 ICB – nivolumab or pembrolizumab from the National Cancer Centre, Singapore (SG cohort n = 32, Table S1, real-world clinical cohort) or nivolumab from the Asan Medical Center, South Korea (KR cohort n = 29, Table S2, NCT03695952) – who provided written informed consent were recruited according to the Institutional Review Board guidelines of each institution. Patients received intravenous nivolumab (3 mg/kg) every 2 weeks or pembrolizumab (200 mg) every 3 weeks. Blood samples were collected at baseline (both cohorts) and during treatment (SG cohort only). Treatment response was monitored using RECIST1.1 and irAEs were assessed using National Cancer Institute Common Terminology Criteria for Adverse Events (NCI CTCAE; version 4.03). Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque Plus (GE Healthcare, UK) (SG Cohort) or Lymphocyte Separation Medium (Corning) (KR cohort). mRNA from pre and 1-week on-treatment tumour biopsies were obtained (n = 10 patients, SG cohort).
HCC model
Male C57BL/6 mice (aged 6–8 weeks; InVivos, Singapore), housed in specific pathogen-free conditions according to guidelines of Institutional Laboratory Animal Care and Use Committee of the National University of Singapore, were inoculated with 1×106 Hepa1-6 murine hepatoma cells via hydrodynamic tail-vein injection.
From day 7, tumour-bearing mice were injected intraperitoneally bi-weekly for 2 weeks with anti-PD-1 (RMP1-14, 250 μg/mouse), anti-TNFR1 (55R-170, 250 μg/mouse), anti-TNFR2 (TR75-54.7, 500 μg/mouse), alone or in combination (anti-PD1+anti-TNFR1, anti-PD1+anti-TNFR2), Armenian hamster IgG (PIP, 500 μg/mouse) and rat IgG2a (1-1, 250 μg/mouse) (all from ichorbio, UK). On day 21, mice were euthanized by CO2 asphyxiation and the numbers of liver tumour nodules and liver weights were recorded. Infiltrating leucocytes from tumour and non-tumour liver tissue were isolated for flow cytometry analysis. Mouse colons were flushed and collected for formalin-fixed paraffin embedding using the Swiss-rolling method.
Cytometry by time-of-flight
Cytometry by time-of-flight (CyTOF) staining was performed as previously described
with a panel of 39 antibodies (Table S3) and analysed using a Helios mass cytometer (Fluidigm, USA). Data were down-sampled to 10,000 viable CD45+ cells for in-house developed Extended Poly-dimensional Immunome Characterisation.
Clustering was performed with the FlowSOM algorithm, dimension reduction by t-distributed stochastic neighbour embedding, and visualisation with the R shiny app ‘SciAtlasMiner’. Enriched clusters were identified by 2-tailed Mann-Whitney U test and validated with manual gating using FlowJo (V.10.5.2; FlowJo, USA).
Flow cytometry
Baseline PBMC samples from 29 patients (KR cohort) were stained with 13 antibodies (Table S4) and analysed using a BD LSR II cytometer. Immune cells from mouse samples were stained with 10 antibodies (Table S5). The Intracellular Fixation/Permeabilization Buffer Set (eBioscience) was used for intracellular staining. All data analysis was conducted using FlowJo V.10.5.2.
Single-cell RNA sequencing
Single-cell RNA sequencing (scRNA-seq) was performed on 10 PBMC samples consisting of 9 on-treatment samples (6 Res vs. 3 Non-Res; 5 Tox vs. 4 Non-Tox) and 1 matched pre-treatment sample (HCC6; Res/Tox) (Table S1). The 5’ gene expression libraries were prepared using the 10x Genomics platform for indexed paired-end sequencing of 2x150 base pairs on an Illumina HiSeq 4000 system at 20,000 read pairs per cell. Reads were aligned to the human GRCh38 reference genome and quantified using cellranger count (10x Genomics, v3.0.2). Data repository ID: EGAS00001004843. Cells with <200 genes and >10% mitochondrial RNA were filtered, followed by analyses using Seurat (v3.0) pipelines. A total of 29 cell clusters were annotated based on the expression of known cell lineage-specific genes (Table S6).
Functional pathway analysis was conducted using Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.8. CellPhoneDB 2.0
was used to analyse ligand-receptor expression and predict cell-cell communications of CXCR3-expressing CD8 T cells using default parameters.
Bulk RNA-seq
mRNA from matched pre- and 1-week on-treatment tumour biopsies (n = 10 patients, Table S1) were obtained using the Qiagen AllPrep DNA/RNA Mini Kit and sequenced using HiSeq 4000 platform. Raw reads were aligned to the Human Reference Genome hg19 via STAR and the expected gene-level counts were calculated using RSEM. Protein-coding genes with >0.5 counts per million were retained and differentially expressed gene (DEG) analyses were conducted using R package DESeq2 with Benjamini-adjusted p <0.05 and |log2(fold-change)|>0.5 (Table S7). Functional pathway analysis was conducted using DAVID v6.8.
Statistical analysis
Statistical analyses were performed using unpaired Mann-Whitney U or Wilcoxon matched-pairs tests with 2-tailed p values using GraphPad Prism7. Cox regression with Wald test analysis and Kaplan-Meier curves with Log-rank tests were performed using the R package survminer.
Results
Early immunological predictors of response in the peripheral blood
Pre- and on-treatment blood samples from patients with HCC receiving anti-PD-1 ICB (SG cohort: n = 32; Table S1) were analysed using CyTOF and scRNA-seq to uncover the mechanisms of response and irAEs (Fig. 1A). An additional KR cohort (n = 29; Table S2) was included as a validation cohort and analysed using flow cytometry for defined biomarkers identified from the SG cohort. Further validation was conducted by bulk RNA-seq analysis of pre-vs. 1-week on-treatment tumour biopsies (SG cohort) and using a murine HCC model (Fig. 1A). The patients were stratified as: responders (Res), those who showed partial response or stable disease for ≥6 months; and non-responders (Non-Res), those who showed progressive disease within 6 months according to RECIST1.1. The 6-month time-point was identified from the Checkmate040 study, in which disease control with stable disease for ≥6 months was reported in 37% of patients with HCC treated with nivolumab.
Nivolumab in patients with advanced hepatocellular carcinoma (CheckMate 040): an open-label, non-comparative, phase 1/2 dose escalation and expansion trial.
Patients were also categorised as: Tox, those who experienced Grade (G) 2 and above irAEs; and Non-Tox, those with G1 or no irAEs according to NCI CTCAE v4.03, where G2 irAEs is the point where therapeutic interventions or ICB interruption would be considered.
(A) Schematic illustration of clinical study design and data analysis. (B) Heatmap showing scaled median expression of key markers in the immune clusters enriched in Res or Non-Res. (C) tSNE plots showing enriched immune clusters (C) in Res or Non-Res. (D,E) Boxplots showing manually gated immune subsets in (D) Res (n = 8) or non-Res (n = 13) from SG cohort; and (E) Res (n = 9) or Non-Res (n = 20) from KR cohort. Median and interquartile range shown. Baseline samples indicated as black circles. ∗, ∗∗ denote 2-tailed p <0.05 and p <0.01, respectively, by unpaired Mann-Whitney U test. (F) Kaplan-Meier graphs showing PFS profiles of SG (n = 21) and KR cohorts (n = 29). Log-rank test p values are shown. APC, antigen-presenting cell; CyTOF, cytometry by time-of-flight; HCC, hepatocellular carcinoma; ICB, immune checkpoint blockade; irAEs, immune-related adverse events; KR, South Korea; MDSCs, myeloid-derived suppressor cells; Non-Res, non-responders; PBMCs, peripheral blood mononuclear cells; PFS, progression-free survival; Res, responders; scRNA-seq, single-cell RNA sequencing; SG, Singapore; TEM, effector memory T; Tregs, regulatory T cells; tSNE, t-distributed stochastic neighbour embedding.
revealed clusters corresponding to major immune lineages and subtypes according to the relative expression of 38 immune markers (Fig. S1A,B). To identify biomarkers for early prediction of response, we selected pre- and early on-treatment samples (<6 weeks from treatment initiation, before the first restaging CT scan for efficacy determination) from the SG cohort (n = 21; Table S1). The initial unsupervised Mann-Whitney U analysis of 6 Res vs. 6 Non-Res clinically matched samples (Table S1) revealed 2 CD4+ clusters – namely, FoxP3+CD4+ T cells (C33) and FoxP3+CTLA4+CD4+ regulatory T cells (Tregs) (C3), and a CD8+CD45RO+CCR7-CXCR3+ TEM (C76) cluster – that were enriched in Res (Fig. 1B,C and S1C). Two distinct CD11c+ myeloid cell clusters, C4 (HLADRhiCD86+ – indicative of antigen presentation capabilities) and C37 (CD14+HLADRlo/- – potentially myeloid-derived suppressor cells
) were enriched in Res and Non-Res, respectively (Fig. 1B,C and S1C). Validation of these clusters by supervised manual gating with FlowJo (Fig. S2A) confirmed the significant enrichment of these immune subsets (n = 21, Fig. 1D). Notably, these clusters showed similar frequencies in pre- or early on-treatment (<6 weeks) blood, particularly in Res (Fig. S2B).
We next validated the enrichment of peripheral Tregs, CXCR3+CD8+ TEM cells and APCs in Res, and myeloid-derived suppressor cells in Non-Res by flow cytometric analysis of an independent anti-PD1-treated KR cohort (n = 29; Fig. 1E, S3A and Table S2). Moreover, Kaplan-Meier analyses showed that higher frequencies of Tregs, APCs and CXCR3+CD8+ TEM cells were significantly associated with superior progression-free survival (PFS) in both cohorts (Fig. 1F). Multivariate analyses of these biomarkers with clinical parameters revealed enrichment of CXCR3+CD8+ TEM cells and APCs as independent predictors of PFS in both cohorts, while (≥G2) irAE incidence showed marginal significance (Table S8). To examine the influence of irAE status on the association of these immune biomarkers with response, we analysed them by segregating the patients according to their Tox status. We observed that CXCR3+CD8+ TEM cells and APCs remained significantly enriched in Res, particularly in Non-Tox patients, from both cohorts (Fig. S2C and S3B). These data show that peripheral CXCR3+CD8+ TEM cells and APCs are independent predictors of response and PFS in patients with HCC treated with anti-PD-1 ICB.
Peripheral immune markers associated with irAEs
Next, we analysed blood samples obtained during or close to (±2-weeks) ≥G2 irAEs (Tox) vs. those at matched post-ICB timepoints from patients who developed no or G1 irAEs (Non-Tox) (Table S1). Due to differences in the study design, this analysis was only performed for the SG cohort. Two CXCR3+CD38+CD16+CD56+ natural killer cell clusters (C89 and 99) showed enrichment in Tox (Fig. 2A,B and S4A). Conversely, 3 CD8+ clusters – namely, C66 and C76 TEM (CD45RO+CCR7-) cells and C96 (V⍺7.2+CD161+CD56+CD8+) mucosal-associated invariant T cells – and a CD11c+CD14+HLADR+ APC cluster (C27) were enriched in Non-Tox (Fig. 2A,B and S4A). Interestingly, the same CXCR3+CD8+ TEM (C76) cluster was also enriched in Res as described earlier (Fig. S1C). Manual gating confirmed their enrichment (Fig. 2C and S4B). All 5 immune subsets displayed similar trends after controlling for response status (Fig. S4C). Thus, these immune subsets provide insights into irAE manifestation, regardless of their response status.
Fig. 2Peripheral immune markers associated with irAEs.
(A) Heatmap showing scaled median expression of key markers in the immune clusters enriched in either Tox (≥Grade 2 irAEs) or Non-Tox (Grade 1 or no irAEs). (B) tSNE plots showing enriched immune clusters (C) in Tox or Non-Tox. (C) Boxplots showing manually gated immune subsets in Tox (n = 9) or Non-Tox (n = 11). Median and interquartile range shown. ∗, ∗∗denote 2-tailed p <0.05 and p <0.01, respectively, by unpaired Mann-Whitney U test. APC, antigen-presenting cells; irAEs, immune-related adverse events; MAITs, mucosal invariant T cells; NK, natural killer; Non-Tox, non-toxicity; TEM, effector memory T; Tox, toxicity; tSNE, t-distributed stochastic neighbour embedding.
Distinct CD11c+ myeloid APC subsets involved in response and irAEs
To obtain deeper molecular and mechanistic insights into on-treatment transcriptomic perturbations in the immune subsets identified, we conducted scRNA-seq on 10 PBMC samples consisting of 9 on-treatment PBMCs (6 Res vs. 3 Non-Res; 5 Tox vs. 4 Non-Tox) and 1 matched pre-treatment sample (Res/Tox) (Table S1). From 59,980 single cells, 29 clusters were identified and annotated according to their respective DEGs (Fig. 3A,B and Table S6).
Fig. 3scRNA-seq analysis of immune subsets associated with response and irAEs.
(A) tSNE plot showing 29 scRNA-seq clusters. (B) Heatmap illustrating the top 10% DEGs of 29 clusters. (C,D) Box plots showing cell clusters enriched in Res (n = 6) or Non-Res (n = 3) and Tox (n = 5) or Non-Tox (n = 4). Median and interquartile range shown. ∗denotes 2-tailed p <0.05 by unpaired Mann-Whitney U test. (E) Violin plots of selected DEGs from the 4 enriched immune subsets in either response or irAEs. (F) Top 10 significant pathways by fold-enrichment for cDC1. (G) Top 5 significant pathways by fold-enrichment for both CD14-1 and CD14-3. (F,G) Vertical axis represents -log10(Benjamini)-adjusted p values and colour gradient represents fold-enrichment of each pathway. cDC, conventional dendritic cells; DEGs, differentially expressed genes; Eff, effector; Imm, immature; irAEs, immune-related adverse events; Non-Res, non-responders; Non-Tox, non-toxicity; pDCs, plasmacytoid DCs; Prolif, proliferative; Res, responders; Th2, T-helper 2 cells; Tox, toxicity; Tregs, regulatory T cells; tSNE, t-distributed stochastic neighbour embedding.
were significantly enriched in Res (Fig. 3C,E and S5A,B). In addition, 2 CD14 and ITGAX (CD11c)-expressing myeloid clusters, CD14-1 and CD14-3, were associated with Non-Tox (Fig. 3D,E and S5A,C). These CD14+ clusters expressed higher levels of KLF4 and CLEC7A, indicating potential polarization to immunosuppressive macrophages.
To decipher the immune mechanisms behind the distinct clinical fates of response and irAEs, we next focused on CD11c+ APCs which were associated with both events. The cDC1 cluster enriched in Res expressed the highest level of HLA genes (Fig. 3E and S5A), suggesting superior antigen presentation capability. Indeed, its enriched functional pathways included antigen processing and presentation via MHC class II, T-cell co-stimulation, and IFNγ-mediated signalling (Fig. 3F), which are important for immune priming. This corroborates other studies associating cDC1s with better survival in cancers, as well as with anti-tumour roles in adoptive T-cell therapy and ICB.
Comparison of the other 2 myeloid clusters associated with non-Tox (CD14-1 and CD14-3) revealed that CD14-1 expressed higher levels of antigen-presenting HLA-related genes than CD14-3 (Fig. 3E and S5A, albeit lower than the cDC1 cluster). Conversely, CD14-3 expressed higher levels of immunosuppressive STAB1 (Clever-1)
(Fig. 3E). Furthermore, among their enriched functional pathways, peptide antigen assembly with MHC class II and the pro-inflammatory IL-1β pathway were enriched in CD14-1 but not in CD14-3 (Fig. 3G). In summary, among these CD14 clusters, CD14-3, that is more significantly associated with Non-Tox (Fig. 3D), displayed reduced antigen presentation/inflammatory characteristics and a more immunosuppressive phenotype than CD14-1.
Distinct phenotypes of CXCR3+CD8+ TEM cells in response and irAEs
Since we identified CXCR3+CD8+ TEM cells as the immune subset common to both Res (Fig. 1D,E) and Tox (Fig. 2C), we next focused on CXCR3-expressing CD8 T cells (n = 863 cells) in the scRNA-seq data (Fig. S5D). Compared to all other T cells, multiple genes involved in antigen presentation, HLA(s), inflammation, granzymes (GZM)s and proliferation (MKI67) were enriched in the CXCR3+CD8+ T cells (Fig. 4A). Conversely, the expression of naïve T-cell markers like CCR7, IL7R and LEF1 was downregulated (Fig. 4A), suggesting an effector memory phenotype. Enriched functional pathways included inflammatory response, cytolysis and antigen processing and presentation via MHC class II (Fig. 4B). Thus, these CXCR3+CD8+ TEM cells display a more inflammatory and cytolytic phenotype compared to other T cells.
Fig. 4CXCR3+CD8+ TEM cells are involved in both response and irAEs.
(A) Volcano plot showing DEGs comparing CXCR3+CD8+ TEM against all T cells. Selected genes are highlighted in boxes as upregulated (Up): p <0.05 & ln(FC)>0.25, downregulated (Down): p <0.05 & ln(FC)<-0.25 or non-significant (n.s.): ln(fold-change)<±0.25. (B) Top 10 significantly enriched functional pathways from DEGs in CXCR3+CD8+ TEM cells. Vertical axis represents the -log10(Benjamini) adjusted p values and colour gradient represents fold-enrichment. (C,D) Dot plots of selected ligand-receptor interacting pairs between CXCR3+CD8+ TEM cells and all other immune clusters computed using CellPhoneDB in (C) Res/Non-Res and (D) Tox/Non-Tox. Point shade reflects log2Mean of average expression levels of interacting molecule 1 from cluster 1 and interacting molecule 2 from cluster 2. Point size indicates the -log10(p value). DEGs, differentially expressed genes; irAEs, immune-related adverse events; NK, natural killer; Non-Res, non-responders; Non-Tox, non-toxicity; pDCs, plasmacytoid DCs; Res, responders; TEM, effector memory T; Tox, toxicity; Tregs, regulatory T cells; tSNE, t-distributed stochastic neighbour embedding.
Given that the systemic immune landscape is a dynamic ecosystem of immune cell cross-talk that could affect their functions in immunity, we employed CellPhoneDB
to identify the expression of receptors and ligands in CXCR3+CD8+ TEM cells and predict their potential interactions with other immune cells. Lymphotoxin alpha and its receptors, tumour necrosis factor receptor superfamily (TNFRSF) 1A, 1B and lymphotoxin beta receptor, which could promote inflammation and oncogenesis,
were enriched in both Res and Tox (Fig. 4C,D). This suggests that CXCR3+CD8+ TEM cells form pro-inflammatory interactions with other cells, leading to both response and irAEs.
Furthermore, we observed distinct TNF interactions between CXCR3+CD8+ TEM and myeloid cell populations, where TNF-TNFRSF1B (TNFR2) was enriched in Res, but TNF-TNFRSF1A (TNFR1) was enriched in Non-Tox (Fig. 4C,D). The interactions of TNF with TNFRSF1A and 1B play important roles in macrophage activation and inflammation.
To validate the protein expression of TNFα, TNFR1 and TNFR2, we performed flow cytometry on PBMCs from ICB-treated patients with HCC (Fig. S6A). Consistent with data shown in Fig. 4C, we found that TNFα expression was significantly higher particularly in CXCR3+CD8+ TEM cells from Res than Non-Res (Fig. S6B). We also observed increased expression of TNFR1 on both CD14+ monocytes and CD14-CD11c+HLA-DR+ dendritic cells in Non-Tox vs. Tox (Fig. S6C). However, there was no significant difference in TNFR2 expression on monocytes and dendritic cells between Res and Non-Res or Tox and Non-Tox (Fig. S6D), indicating that the increased TNF interaction in Res (Fig. 4C) is largely driven by TNFα upregulation; while in Non-Tox (Fig. 4D), it is primarily due to increased TNFR1 expression. This suggests that the different TNF signalling pathways could be harnessed to uncouple response and irAEs in ICB.
Tissue recruitment of APCs and CXCR3+CD8+ TEM cells
The trafficking of immune cells into tumour tissue, as part of the anti-tumour response induced by immunotherapy, could be reflected by changes in their frequencies in the blood.
We compared the frequency of the response-associated immune subsets (Fig. 1D,E) and found a significant reduction in APCs and CXCR3+CD8+ TEM cells in late (>10 weeks) compared to the matched early (<6 weeks) on-therapy blood samples (Table S1), indicating their tumour recruitment in Res (Fig. 5A) but not in Non-Res (Fig. S7A).
Fig. 5Tissue recruitment of APCs and CXCR3+CD8+ TEM cells upon immunotherapy.
(A) Analyses of immune subsets from responders’ matched PBMCs taken from early, <6 weeks (6W) vs. late, >10 weeks (10W) timepoints after anti-PD-1 treatment (n = 7); ∗ denotes p <0.05; NS denotes non-significant 2-tailed p values by Wilcoxon signed-rank test. (B) Heatmap showing DEGs comparing pre-vs. on-treatment matched tumour tissues in responders to immunotherapy (n = 6). Selected DEGs of interest are highlighted. (C) Top 10 significantly enriched functional pathways of upregulated DEGs in matched on-treatment vs. pre-treatment tumour tissues from the responders. Horizontal-axis represents the -log10(Benjamini) adjusted p value and colour gradient represents enrichment fold. (D) Analyses of immune subsets from matched PBMCs taken before (Pre-Tox) and at the point of irAE manifestation (Tox) after anti-PD-1 treatment (n = 6); ∗ and n.s. denotes p <0.05 or non-significant 2-tailed p values by Wilcoxon signed-rank test. APCs, antigen-presenting cells; DEGs, differentially expressed genes; irAE, immune-related adverse event; MAITs, mucosal invariant T cells; MDSCs, myeloid-derived suppressor cells; NK, natural killer; PBMCs, peripheral blood mononuclear cells; TEM, effector memory T; Tox, toxicity; Tregs, regulatory T cells.
To link our observations in the blood to events in the TME, we conducted bulk tissue RNA-seq on pre- and 1 week on-treatment tumour biopsies from 10 ICB-treated patients with HCC (6 Res, 4 Non-Res) (Table S1). DEG analysis comparing on- vs. pre-treatment tumours from Res (Table S7) revealed upregulation of genes related to T-cell activation (GZMA, GZMH) and antigen presentation (HLA-related genes) (Fig. 5B), the same genes that were also upregulated in CXCR3+CD8+ TEM cells and APCs (Fig. 4A and S5A). On-treatment enriched functional pathways from Res included antigen presentation, T-cell co-stimulation, leukocyte chemotaxis, and IFNγ-mediated signalling (Fig. 5C), many of which were again common functional pathways enriched in both cDC1 and CXCR3+CD8+ TEM cells (Fig. 3, Fig. 4B); suggesting that these immune cells are recruited to the tumour tissue following ICB, particularly in Res. Moreover, it was intriguing to note the enrichment of CXCL9, CXCL10 and CXCL11 in Res (Fig. 5B), the key chemokines that bind to CXCR3, further supporting tumour recruitment of CXCR3+CD8+ TEM cells in Res. In contrast, Non-Res showed a completely different set of DEGs that were unrelated to immune activation (Fig. S7B).
Since depletion of CXCR3+CD8+ TEM cells was also related to irAEs (Fig. 2C), we compared their frequencies in matched on-treatment blood samples taken before (Pre-Tox) and during or close to (±2 weeks) irAEs (Tox). We found that CXCR3+CD8+ TEM cells were indeed significantly depleted at the point of Tox manifestation (Fig. 5D), suggesting their recruitment to the tissue. Once again, these data highlight the importance of CXCR3-mediated migration of CXCR3+CD8+ TEM cells in the manifestation of response and irAEs.
TNFR2 inhibition uncouples response and toxicity to anti-PD-1 ICB
Finally, our scRNA-seq data demonstrated that distinct TNF signalling pathways related to Res and Non-Tox (Fig. 4C,D) could be harnessed to uncouple response and irAEs upon ICB. We investigated this in mice inoculated with hepatoma cells via hydrodynamic tail-vein injection
(A) Mice with HCC induced by hydrodynamic tail-vein injection of Hepa1-6 cells were randomly assigned to 6 treatment groups and treated bi-weekly for 2 weeks from day 7 to 21 (n = 5 mice per group). (B) Representative images of livers harvested from the mice at Day-21. (C) Tumour nodules from the mice. NIL, no tumours from all mice treated with anti-PD-1+anti-TNFR2. (D) Liver/body weight ratio of the mice from each condition. (E) Frequencies (%) of CD8 T cells and CD69+ active CD8 T cells in non-tumour liver tissues of the mice. (F) Frequencies (%) of CXCR3+CD8+ T cells and CD11c+MHCII+XCR1+ cDC1 cells in tumour tissues of the mice. NA, no tumours from anti-PD-1+anti-TNFR2 treatment group. (G) Frequencies (%) of CXCR3+CD8+ T cells and cDC1 cells in non-tumour liver tissues from the mice. (C–F) Boxplots show median and interquartile range. ∗, ∗∗ n.s. denote 2-tailed p <0.05, p <0.01 or non-significant p values by unpaired Mann-Whitney U test. cDC1, type 1 conventional dendritic cells; HCC, hepatocellular carcinoma.
At harvest on day 21, all mice receiving combination treatments showed significant reduction in tumour nodules, especially those treated with anti-PD-1+anti-TNFR2, which displayed no tumour burden (Fig. 6B,C). We also observed a significantly higher liver-to-body weight ratio in the mice treated with the anti-PD-1+anti-TNFR1 combination (Fig. 6D) with no significant differences in mouse body weight (Fig. S8A) and reduced tumour burden in this group (Fig. 6C), suggesting liver hypertrophy and inflammation. The higher TNFR1 expression observed in Non-Tox (Fig. 4D and S6C), indicating its role in preventing irAEs, corroborates the enhanced toxicity observed in mice treated with the anti-PD1+anti-TNFR1 combination. This is further supported by increased CD8+ T-cell infiltration, especially of pro-inflammatory CD69+ activated CD8+ T cells, in the non-tumour liver tissue (Fig. 6E and S8B) and enhanced colonic CD4+ T-cell infiltration, indicating colitis and intestinal inflammation
(Fig. S8C,D). Enhanced toxicity was not observed in the anti-PD-1+anti-TNFR2 combination, which displayed the greatest tumour control (Fig. 6C-E and S8C,D), further strengthening our hypothesis that the differential blockade of TNFR1 or TNFR2 combined with anti-PD-1 therapy can uncouple response and irAEs.
The selective enhanced response following TNFR2 inhibition could stem from the preferential expression of TNFR2 on highly immunosuppressive Tregs.
To validate this, we sorted Tregs and non-Tregs from PBMCs, adjacent non-tumour liver and tumour tissues from patients with HCC (Fig. S9A). We found significantly higher expression of TNFRSF1B (TNFR2), but not TNFRS1A (TNFR1) in Tregs compared to non-Tregs in tumour-infiltrating leucocytes (Fig. S9B). TNFRSF1B expression was also higher in Tregs from tumour-infiltrating leucocytes compared to Tregs from PBMCs or non-tumour liver-infiltrating leukocytes (Fig. S9B). These findings demonstrated the specificity of TNFR2 expression on Tregs from HCC tumours, which upon selective inhibition, could enhance anti-tumour response but not systemic toxicity.
Furthermore, we observed intratumoural enrichment of CXCR3+CD8+ T cells and CD11c+MHCII+XCR1+ cDC1 in the mice treated with anti-PD-1, which was further enhanced by the anti-PD-1+anti-TNFR1 combination that corresponded to enhanced tumour control (Fig. 6F and S8B). This corroborates our human data that indicated the recruitment of these cells to tumours in Res (Fig. 5A-C). Notably, the anti-PD-1+anti-TNFR1 combination group, which displayed enhanced irAEs, also displayed a significantly higher infiltration of CXCR3+CD8+ T cells in the non-tumour liver tissue (Fig. 6G and S8B), again validating our observation that these cells are recruited to irAE sites (Fig. 5D).
Thus, using this model, we validated anti-PD-1 and anti-TNFR2 as a novel ICB combination strategy for HCC that results in superior responses without enhanced irAEs.
Discussion
Biomarkers of response to ICB have been explored previously in HCC,
Here, we identified circulating CXCR3+CD8+ TEM cells and CD11c+HLADRhi APCs, which are potentially recruited to the TME upon treatment, as biomarkers for response to anti-PD-1 ICB in patients with HCC. In support to our findings, previous studies in melanoma and lung cancer suggested that peripheral biomarkers like CD14+ classical monocytes and central memory-to-effector T-cell ratios were associated with favourable clinical outcomes following anti-PD-1 therapy.
Circulating T cell subpopulations correlate with immune responses at the tumor site and clinical response to PD1 inhibition in non-small cell lung cancer.
none have explored the immunological trajectories spanning both response and irAEs. In our current study, we found that CXCR3+CD8+ TEM cells with tissue-recruitment capability contributed to both response and irAEs, and demonstrated that local tumour inflammatory cues, specifically the upregulation of the chemokine ligands CXCL9, 10 and 11 upon ICB, induce their recruitment. Our findings are further supported by a recent meta-analysis across several solid tumours, where CD8A and CXCL9 were identified as predictors of response to ICB.
Finally, based on predicted cell-cell communications between CXCR3+CD8+ TEM cells and other immune cells, we found distinct pathways involving TNFR1 and TNFR2 that could be harnessed to uncouple response from irAEs in anti-PD-1 ICB therapy. The benefits of anti-TNF treatment in enhancing the efficacy of ICB through modulation of potential irAEs have recently been discussed.
However, our current data further demonstrated that TNFR1 and TNFR2 each govern distinct pathways underlying response and irAEs. The TNF-TNFR2 interaction was enriched in Res and was largely driven by the increased expression of TNFα on CXCR3+CD8+ TEM rather than TNFR2. Moreover, TNFR2 inhibition was recently proposed as a novel immunotherapeutic strategy that could confer enhanced anti-tumour efficacy due to the preferential expression of TNFR2 on highly immunosuppressive Tregs;
making it a potential immune checkpoint target and a promising candidate for combination immunotherapy, as evidenced in our current HCC murine study.
Conversely, the TNF-TNFR1 interaction was enriched in Non-Tox in our study, indicating it may play an important role in preventing toxicity. Indeed, the enhanced toxicity observed following TNFR1 inhibition in our murine study is consistent with a previous study that linked TNFR1 inhibition to the expansion of pathogenic T-helper 1 and 17 pro-inflammatory cells in mice with collagen-induced arthritis.
further emphasize the need to distinguish the underlying mechanisms of the TNFR1 and TNFR2 receptors in immunotherapy for HCC. The complex effects of these 2 receptors highlight the potential of selective TNFR2 inhibition as a promising immunotherapeutic strategy to uncouple anti-tumour efficacy from autoimmune toxicity in combination with anti-PD-1 ICB for HCC.
In conclusion, we identified early peripheral predictors of response in patients with HCC receiving anti-PD-1 therapy and provided valuable insights on the interface between response and irAEs, which could be harnessed to enhance clinical response without exacerbating the toxicity of HCC treatment.
Abbreviations
APC, antigen-presenting cell; cDC1, type-1 conventional dendritic cells; CTLA4, cytotoxic T lymphocyte-associated protein 4; CyTOF, cytometry by time-of-flight; DEGs, differentially expressed genes; G, grade; GZM, granzyme; HCC, hepatocellular carcinoma; ICB, immune checkpoint blockade; irAEs, immune-related adverse events; KR, South Korea; NCI CTCAE, National Cancer Institute Common Terminology Criteria for Adverse Events; Non-Res, non-responders; Non-Tox, non-toxicity; ORR, objective response rate; PBMCs, peripheral blood mononuclear cells; PD-1, programmed cell death 1; PD-L1, programmed death-ligand 1; PFS, progression-free survival; Res, responders; scRNA-seq, single-cell RNA sequencing; SG, Singapore; TEM, effector memory T; TME, tumour microenvironment; TNF, tumour necrosis factor; TNFRSF, tumour necrosis factor receptor superfamily; Tox, toxicity; Treg, regulatory T cells.
Financial support
This work was supported by the National Medical Research Council (NMRC), Singapore (ref numbers: NMRC/OFLCG/003/2018, NMRC/TCR/015-NCC/2016, NMRC/CSA-SI/0013/2017, NMRC/CSA-SI/0018/2017 and NMRC/STaR/020/2013).
Authors’ contributions
S. Chuah obtained and analysed data and prepared the paper. J. Lee recruited patients and involved in clinical study design. Y. Song performed the murine HCC work. H.D. Kim and K. Bang obtained and analysed clinical data from Korea cohort. M. Wasser performed CyTOF data analysis and discussed the data. N. Kaya analysed the tissue transcriptomic data and discussed the data. Y.J. Lee and S.H. Jeon were involved in sample collection and FACS data analysis of the Korea cohort. S.D/O. Suthen, S. A’Azman, G. Gien and C.J. Lim processed the samples and performed experiments. C. Chua, S. Hazirah, H.K. Lee and J.Q. Lim assisted in transcriptomic data collection and analysis. T.K.H. Lim and J. Yeong prepared and provided tissue samples and discussed the data. J.M. Chen assisted in scRNA-seq data analysis. E.C. Shin supervised and designed the FACS data analysis of the Korea cohort. S. Albani designed the CyTOF pipeline and discussed the data. W. Zhai collected and analysed the tissue transcriptomic data. C. Yoo recruited Korea cohort and discussed the data. H. Liu supervised and provided murine HCC data. S.P. Choo and D. Tai designed the clinical study, recruited patients and discussed the data. V. Chew designed and led the study, performed the analysis and prepared the paper.
Data availability statement
Raw scRNA-seq data has been deposited at the European Genome-phenome Archive (EGA) under accession number EGAS00001004843.
Conflict of interest
The authors declare no conflicts of interest that pertain to this work.
Please refer to the accompanying ICMJE disclosure forms for further details.
Acknowledgments
The authors would like to thank all participating patients, all members of TII, members of Dr. Chen Jinmiao’s team from SIgN and the clinical research coordinators from NCCS and AMC for their contributions. Acknowledgement is also extended to Dr. Jessica Tamanini from Insight Editing London for scientific and language editing of this manuscript.
Supplementary data
The following are the supplementary data to this article:
Nivolumab in patients with advanced hepatocellular carcinoma (CheckMate 040): an open-label, non-comparative, phase 1/2 dose escalation and expansion trial.
Circulating T cell subpopulations correlate with immune responses at the tumor site and clinical response to PD1 inhibition in non-small cell lung cancer.
00233-1&id=ga1.jpg){kind=link}
00233-1&id=gr1.jpg){kind=link}
00233-1&id=gr2.jpg){kind=link}
00233-1&id=gr3.jpg){kind=link}
00233-1&id=gr4.jpg){kind=link}
00233-1&id=gr5.jpg){kind=link}
00233-1&id=gr6.jpg){kind=link}