Single-cell immune signature for detecting early-stage HCC and early assessing anti-PD-1 immunotherapy efficacy

Abstract

Background: The early diagnosis of hepatocellular carcinoma (HCC) can greatly improve patients' 5-year survival rate, and the early efficacy assessment is important for oncologists to harness the anti-programmed cell death protein 1 (PD-1) immunotherapy in patients with advanced HCC. The lack of effective predicting biomarkers not only leads to delayed detection of the disease but also results in ineffective immunotherapy and limited clinical survival benefit.

Methods: We exploited the single-cell approach (cytometry by time of flight (CyTOF)) to analyze peripheral blood mononuclear cells from multicohorts of human samples. Immune signatures for different stages of patients with HCC were systematically profiled and statistically compared. Furthermore, the dynamic changes of peripheral immune compositions for both first-line and second-line patients with HCC after anti-PD-1 monotherapy were also evaluated and systematically compared.

Results: We identified stage-specific immune signatures for HCC and constructed a logistic AdaBoost-SVM classifier based on these signatures. The classifier provided superior performance in predicting early-stage HCC over the commonly used serum alpha-fetoprotein level. We also revealed the treatment stage-specific immune signatures from peripheral blood and their dynamical changing patterns, all of which were integrated to achieve early discrimination of patients with non-durable benefit for both first-line and second-line anti-PD-1 monotherapies.

Conclusions: Our newly identified single-cell peripheral immune signatures provide promising non-invasive biomarkers for early detection of HCC and early assessment for anti-PD-1 immunotherapy efficacy in patients with advanced HCC. These new findings can potentially facilitate early diagnosis and novel immunotherapy for patients with HCC in future practice and further guide the utility of CyTOF in clinical translation of cancer research.

Trial registration numbers: NCT02576509 and NCT02989922.

Keywords: biomarkers; immunotherapy; liver neoplasms; tumor.

Conflict of interest statement

Competing interests: HS and WY are both cofounders of, WC and XZ are the scientific consultants of, HS is the CEO of, and TW is employed by Zhejiang Puluoting Health Technology Co., Ltd. The authors declare the following competing financial interests: Zhejiang University and Zhejiang Puluoting Health Technology Co., Ltd. are jointly filing Chinese invention patents for the Anti-PD-1 Monotherapy Effectiveness Evaluation System Based on Peripheral Blood Immune Cell Atlas (202111513101.5) and An Early Screening System for Hepatocellular Carcinoma Based on Peripheral Blood Immune Cell Atlas (202111513095.3).

© Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Figures

Figure 1

Single-cell CyTOF analysis reveals major peripheral immune compositions of HDs, patients with HH, and patients with HCC. (A) Experimental design and analysis flow for single-cell CyTOF. (B) The t-SNE plots of 25,000 immune cells randomly sampled from the HD, HH, and HCC groups, colored by major immune cell subsets. (C) The t-SNE plots of 25,000 immune cells randomly sampled from the HD, HH, and HCC groups, colored by patient groups. (D) Contours of cell density distributions in each group. (E) Comparisons of the percentages of major immune cell subsets across groups. Unpaired Student t-test is used for statistical analysis. (E) *p

Figure 2

Characterization of peripheral lymphocytes in HDs, patients with HH, and patients with HCC. (A) Heatmaps of the normalized median expressions of four identified B-cell clusters (top) and two identified NK-cell clusters (bottom). The barplot of the relative frequencies for B-cell clusters and NK-cell clusters is displayed by gray bars on the right. (B) Comparisons of the percentages of identified B-cell clusters (left) and NK-cell clusters (right) across groups. (C) Heatmap of the normalized median expressions of 22 identified T-cell clusters. The barplot of the relative frequencies of T-cell clusters is displayed by gray bars on the right. The annotated lineage and functional cell types are color-labeled on the right. (D) The t-SNE plot of T cells for the HD, HH, and HCC groups, colored by the identified T-cell clusters. (E) Comparisons of the percentages of identified CD4+ T-cell and CD8+ T-cell functional subsets across groups. (F) Comparisons of the log2 (fold change) of the percentages of CD8+ effector T-cell cluster (top) and NK2 cluster (bottom) versus regulatory T-cell cluster across groups. (G) Comparisons of the percentages of identified CD8+ and CD4+ EM T-cell clusters across groups. (H) Histograms of the expressions of selected functional markers on identified CD4+ and CD8+ T-cell clusters. Unpaired Student t-test is used for statistical analysis. (B, E, F, G) *p<0.05, **p<0.01, ***p<0.001. BCLC-A, Barcelona Clinic Liver Cancer Stage A; BCLC-C, Barcelona Clinic Liver Cancer Stage C; DN, double negative; FC, fold change; HCC, hepatocellular carcinoma; HD, healthy donor; HH, hepatic hemangioma; IL, interleukin; NK, natural killer; Treg, regulatory T cell; t-SNE, t-distributed stochastic neighbor embedding.

Figure 3

Altered immune interactions in different stages of HCC and integrated immune features for HCC early detection. (A) Heatmaps of the Pearson correlation coefficients between immune cell subsets across groups. (B) Correlations of the percentages of selected immune cell subsets across groups and the regression line with 95% CI. (C) The PCA projection of total PBMC samples, colored by groups, and each ellipse plot represents the CI of 95% confidence coefficient for individual groups. (D) The PCA projection of HD, Hh, and BCLC-A HCC samples, colored by groups, and each ellipse plot represents the CI of 95% confidence coefficient for individual groups. (E) Confusion matrix of classification result of 47 samples (F) The ROC curves and AUC values of trained AdaBoost-SVM classifier or classified by using AFP levels with a cut-off of 20 ng/mL. ACC, accuracy; AUC, area under the curve; AFP, alpha-fetoprotein; BCLC-A, Barcelona Clinic Liver Cancer Stage A; HCC, hepatocellular carcinoma; HD, healthy donor; HH, hepatic hemangioma; NK, natural killer; PBMC, peripheral blood mononuclear cell; PCA, principal component analysis; FOR, false omission rate; FDR, false discovery rate; FPR, false positive rate; ROC, receiver operating characteristic; TNR, true negative rate; TPR, true positive rate.

Figure 4

Personalized peripheral immune changes in patients with HCC after ICB monotherapy. (A) The clinical event lines for 10 patients with BCLC-C HCC treated with anti-PD-1 monotherapy in two groups, colored by distinct clinical responses. (B) Percentage changes in tumor burden from baseline for 10 patients with BCLC-C HCC treated by anti-PD-1 monotherapy over time in two groups. (C) The t-SNE plot of 300,000 immune cells from the 10 patients with BCLC-C HCC, colored by major immune cell subsets. (D) Contours of cell density distributions of each patient based on the t-SNE plot (C). (E) The percentages of major immune cell subsets in durable treatment cycles for each patient in two groups. BCLC-A, Barcelona Clinic Liver Cancer Stage A; DC, dendritic cell; DCB, durable clinical benefit; HCC, hepatocellular carcinoma; ICB, immune checkpoint blockade; NDB, non-durable benefit; NK, natural killer; PD-1, programmed cell death protein 1; PD, progressive disease; PR, partial response; SD, stable disease; t-SNE, t-distributed stochastic neighbor embedding.

Figure 5

Detailed longitudinal changes of peripheral immune compositions for patients with DCB and NDB. (A) Diffusion maps based on the frequencies of circulating immune components for 10 patients with BCLC-C HCC treated with immunotherapy, colored by patients, and shaped by clinical responses. (B) The relative changes in the percentages of major immune cell subsets versus baselines of 10 BCLC-C patients during the treatment cycles. The lines are colored by patients and shaped by clinical responses. (C) Comparisons of the log2 (fold change) of the percentages of major immune cell subsets versus baseline between patients with DCB and NDB during the treatment cycles. (D) The relative changes in the percentages of the identified immune cell subsets versus baselines of 10 patients with BCLC-C during the treatment cycles. The lines are colored by patients and shaped by clinical responses. (E) Comparisons of the log2 (fold change) of the percentages of the identified immune cell subsets versus baseline between patients with DCB and NDB during the treatment cycles. (F) The PCA projection of the Spearman correlation coefficients between the relative changes of identified cell clusters and the treatment cycles. Point plots are labeled by patients and colored with DCB and NDB groups (top). The arrow length and direction represent the dominant cell clusters to the directions of PCs (bottom). (G) Heatmap of the Spearman correlation coefficients between the relative changes of identified cell clusters and the treatment cycles, colored by clinical responses on the right. Unpaired Student t-test is used for statistical analysis. (C, E) *p<0.05, **p<0.01. BCLC-C, Barcelona Clinic Liver Cancer Stage C; CDC, dendritic cell; DCB, durable clinical benefit; NDB, non-durable benefit; NK, natural killer; PCA, principal component analysis; PC, principle components.