An Immuno-Clinic score model for evaluating T cell immunity and predicting early antiviral therapy effectiveness in chronic hepatitis B

Antiviral cytokine and exhausted T cell profiles

To display an overall picture of T cell immunity in naïve CHB, we investigated the detailed immune phenotypes of antiviral cytokines and inhibitory molecules in different disease phases. The disease phases of the training cohort were divided into immune tolerant (IT, n=17), immune active (IA, n=120), inactive carriers (IC, n=20), and gray zone (GZ, n=72) phases according to the American Association for the Study of the Liver guidelines, Supplementary Table 1 [2]. The frequencies of antiviral cytokines (IFN-γ, TNF-α, IL-2) and inhibitory molecules (PD-1, LAG-3, Tim-3, CTLA4, and LAIR-1) of T cells are shown in Figure 1. A significant difference was observed in the frequency of IFN-γ produced by CD4+ T cells, with more frequency in the IA patients than in the IT patients (P = 0.007). No significant difference in the levels of T cells producing IFN-γ was found between the IA, GZ, and IC groups. Also, TNF-α+ CD4+ and CD8+ T cells were not different in CHB patients for all disease phases. TNF-α+ CD8+ T cells of patients in all disease phases were lower than the healthy controls (P = 0.001,0.004, 0.04, and 0.01 in cohorts of IT, IA, IC, and GZ, respectively, Figure 1A). IL-2+ CD4+ and CD8+ T cells were not statistically different between the IT, IA, IC and GZ patients.

Figure 1. Expression of three pairs of antiviral cytokines and five pairs of inhibitory molecules by CD4+ and CD8+ T cells from naïve CHB patients. (A) Expression of antiviral cytokines IFN-γ, TNF-α, and IL-2 by CD4+ and CD8+ T cells derived from the indicated patient groups. The levels were compared among patients in the IT, IA, GZ, IC phases and healthy controls. (B) Expression of inhibitory molecules PD-1, Tim-3, LAG-3, CTLA4, and 2B4 were measured on CD4+ and CD8+ T cells derived from the indicated patient groups. The levels were compared among patients in the IT, IA, GZ, IC phases and healthy controls. Differences between multiple groups were evaluated by the Wilcoxon rank sum test. Data is presented as the median (indicated by a red line). *P < 0.05.

To analyze whether CHB had distinct frequencies of exhausted T cells at different disease phases, we measured the inhibitory receptor PD-1, LAG-3, Tim-3, CTLA4, and LAIR-1 expression on CD4+ and CD8+ T cells. Figure 1B shows lower expressions of CTLA4 and LAG-3 in patient groups of IT and IC, and higher expressions of 2B4 in IA, IC and GZ patients than that of the healthy controls. Comparable levels of PD-1 and Tim-3 by T cell subsets were found among the CHB cohort regardless of disease phase and healthy controls.

Taken together, these data suggest that antiviral cytokines combined with co-inhibitory molecules affects the immuno-dominance in HBV infection across CHB disease phases. Therefore, we explored an evaluation model to distinguish comprehensive T cell immunity with both antiviral cytokines and co-inhibitory molecules in CHB.

Clustering of CHB T cell immunity into immuno-high or immuno-low

To compare the T cell immunity status of individual CHB patients, we used K-means cluster analysis. Before clustering, the correlation was tested among 16 immune markers for CD4+ and CD8+ T cells, including five pairs of co-inhibitory molecules and three pairs of antiviral cytokines. The data shows that 13 immune markers were independent of each other (r<0.8) and were then used to construct a T cell evaluation model (Figure 2A). Clustering was used to classify 229 patients into two groups of immuno-high or immuno-low. We found significantly greater levels of antiviral cytokines by CD4+ and CD8+ T cells—including IFN-γ, TNF-α, and IL-2—in the immuno-high group than in the immuno-low group (Supplementary Table 2). In contrast, we observed that the expression of exhausted T cell molecules—including PD-1 and 2B4—was higher in the immune-low group. These results suggest that the clustering model could help differentiate between high and low T cell immunity in CHB. The number of patients in the immuno-high group (47, 21%) was less than immuno-low group (182, 79%), indicating that most of the CHB patients’ comprehensive T cell immunity was low (Figure 2B).

Figure 2. K-means cluster of CHB patients with three pairs of antiviral cytokines (IFN-γ, TNF-α, and IL-2) and five pairs of inhibitory molecules (PD-1, Tim-3, LAG-3, CTLA4, and 2B4) produced by CD4+ and CD8+ T cells. (A) Correlations among the three pairs of antiviral cytokines and five pairs of inhibitory molecules produced by CD4+ and CD8+ T cells were measured by the Spearman correlation. P<0.05 is colored, and the pseudocolors indicate correlation levels from negative (-1) to positive (1), ranging from a weak (white) to strong (red or blue) association strength. (B) Representative image of CHB patient clustering with three pairs of antiviral cytokines and five pairs of inhibitory molecules simultaneously produced by CD4+ and CD8+ T cells. The green balls represent patients of immuno-high (expressing high levels of antiviral cytokines and low levels of inhibitory molecules), and the red balls represent patients of immuno-low (expressing low levels of antiviral cytokines and high levels of inhibitory molecules), respectively. All of the 229 patients were divided into these two groups.

ICS model

Using clustering plots, we classified patients into immuno-high and immuno-low groups and constructed the ICS model. First, we assessed the clinical characteristics that correlated with immune-markers: Age, body mass index (BMI), HBeAg, Fibroscan value, Log HBV DNA, quantitative hepatitis B surface antigen (qHBsAg), and HBV genotype (Figure 3A). Second, we incorporated the clinical variables into the ICS model. Discrimination was evaluated by analyzing the area under the receiver-operator characteristic (ROC) curve (Figure 3B). We selected the optimal cut-off for the scores based on the clinical variables using the Youden Index. The area under the curve (AUC) of 0.801 (95% CI 0.703-0.900) was generated to discriminate individuals with high T cell immunity from low T cell immunity, with a sensitivity of 72.7% and a specificity of 72.2%. Third, a validated formula was derived for use in clinical practice: ICS= -1.577 + 0.181 Age + 0.088 BMI - 0.373 Fibroscan + 1.363 Sex - 0.250 Log HBV DNA - 0.188 qHBsAg + (1.026 Genotype C + 1.084 Genotype NA - 0.408 Genotype O).

Figure 3. Development of an ICS model for the evaluation of comprehensive T cell immunity. (A) Correlations of antiviral cytokines (IFN-γ, TNF-α, and IL-2) and inhibitory molecules (PD-1, Tim-3, LAG-3, CTLA4, and 2B4) produced by CD4+ and CD8+ T cells with 18 clinical-virological characteristics were measured. Seven clinical-virological variables (Age, Fibroscan value, BMI, HBV genotype, Log HBV DNA, HBeAg, and qHBsAg) showed significant association with different immune variables. The Spearman correlation or Wilcoxon rank sum test was used to test the correlation. P < 0.05 is shown in color, and the red hue depth represents the degree of statistical difference. NS=not significant. (B) ROC curve analysis of different ICS models in all CHB patients. Four ROC curves are shown by different clinical-virological variables based on classifications from all CHB patients, and the ROC curve for the selected ICS model is displayed in green (+ Sex). AUC, sensitivity, and specificity of this ROC curve were 0.801 (95% CI 0.703-0.900), 0.727, and 0.722, respectively, and the cut-off value was 0.442. (C) Re-evaluation of T cell comprehensive immunity with the ICS model in patients at different CHB phases. Numbers in the blue and red proportion indicate the percentage of ICS-low and ICS-high patients, respectively.

The accuracy of the ICS model was 77.7% by cross-validation (leaving out one validation). The cut-off value for discriminating T cell immuno-high from immuno-low was 0.442. Table 1 shows the scores for each item in the formula. Re-evaluation of T cell immunity was performed by the ICS model for patients in different phases of CHB (Figure 3C). In IT patients, 15% were ICS-high and 85% were ICS-low, similar to the percentage of patients in the IA phase where 17% of patients were ICS-high and 83% were ICS-low. In GZ patients, the percentage of ICS-high was 56%, while in IC patients, the percentage of ICS-high was 92%—the highest among the four CHB phases.

Application of the ICS model to predict antiviral therapy efficacy

We applied the model to predict antiviral therapy efficacy by longitudinally analyzing the ICS 48 weeks after therapy in 126 CHB patients, including 77 treated with nucleoside analogs [NAs, 70 with Entecavir (ETV), 6 with Telbivudine (LDT), and 1 with Tenofovir (TDF) and 49 with PEG-interferon (PEG-IFN)] (Supplementary Table 3). Figure 4A, 4B shows a progressive increase in the proportion of ICS-high patients during treatment.

Figure 4. Application of the ICS model in predicting antiviral therapy efficacy. (A) Evaluation of Immuno-Clinic scores at different time points during antiviral therapy within NAs, PEG-IFN, or total CHB patients. Data are presented as mean ± SEM. (B) The percentage of ICS-high or ICS-low patients at different time points during antiviral therapy are shown. Numbers in the blue and red proportion indicated the percentage of ICS-low and ICS-high patients, respectively. (C) Comparison of the ICS values between patients with different antiviral efficacy. In the upper panel, ΔICS at weeks 4 or 8 relative to week 0 were calculated. Data are presented as mean ± SEM. ***P < 0.001; *P < 0.05; NS=not significant. In the lower panel, pairs of the original Immuno-Clinic scores for each individual at weeks 4 or 8 and week 0 are shown.

Analyses of serologic response revealed that 50 out of 126 patients achieved HBV DNA <20 IU/ml at week 48. Longitudinal data showed a significant increase in ICS as early as week 4, and was found in patients achieving HBV DNA <20 IU/ml at week 48, compared with those who did not (P<0.001, Figure 4C). By 48 weeks, 82 out of 126 patients achieved normalization of ALT levels. A significant increase in ICS as early as week 4 was found in patients achieving normalization of ALT levels, compared with those who did not (P<0.001, Figure 4C). By weeks 24 and 48, 64 and 68 out of 126 patients achieved a >0.5 log HBsAg IU/ml reduction, respectively. A significant increase in ICS at week 8 was found in patients with a >0.5 log HBsAg IU/ml reduction by week 24 or week 48 compared with those who did not (P=0.001, P=0.015, respectively, Figure 4C). By week 48, 25 out of 70 patients had HBeAg loss after treatment. However, longitudinal data showed that no significant change of ICS at week 4 was found to correlate with HBeAg loss (Figure 4C).

The main purpose of this study was to construct a practical ICS model to evaluate patients’ comprehensive T cell immunity and use it to predict antiviral efficacy, however, given that both of the HBV DNA and HBsAg levels were included into the ICS model, we compared the prediction efficacy of ICS with HBV DNA and HBsAg levels. The results showed that although the ROCs predicting for HBV DNA and HBsAg decline constructed by HBV DNA and HBsAg themselves had more advantages than ICS, the AUC for ALT normalization at week 48 constructed by Δ0-8w ICS (0.714) was higher than that constructed by Δ0-8w HBV DNA (0.710) and HBsAg levels (0.648). The ROC curve constructed by the Δ0-4w ICS nearly predicted the HBeAg sero-conversion at week 48 (p=0.068, Supplementary Table 4).

These results indicate that a robust increase of ICS as early as week 4 in CHB patients may predict a better outcome of antiviral therapy.

Subjects

Adult patients were recruited from a hepatitis clinic at the Third Affiliated Hospital of Sun Yat-sen University. Written informed consent was obtained from all patients. The study was approved by the Sun Yat-sen University Institute Review Board. Patients were excluded if they received antiviral treatment (interferon or nucleoside analogs [NAs]) within the previous 6 month, had human immunodeficiency virus (HIV), hepatitis C virus, hepatitis D virus co-infection, end-stage liver insufficiency, autoimmune disorders, fatty liver disease, immunosuppressive treatment, cirrhosis, or malignancies.

Two groups of patients were used in the study. The study group (training cohort) included 229 patients with different disease phases for the purpose of developing the above-mentioned T cell profile model. The external group (application cohort) included 126 patients receiving antiviral therapy to confirm this model. The clinical characteristics of these subjects are listed in Table 2 and Supplementary Table 3. In the external group, sequential HBV DNA, HBsAg, HBeAg, and ALT were tested at 4, 8, 12, 24, and 48 weeks after the start of treatment.

Clinical and serologic parameters

HBV e antibody (HBeAb) was tested using commercial kits (Abbott Laboratories, North Chicago, IL). HBV genotype was determined by direct sequencing. Quantitative hepatitis B surface antigen (qHBsAg) was measured by the Elecsys HBsAg II Quant reagent kits (Roche Diagnostics, Indianapolis, IN) according to the manufacturer’s instructions. HBV core antibody (HBcAb) levels were quantified with a chemiluminescence immunoassay (Roche Diagnostics, Indianapolis, IN). Serum HBV DNA levels were measured by Roche COBAS Ampliprep/COBAS Taqman HBV test v2.0 (dynamic range from 20 to 1.7E+08 IU/ml, Roche Molecular Diagnostics, Branchbug, NJ). Fibrosis was defined by liver stiffness measurements (Fibroscan, Echosens, Paris, France).

Cell surface and intracellular staining

Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll density gradients. For the phenotypic analysis, PBMCs were stained with FITC-LAG-3, FITC-CD8, PE-2B4, PE-CD8, PE-CTLA4, PE-CF594-CD3, PE-CY7-CD4, BV421-PD-1, V450-CD8 (BD Biosciences, Franklin Lakes, NJ), and APC-Tim-3 (eBioscience, San Diego, CA, USA). For cytokine analysis, PBMCs were stimulated with a Leukocyte Activation Cocktail (eBioscience, San Diego, CA, USA), at 37° C for 4 h prior to intracellular staining using Pharmingen's staining protocol. Anti-human monoclonal antibodies against FITC-IFN-γ, PE-TNF-α (eBioscience), and PB-IL-2 (Biolegend, San Diego, CA, USA) were used. Corresponding isotype-matched controls were purchased from BD Biosciences and eBioscience. Data was acquired on a Gallios instrument (Beckman Coulter, Brea, CA, USA) and analyzed with FlowJo software (Ashland, OR, USA). The gating strategy for cytokines and exhaustion markers is shown in Supplementary Figure 1.

Statistical analysis

We used a 6-step strategy to develop and validate the ICS model. In step 1, we plotted antiviral cytokines and T cell exhausted molecules in different CHB disease phases by the Wilcoxon rank sum test for continuous variables and χ2 test for categorical variables. In step 2, we tested correlations among 16 immune parameters to identify independent indicators that influenced T cell immunity using the Spearman correlation. In step 3, we used the K-means cluster analysis based on the above selected immune parameters to group parameters with similar patterns of T cell immunity (all the selected immune parameters were scaled in K-means cluster analysis). This method allowed us to partition the data into two groups and develop a T cell evaluation model for classifying individual patients as having high or low T cell immunity. The model was checked through the comparison of immune parameters between the high and low T cell immunity groups. In step 4, the Spearman correlation or Wilcoxon rank sum test was used to test the correlation between the clinical variables and the 16 immune parameters, generating potential clinical variables to be used in the next step. In step 5, discrimination analysis was used to build the ICS models based on the above potential clinical variables and the immunity groups (high or low T cell immunity in step 3), resulting in the Immuno-Clinic scores. The Immuno-Clinic scores and the immunity groups were then used to build the receiver-operator characteristic (ROC) curve. We selected the optimal cut-off for the scores based on the clinical variables using the You den Index. The area under the curve (AUC), sensitivity, and specificity were also computed and compared between the different ICS models. Leave one out validation was also used to confirm the models. In step 6, the selected ICS model was applied to predict the efficacy of antiviral therapy in a longitudinal cohort of CHB patients. ICS model construction and all the other statistical tests were done with R software version 3.2.2 and SAS 9.2. Statistical significance was set at 0.05.