Intratumoral CD3 and CD8 T-Cell Densities Associated with Relapse Free Survival in HCC

Patients and database

The records of 65 consecutive patients with stage I to IV HCC who underwent primary tumor resection at the Medstar/Georgetown University Hospital/Lombardi Comprehensive Cancer Center (Washington, DC) between 2006 and 2015 were reviewed. Post-operative treatment was in accord with generally accepted guidelines and the mean duration of follow-up was 39.7 months (range 9 -84 months). RFS was defined as the length of time from resection of the primary tumor to documented disease progression. Patient tumor samples were obtained from the Ruesch Center for the Cure for Gastrointestinal Cancers' Tissue Bank at Lombardi. Patients provided consent for tissue banking prior to surgical resection of their HCC. A secure database with proper safeguards was constructed for the management of patient data. An Institutional Review Board approved the ethical, legal and social implications of the project.

Immunohistochemistry and semi-quantitative analysis

Pathologic slides from paraffin blocks prepared with surgical tissue specimen were stained with monoclonal antibodies to CD3 (polyclonal rabbit anti-human, Dako North America Incorporated, Carpintaria, CA, Catalog #A0452), CD8 (monoclonal mouse anti-human, Dako North America Inc, Carpintaria, CA, Catalog #M7103), PD-1 (EH12.1 from BD Bioscences/Pharmingen, San Jose, CA, Catalog Number: 561273), PD-L1 (SP142, Spring Biosciences; Atlanta, GA, Catalog #, M4420; SP263, Roche/Ventana Medical Systems, Tucson, AZ, Catalog Number: 790-4905), PD-L1_IHC_22C3 (companion diagnostics kit antibody from Dako North America, Inc. Carpinteria, CA.), and β-catenin (Millipore, Catalog Number: 05-665). All slides were then stained with hematoxylin and eosin (H&E). Slides stained for CD3⁺, CD8⁺, PD-1⁺, and PD-L1⁺ cells were first evaluated by a clinical pathologist. The tumor interior (TI), invasive margin (IM) and noncancerous liver parenchyma were marked for biomarker imaging. Expression of β-catenin expression was determined by intensity (0, +1, +2, or +3) and distribution (+1 as less than 10% of cells, +2 as 10%-50% of cells, and +3 as > 50% of cells).

Image capture and quantification

After the TI and IM were reviewed, images were captured in areas where TIL density was focally high, as well as in regions where the TIL density was representative of the TI and IM as a whole; this was done under high power magnification (20×). Images were captured as a spectral cube with a Nikon E600 Epifluorescence microscope and Nuance FX Multiplex Biomarker Imaging. For both CD3⁺ and CD8⁺ stained slides, two images of high cell density each were taken of the TI and IM. Computer-assisted image analysis was conducted using ImageJ v1.48, a public domain, Java-based, image processing program developed at the National Institutes of Health. Five built-in functions of ImageJ were used to measure the number of positively stained cells per tissue surface unit in square millimeters (See Supplementary Fig. S1 for a flowchart of the image analysis process, Supplementary Fig. S2 for an example of images obtained using computer-assisted evaluation of infiltrate densities, and Supplementary Fig. S3 for ImageJ instructions and plug-in code).

The median immune cell density was used to stratify patients into groups based on the degree of tumor infiltration. This method was independently validated by Galon et al. in his evaluation of the Immunoscore in CRC, and was thus adopted for the current study [21]. The cutoff threshold for CD3⁺ and CD8⁺ cell density was determined to be 273 cells/mm² and 217.5 cells/mm², respectively (Supplementary Fig. S4). Based on the established threshold, each patient was given a binary score (0 as low, 1 as high) for each immune cell type (CD3⁺ and CD8⁺) in each tumor region (TI and IM). An Immunoscore for each patient was obtained by summation of the four binary score values; the scale being from 0 to 4. Five patient groups were defined:patients with low densities of CD3⁺ and CD8⁺ T cells in both tumor regions (All-Low) were classified as Im0; patients with one high (1-High) density for one marker were classified as Im1; and patients with two (2-High), three (3-High), and four (All-High) among these two markers were classified as Im2, Im3, and Im4, respectively. Patients with a high degree of intratumoral immune cell infiltration were those with an Immunoscore (3+), and patients with a low degree of immune cell infiltration were those with an Immunoscore (≤ 2). The corresponding Immunoscore values were correlated with tumor recurrence and RFS, as well as compared to the pathological factors currently indicative of tumor invasiveness, such as vascular invasion, elevated AFP and advanced pTNM stage.

Statistical analysis

Descriptive statistics were used to summarize patients' demographic information. Chi-Square test or Fisher's exact test was used to examine the association between categorical variables. For analysis, patients were grouped by CD3⁺ density, CD8⁺ density, and Immunoscore. The nonparametric Kaplan-Meier methodology was used to analyze the RFS time and the log-rank test was used to compare survival curves. Multivariate logistic regression models were fit to look at the effect of CD3⁺ and CD8⁺ density on the odds of tumor recurrence, when adjusting for other significant covariates.

Demographics of HCC patients treated by primary surgery

The demographic and clinical characteristics of the registered cohort treated at MedStar Georgetown University Hospital are shown in Table 1. The mean age was 61 (range, 30-86), and the patient group was predominantly male (77%). Regarding etiology of HCC, 16.9% of patients had viral hepatitis B (HBV), 32.3% patients had viral hepatitis C (HCV), 12.3% of patients had co-infection of HBV and HCV, and 38.4% of the patients had neither virus. Using the Kaplan-Meier methodology, the estimated probability of relapse-free survival following HCC was found to be 63% at one year and 59% at 3 years, and the estimated probability of overall survival was 82% at one year and 57% at 3 years.

Pathology review and Immunohistochemcal characteristics

Slides stained with CD3 and CD8 antibodies revealed a wide spectrum of immune infiltrate densities in the TI and IM tumor regions (Fig. 1 and Fig. 2A,, B, E, and G). The H&E stain of the tumor is shown in Fig. 2D, and the negative control for CD3 and CD8 staining are shown in Fig. 2F and H, respectively. The pattern of TILs in each region fell into 3 categories: (1) dense nodular clusters of lymphocytes, (2) diffuse, evenly distributed lymphocytes (3) rare, or scattered lymphocytes (see Fig. 1A, B, and E, respectively). Clusters of lymphocytes are found from the IM to the TI in the first two patterns of TILs (Fig. 1C and D, migration marked by red arrows); whereas there are no obvious TILs between the IM and TI in the third pattern of TIL (Fig. 1F). Bivariate analysis showed no significant association between the pattern of TIL and TNM staging (P = 0.219 for CD3, 0.377 for CD8), level of differentiation (P = 0.054 for CD3, 0.283 for CD8), or vascular invasion (P = 0.096 for CD3, 0.104 for CD8) (Supplementary Fig. S5). Similarly, the pattern of TILs showed no association with relapse-free or overall survival in either immune cell type (Supplementary Fig. S6). Tumors with different staging, differentiation, and vascular invasion were found to have either pattern of TILs.

The number of PD-1 positive TILs varied from case to case, ranging from 0 to >1000 cells per 10× magnification field. A moderate number of PD-1 positive TILs (about 160 cells per 20× magnification field), were both within the TI and at the IM (Fig. 2K). In the 60 HCC samples evaluated, only two samples showed positive PD-L1 stain in cancer cells; 19 samples showed PD-L1 expression within or around the tumor, but in nonmalignant (noncancer) cells (Fig. 2C, I, and J), presumably mononuclear cells such as macrophages, dendritic cells and lymphocytes; and 39 samples showed no PD-L1 stain. The IHC analysis of PD-L1 expression using three antibodies (SP142, SP263, and PD-L1_IHC_22C3) gave identical results. PD-L1 expression showed a positive correlation with CD3⁺ and CD8⁺ cell densities. Fourteen of 31 HCC samples with high CD3⁺ density showed PD-L1 expression, while only five of 29 HCC samples with low CD3⁺ density showed PD-L1 expression (Fisher exact test P = 0.011). Fifteen of 31 HCC samples with high CD8⁺ density showed PD-L1 expression, while only four of 28 HCC samples with low CD8⁺ density showed PD-L1 expression (Fisher exact test P = 0.003). Similarly, CD3⁺ cell density showed a positive correlation with CD8⁺ cell densities. Twenty-two of 32 HCC samples with high CD3⁺ density showed high CD8⁺ cell density, whereas only ten of 31 HCC samples with low CD3⁺ density showed high CD8⁺ cell density (Fisher exact test P = 0.0055). IHC analysis of β-catenin expression showed a variable level of cytoplasmic, membranous and nuclear staining (Fig. 2L). Expression of β-catenin and CD3⁺ or CD8⁺ cell density was not correlated. (P= 0.432, P=0.227, respectively, as shown in Supplementary Fig. S7).

Association between clinical risk factors and recurrence

Established risk factors for HCC invasiveness include tumor size, tumor stage, tumor grade, serum AFP level, cellular differentiation, HCV or HBV infection, vascular invasion, and the presence of liver cirrhosis [26-29]. These elements were evaluated in our HCC patient cohort. However, bivariate analysis revealed that only tumor cell differentiation, grade, and the presence of vascular invasion were significantly correlated with HCC recurrence (Table 2). Patients with elevated serum AFP and advanced tumor stage showed an increased rate of recurrence; however the association was not statistically significant. Tumor stage did not correlate with recurrence, but correlated with vascular invasion and cellular differentiation (as shown in supplementary Fig. S8). Patients with well differentiated tumors experienced a significantly reduced incidence of HCC recurrence: 5% of well differentiated tumors recurred after primary resection compared to 33% and 48% for moderate and poorly differentiated tumors respectively (P = 0.011). Furthermore, patients with HCC whose tumors did not reveal vascular invasion also faced a reduction in the rate of recurrence: those exhibiting tumor vascular invasion recurred in 48% of cases compared to only 22% in those without vascular invasion (P= 0.040).

Impact of the immune response on recurrence and RFS

The densities of total (CD3⁺) and cytotoxic T (CD8⁺) TILs were examined in the TI and IM regions. Statistically significant associations were observed between the individual densities of CD3⁺ and CD8⁺ cells in the distinct tumor regions and frequency of HCC recurrence (Table 2). Patients with a high density of CD3⁺ immune infiltrates in the TI and IM experienced recurrence of their HCC in only 15% of cases compared to 44% in those with a low CD3⁺ cell density (P = 0.027). Similarly, 15% of patients with a high density of CD8⁺ immune infiltrates in the TI and IM experienced recurrence of their HCC compared to 45% in those with a low CD8⁺ cell density (P= 0.014). According to the Immunoscore grouping method outlined above, 27%, 14%, 16%, 16%, and 27% of the patients were classified as Im0, Im1, Im2, Im3, and Im4, respectively. The frequency of HCC recurrence in each Immunoscore subgroup was as follows: 65% for Im0, 22% for Im1, 10% for Im2, 10% for Im3, and 11% for Im4. This combined analysis revealed that patients with a high density of CD3⁺ and CD8⁺ cells in one or both tumor regions (TI or IM) experienced a significant reduction in the rate of HCC recurrence (P = 0.007). In keeping with the positive correlation between PD-L1 expression and CD3⁺ and CD8⁺ TIL density, the lack of PD-L1 expression also predicts higher risk of HCC recurrence. Fifteen of 39 HCC samples lacking PD-L1 expression recurred, while only 2 of 19 HCC samples with PD-L1 expression recurred (P=0.034).

The densities of CD3⁺ and CD8⁺ immune cell populations were also examined in relation to relapse-free survival times, our primary endpoint. Different permutations of CD3⁺ and CD8⁺ cell populations were used in the analysis: CD3⁺ density in each tumor region alone, CD8⁺ in each tumor region alone, Total CD3⁺ density in the TI + IM, Total CD8⁺ density in the TI + IM, and the combined Immunoscore measuring both cell types in both tumor regions (Figure 3). Kaplan-Meier survival curves for average CD3⁺ (Fig. 3A and B) and CD8⁺ (Fig. 3C and D) cell density in each tumor region alone were first assessed. When each cell population (CD3TI, CD3IM, CD8TI, CD8IM) was analyzed independently of one another, RFS time between patients with high and low immune cell density in either region were statistically significant different (CD3TI P = 0.044, CD3IM P< 0.001, CD8TI P = 0.002, CD8IM P = 0.01). RFS was then correlated with the total CD3⁺ cell density, total CD8⁺ cell density and the corresponding Immunoscore value. The Kaplan-Meier curves (Fig. 3E) revealed a strong positive association between high total CD3⁺ cell density TI + IM and prolonged RFS (P = 0.006). The Kaplan-Meier curves (Fig. 3F) for total CD8⁺ cell density TI + IM yielded a similar but less significant association with RFS (P = 0.01). Lastly, Kaplan-Meier curves for the Immunoscore were constructed to provide a more comprehensive picture of the two immune cell populations as they relate to survival. The curves (Fig. 3G) show three distinct patient groups with statistically significant differences in RFS times (P = 0.002). Patients with an Immunoscore of 0 experienced the worst post-operative outcome, with all patients having disease recurrence within two years of follow-up. The Immunoscore remained significantly associated with recurrence and RFS, whereas several clinicopathologic factors including TNM staging did not reach significance. PD-L1 expression was also examined in relation to relapse-free survival times. The RFS time between patients with and without PD-L1 expression was significantly different (Fig. 3H, P= 0.029).

Immunoscore predicts risk of recurrence

Analysis of possible links between immune infiltration and clinical risk factors is shown in Table 3. Viral hepatitis B and C were the predominant etiologies for HCC in our study population; however the densities of CD3⁺ and CD8⁺ tumor-infiltrating lymphocytes among subjects with or without HBV and HCV infection were not significantly different. Similarly, vascular invasion, cellular differentiation and cancer stage showed no significant association with CD3⁺ and CD8⁺ cell densities in the TI and IM (Table 3). The expression of PD-L1 showed no significant association with vascular invasion, cellular differentiation, or tumor stage (Table 3). Since vascular invasion and differentiation were the only clinical risk factors that strongly correlated with HCC recurrence in our bivariate analysis, we focused on these factors in the multivariate analysis.

In order to assess the impact of CD3⁺ and CD8⁺ cell densities on the tumor recurrence, when adjusting for the significant covariates in the bivariate analysis (specifically, vascular invasion and differentiation), multivariate logistic regression models were fit. The adjusted odds ratio (OR) of HCC recurrence was 5.8 (95% CI 1.6-21.8) for CD3⁺ cell density. Similar significant results were obtained for CD8⁺ density (the adjusted OR was 3.9 (95% CI 1.1-14.1)). The results showed that after adjusting for other covariates, CD3⁺ and CD8⁺ cell densities remained as significant risk factors for predicting tumor recurrence. Thus, the Immunoscore could predict risk of recurrence independent of the only risk factors that demonstrated predictive value (vascular invasion and cellular differentiation) in our data set. As a result, the Immunoscore seems to be a highly significant prognostic factor in the cohort of patients with HCC treated by primary surgery.

Given the great impact of CD3⁺ and CD8⁺ cell densities, and PD-L1 expression on tumor recurrence and RFS, we examined the relation between CD3⁺ and CD8⁺ cell density and OS. We found that TIL or PD-L1 expression density and OS in patients after curative resection were not significantly correlated (Fig. 4).