Patients’ characteristics
The characteristics of the ninety-six lung adenocarcinoma patients with both 18F-FDG PET and RNA sequencing data are described in Table 1. The median age of the patients was 68 years, and the median follow-up period was 1399 days. The proportion of male patients (n=67) was about 2.3 times higher than that of female patients (n=29), and seventy-five patients (approximately 78%) had a history of smoking. Clinical staging was done according to the 8th edition of the AJCC, and 76% of the patients were classified as clinical stages I and II. The most prevalent location of the primary tumor was right upper lobe, affecting 33 patients.
Association between molecular subtypes and clinical variables, immune scores and 18F-FDG PET parameters
The patients were grouped according to the molecular subtype (PP, TRU, and PI) based on RNA sequencing data, with 30, 41 and 25 patients in each group, respectively. We examined whether there were differences in clinical variables and immune scores between the three groups. We found that sex and stage distribution were significantly different among the three groups (P = 0.004, and 0.002, respectively). In a post hoc analysis, the proportion of women was significantly higher in the TRU group than in the PI group (P = 0.002). Also, the TRU group contained more early-stage patients compared to the PP and PI groups (P = 0.02, and 0.043, respectively) (Table 2, Figure A). Meanwhile, the CYT score of the PI group was significantly higher than those of the PP and TRU groups (P = 0.001, and 0.001, respectively), which indicates that the degree of anticancer immunity is highest in the PI group (Table 2, Figure 2B). We also compared the 18F-FDG PET parameters between the groups, finding that TLR and COV were significantly different between the groups (P = 0.038 and 0.001, respectively). In a post hoc test, TLR tended to be higher in the PI group than in the TRU group (P = 0.061). Moreover, the PI group had significantly higher COV scores than those of the TRU group (P = 0.001) (Table 3, Figure 2C-F). Taken together, the PI group demonstrated the highest levels of anticancer immunity, 18F-FDG uptake and metabolic heterogeneity among the three groups. Given that 18F-FDG uptake can be determined by both infiltrating immune cells and malignant cells, we explored the association between the 18F-FDG PET parameters and immune cell scores further.
Correlation between 18F-FDG PET parameters and immune cell scores
A correlation analysis was conducted considering the 18F-FDG PET parameters, CYT scores, and immune cell scores. With a pair for which absolute value of correlation coefficient |r| > 0.4 and P < 0.05 considered to be significant, there were no pairs of immune cell scores and 18F-FDG PET parameters (Figure 3A, Supplementary Table 1). We also explored the correlation between the immune cell scores and 18F-FDG PET parameters in each genomic subtype group. In the TRU group, Th2, activated dendritic cell, eosinophil, and mast cell scores showed weak negative correlations with the 18F-FDG PET parameters. In the PP group, the gamma delta T cell (Tgd) score showed a weak negative correlation with TLR. In contrast, there were pairs which showed a weak positive correlation in the PI group [Tgd vs. TLR, Tgd vs. TLG, macrophages vs. TLR, and macrophages vs. TLG] (Figures 3B-D, 3A, Supplementary Table 1). CYT showed weak correlations with TLR and COV (r = 0.308 and 0.268 and P = 0.01 and 0.029, respectively). In the PI group, TLR and TLG showed weak positive correlations with CYT. However, there were no significant correlations between the CYT scores and the 18F-FDG PET parameters in the TRU and PP groups (Figure 4B). Overall, there was a trend toward weak negative correlations between immune cell scores and 18F-FDG PET parameters in the TRU and PP groups. However, a weak positive correlation was found for the PI group. This suggests that 18F-FDG uptake in the PI group is more affected by immune cell infiltration than in the other groups.
Prognostic value of 18F-FDG PET parameters, CYT, and immune cell scores
We assessed the abilities of the 18F-FDG PET parameters, CYT scores, immune cell scores and molecular subtypes with regard to predicting patients’ overall survival. Among the molecular subtypes, there were no significant differences in the overall survival rates between the three groups (P = 0.3). However, when the TRU group and the others were compared, the TRU group tended to have better clinical outcomes (P = 0.1) (Figure 5), consistent with a previous report [30]. Among the 18F-FDG PET parameters, low TLR and low COV scores were associated with better overall survival (P = 0.01, and 0.04, respectively) (Figure 6, Supplementary Table 2). Also, a low CYT score was associated with better prognosis (P = 0.05) (Figure 7A, Supplementary Table 2). Among the immune scores, high T follicular helper cell (TFH) and high CD8 T cell scores were associated with better prognosis (P = 0.036 for CD8 T cells; P = 0.005 for TFH cells) (Figures 7B-C, Supplementary Table 2).
Finally, we selected TLR from the 18F-FDG PET parameters and the TFH score from the immune cell scores, which showed the most robust parameters in their groups for predictions of clinical outcomes (lowest P value in a log rank test) to explore the additive value of combining the immune cell score and this 18F-FDG PET parameter. The patients were divided into four groups, as follows: 1) patients with high TLR and high TFH scores, 2) patients with high TLR and low TFH scores, 3) patients with low TLR and high TFH scores, and 4) patients with low TLR and low TFH scores. A Kaplan Meyer analysis demonstrated good stratification of the four groups, and patients in the high TLR group with a low TFH score had the worst clinical outcome (P = 0.002) (Figure 8A). The five-year survival rate was visualized according to different stages in the groups according to the TLR and TFH scores (Figure 8B). As expected, the five-year survival rate declined at higher stages and when the TLR was higher. Interestingly, the high TFH group had a longer five-year survival rate than the low TFH group at a low stage, but the trend was reversed at a high stage. (Figure 8B, Table 4). Furthermore, TLR and TFH had independent prognostic values according to a multivariate Cox regression analysis, even after adjustments with clinicopathologic features and with each other (P = 0.016 for TLR; P = 0.017 for TFH; P = 0.024 for adjusted TLR; P = 0.047 for adjusted TFH) (Table 5).