Patient characteristics and comparison of cytokine levels between cancer and control groups

The clinicalpathological characteristics of the 65 patients at baseline were summarized in Table 1. The male displayed higher circulating levels of IL-6, mitogen-inducible gene (MIG) and CCR than female. No significant differences in cytokine levels were observed in terms of smoking or drinking history. The predominant cancer types were lung cancer [22(33.8%)] and gastroesophageal cancer [16(24.6%)]. 34 (52.3%) patients were administered PD-1 plus chemotherapy, while 14 patients (21.5%) received PD-L1 plus chemotherapy. Of note, 20 (54.1%) patients displayed positive expression of PD-L1 and exhibited elevated levels of CCL2. Tumor response to ICIs therapy, assessed using RECIST 1.1 criteria, showed an objective response rate (ORR) of 30.7% and a disease control rate (DCR) of 87.1%. No significant differences in terms of cytokines were observed between different treatment response groups. There was no difference in age (P = 0.177) and sex (P = 0.062) between the cancer and control groups. The levels of circulating PD-L1, CXCL10, MIG, HGF, CCL-2, and IL-6 were significantly higher in the cancer group compared with those of the control group. (P = 0.012, P = 0.022, P < 0.001, P < 0.001, P < 0.001, and P = 0.030, respectively) (Fig. 1).

Table 1 Clinicalpathological characteristics of the patients (mean ± sd)Fig. 1

The difference between cancer (n = 65) and healthy group (n = 10) in terms of cytokines. The line in the middle of the box shows the mean value. Error bars show the interquartile range. (MIG: P < 0.001, HGF: P < 0.001, CCL2: P < 0.001, IL-6: P = 0.03, PD-L1: P = 0.012, CXCL10: P = 0.022, LAG-3: P = 0.761, GRANB: P = 0.068, IL-18: P = 0.462, CTLA4: P = 0.9441)

Correlation analysis of cytokines and nutritional indexes

A heat-map of the correlation between the soluble cytokines and nutritional indexes were presented in Fig. 2. It’s essential to highlight that the correlation coefficient of CCR with NRI is 0.27 (P = 0.030). The circulating level of CCR was negatively correlated with that of LAG-3 (R = -0.303, P = 0.014) and HGF (R = -0.311, P = 0.012). Likewise, NRI was negatively correlated with PD-L1 (R = -0.423, P < 0.001), IL-6 (R = -0.288, P = 0.02), and CXCL10 (R = -0.306, P = 0.013).LAG-3 was positively correlated with PD-L1 (R = 0.407, P < 0.001), CTLA4 (R = 0.371, P = 0.001), MIG (R = 0.402, P < 0.001), IL-18 (R = 0.321, P = 0.009), CXCL10 (R = 0.349, P = 0.004), CCL2 (R = 0.272, P = 0.028), IL-6 (R = 0.371, P = 0.003) and HGF (R = 0.294, P = 0.018). CCL2 was positively correlated with PD-L1 (R = 0.251, P = 0.044), CTLA4 (R = 0.274, P = 0.027), MIG (R = 0.264, P = 0.034), HGF (R = 0.337, P = 0.006), CXCL10 (R = 0.354, P = 0.004), IL-6 (R = 0.444, P < 0.001) and GRANB (R = 0.435, P < 0.001). The correlation coefficient of HGF and IL-6 is 0.687.The detail of correlation analysis of cytokines and CCR was shown in Supplementary Table S1 and Figure S1.

Fig. 2

The heatmap showing the correlation between cytokines, CCR, and NRI. The color shows the degree of correlation

Associations between cytokines and clinical benefits

We compared serum cytokines and CCR concentrations in DCB and NCB groups. The NCB group had elevated serum concentrations of HGF (2365 pg/mL vs. 1769 pg/mL, P = 0.006) and IL-6(16.6 pg/mL vs. 7.195 pg/mL, P = 0.001) compared with the DCB group (Table 2). Univariate analysis showed that low HGF and low IL-6 were significant prognostic factors for DCB. Multivariate logistics regression analysis revealed that the low level of IL-6 tends to independently predict DCB (P = 0.062, Table 3). The highest AUC (0.743) for DCB was observed in patients with IL-6 alone.The AUCs for DCB in patients of either IL-6low and/or HGFlow levels (n = 39), and IL-6low and HGFlow (n = 25) were 0.729 and 0.629, respectively, which were lower than the AUC of IL-6 alone. (Supplementary Figure S2).

Table 2 Differences of cytokines in patients between NCB and DCB groupsTable 3 Univariate and multivariate logistic regression analysis for NBC/DBCPrognostic value of cytokines and nutritional indexes

We next tested whether survival curves for PFS and OS were stratified by the levels of the 10 cytokines and CCR using the median values as cut-offs. Higher levels of circulating IL-6 and HGF were significantly associated with shorter PFS and OS (P = 0.002 and P < 0.001 for IL-6, P = 0.023 and P = 0.029 for HGF). Patients with CCR levels above the median showed longer PFS and OS (P = 0.031 and P = 0.008 respectively, Fig. 3). Furthermore, we performed ROC analysis to determine the utility of HGF and IL-6 in predicting survival. HGF was determined at cut-off values of 6-month, 12-month, and 18-month OS, and yielded AUCs of 0.728, 0.630, and 0.590 respectively, indicating that the 6- month OS cutoff possessed greater predictive value. These cutoffs were compared with IL-6, which generated AUCs of 0.739, 0.680, and 0.641 at 6-month, 12-month, and 18-month OS cutoffs, respectively. These results demonstrated that IL-6 was superior to HGF in predicting survival (Fig. 4A, B). Notably, to further explore the predictive abilities of these biomarkers, we combined the patients into the following three groups: both IL-6 and HGF low group; either IL-6 or HGF high group; and both IL-6 and HGF high group. Survival curves for PFS and OS were clearly stratified into the two distinct groups, with the worst PFS and OS in the IL-6 and HGF high group (P = 0.002 and P = 0.005, Fig. 4C, D). The AUCs for 6-month, 12-month, and 18-month OS in patients with IL-6low and HGFlow levels were 0.654, 0.640 and 0.595, respectively. The AUCs for 6-month, 12-month, and 18-month OS in patients with IL-6low and/or HGFlow levels were 0.791, 0.655 and 0.636, respectively (Fig. 4E, F). Finally, we analyzed the data using the Cox proportional hazards model with known risk factors for OS and PFS. As shown in Table 4, HGF IL-6 and CCR were shown to be possibly related to OS and PFS in the univariate analysis. When adjusting for HGF, IL-6 or CCR alone, all were independent prognostic factors (PFS: P = 0.028, P = 0.010, and P = 0.016, respectively, OS: P = 0.043, P = 0.003, and P = 0.026, respectively, Tables 5 and 6). However, when adjusting for 3 factors simultaneously including HGF, IL6, and CCR, only IL-6 and CCR were independent factors for prognosis, indicating that the prognostic effect of HGF could be obscured by IL-6 and CCR (Supplement Table S2).

Fig. 3

Kaplan–Meier curve for PFS and OS between patients with high HGF A, D, IL6 B, E, or CCR C, F and those with low HGF, IL6, or CCR

Fig. 4

ROC curves of HGF A and IL6 B predicting 6, 12, and 18 months survival. Kaplan–Meier curves for OS C and PFS D on the basis of HGF and IL-6 levels classified as both low, either high, or both high. E ROC curves of IL-6low and HGFlow predicting 6, 12, and 18 months survival. F ROC curves of IL-6low and HGFlow predicting 6, 12, and 18 months survival

Table 4 Univariate COX regression analysis for PFS and OSTable 5 Multivariate COX regression analysis for PFSTable 6 Multivariate COX regression analysis for OSAssociations between cytokines and irAEs

We next explored cytokines in relation to clinical appearance of irAEs. At the time of analysis, 11 of these patients were identified as having grade 2 irAEs. Patients in DCB group were more prone to occur irAEs (Fig. 5A, B). Low level of lymphocyte activation gene-3 (LAG3) was correlated with the occurrence of irAEs (Fig. 5C). Patients experiencing grade 2 irAEs also expressed high levels of C–C motif chemokine ligand 2 (CCL2) than patients with grade 1 irAEs (Fig. 5D).

Fig. 5

Cytokines in relation to irAEs. A The frequency, severity, and type of irAEs. B Proportion of patients who experienced different grade of irAEs on the basis of DCB and NCB. C Proportion of patients with high or low level LAG3 on the basis of grade of irAEs. D Comparisons of serum CCL2 levels according to the grade of irAEs