Patient characteristics

A total of 60 advanced NSCLC patients with EGFR-TKI resistance were enrolled in the study (Fig. 1). The baseline characteristics of all participants were summarized in Table 1. The median age was 63.5 years old, and Most patients had no history of smoking (85%). All patients were diagnosed with lung adenocarcinoma except four patients with carcinoma not otherwise specified (NOS). The most common subtypes of EGFR mutations were 19DEL and L858R, and three patients with G719X were included. The higher proportions of patients with bone metastases and brain metastases were probably attributed to the fact that all patients had underwent at least one kind of systematic treatment. It was worth noting that only 16.7% of patients in this study had acquired T790M, probably caused by preference selection of clinicians as some research data suggested that negative T790M may be a favorable factor for immunotherapy in NSCLC patients who failed EGFR-TKI treatment [21]. There were 80% of patients who received ICI plus chemotherapy at second or third line, and only 36.7% of patients had received other systematic treatment before immunotherapy. As PD-L1 is not required to be tested for such patients before immunotherapy, the expression of PD-L1 was not evaluated in 86.7% of patients in this study, so we did not make subsequent analysis on PD-L1 expression.

Efficacy and influencing clinical factors of ICI plus chemotherapy

The median follow-up time was 19.7 months, and the median PFS of total population was 6.4 months (95%CI: 4.3–8.6), the overall ORR and DCR were 21.7%, and 86.7%, respectively, and the CB rate as we previously defined was 31.6%.

To explore the potential clinical factors associated with outcome of ICI plus chemotherapy, we firstly compared the baseline characteristics between CB and NB groups, and found no difference in all common characteristics including age, sex, smoking history, pathology, distant metastasis, gene mutation subtype, T790M mutation, PD-L1 expression, and immunotherapy treatment line. Previous studies reported that baseline NLR, eosinophils, platelets, and platelet-to-lymphocyte ratio (PLR) were associated with immunotherapy efficacy in advanced NSCLC patients [22, 23]. Therefore, we further compared these baseline hematologic indicators between the two groups, and found no significant difference in the absolute count of lymphocytes (Fig. 2A), neutrophils (Fig. 2B), eosinophils (Fig. 2C), platelets (Fig. 2D), and the NLR (Fig. 2E), while the PLR in CB group was significantly lower than that in NB group (P = 0.045) (Fig. 2F).

Fig. 2

The association of baseline hematologic indicators with efficacy of ICI plus chemotherapy. Comparisons of baseline hematologic indicators between patients with clinical benefits (CB) and patients with non-benefit (NB) from ICI plus chemotherapy, including absolute lymphocyte count (A), absolute neutrophil count (B), absolute eosinophil count (C), platelet count (D), the ratio of neutrophil-to-lymphocyte (NLR) (E), and the ratio of platelet -to-lymphocyte (PLR)

To analyze the impact of such clinical factors on PFS of ICI plus chemotherapy, we firstly converted the above continuous variables, such as the absolute counts of lymphocytes, neutrophils, eosinophils, platelets, NLR, and PLR, to categorical variables according to the optimal cut-off values of 1.07, 3.5, 215, 4, and 200, respectively, calculated by X-Title. Univariate analysis of PFS in the total population showed that some variables, including the number of distant metastatic organs, liver metastasis, other treatments before immunotherapy, the lines of immunotherapy, lymphocytes, NLR, and PLR, were significantly related with PFS (Table 2). All variables with P < 0.1 in univariate analysis was included in the multivariate analysis and taking collinearity between any two of these variables into account meanwhile, we used Cox regression with forward stepwise (likelihood ratio) to perform multivariate analysis. The results suggested that only “liver metastasis” (P < 0.001) and “PLR” (P = 0.003) were independently associated with PFS of EGFR-TKI resistant NSCLC patients receiving ICI plus chemotherapy.

Table 2 Univariate and multivariate analysis of PFS of all patients receiving ICI plus chemotherapyProfiling of immune cells in peripheral bloodCharacteristics of patients collected PBMC

Among all 60 patients in this study, PBMC samples were collected from 24 patients at the baseline and after 2 cycles of ICI plus chemotherapy. The median interval between the two sampling was 50 days with range of 25–69 days.

In order to find out whether there was large population selection bias in these 24 patients with collected specimens, we summarized their baseline characteristics and found that almost all clinical characteristics were consistent with those of the whole study population (Supplement Table 2), suggesting that there was no significant bias in this population. Moreover, all these 24 patients had no smoking history, and 9 of them were divided into the CB group, while 15 of them were divided into the NB group according to our definition before. The baseline characteristics in the two groups were balanced. The median PFS of all these 24 patients was 5.32 months (95%CI, 3.97–6.68). Both univariate and multivariate analyses suggested that liver metastasis was the independent factor for poor prognosis in patients with EGFR-TKI resistant NSCLC (P = 0.033) (Supplement table 3).

Proportion of immune cell subsets

A total of 19 antibodies were designed to cluster common peripheral blood immune cells according to their development lineage in this study and only 19 immune cell subsets were finally identified after screening out subsets that could not be detected in more than 10% of the patients. Firstly, we compared the proportions of immune cell subsets in patients before and after receiving ICI plus chemotherapy and found that there was no significant change after 2 cycles of treatment (Fig. 3A). Then, we respectively compared the difference in the proportions of immune cell subsets in the CB and NB groups, and found that: at baseline, compared with NB group, the CB group had a significantly higher proportion of effective CD4+T cell (E4, CD4+CD45RA+CD197−), with marginal statistical significance (P = 0.055) (Fig. 3B), while after 2 cycles of treatment, there was no significant difference in the proportions of immune cell subsets (Fig. 3C). In addition, there was also no significant difference in the dynamic changes of immune cell subsets proportions between the CB and NB groups (Fig. 3D).

Fig. 3

Association of immune cell subsets with efficacy of ICI plus chemotherapy. Comparisons of immune cell subsets proportions in different subgroups: Indicators comparison between patients pre- and after-treatment (A). Pre-treat indicators comparison between patients in CB and NB groups (B). On-treatment indicators comparison between patients in CB and NB groups (C). Dynamic changes of indicators comparison between patients in CB and NB groups (D). Abbreviations: CB, clinical benefits, NB, non-benefit. *P < 0.1

Expression of immune checkpoint proteins

Many studies suggested that PD-1 expression on some subsets of peripheral immune cells, like CD8+ T cell [24], CD4+ T cell [25], NK cells [26], etc. were associated with immunotherapy outcomes. Thus, we added PD-1 antibody into our panel. Besides, as novel immune checkpoint molecules on T cells have been discovered continuously and previous studies have suggested their alternative roles for immune escape to PD-1/PD-L1 pathway [27], we would like to supplement some antibodies against these novel ICPs into our multi-color panel. Initially, some of these checkpoint targets under clinical trials [28] were selected, such as lymphocyte activation gene 3 (LAG-3), T cell immunoglobulin and mucin domain-containing protein 3 (TIM-3), V-domain immunoglobulin suppressor of T cell activation (VISTA), and T cell immunoglobulin and ITIM domain (TIGIT). However, TIM-3 was given up due to the unsatisfying staining in preliminary experiment. In addition, we noticed that CD25 [29] and human leukocyte antigen DR (HLA-DR) [30] in our designed panel could not only be used as clustering marker for Tregs and dendritic cells (DC) respectively, but also be regarded as ICPs on immune cell subsets, so we have also evaluated their expressions on immune cell subsets as ICPs using MFI. The overall expressions of these ICPs are shown in supplement Fig. 2. As can be intuitively seen from the figure, the ICPs expressions on immune cell subsets seem higher in the CB group than those in NB group, no matter at baseline or after 2 cycles of treatment.

To find the specific differential variables, we made further statistical analysis. Firstly, we compared the expression levels of ICPs on immune cell subsets before and after receiving ICI plus chemotherapy in all patients, and the data showed that after treatment, HLA-DR on central memory CD8+T cell (CM8, CD8+ CD45RA−CD197 +) were significantly elevated, while PD-1 on effective memory CD4+T cell (EM4, CD4+ CD45RA−CD197 −) was significantly decreased (Fig. 4A). Secondly, we made further comparison between CB group and NB group, and found that: at baseline, CD25 expression on CM8 and DC, and LAG-3 expression on effective memory CD8+T cell (EM8, CD8 + CD45RA−CD197−) were significantly higher in CB group than in NB group (Fig. 4B), while after treatment, CD25 expression on CD8+T/EM8/natural killer (NK) cells were significantly higher in CB group than in NB group (Fig. 4C). Lastly, the dynamic changes comparison suggested that the decrease of TIGIT on Tregs was more significant in NB group than in CB group, and the VISTA on Th1 was increased in CB group but decreased in NB group (Fig. 4D).

Fig. 4

Association of immune checkpoint proteins with efficacy of ICI plus chemotherapy. Comparisons of immune checkpoint proteins (ICPs) expressions on immune cell subsets in different subgroups: Indicators comparison between patients pre- and after-treatment (A). Pre-treat indicators comparison between patients in CB and NB groups (B). After-treat indicators comparison between patients in CB and NB groups (C). Dynamic changes of indicators comparison between patients in CB and NB groups (D). Abbreviations: CB, clinical benefits, NB, non-benefit. *P < 0.1, ** P < 0.05

Construction of model to predict efficacy of ICI plus chemotherapy

The above data suggested that some clinical characteristics and features of peripheral blood immune cells were associated with the efficacy of EGFR-TKI resistant NSCLC patients receiving ICI plus chemotherapy. Therefore, Lasso regression was used to systematically analyze the potential influencing factors, and all differential variables between CB and NB group with P < 0.1 were included. The results showed that when the parameter λ was the minimum value, variables including platelet, pre-treatment E4, and on-treatment CD25 on NK cells were confirmed to be meaningful to predict CB response.

Then, Logistic regression was used to establish a prediction model. Internal cross-validation data suggested that when the model score threshold was 0.598, the sensitivity and specificity of the model were 62.5% and 100%, respectively, with area under curve (AUC) = 0.817 (Fig. 5A). A Nomo diagram was drawn to display this model, and the values of relevant quantitative variables were standardized by z-score transformation (Fig. 5B).

Fig. 5

Prediction model to efficacy of ICI plus chemotherapy. A is the receiver operating characteristic (ROC) curve of the prediction model, with an AUC area of 0.817. When the cut-off value of model score is 0.598, the sensitivity is 62.5% and the specificity is 100%. B is the Nomo diagram to display the prediction model. Factors contributed to the prediction model include PLR, pre-treatment E4 (E4_1), and after-treatment CD25 on NK cells (NK_25_2). Lower value of PLR and higher values of E4_1 and NK_25_2 bring higher points, and higher points predict higher probability to acquire clinical benefit (CB) from ICI plus chemotherapy for patients with EGFR-TKI resistance