Analysis of Peripheral T Cell Profiling and Plasma Proteomics in Advanced NSCLC Patients Treated With Atezolizumab

ABSTRACT

We investigated immunologic biomarkers that predict outcomes of patients with previously treated metastatic non‐small cell lung cancer who were enrolled in the J‐TAIL study, a prospective, observational study of atezolizumab monotherapy. Of 262 patients participating in the J‐TAIL exploratory study, peripheral blood mononuclear cells were obtained from 51 patients and analyzed by T‐cell fractionation analysis using Helios mass cytometry and serum proteomics analysis. Following treatment with atezolizumab, an increase in programmed cell death‐1 (PD‐1)‐expressing CD8 and CD4 T‐cell populations was observed. A more pronounced increase in PD‐1 expression was seen in T cells from patients whose progression‐free survival (PFS) was 100 days or longer compared with those with shorter PFS. The proximity extension assay, which is highly sensitive multiplex analysis technology that combines antibody‐based affinity assays with next‐generation sequencing, showed a significant increase in FOXO1, possibly in response to precursor‐exhausted T‐cell population activation. Immune‐related adverse events were associated with a high percentage of PD‐1‐positive cells on effector memory CD8 T cells, which was thought to be accompanied by extremely high CD8 T‐cell activation. Further analysis distinguished poor prognosis populations with significant differences in CD62L_high Th7R and CXCR3⁺ component of Th7R (CXCR3⁺ Th7R) within the population with PFS < 50 days. Patients with low Th7R or CXCR3⁺ Th7R percentages prior to atezolizumab treatment had significantly poorer overall survival. These findings provide valuable insights regarding T‐cell kinetics and biomarkers in atezolizumab therapy and may offer promising directions for future research.

Trial Registration: UMIN Clinical Trials Registry: UMIN000033133 and UMIN000035567; ClinicalTrials.gov: NCT03645330

This study found immune markers that may predict how well patients with advanced lung cancer respond to atezolizumab. Higher levels of certain T cells were linked to better survival. Specific patterns suggested a worse prognosis for some patients.

Abbreviations

AUC

area under the curve

CI

confidence interval

CR

complete response

EMRA

effector memory cells re‐expressing CD45RA

HR

hazard ratio

ICI

immune checkpoint inhibitor

irAE

immune‐related adverse event

NSCLC

non‐small cell lung cancer

OS

overall survival

PBMC

peripheral blood mononuclear cell

PD

progressive disease

PD‐1

programmed cell death‐1

PD‐L1

programmed cell death‐ligand 1

PEA

proximity extension assay

PFS

progression‐free survival

PR

partial response

ROC

receiver operating characteristic

Tpex

precursor‐exhausted T cells

TTF

time to treatment failure

1. Introduction

Immune checkpoint inhibitors (ICIs) exhibit unprecedented long‐term survival benefits characterized by a tail plateau survival curve pattern [1, 2]. However, few patients experience such benefits, highlighting the need for reliable biomarkers. While tumor programmed cell death‐ligand 1 (PD‐L1) expression serves as a biomarker in lung cancer, its predictive performance remains limited, with recent studies suggesting that host cell PD‐L1 expression may be more relevant to treatment efficacy than tumor cell expression [3, 4, 5].

Recent advances in single‐cell analysis have identified key T‐cell populations crucial for sustained anti‐tumor immunity. These include precursor‐exhausted T cells (Tpex), a distinct CD8⁺ T‐cell population that maintains proliferative potential despite PD‐1 expression [6], and a novel CD4⁺ T‐cell metacluster associated with improved survival after PD‐1 blockade therapy (Th7R) [7, 8, 9]. Additionally, studies utilizing the proximity extension assay (PEA) method, a highly sensitive technique for measuring proteins in blood, have revealed promising proteomic signatures associated with treatment response [10, 11, 12].

Our study analyzed data from the J‐TAIL study [13], a prospective, multicenter observational study of atezolizumab in previously treated patients with advanced non‐small cell lung cancer (NSCLC). Through comprehensive T‐cell cluster and protein analyses using mass cytometry and PEA of peripheral blood, we aimed to identify biomarkers that could facilitate appropriate patient selection for atezolizumab treatment.

2. Material and Methods

2.1. Study Design

Between April and October 2019, 262 patients from 73 institutions provided written informed consent to participate in this exploratory biomarker analysis study of J‐TAIL. Of these, peripheral blood mononuclear cells (PBMCs) were obtained from 51 patients from 20 study sites. Both J‐TAIL and this study were registered at the UMIN Clinical Trials Registry (UMIN000033133 and UMIN000035567) and ClinicalTrials.gov (NCT03645330).

2.2. Patients and Treatment

The J‐TAIL eligibility criteria have been described previously [13]. Briefly, patients aged ≥ 20 years who had received prior systemic therapy for unresectable advanced or recurrent NSCLC and who were scheduled to receive atezolizumab monotherapy were included. Those deemed unsuitable for enrollment by the investigator were not included.

Eligible individuals received atezolizumab monotherapy, guided by the most recent package insert [14]. The physician made the choice to discontinue atezolizumab treatment, guided by the package insert and the Guidelines for the Promotion of Optimal Use [15].

2.3. Effectiveness Assessment

Disease progression and treatment response were assessed by the investigators according to RECIST version 1.1 [16], without confirmatory measurements. As J‐TAIL was an observational study, the timing and method of these assessments were determined at the discretion of the investigators and institutions.

2.4. PBMC Profiling

At baseline (PRE timepoint, before atezolizumab therapy) and before the third dose of atezolizumab (post‐atezolizumab), 8‐mL blood samples were collected using serum separation tubes (Supporting Information S1).

2.5. T‐Cell Fractionation Analysis

Mass cytometry analysis was performed on peripheral blood samples to profile the number of cells per T cell lineage and the percentage of functional molecules expressed in PBMCs. The relationship between the obtained immune cells and the therapeutic effect of atezolizumab was analyzed.

T‐cell fractionation analysis was performed using Helios (Standard BioTools, San Francisco, CA, USA) mass cytometry (Supporting Information S1). Cytobank (https://www.cytobank.org) software was used to analyze > 200,000 cells per sample. Details of the mass cytometry panel, including all antibodies, clones, and metal tags, are detailed in Table S1. Details of the gating strategy are shown in Figure S1.

2.6. PEA

PEA was used to comprehensively quantify the expression levels of 644 proteins in patients from whom serum blood samples could be obtained. The presence of potential statistical outliers was evaluated by assessing the interquartile ranges and sample median across all assays. Data were further investigated by principal component analysis. At baseline (PRE timepoint), before the second and third dose of atezolizumab (2nd and 3rd timepoint), and after the occurrence of an immune‐related adverse event (irAE), 1‐mL blood samples were collected using serum separation tubes (Supporting Information S1). Protein expression was comprehensively quantified using Immune Response, Immuno‐Oncology, Inflammation, Oncology II, Organ Damage, Cell Regulation, and Metabolism panels (Olink Proteomics, Uppsala, Sweden), with some proteins measured across multiple panels. The association between the efficacy/safety of atezolizumab and baseline biomarker levels (or changes from baseline) was evaluated. The investigators assessed all AEs for their potential relationship to treatment. Any AE suspected to be an irAE that occurred at least once during the observation period, defined as the time from the initiation of atezolizumab until 30 days after its discontinuation, was recorded. The highest grade for each irAE was determined per the National Cancer Institute Common Terminology Criteria for Adverse Events version 4.0.

2.7. Statistical Analysis

Descriptive statistics were used to summarize baseline patient characteristics using median (interquartile range or range) for continuous variables and n (%) for categorical variables. For overall survival (OS) and progression‐free survival (PFS), median time‐to‐event was calculated by the Kaplan–Meyer method, and 95% confidence intervals (CIs) were calculated using the Brookmeyer–Crowley method. Hazard ratios (HRs) and 95% CIs were calculated by Cox proportional hazards regression, and between‐group comparisons were performed by log‐rank test. For group comparisons of the percentage of T‐cell subpopulations, data are mean ± SEM unless otherwise indicated. Tests for differences between the two populations were performed using Welch's t‐test. The definition of each outcome is the same as in previous studies (Supporting Information S1) [13, 17].

Receiver operating characteristic (ROC) curve analysis was used to examine the biomarker performance of the percentage change in PD‐1‐positive cells in CCR7⁺CD45RA⁻CD8⁺ T cells from the first dose to before the third dose, and the percentages of CD62L_high Th7R and CXCR3⁺ component of Th7R (CXCR3⁺ Th7R). Detailed explanations of the ROC curves and determination of cutoff value are provided in the Supporting Information S1. The cut‐off value was the median.

Proteins showing between‐group differences were identified using pairwise comparisons. For each of the categorical variables and time points, protein‐wise Wilcoxon tests were performed to compare the groups. Unsupervised clustering was performed using the proteins that showed different levels in non‐responders (progressive disease [PD]) and responders (complete response [CR]/partial response [PR]) at baseline.

For the PBMC analysis, the significance level was 5% without adjustment for multiplicity due to the exploratory nature of the study. For the PEA analysis, p‐values were adjusted for multiple testing using the Benjamini–Hochberg method to account for the large number of comparisons. All statistical analyses were performed using GraphPad Prism software (v.8.0; GraphPad Software, San Diego, CA, USA) and R v4.2.2 (https://www.r‐project.org/).

3. Results

For the 51 patients whose PBMCs were analyzed, the median age was 70 years, 56.9% (29/51) were male, and 80.4% (41/51) had non‐squamous histology (Table 1).

TABLE 1.

Baseline demographic and clinical characteristics of exploratory T‐cell cluster analysis study.

Characteristic	n = 51
Age, years, median (range)	70 (41–85)
Sex, male, n (%)	29 (56.9)
Histology, n (%)
Squamous	10 (19.6)
Non‐squamous	41 (80.4)
Eastern Cooperative Oncology Group performance status, n (%)
0	24 (47.1)
1	24 (47.1)
2	3 (5.8)
Smoking history, n (%)
Former or current smoker	34 (66.7)
Never smoker	17 (33.3)
Driver gene alteration, n (%)
Epidermal growth factor receptor mutation	7 (13.7)
Programmed cell death‐ligand 1 tumor proportion score, n (%)
< 1%	14 (27.5)
≥ 1%, < 50%	18 (35.3)
≥ 50%	10 (19.6)
Unknown	9 (17.6)
Staging, n (%)
IIIB	3 (5.9)
IVA	12 (23.5)
IVB	20 (39.2)
Postoperative recurrence	13 (25.5)
Recurrence after definitive radiotherapy	3 (5.9)
Metastatic sites, n (%)
Brain	14 (27.5)
Bone	16 (31.4)
Liver	5 (9.8)
Adrenal gland	4 (7.8)
Number of prior treatment regimens, n (%)
1	24 (47.1)
2	9 (17.6)
3	8 (15.7)
4	3 (5.9)
5	4 (7.8)
≥ 6	3 (5.9)

The PFS Kaplan–Meier curve shows an inflection point ~100 days after the start of treatment (Figure 1A), consistent with the trend in the overall population of the J‐TAIL study population. Although few patients showed long‐term PFS, OS was relatively good, and median OS was not reached even after a mean follow‐up period of 356.7 days (Figure 1B). Based on the PFS analysis, we designated the group with PFS < 100 days (n = 31), in which PFS events occurred rapidly by day 100 after treatment, as the non‐responder group. In contrast, the PFS curve gradually decreased after day 100 in the group with PFS ≥ 100 days (n = 20), and this group was considered the responder group. The main biomarker exploration was assessed in these groups. These groups also showed a significant difference in OS, demonstrating the validity of considering the group with PFS ≥ 100 days to be the responder group (Figure 1C). Conversely, a tail plateau in OS was observed in the group with PFS < 100 days. This suggests that the group with PFS < 100 days also included cases that benefited from the anti‐tumor effects of atezolizumab treatment in terms of OS.

FIGURE 1.

Kaplan–Meier curves for (A) PFS, (B) OS, (C) OS stratified by PFS at 100 days, and (D) OS stratified by PFS at 50 days. HR, hazard ratio; OS, overall survival; PFS, progression‐free survival.

To specifically isolate the population with the poorest prognosis who likely derived no benefit from atezolizumab, we stratified patients into PFS < 50‐day (n = 13) and PFS ≥ 50‐day (n = 38) groups, based on the early progression pattern observed in the overall J‐TAIL study and analyzed OS. The cases forming the tail plateau of OS were only present in the group with PFS ≥ 50 days (Figure 1D); therefore, we performed an additional comparative analysis of biomarkers for the group with PFS ≥ 50 days and the group with PFS < 50 days. Baseline demographic and clinical characteristics of patients stratified by PFS are provided in Tables S2 and S3, respectively.

3.1. PD‐1 Expression on T Cells Following Atezolizumab Treatment

The phenomenon of increased PD‐1‐expressing cells on T cells following atezolizumab treatment was seen in both CD8⁺ and CD4⁺ T‐cell populations (Figure 2A,B). A more pronounced increase in PD‐1 expression was shown in the population with PFS ≥ 100 days than that with PFS < 100 days (Figures 3 and 4).

FIGURE 2.

PD‐1 expression in (A) central memory (CD45RA⁻CCR7⁺), effector memory (CD45RA⁻CCR7⁻), and EMRA (CD45RA⁺CCR7⁻) CD8⁺. (B) Th1 (CXCR3⁺CCR4⁻CCR6⁻), Th17 (CXCR3⁻CCR4⁺CCR6⁺), and Th1/17 (CXCR3⁺CCR4⁻CCR6⁺) CD4⁺ T cell populations following atezolizumab treatment. Paired plots show values at pre‐ and post‐treatment timepoints, with difference plots showing mean of differences. Adjacent difference plots show individual changes (post‐treatment—pre‐treatment); red line indicates mean of differences, blue dotted line represents line of no change (zero). PD‐1, programmed cell death‐1.

FIGURE 3.

PD‐1 expression in central memory (CD45RA⁻CCR7⁺), effector memory (CD45RA⁻CCR7⁻), and EMRA (CD45RA⁺CCR7⁻) CD8⁺ T cell populations following atezolizumab treatment. Paired plots show values at pre‐ and post‐treatment timepoints, with difference plots showing mean of differences. Adjacent difference plots show individual changes (post‐treatment—pre‐treatment); red line indicates mean of differences, blue dotted line represents line of no change (zero). PD‐1, programmed cell death‐1.

FIGURE 4.

PD‐1 expression in Th1 (CXCR3⁺CCR4⁻CCR6⁻), Th17 (CXCR3⁻CCR4⁺CCR6⁺), and Th1/17 (CXCR3⁺CCR4⁻CCR6⁺) CD4⁺ T cell populations following atezolizumab treatment. Paired plots show values at pre‐ and post‐treatment timepoints, with difference plots showing mean of differences. Adjacent difference plots show individual changes (post‐treatment—pre‐treatment); red line indicates mean of differences, blue dotted line represents line of no change (zero). PD‐1, programmed cell death‐1.

There was a significant change in the composition of the CD8⁺ T‐cell fraction after atezolizumab treatment only in the population with PFS ≥ 100 days. Specifically, there was a decrease in the effector memory fraction (CD45RA⁻CCR7⁻/CD8⁺ T cell) and an increase in terminal effector memory cells re‐expressing CD45RA (EMRA) (i.e., CD45RA⁺CCR7⁻/CD8⁺ T cells) at a rate corresponding to the effector memory decrease (Figure 5). At the 2nd timepoint, PEA showed a significant increase in FOXO1, possibly in response to Tpex population activation (Figure 6).

FIGURE 5.

Central memory (CD45RA⁻CCR7⁺), effector memory (CD45RA⁻CCR7⁻), and EMRA (CD45RA⁺CCR7⁻) CD8⁺ T‐cell expression after atezolizumab treatment in the populations with PFS ≥ 100 and < 100 days. Paired plots show values at pre‐ and post‐treatment timepoints, with difference plots showing mean of differences. Adjacent difference plots show individual changes (post‐treatment—pre‐treatment); red line indicates mean of differences, blue dotted line represents line of no change (zero). CM, central memory; EM, effector memory; PFS, progression‐free survival.

FIGURE 6.

Results of proximity extension assay from the 2nd timepoint. Plot illustrates differences in protein expression between patients with progression‐free survival (PFS) ≥ 100 days and those with PFS < 100 days. x‐axis represents difference in expression levels, y‐axis represents statistical significance (as −log10 p‐value). PFS, progression‐free survival.

PFS and time to treatment failure (TTF) were significantly better in the high versus low PD‐1 expression group (percentage change in PD‐1⁺ on T cells before and after the start of treatment > 2.5 and < 2.5%, respectively) (Figure 7). The CCR7⁻ population was considered an effector memory T cell‐like population, while the CCR7⁺CD45RA⁻ population with increased PD‐1 expression was considered a precursor‐exhausted CD8⁺ T cell‐like population.

FIGURE 7.

ROC and Kaplan–Meier curves for PFS, TTF, and OS according to PD‐1 expression on T cells. ROC curve analysis for predicting PFS based on percentage of PD‐1⁺ cells in CM (CD45RA⁻CCR7⁺) CD8⁺ T‐cell population, showing an optimal threshold. Optimal threshold for patient stratification was determined using the Youden index. Kaplan–Meier curves for PFS, TTF, and OS stratified by this threshold. AUC, area under the curve; HR, hazard ratio; OS, overall survival; PD‐1, programmed cell death‐1; PFS, progression‐free survival; ROC, receiver operating characteristic; TTF, time to treatment failure.

The population with irAEs had a high percentage of PD‐1‐positive cells on effector memory CD8⁺ T cells, which was thought to be accompanied by extremely high CD8⁺ T‐cell activation (Figure 8). Details of irAEs in the PBMC analysis are provided in Table S4.

FIGURE 8.

Population with irAEs according to PD‐1 expression on effector memory CD8 T cells. Percentage of PD‐1+ cells is shown in central memory (CD45RA⁻CCR7⁺), effector memory (CD45RA⁻CCR7⁻), and EMRA (CD45RA⁺CCR7⁻) CD8⁺ T‐cell subsets at pre‐treatment, post‐treatment, and at the time of irAE onset. irAEs, immune‐related adverse events; PD‐1, programmed cell death‐1.

3.2. Pre‐Treatment Peripheral Blood Biomarker Analysis

To identify pre‐treatment peripheral blood biomarkers for the population with the poorest prognosis (PFS < 50 days), the ROC analysis showed significant differences in CD62L_high Th7R and CXCR3⁺ component of Th7R (CXCR3⁺ Th7R) within the poor prognosis population with PFS < 50 (Figure 9). The patients with low Th7R or CXCR3⁺ Th7R percentages prior to atezolizumab treatment, which consisted of roughly 25% of the total population, showed significantly poorer OS.

FIGURE 9.

ROC for PFS < 50 days and Kaplan–Meier curves for OS based on pretreatment CD4+ T‐cell cluster. ROC curve for the CXCR3⁺ Th7R cluster predicting PFS < 50 days, and the corresponding Kaplan–Meier curve for OS stratified by the determined threshold (≥ 1.0 vs. < 1.0). ROC curve for the CD62L_high Th7R cluster predicting PFS < 50 days, and the corresponding Kaplan–Meier curve for OS stratified by its threshold (≥ 2.0 vs. < 2.0). AUC, area under the curve; HR, hazard ratio; OS, overall survival; PFS, progression‐free survival; ROC, receiver operating characteristic.

3.3. Protein Expression Using PEA

A total of 8 samples in 13 panels were identified as outliers and were excluded from the downstream analysis. No samples without QC Warnings were identified as outliers.

In total, 99 proteins at the PRE timepoint, 101 at the 2nd timepoint, and 21 at the 3rd timepoint were found to be significantly different between the populations with PFS < 100 and PFS ≥ 100 days (adjusted p‐value < 0.05) (Figure S2). Those exceeding the set threshold for elevation in the population with PFS ≥ 100 days (adjusted p‐value < 0.05 and −0.5 > estimate or estimate < 0.5) were FOXO1, AIFM1, ERBB2IP, IRAK4, and SCAMP3 at the 2nd timepoint (Figure 6), but not at other timepoints (Figure S2A–D). Proteins with high values in the population with PFS < 100 days were KRT19 and MUC‐16 at the PRE timepoint; IL‐8, KRT19, MUC‐16, and CRKL at the 2nd timepoint; KRT19 and MUC‐16 at the 3rd timepoint; and KRT19 and MUC‐16 at all time points in which high values were extracted. We also analyzed the change in protein expression from baseline to the 2nd and 3rd timepoints (∆2nd and ∆3rd). However, no proteins showed significant changes after adjusting for multiple testing (Figure S2C,D). None of the proteins exceeded threshold values in ∆2nd and ∆3rd (data not shown). The proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) are listed in Tables S5–S7.

To highlight the association with clinical response, we compared responders (CR/PR) and non‐responders (PD), excluding stable disease cases. The number of significantly different proteins between responders and non‐responders varied by timepoint: 32 proteins at PRE (Figure S3A), 66 at the 2nd timepoint (Figure S3B), and 18 at the 3rd timepoint (Figure S3C). These differences were largely similar when comparing the populations with PFS < 100 versus ≥ 100 days and tended to be more numerous and more differentiated. At the 2nd timepoint, proteins elevated in responders included FOXO1, AIFM1, ERBB2IP, IRAK4, SCAMP3, TANK, LAT2, and INPPL1 (Figure S3B). Proteins elevated in non‐responders (MUC‐16, KRT19, and IL‐8) were consistently extracted at the PRE and 2nd and 3rd timepoints (Figure S3A–C). While no proteins showed a statistically significant difference between the groups at the ∆2nd timepoint (2nd − PRE timepoints) (Figure S3D), high levels of KRT19 and IL‐8 were identified at the ∆3rd timepoint (Figure S3E). Key proteomics results are provided in Tables S8–S11. Cluster analysis of Olink's Immunity Panel proteins, which significantly differed between CR/PR and PD cases, showed that CR/PR cases were concentrated in clusters with high levels of TANK, ICA1, IRAK4, IRAK1, and DAPP1, with one PD exception (Figure S4). Conversely, the non‐responder (PD) cluster showed higher expression of proteins such as TREM1 and PRDX1.

Regarding irAEs, no significant differences in protein expression were found when comparing patients with or without irAEs at PRE, or those with grade 3–5 irAEs versus no irAEs (Figure S5A,C). However, a change in protein expression was observed at the PRE timepoint when comparing patients with grade 1–2 irAEs to those without irAEs (Figure S5B). Furthermore, there were no significant associations between irAE status and the change in protein levels from baseline (Figure S5D,E). Key results from the proteomics analysis are provided in Tables S12–S14. Details of irAEs in the PEA analysis are provided in Table S15.

4. Discussion

This exploratory study of the J‐TAIL study evaluated biomarkers in patients with advanced NSCLC to facilitate appropriate patient selection for treatment with atezolizumab.

Stem‐like exhausted T cells and memory T cells have been found to play important roles in cancer immunity and response to immunotherapy [18]. Kamphorst et al. reported that PD‐1 expression of CD45RA⁻CCR7⁻ effector memory‐type CD8⁺ T cells in peripheral blood was increased in patients with lung cancer who were treated with PD‐1 inhibitors and was more pronounced in patients with high antitumor efficacy [19]. Similarly, in the present study, there was a significant increase in PD‐1‐positive cell rate (p = 0.0001) in atezolizumab‐treated lung cancer patients with PFS ≥ 100 days but not in patients with PFS < 100 days (p = 0.17). However, the percentage of CD45RA⁻CCR7⁻ effector memory‐type CD8⁺ T cells was significantly reduced only in patients with PFS ≥ 100 days (p < 0.001). This reduction in CD45RA⁻CCR7⁻ effector memory CD8+ T cells was matched by an increase in CD45RA⁺CCR7⁻ EMRA CD8⁺ T cells. EMRA are considered terminally differentiated effector T cells capable of exerting potent cytotoxic activity without re‐stimulation. This suggests that the change from the effector memory type to the more active, cell‐killing EMRA type occurred in patients who benefited from atezolizumab.

The increase in the PD‐1 positivity rate was observed not only in effector memory‐type CD8⁺ T cells, but also in the CD45RA⁻CCR7⁺CD8⁺ T cell population, which has been reported to contain central memory T cells [20]. However, it is unlikely that PD‐1⁺ T cells are central memory T cells that are resting in the absence of antigen encounter because PD‐1 is expressed through repeated antigen stimulation. Conversely, Tpex, a stem‐like effector cell, is characterized by the expression of PD‐1 along with CCR7 and CD62L, which are homing molecules for secondary lymphoid organs. Therefore, it is likely that Tpex is the main component of the PD‐1⁺ CD45RA⁻CCR7⁺ CD8⁺ T cells that were increased in patients with PFS > 100 days after atezolizumab treatment in this study. Tpex proliferation is thought to be the origin of PD‐1 inhibitor‐induced proliferation, and increased levels of Tpex cells have been associated with the efficacy of NSCLC treatment with anti‐PD‐1 therapy and improved patient survival [21, 22]. Tpex proliferation likely occurred in atezolizumab‐treated patients. Patients with a PD‐1⁺ CD45RA⁻ CCR7⁺ CD8⁺ T cell proliferation rate of 2.5% or higher had significantly better PFS and TTF.

Although the main target of PD‐1 blockade is thought to be CD8⁺ T cells, in this study, atezolizumab had an effect of increasing PD‐1 positivity in CD4⁺ T cells. As with CD8⁺ T cells, increased PD‐1 positivity was more pronounced in patients with PFS ≥ 100 days, suggesting that it is likely related to the antitumor effect (p < 0.0001). This effect was observed not only for Th1 (CXCR3⁺CCR4⁻CCR6⁻) and CXCR3⁺ Th7R, which are associated with antitumor immunity, but also Th17 (CXCR3⁻CCR4⁺CCR6⁺). Therefore, it is highly likely that the effect of atezolizumab on activating CD4⁺ T cells was not tumor antigen‐specific but occurred in all effector CD4⁺ T cells, and that this phenomenon was more pronounced in patients with longer PFS. We consider that the activation of Th17 cells, which have been shown to be associated with inflammatory bowel disease and cytokine release syndrome, may be more closely associated with irAEs than with anti‐tumor immunity.

The response rate to atezolizumab was low, and PFS was poor. However, the median OS was not reached even after > 500 days of follow‐up; OS remained favorable, excluding the early progression group with PFS < 50 days. The search for biomarkers of patients with disease progression at < 50 days despite atezolizumab treatment revealed that early progression was more common in patients with low pretreatment CXCR3⁺ Th7R or Th7R. Th7R is an IL‐7R‐high expressing, TCF7‐positive effector CD4⁺ T‐cell fraction, found by scRNAseq analysis of peripheral blood from patients with pembrolizumab‐responsive lung cancer. The performance of pre‐treatment peripheral blood Th7R percentage in predicting anti‐PD‐1 antibody treatment response has been reported [7]. Peripheral Th7R cells have also been reported to predict disease‐free survival in patients with early‐stage lung cancer [8]. Our finding, that Th7R has similar predictive performance for anti‐PD‐L1 antibodies, supports this. The clinical utility of these pre‐treatment biomarkers lies in identifying patients with ultra‐poor prognosis who may derive minimal benefit from atezolizumab monotherapy, thereby guiding considerations for alternative strategies. A key challenge for clinical feasibility, however, is the need to translate our mass cytometry‐based findings into a validated, standardized assay using conventional flow cytometry, which is more widely available. If established, these biomarkers could complement existing ones like PD‐L1 to refine patient selection.

The role of the proteins IRAK4 and FOXO1 in tumor immunity has been reported previously, each of which had higher PFS ≥ 100 days and CR/PR at the 2nd timepoint. IRAK4 is a protein located downstream of TLR signaling. In the Immunity Panel cluster analysis of CR/PR versus PD, CR/PR cases were enriched in the TANK, ICA1, IRAK4, IRAK1, and DAPP1 high cluster, except for one PD case. IRAK4, IRAK1, and TANK are all expressed on antigen‐presenting cells, such as dendritic cells, and are components of the TLR signaling pathway [23]. The increase in antigen‐presenting cells may play an important role in the later activation of CD8⁺ T cells (increase in PD‐1⁺CD8⁺T). For example, HMGB1, produced by immunogenic cell death, is a TLR4 agonist, and this signal itself plays an important role in activating CD8⁺ T cells by enhancing antigen presentation.

FOXO1 is a transcription factor that regulates TCF1 (TCF7 gene) expression, suppresses cell senescence, and is important for forming CCR7⁺ central memory T cells. It is also considered necessary for the proliferation of PD‐1⁺ TCF1⁺ stem‐like T cells in the drug effect of ICIs, whereas Tcf1‐negative activated T cells are directed toward exhaustion in the short term [24, 25]. High FOXO1 levels after atezolizumab administration may lead to a treatment response and its long‐term maintenance via maintenance/increase of PD‐1⁺TCF1⁺ T cells.

In this study, high levels of KRT19 and IL‐8 were seen in PD cases and were further elevated at 6 weeks compared with pre‐dose, suggesting an association with poor response to treatment. MUC‐16, also known as tumor marker CA125, is assumed to be associated with tumor burden and poor prognosis unrelated to treatment, while reports on the regulation of immune response have been increasing recently. For example, it has been shown that MUC16 reduces T cell and natural killer cell function through binding to the inhibitory receptor Siglec‐9 [26] and that in pancreatic cancer, MUC16 induces CD4⁺ T cell Foxp3 expression and increased tumor‐associated regulatory T cells through IL‐6 production [27]. It may also cause a decrease in CD8⁺ T cell and natural killer cytotoxicity, either directly or indirectly, by increasing regulatory T cells.

KRT19 is a structural protein belonging to the keratin family that plays a vital role in maintaining cell morphology and function. Keratin 19‐positive tumor cells form a CXCL12‐keratin 19 complex on the cell surface that reduces the motility of CXCR4‐positive T cells. This complex reduces the motility of CXCR4‐positive T cells and forms a T cell‐excluded tumor microenvironment in a mouse pancreatic ductal adenocarcinoma tumor model, contributing to the resistance of anti‐PD‐1 antibodies [28].

In patients with NSCLC, an association between ICI resistance and high levels of IL‐8 in blood has been reported previously [29, 30]. Notably, this finding is consistent with a recent comprehensive analysis in a similar cohort, which also identified IL‐8 as a predictor of resistance to immune checkpoint inhibitors [30]. Levels of plasma cytokines such as TNF, IL‐6, and IL‐8 are associated with ICI resistance, suggesting that IL‐8 may be involved in the tumor microenvironment, suppressing T cell responses by inducing myeloid‐derived suppressor cells and M2‐type macrophages [31]. Aligned with this, our cluster analysis of non‐responders identified higher expression of proteins such as TREM1 and PRDX1, which have also been implicated in promoting an immunosuppressive tumor microenvironment.

Interestingly, significant differences in protein expression were observed only in patients with grade 1–2 irAEs, but not in those with grade 3–5 irAEs. This may be attributable to the small number of patients with high‐grade irAEs in our cohort, which limited the statistical power to detect differences. An alternative hypothesis is that the underlying biological mechanisms may differ between low‐grade and high‐grade irAEs. Because of the limited number of specific irAE events, we could not perform a robust analysis to identify proteins associated with individual types of irAEs.

Because the sample size for PBMC analysis was small (n = 51), the statistical power was limited, and the generalizability of the results is reduced. Validation in a larger cohort is needed. The sampling was heterogeneous, and the PBMC and PEA analyses used different numbers of patients. This complicates the interpretation of results and may create bias. The study findings have not been validated in an independent cohort, which reduces their reliability and reproducibility. As this was an exploratory study, it is not possible to determine if the associations found are causal. The median OS was not reached, so its validity as a predictor of long‐term prognosis is unclear. The impact of other important factors, such as PD‐L1 expression status, tumor mutational burden, and comorbidities, may not have been adequately considered. We did not perform histopathological validation for the hypothesis linking KRT19 to a T‐cell excluded phenotype, as the correlation between KRT19 expression and the status of tumor‐infiltrating lymphocytes could not be assessed in this study. Therefore, direct evidence for KRT19‐mediated immune exclusion is lacking and requires validation using spatial tissue analysis. The results of this study should be considered exploratory and require validation in larger, rigorously designed studies.

This study is one of the few that addresses the topic of immune‐monotherapy response evaluation, providing an extensive analysis of T‐cell profiling in Japanese patients with NSCLC. The present study findings provide valuable insights regarding T‐cell kinetics and biomarkers in atezolizumab therapy and may offer promising directions for future research.

Author Contributions

Atsuto Mouri: conceptualization, investigation, resources, visualization, writing – original draft, writing – review and editing. Hiroshi Kagamu: conceptualization, investigation, resources, visualization, writing – original draft, writing – review and editing. Koji Tamada: conceptualization, investigation, resources, visualization, writing – original draft, writing – review and editing. Makoto Nishio: conceptualization, investigation, resources, writing – review and editing. Hiroaki Akamatsu: conceptualization, investigation, resources, writing – review and editing. Yasushi Goto: conceptualization, investigation, resources, writing – review and editing. Hidetoshi Hayashi: conceptualization, investigation, resources, writing – review and editing. Satoru Miura: conceptualization, investigation, resources, writing – review and editing. Akihiko Gemma: conceptualization, funding acquisition, investigation, resources, supervision, writing – review and editing. Ichiro Yoshino: conceptualization, investigation, resources, writing – review and editing. Toshihiro Misumi: formal analysis, writing – review and editing. Ryota Saito: investigation, resources, writing – review and editing. Noriko Yanagitani: investigation, resources, writing – review and editing. Fujita Masaki: investigation, resources, writing – review and editing. Hiroshi Nokihara: investigation, resources, writing – review and editing. Kazumi Nishino: investigation, resources, writing – review and editing. Masahiro Seike: investigation, resources, writing – review and editing. Tetsunari Hase: investigation, resources, writing – review and editing. Osamu Hataji: investigation, resources, writing – review and editing. Hiroaki Takeoka: investigation, resources, writing – review and editing. Yosuke Kawashima: investigation, resources, writing – review and editing. Hirotaka Kuroki: conceptualization, project administration, visualization, writing – original draft, writing – review and editing. Masamichi Sugimoto: project administration, visualization, writing – original draft, writing – review and editing. Hiroshi Kuriki: formal analysis, writing – review and editing. Tetsuya Mitsudomi: conceptualization, funding acquisition, investigation, project administration, resources, supervision, visualization, writing – review and editing.

Funding

This work was supported by the Chugai Pharmaceutical, in collaboration with the Japanese Lung Cancer Association.

Ethics Statement

The study protocol was approved by the institutional review board of each study site.

Consent

All patients provided written informed consent.

Conflicts of Interest

Atsuto Mouri reports honoraria from Chugai, BMS, Eli Lilly, and Ono. Hiroshi Kagamu reports grants from Boehringer Ingelheim; honoraria from MSD, AstraZeneca, BMS, Chugai, and Ono; and has a patent for Immunological biomarkers for predicting the clinical efficacy of cancer immunotherapy. Koji Tamada reports honoraria from Ono and MSD. Makoto Nishio reports honoraria from Ono, BMS, AstraZeneca, Eli Lilly, Pfizer, Chugai, Taiho, Daiichi Sankyo, AbbVie, Takeda, Boehringer Ingelheim, Novartis, Nippon Kayaku, Merck, and Janssen. Hiroaki Akamatsu reports grants from Amgen, Chugai, Novartis, and MSD; honoraria from AstraZeneca, Pfizer, Chugai, and MSD. Yasushi Goto reports grants from IQVIA, MSD, Astellas, AstraZeneca, AbbVie, Amgen, Syneos Health, Sysmex, CMIC, Novartis, Bayer, BMS, MedPace Japan, Janssen, Clinical Research Support Center Kyushu, SATOMI, Ono, Daiichi Sankyo, Takeda, Chugai, NPO Thoracic Oncology Research Group, Eli Lilly, and Preferred Network to his institution and AstraZeneca, AbbVie, Eli Lilly, Pfizer, BMS, Ono, Novartis, Kyorin, Daiichi Sankyo, MSD, Guardant Health, and Preferred Network to himself; honoraria from Eli Lilly, Chugai, Taiho, Boehringer Ingelheim, Ono, BMS, Pfizer, MSD, Novartis, Merck, and Thermo Fisher; monitoring board or advisory board membership for AstraZeneca, Chugai, Boehringer Ingelheim, Eli Lilly, GSK, Taiho, Pfizer, Novartis, Kyorin, Guardant Health, Illumina, Daiichi Sankyo, Merck, MSD, Ono, and Janssen; and a leadership role with Cancer Net Japan and JAMT. Hidetoshi Hayashi reports grants from A2 Healthcare, IQVIA, PPD‐SNBL, SYNEOS HEALTH, MSD, GSK, ICON Clinical Research GK, AbbVie, Astellas, Amgen, AstraZeneca, Taiho, CMIC, Chugai, Boehringer Ingelheim, West Japan Oncology Group, Sysmex, Eisai, PRA Health Sciences, Pfizer, Daiichi Sankyo, EPS Corporation, BMS, Janssen, and Sanofi; honoraria from AstraZeneca, BMS, Chugai, Takeda, Daiichi Sankyo, Eli Lilly Japan, MSD, AbbVie, Ono, and Amgen; manuscript fees from Guardant Health Japan and Ono; and scholarship from Chugai and Eisai. Satoru Miura reports honoraria from Chugai, Ono, AstraZeneca, Takeda, Daiichi Sankyo, and MSD. Ryota Saito reports honoraria from Chugai. Noriko Yanagitani reports honoraria from AstraZeneca, Eli Lilly, Pfizer, Ono, Chugai, and BMS. Hiroshi Nokihara reports grants from MSD, Ono, AstraZeneca, and Chugai; honoraria from MSD, Ono, AstraZeneca, and Chugai. Kazumi Nishino reports grants from Ono, Taiho, MSD, AbbVie, Daiichi Sankyo, Amgen, Eisai, Sanofi, Janssen, Novartis, Pfizer, Eli Lilly, Merck Biopharma, Takeda, AstraZeneca, Merus, Gilead, Chugai, Bayer, Amgen; and honoraria from AstraZeneca, Boehringer Ingelheim, Eli Lilly Japan, MSD, Novartis, Pfizer, Merck Biopharma, Janssen, BMS, Nippon Kayaku, Ono, Takeda, Chugai, Amgen, Daiichi Sankyo, and Boehringer Ingelheim. Masahiro Seike reports grants from Chugai, Taiho, Eli Lilly, Nippon Kayaku, and Kyowa Kirin; honoraria from AstraZeneca, Chugai, Taiho, MSD, and Daiichi Sankyo; and scholarship from Chugai, Taiho, Eli Lilly, Nippon Kayaku, and Kyowa Kirin. Tetsunari Hase reports grants from Chugai, AstraZeneca, Taiho, Novartis, Amgen, and BeiGene; honoraria from Chugai, AstraZeneca, Taiho, MSD, BMS, Eli Lilly, Ono, Kyowa Kirin, Pfizer, Takeda, Merck Biopharma, and Daiichi Sankyo. Osamu Hataji reports grants from Chugai, MSD, IQVIA, AstraZeneca, Ono, Pfizer, Daiichi Sankyo, BMS, AbbVie, Janssen, Syneos Health Japan, EPS Corporation, and Boehringer Ingelheim; and honoraria from AstraZeneca, Novartis, Eli Lilly Japan, Chugai, MSD, Janssen, Takeda, Nippon Kayaku, and Ono. Hirotaka Kuroki, Masamichi Sugimoto, and Hiroshi Kuriki own stock in Chugai. Tetsuya Mitsudomi reports grants from Boehringer Ingelheim, AstraZeneca, Taiho, Bridge Biopharma; honoraria from AstraZeneca, Boehringer Ingelheim, Chugai, Pfizer, BMS, Eli Lilly, MSD, Novartis, Merck Biopharma, Ono, Amgen, and Daiichi Sankyo. Koji Tamada, Yasushi Goto, and Tetsuya Mitsudomi serve as editorial board members of Cancer Science. Other authors do not have a conflicts of interest.

Supporting information

Figure S1: Gating strategy.

Figure S2:. Volcano plot for the results from the protein‐wise Wilcoxon tests between PFS ≥ 100 days and PFS < 100 days.

Figure S3:. Volcano plot for the results of the protein‐wise Wilcoxon tests between non‐responders (PD) and responders (CR/PR).

Figure S4:. Cluster analysis of samples from proteins on Olink's Immunity Panel.

Figure S5:. Volcano plot for the results from the protein‐wise Wilcoxon tests.

Table S1: Details of the mass cytometry panel, including all antibodies, clones, and metal tags.

Table S2: Baseline demographic and clinical characteristics of patients who underwent PBMC analysis, stratified by progression‐free survival (PFS) < 100 days and ≥ 100 days.

Table S3: Baseline demographic and clinical characteristics of patients who underwent PBMC analysis, stratified by progression‐free survival (PFS) < 50 days and ≥ 50 days.

Table S4: Details of irAEs in the PBMC analysis.

Table S5: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S2A.

Table S6: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S2B.

Table S7: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure 6.

Table S8: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S3A.

Table S9: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S3B.

Table S10: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S3C.

Table S11: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S3E.

Table S12: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S5B.

Table S13: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S5D.

Table S14: Proteins that met the thresholds for both statistical significance (adjusted p‐value < 0.05) and a large magnitude of change (estimate < −0.5 or > 0.5) in Figure S5E.

Table S15: Details of irAEs in the PEA analysis.

Data S1: cas70310‐sup‐0003‐Supinfo.docx.

Data Availability Statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request.