Predictive and Prognostic Implications of Circulating CX3CR1⁺ CD8⁺ T Cells in Non–Small Cell Lung Cancer Patients Treated with Chemo-Immunotherapy

Immune checkpoint inhibitor (ICI) therapy has become a major component in the management of non–small cell lung cancer (NSCLC) in recent years. Inhibitory immune checkpoints, such as the programmed cell death 1 (PD-1)/programmed death-ligand 1 (PD-L1) pathway, function to prevent healthy tissue from excessive immune-mediated destruction during inflammatory states. By hijacking this machinery, tumor cells can evade detection and elimination by the immune system. ICIs that block PD-1/PD-L1 axis have demonstrated improvements in survival over chemotherapy alone in NSCLC in multiple randomized trials both when administered as individual agents (1–3) and as combination ICI regimens (4). This has led to FDA approval of pembrolizumab, nivolumab, and atezolizumab in the treatment of NSCLC, and approval of single-agent pembrolizumab as first-line therapy for PD-L1–positive advanced NSCLC (5). In addition, combination chemo-immunotherapy regimens have shown improved outcomes over chemotherapy alone in several recent clinical trials (6–8).

High PD-L1 expression in the tumor microenvironment (TME) has been correlated with response to PD-1/PD-L1 blockade therapy and is used as a biomarker for the initiation of immunotherapy in NSCLC (1, 4, 9). However, no established biomarker presently exists for chemo-immunotherapy. In trials that showed improvements in survival with combination chemo-immune regimes, PD-L1 expression was not found to be correlated with response to treatment, and there were significant improvements in survival even in patients with PD-L1 expression < 1% (6, 8). Other criticisms of PD-L1 expression include variability in different testing platforms and the intratumoral heterogeneity of PD-L1 expression (10). Other biomarkers for immunotherapy, such as tumor mutational burden and tumor-infiltrating lymphocytes (TIL), have been proposed, although these are still under investigation with varying results (10).

Despite the improved responses seen with chemotherapy and anti–PD-1/PD-L1 therapy (chemo-immunotherapy), only a fraction of patients show durable response with improved survival even with high PD-L1 expression (6–8). In addition, there are serious immune-related toxicities associated with ICI therapy due to nonspecific activation of the immune system by ICIs. These toxicities can affect any organ system in the body and are sometimes irreversible, and most commonly include pneumonitis, colitis, hepatitis, dermatitis, hypothyroidism, and hypophysitis (11). In combination chemo-immune regimens, there are chemotherapy-related side effects in addition to those of ICIs. Therefore, establishing an effective biomarker to predict response to chemo-immunotherapy is important not only to optimize treatment, but also avoid or minimize serious toxicity.

CX3C chemokine receptor 1 (CX3CR1) is widely expressed in immune cells including dendritic cells, monocytes, macrophages, natural killer (NK) cells, and T cells (12). Expression of CX3CR1 correlates with the degree of effector CD8⁺ T-cell differentiation (13). Accumulating evidence has revealed the unique features of CX3CR1 suitable for use as a blood-based T-cell biomarker for immunotherapy. First, despite higher expression of cytotoxic effector molecules, granzymes and perforin, and potent cytotoxicity in vitro, CX3CR1⁺CD8⁺ T cells show lower expression of CXCR3 and L-selectin (CD62L; refs. 13–17), trafficking receptors needed for entry across the tumor microvasculature and lymphoid organ high endothelial venules, respectively (18–20), which allows the CX3CR1⁺ subset to remain in circulation after initial response while CX3CR1^– subsets traffic to the TME and mediate antitumor efficacy (13, 15). Second, unidirectional differentiation of CX3CR1^– CD8⁺ T cells to CX3CR1⁺ CD8⁺ T cells (13–15) facilitates stable expression of CX3CR1 in CD8⁺ T cells in the effector phase unlike other molecules transiently upregulated after activation such as PD-1, 4-1BB, ICOS, and Ki67. Consequently, our study and others have demonstrated that tumor-specific and tumor-infiltrating CD8⁺ T-cell repertoires are enriched in the circulating CX3CR1⁺ CD8⁺ subset in preclinical models and patients (16, 21), and the frequency of peripheral blood (PB) CX3CR1⁺ CD8⁺ T cells increases after effective immunotherapy such as adoptive T-cell therapy, neoantigen/in situ vaccination, and ICI therapy (15, 17, 22, 23).

We have recently analyzed longitudinal PB samples from patients with NSCLC undergoing anti–PD-1 therapy, and found that the percent change of the CX3CR1⁺ subset in PB CD8⁺ T cells from baseline (CX3CR1 score) associated with response and prognosis (16). Although it remains unclear whether T-cell CX3CR1 is a predictor of response to combination chemo-immunotherapy, Yan and colleagues demonstrated that CX3CR1⁺CD8⁺ T cells withstand treatment with chemotherapy and are increased in response to chemo-immunotherapy in patients with metastatic melanoma (22). These findings set the stage for us to explore the utility of PB CX3CR1⁺CD8⁺ T cells as a potential biomarker for treatment response to chemo-immunotherapy in patients with NSCLC.

Here, we hypothesize that the change of the CX3CR1⁺ subset in PB CD8⁺ T cells from baseline correlates with response to chemo-immunotherapy in patients with advanced NSCLC. We collected serial PB samples from patients with advanced NSCLC undergoing treatment with chemo-immunotherapy, analyzed levels of the CX3CR1⁺ subset in CD8⁺ T cells, and evaluated its predictive and prognostic values of the CX3CR1 score. Furthermore, we employed T-cell receptor sequencing (TCR-seq) of pretreatment tumor tissue and single-cell (sc) RNA/TCR-seq of serial PB samples to evaluate genomic and transcriptomic signatures of tumor-infiltrating lymphocyte repertoires in PB. Our studies identify a potential role of T-cell CX3CR1 as a predictor of response to chemo-immunotherapy in patients with NSCLC.

Informed written consent was obtained from 29 patients with naïve or previously treated NSCLC, undergoing combination of chemotherapy and anti-PD-1/PD-L1 antibody (pembrolizumab or atezolizumab) for the collection and storage of blood samples, the analysis of archived tumor tissue, and the review of their medical records under the Institutional Review Board of Roswell Park Comprehensive Cancer Center protocol approval (I 188310), in accordance with the Declaration of Helsinki.

The clinical samples were prospectively collected, and selected on the basis of availability during the study window. No statistical methods were used to predetermine sample size. Randomization was neither feasible nor appropriate due to the nature of this study. The observed sample size (n = 29) provides an adequate pool of both responders and non-responders, and produces performance measures (i.e., sensitivity, ROC curves, and the corresponding AUC, etc.) with adequate levels of precision. We evaluated the correlation between a biomarker performance during treatment and response. The investigators were not blinded during experiments and outcome assessment. This study was conducted in accordance with the Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK; ref. 24).

PB was obtained in EDTA-containing tubes. Peripheral blood mononuclear cells (PBMCs) were isolated using Lymphocyte Separation Medium (Corning) density gradient centrifugation. Fresh or cryopreserved PBMCs were incubated with anti-human IgG (Sigma). These antibodies were used for flow cytometry or cell sorting: anti-human CD3 (clone UCHT1; BioLegend), CD4 (clone RPA-T4; BD Biosciences), CD8 (clone RPA-T8; BioLegend), CD45 (clone HI30; BD Biosciences), CD56 (cone HCD56: BioLegend), CD19 (clone HIB19 BioLegend), and CX3CR1 (clone 2A9–1; BioLegend) antibodies. Samples were acquired using LSRFortessa (BD Biosciences), and data were analyzed with FlowJo software v10.1.5 (FlowJo LLC).

Clinical response to chemo-immunotherapy was assessed as best response according to immune-related RECIST (iRECIST; ref. 25) within 12 weeks as described previously (16). Patients who had complete response (CR) and partial response (PR) were classified as responders while ones with stable disease (SD) and progressive disease (PD) as nonresponders. Objective responses were confirmed by at least one sequential tumor assessment. Overall response rates (ORR) were calculated as [(CR + PR) ÷ number of patients] × 100.

The expression of PD-L1 on the surface of tumor cells was reported as a standard of care before treatment, which was performed using the 22C3 PharmDx antibody on the Dako Omnis platform (Agilent) and scored by published guidelines (26).

Single live CD45⁺CD19^–CD3⁺CD56^– cells were enriched from cryopreserved PBMC samples by flow cytometry sorting using a BD FACSAria II (BD Biosciences; Supplementary Fig. S1A and S1B). Cells were counted, hashed with TotalSeq-C0251 (ref: 394661), TotalSeq-C0252 (ref: 394663), TotalSeq-C0253 (ref: 394665), and TotalSeq-C0254 (ref: 394667) anti-human antibodies (BioLegend) and pooled. The pooled cells were then quantified and subsequently loaded onto the chromium chip G using the standard protocol for the Chromium single-cell 5′ kit v2 (10x Genomics, Inc). Following Gel Bead-in Emulsion (GEM) generation, samples were processed according to the standard manufacturer's protocol. After sequencing libraries passed standard quality control metrics, the libraries were sequenced on Illumina NovaSeq6000 S1 100cycle v1.5 kits with the following read structure: read1: 28, read2: 90, index 1: 10, index 2:10. Libraries were sequenced to obtain a read depth greater than 16,000 reads/cell for the gene-expression (GEX) libraries and greater than 4,000 reads/cell for the V(D)J-enriched T-cell libraries.

Raw sequence data demultiplexing, barcode processing, alignment (GRCh38), and filtering for true cells were performed using the Cell Ranger Single-Cell Software Suite (v6.0.0), yielding 8,916 cells (pretreatment: 4,170 cells, 3 weeks: 534 cells, 6 weeks: 3,090 cells, 9 weeks: 1,122 cells) with a mean of 22,239 reads/cell (90.15% mapping rate), median of 1,175 genes/cell, 19,831 total unique detectable genes, and 2,697 median UMI counts/cell. Seurat (v4; ref. 27) was used to perform filtering, normalization, and downstream analyses as previously described (refs. 23, 28; Supplementary Fig. S1C–S1F). Hashtag feature barcoding (TotalSeq-C antibodies, BioLegend) of pooled samples was demultiplexed using a k-medoid clustering approach implemented by Seurat to assign cells to individual samples and remove doublets. After quality control assessment, 4,040 cells were removed (45.32% of total cells), and 4,876 high quality cells (pretreatment: 1,861 cells, 3 weeks: 223 cells, 6 weeks: 2,137 cells, 9 weeks: 655 cells) were included in downstream analyses. VDJ annotations derived from Cell Ranger were analyzed using scRepertoire (29) and custom scripts. Differential TCRB clonotype abundances were determined by Fisher exact test. Gene-set enrichment analysis (GSEA) of cluster-specific gene markers was performed via enrichR (30). Reference gene sets included those from the GO-Biological Processes and Reactome databases, compiled from the Molecular Signatures Database (MSigDB; ref. 31). Gene sets with Benjamini–Hochberg adjusted P < 0.05 were considered as significantly enriched.

We obtained DNA from NSCLC formalin-fixed, paraffin-embedded (FFPE) samples prepared within the 2 months before the initiation of chemo-immunotherapy. TCRβ CDR3 repertoires were profiled using the ImmunoSEQ immune profiling platform at the survey level (Adaptive Biotechnologies) as previously described (16, 23). Repertoire characteristics were analyzed using the LymphoSeq package and custom scripts in the R statistical software environment. The level of similarity (Morisita–Horn Index) between repertoires was calculated using the vegan package.

Patient demographic and clinical characteristics were reported using the mean and range for continuous variables; and frequencies and relative frequencies for categorical variables. The marker expression was compared using the Mann–Whitney U test as described before (16). The maximal percent change of the CX3CR1⁺ subset in PB CD8⁺ T cells from baseline by the given time point was calculated as described previously (16). We assessed the correlation between the maximal percent change in CX3CR1 and objective response as described previously (16). Briefly, we estimated the ROC curves and the corresponding AUC using a logistic regression model, and obtained confidence intervals for the AUC using DeLong method (32). We utilized the Youden index criterion (33) to identify the optimal cut-off point, and examined sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV); with 95% confidence intervals by Jeffrey prior method. Survival outcomes were summarized by groups (i.e., CX3CR1 expression) using standard Kaplan–Meier methods, with comparisons made using the log-rank test (GraphPad Prism 9.4.1). Associations with demographic and clinical factors were assessed using Cox regression models, where HRs were obtained from model estimates. The variables that were found to be P < 0.01 on univariate analysis were included in the multivariable analysis. Fisher exact test was used to assess the association between PD-L1 expression or the CX3CR1 score and objective response.

Raw and processed scRNA/TCR-seq data supporting the findings of this study have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (NCBI-GEO) under accession number GSE213902. Bulk TIL TCRseq clonotype calls derived from Adaptive ImmunoSEQ are available at https://github.com/mdlong-rpccc/Ito_ChemoImmunotherapy_CX3CR1. Other source data are provided with this article. All data generated and analyzed are available from the corresponding author upon reasonable request.

We evaluated 29 patients who were treated with chemo-immunotherapy, and enrolled in this study between February of 2018 and July of 2021. The cut-off date for treatment outcome analysis was October 30, 2022, at which time response evaluations were available for 29 patients. The median time of follow-up was 35.7 months (range 3.3–43.4). Baseline characteristics of 29 patients are shown in Supplementary Table S1. The majority of patients (17 patients; 59%) received carboplatin, pemetrexed and pembrolizumab while 10 patients (34%) received carboplatin, paclitaxel and pembrolizumab, and 2 patients (7%) received carboplatin, paclitaxel, atezolizumab, and bevacizumab. Twelve patients (41%) were classified as “responders” due to partial response (PR) as their best response within 12 weeks from the initiation of therapy by iRECIST criteria while no patients achieved a complete response (CR). Seventeen patients (59%) were classified as “nonresponders”, and of these 13 had stable disease (SD) while 4 demonstrated progressive disease (PD) based on iRECIST criteria, resulting in an ORR of 41.3%.

Next, we analyzed the frequency of the CX3CR1⁺ subset in PB CD8⁺ T cells (Fig. 1A). The median baseline frequency of the CX3CR1⁺ subset among CD8⁺ T cells was 30.9% (4.0–85.1%) with no difference in ORR and progression-free (PFS) and overall survival (OS) between the low- and high-frequency groups using various cut-off points (Supplementary Fig. S2). As described recently, we analyzed responses in terms of the percent change of the CX3CR1⁺ subset in circulating CD8⁺ T cells from baseline (CX3CR1 score; ref. 16). The maximal percent change of the CX3CR1⁺ subset in PB CD8⁺ T cells from baseline by the given time point could differentiate responders from nonresponders as early as 4 weeks from the start of the treatment (Fig. 1B). Next, an ROC curve and the Youden Index (33) were used to determine the optimal cut-off score of the maximal percent changes in CX3CR1 biomarker expression on PB CD8⁺ T cells. We found that an increase of CX3CR1⁺ CD8⁺ T-cell subsets by 9.42%–10.35% from baseline differentiated responders from nonresponders at 6–12 weeks, and was correlated with higher OR, sensitivity, specificity, PPV, and NPV (Supplementary Table S2). In line with this finding, at least 10% increase of the CX3CR1 score was observed in 83.3% (10/12) of responders compared with 10.5% (2/19) of nonresponders (Fig. 1C). Indeed, at least 10% increase of the CX3CR1 score started to correlate with ORR at 4 weeks [P = 0.0228; OR, 10.7; 95% confidence interval (CI), 1.05–109.78], and became more predictive at 9 weeks (P < 0.0001; OR, 15; 95% CI, 2.24–100.48; Fig. 1D; Supplementary Table S3). We then analyzed the sensitivity, specificity, PPV, NPV, and accuracy of our biomarker at various time points and compared them to PD-L1 expression by tumor proportion score (TPS). The maximal CX3CR1 score of at least 10% increase exhibited higher sensitivity, specificity, PPV, and NPV than the PD-L1 TPS, and identified response in 14/20 (70.0%), 22/27 (81.4%), 24/28 (85.7%) and 25/29 (86.2%) at 3, 4, 6, and 9 weeks, respectively (Table 1).

Next, we evaluated relevant variables for correlation with PFS and OS. Kaplan–Meier survival analysis demonstrated significantly improved PFS (P = 0.0051) and OS (P = 0.0138) by log-rank test for patients with at least 10% increase in CX3CR1 score (Fig. 1E). Of note, we previously reported that 20% increase in CX3CR1 score correlated with survival in a patient with NSCLC undergoing anti–PD-1 monotherapy (16). Therefore, we evaluated prognostic value of 20% increase in CX3CR1 score, and found that improved PFS and OS with this cut-off point for chemo-immunotherapy as well (Supplementary Fig. S3). Univariate analysis was performed to compare the characteristics of patients with NSCLC treated with chemo-immunotherapy. This analysis revealed that at least 10% change in CX3CR1 score was the only significant prognostic factors for PFS [HR, 0.28; 95% CI, 0.11–0.75; P = 0.011) and OS (HR, 0.25; 95% CI, 0.08–0.80; P = 0.019) among other variables such as ECOG status, histology, stage III versus IV, prior chemotherapy, presence of brain metastases, and histologic variables including PD-L1 expression (Table 2). When the factors identified by univariate analysis (P < 0.10) were subjected to a multivariate analysis, at least 10% change in CX3CR1 score was the only independent prognostic factor that approached statistical significance for PFS (HR, 0.36; 95% CI, 0.13–1.01; P = 0.051), and female sex (HR, 3.115; 95% CI, 1.12–8.67; P = 0.030) and at least 10% change in CX3CR1 score (HR, 0.25; 95% CI, 0.07–0.91; P = 0.036) were the only independent prognostic factors for OS (Table 2).

Next, we sought to characterize the full spectrum of pre- and early on-treatment circulating T cells early after the initiation of chemo-immunotherapy. We employed scRNA/TCR-seq on serially obtained PBMCs from a patient who received a long-term benefit from chemo-immunotherapy. The patient is a 68-year-old female who was treated with every 3-week carboplatin, pemetrexed, and pembrolizumab for stage IV NSCLC with multiple bilateral pulmonary nodules and mediastinal adenopathy. After two cycles of chemo-immunotherapy, her CT scan at 6 weeks showed an increase in mediastinal and bilateral hilar/infrahilar adenopathy and numerous lung lesions except for a single lesion within the right lower lobe (Supplementary Fig. S4A); however, we found a prompt substantial increase of PB CX3CR1⁺ CD8⁺ T cells and the CX3CR1 score (Supplementary Fig. S4B). Eventually, chemo-immunotherapy was found to be effective with a long-term disease control for 23 months and she lived for longer than 3 years after initiation of the treatment.

To profile PB T cells before and during ICI therapy, we flow-sorted CD45⁺CD19^–CD3⁺CD56^– cells from cryopreserved PBMC samples at pretreatment and 3, 6, and 9 weeks from the initiation of the treatment for scRNA/TCR-seq (Supplementary Fig. S1A and S1B). This yielded data for 4,876 high-quality cells after stringent filtering (pretreatment: 1,861 cells; 3 weeks: 223 cells; 6 weeks: 2,137 cells; and 9 weeks: 655 cells; Supplementary Fig. S1C–S1E). Unsupervised clustering analysis identified 13 distinct lymphocyte clusters [cluster (C)0–12; Fig. 2A; Supplementary Fig. S5; Supplementary Table S4]. Within the clusters expressing CD8A, we found a markedly increased frequency of C4- and C6-expressing CX3CR1 (Fig. 2B–D) consistent with flow cytometric analysis (Supplementary Fig. S4B). Although C6 was notable for high expression levels of GNLV, both C4 and C6 overexpressed NKG7, ZEB2, GZMB, GZMA, GZMH, and PRF1 encoding perforin, and genes associated with effector molecules such as SLAMF6, TBX21 encoding T-bet, chemokine (C-C motif) ligands (CCL4, CCL5), and human leukocyte antigen (HLA) class II molecules, suggesting terminal effector CD8⁺ T cells (Fig. 2D; Supplementary Fig. S5; and Supplementary Table S4). Consistent with this, these clusters were notable for high levels of TCR, PD-1, IFN, proliferative, and cytokine-mediated signaling pathways (Supplementary Fig. S6; Supplementary Table S5A and S5B). There was a gradual increase of C7-expressing markers of memory T cells, TCF7 encoding T-cell factor 1 (TCF1), SELL encoding L-selectin, IL7R and CCR7. In contrast, there was a decrease of C8 that was notable for elevated expression of GZMK and CXCR3. C8 was also positive for genes related to activated (PDCD1 encoding PD-1, IL2RA), effector (NKG7, GZMA, GZMM, CCL5,) and memory (TCF7, SELL, IL7R, CCR7) T cells, but relatively negative for CX3CR1 expression, suggesting activated early effector T cells. C0 expressing CD4, TCF7, SELL, S1PR1, GATA3, and IL7R was the most frequent cluster at pretreatment, but considerably decreased during chemo-immunotherapy. Although this cluster was enriched with TNF-mediated signaling pathway, it also exhibited high levels of apoptosis and programmed cell death pathways. In line with these changes in specific cluster frequencies, differential expression analysis identified that genes associated with terminal differentiation and effector function including CX3CR1, NKG7, GNLY, GZMB, GZMH, CCL4, CCL5 in circulating T cells were markedly increased as early as 3 weeks from the initiation of chemo-immunotherapy (Fig. 2E; Supplementary Table S6A–S6C). Taken together, effective chemo-immunotherapy induces dynamic changes in the composition of circulating T cells early on-treatment including increased frequency of terminal effector CX3CR1⁺ CD8⁺ T cells.

We have recently shown that successful ICI therapy induces an expansion of the peripheral CX3CR1⁺ CD8⁺ T-cell subset that includes an enriched repertoire of tumor-specific and tumor-infiltrating CD8⁺ T cells in preclinical models (16). To evaluate these findings in this patient (Supplementary Fig. S4), we first assessed VDJ gene usage in the PB scRNA/TCR-seq data. This analysis identified 2,541 unique productive TCRB clonotypes, and revealed a higher concentration of larger clonally expanded T cells in C4 and C6 (Fig. 3A). Second, analysis of T-cell repertoire similarities amongst samples revealed a dramatic reshaping of T-cell repertoires, demonstrated by lower TCR overlap score occurring within 3 weeks of treatment (Fig. 3B). Finally, we sought to determine whether circulating clonally expanded T cells could be identified in the tumor. To this end, we extracted DNA from the archival pretreatment tumor samples, and performed TCR-seq to identify TCR clonotypes observed in the tumor (TIL-TCRs). We identified 16,109 unique productive TCRB clonotypes (Supplementary Table S7), of which 1,064 (6.6%) were observed at least 5 counts in sequencing, and were classified as frequent TIL-TCRs (Supplementary Fig. S7). Of these, 122 (11.5%) TIL-TCRs were identified in PB scRNA/TCR-seq data (Fig. 3C). The combined frequency of these TIL-TCRs was increased in the periphery after the initiation of chemo-immunotherapy (Fig. 3D; Supplementary Table S8). The scRNA/TCR-seq analysis identified 10 significantly expanded clonotypes in PB at 6 and 9 weeks compared with pretreatment (Fig. 3E; Supplementary Table S9A–C), including 8 frequent TIL-TCRs (Fig. 3C; Supplementary Table S9). All 10 circulating expanded clones overexpressed CX3CR1 (Fig. 3F), and were enriched in C4 and C6 (Fig. 3G). Collectively, these findings suggest that circulating clonally expanded T cells early after the initiation of chemo-immunotherapy overexpress CX3CR1, and contain repertoires of tumor-infiltrating T cells.

In this study, using prospectively collected longitudinal blood samples, we provide evidence that effective combination chemo-immunotherapy correlates with an early on-treatment expansion of circulating CX3CR1⁺ CD8⁺ T cells. The increase of CX3CR1 score, regardless of whether it was 10% or 20%, was associated with both increased PFS and OS. Our results are consistent with previous findings in patients with renal cell carcinoma and NSCLC treated with PD-1/PD-L1 blockade monotherapy, and patients with melanoma undergoing chemo-immunotherapy (16, 22, 34). Our findings further demonstrated that at least 10% increase of the CX3CR1 score was predictive of response to chemo-immunotherapy with high sensitivity, specificity, PPV, and NPV, compared with pretreatment PD-L1 expression on immune and tumor cells. The CX3CR1 score was predictive of response to anti–PD-1 monotherapy (16) and chemo-immunotherapy in this study; however, the optimal cut-off score was different: 20% for anti–PD-1 monotherapy (15) and 10% for chemo-immunotherapy. It is possible that the chemotherapy component of treatment might have affected T-cell differentiation although direct evidence for this explanation has been lacking. While beneficial immunologic effects of chemotherapy such as induction of immunogenic cell death and depletion of immunosuppressor cells have been recognized (35), chemotherapy could affect proliferation of antitumor T cells induced by anti–PD-1 therapy (36, 37). Additional work is needed to evaluate the impact of chemotherapy on differentiation and cytotoxic function of effector T cells rescued by ICI therapy.

The potential value of an increase in the CX3CR1 score following initiation of therapy was illustrated by a patient who was not considered to be a responder based on the serial imaging but received long-term benefit from chemo-immunotherapy. In this patient, consistent with a remarkable expansion of PB CX3CR1⁺ CD8⁺ T cells, single-cell profiling of longitudinal circulating T cells revealed effective early on-treatment differentiation and expansion of effector T cells that includes tumor resident T-cell repertoires. Moreover, we found a notable reshaping of the circulating TCR clonotypes 3 weeks after initiation of the treatment. These findings suggest that evolution of the immune landscape observed in the TME during ICI therapy (38, 39) could be identified in circulating immune cells, and might be useful for treatment decision-making.

Combining scRNA/TCR-seq of sorted T cells in serial PB samples with TCR-seq of tumor tissue provided insight into the transcriptional landscape of circulating TIL-TCRs, and revealed that expanded circulating TIL-TCRs early on-treatment were enriched in signatures of terminal effector differentiation and marked by high expression of CX3CR1. While the data were obtained from a single patient, these findings are in agreement with our recent study showing that peripheral CX3CR1⁺ CD8⁺ T-cell clones reflect the TCR repertoires in CD8⁺ TILs in preclinical models (15), suggesting that T-cell CX3CR1 expression may act as a dynamic blood-based biomarker of response to immunotherapy. Of note, TIL-TCRs expanded in response to chemo-immunotherapy could be identified within the same cluster expressing CX3CR1 at pretreatment in line with recent studies evaluating the tumor and a time-matched blood sample from patients with melanoma with scRNA/TCR-seq (21, 40). However, this was markedly pronounced in early on-treatment PB samples in this study.

Our study has several limitations. While our findings on CX3CR1 score are statistically significant, our cohort is relatively small, resulting in broad confidence intervals. On the basis of Cox univariate analysis, at least 10% change in CX3CR1 score was the only significant predictor of PFS and OS; however, the lack of effect of other key variables such as stage III versus IV and brain metastasis almost certainly reflects the limited power of our small cohort. To validate our results, follow up studies are necessary in larger sample sizes and with independent cohorts. There was some variability in our chemo-immune regimens, with three different regimens in NSCLC employed across patients, and the number of patients was underpowered to detect any difference in outcome between the different regimens. Our cohort did not include patients who had radiotherapy during chemo-immunotherapy, and therefore it remains unclear how the CX3CR1 score would behave in this patient subset. Because of the limited availability of tumor tissue for TCR-seq, single-cell profiling of circulating TIL-TCRs was limited to a single patient in the current cohort. In addition, intratumor heterogeneity of TIL-TCRs might have not been fully captured because DNA was not extracted from the entire tumor for TCR-seq. It remains unclear whether clonally expanded TIL-TCRs in PB also did in the tumor during chemo-immunotherapy because on-treatment tumor tissue was not available in this patient. More studies are needed to elucidate further genomic and transcriptomic characterization of TIL-TCRs responding to immunotherapy.

Given the curative potential of the treatment and the evolution of the TME during ICI therapy (38, 39), early on-treatment blood-based biomarkers with high NPV would be of great value (41). Because ICI therapy targets host immunity, investigating the potential utility of markers expressed on circulating immune cells in clinical applications as a predictive biomarker has been an intense area of research. There are several mechanistic advantages to the use of CX3CR1 as a biomarker in this regard due to (i) the unidirectional differentiation of CX3CR1^int to CX3CR1^hi subsets, allowing CX3CR1 to be stably expressed on CD8⁺ T cells (13–15); and (ii) the decreased expression of L-selectin and CXCR3 (13–16), essential trafficking molecules for T cells from blood to secondary lymphoid organs and the TME, respectively (18–20), allowing CX3CR1⁺ CD8⁺ T cells to remain in circulation at the end of the primary response (13, 15).

In summary, our findings demonstrate that at least 10% increase of the CX3CR1⁺ subset in PB CD8⁺ T cells identifies patients with NSCLC responding to chemo-immunotherapy early on-treatment. Substantial early on-treatment reshaping of T-cell clonotypes and clonally expanded TIL-TCRs expressing CX3CR1 can be identified in PB of patients undergoing chemo-immunotherapy. Our study provides evidence to plan further trials testing this circulating T-cell differentiation marker in larger prospective trials in a variety of malignancies.

B.H. Segal reports other from Apellis, personal fees and other from NextCure, and grants from NCI outside the submitted work. G.K. Dy reports other from AZ, Mirati Therapeutics, Takeda, Eli Lilly, Amgen, Regeneron outside the submitted work. No disclosures were reported by the other authors.

E. Abdelfatah: Formal analysis, writing-original draft, writing-review and editing. M.D. Long: Formal analysis, writing-original draft, writing-review and editing. R. Kajihara: Data curation, formal analysis, writing-review and editing. T. Oba: Data curation, writing-original draft. T. Yamauchi: Investigation, writing-review and editing. H. Chen: Resources, writing-review and editing. J. Sarkar: Formal analysis. K. Attwood: Formal analysis, writing-review and editing. J. Matsuzaki: Resources, writing-review and editing. B.H. Segal: Resources, writing-review and editing. G.K. Dy: Resources, writing-review and editing. F. Ito: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, project administration, writing-review and editing.

We are very grateful to all patients who generously contributed samples and participated in the study. We thank the clinical study staff, Ms. Elongia Farrell, Ms. Noelle Brunsing, and Ms. Rushka Kallicharan-Smith for assistance in collecting patient samples. We acknowledge Ms. Rachel Reagan, Ms. Courtney Ryan, Ms. Jessie L. Chiello, Ms. Alexandra Corrao, Drs. Prashant Singh, Toshifumi Hoki, and Toshihiro Yokoi (Roswell Park) for technical assistance. This work was supported by NCI grants P50CA159981–07A1 and P30CA016056 involving the use of Roswell Park's Flow and Image Cytometry, Clinical Research and Pathology Network, Biostatistics & Statistical Genomics Shared Resource, Immune Analysis Facility and Data Bank and Biorepository Shared Resource, and by the National Center for Advancing Translational Sciences of the NIH (UL1TR001412) to the University of Buffalo. This work was also supported by Roswell Park Alliance Foundation, Department of Defense Lung Cancer Research Program (LC180245), and NCI grant, K08CA197966, R01CA255240–01A1 (to F. Ito), R01CA188900 and R01CA267690 (to B.H. Segal), and Uehara Memorial Foundation (to T. Oba and R. Kajihara).