T-cell CX3CR1 expression as a dynamic blood-based biomarker of response to immune checkpoint inhibit

Immune checkpoint inhibitors (ICI) have revolutionized treatment for various cancers; however, durable response is limited to only a subset of patients. Discovery of blood-based biomarkers that reflect dynamic change of the tumor microenvironment, and predict response to ICI, will markedly improve current treatment regimens. Here, we investigate CX3C chemokine receptor 1 (CX3CR1), a marker of T-cell differentiation, as a predictive correlate of response to ICI therapy. Successful treatment of tumor-bearing mice with ICI increases the frequency and T-cell receptor clonality of the peripheral CX3CR1+CD8+ T-cell subset that includes an enriched repertoire of tumor-specific and tumor-infiltrating CD8+ T cells. Furthermore, an increase in the frequency of the CX3CR1+ subset in circulating CD8+ T cells early after initiation of anti-PD-1 therapy correlates with response and survival in patients with non-small cell lung cancer. Collectively, these data support T-cell CX3CR1 expression as a blood-based dynamic early on-treatment predictor of response to ICI therapy.

Cancer immunotherapies that target the immune checkpoints, such as cytotoxic T lymphocyte-associated antigen 4 (CTLA-4), programmed cell death protein-1 (PD-1), and PD1 ligand-1 (PD-L1), have transformed the therapeutic landscape of a variety of malignancies1,2,3. However, despite compelling clinical responses seen across diverse tumor types, only a fraction of patients achieve durable responses. Moreover, unusual response patterns such as pseudoprogression and delayed response, a dichotomous outcome, potentially severe toxicity, and high cost indicate a critical need for a reliable predictive biomarker4,5,6,7.

Baseline PD-L1 expression on immune and tumor cells, preexisting infiltrating CD8+ T cells, and tumor mutational burden (TMB) correlate with response2,3,4,5,6,7,8,9,10,11; however, the use of these pretreatment markers are hampered by the significant overlap between responders and non-responders, limited quantity and quality of the tissue, and/or lack of standardization12,13,14. Analysis of serially collected tumor samples could aid in the assessment of the evolution of the tumor microenvironment (TME) during immune checkpoint inhibitor (ICI) therapy15,16,17,18; however, this approach is invasive and challenging for visceral tumors such as non-small cell lung cancer (NSCLC). The discovery of dynamic circulating immune biomarkers that reflect the evolution of adaptive immunity in the TME and are early predictors of clinical response to ICI would be of value to guide the selection of patients most likely to benefit from ICI therapy.

Emerging blood-based biomarkers such as exosomal PD-L1, TMB, and T-cell receptor (TCR) sequence in cell-free DNA, and hypermutated circulating tumor DNA associate with response19,20,21,22,23; however, these approaches require complex platforms, and/or bioinformatics analysis, that limit their widespread application in community-based clinical practice. Since ICI targets T-cell regulatory pathways, the utility of surface and intracellular proteins expressed on T cells have been investigated as a potential biomarker for response8,9,24,25,26,27,28. Of these, the proliferation marker Ki-67 has been extensively investigated. However, most studies showed Ki-67 expression only transiently increased in subsets of peripheral blood (PB) CD8+ T cells after the first cycle with unclear predictive and prognostic value as a stand-alone biomarker for ICI8,9,24,25,26,27.

Recently, CX3C chemokine receptor 1 (CX3CR1) was found to be a marker of T-cell differentiation, where CX3CR1+ CD8+ T cells were the progeny of CX3CR1−CD8+ T cells, and exhibited robust cytotoxicity in anti-viral immunity29,30. Mechanistically, CX3CR1 is stably expressed on CD8+ T cells through unidirectional differentiation from CX3CR1−CD8+ T cells during the effector phase30,31, which theoretically provides an advantage as a biomarker compared with transiently expressed molecules on T cells. Indeed, increased frequency of PB CX3CR1+ CD8+ T cells has been observed in a few patients who responded to anti-VEGF and anti-PD-L1 antibodies (Ab) for renal cell carcinoma32. Furthermore, Yan et al.33 have reported the increased frequency of CX3CR1+ granzyme B+ T cells among PB CD8+ T cells in melanoma patients who responded to anti-PD-1 Ab compared to non-responders. Collectively, evidence from preclinical and clinical studies prompted us to evaluate the role of CX3CR1 as a blood-based T-cell biomarker of response to ICI therapy.

Here, we hypothesize that changes in the frequency of PB CX3CR1+ CD8+ T cells would correlate with response to ICI and help identify responders vs. non-responders early after initiation of therapy. We investigate the frequency of CX3CR1+ CD8+ T cells in PB before and during ICI therapy, and delineate the TCR repertoire in peripheral CX3CR1+ CD8+ T-cell subsets and CD8+ tumor-infiltrating lymphocytes (TILs) using preclinical models. To understand the clinical utility of CX3CR1 as a circulating T-cell biomarker, we analyze longitudinal PB samples from patients with NSCLC undergoing anti-PD-1 therapy and evaluate changes in the frequency of PB CX3CR1+ CD8+ T cells as a correlate of response to anti-PD-1 therapy. Our results support circulating CX3CR1+ CD8+ T cells as an early on-treatment biomarker of clinical response to anti-PD-1 therapy.

Effective ICI therapy correlates with increased frequency of circulating CX3CR1+ CD8+ T cells

While ICI is known to restore cytokine production and proliferation of effector T cells34, a recent study in a murine infection model indicates that ICI also facilitates effector differentiation of T cells35. To further elucidate this mechanism, we evaluated CX3CR1 expression, a marker of T-cell differentiation in combination with CD27 29,30,31, on PB CD8+ T cells before and during treatment with anti-PD-L1 and anti-CTLA-4 Ab or isotype Ab (NT: no-treatment) in two mouse tumor models, MC38 and CT26 colon adenocarcinoma (Fig. 1a and Supplementary Fig. 1). Markedly improved tumor control and survival (Fig. 1b) with increased frequency of CX3CR1+ CD8+ T cells (Fig. 1c) were observed in MC38 and CT26 tumor-bearing mice treated with ICI compared to mice receiving isotype Ab. Next, we used a tetramer (Tet) to detect CD8+ T cells specific for mutated Adpgk protein (AdpgkMut) in MC38 and shared tumor-associated antigen (TAA), gp70 in CT26 tumors36,37, and found substantially increased frequency of CX3CR1+ Tet+ CD8+ T cells in both tumor models (Fig. 1d), suggesting that T-cell differentiation after ICI therapy occurs in tumor-specific CD8+ T cells. We also tested whether the increase of PB CX3CR1+ CD8+ T cells can be seen in mice bearing B16 tumors which exhibit primary resistance to CTLA-4 or PD-1/PD-L1 blockade therapy38,39. There was a non-significant trend toward increased PB CX3CR1+ CD8+ T cells in B16 tumor-bearing mice without improvement of survival by combined CTLA-4/PD-L1 blockade therapy (Supplementary Fig. 2).

Although both anti-PD-L1 Ab and anti-CTLA-4 Ab target subsets of exhausted-like CD8+ T cells, they do so through distinct cellular mechanisms40. Therefore, we evaluated CX3CR1 expression of PB CD8+ T cells in CT26-bearing mice treated with either anti-PD-L1 Ab, anti-CTLA-4 Ab, or both. Increased frequency of the PB CX3CR1+ CD8+ T cells was seen after either monotherapy or combined ICI therapy compared to no treatment (Fig. 1e). Of note, there was no obvious difference in the antitumor efficacy between monotherapy and combined therapy (Fig. 1f), consistent with the frequency of PB CX3CR1+ CD8+ T cells. The increase of circulating CX3CR1+ CD8+ T cells was also observed in MC38 tumor-bearing mice treated with anti-PD-L1 therapy alone with improved tumor control and survival (Fig. 1g, h). Collectively, these findings suggest that effective ICI therapy correlates with increased frequency of PB CX3CR1+ CD8+ and CX3CR1+ Tet+ CD8+ T cells in mice.

Peripheral CX3CR1+ CD8+ T cells display properties of recently activated effector cells with decreased capacity of trafficking across tumor vessels

Next, using a gating strategy based on the expression of CD27 and CX3CR1, we characterized the three subsets of CD8+ T cells (CD27lo CX3CR1−, CD27hi CX3CR1− and CX3CR1+) in mice treated with ICI therapy (Fig. 2a). In CT26 tumor-bearing mice treated with CTLA-4 and PD-L1 blockade therapy, PB CX3CR1+ CD8+ T cells had low expression of L-selectin (CD62L) and CXCR3, trafficking receptors required for entry of blood-borne T cells across lymphoid organ high endothelial venules (HEV) and the tumor microvasculature, respectively (Fig. 2b)41,42. These findings are in parallel to our recent study and others showing that peripheral virus- and tumor-specific CD8+ T cells expressing high levels of CX3CR1 exhibit decreased CD62L and CXCR3 expression, and are predominantly located within the circulation30,43. Of note, the CX3CR1+ fraction was largely positive for PD-1 which is reported to be expressed in PB neoantigen-specific CD8+ T cells in patients44. Further profiling of three subsets of CD8+ T cells in CT26 tumor-bearing in Balb/c mice treated with anti-CTLA-4/PD-L1 therapy and MC38 tumor-bearing C57BL/6 mice treated with anti-PD-L1 monotherapy in spleen revealed that the CX3CR1+ subset was notable for increased expression of granzyme A, 4-1BB, TIM3, and KLRG1 regardless of mouse strain, tumor types, and monotherapy or combination ICI therapy (Fig. 2c and Supplementary Fig. 3), and that CX3CR1+ CD8+ T cells represent a subset of recently activated effector cells in agreement with prior studies29,30,31,33.

CX3CR1 but not Ki-67 is stably upregulated in PB CD8+ T cells during ICI therapy

To gain further insight into the PB CX3CR1+ CD8+ T cells, we evaluated the expression of the nuclear protein Ki-67, a marker for proliferation that is upregulated in subsets of PB CD8+ T cells in response to ICI8,9,24,25,26,27. The levels of Ki-67 expression in the CX3CR1+ CD8+ T cells were significantly higher than in the CX3CR1−(CD27lo CX3CR1− and CD27hi CX3CR1−) subsets 2 weeks after anti-CTLA-4/PD-L1 therapy in CT26 tumor-bearing mice (Fig. 3a). We next examined whether increased expression of CX3CR1 on PB CD8+ T cells is transient or sustained during anti-CTLA-4/PD-L1 therapy. Increased frequency of CX3CR1+ subset was seen in both CD8+ and Tet+ CD8+ T cells starting from day 7, which remained high during treatment (Fig. 3b). In contrast, Ki-67 expression peaked at day 14, and returned to the baseline at day 21 (Fig. 3b).

Tumor-specific CD8+ T cells are enriched in the CX3CR1+ subset in PB

Identification of circulating T-cell biomarkers that enrich tumor-reactive T cells may facilitate the discovery of dynamic predictive markers of response to ICI. First, we assessed the frequency of Tet+ CD8+ T cells within the PB CX3CR1+ and CX3CR1− subsets in MC38 and CT26 tumor-bearing mice (Supplementary Fig. 4). We found more Tet+ CD8+ T cells in the CX3CR1+ subset than in the CX3CR1− subsets even before treatment although the frequency varied between individual mice (Fig. 3c). Moreover, the frequency of Tet+ CD8+ T cells remained higher in the CX3CR1+ subset than in the CX3CR1− subsets in both tumor models (Fig. 3c), suggesting that the PB CX3CR1+ subset is enriched with tumor-specific CD8+ T cells.

Clonally expanded TCR repertoires of CD8+ TILs are enriched in the peripheral CX3CR1+ subset during ICI therapy

The high frequency of tumor-specific CD8+ T cells in the CX3CR1+ subset before and during ICI treatment is suggestive that heterogeneous tumor-infiltrating CD8+ T cells are also enriched in the CX3CR1+ subset. To this end, we performed TCR sequencing on isolated CD8+ TILs and splenic CD27lo CX3CR1−, CD27hi CX3CR1−, and CX3CR1+ CD8+ T cells from MC38 tumor-bearing mice treated with combined anti-CTLA-4/PD-L1 therapy (Supplementary Fig. 5a, b). Comparison of the TCR repertoire in CD8+ TILs and three subsets of splenic CD8+ T cells demonstrated a high degree of overlap in TCR usage between CD8+ TILs and splenic CX3CR1+ CD8+ T cells (Fig. 4a and Supplementary Fig. 6) as determined by the Morisita’s overlap index45.

Next, we evaluated TCR clonality in CD8+ TILs and three subsets of splenic CD8+ T cells 2 weeks after anti-CTLA-4/PD-L1 therapy (Fig. 4b). The 100 most abundant TCR clones (colored) comprised more than 50% of TCR sequences in the splenic CX3CR1+ subset and CD8+ TILs while the majority of TCR sequences in the splenic CX3CR1−subsets were constituted of less frequent clones (purple). To quantify the skewness of the clonal distribution, we measured the Gini index, and found similarly higher clonality in the splenic CX3CR1+ subset and CD8+ TILs compared to splenic CX3CR1−subsets. Lorenz curves for splenic CX3CR1+ subset and CD8+ TILs were far from the equidistribution line, suggesting unequal distribution and skewing of the TCR repertoire.

ICI therapy induces a high degree of TCR sequence similarity and clonality between CD8+ TILs and the peripheral CX3CR1+ CD8+ T cells

TCRs that recognize the same antigen may not be the exact TCR clonotypes but have highly homologous sequences and share similar sequence features46,47. Unlike Morisita’s overlap index, a bioinformatics program, ImmunoMap48, allows us to analyze biological sequence similarity between peripheral and intratumoral CD8+ T cells. The structural clones expanded in the splenic CX3CR1+ CD8+ T-cell subsets and CD8+ TILs were generally shared compared to the CX3CR1−subsets and CD8+ TILs as visualized by the dendrograms (Fig. 4cand Supplementary Fig. 7a, b) and by tracking dominant motifs (Fig. 4d) in 2 weeks after anti-CTLA-4/PD-L1 therapy.

Analysis of the top six dominant CDR3β amino acid (AA) sequences in splenic CX3CR1+ CD8+T cells and CD8+ TILs revealed they shared a highly frequent AA sequence (CASSLVGNQDTQYF) in all three independent experiments (Supplementary Table 1, yellow highlighted and Supplementary Data 1). Although the most frequent AA sequence, CASSPRLGDNYAEQFF, in splenic CX3CR1+ CD8+ T cells was not identified in CD8+ TILs in the same experiment (Exp. #1 in Supplementary Table 1a), this AA sequence had a high degree of sequence homology with dominant AA sequences, CASSPGYAEQFF and CASSPGQGYAEQFFin CD8+ TILs, located in the same branch of the dendrogram (Supplementary Table 1b, blue highlighted and Supplementary Fig. 7a). Similarly, abundant AA sequences, CASSPGRGYEQYF in splenic CX3CR1+ CD8+ T cells and CASSSGTYEQYF in CD8+ TILs clustered largely in the same branch, indicating that they shared a high degree of similarity (Exp. #2 in Supplementary Table 1, green highlighted and Supplementary Fig. 7b). Collectively, these findings suggest TCR repertoires in peripheral CX3CR1+ CD8+ T-cell clones reflect the TCR repertoires in CD8+ TILs, and CX3CR1 on PB CD8+ T cells may act as a dynamic biomarker during the course of effective anti-CTLA-4/PD-L1 therapy.

Expansion of the CX3CR1+ subset in PB CD8+ T cells correlates with improved response to anti-PD-1 therapy and survival in patients with NSCLC

We next explored whether changes in the frequency of the CX3CR1+ subset in PB CD8+ T cells correlate with response to ICI in patients. We evaluated 36 patients with NSCLC treated with anti-PD-1 Ab (pembrolizumab or nivolumab). PB was prospectively obtained at baseline (before treatment initiation) and every 3–6 weeks during therapy for 12 weeks. All patients had pretreatment tumor tissue available to assess PD-L1 expression. When additional tumor tissues were available, the frequency of TILs and TMB was also analyzed as described before49. Baseline characteristics of 36 NSCLC patients are described in Supplementary Table 2. Clinical response in individual patients was derived from investigator-reported data per iRECIST criteria50 at the 12 week time point. Overall response rates (ORR) which include a complete response (CR) and partial response (PR) were 36.7% and 20% for patients with a PD-L1 tumor proportion score (TPS) of 50% or greater and 1–49%, respectively, in line with previous studies51,52.

We analyzed the frequency of the CX3CR1+ subset in PB CD8+ T cells from 36 NSCLC patients treated with anti-PD-1 Ab (Fig. 5a). The median baseline frequency of the CX3CR1+ subset among CD8+ T cells was 32.3% (6.1–76.3%) with no difference of overall survival (OS) between the high- and low-frequency groups (Fig. 5b). Because the pretreatment frequency of CX3CR1+ CD8+ T cells was variable between patients, to assess the effect of anti-PD-1 therapy on NSCLC patient’s CD8+ T cells, we calculated the percent change from baseline in the frequency of the CX3CR1+ subset in PB CD8+ T cells at all post-treatment time points available (3–12 wk post-treatment initiation). We evaluated the largest change (maximal percent change) of the CX3CR1+ subset in PB CD8+ T cells from baseline by the given time point in responders and non-responders. The maximal percent change of the CX3CR1+subset was substantially higher in responders than non-responders as early as 3 weeks from the initiation of the treatment (Fig. 5c). Next, we obtained estimates of the area under the curve (AUC) and corresponding 95% confidence interval (CI) using a logistic regression model, and the optimal cutoff score for discriminating between groups using the Youden’s index criterion53. These analyses revealed that an increase of CX3CR1+ CD8+ T-cell subsets by 15.5–21.2% from baseline segregated responders from non-responders at 3–12 weeks associated with higher odds ratio, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) (Supplementary Table 3 and Supplementary Fig. 8). Hereafter the percent change of the CX3CR1+ subset in PB CD8+ T cells from baseline are designated as a “CX3CR1 score”. Figure 5d shows longitudinal CD8 T-cell responses in individual patients. We found at least 20% increase of the CX3CR1 score in 92.3% (12/13) of responders compared to only 13.0% (3/23) of non-responders.

Based on these results, we hypothesized that a cutoff of at least 20% increase of the CX3CR1 score would correlate with response to anti-PD-1 therapy and assessed the association between the CX3CR1 score and objective response. The maximal CX3CR1 score of at least 20 started to associate with ORR at 3 weeks (P = 0.0152) (odds ratio: 16.0; 95% CI: 1.5–171.2), and became a strong correlate of response at 9 weeks (P < 0.0001) (odds ratio: 36.7; 95% CI: 5.3–253.8) (Fig. 5e). Next, we analyzed the corresponding sensitivity, specificity, PPV, and NPV of the CX3CR1 score and the PD-L1 TPS. The maximal CX3CR1 score of at least 20 demonstrated remarkably high PPV, NPV, sensitivity, and specificity, and identified response in 21/27 (77.8%), 28/36 (77.8%), 31/36 (86.1%), and 32/36 (88.9%) at 3, 6, 9, and 12 weeks, respectively, while a PD-L1 TPS of at least 50% had suboptimal PPV and specificity, and correctly identified response only in 16/36 (44.4%) (Table 1). Notably, tumor tissues were available for assessing the frequency of TILs and TMB for only 66.6% (24/36) and 61.1% (22/36) of NSCLC patients, respectively (Supplementary Table 4), suggesting limitation of these analyses, in line with previous reports54,55. Lastly, we evaluated the correlation between the CX3CR1 score and survival. The median time of follow-up was 22.3 months (range 3.3–35.5). We found that at least 20% increase of the CX3CR1 score by 12 weeks was associated with better progression-free survival (PFS) (hazard ratio for death or disease progression: 0.28; 95% CI: 0.13–0.62; P = 0.0033) and OS (hazard ratio for death, 0.24; 95% CI: 0.09–0.61; P = 0.0136) (Fig. 5f). The median PFS and OS among patients with a CX3CR1 score of <20% were 5.7 and 8.6 months, respectively, while the median PFS and OS among patients with a CX3CR1 score of at least 20% by 12 weeks were 19.5 months and not reached.

The majority of patients (86.1%: 31/36) in our cohort had NSCLC with a PD-L1 TPS of at least 50% and was treated with pembrolizumab, which was approved by the U.S. Food and Drug Administration in 2016. Therefore, we evaluated the relationship between the CX3CR1 score and response to anti-PD-1 therapy in this population. The CX3CR1 score was a correlate of response, and at least 20% increase of the CX3CR1 score by 12 weeks was associated with better clinical outcome by ORR, PFS, and OS (Supplementary Fig. 9a–c). Although the number is small, the CX3CR1 score was a correlate of response in patients with a PD-L1 TPS < 50% (Supplementary Fig. 9d). We also evaluated whether prior treatment with chemotherapy or radiation might have affected the utility of the CX3CR1 score, but did not observe the impact of either prior chemotherapy or radiation on the biomarker performance (Supplementary Fig. 9e). Taken together, the CX3CR1 score is highly correlated with a patient’s clinical response and survival early on-treatment.

Of note, there was no significant difference in the maximal percent change of CD8+ T cells in PB CD3+ T cells from baseline between responders and non-responders at any time point (Supplementary Table 5 and Supplementary Fig. 10a). Accordingly, changes in the frequency of CD8+ T cells among PB CD3+ T cells did not correlate with OS and PFS in all NSCLC patients (Supplementary Fig. 10b) and NSCLC patients who had a PD-L1 TPS of at least 50% and were treated with pembrolizumab (Supplementary Fig. 10c).