Plasma fractalkine contributes to systemic myeloid diversity and PD‐L1/PD‐1 blockade in lung cancer

Abstract Recent studies highlight the importance of baseline functional immunity for immune checkpoint blockade therapies. High‐dimensional systemic immune profiling is performed in a cohort of non‐small‐cell lung cancer patients undergoing PD‐L1/PD‐1 blockade immunotherapy. Responders show high baseline myeloid phenotypic diversity in peripheral blood. To quantify it, we define a diversity index as a potential biomarker of response. This parameter correlates with elevated activated monocytic cells and decreased granulocytic phenotypes. High‐throughput profiling of soluble factors in plasma identifies fractalkine (FKN), a chemokine involved in immune chemotaxis and adhesion, as a biomarker of response to immunotherapy that also correlates with myeloid cell diversity in human patients and murine models. Secreted FKN inhibits lung adenocarcinoma growth in vivo through a prominent contribution of systemic effector NK cells and increased tumor immune infiltration. FKN sensitizes murine lung cancer models refractory to anti‐PD‐1 treatment to immune checkpoint blockade immunotherapy. Importantly, recombinant FKN and tumor‐expressed FKN are efficacious in delaying tumor growth in vivo locally and systemically, indicating a potential therapeutic use of FKN in combination with immunotherapy.


Introduction
Non-small-cell lung cancer (NSCLC) remains a leading cause of cancer death. Immune checkpoint blockade (ICB) immunotherapies such as anti-PD-L1/PD-1 therapies have yielded remarkable clinical results. Nevertheless, these treatments fail in most patients. The mechanisms of resistance to ICB therapies have traditionally been studied within the tumor microenvironment (TME). However, PD-L1/PD-1 blocking antibodies are administered systemically, acting over a variety of systemic immune populations that may contribute to clinical outcomes Rashidian et al, 2019;Peng et al, 2020). Accumulating evidence is showing that functional systemic immunity in cancer patients before starting immunotherapies is required for ICB success (Mathios et al, 2016;Spitzer et al, 2017;Rashidian et al, 2019;Zuazo et al, 2019Zuazo et al, , 2020Kagamu et al, 2020;Ferrara et al, 2021;Horton et al, 2021). Indeed, we and others have recently demonstrated that functional systemic T-cell immunity before ICB is a major requirement for its efficacy (Spitzer et al, 2017;Zuazo et al, 2019;Kagamu et al, 2020;Horton et al, 2021). For instance, NSCLC patients with dysfunctional T cells failed to respond to PD-L1/PD-1 monotherapies .
T-cell activities are largely regulated through antigen presentation provided by different myeloid cell types. In turn, myeloid cell differentiation and activities are disturbed by cytokines and plasma factors in cancer patients (Robb, 2007). These factors can alter the myeloid cell compartment, and hence disturb T-cell immunity. For example, tumors secreting different cytokines such as GM-CSF, G-CSF, M-CSF, or IL6 cause systemic expansion of myeloid-derived suppressor cells (MDSCs) that inhibit anti-tumor T cells and NK cells (Liechtenstein et al, 2014;Gato-Canas et al, 2015;Ortiz-Espinosa et al, 2022). By contrast, elevated numbers of circulating HLA-DR + monocytes and decreased neutrophils are generally good prognostic factors for ICB therapies (Krieg et al, 2018). Therefore, specific cytokine profiles may strongly influence ICB outcomes by altering systemic immunity. Indeed, some of these cytokines have been associated with good prognosis such as IL2, IFNa, or IL12 and have even been used in immunotherapies (Propper & Balkwill, 2022).
A cytokine with ambivalent roles in cancer is fractalkine (FKN), a membrane-bound chemokine of the CX3C family, which mediates cellular adhesion. This protein is mainly expressed by endothelial cells and signals through its receptor CX3CR1 and integrins (Fujita et al, 2012). A soluble form is released by proteolysis mediated by ADAM-10 and ADAM-17 (Garton et al, 2001;Hundhausen et al, 2003), which regulates immune chemotaxis. FKN has been reported to elicit anti-tumor responses by facilitating immune cell infiltration into tumors (Lavergne et al, 2003;Ohta et al, 2005;Xin et al, 2005;Tang et al, 2007;Hyakudomi et al, 2008;Park et al, 2012;Kee et al, 2013;Yamauchi et al, 2020Yamauchi et al, , 2021. On the other hand, FKN promotes tumor migration and invasion of circulating tumor cells towards tissues displaying high FKN expression (Korbecki et al, 2020) (Shulby et al, 2004;Marchesi et al, 2008;Erreni et al, 2010;Gaudin et al, 2011;Jamieson-Gladney et al, 2011;Kim et al, 2012;Marchica et al, 2019). Thus, FKN plays context-dependent roles in anti-tumor immunity, and its role in ICB immunotherapies is still poorly understood.
Here, we performed an extensive profiling of systemic myeloid cell populations and plasma soluble factors in NSCLC patients undergoing PD-L1/PD-1 blockade therapies, before and during treatment. Responder patients showed an elevated diversity of myeloid cell types, enriched in activated monocytic cells, and decreased granulocytic populations. Plasma FKN concentration correlated with myeloid cell diversity and response to PD-L1/PD-1 blockade. The mechanisms underlying its implication in PD-L1/PD-1 blockade were explored both in vitro and in vivo, as well as its potential use as a therapeutic approach in NSCLC.

Baseline plasma FKN is associated with myeloid diversity and objective clinical responses
An extensive characterization of circulating immune cell populations was performed by high-dimensional flow cytometry (HDFC) (Qiu et al, 2011;Krieg et al, 2018;Kagamu et al, 2020) in a wellcharacterized cohort of 112 NSCLC patients (Table EV1A-D) undergoing PD-L1/PD-1 blockade immunotherapy. A panel of 43 lineage, differentiation, and activation markers was used to label freshly isolated peripheral blood mononuclear cells (PBMCs) before the initiation of immunotherapies (baseline) and after the first cycle of treatment. Samples from age-matched healthy donors were used as controls. Patient groups were classified into long-term responders, stable disease, and progressors, with some patients classified as hyperprogressors in our previous studies Arasanz et al, 2020).
High-dimensional hierarchical clustering was performed over the phenotypes of baseline CD11b + myeloid immune populations. Each response group was characterized by distinct cluster profiles ( Fig  EV1). Responders showed high phenotypic diversity with a dominance of the monocytic lineage, in contrast to nonresponders who exhibited elevated percentages of neutrophils and granulocytic myeloid-derived suppressor cells (G)-MDSCs (Fig EV1A and C). A diversity index (DI) was calculated for each patient as the number of terminal phenotype clusters in HDFC profiles generated with 200,000 CD11b + cells. Objective responders exhibited increased DI (P < 0.0001) compared with nonresponders before starting immunotherapy (DI = 18.63 AE 2.36 vs. 12.21 AE 3.31; Mean AE SD), although still inferior to that of healthy donors (DI = 29.86 AE 3.71; Mean AE SD) ( Fig 1A). Elevated diversity indexes were significantly associated with objective clinical responses as evaluated by ROC analysis (area under the curve, AUC, 0.9216 [95% confidence interval = 0.8601 to 0.9832]; P < 0.0001; Fig 1B). A DI cut-off value of 13.5 identified responders with 100% sensitivity and 63% specificity, which stratified patients with significantly (P < 0.0001) increased progression-free survival (PFS) ( Fig 1C) and overall survival (OS; P = 0.007; Fig 1D).
Patients classified according to clinical responses showed differential myeloid profiles (Fig EV2A and B). Objective responders presented a myeloid signature dominated by monocytic CD14 + HLA-DR + cells and a low abundance of granulocytic CD66b + myeloid cells (neutrophils and G-MDSCs; Fig EV2C and D). Relative percentages of these myeloid lineages were significant predictors of responses by ROC analyses (Fig EV2E and F). Then, we tested whether DI correlated with relative percentages of monocytes and granulocytes. Albeit data dispersion expected from the inherent heterogeneity of clinical samples, elevated DI was a positive correlator with increased percentages of CD14 + HLA-DR + monocytes within CD11b + cells (Fig 1E), and a negative correlator with increased percentages of systemic CD66b + granulocytic cells (Fig 1F).
In light of these results, we wondered whether plasma factors correlating with myeloid function and clinical responses could be identified in our cohort. Therefore, a panel of 65 cytokines, chemokines, and soluble immune checkpoints was evaluated. While most of the quantified analytes showed increased levels in patients who rendered no objective clinical responses to anti-PD-1/PD-L1 immunotherapy (Table EV2), only FKN concentrations were significantly (P = 0.0037) elevated in a sample of responders before treatment (Fig 1G; Table EV2). A validation cohort independently confirmed our results in a limited sample of patients ( Fig EV3A). FKN plasma concentrations in clinical samples positively correlated with myeloid diversity in our discovery cohort ( Fig 1H). A cut-off value for FKN concentration above 55 pg/ml was identified by ROC analysis (Fig 1I) for significant benefit in PFS (P = 0.003) and OS (P < 0.001; Fig 1J and K). To find out whether FKN had prognostic value in our cohort for treatments other than immunotherapies, we analyzed the time elapsed between diagnosis and progression in our discovery cohort during which patients were undergoing chemotherapy only. No significant differences were found, thus discarding that plasma FKN had significant prognostic value in our cohort ( Fig 1L).

Secreted FKN possesses potent anti-oncogenic and immunemodulating activities
The better therapeutic outcome of patients with increased plasma FKN concentrations suggested that FKN could possess potential anti-oncogenic properties in lung cancer. Therefore, we decided to assess its potential use as a therapeutic agent. Fractalkine is known to be produced by healthy tissue. To find out whether lung cancer cells could also be a source of FKN in lung cancer patients, we quantified in vitro FKN secretion by a collection of 10 human lung adenocarcinoma (LUAD) and lung squamous carcinoma (LUSC) cell lines. FKN production was very heterogeneous, and only two cell lines out of 10 clearly overexpressed FKN (Fig 2A). This was in agreement with our clinical observations since a minor percentage of NSCLC patients showed plasma FKN concentrations above the cut-off value ( Fig 1I). Due to the unavailability of tumor tissue biopsies from our discovery cohort, we evaluated FKN expression in a limited sample of biopsies from another cohort of LUAD patients. Double immunostainings were performed to detect the lung tumor marker Pan-Cytokeratin (PanCK) together with FKN in tumor and adjacent healthy tissue (Fig EV3B and C). FKN expression was inversely correlated with PanCK expression within both adjacent nontumor and LUAD tissues (Pearson's correlation: r = À0.9016, P < 0.001 [adjacent nontumor tissues] and r = À0.8518; P < 0.002 [LUAD tissues;Fig EV3D and E]). In vitro results together with biopsy data were indicators that FKN was poorly expressed in tumor tissue in the majority of lung cancer patients. These data were confirmed by evaluating FKN transcriptomic expression in clinical samples from lung cancer patients registered in The Cancer Genome Atlas (TCGA) database ( Fig EV3F).
To find out whether FKN could trigger anti-tumor mechanisms in NSCLC patients, the FKN-CX3CR1 signaling axis was systemically targeted in a mouse model of lung adenocarcinoma. In agreement with our hypothesis, the pharmacological inhibition of CX3CR1 by the administration of the allosteric noncompetitive FKN antagonist AZD8797 accelerated tumor progression (P < 0.05; Fig 2B). This result was supported by the systemic administration of recombinant FKN, which delayed tumor growth, compared with the null effect of a nonrelevant control protein (IgG) (Fig 2C). To overcome the challenging therapeutic schedule with recombinant FKN in mouse models, we overexpressed FKN directly from cancer cell lines, as observed for some human lung adenocarcinoma cell lines such as H358 and SW900 (Fig 2A). To that end, murine adenocarcinoma cell lines Lacun-3 (Bleau et al, 2014) and 3LL were engineered to express FKN (Fig EV3G). Since a tendency for increased concentrations of hematopoietic growth factors such as GM-CSF and G-CSF were identified in the plasma of responder patients (Table EV2), and a correlation between FKN levels and myeloid diversity was also found (Fig 1H), we included cancer cell lines overexpressing key hematopoietic growth factors as controls ( Fig EV3G). The two FKNexpressing LUAD cell lines showed impaired cell growth in vitro (Figs 2D and EV3H). Interestingly, GM-CSF expression from cancer cells significantly enhanced 3LL proliferation (Figs 2E and EV3H). By contrast, FKN was a strong inhibitor of tumor cell growth rate and an inductor of apoptosis of cancer cells .
To evaluate the growth dynamics of tumors expressing FKN and control molecules, 3LL cell lines expressing FKN, GM-CSF, G-CSF, and M-CSF were subcutaneously transplanted in groups of mice. FKN secretion strongly impaired (P = 0.0002) 3LL tumor growth ( Fig 2F) and increased median survival from 17 days in the control group to 30.5 days (P = 0.0005; Fig 2G). G-CSF and M-CSF expression also delayed tumor growth and increased survival but to a lesser extent. In agreement with in vitro data, GM-CSF expression by 3LL cells significantly accelerated tumor progression. These A Diversity indexes as a function of clinical responses. Each dot represents a biological replicate. B ROC curve of the diversity index as a function of objective responses vs disease progression. C Kaplan-Meier plot of PFS stratifying the patients according to high or low diversity index. N = 69 (33 above the cut-off, 36 below). D As in (C) but plotting OS. N = 72 (34 above the cut-off, 38 below). E Correlation between the percentage of monocytes and diversity index. Each dot represents a biological replicate. F Correlation between the percentage of granulocytic myeloid cells and diversity index. Each dot represents a biological replicate. G Baseline plasma FKN concentrations in responders (R, n = 26), progressors (PR, n = 62), hyperprogressors (HPR, n = 3), age-matched healthy donors (H, n = 32), and patients who were not eligible for immunotherapy (NO IT, n = 28). Sample sizes represent biological replicates. Error bars are shown (standard deviations, SD), and relevant statistical comparisons by Mann-Whitney U tests are shown in the graphs. H Correlation between FKN plasma concentration and diversity index with the Spearman's test. Each dot represents a biological replicate. I ROC analysis of FKN concentration as a predictor of objective clinical responses. The calculated cut-off value for the statistics provided in the graph is shown. J Kaplan-Meier plot of PFS stratifying the patients according to the FKN cut-off value identified in the ROC curve (n = 42). The associated P-value is shown. K As in (J) but plotting OS. n = 46 (17 above the cut-off, 29 below). L Time elapsed from disease diagnosis to the beginning of immunotherapy in second-line treated patients. Each dot represents a biological replicate.
Data information: Multicomparisons in dot plots were carried out by the Wilcoxon test. Pairwise comparisons were performed by the Mann-Whitney's U test. Survival differences were tested with the log-rank test and correlation plots with the Spearman's test. *, **, ***, indicate significant (P < 0.05), very significant (P < 0.01) and highly significant (P < 0.001) differences. ns, nonsignificant differences. We then studied whether FKN production by tumors altered the composition of circulating immune cells. First, we confirmed that mice carrying FKN-producing 3LL tumors had elevated FKN concentrations in plasma compared with mice bearing unmodified 3LL tumors (0.85 ng/ml AE 0.11 vs. 0.36 ng/ml AE 0.05; Means AE SD). Even so, mice with tumors showed a reduction in circulating FKN compared with control mice (1.32 ng/ml AE 0.03) at least at early time points with relatively small tumors (67.47 AE 25.39 mm 3 in FKN-producing tumor-bearing mice vs 516.3 AE 194.1 mm 3 in wildtype control group; Fig 3A).
Profiling of circulating immune cells was performed 15 days after tumor engraftment, once significant differences in tumor volume were observed ( Fig 3B). In our murine experimental models, the composition of myeloid and lymphoid compartments in tumorbearing mice was apparently similar to that of healthy control mice with the exception of GM-CSF-and G-CSF-producing tumors ( Fig 3C). When looked at in detail and in agreement with our clinical data, mice with 3LL-FKN tumors had elevated circulating DC, macrophages, and monocytes, accompanied by a significant decrease (P < 0.01) in the percentage of Ly6G + granulocytes (12.25% AE 2.36) compared with control mice with 3LL tumors (29.2% AE 9.9; Fig 3D). By contrast, mice bearing 3LL/GM-CSF tumors showed profiles comparable to that of human hyperprogressors, characterized by an exponential acceleration of tumor growth (P < 0.0001) tumor growth, shortened survival (median survival of 14 days vs. 17 days of wild-type control, P = 0.009) and overwhelming numbers of MDSCs (58.52% AE 9.17 of Ly6G + cells).

FKN sensitizes lung adenocarcinoma tumors intrinsically resistant to anti-PD-1 immunotherapies
Our clinical data associated elevated plasma FKN concentrations with objective clinical responses to PD-L1/PD-1 blockade. Therefore, we investigated whether FKN contributed to ICB efficacy in the murine 3LL lung adenocarcinoma model, which is intrinsically refractory to these immunotherapies. Indeed, while unmodified 3LL tumors were fully resistant to PD-1 blockade in vivo, FKN expression sensitized tumors to PD-1 blockade. This sensitization led to a highly significant (P < 0.001) reduction in tumor size ( Fig 4A) although with limited impact on survival (Fig 4B and C). It has nevertheless to be noted that in our mouse models, FKN expression is a major driving force of anti-tumor activities, causing a very highly significant delay (P < 0.0001) in tumor growth ( Fig 4B) associated with a very significant (P = 0.001) increase in median survival from 20 to 35 days ( Fig 4C).
Then, we studied whether the combination of FKN expression and PD-1 blockade had a further impact on the composition of systemic immune cells. These studies were carried out at a time point with significant differences in tumor volumes (15 days after tumor engraftment). We characterized the relative composition of major immune cell types within CD45 + cells in spleens, to detect systemic changes in our mouse models. No major impact of the anti-PD-1 immunotherapy was observed on top of FKN at least within myeloid cell types ( Fig EV4A). In agreement with our clinical data and previous experiments, FKN expression increased the relative percentage of classical and nonclassical monocytes (Fig 4D and E), and a reduction in neutrophils ( Fig 4F). Anti-PD-1, however, increased the percentage of PD-1 + T cells ( Fig 4G) but not of NK cells (Fig 4H). Then, we evaluated whether these changes had a reflection on tumor infiltration. FKN expression significantly enhanced immune infiltration of the tumor (Fig 4I), mainly with CD4 T and NK cells (Fig 4J and K). No significant differences were observed for the rest of the main cell types ( Fig EV4B). As we lacked biopsy data from our cohort of patients, we complemented the results in mouse models with analyses of transcriptional data from lung adenocarcinoma samples in the TCGA database. The Tumor Immune Estimation Resource tool (TIMER2.0) was used to estimate immune infiltration and perform correlations with FKN transcription (Li et al, 2020). Positive correlations of FKN expression were found with infiltration by CD8 (Rho = 0.222, P = 6.28e-07), CD4 (Rho = 0.185, P = 3.66e-05), NK (Rho = 0.244, P = 4.07e-08) cells, neutrophils (Rho = 0.211, P = 2.39e-06) and monocytes (Rho = 0.263, P = 2.96e-09) (Fig EV5A). By contrast, a negative correlation with MDSC infiltration was observed (Rho = À0.269, P = 1.21e-09). The correlation of FKN with PD-L1 tumor expression was also confirmed, as PD-L1 is a clinically-◀ Figure 2. FKN inhibits lung adenocarcinoma cell growth.
A Secreted FKN in cell cultures of a collection of 10 LUAD and LUSC cancer cell lines quantified by ELISA. Results are presented as mean AE SD. Samples were assayed in duplicates from a pool of three independent replicates. Statistical differences among groups were analyzed by ANOVA followed by Tukey's tests. B Tumor growth curves of 3LL tumor-bearing mice treated with recombinant FKN (rFKN), the CX3CR1 inhibitor AZD8797, a nonrelevant protein (IgG2a) or saline buffer as vehicle control (WT). Arrows indicate the time of intraperitoneal injections of rFKN (blue) and AZD8797 (black). Data are expressed as mean AE SD (n = 6 mice per group). Statistical comparisons by ANOVA followed by Tukey's pairwise comparison tests are provided. C Dot plot of 3LL tumor size in the indicated groups of mice at day 10 after cancer cell inoculation. Statistical comparisons by ANOVA and Tukey's pairwise comparison tests are indicated. D Proliferation of two LUAD murine cell lines overexpressing FKN and their parental unmodified counterparts (WT), as indicated. Real-time cell growth was monitored by RTCA. Delta-cell indexes are expressed as mean AE SD from three independent replicates. Differences in delta-cell index proliferation data were tested by ANOVA following Tukey's pairwise comparison tests after 40 h of culture. E Bar graphs with cell growth rates for the indicated 3LL cell lines relative to the growth of unmodified 3LL cells. Error bars are shown (SD). Statistical comparisons among three independent biological replicates by ANOVA and Tukey's pairwise comparison tests are indicated. F In vivo tumor growth of engrafted 3LL cell lines producing the indicated cytokines. Data are expressed as mean AE SD (n = 6 mice per group). Comparisons between groups were performed by ANOVA and Tukey's pairwise comparison tests. G Kaplan-Meier survival plots. Differences between the control group (WT) and the FKN group were evaluated by a two-sided log-rank test.
Most of the experiments in mouse models were carried out by expressing FKN from cancer cells. It could be argued that anti-tumor effects were restricted to the FKN-expressing tumors, while the clinical data ( Fig 1G) and treatments with recombinant FKN (Fig 2B and  C) strongly suggested systemic anti-tumor activities by circulating FKN. Therefore, to demonstrate systemic activities for secreted FKN, 3LL-FKN cells were subcutaneously implanted in mice on the left flank (inoculation side), followed by injection of unmodified 3LL cells on the right flank (target tumor side) 1 week later (Fig 5A). Mice were then treated with anti-PD-1 immunotherapy and tumor growth was monitored at both flanks. Interestingly, the growth of 3LL tumors was delayed and comparable to 3LL-FKN tumors in the same mice (Fig 5B). These results demonstrated that secreted FKN associated with anti-PD-1 blockade could delay the growth of an immunotherapy-resistant distal tumor that did not express FKN. Therefore, FKN secreted from modified tumor cells showed efficacious systemic therapeutic activities.
To identify the anti-tumor effector cells in our experimental systems, CD4, CD8, and NK cells depletions were carried out following standard procedures in our group (Ortiz-Espinosa et al, 2022), and the growth of 3LL-FKN tumors was monitored. Interestingly, NK cell abrogation restored the growth of 3LL-FKN tumors to rates comparable to unmodified 3LL tumors (Fig 5C). Limited effects were observed by CD4 and CD8 T-cell depletion. These results suggested that the main effector immune cell types were NK cells with a minor contribution by T cells. NK depletion shortened median mice survival down to 24 days, compared with 40.5 days of median survival in the FKN control group (P = 0.0051; Fig 5D). We hypothesized that FKN and anti-PD-1 could be systemically reducing the percentage of NK cells expressing inhibitory checkpoints and exhaustion markers other than PD-1. Indeed, a significant decrease in KLRG1 + , TIGIT + , and TIM3 + NK cells was observed (Fig 5E-G). No significant differences were observed in T cells, in agreement with our depletion data (Fig 5H-J).  (Kargl et al, 2019;Koh et al, 2020;Veglia et al, 2021). MDSCs and granulocytes are strong T-cell suppressors in cancer through several mechanisms, as extensively reviewed elsewhere (Veglia et al, 2018;Grover et al, 2021). On the other hand, elevated percentages of HLA-DR + monocytes correlate with efficacious PD-1 blockade immunotherapy (Krieg et al, 2018). Hence, profiling of immune cell composition and functionality in peripheral blood could be used for the selection of potential responder patients in PD-L1/PD-1 blockade immunotherapies. Following this reasoning, we performed detailed analyses of myeloid cell subsets and their differentiation stages in a discovery cohort of NSCLC patients before starting immunotherapies by HDFC. Here, we confirmed that an elevated diversity of circulating myeloid cell types was characteristic of objective responders to PD-L1/PD-1 blockade. It could be argued that a diverse composition of peripheral immune cell phenotypes could reflect functional myelopoiesis (Schultze et al, 2019) in patients before starting ICB. Indeed, the elevated diversity of phenotypes was represented by an expansion of activated classical monocytes with a concomitant decrease in granulocytic myeloid cells. Our data confirmed that in addition to their classical prognostic value, these myeloid cell types show significant predictive power in PD-L1/PD-1 blockade immunotherapies by ROC analyses, in agreement with previous studies with highdimensional analytical techniques (Gubin et al, 2018;Krieg et al, 2018;Arasanz et al, 2022). Our high-dimensional profiles included (but were not restricted to) DC, monocytes, macrophages, neutrophils, and MDSCs at various differentiation and activation stages, and demonstrated that the balance between these immune cell types is critical for ICB outcome before starting immunotherapies.

Discussion
On the lookout for plasma factors associated with (i) high systemic myeloid cell diversity, (ii) elevated monocyte populations, and (iii) increased survival, we found that only FKN fulfilled these three conditions. Hence, in our discovery cohort FKN plasma concentration showed good predictive value, which was confirmed in a ◀ Figure 3. Alteration of immune composition of peripheral blood by tumors expressing cytokines.
A FKN plasma concentration at day 15 in the indicated groups of tumor-bearing mice. Data are expressed as mean AE SD from a pool of 6 mice/group. Statistical comparisons were performed by ANOVA and Tukey's pairwise comparison tests. B Tumor volumes 15 days after tumor inoculation. Data are expressed as mean AE SD (n = 6 mice per group), and comparisons between groups were performed by ANOVA and Tukey's pairwise comparison tests. C Relative myeloid and lymphoid composition of peripheral blood at day 15 in mice transplanted with 3LL-expressing the indicated cytokines. WT, unmodified 3LL cells; Healthy, mice without tumors. D Percentage of the indicated peripheral immune cell populations at day 15 after injection of groups of mice (n = 6 mice per group) with 3LL cells overexpressing the indicated myeloid-regulating cytokines. The relevant immune populations were quantified as percentages of total leukocytes (CD45 + cells), specifying DCs (CD11c), macrophages (F4/80), Ly6C + monocytes, and Ly6C À monocytes, granulocytes (Ly6G), B cells (CD19), NKs (NK1.1), CD4 T cells (CD3 CD4) and CD8 T cells (CD3 CD8). Relevant statistical comparisons were performed by the Wilcoxon's test followed by pairwise comparisons of relevance by the Mann-Whitney U test. Box and whisker plots indicate median (central line), 25 th to 75 th percentiles (box), and minimum to maximum values (whiskers).

of 16
EMBO reports 24: e55884 | 2023 Ó 2023 The Authors validation cohort. In agreement with the clinical data, mice engrafted with FKN-producing tumors showed increased OS with elevated percentages of monocytes and reduced Ly6G + granulocytes in peripheral blood and spleen. Fractalkine was demonstrated to have anti-cancer properties both in vitro and in in vivo models for murine lung adenocarcinoma. Therapeutic activities were observed for injected recombinant FKN or produced by 3LL cells. Indeed, systemic pharmacological disruption of the FKN-receptor signaling axis accelerated the growth of adenocarcinoma tumors. Our results showed that local FKN expression by the tumor boosted stronger anti-tumor immune responses that resulted in faster tumor rejection.
It is important to note that FKN possessed systemic anti-tumor activities without the need to be expressed within the tumor Unmodified and FKN-expressing 3LL cells were subcutaneously inoculated in mice and tumors were allowed to grow for 7 days. Tumor-engrafted mice were treated intraperitoneally with anti-PD-1 antibody or vehicle (days 7, 11, and 15, as indicated by arrows) following randomization into two groups. Tumor growth was monitored. Data are presented as mean AE SD (n = 6 mice per group). C Kaplan-Meier survival plots of the indicated groups of mice. Survival differences between the control group (WT + aPD-1) and FKN + aPD-1 were evaluated by a two-sided log-rank test. D-H (D) Percentage of splenic Ly6C+ monocytes, (E) Ly6G À monocytes, (F) Ly6G + neutrophils, (G) PD-1 + T cells (CD3 + ), and (H) PD-1 + NK cells (NK1.1 + ). Data are presented as mean AE SD (n = 6 mice per group). I-K (I) Percentage of tumor infiltration with total leukocytes (CD45 + ), (J) CD4 T cells, and (K) NK cells (NK1.1 + ). Infiltration data are shown as mean AE SD (n = 8 mice per group).
Ó 2023 The Authors EMBO reports 24: e55884 | 2023 microenvironment. For example, FKN expressed by a tumor can exert inhibitory activities over a distal tumor that does not express FKN. This is important because it was previously thought that FKN expressed by tumors could exert anti-tumor effects through the recruitment of T cells, DCs, and NK cells to the tumor microenvironment (Lavergne et al, 2003;Ohta et al, 2005;Xin et al, 2005;Tang et al, 2007;Hyakudomi et al, 2008;Park et al, 2012;Kee et al, 2013;Yamauchi et al, 2020Yamauchi et al, , 2021. Nevertheless, we found that recombinant FKN was very unstable in vivo. The systemic administration of recombinant FKN as a therapeutic approach might require experimental strategies to effectively deliver cytokines to the TME in a targeted way while improving their stability. Anti-tumor systemic activities were mediated mainly by circulating NK cells, with a minor contribution by CD4 and CD8 T cells. It has been shown in some tumor types that the membrane-bound FKN can promote metastasis of CX3CR1 + circulating tumor cells towards tissues displaying a high CX3CL1 expression, such as bones, lungs, and nervous tissues (Shulby et al, 2004;Marchesi et al, 2008;Erreni et al, 2010;Gaudin et al, 2011;Jamieson-Gladney et al, 2011;Kim et al, 2012;Marchica et al, 2019;Korbecki et al, 2020). Our data clearly showed that at least in LUAD cancer models this was not the case. This dual behavior of FKN may probably be a consequence of the two different bioactive forms of the protein (Vitale et al, 2007;D'Haese et al, 2010;Winter et al, 2020). The membrane-bound form mediates cell adhesion processes and migration while the soluble version participates in the chemoattraction of effector cells to the tumor, but also in the modulation of systemic NK cells, which show less exhausted phenotypes.
Importantly, we demonstrated that FKN expression in adenocarcinoma cells refractory to ICB therapy (Ajona et al, 2020) could sensitize tumors to PD-1 blockade. This sensitization further retarded the growth of tumors, although in our mouse models, this did not translate into a benefit in survival. Nevertheless, it has to be mentioned that FKN exerted very potent anti-tumor activities, and improvements in survival would likely require blockers even stronger than classical anti-PD-1 antibodies.
Concluding, plasma FKN is a biomarker of systemic myeloid cell diversity, characterized by an elevation in circulating monocytic myeloid cells and a strong decrease in granulocytes. FKN plasma concentrations also correlate with increased survival in human lung adenocarcinoma patients treated with PD-L1/PD-1 blockade monotherapies. Our results demonstrate that the therapeutic activities of FKN are not restricted to the tumor microenvironment. These results open an avenue towards either utilizing FKN concentration as a biomarker or its development as a therapeutic agent.

Clinical samples and study design
One hundred and twelve patients diagnosed with locally advanced or metastatic NSCLC treated with the immune checkpoint inhibitors (ICI) nivolumab, pembrolizumab, and atezolizumab, or with the combination of chemo-immunotherapy (pembrolizumab + platinum-based chemotherapy) were recruited between December 2017 and October 2020 at the Oncology Department of the University Hospital of Navarra (Pamplona, Spain) (Table EV1). Hyperprogressive disease was defined by radiological criteria and immunological profiles as described in Arasanz et al (2020). The exclusion criteria consisted of concomitant administration of chemotherapy or previous immunotherapy treatment.
The current prospective observational study was approved by the Ethics Committee of Clinical Investigations at the University Hospital of Navarre (reference number: PI_2020/115). REMARK reporting guidelines were followed for the study. Informed consent was obtained from all subjects and all experiments conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. Samples were collected through the Blood and Tissue Bank of Navarre, Health Department of Navarre, Spain. Eligible patients were 18 years of age or older who received immunotherapy targeting PD-1/PD-L1 as their current standard of care. Tumor PD-L1 expression was quantified in 103 of these patients. Thirty-two age-matched healthy donors were recruited from the Valle de Salazar Nursing Home (Navarra, Spain) from whom written informed consents were obtained. The total number of donors was calculated a priori to ensure a power of 0.95 for F tests taking into consideration a large effect size (f = 0.4). Power calculations were carried out with Gpower 3.1.9.7 (Faul et al, 2009). Serum samples from 28 LUAD patients not eligible for immunotherapy treatment were obtained from the Hospital Fundaci on Jimenez D ıaz (Madrid, Spain). All patients provided written informed consent for participation in the study, approved by the Fundaci on Jim enez D ıaz (CEImJGD, ref. ER_EO180-19_FJD-HGV). As a validation cohort, 18 patients diagnosed with locally advanced or metastatic NSCLC treated with the combination of chemoimmunotherapy (pembrolizumab + platinum-based chemotherapy) as first-line treatment were recruited between October 2020 and October 2022 at the Oncology Department of the University Hospital of Navarra (Pamplona, Spain), from whom written informed consents were obtained. Samples were processed by M.G., and plasma FKN concentrations were determined by Luminex assays by B.T. Data are presented as mean AE SD (n = 6 mice per group). Comparisons among groups were performed by ANOVA and Tukey's pairwise comparison tests.
Ó 2023 The Authors EMBO reports 24: e55884 | 2023 Eight milliliters of peripheral blood samples were obtained prior to and during immunotherapy before administration of each cycle. PBMCs were isolated as described (Escors et al, 2008;Bocanegra et al, 2019;Zuazo et al, 2019), and myeloid cells were analyzed by flow cytometry. The participation of each patient concluded when a radiological test confirmed response or progression, with the withdrawal of consent or after death of the patient. Tumor responses were evaluated according to RECIST 1.1 (Eisenhauer et al, 2009) and Immune-Related Response Criteria (Wolchok et al, 2009). Objective responses were confirmed by at least one sequential tumor assessment.
When indicated, cell lines were engineered to constitutively overexpress soluble forms of GM-CSF, G-CSF, M-CSF, and FKN. Their coding sequences were synthesized (GeneArt Thermo Fisher) and cloned into pDUAL-Puromycin lentivectors (Gato-Canas et al, 2017) under the transcriptional control of the SSFV promoter. Lentivector production, titration, cell transduction, and selection with puromycin were carried out as described elsewhere (Gato-Canas et al, 2017). Cytokine production was confirmed by ELISA. Cell growth and survival were monitored in real time using xCELLigence real-time cell analysis (RTCA ACEA Biosciences) as described before (Gato-Canas et al, 2017). All samples were acquired in a FACS Canto II flow cytometer (Becton Dickinson). Flow cytometry data were exported as FCS3.0 files and analyzed using FlowJo or SPADE software.
The clustering algorithm SPADE (spanning tree progression analysis of density-normalized events, Stanford University) was used to integrate and analyze multiple flow cytometry panels using internal FSC-SSC patterns and common CD11b and CD14 as overlapping markers to reconstruct SPADE trees (Qiu et al, 2011). Diversity indexes were defined as the total number of terminal branches within each hierarchical cluster tree.
Quantification of plasma soluble FKN in murine plasma samples was carried out by ELISA following the manufacturer's instructions (R&D DuoSet DY472 ELISA kit). Quantification of secreted M-CSF, G-CSF, FKN, and GM-CSF from cell cultures was carried out by ELISA (R&D DuoSet ELISA kits; DY415-05 for GM-CSF; DY414-05 for G-CSF; DY416-05 for M-CSF; MCX310 for FKN). Secreted FKN concentration was quantified in supernatants of human cancer cell lines by ELISA (DY365 R&D DuoSet ELISA kit).

Mouse lung cancer models and therapies
Approval for animal studies was obtained from the Animal Ethics Committee of the University of Navarra (Pamplona, Navarra, Spain. Reference 077-19 and 064-22) and from the Government of Navarra. All animals were housed at CIMA's animal house facilities (conventional biosafety 2 housing conditions with environmental enrichment, ES31 2010000132, University of Navarre). When males were used for experiments, they were housed in individual cages if dominant behaviors and barbering were observed. Randomization was used to allocate mice into cages. ARRIVE guidelines were followed for animal experimentation. The indicated 3LL cell lines (1.5 × 10 6 cells/mouse; n = 6 mice/group) were subcutaneously injected in the flanks of 10-week-old C57BL/6 female and male mice (Envigo). 3LL engraftments were allowed to grow for 7 days, then mice were intraperitoneally treated with 100 lg of anti-PD-1 (RPMI-14, BioX-Cell) or saline buffer as control vehicle at days 7, 11, and 15 after tumor inoculation. Tumor size was measured three times per week with a digital caliper until humane endpoint was reached (tumor large diameter superior to 14 mm). Tumor volumes (V) were calculated as V (mm 3 ) = [(short diameter) 2 × (long diameter)]/2. Analyses of circulating immune populations were carried out on day 15 after tumor engraftment. For immunophenotyping of tumor immune infiltrates and spleens, tumors and spleens were harvested and mechanically disaggregated 14 days after engraftment. To assess the abscopal effects of systemic FKN, 3LL or 3LL-FKN cells were subcutaneously injected (1.5 × 10 6 cells/mouse; n = 6 mice/group) on the left flank of mice. Seven days after tumor engraftment, 3LL cells were injected on the right flank. 3LL engraftments were allowed to grow for 7 days, then mice were intraperitoneally treated with 100 lg of anti-PD-1 (RPMI-14, BioXCell) or saline buffer as control vehicle at days 7, 11, and 15 after tumor inoculation. In vivo NK, CD4, and CD8 T-cell depletions were carried out by intraperitoneal administration of 100 lg of anti-mouse CD8a (clone 2.43; BioXCell), CD4 (clone GK1.5; BioXCell) or NK1.1 (clone PK136; BioXCell) antibodies, respectively, at days 6, 10, 14 and 18 after 3LL tumor inoculation.

Immunostaining analysis
Formalin-fixed paraffin-embedded tissues from 10 LUAD patients were obtained from the Hospital San Pedro (Logroño, Spain). All patients provided written informed consent for participation in the study, approved by the Ethics Committee of Clinical Research of La Rioja (CEICLAR, ref. PI-205). Paraffin-embedded lung tissues from LUAD patients were cut into 3 lm sections. Immunostaining analysis to evaluate PanCK and FKN positive areas (%) was performed using Pan-Cytokeratin (Clone AE1/AE3 1:330, Santa Cruz Biotech. Inc., Dallas, TX) and fractalkine (1:230, Abcam, Cambridge, UK) primary antibodies and the Novolik Polymer Detection System (Leica Biosystems, Nussloch, Germany). Fiji open-source image processing software package v1.48r (http://fiji.sc) was used to quantify PanCK and FKN positive areas using the color deconvolution tool. Statistical analyses were performed using SPSS Statistics Software v21 for Windows (IBM, Armonk, NY, USA). According to the sample distribution (Shapiro-Wilk normality test), either a Mann-Whitney or unpaired t-tests were used to compare differences between 2 independent groups. Pearson's correlation coefficients between PanCK and FKN positive areas were calculated using GraphPad Prism v8.3 (GraphPad Software Inc., San Diego, CA, USA). For all analyses, a P-value < 0.05 was considered statistically significant.

Transcriptomic data analyses
To assess the association of FKN mRNA expression with infiltration of different immune populations in human lung adenocarcinomas, we used Tumor Immune Estimation Resource or TIMER2.0 (http:// timer.cistrome.org/; Li et al, 2020). This resource contains transcriptional data from samples included in the TCGA database, including the abundance of immune cells across multiple tumor types. It generates a graphical representation of user-selected gene expression and Spearman correlations with immune populations of interest as identified by CIBERSORT, quanTIseq, xCell, and TIDE algorithms. We restricted our analysis to lung adenocarcinomas and performed systematic correlation studies with CD8 lymphocytes, CD4 lymphocytes, regulatory T lymphocytes, B cells, NK cells, neutrophils, monocytes, macrophages, dendritic cells, and MDSCs. A partial purity-adjusted Spearman's correlation was used. Correlation between FKN, PD-1, and PD-L1 expression was performed by Spearman's correlation tests within GEPIA2 (Tang et al, 2019).

Statistical analysis and study design
No data were considered an outlayer, or removed from the analyses. For mouse experiments, sample sizes were calculated to achieve a minimum power of 0.8 for F-based tests taking into consideration a large effect size (f = 04). Power calculations were carried out with Gpower 3.1.9.7. Blinding was used for data analysis, and double blinding for the validation cohort in FKN plasma quantification and correlation with survival. All variables were tested for normality using the Kolgomorov-Smirnov test. Homogeneity was assessed with Spearman's coefficient of variations. Samples were considered homogeneous with CV < 25%. Homogeneity of variances was tested by the F-Fisher test. Homogeneous groups with comparable variances and fulfilling normality were compared using parametric tests. For multicomparisons, one-way ANOVAs were used followed by pairwise comparisons with a posteriori Tukey's tests. Variables fulfilling these criteria were (1) diversity indexes as calculated below; (2) cell growth rates and cell index data from ACEA Real-Time Cell monitoring (RTCA) data; (3) Tumor volumes as measured at the indicated time points; (4) FKN plasma concentrations in tumor-bearing mice, and (5) secreted FKN concentrations from cultures of cancer cell lines.
The rest of the variables were tested with nonparametric tests. For multicomparisons, one-way Kruskal-Wallis tests were used followed by a posteriori pairwise comparisons with the Dunn's test. Variables that were normally distributed but did not fulfill homogeneity of variances were also tested with nonparametric tests. These included (1) percentages of monocytes and neutrophils quantified by high-dimensional flow cytometry; (2) human serum cytokine concentrations, and (3) percentages of circulating and tumorinfiltrating immune cell types quantified by flow cytometry, in mouse tumor models.
DI in circulating CD11b + cells was calculated as the number of clusters with terminal phenotypes from SPADE 3 hierarchical cluster analyses using high-dimensional flow cytometry data. Correlations between DI and other variables (which included percentages of monocytes, neutrophils, and serum concentrations of relevant factors) were evaluated by Spearman's tests. The predictive capacities of the selected indicated variables were assessed by ROC analyses as described before . PFS and OS in human patients and mouse tumor models were compared with log-rank tests. Tumor experiments were independently replicated between 2 and 7 times (sensitization of PD-1 refractory tumors by FKN). In vitro experiments were replicated at three independent times at least.

Data availability
This study includes no data deposited in external repositories.
Expanded View for this article is available online.