Interferon-stimulated neutrophils as a predictor of immunotherapy response

Summary Despite the remarkable success of anti-cancer immunotherapy, its effectiveness remains confined to a subset of patients—emphasizing the importance of predictive biomarkers in clinical decision-making and further mechanistic understanding of treatment response. Current biomarkers, however, lack the power required to accurately stratify patients. Here, we identify interferon-stimulated, Ly6Ehi neutrophils as a blood-borne biomarker of anti-PD1 response in mice at baseline. Ly6Ehi neutrophils are induced by tumor-intrinsic activation of the STING (stimulator of interferon genes) signaling pathway and possess the ability to directly sensitize otherwise non-responsive tumors to anti-PD1 therapy, in part through IL12b-dependent activation of cytotoxic T cells. By translating our pre-clinical findings to a cohort of patients with non-small cell lung cancer and melanoma (n = 109), and to public data (n = 1440), we demonstrate the ability of Ly6Ehi neutrophils to predict immunotherapy response in humans with high accuracy (average AUC ≈ 0.9). Overall, our study identifies a functionally active biomarker for use in both mice and humans.


In brief
A lack of reliable, highly predictive biomarkers remains a major obstacle in immuno-oncology.In this study, Benguigui et al. discover a promising new biomarker: interferon-stimulated, Ly6E hi neutrophils-whose frequency in the blood of both mice and patients strongly correlates with immunotherapy outcomes across cancer types.

INTRODUCTION
In the era of personalized medicine, predictive biomarkers play a critical role in the clinical decision-making process by identifying optimized treatments tailored to each individual patient, and toward the particular characteristics of each tumor.In cancer, the integration of biomarkers into anti-cancer clinical trials significantly improves response rates. 1 Despite this, robust, predictive biomarkers for newer front-line cancer treatments remain underdeveloped or elusive.Immune checkpoint inhibitors (ICIs) (e.g., anti-PD1 and anti-CTLA4), a revolutionary form of immunotherapy, drastically improve 5-year survival rates in patients with advanced metastatic disease 2 ; yet, only a fraction of patients exhibit durable response. 3Pre-existing biomarkers for ICI outcome, including those used in clinical practice such as PDL1 immunohistochemistry (IHC), tumor mutational burden, or a variety of gene-signatures, are limited in their predictive power (AUC z 0.6-0.75)and require, often inaccessible, tissue biopsies to profile. 4 Notably, these biomarkers are all tumor-intrinsic, yet, immunotherapy response depends on a complex, dynamic interplay between the tumor and the host. 5Newer efforts to define biomarkers have therefore centered on varying aspects of the immune system, such as the rate of tumor-infiltrating T cells 6,7 or levels of myeloid-derived suppressor cells (MDSCs). 8Nevertheless, accurate biomarkers for ICI outcome-applicable to multiple cancer types-remain a crucial but unfulfilled need in clinical oncology.
Biomarkers that integrate tumor-and host-dependent factors may, theoretically, outperform pre-existing markers.Thus, here, we combine single-cell RNA-sequencing (scRNA-seq) and a pre-clinical tumor model encompassing clones with intrinsically low and high immunogenicity, as generated through mutagenesis, to identify cellular states predictive of response that also reflect tumor-intrinsic patterning of host cells (Figure 1A).Specifically, we identify interferon-simulated, Ly6E hi neutrophilsinduced by tumor-intrinsic STING-signaling-as a tumor-infiltrating and blood-borne predictive biomarker for immunotherapy response in both mice and humans (AUC z 0.9, humans) across a multitude of additional models and cancer types, respectively (Figures 1B and 1C).Moreover, we derive a 15-gene Ly6E hi signature that accurately stratifies responders and non-responders in human, bulk RNA-seq data (average AUC > 0.9).Finally, we expand upon the functional characteristics of this neutrophil subtype, revealing its ability to directly sensitize otherwise non-responsive tumors to anti-PD1 in mice-in part through the modulation of cytotoxic CD8 + T cell activity.

RESULTS
An interferon-stimulated subtype of neutrophil marks response to anti-PD1 in 4T1 breast cancer models A prerequisite of biomarker discovery at baseline (i.e., pre-treatment) is the use of a stable and predictable model whose response outcome is known a priori-a requirement notably diffi-cult and time-consuming to fulfill in humans. 9,10In order to search for a biomarker to predict immunotherapy response, we therefore focused our initial efforts on pre-clinical models.Specifically, we generated 4T1 breast carcinoma cell lines, comprising a mutagenized clone (4T1 M ), that is responsive to anti-PD1, derived from a non-responsive parental cell line (4T1 P ), thereby facilitating a biologically relevant comparison between the two related clones (Figure 2A, see STAR Methods and Figure S1A).The exposure of tumor cells to carcinogen results in increased total tumor mutational burden (tTMB), as previously demonstrated, 11,12 mimicking at least one potential tumordependent aspect of immunotherapy response.Using this model, and in-conjunction with mass cytometry (CyTOF) and flow cytometry, we confirmed that mutagenesis resulted in tumors with a higher degree of immunogenicity, characterized by reduced numbers of immunosuppressive cells (e.g., G-and M-MDSCs and PDL1 + cells), increased numbers of anti-tumor immune cells (e.g., activated B and T cells) and elevated Granzyme B levels (Figures 2B-2D and S1B-S1I).4T1 M and 4T1 P therefore constitute suitable models to initially study immunotherapy response.
Preclinical and clinical studies, thus far, have focused on a variety of immune cells as potential biomarkers for immunotherapy-most notably and primarily T cells 6,13 but also MDSCs. 14MDSCs, which include granulocytic and monocytic subtypes, constitute a widely variable, heterogeneous group of cells that are strongly linked to negative outcomes in patients A B C Figure 1.A multi-model approach to identify a clinically relevant biomarker for immunotherapy response A schematic overview of the paper.(A and B) In brief, several mouse strains in combination with multiple cancer cell lines and clones were used to initially screen for and subsequently cross-validate a biomarker for immunotherapy response in mouse.(C) To clinically translate our findings, public data and data from a cohort of patients with non-small cell lung cancer (NSCLC) and skin cutaneous melanoma (SKCM) were used to assess the accuracy and utility of the identified biomarker in humans (see STAR Methods and introduction for additional, step-by-step details).Mouse strains: BALB/c, C57BL/6, and a C57BL/6 x CBA backcross.Cancer cell lines: 4T1 breast cancer, Lewis lung carcinoma (LLC), renal cell carcinoma (RENCA), and EMT6 breast cancer.P = parental cell line, M = mutagenized clone.
with cancer. 15However, contrasting reports suggest different myeloid subsets are associated with anti-tumor activity. 16Given the pleiotropic roles myeloid cells evidently play in tumor biology, and the prior efforts of the community to characterize T cellbased biomarkers, we focused our search on myeloid cells.To this end, we isolated GR1 + cells, representing both monocytic and granulocytic immune cells in mice, from non-responsive and responsive 4T1 tumors and subsequently performed scRNA-seq to map GR1 + subpopulations in detail.UMAP analysis of all GR1 + cells isolated from these tumors revealed two major, coherent populations, representing monocytic and granulocytic phenotypes, as is congruent with previous literature 17 (Figure S2A).Within the monocytic compartment, we observed significant differences in macrophage subsets consistent with previous findings. 18Namely, non-responsive mice were enriched for immunosuppressive, M2 macrophage-like cells while mice responsive to anti-PD1 were enriched for inflammatory, M1 macrophage-like cells (Figure S2B).No further differences were noted, prompting us to search for biomarkers within the granulocytic population.Counterintuitively and in contrast to the known immunosuppressive effects of myeloid cells, including granulocytic MDSCs, 19 we identified a subpopulation of neutrophil whose abundance significantly increases as a function of anti-PD1 response (Figures 3A and S2C).Moreover, this subpopulation exhibited up-regulated expression of 192 different genes (>1.5 log 2 fold-change), including 30 genes with a >2 log 2 fold-change-providing a large pool of candidate biomarkers, associated with this cellular subpopulation, for immunotherapy response within the phenotypically stable 4T1 model (Table S1A).
To narrow down the selection, and aid clinical translation, we reasoned that a successful biomarker must fit the following criteria: (1) the marker is mechanistically understood e.g., it is induced by a, or a series of, signaling pathway(s) identifiable in the data or active in the tumor microenvironment, (2) the marker is present on a host-cell (e.g., neutrophil), but is induced by tumor-intrinsic activity (see: introduction for rationale), (3) expression of the marker is predominately found in a metastable or highly differentiated cell state, as opposed to a transient state that may be difficult to consistently detect in vivo and finally, for practicality (4) the marker gene encodes a cell surface marker, permitting cost-effective analysis by simple flow cytometry.
Following this logic, we initially performed trajectory analysis for all neutrophilic cells and deduced the trajectory's directionality by calculating vectors of RNA velocity (Figures 3B and  S2D) (see STAR Methods for details).Such an approach not only determines the endpoint(s) of cellular differentiation, but also provides mechanistic insight by revealing which biological processes and transcription factors are periodically activated as differentiation proceeds.Interestingly, we observed a branched, dual-lineage trajectory that diverges at an early stage (i.e., at the progenitor level) yet ultimately converges upon a single cell state (Figure 3B).Importantly, and consistent with the RNA-velocity-inferred direction, known progenitor genes are expressed at early pseudotime values (Figures S2E-S2I) 20 -suggesting this trajectory captures pathways of legitimate cellular development.We hypothesized that the convergence of two lineages toward the same cell state may reflect a common, underlying differentiation program.We therefore used a trajectory alignment algorithm 21 to match ''homologous'' segments of each lineage to one another and identified multiple, shared modules of genes-present in both branches-whose expression alters as a function of pseudotime (Figure 3C).By calculating the density of cells across the resulting aligned trajectory on a persample basis (Figure 3C -top) and characterizing each shared module (Figure 3C -bottom), we observed that neutrophils from non-responsive mice typically fail to progress beyond a progenitor-like, apoptotic state.In contrast, neutrophils from responsive mice differentiate further to a terminal state marked by response to interferon a/g (IFNa/g) and NFkB/TNFa signaling, suggesting exposure to IFN is a major driving force behind this differential progression.Consistent with this, previous studies have shown a link between IFNa/g levels and response to immunotherapy. 22When examining 4T1 tumors at the protein level, we observed a similar correlation between IFNa/IFNg/TNFa levels and response, validating our scRNA-seq results (Figure 3D, see STAR Methods).Therefore, in order to select a candidate biomarker that fulfills our criteria (see previous text), we screened all 192 differentially expressed genes (Table S1A) for IFN-stimulated genes (ISGs) residing at the cell surface.Ly6E, a known ISG 23 and the only cell-surface marker to fulfill our criteria, was found to have a high expression-weighted pseudo-time value and constitute a prime biomarker candidate by which to assay this subtype of neutrophil (Figure 3E).Specifically, a high frequency of Ly6E hi neutrophils in the tumor significantly correlates with immunotherapy response in the 4T1 model (Figure 3F and see S3A; gating strategy).Despite the discovery of Ly6E hi neutrophils in tumor samples, we hypothesized that these cells may also additionally form in or cycle back into the blood.Indeed, Ly6E hi neutrophils similarly mark response when assayed in the blood of mice bearing 4T1 tumors (Figure 3G), and importantly, the ability of Ly6E hi neutrophils to distinguish responsive and non-responsive mice is established at early stages of tumor growth (50 mm 3 ) -collectively suggesting Ly6E hi neutrophils may serve as a predictive, bloodborne biomarker of anti-PD1 response in this model.

Ly6E hi neutrophils overcome resistance to anti-PD1 therapy
Biomarkers can be surrogate-that is, passive bystanders generated as a byproduct of the main biological mechanism(s) underpinning immunotherapy response (e.g., the presence of IFNa/g in the microenvironment of responding tumors) -or they may be functionally involved in response itself.We reasoned that functionally active biomarkers may possess a wider degree of applicability, beyond a single preclinical model or cancer-type.Thus, to distinguish between these two possibilities, we artificially generated Ly6E hi neutrophils in vitro by exposing GR1 + cells to a cocktail of IFNa/g (Figure 4A), as informed by scRNA-seq analysis (Figures 3C and 3D).To ensure that the resulting cells resemble the Ly6E hi phenotype observed in our scRNA-seq data, we analyzed the induction of Ly6E at the protein level and the mRNA expression levels of selected differentially expressed, secreted factors by RT-qPCR, based on our scRNA-seq data.Firstly, we observed a strong induction of Ly6E on the surface of neutrophils following IFN treatment (Figure 4B).Secondly, we observed a striking correlation between the log 2 fold-changes of the RT-qPCR (treated vs. untreated) and the scRNA-seq (response vs. non-response) (Figure 4C), collectively suggesting these cells are analogous.Subsequently, we sought to test, in vivo, the effect of these generated cells on tumors resistant to anti-PD1.We, therefore, administered Ly6E hi neutrophils, by adoptive transfer (see Figure S4A, for treatment protocol), to mice bearing non-responsive 4T1 tumors, and observed a significant reduction in tumor growth following anti-PD1 therapy but no efficacy of these cells as a monotherapy (Figures 4D and S4B).Consistent with these results, we examined the levels of various immune cells in a separate experiment and found that the frequency of bloodborne and tumor-infiltrating activated cytotoxic CD8 + T cells was significantly higher in mice treated with both anti-PD1 and Ly6E hi neutrophils relative to those treated with either monotherapy alone (Figures S4D-S4F).This trend was further recaptured when measuring intra-tumoral granzyme B levels (Figure S4G).Of note, we identified fluorescently labeled Ly6E hi neutrophils in treated tumors (Figure S4H), further indicating that Ly6E hi neutrophils successfully infiltrate and play a role in the responding tumor microenvironment.
Given the IFN-stimulated phenotype of Ly6E hi neutrophils, we next evaluated whether IFN-g and IFN-a (IFNg/a) alone can sensitize resistant tumors to the same extent.While mice treated with a combination of IFNg/a and anti-PD1 display a marginal reduction in tumor growth, it is not significant (Figures S4C and  S4I) and no change in the levels of activated CD8 + T cells was observed in either the blood or the tumor (Figures S4J and  S4K).Nevertheless, despite the lack of response, the levels of Ly6E hi neutrophils in the tumor were significantly elevated by IFNg/a treatment, reinforcing the fact that IFN induces Ly6E hi neutrophils both in vivo and in vitro (Figures S4L and S4M and  4B, respectively).This apparent paradox suggests IFN-g and/ or IFN-a, when given systemically, mediate additional pleiotropic effects beyond the generation of Ly6E hi neutrophils-effects which inhibit immunotherapy response and overwrite the ability of Ly6E hi neutrophils to overcome non-responsiveness.Furthermore, our results suggest that Ly6E hi neutrophils themselves represent an isolated, distinctly anti-tumorigenic effect of IFN.Such findings are consistent with but potentially build upon the apparent ineffectiveness of systemic IFN-treatment in augmenting ICI therapy in humans. 24e STING signaling pathway accounts for IFN-induced Ly6E hi neutrophils which in-turn directly support antitumor immunity Cytosolic double-stranded DNA (dsDNA), generated under conditions of cellular stress, hypoxia or chromosomal instability, is known to induce tumor-intrinsic STING pathway activity and the subsequent secretion of IFNs (e.g., IFNa) from cancer cells. 25,26Given the IFN-stimulated phenotype of Ly6E hi neutrophils, the use of a model with high mutational burden, and our desire to identify a biomarker patterned by tumor-intrinsic properties, we asked whether STING signaling is responsible for the generation of these cells in the tumor microenvironment.To this end, we quantified the levels of STING-pathway associated factors in non-responsive 4T1 P and responsive 4T1 M clones.We observed significantly higher levels of cytosolic dsDNA and a significant up-regulation of STING and its downstream signaling components (IRF3, NF-kB, and native ISG15 [15 kDa]) in 4T1 M relative to 4T1 P (Figures 5A, 5B, and S5A).Consistent with this, 4T1 M cells secrete higher levels of IL-6, up-regulate cell-surface MHCI, and down-regulate PDL1-all known readouts of STING activity (Figures S5B-S5D). 27,28Importantly, these trends are reversed with use of the STING inhibitor, H151 (Figures S5B-S5D).Interestingly, 4T1 M tumors show a reduced level of (D) Averaged tumor growth profiles for mice bearing parental, non-responsive 4T1 breast tumors treated with either a monotherapy (control IgG or aPD1) or a combined therapy, with GR1 + or Ly6E (hi) neutrophils, as specified (n = 6 mice/group).A time-course of the adoptive transfer is depicted in (Figure S4A).Raw data are available in (Figure S4B).Treatment was initiated at a tumor size of 50 mm 3 (arrow).Significance was assessed by means of two-sample KS-test (***, p < 0.0001).
ISGylated proteins (Figure 5B), suggesting ISGylation machinery is suppressed or abnormal in these tumors despite robust STING and ISG15 induction.Nevertheless, and critically, conditioned media of 4T1 M cells strongly induces the Ly6E hi neutrophil phenotype in vitro, in a STING-dependent manner, and this induction is reversed when blocking IFN receptors (IFNRa/g) (Figure 5C).In contrast, no such dynamics are seen with media derived from 4T1 P .Consistent with these results, IFNRa/g are expressed at a high level on GR1 + cells (Figure S6A) -confirming their ability to respond to IFN.Interestingly, receptor expression is maintained on Ly6E hi cells (Figure S6A) and we further show that the higher levels of IFNa/g previously observed within 4T1 M tumors (see: Figure 3D) are entirely STING-dependent (Figures S6B and S6C).In order to expand upon these observations, we assessed the effects of IFNR-a/g inhibition in vivo.Mice bearing responsive 4T1 M tumors-treated with aIFNR-a/gwere no longer able to mount an effective response to anti-PD1 (Figures S6D and S6E) and this lack of response was marked by lower levels of Ly6E hi neutrophils (Figure S6F).Importantly, adoptive transfer of Ly6E hi neutrophils was able to rescue immunotherapy response despite IFNR-a/g blockade (Figures S6D-S6F).Taken together, our results strongly suggest that STING activation-intrinsic to 4T1 M responsive cancer cells-accounts for the induction of Ly6E hi neutrophils, as mediated by IFN, and in-turn the ability of these cells to predict but also induce immunotherapy response.
Given the ability of Ly6E hi neutrophils to mediate immunotherapy response in 4T1-bearing mice (see Figures 4D, S4, and S6D-S6F), we sought to uncover the Ly6E hi -dependent molecular mechanisms responsible for this.Since adoptive transfer of Ly6E hi neutrophils into 4T1 P -bearing mice induced cytotoxic CD8 + T cell activity (see Figures S4E and S4F), we explored whether Ly6E hi neutrophils directly mediate this activation and whether this activity is dependent on Ly6E itself, or through secreted factors induced post-IFN-stimulation.To address these two questions, we first co-cultured Ly6E hi neutrophils or unstimulated GR1 + cells with CD8 + T cells.While Ly6E hi neutrophils promote the proliferation and activation of cytotoxic CD8 + T cells, GR1 + cells substantially inhibit such activities (Figures S7A-S7D).Consistent with this, Ly6E hi neutrophils significantly promote T cell mediated tumor cell killing in vitro, relative to control cultures (Figure S7E).We subsequently  (hi) phenotype and in-turn supports activation of effector T cells (A) Density plots of dsDNA levels in cultured 4T1 P and 4T1 M cell-lines, as determined by a-dsDNA staining and flow cytometry.dsDNA levels were quantified relative to an unstained, IgG2a isotype control (CTRL) (n = 5 biological repeats/group).(B) Densitometry quantification of western blots (see Figure S5A) for STING-pathway related proteins in 4T1 P and 4T1 M tumor lysates (n = 3-4 biological repeats/ group).Each protein was normalized relative to an actin control.(C) Isolated GR1 + cells were cultured in vitro with conditioned media generated from 4T1 P (P) or 4T1 M (M) tumors in the presence or absence of the STINGinhibitor H151 or aIFNR-a/g, and the frequencies of Ly6E (hi) neutrophils were determined by flow cytometry (n = 6 biological repeats/group).CTRL = GR1 + cells only.(D and E) Conditioned media was generated from GR1 + cells or IFNag-induced Ly6E (hi) neutrophils, and subsequently assayed on a cytokine array (n= 3 mice pooled/group).Hyper-geometric, over-representation tests and the Gene Ontology (GO) database were used to determine enriched pathways for Ly6E (hi) neutrophils (D); and GR1 + cells (E).Only differentially expressed proteins with a log 2 FC > 0.35 were included and only significant pathways (FDR < 0.01) are shown.(F) Isolated CD8 + T cells were cultured in vitro with a-IL-12b or a-IL23a neutralizing antibodies, with or without conditioned media from IFNa/g-induced Ly6E (hi) neutrophils (L), and the levels of activated CD25 + CD8 + T cells were determined by flow cytometry (n = 5 mice/group).CTRL = CD8 + T cells only.In (B, C, and F), significance was assessed by means of a one-way ANOVA and Tukey's post-hoc HSD test (NS, p > 0.01; *, p < 0.01; **, p < 0.001; ***, p < 0.0001).(G) Schematic of the proposed mechanism.Tumor-intrinsic STING activity, as induced by cytosolic dsDNA as a result of hypoxia, genomic instability and/or cell stress, transcriptionally activates an IFN response.Tumor-secreted IFNa, for example, subsequently binds to Ifnar-expressing Neutrophils in the TME, inducing the Ly6E (hi) phenotype and in-turn activation and proliferation of CD8 + T cells through IL-12b.Collectively, this supports immunotherapy response and anti-tumor activity.It is important to note that this mechanism is STING-specific, but that Type II IFNs (e.g., IFNg)-derived from other sources or mechanisms-are also able to elicit equivalent effects, as shown in our work.knocked Ly6E down in the bone-marrow of mice (Figure S8A, see STAR Methods), and repeated these experiments.Importantly, no change in response to anti-PD1 was observed in vivo (Figure S8B) and all positive effects of IFN-induced Ly6E hi neutrophils on T cells and T cell-mediated tumor killing were retained regardless of Ly6E status (Figures S8C-S8G)-suggesting that Ly6E has no functional role in the mechanism of response, but rather serves solely, in our study, as means to assay this subpopulation of neutrophil.
Therefore, we next compared the secretome of Ly6E hi neutrophils, relative to all other GR1 + cells, in order to determine the potential mechanism(s) underlying the induction of T cell activation.Based on pathway analysis of differentially expressed proteins, Ly6E hi neutrophils support the activation and positive regulation of CD8 + T cells-through cytokines such as IL-12b, IL-1b, IL-6, and IL-10-while unstimulated GR1 + cells support an immunosuppressive tumor microenvironment through the recruitment of additional immunosuppressive myeloid cells (Figures 5D and  5E).Consistent with this and relative to all other GR1 + neutrophil subsets, Ly6E hi neutrophils are significantly down-regulated at the mRNA level for secreted, immunosuppressive factors such as S100A8, S100A9, and CCL6, [29][30][31] while up-regulated for pro-inflammatory factors such as TNF-a, IL23a, IL-12b, and IL-1a (Table S1B).To validate this further, we co-cultured Ly6E hi neutrophils with CD8 + T cells in the presence or absence of neutralizing antibodies targeting IL-12b and IL23a, both of which were up-regulated in Ly6E hi neutrophils compared to GR1 + cells.Interestingly, and in line with a recent publication, 32 we found that IL-12b but not IL23a induced the activity of CD8 + T cells (Figure 5F).These results therefore suggest that Ly6E hi neutrophils may augment cytotoxic CD8 + T cell activity, through secretion of IL-12b.
To establish a clear order of events, we further tested if the levels of Ly6E hi neutrophils are reciprocally dependent upon T cell activity by utilizing SCID mice lacking an adaptive immune system and found this not to be the case.Instead, we observed that the ability of blood-borne Ly6E hi neutrophils to distinguish responding and non-responding 4T1 tumors in immunocompetent mice remains intact within SCID mice (Figures S7F-S7G).Collectively, our results suggest that Ly6E hi neutrophils not only serve as a predictive biomarker for immunotherapy response in mice bearing 4T1 tumors but also: (1) are functionally involved in the mechanism of response; (2) operate upstream of T cells; (3) can be induced by an entity other than the adaptive immune system or host (e.g., tumor-intrinsic STING signaling, via IFNa or via IFNg through yet-to-be characterized mechanisms); and (4) contribute to anti-tumor immunity by directly activating cytotoxic CD8 + T cells via IL-12b (Figure 5G).

Cross-validation of Ly6E hi neutrophils as a biomarker for response in various preclinical tumor models
The majority of translational studies, to their detriment, continue to employ simplistic approaches involving only single mouse strains or cancer types.Therefore, the identification of Ly6E hi neutrophils in one preclinical model prompted us to validate them as blood borne biomarkers in a diverse array of additional models capturing both tumor-and host-dependent variation, as both aspects play a key role in drug efficacy. 33We therefore employed tumor models based on cell lines, encompassing: (1)   clones with or without mutagenesis in two strains of mice (RENCA renal cell carcinoma, and Lewis Lung carcinoma (LLC)), as in our previous 4T1 approach; (2) cell lines that spontaneously respond to immunotherapy (EMT6 breast cancer); and (3) mixed background mice, containing variable baseline immune states, implanted with LLC tumors (Figure S9).In all cases, we observed that the frequency of Ly6E hi neutrophils predicts response to anti-PD1 prior to treatment-to a significant degree and in a model agnostic manner (Figures S9A-S9D).Collectively, our data suggest that IFN-stimulated, Ly6E hi neutrophils are a potential ''pan-mechanistic'' marker for therapy outcome in mouse, whether the response is driven by tumor-, host-dependency or strain-specific differences and that IFN-secretion into the tumor microenvironment may therefore be a common step in the mechanism of response.

Ly6E hi neutrophils predict immunotherapy response in human
Species-specific differences typically hinder the ability to translate findings, such as a biomarker, from mouse to human. 34To help overcome this, we employed a set of pre-clinical models (see Figure 1) to identify Ly6E hi neutrophils as a potential ''panmechanistic'' biomarker in mouse with a greater degree of confidence that the marker may be conserved in humans.Nevertheless, it remained unclear whether Ly6E would be a marker of the same, IFN-stimulated cell state in human.To address this limitation and further bridge the cross-species gap, we first built a functional signature based upon the biological processes that mark response in mouse (see Figure 3C), namely IFNa/g response and NF-kB/TNFa signaling.Subsequently, we analyzed public, scRNA-seq data from the blood of 8 patients with non-small cell lung cancer (NSCLC) obtained prior to treatment and applied the mouse-derived signature to all 6607 identifiable human neutrophils 35 (Figure 6A).We observed a cluster of cells highly enriched for our signature, marked by genes induced by IFN (Figure 6B).Notably, this cluster displayed a high level of Ly6E expression, suggesting Ly6E is an appropriate marker by which to assay these cells in human (Figure 6C).Subsequently, to test whether Ly6E hi neutrophils predict response to immunotherapy in humans, we obtained pre-treatment peripheral blood mononuclear cells (PBMCs) from a limited, independent mixed cohort of patients with advanced metastatic NSCLC (n = 50) and malignant melanoma (n = 59) predominately treated with ICI-based therapy and quantified the levels of Ly6E hi neutrophils.For the sake of clarity, it is important to note that low-density neutrophils found in chronic disease states are present in PBMC fractions. 36As in mouse, high levels of Ly6E hi neutrophils were strongly correlated with response and positive, clinical outcome (Figures 6D, 6E, and S3B for gating strategy).Remarkably, Ly6E hi neutrophils stratify between non-responder and responder groups (AUC z 0.9) in both cancer types, whereas pre-existing biomarkers, namely, PDL1 IHC and total neutrophil count measured in the same group of patients with NSCLC, underperformed (AUC z 0.6 and 0.75, respectively) (Figure 6F).8][39][40] We observed that, in all but one dataset, neutrophils in responders relative to non-responders are highly enriched for a Ly6E hineutrophil derived, IFN-stimulated signature (Neut IFN -15, genes: IFIT1, MX1, HERC5, IFI6, ISG15, IFIT3, RSAD2, GBP1, IFIT2, XAF1, PARP9, UBE2L6, IRF7, PARP14, and APOL6)-including in urothelial bladder carcinoma, glioblastoma, NSCLC, renal cell carcinoma, melanoma, and stomach adenocarcinoma datasets, at the pre-treatment stage (Figure 7A, top).Conversely, the previously published IFN-g 6 signature, 41 which has no overlap in genes with Neut IFN -15, underperforms on these datasets (average AUC 0.62 vs. AUC 0.88, respectively) (Figure 7A, bottom and S10 for raw data).Moreover, in one dataset where pre-existing biomarkers (PDL-1 IHC, tTMB, and STK11/KEAP1 status) were measured, Neut IFN -15 predicted outcome with significantly higher accuracy (Figure 7B).Of note, in 203 samples taken post ICI therapy, the ability of Ly6E hi neutrophils to stratify between responders and non-responders is weakened (Figure 7A).These results, taken together, suggest that the levels of Ly6E hi neutrophils-whether measured in the blood or the tumor-serve as a predictive biomarker for immunotherapy response in both mice and humans across a multitude of different tumor types.

DISCUSSION
The efficacy of immunotherapy is governed by complex mechanisms dependent upon the interactions between host (e.g., the immune system) and malignant cells.By narrowing our search to host cell biomarkers not only predictive of response, but for which we also have a tumor-dependent mechanism, we discovered interferon-stimulated, Ly6E hi neutrophils as a blood borne, predictive biomarker with potentially high predictive power in both mice and humans (AUC z 0.9 in humans).Importantly, Ly6E hi neutrophils appear to remain predictive in a diverse array of cancer types.Our approach may therefore have revealed a ''pan cancer'' biomarker that can be assayed in a cost-effective manner by liquid biopsy, however further clinical validations are required.
Neutrophilic GR1 + cells or MDSCs are ordinarily and strongly pro-tumorigenic, acting to suppress anti-tumor immunity. 19et, Ly6E hi neutrophils exhibit anti-tumorigenic properties, induce immunotherapy response in mice and enhance immunity against tumors, further highlighting the plasticity and importance of myeloid cell state in the tumor microenvironment. 42In highly mutated, murine 4T1 tumors, induction of tumor-intrinsic STING activity is responsible for IFN-secretion and the generation of Ly6E hi neutrophils in the tumor microenvironment, mediated specifically by IFNa/g.Due to technical limitations, it remains unclear if STING activity is the driving force behind the Ly6E hi phenotype in all cases and cancers.Nevertheless, given the broad predictive power of our biomarker, and the fact that the Ly6E hi phenotype is induced by IFN, localized IFN activity in the tumor microenvironment may prove to be a crucial and common step in the mechanism of immunotherapy response regardless of the exact source of IFN or the exact IFN involved in a given case (IFNa or IFNg). 43Consistent with this, studies demonstrate that IFNg or its related pathways serve as predictors of immunotherapy response 41,44,45 -albeit with a lower predictive power than Ly6E hi neutrophils.Moreover, up-and downregulation of MHCI and PDL1 respectively, due in part to IFN stimulation, can also stratify between responsive and nonresponsive tumors [46][47][48]   evaluation as a combinatorial therapy with ICIs. 49However, IFN has also been shown to counter-intuitively exhibit pro-tumorigenic effects and promote resistance to anti-PD1 therapy. 50ndeed, our study demonstrates that systemic IFNg/a treatment in combination with anti-PD1 resulted in a non-significant reduction in tumor growth and no change in cytotoxic CD8 + T cell activation.It is plausible that IFN acts, in part, via Ly6E hi neutrophils to augment immunotherapy outcome but that additional, negative effects of IFN-or chronic, systemic IFN treatment 51 -''tip the scales'' and counterbalance this.Regardless, our study further provides mechanistic insights into the complex role IFNs play in cancer biology and by ''zooming in'' and identifying a specific anti-tumorigenic effect of IFN, i.e., generation of Ly6E hi neutrophils, it may be possible to develop therapeutic approaches that lack the negative aspects of IFN-as our adoptive transfer results suggest.
We show that Ly6E hi neutrophils not only act as a biomarker but also function as an immunomodulator-sensitizing otherwise resistant tumors to anti-PD1 therapy, in part, by creating an environment permissive to CD8 + T cell activation through secretion of known activating factors such as IL-12b. 52Critically, Ly6E hi neutrophils appear to act upstream of the central anti-tumor T cell response, potentially ''priming'' tumors to respond.Consistent with this, treatment-elicited neutrophils acquire an IFN-gene signature following treatment with anti-PD1, and are essential to the response process in humans. 32Our work thus expands upon this study, by demonstrating the presence of predictive, IFN-stimulated neutrophils prior to treatment.
It is important to note that, while our preclinical work focused entirely on anti-PD1, our clinical cohort is mixed-comprising patients with metastatic NSCLC and melanoma (n = 109) treated with either ICI monotherapy (anti-PD1, anti-CLTA4, or anti-PDL1) or ICIs in combination with other treatment modalities (e.g., chemotherapy).Therefore, while Ly6E hi neutrophils remained highly predictive in all cases, further prospective clinical studies, including those related to neoadjuvant settings, 53 should be designed to validate the robustness of these results within each treatment arm and the ability of Ly6E hi neutrophils to differentiate between non-responders and responders in a variety of treatment scenarios and tumor backgrounds e.g., specific mutations.Nevertheless, our main conclusions were further supported by the analysis of publicly available bulk RNA-seq datasets taken pre-treatment from 1,237 patients with cancer who underwent ICI therapy.In all samples analyzed, except one, the enrichment of a Ly6E hi neutrophil-derived gene signature (Neut IFN -15) correlated strongly (average AUC > 0.9) with patients who responded to immunotherapy-suggesting Ly6E hi neutrophils are widely applicable as a biomarker.Furthermore, additional limitations exist in this study.First, our preclinical models were based on cancer cell lines and did not include genetically engineered mouse models or patient-derived xenografts.While this is a preclinical limitation, clinically, we demonstrate the validity of Ly6E hi neutrophils as a potential biomarker regardless of these limitations.Moreover, and consistently with previous publications, 11 we demonstrate that high mutational burden contributes to ICIresponsive tumors, as mutagenesis induces a high degree of immunogenicity.It is worth mentioning that high mutational burden does not necessarily correlate with ICI outcome 54 ; however, for our preclinical approach, the use of this artificial model aided prospective prediction of ICI therapy outcome.Second, the detection of Ly6E hi neutrophils in clinical samples lacks a clear, demarcated population of cells when analyzed by flow cytometry.Rather, the Ly6E hi phenotype is defined in relative terms compared to other samples and the expression of Ly6E itself occupies a continuum as opposed to discrete positive or negative states.Thus, future clinical studies should focus on refining Ly6E hi neutrophil identification by, for example, utilizing internal markers expressed by these cells, CITE-seq or cell surface marker screening in order to adequately stratify between responders and non-responders using predefined, absolute thresholds.Alternatively, machine-learning classifiers may be able to determine an appropriate flow-based threshold-given a sufficiently large test cohort.Third, owing to their short halflife and fragility in peripheral blood, 55 neutrophils are typically overlooked or discarded as a source of potential biomarkers or biology, and the methodologies used to collect PBMCs in human often exclude neutrophils due to their high density.However, in diseased states such as cancer, a subset of neutrophils adopts a low-density state, 36 making them clinically accessible and warranting further studies.Our study, itself, further demonstrates that neutrophils can be reliably detected in frozen PBMCs obtained from human samples.Overall, while there are a number of limitations to our study which deserve further exploration and clinical validation, we nevertheless provide strong evidence that IFN-stimulated, Ly6E hi neutrophils predict ICI outcome and are functionally involved in the generation of response.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Clinical translation
To translate the use of this cellular biomarker into humans, a functionally equivalent cell state can be identified through public data mining (Figure 1C).Functional equivalence is superior to the use of direct orthologues (e.g.''Gene-A'' in both mice and humans) as they may not necessarily mark the same cell state in a different species.Here, we analyzed published scRNA-seq data from the blood of patients with non-small cell lung carcinoma (NSCLC) to identify cells undergoing similar biological processes to the cells identified in mice.Subsequently, the cellular biomarker was validated in a separate retrospective cohort of patients with NSCLC and melanoma treated with ICI-based therapy, as well as in publicly available datasets of additional tumor types, as outlined below.

Mouse tumor models
The use of animals and experimental protocols were approved by the Animal Care and Use Committee of the Technion.Female BALB/c, C57BL/6, and combined immunodeficient (SCID) mice (8 weeks of age) were purchased from Envigo, Israel.Mixed background mice were created by backcrossing female C57BL/6 and CBA female mice with pure C57BL/6 male mice for 5 generations.All mice were maintained under specific pathogen-free conditions in the animal facility.4T1 P , 4T1 M and EMT6 (5x10 5 /50mL in serum free medium) were orthotopically injected into the mammary fat pad of 8-10-week-old female BALB/c mice or SCID mice.RENCA P , RENCA M , LLC P and LLC M (5x10 5 /50mL in serum free medium) were subcutaneously injected into the flanks of 8-10-week-old female BALB/c and C57BL/6 mice, respectively.Mice were randomly grouped before therapy.Typically, the number of mice per group was set to 5, to reach statistical power, unless indicated otherwise in the text.In all experiments, when tumors reached 50 mm 3 mice were treated with anti-mouse anti-PD-1 (clone RMP1-14, BioXCell Cat# BE0146 or ichorbio Cat# ICH1132) antibody.The antibody was given twice a week in a dose of 100mg/mouse for up to 2-week period.The control groups were injected with IgG isotype control (BioXCell Cat# BE0089 or ichorbio Cat# ICH2244).In some experiments mice were treated with the combination of IFNa (BioLegend, Cat# 752802) and IFNg (Peprotech, Israel, Cat# 315-05) in a total dose of 2mg/mouse for 10 days or with antibodies blocking IFN-Ra (Clone MAR1-5A3,BioXCell, Cat# BE0241) and IFN-Rg (Clone GR-20, BioXCell, Cat# BE0029), at a dose of 50mg/mouse twice a week, as previously described. 76ood collection from patients with cancer Blood collection from human subjects was approved by ethic committees at Sheba medical center, Tel Hashomer (Ramat Gan, Israel) (IRB: 0226-13), Yale University School of Medicine (New Haven, CT, USA) (IRB: 0609001869) as well as Rambam Heath Care campus (Haifa, Israel) and Hadassah Medical Center (Jerusalem, Israel) through the national bio-bank (Midgam, Israel) (IRB: RMB-0631-17).All patients signed informed consent.Blood was drawn at baseline, before immunotherapy, from patients with non-small cell lung cancer (n=50) and melanoma (n=59).Patient characteristics are indicated in Table S2.Peripheral blood mononuclear cells (PBMCs) were isolated from ficoll tubes and stored in freezing medium at -80 C, until further analyzed.PBMCs were then thawed and analyzed by flow cytometry using a mixture of antibodies indicated above.Patients were stratified to responders and non-responders based on RECIST criteria at 3 and/or 6 months where partial/complete response and stable disease patients were considered responders, and progressive disease patients were considered non-responders.The correlations of Ly6E hi neutrophil levels with response rates were then calculated.

Cell line mutagenesis
Parental cell lines that are resistant to ICI therapy were cultured with 1-methyl-3-nitro-1-nitrosoguanidine (MNNG, Apollo Scientific, Cat# OR301388) for 2 hours.After the cells were washed with PBS and growth medium was added, cells were allowed to recover over 5 days and multiclonal mutational cells were created.Mutagenized cells were validated in-vivo for their response to ICI therapy.Using this procedure, we have generated responsive clones to ICI therapy including 4T1 parental cells (4T1 P ) and its mutagenized clone (4T1 M ), LLC parental cell line (LLC P ) and its mutagenized clone (LLC M ), and RENCA parental cell line (RENCA P ) and its mutagenized clone (RENCA M ).4T1 tumors were also evaluated for immunogenicity as described in Figure 2.
Cytokine array and biological pathway enrichment GR1 + cells or IFN-induced Ly6E hi neutrophils were cultured in serum-free medium for 24 hours to generate conditioned medium (10 6 cells/ml).The conditioned medium was applied to a proteomic profiler mouse XL cytokine array (R&D, MN, Cat# ARY028), in accordance with the manufacturer's instructions.Relative levels of the different proteins were calculated based on densitometry and compared between GR1 + and Ly6E hi neutrophils to obtain log 2 (fold changes).Over-representation tests were performed using clus-terProfiler [v4.0.0] 72 and gene-lists from the Gene Ontology (GO) database to characterize all differentially expressed proteins with an absolute log 2 FC > 0.35.Only significantly enriched (FDR < 0.01, Bonferroni correction method) pathways were retained.
Time of flight mass cytometry (CyTOF) 4T1 P and 4T1 M (5x10 5 /50mL in serum free medium) were orthotopically injected into the mammary fat pad of 8-10-week-old female BALB/c mice (n=5 mice/group).When tumors reached 50 mm 3 , mice were treated with anti-mouse anti-PD-1 or IgG control for 2 weeks, as described above.At endpoint, mice were sacrificed and tumors were prepared as single cell suspensions.The cells were acquired by CyTOF as previously described. 79Briefly, an equal number of tumor cells were pooled per group (5 mice/group) and 3x10 6 cells were collected from each pool for CyTOF acquisition.The cells were washed with cell staining media (PBS without Ca 2+ /Mg 2+ , 2% bovine serum albumin, and 0.09% Azide) and immunostained with a mix of metal tagged antibodies (See: key resources table).Following acquisition, the cells were gated and analyzed, as described below.
Adoptive transfer of Ly6E hi neutrophils experiments GR1 + cells were isolated (positive isolation, EasySep Mouse PE, Stemcell Technologies, Cat# 17666) from the spleens of 4T1 tumor bearing mice and cultured overnight with 5% medium containing IFNa and IFNg (10 ng/ml each, BioLegend Cat# 752802 and Peprotech Cat# 315-05).Subsequently, cells were collected, centrifuged and washed twice with PBS.Ly6E hi neutrophils were analyzed by flow cytometry and by RT-qPCR as described below.The experimental procedure was carried out as described in the schematic illustration (Figures S4A and S6D) Specifically, Ly6E hi neutrophils (1x10 6 cells per mouse) were intravenously injected into mice bearing 50 mm 3 4T1 P or 4T1 M tumors (n=6-7 mice/group), and 4 hours later, mice were treated with anti-PD-1 or IgG control.Ly6E hi neutrophils were adoptively transferred for a total of 3 times.In some experiments, at the time of the last injection, the cells were first labelled with Live Cell Labeling -Red Fluorescence -(Cytopainter, abcam, Cat# ab187965), in accordance with the manufacture's protocol.Tumor volume was measured twice a week.When tumors reached endpoint, the experiment was terminated.

Real-Time quantitative PCR (RT-qPCR)
RNA was extracted from the in-vitro Ly6E hi induced cells using Total RNA Purification Kit (Norgen, Ontario, Canada, Cat# 35300).cDNA was synthesized using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, California, USA, Cat# 4374966).RT-PCR reaction was performed using SYBR Green Master Mix and ran in the CFX Connect Real-Time PCR Detection System (Bio-Rad, Cat# 4385612).Analysis was performed using the DDCt method.Five biological repeats were carried out.Primers are listed in Table S4.
Tumor cell killing assay 4T1 cells were seeded in a 24-well plate (20,000 cells/well) along with CD8 + T cells and IFN-induced Ly6E hi neutrophils (200,000 cells, 1:1 ratio) for 24 hours.Subsequently, PI (500 nM) was added to cultures in order to identify dead cells.T cell killing effect was analyzed by flow cytometry.The results are representative of five biological replicates.
To evaluate the induction of Ly6E hi neutrophils from GR1 + cells in the presence of tumor cells, 4T1 P and 4T1 M cells (2x10 6 /ml) were cultured in the presence or absence of H151.After 24 hours the cells were washed and cultured in serum-free medium to generate conditioned medium (CM).Subsequently, CM was cultured with GR1 + isolated cells (positive isolation, EasySep Mouse PE, Stemcell Technologies) from the spleens of 4T1 tumor bearing mice under a humidified 5% CO2 atmosphere at 37 C overnight, and analyzed for the expression of Ly6E by flow cytometry.All experiments were performed with at least 3 biological repeats.

Ly6E knock-down in bone marrow cells
Constitutive Cas9-expressing mice (JAX mice, #026179), were used to create CRISPR-Cas9 knockout of immune cells using CHIME (CHimeric IMmune Editing), in line with a previous publication, 80 with some modifications.Specifically, Ly6E gRNA (forward: 5'CACCG AGCAAGCTAAGCCTGCGCAC 3'; reverse: 5'AAAC GTGCGCAGGCTTAGCTTGCT C 3') was cloned into the pXPR_053 vector plasmid containing GFP. Next, lentiviral particles were generated by co-transfecting HEK-293FT cells with packaging (psPAX2, Addgene Plasmid #12259) and envelope (pMD2.G, Addgene Plasmid #12259) plasmids together with pXPR_053 (control vector, Addgene Plasmid # 113591) or pXPR_053 vector containing gRNA specific for the Ly6E gene in StemSpam SFEM II media (Stemcell technologies, Cat# 09605).After 24 hours, fresh SFEM II media was added, and two days later, supernatants were centrifuged at 3000 x RPM for 10 minutes and filtered through a 0.45 mm syringe filter.The viral particles were then transduced into the Lineage negative cells (Lin -) harvested from the bone marrow of donor Cas9 overexpressing mice using MagCellect mouse hematopoietic cell lineage depletion kit (R&D systems, Cat# MAGM209).Transduced Lin -cells were then allowed to grow in-vitro for 24 hours with effective transduction of over 80%.Subsequently, the cells were intravenously injected (100,000 cells/mouse) into lethally irradiated (1000rad) wild type C57BL/6 mice and allowed to reach bone marrow reconstitution by week 8 post bone marrow transplantation.The recipient mice were implanted with LLC M cells.When tumor size reached 50 mm 3 , blood was drawn and Ly6E expression was assessed on different immune cells.Subsequently mice were treated with aPD1 or control IgG antibodies (n=6-7 mice/group), and tumor growth was assessed.In some experiments, the recipient mice were used to study Ly6E knockdown in-vitro as outlined in the text.

Single cell RNA sequencing on GR1 + cells
The evaluation of GR1 + myeloid cells in responsive and non-responsive tumors was performed by single-cell RNA sequencing (scRNA-seq).Briefly, 4T1 P and 4T1 M tumors were prepared as single cell suspensions.Subsequently, GR1 + cells were isolated by positive isolation (EasySep Mouse PE, Stemcell Technologies, Cat# 17666).The cells were then washed in PBS with 0.04% BSA and resuspended in 1000 cells/mL PBS.RNA was extracted and immediately was acquired by the 10X Genomics single cell sequencing system, as per manufacture's instructions.Bioinformatic analysis was then carried out as described below.

QUANTIFICATION AND STATISTICAL ANALYSIS
CyTOF pre-processing and analysis CD45 + gated FCS files were imported into R [v4.1.0]for unsupervised, cluster-based analysis using CATALYST 56 [1.24.0].Aggregated data was transformed (arcsinh) and clustered using FlowSOM (10 x 10 grid) and ConsensusClusterPlus meta-clustering to yield 25 distinct clusters, annotated based on the expression levels of all markers inspected in parallel.To detect clusters differentially abundant between responders (4T1 M ) and non-responders (4T1 P ), generalized linear models (GLMs) were fit using the adapted edgeR protocol in diffcyt 57 [v1.20.0]-which reports adjusted, FDR-corrected p-values for each comparison.
Single cell RNA-seq alignment and pre-processing Raw, Illumina base calls (BCLs) were demultiplexed and the resulting FASTQ files were aligned to the mm10 (GRCm38, Ensembl 93) murine reference genome and normalized for sequencing depth using 10x Genomics CellRanger [v 5.0.1] to generate expression matrices.82.8-85.7% of reads mapped to the transcriptome across all samples.A median of 3,252 and 2801 unique molecular identifiers (UMI) per cell for 4T1 P and 4T1 M were observed respectively.R and Python [v3.8.5] were used for all downstream analyses.Genes expressed in <10 cells were discarded.High-quality cells were retained by excluding: (i) cells expressing <500 or >5000 unique genes and (ii) cells with a mitochondrial UMI proportion of >10% -yielding 4711 cells and a total of 14214 detectable genes.SCTransform [v0.3.2], 58accessed via Seurat [v4.0.3], 59 was utilized to normalize and scale the data, select 3000 variable features and linearly regress out any remaining influence of mitochondrial UMI% on downstream analyses.SCTransform specifically mitigates technical factors, but retains biological heterogeneity, improving downstream analysis.

Dimensionality reduction, unsupervised clustering and differential abundance analysis
Data from all samples was aggregated and, as calculated by the Seurat [v4.0.3] functions RunPCA and RunUMAP respectively (default parameters), the top 3000 variable features and 25 principal components were utilized to generate a uniform manifold approximation and projection (UMAP) for visualization of the data.To assess globular, cellular heterogeneity, transcriptionally distinct cell states were defined by shared k-nearest-neighbour (s-KNN) analysis and Louvain-Jaccard clustering via the Seurat [v4.0.3] functions FindNeighbors and FindClusters respectively, using a resolution of 0.75.Cellular neighborhoods displaying differential abundance between conditions were defined by DASeq [v1.0.0] 61 using the top 10 principal components and k-values of [50-1000] at 50 step-wise intervals.Non-significant neighborhoods were discarded, as determined by a random permutations test.

Differential gene expression analysis
All differentially expressed genes were identified using the scRNA-seq-specific tool MAST [v1.18.0] 65 accessed via the Seurat [v4.0.3]FindMarkers function.Significance was assessed by calculating adjusted FDR p-values using the Bonferroni correction method and a gene was considered to be differentially expressed if its log 2 fold-change was >±0.35.

Gene modules and pathway analysis
To identify genes with pseudotime-associated patterns of expression, negative binomial generalized additive models (NB-GAMs) were fit to 14,000 genes and the significance of association was statistically tested by tradeSeq [v1.6.0]. 70NB-GAMs were fit using the parameter nknots=6 -a conservative estimate, as determined by the tradeSeq function, evaluateK, to avoid overfitting.Expression patterns were binned (n=20) along pseudotime and clustered via clusterExperiment [v2.12.0] 71 to define distinct gene modules.To characterize each module, over-representation tests were performed using clusterProfiler [v4.0.0] 72 and gene-lists from the HALLMARK database 75 (biological processes) and msigdbr [v7.4.1] (category = C3, transcription factors).The latter determines which, if any, transcription factors (TFs) regulate the genes present in each module.Only significantly enriched (FDR < 0.01, Bonferroni correction method) processes and TFs were retained.

Trajectory alignment
To compare trajectory lineages, a common pseudotemporal axis was defined using cellAlign [v0.1.0] 21-set to default, globalAlignment parameters as specified here: https://github.com/shenorrLab/cellAlign. In brief, inferred pseudotime values (defined by PAGA/ RNA velocity), and the normalized expression values of all genes in modules common to both lineages were utilized to align the trajectories across 200 interpolated points and module enrichment values were averaged at corresponding, aligned pseudotime values.
Human analysis of blood scRNA-seq Raw, scRNA-seq expression matrices were downloaded from the GEO Omnibus database (GSE127465) 35 (n=8, blood, patients with NSCLC at baseline).Data was imported into Seurat [v4.0.3] and pre-processed using SCTransform [v0.3.2] with identical filtering criteria to mouse -yielding 13403 cells and a total of 22413 detectable genes.To classify all 13404 cells in an unsupervised manner, SingleR [v1.6.1] was utilized to compare the transcriptome of each cell to the Human Primary Cell Atlas reference, as provided by celldex [v1.2.0].1701 (14.4%) non-immune cells or cells with ambiguous or poor-quality classifications were excluded.Human-specific gene-lists from the HALLMARK database, as accessed in R via msigdbr [v7.4.1], for (i) interferon_alpha_response (ii) interferon_ gamma_response and (iii) tnfa_signalling_via_nfkb were combined to generate a functional signature representative of Ly6E hi neutrophils.The enrichment of each, individual cell for the resulting signature was scored using the Seurat [v4.0.3] ssGSEA-like function, AddModuleScore.

Statistical analysis
All statistical tests were performed in R [v4.1.0].Statistical, pairwise comparisons for ELISA, LEGENDplex and Flow Cytometry data were performed using unpaired, two-sample Mann-Whitney tests (R function: wilcox.test)for n=2, or by one-way ANOVA coupled with Tukey's post-hoc HSD test for n>2 (R functions: aov and TukeyHSD).Two-sample Kolmogorov-smirnov tests (R function: ks.test) were utilized to compare tumor growth curves.Mice were randomized before tumor implantation.The analysis of the results was performed blindly.At least 5 mice per group were used in order to reach statistical power considering a Gaussian distribution.For in-vitro studies, at least three biological repeats were carried out, unless indicated otherwise in the text.Where appropriate (e.g.differential gene expression analysis), p-values were adjusted using the Bonferroni correction method to control for type I error rates i.e. false discovery rate (FDR).In all cases, significant differences were considered if p-values or FDR were <0.01.The number of samples or independent experiments are indicated in the text.For patients with NSCLC and melanoma, the investigators were blinded to allocation (i.e.RECIST categories) during experiments and outcome assessment.Co-variates including age, sex and stage were not controlled for.

B FIGURE S3
Figure S3.Gating strategy of Ly6E (hi) neutrophils in mouse and human, related to Figures 3 and 6. (A) Representative flow cytometry plots taken from the blood of BALB/c mice bearing parental 4T1P (NR) or mutagenized 4T1M (R) breast carcinoma.The analysis was performed on whole blood cells gated for CD45 + /CD11b + , Ly6C (lo) /Ly6G + , and Ly6E (hi) .
Notably, the percentage of Ly6E (hi) gated cells is higher in mice bearing 4T1 M tumors.(B) Representative flow cytometry plots taken from peripheral blood mononuclear cells (PBMCs) of patients with NSCLC at baseline.Data is stratified into non-responders (NR) and responders (R) based on RECIST criteria at 3 and/or 6 months.The analysis was performed on PBMCs gated for CD45 + , Lin -/HLA-DR -, CD14 -/CD15 + , and Ly6E (hi) .Notably, the percentage of Ly6E (hi) gated cells is higher in responding patients.

Figure 5 .
Figure 5. Tumor-intrinsic STING activity induces the Ly6E (hi) phenotype and in-turn supports activation of effector T cells

Figure S2 .
Figure S2.scRNA-seq of GR1 + cells from 4T1 tumors, related to Figure 3. 10X scRNAseq was performed on GR1 + cells obtained from 4T1 breast cancer parental (4T1P) (nonresponsive) and mutagenized (4T1M) (responsive) tumors (n=3 mice pooled/group).Data is identical to that used in Fig 3. (A) UMAP plot of 4711 filtered, GR1 + cells from which 1825 cells are monocytes (4T1 P = 815 cells, 4T1 M = 1010 cells).Cells are colored by cell type.(B) Differential abundance plot of monocytic cells (dotted box in (A)), highlighting two significantly enriched, cellular neighborhoods (dotted lines).The top 10, most significant marker genes of each neighborhood are listed (FDR<0.001,log2 fold-change >1.5).Genes underlined are classical markers of M1 (responders) and M2 (non-responders) macrophages.(C-I) scRNA-seq analysis of the neutrophil subpopulation displaying (C) a UMAP plot of 2886 GR1 + neutrophils, highlighting two significant, differentially abundant cellular neighborhoods found in 4T1P and 4T1M tumors.Data is identical to that shown in Fig 3. Significance of each neighborhood was determined via a randomized, permutations test; (D) Raw, RNA-velocity flow-field vectors projected back onto the UMAP; (E-I) Binned, normalized expression of known granulocytic progenitor genes.Data was imputed for visual clarity.

Figure S8 .Figure 5 .
Figure S8.Ly6E itself has no functional role in immunotherapy response, related toFigure 5. (A-G) Lin -bone-marrow progenitor cells, isolated from C57BL/6 strains overexpressing Cas9, were transfected with lentivirus carrying a single guide RNA GFP-CRISPR-Cas9 construct targeting the Ly6E locus.A transfection rate of ~80%, as measured by GFP expression, was observed.Transfected cells were subsequently implanted into irradiated, otherwise WT C57BL/6 mice (C57BL/6Ly6EKO).(A) Frequencies of Ly6E (hi) cells across all major immune subtypes were determined by flow cytometry (n=5 mice/group).CTRL = single guide RNA without the Ly6E-targeting.(B) Averaged tumor growth profiles for C57BL/6CTRL or C57BL/6Ly6EKO mice implanted with mutagenized Lewis-lung carcinoma (LLCM), and treated with αPD1 or control IgG antibodies (n=6-7 mice/group).(C-G) CD8 + T cells, obtained from the spleens of LLC tumor bearing mice, were cultured in-vitro either alone or with conditioned media derived from GR1 + cells or IFNαγ-induced, Ly6E (hi) neutrophils (n=4-5 biological repeats/group).Frequencies or levels of: activated, CD8 + CD25 + T cells (C); effector CD8 + T cells (D); proliferating CD8 + T cells (E); and Granzyme B + CD8 + T cells (F), were determined by flow cytometry.(G) In a separate experiment, 4T1P cells were cultured in the presence of CD8 + T cells or Ly6E (hi) neutrophils for 24 hours in a ratio of 1:10:10.T cell killing efficacy was analyzed by flow cytometry using PI to detect dead tumor cells (n=4 biological repeats/group).Significance was assessed by means of a two-sample KS-test (growth) and one-way ANOVA and Tukey's post-hoc HSD test (flow and ELISA) (NS, p>0.01).

TABLE
Human analysis of blood scRNA-seq B Cell-specific deconvolution B Statistical analysis SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.ccell.2023.12.005.ACKNOWLEDGMENTS This work is supported primarily by an H2020 European Research Council Grant (771112) and the Israel Science Foundation (194/18) as awarded to Y.S., and NIH CA231325 awarded to S.S.O.M.B. is supported by Ariane de Rothschild Women Doctoral Program.T.J.C. is supported by RTICC-Rubinstein fellowship.A select number of illustrations were drawn using BioRender.
d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d EXPERIMENTAL MODEL AND SUBJECT DETAILS B The establishment of diverse models to study predictive biomarkers for immunotherapy B Cell lines and culture B Mouse tumor models B Blood collection from patients with cancer d METHODS DETAILS B Cell line mutagenesis B Tumor lysate preparation and protein measurement B Cytokine array and biological pathway enrichment B Flow cytometry acquisition and analysis B Time of flight mass cytometry (CyTOF) B Adoptive transfer of Ly6E hi neutrophils experiments B Real-Time quantitative PCR (RT-qPCR) B CD8 + T cell assay B Tumor cell killing assay B dsDNA acquisition B STING signaling pathway analysis B Ly6E knock-down in bone marrow cells B Single cell RNA sequencing on GR1 + cells d QUANTIFICATION AND STATISTICAL ANALYSIS B CyTOF pre-processing and analysis B Single cell RNA-seq alignment and pre-processing B Classification of cell types B Dimensionality reduction, unsupervised clustering and differential abundance analysis B Data visualization B Differential gene expression analysis B RNA-velocity and trajectory inference B Gene modules and pathway analysis B Trajectory alignment B