Metastasis-Entrained Eosinophils Enhance Lymphocyte-Mediated Antitumor Immunity

Over the past two decades, it has been gradually acknowledged that the tumor microenvironment (TME) has a central role in tumor growth and progression (1). Specifically, the composition and activity of tissue resident and recruited immune cells have gained significant attention and are pursued therapeutically in experimental and clinical settings (2).

Eosinophils are bone marrow (BM)–derived granulocytes that have been predominantly studied in the context of allergic diseases (3). Accumulating evidence suggests that eosinophils are an integral part of the immune landscape of various tumor types, especially in mucosal-barrier organ tumors (4). Furthermore, eosinophils accumulate in the TME in response to distinct therapeutic strategies such as immune checkpoint blockade (5), dipepitidyl peptidase 4 inhibitors (6), and cytokines (e.g., IL33, IL4, and GMCSF; 7). Despite these data, the roles and environmental cues that direct eosinophil activities in the TME remain largely obscure.

The presence of increased tumor-infiltrating or peripheral blood eosinophilia has been clinically associated with both favorable or poor prognosis in different tumor types (7). Thus, eosinophils have been attributed with pro- and antitumorigenic activities. These apparently opposing clinical associations and activities led to controversy regarding the roles of eosinophils in the TME (8). However, similar to other myeloid cells, such as macrophages (9) and neutrophils (10), eosinophils likely display pleiotropic functions, which are dictated by the specific mechanisms involved in their recruitment and activation into a given TME. Recent data suggest that eosinophils are a heterogenous population displaying diverse functional activity. For example, two distinct eosinophil populations have been described in settings of experimental asthma; tissue-resident eosinophils (characterized as CD125^intSiglec-F^intCD62L⁺CD101^low cells) and recruited eosinophils (characterized as CD125^intSiglec-F^hiCD62L⁻CD101^hi cells). Notably, tissue-resident eosinophils displayed a regulatory phenotype, whereas recruited eosinophils promoted type 2 (Th2-driven) inflammatory responses (11). Whether such eosinophil populations exist in the TME and whether they respond differentially to environmental cues is currently unknown.

Herein, we aimed to study eosinophils in the TME of an anatomically relevant location that is comprised of resident eosinophils and is potentially capable of recruiting eosinophils. Employing these criteria led us to investigate eosinophils in breast cancer–driven lung metastasis, which is one of the most common metastatic sites, a major cause of cancer-related death (12) and a natural anatomical niche for eosinophils (11).

We report that human metastatic lung biopsies from patients with breast cancer are characterized by infiltrating eosinophils. Consequently, we show that eosinophils are actively recruited to the metastatic lung via a G protein–coupled receptor (GPCR)–dependent, CCR3-independent manner and that tumor secreted factors can promote eosinophil migration independent of eotaxins. Eosinophils in the TME displayed antitumorigenic activities and markedly altered the cellular composition of lymphocytes in the TME. Transcriptome and proteome analysis of eosinophils revealed that the TME was the main instructing factor for eosinophils to acquire an antitumorigenic activities independent of their origin (i.e., resident vs. recruited). Furthermore, eosinophils displayed a transcriptome and proteomic signature that was enriched for TNFα and IFNγ signaling. Adoptive transfer of eosinophils facilitated the infiltration of CD8⁺ and CD4⁺ T cells and promoted lymphocyte-mediated antitumor immunity. These results demonstrate a key role for eosinophils in breast cancer–driven lung metastasis and introduce TNFα and IFNγ as key drivers of their antitumor response.

Breast cancer patient FFPE samples of lung metastasis biopsies (n = 22) and corresponding patient characteristics were collected and processed at the Tel Aviv Sourasky Medical Center (Ichilov, Israel), in accordance with the required ethical guidelines, under an approved Institutional Review Board authorization (0471–18-TLV). All of the patient clinical information is summarized in Fig. 1A.

C57BL/6 WT, BALB/c WT, and R2G2 (B6;129-Rag2^tm1Fwa Il2rg^tm1Rsky/DwlHsd) mice were purchased from Harlan Laboratories. GREAT mice (B6.129S4-Ifng^tm3.1Lky/J) were purchased from the Jackson Laboratory. CD3-Il5 transgenic mice (NJ.1638, Il5^Tg) mice were kindly provided by Dr. Jamie Lee (Mayo Clinic, Scottsdale, AZ). ΔdblGATA mice were kindly provided by Dr. August Avery (Cornell University, Ithaca, NY). OT-I (C57BL/6-Tg(TcraTcrb)1100Mjb/J) mice were kindly provided by Prof. Nir Friedman (Weizmann Institute, IL). Experiments on Ccr3^−/− mice (13) and BALB/c controls were conducted in the animal facilities of Cincinnati Children's Hospital Medical Center (CCHMC). All experiments were reviewed and approved by the Animal Care Committee of the Tel Aviv University or CCHMC and were performed in accordance with the regulations and guidelines regarding the care and use of animals for experimental procedures (01–18–086). All experiments were conducted in the specific pathogen-free facilities and age-, weight-, and sex-matched mice were used.

A classifier for identification of eosinophils was developed using the Mouse Hemopedia Gene Expression dataset (14), which is available on Gene Expression Omnibus (GEO) through accession number GSE116177. Raw counts were downloaded from the corresponding GSE accession, logCPM transformed and TMM normalized (15) using edgeR (16). An extreme gradient boosting classifier (17) was trained to 70% of the data using linear boosting (booster = “gblinear”) with a binary logistic loss function (objective = “binary:logistic”), 5 rounds of iteration (nrounds = 5), and a moderate L2 penalty of 0.01 (lambda = 0.01). The classifier was tested on the remaining 30% of the dataset. The model was constructed on the basis of top 5,000 highly variable genes (HVG) from a single-cell lung eosinophil RNA-seq dataset (GSE153632), where HVG genes were determined by Seurat v3 SCTransform workflow (18).

Normalized gene counts for 20 primary and matched lung/pleural metastatic breast tumor RNA-seq profiles were downloaded from GSE147322 (19), corresponding to gene expression profiles of 10 patients. The logCPM transformation was applied to the data as described earlier. The abovementioned classifier was applied to logCPM values to compute a confidence score (prediction) for the presence of infiltrating eosinophils in the primary and metastatic tumor of each patient.

The fry test (Fast Approximation to ROAST; ref. 20) in limma R/Bioconductor package was applied to test for enrichment of macrophage and T-cell signatures published by Azizi and colleagues (21) and Szabo and colleagues (22) in lung metastatic samples with high infiltrating eosinophils compared with low eosinophil-infiltrating metastatic tumors. The infiltration level in a given metastatic sample was determined relative to infiltration in the matched primary sample.

Mouse eosinophils were isolated from the spleen or peritoneal cavity of Il5^tg mice under sterile conditions; For splenic eosinophils, the spleen was dissected, minced, and passed through a 70-μm cell strainer into PBS. Cell pellets were then incubated with RBC lysis buffer for 10 minutes at room temperature. For peritoneal eosinophils, peritoneal cavity was washed with 10 mL of PBS. Thereafter, negative selection of eosinophils was performed using anti-Ty1.2 (11443D, Invitrogen) and anti-B220 (11331D, Invitrogen) Dynabead-conjugated antibodies according to the manufacturer's instructions. Eosinophil purity was validated using flow cytometry. For functional assays, the purity of eosinophils was >90% and viability >95%. For RNA sequencing (RNA-seq) eosinophil purity was >99%. Human eosinophils were isolated from healthy donor peripheral blood using MACS eosinophil isolation kit (130–092–010, Miltenyi Biotec) according to the manufacturer's instructions. Eosinophil purity was validated using flow cytometry; APC-conjugated anti-CD45, PE-conjugated anti–Siglec-8, APC-Cy7-conjugated CCR3 antibodies and DAPI. Eosinophils were used when purity >90% and viability >95%.

Intravenous eosinophil adoptive transfer was conducted by obtaining splenic eosinophils (10⁸ eosinophils/per mouse), which were resuspended in 200-μl sterile saline and injected to the tail vein of ΔdblGATA mice. Forty-eight hours later, the mice were euthanized, and lung eosinophilia was assessed using flow cytometry. For intranasal eosinophil adoptive transfer, 5×10⁶−10×10⁶ eosinophils/per mouse were incubated in 37°C for 1 hour in DMEM supplemented with 10% FCS with 50 ng/mL of IFNγ and TNFα or media only. Cells were then washed twice with PBS, resuspended in 50-μL sterile saline and applied to the mouse nasal cavity following anesthesia by inhalation of isoflurane on days 3 (media only) and 1, 3, 5, 9, 11 (IFNγ/TNFα activated) from the injection of tumor cells. Mice were euthanized on day 14 and tumor burden as well as immune cell infiltration was assessed.

Tumor cell–conditioned media (CM) were prepared by growing tumor cells until confluent. The cells were then washed and cultured with serum-free media for 72 hours. Thereafter, media were collected, cleared of residual cells and debris via centrifugation and loaded into 3,000 MWCO Vivaspin centrifugation columns (VS2092, Sartorius) for 2 hours at maximum speed.

A Transwell system of 3- or 5-μm pores was used for mouse and human cells, respectively. For GPCR inhibition, eosinophils were preincubated with pertussis toxin (10 mg/mL, LBL-180, List Biological Laboratories) for 2 hours in 37°C and thereafter washed 3 times with PBS. Tumor cell CM was inserted in the lower chamber and 10⁶/mL eosinophils were added to the upper chamber. The cells were then incubated in for 3–6 hours in 37°C. Thereafter, the lower chamber fluid was collected, and relative cell count was determined by flow cytometry.

To determine the ability of tumor-secreted factors to induce eosinophil recruitment in vivo a tumor-free model was established (23). C57BL/6 female mice received intraperitoneal injections of 200-μL PyMT cell CM or serum-free media control for five sequential days. Three hours following the last injection, the mice were euthanized, and the peritoneal cavity was washed with 10 mL of PBS. Eosinophil accumulation in the peritoneal wash was then assessed by flow cytometry.

To pharmacologically deplete eosinophils, mice were intraperitoneally injected with 20 μg of rat anti–mSiglec-F (clone 238047, R&D Systems) or the respective isotype-matched control antibody rat IgG2a (clone 2A3, BioXcell). Intraperitoneal injections were performed 6, 4, and 2 days before and every 2 days after tumor-cell inoculation. To pharmacologically deplete CD4 and CD8 T cells, mice were intraperitoneally injected with 100 μg of anti-CD8a (clone 53–6.7, BioXcell) and anti-CD4 (clone GK1.5, BioXcell) as previously described (24). Intraperitoneal injections were performed 1 day before and 5 days after tumor-cell inoculation.

Spleens were collected from naive OT-I mice and were dissociated to obtain single-cell suspensions. Red blood cells were lysed with ACK buffer. Cells were resuspended at a density of 2 × 10⁶ cells per mL in complete RPMI medium with 1 μg/mL SIINFEKL (S7951 Sigma) for 72 hours. Thereafter, Ficoll-Paque separation was conducted, and monolayer cells were resuspended in RPMI supplemented with 10% FBS, 1% sodium pyruvate (03–042–1B, Biological Industry), 1% l-glutamine (03–020–1B, Biological Industry), 100 U/mL penicillin, 100 μg/mL streptomycin, 1% non-essential amino acids (01–340–1B, Biological Industry), 0.05 mmol/L 2-mercaptoethanol and 50U/mL IL2 (212–12, Peprotech), and incubated for 24 hours. CD8⁺ T-cell purification was verified >98% using flow cytometry. 5 × 10⁵ CD8⁺ T cells were loaded in Transwell filters (5 μmol/L pores), which were placed in 24-well plates containing 600 μL medium. After incubation for 3 hours at 37°C, cells in the bottom well were collected and counted by flow cytometry.

PyMT breast cancer cell lines were derived from primary tumor-bearing transgenic mice expressing polyoma middle T-antigen (PyMT) under the control of the murine mammary tumor virus promoter, and were kindly provided by Prof. Yuval Shaked (Technion, Haifa, IL). 4T1 breast cancer cell line was kindly provided by Prof. Neta Erez (Tel Aviv University, Tel Aviv, IL). MDA-MB-231 and MCF7 cells were kindly provided by Prof. Adit Ben-Baruch (Tel Aviv University, Tel Aviv, IL). Cell lines were not authenticated in our laboratory. All cell lines were routinely tested for Mycoplasma using the EZ-PCR-Mycoplasma test kit (Biological Industries; 20–700–20). All cells were cultured in DMEM supplemented with 10% FCS and 1% penicillin–streptomycin solution in a 5% CO₂ humidified incubator at 37°C. PyMT-mCherry-Luc cells were generated by transduction with lentivirus encoding mCherry and Luciferase. For transduction, cells were seeded (5 × 10⁴ cells per 0.5 mL) and 300 μL of lentivirus were added with 10 μg/mL polybrene (TR-1003-G, Sigma-Aldrich), centrifuged (1,500 × g, 90 minutes) and transferred to incubation at 37°C (overnight). Transfected mCherry⁺ cells were sorted and seeded.

All cell lines used in this study tested negative for Mycoplasma. PyMT, PyMT-mCherry-Luc, as well as 4T1 tumor cells were grown in DMEM with 10% FCS. Tumor cells were detached using 0.25% Trypsin- 0.05% EDTA (Biological Industries) and washed twice in PBS before injection. For experimental lung metastasis, 1 × 10⁵ PyMT, PyMT-mCherry-Luc or 4T1 (Passage 2–4) were resuspended in saline (200 μL) and used for tail vein injections, saline only was used as control. Lung tumors were quantified using in vivo imaging and histological tumor quantification. For in vivo imaging, PyMT-mCherry-Luc–injected mice were anesthetized using 2.5% isoflurane and administered with D-luciferin (150 mg/kg, Regis) by intraperitoneal injection at the indicated time points. The bioluminescence signals were detected using an IVIS Lumina III system (PerkinElmer). Spontaneous breast cancer metastasis to the lungs was induced as follows; 2 × 10⁵ transplantable 4T1 tumor cells were suspended in PBS and mixed 1:1 with Matrigel (BD Biosciences, 354230). A total of 100 μL of cell mixture was injected into the right inguinal mammary glands of 8-weeks-old female BALB/c mice. Tumors were resected 3 weeks following the injection, 3 weeks later, mice were euthanized for metastatic lung evaluation.

For experimental models of primary tumors, orthotropic injections were performed to the 4th inguinal mammary fat pad using 10⁵ PyMT tumor cells resuspended in 100 μL PBS. Primary tumors were palpable after 7 days and tumor size was monitored starting from day 10 using a caliper. Asthma was induced using nine intranasal challenges of house dust mite.

The differential expression (DE) of the sequenced genes was calculated using the DEseq2 (25) package in R language on Rstudio. Genes were regarded as differentially expressed if they presented false discovery rate (FDR) adjusted P value below 0.05, and |${\rm log2( {fold\ change} ) \pm 1.5}$|⁠. The samples were presented using principal component analysis (PCA) created with ggplot R package. DE genes (DEG) between eosinophil groups were reduced to two principal components, who visualize the two most variant planes in the multidimensional space. Volcano plots were created using EnhancedVolcano package in R. Volcano plots visualize results of DE analyses. The filtering parameters for DEGs [P value and |${\rm log2( {fold\ change} )}$|] are presented as the x and y axes. In this manner, the overall difference between the compared groups is clear and selected genes can be highlighted to observe their differential state.

Following DE analysis, the differentially expressed genes were separated to transcription factors (TF) and targets. The targets of all DE TFs were determined using ChIP-seq data from the database GTRD (26). TFs and targets defined as DE were added as network nodes. Edges (interactions) were added as connections between them. The network figures were created using Cytoscape. DE TFs were ranked to identify which are the most dominant ones. HG stands for hypergeometric, the value portrayed in the heatmap is the -log10(P value) of a hypergeometric enrichment test. The targets of each TF are tested for enrichment of DE targets, relative to targets that are not DE. The HG calculation was conducted using the python SciPy package. Cent stands for centrality, which is a parameter that is given to each node, based on the shortest path from the node to the other nodes in the network. It represents how central and connected a node is to the rest of the network. The centrality calculation was conducted using the python NetworkX package. Both Cent and HG were normalized, the Rank column is an average of the normalized HG and Cent values.

Selected genes of interest (GOI) were subsetted from the sequencing expression data. The expression data in the heatmap are scaled per gene (row). The heatmaps were created using the R package Pheatmap. The heatmaps were hierarchically clustered using the built-in Pheatmap hierarchical clustering method.

Using a gene set enrichment analysis (GSEA) API (27) for python, the eosinophil sequencing data were tested for enrichment of GO (28) “biological process” terms. Several annotations from the gene sets of the GO biological process were revealed to be significantly enriched, whereas several others were depleted. A bar plot presenting the GO terms and their enrichment FDR P value was created using matplotlib (29) package in python (30).

To identify molecules upstream of the genes in the dataset that potentially explain the observed expression changes, ingenuity pathway upstream regulator analysis (URA) was performed using IPA software (Ingenuity System; ref. 31). IPA was carried out for genes whose expression was considered to be differentially expressed. Top five significant overlap P value cytokine and TFs upstream regulators were selected, and their activation Z-score is presented.

Sample sizes were determined by prior experience and to achieve a confidence level of at least 95%. All summary statistics were calculated using GraphPad Prism v.9.0. Statistical significance between groups was determined by unpaired, two-tailed student t test or one/two-way ANOVA (P ≤ 0.05 was considered statistically significant). Statistical analysis of the proteomic profiling was performed using the Perseus software version 1.6.2.3. For statistical analyses, the proteinGroups output table was used. Reverse proteins, proteins that were only identified by site, and potential contaminants were filtered out, and proteins intensities were LFQ normalized and log₂-transformed. Protein groups were filtered to have valid values in at least 70% of the samples. Missing value imputation was performed sample-wise by drawing values from a normal distribution with a downshift of 1.6 standard deviations and a width of 0.4. Network analysis was based on student t test results (FDR < 0.1, S0 = 0.1) and on the STRING (32) algorithm and visualization was done using the Cytoscape (33) software. Biological annotations were taken from Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG).

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD027596. RNA-seq raw and processed gene expression values used in this study were deposited on the NCBI GEO website with the accession code GSE181491 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE181491).

Breast cancer–associated mortality is primarily a result of tumor metastasis (12). Because the lungs are a main metastatic site (12) and given that eosinophils mainly infiltrate tumors in mucosal barrier tissues (4), we aimed to determine the level of eosinophil infiltration into metastatic lungs. We examined the presence of eosinophils in lung biopsies from a cohort of 22 patients with breast cancer–driven lung metastasis (Fig. 1A) using highly specific anti-eosinophil peroxidase (EPX) staining (Fig. 1B). Analysis of this cohort showed that patient biopsies exhibited tumor-infiltrating eosinophils (21.71 ± 18.5 eos/mm², Fig. 1C). Furthermore, varying levels of eosinophil degranulation, a hallmark characteristic of activated eosinophils (3) was observed. The metastatic lungs in our cohort displayed either intact eosinophils (Fig. 1B, left), degranulated eosinophils (Fig. 1B, right) or both intact and degranulated cells (Fig. 1D). Overall, 9 out of the 22 biopsies (40.90%) displayed eosinophil degranulation (Fig. 1E). Thus, eosinophils infiltrate and degranulate in breast cancer–driven lung metastasis.

Next, we aimed to determine whether eosinophils accumulate in the lungs in experimental breast cancer–driven lung metastasis. To this end, BALB/c and C57BL/6 mice were tail-vein injected with 4T1 or PyMT breast cancer cells, respectively, and eosinophils in the lungs were detected using anti-eosinophil major basic protein (MBP) staining. Eosinophils were localized within the tumor nodules and in the surrounding lung parenchyma (Fig. 2A; Supplementary Fig. S1A). No eosinophils were detected in the bronchoalveolar lavage fluid (Supplementary Fig. S1B). Flow cytometric analysis demonstrated increased eosinophil numbers (defined as viable CD45⁺/CD11b⁺/Siglec-F⁺/CD125^int/SSC^hi cells, Supplementary Fig. S1C and S1D) in the lungs following seeding of PyMT and 4T1 cells (Fig. 2B and C, gated on viable CD45⁺/CD11b⁺ cells). Increased lung eosinophil numbers were accompanied with a 2-fold decrease in peripheral blood eosinophil levels (Supplementary Fig. S1E). Assessment of eosinophil progenitor (EoP) levels revealed no alteration in BM EoP numbers (Supplementary Fig. S1F and S1G) whereas lung EoPs were reduced (Supplementary Fig. S1H and S1I). In vivo EdU-labeling experiments revealed no proliferating EoPs in the BM (Supplementary Fig. S1J). In addition, while gating on total lung CD45⁻ or CD45⁺ cells revealed EdU⁺ cells (Supplementary Fig. S1K and S1L), no EdU⁺ eosinophils were observed (Supplementary Fig. S1M). Collectively, these data suggest active recruitment of eosinophils to the metastatic lung.

Recently two eosinophil populations were described in the asthmatic lung (11), namely resident (defined as Siglec-F^int/CD125^int/CD101^low) and recruited eosinophils (defined as Siglec-F^hi/CD125^int/CD101^hi; Fig. 2D; Supplementary Fig. S2A and S2B). Similarly, following seeding of PyMT and 4T1 cells in the lungs we observed two eosinophil subsets. The Siglec-F^int resident population was seen in the naïve and metastatic lung, whereas the Siglec-F^hi population was solely observed in the metastatic lung (Fig. 2D). Intermediate levels of CD125 and Siglec-F on eosinophils were verified by assessing their expression in eosinophils versus CD45⁺ cells (Supplementary Fig. S2C and S2D). In accordance with the phenotype of recruited eosinophils, the Siglec-F^hi population displayed low, albeit evident levels of CD101, whereas the Siglec-F^int population was CD101⁻ (Fig. 2E). The expression of CD62 L was not detected on either population. Given the shared expression of Siglec-F and CD125 on alveolar macrophages and because alveolar macrophages were CD62L⁻/CD101⁺ (Supplementary Fig. S2E and S2F), we aimed to verify that CD45⁺/CD11b⁺/Siglec-F^int/CD125^int and CD45⁺/CD11b⁺/Siglec-F^hi/CD125^int cells were eosinophils. Indeed, the aforementioned cells displayed a ring-shaped nucleus and eosinophilic cytoplasm corresponding with eosinophils (Supplementary Fig. S2G and S2H). Sorted CD45⁺/CD11b⁻/Siglec-F⁺/CD125^int cells displayed macrophage-like appearance (Supplementary Fig. S2I). We further distinguished between alveolar macrophages and eosinophils using an established gating strategy with additional cell surface markers (i.e., CD11c and Gr-1; Supplementary Fig. S2J; ref. 34).

To further translate our findings, lung eosinophils were assessed in an additional model of orthotopic transplantation of 4T1 cells followed by resection of the primary tumor, which results in spontaneous lung metastasis (35). Eosinophil infiltration was significantly increased in the metastatic lung. Similarly, a Siglec-F^int eosinophil population was seen in the naïve and metastatic lung, whereas the Siglec-F^hi population was observed only in the metastatic lung (Fig. 2F and G). Assessment of the ratio between the Siglec-F^int and Siglec-F^hi eosinophil s in the different experimental models revealed that the Siglec-F^hi eosinophil population comprised approximately 50% of lung eosinophils in the different experimental models (Fig. 2H).

To assess whether eosinophils are actively recruited to the metastatic lung, eosinophils were adoptively transferred via tail vein injection to control and metastasis-bearing ΔdblGATA (eosinophil deficient) mice. Metastasis-bearing mice had a 10-fold increase in lung eosinophil numbers in comparison with control mice (Fig. 2I and J). Recruited eosinophils following adoptive transfer displayed increased Siglec-F expression in comparison with eosinophils that were recruited to the naïve lung (Fig. 2K).

Increased infiltration of eosinophils to the metastatic TME suggests the presence of eosinophil chemotactic factors that facilitate their migration toward tumor nodules. The CCL11/CCL24-(eotaxins)-CCR3 signaling pathway has a critical role in recruitment of eosinophils to mucosal sites at baseline and in allergic inflammation (36). Furthermore, CCL11 and/or CCL24 expression has been correlated with eosinophil levels in multiple tumor types (37). Therefore, we hypothesized that the recruitment of eosinophils to the metastatic lung would be CCR3 dependent. The expression of CCL11, but not CCL24, was increased in the metastatic lung (Supplementary Fig. S3A and S3B). To functionally examine the contribution of CCR3, wild-type (WT) and Ccr3^−/− mice were subjected to the experimental lung metastasis model. No statistically significant difference was observed in lung eosinophil numbers between WT and Ccr3^−/− mice and their tumor burden was similar (Fig. 3A and B; Supplementary Fig. S3D and S3E). Moreover, eosinophil activation (as determined by degranulation and expression of CD11b and Siglec-F,) was similar in tumor-bearing WT and Ccr3^−/− mice (Fig. 3A; Supplementary Fig. S3F).

We further hypothesized that tumor-secreted factors can directly induce eosinophil migration. Thus, the migration of eosinophils toward PyMT CM was assessed in vitro, resulting in increased eosinophil migration (Fig. 3C). Although CCL11 and CCL24 were not detected in PyMT CM (Supplementary Fig. S3C), in vitro migration of eosinophils toward tumor-secreted factors was pertussis toxin-dependent (Fig. 3D). Subsequently, we determined the ability of tumor-secreted factors to recruit eosinophils in vivo, using a tumor-free model (23). Sequential PyMT CM injections to the peritoneal cavity of WT mice resulted in an approximately 4-fold increase in peritoneal eosinophil numbers in comparison with media control (Fig. 3E). Similar to our findings in vitro, adoptive transfer of eosinophils pretreated with pertussis toxin to the tail vein of metastasis bearing ΔdblGATA mice, resulted in decreased recruitment of eosinophils to the metastatic lung in comparison with control (Fig. 3F). Finally, CM from the human breast cancer cell line MDA-MB-231 and MCF7 induced the migration of human peripheral blood eosinophils (Fig. 3G).

Taken together, these results suggest a CCR3-independent tumor-secreted factor-induced GPCR-dependent pathway for recruitment of eosinophils to the breast cancer–driven metastatic lung.

To delineate the role of eosinophils in lung metastasis we used two independent approaches of eosinophil depletion. Induction of experimental lung metastasis using PyMT cells in ΔdblGATA mice resulted in increased tumor burden in comparison with WT mice. Increased tumor burden was evident as early as day 5 from injection and increased by day 7 (Fig. 4A and B). These results were recapitulated by pharmacological depletion of eosinophils using anti–Siglec-F antibody (38). Specific blood and lung eosinophil depletion was confirmed (Fig. 4C and D) using an established gating strategy that does not require Siglec-F for identification of eosinophils (i.e., viable CD45⁺/CD11b⁺/CD11c⁻/MHC-II⁻/Gr-1⁻/SSC^hi cells, Supplementary Fig. S4A; ref. 39). Consequently, eosinophil-depleted mice displayed increased tumor burden in comparison with isotype-treated control mice (Fig. 4E). Interestingly, eosinophil accumulation did not affect primary tumor growth: Despite increased accumulation of eosinophils in a transplantable model of breast cancer (by orthotopic injection of PyMT cell to the mammary gland; Supplementary Fig. S5), primary tumor growth was comparable between WT and ΔdblGATA mice (Fig. 4F).

Recent data suggest an intricate crosstalk between eosinophils and antitumor-adaptive immune responses driven by T cells (40, 41). Thus, we were interested to determine whether the antitumorigenic activities of eosinophils were associated with alterations in the cellular composition of CD8⁺ and CD4⁺ lymphocytes in the metastatic lung. Assessment of intratumoral infiltration of CD8⁺ and CD4⁺ T cells in PyMT-injected ΔdblGATA mice revealed that the absence of eosinophils resulted in decreased numbers of intratumoral CD8⁺ and CD4⁺ cells (Fig. 5A–D). Moreover, increased tumor cell proliferation was observed in ΔdblGATA mice as indicated by enhanced Ki67⁺ PyMT cells in the lungs (Fig. 5E and F). No differences were observed in T-cell abundance in the blood and thymus (Supplementary Fig. S6) whereas the spleen of PyMT-injected ΔdblGATA mice displayed significantly increased CD4⁺ and CD8⁺ T cells (Supplementary Fig. S6).

No statistically significant differences were observed in the composition of monocytes (WT = 11.58% ± 1.53%, ΔdblGATA = 15.58% ± 4.88%, P = 0.27) or macrophages in the lungs (WT = 6.38% ± 0.85%, ΔdblGATA = 8.35% ± 1.82%, P = 0.78). Yet, a trend toward decreased neutrophils in ΔdblGATA mice was observed (WT = 22.89% ± 3.7% of live leukocytes, ΔdblGATA = 16.80% ± 0.93%, P = 0.055).

These data suggested that the antitumor activity of eosinophils may be mediated by T cells. To examine this possibility, eosinophils were pharmacologically depleted in Rag2^−/−/Il2rg^−/− mice, which lack functional lymphocytes using anti–Siglec-F antibodies (Supplementary Fig. S4B and S4C) and Rag2^−/−/Il2rg^−/− mice were subjected to experimental lung metastasis (Fig. 5G). In contrast with increased tumor burden, which was observed following depletion of eosinophils in WT mice (Fig. 4C–E), no difference in tumor burden was observed in eosinophil-depleted Rag2^−/−/Il2rg^−/− mice (Fig. 5G). Furthermore, pharmacological depletion of eosinophils and CD4⁺/CD8⁺ T cells in WT C57BL/6 mice using anti-CD4 and anti-CD8 antibodies (Supplementary Fig. S4D and S4E) revealed no difference in tumor burden between eosinophil-depleted and control mice (Fig. 5H).

The lymphocyte-dependent antitumor effect of eosinophils in our experimental model, prompted us to define whether infiltration of eosinophils in human breast cancer–driven lung metastasis is associated with alterations in the cellular composition of the TME as well. Using a trained eosinophil bioinformatics classifier (Supplementary Table S1), we assessed the gene set enrichment of immune cell population signatures in lung/pleural metastasis samples from patients with breast cancer that are predicted to have high or low eosinophil-infiltration (obtained from dataset GSE147322 (Fig. 5I–M; Supplementary Table S2; ref. 19). The only signature, which displayed a significant enrichment in high eosinophil-infiltrating lung metastasis, was the gene signature of activated CD8⁺ T cells (Fig. 5I).

The environmental triggers, which direct eosinophils to adopt antitumorigenic activities, are largely unknown. To better understand how the TME instructs eosinophils, we undertook an unbiased approach comparing the transcriptional and proteomic profiles of naïve versus metastatic lung eosinophils. Because at least two eosinophil populations were present in the metastatic lungs (Fig. 2D), naïve resident lung eosinophils as well as Siglec-F^int and Siglec-F^hi eosinophils (Fig. 2D) from the metastatic lung were sorted subjected to RNA-sequencing and mass spectrometry analysis. PCA of both RNA-sequencing (Fig. 6A; Supplementary Fig. S7A) and proteomics data (Fig. 6B) revealed that naïve eosinophils are markedly distinct from metastatic lung eosinophils. Similar to our previous analyses, which revealed heterogeneity in intratumoral eosinophils from colon cancer (42), we observed substantial heterogeneity within eosinophils that were obtained from different mice in the metastatic TME (Fig. 6D and E). Despite this, no transcriptional or proteomic difference was observed between the resident and the recruited eosinophils in the metastatic lung environment (Fig. 6A and B; Supplementary Fig. S7B and S7C). In fact, assessment of the differentially expressed genes of these two eosinophil populations (11) did not cluster in our sorted populations (Supplementary Fig. S7D and S7E). Thus, in contrast with the phenotype of eosinophils in asthma (11), the major factor affecting the phenotype of eosinophils in the metastatic TME was the microenvironment rather than the origin of the eosinophil subset.

Subsequently, we analyzed the differentially expressed transcripts between naïve and metastasis-entrained eosinophils (MEE) containing both resident and recruited eosinophil populations (Fig. 6C). Our analysis identified 3,732 genes, which were differentially expressed (P_adj < 0.05, log₂ fold change >±1.5); of these transcripts, 2,280 were upregulated in MEEs and 1,452 downregulated, compared with naïve eosinophils (Supplementary Table S3). Specifically, Irf7, Itgam, Ifit1b, Rsad2, and H2-Eb2, which were recently associated with an IFNγ-stimulated neutrophil signature (43), were upregulated in MEEs (Fig. 6C). Moreover, Pten and Cd86, which are associated with M1 macrophage polarization, were upregulated in MEEs as well. In contrast, Akt1, which is associated with an M2 macrophage phenotype, was downregulated in MEEs (Fig. 6C). Correspondingly, single-sample gene-set enrichment analysis revealed that MEEs were enriched (normalized enrichment score = 2.11, adjusted P = 0.008) for an M1 gene expression signature (Supplementary Fig. S7F; Supplementary Table S4; ref. 44).

To further characterize the transcriptional phenotype of MEEs, we specifically focused on differential cell surface receptors, secreted factors such as cytokines and/or chemokines and finally expression of TFs. MEEs displayed increased expression of various cell surface receptors that were previously associated with eosinophil activation, including Cd86 and Cd69. Furthermore, various cytokine or immune receptors, which were previously implicated in regulating the activities of myeloid cells in cancer (e.g., Ifngr1, Il1rl1/St2, Il6r, Il18rap, Cd86), were increased (Fig. 6D; Supplementary Table S5). In addition, various immunoglobulin superfamily receptors such as Lair1, Cd300c were increased in MEEs. The pattern-recognition receptors Tlr2/6/7 were upregulated in MEEs (Fig. 6D).

Analysis of the expression of secreted factors (Fig. 6E; Supplementary Table S6) in MEEs revealed increased expression of various chemokines and cytokines. Specifically, T-cell recruiting chemokines (e.g., Cxcl9, Cxcl16), and chemokines involved in recruitment of myeloid cells (e.g., Cxcl12, Cxcl16, Ccl2, Ccl3 and Ccl4) were increased (Fig. 6E). MEEs displayed increased expression of “hallmark” inflammatory cytokines, including Il1b, Il6 and Tnfa. The type 2–promoting cytokines Il33 and thymic stromal lymphopoietin (Tslp) were also increased in MEEs.

Further analysis assessing the TF networks identified several factors that are differentially regulated in MEEs. The network depicts nodes representing the differentially expressed TFs, with red edges governing upregulated targets, and blue edges governing downregulated targets. We calculated the ranking of the TFs based on network centrality and hypergeometric enrichment. Among the highly ranked upregulated TFs were Distal-less homeobox 2 (Dlx2), androgen receptor (Ar), and estrogen receptor 1 (Esr1; Fig. 6F and G), which was previously shown to promote type 1 IFN production and signaling in innate immune cells (45). In contrast, microphthalmia-associated TF (Mitf) and serum response factor (Srf) were highly ranked as downregulated (Fig. 6H and I).

GSEA of the MEE transcriptome in comparison with naïve lung eosinophils revealed an enrichment in IFNγ and TNFα signaling pathways in MEEs in comparison with naïve lung eosinophils (Fig. 7A; Supplementary Table S7). In addition, MEEs were enriched in pattern recognition receptor signaling pathway, previously shown to be upregulated in colon tumor-infiltrating eosinophils (42). Pathways such as arachidonic acid metabolism and cytochrome c release, which were previously associated with eosinophil-mediated inflammation (46) or apoptosis (47), respectively, were reduced.

To further identify molecules that are central upstream regulators of the differentially expressed transcripts of MEEs, the URA was applied using ingenuity pathway analysis (31). The URA defines an overlap P value measuring the enrichment of network-regulated transcripts in the dataset, as well as an activation Z-score, which can be used to find molecules that likely regulate the expression and/or activation state of a given set of transcripts. URA analysis of the differentially expressed transcripts in MEEs revealed IFNγ and TNFα to be among the top five statistically significant, upregulated, cytokine upstream regulators (Fig. 7B; Supplementary Table S8). Furthermore, STAT-1, a key signaling molecule of IFNγ signaling, was among the top 5 upregulated upstream transcription regulators (Fig. 7C; Supplementary Table S8). In addition, TP53 that was previously shown to reduce expression of M2 transcripts in macrophages (48), was identified as an upstream regulator as well (Fig. 7C).

Proteomic analysis of MEEs revealed that 137 proteins were upregulated (Fig. 7D) and 85 proteins were downregulated (Fig. 7E) in comparison with naïve lung eosinophils (FDR < 0.1, S0 > 0.1, Supplementary Table S9). The upregulated protein–protein interaction network (Fig. 7D) identified that proteins, which are involved in response to IFNγ, including STAT1, guanylate-binding protein 2 (GBP2), interferon gamma-induced GTPase (IGTP), were enriched. As well as B2M and transporter 2 ATP-binding cassette subfamily B member (TAP2) and TAP-binding protein (TAPBP). Additional proteins, which were previously shown to be upregulated in tumor-associated eosinophils such as S100A9 (42) or increased in recruited MEEs such as SIGLEC-F (Fig. 2D), were also identified. The protein network, which was downregulated in MEEs, shows elevation of cell movement and adhesion proteins (Fig. 7E). This network featured downregulation of several cell surface receptors such as CD84, CD36, and carcinoembryonic antigen-related cell adhesion molecule 1 (CEACAM1; Fig. 7E). In addition, MEESs displayed downregulation of arachidonate 15-lipoxygenase (ALOX15) that is involved in arachidonic acid metabolism. ANXA2 and ANXA3 were downregulated in MEEs relative to naïve lung eosinophils (Fig. 7E). Analysis of the concordance between the DE in RNA-sequencing and proteomics (Supplementary Fig. S7G) revealed that MEEs displayed increased expression of genes related to IFNγ response, including GBP2 and IGTP and genes, involved in antigen presentation such as B2M and endoplasmic reticulum aminopeptidase 1 (ERAP1).

Collectively, our bioinformatics analyses highlight TNFα and IFNγ as key regulators in the activities of MEEs.

The suggested role for IFNγ-activated eosinophils in the metastatic lung and our previous data on IFNγ-activated eosinophils in colorectal cancer (42), prompted us to identify the cellular source of IFNγ in the metastatic lung. To this end, IFNγ reporter mice (GREAT mice; ref. 49) were injected with PyMT cells and IFNγ production in lung leukocytes was determined (Fig. 8A). CD4⁺ T cells, NK, and NKT cells were identified as the major IFNγ-producing cells.

To further determine whether the TME entrains eosinophils to adopt an antitumorigenic phenotype via TNFα/IFNγ, we were interested to define the cytokine profile secreted by IFNγ/TNFα-activated eosinophils. Purified eosinophils were stimulated with TNFα and IFNγ and the levels of 111 cytokines/chemokines were assessed in the culture media (Fig. 8B; Supplementary Table S10). TNFα/IFNγ-activated eosinophils significantly increased the secretion of T-cell recruiting chemokines CCL5, CXCL9, and CXCL16 (Fig. 8B), CXCL9 and CXCL16 were also upregulated in MEEs (Fig. 6E). Secretion of CXCL9 was further validated and quantitated by ELISA (Supplementary Fig. S8A). In addition, IFNγ-activated eosinophils presented increased STAT-1 phosphorylation (Supplementary Fig. S8B). Co-culture of eosinophils with PyMT cells in vitro, resulted in cytolytic activity of eosinophils (Supplementary Fig. S9C). Interestingly, IFNγ or IFNα activation did not potentiate eosinophil-mediated tumor cell killing (Supplementary Fig. S9D).

The observed secretion of T-cell recruiting chemokines by IFNγ/TNFα-activated eosinophils raised the hypothesis eosinophils can induce T-cell migration. To examine this hypothesis CD8⁺ T cells were isolated from OT- I transgenic mice and their migration toward eosinophil-secreted factors was determined (Fig. 8C). The culture media of IFNγ/TNFα-activated eosinophils were able to induce T-cell migration in vitro (Fig. 8C). Next, we hypothesized that adoptive transfer of eosinophils into PyMT-injected ΔdblGATA mice would be capable of alleviating tumor growth. To properly control for eosinophil activation, eosinophils were stimulated ex vivo with TNFα/IFNγ and intranasally transferred to PyMT-injected ΔdblGATA mice. Adoptive transfer of eosinophils resulted in decreased tumor burden compared with control mice (Fig. 8D–F). Consequently, the transfer of eosinophils was accompanied with increased infiltration of CD8⁺ (Fig. 8G and H) and CD4⁺ T cells (Fig. 8I and J).

Understanding the biology of eosinophils in the TME bears several fundamental questions; for example, what is their role in the TME, and how does the TME instruct them to acquire these functions. Because the lungs are a physiological anatomical habitat for eosinophils, we addressed these queries in human samples and experimental breast cancer–driven lung metastasis. Assessment of human patient biopsies as well as geonomically profiled specimens, demonstrated that eosinophils infiltrate and degranulate in lung metastasis, and that high eosinophil-infiltrating samples are enriched with an activated CD8⁺ T-cell signature. Experimentally, we characterized a CCR3-independent GPCR-dependent pathway involved in eosinophil recruitment to the metastatic lung, which is composed of resident and recruited eosinophils. Subsequently, we show that the TME entrains both eosinophil populations to acquire antitumorigenic activities that are largely dictated by IFNγ and TNFα and dependent on lymphocytes. Consequently, activated eosinophils facilitate the infiltration of T cells and promote antitumor immunity.

Multiple chemotactic factors were previously linked with the accumulation of eosinophils in the TME. The eotaxin–CCR3 pathway, which is responsible for the homing of eosinophils in mucosal tissues under baseline conditions and in allergic inflammation (36), has been implied in their migration in cancer as well. Indeed, the number of tumor-infiltrating eosinophils was correlated with immunohistochemical staining of eotaxins in colorectal cancer (50), Hodgkin lymphoma (51), and oral squamous cell carcinoma (37). Yet, our results show a surprising CCR3/eotaxin-independent recruitment mechanism of eosinophils to the experimental lung TME. Our data further suggest that eosinophils may be attracted to the TME by tumor-secreted factors. The finding that CCR3 deficiency did not abolish eosinophil recruitment to the TME, suggests that additional pathways promote eosinophil accumulation to the TME at least in experimental metastasis using 4T1 cells. Whether CCR3-independent eosinophil recruitment is limited to the experimental model using 4T1 cells or BALB/c mice should be further determined. Our finding that eosinophils migrate toward tumor cell-secreted factors (which do not contain CCL11/CCL24) suggests that this may occur in additional settings. Similarly, it would be important to establish whether increased eosinophilia in the metastatic lung is not due to local proliferation. Because of multiple additional pathways capable of regulating eosinophil dynamics (7), the specific factors that facilitate this phenomenon remain to be explored.

Using genetic and pharmacological depletion strategies, we observed an antitumorigenic function for eosinophils. These findings are in agreement with epidemiological studies showing a positive association between the levels of blood eosinophils with improved breast cancer–specific survival (52), better time to treatment failure (53) and decreased risk for recurrent disease (52). Nonetheless, in our study, eosinophils had no role in the regulation of tumor growth in an experimental model of primary breast cancer. This may be due to the fact that only a modest increase was observed in eosinophil levels in the primary tumor in comparison with breast cancer lung metastasis. In support of these findings, using an RNA profiling algorithm followed by eosinophil-specific IHC we recently showed that primary breast tumors had fewer infiltrating eosinophils compared with other anatomical sites and especially in comparison with mucosal sites such as the lungs (4). Alternatively, the different roles of eosinophils in primary versus metastatic tumors can result from differential environmental cues that can shape the activity of eosinophils in a given environment or organ-specific molecular and cellular mechanisms of breast cancer cells. In favor of the latter notion, differential activities were observed for neutrophils in primary versus metastatic breast tumors, which were largely dictated by the different TMEs (54). To assess this hypothesis, comparative studies should examine the phenotype of eosinophils in a primary versus metastatic tumors or in different tumors in the same metastatic site.

Previous studies have shown that following induction of experimental asthma, two populations of eosinophils exist in the lungs, namely resident and recruited eosinophils. These eosinophil populations were shown to be functionally distinct because the resident population displayed regulatory activities whereas the recruited population promoted type 2 inflammation. These data suggest that in asthma, the eosinophil subset is the major factor that governs eosinophil activities as opposed to the inflammatory microenvironment. In sharp contrast, our bioinformatic approach of RNA-seq and proteomics established that the TME (and not the eosinophil subset per se) is the major factor, which regulates eosinophil activities and characteristics. Our RNA-seq and proteomic approaches identified markedly different numbers of transcripts/proteins that were upregulated in eosinophils (2,280 vs. 137, respectively). This is likely due to the differential coverage in the RNA-seq versus mass spectrometry methods that were used. For mass spectrometry, one mouse was used per sample with 10,000 sorted eosinophils whereas in the RNA-seq experiments, pooling of mice was used to reach approximately 200,000 eosinophils per sample. Despite the methodological differences, our data demonstrate that the TME instructs similar transcriptional and proteomic networks in eosinophils in settings of breast cancer–driven lung metastasis. Though resident and recruited eosinophils in the TME display similar phenotypes, we cannot exclude the possibility that additional eosinophil populations that display functional heterogeneity exists. Our analyses further identified IFNγ and TNFα as key upstream regulators that dictate the activation status of eosinophils and entrains them to display antitumorigenic activities. Our data further indicate that STAT-1 is a major signaling pathway regulating eosinophil activation in the TME. IFNγ and TNFα were formerly linked to the antitumorigenic activities of eosinophils in diverse TMEs (42, 40), and have been previously shown to act synergistically to enhance secretion of CXCL9 and CXCL10 from human eosinophils (55). Our data corroborate these previous findings and further demonstrate that IFNγ/TNFα-treated eosinophils are capable of facilitating CD8⁺ and CD4⁺ T-cell recruitment into the metastatic lung and subsequent antitumorigenic activities. The involvement of IFNγ in the antitumorigenic activities of eosinophils is of specific interest because we have recently shown that IFNγ plays a key role in tumor-associated eosinophil activation in colorectal cancer (42). Despite this similarity, it appears that IFNγ can induce differential effects on eosinophils in cancer. In lung MEEs, IFNγ likely initiated tumor rejection via a T-cell–mediated mechanism whereas in colorectal cancer, the antitumorigenic activities of eosinophils were independent of CD8⁺ T cells and likely mediated via eosinophil-driven cytotoxicity toward tumor cells (42). Importantly, additional antitumorigenic roles for eosinophils may exist that can be mediated by direct cytolytic activity as demonstrated in vitro. In this scenario, eosinophil-mediated cytotoxicity and subsequent secretion of tumor neoantigens and/or danger signals from tumor cells could be an alternative indirect pathway of eosinophil-mediated T-cell recruitment (56). These findings support the notion that activation state of eosinophils and their subsequent function in a specific tumor is dictated by the environmental cues they receive from the TME.

Accumulating evidence highlights TNFα/IFNγ eosinophil-derived T-cell chemoattractants and subsequent recruitment of T cells as key pathways mediating the antitumor activities of eosinophils. It has been recently shown that, depletion of regulatory T cells (T_reg cells) resulted in eosinophil-dependent tumor rejection due to the ability of eosinophils to attract CD8⁺ T cells via expression of CCL5, CXCL9 and CXCL10. Adoptive transfer of CD8⁺ T cells alone failed to induce tumor rejection, whereas co-transfer of T cells with TNFα and IFNγ-activated eosinophils led to infiltration of T cells and antitumor immunity (40). Furthermore, the proposed crosstalk between activated eosinophils and T cells has gained recent attention especially in cancer immunotherapy following immune checkpoint blockade (ICB; refs. 7, 41). Clinical and experimental data demonstrate that increased blood or tumor eosinophilia is associated with positive prognosis of patients with metastatic melanoma that have been treated with ICB and especially anti–CTLA4 (5). Mechanistically, it has been proposed that eosinophil-mediated the antitumor effects of ICB via IFNγ-dependent vessel normalization (41). Collectively, these data highlight eosinophils as potential cellular targets in cancer immunotherapy.

In summary, our findings reveal the phenotypic landscape of eosinophils in breast cancer–driven lung metastasis. Furthermore, our analyses identify TNFα and IFNγ as key instructing cytokines, which activate eosinophils to acquire lymphocyte-dependent antitumorigenic activities in lung metastasis. These results could be exploited therapeutically by pharmacologically modulating eosinophil activities or adoptive-transfer procedures in combination with ICB for the management of metastatic disease.

A. Munitz reports grants from Israel Science Foundation, Binational Israel US Science Foundation, and grants and personal fees from GlaxoSmithKline, grants from Emerson Collective, Israel Cancer Research Fund, and grants and personal fees from AstraZeneca outside the submitted work. M.E. Rothenberg reports grants from National Institute of Allergy and Infectious Diseases and National Institutes of Health during the conduct of the study; and personal fees and other support from Pulm One, Spoon Guru, ClostraBio, Serpin Pharm, Allakos, and personal fees from Celgene, AstraZeneca, Arena Pharmaceuticals, GlaxoSmithKline, Guidepoint, Suvretta Capital Management, Teva Pharmaceuticals, Mapi Research Trust, and UpToDate outside the submitted work. M.J. Davis reports other support from CTX-One (Now Oncology One) outside the submitted work. No disclosures were reported by the other authors.

S. Grisaru-Tal: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft. S. Dulberg: Software, validation, investigation, methodology, writing–original draft. L. Beck: Validation, investigation, methodology, writing–original draft. C. Zhang: Validation, investigation. M. Itan: Formal analysis, validation, investigation, project administration. S. Hediyeh-zadeh: Formal analysis, investigation, methodology. J. Caldwell: Validation, investigation, methodology. P. Rozenberg: Conceptualization, validation, investigation, methodology. A. Dolitzky: Validation, investigation, methodology. S. Avlas: Validation, investigation, methodology. I. Hazut: Validation, investigation, visualization, methodology. Y. Gordon: Validation, investigation. O. Shani: Investigation, methodology. S. Tsuriel: Validation, visualization, methodology. M. Gerlic: Visualization, writing–original draft. N. Erez: Conceptualization, resources, validation, investigation. N. Jacquelot: Validation, investigation, visualization, methodology. G. Belz: Resources, formal analysis, validation, investigation, methodology, writing–original draft. M.E. Rothenberg: Resources, validation, investigation, visualization, methodology, writing–original draft. M.J. Davis: Software, validation, investigation, visualization, methodology, writing–original draft. H. Yu: Funding acquisition, validation, investigation, visualization, methodology. T. Geiger: Software, validation, investigation, visualization, methodology, writing–original draft. A. Madi: Conceptualization, software, formal analysis, supervision, validation, investigation, visualization, methodology. A. Munitz: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, writing–original draft, project administration.

The authors thank all participants in the current study. The authors would like to thank Dr. Z. Gavish for performing part of the histological work. A. Munitz is supported by the US-Israel Binational Science Foundation (grant no. 2015163), by the Israel Science Foundation (grants no. 886/15 and 542/20), the Israel Cancer Research Fund, the Richard Eimert Research Fund on Solid Tumors (TAU), the Israel Cancer Association Avraham Rotstein Donation, the Cancer Biology Research Center (TAU), the Emerson Collective and GSK Investigator Supported Studies Program.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.