Uncovering common transcriptional features shared by mature peripheral blood PMN-MDSCs and tumor-infiltrating neutrophils in humans

ABSTRACT

Human polymorphonuclear-myeloid-derived suppressor cells (PMN-MDSCs) consist of circulating low-density neutrophils (LDNs) characterized by the ability to inhibit T-cell responses. In previous studies, we demonstrated that the mature fraction of PMN-MDSCs (i.e. mPMN-MDSCs) exerts more potent immunosuppressive functions than its immature counterpart. More recently, we defined a specific gene signature of mPMN-MDSCs from cancer patients and G-CSF-treated donors (GDs) by bulk RNA sequencing (RNA-seq) experiments. In this study, by performing single-cell RNA-seq (scRNA-seq) experiments of circulating mPMN-MDSCs from non-small cell lung cancer (NSCLC) patients, we identified a major scRNA-seq cell cluster (arbitrarily named NSCLC c6) specifically displaying immunosuppressive and protumor transcriptomic features. Then, by analyzing publicly available scRNA-seq datasets from human tumor-associated neutrophils (TANs), we uncovered three TAN clusters (arbitrarily named TAN c6-c8) that were found to share with NSCLC c6 several common genes and transcription factor (TF) regulons associated with response to hypoxia, positive regulation of angiogenesis, and metabolic reprogramming. Furthermore, by performing scRNA-seq experiments of GD mPMN-MDSCs, we uncovered four scRNA-seq cell clusters (arbitrarily named GD c4-c7) that were enriched for the same genes and pathways characterizing NSCLC c6 and TAN c6-c8 cells. Altogether, these data uncover that human circulating mPMN-MDSCs and TANs from different cancer types share scRNA-seq cell clusters with transcriptomic similarities, supporting the notion that they might be strictly related.

Introduction

Low-density neutrophils (LDNs) in humans consist of CD66b⁺ cells that abnormally accumulate within the mononuclear cell fraction after density gradient centrifugation of blood from patients with various pathological conditions, including cancer.^1–3 LDNs able to suppress T cell responses, such as proliferation and/or IFNγ production, are referred to as polymorphonuclear-myeloid-derived suppressor cells (PMN-MDSCs).^4–8 The latter cells have gained great interest in cancer research as their frequency has been shown to correlate with prognosis, tumor stage, and response to treatment.^5,7,9 A significant advance in the field has been the demonstration that the mature fraction of circulating PMN-MDSCs (mPMN-MDSCs) from patients with head and neck cancer (HNC),¹⁰ urological cancer,¹⁰ non-small cell lung cancer (NSCLC),¹¹ and multiple myeloma¹² display much more potent immunosuppressive functions than the immature PMN-MDSCs. In line with this notion, our group had previously reported that also in healthy subjects receiving G-CSF for stem cell mobilization (named GDs), the mature (but not the immature) fractions of both LDNs and normal-density neutrophils (NDNs) display potent immunosuppressive functions,¹³ indicating that they represent reliable cellular models to study the phenotypical and molecular features of mPMN-MDSCs.^13,14 Accordingly, we recently identified a distinct gene signature shared by mature immunosuppressive neutrophils from GDs (i.e., hereafter referred to as GD mPMN-MDSCs) and mPMN-MDSCs isolated from NSCLC and HNC patients by performing bulk RNA sequencing (RNA-seq) experiments.¹⁴ Interestingly, this arbitrarily named “mPMN-MDSC gene signature” contained genes associated not only with immunosuppressive but also with protumor features.¹⁴ However, despite a better understanding of their molecular features through bulk RNA-seq studies, purified mPMN-MDSCs have never been thoroughly characterized at single-cell RNA-seq (scRNA-seq) levels. In fact, most published scRNA-seq studies have focused on human tumor-associated neutrophils (TANs), revealing that they include distinct proinflammatory/antitumor or immunosuppressive/protumor scRNA-seq cell clusters.^15–19 Currently, the most credited hypothesis is that the TAN transcriptomes result from the reprogramming of neutrophils already present within the tumor microenvironment (TME) and not in the blood, as suggested by some mouse and human studies.^16,18–23 However, at least in humans, no definitive consensus exists on such a hypothesis due to the limited analysis performed on circulating neutrophils in these studies.

To gain novel insight into the transcriptional profile of circulating mPMN-MDSCs at a single-cell level, we performed scRNA-seq experiments of mPMN-MDSCs from NSCLC patients. For several reasons, we specifically chose NSCLC mPMN-MDSCs as a reference model. First, NSCLC is one of the most common cancer types and a leading cause of cancer-related mortality.^24,25 Second, PMN-MDSCs have emerged as critical mediators of immunosuppression and resistance to immunotherapy among the key immune regulators in NSCLC.^26–28 Third, we have recently demonstrated that NSCLC mPMN-MDSCs display molecular and phenotypic features common to mPMN-MDSCs from HNC, diffuse large B-cell lymphoma, renal cell carcinoma, transitional cell carcinoma, and breast cancer.¹⁴ Hence, by performing scRNA-seq experiments, we identified a major NSCLC mPMN-MDSC cluster (named NSCLC c6) characterized by immunosuppressive and protumoral transcriptomic features. Then, by analyzing publicly available scRNA-seq data of human TANs from various types of tumor tissues (including NSCLC, colon adenocarcinoma, gallbladder cancer, hepatocellular carcinoma, renal cell carcinoma, oral squamous cell carcinoma, pancreatic adenocarcinoma),^15,17,19 we identified three distinct scRNA-seq cell clusters (named TAN c6, c7, and c8; in short TAN c6-c8), which shared transcriptomic features with NSCLC c6. Such common transcriptional features mainly included activation of the hypoxia signaling pathway, positive regulation of angiogenesis, and metabolic reprogramming. Finally, by performing scRNA-seq experiments of circulating GD mPMN-MDSCs, we identified four discrete cell clusters (namely GD c4, c5, c6, c7; in short, GD c4-c7) that also displayed a significant enrichment of the genes common to NSCLC c6 and TAN c6-c8, as well as deregulations of the hypoxia and metabolic pathways. Taken together, our data suggest that human circulating mPMN-MDSCs from cancer patients and GDs include discrete scRNA-seq cell clusters that share protumor transcriptomic features with TANs, suggesting a potential relationship between these cell populations.

Materials and methods

Study participants

Peripheral blood was collected from: i) NSCLC patients (n = 6; Table 1); ii) GDs, peripheral blood stem cell donors receiving 8–10 μg/kg/day rh G-CSF (Lenogastrim, Italfarmaco, Italy) for 5 days^13,14, (n = 6; Table 1), selected according to the “Italian bone marrow donor registry” criteria, and based on their HLA-compatibility with related or unrelated recipients; iii) sex- and age- matched healthy donors (HDs) (n = 8; Table 1), offering their blood to the Blood Bank Unit of the Verona “Azienda Ospedaliera Universitaria Integrata (AOUI)”. Criteria for NSCLC patients’ selection were: 18 years or older patients; no systemic glucocorticoids and/or immunosuppressive treatment; no active infective and/or autoimmune diseases; no ongoing radio- or chemo- or immunotherapy. See Table 1 for detailed characteristics of HDs, GDs, and NSCLC patients.

Table 1.

Hds, GDs, and NSCLC patient characteristics.

Characteristics		n.	%
HDa Cohort
All samples		8	100
Gender/Age
	Male	5	63
	Female	3	37
	Male mean age	60 ± 12,4
	Female mean age	64 ± 16,3
GDb Cohort
All samples		6	100
Gender/Age
	Male	4	66
	Female	2	34
	Male mean age	56 ± 11,3
	Female mean age	55 ± 19,1
NSCLCc Cohort
All samples		6	100
Gender/Age
	Male	4	66
	Female	2	34
	Male mean age	73 ± 13,6
	Female mean age	64 ± 0,7
Tumor Type
	Adenocarcinoma	4	66
	Squamous cell carcinoma	2	34
Stage
	IIA	1	17
	IVA	1	17
	IIIB	2	33
	IVB	2	33

^aHD, healthy donor; ^bGD, G-CSF-treated healthy donor; ^cNSCLC, non-small cell lung cancer.

Cell isolation

Blood samples from study participants were collected in EDTA- or sodium citrate- treated tubes by venipuncture and processed within 1 h. Mononuclear cells and granulocytes were separated by density gradient centrifugation of blood onto Ficoll-Paque (GE Healthcare) as previously described.^13,14 CD66b⁺ LDNs and CD66b⁺ NDNs were isolated from the mononuclear and granulocyte cell fractions, respectively, by fluorescence-conjugated anti-CD66b mAbs (Table S1) followed by incubation with specific anti-fluorochrome microbeads (Miltenyi Biotec), according to the manufacturer’s protocol.¹³

Single-cell RNA-seq (scRNA-seq)

Total CD66b⁺LDNs and CD66b⁺NDNs were isolated from the peripheral blood of NSCLC patients (n = 6) and included: autologous LDNs and NDNs from 3 patients, LDNs only from 2 patients, and NDNs only from 1 patient. Additionally, autologous LDNs and NDNs were isolated from the peripheral blood of GDs (n = 6), while NDNs only were isolated from the peripheral blood of HDs (n = 8) as controls. Isolated CD66b⁺LDNs and NDNs were processed with BD Rhapsody Single-Cell Analysis System for scRNA-seq as previously described¹⁴ (and Supplemental Materials and Methods). In selected experiments, profiling of surface CD10 expression by AbSeq was run in parallel. Briefly, cells were co-labeled with the BD CD10 AbSeq antibody-oligo (type mouse anti-human, CD10 mAb, sequence ID: AHS005, clone: HI10a, Catalog: 940045), leaving a sample untagged to control the antibody signal background noise.

Statistics

Dataset distribution was evaluated using the Shapiro-Wilk normality test. According to normality test results, the comparison of variables was performed using the unpaired two-tailed Mann-Whitney test or the Wilcoxon matched-pairs signed rank test (for comparison between two groups) and either the Friedman or the Kruskal-Wallis test followed by pairwise either Dunn’s or Wilcoxon’s test (for comparison between three or more groups). P-values lower than 0.05 were considered significant and asterisks indicate significant increases: **** = p-value <0.0001. Graphs were elaborated using R Studio and R environment v4.3 (The R project for Statistical Computing).

Supplemental information

This includes Supplemental Materials and Methods, two Supplemental Tables, seven Supplemental Figures, and one Supplemental Dataset.

Results

Identification of a distinct scRNA-seq cell cluster of NSCLC mPMN-MDSCs that displays immunosuppressive and protumor transcriptomic features

To investigate whether circulating NSCLC mPMN-MDSCs include distinct cell clusters, we performed scRNA-seq experiments on them and, as controls, on NSCLC and HD mNDNs. To reduce cell loss and minimize cell activation,^29–31 all neutrophil populations were isolated by positive selection of CD66b⁺cells by magnetic beads. After normalization and filtering, our datasets included transcriptomes of 14,085 cells from NSCLC patients (Figure S1a) and 7,875 cells from HDs (Figure S1b). Since CD66b⁺LDNs/PMN-MDSCs include both mature and immature neutrophils, we selected the mature neutrophil component from all samples for further analysis. This was done based on the expression levels of a previously defined “maturation gene module” of human mature and immature neutrophils (i.e., r2 gene module of Figure 1 in Grassi et al.’s study).³² Hence, by evaluating the expression of the r2 gene module in the CD66b⁺cells from NSCLC patients and HDs, we selected: (i) 5,291 and 4,258 mPMN-MDSCs and mNDNs, respectively, of NSCLC patients (black gate of Figure S1a); (ii) all 7,875 mNDNs from HDs (black gate of Figure S1b), as expected. To further validate the terminal maturation status of our so-defined neutrophil populations, we profiled surface CD10 expression by CITE-seq in two of the above-described experiments. CD10 is, in fact, a surface antigen expressed on neutrophils only at the latest differentiation stages.^13,33 The Uniform Manifold Approximation and Projection (UMAP) graphs in Figure S1c,d show CD10⁺cells within the merged neutrophil samples from NSCLC patients (black gate of Figure S1c) and HDs (black gate of Figure S1d), which exhibited a significantly higher maturation score than CD10⁻cells in each model (Figure S1e,f). Notably, the frequency of CD10⁺neutrophils measured by CITE-seq technology was very similar to that obtained by flow cytometry analysis prior to the CITE-seq experiments (Table S2). Of note, we also evaluated the expression of MME mRNA (i.e., CD10 encoding gene)^34–37 in our CITE-seq experiments and found that it only partially overlapped with surface CD10 (Figure S1c,d and Figure S1g,h), advising against its use to select mature neutrophils in scRNA-seq datasets. Figure 1a shows the resulting UMAP plots of the mature NSCLC and HD neutrophil populations (identified as shown in Figure S1a-d, back gates), either merged (top left panel) or divided into the groups of origin (top right and bottom panels). UMAP plots confirmed that NSCLC mPMN-MDSCs (Figure 1a, top right panel) distinctly segregated from both autologous mNDNs (Figure 1a, bottom left panel) and HD mNDNs (Figure 1a, bottom right panel), in line with their discrete immunosuppressive properties. By contrast, NSCLC and HD mNDNs exhibited substantial, though not complete, similarity (Figure 1a, bottom left and right panels), likely reflecting their exposure to different microenvironments, with the former being influenced by the presence of a tumor. Next, by performing graph-based clustering analysis by the Louvain algorithm³⁸ of the NSCLC/HD merged scRNA-seq dataset, we identified seven discrete cell clusters (i.e., c0-c6) (Figure 1b) whose distribution across the different NSCLC and HD neutrophil samples is reported in Figure S2a. Among these cell clusters, c5 was found to mainly include NSCLC mNDNs and NSCLC mPMN-MDSCs (>66.2 % and > 33.2 % of the total cells, respectively, therefore named NSCLC c5), while c6 was found primarily attributable to NSCLC mPMN-MDSCs (>83 % of the total cells, therefore named NSCLC c6) (Figure 1c). By contrast, the remaining cell clusters were found to include a mixture of all the different NSCLC and HD neutrophil populations in variable proportions (therefore named HD/NSCLC c0-c4) (Figure 1c). Valuable insights into the biological features of each cell cluster were then provided by the dot plot of Figure 1d, displaying selected upregulated genes (among the top 100, Dataset S1). Regarding the NSCLC/HD cell mixed clusters, NSCLC/HD c0 (which was found to include about 63 % of total HD mNDNs, 23 % of total NSCLC mNDNs, and 13 % of total NSCLC mPMN-MDSCs, Figure 1c), for instance, due to its less distinct pattern of gene expression, recalled similar cell clusters previously described in other scRNA-seq studies of HD neutrophils.^{35,36,39–41} Similarly, HD/NSCLC c1 was found to be characterized by an elevated expression of interferon-stimulated genes (ISGs, such as IFI6, IFIT3, IFIT2, and RSAD2), and thus thought to correspond to the previously described ISG-expressing neutrophil clusters.^{35–37,39–42} In line with the fact that type I IFN signaling promotes proinflammatory/antitumor but restricts immunosuppressive/protumor functions,^43–47 HD/NSCLC c1 was found to include a smaller proportion of NSCLC mPMN-MDSCs (17 % of total cells) than NSCLC or HD mNDNs (38.2 % and 44.8 % of total cells, respectively) (Figures 1c and S2a) and thus of poor interest for our purposes. HD/NSCLC c2 cells (present in all the different HD and NSCLC neutrophil populations in a similar proportions between 5 and 11 % of total cells; Figure S2a) likely represented the previously described neutrophil clusters enriched with CC chemokine ligand (CCL) family genes (CCL4, CCL3 and CCL4L2),^34,35 while HD/NSCLC c4 cells (present in all the different HD and NSCLC neutrophil populations in a similar proportions – about 5–6 % of total cells; Figure S2a) likely corresponded to the cell cluster of neutrophils expressing elevated levels of PI3 and SLPI genes^34,39,42 (Figure 1d). Under our experimental conditions, HD/NSCLC PI3/SLPI-expressing neutrophils were of poor interest as they resembled those typically found in circulating neutrophils from burn or COVID-19 patients,^34,39,42 and not those that expand, for instance, in TANs from cancer patients^15,19 (and this manuscript Figure 2d below). Only in TANs, indeed, PI3/SLPI-expressing neutrophils acquire a protumorigenic/immunosuppressive transcriptomic profile.^15,19 Finally, despite their terminal maturation status, HD/NSCLC c3 cells were found to display an elevated expression of primary granule genes, such as DEFA3, DEFA4, AZU1, ELANE ³⁶ (Figure 1d) but, being very poorly represented (less than 3 % of all the different neutrophil groups, Figure S2a) were disregarded for further analysis.

Figure 1.

Identification of a distinct NSCLC mPMN-MDSC cluster by scRNA-seq analysis. (a) UMAP visualization of scRNA-seq profiles of mature NSCLC and HD neutrophil populations, either merged (top left panel) or divided into groups of origin (top right and bottom panels): NSCLC mPMN-MDSCs (n = 5; top right panel), NSCLC mNdns (n = 4; bottom left panel) and HD mNNDs (n = 8; bottom right panel). (b) scRNA-seq cell clusters (c0-c6) of the UMAP shown in (a) Identified by graph-based clustering analysis by the Louvain algorithm.³⁸ (c) Stacked bar graph showing the relative abundance of HD mNDNs, NCSLC mNDNs, and NCSLC mPMN-MDSCs in the c0-c6 cells. (d) Dot plot showing the average mRNA expression levels of selected upregulated DEGs (among the top 100) for each cluster (c0-c6) (Bonferroni-corrected p < 0.05; two-sided wilcoxon rank sum test); the dot size depends on the percentage of cells expressing that specific gene.

Figure 2.

Analysis of published human pan-cancer TAN scRNA-seq datasets. (a–c) UMAPs derived from the integration of the TAN and NAN datasets from the studies by Salcher S. et al.,¹⁵ Zilionis R. et al.,¹⁹ and Wu Y. et al.,¹⁷ displaying: (a) The type of tumor tissue; (b) The maturity score calculated as in Figure S1; (c) The cell clusters (c0-c8) identified by graph-based clustering analysis by the Louvain algorithm.³⁸ (d-i) violin (d,f,h,i) or UMAP (e, g) plots showing the NAN (d,e,h) and the TAN (f,g,i) “signature scores” across the NAN (d,e) and TAN (f, g) single-cell transcriptomes, or across the NAN (c0-c3) (h) and TAN (c6-c8) (i) specific clusters, calculated as described in Supplemental material and Methods. ****p < 0.0001, by Kruskal-Wallis. (j) dot plot showing the average mRNA expression levels of selected upregulated DEGs (among the top 100) for each cluster (c0-c8) (Bonferroni-corrected p < 0.05; two-sided wilcoxon rank sum test); the dot size depends on the percentage of cells expressing that specific gene.

NSCLC c5, one of the two clusters specifically including NSCLC cell populations (in fact, representing 9 and 22 % of total NSCLC mPMN-MDSCs and NSCLC mNDNs, respectively, and only 0,2 % of total HD mNDNs; Figure S2a), was found to exhibit elevated expression of genes such as IL2R, FKBP5, and RNF144B (Figure 1d). However, NSCLC mPMN-MDSC representation within NSCLC c5 varied markedly across patient samples, as these cells constituted a substantial proportion (37 % of total cells) in a single NSCLC patient, while they were sparsely represented (<1.3 %) in the remaining samples (Figure S2b). We therefore focused on NSCLC c6, since it represented the other larger cell cluster containing NSCLC mPMN-MDSCs (about 50 % of total cells with a relatively consistent proportion across different NSCLC patient samples; Figure S2a,b) and was found to express elevated levels of PLAU, PLAUR, PFKFB3, CST7, CD83, VEGFA, HIF1A, CD44, and C15orf48 mRNAs (Figure 1d). Notably, most of these genes were shown to be expressed by previously described scRNA-seq cell clusters of TANs characterized by immunosuppressive/protumor transcriptomic features (i.e., hN5,¹⁹ TAN-1,¹⁶ TAN-1/TAN-2,¹⁵ VEGFA⁺SPP1⁺/NFKBIZ⁺HIF1A⁺¹⁷). NSCLC c6 was therefore pointed out as the major candidate of the NSCLC neutrophils displaying immunosuppressive/protumor features. Consistent with such a hypothesis, we found that our “mPMN-MDSC gene signature”¹⁴ was -significantly enriched in the transcriptome of total NSCLC mPMN-MDSCs as compared to HD and NSCLC mNDNs (Figure S2c, top panel), as expected, but such enrichment was found to occur only in the NSCLC c6 (Figure S2c, bottom panel). Moreover, by Gene Ontology (GO) term enrichment analysis, we found that the differentially expressed genes (DEGs) of NSCLC c6 cells were associated with several biological processes typically characterizing myeloid cells in cancer, such as metabolic processes (i.e, “glycolytic process”, “pyruvate metabolic process”), “response to oxidative stress and hypoxia”, “angiogenesis”, “negative regulation of cytokine production” or “cellular response to lipid” (Figure S2d).^48-56 Notably, only the latter two GO-terms resulted enriched also in NSCLC c5 cells, which, on the other hand, resulted more enriched in GO-terms such as “response to nitrogen compound” and “response to growth factor” (Figure S2d), in line with its lack of enrichment of the mPMN-MDSC gene signature (Figure S2c, bottom panel). These findings ultimately indicate that NSCLC c5 was poorly representative of immunosuppressive/protumor cells. Moreover, the DEGs of HD/NSCLC c1-c4 cells were mainly enriched in GO-terms such as “response to the virus” and “leukocyte migration” (Figure S2d), thus confirming that they were not representative of the immunosuppressive features of NSCLC mPMN-MDSCs.

In sum, these scRNA-seq experiments uncovered that a major cell cluster within circulating NSCLC mPMN-MDSCs, namely NSCLC c6 (including 50 % of its total cells), transcriptionally recapitulates their immunosuppressive/protumor features.

NSCLC c6 shares transcriptomic similarities and transcriptional regulatory networks with discrete clusters of TANs from 14 cancer types

Since circulating NSCLC c6 cells were found to share several immunosuppressive/protumor genes (Figure 1d) with previously reported protumor/immunosuppressive TAN clusters,^15–17,19 we decided to investigate the transcriptomic similarities among these cell populations further. To broaden our findings to various tumor types, we merged two published scRNA-seq TAN dataset from lung cancer patients only^15,19 with a recently published human pan-cancer scRNA-seq TAN dataset obtained from tissue biopsies of primary tumors, normal adjacent tissue, and metastases of 14 different cancer types (including NSCLC, colon adenocarcinoma, gallbladder cancer, hepatocellular carcinoma, renal cell carcinoma, oral squamous cell carcinoma, pancreatic adenocarcinoma)¹⁷ (Figure 2a). We found that all cells in this integrated pan-cancer scRNA-seq TAN dataset display a mature phenotype, as revealed by the remarkable expression of the r2 gene module defined by Grassi et al.³² (Figure 2b), and then identified nine discrete cell clusters (c0-c8), as revealed by graph-based clustering analysis by the Louvain algorithm³⁸ (Figure 2c). As a consequence of the integration of the scRNA-seq datasets from the studies by Salcher S. et al.¹⁵ and Zilionis R. et al,¹⁹ the nine scRNA-seq cell clusters identified in the current study substantially, but not completely, overlapped with the ten cell clusters identified in the original study by Wu Y. et al.¹⁷ Normal adjacent tissue-associated neutrophils (NANs; Figure 2d,e) and TANs (Figure 2f,g) were then identified based on the expression levels of recently published NAN and TAN signatures,¹⁵ whose enrichments across the different cell clusters revealed that: i) c0-c3 cells were significantly enriched of the NAN signature (therefore named NAN c0-c3; Figure 2h); ii) c6-c8 were significantly enriched of the TAN signature (TAN c6-c8; Figure 2i); iii) c4 and c5 were modestly, but not significantly enriched of both signatures, indicating a mixed composition of NAN and TAN cells (NAN/TAN c4-c5; Figure 2h,i). As shown in Figure 2j (reporting selected upregulated genes among the top 100, Dataset S1), NAN c0-c2 cells were enriched with genes such as S100A9, S100A4, PTGS2, and SELL, which recalled the previously described mature NAN clusters without a defined terminal state,⁵⁷ while NAN c3 cells were instead enriched with ISGs, and recalled the previously reported interferon-responsive NAN clusters with antitumor features.^15–19,57 NAN/TAN c4 were enriched with CCL family genes,^15,18,19 while NAN/TAN c5 corresponded to the previously described HLA-DR⁺CD74⁺cell cluster with antigen-presenting cell/antitumor features.^15,17 All the previous cell clusters were, therefore, of scarce interest in this context. On the other hand, the TAN c6-c8 cells caught our interest since they were collectively enriched, albeit at variable expression levels, with genes such as BHLHE40, VEGFA, CD83, C15orf48, OLR1, HIF1A, LDHA (Figure 2j), that were previously associated with human scRNA-seq TAN clusters with immunosuppressive/protumor features.^15–17,19 Notably, the selected enriched genes of the TAN c6-c8 cells (Figure 2j) were found to substantially overlap with those of NSCLC c6 cells (Figure 1d). In line with this observation, our “mPMN-MDSC gene signature”¹⁴ resulted significantly enriched in the integrated pan-cancer scRNA-seq TAN dataset (Figure S3a), more specifically in TAN c6-c8 cells (Figure S3b,c). Similarly, the upregulated DEGs of the TAN c6-c8 resulted significantly enriched in the single-cell transcriptome of NSCLC mPMN-MDSCs (Figure S3d), specifically in NSCLC c6 cells (Figure S3e,f), therefore evidencing transcriptional similarities between TAN c6-c8 and NSCLC c6 cells.

To identify the genes responsible for the transcriptional similarities between TAN c6-c8 cells and NSCLC c6 cells, we compared all their upregulated DEGs. By doing so, we extrapolated 60 commonly upregulated genes (Figure 3a,b and Dataset S1), among which we found several genes associated with processes such as “wound healing” (i.e., VEGFA, IL1B, NINJ1, CD44, FCER1G, CHMP4B, PLAU and HIF1A), “response to hypoxia” (i.e., HIF1A, PLAU, VEGFA, FOSL2, ENO1, PLK3), “glycolysis” (i.e., PKM, GAPDH, ENO1, IER3, HIF1A, and LDHA), “positive regulation of angiogenesis” (i.e., HIF1A, NFE2L2, SP1, ATF2, XBP1, GATA6, TWIST1) (Figure 3b,c). Interestingly, as illustrated in Figure S3g, NSCLC c6 and TAN c6-c8 were found to also share nearly half of the top 15 GO-terms enriched in their specific, non-overlapping DEGs. These shared terms were, once again, predominantly related to immunoregulatory functions, such as “negative regulation of immune effector process”, “negative regulation of responses to external stimulus”, “leukocyte mediated immunity”, “negative regulation of cytokine production”, “negative regulation of immune response”. This suggests that NSCLC c6 and TAN c6-c8 transcriptomes, while comprising only 60 common genes, substantially converge on similar regulatory and functional pathways.

Figure 3.

Genes commonly expressed by NSCLC c6 and TAN c6-c8 cells. (a) Venn diagram highlighting (light orange circle) that 60 are the genes shared between the NSCLC c6 (green circle) and the TAN c6-c8 (violet circle) clusters. (b) Heatmap showing the average expression [ln(UMI)] of the 60 upregulated genes shared by NSCLC c6 and TAN c6-c8 across the HD/NSCLC (left panel) or the NAN/TAN (right panel) scRNA-seq datasets. Selected genes are listed on the right of the heatmap. (c) Graph depicting the GO terms significantly (false discovery rate [FDR] < 0.05) over-represented in the NSCLC c6 and TAN c6-c8 common genes (a,b). Orange bars indicate the number of genes composing the enriched GO terms.

To validate these latter findings, we also investigated whether NSCLC c6 and TAN c6-c8 cells share commonly active transcriptional regulatory networks by using CollecTRI, a meta-resource that compiles transcription factor (TF)-gene information to estimate TF regulon activities.⁵⁸ By applying a univariate linear model, we identified 249 and 243 TF regulons differentially active in NSCLC c6 and TAN c6-c8 cells, respectively, compared to their respective controls (i.e., NSCLC/HD c0-c5 versus NSCLC c6; TAN/NAN c0-c5 versus TAN c6-c8), of which 123 (i.e., nearly half of the total for both cell types) were shared (Figure 4a and Dataset S1). These common TF regulons (Figure 4b showing the top 30) were associated with GO-terms related to biological processes such as “hypoxia” (i.e., HIF-1α, NFE2L2, EGR1, ATF2, ARNT, PPARA, NOTCH1), “oxidative stress” (i.e., FOS, JUN, HIF-1α, NFE2L2, SP1, EPAS1, ATF2, ARNT), “regulation of inflammatory responses” (i.e., PPARD, PARK7, PPARA, SMAD3, and STAT5B), “response to endoplasmic reticulum stress” (i.e., JUN, NFE2L2, ATF3, PARK7, DDIT3, XBP1, and ATF6), “regulation of angiogenesis” (i.e., HIF-1α, NFE2L2, SP1, ATF2, XBP1, GATA6, and TWIST1), “regulation of lipid metabolic process” (i.e., EGR1, PPARD, SNAI1, CREB1, PPARA, KAT5, STAT5B, and TWIST1) and “glycolytic process” (i.e., HIF-1α, ARNT, EP300, and PPARA) (Figure 4c). Similar findings were obtained by utilizing VIPER, a statistical framework that employs a three-tailed enrichment score performed by the aREA algorithm to infer TF regulon activities.⁵⁹ By using this tool, we found 91 TF regulons (Figure S3h) commonly activated in NSCLC c6 and TAN c6-c8, which resulted associated with GO-terms related to biological processes highly similar to those identified by using the univariate linear model (Figure S3i).

Figure 4.

Transcription factor (TF) regulons commonly activated in NSCLC c6 and TAN c6-c8 cells. (a) Venn diagram showing the TF regulons differently active in NSCLC c6 (green circle) and TAN c6-c8 (violet circle). The intersection in light orange highlights the 123 TF regulons commonly active in NSCLC c6 and TAN c6-c8 cells. (b) Heatmap showing the average activity score of the top 30 shared TF regulons (a) across the HD/NSCLC (left panel) or the NAN/TAN (right panel) scRNA-seq datasets. Selected TFs are listed on the right of the heatmap. (c) Graph depicting the GO terms significantly (false discovery rate [FDR] < 0.05) over-represented in NSCLC c6 and TAN c6-c8 shared TF regulons. Orange bars indicate the number of TF regulons composing the most significant GO terms.

These findings demonstrate that circulating NSCLC mPMN-MDSCs contain a major cell cluster that shares transcriptomic features and transcriptional regulatory networks with protumor/immunosuppressive TAN clusters from various tumor types.

NSCLC c6 and TAN c6-c8 cells display activation of the hypoxia signaling pathway and metabolic reprogramming

To further validate the previous findings, we utilized Signaling Pathway RespOnsive GENes (PROGENy), a software that uses consensus gene signatures to measure signaling pathway activities⁶⁰ and confirmed that the hypoxia signaling pathway results significantly activated in both NSCLC mPMN-MDSCs (Figure 5a) [more specifically in NSCLC c6 (Figure 5b,c) cells] and TANs (Figure 5d) [more specifically in TAN c6-c8 cells (Figure 5e,f)].

Figure 5.

NSCLC c6 and TAN c6-c8 cells display activation of hypoxia signaling pathway and metabolic reprogramming. (a–f) Violin (a,c,d,f) or UMAP (b,e) plots showing the activity scores of the hypoxia pathway across the single-cell transcriptome of HD/NSCLC neutrophils (a,b) and NANs/TANs (d,e) or across their specific clusters (HD/NSCLC c0-c6, c; NAN/TAN c0-c8, f), calculated as described in Supplemental material and Methods. ****p < 0.0001, by Kruskal-Wallis test followed by pairwise Wilcoxon test. (g-h) graphs show the activity of metabolic pathways such as glycolysis (left panels) and arginine and proline metabolism (right panels) in NSCLC c6 as compared to HD/NSCLC c0-c5 (g) or TANs c6-c8 as compared to TAN/NAN c0-c5 (h) in terms of Cohen’s d. Each dot is a reaction of the metabolic pathways, and the main reaction is labeled with the enzymes catalyzing the reaction. Significant pathways p < 0.05, by pairwise Wilcoxon test followed by Benjamini Hochberg correction procedure.

Besides the activation of the hypoxia signaling pathway, several GO-terms related to metabolic reprogramming (namely “glycolitic process”, “pyruvate metabolic process”, “regulation of lipid metabolic process” and “lactate metabolic process”) were also associated with DEGs (Figure 3c) and TF regulons (Figure 4c) shared by NSCLC c6 and TAN c6-c8 cells. Therefore, we applied the Compass algorithm⁶¹ to our NSCLC/HD and TAN/NAN scRNA-seq datasets to infer the metabolic status of NSCLC c6 and TAN c6-c8 cells and, in turn, to detect their most active metabolic pathways. By doing so, NSCLC c6 (Figure S4a) and TAN c6-c8 (Figure S5a) cells were found to display a significantly increased metabolic activity compared to the other NSCLC/HD and NAN/TAN clusters, respectively. Specifically, metabolic pathways such as glycolysis, fatty acid oxidation, and arginine and proline metabolism, which are known to characterize myeloid cells in cancer,^54,62–66 resulted to be all significantly more active in both NSCLC c6 (Figure 5g and S4a) and TAN c6-c8 (Figure 5h and S5a) cells than in the other NSCLC/HD and NAN/TAN cell clusters. Similar findings were obtained by utilizing the MetaFLUX algorithm, another bioinformatic tool utilized to infer the metabolic status of the cell populations of interest (Figure S4b and S5b).⁶⁷

These findings suggest that activation of the hypoxia signaling pathway and metabolic reprogramming are among the main transcriptomic features shared by NSCLC c6 cells and TAN c6-c8.

GD mPMN-MDSCs include distinct scRNA-seq cell clusters displaying activation of the hypoxia signaling pathway and metabolic reprogramming

We then performed scRNA-seq experiments of GD mPMN-MDSCs to verify whether our findings extend to other models of circulating mPMN-MDSCs. By using the same approach utilized for NSCLC mPMN-MDSCs (see Figure S1a and S1c), we evaluated the expression of the r2 maturation gene module³² (Figure S6a, left panel) and profiled surface CD10 expression by CITE-seq (Figure S6a, right panel) to define the mature component of GD mPMN-MDSCs. We obtained 12,844 GD mPMN-MDSCs in total (Figure S6a) and then merged them with HD mNDNs (Figure S1b) to perform data analysis. Figure 6a shows the resulting UMAP plots of the mature GD and HD neutrophil populations, either merged (top left panel) or divided into the groups of origin (top right and bottom panels). Graph-based cell clustering analysis by the Louvain algorithm³⁸ identified eight cell clusters (Figure 6b) that were distributed across the different GD and HD samples, as displayed in Figure S6b. Among these cell clusters: i) c0 and c1 mainly consisted of HD mNDNs (>80 % of the total cells; therefore, named HD c0-c1); ii) c4, c5, c6 and c7 mostly included GD mPMN-MDSCs (>90 % of the total cells; therefore, named GD c4-c7); iii) c2 and c3 included a mixture of all the different GD and HD neutrophil populations in substantially similar proportions (therefore named HD/GD c2-c3) (Figure 6c). According to the dot plot of Figure 6d, reporting selected upregulated genes for each cell cluster (among the top 100, Dataset S1), HD c0 cells (displaying a less distinct gene expression pattern), HD c1 cells (expressing elevated levels of ISGs), HD/GD c2 cells (expressing elevated levels of CCL ligands), and HD/GD c3 cells (expressing elevated levels of primary granule genes), displayed transcriptomic similarities with the HD/NSCLC c0-c3 cells identified in the scRNA-seq experiments of NSCLC and HD samples (Figure 1d) and were therefore of poor interest for our purposes. Notably, similarly to NSCLC mPMN-MDSCs (Figure 1c and S2a), also GD mPMN-MDSCs were very poorly represented (13.7 % of the total cells) in the ISG-expressing neutrophil cluster (HD c1 cells) (Figure 6c and S6b), while HD/GD c3 cells were disregarded for further analysis because they were poorly represented (less than 2 % of the total cells) within both HD and GD neutrophils (Figure S6b).

Figure 6.

Identification of four distinct GD mPMN-MDSC clusters by scRNA-seq analysis. (a) UMAP visualization of scRNA-seq profiles of mature GD and HD neutrophil populations, either merged (top left panel) or divided into the groups of origin (top right and bottom panels): HD mNdns (n = 8; top right panel) and GD mPMN-MDSCs (n = 6; bottom panel). (b) IAdentification of cell clusters (c0-c7) in the UMAP shown in (a) by graph-based clustering analysis by the Louvain algorithm.³⁸ (c) Stacked bar graph showing the relative abundances of HD mNdns and GD mPMN-MDSCs in c0-c7 clusters. (d) Dot plot showing the average mRNA expression levels of selected upregulated DEGs (among the top 100) for each cluster HD/GD cluster (c0-c7) (Bonferroni-corrected p < 0.05; two-sided wilcoxon rank sum test); the dot size depends on the percentage of cells expressing that specific gene.

On the other hand, the clusters enriched of GD mPMN-MDSCs (precisely, GD c4-c7; Figure 6c), collectively included, albeit at variable expression levels, genes such as PLAUR, IL1B, CD83, CXC3CR1, BHLHE40, EGR1, PI3, SLPI, CST7, CD177, CD44, PLAC8, LDHA, and PFKFB3 mRNAs (Figure 6d). Notably, most of these DEGs, including PLAUR, CD83, BHLHE40, EGR1, CST7, CD44, LDHA and PFKFB3 were also expressed by NSCLC c6 (Figure 1d) and TAN c6-c8 (Figure 2j). Hence, GD c4-c7, which altogether represented more than 80 % of total GD mPMN-MDSCs (with a relatively consistent proportion across different GD samples; Figure S6b,c), displayed features reminiscent of the immunosuppressive/protumor features common to NSCLC c6 and TAN c6-c8. Consistent with such a hypothesis, not only our “mPMN-MDSC gene signature”¹⁴ (Figure S6d), but also the DEGs of NSCLC c6 (Figure S6e) and of TAN c6-c8 (Figure S6f) were found significantly enriched in GD c4-c7 cells.

The finding that GD mPMN-MDSCs (Figure 7a), and more specifically GD c4-c7 (Figure 7b,c), resulted significantly enriched with the 60 genes commonly expressed by NCSCL c6 and TAN c6-c8 further confirmed the existence of transcriptomic similarities between GD c4-c7 cells, NSCLC c6, and TAN c6-c8. We also observed that GD mPMN-MDSCs, and more specifically GD c4-c7, displayed a remarkable activation of the hypoxia signaling pathway (Figure 7d–f), as well as a significantly increased metabolic activity (Figure S7a,b), with glycolysis, fatty acid oxidation, and arginine and proline metabolism being among the most active metabolic pathways (Figure 7g and S7a,b). These findings suggest that circulating mPMN-MDSCs from cancer patients and GDs are characterized by specific scRNA-seq cell clusters displaying immunosuppressive and protumor transcriptomic features in common with immunosuppressive/protumor TAN clusters.

Figure 7.

GD c4-c7 cells are enriched of the genes common to NSCLC c6 and TAN c6-c8 and display activation of hypoxia signaling pathways and metabolic reprogramming.

(a-f) Violin (a,c,d,f) or UMAP (b,e) plots showing the score of the enrichment of the 60 genes common to NSCLC c6 and TAN c6-c8 (a-c) or of the hypoxia pathway activity (d-f) across the single-cell transcriptome of HD/GD neutrophils (a,b,d,e) or across their specific clusters (c0-c7; C,F), calculated as described in Supplemental Materials and Methods. ****p < 0.0001, by Kruskal-Wallis test followed by pairwise Wilcoxon test. (g) graphs show the activity of metabolic pathways such as glycolysis (left panels) and arginine and proline metabolism (right panels) in GD c4-c7 as compared to HD/GD c0-c2 in terms of Cohen’s d. Each dot is a reaction of the metabolic pathways, and the main reaction is labeled with the enzymes catalyzing the reaction. Significant pathways p < 0.05, by pairwise Wilcoxon test followed by Benjamini Hochberg correction procedure.

Discussion

Our previous studies demonstrated that circulating human mature PMN-MDSCs (mPMN-MDSCs) display much more potent immunosuppressive properties than their immature counterparts.^10,13 Herein, we performed scRNA-seq experiments of: i) mPMN-MDSCs and mature NDNs (mNDNs) from patients with NSCLC (chosen as a reference cancer model); ii) mPMN-MDSCs (from both the LDN and NDN fractions) from healthy subjects receiving G-CSF for stem cell mobilization (named GDs); iii) mNDNs from healthy donors (HDs). Then, we took advantage of publicly available scRNA-seq TAN datasets from 14 different tumor types, including NSCLC,^15,17,19 to determine whether mPMN-MDSCs and TANs contain discrete cell clusters with similar transcriptional features. Our bioinformatic analysis suggested that human TANs exhibit transcriptomic features of mature neutrophils, ensuring that in our mPMN-MDSC versus TAN comparison, we were utilizing cells at comparable maturation levels. By doing so, we identified seven distinct cell clusters in circulating mPMN-MDSCs from NSCLC patients, one of them found specifically enriched with immunosuppressive and protumor genes (i.e, NSCLC c6) and representing 50 % of the total cells. We could then identify nine distinct pan-cancer TAN clusters, three of them specifically enriched with protumor and immunosuppressive genes (named TAN c6-c8). By comparing the DEGs between NSCLC c6 and TAN c6-c8, we could subsequently identify 60 commonly expressed genes, including VEGFA, IL1B, PLAU, HIF1A, ENO1, PKM, LDHASP1, ATF2, XBP1, GATA6, TWIST1, as well as 123 commonly active TF regulons. Most of them were associated with biological processes, such as “responses to hypoxia and oxidative stress”, “positive regulation of angiogenesis”, and “altered cell metabolism” (namely “glycolytic process”, “pyruvate metabolic process”, “regulation of lipid metabolic process” and “lactate metabolic process”). The fact that TAN c6-c8 were identified in different tumor types suggests that this transcriptional program is conserved across multiple tumor types. Nonetheless, when we analyzed scRNA-seq experiments of circulating GD mPMN-MDSCs, we identified eight distinct cell clusters, four of which resulted significantly enriched of the 60 genes commonly expressed by NSCLC c6 and TAN c6-c8 (named GD c4-c7, representing 80 % of the total cells), and displayed activation of the hypoxia signaling pathway and metabolic reprogramming. These data support the notion that circulating mPMN-MDSCs share distinct transcriptional features with immunosuppressive/protumor TAN clusters and, as also highlighted in a recent review by Eruslanov E. et al.,⁶⁸ are in line with our previous findings on the enrichment of the mPMN-MDSC signature in scRNA-seq TAN datasets from lung cancer patients.¹⁴

Recent scRNA-seq studies of neutrophils from cancer patients concluded that they do not exhibit noticeable transcriptomic reprogramming while in circulation.^16,18,19 However, it is crucial to note that, while in our study mPMN-MDSCs were isolated from LDNs using blood density gradient centrifugation, all previous scRNA-seq studies included the entire population of circulating CD66b⁺ CD15⁺ neutrophils, leading to transcriptomic profiles that likely underestimate the contributions from the small fraction of mature PMN-MDSCs.^16,18,19 On top of this, most scRNA-seq studies have integrated, within the same dataset, circulating neutrophils and autologous TANs from cancer patients with circulating neutrophils from HDs, an approach that favors the detection of gene expression differences driven by tissue origin rather than by the intrinsic transcriptomic features of the distinct cell populations.^16,18,19,69 By contrast, we initially compared each neutrophil population to its appropriate internal control (e.g., autologous mNDNs and HD mNDNs versus NSCLC mPMN-MDSCs; circulating HD mNDNs versus GD mPMN-MDSCs; NANs versus TANs) to first identify specific NSCLC mPMN-MDSC, GD mPMN-MDSC and TAN clusters, and then to analyze their specific transcriptomic similarities. Overall, these approaches enabled us to detect more subtle transcriptomic features of the relatively rare subset of mPMN-MDSCs.

It is noteworthy to mention that, in line with our findings, those scRNA-seq studies focusing on total circulating neutrophils only have reported transcriptomic differences between circulating neutrophils from cancer patients versus those from HDs.^35,69,70 For example, Veglia F. et al.⁷⁰ identified a scRNA-seq cell cluster (named c0) that included 40 % of NSCLC circulating neutrophils versus 8 % of HD neutrophils, without however investigating whether this cell cluster contained mature or immature neutrophils, or reporting its specific transcriptomic features. Similarly, a recent scRNA-seq study by Marteau V. et al.⁶⁹ reports that total circulating neutrophils from colorectal cancer (CRC) patients display a metabolic rewiring compared to those from HDs, but a detailed cell cluster analysis was not performed.

That mPMN-MDSCs exhibit metabolic alterations, such as glycolysis, arginine and proline metabolism, and fatty acid oxidation, is well documented and thought to be closely linked to their immunosuppressive functions.^54,62–66 However, while immunosuppression is a recognized function of human mPMN-MDSCs,^5,7,71 a T cell-independent direct tumor-promoting activity is less expected. Nonetheless, in mice, neutrophils classified as PMN-MDSCs – which are typically isolated from bone marrow (BM) and spleen, rather than from the blood – were reported to contribute to tumor progression by promoting angiogenesis,⁷² epithelial-to-mesenchymal transition (EMT),⁷³ and pre-metastatic niche formation.⁷⁴ Additionally, a previous study suggested that CD66b⁺PMN-MDSCs from breast cancer patients promote tumor growth in a mouse xenograft model.⁷⁵ Given that several genes and pathways linked to tumor progression and angiogenesis – such as those regulating angiogenesis and EMT – are included in the mPMN-MDSC signature¹⁴ and shared between circulating mPMN-MDSCs and TAN scRNA-seq cell clusters, further research is needed to elucidate the protumor functions of circulating human mPMN-MDSCs.

An additional, apparently unexpected, finding that we report here is that circulating mPMN-MDSCs exhibit evidence of hypoxia signaling activation. Hypoxia is indeed a hallmark of tumors,^52,56 and it is also known to profoundly influence TAN metabolism, polarization, and functions, in turn shaping their immunosuppressive and protumor activities within the TME.^16,21,22,76 Previous studies performed in tumor-bearing mice have suggested a role for hypoxia and HIF-1α in the differentiation of PMN-MDSCs and the modulation of their immunosuppressive functions.^51,77,78 However, these studies mostly referred to immunosuppressive neutrophils infiltrating the tumor tissue and not to circulating cells.^51,77,78 Three recent studies – one conducted in a mouse model of pancreatic ductal adenocarcinoma (PDAC),²¹ one conducted in a mouse brain tumor model²² and the other in PDAC patients¹⁶ - have proposed that it is only within the TME – and not in the circulation, spleen or BM – that the hypoxic niche drives the differentiation of protumor TAN clusters (T3 in study by Ng MSF. et al.,²¹ N3 in the study by Ugolini A. et al.²² and TAN-1 in the study by Wang L. et al.¹⁶ with hyperactivated glycolysis, and immunosuppressive and protumor/proangiogenic transcriptomic features. However, once again, in these studies, neutrophils from different tissue origins (e.g., blood, BM, spleen, and tumor tissue) were integrated into the same scRNA-seq dataset for bioinformatic analysis, therefore potentially masking the specific transcriptomic characteristics of circulating cells due to dominant transcriptional differences dictated by tissue origin.^16,21,22

It remains unclear what may drive hypoxia signaling in mPMN-MDSCs, even in those from GDs who do not have tumors. In our previous study, we speculated that the transcriptomic similarities between mPMN-MDSCs from cancer patients and GD mPMN-MDSCs might depend on the fact that in both conditions an emergency granulopoiesis occurs, causing their mobilization into circulation.¹⁴ During this process, neutrophil progenitors are exposed not only to G-CSF but also to multiple factors that might mimic the tumor microenvironment.¹⁴ It has been previously shown that G-CSF-induced emergency granulopoiesis can create low-oxygen conditions in hematopoietic progenitors within the BM.⁷⁹ Accordingly, mPMN-MDSCs are typically detected in circulation only in advanced-stage cancer patients, when tumor-induced emergency granulopoiesis is activated.^5,80,81 This suggests that, in both cancer patients and GDs, circulating mPMN-MDSCs may retain the hypoxia-driven imprinting and consequent metabolic reprogramming acquired by their progenitors in BM niches under emergency granulopoiesis conditions. Additionally, inflammatory cytokines and tumor-derived mediators are known to activate HIF-1α under oxygen-independent mechanisms, which could also contribute to sustaining hypoxia signaling activation and metabolic rewiring of these cells while in circulation.^82–85

In sum, our study uncovers that circulating mPMN-MDSCs and TANs share previously underestimated transcriptomic similarities. Although deriving from bioinformatic analyses and requiring further validation, these data emphasize the need for a more precise characterization of mPMN-MDSCs and TANs in terms of functional and phenotypical similarities to better define their eventual relationship. In murine tumor models, it is assumed that PMN-MDSCs are directly recruited to the TME during disease progression, increasing their protumor and immunosuppressive capacities and further differentiating into TANs.^5,68,86 A more detailed understanding of the interconnections between mPMN-MDSCs and TANs in humans could provide novel neutrophil-targeted immunomodulatory approaches for cancer treatment.

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Funding Statement

This work was supported by grants from: Associazione Italiana per la Ricerca sul Cancro [AIRC, IG27613] to M.A.C.; Ministero dell’Istruzione, dell’Università e della Ricerca (MIUR) [MIUR-PRIN20227YR8AW to M.A.C., MIUR-PRIN20174T7NXL to F.B and PRIN 2022W839FM to N.T.); Università di Verona [RIBA2022Scapini] to P.S; Associazione Mogli Medici Italiani (AMMI) to B; Ministry of Health, grant no. GR-2016-02361263 to N.T.; AIRC, Next Gen Clinician Scientist 2023 n° 30204 to S.P.; German Research Foundation (DFG) through Collaborative Research Center (TRR) 332 (project A4 to S.B.). This work was also supported by European Cooperation in Science and Technology (COST) Actions BM1404 Mye-EUNITER (https://www.cost.eu/actions/BM1404/.) and CA20117 Mye-InfoBank (www.mye-infobank.eu.); COST is supported by the EU Framework Program Horizon Europe.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

scRNA-seq data included in this study, which were either partially retrieved and modified from our previous study¹⁴ (GEO: GSE250002, subseries GSE250001) or newly generated, have been collectively deposited in the Gene Expression Omnibus database under the accession number GSE293018 and are publicly available as of the date of publication. This paper does not report any original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Ethics approval and consent to participate

The study was conducted according to the Declaration of Helsinki principles and cleared by the Ethics Committee of the Azienda Ospedaliera Universitaria Integrata di Verona (Italy) (CMRI/55742). All patients signed informed consent forms.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/2162402X.2025.2521396

Abbreviations

PMN-MDSCs

polymorphonuclear-myeloid-derived suppressor cells;

RNA-seq

RNA sequencing;

scRNA-seq

single-cell RNA-seq;

UMAP

Uniform Manifold Approximation and Projection;

GDs

G-CSF-treated donors;

NSCLC

non-small cell lung cancer;

HNC

head and neck cancer;

TANs

tumor-associated neutrophils;

TF

transcription factor;

LDNs

low-density neutrophils;

TME

tumor microenvironment;

NDNs

normal-density neutrophils;

HDs

healthy donors;

CCL

CC chemokine ligand;

GO

gene ontology;

DEGs

differentially expressed genes;

NANs

normal adjacent tissue-associated neutrophils;

BM

bone marrow;

EMT

epithelial-to-mesenchymal transition.