Single-cell resolution characterization of myeloid-derived cell states with implication in cancer outcome

Guimarães, Gabriela Rapozo; Maklouf, Giovanna Resk; Teixeira, Cristiane Esteves; de Oliveira Santos, Leandro; Tessarollo, Nayara Gusmão; de Toledo, Nayara Evelin; Serain, Alessandra Freitas; de Lanna, Cristóvão Antunes; Pretti, Marco Antônio; da Cruz, Jéssica Gonçalves Vieira; Falchetti, Marcelo; Dimas, Mylla M.; Filgueiras, Igor Salerno; Cabral-Marques, Otavio; Ramos, Rodrigo Nalio; de Macedo, Fabiane Carvalho; Rodrigues, Fabiana Resende; Bastos, Nina Carrossini; da Silva, Jesse Lopes; Lummertz da Rocha, Edroaldo; Chaves, Cláudia Bessa Pereira; de Melo, Andreia Cristina; Moraes-Vieira, Pedro M. M.; Mori, Marcelo A.; Boroni, Mariana

doi:10.1038/s41467-024-49916-4

Download PDF

Article
Open access
Published: 07 July 2024

Single-cell resolution characterization of myeloid-derived cell states with implication in cancer outcome

Nature Communications volume 15, Article number: 5694 (2024) Cite this article

7674 Accesses
6 Altmetric
Metrics details

Subjects

Abstract

Tumor-associated myeloid-derived cells (MDCs) significantly impact cancer prognosis and treatment responses due to their remarkable plasticity and tumorigenic behaviors. Here, we integrate single-cell RNA-sequencing data from different cancer types, identifying 29 MDC subpopulations within the tumor microenvironment. Our analysis reveals abnormally expanded MDC subpopulations across various tumors and distinguishes cell states that have often been grouped together, such as TREM2+ and FOLR2+ subpopulations. Using deconvolution approaches, we identify five subpopulations as independent prognostic markers, including states co-expressing TREM2 and PD-1, and FOLR2 and PDL-2. Additionally, TREM2 alone does not reliably predict cancer prognosis, as other TREM2+ macrophages show varied associations with prognosis depending on local cues. Validation in independent cohorts confirms that FOLR2-expressing macrophages correlate with poor clinical outcomes in ovarian and triple-negative breast cancers. This comprehensive MDC atlas offers valuable insights and a foundation for futher analyses, advancing strategies for treating solid cancers.

Single-cell profiling of myeloid cells in glioblastoma across species and disease stage reveals macrophage competition and specialization

Article 29 March 2021

Identification of leukemic and pre-leukemic stem cells by clonal tracking from single-cell transcriptomics

Article Open access 01 March 2021

Tumor-associated macrophage clusters linked to immunotherapy in a pan-cancer census

Article Open access 09 August 2024

Introduction

The tumor microenvironment (TME) represents a dynamic network consisting of diverse cell types and acellular components that intricately interact with malignant cells. The understanding of immune cells within the TME, particularly lymphocytes, has paved the way for the development of immunotherapies capable of inducing long-lasting responses across a wide spectrum of cancers¹. However, the current immune-checkpoint inhibitor therapies only confer benefits to a subset of individuals², justifying a better comprehension of other key immune cells with protumoral activities. Among the crucial players in the TME are myeloid-derived cells (MDCs), comprising a heterogeneous array of populations including monocytes, macrophages, conventional dendritic cells (cDC), and polymorphonuclear granulocytes. These MDCs exhibit remarkable plasticity, allowing them to adopt various cellular states and perform a wide range of functions when exposed to different niches³. Intriguingly, while MDCs possess the capacity to enhance cancer cell phagocytosis and induce cytotoxic tumor death when properly activated^4,5, they can also foster tumor growth by facilitating and sustaining cancer hallmarks⁶.

The majority of MDCs originate from bone marrow progenitor cells and subsequently migrate into the TME. Macrophages, however, can derive from two primary sources: 1) erythro-myeloid progenitors in the yolk sac and fetal liver before birth, which exhibit enriched expression of FOLR2, PLTP, and LYVE1 markers^7,8,9; and 2) circulating monocytes originating from the bone marrow and recruited into the tumor¹⁰. Monocyte-derived macrophages have a transient lifespan and necessitate constant replenishment from circulating monocytes¹¹. Their ontogeny can be indicated by the expression of monocyte-related markers such as FN1, SELL, and VCAN¹².

Macrophages, regardless of their origin, can exhibit distinct phenotypes. Macrophages have been classified into two main polarization states: classically activated M1, characterized by a proinflammatory phenotype, and alternatively activated M2, associated with tissue remodeling and/or anti-inflammatory properties¹⁰. In the context of the TME, M2 macrophages are commonly referred to as tumor-associated macrophages (TAMs), and their presence has been correlated with poor prognosis in several tumor types^13,14,15. However, it is important to note that not all TAMs display a clear M1 or M2 phenotype, as they often express markers associated with both polarization states^16,17. This underscores the need to move beyond the simplistic M1/M2 dichotomy and delineate distinct TAM states.

TAMs play a multifaceted role within the TME, influencing tumor growth, epithelial-mesenchymal plasticity, extracellular matrix remodeling, cell invasion, and angiogenesis, subsequently impacting tumor progression, metastasis, and therapy resistance. Furthermore, the interaction between TAMs and malignant cells can facilitate immune evasion through TME shaping¹⁸. Given the pivotal role of TAMs and other MDCs in promoting tumor growth and metastasis, they have emerged as promising targets for cancer therapy. Numerous strategies aimed at depleting or modulating the functional/phenotypic reprogramming, infiltration, or activation of TAMs are being explored¹⁹. However, the lack of precise markers to distinguish between MDC subpopulations and states poses a challenge to the effectiveness of these treatments. Therefore, a thorough characterization of MDC populations within the TME is urgently required to overcome this limitation.

While recent studies have utilized high-resolution technologies such as single-cell RNA-sequencing (scRNA-Seq) and spatial transcriptomics to investigate the immune landscape of different cancer types, a comprehensive pan-cancer integrated analysis encompassing MDCs in the TME with associations with cancer prognostic has been lacking. To address this critical gap, we employed data from three distinct scRNA-Seq technologies, integrating them to construct a comprehensive pan-cancer catalog of MDC subpopulations. Our analysis incorporated data from seven solid tumor types, enabling a thorough characterization of tumor-associated cells. Consequently, we identified and characterized MDC populations/states within these tumors uncovering abnormally expanded MDC subpopulations associated with a poor prognosis across various tumor origins. These findings pave the way for the development of effective targeted immunotherapy strategies aimed at specific MDC subpopulations.

Results

A thorough integration strategy recovers an in-depth pan-cancer repertoire of MDCs in the tumor microenvironment

To define the landscape of cell subtypes in the TME, we integrated scRNA-Seq data comprising 392,204 high-quality cells from 13 public datasets (Supplementary Data 1 and Supplementary Fig. 1a). The integrated datasets comprised three different technologies (10x Genomics, InDrop, and Smart-Seq2) (Supplementary Fig. 1b) and samples from different anatomical sites of healthy donors and patients diagnosed with different cancer types, totaling 138 individuals (Supplementary Data 2). Our pipeline allowed for the integration of a heterogeneous group of datasets, and even though we applied stringent quality control, it retained crucial biological variation while efficiently removing typical contaminants such as doublets and ambient RNA, as well as noise due to batch effect (Fig. 1a). The final integrated dataset encompassed seven tumors, including breast, colorectal, liver, lung, ovary, skin, and uveal melanomas, as well as their adjacent tissue counterparts, metastatic samples, blood samples, and adjacent lymph nodes (Fig. 1b).

**Fig. 1: Multi-tissue single-cell atlas of tumor and healthy samples.**

Using our approach, we identified homogeneous cell groups distributed in clusters. The clusters were annotated based on canonical gene markers and functional signatures (Supplementary Data 3 and 4) of major cell types and differentially expressed genes (DEGs) (Supplementary Data 5 and Supplementary Fig. 1c–f), yielding 10 broadly recognized cell types. Malignant cells were identified by exhibiting a copy number variation (CNV) estimate above the median value of non-malignant cells in the TME (Supplementary Fig. 1c). Low-quality cells were annotated as non-identifiable, and were removed for further analysis (Supplementary Fig. 1d). MDCs were the second-largest group of cells in the TME (Supplementary Fig. 1e), and their proportion in tumor samples was 1.74 times greater than that in normal samples (Supplementary Fig. 1g). The percentage of these cells varied across sample types from 1.7% to 25%, except for metastatic lung samples, where over 60% of TME cells matched mononuclear phagocytes (Supplementary Fig. 1h).

Among the MDC subpopulations (Fig. 1c–g), mononuclear phagocytes (n = 51,687), positive for LYZ, AIF1, and HLA-DRA, were the majority. Other populations, such as neutrophils, marked by CXCR2 and ICAM1, and mast cells, identified by TPSAB1 and CPA3 expression, were also largely detected. Megakaryocytes were the smallest group among MDCs (n = 185) and were not further investigated in this study (Supplementary Fig. 1d).

To characterize the MDC subpopulations, we conducted unsupervised clustering and identified five major lineages (Fig. 1e, f): mast cells (n = 4539), neutrophils (n = 12,033), DCs (n = 6212), monocytes (n = 10,321), and macrophages (n = 38,517). Additional unsupervised hierarchical clustering analysis of the MDCs subpopulations, anchored in 750 highly variable genes, substantiates the delineation of the principal five broad cell types (Fig. 1g). The dendrogram reinforces the proximity of mononuclear phagocytes and elucidates a closer gene expression similarity between macrophages and dendritic cells compared to their expression relationship with monocytes. This observation suggests shared functional or ontogenic pathways between macrophages and dendritic cells. To avoid subjective and arbitrary definitions of cell clusters, we employed entropy-based statistics to accurately quantify the purity of each cluster. Our approach improved the ability to detect and define robust signatures, ranging from common cell states to less frequent ones, performing better than previous strategies that evaluated small and isolated datasets^20,21. Broad cell types displayed on average purity scores of 0.95 (values above 0.9 are recommended as a pure cluster) (Supplementary Fig. 1i,j). Our findings provide new insights into the MDC subpopulations and contribute to a better characterization of their distribution in the TME.

MDCs are heterogeneous and phenotypically diverse across tumor samples

Mast cells and neutrophils (Fig. 1c–g, and Supplementary Fig. 2) corresponded to the less frequent MDC subpopulations identified in our study, and given their low pan-cancer representativeness, they have not been further studied here in detail.

DCs were the fourth most prevalent subpopulation of MDCs, accounting for 9% of these cells (Fig. 1f). Guided by foundational studies²², we identified six primary DC groups (Fig. 2a). The DC1 subpopulation, marked by CLEC9A and CADM1 expression, included 580 cells, while the larger group, DC2, was distinguished by CD1C expression and was further divided into three phenotypes as shown in Fig. 2a–d. The DC3 subpopulation, notable for CD14 and S100A9 markers, included 765 cells enriched within tumor samples. In contrast, the DC4 subpopulation had 314 cells expressing high levels of FCGR3A and SERPINA1, predominantly found in blood and indicative of Monocyte-derived Dendritic Cells (MoDC) (Supplementary Fig. 3a, b). Additionally, we identified the DC_CXCL8 subpopulation (n = 993) (Fig. 2c, d), expressing chemokines like CXCL8 and CXCL2, and the DC_LAMP3 subpopulation (n = 285), characterized as migratory mature DCs expressing LAMP3 and CCR7 (Fig. 2a-d). The DC2_AREG was enriched in the tumor samples, especially in ovary, lung and breast tumors, while DC_CXCL8 was abundant in skin samples (Fig. 2e, Supplementary Fig. 3a, b). These cells showed increased expression of genes coding for PD-1 and PD-2 ligands, suggesting a potential immunosuppressive role (Supplementary Fig. 3d). Additionally, functional enrichment analysis further supported the cell’s profile (Fig. 2f).

**Fig. 2: Characterization of dendritic cell subpopulations.**

The DC2 subpopulation was subdivided into three subpopulations/states DC2_AREG (n = 2044), enriched in tumors and lymph nodes, DC2_FCER1A (n = 726), found mostly in the blood (Fig. 2c–f and Supplementary Fig. 3a,b), and a subpopulation expressing CD207, a langerin protein encoder (DC2_CD207, n = 505), which was abundant in tumor samples, particularly in lung and ovarian tumors and lymph nodes (Fig. 2e and Supplementary Fig. 3b,c). Although langerin is primarily expressed by epidermal macrophages known as Langerhans cells^23,24, DC2_CD207 is a type 2 cDC subpopulation (expressing CD1A) distinct from the aforementioned cell (Fig. 2c, d). DC-activating signals, such as pro-inflammatory cytokines (interferon-alpha/beta and gamma) and antigen presentation, were observed in DC1_CLEC9A and DC2_CD207. DC2_AREG appeared to have regulatory properties, including the modulation of IL-10, -4, and -13 signaling pathways (Fig. 2g and Supplementary Data 6).

Monocytes are precursors of tumor-infiltrating myeloid cells, and we identified four major groups (Non-Classical, Classical, Inflammatory, and Intermediate) (Fig. 3a, b), reflecting six states with distinct gene programs (Fig. 3c–e). The Mono_FCG3RA (n = 2018), exhibited a non-classical phenotype characterized by the co-expression of FCGR3A and FAM110A within a single cluster (Fig. 3b–e). The classical phenotype was marked by the expression of common monocyte markers such as CD14 and SELL (Fig. 3b, e), and was further differentiated into two states based on FOS expression: Mono_CD14_FOS⁻ (n = 2150) and Mono_CD14_FOS⁺ (n = 1497) (Fig. 3c-e). Mono_CD14_FOS⁺ was enriched in lymph nodes and normal tissues, such as the ovary (Supplementary Fig. 4a–c), while Mono_CD14_FOS^- was more prevalent in blood samples (Fig. 3f and Supplementary Fig. 4a–c). These two states are poorly described, and their functions remain unclear, although both are enriched in Toll-like receptor activation pathways (Fig. 3g and Supplementary Data 6).

**Fig. 3: Characterization of monocytes subpopulations.**

Additionally, among the Mono_Inflammatory group, the Mono_IL1B (n = 1739) was identified as displaying a pro-inflammatory profile with high expression levels of IL1B, CXCL2, and EREG (Fig. 3c–e), as well as other inflammatory genes (Supplementary Fig. 4d and Data 4) with increased expression of genes involved in Toll-like receptor pathways and death receptor ligands that activate caspase cascades (Fig. 3g).

We were also able to identify two monocyte clusters displaying an intermediate phenotype between monocytes and antigen-presenting cells, expressing both genes coding for CD16 and CD14. We called this group Mono_Inter, from which CXCL10 and CLEC10A expressions divided this monocytes into MonoInter_CXCL10 (n = 1238) and MonoInter_CLEC10A (n = 1559), respectively (Fig. 3c–e). These cells expressed markers related to antigen presentation, such as HLA-DMB and HLA-DRB6 (Fig. 3d), and were enriched in MHC class II antigen presentation, interferon/gamma signaling, and costimulation by CD28 pathways, as well as PD-1 signaling (Fig. 3g and Supplementary Data 5).

Macrophages were the largest group of mononuclear phagocytes in this study, corresponding to 51.6% of all myeloid cells (Fig. 1f). We identified three main clusters of macrophages based on their ontogeny markers²⁵: 1) monocyte-derived macrophages distinguished by the expression of monocyte markers such as CCR2 (an embryonic cell marker indicative of monocyte lineage), VCAN, S100A6, and CD52; 2) likely resident-tissue macrophages (RTM), expressing CD163 (highly expressed in the initial phase (first wave) of embryonic hematopoiesis, along with markers associated with the second wave of macrophage generation, such as MPO, and other tissue residency markers such as TCF12, MS4A4A, GATM, LYVE1, FOLR2, and PLTP; and 3) resident-like macrophages (RTM-like), resembling resident macrophages, they comprise of monocyte-derived macrophages that exhibit a combination of markers from both the aforementioned groups (Fig. 4a–c). Although monocyte-derived macrophages were the most common type in tumors, RTMs were surprisingly shown to be expanded in tumors (Supplementary Fig. 5a, b), particularly in lung samples (Supplementary Fig. 5c-f). RTM-like cells were found in all conditions except uveal melanoma, melanoma, liver tumors, and normal breast tissue (Supplementary Fig. 5c–f). Also, we observed an RTM-like enrichment in ovarian samples predominantly in normal tissue compared to tumors (Supplementary Fig. 5e).

**Fig. 4: Characterization of macrophage subpopulations.**

In addition to classifying macrophages based on origin, we also distinguished 12 clusters based on gene programs (Fig. 4d–g), across the technologies (Supplementary Fig. 5g-i) with greater diversity in tumors than normal samples (Fig. 4e and Supplementary Fig. 5d).

To ascertain the optimal method for mitigating batch effects while conserving the biological variation inherent to distinct cell types, we conducted an extensive analysis of six data integration methods on the macrophage data (Supplementary Fig. 6a), employing a benchmarking framework²⁶. This framework assesses the quality of integration, the retention of cell type-specific variation, and the preservation of biological meaning within the integrated data. When considering the metrics laid out by this framework, Harmony²⁷ excelled in batch correction but fell short in maintaining cell identity. In contrast, both scVI²⁸ and scANVI²⁹ achieved high rankings in conserving cellular biology. Among these, scANVI marginally outperformed scVI, benefiting from the use of pre-existing cell type annotations to refine cell type distinctions post-integration. Despite this, scVI was chosen for further analyses due to its robust capability to process unlabeled data, striking a balance between effective integration and biological fidelity (Supplementary Fig. 6b). Employing scVI, we generated an embedding that accurately integrates and represents all 37,891 macrophage cells from 11 studies across our datasets (Supplementary Fig. 6c).

The macrophages subpopulations were stratified according to their functional gene signatures, mostly following a consensus model for TAM diversity⁹. The alveolar-like macrophage (Mac_Alv-like, n = 8816), marked by the expression of PPARG and MCEMP1, was mostly found in lung samples, as expected (Fig. 4e and Supplementary Fig. 5d-f). The pro-angiogenic macrophage subpopulation (Mac_Angio, n = 3108) displayed high expression of genes associated with angiogenesis, such as VEGFA, VCAN, and EREG, and was found enriched in lung and ovarian tumors (Supplementary Fig. 5d–f). Hypoxia-associated macrophages (Mac_Hypo, n = 1802) were mostly found in tumor samples (lung and ovary) (Supplementary Fig. 5d–f) and presented an enrichment of genes associated with hypoxia, such as SLC2A1 and ERO1 (Fig. 4f, g).

Two macrophage subpopulations expressing high levels of interferon-primed genes, such as CCL8, IDO1, and CXCL9, were named RTM_IFN (n = 2061) and Mac_IFN (n = 1981), which were abundant in ovarian and lung tumor samples, respectively (Fig. 4d–g and Supplementary Fig. 5d–f). Another two macrophage clusters with a lipid-associated (LA) metabolism transcriptional signature (Fig. 4d–g), including genes such as FABP5, LPL, and LIPA, named RTM_LA (n = 3569) and Mac_LA (n = 1249), were found enriched in breast, lung, and colorectal tumors (Supplementary Fig. 5d–f). Another subpopulation expressing high levels of MKI67 and STMN1 was classified as Mac_Prolif (n = 2234), and found to be enriched in liver, melanoma, and ovary tumor samples (Fig. 4d-g, Supplementary Fig. 5e,f). We also identified a large population of macrophages in uveal melanoma metastatic samples (Supplementary Fig. 5e), which expressed high levels of CCR2 (Fig. 4d, f), a receptor for monocyte chemoattractant protein-1, and LGALS2. Therefore, this state was named early-stage macrophage (Mac_ES n = 4073).

We identified one monocyte-derived macrophage subpopulation displaying a pro-inflammatory phenotype, expressing the receptor of the chemokine CX3CL1 (CX3CR1) and genes from the antigen-presentation pathway (HLA-A/C, HLA-DQA1/B1). This population was named Mac_AgPres (n = 3965) (Fig. 4d–g). Additionally, a population expressing high levels of resident markers such as LYVE1, FOLR2, and PLTP, which resembled interstitial macrophages with specific expressions of genes such as MAF and SEPP1, was identified. This subpopulation was named RTM_Int (n = 3507) and was found to be enriched in normal tissues (Fig. 4e, Supplementary Fig. 5d). The RTM-like macrophage subpopulation was found to express high levels of metallothioneins (Fig. 4d–g), named RTM-like_MT (n = 1209), and was found predominantly in ovary samples (Supplementary Fig. 5e,f). Importantly, the signatures identified were reproducible among different platforms (Supplementary Fig. 7a-b).

By examining the enriched pathways in each macrophage subpopulation, we were able to identify distinct functional roles. Enriched pathways in Mac_IFN, Mac_ES, RTM_IFN, and Mac_AgPres subpopulations were linked to the interferon signaling pathway. Antigen presentation genes were enriched in RTM_IFN, RTM_LA, Mac_LA, and Mac_AgPres, with the latter also expressing genes related to phagocytosis, such as the MyD88-dependent cascade initiated on endosomes and FCGR activation (Fig. 4h). Interestingly, Mac_ES, Mac_IFN, Mac_Hypo, RTM_Int, Mac_Angio, RTM-like_MT, and RTM_IFN showed enrichment of genes involved in the IL-10, −4, or 13 pathways, which have been associated with pro-tumoral processes (Fig. 4h). Mac_Hypo, RTM_LA, Mac_LA, and RTM_IFN displayed pathways related to extracellular matrix remodeling (Fig. 4h and Supplementary Data 6).

Macrophages have the ability to utilize diverse energy sources to fulfill their functions (Fig. 4i). Certain subpopulations, namely Mac_Hypo and Mac_Angio, appear to rely on glucose metabolism since genes related to the glycolytic pathway are enriched. Among these, Mac_Angio also exhibits enrichment in genes involved in the oxidative phosphorylation pathway. Clusters such as the Mac_LA, RTM_LA, and RTM-like_MT expressed more genes associated with fatty acid metabolism. Notably, NAD metabolism is highly enriched in the Mac_ES and Mac_IFN. The Mac_LA, RTM_IFN, and Mac_AgPres expressed an enrichment in genes related to amino acid metabolism. The distinct metabolic profiles of RTM subpopulations appeared to vary significantly from one another. For a comprehensive overview of the major metabolic pathways, please refer to Supplementary Fig. 8.

Our analysis has refined the understanding of macrophage heterogeneity by distinguishing between profiles that were previously grouped together. For instance, we have identified four unique macrophage clusters (Mac_LA, Mac_AgPres, RTM_LA, and Mac_IFN) that express high levels of TREM2. Unlike earlier classifications that lumped these as generic TREM2⁺ macrophage subpopulation (see Supplementary Fig. 9a–h), the subpopulations presented distinct signatures and functional profiles. Similarly, our study challenges the traditional view of RTMs as a uniform group. We have identified at least three diverse RTM profiles, including interstitial, lipid-associated metabolism, and interferon-primed macrophages, each characterized by different tumorigenic behaviors. By assessing cluster purity, we found that our technique produced highly homogeneous clusters compared to previous integration approaches, indicating its effectiveness (Supplementary Fig. 9i–k and Supplementary Data 7). Overall, our approach expanded the catalog of mononuclear phagocytes in the TME, enabling the identification of rare populations. This led to valuable insights into the diversity and complexity of mononuclear phagocyte populations in the TME, especially macrophages (Fig. 4i), underscoring the importance of employing deep scRNA-Seq analysis to uncover poorly studied subpopulations.

In order to validate our findings, we extended our analysis to include 13 independent datasets, creating an additional pan-cancer atlas. This expanded atlas encompasses a comprehensive array of samples (n = 233) across 10 tumor types, incorporating a significant number of myeloid cells (n = 96,514) (Supplementary Fig. 10a–d and Supplementary Data 7). Within this broader dataset, we validated the presence of several DC subpopulations, including the DC_CXCL8 phenotype. The DC4 subpopulation was the only exception, probably because this version of the atlas does not include blood samples (Supplementary Fig. 10e, f). For monocytes, we successfully identified all six subpopulations as initially characterized (Supplementary Fig. 10g, h). In the case of macrophages, we confirmed nine out of ten signatures (Supplementary Fig. 10i–k). The one exception, the RTM-like_MT subpopulation, was noted for its low expression of the metallothionein gene set, indicating a distinct transcriptional profile. Of note, Mac_Alv-like signature was not so evident. This may be attributed to the lack of lung samples within our new integrated dataset. This cross-checking of our results with independent datasets strengthens the trustworthiness of our identified mononuclear phagocyte groups.

A plethora of macrophage phenotypes and functional states co-exist in the TME

Since M1 and M2 polarization gene programs are well-known^30,31,32 and reflect important macrophage-related features, we attempted to assign these profiles to the 12 macrophage subpopulations identified (Fig. 5a, b). Although the RTM-IFN and Mac_ES were, respectively, the subpopulations displaying the higher scores for M1 and M2 gene signatures (Supplementary 11a, b and Supplementary Data 8), there was no clear overrepresentation of genes associated with either polarization profile in any of the subpopulations. In effect, subpopulations such as the RTM-IFN express high levels of genes encoding pro-inflammatory cytokines such as CXCL9, CXCL10, and CXCL11, as well as immunosuppressive molecules such as PDL-1 and PDL-2 (Fig. 5a, b). Although M1/M2 ratio scores (Supplementary Fig. 11c) discriminate the subpopulations regarding their polarization states, this dichotomy oversimplifies the highly specialized, transcriptomically dynamic, and extremely heterogeneous nature of macrophages in vivo.

**Fig. 5: Macrophages exhibited diverse phenotypes and functional states.**

To gain further insights into the roles of macrophage subpopulations in the TME, we investigated their associations with cancer hallmark signatures³³ (Supplementary Data 4). Consistent with their names, Mac_Angio exhibited higher scores for the Angiogenesis signature, while Mac_Hypo showed elevated scores for the Hypoxia signature (Fig. 5c, d). Notably, RTM_Int demonstrated enrichment in signatures associated with the induction of epithelial-mesenchymal transition (Fig. 5e). The extracellular matrix signature in macrophages³⁴ was predominantly observed in Mac_LA, Mac_Alv-like, and RTM_LA subpopulations (Fig. 5f and Supplementary Data 4). The RTM_IFN cluster exhibited the highest scores for antigen presentation³³ as well as for FcγR-mediated phagocytosis signatures (Fig. 5g, h).

To explore the potential immunosuppressive role of macrophages, we examined the co-expression of checkpoint genes and immune regulators. We found that TREM2, a marker for immunosuppressive TAMs³⁵, was highly expressed in both resident and recruited LA populations, as well as in Mac_AgPres (Fig. 5i). However, only Mac_LA exhibited high levels of co-expression with the PDCD1 gene (programmed death 1, PD-1). Furthermore, Interferon-primed subpopulations demonstrated elevated expression of CD274 and PDCD1LG2, the genes encoding PD-L1 and PD-L2, respectively (Fig. 5i). RTM_Int cells expressed high levels of both FOLR2 and PD-L2 coding-gene. TREM1, typically highly expressed in macrophages in inflamed tissues^36,37, was mainly expressed by the Mac_Alv-like, Mac_Angio, and Mac_Hypo subpopulations (Fig. 5i). We also evaluated the myeloid-derived suppressor cell (MDSCs) signature proposed by Alshetaiwi et al (2020) in all immature MDSCs. However, none of the subpopulations resembled that particular signature (Supplementary Fig. 12). Nevertheless, monocytes subpopulations such as Mono_CD14_FOS^- and Mono_CD14_FOS⁺ exhibited likely suppressive profiles, based on CTSD and PLA2G7 expression, as well as neutrophils states/subpopulations, based on CD84, AGR2 and CLE4E expressions (Supplementary Fig. 12).

To elucidate the intricate differentiation trajectory in the monocyte-macrophage and monocyte-DC axes, we employed an unsupervised pseudotime analysis strategy. By using Mono_CD14_FOS^- as the “root” for trajectory inference, our analysis supported the notion that CD14^hi monocytes serve as precursor cells for both Mono_FCG3RA and monocyte-derived cells in tissues (Fig. 5j–m). Intermediate monocytes (MonoInter_CXCL10 and MonoInter_CLEC10A) can potentially give rise to both DCs and Mac_ES subpopulations³⁸. The Mac_ES subpopulation appears to be the earliest monocyte-derived macrophage subpopulation ontologically, acting as a crucial precursor for other macrophages, including RTM-like. Two distinct phenotypes, originating from two main branches, were identified as the most distant from the root: 1) Mac_AgPres and 2) Mac_LA, both of which are characterized by high levels of TREM2 expression. Through the analysis of the genes that co-vary over pseudo-time, we were able to identify a set of monocyte-related genes (VCAN, SELL, and FCN1) at the beginning of the trajectory and both specific and general phenotype markers of the macrophage lineages (APOE, TREM2, SPP1, HLA-DRA, C1QA, and CD63) in late branches (Fig. 5k–m).

Deciphering the clinical impact of TREM2⁺ macrophages in the TME

The clinical significance of MDC subpopulations as potential targets has garnered increased interest due to their higher abundance in the TME. To further explore their clinical impact, we conducted an investigation using larger cohorts to assess MDC subpopulations in different tumor types. We utilized a set of 47 subpopulation signatures to estimate the proportions of cell types through deconvolution analysis of bulk RNA-Seq samples obtained from nine tumor types available in The Cancer Genome Atlas (TCGA) database. To validate the applicability of our signature matrix in estimating cell populations within the TCGA data, we calculated the correlation between bulk RNA-Seq and scRNA-Seq data. We observed a significant correlation for all TCGA cohorts (Pearson R² > 0.7, p value ≤ 0.05) (Supplementary Fig. 13a–i).

All the tumor samples were enriched with malignant cells (~50%), except for lung adenocarcinoma (LUAD), which had very similar proportions of malignant cells (16% of total cells) and fibroblasts (19.9% of total cells) and higher content of epithelial cells (43.4% of total cells) (Fig. 6a). The proportions of cell types across tumors are shown in Supplementary Fig. 14a–m. Breast cancer (BRCA) presented a higher proportion of non-epithelial cells, and together with LUAD, these tumors were the most enriched in MDCs (Fig. 6a). Stratifying for mononuclear phagocyte subpopulations, macrophages were among the most abundant cell types (~12% of total) (Fig. 6b). The RTM_IFN subpopulation was more commonly found in different tumor types, especially in skin cutaneous melanoma (SKCM) (7.6% for metastatic and 5.6% for primary tumor) and uveal melanoma (UVM) samples (1.9% of total cells). In contrast, RTM_Int was predominantly found in lung tumors (6.06% in LUAD and 3.9% in lung squamous cell carcinoma) (Fig. 6b). The Mac_LA subpopulation was mainly found in LUAD and BRCA tumors, while Mac_AgPres was mainly found in colorectal tumors (2.6% of total cells) (Fig. 6b).

**Fig. 6: Impact of Mac_LA (TREM2⁺) on clinical outcomes.**

We then conducted an investigation to assess the impact of TREM2⁺ macrophage subpopulations (Fig. 6c), particularly Mac_LA, due to its high abundance in tumors (Fig. 6d). Patients with triple-negative breast cancer (TNBC) were categorized into groups based on high and low estimated abundance of Mac_LA in their TME. Those with a high quantity of Mac_LA demonstrated significantly worse overall survival (OS) and progression-free survival (PFS) compared to those with low levels of Mac_LA (OS, log-rank test, p-value = 0.0076; PFS, log-rank test, p value = 0.033) (Fig. 6e). However, this trend was divergent in other BRCA subtypes (Supplementary Fig. 15a–c) as well as high-grade serous ovarian carcinoma (HGSOC) samples (Supplementary Fig. 16a, b). We also evaluated global TREM2 expression and protein levels as clinical biomarkers in different cohorts. Global TREM2 gene expression was associated with poorer OS compared to the lower expression group in patients diagnosed with different BRCA subtypes (TNBC, log-rank test, p-value = 0.0031, Supplementary Fig. 15d-g) and HGSOC (OS, log-rank test, p-value = 0.013, Supplementary Fig. 16c, d). Upon evaluating TREM2 protein expression globally and specifically within peritumoral macrophages (identified by CD68⁺ and TREM2⁺ co-staining) via immunohistochemistry, we observed that high levels of lipid-associated macrophages (Mac_LA) tend to correlate with poorer clinical outcomes in TNBC. This observation aligns with our survival analysis conducted using deconvolution estimates. However, it was only the overall TREM2 expression levels in the TNBC-INCA cohort that were significantly associated with adverse outcomes (Fig. 6f-h). No significant difference was found in the HGSOC-INCA cohort (Supplementary Fig. 16e, f).

In HGSOC-INCA, the higher intratumoral expression of CD68⁺ and TREM2⁺ was accompanied by increased PD-1 and CD8 levels, suggesting the involvement of the PD-1 receptor in the clinical response associated with TREM2-expressing macrophages (Supplementary Fig. 16g-h). In TNBC-INCA, a divergent pattern was noted. The High TREM2 group exhibited a lower proportion of CD8, suggesting an inverse relationship compared to HGSOC-INCA (Fig. 6i). These distinct predictions underscore the context-dependent variations in the immune microenvironment and the nuanced role of TREM2-expressing macrophages in different cancer types. PD-1 expression was significantly associated with poorer OS and recurrence-free survival (RFS) in TNBC-INCA patients (OS, log-rank test, p-value = 0.0053; RFS, log-rank test, p-value = 0.03) (Supplementary Fig. 17a). No difference in OS and RFS were observed for CD8 and PD-L1 (Supplementary Fig. 17b, c).

To identify communication axes involving Mac_LA, we conducted a ligand-receptor inference analysis between these cells and other key players in tumor progression, including malignant cells and T/NK cells. We first clustered and annotated these cells based on canonical gene markers and DEGs, resulting in 11 cell types (Supplementary Fig. 18a, b). Mac_LA exhibited 714 putative intercellular communications, as indicated by ligand-receptor pairings between Mac_LA and lymphocytes, as well as malignant cells (Supplementary Data 9). Among these interactions, CD8⁺ T cells emerged as the most representative cell type (Supplementary Fig. 18c). The axes of major expression in these interactions were SPP1_CD44, MIF_CD74, ANXA1_CXCR4 and GSTP1_CD44 (Fig. 6i). These findings shed light on potential communication pathways involving Mac_LA and provide valuable insights into its role in tumor microenvironment interactions.

MDC subpopulations display distinct clinical outcomes depending on their niche

To further investigate the clinical impact of other MDC subpopulations inferred in the TME across different tumor types, we conducted a univariate Cox regression analysis (Supplementary Fig. 19a). Mono_FCG3RA subpopulation, enriched in BRCA tumors, especially the Luminal B subtype (representing 1.22% of total cells) (Fig. 6b), exhibited a significant association with poor OS in TNBC when assessed through uni- and multivariate regression analysis (log-rank p-value = 0.0027, Supplementary Fig. 19a–c), reinforcing its independent value as a prognostic marker. Conversely, in patients with the Luminal A subtype, a higher percentage of this population was correlated with improved OS (log-rank p-value = 0.0011, Supplementary Fig. 19a,c,e), indicating that its clinical impact depends on the specific TME context. Furthermore, Mono_FCG3RA was markedly reduced in TNBC patients with “tumor-free” status (p-value = 0.0095), whereas this was not observed in the Luminal A subtype (Supplementary Fig. 19f).

Another subpopulation with distinct impacts depending on the TME niche is the RTM_Int. In HGSOC and TNBC, RTM_Int is associated with poor OS, while in lung adenocarcinoma (LUAD) it is linked to improved OS (univariate Cox regression and log-rank, p-value <0.05, Fig. 7a and Supplementary Fig. 19a). No significant clinical impact was observed in other BRCA subtypes (Supplementary Fig. 20a–c). Patients with TNBC and HGSOC were categorized into groups, and those with a high quantity of RTM_Int demonstrated significantly worse OS compared to those with low levels of RTM_Int (log-rank test, p-value = 0.0025, and log-rank test, p-value = 0.017, respectively), but there was no significant difference for progression-free survival (PFS) (log-rank test, p-value = 0.34, and log-rank test, p-value = 0.65, respectively) (Fig. 7b, c). When analyzing FOLR2 as a biomarker, high protein levels were associated with poor OS in TNBC-INCA (log-rank test, p-value = 0.0088, Fig.7d) and poor PFS in HGSOC-INCA (log-rank test, p-value = 0.0034, Fig. 7e). The global expression of FOLR2 showed distinct results depending on the analyzed cohort (Supplementary Fig. 21a-c). In order to validate the impact of RTM_Int we performed dual staining with FOLR2 and PDL-2, another marker highly present in this macrophage subpopulation. For TNBC-INCA the dual staining showed no significant differences, although the same pattern was observed (Fig. 7f). In HGSOC-INCA the high levels of FOLR2 and PDL-2 were associated with a poor OS (log-rank, p-value = 0.019) and PFS (log-rank, p-value = 0.014), reinforcing the previous findings (Fig. 7g). Representative images of tissue microarrays and the association between FOLR2 and CD8 or PD-1 are provided in Fig. 7h–i.

**Fig. 7: Impact of RTM_Int (FOLR2⁺) on clinical outcomes.**

Due to its dual role in prognosis associated with niche variability, we compared DEGs between RTM_Int cells found in normal and tumor samples. Protumorigenic factors associated with hypoxia and angiogenesis, like HIF1A and VEGFA, as well as genes related to a pro-tumoral phenotype, such as SPP1 and TGFB2, were highly expressed in this macrophage found in the tumor context (Fig. 7l, m). Since RTM_Int exhibits high levels of FOLR2 and the immunosuppressive gene PDCD1LG2 (Fig. 7n), we also investigated the association with other immune markers. Interestingly, an increase in CD8⁺ T cell frequency was observed in tumor samples displaying high global levels of FOLR2 (Fig. 7j) and intratumoral FOLR2 macrophages (Fig. 7k) accompanied by higher levels of PD-1⁺. No significant difference in OS and PFS was observed for PD-1⁺ and CD8⁺ cells, or CD68⁺ macrophages (Supplementary Fig. 22a–h). These findings support the notion that FOLR2⁺ macrophages are reprogrammed during tumor formation by activating immunosuppressive gene programs, potentially leading to the exhaustion of CD8⁺ T lymphocytes in the context of TNBC and ovarian tumors. This highlights RTM_Int as a promising therapeutic target. Analyses of OS (uni and multivariate) from other tumor types are described in Supplementary Data 10 and 11. Representative images of tissue microarrays depicting CD8^+, PD1⁺, and KI67 markers and FOLR2⁺ PDL-2⁺ co-expression in both HGSOC-INCA and TNBC-INCA samples are provided in Supplementary Fig. 22i and Supplementary Fig. 23a, b.

In our study, we further explored the clinical impact of RTM_Int by investigating the communication axes between this subpopulation and malignant cells, as well as T/NK cells. We identified 889 intercellular communications involving the RTM_Int cell type (Supplementary Fig. 18c, Supplementary Data 9). The connections with CD8⁺ T and CD4⁺ T cells were particularly enriched, with ANXA1_CXCR4, CXCL2_CXCR4, and CXCL3_CXCR4 axes being prevalent in these communications (Fig. 7o). Additionally, similar to Mac_LA, MIF_CD74 appeared to be an important axis ligand involved in cell-cell communication between RTM_Int and tumor cells (Fig. 7o).

In comparison to previous studies, our method of integrating scRNA-Seq datasets, clustering, and characterizing cell states/subpopulations significantly expanded the catalog of mononuclear phagocytes in the TME. This approach allowed us to identify rare and poorly described subpopulations. Furthermore, our analysis led to the identification of distinct RTM phenotypes, such as interstitial, lipid-associated, and interferon-primed macrophages, as well as to the characterization of a phenotype (RTM-like_MT) where cells simultaneously express resident-tissue and monocyte-derived markers. Moreover, we unveiled the clinical impact of specific TREM2⁺ (Mac_LA) and FOLR2⁺ (RTM_Int) subpopulations in both TNBC and HGSOC patients, which may lead to new possibilities for targeted treatments.

Discussion

Integration of single-cell data has proven to be a game changer for uncovering valuable biological information from large and complex volumes of data. By pooling data from diverse sources, it is possible to unveil cellular heterogeneity and phenotypes by enhancing the robustness and reliability of the findings. Here we have thoroughly integrated and characterized cell subpopulations in the TME, yielding a large and comprehensive dataset comprising different tumors, technologies, and sample types, generated across multiple conditions and donors. This diversity introduces dataset-specific batch effects. The selection of scVI as the integration tool was based on comprehensive analyses, which included the application of various integration platforms and the evaluation of metrics from recent benchmarking integration studies^26,39,40. These studies highlight the importance of selecting an appropriate data integration method to ensure the reliability of the resulting atlas. scVI²⁸, employing variational autoencoders, is recognized for its capability to process complex datasets by identifying latent variables and conserving biological variance. Nonetheless, scVI may not completely eliminate batch effects or may inadequately delineate batches within certain cell types, reflecting a trade-off between batch effect correction and the preservation of biological diversity. To bolster the robustness and generalizability of our findings, cross-validation with alternative integration methods was performed, and results were corroborated using independent datasets.

Notably, we expanded the classification of myeloid-derived cell types including monocytes, macrophages, DCs, neutrophils, and mast cells to 29 distinct clusters. We endeavored to infer the likely origins of macrophages within our study, differentiating between those that are suggestive of tissue residency and those potentially derived from monocytes while acknowledging the inherent challenges in classifying macrophage ontogeny in human samples. This distinction was drawn using the best available evidence and methodologies, including analysis of specific markers as guided by the current literature^25,41. However, we recognize the limitations of these approaches, particularly the absence of ways to unequivocally trace cell lineages in human studies. Through a rigorous methodology and extensive curation, we identified more homogeneous clusters compared to previous approaches^20,21, unraveling high-quality clusters representing a wide range of cells found in the TME of solid tumors, including states less characterized such as Mono_CD14_FOS^+/− and RTM-like_MT. These findings offer valuable insights for future research and development of immune checkpoint blockades and target therapies.

Mononuclear phagocytes play a pivotal role in development and homeostasis maintenance⁴². Their remarkable plasticity enables them to actively participate in various critical processes, including organogenesis, tissue repair, and immune defense. In addition to producing essential growth and inflammatory factors that orchestrate lymphocyte responses, they are involved in extracellular matrix maintenance, angiogenesis, tissue enervation, and the clearance of apoptotic cells^43,44. In the context of tumor development, which shares similarities with organ development, these functions are also highly prevalent. In lung metastasis, mononuclear phagocytes account for more than 60% of the TME cell population, underscoring their significant involvement in the tumor microenvironment and potentially impacting disease progression.

In our study, we identified six distinct states of DCs based on their unique gene signatures. Among them, one cDC type 1, expressing CLEC9A and CADM1, four subpopulations of cDC type 2, three of them DC2, distinguished by CD1C expression, and one DC3, one DC4, one mature subpopulation expressing LAMP3 and CCR7 and a poorly described phenotype, DC_CXCL8, that exhibits contrasting prognostic implications in different tumor types. Moreover, we identified six monocytes displaying either pro- or anti-tumoral properties, regulating a variety of processes from angiogenesis to immune modulation in a context-dependent manner. The transcription factor c-Fos has been shown to positively regulate monocyte development^45,46. Resting monocytes typically have low or undetectable FOS mRNA levels, and their activation is associated with a rapid and transient increase in FOS mRNA levels⁴⁵. Furthermore, FOS overexpression stimulates growth arrest and differentiation into macrophages^45,47. In addition, the Mono_FCGR3A is associated with a worse prognosis in the TNBC subtype, aligning with the findings of Chen et al., who have linked non-classical monocytes with immunosuppressive functions⁴⁸. However, an opposite impact is observed in the LumA subtype, where the Mono_FCG3RA enrichment is associated with a better prognosis, reinforcing that MDC association with cancer prognosis is context-dependent.

Macrophages are versatile immune cells present in various tissues and their ontogeny and functional diversity have been the subject of extensive research. Studies have revealed that TAMs can be derived from circulating monocytes recruited to the tumor site or from RTMs that adapt to the tumor milieu⁴⁹. These TAMs undergo polarization into various phenotypes influenced by their environment, such as the M1 and M2 states, classically associated with pro-inflammatory and immunosuppressive functions, respectively^50,51,52. However, our study challenges the traditional binary division of macrophages since none of the states identified clearly displayed M1 or M2 signatures. Instead, we observed that some populations simultaneously expressed genes associated with both signatures, highlighting the complex nature of macrophage plasticity in the TME. To address this complexity and to follow recent efforts to standardize nomenclatures, our in-depth analysis identified twelve distinct TAM states, each characterized by a unique transcriptional profile and ontogeny. We based our annotation mainly on a recently proposed consensus model of TAM diversity gene signatures⁹, and better-characterized phenotypes that have not been extensively described, such as Mac_LA co-expressing TREM2 and PD1, and three different profiles of RTMs (RTM_LA, RTM_IFN, and RTM_Int), which were frequently lumped together in a single group of cells in multiple TME studies^9,53. We also reported RTM-like_MT, a subpopulation that displays markers associated with both ontogeny types, supporting the notion that under steady-state conditions, RTMs are mostly maintained via self-renewal processes. However, circulating monocytes can give rise to self-renewing RTMs if an adequate habitat is provided^54,55.

We also investigated the expression of the members of the Triggering Receptor Expressed on Myeloid cells (TREM) family, TREM1 and TREM2, associated with TAMs^56,57. Our catalog revealed that TREM2⁺ macrophages²⁰ can be further expanded into at least four different cell states including, RTM_LA, Mac_LA, Mac_IFN, and Mac_AgPres. Similarly, TREM1 was mainly found expressed in into three subpopulations (Mac_Alv-like, Mac_Angio, and Mac_Hypo). These findings indicate that there is no single phenotype for TREM1⁺ or TREM2⁺ macrophages.

The Mac_LA subpopulation stands out for its significant expression of TREM2 among the other Mac subpopulations, and its specific implication in clinical outcome is notable in TNBC, but not in other BRCA subtypes or HGSOC, despite the association of TREM2 expression or prtein levels with poor overall survival, consistent with findings from other research groups^58,59,60. These results suggest that in different tumor types, other TREM2-expressing macrophage subpopulations might play central roles in tumor progression within their respective microenvironments.

Other subpopulations expressing high levels of TREM2, such as RTM_LA and Mac_AgPres, are respectively associated with poor and good prognosis in LIHC. The co-expression of TREM2 and the receptor PD-1 in Mac_LA raises the possibility that this subpopulation may indeed have immunosuppressive roles in the TME, in line with previous findings⁶¹. An important axis of communication between malignant cells and these macrophages is the MIF:CD74 interaction axis. The activation of CD74 receptors in Mac_LA may contribute to the activation of an anti-inflammatory profile^62,63. In mouse models of sarcoma, colorectal, breast, and gynecological cancers, TREM2 deletion or blockade with a monoclonal antibody has been shown to reduce tumor growth, enhance antitumor CD8⁺ T cell responses, including the effectiveness of anti-PD–1 treatment, and modified the TAM landscape^61,64. Our findings strongly support the notion that distinct TAM phenotypes expressing TREM2 can coexist in the tumor microenvironment, exerting different roles, and contributing to the unclear dual function of TREM2 in cancer prognosis⁶⁵. Phase I clinical trial has been leading in advanced refractory tumors using specific TREM2 mAb against tumor-associated macrophages expressing TREM2⁶⁶. The complex and context-dependent function of TREM2-expressing macrophages in the tumor microenvironment underscores the need for further research to fully understand its contribution to cancer progression and its potential as a therapeutic target. The identification of distinct TAM phenotypes expressing TREM2 and their differential impact on cancer prognosis opens new avenues for precision medicine approaches and personalized treatment strategies.

In addition to TREM2-expressing macrophages, we also investigated the clinical impact of RTM_Int, a subpopulation expressing high levels of folate receptor β (FRβ), which is encoded by the FOLR2 gene and serves as a marker for resident macrophages in the TME. High levels of this subpopulation are associated with a poor prognosis in TNBC and HGSOC, two very lethal tumor types, although a beneficial role has been seen for other breast cancer subtypes⁸. The presence of cells co-marked for FOLR2 and PDL-2 was significantly associated with poor outcomes in HGSOC-INCA highlighting its importance as a target for immunotherapies in a context-dependent maner^67,68,69.

In addition, RTM_Int expresses β1 integrins (ITGB1), Insulin-like Growth Factor 1 (IGF1), and Tumor necrosis factor (TNF) that are involved in interaction with the receptor for activated C kinase (RACK1). These pathways are involved, in general, in therapy resistance and cell progression^70,71,72. Our results also show the potential of utilizing FOLR2 as a biomarker for patient stratification and prognosis in these tumor types, in line with other studies that evaluated other tumor types^8,64,73,74. Several trials involving targeting FOLR2 are currently ongoing, reflecting the current level of research interest using this approach^75,76.

In conclusion, we have characterized in high-resolution the landscape of MDCs in solid tumors by effectively integrating a large dataset through a pipeline that uses cluster-specific gene expression analysis to infer cell type, distribution, function, ontogeny, phenotype, cellular interactions, and disease associations. This comprehensive reference atlas serves as a resource for the scientific community and provides insights into the identities and characteristics of mononuclear phagocytes in the TME, presenting new avenues for understanding and manipulating their behavior in cancer. Our findings also demonstrate that the reprogramming of TAMs is highly dependent on the microenvironment, leading to varied TAM expression profiles across different cancer types, with distinct clinical impacts. This highlights the importance of understanding the specificities of subpopulations/states in the TME context. By shedding light onto the intricate landscape of mononuclear phagocytes in the TME, our work pushes the development of more effective and targeted therapeutic interventions against cancer.

Methods

Ethical considerations

This retrospective study received approval from the Ethics in Human Research Committee of the Brazilian National Cancer Institute (INCA) (CAAE 52409221.4.0000.5274 and 61675516.9.0000.5274) in Rio de Janeiro, Brazil, and was conducted in accordance with all applicable regulations concerning the involvement of human study participants. Written informed consent was obtained from each patient prior to the commencement of study procedures. The study adhered to the Good Clinical Practice Guidelines.

Data download and pre-processing

Pre-processed scRNA-Seq data (Supplementary Data 1) from patients with breast cancer (GSE114727), hepatocellular carcinoma (GSE140228, GSE125449), lung cancer (GSE127465), melanoma (GSE115979), ovarian cancer (GSE154600, GSE72056), uveal melanoma (GSE139829), skin (GSE130973) and metastasis from uveal melanoma and lung (GSE158803) samples were obtained from the public repository Gene Expression Omnibus (GEO) using the GEOquery Bioconductor package⁷⁷. Additionally, datasets from lung, ovarian, and breast cancer were downloaded from Qian et al.⁷⁸, on its platform (blueprint.lambrechtslab.org/). We downloaded datasets from the Human Lung Cell Atlas project⁷⁹ on the Synapse platform (SYN21560407). Finally, the dataset containing PBMCs (10x Genomics standard) was downloaded from the company platform (support.10xgenomics.com/single-cell-gene-expression/datasets/1.1.0/pbmc3k).

Raw gene expression matrices were analyzed with the Seurat package (v4.0)⁸⁰ for each study. We converted all gene symbols to Ensembl format (hg v38). Different filters of quality control were applied by technology: for 10x Genomics data the cutoff points were percentage of mitochondrial genes expressed <10, number of counts (nCounts) per cell >200, and ratio of nCounts by number of features (nFeatures) <5; for Smart-Seq2 were percentage of mitochondrial genes expressed <15, nCounts >200, and ratio nCounts/nFeature <1000; and for inDrop were percentage of mitochondrial genes expressed <15, nCount >200, and ratio nCounts/nFeature <1000. The remaining cells were submitted to the doublet removal step through the Scrublet package v0.2.3⁸¹. Through the doublet score histogram distribution, a cutoff value was determined for each library by manually setting the threshold (Optimal pK), eliminating cells with a higher probability of being doublets.

Integration and cluster annotation

The datasets were concatenated with Scanpy (v1.7.2)⁸², and the 3000 highly variably expressed genes were selected using flavor = “Seurat”. The integration was performed with scVI (v0.6.8)²⁸ from scvi-tools (v0.19.0)⁸³. For training the variational autoencoder neural network, we used the following hyperparameters: n_latent = 20, n_layers=4, dropout_rate = 0.1. After training the scVI model and correcting for batch effects using “study” and “tissue origin” as co-variables in order to remove the technical bias, the clusters were calculated through the Leiden algorithm⁸⁴. Different resolutions were evaluated, ranging from 0.3 to 2.0, and the resolution of 0.6 was selected for the annotation of broad cell types based on gene profiles. The latent space generated by scVI was then projected to a two-dimensional space using the UMAP dimensional reduction method⁸⁵. To determine the broad cell types, we use canonical markers (Supplementary Data 3). We found and removed three undefined/contaminated subsets with low quality (clusters 17, 19, and 29). After this first level of annotation, we applied DecontX⁸⁶, from the celda package (v1.14.1), using default parameters, to estimate and remove ambient RNA contamination in individual droplets. Post DecontX, we repeated the step of first-level cell type annotation, refining the cell type annotations after contamination removal. This approach was designed to ensure the preservation of clusters with minimal contamination and to maintain the integrity of genuine biological signals that could potentially be mistaken for contamination.

For the other levels of annotation, we subset the broad cell types and then rerun highly variable genes and scVI models with the same parameters, except for batch correction, which was performed by each sample. We also exclude samples with a contribution of fewer than ten cells. In order to classify clusters in cell types or states, we apply DEG analysis by Wilcoxon rank-sum and MAST v1.16.0⁸⁷, using logfc.threshold = 0.25, and min.pct = 0.1. The FDR method was used to adjust p-values. In addition, we evaluated the purity of the clusters through the ROGUE algorithm (v1.0)⁸⁸ with default parameters.

Integration benchmarking

We performed an Integration benchmark on macrophages data comparing six integration tools: Scanorama v1.7.4⁸⁹, Harmony v.1.2.0²⁷, BBKNN v.0.2.0⁹⁰, FastMNN v4.0⁹¹, scanVI²⁹ and scVI²⁸, that we used to integrate our dataset, and also the unintegrated data. For that we used the scib tool²⁶ and its metrics. Those metrics are described below:

Bio conservation metrics:

- Isolated labels: score how well isolated labels are distinguished from all other labels using the average-width silhouette score (ASW);

- KMeans NMI: the normalized mutual information (NMI) is a version of the mutual information corrected by the entropy of clustering and ground truth labels (e.g., cell type). The score ranges between 0 and 1, with 0 representing no sharing and 1 representing perfect sharing of information between clustering and annotated cell labels;

- KMeans ARI: the adjusted rand index (ARI) is a chance-adjusted rand index, which evaluates the pair-wise accuracy of clustering vs. ground truth label assignments. The score ranges between 0 and 1 with larger values indicating better conservation of the data-driven cell identity discovery after integration compared to annotated labels;

- Silhouette label: wrapper for sklearn silhouette function, including average silhouette width (ASW). Values range from [−1, 1] with 1 indicates distinct, compact clusters, 0 indicates overlapping clusters, and −1 indicates core-periphery (non-cluster) structure. By default, the score is scaled between 0 and 1.

Batch correction metrics:

- Silhouette batch: this metric measures the silhouette of a given batch. It assumes that a silhouette width close to 0 represents perfect overlap of the batches, thus the absolute value of the silhouette width is used to measure how well batches are mixed.;

- Graph connectivity: quantify the connectivity of the subgraph per cell type label.

Copy number variation inference

To differentiate between malignant and non-malignant epithelial cells, our study utilized an analysis of Copy Number Variations (CNVs). We estimated the CNV score for individual cells by applying inferCNVpy software (v0.4.0), available at github.com/icbi-lab/infercnvpy. This tool is an extension of the principles used in inferCNV⁹². We defined the malignancy status of epithelial cells based on their relative CNV scores that were calculated considering all cells in the dataset. Specifically, epithelial cells with a CNV score below the established median threshold of 0.004 were deemed “non-malignant.” In contrast, those exhibiting a CNV score above this median were classified as “malignant,” suggesting the presence of genomic changes commonly associated with cancer cells.

Unsupervised hierarchical clustering analysis

For the clustering analysis and dendrogram construction, we calculated the mean expression of the 750 most variable genes in MDCs using Seurat’s AverageExpression function. Pearson correlation coefficients were obtained and used to calculate the distance by the formula (1 - Pearson correlation)/2. We then constructed the dendrogram using the pvclust R package⁹³, with the correlation method set to “cor”, hierarchical clustering to “average”, and bootstrap replications to 10,000 for stability assessment. Jaccard bootstrap mean values were calculated with the fpc R package’s⁹⁴ cluster boot function, employing 10,000 bootstrap samples and average linkage clustering with k = 5 clusters.

Comparison with previously identified macrophage subpopulations

In order to validate our findings, we correlated the populations we have defined with those from previous works. Another set of pre-processed scRNA-Seq data was downloaded^20,21 (Supplementary Data 6). Those two datasets were selected because they are considered robust amounts of data, well-annotated regarding mononuclear phagocytes, and with a great diversity of tumor types, including those we have worked with in this paper. We used the same strategy and quality control filters that were used with our main datasets. The datasets were analyzed separately, using the authors’ annotations for macrophages, monocytes, and DC subpopulations. To perform the correlation, expression data was log-normalized using the NormalizedData() function of the Seurat package (v4.0)⁸⁰, followed by the AverageExpression function from the same package to determine the average expression of each gene for each subpopulation. Cor() function⁹⁵ from the stats R library was used to correlate the expression of all genes between subpopulations, and ggplot2 v3.4.4⁹⁶ was used to plot the correlation graphs. The purity of clusters was verified using ROGUE⁸⁸ with default parameters.

External validation: construction of a secondary pan-cancer atlas with updated datasets

Data download and pre-processing

Pre-processed single-cell RNA-sequencing (scRNA-Seq) data from various cancer types^{57,97,98,99,100,101,102,103,104,105,106,107,108} were acquired (Supplementary Data 7) from GEO using the GEOquery Bioconductor package. The data encompassed head and neck cancer (GSE181919), renal cell carcinoma (GSE159115), pancreatic adenocarcinoma (GSE205013), prostate cancer (GSE185344), endometrial cancer (GSE251923), and multiple datasets for breast cancer (GSE176078, GSE161529), ovarian cancer (GSE140819, GSE147082), gastric cancer (GSE163558), and thyroid cancer (GSE184362). Additional datasets for colorectal cancer (E-MTAB-8410) and more datasets for breast (EGAD00001006608) and ovarian cancers (EGAS00001004935) were retrieved from alternative databases.

For the creation of the validation pan-cancer atlas, each dataset was subjected to uniform filtering standards. Quality control parameters were adapted to each sequencing technology: for data generated by 10x Genomics, cells were retained if they displayed less than 10% mitochondrial gene expression, over 200 detected genes (nCounts), and a nCounts to nFeatures ratio of less than 5. For inDrop and Drop-seq technologies, the criteria were less than 15% mitochondrial gene expression, more than 200 nCounts, and a nCounts to nFeatures ratio below 1000. Subsequent to these filters, we applied the Scrublet package for doublet removal, setting the cutoff (Optimal pK) manually based on the distribution of doublet scores to exclude cells with a high probability of being doublets.

Integration and myeloid cells subset

For processing, samples originating from the same study were merged using the Scanpy software (version 1.7.2). We then identified the 3000 most variably expressed genes using the “Seurat” flavor setting for selection. Integration across datasets was carried out using scVI (version 0.6.8) as part of the scvi-tools suite (version 0.19.0). The variational autoencoder neural network was trained with specific hyperparameters: the number of latent dimensions was set to 20, the model was constructed with 4 layers, and a dropout rate of 0.1 was implemented to prevent overfitting. Batch effects correction was performed using each sample as co-variables. Cluster determination was subsequently performed using the Leiden algorithm. To visualize the data in a two-dimensional framework, we employed UMAP technique on the latent space generated by scVI.

In our analysis, we explored a range of resolutions from 0.3 to 2.0 to accurately cluster myeloid cells within various datasets. The evaluation revealed that a resolution setting of 0.3 suited most datasets effectively. Nonetheless, for certain studies—namely the Thyroid cancer¹⁰⁰, the Breast Cancer¹⁰⁴, and the Endometrial Cancer¹⁰⁹—alternate resolutions of 0.4, 0.4 and 0.6, respectively, provided better clustering results. Myeloid cell populations were identified using the canonical markers LYZ, AIF1, and CD68. To maintain consistency in gene annotations across all datasets, we standardized them to the human genome hg38 assembly.

Myeloid cell clusters were subsetted and the process was rerun for identifying highly variable genes and the scVI model adhering to the previously established parameters. Additionally, we removed samples contributing fewer than ten cells to refine our dataset. In order to classify mononuclear phagocytes, we apply DEG analysis by Wilcoxon rank-sum, using logfc.threshold = 0.25, and min.pct = 0.1. The FDR method was used to adjust p values (<0.05).

To confirm the robustness of the myeloid signatures identified, we integrated the discovery (reference) cohort with the validation cohort using the scANVI algorithm, which leveraged predefined cell type labels to facilitate cell type prediction. After integrating and confirming the cell types within the validation cohort, we recalculated the nearest neighbors and UMAP, and then conducted a reanalysis of gene expression, taking into account the previously identified differentially expressed genes.

Metabolic characterization

We characterized the metabolic pathways of each cluster using Compass v.0.9.7.1¹¹⁰. Cells were subsetted into clusters prior to running. The following parameters were used: --calc-metabolites --microcluster-size 10 --lambda 0. Raw reaction penalties were converted to reaction scores as described in the original publication, and reactions with zero variance were removed. Pairwise comparison was performed for each cluster (e.g., Cluster 1 vs. Cluster 2; Cluster 1 vs. Cluster n; etc) applying Wilcoxon’s test and calculating Cohen’s d effect size between means. The FDR method was used to adjust p-values. Reactions with confidence scores below 4 (most confident) or with an adjusted p-value greater than 0.1 were removed. Cohen’s d median scores were calculated for each pairwise comparison and metabolic pathway. Those median scores were again aggregated by median comparing the median Cohen’s d value across all clusters to obtain a single value per cluster and metabolic pathway, which were z-scored to compare across macrophage clusters.

Reaction scores calculated for each myeloid cluster were pairwise compared between all clusters. Reactions were filtered for adjusted p-value below 0.1 and a measure of the effect size was calculated using the median of Cohen’s values for each metabolic pathway. Next, Cohen’s medians for each metabolic pathway were summarized for each cluster to reflect the activity of the respective pathway in the cluster. The metabolic reactions Glycolysis/Gluconeogenesis, Citric Acid Cycle, NAD Metabolism, Oxidative Phosphorylation, Arginine and Proline Metabolism, Transport, Mitochondrial, Fatty Acid Oxidation, Fatty Acid Synthesis, Glycerophospholipid Metabolism, Glycosphingolipid Metabolism, Sphingolipid Metabolism are used to classified the macrophage based in the metabolic pathway.

Trajectory analysis through pseudo-time

We employed Monocle 3 (v.1.3.1)¹¹¹ with the UMAP generated by the integration with scVI as a “partition”. We set the parameter use_partition = FALSE of the function “learn graph” in order to create a linear trajectory, with default parameters. To order cells, the root used to infer the pseudotime was the Mono_CD14_FOS^- cluster, with default parameters. We also used Monocle 3 to perform gene module analysis based on graph autocorrelation to identify genes that co-vary over pseudotime.

Scores and pathway enrichment

The top hundred genes identified in the DEGs analysis, sorted by the log fold change for each cluster, were selected to perform the functional enrichment analysis with the curated database Reactome (2021)¹¹² through overrepresentation analysis (ORA), which consists of a one-sided Fisher’s exact test, using the clusterProfiler R package (v3.0)¹¹³. We considered significant pathways those with a Benjamini-Hochberg (BH) adjusted p value ≤0.05. The comparisons were made among the following cell types: Mac_Angio (n = 3108), Mac_Hypo (n = 1802), Mac_ES (n = 4073), Mac_IFN (n = 1981), Mac_Prolif (n = 2234), Mac_LA (n = 1249), Mac_AgPres (n = 3965), Mac_alv-like (n = 8816), RTM_like-MT (n = 1209), RTM_IFN (n = 2061), RTM_LA (n = 3569), RTM_Int (n = 3507). Some pathways were selected for plotting based on the scientific experience of the authors. The redundancy was removed from the enriched pathway set by selecting the pathway with the lowest p-value. All enriched pathways by subpopulations are available in Supplementary Data 5. The gene signature scores were computed by the AddModuleScore_UCell¹¹⁴ function. The gene set used for the scores is available in Supplementary Data 4.

Cell–cell communication

To perform cell communication analyses, we only considered cells detected with more than 200 genes and genes with detection in more than 50 cells. The CellComm algorithm v.1¹¹⁵ was used to infer communications between pairs of cell types including macrophages and tumors or lymphocytes. We used as a set of possible intercellular communications the 688 ligands and 857 receptors contained in the NicheNet database¹¹⁶ resulting in 12650 possible pairs. In communications between macrophages and tumors, we used cells classified as “MacAlv-like”, “Mac_Angio”, “Mac_Hypo”, “Mac_IFN”, “Mac_LA”, “Mac_ES”, “Mac_AgPres”, “RTM-like_MT”, RTM_IFN”, “RTM_Int” and “RTM_LA” and breast, colorectal, liver, lung, melanoma and ovarian tumor cells. In communications between macrophages and lymphocytes, we used cells classified as macrophages “Mac_IFN”, “Mac_LA”, “Mac_ES”, “RTM_IFN” and “RTM_Int” and lymphocytes “NKT”, “TCD4ex”, “TCD4reg”, “TCD8em”, “TCD8ex” and “TGD”. For consideration of intercellular communication, the expression of ligands and receptors should be greater than 0.25 normalized counts and show a p-value less than 0.01 in a permutation test of cluster annotation containing 1000 permutations. In order to display the main figures we applied the following filters: expression values of the ligands, receptors and mean of both higher than 1 and the p-value lower than 0.002. All interactions are shown in the Supplementary Data 9.

Deconvolution analysis

To comprehensively understand the heterogeneity of the TME, as well as the cellular composition in each tumor type, we applied deconvolution methods to predict the abundance of cell types by modeling gene expression levels through bulk RNA-Seq data. For these analyses, gene expression data (HTSeq counts), and clinical-pathological information from the database portal TCGA (http://cancergenome.nih.gov/) were downloaded using the TCGAbiolinks v.2.28¹¹⁷ package of nine different tumor types, using the identifiers: OV (n = 354 samples), LUAD (n = 510), LUSC (n = 496), UVM (n = 77), SKCM (divided into Metastatic (n = 385) and Primary (n = 103) tumor samples), COAD (n = 454), READ (n = 170), LIHC (n = 369) and BRCA (n = 1195) which comprises Luminal A (n = 569), Luminal B (n = 210), HER2 (n = 81) and TNBC (n = 188) subtypes. The single-cell signatures identified in this work was used as a reference using BayesPrism package¹¹⁸ (v2.0) to predict the relative frequency of cell types for each patient of each tumor type. We considered 47 cell signatures, including cell types identified in the first level of annotation, such as fibroblasts, endothelial cells, B lymphocytes, mast cells, pDC, malignant cells, and normal epithelial cells, and cell states/subtypes identified in other levels of clustering, including macrophages (n = 12, except the Mac_Prolif subpopulation), monocytes (n = 6), DC (n = 8), T cells (n = 9), NK cells (n = 2), and neutrophil (n = 3) subpopulations. The gene expression counts matrix was filtered to only contain the 12,854 genes that make up the signatures of the cell types. Only cell types representing more than 1% of in each tumor type were considered. Ribosomal and mitochondrial genes with high expression were removed using the cleanup.genes() function. In addition, genes expressed in less than 5 cells were excluded. After that, to get more consistent results, we filtered for protein-coding genes and used the function get.exp.stat() to select differentially expressed genes between cell states of different cell types using the default parameters, except for the parameter pseudo.count (0.1 for 10x data and 10 for Smart-seq data). Finally, we set up the prism object using the new.prism() function for each tumor type by setting the following parameters:

reference=scRNAfiltered_matrix,mixture=bulk_matrix,input.type = “count.matrix”,cell.type.labels = celltype, cell.state.labels = states, key = “Malignant Cells”, outlier.cut = 0.01, outlier.fraction

= 0.1. And then running BayesPrism, run.prism(prism = prism object, n.cores = 70).

Study design and patient selection

For the ovary cohort, 193 women diagnosed with HGSOC at INCA between 2001 and 2017, regardless of adjuvant or neoadjuvant treatment, were included in our study. The TNBC-INCA cohort, previously described by our collaborators¹¹⁹, consisted of 112 women who underwent neoadjuvant chemotherapy followed by curative surgery between 2010 and 2014. Clinical data regarding sex, age at diagnosis, staging, surgery, histological subtype, chemotherapy, and survival were retrospectively obtained from medical records. Both Brazilian cohorts are described in more detail in Supplementary Data 12. Patient sex information was sourced from medical records for both conditions.

Immunohistochemistry analysis

The blocks were cut with a microtome into fine slivers of 4-mm sections to form the tissue microarray which encompasses the most representative areas of greatest tumor cellularity in formalin-fixed paraffin-embedded tissue specimens. The immunohistochemistry was processed using a NOVOLINK™ Polymer Detection Systems (Leica), followed by 3,3′-Diaminobenzidine (DAB). Antigen retrieval was performed with the Trilogy (Cell Marque) reagent. The samples were immunostained for Rabbit Monoclonal anti-Ki67 (clone 30-9 at 1:7, Ventana- Roche), Rabbit monoclonal anti-CD8 (cat. 108R-14, clone SP57 at 1:7, Ventana- Roche), Mouse monoclonal anti-PD-1 (cat. 315M-94, clone NAT105, Cell Marque, diluted 1:100), Mouse monoclonal anti-CD68 (cat. 168 M-95, clone KP-1 at 1:20, Ventana- Roche), Mouse monoclonal anti-FOLR2 (cat. MA5-26933, OTI4G6 at 1:200, ThermoFisher), Rabbit monoclonal anti-TREM2 (cat. 91068T, clone D814C, 1:100, Cell Signaling Technology) and Rabbit polyclonal anti-PDL-2 (ab200377, at 1:200, Abcam). For double staining, TREM2 was combined with anti-CD68, and anti-FOLR2 with anti-PDL-2. Briefly, after completing the first immune reaction (TREM2 or PDL-2) with DAB, the second immune reaction (CD86 or FOLR2) was visualized using MACH4 MR-AP (Biocare Medical) green subtract (Polydetector HRP Green Substrate-chromogen Bio-SB - ref BSB0129t). The entire analysis was carried out at the Division of Pathology - INCA by experienced pathologists and the slides were scanned by Aperio ImageScope v12.4.6.5003.

Tissue microarrays were assessed by determining the percent of immune cells and their intra or peritumoral distribution under the supervision of a qualified pathologist. The cutoffs for positive staining were established based on positive and negative controls, as well as her experience as a pathologist. These cutoffs were chosen based on their clinical and pathological significance (Table 1).

Table 1 Cutoffs established for high and low expression groups in immunohistochemistry analysis

Full size table

FOLR2 and the TREM2 were accessed by ImageJusing the IHC Profiler ImageJ Plugin¹²⁰.The separation of the color layers was based on hematoxylin and 3,3′-Diaminobenzidine. For each patient, we calculated the average of the percentage values obtained from ImageJ for all the spots. Then, the score for each average value was calculated based on the IHC Profiler tutorial, as follows: percentage of high positive (4 × % area of high positive staining), positive (3 × % area of positive staining), and low positive (2 × % area of low positive staining) staining. Finally, the score per sample was defined as the higher value among them.

The expression of the co-markers FOLR2 and PDL-2 in TNBC-INCA and HGSOC-INCA was measured by HALO software version 3.6 (Indica Labs, Albuquerque, NM, USA). Spots were segmented using the TMA module and positive cells (displaying both markers) were quantified via the Multiplex IHC module by defining thresholds for each marker.

In the immunohistochemistry analysis of the TNBC-INCA cohort, each patient was represented by three tissue spots, whereas in the HGSOC-INCA cohort, each patient was represented by two spots. The analysis involved calculating the average staining intensity per patient, ensuring that the data represented a robust measure of the marker being studied across multiple tissue samples.

Survival and statistical analysis

To assess differences in means (scores and clinical impact), we employed the Kruskal-Wallis method for overall comparisons, complemented by the Wilcoxon test for specific paired comparisons. Survival analysis was visualized using the Kaplan-Meier method, facilitated by the Survival (v3.5) and Survminer (v0.4.9) packages^121,122. Branches were defined using two approaches depending on how the data was collected. The first approach was the use of the “surv_cutpoint” function from the “survminer” package to define groups. This function determines optimal cutpoints for continuous variables that are used to stratify patients into different risk groups. The “surv_cutpoint” function applies a maximally selected rank statistics method to identify the best-performing threshold based on survival outcomes, ensuring an unbiased separation of groups despite the uneven distribution of sample sizes. This approach was used to define groups in the following cases:

Macrophage subpopulations estimations in bulk RNA-Seq samples from TCGA that were derived through deconvolution analysis. The cut-off points for Mac_LA, as determined by “surv_cutpoint”, were a score of 0.11 for TNBC samples and 3.26 for HGSOC samples. As for RTM_Int, the cut-off was 9.67 for TNBC samples and 0.0078 for HGSOC samples.
FOLR2 and TREM2 global levels quantified using the ImageJ software with the IHC Profiler plugin. The cut-off points determined by “surv_cutpoint” were scores equal to 71.5 and 83.8 for TNBC-INCA samples when observing global expression levels of FOLR2 and TREM2, respectively.
Tissue microarray slides of HGSOC-INCA and TNBC-INCA patients stained with hematoxylin, eosin, and co-marked with PDL-2 and FOLR2 analyzed using HALO software version 3.6 (Indica Labs, Albuquerque, NM, USA). The cut-offs determined by “surv_cutpoint” were 7.5 cells for HGSOC samples and 53.5 for TNBC samples.

For the evaluation of double staining of TREM2 and CD68, the quantification of markers was performed under the supervision of a qualified pathologist. The cutoffs for positive staining were established based on positive and negative controls, as well as her experience as a pathologist. These cutoffs were chosen based on their clinical and pathological significance and described in the Immunohistochemistry section. Hazard ratios for survival analysis were computed using both univariate and multivariate Cox proportional hazards models. Statistical testing throughout the study was conducted on a two-sided basis. To address the challenge of multiple hypothesis testing, the Benjamini–Hochberg procedure was applied to adjust P values, and False Discovery Rate (FDR) Q values were subsequently calculated. Results were considered statistically significant at P value or FDR Q value of <0.05.

Regarding the clinical parameters used, the clinical annotation of TCGA dataset, and the times calculated for OS and PFS were downloaded from the PanCanAtlas study. For the INCA cohorts, OS was calculated considering the day from the diagnosis to the date of death of any cause if the patient was known to be alive on the last day of data collection. The PFS was calculated from the last chemotherapy data or the last registration to recurrence data if the patient was known to be alive or the date of death. The Kaplan-Meier method was used to visualize the difference between the curves of the high and low groups. The statistical analyses were conducted using R project version 3.5.3.

Inclusion and ethics

The research process involved local researchers in all stages, and all collaborators meeting the authorship criteria mandated by Nature Communication journals have been included as authors, given their vital roles in designing and executing the study. Roles and responsibilities were established among collaborators prior to the research. The study incorporates locally relevant findings, identified in collaboration with local partners. Ethical approval was sought from local ethics review committees where appropriate. Research restrictions or prohibitions in the researchers’ setting did not apply, and participant risks such as stigmatization, incrimination, or discrimination were avoided. Our citations acknowledge the relevance of local and regional research to our study.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The published datasets related to the main analysis can be accessed under GEO accession numbers: GSE114727, GSE140228, GSE125449, GSE115978, GSE154600, GSE130973, GSE139829, GSE158803; URLs for additional data are www.synapse.org/#!Synapse:syn21041850/wiki/600865 and lambrechtslab.sites.vib.be/en/dataaccess, as well as support.10xgenomics.com/single-cell-gene-expression/datasets/1.1.0/pbmc3k. The public datasets related to the validation analysis can be accessed under GEO accession numbers: GSE181919, GSE159115, GSE205013, GSE251923, GSE185344, GSE176078, GSE161529, GSE140819, GSE147082, GSE163558, GSE184362; European Genome-phenome Archive study IDs EGAD00001006608 [ega-archive.org/studies/EGAS00001004809], EGAS00001004935 [ega-archive.org/studies/EGAS00001004935] and ArrayExpress accession number E-MTAB-8410 (see also Supplementary Data 1 and 6). The integrated dataset is publicly available and can be downloaded via CELL×GENE | Collections through the following link: cellxgene.cziscience.com/collections/3f7c572c-cd73-4b51-a313-207c7f20f188. Source data are provided with this paper.

Code availability

The code for reproducing the analyses and figures of this paper is available in the GitHub repository: github.com/bioinformatics-inca/MyeloidDerivedCells_in_TME¹²³.

References

Waldman, A. D., Fritz, J. M. & Lenardo, M. J. A guide to cancer immunotherapy: from T cell basic science to clinical practice. Nat. Rev. Immunol. 20, 651–668 (2020).
Article CAS PubMed PubMed Central Google Scholar
Sharma, P. et al. The next decade of immune checkpoint therapy. Cancer Discov. 11, 838–857 (2021).
Article CAS PubMed Google Scholar
Greene, J. T., Brian, B. F. 4th, Senevirathne, S. E. & Freedman, T. S. Regulation of myeloid-cell activation. Curr. Opin. Immunol. 73, 34–42 (2021).
Article CAS PubMed PubMed Central Google Scholar
Mantovani, A., Marchesi, F., Jaillon, S., Garlanda, C. & Allavena, P. Tumor-associated myeloid cells: diversity and therapeutic targeting. Cell. Mol. Immunol. 18, 566–578 (2021).
Article CAS PubMed PubMed Central Google Scholar
Zhou, X., Liu, X. & Huang, L. Macrophage-mediated tumor cell phagocytosis: opportunity for nanomedicine intervention. Adv. Funct. Mater. 31, 2006220 (2021).
Article CAS PubMed Google Scholar
Hanahan, D. Hallmarks of cancer: new dimensions. Cancer Discov 12, 31–46 (2022).
Article CAS PubMed Google Scholar
Rizzo, G. et al. Dynamics of monocyte-derived macrophage diversity in experimental myocardial infarction. Cardiovasc. Res. https://doi.org/10.1093/cvr/cvac113 (2022).
Nalio Ramos, R. et al. Tissue-resident FOLR2+ macrophages associate with CD8+ T cell infiltration in human breast cancer. Cell 185, 1189–1207.e25 (2022).
Article CAS PubMed Google Scholar
Ma, R.-Y., Black, A. & Qian, B.-Z. Macrophage diversity in cancer revisited in the era of single-cell omics. Trends Immunol. 43, 546–563 (2022).
Article CAS PubMed Google Scholar
Larionova, I. et al. Interaction of tumor-associated macrophages and cancer chemotherapy. Oncoimmunology 8, 1596004 (2019).
Article PubMed PubMed Central Google Scholar
Patel, A. A., Ginhoux, F. & Yona, S. Monocytes, macrophages, dendritic cells and neutrophils: an update on lifespan kinetics in health and disease. Immunology 163, 250–261 (2021).
Article CAS PubMed PubMed Central Google Scholar
Aegerter, H., Lambrecht, B. N. & Jakubzick, C. V. Biology of lung macrophages in health and disease. Immunity 55, 1564–1580 (2022).
Article CAS PubMed PubMed Central Google Scholar
Kübler, K. et al. Prognostic significance of tumor-associated macrophages in endometrial adenocarcinoma. Gynecol. Oncol. 135, 176–183 (2014).
Article PubMed Google Scholar
Gordon, S. R. et al. PD-1 expression by tumour-associated macrophages inhibits phagocytosis and tumour immunity. Nature 545, 495–499 (2017).
Article ADS CAS PubMed PubMed Central Google Scholar
Yuan, X. et al. Prognostic significance of tumor-associated macrophages in ovarian cancer: a meta-analysis. Gynecol. Oncol. 147, 181–187 (2017).
Article CAS PubMed Google Scholar
Mantovani, A., Biswas, S. K., Galdiero, M. R., Sica, A. & Locati, M. Macrophage plasticity and polarization in tissue repair and remodelling. J. Pathol. 229, 176–185 (2013).
Article CAS PubMed Google Scholar
Sanchez, L. R. et al. The emerging roles of macrophages in cancer metastasis and response to chemotherapy. J. Leukoc. Biol. 106, 259–274 (2019).
Article CAS PubMed Google Scholar
Kim, H. J., Ji, Y. R. & Lee, Y. M. Crosstalk between angiogenesis and immune regulation in the tumor microenvironment. Arch. Pharm. Res. 45, 401–416 (2022).
Article CAS PubMed PubMed Central Google Scholar
Li, M., He, L., Zhu, J., Zhang, P. & Liang, S. Targeting tumor-associated macrophages for cancer treatment. Cell Biosci 12, 85 (2022).
Article PubMed PubMed Central Google Scholar
Mulder, K. et al. Cross-tissue single-cell landscape of human monocytes and macrophages in health and disease. Immunity 54, 1883–1900.e5 (2021).
Article CAS PubMed Google Scholar
Cheng, S. et al. A pan-cancer single-cell transcriptional atlas of tumor infiltrating myeloid cells. Cell 184, 792–809.e23 (2021).
Article CAS PubMed Google Scholar
Villani, A.-C. et al. Single-cell RNA-seq reveals new types of human blood dendritic cells, monocytes, and progenitors. Science 356, eaah4573 (2017).
Article PubMed PubMed Central Google Scholar
Tamoutounour, S. et al. Origins and functional specialization of macrophages and of conventional and monocyte-derived dendritic cells in mouse skin. Immunity 39, 925–938 (2013).
Article CAS PubMed Google Scholar
Doebel, T., Voisin, B. & Nagao, K. Langerhans cells—the macrophage in dendritic cell clothing. Trends Immunol. 38, 817–828 (2017).
Article CAS PubMed Google Scholar
Bian, Z. et al. Deciphering human macrophage development at single-cell resolution. Nature 582, 571–576 (2020).
Article ADS CAS PubMed Google Scholar
Luecken, M. D. et al. Benchmarking atlas-level data integration in single-cell genomics. Nat. Methods 19, 41–50 (2022).
Article CAS PubMed Google Scholar
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with harmony. Nat. Methods 16, 1289–1296 (2019).
Article CAS PubMed PubMed Central Google Scholar
Lopez, R., Regier, J., Cole, M. B., Jordan, M. I. & Yosef, N. Deep generative modeling for single-cell transcriptomics. Nat. Methods 15, 1053–1058 (2018).
Article CAS PubMed PubMed Central Google Scholar
Xu, C. et al. Probabilistic harmonization and annotation of single-cell transcriptomics data with deep generative models. Mol. Syst. Biol. 17, e9620 (2021).
Article PubMed PubMed Central Google Scholar
Sjoelund, V., Smelkinson, M. & Nita-Lazar, A. Phosphoproteome profiling of the macrophage response to different toll-like receptor ligands identifies differences in global phosphorylation dynamics. J. Proteome Res. 13, 5185–5197 (2014).
Article CAS PubMed PubMed Central Google Scholar
Daniels, C. M. et al. Dynamic ADP-ribosylome, phosphoproteome, and interactome in LPS-activated macrophages. J. Proteome Res. 19, 3716–3731 (2020).
Article CAS PubMed PubMed Central Google Scholar
He, L. et al. Global characterization of macrophage polarization mechanisms and identification of M2-type polarization inhibitors. Cell Rep. 37, 109955 (2021).
Article CAS PubMed PubMed Central Google Scholar
Liberzon, A. et al. The molecular signatures database (MSigDB) hallmark gene set collection. Cell Syst. 1, 417–425 (2015).
Article CAS PubMed PubMed Central Google Scholar
Etich, J. et al. Gene expression profiling of the extracellular matrix signature in macrophages of different activation status: relevance for skin wound healing. Int. J. Mol. Sci. 20, 5086 (2019).
Article CAS PubMed PubMed Central Google Scholar
Katzenelenbogen, Y. et al. Coupled scRNA-Seq and intracellular protein activity reveal an immunosuppressive role of TREM2 in cancer. Cell 182, 872–885.e19 (2020).
Article CAS PubMed Google Scholar
Ho, C.-C. et al. TREM-1 expression in tumor-associated macrophages and clinical outcome in lung cancer. Am. J. Respir. Crit. Care Med. 177, 763–770 (2008).
Article CAS PubMed Google Scholar
Yang, F.-C., Chiu, P.-Y., Chen, Y., Mak, T. W. & Chen, N.-J. TREM-1-dependent M1 macrophage polarization restores intestinal epithelium damaged by DSS-induced colitis by activating IL-22-producing innate lymphoid cells. J. Biomed. Sci. 26, 46 (2019).
Article PubMed PubMed Central Google Scholar
Sheng, J. et al. A discrete subset of monocyte-derived cells among typical conventional type 2 dendritic cells can efficiently cross-present. Cell Rep. 21, 1203–1214 (2017).
Article CAS PubMed Google Scholar
Ryu, Y., Han, G. H., Jung, E. & Hwang, D. Integration of single-cell RNA-Seq datasets: a review of computational methods. Mol. Cells 46, 106–119 (2023).
Article CAS PubMed PubMed Central Google Scholar
Sikkema, L. et al. An integrated cell atlas of the lung in health and disease. Nat. Med. 29, 1563–1577 (2023).
Article CAS PubMed PubMed Central Google Scholar
Zheng, X. et al. Single-cell analyses implicate ascites in remodeling the ecosystems of primary and metastatic tumors in ovarian cancer. Nat Cancer 4, 1138–1156 (2023).
Article CAS PubMed PubMed Central Google Scholar
Jones, C. V. & Ricardo, S. D. Macrophages and CSF-1: implications for development and beyond. Organogenesis 9, 249–260 (2013).
Article PubMed PubMed Central Google Scholar
Arroyo Portilla, C., Tomas, J., Gorvel, J.-P. & Lelouard, H. From species to regional and local specialization of intestinal macrophages. Front. Cell Dev. Biol. 8, 624213 (2020).
Article PubMed Google Scholar
Dutertre, C.-A. et al. Single-cell analysis of human mononuclear phagocytes reveals subset-defining markers and identifies circulating inflammatory dendritic cells. Immunity 51, 573–589.e8 (2019).
Article CAS PubMed Google Scholar
Sariban, E., Luebbers, R. & Kufe, D. Transcriptional and posttranscriptional control of c-fos gene expression in human monocytes. Mol. Cell. Biol. 8, 340–346 (1988).
CAS PubMed PubMed Central Google Scholar
Huber, R. et al. Regulation of monocyte differentiation by specific signaling modules and associated transcription factor networks. Cell. Mol. Life Sci. 71, 63–92 (2014).
Article CAS PubMed Google Scholar
Lord, K. A., Abdollahi, A., Hoffman-Liebermann, B. & Liebermann, D. A. Proto-oncogenes of the fos/jun family of transcription factors are positive regulators of myeloid differentiation. Mol. Cell. Biol. 13, 841–851 (1993).
CAS PubMed PubMed Central Google Scholar
Chen, J.-Y. et al. Association of FCGR3A and FCGR3B copy number variations with systemic lupus erythematosus and rheumatoid arthritis in Taiwanese patients. Arthritis Rheumatol. 66, 3113–3121 (2014).
Article CAS PubMed PubMed Central Google Scholar
Lavin, Y. et al. Tissue-resident macrophage enhancer landscapes are shaped by the local microenvironment. Cell 159, 1312–1326 (2014).
Article CAS PubMed PubMed Central Google Scholar
Martinez, F. O. & Gordon, S. The M1 and M2 paradigm of macrophage activation: time for reassessment. F1000Prime Rep. 6, 13 (2014).
Article PubMed PubMed Central Google Scholar
Chen, Y., Hu, M., Wang, L., Chen, W. Macrophage M1/M2 polarization. Eur. J. Pharmacol. 877, 173090 (2020).
Mills, C. D., Kincaid, K., Alt, J. M., Heilman, M. J. & Hill, A. M. M-1/M-2 macrophages and the Th1/Th2 paradigm. J. Immunol. 164, 6166–6173 (2000).
Article CAS PubMed Google Scholar
Casanova-Acebes, M. et al. Tissue-resident macrophages provide a pro-tumorigenic niche to early NSCLC cells. Nature 595, 578–584 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Hashimoto, D. et al. Tissue-resident macrophages self-maintain locally throughout adult life with minimal contribution from circulating monocytes. Immunity 38, 792–804 (2013).
Article CAS PubMed Google Scholar
Blériot, C., Chakarov, S. & Ginhoux, F. Determinants of resident tissue macrophage identity and function. Immunity 52, 957–970 (2020).
Article PubMed Google Scholar
Zhang, Q. et al. Landscape and dynamics of single immune cells in hepatocellular carcinoma. Cell 179, 829–845.e20 (2019).
Article CAS PubMed Google Scholar
Hornburg, M. et al. Single-cell dissection of cellular components and interactions shaping the tumor immune phenotypes in ovarian cancer. Cancer Cell 39, 928–944.e6 (2021).
Article CAS PubMed Google Scholar
Schwarz, H. et al. The role of triggering receptor expressed on myeloid cells 2 in Parkinson’s disease and other neurodegenerative disorders. Behav. Brain Res. 433, 113977 (2022).
Article CAS PubMed Google Scholar
Nakamura, K. & Smyth, M. J. TREM2 marks tumor-associated macrophages. Signal Transduct Target. Ther. 5, 233 (2020).
Article CAS PubMed PubMed Central Google Scholar
Wolf, E. M., Fingleton, B. & Hasty, A. H. The therapeutic potential of TREM2 in cancer. Front. Oncol. 12, 984193 (2022).
Article CAS PubMed PubMed Central Google Scholar
Binnewies, M. et al. Targeting TREM2 on tumor-associated macrophages enhances immunotherapy. Cell Rep. 37, 109844 (2021).
Article CAS PubMed Google Scholar
White, E. S., Strom, S. R., Wys, N. L. & Arenberg, D. A. Non-small cell lung cancer cells induce monocytes to increase expression of angiogenic activity. J. Immunol. 166, 7549–7555 (2001).
Article CAS PubMed Google Scholar
Noe, J. T. & Mitchell, R. A. MIF-dependent control of tumor immunity. Front. Immunol. 11, 609948 (2020).
Article CAS PubMed PubMed Central Google Scholar
Molgora, M. et al. TREM2 modulation remodels the tumor myeloid landscape enhancing Anti-PD-1 immunotherapy. Cell 182, 886–900.e17 (2020).
Article CAS PubMed PubMed Central Google Scholar
Zhang, X., Wang, W., Li, P., Wang, X. & Ni, K. High TREM2 expression correlates with poor prognosis in gastric cancer. Hum. Pathol. 72, 91–99 (2018).
Patnaik, A. et al. A phase 1a dose-escalation study of PY314, a TREM2 (triggering receptor expressed on macrophages 2) targeting monoclonal antibody. J. Clin. Oncol. 40, 2648–2648 (2022).
Article Google Scholar
Kurahara, H. et al. Clinical significance of folate receptor β-expressing tumor-associated macrophages in pancreatic cancer. Ann. Surg. Oncol. 19, 2264–2271 (2012).
Article PubMed Google Scholar
Tie, Y. et al. Targeting folate receptor β positive tumor-associated macrophages in lung cancer with a folate-modified liposomal complex. Signal Transduct. Target. Ther. 5, 6 (2020).
Article CAS PubMed PubMed Central Google Scholar
Moharil, P. et al. Engineering a folic acid-decorated ultrasmall gemcitabine nanocarrier for breast cancer therapy: dual targeting of tumor cells and tumor-associated macrophages. Acta Pharm. Sin. B 12, 1148–1162 (2022).
Article CAS PubMed Google Scholar
Mojic, M., Takeda, K. & Hayakawa, Y. The dark side of IFN-γ: its role in promoting cancer immunoevasion. Int. J. Mol. Sci. 19, 89 (2017).
Article PubMed PubMed Central Google Scholar
Kiely, P. A., Sant, A. & O’Connor, R. RACK1 is an insulin-like growth factor 1 (IGF-1) receptor-interacting protein that can regulate IGF-1-mediated Akt activation and protection from cell death. J. Biol. Chem. 277, 22581–22589 (2002).
Article CAS PubMed Google Scholar
Das, S. et al. A novel computational predictive biological approach distinguishes Integrin β1 as a salient biomarker for breast cancer chemoresistance. Biochim. Biophys. Acta Mol. Basis Dis. 1869, 166702 (2023).
Article CAS PubMed Google Scholar
Zhu, J. et al. A novel 12-gene prognostic signature in breast cancer based on the tumor microenvironment. Ann. Transl. Med. 10, 143 (2022).
Article PubMed PubMed Central Google Scholar
Yu, Y. et al. Single-cell profiling reveals the trajectory of FOLR2-expressing tumor-associated macrophages to regulatory T cells in the progression of lung adenocarcinoma. https://doi.org/10.21203/rs.3.rs-2427294/v1 (2023).
Scaranti, M., Cojocaru, E., Banerjee, S. & Banerji, U. Exploiting the folate receptor α in oncology. Nat. Rev. Clin. Oncol. 17, 349–359 (2020).
Article PubMed Google Scholar
Rodriguez-Garcia, A. et al. CAR-T cell-mediated depletion of immunosuppressive tumor-associated macrophages promotes endogenous antitumor immunity and augments adoptive immunotherapy. Nat. Commun. 12, 877 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Davis, S. & Meltzer, P. S. GEOquery: a bridge between the Gene Expression Omnibus (GEO) and BioConductor. Bioinformatics 23, 1846–1847 (2007).
Article PubMed Google Scholar
Qian, J. et al. A pan-cancer blueprint of the heterogeneous tumor microenvironment revealed by single-cell profiling. Cell Res. 30, 745–762 (2020).
Article CAS PubMed PubMed Central Google Scholar
Travaglini, K. J. et al. A molecular cell atlas of the human lung from single-cell RNA sequencing. Nature 587, 619–625 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587.e29 (2021).
Article CAS PubMed PubMed Central Google Scholar
Wolock, S. L., Lopez, R. & Klein, A. M. Scrublet: computational identification of cell doublets in single-cell transcriptomic data. Cell Syst. 8, 281–291.e9 (2019).
Article CAS PubMed PubMed Central Google Scholar
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Article PubMed PubMed Central Google Scholar
Gayoso, A. et al. A Python library for probabilistic analysis of single-cell omics data. Nat. Biotechnol 40, 163–166 (2022).
Traag, V. A., Waltman, L. & van Eck, N. J. From Louvain to Leiden: guaranteeing well-connected communities. Sci. Rep. 9, 5233 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Sainburg, T., McInnes, L. & Gentner, T. Q. Parametric UMAP Embeddings for Representation and Semisupervised Learning. Neural. Comput. 33, 2881–2907 (2021).
Yang, S. et al. Decontamination of ambient RNA in single-cell RNA-seq with DecontX. Genome Biol. 21, 57 (2020).
Article PubMed PubMed Central Google Scholar
Finak, G. et al. MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data. Genome Biol. 16, 278 (2015).
Article PubMed PubMed Central Google Scholar
Liu, B. et al. An entropy-based metric for assessing the purity of single-cell populations. Nat. Commun. 11, 3155 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Hie, B., Bryson, B. & Berger, B. Efficient integration of heterogeneous single-cell transcriptomes using Scanorama. Nat. Biotechnol. 37, 685–691 (2019).
Article CAS PubMed PubMed Central Google Scholar
Polański, K. et al. BBKNN: fast batch alignment of single-cell transcriptomes. Bioinformatics 36, 964–965 (2020).
Article PubMed Google Scholar
Haghverdi, L., Lun, A. T. L., Morgan, M. D. & Marioni, J. C. Batch effects in single-cell RNA-sequencing data are corrected by matching mutual nearest neighbors. Nat. Biotechnol. 36, 421–427 (2018).
Article CAS PubMed PubMed Central Google Scholar
Tickle, T., Tirosh, I., Georgescu, C., Brown, M. & Haas, B. inferCNV of the Trinity CTAT Project. (Klarman Cell Observatory, Broad Institute of MIT and Harvard, 2019).
Suzuki, R. & Shimodaira, H. Pvclust: an R package for assessing the uncertainty in hierarchical clustering. Bioinformatics 22, 1540–1542 (2006).
Hennig, C. Calibrating and aggregating cluster validity indexes for context-adapted comparison of clusterings. Stat. Comput. 30, 1523–1544 (2020).
Becker, R. A., Chambers, J. M. & Wilks, A. R. The New S Language. (Wadsworth and Brooks/Cole Advanced Books & Software, Monterey, CA, United States, 1988).
Wickham, H. ggplot2: Elegant Graphics for Data Analysis (Springer, 2016).
Choi, J.-H. et al. Single-cell transcriptome profiling of the stepwise progression of head and neck cancer. Nat. Commun. 14, 1–13 (2023).
Article ADS CAS Google Scholar
Werba, G. et al. Single-cell RNA sequencing reveals the effects of chemotherapy on human pancreatic adenocarcinoma and its tumor microenvironment. Nat. Commun. 14, 1–16 (2023).
ADS Google Scholar
Wong, H. Y. et al. Single-cell analysis of cribriform prostate cancer reveals cell intrinsic and tumor microenvironmental pathways of aggressive disease. Nat. Commun. 13, 1–21 (2022).
Article ADS CAS Google Scholar
Pu, W. et al. Single-cell transcriptomic analysis of the tumor ecosystems underlying initiation and progression of papillary thyroid carcinoma. Nat. Commun. 12, 1–18 (2021).
Article ADS Google Scholar
Pal, B. et al. A single-cell RNA expression atlas of normal, preneoplastic and tumorigenic states in the human breast. EMBO J 40, e107333 (2021).
Article CAS PubMed PubMed Central Google Scholar
Jiang, H. et al. Revealing the transcriptional heterogeneity of organ-specific metastasis in human gastric cancer using single-cell RNA Sequencing. Clin. Transl. Med. 12, e730 (2022).
Article CAS PubMed PubMed Central Google Scholar
Lee, H.-O. et al. Lineage-dependent gene expression programs influence the immune landscape of colorectal cancer. Nat. Genet. 52, 594–603 (2020).
Article CAS PubMed Google Scholar
Bassez, A. et al. A single-cell map of intratumoral changes during anti-PD1 treatment of patients with breast cancer. Nat. Med. 27, 820–832 (2021).
Article CAS PubMed Google Scholar
Wu, S. Z. et al. A single-cell and spatially resolved atlas of human breast cancers. Nat. Genet. 53, 1334–1347 (2021).
Article CAS PubMed PubMed Central Google Scholar
Slyper, M. et al. A single-cell and single-nucleus RNA-Seq toolbox for fresh and frozen human tumors. Nat. Med. 26, 792–802 (2020).
Article CAS PubMed PubMed Central Google Scholar
Olalekan, S., Xie, B., Back, R., Eckart, H. & Basu, A. Characterizing the tumor microenvironment of metastatic ovarian cancer by single-cell transcriptomics. Cell Rep. 35, 109165 (2021).
Article CAS PubMed Google Scholar
Zhang, Y., Narayanan, S. P. & Mannan, R. Single-cell analyses of renal cell cancers reveal insights into tumor microenvironment, cell of origin, and therapy response. PNAS 118, e2103240118 (2021).
Article CAS PubMed PubMed Central Google Scholar
Yu, X. et al. Integrating single-cell RNA-seq and spatial transcriptomics reveals MDK-NCL-dependent immunosuppressive environment in endometrial carcinoma. Front. Immunol. 14, 1145300 (2023).
Article CAS PubMed PubMed Central Google Scholar
Wagner, A. et al. Metabolic modeling of single Th17 cells reveals regulators of autoimmunity. Cell 184, 4168–4185.e21 (2021).
Article CAS PubMed PubMed Central Google Scholar
Cao, J. et al. The single-cell transcriptional landscape of mammalian organogenesis. Nature 566, 496–502 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Fabregat, A. et al. The reactome pathway knowledgebase. Nucleic Acids Res. 46, D649–D655 (2018).
Article CAS PubMed Google Scholar
Yu, G., Wang, L.-G., Han, Y. & He, Q.-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 16, 284–287 (2012).
Article CAS PubMed PubMed Central Google Scholar
Andreatta, M. & Carmona, S. J. UCell: robust and scalable single-cell gene signature scoring. Comput. Struct. Biotechnol. J. 19, 3796–3798 (2021).
Article CAS PubMed PubMed Central Google Scholar
Lummertz da Rocha, E. et al. CellComm infers cellular crosstalk that drives haematopoietic stem and progenitor cell development. Nat. Cell Biol. 24, 579–589 (2022).
Article CAS PubMed Google Scholar
Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat. Methods 17, 159–162 (2020).
Article CAS PubMed Google Scholar
Colaprico, A. et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. Nucleic Acids Res. 44, e71 (2016).
Article PubMed Google Scholar
Chu, T., Wang, Z., Pe’er, D. & Danko, C. G. Cell type and gene expression deconvolution with BayesPrism enables Bayesian integrative analysis across bulk and single-cell RNA sequencing in oncology. Nat. Cancer 3, 505–517 (2022).
Article CAS PubMed PubMed Central Google Scholar
da Silva, J. L. et al. Prognostic influence of residual tumor-infiltrating lymphocyte subtype after neoadjuvant chemotherapy in triple-negative breast cancer. Front. Oncol. 11, 636716 (2021).
Article PubMed PubMed Central Google Scholar
Varghese, F., Bukhari, A. B., Malhotra, R. & De, A. IHC Profiler: an open source plugin for the quantitative evaluation and automated scoring of immunohistochemistry images of human tissue samples. PLoS One 9, e96801 (2014).
Article ADS PubMed PubMed Central Google Scholar
Kassambara, A., Kosinski, M., Biecek, P. & Fabian, S. Package ‘survminer’. Drawing Survival Curves using’.
Therneau, T. M. Survival Analysis [R package survival version 3.5-5] (2023).
Guimarães, G. R. et al. bioinformatics-inca/MyeloidDerivedCells_in_TME: 2024. Zenodo https://doi.org/10.5281/zenodo.11236423 (2024).

Download references

Acknowledgements

The authors are indebted to the patients and their families for their participation and for providing biological material for research purposes. We extend our gratitude to Nicole Scherer from Bioinformatics Core Facility, Diego José Gomes de Paula and Luciana Medeiros de Castro from National Tumor Bank, and the Division of Pathology, all located at the Brazilian National Cancer Institute (INCA-RJ), for their continuous support. This work was supported by INCA/MS fellowships, Chan Zuckerberg Initiative - Inflammation Network (M.A.M., M.B., and P.M.M.M.V.), and Fundação de Amparo à Pesquisa do Estado do Rio de Janeiro (FAPERJ - E-26/205.530/2022 (284921), E-26/201.322/2022 (272260), E-26/211.648/2021 (269502), and E-26/210.302/2022 (270388) for M.B.). We thank the São Paulo Research Foundation (FAPESP grants 2018/18886-9, 2020/01688-0, and 2020/07069-0 to O.C.M., and 2023/07806-2 to I.S.F.), the Brazilian National Council for Scientific and Technological Development (CNPq) (grants: 309482/2022-4 to O.C.M., 312898/2023-1 to R.N.R. and 310376/2022-0 to M.B.), Serrapilheira (grant R-2111-39828 for R.N.R and R-2111-39726 to E.L.R.) and Fundação de Amparo à Pesquisa e Inovação de Santa Catarina (grant FAPESC FAB2019121000026 to E.L.R.) for their financial support. This publication is part of the Human Cell Atlas—www.humancellatlas.org/publications.

Author information

These authors contributed equally: Gabriela Rapozo Guimarães, Giovanna Resk Maklouf.

Authors and Affiliations

Laboratory of Bioinformatics and Computational Biology, Division of Experimental and Translational Research, Brazilian National Cancer Institute (INCA), Rio de Janeiro, RJ, Brazil
Gabriela Rapozo Guimarães, Giovanna Resk Maklouf, Cristiane Esteves Teixeira, Leandro de Oliveira Santos, Nayara Gusmão Tessarollo, Nayara Evelin de Toledo, Alessandra Freitas Serain, Cristóvão Antunes de Lanna, Marco Antônio Pretti, Jéssica Gonçalves Vieira da Cruz & Mariana Boroni
Department of Microbiology, Immunology, and Parasitology, Federal University of Santa Catarina, Florianópolis, SC, Brazil
Marcelo Falchetti, Mylla M. Dimas & Edroaldo Lummertz da Rocha
Department of Immunology, Institute of Biomedical Sciences, University of São Paulo,(USP), São Paulo, Brazil
Igor Salerno Filgueiras, Otavio Cabral-Marques & Rodrigo Nalio Ramos
Instituto D’Or de Ensino e Pesquisa, São Paulo, Brazil
Otavio Cabral-Marques & Rodrigo Nalio Ramos
Department of Medicine, Division of Molecular Medicine, Laboratory of Medical Investigation 29, School of Medicine, University of São Paulo (USP), São Paulo, Brazil
Otavio Cabral-Marques
Laboratory of Medical Investigation in Pathogenesis and Directed Therapy in Onco-Immuno-Hematology (LIM-31), Departament of Hematology and Cell Therapy, Hospital das Clínicas HCFMUSP, School of Medicine, University of São Paulo (USP), São Paulo, Brazil
Rodrigo Nalio Ramos
Division of Pathology, Brazilian National Cancer Institute (INCA), Rio de Janeiro, RJ, Brazil
Fabiane Carvalho de Macedo, Fabiana Resende Rodrigues & Nina Carrossini Bastos
Division of Clinical Research and Technological Development, Brazilian National Cancer Institute (INCA), Rio de Janeiro, RJ, Brazil
Jesse Lopes da Silva, Cláudia Bessa Pereira Chaves & Andreia Cristina de Melo
Gynecologic Oncology Section, Brazilian National Cancer Institute (INCA), Rio de Janeiro, RJ, Brazil
Cláudia Bessa Pereira Chaves
Laboratory of Immunometabolism, Department of Genetics, Evolution, Microbiology, and Immunology, Institute of Biology, Universidade Estadual de Campinas, Campinas, SP, Brazil
Pedro M. M. Moraes-Vieira
Obesity and Comorbidities Research Center (OCRC), Universidade Estadual de Campinas, Campinas, SP, Brazil
Pedro M. M. Moraes-Vieira & Marcelo A. Mori
Experimental Medicine Research Cluster (EMRC), Universidade Estadual de Campinas, Campinas, SP, Brazil
Pedro M. M. Moraes-Vieira & Marcelo A. Mori
Laboratory of Aging Biology, Department of Biochemistry and Tissue Biology, Universidade Estadual de Campinas, Campinas, SP, Brazil
Marcelo A. Mori

Authors

Gabriela Rapozo Guimarães
View author publications
You can also search for this author in PubMed Google Scholar
Giovanna Resk Maklouf
View author publications
You can also search for this author in PubMed Google Scholar
Cristiane Esteves Teixeira
View author publications
You can also search for this author in PubMed Google Scholar
Leandro de Oliveira Santos
View author publications
You can also search for this author in PubMed Google Scholar
Nayara Gusmão Tessarollo
View author publications
You can also search for this author in PubMed Google Scholar
Nayara Evelin de Toledo
View author publications
You can also search for this author in PubMed Google Scholar
Alessandra Freitas Serain
View author publications
You can also search for this author in PubMed Google Scholar
Cristóvão Antunes de Lanna
View author publications
You can also search for this author in PubMed Google Scholar
Marco Antônio Pretti
View author publications
You can also search for this author in PubMed Google Scholar
Jéssica Gonçalves Vieira da Cruz
View author publications
You can also search for this author in PubMed Google Scholar
Marcelo Falchetti
View author publications
You can also search for this author in PubMed Google Scholar
Mylla M. Dimas
View author publications
You can also search for this author in PubMed Google Scholar
Igor Salerno Filgueiras
View author publications
You can also search for this author in PubMed Google Scholar
Otavio Cabral-Marques
View author publications
You can also search for this author in PubMed Google Scholar
Rodrigo Nalio Ramos
View author publications
You can also search for this author in PubMed Google Scholar
Fabiane Carvalho de Macedo
View author publications
You can also search for this author in PubMed Google Scholar
Fabiana Resende Rodrigues
View author publications
You can also search for this author in PubMed Google Scholar
Nina Carrossini Bastos
View author publications
You can also search for this author in PubMed Google Scholar
Jesse Lopes da Silva
View author publications
You can also search for this author in PubMed Google Scholar
Edroaldo Lummertz da Rocha
View author publications
You can also search for this author in PubMed Google Scholar
Cláudia Bessa Pereira Chaves
View author publications
You can also search for this author in PubMed Google Scholar
Andreia Cristina de Melo
View author publications
You can also search for this author in PubMed Google Scholar
Pedro M. M. Moraes-Vieira
View author publications
You can also search for this author in PubMed Google Scholar
Marcelo A. Mori
View author publications
You can also search for this author in PubMed Google Scholar
Mariana Boroni
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

Conceptualization: M.B., M.A.M., P.M.M.M.V. Single-cell integration, and cells annotation: G.R.M., G.R.G. Pathways analysis: M.A.P., NET Pseudo-time analysis: G.R.G. Benchmarking analysis: L.O.S., C.A.L. External Validation: G.R.G., C.E.T., C.A.L., N.G.T. Cell-Cell communication analysis: M.F., M.M.D., L.O.S., E.L.R. Deconvolution and clinical association: C.E.T., N.G.T., G.R.M., G.R.G. Immunohistochemistry analysis: N.G.T., I.S.F., O.C.M., R.N.R., N.C.B., F.C.M., F.R.R., C.E.T., J.G.V.C., A.F.S., C.A.L. Cohort obtention: C.B.P.C., A.C.M., J.L.S., N.G.T., N.E.T. Funding acquisition: M.B., M.A.M., P.M.M.M.V. Supervision: M.B., M.A.M., P.M.M.M.V. Writing – original draft: G.R.G., G.R.M., N.G.T., C.E.T., L.O.S., N.E.T., C.A.L. Writing – review & editing: M.B., M.A.M., P.M.M.M.V., A.C.M., E.L.R.

Corresponding author

Correspondence to Mariana Boroni.

Ethics declarations

Competing interests

During the preparation of the manuscript M.A.P. was hired by Egle Therapeutics. The other authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Jo Van Ginderachterand the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Peer Review File

Description of Additional Supplementary Files

Dataset 1

Dataset 2

Dataset 3

Dataset 4

Dataset 5

Dataset 6

Dataset 7

Dataset 8

Dataset 9

Dataset 10

Dataset 11

Dataset 12

Reporting Summary

Source data

Source Data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Guimarães, G.R., Maklouf, G.R., Teixeira, C.E. et al. Single-cell resolution characterization of myeloid-derived cell states with implication in cancer outcome. Nat Commun 15, 5694 (2024). https://doi.org/10.1038/s41467-024-49916-4

Download citation

Received: 31 July 2023
Accepted: 19 June 2024
Published: 07 July 2024
DOI: https://doi.org/10.1038/s41467-024-49916-4

Subjects

Abstract

Similar content being viewed by others

Introduction

Results

A thorough integration strategy recovers an in-depth pan-cancer repertoire of MDCs in the tumor microenvironment

MDCs are heterogeneous and phenotypically diverse across tumor samples

A plethora of macrophage phenotypes and functional states co-exist in the TME

Deciphering the clinical impact of TREM2+ macrophages in the TME

MDC subpopulations display distinct clinical outcomes depending on their niche

Discussion

Methods

Ethical considerations

Data download and pre-processing

Integration and cluster annotation

Integration benchmarking

Copy number variation inference

Unsupervised hierarchical clustering analysis

Comparison with previously identified macrophage subpopulations

External validation: construction of a secondary pan-cancer atlas with updated datasets

Data download and pre-processing

Integration and myeloid cells subset

Metabolic characterization

Trajectory analysis through pseudo-time

Scores and pathway enrichment

Cell–cell communication

Deconvolution analysis

Study design and patient selection

Immunohistochemistry analysis

Survival and statistical analysis

Inclusion and ethics

Reporting summary

Data availability

Code availability

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding author

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Supplementary information

Source data

Rights and permissions

About this article

Cite this article

Share this article

Search

Quick links

Deciphering the clinical impact of TREM2⁺ macrophages in the TME