Although the majority of patients with advanced lung adenocarcinoma (LUAD) are eligible to receive immune checkpoint blockade, approximately 80% of these tumors are resistant to this therapeutic approach. Insights at the single-cell level into mechanisms that drive LUAD tumorigenesis and the relationship of LUAD histologic heterogeneity to response to immune checkpoint blockade could help identify biomarkers and potential combinational approaches to improve immunotherapy efficacy. Here, we used a genetically engineered mouse model that replicates the development of human LUAD through a spectrum of preinvasive to invasive adenocarcinoma histologic subtypes. A systems onco-immunology approach of integrating the analytical power and unique, complementary capabilities of time-of-flight mass cytometry (CyTOF) and imaging mass cytometry was leveraged to identify cellular and spatial immune contextures in LUAD. Comprehensive investigation of mouse and human LUAD using these single-cell proteomics platforms showed that LUAD progression is associated with spatiotemporal evolution of tumor-associated macrophages in the tumor-immune microenvironment, which governs tumor response to immunotherapy. PD-1 was expressed in a highly plastic tumor-promoting subtype of tumor-associated macrophages that develops during tumor progression from preinvasive to invasive adenocarcinoma, controls the lymphocyte-depleted niche of invasive tumors, and protects tumor cells in the solid histologic components of the tumor. Longitudinal, multidimensional single-cell analyses of LUAD tumorigenesis revealed dynamic alteration of immunoregulatory PD-1–expressing tumor-associated macrophages that can be targeted to overcome resistance to checkpoint immunotherapy.

Significance:

Comprehensive single-cell proteomics analyses of lung adenocarcinoma progression reveal the role of tumor-associated macrophages in resistance to PD-1 blockade therapy.

See related commentary by Lee et al., p. 2515

Lung cancer is responsible for the most cancer-related deaths in the United States and only 21% of patients who receive this diagnosis are alive five years later (1). Immune checkpoint inhibitors (ICI) against PD-1 and PD-L1 have shown encouraging clinical results and tolerable safety in patients with advanced non–small cell lung cancer (NSCLC; refs. 2–8). Unfortunately, the majority of NSCLCs are resistant to ICI therapy and the rate of response is only about 20% (8, 9).

Lung adenocarcinoma (LUAD) is the most common form of lung cancer. In recent years, it has become evident that LUAD is actually comprised of a number of histologic subsets with variable biologic behavior. Among them, several share lepidic pattern growth in which tumor cell proliferation occurs along the preexisting framework of lung alveolar structures, initially without destruction of its architecture. Preinvasive lesions called adenocarcinoma in situ (AIS) progress to minimally invasive adenocarcinoma (MIA) and to other invasive subtypes including lepidic predominant adenocarcinoma (LPA) and solid subtype adenocarcinoma (10, 11). A number of studies have reported better prognosis in patients with AIS and MIA forms of LUAD, and worse prognosis in patients with invasive subtypes of LUAD including LPA, solid, and others (10, 12, 13). There remains a lack of deep insight into mechanisms that drive progression of AIS to invasive adenocarcinoma at the single-cell level, and the relationship of the histologic heterogeneity of LUAD to response of ICI therapy is poorly understood.

To address these gaps, we used a genetically engineered mouse model that replicates the development of human lung adenocarcinoma through a spectrum of preinvasive (AIS) to invasive adenocarcinoma histologic subtypes (11). We leveraged a systems onco-immunology approach of integrating the analytical power and unique, complementary capabilities of time-of-flight mass cytometry (CyTOF) and imaging mass cytometry (IMC) to identify cellular and tissue architectural determinants of an immunologic state that is conducive to productive anti-tumor immunity in LUAD. Through longitudinal analyses, we discovered the emergence of population of immuno-regulatory, tumor-protective macrophages identified by their expression of PD-1, and that are phenotypically, functionally, and spatially dynamic. We show that these cells are intricately related to the tumor architecture of LUAD subtypes, compromise antitumor immunity, and are associated with maintenance of the immune-depleted milieu of invasive LUAD. Targeting tumor-associated macrophages (TAM) decreased the extent of invasive lepidic and solid LUAD histology, and combining the macrophage-targeting agent with PD-1 blockade further improved these therapeutic responses, particularly for solid subtype LUAD. These data provide a novel spatiotemporal immunologic and histologic framework to approach the treatment of patients suffering from LUAD.

Mouse model

Animal experiments were performed in accordance with the Institutional Animal Care and Use Committees and in accordance with all relevant animal use guidelines and ethical regulations at Baylor College of Medicine (IRB. No. AN-6685). We utilized a doxycycline-inducible EGFRL858R bi-transgenic mouse model of lung adenocarcinoma driven by an organ-specific human EGFR gene mutation as previously described (11, 14). Both strains of mice were gifts from Dr. Katerina Politi, Yale School of Medicine (New Haven, CT) and are of FVB background. The mutant human EGFR (L858R) transgene was activated by feeding male bitransgenic (CCSP/EGFR) mice with doxycycline-impregnated food pellets (625 ppm; Harlan Laboratories). This mouse model results in progressive development of lung adenocarcinomas with preinvasive lepidic histology that progress to invasive adenocarcinoma in both whole lungs (11, 14), and is very similar to the progression of these tumors in humans. Histologic changes are present in the lungs as early as two weeks after starting their doxycycline diet, resembling human lepidic type of lung adenocarcinomas with mainly diffuse thickening of the alveolar walls (11). Invasive adenocarcinomas develop five weeks after doxycycline feeding, and discontinuing the doxycycline diet induces tumor regression (11).

Human samples

This study was performed in accordance with an Institutional Review Board protocol at Baylor College of Medicine (IRB. No. H-42636). A prospectively maintained single institution database was retrospectively reviewed. Twenty patients with LUAD who received nivolumab or pembrolizumab from 2015 to 2019 were evaluated. All anonymous human samples were collected with written informed patient consent.

Single-cell preparations

Tumor-bearing lungs were finely minced and digested in unsupplemented RPMI1640 (without glutamate) using a mouse Tumor Dissociation Kit (Miltenyi Biotec Inc., cat. no. 130–096–730) in 50 mL Falcon tubes for 20 minutes in the 37°C rotating incubator. Cells were then filtered through a 70-μm cell strainer (Corning Life Sciences Plastic, cat. no. 07201431), washed, and lysed in ACK lysing buffer (Life Technologies, cat. no. A049201). After centrifuging cells with unsupplemented RPMI media for 5 minutes at 400 × g in 20°C and washing twice, supernatants were carefully suctioned off. The cell pellet underwent a final washing with 40 mL supplemented cell culture media (RPMI with FBS), and cells were refiltered through a 70-μm cell strainer. After centrifuging the cells for 5 minutes at 400 × g in 20°C, the supernatant was carefully suctioned off. Cells were placed in freezing media (FBS with 7% DMSO) and cryopreserved in −80°C freezer storage after cell counting with a cell counter (Countess II FL, Life Technologies, cat. no. AMQAF1000). For long-term preservation, the cryovials were transferred into a liquid nitrogen tank at −196°C.

Flow cytometry and cell sorting

To sort cancer cells, PD-1HIGH TAMs, and PD-1LOW TAMs from single-cell suspensions of mouse lung adenocarcinoma, cells were incubated with TruStain FcX (anti-mouse CD16/32) antibodies (BioLegend, cat. no. 422301) as an Fc Receptor Blocking Solution. The panel of antibodies used in the fluorescence-activated cell sorting (FACS) of macrophages included anti-mouse CD45-BUV-395 (clone 30-F11, BD Biosciences, cat. no. 564279), CD11b-BV711 (clone M1/70, BioLegend, cat. no. 101242), F4/80-APC (clone BM8, BioLegend, cat. no. 123116), CD3-AF488 (clone 17A2, BioLegend, cat. no. 100212), CD19-AF488 (clone 6D5, BioLegend, cat. no. 115524), and PE-conjugated PD-1 antibody. To generate PE-conjugated PD-1 antibody, unconjugated anti-mouse PD-1 antibodies (clone 29F.1A12, Bio X Cell, cat. no. BE0273) were conjugated to DIBO-modified R-phycoerythrin (R-PE) using the SiteClick Antibody Labeling Kit (Invitrogen, cat. no. S10467) following the manufacturer's protocol. Dead cells were excluded using 4,6-diamidino-2-phenylindole dihydrochloride (DAPI; Sigma-Aldrich). Anti-human EGFR was purchased from BioLegend (352906). Cell viability assay was performing with Annexin V (BioLegend, cat. no. 640922) and DAPI (Life Technologies, cat. no. D1306) staining. Flow cytometry was performed by an LSR Fortessa cytometer equipped with FACS Diva software (BD Biosciences). Data were analyzed using FlowJo (FlowJo, LLC). PD-1HIGH TAMs (CD45+F4/80+CD11b+PD-1+), PD-1LOW TAMs (CD45+F4/80+CD11b+PD-1−), and EGFR-mutant epithelial cells (CD45-human EGFR+) were sorted from single lung suspensions of bitransgenic mice fed doxycycline for 7 weeks using a FACSAria I Cell Sorter (BD Biosciences).

CyTOF experiments

Single cells were stabilized for 6 hours in media at 37°C. Five hundred thousand cells were resuspended in Maxpar Cell Staining Buffer (Fluidigm, cat. no. 201068) in individual 5 mL tubes for each sample to be barcoded. Mass-tag cellular barcoding using the Cell-ID 20-Plex Pd Barcoding Kit (Fluidigm, cat. no. 201060) was performed as described previously (15). Cell-ID Intercalator-Ir is a cationic nucleic acid intercalator that contains naturally abundant Iridium (191Ir and 193Ir) and is used for identifying nucleated cells in CyTOF analysis according to standard protocol. For measurement of intracellular cytokines by CyTOF, cells from lungs with tumor-bearing mice were incubated in 1 μL/mL Golgistop (BD cell analysis, cat. no. BDB554724A), 2 μL/mL PMA/ionomycin (eBioscience Cell Stimulation Cocktail, cat. no. 00–4970–93) for 6 hours at 37°C, according to standard protocol. The samples were then washed and incubated with cell surface antibodies for 30 minutes at room temperature and washed. After overnight at 4°C with resuspension in 1× Fix I buffer, the samples were stained with intracellular antibodies against cells cytokines for 30 minutes at room temperature and washed. Stained cells were analyzed on a mass cytometer (CyTOF3 mass cytometer, Fluidigm) at an event rate of 400 to 500 cells per second. All mass cytometry files were normalized together using the mass cytometry data normalization algorithm (16), which uses the intensity values of a sliding window of these bead standards to correct for instrument fluctuations over time and between samples.

CyTOF analysis

The scheme of data analysis is illustrated in Supplementary Fig. S1. Total live nucleated cells were used for all analyses and visualized using Spanning-tree Progression Analysis of Density-normalized Events (SPADE)–based SCAFFOLD (Single-Cell Analysis by Fixed Force- and Landmark-Directed) map generation as a mixture of human guided knowledge and automated clustering algorithm (17, 18). To characterize all cells obtained from harvested lungs, all 733 nodes, or clusters of single cells with similar phenotypes, were organized in 17 phenotypes. Eighteen cellular phenotypes were manually defined by a panel of 38 antibodies (Supplementary Fig. S2): B cells (CD45+CD19+B220+), CD4 T cells (CD45+CD3+TCR-β+CD4+), CD8 T cells (CD45+CD3+TCR-β+CD8+), double negative (DN) T cells, (CD45+CD3+TCR-β+CD4−CD8−), γδT cells CD45+CD3+TCR-β−TCR-γδ+), monocytes (CD45+CD3−CD11b+CD115+), TAMs (CD45+CD3−CD64+F4/80+SiglecF−), alveolar macrophages (AM; CD45+CD3−CD64+F4/80+SiglecF+), plasmacytoid dendritic cells (CD45+CD3−CD19−CD11c+B220+CD317+), conventional DC (CD45+CD3−CD19−CD11c+B220−MHCII+), neutrophils (CD45+CD3−CD19−CD11b+Ly-6G+), natural killer (NK) cells (CD45+CD3−CD19−CD64−CD49b+), cancer cells [CD45−Pan-cytokeratin (CK)+Ki-67+], cancer-associated fibroblasts (CD45−Pan-CK−Vim+), epithelial cells (CD45−Pan-CK−EpCAM+), endothelial cells (CD45−Pan-CK−CD31+), and stromal cells (CD45−Pan-CK−Vim−EpCAM−CD31−). Furthermore, through deep phenotyping of lung macrophages, AMs and TAMs were distinctly separated by SiglecF expression. AMs were divided into typical SiglecF+CD11c+CD11b− AMs with low PD-1 expression and monocyte-derived AMs (SiglecF+CD11c+CD11b+). TAMs were divided into SiglecF−CD11b+CD11c− MHCIIhigh TAMs with low PD-1 expression (PD-1LOW TAMs) and SiglecF−CD11b+CD11c−MHCIIlow TAMs with high PD-1 expression (PD-1HIGH TAMs).

Based on SPADE results, we generated hierarchical inferences combined with the representative node. A cluster of similar patterns of protein expression was regarded as the same phenotype, and the most fitted node to each phenotype was regarded as the representative node. After assignment of all nodes to each phenotype, nodes were connected and displayed with SCAFFOLD maps for creating a reference map from high-dimensional single-cell data, facilitating comparisons across samples (19). SCAFFOLD maps were generated as previously reported (19). Briefly, a graph is constructed by connecting together the nodes representing the manually gated landmark populations and then connecting to them the nodes representing the cell clusters as well as connecting the clusters to one another. Each node is associated with a vector containing the median marker values of the cells in the cluster (unsupervised nodes) or gated populations (landmark nodes). Edge weights are defined as the cosine similarity between these vectors after comparing the results from the implementation of several distance metrics. Each circle represents a node, a population of cells with the same pattern of expression, and the size of the node represents the percentage of cells in each experiment. The clusters for all the tissues are combined in a single graph with edge weights defined as the cosine similarity between the vectors of median marker values of each cluster. All the pair-wise distances are calculated. The graph is then laid out using the ForceAtlas2 algorithm in Gephi 0.9.1 (https://gephi.org). In this representation of the cellular immune system of the mouse lung, each node is a population of cells with a similar pattern of protein expression, and the size of the node correlates with the average number of cells in the corresponding cell population. To overlay the additional samples on the SCAFFOLD map, the position and identity of the landmark nodes were fixed and the clusters of each sample were connected to the landmark nodes as described above.

Imaging mass cytometry

The formalin-fixed paraffin-embedded (FFPE) tissues were sectioned at a 5-μm thickness for IMC. FFPE tissues on charged slides were stained with 1:100 diluted antibody cocktails (concentration of each antibody = 0.5 mg/mL) as recommended by the user's manual. The slides were scanned in the Hyperion Imaging System (Fluidigm). They were scanned at least four regions of interest in >4 mm2 width at 200 Hz. By applying the novel approach of IMC to complex immunological systems such as the tumor microenvironment, multiplexed measurements of up to 35 protein expressions in FFPE tissue samples can be analyzed all at once in a single experiment. The value in IMC with FFPE as a multiplex platform lies in its efficiency and practicality in displaying tissue architecture. We are able to investigate the cellular phenotypes and biomarkers of a single tissue sample, observe each components’ spatial distribution simultaneously, and compare the qualities of each. This unique and comprehensive analysis of a single tumor sample affords us to draw conclusions between the identification and spatial relationships of the significant cellular components within a tumor microenvironment. The samples targeted for study are FFPE tissues that make this kind of architectural analysis easy to handle and highly convenient for use in the lab. We require some other image processing and computational methods in order to quantify the .mcd imaging files (Fluidigm) yielded from IMC and to determine the overall spatial architecture of the tumor immune microenvironment (20). We introduce the free, open-source, image-analyzing platform Fiji as the best resolution to fulfill this need. The advantages offered by Fiji (21) include a fast, robust, unsupervised, automated cell segmentation method requiring minimal expertise in computers. Thirty-two-bit TIFF stacked images were loaded in Fiji and we applied our novel method of automated cell segmentation that estimates cell boundaries by expanding the perimeter of their nuclei, identified by Cell ID Intercalator-iridium (191Ir). Once images from the IMC methodology are acquired, they can be quantified through Fiji's threshold and watershed tools. Protein expression data were then extracted at the single-cell level through mean intensity multiparametric measurements performed on individual cells (22), and acquired single-cell data were transferred into additional cytometric analysis in FlowJo and Cytosplore (23, 24), a software for interactive immune cell phenotyping for large single-cell datasets.

Spatial Cancer Evolution map

Quantified IMC data are adjusted with 191Ir and 193Ir intensities and normalized with CytoNorm (25) across IMC ROIs. CytoNorm is a normalization method for cytometry data applicable to large clinical studies. CytoNorm allows reducing mass cytometry signal variability across multiple batches of barcoded samples. Normalized IMC data are combined by using FlowJo. To visualize temporal trajectories during tumor progression, the Uniform Manifold Approximation and Projection (UMAP) for dimensional reduction is applied (26).

Three-dimensional coculture system of TAMs and cancer cells

We performed the three-dimensional coculture system of TAMs and cancer cells to form spheroid structures. For in vitro spheroid formation assay using a 3D coculture system with Nunclon Sphera 96-Well, Nunclon Sphera-Treated, U-Shaped-Bottom Microplate (Thermo Fisher Scientific, cat. no. 174925), single cells obtained from invasive adenocarcinoma were cultured in medium containing 2% Matrigel and seeded onto the plate precoated with Matrigel. In this model, we detected spheroid formation at 7 days. Spheroid size was evaluated by measuring spheroid perimeter using Fiji software (NIH, Bethesda), which area was converted into % relative area based on the size of the intact spheroid. Results are expressed as mean pixel measure ± SEM from three independent experiments in four replicates. Pictures for spheroid size evaluation were taken with a Zeiss Axiovert 200/M-based phase-contrast microscope.

Treatment

Trabectedin (Yondelis, Manufactured by Baxter Oncology GmbH) was purchased for research and was dissolved in 0.9% sodium chloride, USP to 2 mmol/L and kept at −20°C. Bitransgenic mice carrying human EGFR transgene were fed doxycycline impregnated food pellets and trabectedin (0.15 mg/kg/body weight) was administrated intravenously 3 times biweekly for adenocarcinoma in situ or weekly for invasive adenocarcinoma after starting their doxycycline diet. After completion of the trabectedin treatment, the doxycycline diet was continued until day 21 for measurement of the tumor burden. A total of 50 μL of anti-mouse PD-1 Abs (10 mg/kg, InVivoMAb anti-mouse PD-1 (CD279), clone 29F.1A12, Bio X Cell, cat. no. BE0273) diluted with InVivoPure pH 7.0 Dilution Buffer (cat. no. IP0070, Bio X Cell) are injected intraperitoneally. As a control group, mouse IgG isotype (InVivoMAb rat IgG2a isotype control, anti-trinitrophenol, Clone 2A3, Bio X Cell, BE0089) was used.

MTT assay

All isolated cells were cultured in RPMI media as recommended by the providers in a humidified incubator at 37°C and 5% CO2. The cytotoxicity of trabectedin in all three types of cells was determined by MTT assay (CellTiter 96 Non-Radioactive Cell Proliferation Assay, Promega). Compactly, cells (1 × 104 cells/well) were seeded in a 96-well plate with fresh medium and incubated overnight. MTT solution was added into each well at the indicated time points for 24 hours with 9 different concentrations of trabectedin diluted in fresh medium. As a positive control, cells were cultured in fresh medium, and medium only was used as a negative control. Cell viability was evaluated at different time points following the manufacturer's instructions through the spectrophotometric reading at two different wavelengths (570 and 650 nm). All experiments were repeated in at least triplet.

Detailed experimental methods of protein profiling by mass spectrometry are described in the supplementary documents.

Statistical analysis

Student t tests, χ2 tests, and Fisher exact tests were used to compare the data. All in vitro experiments were performed at least in triplicate. Differences between treatment groups were analyzed by the two-tailed Student t test using Prism 5.0 (GraphPad Software, Inc) or SPSS 28.0 (SPSS, Inc.). Mean mass intensities (MMI) of proteins were compared with t tests accordingly among phenotypes before and after treatment. To generate the statistical values of phenotypic differences between two subsets, the internodal differences in the same phenotypes were analyzed using Significance Analysis of Microarrays according to the percentages of cell frequency in corresponding nodes (27). Biometric Research Branch-Array Tools 4.3.0 (National Cancer Institute, Bethesda, MD) was used for statistical analysis of the internodal differences. To compare the longitudinal alteration of protein markers between macrophage subtypes, we performed analysis of variance (ANOVA) test of difference of proteins of interest. CyTOF data were additionally analyzed with FlowJo V10 software (FlowJo, LLC) to confirm consistency of results by CyTOF analysis. Statistical significance was accepted for P < 0.05, and all tests were two-tailed.

Data availability

Expression profile data analyzed in this study were obtained from NCBI's Gene Expression Omnibus (GEO) database at GSE52248 (48), GSE58772 (49), GSE93157 (51), and GSE136961 (52). Protein profiling data generated in this study are available within the article and its supplementary data files. The CyTOF and IMC data generated in this study are available upon request from the corresponding author.

The tumor-immune microenvironment transforms during histologic progression of invasive lung adenocarcinoma

Lung adenocarcinoma exists along a continuum of preinvasive to invasive histology (10). We hypothesized that the evolving histologic stages of LUAD harbor distinct tumor-immune microenvironments (TiME) that differentially regulate response to immunotherapy. To test this hypothesis, we used a genetically engineered mouse model that replicates the development of human LUAD (Fig. 1A; ref. 11) and compared the TiME of progressing tumors through early lepidic (largely preinvasive), invasive lepidic, and invasive solid adenocarcinoma, using a 38-channel CyTOF platform (Supplementary Figs. S1 and S2; Supplementary Table S1). Ten major phenotypes of immune cells were overlaid by T-distributed stochastic neighbor embedding (t-SNE; ref. 28) and demonstrated striking differences between the TiME of these tumors (Fig. 1B). B-cell and T-cell populations diminished steadily over the course of tumor development (Fig. 1C). With progression to invasive LUAD, CD8 T cells acquired an immunoregulatory phenotype with high expression of TIM-3 and PD-1 and became functionally exhausted, decreasing production of IFNγ and TNFα in response to in vitro stimulation (Supplementary Fig. S3), while the frequency of regulatory T cells steadily increased (Supplementary Fig. S4).

We found the majority of macrophages in the normal lung to be AMs (CD64+F4/80+SiglecF+CD11c+) and the remaining macrophages to be of the interstitial phenotype (IMs, CD64+F4/80+SiglecF−CD11c−; refs. 29–32). As we have previously shown in this model (14), the total macrophage population (CD45+CD64+F4/80+) more than doubled during the development of invasive LUAD. The proportion of AMs, however, contracted, and the majority of macrophages were replaced by CD11b-expressing macrophages not present in the nontumor-bearing lung, referred to hereafter as TAMs (CD64+F4/80+CD11b+SiglecF−; Fig. 1D).

To understand these changes within the context of the evolving TiME of LUAD, we applied a semisupervised approach and constructed SCAFFOLD maps (Fig. 1E; Supplementary Figs. S1 and S2; refs. 19, 27). Briefly, the SPADE unsupervised clustering algorithm (17) was applied to enumerate 1,472 cellular subpopulations, each defined by its unique protein expression and mapped based on their relative similarities to 14 landmark populations defined by prior knowledge of the immune system (Fig. 1E). Early lepidic histology adenocarcinoma contained abundant CD4 effector T cells (CD4 Teff), CD8 effector T cells (CD8 Teff), and B cells, which contracted in invasive solid adenocarcinoma that contained TAMs predominantly (Fig. 1E). These data indicate that a series of potential immune evasive mechanisms evolve in parallel to histologic tumor progression in LUAD, including reduction of lymphocytes, increase in the number of dysfunctional CD8 T cells, and expansion of a potential tumor-promoting macrophage population present only in tumors.

PD-1 distinguishes highly plastic TAM subtypes in lung adenocarcinoma

Given the striking increase in TAMs in invasive LUAD, we scrutinized the TAM phenotype by hierarchical unsupervised clustering of the 449 nodes within the TAM gate (CD64+F4/80+CD11b+SiglecF−). Two subtypes of TAMs were segregated by expression of PD-1 (Fig. 2A; Supplementary Fig. S5). Compared with PD-1HIGH TAMs, PD-1LOW TAMs demonstrated high expression of MHC-II and PD-L1. PD-1HIGH TAMs highly expressed the proliferation markers, Ki-67 and IdU (5-Iodo-2′-deoxyuridine; Fig. 2A) and also highly expressed the chemokine receptor CX3CR1 (Fig. 2B), suggesting a proliferative, tissue-resident ontology (33–35).

In normal, nontumor-bearing lungs, PD-1HIGH macrophages (interstitial macrophages) were rare (≤3% of all macrophages) and PD-1LOW macrophages comprised 17.3 ± 15.4% of all macrophages. In tumor-bearing lungs, both PD-1HIGH and PD-1LOW TAMs increased during tumor progression to invasive LUAD, and PD-1HIGH TAMs constituted 37.5 ± 10.9% of all macrophages at 6 weeks of tumorigenesis when invasive lepidic and solid invasive histology is present (Supplementary Fig. S6). To visualize longitudinal changes of PD-1HIGH TAMs and PD-1LOW TAMs during tumor progression, we used UMAP for dimensional reduction (26). Early in tumorigenesis during the development of lepidic histology LUAD, PD-1HIGH TAMs and PD-1LOW TAMs had divergent phenotype. Late in tumorigenesis during the development of invasive solid histology LUAD, PD-1HIGH TAMs were juxtaposed with PD-1LOW TAMs on UMAP plots, implying more similar protein expression (Fig. 2C). To more completely understand the evolution of phenotype and function of PD-1HIGH TAMs and PD-1LOW TAMs during LUAD progression, we investigated cytokine production and phospho-signaling in TAMs using CyTOF. In early tumorigenesis, PD-1HIGH TAMs increased their expression of Arg1, cMyc, Ki-67, vimentin, mTOR, and phospho-NF-κβ, then decreased expression of these proteins as tumor histology became invasive. PD-L1, MHC-II, CD44, and CD80 expression on PD-1HIGH also declined dramatically at later time points of tumor progression (Fig. 2D).

We next performed protein profiling of sorted PD-1HIGH TAMs and PD-1LOW TAMs early and late in tumorigenesis using mass spectrometry (Supplementary Table S2). In early lepidic LUAD, PD-1HIGH TAMs upregulated signaling pathways associated with mitochondrial dysfunction, activated oxidative phosphorylation, the TCA cycle, and the JAK/Stat pathway (Fig. 2E), similar to the metabolic profile seen in IL4-activated macrophages (36). In contrast, the protein profile of PD-1HIGH TAMs isolated in later tumorigenesis were engaged in different metabolic pathways including fatty acid β-oxidation, protein synthesis, and branched-chain amino acids catabolism (Fig. 2E). These data demonstrate that high PD-1 expression is associated with a TAM subtype in lung LUAD that has a plastic, immune-regulatory phenotype, and that undergo functional and metabolic transformation during tumor development.

PD-1HIGH TAMs are enriched in the solid component of invasive lung adenocarcinoma

We investigated the spatial contexture of PD-1HIGH and PD-1LOW TAMs in invasive LUAD at 7 weeks of tumorigenesis, a time when both invasive lepidic and invasive solid histologic components are present, using IMC (Supplementary Table S3). In these experiments, CD11b was used to identify macrophages, as greater than 90% of CD45+CD11b+ cells were TAMs according to our CyTOF data (Supplementary Fig. S7). From the same invasive LUAD tumors, we compared the immune composition of invasive lepidic tumor architecture with that of invasive solid tumor architecture by selecting separate regions of interest (ROI; Fig. 3A). From each ROI, we extracted cellular information at the single-cell level using Fiji software-based cell segmentation (Supplementary Fig. S8; ref. 21). Density plots were generated using data from both lepidic and solid ROIs (9427 cells) by hierarchical stochastic neighbor embedding (HSNE) within the single-cell analysis tool, Cytosplore (24). Using unsupervised Gaussian mean shift (GMS) clustering (37), we identified three clusters, based upon local probabilistic cell density (Fig. 3B). Cells from invasive solid LUAD (ROI-2) contained high densities of PD-1HIGH TAMs (CD11b+PD-1+; Cluster 3) and high densities of cancer cells (Pan-cytokeratin+PD-L1+; Cluster 1; Fig. 3B). In contrast, cells from invasive lepidic LUAD (ROI-1) contained abundant PD-1LOW TAMs (Cluster 2) and cancer cells (Cluster 1; Fig. 3B). Consistent with our CyTOF data, PD-1LOW TAMs expressed PD-L1 and the phagocytic marker, CD44 (Fig. 3C; ref. 38). These analyses indicate that PD-1HIGH TAMs are localized to the solid components of LUAD tumors and PD-1LOW TAMs are distributed mainly within their lepidic components.

PD-1HIGH TAMs evolve through space and time to protect cancer cells

Longitudinal IMC analyses were performed to track the spatiotemporal determinants of PD-1HIGH TAMs during the histologic progression of LUAD. During early tumor development, PD-1HIGH TAMs were diffusely distributed among these tumors of preinvasive lepidic histology; however, at later time points, PD-1HIGH TAMs localized predominantly within the invasive solid histologic components that developed in these tumors (Fig. 3D). Similar to our CyTOF findings, IMC showed that PD-1HIGH TAMs highly expressed the proliferation marker Ki67, which additionally demonstrated its expression on PD-1HIGH TAMs during all stages of tumor development. Given the congregation of PD-1HIGH TAMs with high densities of cancer cells in the solid components of invasive LUAD tumors, we used IMC to dissect this spatial relationship. At later time points in tumor development, PD-1HIGH TAMs colocalized with cancer cells, displaying an appearance similar to hook and loop fasteners within invasive lepidic and solid components of the tumor (Fig. 3E). We leveraged the nuances of cell segmentation during IMC to develop a method to quantify the proximity of two defined cell populations to each other within a defined area of tumor, based on signal interference of protein expression at the points of contact of their cell borders. We created a series of SPACE (Spatial Cancer Evolution) maps at incremental time points of tumor progression to define the regional relationship of PD-1HIGH TAMs with cancer cells. These SPACE maps showed the steady emergence of PD-1HIGH TAMs during progression to invasive LUAD and demonstrate their progressive colocalization with cancer cells (Fig. 3F). Shown also in these SPACE maps is an increase in cellular TGFβ towards later time points in tumor development (Fig. 3G), similar to the trends in cellular cytokine production by PD-1HIGH TAMs shown in our longitudinal CyTOF analyses (Fig. 2C).

Given the intimate spatial network of PD-1HIGH TAMs and cancer cells in invasive LUAD, we investigated whether PD-1HIGH TAMs were tumor-promoting. Coculture of TAMs and cancer cells sorted from 4-week point tumors demonstrated that PD-1HIGH TAMs protected cancer cells from apoptosis and to a significantly greater degree than PD-1LOW TAMs (Fig. 3H). To investigate PD-1/PD-L1 interactions between PD-1HIGH TAMs and PD-L1–expressing cancer cells, we set up a three-dimensional spheroid culture system (39–41). PD-1HIGH TAMs, PD-1LOW TAMs, and PD-L1–expressing cancer cells sorted from 7-week point invasive solid adenocarcinoma tumors (at the 7-week time point) were cultured at a ratio of 1:2:2 for 7 days to allow cells to form spheroid structures (42). Spheroid size was measured by its cross-sectional area on the 7th day after seeding, and smaller size was considered to have better spheroid integrity, reflecting stronger cell-to-cell interactions (43). In an effort to disrupt the interaction of PD-1 on TAMs and PD-L1 on tumor cells, we added anti–PD-1 or anti–PD-L1 antibodies to the cultures, and each compromised spheroid formation (Fig. 3I and J). These results indicate that PD-1HIGH TAMs colocalize with PD-L1–expressing tumor cells, particularly in the solid histologic components of LUAD, and that PD-1HIGH TAMs interact with PD-L1(+) tumor cells and support their survival.

PD-1HIGH TAMs and PD-1LOW TAMs have differential impact on PD-L1–expressing cancer cells

To further investigate the interaction between PD-1HIGH TAMs and PD-L1–expressing cancer cells, we performed functional CyTOF to detect cytokines and phosphoproteins. PD-1HIGH TAMs, PD-1LOW TAMs, and PD-L1(+) cancer cells were sorted from lungs harboring invasive adenocarcinoma at the 7-week time point and cultured at a ratio of 1:1:1 for 24 hours (Fig. 4A; Supplementary Fig. S9). Sorted PD-L1–expressing cancer cells were cultured alone or with either PD-1LOW TAMs or PD-1HIGH TAMs, in the presence or absence of blocking antibodies to PD-1. In the presence of PD-1HIGH TAMs, cancer cells increased their expression of EPCAM and phospho-STAT3 (Fig. 4A), molecules that are related to initiation and maintenance of cancer stem cells (44, 45). UMAP plots demonstrated that cancer cell phosphosignaling and cytokine production were drastically and differentially affected by PD-1HIGH TAMs compared with PD-1LOW TAMs. Expression of EPCAM and pSTAT3 in cancer cells were abrogated by the presence of blocking antibodies to PD-1 (Fig. 4B). Furthermore, the addition of PD-1 blocking antibodies to the cultures resulted in increases in cancer cell expression of pSTAT1, pSTAT5, phospho-p38, pS6, and IL10, implying loss of stemness with concomitant gain of proliferative properties (Fig. 4B). In contrast, in the presence of PD-1LOW TAMs, cancer cells had relative high expression of pSTAT1, pSTAT5, pS6, and phospho-p38 and production of IL10 compared with PD-1HIGH TAMs, but anti–PD-1 treatment did not induce functional changes of cancer cells (Fig. 4A). These data demonstrate that PD-1HIGH TAMs have a distinct impact on PD-L1–expressing tumor cells, compared with PD-1LOW TAMs.

Targeting PD-1HIGH TAMs reduces tumor burden and restores immunity in lung adenocarcinoma

We evaluated the efficacy of TAM-targeting strategies in LUAD using trabectedin, an FDA-approved alkylating agent with macrophage/monocyte-depleting properties (46). We administered trabectedin to tumor-bearing mice on two different schedules: (i) beginning at 2 weeks of tumorigenesis when lungs have early lepidic (predominantly preinvasive) histology (Fig. 5A), and (ii) at 6 weeks of tumorigenesis when lungs contain tumors with invasive solid components (Fig. 5B). At the early time point, trabectedin significantly decreased the extent of lepidic tumor formation and prevented the development of invasive solid components (Fig. 5C and D). At the late time point, trabectedin also reduced the extent of lepidic tumor formation, but had only a modest effect on reducing invasive solid foci (Fig. 5E and F). Tumor-immune SCAFFOLD maps revealed that trabectedin reduced TAMs and recovered AM populations when administered at either early (Fig. 5G) or late time points (Fig. 5H). Treatment with trabectedin also resulted in the maintenance of B cells and CD4 T cells when delivered at the early timepoint, and recruitment of B cells, CD4 T cells, CD8 T cells, and NK cells when delivered at the late time point (Fig. 5G and H). To determine whether TAMs were directly affected by trabectedin, we cultured sorted TAMs and cancer cells with trabectedin or vehicle for 24 hours. We found that trabectedin had a greater direct effect on PD-1HIGH TAMs than PD-1LOW TAMs (Fig. 5I). PD-1HIGH TAMs underwent a higher degree of apoptosis than PD-1LOW TAMs (Fig. 5J) and trabectedin downregulated the expression of Arg1, c-myc, and Ki67 on PD-1HIGH TAMs (Fig. 5K). These data indicate that trabectedin directly targets PD-1HIGH TAMs in vitro and reduces the extent of lepidic and invasive solid components of LUAD.

Combining anti–PD-1 therapy with trabectedin eradicates invasive LUAD

We reasoned that concomitant boosting of the innate and adaptive immune system by combining trabectedin with a PD-1 blockade, respectively, could improve antitumor efficacy over single-agent therapy. We treated mice with trabectedin and anti–PD-1 antibodies beginning at 6 weeks of tumorigenesis when tumors contained invasive lepidic and solid adenocarcinoma histologies (Fig. 6A). PD-1 blockade alone reduced invasive solid foci but did not decrease the extent of lepidic adenocarcinoma (Fig. 6B and C). Interestingly, PD-1 blockade alone increased the proportion of proliferative (Ki-67 expressing) TAMs (Fig. 6D) and resulted in a spatial remodeling of the tumor-immune architecture from one in which proliferative TAMs were localized predominantly within the solid component of tumors before treatment to one in which they were diffusely disturbed throughout lepidic components after treatment (Fig. 6E). The majority of these proliferative TAMs but were quantified here based on their expression of Ki-67 to eliminate the confounding effects of anti–PD-1 antibody therapy on PD-1 expression; however, the majority did express PD-1 (Supplementary Fig. S10). Whereas anti–PD-1 monotherapy did not increase lymphocyte infiltration into invasive adenocarcinoma tumors, combination therapy resulted in a marked increase in B cells, CD4 effector T cells, and CD8 effector T cells (Fig. 6F; Supplementary Fig. S11), decreased the fraction of exhausted phenotype (PD-1+TIM-3+) CD8 T cells, and increased the fraction of INFγ and TNFα releasing CD8 T cells (Fig. 6G).

PD-1HIGH TAMs in human LUAD are associated with unfavorable clinical outcomes to PD-1 blockade

To evaluate whether the enrichment of PD-1HIGH TAMs in invasive LUAD was relevant to humans, we generated new macrophage gene signatures based on the differential protein profiles obtained from early PD-1HIGH TAMs from mouse early preinvasive lepidic adenocarcinoma, late PD-1HIGH TAMs from invasive solid adenocarcinoma, and PD-1LOW TAMs (Fig. 2E; Supplementary Fig. S12; Supplementary Table S2). Using CIBERSORT (http://cibersort.stanford.edu/; refs. 14, 47), we deconvoluted mRNA microarray datasets from laser-captured microdissected fresh-frozen tumor samples obtained from 60 patients with LUAD, which were deposited in the NCBI's Gene Expression Omnibus database (GEO GSE52248 and GSE58772; refs. 48, 49). These datasets include transcriptome data from normal lung (n = 6), adenocarcinoma in situ (AIS, n = 6), lepidic (n = 10), acinar (n = 10), papillary (n = 9), micropapillary (n = 9), and solid histology LUAD (n = 10). We replaced the original M0/M1/M2 macrophage populations in CIBERSORT with compositions of PD-1HIGH TAMs and PD-1LOW TAMs (Supplementary Fig. S13). We found that PD-1LOW TAMs were present among all histologic LUAD subtypes and that early PD-1HIGH TAMs were abundant in AIS and lepidic LUAD, subtypes. Late PD-1HIGH TAMs were more frequent in invasive LUAD including solid, acinar, micropapillary, and papillary subtypes, when compared with AIS and lepidic subtypes (P = 1.2 × 10–7; Fig. 7A).

To investigate tissue distribution patterns of PD-1HIGH TAMs in human LUAD, and the relationship of these cells to responses to immunotherapy in patients, we performed IMC on pretreatment tumor specimens from 20 patients with metastatic LUAD who underwent PD-1 blockade by either nivolumab (n = 13) or pembrolizumab (n = 7; Supplementary Table S4). Consistent with our findings in mouse LUAD, representative regions from core needle biopsies of these tumors showed that PD-1HIGH TAMs (PD-1+CD68+CD11b+) aggregated in the solid components of human LUAD (Fig. 7B; Supplementary Table S3) and that lepidic-predominant LUAD histology demonstrated a lower proportion of PD-1HIGH TAMs. Also, in line with our mouse data, invasive solid histology had a paucity of CD3 T-cell infiltration, while lepidic histology showed abundant stromal infiltration of B cells and T cells (Fig. 7C). We next quantified TAM and lymphocyte populations in this cohort using IMC (Supplementary Fig. S8) and found an inverse correlation between PD-1HIGH TAM density and tumor infiltration of T cells and B cells (Fig. 7D), as seen in other human tumors in which decreased infiltration of cytotoxic T cells is accompanied by increased numbers of immunosuppressive myeloid cells (50). Patients whose tumors contained abundant PD-1HIGH TAMs were more likely to have disease progression during PD-1 blockade (Fig. 7D), validated with an independent cohort [n = 32: GSE93157 (51) and GSE136961 (52); Fig. 7E], as well as decreased overall survival (Fig. 7F). These findings indicate that PD-1HIGH TAMs are enriched in the solid histologic components of human LUAD, are associated with a noninflamed TiME that is resistant to PD-1 blockade, and may be potential targets for therapies combining macrophage-targeting agents and ICIs.

Lung cancer is a complex and dynamic ecosystem of epithelial, stromal, and immune cells that regulate its progression and response to therapy (53–55). Improvement in our understanding of the immunobiology of lung cancer can be expected to result in improved therapeutic options for the many patients suffering from this malignancy. Our comprehensive single-cell analytic approach revealed that the progression of LUAD is associated with a macrophage plasticity-dependent spatiotemporal evolution of the TiME that governs response to immunotherapy. We discovered that PD-1 is expressed on a highly plastic and tumor-promoting subtype of TAMs that are intricately associated with cancer cells and that may play a role in the formation of invasive tumor histology. We targeted PD-1HIGH TAMs with the FDA-approved agent, trabectedin, and found that this approach favorably remodels its TiME and improves response to PD-1 inhibition.

PD-1 is an inhibitory surface protein found on a variety of intratumoral immune cells including T cells, B cells (43), NK cells (56), dendritic cells (57), and macrophages (58). Macrophages have previously been reported to express PD-1 following pathogen infection (59); however, PD-1 expression on tumor macrophages has only recently been reported. Gordon and colleagues demonstrated that murine TAMs can express surface PD-1 and that this is associated with impaired phagocytic activity (58). In their model of murine colon cancer, the majority of PD-1(+) TAMs displayed an M2-like, protumorigenic phenotype, whereas PD-1(−) TAMs had an M1-like, phagocytic phenotype. Our data demonstrate the novel roles of PD-1HIGH TAMs in mediating the histologic progression of cancer and in promoting resistance to checkpoint blockade by their maintenance of an immune-depleted tumor milieu. We show that TAMs highly expressing PD-1 are not present in normal lung tissue and develop over the course of tumor progression from preinvasive to invasive adenocarcinoma. Furthermore, our data demonstrate that these PD-1HIGH TAMs have the plastic phenotype, function, and tissue distribution. We describe a striking spatiotemporal evolution in these cells that can be modified by therapy, and that includes dynamic changes in expression of MHC and immunoregulatory molecules, proliferation and phagocytic activity, release of cytokines and enzymes, and tissue architectural location. Interestingly, PD-1HIGH TAMs had a proliferative phenotype and expressed CX3CR1, and consistent with our understanding that cellular proliferation can sustain certain populations of tissue-resident macrophages (60, 61), could play an important role in supporting a niche for TAM maintenance in LUAD tumors (Supplementary Fig. S14).

Trabectedin is an antineoplastic agent that is active against several human tumors including sarcoma, ovarian, and breast adenocarcinoma in phase II clinical trials, and it is FDA approved for use in patients with advanced soft tissue sarcoma. Trabectedin is a DNA minor groove binder that blocks cell cycle and has been shown to suppress monocyte recruitment to tumor sites, inhibit macrophage differentiation, and initiate caspase-8–dependent apoptosis in tumor macrophages in transplantable mouse models of cancer (46, 62). In our genetically engineered mouse model of LUAD, treatment of tumor-bearing mice with trabectedin resulted in TAM depletion and decrease in tumor burden of lepidic and solid adenocarcinoma histology. In vitro, PD-1HIGH TAMs were directly killed by trabectedin and were more sensitive to killing than PD-1LOW TAMs. These findings are consistent with our finding that PD-1HIGH TAM had a highly proliferative phenotype, and with the known ability of trabectedin to block cell cycle on cells with rapid turnover including myeloid bone marrow progenitor cells (46). Combining trabectedin with PD-1 inhibition resulted in improved antitumor immunity, and decreased both the solid and lepidic components of LUAD, likely by several mechanisms (Supplementary Fig. S15). PD-1 blockade alone resulted in a striking increase in the number of proliferative TAMs, potentially increasing their susceptibility to trabectedin. Possibly as a result of its effects on immunoregulatory TAMs, trabectedin remodeled the TiME of LUAD to an environment rich in lymphocytes and INFγ-producing T cells, which would be expected to improve responses to PD-1 inhibition from an “inflamed” tumor (50, 63). Furthermore, our data show a trajectory of spatiotemporal development of PD-1HIGH TAMs during the progression of LUAD, resulting in their clustering within the invasive solid components of advanced LUAD and intimate colocalization with PD-L1 expressing tumor cells. PD-1 blockade caused a disassociation of this solid tumor architecture in vivo with a decrease in the number of solid invasive foci. We explored this further in a spheroid culture system of TAMs and PD-L1(+) tumor cells within which the addition of PD-1 or PD-L1 antibodies decreased the integrity of tumor spheroid structures. This suggests that PD-1 inhibition can interrupt the cell–cell interactions important for tumor integrity, and could also potentially facilitate trabectedin delivery within these tumors. In line with these findings, combination therapy with trabectedin and PD-1 blockade was found to have efficacy against a mouse model of fibrosarcoma (64). By way of its pleiotropic anti-tumor activity that includes suppression of tumor-promoting immune cells and direct cytotoxicity against lung cancer cells, trabectedin or its derivative, lurbinectedin, combined with checkpoint immunotherapy may be a viable therapeutic option for immunoresistant NSCLC tumors, alone or in combination with PD-1 blockade.

This mouse model shares features with human AIS, MIA, LPA, and solid subtype LUAD, and shares several similarities with our human data that support its translational relevance. Broadly, the TiME of mouse LUAD steadily became lymphocyte-depleted and enriched for phenotypically and functionally exhausted T cells with increasing elements of histologic invasion. Reciprocally, PD-1HIGH TAMs were minimally present in preinvasive mouse tumors (AIS), were highly enriched in mouse tumors with invasive solid histology, and localized most densely within solid histologic components. In human datasets, we found a similar inverse correlation between the number of tumor-infiltrating lymphocytes and the number of PD-1HIGH TAM. We also found a lower frequency of PD-1HIGH TAMs in preinvasive and early invasive human LUAD tumors (AIS and MIA) and a higher frequency of PD-1HIGH TAMs in the solid histology subtype. Furthermore, IMC of human LUAD tumors additionally showed increased density of PD-1HIGH TAMs in solid components of human LUAD tumors. Importantly, an abundance of PD-1HIGH TAMs was associated with resistance or hyperprogression, disease progression, and decreased overall survival in patients with LUAD who received anti–PD-1 therapy. These findings are highly suggestive that PD-1HIGH TAMs, and adenocarcinoma subtype histology are important determinants of response to immunotherapy in human LUAD.

In summary, our spatiotemporal investigation of lung adenocarcinoma with single-cell proteomic platforms identified histologic subtype-specific elements of its TiME that have important considerations for immunotherapy. We identified an immunoregulatory subtype of TAM in LUAD that expressed PD-1, that expanded during tumor progression from preinvasive to invasive adenocarcinoma histology, that dominated the lymphocyte-depleted niche of invasive tumors, and that colocalized with tumor cells in the solid histologic components of the tumor. These tumor-protecting TAMs were sensitive to killing by trabectedin, which remodeled the TiME and which improved response to PD-1 blockade. Ultimately, our data support pursuing PD-1HIGH TAMs targeting therapies in patients with immunoresistant lung cancer.

H. Lee reports grants from the Cancer Prevention and Research Institute of Texas (PI), grants from NCI P30CA125123 (Co-I), grants from NCRR S10RR024574 (Core), grants from CPRIT-RP180672 (Core), grants from R50 CA243707-01A1 (Core), and grants from M.D. Anderson Cancer Center Support Grant CA016672 (Core) during the conduct of the study; grants from Duncan Cancer Center pilot award (PI), grants from Michael E. DeBakey Department of Surgery Faculty Research Grants (PI), grants from Samyang Biopharmaceutical USA (PI), and non-financial support from the Gutenstein Family Foundation outside the submitted work. D.Y. Wang reports personal fees from Regeneron and personal fees from Sanofi Genzyme outside the submitted work. C.I. Amos reports grants from National Cancer Institute and grants from Cancer Prevention Research Institute of Texas during the conduct of the study. B.M. Burt reports grants from Cancer Prevention Research Institute of Texas during the conduct of the study, grants from AstraZenica, grants from Momotero-Gene, grants from Novartis, and personal fees from AstraZenica outside the submitted work. No disclosures were reported by the other authors.

H.-J. Jang: Conceptualization, resources, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft. H.-S. Lee: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. W. Yu: Formal analysis, supervision, investigation. M. Ramineni: Formal analysis, validation, investigation, methodology. C.Y. Truong: Data curation, formal analysis, investigation. D. Ramos: Methodology. T. Splawn: Methodology. J.M. Choi: Data curation, formal analysis, visualization. S.Y. Jung: Data curation, formal analysis, writing–review and editing. J.-S. Lee: Formal analysis, supervision, visualization. D.Y. Wang: Resources, supervision, validation, writing–review and editing. J.M. Sederstrom: Methodology. M. Pietropaolo: Supervision, writing–review and editing. F. Kheradmand: Supervision. C.I. Amos: Data curation, supervision, methodology, writing–review and editing. T.M. Wheeler: Supervision. R.T. Ripley: Supervision. B.M. Burt: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, writing–original draft, writing–review and editing.

This work was supported by the Cancer Prevention and Research Institute of Texas (CPRIT RP200443 to H.-S. Lee), the NIH R37 MERIT Award (R37CA248478 to B.M. Burt), the Gutenstein Family Foundation, and Samyang Biopharm USA research grant. C.I. Amos is a Research Scholar of the Cancer Prevention Research Institute of Texas, with funding from the NIH (U19CA203654) and the Cancer Prevention and Research Institute of Texas (CPRIT RR170048). This project was also supported by the Cytometry and Cell Sorting Core at Baylor College of Medicine, with funding from the CPRIT Core Facility Support Award (CPRIT-RP180672), the NIH (P30 CA125123 and S10 RR024574), and the expert assistance of Joel M. Sederstrom. This research was performed in the Flow Cytometry & Cellular Imaging Facility, which is supported in part by the National Institutes of Health through M.D. Anderson Cancer Center Support Grant CA016672, the NCI's Research Specialist 1 R50 CA243707-01A1, and a Shared Instrumentation Award from the Cancer Prevention Research Institution of Texas (CPRIT), RP121010.

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

1.
Siegel
RL
,
Miller
KD
,
Fuchs
HE
,
Jemal
A
.
Cancer statistics, 2021
.
CA Cancer J Clin
2021
;
71
:
7
33
.
2.
Brahmer
J
,
Reckamp
KL
,
Baas
P
,
Crinò
L
,
Eberhardt
WE
,
Poddubskaya
E
, et al
.
Nivolumab versus docetaxel in advanced squamous-cell non–small-cell lung cancer
.
N Engl J Med
2015
;
373
:
123
35
.
3.
Borghaei
H
,
Paz-Ares
L
,
Horn
L
,
Spigel
DR
,
Steins
M
,
Ready
NE
, et al
.
Nivolumab versus docetaxel in advanced nonsquamous non–small-cell lung cancer
.
N Engl J Med
2015
;
373
:
1627
39
.
4.
Herbst
RS
,
Baas
P
,
Kim
D-W
,
Felip
E
,
Pérez-Gracia
JL
,
Han
J-Y
, et al
.
Pembrolizumab versus docetaxel for previously treated, PD-L1-positive, advanced non-small-cell lung cancer (KEYNOTE-010): a randomised controlled trial
.
Lancet
2016
;
387
:
1540
50
.
5.
Reck
M
,
Rodríguez-Abreu
D
,
Robinson
AG
,
Hui
R
,
Csőszi
T
,
Fülöp
A
, et al
.
Pembrolizumab versus chemotherapy for PD-L1–positive non–small-cell lung cancer
.
N Engl J Med
2016
;
375
:
1823
33
.
6.
Fehrenbacher
L
,
Spira
A
,
Ballinger
M
,
Kowanetz
M
,
Vansteenkiste
J
,
Mazieres
J
, et al
.
Atezolizumab versus docetaxel for patients with previously treated non-small-cell lung cancer (POPLAR): a multicentre, open-label, phase 2 randomised controlled trial
.
Lancet
2016
;
387
:
1837
46
.
7.
Reck
M
,
Rodríguez-Abreu
D
,
Robinson
AG
,
Hui
R
,
Csőszi
T
,
Fülöp
A
, et al
.
Updated analysis of KEYNOTE-024: pembrolizumab versus platinum-based chemotherapy for advanced non-small-cell lung cancer with PD-L1 tumor proportion score of 50% or greater
.
J Clin Oncol
2019
;
37
:
537
46
.
8.
Antonia
SJ
,
Borghaei
H
,
Ramalingam
SS
,
Horn
L
,
De Castro Carpeño
J
,
Pluzanski
A
, et al
.
Four-year survival with nivolumab in patients with previously treated advanced non-small-cell lung cancer: a pooled analysis
.
Lancet Oncol
2019
;
20
:
1395
408
.
9.
S
Schoenfeld
AJ
,
Hellmann
MD
.
Acquired resistance to immune checkpoint inhibitors
.
Cancer Cell
2020
;
37
:
443
55
.
10.
Travis
WD
,
Brambilla
E
,
Noguchi
M
,
Nicholson
AG
,
Geisinger
KR
,
Yatabe
Y
, et al
.
International association for the study of lung cancer/american thoracic society/european respiratory society international multidisciplinary classification of lung adenocarcinoma
.
J Thorac Oncol
2011
;
6
:
244
85
.
11.
Politi
K
,
Zakowski
MF
,
Fan
P-D
,
Schonfeld
EA
,
Pao
W
,
Varmus
HE
.
Lung adenocarcinomas induced in mice by mutant EGF receptors found in human lung cancers respond to a tyrosine kinase inhibitor or to down-regulation of the receptors
.
Genes Dev
2006
;
20
:
1496
510
.
12.
Warth
A
,
Muley
T
,
Meister
M
,
Stenzinger
A
,
Thomas
M
,
Schirmacher
P
, et al
.
The novel histologic international association for the study of lung cancer/American thoracic society/European respiratory society classification system of lung adenocarcinoma is a stage-independent predictor of survival
.
J Clin Oncol
2012
;
30
:
1438
46
.
13.
Tsao
M-S
,
Marguet
S
,
Le Teuff
G
,
Lantuejoul
S
,
Shepherd
FA
,
Seymour
L
, et al
.
Subtype classification of lung adenocarcinoma predicts benefit from adjuvant chemotherapy in patients undergoing complete resection
.
J Clin Oncol
2015
;
33
:
3439
46
.
14.
Wang
D-H
,
Lee
H-S
,
Yoon
D
,
Berry
G
,
Wheeler
TM
,
Sugarbaker
DJ
, et al
.
Progression of EGFR-mutant lung adenocarcinoma is driven by alveolar macrophages
.
Clin Cancer Res
2017
;
23
:
778
88
.
15.
Zunder
ER
,
Finck
R
,
Behbehani
GK
,
El-ad
DA
,
Krishnaswamy
S
,
Gonzalez
VD
, et al
.
Palladium-based mass tag cell barcoding with a doublet-filtering scheme and single-cell deconvolution algorithm
.
Nat Protoc
2015
;
10
:
316
33
.
16.
Finck
R
,
Simonds
EF
,
Jager
A
,
Krishnaswamy
S
,
Sachs
K
,
Fantl
W
, et al
.
Normalization of mass cytometry data with bead standards
.
Cytometry A
2013
;
83
:
483
94
.
17.
Qiu
P
,
Simonds
EF
,
Bendall
SC
,
Gibbs
JKD
,
Bruggner
RV
,
Linderman
MD
, et al
.
Extracting a cellular hierarchy from high-dimensional cytometry data with SPADE
.
Nat Biotechnol
2011
;
29
:
886
91
.
18.
Bendall
SC
,
Simonds
EF
,
Qiu
P
,
Amir
E-a
,
Krutzik
PO
,
Finck
R
, et al
.
Single-cell mass cytometry of differential immune and drug responses across a human hematopoietic continuum
.
Science
2011
;
332
:
687
96
.
19.
Spitzer
MH
,
Gherardini
PF
,
Fragiadakis
GK
,
Bhattacharya
N
,
Yuan
RT
,
Hotson
AN
, et al
.
An interactive reference framework for modeling a dynamic immune system
.
Science
2015
;
349
:
1259425
.
20.
Giesen
C
,
Wang
HA
,
Schapiro
D
,
Zivanovic
N
,
Jacobs
A
,
Hattendorf
B
, et al
.
Highly multiplexed imaging of tumor tissues with subcellular resolution by mass cytometry
.
Nat Methods
2014
;
11
:
417
22
.
21.
Schindelin
J
,
Arganda-Carreras
I
,
Frise
E
,
Kaynig
V
,
Longair
M
,
Pietzsch
T
, et al
.
Fiji: an open-source platform for biological-image analysis
.
Nat Methods
2012
;
9
:
676
.
22.
Schwendy
M
,
Unger
RE
,
Bonn
M
,
Parekh
SH
.
Automated cell segmentation in FIJI® using the DRAQ5 nuclear dye
.
BMC Bioinf
2019
;
20
:
39
.
23.
Höllt
T
,
Pezzotti
N
,
van Unen
V
,
Koning
F
,
Eisemann
E
,
Lelieveldt
B
, et al
.
Cytosplore: interactive immune cell phenotyping for large single-cell datasets
.
Comput Graphics Forum
2016
.
Wiley Online Library
. p.
171
80
.
24.
van Unen
V
,
Höllt
T
,
Pezzotti
N
,
Li
N
,
Reinders
MJ
,
Eisemann
E
, et al
.
Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types
.
Nat Commun
2017
;
8
:
1740
.
25.
Van Gassen
S
,
Gaudilliere
B
,
Angst
MS
,
Saeys
Y
,
Aghaeepour
N
.
CytoNorm: a normalization algorithm for cytometry data
.
Cytometry A
2020
;
97
:
268
78
.
26.
Becht
E
,
McInnes
L
,
Healy
J
,
Dutertre
C-A
,
Kwok
IWH
,
Ng
LG
, et al
.
Dimensionality reduction for visualizing single-cell data using UMAP
.
Nat Biotechnol
2019
;
37
:
38
44
.
27.
Lee
H-S
,
Jang
H-J
,
Choi
JM
,
Zhang
J
,
de Rosen
VL
,
Wheeler
TM
, et al
.
Comprehensive immunoproteogenomic analyses of malignant pleural mesothelioma
.
JCI Insight
2018
;
3
:
e98575
.
28.
M
Lvd
H
G
.
Visualizing data using t-SNE
.
J Mach Learn Res
2008
;
9
:
2579
605
.
29.
Bedoret
D
,
Wallemacq
H
,
Marichal
T
,
Desmet
C
., Quesada Calvo F, Henry E, et al
.
Lung interstitial macrophages alter dendritic cell functions to prevent airway allergy in mice
.
J Clin Invest
2009
;
119
:
3723
38
.
30.
Lagranderie
M
,
Nahori
MA
,
Balazuc
AM
,
Kiefer-Biasizzo
H
,
Lapa e Silva
JR
,
Milon
G
, et al
.
Dendritic cells recruited to the lung shortly after intranasal delivery of Mycobacterium bovis BCG drive the primary immune response towards a type 1 cytokine production
.
Immunology
2003
;
108
:
352
64
.
31.
Davies
LC
,
Jenkins
SJ
,
Allen
JE
,
Taylor
PR
.
Tissue-resident macrophages
.
Nat Immunol
2013
;
14
:
986
95
.
32.
Misharin
AV
,
Morales-Nebreda
L
,
Mutlu
GM
,
Budinger
GS
,
Perlman
H
.
Flow cytometric analysis of macrophages and dendritic cell subsets in the mouse lung
.
Am J Respir Cell Mol Biol
2013
;
49
:
503
10
.
33.
Svedberg
FR
,
Brown
SL
,
Krauss
MZ
,
Campbell
L
,
Sharpe
C
,
Clausen
M
, et al
.
The lung environment controls alveolar macrophage metabolism and responsiveness in type 2 inflammation
.
Nat Immunol
2019
;
20
:
571
80
.
34.
Chakarov
S
,
Lim
HY
,
Tan
L
,
Lim
SY
,
See
P
,
Lum
J
, et al
.
Two distinct interstitial macrophage populations coexist across tissues in specific subtissular niches
.
Science
2019
;
363
:
eaau0964
.
35.
Ishida
Y
,
Kimura
A
,
Nosaka
M
,
Kuninaka
Y
,
Hemmi
H
,
Sasaki
I
, et al
.
Essential involvement of the CX3CL1-CX3CR1 axis in bleomycin-induced pulmonary fibrosis via regulation of fibrocyte and M2 macrophage migration
.
Sci Rep
2017
;
7
:
16833
.
36.
Vitale
I
,
Manic
G
,
Coussens
LM
,
Kroemer
G
,
Galluzzi
L
.
Macrophages and metabolism in the tumor microenvironment
.
Cell Metab
2019
;
30
:
36
50
.
37.
Comaniciu
D
,
Meer
P
.
Mean shift: A robust approach toward feature space analysis
.
IEEE Trans Pattern Anal Mach Intell
2002
;
24
:
603
19
.
38.
Vachon
E
,
Martin
R
,
Plumb
J
,
Kwok
V
,
Vandivier
RW
,
Glogauer
M
, et al
.
CD44 is a phagocytic receptor
.
Blood
2006
;
107
:
4149
58
.
39.
Long
L
,
Yin
M
,
Min
W
.
3D Co-culture system of tumor-associated macrophages and ovarian cancer cells
.
Bio Protoc
2018
;
8
:
e2815
.
40.
Yin
M
,
Li
X
,
Tan
S
,
Zhou
HJ
,
Ji
W
,
Bellone
S
, et al
.
Tumor-associated macrophages drive spheroid formation during early transcoelomic metastasis of ovarian cancer
.
J Clin Invest
2016
;
126
:
4157
73
.
41.
Rebelo
SP
,
Pinto
C
,
Martins
TR
,
Harrer
N
,
Estrada
MF
,
Loza-Alvarez
P
, et al
.
3D-3-culture: A tool to unveil macrophage plasticity in the tumour microenvironment
.
Biomaterials
2018
;
163
:
185
97
.
42.
Akbay
EA
,
Koyama
S
,
Carretero
J
,
Altabef
A
,
Tchaicha
JH
,
Christensen
CL
, et al
.
Activation of the PD-1 pathway contributes to immune escape in EGFR-driven lung tumors
.
Cancer Discov
2013
;
3
:
1355
63
.
43.
Agata
Y
,
Kawasaki
A
,
Nishimura
H
,
Ishida
Y
,
Tsubat
T
,
Yagita
H
, et al
.
Expression of the PD-1 antigen on the surface of stimulated mouse T and B lymphocytes
.
Int Immunol
1996
;
8
:
765
72
.
44.
Kuan
II
,
Liang
KH
,
Wang
YP
,
Kuo
TW
,
Meir
YJ
,
Wu
SC
, et al
.
EpEX/EpCAM and Oct4 or Klf4 alone are sufficient to generate induced pluripotent stem cells through STAT3 and HIF2α
.
Sci Rep
2017
;
7
:
41852
.
45.
Walcher
L
,
Kistenmacher
A-K
,
Suo
H
,
Kitte
R
,
Dluczek
S
,
Strauß
A
, et al
.
Cancer stem cells—origins and biomarkers: perspectives for targeted personalized therapies
.
Front Immunol
2020
;
11
:
1280
.
46.
Germano
G
,
Frapolli
R
,
Belgiovine
C
,
Anselmo
A
,
Pesce
S
,
Liguori
M
, et al
.
Role of macrophage targeting in the antitumor activity of trabectedin
.
Cancer Cell
2013
;
23
:
249
62
.
47.
Newman
AM
,
Liu
CL
,
Green
MR
,
Gentles
AJ
,
Feng
W
,
Xu
Y
, et al
.
Robust enumeration of cell subsets from tissue expression profiles
.
Nat Methods
2015
;
12
:
453
7
.
48.
Zabeck
H
,
Dienemann
H
,
Hoffmann
H
,
Pfannschmidt
J
,
Warth
A
,
Schnabel
PA
, et al
.
Molecular signatures in IASLC/ATS/ERS classified growth patterns of lung adenocarcinoma
.
PLoS One
2018
;
13
:
e0206132
.
49.
Morton
ML
,
Bai
X
,
Merry
CR
,
Linden
PA
,
Khalil
AM
,
Leidner
RS
, et al
.
Identification of mRNAs and lincRNAs associated with lung cancer progression using next-generation RNA sequencing from laser micro-dissected archival FFPE tissue specimens
.
Lung Cancer
2014
;
85
:
31
39
.
50.
Chen
DS
,
Mellman
I
.
Elements of cancer immunity and the cancer–immune set point
.
Nature
2017
;
541
:
321
30
.
51.
Prat
A
,
Navarro
A
,
Paré
L
,
Reguart
N
,
Galván
P
,
Pascual
T
, et al
.
Immune-Related Gene Expression Profiling After PD-1 Blockade in Non-Small Cell Lung Carcinoma, Head and Neck Squamous Cell Carcinoma, and Melanoma
.
Cancer Res
2017
;
77
:
3540
50
.
52.
Hwang
S
,
Kwon
AY
,
Jeong
JY
,
Kim
S
,
Kang
H
,
Park
J
, et al
.
Immune gene signatures for predicting durable clinical benefit of anti-PD-1 immunotherapy in patients with non-small cell lung cancer
.
Sci Rep
2020
;
10
:
643
.
53.
Chen
H
,
Carrot-Zhang
J
,
Zhao
Y
,
Hu
H
,
Freeman
SS
,
Yu
S
, et al
.
Genomic and immune profiling of pre-invasive lung adenocarcinoma
.
Nat Commun
2019
;
10
:
5472
.
54.
Zhang
C
,
Zhang
J
,
Xu
F-P
,
Wang
Y-G
,
Xie
Z
,
Su
J
, et al
.
Genomic landscape and immune microenvironment features of preinvasive and early invasive lung adenocarcinoma
.
J Thorac Oncol
2019
;
14
:
1912
23
.
55.
Jang
H-J
,
Lee
H-S
,
Ramos
D
,
Park
IK
,
Kang
CH
,
Burt
BM
, et al
.
Transcriptome-based molecular subtyping of non-small cell lung cancer may predict response to immune checkpoint inhibitors
.
J Thorac Cardiovasc Surg
2019
:
S0022–5223:32515–2
.
56.
Benson
DM
,
Bakan
CE
,
Mishra
A
,
Hofmeister
CC
,
Efebera
Y
,
Becknell
B
, et al
.
The PD-1/PD-L1 axis modulates the natural killer cell versus multiple myeloma effect: a therapeutic target for CT-011, a novel monoclonal anti–PD-1 antibody
.
Blood
2010
;
116
:
2286
94
.
57.
Karyampudi
L
,
Lamichhane
P
,
Krempski
J
,
Kalli
KR
,
Behrens
MD
,
Vargas
DM
, et al
.
PD-1 blunts the function of ovarian tumor–infiltrating dendritic cells by inactivating NF-κB
.
Cancer Res
2016
;
76
:
239
50
.
58.
Gordon
SR
,
Maute
RL
,
Dulken
BW
,
Hutter
G
,
George
BM
,
McCracken
MN
, et al
.
PD-1 expression by tumour-associated macrophages inhibits phagocytosis and tumour immunity
.
Nature
2017
;
545
:
495
9
.
59.
Shen
L
,
Gao
Y
,
Liu
Y
,
Zhang
B
,
Liu
Q
,
Wu
J
, et al
.
PD-1/PD-L pathway inhibits M. tb-specific CD4+ T-cell functions and phagocytosis of macrophages in active tuberculosis
.
Sci Rep
2016
;
6
.
60.
Franklin
RA
,
Li
MO
.
Ontogeny of tumor-associated macrophages and its implication in cancer regulation
.
Trends Cancer
2016
;
2
:
20
34
.
61.
Bonapace
L
,
Coissieux
M-M
,
Wyckoff
J
,
Mertz
KD
,
Varga
Z
,
Junt
T
, et al
.
Cessation of CCL2 inhibition accelerates breast cancer metastasis by promoting angiogenesis
.
Nature
2014
;
515
:
130
3
.
62.
Carminati
L
,
Pinessi
D
,
Borsotti
P
,
Minoli
L
,
Giavazzi
R
., D'Incalci M, et al
.
Antimetastatic and antiangiogenic activity of trabectedin in cutaneous melanoma
.
Carcinogenesis
2019
;
40
:
303
12
.
63.
Tumeh
PC
,
Harview
CL
,
Yearley
JH
,
Shintaku
IP
,
Taylor
EJM
,
Robert
L
, et al
.
PD-1 blockade induces responses by inhibiting adaptive immune resistance
.
Nature
2014
;
515
:
568
71
.
64.
Belgiovine
C
,
Frapolli
R
,
Liguori
M
,
Digifico
E
,
Colombo
FS
,
Meroni
M
, et al
.
Inhibition of tumor-associated macrophages by trabectedin improves the antitumor adaptive immunity in response to anti-PD-1 therapy
.
Eur J Immunol
2021
;
51
:
2677
86
.

Supplementary data