Clustered macrophages cooperate to eliminate tumors via coordinated intrudopodia

Significance

Solid tumors are often replete with macrophages, but multiple observations have also indicated that macrophage clusters somehow associate with patient survival. We find that conditions which maximize cancer cell phagocytosis lead to dynamic macrophage clusters in solid tumor models. Our reductionist approaches ultimately reveal pathways and roles for tumor-disruptive pseudopodia, which we term “intrudopodia.”

Abstract

Macrophages often pervade solid tumors, and clusters of macrophages sometimes associate with longer survival of patients. However, clustering mechanisms and impacts on key functions such as phagocytosis remain obscure. Here, under conditions that maximize cancer cell phagocytosis within cohesive tumors, we uncover pathways that favor dynamic clusters and find a colocalization of tumor-intrusive pseudopodia which we term “intrudopodia.” Cluster formation over hours on low-adhesion substrates occurs after macrophage induction to a state colloquially referred to as M1 after exposure to interferons and T cell–derived cytokines. Clusters prove fluid on timescales of minutes and also sort from interleukin-4-treated, so-called M2 macrophages that tend to disperse. M1 macrophages upregulate specific cell–cell adhesion receptors but suppress actomyosin contractility, with both pathways contributing to cluster formation. Decreased cortical tension was not only reflected in a low level of nuclear lamin-A that downregulates cytoskeletal targets of serum response factor and tends to soften the nucleus but was also predicted to unleash pseudopodia. Macrophage neighbors in tumor spheroids indeed coextend intrudopodia between cancer cell junctions—at least when phagocytosis conditions are maximized. Intrudopodia from neighbors help detach and individualize cancer cells for rapid engulfment. Juxtaposition of a macrophage cluster with tumor cell nests defines a broad interface that minimizes cancer cell nearest neighbor interactions and maximizes coordination of macrophage intrudopodia. Cooperative phagocytosis thus overcomes solid tumor cohesion—and might explain why the macrophage clustering factor ITGAL associates with patient survival.

Adhesion receptors and cytoskeleton are key to any dynamic interactions of a cell with other cells in a microenvironment. Macrophages are an abundant cell type in most solid tissues and tumors, and they make direct contact with numerous cell types—including other macrophages. Clusters or aggregates of macrophages have indeed been described in multiple contexts including cancer (1–5) and wound healing (6, 7), with some granulomas and foreign bodies even stimulating macrophage fusion (8). In human head and neck cancer, a macrophage subtype identified by single-cell transcriptomics as being CXCL9-high was described as forming antitumor “clusters” that accumulate at interfaces between tumor nests and stroma (1). Cell–cell interactions were unstudied, but these macrophages correlated with interferon-γ (IFNγ) signaling and favorable patient outcomes relative to SPP1-high macrophages that associated with poor survival. Further analysis of this intriguing human data more specifically indicates that CXCL9-high macrophages partition about equally in the tumor nest and stroma and are also about as abundant as unpolarized macrophages in the nest (SI Appendix, Fig. S1 A and B). The favorable prognosis for CXCL9-high macrophages might thus be explained by their clustering at the tumor interface. These findings and other descriptions of tumor macrophage “aggregates” (2, 3) or “nests” (4) including aggregates of chimeric antigen receptor–macrophages (CAR-M) (5) motivate the mechanistic study of how such clusters might form and what, if any, unique functions they might possess.

Many cell types retain the capacity for multicellular organization in vitro, which is perhaps best illustrated by sorting of cell mixtures (9–14). Differential adhesion and differential contractility often provide cogent explanations for segregation of distinct cell types (15), but macrophages are poorly studied in such contexts. Macrophage heterogeneity in human cancers is evident from single-cell sequencing studies that distinguish multiple macrophage subtypes (1, 16–19), which may also be spatially segregated to some extent in tumors (1, 19, 20). These subtypes are at least conceptually related to in vitro polarized proinflammatory M1 macrophages and alternatively activated M2 macrophages, but in general tumor-associated macrophage (TAM) subtypes exhibit greater complexity and their gene expression profiles do not map directly to the M1 or M2 states. Although macrophage subtypes are frequently identified based on surface marker expression, secretion profiles, and enzymatic activity, distinguishing characteristics likely also extend to phenotypic “polarizations” of cell morphology (21, 22) and motility (23–25) that reflect differences in membrane receptors and cytoskeleton. An impact on phagocytosis as the defining adhesive-motility function of a macrophage (26–30) seems likely but has remained obscure.

Here, we investigated macrophage clustering in response to various immune stimuli including activation of Fcγ receptors during phagocytosis (31, 32) as well as polarizing cytokines. We show that different macrophage subtypes will cluster to varying extents in “immuno-tumoroid” mixtures with cancer cells, as well as in monocultures. We find and evaluate differential expression of key adhesion receptors among macrophages and also a counterintuitive role for the actomyosin cytoskeleton and nucleoskeleton system in macrophage cluster formation. Live imaging of phagocytosis in tumor spheroids ultimately shows that neighboring macrophages extend intrusive pseudopods or “intrudopodia” that coordinate with each other to disrupt tumor cohesion and thereby target individualized cancer cells for engulfment.

Results

Macrophages Cluster under Prophagocytic Conditions in Tumors and Tumoroids.

To investigate the spatial organization of macrophages in a therapeutically relevant tumor model, we generated lung metastases of syngeneic B16 melanoma cells and treated tumor-bearing mice with intravenous injections of an opsonizing IgG antibody targeting the melanocyte-specific membrane antigen Tyrp1 (Fig. 1A). We had deleted CD47, a key inhibitory ligand that forms an immune checkpoint with signal regulatory protein α (SIRPα) on macrophages (33), and recently shown that anti-Tyrp1 injections reduce the number and size of B16 lung nodules, with greater responses for CD47 knockout (KO) B16 compared to wild-type (WT) B16 (34). Following antibody treatment, we harvested the lungs of treated mice and untreated control mice and stained for the mouse macrophage marker F4/80 (Fig. 1A and B and SI Appendix, Fig. S2 A and B). Macrophages in tumor nodules from treated mice were most often present in clusters quantified as containing N = 1, 2, 3, or 4+ macrophages, whereas macrophages in tumor nodules from untreated control mice were most frequently present as isolated cells (Fig. 1B and SI Appendix, Fig. S2C). From the distribution of N (SI Appendix, Fig. S2C), we computed the mean (N_avg) and the N-weighted mean cluster size (W_avg) for five nodules each from treated and untreated mice. Both methods of averaging revealed a statistically significant difference (Fig. 1B). There was also a statistically significant difference in the cumulative distributions of N (SI Appendix, Fig. S2 C, Inset). Importantly, macrophages in clusters frequently appeared to be phagocytosing B16 cells based on the internalization of both large nuclei characteristic of B16 and melanin pigments visible in bright-field (Fig. 1B). Together, these results suggest an association between macrophage clustering and phagocytosis in tumors.

Fig. 1.

Macrophages (Mφ) cluster under maximum phagocytic conditions and when polarized with IFNγ. (A) Establishing and treating B16 CD47 KO lung metastases in mice. (B) Bright-field and confocal fluorescence microscopy of metastatic nodules immunostained for F4/80. Yellow dotted outlines denote macrophage clusters. Zoomed confocal slices of two macrophage clusters labeled 1 and 2. Pie-charts: Percentage macrophages that are part of clusters of N = 1, 2, 3, and 4+ macrophages in nodules. Number average cluster size (N_avg) and N-weighted average cluster size (W_avg) in nodules. Mean ± SD, n = 5 nodules per condition, Mann–Whitney test (two-tailed). (C) Immuno-tumoroids of growing B16 CD47 KO cells (green) to which macrophages (magenta) primed with IFNγ or IL4 were added at varying ratios with an opsonizing antibody (anti-Tyrp1). The yellow dotted outlines depict the perimeter of a 2D convex hull enclosing all macrophages. (D) The projected GFP⁺ area of B16 CD47 KO in immuno-tumoroids plotted against the macrophage convex hull perimeter for different time points and cytokine priming conditions. Inset: Convex hull perimeter as a function of time. Mean ± SD, n = 7 or 8 wells per condition, repeated measures two-way ANOVA with Dunnett’s multiple comparisons test. (E) Macrophage differentiation from bone marrow (BM) or conditionally immortalized macrophage (CIM) progenitors and clustering assay on low-adhesion U-bottom well plates. (F and G) Representative fluorescence images after 3 d (F) and convex hull perimeters as a function of time (G) of macrophages on low-adhesion surfaces in the presence of 20 ng mL⁻¹ IFNγ (= M1), no cytokine (= M0), or 20 ng mL⁻¹ IL4 (= M2). The orange dotted outlines depict the 2D convex hull enclosing all macrophages. The fluorescence intensity profile corresponds to a line across the image at the position denoted by the arrowhead. Mean ± SD, n = 9 or 10 wells per condition, repeated measures two-way ANOVA with Dunnett’s multiple comparisons test. (H) Bright-field images of micropipette aspiration and creep profiles from aspiration of an M1 macrophage cluster (Top) and a B16 tumoroid (Bottom). The strain (ε = aspiration length L/pipette radius R_P) is normalized by a critical strain ε_c corresponding to the transition to the viscous regime. Solid lines are linear fits of the viscous regime where the slope is inversely proportional to viscosity, η. (I) Outlines of M1 macrophage clusters and B16 tumoroid clusters (4 each) and outlines of B16 and M1 macrophages in immuno-tumoroids at t = 12 h depicting the centralization of macrophage clusters within B16 tumoroids. (J) Radial fluorescence intensity profiles and fluorescence image of a coculture of M1 (magenta, CellTracker Deep Red) and M2 (green, CFDA) macrophages in a low-adhesion U-bottom well after 1 d. Shannon entropy computed from the M1 and M2 radial profiles for n = 8 different cocultures, Student’s t test (two-tailed, paired).

Recently described clusters and nests of macrophages in human and murine tumors have been related to proinflammatory signaling pathways activated in subsets of tumor macrophages (1, 4). To begin to address the effects of tumor macrophage heterogeneity on clustering and phagocytosis rates relative to tumor growth, we performed multiday immuno-tumoroid assays in which mouse bone marrow–derived macrophages (BMDMs) were activated with IFNγ (“M1”) or interleukin-4 (IL4, “M2”) and added together with anti-Tyrp1 opsonizing IgG to cohesive aggregates of B16 cells on low-adhesion, U-bottom well plates (35). Being aware of the limitations of M1/M2 polarization to adequately describe macrophage heterogeneity in vivo, we reserve these terms only for macrophages activated in vitro under the well-defined conditions of our reductionist assays and address the relevance to human macrophage subtypes later. We confirmed BMDM expression of macrophage markers including FcγRI and SIRPα, which positively and negatively regulate antibody-dependent cellular phagocytosis, respectively (SI Appendix, Fig. S3A). We also confirmed the expected upregulation of MHCII and CD206 on BMDMs that were “primed” for 2 d in standard two-dimensional (2D) cell culture conditions with IFNγ and IL4, respectively (SI Appendix, Fig. S3 B and C). We then added primed macrophages to preformed B16 CD47 KO tumoroids and quantified the GFP⁺ area versus time as a proxy for the number of B16 cells (Fig. 1C). The resulting growth curves were fit to a simple exponential model with an effective growth rate, k_eff, that is positive for net B16 growth and negative for net repression by macrophages. Consistent with our earlier study (35), elimination of B16 by unprimed control (M0) macrophages required opsonization with anti-Tyrp1, and the nonlinear dependence of k_eff on the number of added macrophages was well fit by a cooperative phagocytosis model (SI Appendix, Fig. S4A). Tumoroid growth suppression was enhanced by M1-priming: Although still cooperative with respect to the number of macrophages added, k_eff was lower (i.e., faster elimination) with M1-primed macrophages than an equivalent number of unprimed macrophages and there was a shorter delay time (SI Appendix, Fig. S4A). Additionally, the requirement for anti-Tyrp1 was relaxed, and M1-primed macrophages could eliminate unopsonized tumoroids when added at a 3:1 or 9:1 ratio to B16 whereas M0 macrophages could not (SI Appendix, Fig. S4A). Elimination of B16 tumoroids with WT CD47 levels still required opsonization with anti-Tyrp1 regardless of macrophage priming, implicating phagocytosis that is inhibited by CD47-SIRPα as the predominant effector function (SI Appendix, Fig. S4B). M2-primed macrophages were less effective against tumoroids than either M1-primed or M0 macrophages (Fig. 1C and SI Appendix, Fig. S4A).

In addition to the differences in their ability to suppress tumoroid growth, M1- and M2-primed macrophages exhibited differences in clustering in immuno-tumoroids. This was most evident at a nominal BMDM:B16 ratio of 1:1 where M1-primed macrophages clustered even more strongly than M0 macrophages while M2 macrophages dispersed throughout the tumoroid. To quantify the extent of clustering or dispersion, we computed a convex hull—i.e., the smallest convex shape that encapsulates all macrophages—and used the convex hull perimeter as a metric for comparing different experimental conditions (Fig. 1C yellow outline). The convex hull perimeter was smallest for M1-primed macrophages and largest for M2-primed macrophages (Fig. 1D). Importantly, M0 were statistically similar to M1 by this measure at 24 h and 48 h. Skewing of macrophages toward an antitumor phenotype has been reported in previous coculture models (36), and we recently showed in the immuno-tumoroid model that pretreatment of B16 cells with a drug that drives chromosomal instability tends to suppress tumoroid growth and increase macrophage M1 markers (i.e., MHCII, CD86) while suppressing M2 markers (i.e., CD206, CD163) (37). The gradual acquisition of M1-like clustering by M0 macrophages in immuno-tumoroids might therefore be part of a broader phenotypic change.

The convex hulls of the M2-primed macrophages increased with time as the tumoroid expanded due to B16 proliferation that was not effectively suppressed by phagocytosis (Fig. 1 D, Inset). On the other hand, the convex hulls of M1-primed and M0 macrophages decreased in size over time, which likely reflects a combination of macrophage clustering and tumoroid shrinkage as phagocytosis dominates proliferation. In immuno-tumoroids treated with a mixture of equal parts M1-primed and M2-primed macrophages, the B16 area and macrophage convex hull perimeter had intermediate average values between the larger M2 tumoroids and smaller M1 tumoroids (SI Appendix, Fig. S4C). Notably, the macrophage convex hull perimeters in M1+M2 tumoroids were determined almost entirely by the extent of dispersion of the M2-primed macrophages (SI Appendix, Fig. S4D), consistent with the differences observed between M1 tumoroids and M2 tumoroids (Fig. 1D). From these observations, we conclude that macrophage priming with IFNγ enhances both macrophage clustering and elimination of B16 cells from tumoroids while priming with IL4 has the opposite effect on both processes.

Differential Adhesion, Fluidity, and Sorting of Macrophage Subtypes.

We next examined clustering and dispersion of M1 and M2 monocultures on the same low-adhesion well plates used for immuno-tumoroids as a further reductionist approach to understanding differences in macrophage spatial organization (Fig. 1E). We observed that M1 macrophages form compact clusters whereas M2 macrophages instead disperse (Fig. 1F), which is similar to their respective behaviors in immuno-tumoroids (Fig. 1C). Unpolarized M0 macrophages have an intermediate phenotype and are more dispersed than M1 macrophages despite local regions where cells adhere to one another. To quantify these differences, we again computed the perimeter of a convex hull enclosing all cells (Fig. 1 F and G). The trend in the average perimeter—M1 < M0 < M2—is roughly the same as for the immuno-tumoroids, except that M0 was much closer to M1 under conditions of phagocytosis (Fig. 1D)—as emphasized. This trend is certainly consistent with the polarizing effects of IFNγ and IL4 that give rise to diametrically opposed macrophage states, which manifest here as a tendency to cluster or disperse, respectively. M1 clusters could be easily dislodged by inverting the well plate or by pipetting with a wide-bore tip but remained intact and cohesive during these manipulations. M2 macrophages tended to attach more strongly to the substrate, although this depended somewhat on the specific surface treatment. M2 macrophages on surfaces passivated by poly(2-hydroxyethylmethacrylate) tended to remain round while M2 macrophages on surfaces passivated by a commercial antiadherence surfactant treatment were frequently observed to elongate with prominent lamellipodia, particularly at the leading edge (SI Appendix, Fig. S5A).

We performed further experiments to rigorously characterize the clustering behavior of macrophage subtypes (SI Appendix, Fig. S5 and Results). Clustering induced by IFNγ is rapid but does not appear to depend on chemotactic receptors (SI Appendix, Fig. S5 B and C). The effects of IFNγ and IL4 on clustering and dispersion of macrophages are generally reversible, but the kinetics differ for transitions between different states (SI Appendix, Fig. S5 D and E) and even subtly on whether the macrophages were derived from female versus male donor mice (SI Appendix, Fig. S5F). Finally, we observed nearly identical clustering and dispersion of macrophages derived from a conditionally immortalized progenitor cell line (CIMs) (38, 39) as seen with primary BMDMs (SI Appendix, Fig. S5 G–I).

To characterize the physical state of M1 clusters, we performed micropipette aspiration using pipette tips with diameters of approximately 50 μm, which is wide enough to accommodate many cells. M1 clusters behaved as viscoelastic liquids with a short-time elastic response that gave way to cohesive flow on longer time scales (Fig. 1H), which is qualitatively similar to the behavior of B16 tumoroid aggregates (35). M1 clusters generally have a smoother surface than the more fractal perimeters of B16 aggregates, suggesting a higher surface tension (σ). Consistent with previous descriptions of liquid-like tissue aggregates (10, 15), the higher σ macrophages are located more centrally than the lower σ B16 in immuno-tumoroid cocultures Fig. 1 C and I. Moreover, in mixtures of M1 and M2 macrophages labeled with different CellTracker fluorescent dyes, both states retained the behavior observed in monocultures, thereby producing a sorting effect in which M1 macrophages occupied the center of the coculture while M2 macrophages had a more uniform distribution (Fig. 1J). From radial fluorescence profiles approximating the probability distributions of locating M1 and M2 macrophages in the mixture, we computed a Shannon entropy that was always lower for the M1 distribution compared to the M2 distribution (Fig. 1J), indicating greater order or structure of M1 macrophages.

The clustering of macrophages activated by a proinflammatory stimulus is at least somewhat generalizable. A second M1 stimulus, the toll-like receptor 4 agonist lipopolysaccharide (LPS), produced an effect similar to IFNγ on both BMDMs and CIMs, and there was no additive or synergistic effect of combining IFNγ and LPS (Fig. 2A and SI Appendix, Fig. S5 J–L). IFNα drove macrophage clustering but to a lesser extent than IFNγ or LPS. Macrophage clustering driven by IFNγ and IFNα were dose-dependent with EC50 values of ~50 pg mL⁻¹ and 1 pg mL⁻¹, respectively (SI Appendix, Fig. S6 A and B), suggesting that specific signaling events downstream of receptor activation underlie the macrophage clustering observed here.

Fig. 2.

Macrophage clustering with proinflammatory stimuli and roles for differential expression of adhesion receptors. (A) Fluorescence images and convex hull perimeters of BMDMs after 1 d on low-adhesion U-bottom well plates with 20 ng mL⁻¹ IFNα, 100 ng mL⁻¹ LPS, 20 ng mL⁻¹ IFNγ, or 20 ng mL⁻¹ IFNγ + 100 ng mL⁻¹ LPS. Mean ± SD, n = 7 to 10 wells per condition, ordinary one-way ANOVA with Tukey’s multiple comparisons test. (B) Convex hull perimeters of BMDMs after 1 d on low-adhesion U-bottom well plates in the presence of conditioned media from concanavalin-A (ConA)-activated or nonactivated mouse CD4⁺ T cells, or in unconditioned media with ConA or IFNγ. Mean ± SD, n = 3 or 4 wells per condition, ordinary one-way ANOVA with Šidák’s multiple comparisons test between indicated groups. (C) Heat map of fold changes (FC) in expression of genes encoding adhesion receptors in polarized (M1, M1’, or M2) versus unpolarized (M0) BMDMs from datasets listed in SI Appendix, Table S1. * indicates an adjusted P < 0.05 across all datasets in which the gene was detected (Top). n.d., not detected. Macrophage integrin heterodimers comprising an αL, αM, or αX subunit and the β2 subunit interacting with an ICAM-1 homodimer (Bottom). (D) Flow cytometry of M0, M1, and M2 BMDMs stained for integrin αL, αM, or αX. Histograms show staining of a representative BMDM culture and bar graphs report the median fluorescence intensity (MFI) of stained cells corrected by subtracting the MFI of an unstained sample from the same culture. Mean ± SD, n = 3 BMDM cultures from different mice, ordinary one-way ANOVA and Tukey’s multiple comparisons test on log-transformed MFI values. (E, Top) Correlation between fold changes in surface protein expression (from flow cytometry in panel D and SI Appendix, Fig. S7C) and fold changes in gene expression (from GSE69607 dataset in panel C). (Bottom) Correlation between fold changes in surface protein expression on BMDMs (panel D and SI Appendix, Fig. S7C) and CIMs (SI Appendix, Fig. S7D). (F) Generation of KO CIM progenitor lines using lentiviral transduction of single guide RNA (sgRNA) constructs to Cas9⁺ CIMs. (G) Fluorescence images of macrophages differentiated from guide control and KO CIM progenitors after 24 h on low-adhesion U-bottom surfaces under M0 and M1 conditions. (H) Heat map and hierarchical clustering dendrogram for the ratio of mean convex hull perimeters of M1 and M0 cultures on low-adhesion U-bottom surfaces. n = 3 replicate experiments.

Lymphocytes are the most likely source of IFNγ in vivo, and lymphocyte-secreted factors have long been known to promote macrophage aggregation (40). Therefore, we isolated naïve CD4⁺ and CD8⁺ T cells from mouse spleens and activated them ex vivo with concanavalin A (ConA) (Fig. 2B and SI Appendix, Fig. S6C). Adding conditioned media from activated T cells to M0 macrophages resulted in statistically significant clustering based on measurement of the convex hull perimeters, although not to the extent observed with recombinant IFNγ. Neither conditioned media from naïve T cells nor an equivalent concentration of ConA added to unconditioned media had significant effects on clustering.

Differences in Adhesion Receptor Repertoires in Polarized Macrophages.

Differences in adhesion receptors could help to explain the clustering and sorting behavior of different macrophage subsets. Therefore, we analyzed public transcriptomic data for differential expression of genes encoding such receptors in BMDMs polarized to M1 or M2 states (SI Appendix, Table S1). Of the three M1 datasets analyzed, one included macrophages polarized with IFNγ for 18 h and the two others included macrophages polarized with IFNγ + LPS for 24 h. Because IFNγ + LPS produced a similar level of macrophage clustering as IFNγ alone (Fig. 2A and SI Appendix, Fig. S5 J–L), we presumed that any key differences in gene expression underlying this effect would be reflected in both conditions. Nonetheless, we refer to macrophages activated with IFNγ + LPS as M1’ macrophages to avoid confusion. In both M2 datasets analyzed, BMDMs were polarized with IL4 for 24 h.

Although multiple types of cell-surface receptors have been implicated in macrophage adhesion or fusion, we chose to focus on differential expression of integrins considering they mediate both cell–cell and cell–ECM interactions (Fig. 2C). Macrophages express multiple integrins including heterodimers comprising αL, αM, αX, or αD subunits and the leukocyte-specific β2 subunit. Expression of Itgam and Itgb2 encoding the αM and β2 subunits, respectively, did not change significantly upon M1 or M2 polarization, and Itgad encoding the αD subunit was of very low abundance in the RNA-seq data and not detected in either microarray dataset. On the other hand, expression of Itgal encoding the integrin αL subunit was upregulated >fivefold in the two M1’ datasets in which it was detected, while expression of Itgax encoding the integrin αX subunit was upregulated >fourfold in both M2 datasets. The αLβ2 integrin (also known as LFA-1) is expressed widely across immune cells and best known to mediate cell–cell adhesion in the context of leukocyte transendothelial migration (41) and the immune synapse (42). Expression of Icam1 encoding the integrin ligand intercellular adhesion molecule-1 (ICAM-1) was also upregulated >threefold in all three M1 datasets, providing a candidate receptor–ligand pair that could mediate homotypic cell–cell adhesion between M1 macrophages (Fig. 2 C, Bottom).

Gene expression measurements in the M1’ and M2 RNA-seq dataset were made at multiple time points (43), allowing us to compare the kinetics of gene expression changes to the kinetics of macrophage clustering (SI Appendix, Fig. S5C). Itgal and Icam1 expression reached peak levels after only 4 to 6 h in M1’ macrophages but remained elevated after 24 h relative to M0 macrophages (SI Appendix, Fig. S7A). Such a rapid change in the levels of genes encoding adhesion receptors is consistent with our studies of macrophage clustering kinetics in which statistically significant differences between M1 and M0 macrophages emerged within 6 h after plating in U-bottom wells (SI Appendix, Fig. S5C). The same rapid induction of Itgal and Icam1 occurred following M2 to M1 repolarization (SI Appendix, Fig. S7B), which is also consistent with our clustering results with repolarized BMDMs (SI Appendix, Fig. S5D). Therefore, increased cell–cell adhesion mediated by increased levels of αLβ2 integrin and ICAM-1 could account for M1 clustering, at least based on changes in gene expression.

Gene expression changes reported by others were reflected in changes to surface protein levels in our polarized BMDMs and CIMs. Flow cytometry revealed upregulation of integrin αL and ICAM-1 in M1 macrophages and upregulation of integrin αX in M2 macrophages (Fig. 2D and SI Appendix, Fig. S7C). There was a high degree of correlation between changes at the transcript level and surface protein level and also between surface protein level changes in BMDMs and CIMs (Fig. 2E and SI Appendix, Fig. S7D). BMDMs treated with LPS for 24 h upregulated integrin αL and ICAM-1 to a similar extent as BMDMs treated with IFNγ, although the change in ICAM-1 was not statistically significant (SI Appendix, Fig. S7E). BMDMs treated with IFNα had only a small increase in integrin αL surface expression and no significant change in ICAM-1 (SI Appendix, Fig. S7E). Overall, these measurements are consistent with the similar clustering observed for macrophages activated with IFNγ, LPS, or IFNγ + LPS, which was greater than clustering of macrophages activated with IFNα.

We undertook two approaches to determine the functional significance of differential integrin expression on macrophage clustering. First, we assessed clustering and dispersion of M0, M1, and M2 BMDMs in the presence of blocking antibodies against the αL, αM, and β2 integrin subunits and ICAM-1 as well as anti-integrin β1 and isotype control antibodies. None of these blocking antibodies had a significant effect on M1 clustering (SI Appendix, Fig. S8A). Instead, anti-integrin αM and β2 drove tight clustering of both M0 and M2 BMDMs. Similar results were obtained with anti-integrin αM and β2 F(ab’)₂ fragments, indicating macrophage Fc receptors were not likely involved (SI Appendix, Fig. S8B). While such clusters could result from disrupting cell–substrate adhesions involving αMβ2 integrin, the relatively high expression of αM compared to αL and αX led us to instead suspect that the antibodies were agglutinating macrophages.

The inability of any blocking antibodies to prevent M1 clustering and the likely agglutinating activity of some anti-integrin antibodies prompted us to seek an alternative approach. We therefore took advantage of Cas9 nuclease expression in CIMs (39) to knock out the αL, αM, αX, or β2 integrin subunits or ICAM-1. We generated lentiviruses to express single guide RNA sequences targeting the genes encoding those proteins and transduced Cas9⁺ CIM progenitors to generate stable KO lines (Fig. 2F and SI Appendix, Fig. S7F). Clustering in the presence of IFNγ was unaffected in αM KO or αX KO CIMs but was significantly abrogated in αL, β2, or ICAM-1 KO CIMs (Fig. 2G). To quantify differences in M1 clustering, we computed the ratio of the average M1 and average M0 convex hull perimeters for each KO line (Fig. 2H). Given that none of the KO lines exhibited obvious differences in the M0 state, a higher ratio indicates less compact clustering induced by M1 polarization whereas a lower ratio indicates more compact clustering. We performed hierarchical cluster analysis of the M1/M0 ratios for the different cell lines and found the longest distance to be between the cluster containing the αL KO, β2 KO, and ICAM-1 KO lines and the cluster containing the αM KO, αX KO, and nontargeting guide control lines. These results strongly support a role for αLβ2 integrin binding to ICAM-1 in mediating cell–cell adhesions among macrophages in M1 clusters.

Upregulation of Adhesion Receptors Under Phagocytic Conditions in Tumoroids.

Macrophage clustering in IgG-opsonized immuno-tumoroids is enhanced by M1-priming, but clustering is still observed with unprimed macrophages (Fig. 1 C and D). In addition to well-known effects on actin polymerization, Fc receptor activation initiates signaling that leads to transcriptional changes on the order of hours (32), a relevant timescale for immuno-tumoroids. This raises the possibility that signaling associated with phagocytosis upregulates surface receptors to promote macrophage cell–cell adhesion as observed with IFNγ- and LPS-treated macrophages. To investigate this, we collected and pooled macrophages from immuno-tumoroids after 18 h and stained them for integrin αL and ICAM-1 as well as MHCII. As expected from our recent study (35), there were significantly more phagocytic macrophages (CD45⁺GFP⁺) in opsonized immuno-tumoroids compared to unopsonized immuno-tumoroids (SI Appendix, Fig. S9 A–C). Parallel cultures in which macrophages were fluorescently labeled with a CellTracker dye and added to immuno-tumoroids for imaging over the course of several days revealed a near-complete elimination of B16 from opsonized immuno-tumoroids with evidence for macrophage clustering followed by dispersion based on the convex hull perimeters (SI Appendix, Fig. S9 D–F). However, the kinetics for elimination, clustering, and dispersion varied considerably from well to well, which is expected to introduce significant variability in measurements of pooled immuno-tumoroids. Nonetheless, we proceeded to compare integrin αL, ICAM-1, and MHCII surface expression on phagocytic macrophages versus nonphagocytic macrophages (CD45⁺GFP⁻). The histogram distributions of fluorescence intensity levels were shifted rightward toward higher values for all three markers on phagocytic macrophages relative to nonphagocytic ones (SI Appendix, Fig. S9G). Considering these rightward shifts, we plotted the median, 75th percentile, and 95th percentile values against the percentage of phagocytic macrophages and determined that the 95th percentile of ICAM-1 and integrin αL and the 75th percentile of ICAM-1 are significantly correlated with the level of phagocytosis (SI Appendix, Fig. S9G). Together, these findings suggest that phagocytic conditions in immuno-tumoroids might enhance macrophage clustering in part through increased surface expression of the adhesion receptors αLβ2 and ICAM-1 (SI Appendix, Fig. S9H).

Differential Actomyosin Contractility Can Influence Clustering.

To determine how macrophage clustering via these receptors might enhance phagocytosis, we next considered their linkage to the actin cytoskeleton via their cytoplasmic domains. Changes in actin organization could conceivably contribute to the differential clustering behavior of macrophage subtypes and would also affect Fcγ receptor–mediated phagocytosis, which is driven by actin polymerization that pushes pseudopod extensions around the target (44). To investigate this possibility, we performed the macrophage clustering assay in the presence of latrunculin A (LatA), an actin polymerization inhibitor, or para-amino-blebbistatin (NH₂-blebb), an inhibitor of myosin-II-mediated contractility (Fig. 3 A and B). M1 clustering was abrogated in the presence of LatA while the dispersion of M0 macrophages was modestly but significantly decreased (Fig. 3A). On the other hand, M0 macrophages were significantly more clustered in the presence of NH₂-blebb while clustering of M1 macrophages was unaffected (Fig. 3B). Thus, while our results demonstrate a clear requirement for actin filaments in macrophage clustering, they also reveal that clustering is opposed by myosin-II contractility and suggest that myosin activity might be lower in M1 macrophages relative to M0 and M2 macrophages.

Fig. 3.

SRF contractile-adhesion pathway is suppressed in M1 macrophages and promotes cell clusters. (A and B) Representative fluorescence images and convex hull perimeters of M0 and M1 BMDMs as a function of time on low-adhesion U-bottom well plates with 1 μM LatA, 50 μM NH₂-blebb (myosin-II inhibitor), or DMSO control. Bar graphs are after 24 h. Mean ± SD, n = 7 to 10 wells per condition, ordinary two-way ANOVA with Tukey’s multiple comparisons test. (C) Heat maps of fold changes in expression of (Left) SRF target genes including actin bundling proteins, adhesion proteins, and nuclear lamina/LINC complex proteins and (Right) genes encoding actin branching-associated factors between polarized (M1, M1’, or M2) and unpolarized (M0) BMDMs from datasets in SI Appendix, Table S1. Schematic (Center) of the hypothesized antagonism between cortical actomyosin cytoskeleton and pseudopod protrusions driven by branched actin. Fluorescence images of M1 and M2 BMDMs stained for F-actin. (D) Immunoblots of lysates from BMDMs polarized in M0, M1, or M2 medium for 2 d probed with anti-α-actinin, anti-vinculin, or anti-lamin-A,C. The band densities normalized by total protein staining were used to compute M1/M0 and M2/M0 fold changes, which are visualized for α-actinin and vinculin as a scatterplot to emphasize coregulation as SRF targets. Mean ± SD, lysates from each donor (depicted by shape) were loaded in triplicate, Pearson correlation (two-tailed). (E) Immunoblots of lysates from macrophages differentiated from lamin-A,C KO and nontargeting control guide CIM progenitors probed with anti-vinculin and anti-lamin-A,C. The densities of vinculin bands were normalized by total protein staining and then by the density of nontargeting control guide 1. Mean ± SD, n = 3 blots, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. (F) Fluorescence images and convex hull perimeters of macrophages differentiated from lamin-A,C KO and nontargeting control guide CIM progenitors after 1 d on low-adhesion U-bottom well plates. Mean ± SD, n = 8 wells lamin-A,C KO and 9 wells nontargeting control guide, Student’s t test (unpaired, two-tailed). (G) Quantification of phagocytosis of 7 μm diameter IgG-opsonized polystyrene beads or WT B16 by macrophages differentiated from lamin-A,C KO or WT CIM progenitors. Mean ± SD, n = 7 to 9 fields of view per condition, Mann–Whitney test, two-tailed. (H) Summary of differences between lamin-A,C KO and WT macrophages.

F-actin organizes into multiple types of networks in cells (45). To begin to understand which actin structures might be most affected by macrophage polarization, we again analyzed public transcriptomic datasets (SI Appendix, Table S1) for genes involved in cytoskeleton organization, adhesion complexes, and mechanosensing, which revealed differential expression of many such genes including Actn1, Vcl, Lmna, and others in M1 and M2 BMDMs. In other cell types including mesenchymal stem cells (MSCs), such genes are part of the serum-response factor (SRF) pathway and are downregulated on soft substrates that generally limit cellular contractility (46, 47). Therefore, we specifically examined the expression of actin- and cytoskeleton-associated genes that were previously shown to be differentially expressed in peritoneal macrophages elicited from WT and Srf KO mice (48). Many of these SRF-regulated genes are indeed downregulated in M1 or M1’ BMDMs and upregulated in M2 BMDMs after 1 d of polarization, although kinetic profiles suggest more complicated regulation including increased expression within ~1 to 2 h in both M1’ and M2 BMDMs (Fig. 3 C, Left and SI Appendix, Fig. S10A). We focused on two of the most strongly differentially expressed genes, Actn1 and Vcl that encode the actin crosslinking protein, α-actinin, and the mechanosensitive adhesion protein, vinculin, respectively. In addition to differential expression observed in WT versus Srf KO mouse macrophages (48), human ACTN1 and VCL have SRF binding peaks in their promoter or intronic regions in the ENCODE ChIP-seq database (SI Appendix, Fig. S10B). Western blotting of lysates from M0, M1, and M2 BMDMs across multiple donor mice revealed differential expression at the protein level, and the correlation between α-actinin and vinculin levels strongly suggests coregulation (Fig. 3D and SI Appendix, Fig. S11A). In contrast to M1 BMDMs activated with IFNγ, BMDMs treated with interferon-β from the same microarray dataset do not show downregulation of SRF target genes (SI Appendix, Fig. S10C), which would be consistent with the relative lack of macrophage clustering we observed with IFNα (Fig. 2A and SI Appendix, Fig. S5 J and K), another type-I interferon that signals through shared receptors.

SRF regulates its own expression, and Srf and its cofactors Mkl1 and Mkl2 show slight but generally nonsignificant downregulation in M1 BMDMs and upregulation in M2 BMDMs after 1 d of polarization (Fig. 3 C, Left) with similar kinetic profiles to Vcl and Actn1 (SI Appendix, Fig. S10A). In contrast to these genes, genes encoding proteins involved in actin branching including components of the Arp2/3 complex and SCAR/WAVE complex are not significantly affected by macrophage polarization, at least at the 24 h timepoint analyzed here (Fig. 3 C, Right).

The Lmna gene encoding the nuclear envelope protein lamin-A and the splicing product lamin-C is among the genes differentially regulated across all the transcriptomic datasets we analyzed, and genes encoding components of the linker of nucleoskeleton and cytoskeleton (LINC) complex (Sun2 and Syne2) are also downregulated in M1’ macrophages (Fig. 3C). We previously showed that lamin-A,C knockdown in MSCs feeds back to downregulate SRF target gene expression (47), and Lmna was also shown to be downregulated in BMDMs in response to LPS with further regulation of lamin-A,C levels by protein degradation (49). We confirmed differential expression of lamin-A,C among M0, M1, and M2 BMDMs by Western blotting (Fig. 3D) and proceeded to determine the functional consequences by knocking out lamin-A,C entirely using Cas9⁺ CIM progenitors (Fig. 3E and SI Appendix, Fig. S11B). Similar to M1 BMDMs, macrophages differentiated from lamin-A,C KO CIM progenitors had decreased levels of vinculin and α-actinin protein and were more clustered on low-adhesion substrates than control macrophages (Fig. 3 E and F and SI Appendix, Fig. S11C). Lamin-A,C KO macrophages also exhibited increased phagocytosis of both IgG-opsonized latex particles and anti-Tyrp1-opsonized B16 cells (Fig. 3G). These results suggest that lamin-A,C levels regulate SRF target genes in macrophages as observed for MSCs and that this influences macrophage functions including phagocytosis (Fig. 3H), with similar pathways potentially operating in M1 macrophages.

Based on our transcriptomic analysis and functional studies, we hypothesized that weaker cortical actomyosin in M1 macrophages would pose less of a barrier to the generation of cellular protrusions, including pseudopod protrusions involved in phagocytosis (Fig. 3 C, Center). Addition of blebbistatin to mesenchymal cells often leads to more dendritic-shaped cells (50), consistent with the hypothesis. Likewise, M1 BMDMs appear more dendritic and less spread than M2 BMDMs on rigid glass substrates (Fig. 3C). Addition of NH₂-blebb promotes clustering of M0 BMDMs on low-adhesion surfaces (Fig. 3B) but does not impede elimination of the B16 CD47 KO cells from immuno-tumoroids (SI Appendix, Fig. S12 A–C). Together, this all suggests that the cytoskeleton organization that favors clustering of macrophages might also be highly conducive to phagocytosis.

Macrophages Extend Intrusive Pseudopods as Intrudopodia to Disrupt Solid Tumor Adhesions.

The cohesion of solid tumors stems in part from cell–cell adhesions between cancer cells, which vary considerably among different tumors and are of particular interest in governing invasion and metastasis (51). Adhesions between cancer cells are also likely to act as physical barriers to the extension of pseudopods from phagocytes, which would thereby limit engulfment, but studies of phagocytosis of adherent or cohesive targets are generally lacking (35). It is conceivable that macrophages first disrupt target cohesion and then begin to engulf detached cells, which we hypothesized might occur through either proteolysis of cell–cell adhesion molecules or myosin-II-dependent pulling forces (SI Appendix, Fig. S12A). However, we observed that immuno-tumoroid clearance proceeds normally in the presence of a protease inhibitor cocktail and in the presence of NH₂-blebb (SI Appendix, Fig. S12 B–D), suggesting that neither mechanism contributes significantly to phagocytic elimination of cohesive solid tumor cells.

To better visualize phagocytosis of cohesive targets, we generated three-dimensional (3D) spheroids of B16 CD47 KO cells using growth medium containing methylcellulose and then transferred spheroids onto coverslips for time-lapse imaging over the course of several hours after adding macrophages and anti-Tyrp1 (Fig. 4 A and B and SI Appendix, Fig. S13). After 1 to 2 d, spheroids can be eliminated similar to tumoroids (Fig. 1B), even though spheroids are more cohesive and less fractal than tumoroids. At early time points, macrophages were seen to form small clusters on the spheroid periphery rather than forming large central aggregates per tumoroid clusters (Fig. 1C), but such interfacial accumulation is similar to that observed for antitumor macrophages in human tumors (SI Appendix, Fig. S1). Macrophage interactions with B16 spheroids included protrusions wedging between B16 cells and pseudopod-like protrusions several microns in height (SI Appendix, Fig. S13 A, ii) that with time disrupted B16 cell–cell adhesions (Fig. 4B and SI Appendix, Fig. S13 A–C). We term the latter structures intrusive pseudopods or intrudopodia and propose that they play key roles in disrupting adhesions among solid tumor cells and thereby facilitate phagocytosis.

Fig. 4.

Intrusive pseudopods or intrudopodia from clustered macrophages can help disrupt cohesive tumors. (A) Spheroids of B16 CD47 KO cells imaged during phagocytosis. (B, i) Max intensity projection of a confocal stack showing a B16 CD47 KO (green) spheroid to which BMDMs (magenta) and Hoechst 33342 DNA stain (blue) were added. (ii) Single confocal slice at multiple time points corresponding to the region of interest (ROI) in the yellow box in i. Cell outlines of B16 and macrophages in green and magenta, respectively, are shown beneath the zoomed slices. Labeled events of interest include phagocytosis (p), nascent intrudopodia wedging between multiple B16 (>), and intrudopodia between B16 (*) including coordinated intrudopodia extended by neighboring macrophages labeled M_a, M_b, and M_c. (iii) 3D-rendering of the region outlined in ii. Purple arrows point to intrudopodia extended by macrophages M_a, M_b, and M_c. See SI Appendix, Fig. S13A for rendering without B16 cells and rotation to visualize a 3D intrudopod. (iv) Time series of macrophage coordinated detachment of a B16 cell from the periphery of a spheroid. See SI Appendix, Fig. S13B for max intensity projection of this spheroid and a time series at a second ROI. (C) Summary of phagocytic events and intrudopodia observed for n = 4 spheroids as a function of initial spheroid volume calculated from diameters. (D) Maximum intensity projection confocal microscopy images of metastatic nodules immunostained for F4/80 (same as SI Appendix, Fig. S2A). Macrophage cell outlines (magenta) and B16 nuclei outlines (blue) from zoomed regions of interest with putative intrudopodia denoted by *’s. (E) Hypothesized advantage of clustering macrophages for phagocytosis of cohesive targets.

Analysis of multiple spheroids reveals that intrudopodia and phagocytosis are relatively common. Moreover, two nearby macrophages sometimes extend intrudopodia toward one B16 cell, which is ultimately dislodged and engulfed by one of the macrophages (Fig. 4B and SI Appendix, Fig. S13 A–C). The overall number of intrudopodia over time indeed exceeded the number of successful phagocytic events by ~fivefold for each spheroid (Fig. 4C). Given that individual macrophages exhibit no more than 1 to 2 pseudopodia or intrudopodia at any given time, this overall excess relative to phagocytic events is consistent with multiple intrudopodia cooperating for phagocytosis. Also, some intrudopodia meander into the spheroid but do not lead to productive engulfment (SI Appendix, Fig. S13D).

To determine whether intrudopodia might be relevant to phagocytosis in vivo, we reexamined F4/80-stained sections of B16 CD47 KO lung metastases from mice treated with anti-Tyrp1 injections and focused on regions with aggregated macrophages. We observed macrophages that appeared to form wedges and intrudopodia between B16 cells based on their extensions between B16 nuclei (Fig. 4D). While fixed sections provide only a single time point that does not reveal intrudopod dynamics, the extension of such structures is nonetheless consistent with our observations of phagocytic macrophages in spheroids.

Antitumor Macrophages in Human Tumors Express High Levels of ITGAL and Low Levels of Actin Crosslinking Genes.

To begin to extend our finding to human cancers, we next sought to compare in vitro polarized M1 and M2 macrophages with recently identified macrophage subtypes in human tumors. We focused on CXCL9⁺ TAMs (1) that accumulate at the interface between tumor nests and stroma (SI Appendix, Fig. S1). These macrophages associate with favorable patient outcomes and are therefore considered antitumor TAMs in contrast with protumor SPP1⁺ TAMs. We again analyzed BMDM transcriptomic data and observed significant upregulation of Cxcl9 and significant downregulation of Spp1 in all three M1 datasets whereas neither gene changed significantly in the M2 datasets (Fig. 5A). The “CS ratio” of Cxcl9 counts to Spp1 counts from the RNA-seq dataset was very low in M0 and M2 BMDMs but was of order one in M1 BMDMs (Fig. 5B and SI Appendix, Fig. S14A). It should be noted, however, that Spp1 expression is high across all three in vitro macrophage types and changes much less drastically than Cxcl9 upon polarization (SI Appendix, Fig. S14A). In that sense, in vitro polarized macrophages differ from bona fide CXCL9⁺ TAMs and SPP1⁺ TAMs in which CXCL9 and SPP1 expression appeared mutually exclusive by single-cell RNA-seq and immunofluorescence analyses (1). Furthermore, both CXCL9⁺ TAMs and SPP1⁺ TAMs express a combination of M0, M1, and M2 gene signatures rather than a gene expression profile dominated by one particular type of in vitro polarized macrophage (1). Based on these observations, we consider M1 BMDMs used in our tumoroid and clustering assays to be a type of CXCL9⁺ macrophage but distinct from antitumor CXCL9⁺ TAMs (Fig. 5C and SI Appendix, Fig. S14B). Likewise, M0 and M2 BMDMs could be considered CXCL9⁻ SPP1⁺ macrophages but are almost certainly distinct from the protumor SPP1⁺ TAMs (Fig. 5C and SI Appendix, Fig. S14B).

Fig. 5.

ITGAL expression in human tumors correlates with antitumor macrophage abundance and patient survival. (A) Heat map of fold changes in expression of Cxcl9 and Spp1 between polarized (M1, M1’, or M2) and unpolarized (M0) BMDMs from datasets listed in SI Appendix, Table S1. * indicates an adjusted P < 0.05 across all three M1 or M1’ datasets. (B) Ratio of Cxcl9 to Spp1 counts in in vitro polarized BMDMs calculated from fragments per kilobase million (FPKM) reported in GSE158094 in ref. 43. (C) Visualization of CXCL9 expression in in vitro polarized BMDMs and TAMs from human cancers. DN TAMs = double-negative (CXCL9⁻SPP1⁻) TAMs. (D) Spearman correlation coefficients between gene expression in macrophages and the CS TAM ratio in human head and neck squamous cell carcinoma produced from data reported in ref. 1. Correlation coefficients are arranged by rank. The CS TAM ratio describes the relative expression of CXCL9 in TAMsto SPP1 in TAMs in a patient's tumor. Other genes with macrophage expression levels that correlate with the CS TAM ratio and that encode proteins implicated in phagocytosis or actin crosslinking and contractility are indicated. (E and F) Kaplan–Meier survival plots for human liver cancer (E) as a representative cancer type and summary plots of hazard ratios (HR) across additional types of human cancers (F). Patients were stratified based on the ratio of ITGAL to SPP1 expression (top three plots in E and left plot in F) or the ratio of CXCL9 to SPP1 expression (bottom plot in E and right plot in F). ITGAL/SPP1 ratio analyses were performed separately for all patients, for patients with tumors enriched in macrophages, and for patients with tumors having low numbers of CD8⁺ T cells using options available in the KMplot pan-cancer RNA-seq webtool. CXCL9/SPP1 ratio analyses were performed for all patients using KMplot and are shown together with data from ref. 1 to demonstrate comparable effects on survival in both sets of patient data. n.d., no data.

Despite these differences, the shared role for IFNγ in activating M1 BMDMs and CXCL9⁺ TAMs prompted us to consider whether human CXCL9⁺ TAMs might exhibit some of the features suggested by our studies to be important for solid tumor phagocytosis. We therefore analyzed the correlations reported in ref. 1 between gene expression in specific cell types and the abundance of antitumor TAMs relative to protumor TAMs (CS TAM ratio) across different patient tumors. This analysis revealed that macrophage expression of ITGAL and phagocytosis-related genes (SLAMF7 and FCGR3A) correlate positively, indicating that these genes are likely expressed by antitumor macrophages (Fig. 5D). On the other hand, macrophage expression of multiple genes encoding actin crosslinking and bundling factors (FLNB, TNS1, FSCN1, and ACTN4) correlate negatively. A positive correlation was also observed for ITGAL expression in three other cell types (dendritic cells, B cells, and T cells), which is similar to the positive correlation across multiple immune cell types observed for the transcription factor STAT1 that lies downstream of IFNγ signaling (SI Appendix, Fig. S14C). We therefore analyzed the promoter region of human ITGAL by querying ChIP-seq experiments in the ENCODE database. Our analysis revealed the presence of binding peaks for STAT1 as well as the macrophage master regulator transcription factor PU.1 and known ITGAL regulators (52) RUNX3 and CEBPβ (SI Appendix, Fig. S14D). Importantly, high expression of ITGAL associates with patient survival across many different human cancer types (Fig. 5 E and F). Although other immune cells including T cells would be expected to contribute to this effect, the trend remained even when the analyses were limited to subsets of patients with high numbers of macrophages or decreased numbers of CD8⁺ T cells. Thus, despite the limitations of the M1/M2 polarization paradigm in capturing the full complexity of macrophage subtypes in tissues, differences in adhesion receptor expression and actin organization that promote clustering of mouse macrophages in our reductionist approaches are also potential features of antitumor macrophages in human cancer and associate with favorable patient outcomes.

Discussion

Macrophage clustering and possible fusion to multinucleated giant cells are ways in which macrophages can contend with objects that are large and difficult to engulf. Descriptions of clusters, aggregates, and nests of tumor macrophages including adoptively transferred CAR-M (1–5) raise many questions about the mechanisms by which they form and any unique functions enabled by this spatial organization. Here, we associate macrophage clustering with phagocytosis of solid tumors and tumoroid models, determining that clustering results from upregulation of specific adhesion receptors on macrophages (Fig. 2) and from decreased actomyosin contractility (Fig. 3). Low actomyosin contractility and macrophage clustering seem to promote phagocytosis of cohesive targets in two ways: i) by enhancing pseudopod formation or extension (Fig. 3) and ii) by increasing the density of macrophages locally to permit coordination between intrusive pseudopods or intrudopodia for more effective disruption of target cell adhesions (Fig. 4). An analogy for such coordination might be found in American football where multiple defensive lineman use their arms and legs to pull and push through opposing players in an offensive line. Such analogies always have their limits, but immune cells are the body's defense against a tumor's evolving offense.

To better position our findings within a contemporary view of macrophage polarization, we focused on recently identified CXCL9⁺ TAMs that associate with favorable patient prognosis across many cancer types (1). CXCL9⁺ TAMs have a strong IFNγ activation signature consistent with their localization near T cells and also cluster with one another at the tumor nest–stroma interface in head and neck cancer and non–small cell lung cancer. CXCL9⁺ TAMs may represent a subset of other IFNγ-activated TAMs identified on the basis of IL4I1 (interleukin-4-induced gene 1) expression (19, 20). In colorectal cancer, IL4I1⁺ TAMs associate with increased patient survival and also exhibit phagocytic activity (20), although phagocytosis was most likely directed against apoptotic cell corpses rather than proliferating tumor cells as we investigated here using immuno-tumoroids. Our analysis of cell-type–gene correlations reported in ref. 1 revealed that macrophage ITGAL expression correlates with a high ratio of antitumor CXCL9⁺ TAMs to protumor SPP1⁺ TAMs whereas several actomyosin-related genes (ACTN4, FLNB, and FSCN1) anticorrelate with this ratio. These findings align with our observations that M1 BMDMs upregulate Itgal (Fig. 2 C and D) and downregulate actin-bundling genes (Fig. 3C). Increased expression of integrin αL has been reported in several studies of in vitro macrophages activated with inflammatory stimuli (53–56) and may involve several signaling pathways such as STAT1, STAT5, and MAPK-family members JNK and ERK1/2. The latter is suggested by studies of T cell receptor activation (57) and may have relevance to ERK1/2 signaling downstream of macrophage Fc receptor activation (32, 58). Importantly, a high ITGAL:SPP1 ratio associates with survival across many cancer types similar to the CXCL9:SPP1 ratio, and this association applies to patients with high macrophage density or low CD8⁺ T cell density (Fig. 5 E and F). Thus, while the antitumor activity of CXCL9⁺ TAMs has been interpreted primarily through interactions with T cells, our experiments suggest such TAMs might also have advantages in tumor cell phagocytosis and as important effectors of macrophage checkpoint blockade and related therapies.

Our study provides evidence for extension of cellular protrusions between solid tumor cells as a necessary step for phagocytosis of such targets. Furthermore, the possibility for macrophage clusters to extend multiple intrudopodia in a temporally coordinated and spatially cooperative manner would lead to more efficient disruption of target adhesions than intrudopodia from an isolated macrophage (Fig. 4E). Intrudopodia seem distinct, but filopodia-like protrusions and cellular “arms” have been proposed to i) overcome spatial confinement and enable cooperativity between phagocytic cells (59, 60), and also ii) detach surface-bound Escherichia coli for phagocytosis (61). The disruption of cell adhesions for phagocytosis does have parallels to the disruption of endothelial junctions during leukocyte diapedesis through a blood vessel wall, which can involve mechanical sensing by leukocytes to identify the route that offers the least mechanical resistance (62). Live imaging of macrophages in fly embryos and zebrafish demonstrates how macrophages exploit transient disruption of cell adhesions during mitotic rounding to invade into a tissue (63) and conversely how macrophage protrusions disrupt tumor cell adhesions to enhance invasiveness (64). The immuno-tumoroid models described here should be useful to further dissect relationships between target adhesion and macrophage phagocytosis, which can help inform macrophage reprogramming for cancer therapy (25, 65, 66).

Materials and Methods

Additional Methods and Methods available in SI Appendix.

Tumoroid Assays and Macrophage Aggregation Assays.

For immuno-tumoroid assays, B16 cells were prepared as single-cell suspensions in Roswell Park Memorial Institute (RPMI) 1640 growth medium at a density of 1 × 10³ or 1 × 10⁴ mL⁻¹ depending on the experiment, and 100 μL was added per well to previously prepared low-adhesion 96-well plates. The next day, BMDMs were detached from Petri dishes with 1× TrypLE Express and labeled with 1 μM amine-reactive CellTracker Deep Red dye (Invitrogen C34565) in phosphate-buffered saline (PBS). Suspensions were incubated at 37 °C for 20 min, centrifuged at 300× g for 5 min, and washed with growth medium. In some experiments, BMDMs were primed for 48 h in differentiation medium containing IFNγ or IL4 prior to use in immuno-tumoroid assays. Macrophage suspensions were adjusted to densities between 1.5 × 10³ and 5 × 10⁴ mL⁻¹ in RPMI growth medium containing 120 ng mL⁻¹ macrophage colony-stimulating factor (M-CSF). For opsonized tumoroids, 120 μg mL⁻¹ anti-Tyrp1 was added to the macrophage suspension. After acquiring images of B16 tumoroids corresponding to t = 0, 20 μL of the macrophage suspension was added to each well and the plates were incubated at 37 °C for up to 4 d.

For macrophage clustering assays, BMDMs and differentiated CIMs were detached from Petri dishes with TrypLE Express and labeled with CellTracker Deep Red or Vybrant carboxy-fluorescein diacetate succinimidyl ester Cell Tracer (Invitrogen V12883) as described above. The cell density was adjusted to 1 × 10⁴ mL⁻¹ in RPMI growth medium containing 20 ng mL⁻¹ M-CSF, and 100 μL was added to each well of previously prepared low-adhesion 96-well plates. Cell suspensions and other solutions were passed through 40 μm nylon cell strainers (Falcon 352340) to minimize interference for particulate debris.

3D Spheroid Phagocytosis.

To generate spheroids, B16 CD47 KO cells were added at a density of 50 cells per well into nonadherent 96-well U-bottom plates pretreated with 1% F-127 pluronic (Sigma-Aldrich P2443) solution and cultured in RPMI-based growth medium. After 24 h, the culture medium was replaced with RPMI growth medium containing 1% w/v methylcellulose (Thermo Scientific Chemicals 428430500) to promote spheroid formation. Spheroid formation was monitored, and 24 h after the addition of methylcellulose, spheroids were washed with PBS to remove methylcellulose and moved to a 384-well plate amenable to time-lapse microscopy. The spheroids were then opsonized with 20 µg mL⁻¹ anti-Tyrp1, and BMDMs were added to each well at a 5:1 BMDM:B16 ratio. Time-lapse imaging of spheroids and BMDMs was performed using a Zeiss LSM 880 laser scanning confocal microscope equipped with an LD C-Apochromat 40×/1.1 NA water immersion objective and an environmental chamber maintaining the atmosphere at 37 °C and 5% CO₂. Imaging was initiated immediately following the addition of BMDMs, with time-lapse images captured every ~20 min for up to 8 h. The imaging parameters, including laser settings and acquisition conditions, were optimized to minimize photobleaching and maintain cellular viability throughout the experiment.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

There are no data underlying this work.