A central mechanism by which TAM can affect cancer progression is through phagocytosis of tumor cells. Most eukaryotic cells can engulf small particles by endocytosis, but only professional phagocytes, including macrophages and DC, uptake particles bigger than ~0.5 µm by phagocytosis.19 Molecular mechanisms of phagocytosis have been reviewed in detail elsewhere20; in this review, we focus on the factors modulating phagocytic activity and its impact on tumor control.

Interactions between tumor cells and TAM that regulate phagocytosis are the result of “eat-me” ligands (eg, calreticulin, SLAMF7, opsonizing antibodies, phosphatidylserine (PtdSer)) and “don’t eat-me” ligands (eg, CD47, PD-L1, major histocompatibility complex (MHC) I)21 expressed on the surface of tumor cells, which bind to specific receptors on macrophages (see figure 2 for details). The balance between these different functional classes of molecules exposed on cancer cells dictates initiation of phagocytosis. If “eat-me” signals prevail, they induce the rearrangement of the actin cytoskeleton in phagocytes, driving cell engulfment. Among the “don’t eat-me” ligands, CD47 binding to its receptor signal-regulatory protein α (SIRPα) on macrophages inhibits myosin II conformational changes and thereby inhibits phagosome formation.22 Inhibition of CD47 or SIRPα with antagonistic antibodies increases the phagocytic activity of TAM and decreases tumor growth in various preclinical models such as glioblastoma,23 melanoma,24 lymphoma,25 breast26 and colorectal cancer.27 Encouraged by these promising results, more than a dozen of phase I clinical trials are ongoing.28 However, it is still unknown whether TAM alone are sufficient to control tumor growth or whether an involvement of T cells is required, as we will now discuss.

The use of antagonistic anti-CD47 antibody to control growth of human tumors xenografted in NOD/SCID/γ (NSG) mice lacking T cells was efficacious for medulloblastoma and pediatric glioblastoma,23 adult glioblastoma,29 as well as ovarian, bladder, colorectal and breast tumors.30 These different studies confirmed the effect in immunocompetent mice, without, however, formally excluding the involvement of T cells. Using an antagonistic SIRPα antibody in the Raji lymphoma model,25 a synergy with rituximab was observed, resulting in increased phagocytosis, highlighting the key role of antibody-dependent cell phagocytosis (ADCP) in the context of CD47-SIRPα inhibition. Similar involvement of ADCP was reported in the context of direct CD47 blockade.31 Of note, in the aforementioned xenograft models, antihuman anti-CD47 antibody was used, enabling administration of high antibody doses without on-target off-tumor toxicity. The latter would be expected in a syngeneic setting because of ubiquitous CD47 expression by normal murine cells, such as erythrocytes. Xenograft studies therefore potentially over-estimate the efficacy of this approach. To limit the impact of anti-CD47 on healthy tissues, Dheilly et al have developed bispecific antibodies, one arm targeting CD47 and the other a tumor specific antigen.32 In patients, this approach should provide a better specificity of anti-CD47 for cancer cells and reduce side effects. Besides CD47 inhibition, other strategies increasing TAM phagocytic activity have been tested. In an immunocompetent model of pancreatic ductal adenocarcinoma (PDAC), the toll-like receptor (TLR) 9 agonist CpG impacted TAM lipid metabolism and thereby their membrane deformity, resulting in increased phagocytic activity and inhibition of tumor growth.33 Consistent with a CpG-mediated promotion of macrophage phagocytosis, the effect was conserved in T-cell deficient Rag2^−/− mice and after DC depletion using CD11c-DTR mice. Overall, these studies suggest that, in certain mouse models, the phagocytic activity of TAM alone is sufficient to eradicate tumor cells, or at least inhibit tumor growth. In humans, CD47 expression by cancer cells is correlated with poor prognosis,30 suggesting that constitutive phagocytosis might impact immunosurveillance. However, direct quantification of phagocytic activity in patient samples is not feasible using the techniques used in mice (principally quantification of fluorescently labeled cancer cell uptake by macrophages). Therefore, no definitive evidence exists regarding TAM capacity to phagocytose cancer cells without anti-CD47 or other phagocytosis inducers in patients.

The obligatory involvement of T cells in the context of CD47-SIRPα inhibition has been shown in other syngeneic models. In the MPA/DMBA-induced breast cancer model, the positive impact of anti-CD47 on survival was abolished when CD4 and CD8 T cells were depleted.26 In B16F10 melanoma, inhibition of CD47 was efficient only when associated with PD-L1 targeting, suggesting that both an increase of phagocytosis by TAM and reinvigoration of T cells are required for optimal tumor control.34 However, this interpretation has to be considered with caution, since PD-L1 binding to programmed cell death protein 1 on macrophages also inhibits phagocytosis,35 potentially ruling out the need for T cells in this anti-CD47/anti-PD-L1 combination. Nevertheless, another study using the same melanoma model demonstrated synergistic effects between anti-CD47 and CTLA-4.36 As CTLA-4 is not a regulator of phagocytosis, this confirms that, in melanoma, efficient antitumor immunity requires not only TAM phagocytic activity, but also T cells. Finally, in a fibrosarcoma model, efficient control of tumor growth by combining irradiation and CD47 blockade was CD8 T cell-dependent.37

Overall, while the increase of TAM phagocytic activity in T-cell deficient NSG mice was sufficient to control tumor growth, a T-cell dependency was observed in some immunocompetent models, highlighting the importance of understanding the interactions between these cell populations. Indeed, TAM can favor T-cell responses by not only secreting proinflammatory cytokines, but also by presenting tumor antigens on MHC I and II to CD8 and CD4 T cells, respectively.38 It is established that macrophages efficiently present engulfed exogenous antigenic material on MHC II,39 while antigen presentation on MHC I was initially described for endogenous cytosolic antigens in every nucleated cell. However, APC have the ability to load engulfed antigens on MHC I through a mechanism called cross-presentation. This latter pathway is critical for CD8 T-cell mediated antitumor immunity, but it is not clearly established whether it is only promoted by DC or whether TAM also play a role.

The importance of cross presentation for CD8 T-mediated immunity in cancer was shown by Huang et al,40 but without identifying the specific role of DC or TAM. In recent years, the balance of evidence points to DC as being the most efficient cross-presenting cells in vivo (as reviewed in Cruz et al 41), although macrophages also possess cross-presentation potential. A key role of cross presentation is to prime naïve T cells, which takes place principally in the draining lymph node (LN).42 While DC migration and CCR7-mediated LN homing is well documented, the limited migration capacity of TAM and their low CCR7 expression prevent their trafficking to draining LN.10 Thus, the activity of TAM as APC takes place exclusively in the TME, where the evidence of naïve T-cell priming is controversial. In LN-deficient mice implanted with different tumors, T-cell activation in the TME was observed, but appeared to be TAM independent,43 whereas another study showed the capacity of CD11c⁺ F4/80⁺ cells to prime naïve T cells in the TME.44 Tertiary lymphoid structures play a key role for naïve T-cell priming directly in the TME,45 but the participation of TAM in this process has not been described. In this context, the role of TAM as APC in the TME seems to be primarily for T-cell reactivation, and any capacity to prime naïve T cells remains to be documented. For immunotherapy, the importance of T-cell reactivation in the TME was suggested in the context of adoptive T-cell transfer, where tumor specific T-cell retention46 and proliferation47 in situ were dependent on cross-presentation, without, however, an exploration of the exact involvement of DC or TAM. Recently, the importance of reactivating T cells in the TME has been highlighted by the demonstration that tissue-resident memory T cells are important for robust antitumor immunity.48 Regarding TAM, several studies have shown that cross-presentation function can be revealed using different approaches, as will be discussed. Indeed, in view of the abundance of TAM in many solid tumors and the clinical opportunities to manipulate their function, it is timely to reconsider evidence for and against their involvement in this process.

Several studies have shown the importance of DC and the dispensable role of TAM in CD8 T-cell priming. In the immunogenic GL261-OVA GBM model, Malo et al showed that a tumor-specific T-cell response was dependent on DC but not on TAM.49 In mice with conditional MHC I knock-out (KO) in DC (under the CD11c promoter) or in macrophages (under the lysozyme M promoter), T-cell infiltration in tumors was abolished in DC MHC I KO mice after ovalbumin vaccination, but not in macrophage MHC I KO mice. An important study by Xu et al showed that anti-CD47 treatment of MC38 colon cancer promoted preferential cross-presentation by DC rather than by TAM, although the latter showed enhanced phagocytosis.50 The mechanism involved activation of the nicotinamide adenine dinucleotide phosphate (NADPH) oxidase NOX2 in the phagosome, thereby limiting phagosome acidification and facilitating cytosolic translocation of proteins and DNA. Cytosolic DNA activated the cyclic GMP-AMP synthase (cGAS)/stimulator of interferon genes (STING) pathway and promoted cross-presentation of tumor antigens by CD11c⁺ DC. In a PDAC model, TAM activation promoted a proinflammatory response, but T-cell activation was cDC1 dependent.51 In one of the rare studies in patients addressing the role of TAM versus DC, moDC and TAM isolated from peritoneal ascites of patients with cancer were used as APC to present the Melan-A/MART1 melanoma antigen to CD8 T cells; only moDC could efficiently activate T cells by secretion of the proinflammatory cytokine IL-12, which was not secreted by macrophages.52

Unlike the above-mentioned studies excluding TAM cross-presentation, others have shown that, by manipulating TAM phagocytic activity, tumor-derived antigens were efficiently presented to CD8 T cells. Using a human colon cancer cell line transfected with ovalbumin, Tseng et al showed in vitro that anti-CD47 antibodies promoted phagocytosis by murine macrophages and presentation of the ovalbumin-derived peptide to MHC I restricted OT-1 T cells53; similar results were achieved with murine GBM cells in vitro.54 In other in vitro studies using human cells, Barrio et al showed that interferon (IFN) γ activated human macrophages presented the Melan-A/MART-1 antigen to T cells, resulting in activation.55 In a fibrosarcoma model, Muraoka et al used a cholesteryl pullulan nanogel to target tumor antigen to TAM, which then induced proliferation of specific CD8 T cells ex-vivo.56 However, TAM were defined as CD11b⁺ F4/80⁺, which would also be compatible with cDC2. Recently, Klichinsky et al developed chimeric antigen receptor macrophages targeting the HER2 tumor antigen that efficiently phagocytosed cancer cells, cross-presented tumor antigens in vitro, and prolonged NSG mouse survival in lung metastasis and peritoneal carcinosis models,57 opening new perspectives for TAM manipulation in cancer therapeutics. These different observations, mostly in vitro, suggest that macrophages have the capacity to cross-present tumor antigens to CD8 T cells, when their phagocytic activity is appropriately enhanced.

Overall, macrophages have a constitutively lower capacity than DC to cross-present tumor antigens. However, since the cross-presenting machinery exists in macrophages, it is possible that in some conditions or after therapeutic targeting, they could play a role in enhancing T-cell responses.

The details of the intracellular cross-presentation pathway have been mostly elucidated in vitro with DC. We will discuss the general features and highlight those that differ in macrophages (figure 3). A better understanding of the limited capacity of macrophages for antigen processing at the molecular level could provide new targets to unlock the cross-presentation capacity of TAM.

Cross-presentation of tumor antigens relies on the capacity of APC to engulf cancer cells or cancer cell debris, mainly by phagocytosis, then to process their protein content in order to generate peptides of 8–10 residues in length that can be loaded on MHC I. The proteolytic activity of the phagosome is initiated rapidly after cell engulfment when its fuses with early endosomes, inducing acidification to pH 6.5. Then, the phagosome fuses with late endosomes and lysosomes to mature into a phagolysosome, a step that induces further acidification (to pH 5), mainly due to the recruitment of the V-ATPase proton pump.20 This low pH promotes activity of proteolytic enzymes that start to cleave large proteins into smaller peptides. Besides low pH, proteases are also activated through oxidation by the NADPH oxidase NOX2 located on the phagolysosome membrane.58 Importantly, proteolytic activity was shown to be higher in macrophages than in DC.59 This would be consistent with longer antigen persistence in DC, a feature that was detected particularly in cross-presenting DC subsets.60 Thus, higher proteolysis that reduces antigen persistence is the first antigen processing limitation that is encountered for efficient cross-presentation by macrophages.

Once proteolysis is initiated in the phagosome, peptides are processed to a length that is compatible with MHC I loading. Here, two different mechanisms have been described: the cytoplasmic and the vacuolar pathways, the former being dominant for cross-presentation during cancer immunosurveillance in vivo.61

The cytoplasmic pathway is characterized by the translocation of ingested and partially proteolyzed peptides from the phagolysosome to the cytoplasm. The endoplasmic reticulum (ER)-associated degradation (ERAD) machinery that normally facilitates retro-translocation of misfolded proteins from the ER lumen into the cytosol plays an important part in this process. Indeed, the ERAD proteins SEC61 and p97 are required for cross-presentation in DC,62 suggesting that ER components are recruited in the phagosome. In addition, the heat-shock protein HSP90 plays a central role by refolding proteins in the cytoplasm after their translocation through the phagosome membrane.63 Interestingly, cytosolic translocation appeared be less efficient in macrophages than in DC,64 providing a second mechanism limiting their cross-presentation capacity. In the cytoplasm, proteasome cleavage finalizing antigen processing was shown to be essential for cross-presentation by macrophages in in vitro studies.65 Proteasome-generated peptides then follow the classical MHC I pathway, generally through translocation to the ER lumen but also via retro-translocation in the phagosome, as the latter compartment was shown to possess the machinery required for MHC I-antigen complex generation.66 This step is dependent on the transporter associated with antigen processing (TAP) 1 and TAP2 proteins, as their loss of function prevents peptide loading on MHC I.65 Finally, once in these compartments, proteasome generated peptides are trimmed by amino-peptidases to be loaded on MHC I, ER-associated aminopeptidase 1 (ERAP1) in the ER67 and endosomal insulin-responsive aminopeptidase (IRAP) in the phagosome.68 Once loaded with antigens, MHC I molecules are translocated to the cell surface to be recognized by CD8 T cells.

Unlike the cytoplasmic pathway, the vacuolar pathway is resistant to proteasome and TAP inhibition. On the other hand, antigen presentation is prevented by cathepsin S inhibition,69 highlighting the key role of phagosomal proteolysis in this pathway. With the presence of MHC I in this compartment, this suggests that antigen processing and MHC I loading can both occur in the phagolysosome. Once antigenic peptide is bound, the MHC I-peptide complex can translocate directly to the cell surface.

In addition to protein antigens, DNA from phagocytosed cells can also escape into the cytosol where it activates the cGAS-STING pathway.70 cGAS is a cytosolic DNA sensor that catalyzes the formation of cyclic dinucleotide guanosine monophosphate-adenine monophosphate (cGAMP), which is recognized by the STING protein. Once activated, STING promotes the secretion of proinflammatory cytokines including type I IFNs and tumor necrosis factor (TNF)-α by DC and macrophages.71 Type I IFNs were shown to increase the pH of the phagolysosome of DC and to extend the persistence of ingested material, thus increasing the capacity of DC to cross-present.72 In T-cell mediated antitumor responses, cGAS-STING activation in DC is essential, as T-cell responses are abolished in STING-deficient mice, resulting in loss of tumor control.73 In this context, the activity of the cytosolic exonuclease three-prime repair exonuclease 1 (TREX1) induced by irradiation dampens activation of cGAS-STING by degrading cytosolic DNA, and potentially placates the immune response.74 Besides irradiation, TLR4 agonists increase TREX1 expression to a greater extent in macrophages than in DC.75 Interestingly, this study also showed that TREX1 limits upregulation of the costimulatory molecule CD86 in lipopolysaccharide-stimulated macrophages. Thus, increased degradation of cytosolic DNA in macrophages could be a third mechanism limiting their cross-presentation ability.

Overall, it appears that cross-presentation requires a low level of protein degradation in order to preserve antigens to be presented to CD8 T cells. With a higher proteolytic activity in the phagolysosome, antigen persistence in macrophages is reduced, thereby limiting cross-presentation.59 Furthermore, the translocation of antigen into the cytosol, a critical step for the cytoplasmic pathway, is less efficient in macrophages than in DC.64 Finally, the higher propensity of macrophages to increase TREX1 expression reduces the availability of cytosolic DNA for cGAS-STING activation, thereby limiting macrophage activation and capacity for efficient cross-presentation (figure 3).

The different studies presented earlier underline that constitutive cross-presentation by macrophages is less efficient than by DC, even though it can be enhanced using different approaches. The functional limitations of TAM might therefore be potentially counterbalanced by quantitative advantages. Indeed, in comparison with DC, TAM are generally more abundant9 and their phagocytic activity is generally higher.50 It would be of considerable interest to investigate whether the respective antigen-presenting roles of DC and TAM are related to the relative abundance of these APC at the tumor site. Indeed, this ranges from the relatively DC-rich TME of melanoma or lung cancer76 77 to the DC-poor, but microglia-rich and TAM-rich TME of brain tumors,78 79 and we can therefore speculate that TAM may play a more important role in glioblastoma immune surveillance. Whether priming of naïve T cells by TAM in the TME is an important phenomenon in vivo remains to be demonstrated; it is more probable that T-cell reactivation occurs at this site, the former being a less stringent process. Thus, the capacity of TAM to cross-present tumor antigens could be a key mechanism for adoptive T cell transfer or when tissue resident memory T cells are harnessed for anticancer immunity.