The macrophage checkpoint CD47 : SIRPα for recognition of ‘self’ cells: from clinical trials of blocking antibodies to mechanobiological fundamentals

Abstract

Immunotherapies against some solid tumour types have recently shown unprecedented, durable cures in the clinic, and the most successful thus far involves blocking inhibitory receptor ‘checkpoints’ on T cells. A similar approach with macrophages is emerging by blocking the ubiquitously expressed ‘marker of self’ CD47 from binding the inhibitory receptor SIRPα on macrophages. Here, we first summarize available information on the safety and efficacy of CD47 blockade, which raises some safety concerns with the clearance of ‘self’ cells but also suggests some success against haematological (liquid) and solid cancers. Checkpoint blockade generally benefits from parallel activation of the immune cell, which can occur for macrophages in multiple ways, such as by combination with a second, tumour-opsonizing antibody and perhaps also via rigidity sensing. Cytoskeletal forces in phagocytosis and inhibitory ‘self’-signalling are thus reviewed together with macrophage mechanosensing, which extends to regulating levels of SIRPα and the nuclear protein lamin A, which affects phenotype and cell trafficking. Considerations of such physical factors in cancer and the immune system can inform the design of new immunotherapies and help to refine existing therapies to improve safety and efficacy.

This article is part of a discussion meeting issue ‘Forces in cancer: interdisciplinary approaches in tumour mechanobiology’.

1. Introduction

Specific molecular interactions between two cells or a cell and extracellular matrix are often viewed as pro-adhesive and ultimately favouring attachment. However, specific interactions can also be inhibitory, as is the case for several targets for cancer therapy in the clinic. Both pro-adhesive and inhibitory interactions can also involve important mechanobiological factors. Immune cells provide particularly illustrative examples as they frequently contact cells that either belong to ‘self’ (the same organism) or are ‘foreign’ (e.g. microbes that breach epithelia). Specific molecular interactions at immune cell surfaces lead to recognition of ‘self’ or else result in forceful attack and elimination of ‘foreign’. An important example with T cells is the protein PD-1, which interacts with PD-L1 on multiple ‘self’ cells in parallel with T cell receptor interactions; if PD-L1 activates the T cell to attack, PD-1 can effectively passivate it. In cancer, blocking this PD-1 : PD-L1 checkpoint by systemic injections of antibodies to either of these two proteins leads to T cell elimination of tumours in a minor fraction (approx. 10–30%) of otherwise untreatable patients, and the patients that respond best are those with the most mutated (i.e. ‘foreign’) tumours [1–4]. In simplest molecular terms, the T cell receptor activates kinases that signal activation while PD-1 : PD-L1 activates a phosphatase (e.g. SHP isoform) that dominates in its inhibition—although there remains much to learn. Mechanobiology is involved at least via the kinases and/or phosphatases that regulate local membrane mechanics on the small scale and/or cytoskeletal function at a larger scale [5,6]. Importantly, this paradigm of activation-dominated-by-inhibition applies not only to other lymphocytes (e.g. natural killer (NK) cells [7]) but also to macrophages, which are the focus of this review.

2. CD47 : SIRPα as a macrophage checkpoint in cancer

‘Marker of self’ membrane protein CD47 is normally expressed on all cells and binds with weak, sub-micromolar affinity to ‘signal regulatory protein’ SIRPα on macrophages, including precursor monocytes. CD47 : SIRPα binding leads to local accumulation of SIRPα at phagocytic synapses and ultimately to inhibition of engulfment of ‘self’ cells (figure 1) [8,9]. This inhibitory interaction occurs in parallel with various activating interactions, only some of which are well characterized. The clearest example of activation is through immunoglobulin G (IgG) antibodies which bind to a target cell and which also engage activating Fc receptors (FcRs) on macrophages. Some of the key FcRs signal via kinases in very similar ways to integrins activated by extracellular matrix, with a downstream accumulation of focal adhesion proteins such as phospho-paxillin and talin as well as sensitivity to whether the adhesive substrate—i.e. target for phagocytosis—is soft or stiff [10,11]. However, adhesion and phagocytosis of stiff targets more than soft is just one aspect of the mechanosensitivity of macrophages.

Figure 1.

‘Self’-signalling and opsonization in phagocytosis. Binding of CD47 expressed by a target cancer cell to SIRPα on the macrophage surface signals ‘self’ to the immune cell and inhibits phagocytic clearance (a). Inhibition of ‘self’-signalling with an anti-CD47 antibody (Ab) in combination with a tumour-opsonizing Ab that engages macrophage Fc receptors leads to phagocytosis of the target (b).

CD47 : SIRPα blockade strategies have revitalized decades of interest in macrophages as effector cells for cancer therapy. CD47 is expressed on cancer cells [12,13] and was originally described as the OA3 antigen, which is highly upregulated on ovarian cancer cells [14]; CD47 should in principle shield cancer cells from immune surveillance and removal by phagocytic cells. However, solid tumours possess mechanical properties that can influence cancer phenotypes [15] and that might also influence immune cells, including macrophages. Studies in mice, for example, suggest high collagen microenvironments (associated with stiffness) cause macrophages to upregulate SIRPα expression and also cause macrophages to switch off a pro-phagocytic phenotype [16]. These factors motivate a renewed focus on the mechanobiology of phagocytic cells which certainly include macrophages but also neutrophils [17] and dendritic cells [18], which both express SIRPα.

Patient safety is always a concern in therapy, and the main goal of any Phase 1 clinical trial in the US is to establish a safety window for dosing. Because CD47 is expressed on all cells, anti-CD47 antibodies injected intravenously are readily predicted to bind blood cells. Indeed, the pioneering studies that first described ‘marker of self’ function showed that CD47-deficient mouse red blood cells (RBCs) are engulfed within hours of injection into normal mice by macrophages in the spleen [9], with CD47-deficient platelets exhibiting similar clearance [19]. CD47 knockout in at least one strain of mouse (non-obese diabetic, NOD) leads not only to anaemia (RBC loss) but also to premature death of mice with autoimmunity against RBCs [20]. None of these mouse studies clearly determined what factors on a CD47-deficient RBC or platelet activate the engulfment by normal splenic macrophages, but anti-CD47 IgG will engage FcRs and tend to activate phagocytosis. As with integrin-based adhesion, however, a high density of adhesive ligands (IgG in this case) favours focal adhesion complex formation and activated adhesion, and CD47 is more of a low-density signalling molecule on RBCs [8]. Regardless, the various findings in mice underscore the importance of safety studies, and the report of autoimmunity also has intriguing implications for the development of anti-cancer immunity.

3. Clinical blockade of the anti-adhesive CD47-SIRPα interaction—a current snapshot

According to the National Institutes of Health (NIH) database of clinical trials (clinicaltrials.gov), there are presently 15 ongoing anti-CD47 interventional clinical trials, with all but two in Phase 1 [21–35]. Although tumour cells are unlikely to display sufficiently strong ‘eat me’ factors to activate macrophages [16], and modest upregulation of CD47 on tumours (e.g. ovarian cancer [14]) is unlikely to present sufficient anti-CD47 to activate phagocytosis, clinical trials all must start with injections of just anti-CD47 to test safety without interference. All current clinical trials targeting CD47 consist of monoclonal antibodies or engineered fusion proteins and are being driven by companies—no doubt because of the costs involved. Designations for these agents are Hu5F9-G4 (Forty-Seven Inc., Menlo Park, CA), CC-90002 (Celgene Corporation, Summit, NJ), TTI-621 and TTI-622 (Trillium Therapeutics Inc., Mississauga, Ontario, Canada), SRF231 (Surface Oncology, Cambridge, MA), ALX148 (ALX Oncology Inc., Burlingame, CA) and IBI188 (Innovent Biologics, Suzhou, PR China). Clinical trial details are summarized in table 1 (complemented by a thorough review of preclinical information in [36]). Reporting on these trials remains sparse, but details are emerging in meeting abstracts and on company websites. Such information is public but should be treated with caution as it is not clear that companies will publish in peer-reviewed forums, though publications are now beginning to appear. Nonetheless, the information provides some initial insight into the potential safety and efficacy of CD47 blockade therapies in humans.

Table 1.

Summary of ongoing anti-CD47 clinical trials.

drug name	conditions	company	trial identifiera	study phase	status	intervention type
Hu5F9-G4	acute myeloid leukaemia	Forty-Seven, Inc.	NCT02678338	Phase 1	completed	monotherapy
	myelodysplastic syndrome
	acute myeloid leukaemia,		NCT03248479	Phase 1	recruiting	combination with azacitidine
	myelodysplastic syndrome
	colorectal neoplasms, solid tumours		NCT02953782	Phase 1b/2	recruiting	combination with cetuximab
	solid tumours		NCT02216409	Phase 1	recruiting	monotherapy
	non-Hodgkin lymphoma, indolent lymphoma, diffuse large B-cell lymphoma		NCT02953509	Phase 1b/2	recruiting	combination with rituximab
	solid tumours, ovarian cancer	Forty-Seven, Inc., with Merck Celgene	NCT03558139	Phase 1	recruiting	combination with avelumab
CC-90002	acute myeloid leukaemia, myelodysplastic syndrome		NCT02641002	Phase 1	terminated (Oct. 2018)	monotherapy
	haematologic neoplasms		NCT02367196	Phase 1	recruiting	combination with rituximab
TTI-621	haematologic malignancies, solid tumours	Trillium Therapeutics, Inc.	NCT02663518	Phase 1	recruiting	monotherapy, combination with rituximab, combination with nivolumab
	solid tumours, mycosis fungoides, melanoma, Merkel-cell carcinoma, squamous cell carcinoma, HPV-related malignant neoplasm, soft tissue sarcoma		NCT0289036	Phase 1	recruiting	monotherapy vs various combination therapies (PD-1 : PD-L1 inhibitor, pegylated interferon-α2a, T-Vec, radiation)
TTI-622	lymphoma, myeloma		NCT03530683	Phase 1	recruiting	monotherapy vs various combination therapies (PD-1 : PD-L1 inhibitor, rituximab, proteasome-inhibitor regimen)
SRF231	advanced solid cancers, haematologic cancers	Surface Oncology	NCT03512340	Phase 1	recruiting	monotherapy
ALX148	metastatic cancer, solid tumour, advanced cancer non-Hodgkin lymphoma	ALX Oncology, Inc.	NCT03013218	Phase 1	recruiting	combination with pembrolizumab, trastuzumab, or rituximab
IBI188	advanced malignancies	Innovent Biologics	NCT03763149	Phase 1	recruiting	monotherapy
	advanced malignancies		NCT03717103	Phase 1	recruiting	monotherapy vs combination with rituximab

^aNCT number from clinicaltrials.gov database. Table data current as of March 2019.

Hu5F9-G4 is a humanized anti-CD47 monoclonal IgG4 antibody and is the most advanced therapeutic in this class to progress through the clinic [37,38]. The Fc end of an IgG4 engages FcRs only weakly; one might therefore expect minimal interactions with FcRs on splenic macrophages and perhaps other macrophages. In one of the most advanced trials of Hu5F9-G4, the Phase 1b/2 study for relapsed/refractory non-Hodgkin lymphoma (r/r NHL), patients combined anti-CD47 with rituximab (anti-CD20) to provide an ‘eat me’ signal to activate macrophages; the reported objective response rate (ORR) was 50% and the complete response rate (CRR) was 32% [39]. These numbers were revised in late 2018 to 40% ORR and 33% CRR in 15 diffuse large B-cell lymphoma patients and 71% ORR and 43% CRR in seven follicular lymphoma patients [40]. The investigators also noted that they were able to limit anaemia by the administration of a priming dose to clear older RBCs prior to maintenance doses, though to what extent is unclear.

In Phase 1a of the trial for relapsed/refractory acute myeloid leukaemia (r/r AML), 15 patients tolerated Hu5F9-G4 well with no maximum tolerated dose (MTD) reported. However, just two patients exhibited biological activity in response, which the authors defined as ‘significant reduction in marrow cellularity observed similarly in pre-clinical models'. The majority of the studied patients (79%) showed grade 3 anaemia prior to dosing in the study and also received transfusions of RBCs [41]. In prior reported safety results of 13 patients [42], the maximum observed haemoglobin (Hb) drop was 5.2 g dl⁻¹, representing a loss of about 40% of RBCs, though the median decrease was 1.2 g dl⁻¹. Antibodies are of course bivalent, and some anti-RBC antibodies have long been known to drive cross-linking of RBCs (haemagglutination); such adhesive clusters of cells can be expected to have impaired circulation, especially through the narrow slits of the spleen—which would tend to favour macrophage interactions and anaemia. Forty-Seven, Inc. has also initiated additional trials against other malignancies to test Hu5F9-G4 in combination with the chemotherapeutic azacitidine, the anti-epidermal growth factor receptor (EGFR) monoclonal antibody cetuximab, and the PD-L1 targeting antibody avelumab [22,25,28]. Later reports in the first-in-human trial with advanced solid tumours indicated a priming dosing led to 57 and 36% of patients showing transient anaemia and haemagglutination, respectively, mostly grade 2 [43,44]. Whether the loss of RBCs can ever contribute to an autoimmune response to RBCs or other cell types has not been addressed.

TTI-621 is a fully human recombinant SIRPα-Fc fusion protein developed by Trillium Therapeutics, Inc. with preclinical evidence for enhanced phagocytosis in AML and B lymphoma xenograft models [45]. It consists of the N-terminal V-type immunoglobulin-like domain of human SIRPα as a fusion with the Fc portion of human IgG1. IgG1 is expected to bind FcRs more strongly than IgG4 and may increase both antibody-dependent cell-mediated cytotoxicity and complement-dependent cytotoxicity effects. It is being administered intravenously for haematologic malignancies and intratumorally for solid tumours and mycosis fungoides [46]. Early reports in late 2016 showed two patients with dose-limiting toxicities in grade 3 elevated liver enzymes and grade 4 thrombocytopenia (loss of platelets). This result raises questions of the drug safety profile, although later the company stated that this thrombocytopenia was transient and reduced after multiple infusions. The company also reported that 9 out of 10 patients in the intratumorally administered trial saw a reduction of lesions [46], and later, 15 out of 17 r/r mycosis fungoides and Sézary syndrome patients were described as having measurable improvement in lesion severity [47]. They have expanded their study to new cohorts receiving TTI-621 in combination with approaches such as a PD-1 : PD-L1 inhibitor or radiation therapy. A second candidate, linked to an IgG4 domain instead of IgG1, TTI-622, was initiated in a Phase 1a/1b clinical trial in mid-2018 [33].

SRF231 is a fully human anti-CD47 antibody being evaluated against advanced solid tumours and lymphoma/chronic lymphocytic leukaemia [48]. CC-90002 is a humanized anti-CD47 monoclonal antibody of the IgG4 subclass in early clinical development, with preclinical work presented at the 2017 ASCO Annual Meeting [49]. Both CC-90002 and SRF231 seem to avoid haemagglutination, but how this is achieved is unclear [50]. Notably, despite the expansion of anti-CD47 efforts by pharmaceutical companies, Celgene terminated its monotherapy trial of CC-90002 in October 2018 while continuing to recruit in its combination trial with rituximab [34,35]. This further highlights the importance of combining an activating signal for phagocytosis with blockade of the macrophage checkpoint.

ALX148 is a fusion protein with two CD47 binding domains derived from the SIRPα N-terminal D1 domain and an inactive Fc region designed to mitigate haemagglutination and anaemia [51]. The presence of an inert Fc region also indicates its design as a combination therapeutic with tumour-opsonizing antibodies, unlike agents discussed thus far which can directly engage FcRs on the macrophage surface to mediate effector function [52]. Reports at the 2018 ASCO Annual Meeting showed data from the first 30 patients enrolled, 25 of whom received only ALX148, with the remaining five patients receiving combination regimens with pembrolizumab (three patients), trastuzumab (one patient) or rituximab (one patient). ALX148 was generally tolerated, with four combination patients achieving stable disease states though two patients exhibited grade 3 thrombocytopenia [53].

IBI188 is a fully human anti-CD47 monoclonal IgG4 antibody and is the most recent drug to move from the preclinical phase to Phase 1 trials, where it is currently being evaluated against advanced malignancies as a monotherapy and in combination with rituximab [26,27]. The trials dosed their first patients in late February 2019. Blockade of CD47 : SIRPα in patients is thus being pursued by a growing list of companies even though single agent efficacy is not especially compelling. Safety issues that are most easily measured and that are widely reported include significant anaemia, thrombocytopenia and haemagglutination, which must be balanced in turn with establishing efficacy—most often by combining with an IgG that binds to abundant antigens on a tumour cell and thereby strongly activates adhesion-initiated phagocytosis. Consequently, the drug space aimed at CD47 : SIRPα has expanded dramatically, with significant investment in new antibodies, small molecules and peptides.

4. Macrophage checkpoint blockade plus opsonizing antibodies—initial success with liquid tumours, but are solid tumours next?

Targeting the SIRPα side of the CD47 : SIRPα interaction [54,55] rather than CD47 has not yet been attempted in the clinic but could prove analogous to blocking PD-1 (on the T cell), which shows clinical efficacy, and certainly, preclinical anti-SIRPα development efforts are encouraging [50]. What is appearing clinically is the utility of combination therapies with tumour-specific opsonizing antibodies that seek to strongly shift from anti-phagocytic signals to pro-phagocytic activity. This is especially interesting for patients refractory to tumour-specific monotherapies such as rituximab. The combination approach is now seen in virtually all of the trials discussed above: Hu5F9-G4 with rituximab, cetuximab and avelumab, CC-90002 with rituximab, and ALX148 with trastuzumab (anti-HER2) and rituximab, to name a few (complete list in table 1). It is increasingly clear that elimination of non-‘self’ requires an ‘eat me’ cue, often by IgG opsonization and subsequent engagement of macrophage FcRs, in addition to the ‘don't eat me’ signal of CD47 : SIRPα. In further recognition of the merits of this phenomenon, bispecific antibodies containing CD47-specific and tumour-specific (e.g. mesothelin, CD20 or CD33) domains are in preclinical development (reviewed in-depth in [56]), while anti-SIRPα antibodies may also provide opsonization on the tumour cell surface.

Effective clearance by innate immune effector cells relies on much more than one or two cell surface interactions. Some immunotherapies have shown unprecedented success in generating durable cures to blood cancers, but have faltered in treatment of solid, stiff tumours. Macrophages engulf foreign material in all types of native tissues and thus seem tailor-made to enter and eat solid tumours; it will therefore be very interesting to see whether the success with anti-CD47 and rituximab against liquid tumours translates to solid tumours. CD47 : SIRPα's regulatory role in phagocytosis and physical differences in the tumour microenvironment between haematological and solid malignancies motivate a deeper understanding of the mechanobiological underpinnings of phagocytosis in healthy and pathological processes. Cells in solid tumours generally adhere strongly to each other and/or to the extracellular matrix—which is not true of liquid tumours and which makes it physically more challenging for a macrophage to completely engulf a cancer cell integrated into a solid tumour. Knowledge of mechanosensing by phagocytes is likely to help maximize the utility of macrophage checkpoint therapies and macrophage functions more broadly. The following sections describe some of the physical factors to consider in macrophage biology (figure 2a) and how that information might be used to inform therapeutic design.

Figure 2.

Summary of macrophage mechanobiology and forces in phagocytosis. (a) Many factors play roles in macrophage interactions and the immune cells' ability to clear ‘foreign’ matter. In solid tumours, macrophages must contend with cell–matrix and cell–cell adhesions that can influence their ability to engage and eat cancer cells. Macrophages sense matrix properties through podosome adhesions and exhibit lamin A levels that scale with matrix stiffness. Such matrix to nucleus mechanosensing might influence macrophage polarization characterized by different cytokine secretions and surface marker expression. (b) Engagement of CD47 by SIRPα recruits the phosphatase SHP-1 to the phagocytic synapse in the macrophage where it inhibits myosin IIa assembly (top). In blockade therapy, activating signals from the clustering of Fc receptors drive cytoskeletal reorganization and myosin IIa assembly which promote phagocytosis. (Online version in colour.)

5. Forces in phagocytosis and ‘self’-signalling

Phagocytosis is an ancient process that proceeds by the extensive reorganization of the actin cytoskeleton and (often) via contractions mediated by myosin motors. Myosins bind, cross-link and pull on actin filaments, and myosins are well known to be load-sensitive (i.e. contractile velocity decays with force), which provides a basis for mechanosensitive interactions. As with integrin-mediated adhesion, target particle or cell binding to surface receptors frequently leads to receptor clustering and activates signalling pathways that drive maturation of adhesions via actomyosin activity: for macrophages, this involves the formation of pseudopod protrusions around the target, with actin polymerization and branching generating forces of protrusion [57]. Although the pro-phagocytic signals mediating cancer cell engulfment in the presence of CD47 : SIRPα blocking agents have not been fully elucidated, it is clear that phagocytosis is improved with an opsonizing antibody, suggesting an FcR-mediated mechanism [55,58]. Other pro-phagocytic signals may also be involved, depending on the target. Phagocytosis of haematopoietic cancer cells by macrophages is reported to depend on Mac-1 (complement receptor-3, CR3 or α_Mβ₂ integrin), which binds to an unknown receptor on its target, along with a homotypic SLAMF7 interaction and signalling through ITAM-containing co-receptors [59]. Calreticulin is also proposed to be a pro-phagocytic signal recognized by scavenger receptor LRP-1 on macrophages [60]. These pro-phagocytic cues are an emerging theme in CD47 : SIRPα blockade trials and may be especially critical for the success of these agents in shrinking solid tumours.

Several myosin motor proteins localize at phagocytic synapses (figure 2b) [61–64]. The role of myosin in FcR- and complement receptor-mediated phagocytosis has been investigated by pharmacological inhibitors as well as overexpression and knockdown of non-muscle myosin IIa (NMIIA). Target engulfment is deficient in mouse bone marrow-derived macrophages and macrophage cell lines treated with inhibitors of myosin light chain kinase (i.e. ML7) or myosin (i.e. 2,3-butanedione monoxime and blebbistatin) [8,62,65,66]. Similarly, the human monocytic cell line THP-1 internalizes ‘non-self’ sheep RBCs less efficiently when transfected with small interfering (siRNA) against NMIIA and more efficiently when NMIIA is overexpressed [8]. Scanning electron microscopy revealed that ML7-treated cells still form phagocytic cups bound to IgG-opsonized sheep RBCs, but the membrane protrusions are not as closely associated with the target as in the untreated control [67]. Thus, it appears that actin assembly and membrane protrusion occur independently of myosin IIA in FcR-mediated phagocytosis, but mature or stable adhesions might require contractility through NMIIA. By contrast, in complement receptor-mediated phagocytosis, NMIIA is implicated even earlier in the process and mediates actin organization with a requirement for Rho-ROCK signalling [65]. Other myosins, including class I myosins and myosin X, have been observed to participate in various stages of phagocytosis and have potential roles in phagosome formation and closure [62–64].

Cell spreading on matrix-coated substrates uses many of the same cellular components as phagocytosis, and cells typically exhibit greater spreading on stiff substrates than on soft ones. Similarly, the efficiency of phagocytosis is determined in part by the stiffness of the target, with softer targets regarded as more difficult to phagocytose than stiff targets, a factor to consider in cancer cell engulfment. This has been demonstrated with engineered hydrogel microbeads [10,68] and chemically stiffened RBCs [11]. In the light of this effect, control of particle stiffness has risen as a strategy to increase the circulation time of drug-loaded nanocarriers through delaying phagocytic clearance by splenic and liver macrophages [68,69]. Prolonging clearance has also been accomplished by decorating particles or viruses with recombinant CD47 or small ‘self’-peptide mimetics [70,71].

When CD47 on a target cell (or engineered particle or virus) engages SIRPα on macrophages, the cytoplasmic immunoreceptor tyrosine inhibitory motif (ITIM) of SIRPα is phosphorylated and recruits the phosphatase SHP-1 (figure 2b) [72]. NMIIA is a direct target of SHP-1, which inhibits NMIIA accumulation at the phagocytic synapse and prevents the macrophage from efficiently engulfing the target [8]. For very stiff opsonized targets, however, such as dialdehyde-cross-linked RBCs, NMIIA becomes hyperactivated at the phagocytic synapse in macrophages and inhibitory signalling is unable to prevent target engulfment [11]. This could have implications for the phagocytosis of cancer cells and for cancer treatment. Cancer cells can be softer than normal cells [73], which together with increased expression of CD47 [12,13] could allow transformed cells to evade phagocytosis. Alternatively, stiffening of cancer cells following chemotherapy could make pre-treated tumours more susceptible to phagocytosis and macrophage checkpoint therapies [74]. Future progress on this subject could identify biophysical signatures of cancer cells that relate to response to anti-CD47 therapy for predictive or prognostic purposes.

The interaction of CD47 and SIRPα on juxtaposed membranes is also governed by physical forces, including fluctuations of the flexible plasma membrane. In cell-derived giant vesicles displaying human CD47, vesicle spreading and CD47 accumulation was observed on SIRPα-coated coverslips [75]. The apparent binding affinity increased with the concentration of CD47 : SIRPα complexes, indicating a cooperative interaction that arises from a decrease in out-of-plane membrane fluctuations. As a result, even low expression levels of CD47 or SIRPα alleles with low binding affinity may be able to achieve sufficient levels of ‘self’-signalling to limit phagocytosis. Decreasing the pH to mimic acidosis that is frequently observed in tumours rigidifies the vesicle membrane and decreases cooperativity. It is speculated that such a loss of cooperativity could permit macrophages to phagocytose cells expressing low levels of CD47, and in doing so select for cancer cells expressing high levels of CD47 [75].

6. Macrophage mechanosensing from matrix to nucleus and SIRPα

In cancer clearance, monocytes and macrophages face a number of challenges presented by the tumour microenvironment; they must be able to extravasate through small or large pores, differentiate to a phagocytic phenotype, resist reprogramming to an anti-phagocytic state, and ultimately, eat their targets. This process requires the ability to probe and respond to the mechanical forces surrounding the cells. Macrophages lack focal adhesions and stress fibres but form podosome adhesions with the extracellular matrix that are capable of sensing the environment [76]. These structures contain a protruding branched actin core and peripheral actomyosin cables attached to integrins through adapter proteins, and multiple podosomes can be connected into higher-order structures. Force generation in podosomes involves actin polymerization in the core and myosin contractility, which allows cells to probe substrate stiffness [76]. Human monocyte-derived macrophages form more podosomes on stiff substrates than on soft substrates, and phosphorylation of myosin light chain and force generation are also correlated with substrate stiffness [77,78]. How the mechanical stimuli sensed at podosomes are transduced to alter gene expression remains unclear but is likely to be critical in understanding macrophage behaviour and how the tumour microenvironment can directly or indirectly affect the cells' ability to clear ‘foreign’ targets.

Mechanosensing macrophages exhibit differential spreading, migration and polarization, depending on the physical properties of the matrix. Alveolar macrophages from rats are rounded when cultured on soft epithelial monolayers but flatten and spread on polyacrylamide gels of intermediate compliance and stiff glass substrates [79]. Similarly, human monocyte-derived macrophages are more spread on stiff polyacrylamide gels coated with fibronectin than on soft ones [80]. A biphasic relationship between cell area and substrate stiffness has also been reported for human macrophages, with a maximal area reached at an intermediate stiffness [81]. Both studies observed stiffness effects on macrophage migration in two dimensions. In three dimensions, human monocyte-derived macrophages exhibit different migration modes, depending on the structure of the matrix, but seem to be less sensitive to matrix stiffness than to matrix organization [82]. In Matrigel or a dense collagen gel, macrophages underwent mesenchymal migration that was slow and dependent on protease activity. In a fibrillar collagen matrix, macrophages underwent amoeboid migration, which was faster and protease-independent. Podosome-like protrusions with collagen-degrading activity were observed during mesenchymal migration through collagen gel but not during amoeboid migration through fibrillar collagen.

Macrophages exhibit considerable plasticity with respect to their polarization or activation states [83]. Very broadly, macrophages can be polarized to a classical, inflammatory state or to alternatively activated states with anti-inflammatory or wound healing properties. Biophysical cues, including matrix stiffness, topography and external forces, can potentially influence macrophage polarization (reviewed in [84,85]), although further research into their effects is still needed. In particular, investigations of the effects of matrix stiffness on macrophage polarization have reached very different conclusions as to whether stiffness promotes an inflammatory or alternatively an activated phenotype [86–91]. This may be due in part to different sources of macrophages (primary cells versus cell lines), to different substrates (synthetic polymer scaffolds, collagen gels, etc.) and adhesive ligands (fibronectin, collagen), and to simultaneous activation with different cytokines (e.g. IFNγ, IL-4/IL-13) or inflammatory signals (e.g. LPS). Alternatively, differences could be due to the inadequacy of current classification methods to describe the full range of macrophage polarization or to the relatively weak effect of substrate stiffness on this phenotype. Given the altered matrix and different physical cues potentially present in solid tumours, the efficacy of macrophage checkpoint blockade may be strongly influenced by the macrophages' ability to sense and respond to this matrix while maintaining a pro-phagocytic phenotype.

Several studies have noted a relationship between macrophage shape and polarization [92,93]. Polarization of macrophages toward an inflammatory phenotype with lipopolysaccharide (LPS) and interferon-γ (IFNγ) produces round cells while alternative activation with interleukin-4 and interleukin-13 produces elongated cells [92]. The converse also appears to be true. When mouse bone marrow-derived macrophages are cultured on micropatterned lines of fibronectin, they become elongated and upregulate expression of anti-inflammatory markers Arg-1 (arginase-1) and CD206 while downregulating inflammatory markers iNOS (inducible nitric oxide synthase) and IFNγ [92]. Pharmacological inhibition of the actin cytoskeleton with cytochalasin D or cellular contractility with blebbistatin, ML-9 or Y27632 attenuates this polarization. Perhaps consistent with the effects of shape on macrophage polarization, the moderate cyclic strain of macrophages on polymeric scaffolds [94] and interstitial flow shear forces [95] seem to promote alternative activation.

With some exceptions, tissue-resident macrophages originate from yolk sac- or fetal liver-derived haematopoietic cells during embryonic development [96]. In response to injury or inflammation, monocytes can also be recruited to tissues from circulation and differentiated into macrophages. Transcriptomic and epigenetic analyses have revealed that macrophage phenotype is tailored by the local microenvironment [97,98]. Whether biophysical cues including matrix stiffness contribute to tissue-specific phenotypes in macrophages remains unclear but plausible.

Macrophages reside in tissues that span a wide range of stiffnesses. For example, microglia are the resident macrophages in brain tissue, which is very soft with an elastic modulus of approximately 0.3 kPa. Osteoclasts are the resident macrophages on bone tissue, which is much stiffer with an elastic modulus of greater than 30 kPa (non-calcified, and far higher for calcified). Using mass spectrometry-based proteomics, it was demonstrated that the nuclear intermediate filament protein lamin A, but not lamin B, is mechanosensitive [99]. Specifically, lamin A protein levels measured in mouse tissue lysates exhibit a power-law scaling with tissue elasticity E while lamin B levels remain nearly constant. In a meta-analysis of transcriptomic data from different mouse tissue-resident macrophages, the ratio of lamin A to lamin B mRNA (LMNA : LMNB) also exhibited power-law scaling with E [100]. Consistent with this finding, immunofluorescence staining of lamin A was greater in phorbol myristate acetate (PMA)-differentiated THP-1 cells when cultured on stiff polyacrylamide gels than when cultured on soft gels [16].

The increased stiffness of collagenous tumours relative to normal tissue has implications for lamin A expression in tumour-associated macrophages (TAMs) and for macrophage-adoptive transfer therapies. For example, the LMNA : LMNB transcript ratio in macrophages isolated from subcutaneous A549 lung cancer xenografts conformed to the power-law scaling relationship with the measured tumour stiffness [16]. This suggests that TAMs are mechanosensitive and that increased stiffness in tumours could in principle contribute to the reprogramming of monocytes recruited from circulation or tissue-resident macrophages to TAMs. In the macrophage checkpoint context, high-collagen microenvironments in vivo (associated with stiffness) and stiff gels in vitro cause macrophages to upregulate SIRPα expression and also cause macrophages to switch off a pro-phagocytic phenotype [16]. For therapy, this likely means that more blockade antibody will be required and perhaps more opsonizing antibody to activate a macrophage.

7. Macrophage infiltration and differentiation

Within dense, fibrotic tissues, including collagenous tumours, extracellular matrix fibres create constrictions that are generic barriers to three-dimensional migration. As the largest organelle in the cell, the nucleus limits migration through narrow constrictions and must be deformed to pass through pores smaller than the nuclear diameter. Thus, by controlling the deformability and mechanical integrity of the nucleus, lamin expression and scaling with tissue elasticity have implications for tumour infiltration and macrophage migration as well as egress from marrow. Lamin levels in haematopoietic cells are indeed relevant to migration through small pores such as those encountered while trafficking from the bone marrow into circulation or from circulation into tissue [99,101].

Three-dimensional migration through Transwell filters with 3–8 µm diameter pores has served as an in vitro model of constricted migration [102,103]. High levels of lamin A protect the nucleus but can also limit migration [102]. Nuclear rupture and cell death have been observed in human monocyte-derived dendritic cells migrating through 2 µm constrictions in microfluidic channels [104]. Whether nuclear rupture occurs frequently during macrophage migration in vivo is unknown, but the leakage of DNA into the cytoplasm or formation of micronuclei could have significant inflammatory effects by activating the cGAS/STING pathway of interferon activation [104,105]. Infiltration is thus physically modulated but has implications for cell fate and inflammation—which are likely to impact the success of checkpoint blockade.

Lamin A has also been described as a differentiation marker in the monocyte lineage. Early studies showed lamin A and the alternative splicing product lamin C are undetectable in rat bone marrow-derived precursor cells but increase significantly following in vitro differentiation to monocytes and macrophages [106]. A similar increase was observed during the in vitro differentiation of human peripheral blood monocytes to macrophages. Differentiation of the HL-60 cell line to monocyte- and macrophage-like cells by PMA increases levels of both lamin A and B [107]. Lamin A and B expression and the A : B ratio vary greatly during haematopoietic maturation in vivo as measured by mass spectrometry-calibrated, intracellular flow cytometry of freshly isolated human marrow and blood cells [108]. While the total lamin levels decrease slightly or remain approximately constant between human marrow progenitor cells (CD34+ CD38+, CD34+ CD38−) and peripheral blood or marrow monocytes and granulocytes, the lamin A : B ratio increases due to downregulation of lamin B and upregulation of lamin A. Flow cytometry measurements also revealed that the total amount of lamin A and B is lower in haematopoietic cells that traffic into circulation (e.g. lymphocytes and peripheral blood granulocytes and monocytes) than cells residing primarily in the bone marrow compartment (e.g. CD34+ progenitor cells, erythroblasts, megakaryocyte lineages and mesenchymal stem cells). Like cancer cells, the constricted migration of haematopoietic cells through Transwell pores approximating marrow sinusoidal capillaries is strongly influenced by lamins [108]. Through such processes and others, lamin A levels can also change during macrophage activation in vivo. Mouse peritoneal macrophages collected 5 days after stimulation with thioglycollate stain strongly for lamin A/C whereas unstimulated macrophages do not [106]. Together, these results indicate that the nuclear lamins, which are key components of the matrix to nuclear mechanosensing pathway, exhibit different expression during lineage maturation and across different tissues. It will be important to determine whether such changes are accompanied by changes in levels of SIRPα.

8. Conclusion

The CD47 : SIRPα axis is an intriguing, rapidly emerging therapeutic target involving innate immune macrophages. It is likely dependent not only on strongly opsonizing antibodies (such as rituximab) but also on other processes and factors, including mechanical ones. Safety issues include clearance of RBCs and platelets, and these mechanically related processes and factors could play a role. Phagocytic effector cells are well known to be sensitive to their microenvironment, which motivates the careful study of the physical forces involved in phagocytosis and the broader mechanobiology of macrophages, including precursor monocytes. Matrix mechanics in solid tumours affect cancer cells and likely immune cells as well, and physical forces also govern the level of SIRPα as well as the engagement of CD47 : SIRPα, which inhibits phagocytosis. Moreover, the inflammatory state of a macrophage depends on biophysical cues including local matrix stiffness and external forces such as fluid shear, which also influence the ability of these cells to infiltrate tumours and phagocytose cancer cells. Therapeutic designs that use macrophages and other phagocytes as effector cells are likely to be advanced by recognition and careful consideration of the relevant forces and related physical factors.

Data accessibility

Clinical trial information is available at https://clinicaltrials.gov. Screenshots of cited conference abstracts and press releases are available in the electronic supplementary material.

Authors' contributions

All of the authors contributed to drafting and revising the manuscript and gave their approval for the final version to be published. J.C.A. and L.J.D. contributed equally to this work.

Competing interests

The authors declare no competing interests.

Funding

This work was supported by the National Cancer Institute of the National Institutes of Health under U54CA193417 (to D.E.D.) and F32CA228285 (to L.J.D). J.C.A. was supported by the National Science Foundation Graduate Research Fellowship Program under DGE-1845298. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health nor the National Science Foundation.