Anti-colony-stimulating factor therapies for inflammatory and autoimmune diseases

Key Points

• Colony-stimulating factors (CSFs) are pleiotropic, but each has its own unique biological role.

• All CSFs control myeloid cell numbers, but each has a level of specificity in regard to its target cells and effects.

• Preclinical and/or clinical data suggest that targeting granulocyte–macrophage CSF (GM-CSF), CSF1, granulocyte CSF (G-CSF) or IL-3 could be useful in the treatment of numerous autoimmune and/or inflammatory pathologies, including rheumatoid arthritis and multiple sclerosis. Further investigation of the potential of targeting CSFs in other pathologies is warranted.

• CSF targeting, as well as associated patient stratification, should be based on the specific CSF biology.

• There is not necessarily an inverse relationship between the effects of CSF blockade and administration of the protein itself on a particular pathology.

Abstract

Granulocyte–macrophage colony-stimulating factor (GM-CSF), macrophage colony-stimulating factor (M-CSF; also known as CSF1), granulocyte colony-stimulating factor (G-CSF) and interleukin-3 (IL-3) can each play a part in the host response to injury and infection, and there is burgeoning interest in targeting these CSFs in inflammatory and autoimmune disorders, as well as in cancer. For success in clinical medicine, therapeutic targeting will need to be delineated from current strategies. The individual CSFs have unique biological roles, suggesting that they could be used to target specific conditions. This Review compares the CSFs, with a focus on how they could be targeted, discusses the relevant clinical trial data and summarizes the potential clinical applications of targeting each CSF. Importantly, we discuss the novelty of CSF biology and attempt to clarify some of the surrounding misconceptions and issues that can affect therapeutic decisions.

Main

Granulocyte–macrophage colony-stimulating factor (GM-CSF; also known as CSF2), macrophage colony- stimulating factor (M-CSF; also known as CSF1), granulocyte colony-stimulating factor (G-CSF; also known as CSF3) and interleukin-3 (IL-3; also known as multi- CSF) were originally defined by their respective abilities to generate different types of myeloid populations from precursor bone marrow cells in vitro¹. It later became apparent that they could also affect more mature populations in these lineages, promoting their survival and/or proliferation, activation and differentiation^2,3, all of which could be relevant to inflammation⁴. The effects of Csf gene deficiency in mice and of specific neutralizing monoclonal antibodies (mAbs) have spurred a reassessment of the roles that these CSFs have in steady-state haematopoiesis. Each of these CSFs can play a part in the host response to tissue injury and infection, which has potential implications for inflammatory and autoimmune diseases.

The receptors for the various CSFs are quite distinct (see Fig. 1 and below) and are expressed on different myeloid populations: for example, the GM-CSF receptor (GM-CSFR) is expressed on monocytes, macrophages, eosinophils and neutrophils; the CSF1 receptor (CSF1R; also known as FMS) on monocytes and macrophages; the G-CSF receptor (G-CSFR) on neutrophils; and the IL-3 receptor (IL-3R) on monocytes, macrophages, eosinophils, basophils, plasmacytoid dendritic cells (pDCs) and mast cells (Fig. 2). CSF receptor distribution is discussed in more detail in the sections describing the individual CSFs. Each receptor is stimulated by a specific CSF, except for CSF1R, which binds to both CSF1 and IL-34. CSF1R is also unique among the CSF receptors as it is a tyrosine kinase, whereas the others activate downstream signalling, for example, via Janus kinase (JAK)-dependent or signal transducer and activator of transcription (STAT)-dependent pathways.

Figure 1: The structures of the CSF receptors.

Highly schematic representation of the human colony-stimulating factor (CSF) receptors for the ligands granulocyte–macrophage CSF (GM-CSF), CSF1, granulocyte CSF (G-CSF) and interleukin-3 (IL- 3). The GM-CSF receptor subunit (depicted) consists of a unique ligand-binding α-chain and a β-common (β_c) chain through which signalling occurs; the CSF1 receptor is a homodimeric type III receptor tyrosine kinase with intracellular signalling domains as depicted; the homodimeric G-CSF receptor is typical of the type I cytokine receptor family; and the IL-3 receptor subunit, as for the GM-CSF receptor, consists of a unique α-chain and the β_c subunit. Note that both the GM-CSF receptor and the IL-3 receptor signal as multimers.

PowerPoint slide

Figure 2: CSFs and control of target cell numbers and function in inflammation.

During an inflammatory response, it is likely that the major functions of endogenous colony-stimulating factors (CSFs) are in the inflamed tissue, in which CSFs increase the cell numbers of selected myeloid populations by enhancing their survival and/or modifying their trafficking in or out of the tissue. The CSFs are also likely to cause activation or maturation of the target cells to enhance the pro-inflammatory and possible resolving functions of these cells. Tissue macrophages may also proliferate (not shown). The tissue-derived CSFs may also act systemically to activate specific myeloid cells in the blood before their migration into inflamed tissue and contribute to this migration, as well as to the myeloid cell migration (mobilization) from the bone marrow. In the bone marrow, they may also promote lineage-specific myelopoiesis from progenitor cells by proliferation and differentiation. CSF targeting (depicted by red inhibitory lines) could have effects locally in the inflamed tissue and systemically to control disease. G-CSF, granulocyte CSF; GM-CSF, granulocyte–macrophage CSF; IL-3, interleukin-3; pDC, plasmacytoid dendritic cell.

PowerPoint slide

CSF1, unlike the other CSFs, is present at high levels in the blood and is ubiquitously and constitutively expressed in many tissues and by many cell types; the other CSFs are usually produced only after stimulation: for example, GM-CSF and G-CSF are produced by stimulated haemopoietic and non-haemopoietic cells, and IL-3 is produced by stimulated T cells and mast cells. The diverse receptor distribution and potential sources of the CSFs are key to the delineation of their separate biological functions and their potential roles in pathology.

Roles for the different CSFs and their targeting, particularly GM-CSF and CSF1, are increasingly being investigated in preclinical models and clinical trials of inflammatory and autoimmune diseases. In this Review, the CSFs are discussed in an order that reflects our understanding of these molecules as potential therapeutic targets in these diseases. The preclinical models of immune-mediated inflammatory diseases in which CSF targeting (either by deletion or by blockade) has been therapeutically beneficial are listed in Table 1 and are discussed in the individual CSF subsections below. Other agents that can promote myelopoiesis, such as FMS-related tyrosine kinase 3 ligand (FLT3L), are not discussed in this Review.

Table 1 Preclinical models showing therapeutic benefit of CSF depletion or inhibition

Clinical trials investigating anti-GM-CSF and anti-CSF1 therapies in immune-mediated inflammatory diseases have recently been completed^5,6,7,8 and have provided initial evidence of efficacy and safety of these agents. This Review puts the trials into perspective and also summarizes the potential clinical applications of targeting each of the CSFs; importantly, it discusses the novelty of CSF biology and attempts to clarify some of the surrounding misconceptions and issues that can affect therapeutic decisions. It is increasingly being appreciated that inflammation affects numerous pathologies; thus, the development of novel drugs that alter the course of the inflammatory response has therapeutic relevance for a diverse array of diseases.

GM-CSF

GM-CSF biology and downstream signalling

These topics have been referred to briefly above and are well addressed in recent reviews^3,9,10,11,12. Therefore, only a brief summary is provided here with some recent updates. The widespread clinical use of GM-CSF in cancer therapy is not discussed in this Review.

Briefly, most of the published work on GM-CSF biology describes its interaction with myeloid cell populations, such as macrophages, neutrophils and eosinophils. Based on studies with gene-deficient mice, GM-CSF is required for the steady-state development of alveolar macrophages, invariant natural killer T cells, an intestinal lamina propria dendritic cell (DC) population and possibly dermal CD103⁺ dendritic cells (CD103⁺ DCs)¹¹, which is discussed in more detail below. GM-CSF has also been considered to control the differentiation and activation of mouse CD8⁺ splenic DCs¹³ and human pDCs¹⁴ and to act as the principal CD8⁺ T cell-derived 'licensing factor' for DC activation in mice¹⁵.

GM-CSF, particularly during inflammatory responses, can be produced by a number of both haemopoietic and non-haemopoietic cell types upon their stimulation; moreover, the ligand can activate myeloid populations to produce inflammatory mediators. Therefore, GM-CSF is a multifunctional cytokine that acts at the interface between innate and adaptive immunity. As some molecules, such as IL-1β, tumour necrosis factor (TNF) and IL-23, can be involved in both GM-CSF production and as mediators of its downstream effects, it can be difficult to decide whether GM-CSF is an upstream or downstream regulator of such mediators. As a result, GM-CSF 'networks' have been proposed in an attempt to clarify matters and explain lesion chronicity^{9,16,17,18,19,20,21}. If such networks exist, therapeutic interventions that target GM-CSF during inflammation will restrict not only the inflammatory myeloid cells that respond to this cytokine but also the cells that produce GM-CSF, such as autoaggressive T cells¹⁹.

The basic GM-CSFR subunit, which is expressed mainly on myeloid populations, consists of a unique ligand-binding α-chain that contains three extracellular domains and a signalling β-chain (Fig. 1); this β-common (β_c) chain is also a component of the IL-3R. The crystal structure of the human GM-CSFR bound to GM-CSF is unusual: it consists of a hexamer (containing two molecules each of the ligand, the receptor α-chain and the receptor β-chain) and a dodecamer, which is formed from the association of two hexamers^22,23. The different biological responses (namely, survival, proliferation, activation and/or differentiation) that occur at different GM-CSF concentrations in myeloid populations have been explained by a sequential model of GM-CSFR activation according to the proposed structure; although the hexamer can bind the ligand, the dodecamer activates intracellular signalling pathways²². In myeloid populations and during DC development, engagement of the GM-CSFR activates STAT5, a key transcription factor that is most likely phosphorylated by activated JAK2 (Refs 22,24); the RAS–RAF–MEK–ERK mitogen- activated protein kinase (MAPK), nuclear factor-κB (NF-κB) and phosphoinositide 3-kinase (PI3K)–AKT pathways have also been reported to be activated as a result of GM-CSFR engagement²⁴.

GM-CSF in preclinical models

The effects of GM-CSF administration, depletion or deletion in preclinical models of inflammatory and autoimmune diseases have been reviewed recently^11,12. Notably, the effects of high doses of systemically administered GM-CSF on a disease may not necessarily be informative about the role of endogenous, potentially locally acting, GM-CSF in that disease^3,11. It is also worth noting that data obtained using a Csf2^−/− mouse may not be equivalent to results obtained using a neutralizing mAb approach to reduce GM-CSF activity; in addition, mouse strain differences in macrophage biology²⁵ could affect the efficacy of preclinical anti-GM-CSF therapy.

As GM-CSF can function as an adjuvant, it is frequently used to potentiate the immune system. For example, it can exacerbate experimental arthritis²⁶ and lung inflammation²⁷, as well as induce disseminated histiocytosis²⁸. However, GM-CSF administration can also improve outcomes, for example, in myasthenia gravis, diabetes, colitis and wound-healing models^3,9,29,30.

Neutralization of GM-CSF by mAb treatment or genomic deletion (Csf2^−/− mice) was initially found to be effective in ameliorating inflammatory arthritis and experimental autoimmune encephalomyelitis (EAE)^31,32,33. These strategies have been effective in various other inflammatory and autoimmune models^3,9,11,12, including asthma, chronic obstructive pulmonary disease (COPD), nephritis, psoriasis, atherosclerosis, peritonitis, cancer, Alzheimer disease (AD), myocardial infarction, peripheral insulin resistance, inflammatory pain, interstitial lung disease, aortic aneurysm, epidermolysis bullosa acquisita, periodontal disease³⁴, aortic dissection in mice³⁵, intramural haematoma and myocarditis²¹ (Table 1).

T_H17 cells are a subset of T helper cells (T_H cells) that were originally defined by their production of IL-17. In EAE, it was reported that T_H17 cells, which predominantly produce IL-17, are pathogenic and that GM-CSF is crucial for this function^17,18. GM-CSF, but not IL-17, is required for the development of severe interstitial lung disease in SKG mice, a strain that spontaneously develops T cell-dependent inflammatory polyarthritis as a result of a spontaneous germline point mutation in zeta chain T cell receptor- associated protein kinase 70 (Zap70)³⁶. The complementary action of GM-CSF and IL-17 inhibitors in the suppression of arthritis in mice has been proposed as a rationale for combination therapy in the treatment of rheumatoid arthritis (RA)³⁷.

In contrast to these examples, in which GM-CSF depletion or deletion results in a beneficial, anti- inflammatory effect, exacerbation of the phenotype has been noted when such strategies were tested in some models of atherosclerosis and colitis^3,9,11. Interestingly, in dextran sodium sulfate (DSS)-induced colitis, Csf2^−/− mice are more severely affected than wild-type mice³⁸, even though exogenous GM-CSF has been suggested to be pro-inflammatory³⁹. In the 2,4,6-trinitrobenzenesulfonic acid (TNBS)-induced mouse model of colitis³⁹, genetic deletion of GM-CSFR did not affect pathogenesis, but anti-GM-CSF mAb treatment led to disease suppression⁴⁰. Furthermore, anti-GM-CSF mAb treatment attenuated IL-23-driven T cell transfer-induced colitis⁴¹, and the same group recently identified a GM-CSF-dependent role for eosinophils in the pathogenesis of IL-23-induced and T_H17 cell-induced colitis⁴². Recent findings showed that anti-CD40-induced innate immune colitis is dependent on group 3 innate lymphoid cell (ILC)-derived GM-CSF⁴³, and an IL-23R–GM-CSF axis within group 3 ILCs, which has a pathogenic role in colitis, has been identified as a key control point in the accumulation of innate effector cells in the intestines⁴⁴. As discussed elsewhere¹¹, these model-specific differences in the role of GM-CSF in colitis are likely to be due to different pathogenic mechanisms and/or to differences in the extent of GM-CSF depletion⁴⁰.

GM-CSF biology: outstanding questions

Does GM-CSF promote DC development? GM-CSF has been widely used to generate DC-like populations from mouse and human precursors, particularly in vitro, either alone or in combination with other cytokines, such as IL-4 (Ref. 24); however, this might not reflect DC development in vivo.

Mouse CD103⁺ DCs from different tissues have distinct functional activities. Migratory CD103⁺ DCs derived from the skin, lungs and intestines efficiently present exogenous antigens to specific CD8⁺ T cells in their corresponding draining lymph nodes, intestinal CD103⁺ DCs can drive the induction of gut-homing receptors, and CD103⁺ DCs can also control inflammatory responses and intestinal homeostasis by fostering the conversion of naive T cells into FOXP3⁺ regulatory T cells⁴⁵. There has been disagreement about the role of GM-CSF in CD103⁺ DC development in mice in vivo^11,46,47,48; perhaps varying levels of GM-CSF can explain the discrepancies between different studies⁴⁹. As discussed elsewhere¹¹, there is also disagreement as to whether GM-CSF controls monocyte-derived DC development, particularly during inflammation, probably, in part, owing to the difficulty in defining inflammatory DC and macrophage populations categorically^{13,47,50,51,52,53}. To add to this debate, GM-CSF evidently plays no part in the phenotypic transition from monocytes to monocyte-derived DCs in EAE, as the critical GM-CSFR-mediated signalling occurs before this terminal differentiation step¹⁹. The debate about the phenotype of cultured cells derived from GM-CSF-treated mouse bone marrow continues: indeed, they seem to comprise a heterogeneous population of CD11c⁺MHCII⁺ macrophages and DCs^54,55.

Cell numbers and/or activation. GM-CSF can function as a survival factor — or even as a mitogen — for macrophage lineage populations and can also control differentiation and activation of myeloid cells^3,56. It is unclear as to what extent the pro-inflammatory effects of GM-CSF are due to altered cell numbers and/or phenotypic changes in vivo (Fig. 2) and could be context dependent, as discussed above for DC development and 'CSF networks'. Macrophages isolated from RA synovial fluid exhibit a GM-CSF-like transcriptomic and phenotypic pro-inflammatory state, which suggests the involvement of GM-CSF in macrophage activation in RA⁵⁷. Reduced local survival, reduced proliferation, altered trafficking in or out of the lesion or a combination of these could explain situations in which cell infiltration into an inflammatory lesion is lowered upon GM-CSF depletion^53,56 (Fig. 2). As an example, the pro-survival function of GM-CSF in eosinophils in asthma may contribute to the disease⁵⁸. GM-CSF augments glycolytic flux in human macrophages in vitro in a TNF-dependent manner, and GM-CSF increases fludeoxyglucose (¹⁸F) uptake within inflamed atheroma in vivo⁵⁹. Strangely, given the well-documented pro-survival function of GM-CSF in monocytes and macrophages, the presence of GM-CSF promotes plaque progression in low-density lipoprotein- driven atherosclerosis in mice by increasing macrophage apoptosis susceptibility via IL-23 signalling⁶⁰.

GM-CSF: an M1 or M2 cytokine? The different activation states of macrophages are widely referred to as M1 ('classical activation') and M2 ('alternative activation'). GM-CSF-dependent macrophage polarization has been considered to be M1-skewing⁶¹ and, at least in terms of cytokine expression, has been likened to M1 polarization^62,63; however, the presence of GM-CSF in the lungs has been associated with T_H2 immunity (and therefore M2 polarization)^64,65, GM-CSF can regulate inflammation via the typical M2 cytokine C-C motif chemokine ligand 17 (CCL17)⁵⁵ and, based on preclinical and clinical studies in glioma, GM-CSF can trigger and drive the alternative activation of tumour-infiltrating microglia and/or macrophages by which these cells support tumour growth⁶⁶. GM-CSF signalling may also play a crucial part in macrophage repair functions^67,68,69. Thus, it has been recommended that the M1–M2 paradigm should not be applied too rigidly to GM-CSF-treated monocytes and/or macrophages^70,71, or should even be abandoned⁷².

T_H cell subsets and GM-CSF formation. In mouse models of autoimmunity, such as EAE, the non-redundant pathogenicity of T_H17 cells has been associated with GM-CSF production^17,18,73. However, it has recently been suggested that a novel subset of T_H cells named 'T_H-GM' — which are IL-7–STAT5 dependent — predominantly produces GM-CSF but not IL-17, develops independently of the mechanisms required for T_H1 or T_H17 differentiation and is essential for the development of EAE⁷⁴. Furthermore, in a recent study using human cells, distinct and counter-regulatory pathways for the generation of IL-17-producing and GM-CSF-producing T_H cells were demonstrated as evidence for distinct T_H17 and T_H-GM lineages⁷⁵; the number of GM-CSF-producing T_H cells was increased in the cerebrospinal fluid of patients with multiple sclerosis (MS). To add to the complexity, the notion that IL-23 and T_H17 plasticity are universally required for the development of EAE has been challenged: IL-12-polarized T_H1 cells produce GM-CSF and induce EAE independently of IL-23 (Refs 76,77). Libraries that were derived from C-C motif chemokine receptor 6 (CCR6)-positive, myelin- reactive memory T cells from individuals with MS exhibited significantly enhanced GM-CSF production compared with those derived from healthy controls⁷⁸. It has been reported in vitro that GM-CSF-treated human macrophages have a phenotype similar to that observed in MS lesions⁷⁹. Evidence that either supports or refutes the role of T_H-GM cells in autoimmune disease pathogenesis is awaited with interest, although GM-CSF production in stimulated resident tissue cells, such as endothelial cells and fibroblasts, could also be important in some of these diseases^3,11,80.

Although T_H-GM cells are yet to be studied in other diseases, an enrichment of GM-CSF-expressing T cells in the joints of patients with juvenile idiopathic arthritis has been reported, which correlates with the degree of systemic inflammation in these individuals⁸¹. Synovial CD4⁺ T cell-derived GM-CSF supports the differentiation of an inflammatory DC population⁸² and a significant increase in the number of synovial macrophages that express the α-chain of GM-CSFR has been observed in individuals with RA or psoriatic arthritis⁸³.

B cells and GM-CSF. A GM-CSF-producing B cell population, termed innate response activator (IRA) B cells, has been identified in a sepsis model, and these cells protect against sepsis through an uncharacterized GM-CSF-dependent mechanism⁸⁴; they also protect against pneumonia via a GM-CSF–IgM axis in the lungs⁸⁵. However, these B cells shift the leukocyte response, via GM-CSF secretion, towards a T_H1-associated milieu that aggravates atherosclerosis⁸⁶. Memory B cells from patients with MS express high levels of GM-CSF, thereby rationalizing B cell depletion as a treatment option for MS⁸⁷. Thus, production of GM-CSF by B cells seems to enhance the immune response, which is beneficial when fighting infections but is potentially detrimental in chronic inflammatory and/or autoimmune diseases. Splenic ILCs release GM-CSF, which co-opts neutrophils, resulting in increased antibody production from marginal zone B cells⁸⁸. These unexpected links to B cell biology are intriguing and await further analysis.

GM-CSF and the nervous system. In addition to the involvement of GM-CSF in the nervous system in EAE and MS, as discussed above, there is a body of literature that suggests neuroprotective properties of GM-CSF in models of neurological diseases and injury⁸⁹. It is possible that GM-CSF-induced neuroprotection is mediated through the interaction with neuronal GM-CSFR^90,91; GM-CSF has also been suggested to sensitize nerves to mechanical stimuli directly via its receptor on neurons⁹¹, and a GM-CSF-stimulated neuronal transcriptome has been reported⁹². Both ligand and receptor are reported to be expressed in the adult human brain⁹³.

Consistent with the notion that GM-CSF has a role in pain, blocking the α-chain of GM-CSFR suppressed bone cancer pain⁹¹. GM-CSF is also required for the development of inflammatory and arthritic pain in mice, as well as pain in a mouse model of osteoarthritis (OA)^94,95, although the contribution of a direct GM-CSF–neuron interaction in these studies is unknown. Interestingly, in one of the models of GM-CSF-driven arthritic pain, a cyclooxygenase inhibitor was very effective at blocking pain, which suggests the involvement of an eicosanoid(s). As a follow-up to these preclinical findings in osteoarthritic pain, a phase II trial in inflammatory hand OA has been initiated, which is investigating the potential of anti-GM-CSF mAb treatment for disease modification and analgesic activity (see below). It is unknown whether targeting GM-CSF as a therapy for pain or MS is likely to compromise neural function, although it should be mentioned that most of the studies that examined the beneficial role of GM-CSF-mediated neuroprotection involved administered GM-CSF; the potential difficulties in drawing conclusions about the role of endogenous GM-CSF in such cases have been highlighted above.

Other outstanding questions. There is disagreement from studies using Csf2^−/− mice as to whether GM-CSF is essential for normal weight maintenance during development^96,97. GM-CSF may have a protective role in obesity, as Csf2^−/− mice are more obese than wild-type mice are when both are fed a high-fat diet; however, Csf2^−/− mice have decreased adipose tissue inflammation, which leads to amelioration of peripheral insulin resistance in response to a high-fat diet, despite the increase in adiposity⁹⁸. There are also conflicting reports about the fertility of Csf2^−/− mice^97,99.

Clinical studies targeting GM-CSF

Several clinical trials that used neutralizing mAbs to target GM-CSF or its receptor in inflammatory diseases have been carried out. As most of these trials have been reviewed recently^11,12, a brief summary of this information is given here, as well as information on the most recent human studies (Table 2).

Table 2 Clinical trials targeting GM-CSF or its receptor in inflammatory diseases

Mavrilimumab. Mavrilimumab (previously known as CAM-3001) is a high-affinity IgG4 mAb, which was developed by MedImmune against the α-chain of the GM-CSFR. The first study was a randomized, double- blind, placebo-controlled, dose-escalating phase I study (NCT00771420) in 32 patients with mild to moderate RA and, even though the study was not designed to assess efficacy, the results were encouraging and showed a satisfactory safety and tolerability profile¹⁰⁰. Next, the multicentre, randomized, double-blind, placebo- controlled, dose-escalating EARTH study (NCT01050998) was a phase IIa trial over a 12-week period in individuals with moderate to severe RA (n = 239); for individuals on stable background methotrexate therapy at week 12, 56% of those treated with mavrilimumab met the primary end point, manifested as a reduced disease activity score, as did 35% of those treated with placebo¹⁰¹. Response rate differences between mavrilimumab-treated and placebo-treated individuals were observed at week 2 and increased throughout the treatment period, and adverse reactions were mild to moderate. In a Japanese cohort (n = 51)¹⁰², the results and the profile of adverse events were similar to those of the European cohort discussed above.

Data from a subsequent 24-week, phase IIb study (NCT01706926) in patients with RA (n = 326) who had moderate to severe RA and received background methotrexate treatment have also been reported^{103,104,105,106}. The following conclusions were drawn: mavrilimumab produced rapid and clinically meaningful responses across a range of disease activity parameters, including pain, meeting both co-primary end points with a clear dose–response effect; early and sustained improvements were observed in markers of disease activity, including the disease activity score evaluated in 28 joints that included analysis of C-reactive protein levels (a score known as DAS28-CRP) and the American College of Rheumatology (ACR) 20 response; an acceptable safety and tolerability profile was shown over the 24-week period.

A non-randomized open-label extension (OLE) study (NCT01712399)⁸ to investigate long-term (at 74 weeks) safety and tolerability in patients with RA (n > 300) who completed the EARTH Explorer 1 study has recently been reported. Of note, no radiographic progression was observed in 68% of the cases and clinical parameters improved without evidence of increased adverse events. There is a theoretical risk of pulmonary alveolar proteinosis (PAP), as anti-GM-CSF autoantibodies have an important role in the pathogenesis of this disease. However, there have been no significant pulmonary signals in the clinical trial programme so far. This finding is consistent with the recent proposal¹⁰⁷ that PAP is unlikely to be induced by mAbs against GM-CSF, but only by polyclonal antibodies, as found in autoimmune PAP. A randomized, double-blind, placebo-controlled phase II study (NCT01715896) of 120 individuals with moderate to severe RA, which set out to compare the efficacy and safety of a subcutaneous dose of mavrilimumab with the commercial anti-TNF mAb golimumab, has been completed; however, these results have not yet been published.

GSK3196165. A randomized, double-blind, placebo- controlled, dose-escalating phase Ib/IIa trial (NCT01023256) in patients with moderate RA (n = 96) has been completed with GSK3196165 (previously known as MOR103), a human anti-GM-CSF mAb developed by MorphoSys AG and in-licensed by GlaxoSmithKline¹⁰⁸. GSK3196165 was well tolerated by participating individuals and showed evidence of rapid efficacy (clinical responses) and a sustained response up to 10 weeks beyond the 4-week treatment period. Trials with larger sample sizes and longer treatment periods are ongoing to fully assess this therapeutic compound.

A phase Ib study (NCT01517282)¹⁰⁹ with this mAb has also been carried out in patients with MS to investigate drug safety; the treatment was generally well tolerated in these individuals with relapsing-remitting MS and secondary progressive MS.

Furthermore, a phase II trial in hand OA (NCT02683785; EU Clinical Trials Register EudraCT number: 2015-003089-96) has just commenced, which explores the potential of anti-GM-CSF treatment for disease modification and analgesic activity.

KB003. A randomized phase II clinical trial (NCT00995449) in individuals with RA that tested a humanized anti-GM-CSF mAb termed KB003 (developed by KaloBios) has been terminated owing to a programme refocus; KB003 was generally safe and well tolerated over approximately 3 months of repeated dosing. The findings from a phase II, randomized, double-blind, placebo-controlled 24-week study (NCT01603277) in patients with severe asthma (n = 160) have just been reported⁶; although the primary objective of improved lung function was not met in the overall population, some improvement was noted in pre-specified groups, and there were no safety signals, including no evidence of PAP. It was suggested that higher dosing and/or further asthma phenotyping may be required.

Namilumab. Namilumab (previously known as MT203), a human IgG1 anti-GM-CSF mAb, has been investigated for safety and tolerability in a double- blind, placebo-controlled, randomized, dose-escalating phase Ib study (NCT01317797)¹¹⁰ in patients with mild to severe RA (n = 24). Initial evidence of good tolerability and safety were observed along with efficacy¹¹¹. Additional phase II trials (NCT02393378 and NCT02379091) are underway, which will provide more information on efficacy compared with treatment that uses the anti-TNF antibody adalimumab. Namilumab is also being tested in a 12-week phase II, randomized, double-blind, placebo-controlled, dose-escalating trial (NCT02129777) in individuals with moderate to severe plaque psoriasis.

MORAb-022. A phase I trial (NCT01357759) of this human mAb against GM-CSF (developed by Morphotek) in patients with RA (n = 20) has been completed; however, no results have been disclosed to date.

GM-CSF targeting: perspectives

The clinical trial data on therapies that target GM-CSF or its receptor in inflammatory diseases are encouraging so far with respect to safety, efficacy and speed of response. Larger and longer trials are needed in RA and other indications and, in our view, are worth carrying out. Additional animal studies are still warranted to explore the mechanisms in GM-CSF-driven biology but are probably not necessary to determine future clinical applications. In regard to the broader potential of GM-CSF and/or GM-CSFR targeting, determining factors will include the competitive landscape, unmet medical need and the relevance of GM-CSF in human biology. Because myeloid cell populations seem to be the main targets of GM-CSF activity during inflammation, the functions of GM-CSF are likely to be more restricted than those of pro-inflammatory cytokines with relatively broad effects, such as TNF. These functional differences could mean that GM-CSF represents a unique targeting opportunity. Furthermore, some evidence suggests that GM-CSF is expressed earlier than TNF in the course of RA disease progression^112,113,114. Clarification is still needed on whether targeting the ligand or the receptor will be more beneficial for clinical intervention.

CSF1

CSF1 biology and downstream signalling

CSF1 is unique among the CSFs for a few reasons. First, CSF1 is ubiquitously expressed in the steady state; second, its receptor, CSF1R, is a tyrosine kinase, which CSF1 shares with IL-34, thereby increasing complexity; third, among myeloid populations, the receptor is thought to be expressed on cells of the mononuclear phagocyte system (MPS); and, fourth, CSF1 exists in both secreted and membrane-bound forms, which have different effects. As CSF1R is a kinase, oral inhibitors have been developed (see below). Two unanswered questions in CSF1 biology are to what extent CSF1R is expressed on cells outside the MPS and whether CSF1-stimulated macrophages can interconvert into cells of mesenchymal lineages¹¹⁵.

Recent studies have demonstrated that homeostasis in the adult MPS involves local tissue control and/or a contribution from the blood monocyte pool, depending on the tissue¹¹⁶. Based on depletion studies, CSF1 can have an important role in tissue MPS homeostasis¹¹⁷. It seems that many of the effects of CSF1 and CSF1R mutations reflect developmental roles that are redundant or functionally compensated in an adult mouse¹¹⁷. The divergent data on how fast CSF1R and/or CSF1 neutralization can reduce MPS cell numbers^115,118,119 has been discussed recently⁵³.

In the intestine, crosstalk between muscularis macrophages and enteric neurons regulates gastrointestinal motility, and enteric neuron-derived CSF1 is required for the development of these macrophages¹²⁰. There is evidence that CSF1R directly supports small intestine Paneth cell maturation and that, in turn, these cells determine the intestinal stem cell niche¹²¹. The maintenance of different tissue macrophage populations at steady state by CSF1 means that it plays a part in both maintaining tissue patency and also in restoring tissue homeostasis following injury or inflammatory damage^3,9. To what extent CSF1-dependent local macrophage proliferation contributes to macrophage steady-state numbers seems to be tissue dependent^{53,122,123,124,125}. Although CSF1 can instruct haemopoietic stem cells to generate macrophage lineage progeny in vitro and in vivo^126,127, based on studies of CSF1R and/or CSF1 neutralization^{53,118,119,128}, endogenous CSF1 may not be necessary in the steady state for MPS development in the adult mouse in cells that are more immature than LY6C⁻ blood monocytes.

Some genetic evidence supports this perhaps surprising conclusion. Mutations in CSF1R that abrogate the expression of the affected allele or lead to the expression of mutant receptor chains that are devoid of kinase activity have been identified in both familial and sporadic cases of adult-onset leukodystrophy with axonal spheroids and pigmented glia (ASLP)^129,130,131. Consistent with a role for CSF1 in controlling the numbers of more mature blood monocytes, CD16⁺ monocytes are depleted in individuals with ASLP¹³². Inactivation of one Csf1r allele is sufficient to cause ASLP in a mouse model¹³³; aberrant activation of microglia in Csf1r^+/− mice may play a central part in this pathology and, of relevance to this Review, it was speculated that the observed increase in GM-CSF levels could have an important role in disease pathology, indicating that GM-CSF inhibition may therefore be therapeutically important. These ASLP studies form part of the evidence that there are important roles for CSF1R and its ligands in the regulation of both microglial and neuronal lineages in the brain, as discussed in more detail below.

In addition to the MPS, it has been reported, albeit not uniformly¹¹⁷, that CSF1R is expressed on other cells, such as trophoblasts, neural progenitor cells and epithelial cells¹³⁴. CSF1R signalling, via its kinase activity, promotes MPS development, as well as the survival, proliferation and chemotaxis of macrophages, by regulating tyrosine phosphorylation, activation or expression of multiple proteins¹³⁴. Many of these proteins (for example, GAB2) regulate downstream signalling pathways in a developmental, stage-specific manner.

CSF1R is the only receptor tyrosine kinase that is activated by two ligands of unrelated sequence: CSF1 and IL-34. Despite sharing low sequence similarity, the biologically active regions of CSF1 and IL-34 have similar cytokine folds. Even though there are similarities, the IL-34–CSF1R complex differs from the CSF1–CSF1R complex in many structural ways^135,136,137. Consistent with these structural data, in many aspects IL-34 functions similarly to CSF1. However, some differences in their biology have been identified: the most notable example is their expression patterns in embryonic and adult tissues¹³⁸. IL-34, rather than CSF1, is essential for the development of epidermal Langerhans cells and microglia⁹; it also interacts with receptor-type tyrosine-protein phosphatase-ζ (RPTPζ), which is co-expressed with CSF1R in several cell types and may provide additional mechanisms for fine-tuning CSF1R signalling^134,139.

CSF1 in preclinical disease models

There is some dispute as to whether CSF1R blockade can reduce the population size of inflammatory macrophages^{53,115,118,119}. Part of the debate centres around the possible differences in modes of action of anti-CSF1R mAb isotypes that have been used in various studies^53,115,119. Whether macrophage numbers are lowered may depend on whether resident tissue macrophages are required for the inflammatory lesion in question⁵³ and/or depend on how chronic the particular inflammatory reaction is. Even though recruited macrophages are generally more prominent in chronic inflammatory conditions than in acute ones, both acute neutrophil-dependent and chronic inflammatory responses may require the presence of resident tissue macrophages¹⁴⁰. One potentially confounding factor in targeting CSF1R is that its neutralization in vivo leads to dramatic increases in CSF1 levels, presumably due to reduced ligand internalization¹¹⁸. Thus, it is possible that this elevation in systemic CSF1 levels could compromise the effectiveness of receptor neutralization. A range of low-molecular-weight kinase inhibitors have been used successfully in various inflammatory, autoimmune and cancer metastasis models⁹. However, as always with this type of approach, the question of specificity arises and caution should be exercised when interpreting the findings of such studies^9,115. In addition, CSF1R-targeting strategies for a particular indication may need to take into account the effects on alterations in IL-34-mediated signalling^9,53.

CSF1 and/or CSF1R targeting can have beneficial effects in models of immune-mediated inflammatory diseases (Table 1). Improvement of conditions, such as arthritis (including bone loss), nephritis, lung fibrosis, atherosclerosis, obesity, colitis, diabetic nephropathy, orthodontic tooth movement and cancer metastasis, have been observed in Csf1^op/Csf1^op mice or upon neutralization of CSF1 or CSF1R^{3,9,115,141,142}; however, such strategies have deleterious effects in colitis¹⁴³ and skeletal muscle regeneration¹⁴⁴. Recently, both anti-CSF1R mAb and oral CSF1R antagonists showed efficacy in mouse models of arthritis and inflammation-induced bone loss^{140,145,146,147}, as well as in nephritis¹⁴⁸; the bone loss is perhaps to be expected, given the key role of CSF1 in osteoclastogenesis. However, the resolution of inflammation in post-myocardial infarction and the preservation of ventricular function were compromised by an oral antagonist¹⁴⁹. Therefore, it is unclear whether CSF1R is a good target for anti-inflammatory therapy^115,119. CSF1 depletion or blockade has been reported to suppress tumour metastasis in some preclinical models owing to its effects on tumour-associated macrophages, and clinical trials have incorporated this strategy¹⁵⁰ (see below). However, in two breast cancer models, CSF1 neutralization exacerbated metastasis¹⁵¹.

Emerging roles for CSF1R and its ligands in the nervous system have recently been identified¹⁵². It has been reported that microglial cell proliferation is a core component of the neuroinflammatory response in models of prion disease and is controlled by CSF1R activation¹⁵³. Subsequently, in a model of AD, another chronic neurodegenerative condition, pharmacological targeting of CSF1R inhibited microglial proliferation and prevented the progression of the pathology¹⁵⁴; it also prevented microglial plaque association and improved cognition in 3xTg-AD mice¹⁵⁵. Paradoxically, it had previously been found that systemic administration of CSF1 ameliorated memory deficits in another transgenic model of AD and selective deletion of Csf1r in forebrain neurons in mice exacerbated excitotoxin-induced neurodegeneration¹⁵⁶. From that study, it was concluded that CSF1 (and IL-34) provide powerful neuroprotective signals that involve CSF1R expression on neurons; however, in this context, it should be noted that in the postnatal mouse brain CSF1R has been reported to be expressed on microglia but not on neurons¹⁵⁷. Peripheral nerve injury induces de novo expression of CSF1 in injured sensory neurons, which, after being transported to the spinal cord, induces microglial proliferation and DNAX-activation protein 12 (DAP12; also known as TYROBP)-dependent pain. CSF1 is therefore a potential target for neuropathic pain therapy¹⁵⁸, a conclusion that was also reached in a separate study¹⁵⁹.

In two models of Charcot–Marie–Tooth type 1 demyelinating neuropathy, systemic short-term and long-term inhibition of CSF1R by oral administration led to a robust ∼70% decline in the nerve macrophage population and a substantial reduction in the typical histopathological and functional alterations, which could have clinical ramifications¹⁶⁰. In one of these models, transgenic expression of different CSF1 isoforms intriguingly led to opposing effects on macrophage activation and disease progression¹⁶¹.

CSF1 biology: outstanding questions

Based on the discussion above, several research questions can be raised, including to what extent CSF1R is expressed on non-MPS populations, whether CSF1-stimulated macrophages can convert into mesenchymal cells, at what stage CSF1 is required for steady-state monopoiesis, how rapidly CSF1R and/or CSF1 neutralization can deplete steady-state MPS populations (with implications for the mechanisms behind such depletion, namely reduced survival and/or modified trafficking), and when CSF1 and/or IL-34 is the relevant CSF1R ligand. Additional questions are discussed in the following sections.

CSF1: a pro-inflammatory or pro-repair cytokine? Even though CSF1 can maintain tissue macrophage numbers in the steady state, whether its blockade can lower such numbers during an inflammatory reaction is debated, as already discussed elsewhere^{53,115,118,140,162}. There is evidence that during an inflammatory reaction, CSF1 can increase tissue macrophage numbers by inducing local proliferation in some situations¹²³, but not in all⁵³. It has been reported that under hypoxic conditions in vitro — that is, conditions similar to those at many sites of inflammation — the mitogenic activity of CSF1 towards macrophages is enhanced¹⁶³.

In addition to being a pro-survival, mitogenic cytokine for macrophages, CSF1 can 'activate' them. Such activation, possibly via the transcription factor interferon regulatory factor 4 (IRF4), has been termed M2 or M2-like, which is consistent with a role in the resolution of an inflammatory reaction¹⁶⁴. However, it has been recommended that such nomenclature should not be used for CSF1-mediated activation, as this term usually refers to macrophages stimulated in vitro with IL-4 (Ref. 72) and, in addition, the role of IRF4 in this process has been questioned^9,71. Furthermore, among human monocytes polarized to different phenotypes by various cytokines, the CSF1-oriented cells have the highest pro-inflammatory response to RA-specific immune complexes that contain anti-citrullinated protein antibodies (ACPAs)¹⁶⁵, and Fcγ receptor–Toll-like receptor (TLR) crosstalk elicits pro-inflammatory cytokine production in these cells¹⁶⁶.

As CSF1 is important for macrophage homeostasis and can elicit some opposing responses to pro-inflammatory cytokines in these cells, it has been proposed that steady-state macrophages are in a state of 'CSF1 resistance', which must be overcome for an inflammatory profile to manifest³. This concept is consistent with evidence showing that exogenous CSF1 can promote repair, presumably in an attempt to restore tissue patency^3,167. For example, it has been reported that proximal tubule production of CSF1 is important for renal macrophage proliferation and M2 polarization during the kidney repair that occurs after acute injury^168,169, as well as for liver homeostasis and repair^170,171. Whether these homeostatic or repair functions of CSF1 have implications for its clinical targeting is currently unknown. It is intriguing to note that GM-CSF has also been reported to have these functions⁶⁸. CSF1 administration has been proposed as a potential prophylactic therapy to limit acute graft-versus-host disease (GVHD)¹²⁸, although targeting CSF1R signalling for chronic GVHD has also been suggested¹⁷². However, as for GM-CSF, findings that result from systemic CSF1 administration may not necessarily provide information about the role of endogenous CSF1 in a particular tissue or condition, in our view^3,9.

CSF1 and the nervous system. In the nervous system, CSF1R is expressed on microglia, but there is disagreement regarding its expression in the neuronal lineage¹⁵². The selection of an appropriate CSF1R detection system is important for resolving this issue¹⁵². Although CSF1R has a role in ASLP pathology, more information is needed on the significance of CSF1R-regulated functions in microglia, including the role of IL-34, for brain development and function as well as whether there are additional CSF1R ligands^152,157.

Clinical studies targeting CSF1

CSF1R expression has been shown to be increased in the synovium of patients with RA compared with individuals with OA or healthy controls¹⁴⁰. In a recent study, CSF1 and IL-34 were expressed in the RA synovium, but not at a higher level than in the synovia of patients with psoriatic arthritis or inflammatory OA¹⁴⁵. Independent neutralization of CSF1 or IL-34 had no effect on RA synovial explant IL-6 production¹⁴⁵; however, as an anti-CSF1R mAb reduced the production of IL-6 and other inflammatory mediators in RA synovial explants, it was therefore suggested that CSF1R blockade could be a novel therapeutic strategy. CSF1 levels are elevated in the serum of patients with systemic lupus erythematosus (SLE)¹⁷³ and in patients with lupus nephritis who experience renal flares¹⁷⁴. In addition, elevations in serum CSF1 in patients with SLE correlate with renal disease activity and are predictive of flares¹⁷⁵.

There have been several clinical trials targeting CSF1 or its receptor; CSF1R has been targeted with mAbs or oral kinase antagonists. The trials for inflammatory diseases are listed in Table 3, and the many studies of cancer growth or metastasis are mentioned in the text.

Table 3 Clinical trials targeting CSF1 or CSF1R in inflammatory diseases

JNJ-40346527. Following a phase I study in healthy volunteers, the oral CSF1R kinase inhibitor JNJ-40346527 (developed by Janssen) was tested in a 12-week, randomized, double-blind, placebo-controlled phase II trial (NCT01597739) in patients (n = 95) with disease- modifying antirheumatic drug (DMARD)-refractory RA⁷. Efficacy was not observed; the most common adverse events were increased levels of liver enzymes. Caveats of the study were the high placebo response rate, the low disease activity and the relatively small number of patients. Interestingly, CD14^lowCD16^hi blood monocytes were reduced, which paralleled the reduction in the blood LY6C⁻ population that followed CSF1R or CSF1 neutralization in mice^{53,118,119,128}; these respective blood populations are considered to be analogous.

PLX5622. This oral CSF1R antagonist (developed by Plexxikon) has been tested in a phase I study (NCT01282684) in healthy volunteers and a phase Ib study (NCT01329991) in patients (n = 26) with RA receiving methotrexate. No results have been disclosed to date.

FPA008. This IgG4 mAb against CSF1R (developed in a partnership between Five Prime and Bristol–Myers Squibb (BMS)) is being tested in a phase I RA trial (NCT01962337) (n = 89) and recruitment for a phase I/II trial (NCT02471716) in pigmented villonodular synovitis (PVNS) and diffuse type tenosynovial giant cell tumour (dtTGCT) (n = 45) has commenced.

PD-0360324. A phase I study (NCT00550355) with this anti-CSF1 IgG2 mAb (developed by Pfizer) has been completed in patients with RA who receive methotrexate (n = 78). No results have been disclosed to date, although the monocyte subsets were studied in one anti-CSF1 mAb-treated and one placebo-treated patient and the anti-CSF1 mAb was found to reduce the number of circulating CD16⁺ monocytes¹⁷⁶. Subsequently, in a randomized, 12-week phase I study (NCT01470313) in patients with cutaneous lupus erythematosus (CLE) (n = 8), PD-0360324 lowered CD16⁺ blood monocyte numbers, raised liver enzyme levels, possibly by reducing Kupffer cell numbers, and suppressed osteoclast marker activity, without improving clinical end points⁵. A phase II trial (NCT01732211) in chronic pulmonary sarcoidosis was terminated in 2014 owing to a business decision. The decision to terminate the trial was not based on any clinical safety or efficacy concerns.

MCS110. A phase II trial (NCT01643850) in PVNS and giant cell tumour of the tendon sheath (GCTTS) with this anti-CSF1 IgG1 mAb (developed by Novartis) is ongoing (n = 18).

CSF1R antagonists and cancer. Clinical trials have been completed or are ongoing in several solid tumour types in which macrophages (including tumour-associated macrophages) are now being targeted via CSF1R inhibitors or by blocking mAbs. These trials include: AMG 820 (developed by Amgen; NCT01444404 and NCT02713529), IMC-CS4 (Eli Lilly; NCT01346358, NCT02265536 and NCT02718911), ARRY-382 (Array Biopharma; NCT01316822), RG7155 (Roche; NCT01494688 (Ref. 150) and NCT02323191), FPA008 (Five Prime and BMS; NCT02526017), MCS110 (Novartis; NCT00757757 and NCT02435680) and PLX3397 (Plexxikon; NCT01004861 (Ref. 177), NCT02371369 and NCT01349036 (Ref. 178)).

CSF1 targeting: perspectives

As mentioned, a key issue is whether to target CSF1R or its ligands, especially given the possible ligand redundancy. Another potential issue is whether the depletion of resident, pro-resolving tissue macrophages may compromise the benefit in some situations. In inflammatory diseases, the clinical data gathered so far are not particularly encouraging. However, there are caveats, and some of the preclinical data on improved bone function¹⁴⁰ and pain alleviation^158,159 could perhaps be worth further investigation in a clinical setting.

G-CSF

G-CSF biology and downstream signalling

Like GM-CSF, G-CSF circulates at low concentrations, but its levels are elevated as part of the host response to infection or injury^3,179,180; there are high, constitutive levels in the colon⁹. In vitro, G-CSF can be produced by many cell types upon addition of pro-inflammatory stimuli such as TNF, IL-1 or lipopolysaccharide (LPS)³. G-CSF is considered to be a key regulator of granulopoiesis by inducing the development of neutrophils from progenitors and by promoting neutrophil release from the bone marrow. The role of neutrophils in inflammation, even in chronic situations, is becoming widely recognized^180,181. Given that G-CSF is a key ligand that controls neutrophil numbers (through its pro-survival and trafficking regulation functions) and activation¹⁸² (Fig. 2), it is likely to be crucial for many aspects of neutrophil function, thereby providing clues as to potential therapeutic indications^3,9,179,180. An important role for an IL-17–G-CSF pathway in the regulation of neutrophil homeostasis has been proposed, as IL-17 can regulate G-CSF levels¹⁸³, which has potential implications for anti-IL-17 therapies in development⁹.

The homodimeric G-CSFR (Fig. 1) is a transmembrane protein that is expressed on myeloid cells: the highest expression levels have been reported on neutrophils, but G-CSFR is also found on other cell types, including monocytes^{3,180,184,185}. Therefore, it cannot be assumed that direct interactions with G-CSF only occur in granulocyte lineage populations. Ligand-induced dimerization of G-CSFR rapidly triggers numerous downstream signal transduction pathways¹⁸⁶. The proximal cytoplasmic domain of G-CSFR, which contains box 1 and box 2 motifs, activates JAK transautophosphorylation, which in turn can lead to STAT3 and STAT5 phosphorylation and enhanced expression of genes involved in proliferation and differentiation; the distal cytoplasmic domain contains key tyrosine residues that can be phosphorylated and serve as docking sites for Src homology domain 2 (SH2)-containing proteins, thereby activating, for example, RAS–MAPK and PI3K–AKT signal transduction pathways. In the clinic, G-CSF has been widely used to treat neutropenia associated with chemotherapy and to mobilize haemopoietic stem cells for transplantation¹⁸⁷.

G-CSF in preclinical models

Consistent with the arthritis flares noted upon treatment of Felty syndrome with G-CSF^188,189, G-CSF administration can exacerbate arthritis in mice¹⁶². Arthritis was ameliorated upon G-CSF neutralization or deletion in two T cell-dependent mouse models of arthritis^190,191 and recently in a lymphocyte-independent, immune-complex-driven arthritis model¹⁹². G-CSF blockade also suppressed hapten-induced skin inflammation¹⁹³. Like GM-CSF, G-CSF can induce and exacerbate pain, reportedly by acting on its receptor on neurons⁹¹ (see 'G-CSF and the nervous system' below); G-CSFR blockade alleviates bone cancer pain⁹¹. G-CSFR-deficient (Csf3r^−/−) mice are relatively resistant to EAE that is induced by adoptive transfer, which is consistent with a role for neutrophils during the effector phase¹⁹⁴. Recombinant G-CSF can exacerbate MS in some patients, although its administration at pharmacological doses during EAE has yielded conflicting results¹⁹⁴. Recently, endogenous G-CSF has also been implicated in the pathogenesis of mouse experimental autoimmune uveitis, possibly by acting at multiple levels: G-CSF increases neutrophil differentiation and production, mobilizes neutrophils from the bone marrow, regulates expression of chemokine receptors on circulating neutrophils, activates endothelial cells and promotes tissue neutrophil survival and polarization of the T_H17 response¹⁹⁵. An excessive IL-1–G-CSF response has been identified as a major driver of enhanced inflammation in chronic granulomatous disease as a response to damaged cells¹⁹⁶. Endogenous G-CSF has also been implicated in the promotion of breast cancer metastasis^151,197, possibly via IL-17 that is produced by γδ T cells¹⁹⁸. G-CSF and/or G-CSFR targeting has shown benefit in models of immune-mediated inflammatory diseases (Table 1). However, G-CSF itself can have a beneficial role in some circumstances. Although neutrophil numbers and activation have been associated with obesity^199,200, G-CSF administration led to beneficial effects on obesity-associated cardiac impairment²⁰¹. Such administration also reduced pathology in many indications⁹, including lung inflammation, AD, cardiac hypertrophy and colitis^202,203. A potential explanation for these findings may be the ability of exogenous G-CSF to mobilize haemopoietic cells from the bone marrow, which changes the nature of the infiltrating cell populations, as discussed in more detail below.

G-CSF biology: outstanding questions

G-CSF depletion and neutropenia. G-CSF-deficient mice have been reported to be neutropenic²⁰⁴, so an obvious theoretical concern with targeting G-CSF is to what extent neutrophil numbers are reduced, which could have consequences for host defence against infections. Even though G-CSF can promote granulopoiesis in vitro and in vivo, granulopoiesis can still occur in the steady state and during infection (referred to as 'emergency' granulopoiesis) in a G-CSF-independent manner^{205,206,207,208}. Notably, a humanized, neutralizing anti-G-CSFR mAb (CSL324) was well tolerated in primates and did not result in neutropenia²⁰⁹. In addition, a neutralizing anti-mouse G-CSFR mAb suppressed arthritis and significantly inhibited neutrophil accumulation in the joints, which occurred without rendering the mice neutropenic, thereby suggesting that G-CSFR blockade affects the homing of neutrophils to inflammatory sites²⁰⁹.

G-CSF: a pro-inflammatory or pro-repair cytokine? As mentioned above, G-CSF depletion or administration can have beneficial or detrimental consequences in inflammatory and autoimmune diseases, depending on the model. Csf3- or Csf3r-deficient mice may be more susceptible to microorganismal insult: they are hypersensitive to DSS-induced colitis²¹⁰, possibly on account of compromised barrier function, which could help to explain the amelioration of experimental colitis²⁰² and the amelioration of symptoms in some patients with Crohn's disease who received G-CSF treatment²¹¹. G-CSF can modify the plasticity of the vasculature, as its administration can lead to both arteriogenic and angiogenic effects, an associated increase in blood flow and improved local oxygen diffusion²¹²; however, it is debated whether this improved vessel function is caused by direct effects on G-CSFR in endothelial cells or by indirect effects²¹².

In some situations, G-CSF has even been considered to be an anti-inflammatory immunomodulator¹⁸⁴. As for GM-CSF and CSF1, the effects of pharmacological doses of G-CSF on a particular disease may not necessarily be informative about the role of endogenous G-CSF in that disease. What has to be taken into account and which occurs in addition to local lesion-specific effects and enhanced antimicrobial immunity, is that systemic, exogenous G-CSF can mobilize haemopoietic cells from the bone marrow. This, in turn, probably changes the composition of peripheral blood cell populations to less mature and perhaps less inflammatory phenotypes, thereby changing the nature of the dynamic cell populations available to migrate into a site of inflammation.

G-CSF and the nervous system. G-CSF can induce pain, as measured by mechanical and thermal hyperalgesia^{91,213,214,215}; however, the induction of thermal hyperalgesia in mice could not be independently confirmed²¹⁴. G-CSF can also exacerbate pain-related behaviours after peripheral nerve injury²¹⁶. A G-CSF-dependent neuronal transcriptome has also been identified, which is consistent with a direct effect of G-CSF via its receptor on neurons⁹². These data are all consistent with findings showing that G-CSF blockade has analgesic effects⁹¹. However, G-CSF itself can attenuate neuropathic pain^217,218. Therefore, further research is required to understand the role or roles of G-CSF in the control of different types of pain.

G-CSF could also have a role in neuroprotection and neural tissue repair, as well as in improving functional recovery²¹⁹. This enhanced neurogenesis upon G-CSF administration may be direct or indirect via mobilized haemopoietic cells²¹². Protective and recovery effects have been noted in animal models of stroke, amyotrophic lateral sclerosis (ALS), AD, Parkinson disease, traumatic brain injury and spinal cord injury (SCI)²²⁰. Clinical trials have also begun in this area: for example, in ALS and in SCI²¹².

Clinical studies targeting G-CSF

A phase I study in healthy volunteers with an anti-G-CSFR mAb (CSL324; developed by CSL) (Australian and New Zealand Clinical Trials Registry (ANZCTR) ID: ACTRN12616000846426) has commenced.

G-CSF targeting perspectives

Contrary to what was first feared, it is possible that circulating neutrophil numbers will not drop to critical levels upon G-CSF and/or G-CSFR targeting, although neutrophil numbers will need to be carefully monitored in future trials. Indications with obvious neutrophil involvement — for example, arthritis and uveitis — would be reasonable starting points. Additional preclinical data would be informative, including in the area of pain control.

IL-3

IL-3 biology and downstream signalling

Along with GM-CSF and IL-5, IL-3 belongs to the β_c subfamily of the cytokines, which signal through heterodimeric cell surface receptors that are composed of a cytokine-specific, major binding subunit (α-chain) with three extracellular domains and a shared signalling subunit (β_c)¹⁰ (Fig. 1). The crystal structure of the human α-chain has been solved²²¹ and, like GM-CSFR, IL-3R is a dodecamer (A. Lopez, personal communication). The human IL-3R is highly expressed on mast cells, basophils and pDCs, and also on monocytes and macrophages, eosinophils, DCs, endothelial cells and haemopoietic progenitor cells (partially depicted in Fig. 2). Although the IL-3R can only contain the β_c chain in most species, the receptor composition in mice is atypical, as mouse IL-3R can contain either the β_c chain or a β-chain that is specific for IL-3 (Ref. 222). Despite the direct binding of IL-3 by this additional chain, both the α- and β-subunits are required for signal transduction²²³. At least in vitro, IL-3 regulates the production and function of a wide range of haemopoietic cells, including stem cells (as mentioned, IL-3 has also been called multi-CSF)². In spite of these observations in vitro, the Il3^−/− mouse has no dramatic phenotype, perhaps indicating that IL-3 is unlikely to be substantially involved in steady-state haemopoiesis. IL-3 expression is relatively restricted: it is mainly expressed by activated T cells and mast cells²²⁴ and is important for the generation and function of mast cells and basophils, and it is thereby associated with allergic inflammation^225,226,227. It can also promote the survival and activation of pDCs and eosinophils^227,228 (Fig. 2). It has been claimed that IL-3 regulates brain development via IL-3R on neural progenitors²²⁹. Given that the β_c chain is shared with GM-CSFR, many signalling outcomes for IL-3 are expected to be similar (but not identical) to those described for GM-CSF, although IL-3 signalling has not been as widely studied^10,230. With some differences, IL-3 and GM-CSF can activate human monocytes, generate overlapping responses and, usually together with IL-4, can differentiate these monocytes into cells with DC-like features^231,232,233.

Elevated numbers of cells expressing CD123 (IL-3R α-chain; IL-3Rα), which presumably are pDCs, as well as raised IL-3 levels, have been noted in the skin lesions of individuals with leprosy type 1 reactions²³⁴. There are conflicting data on whether IL3 mRNA is present in the RA synovium^235,236. Some, but not all individuals with RA have detectable IL-3 in the circulation²³⁷, and there is an association between a single-nucleotide polymorphism (SNP) in the IL-3 promoter and RA²³⁸. The significance of the elevated plasma IL-3 levels in coronary artery disease²³⁹ and schizophrenia²⁴⁰ is unclear.

IL-3 in preclinical models

The effect of IL-3 depletion or blockade has indicated a role for IL-3 in hypersensitivity models^241,242 and in the early phase of collagen-induced arthritis in mice²⁴³. However, in contrast to GM-CSF, exogenous IL-3 attenuated arthritis, apparently by modulating the development of FOXP3⁺ regulatory T cells²⁴⁴, and IL-3 pretreatment prevented arthritis²⁴⁵. IL-3 administration aggravated, whereas its blockade inhibited, lupus nephritis in MRL/lpr mice²⁴⁶. Similar findings were observed in a mouse model of abdominal sepsis and, combined with the association of high plasma IL-3 levels with mortality in human sepsis, led to the suggestion that IL-3 is a potential therapeutic target in sepsis²⁴⁷. IRA B cells have been proposed to be major sources of IL-3 in humans and mice with sepsis²⁴⁸. The data that show benefit in models of immune-mediated inflammatory diseases upon IL-3 targeting are listed in Table 1. In vitro, IL-3 has been reported to have neuroprotective activity, which leads to the suggestion that it could have such a role in AD^249,250. The relevance of IL-3 administration in a particular pathology to the pharmacological targeting of IL-3 in that pathology is unclear, as discussed for other CSFs.

In β_c-deficient mice, allergen-induced eosinophilia and allergic airway inflammation were significantly reduced, probably because IL-3, GM-CSF and/or IL-5 signalling was compromised²⁵¹. A potent therapeutic mAb (CSL311), which binds to a unique epitope that is specific to the cytokine-binding site of the human β_c chain, has recently been developed²⁵². This strategy of simultaneously suppressing the activity of IL-3, GM-CSF and IL-5 (Ref. 253) may be useful in the treatment of inflammatory diseases, as CSL311 decreased mucus production in a model in which human nasal polyps have been transplanted into mice transgenic for the human IL3 and CSF2 genes²⁵⁴.

IL-3 biology: outstanding questions

There have not been many studies that have assessed the role of IL-3 in inflammation, possibly owing to its limited expression, and perhaps because it has historically been viewed as a multi-CSF in myelopoiesis. However, this could be about to change, as there may be relatively few safety issues with targeting IL-3 and the focus of IL-3 research may shift to less well-studied IL-3-responsive cell types, such as mast cells, basophils and pDCs.

Clinical studies targeting IL-3

IL-3Rα is differentially and significantly overexpressed in a large proportion of individuals with acute myeloid leukaemia (AML)²⁵⁵. First-generation mAbs that bound to and blocked IL-3 signalling, but did not mediate antibody-dependent cellular toxicity (ADCC), were safe; nevertheless, they lacked clinical activity²⁵⁶. As a result, JNJ-56022473 (previously known as CSL362) was developed, which is a humanized therapeutic mAb that binds to IL-3Rα and incorporates two mechanisms of action. It inhibits binding of IL-3 to IL-3Rα, thereby antagonizing IL-3 signalling. JNJ-56022473 also enhances ADCC, owing to a mutation in the Fc region that increases the affinity for FcγRIIIa (also known as CD16)²⁵⁷. A phase I trial (NCT01632852) in AML has recently been completed (n = 25)²⁵⁸, and there was no increase in infections, despite rapid, complete and sustained pDC and basophil depletion. A phase II study (NCT02472145) is underway, which is investigating JNJ-56022473 treatment in combination with the DNA methylation inhibitor decitabine. The potential utility of JNJ-56022473 in primary human blood cells that are derived from patients with SLE has also been assessed²⁵⁹. Ex vivo, pDCs and, to a lesser extent, basophils were depleted, but other cell types were not. JNJ-56022473 also inhibited TLR7- and TLR9-stimulated interferon-α (IFNα) production and IFNα-inducible gene expression ex vivo; this observation was also confirmed in vivo in cynomolgus monkeys²⁵⁹. It was concluded that these data provide a preclinical rationale for the planned therapeutic evaluation of JNJ-56022473 in SLE (NCT02920424).

A phase I study (NCT00397579) in patients with blastic plasmacytoid dendritic cell neoplasm (BPDCN) (n = 11) with SL-401, a recombinant fusion protein that is composed of the catalytic and translocation domains of diphtheria toxin and fused to IL-3, has been completed with evidence of some efficacy²⁶⁰.

IL-3 targeting perspectives

In regard to inflammatory diseases, it would not seem unreasonable to target IL-3 and/or IL-3R in conditions that involve mast cells, basophils or pDCs, either by ADCC, or ligand or IL-3R neutralization^261,262. SLE, BPDCN, leprosy type 1 and sepsis would seem to be worth pursuing. Anti-β_c strategies are definitely worthy of consideration, owing to the number of cytokines that are potentially targeted at once, although this approach might not always be necessary or even desirable. Again, additional preclinical data would be informative.

Conclusions and future perspectives

As judged by the updated basic research literature and clinical trial activity outlined in this Review, there is burgeoning interest in targeting CSFs in inflammatory and autoimmune disorders (and cancer). For success in clinical medicine, such targeting will need to be delineated from current therapeutic strategies. Whether and when a particular CSF and its receptor are expressed sufficiently in the appropriate lesion or lesions are important considerations. The uniqueness of the biological roles of the individual CSFs holds promise that targeting CSFs could be beneficial for a diverse range of conditions. As part of this consideration, it is important that any links or 'networks' between the CSFs and other mediators — for example, TNF^3,9,16 — are defined thoroughly. In this context, the fact that the main cellular targets are myeloid populations, but with some additional specificity within this framework, augurs well in our view. It could also turn out that, because there is some cell-type specificity with which CSFs can control myeloid cell numbers, clues as to which CSF should be targeted for a given condition could come from the increased numbers of a particular myeloid population in patients with a particular disease. As a result, additional indications could be forthcoming, particularly if such increases in cell numbers correlate with disease severity⁵⁶.

Patient stratification along these lines may be a useful strategy. Furthermore, it will be important to determine the benefit/risk ratio of targeting CSFs for different stages of immune-mediated inflammatory diseases. For instance, anti-GM-CSF treatment with GSK3196165 is not only being evaluated in established RA but also as an induction regimen in early RA²⁶³. As with most — if not all — anti-inflammatory therapies, the potential for development of a compromised host defence is an issue that must be considered carefully; information on issues ranging from the contribution of endogenous CSFs to the regulation of myelopoiesis continues to be important. The lack of serious adverse events in the 74-week OLE mavrilimumab study⁸ is encouraging in that respect.

It is important that any decrease in cell numbers upon CSF signalling blockade does not affect host defence; the developmental stage at which each CSF acts during inflammation is still being assessed (Fig. 2), so the potential effects on infection are not yet fully understood. It would also be beneficial to conclusively delineate which upstream and downstream components regulate CSF expression and function, including putative cytokine networks^3,9,16,18,55, some of which may involve two or more CSFs.

Although the roles of the CSFs are best understood in myeloid populations, there is evidence that CSF receptors are expressed on non-myeloid cell types: for example, neurons and endothelial cells. The importance of such reported CSF receptor distribution is awaited with interest. It is obviously essential to ensure in any assessment of the data that receptor expression is monitored in well-characterized single-cell assays or on purified and well-characterized populations. The roles of the CSFs in the nervous system and as possible targets in pain are intriguing.

Combinatorial anti-CSF therapies could also be useful. There is some evidence that CSFs can regulate the expression of each other: for example, as reported in human monocytes²⁶⁴ and mouse neutrophils²⁶⁵. As a generalization, the CSFs have distinct biological roles compared with each other and other cytokines, so combination therapies that target more than one CSF or a CSF and another mediator (such as a cytokine)³⁷ may be beneficial. The anti-β_c strategy for targeting potentially GM-CSF, IL-3 and IL-5 should be explored further to see in which clinical situations it may be suitable.