Introduction

The development of experimental medicine from the end of the XIX° century to the present time was the main source of the discoveries in physiology and pathophysiology in most of the fields in medicine [1]. During the same times, the similarities of targets between men and animal allowed therapeutic developments many years before the involved targets and mechanisms were uncovered. Many animal species were used in all the fields of research, depending on availability, reproductive capacities, costs, similarities, or organs. Inbred strains of rats were popular in pharmacology, but guinea pigs, rabbits, dogs, and cats were also largely used. The development of immunology was largely associated with the use of mice: these animals are smaller (about 10 times than a rat) and need less material for pharmacologic issues, allowing to perform experiments in many individuals. Moreover, mice have major advantages: their immune system was rapidly extensively dissected; the reagents were available for many purposes; genotyping grew quickly and genetic variants became available; and genetic manipulation was possible as soon as the 1980s [2]. We will focus this review on models of inflammatory arthritis in mice (Table 1) mainly the models of rheumatoid arthritis (RA). Literature is particularly rich on this field. Even if some controversies are sometimes raised, it appears that human inflammatory diseases are greatly mimicked by mouse models [3].

Table 1 Most commonly used RA models in mice
Full size table

Animal modelizing: exploring novelties or reproducing the human diseases?

An animal model is only an experimental tool to address a question; it mimics a part of a disease and never duplicates or reproduces the entire human pathology. Quantitatively speaking, animal models are extensively used for screening molecules by pharmaceutical companies and sometimes in academic laboratories to test the hypothesis of the global effect of a molecule. At this level of research, what is important is the reproducibility and the reliability of the effect. Effects of non-steroidal anti-inflammatory drugs (NSAIDs) on adjuvant arthritis were known before the mode of action of these molecules was uncovered (anti-prostaglandin effect; then effect on cyclo-oxygenases). Adjuvant arthritis in rats was largely the most popular animal model during 40 years, corresponding to the period of a huge development of NSAIDs. This model, described in a preliminary form by Stoerk but rationalized by Pearson, is a severe polyarthritis induced by intradermal injection of complete Freund’s adjuvant, including killed mycobacteria emulsified in oil [4, 5]. It took 25 years to show that T cells were involved, with a hypothesis of molecular mimicry between hsp65 of mycobacteria and cartilage proteoglycans; the precise role of this mimicry remained quite obscure, and this mechanism was never proven in human arthritis. The model was described only in some strain of rats and is not described in mice (in some laboratories, less than 10% of mice might be sensible to Freund’s adjuvant). For the researcher, this model remains a rat model of extremely severe inflammation depending on T cell activation. Other models of T-dependent inflammation are nevertheless available and better characterized in terms of cellular and molecular involvement and should be preferred. In rat adjuvant arthritis, intense inflammation results in periarticular demineralization that was specifically studied in this model. Direct injection in the paw tissue, although effective in terms of arthritis occurrence, is not recommended because of ethical issues, being an intensely painful procedure.

Humans are not big mice! Thus, there is no model reproducing a human rheumatic disease. That means that therapeutic intervention on models always provides partial information. Reciprocally, it appears useless to test in animals any therapeutics that is already approved in humans. The absence of reproduction of the human diseases does not imply the uselessness of these models: they all mimic a part of the supposed mechanism. For instance in RA, the choice of the model depends on the phase of the disease to be studied. A quite longlasting animal disease (for instance collagen-induced arthritis (CIA)) is necessary if the immune system or the whole T cell response is explored. Only SKG mice have rheumatoid factor. Rapidly induced with inflammation and joint destruction is the K/BxN model, although involving a limited sequence of phenomena (which are nevertheless critical for arthritis). Single pathways are better explored by genetically modified mice: the most popular mice here are the human TNF-transgenic (TTG) mice, a reliable model of destructive inflammation induced by an overexpression of TNFα. The models are used for studying a given mechanism, for testing a therapeutic intervention (usually blockade of a target), or for both purposes, as the scientific design often includes both these aspects.

Animal studies are a hot topic of debate in medical literature. Some argue that animal studies have a poor level of methodology and that positive results in animals are not always observed when clinical trials are conducted. Finally, the conclusions mix all these points, to consider that studies in animal models are unethical, useless, and should be banned from the scientific community and the whole public community. Without delving into the undisputed necessity to consider only the well-designed studies, it is necessary to address the question of how to move up the ranks from a concept (or an idea) to the practical application, in other words, how to proceed from a drug design to its first administration in human. Animal models appear, in the field of the rheumatic diseases, as a cornerstone of biomedical research to test the concepts and their association with others, to excite the mind to find other pathways, to dissect the disease into simpler mechanisms, to test targets, and finally to dare to propose to a patient to be the first (or among the first) to test a drug. Nevertheless, examples are numerous of treatments that have been developed without preclinical experimental design in animal model (first rabbies vaccination as a popular one; corticosteroid administration in RA, which was awarded a Nobel prize). We know also treatments that were not effective in any animal model (gold salts) or treatments that were initially used in patients without experimental preclinical design (rituximab in RA; and TNF blockers in spondyloarthritis).

A paper by Seok et al. [6] enkindled the spark of the polemic in the world of animal models of inflammation, claiming that genomic responses in mouse models poorly mimic human inflammatory diseases; this paper resulted in concerns to keep a major role for animal models whatever the purposes: pathophysiology, molecular or cellular interactions, and treatment evaluation. Two years later, a reevaluation was performed of the same expression datasets used in the previous study, by focusing on genes whose expression had been significantly changed in both mice and humans; results were a complete refutation of the previous findings, showing high correlations between mice and humans [3]. This later paper demonstrated clearly that gene expression patterns in mouse models closely recapitulate those in human inflammatory conditions. These two groups of authors did not evaluate any model of arthritis. What seems important is to perform studies that are conducted with a clear hypothesis, with reproducible models, with well-conducted and replicated experiments. Moreover, the analysis of data should use the same high standards as clinical trials in human; overinterpretation should be chased. Meta-analysis of animal studies should be correctly and extensively performed [7].

Contribution of the experimental models to the understanding of pathophysiological mechanisms of RA

Innate immunity

Although RA is an autoimmune disease (meaning a disease involving T and B lymphocytes leading to autoantibody production), innate immunity also plays a key role in RA development.

Cells and associated cytokines

In addition to their role as first line of defense against invading pathogens or in tissue repair, some innate immune cells are also involved in mounting adaptive immune responses and can moreover be pathogenic in situations like RA. In RA synovial fluid, a pro-inflammatory sub-population of myeloid dendritic cells—typical antigen-presenting cells highly efficient in T cell priming—has been reported [8]. Those cells support Th17 differentiation and secretion of the pro-inflammatory cytokine IL-17A. In mice, dendritic cells activated in vitro by an inflammatory stimulus induce arthritis when injected intra-articularly [9]. Likewise, antigen-presenting cells from rats with pristane-induced arthritis trigger inflammatory arthritis after adoptive transfer, but only after in vitro stimulation [10].

Inflammatory cells are also involved in RA pathogenesis. Among them, macrophages act through the release of diverse cytokines, inflammatory mediators, and enzymes. Particularly, synovial resident macrophages are activated in inflamed RA joints. The level of synovial macrophage infiltration correlates with pain scores in patients [11]. Pro-inflammatory (classically activated) M1 macrophages (secreting TNF and IL-1) are believed to mediate articular destruction, in contrast to M2 macrophage sub-populations. It should be noted that both blocking of TNF and IL-1 is beneficial in CIA [12]. In vitro, culture of RA macrophage precursors with purified RA-specific autoantibodies (ACPA, see below) induces differentiation into M1 macrophages [13]. The role of macrophages has been confirmed in animal models. In K/BxN serum transfer-induced arthritis, a sub-population of blood monocytes (precursors of tissue macrophages), namely, non-classical Ly6G− monocytes, is required for initiation of joint inflammation upon recruitment and differentiation into M1 macrophages [14]. Importantly, reshaping macrophages to decrease the M1 population lowers Th17 cell count and suppresses CIA [15]. Actually, the IL-1R2 decoy receptor on macrophages is pivotal in controlling CIA [16]. Blocking granulocyte macrophage colony-stimulating factor (GM-CSF), a cytokine involved in macrophage differentiation and associated with RA, also inhibits CIA development [17]. Likewise, in vivo inhibition of macrophage colony-stimulating factor (M-CSF) protects from CIA progression [18], supporting a role not only for macrophages but also for osteoclasts (see below) in the disease.

RA is an inflammatory disease, but surprisingly the typical pro-inflammatory cells, i.e., polymorphonuclear neutrophils (PMN), have not been deeply investigated, until recently with the discovery of NETosis (see below). The reason is probably because PMN have for a long time been considered as basic cells; however, we know now they are much more complex than believed. PMN have an in vivo lifespan of 5.4 days in humans [19], represent a heterogeneous population, with blood vs. tissue sub-populations, have immunoregulatory functions, like B cell-helper PMN via BLyS production [20] vs. regulatory PMN via IL-10 secretion [21], secrete IFN-α [22, 23], produce (TNF, IL-8) or respond to several cytokines/chemokines (TNF, IL-6, IL-8, GM-CSF) associated with RA, and are activated in RA, recruited in inflamed joints, and highly represented in RA synovial fluid. PMN have non-classical functions in RA, like differentiation in dendritic cells [24] or receptor activator of nuclear factor kappa-B ligand (RANKL) expression [25]. PMN were also analyzed in arthritis models. In mice, PMN depletion [26] or blockade of the PMN cell surface marker Ly6G [27] reduces collagen antibody-induced arthritis (CAIA) or K/BxN serum transfer-induced arthritis, respectively. Moreover, inflammatory arthritis development is impaired in granulocyte colony-stimulating factor (G-CSF)-KO mice or after G-CSF blocking in different models [28]. G-CSF is involved both in PMN production and in recruitment to inflamed joint in CIA [29]; therapeutic blockade of the G-CSF receptor lowers PMN infiltration and reduces established CAIA [30]. In the K/BxN serum transfer model of arthritis, PMN recruitment in joints and subsequent delivery of IL-1 and arthritis initiation are dependent on leukotriene B4 [31]. In agreement with those data, human IL-1α-transgenic mice develop polyarthritis with joint accumulation of active PMN [32]. Moreover and as mentioned above, IL-1R2 decoy receptor deficiency exacerbates K/BxN serum transfer-induced arthritis in mice, without affecting systemic inflammation; interestingly, PMN strongly express IL-1R2 and regulate fibroblasts, as IL-1R2-deficient PMN increase the response of normal fibroblasts to IL-1 [33]. However, the role of PMN plasticity and PMN regulatory functions will have to be clarified in the future. Recently, special attention has been paid to PMN due to their ability to produce neutrophil extracellular traps (NET) upon activation (NETosis). NET are chromatin fragments expelled by activated PMN, originally described as an innate immune mechanism against bacteria [34]. However, NET might be pathogenic in certain circumstances. For instance, NETosis depends on citrullination [35] (a post-translational modification of proteins mediated by peptidylarginine deiminase (PAD) enzymes) and might therefore be involved in triggering the production of the RA-specific anti-citrullinated protein antibodies (ACPA, see below) via autoantigen citrullination. NETosis is increased in RA patients [36]; some RA autoantibodies recognize NET [37], and active PAD is released in RA synovial fluid [38]. Interestingly, in vivo preventive or therapeutic PAD inhibition reduces NETosis [39] as well as CIA, but not CAIA, severity [40, 41], supporting a role of NETosis in autoimmunization; likewise, PAD4-deficient mice develop inflammatory arthritis in the K/BxN serum transfer model [42], suggesting that citrullination/PAD4 is dispensable in effector phases. Nevertheless, the crucial role of NETosis in arthritis development, its consequences, the mechanisms triggered, and especially the physiopathological stimuli involved require further investigation. One possibility is that PMN are activated by endogenous ligands or damage-associated molecular patterns (DAMP), leading to sterile inflammation. For instance, we have previously demonstrated that extracellular chromatin (which is present in RA joints [43, 44]) triggers PMN activation [45, 46] as well as dendritic cell maturation [47].

Among bone cells, osteoclasts are also actively involved in RA joint destruction via bone resorption [48]. Interestingly, TNF and IL-17A are known to activate osteoclasts whereas regulatory T cells (Treg) suppress osteoclast formation [49]. M-CSF and RANKL participate in promoting osteoclast differentiation from monocytes (osteoclastogenesis) and maturation. Importantly, RA synovial fibroblasts and B lymphocytes express RANKL [50, 51] and ACPA induce osteoclastogenesis [52]. The CAIA and CIA mouse models have demonstrated that RANKL expression by synovial fibroblasts is important for osteoclast formation and bone erosion [53]. IL-6 induces RANKL expression by RA fibroblasts under the control of the SOX5 transcription factor, and in CIA mice, silencing SOX5 expression in joints lowers RANKL expression and bone erosion [54]. Moreover, in vivo inhibition of Janus kinase signaling decreases RANKL production and reduces bone resorption in adjuvant-induced arthritis in rats [55]. Surprisingly, in the SKG model, osteoclast precursors with T cell suppressive activity are able to impair arthritis development [56].

Animal models also helped in understanding the role of natural killer (NK) cells in RA pathophysiology. NK cells are present in inflamed RA joints, express RANKL and M-CSF, and trigger differentiation of monocytes into dendritic cells [57] and osteoclasts. Actually, depletion of NK cells impairs CIA development [58]. The activating NK receptor NKG2D and its ligands are expressed in paws of CIA mice, and blocking NKG2D ameliorates the disease [59]. Interestingly, NKG2D is also expressed on autoreactive CD4+ T lymphocytes of RA patients in response to TNF and is involved in their stimulation by RA synoviocytes expressing NKG2D ligands [60].

Among the latter, RA synovial fibroblasts are key players in joint destruction through production of matrix-degrading enzymes, cytokines, and chemokines, like CXCL12 which supports migration of CD4+ and CD8+ T lymphocytes to the synovial tissue [61]. Actually, RA synovial fibroblasts have been shown to participate to disease progression and spreading via migration to unaffected joints in SCID mice [62].

Innate sensors

Several cellular receptors as well as plasma receptors are involved in RA pathogenicity and have been studied in animal models.

Among cellular receptors, toll-like receptors (TLR) have been particularly investigated. They recognize pathogen-associated molecular patterns (PAMP) and might therefore link infections to RA induction. Indeed, although RA etiology is still unknown, environmental factors are known to be involved. However, the ligands recognized are still mostly unknown since those receptors may recognize endogenous ligands in addition to PAMP.

RA synovial fibroblasts express TLR2, 3, 4, and 9 [63]. Particularly, RA synovium strongly expresses TLR3 [50]. Likewise, TLR3 is upregulated in synovial fibroblasts during pristane-induced arthritis in rats and local TLR3 stimulation aggravates the disease [64]. RA joint macrophages express high levels of TLR2 and 4, and ligation of those receptors triggers a greater IL-8 secretion than on normal macrophages [65]. Accordingly, TLR4 inhibition in vivo reduces arthritis in CIA- and IL-1Ra-deficient mice, probably through inhibition of dendritic cell maturation and of IL-1β expression [66]. The use of TLR4-deficient mice confirmed that TLR4 is involved in CIA development and IL-17 production [67] and in the K/BxN serum transfer model [68]. One potential endogenous pro-inflammatory TLR4 ligand is HMGB1, which is present at high concentrations in RA synovial fluid [69]. In contrast, studies in IL-1Ra-deficient mice suggest TLR2 is protective against inflammatory arthritis [70]. Less works focused on TLR7. Nevertheless, inducing TLR7 hyporesponsiveness reduces K/BxN serum-induced arthritis [71], a result confirmed in the CIA model using TLR7-deficient mice [72]. A natural endogenous TLR7 ligand (microRNA let-7b) has been recently identified in the synovial fluid of RA patients and is able to provoke arthritis in mice when injected intra-articularly [73]. TLR9 is expressed at higher levels in RA monocytes, especially in the synovial fluid [74]. TLR9 was first described as recognizing bacterial DNA [75]. Intra-articular injection of bacterial DNA triggers arthritis in mice [76]. However, we know now that TLR9 might also recognize self DNA, that other receptors recognize DNA, and that TLR9 recognizes non-DNA ligands. Moreover, the control of TLR9 activation might be more complex than expected. We demonstrated that, in addition to endosomal TLR9, this receptor is also expressed in an active form at the cell surface [77]. Importantly, the role of this receptor in inflammatory arthritis has not been investigated in TLR9-deficient CIA mice so far. However, in vivo inhibition of cathepsin K leads to defective TLR9 signaling and protects from adjuvant-induced arthritis in rats [78]. In addition to membrane TLR3/7/8/9, cytosolic helicases MDA5 and RIG-I are also involved in nucleic acid sensing. They are both expressed in RA synovial fibroblasts and upregulated when activated, leading to IFN-β, IL-8, and MMP3 production [79]. It should be noted that triggering type I IFN production by RNA can either induce [80] or suppress arthritis [81] in different models. Interestingly, when macrophages are unable to digest engulfed self DNA as a consequence of defective DNase II, mice spontaneously develop an RA-like disease [82]. Of note, another nuclease, Trex-1, is deficient in RA synovial fibroblasts [83].

Because autoantibodies, and especially immune complexes (IC), are involved in RA pathogenesis, Fc receptors (FcR) are also key cellular receptors. RA IC trigger macrophage TNF production in an FcγR- (especially FcγRIIa [84]) and TLR4-dependent manner, when IC contain citrullinated fibrinogen [85]. Actually, blocking human FcγRIIa in transgenic mice inhibits CAIA [86]. In CIA mice and K/BxN serum-induced arthritis, all activating FcγR (RI, RIII, RIV) are involved in arthritis development, at different levels, but arthritis can occur in the absence of FcγRI and RIII [87]. More precisely, FcγRIV expression by osteoclasts is critical for bone erosion in the K/BxN serum transfer model [88]. In contrast to other FcγR, FcγRIIb negatively regulates IC signaling and animal models have shown that this receptor controls CIA as mice which fail to upregulate FcγRIIb develop a more severe disease [89].

Some plasma immune sensors are also involved in RA. Among them, proteins from the complement system have been particularly studied. Indeed, IC are known to be recognized by the C1q complement protein, activating the complement cascade and downstream effector mechanisms. Several C1q polymorphisms are associated with RA and with C1q serum concentrations in healthy donors [90]. Surprisingly, ACPA activate not only the classical but also the alternative complement pathways in vitro [91]. Studies in the passive K/BxN serum-induced arthritis and CAIA models have shown that C1q is not required for arthritis, in contrast to the C5 downstream effector [92, 93]. Nevertheless, C1q has not been evaluated in the active CIA model (in which autoimmunization is required) and the numerous additional non-classical functions of C1q should be taken into account. All complement pathways converge to C3 activation, leading to production of the C3a and C5a anaphylatoxins. The latter are not only chemoattractants but also T cell costimulatory molecules [94]. They are both present in RA synovial fluid [95]. In K/BxN serum-induced arthritis, C5a receptor and FcγR are involved in PMN recruitment [96]. In transgenic mice for human FcγRIIa, the latter receptor is crucial for arthritis development in this model where PMN recruitment and activation are dependent on the C5a receptor, demonstrating a crosstalk between those two innate immune receptors [97]. In vivo blocking of C5a [98] or the C5a receptor [99] reduces CIA severity. Importance of the complement system was confirmed in mice deficient for properdin, which normally amplifies activation of the alternative pathway, where less severe CAIA develops in association with impaired osteoclast differentiation [100].

Adaptive immunity

B lymphocytes and autoantibodies

Because RA is associated with autoantibody production, B cells are key players. This was confirmed by showing that B cell-deficient mice do not develop CIA [101]. B lymphocytes work in addition as antigen-presenting cells and secrete cytokines. Indeed, RA synovial fluid B cells strongly express several RA-associated cytokines, such as IL-1, IL-6, and TNF, and moreover produce RANKL, suggesting a role in osteoclastogenesis [51].

B lymphocytes are activated in RA and produce several autoantibodies, some of which are detectable decades before disease clinical onset. The CAIA model by itself demonstrates the role of antibodies in arthritis development. ACPA are detected in two thirds of patients; they specifically recognize citrullinated antigens [102, 103] and have a diagnostic value. Some of them are pathogenic. In RA patients, two classical targets of ACPA are citrullinated fibrinogen as well as citrullinated vimentin. Citrullinated fibrinogen-containing IC are particularly efficient in inducing TNF secretion by M-CSF-polarized macrophages [104]. Moreover, anti-citrullinated vimentin autoantibodies bind to osteoclasts, enhancing their bone-resorptive activity and TNF secretion [105]. ACPA purified from RA patients induce the differentiation of human osteoclasts in an IL-8-dependent manner, and transfer of ACPA into mice triggers bone loss in vivo [52]. The role of antibodies against citrullinated proteins is less clear in animal models. Such antibodies develop in CIA over time after epitope spreading [106] and tolerization with citrullinated peptides reduces disease severity and incidence [107]. Moreover, citrullinated type II collagen-specific monoclonal antibodies bind synovial tissue and are arthritogenic in mice [108]. However, another study has shown that although PAD4 is expressed in inflamed synovial tissue in the CIA and SCW models and that some synovial proteins are citrullinated, no anti-citrullinated protein antibodies were detected [109]. Likewise, even if some immunization protocols may lead to the production of anti-citrullinated protein antibodies in mice, those antibodies fail to induce or amplify CAIA and CIA [110]; similar results were obtained in rats [111]. Nevertheless, whatever the mechanism involved, the pathogenic role of citrullination has been confirmed recently by demonstrating the beneficial effect of an immunotherapy based on tolerization towards citrullinated peptides in RA patients [112]. And recently, a new antibody population has been described in RA: a subset of anti-PAD4 autoantibodies that activate PAD4 and are associated with erosive disease [113]. Anti-PAD4 might therefore amplify ACPA response, but no data is available on those antibodies in arthritis models so far. Although rheumatoid factors (RF), autoantibodies against the Fc portion of immunoglobulins, are often present in RA patients, nearly no data exist in animal models. RF are usually not detected in mouse CIA, but one study has shown they are produced in a modified CIA model in rats [114]. RF are however potentially pathogenic in RA as they amplify macrophage activation [115] and complement activation [116] in response to ACPA IC. Finally, anti-carbamylated protein antibodies have been identified in RA [117]; their concentration is increased in presymptomatic individuals, their presence is related to radiological destruction, and they are detected in a fraction of ACPA-negative RA patients [118]. Those antibodies could also be detected in active models of arthritis [119], especially in mouse CIA where they precede clinical signs [120]. However, in most cases, the stimuli and the autoantigens leading to autoantibody production in RA are still unknown.

On the opposite, IL-10-producing B lymphocytes are known as regulatory B cells (Breg). Importance of Breg was demonstrated in the methylated bovine serum albumin (mBSA)-induced arthritis model, in which mice lacking Breg develop a more severe disease with more Th1/Th17 cells but less Treg [121]. Actually, Breg are decreased in RA patients and Breg counts inversely correlate with disease activity [122].

Involvement of T lymphocytes

The synovium of RA patients contains CD4+ T cells, and the association between RA susceptibility and HLA-DRB1 alleles on MHCII indicates the key role of those cells in the initiation and the continuation of the disease. Antigen-presenting cells such as dendritic cells, macrophages, and B cells first activate T cells that in turn secrete various cytokines. This leads to a subsequent activation of innate immune cells and B cells and eventually ends in synovial fibroblast, chondrocyte, and osteoclast activation.

At the initiation of the process, it is thought that autoimmunity in RA is characterized by defects in the normal process of positive and negative selection of T cells in the thymus, thus generating the escape of autoreactive T cells into the peripheral repertoire.

To what extent can animal models be helpful to highlight the role of T cells in RA? CD4+ T cell depletion before or after arthritis development leads to partial or complete inhibition of arthritis development in some, but not all, models of RA. In CIA, CD4-depleting antibodies suppress the disease when administered before, but not after, arthritis development [123, 124], as in proteoglycan-induced arthritis (PGIA) [125] and in SKG mice [126]. Resolution of clinically established arthritis is possible in glucose-6-phosphate isomerase (G6PI)-induced arthritis in DBA/1 [127] with CD4 depletion, as in the SCW model [128]. Other direct evidence of CD4+ T cell involvement in arthritis models is also provided by disease induction by CD4+ T cell transfer experiments, for instance in SCID mice receiving CD4+ T cells from SKG mice [129]. Eventually, K/BxN spontaneous arthritis depends on the generation of CD4+ T cells specific for the systemically distributed self-protein G6PI.

RA is highly characterized by the presence of antibodies against self IgG (RF) and against citrullinated proteins (ACPA). Their production by B cells requires T cell help, but specific antigens of the arthritogenic T cells have not been identified yet. Animal models of RA help to identify T cell specificity in RA and to show that antigen-specific T cells are necessary to mediate B cell expansion and function in arthritis development. This feature is illustrated by the transient nature of serum or antibody-induced arthritis such as CAIA, while it becomes chronic and progressive in the presence of autoantigen-specific T cells [130].

The role of T lymphocytes in RA is not black or white, since CD4+ subsets (Th1, Th2, Th9, Th17, Th22, Treg, and TFH) differ not only in their cytokine secretion pattern and their extra- and intra-cellular markers [131] but also in their epigenetic programs and metabolic pathways. Moreover, these T cell subsets are not all terminally differentiated cells, but are generally plastic since one subset can switch into another one, depending on the cytokine milieu and the nature and intensity of antigen stimulation [132]. RA has long been considered a Th1-mediated disease. However, studies in animal models of RA using either recombinant IFNγ administration or blocking antibodies or mice deficient in IFNγ or IFN receptor led to various conclusions about the role of IFNγ in arthritis development. It depended on various parameters, notably the timing of IFNγ overexpression or blockade [133, 134]. Moreover, no direct therapy targeted against Th1 cells or Th1 cytokines has proven to be successful in RA. In contrast, while the deleterious role of Th1 cells has progressively become ambiguous, the role of Th17 cells in RA pathogenesis is now clearly established. There is an increase in IL-17A-expressing Th17 cell subset in the peripheral blood of patients with early RA compared with healthy controls, with an association with disease activity [135]. Moreover, RA synovial tissue produces bioactive IL-17 in vitro and IL-17-positive cells are identified in RA synovial biopsies. This Th17 response is specifically promoted by activated monocytes from the site of inflammation in humans [136].

Animal models of RA have been particularly helpful to demonstrate the crucial role of Th17 cells in the pathogenesis of arthritis. This was first confirmed in the CIA model where IL-17 mRNA levels in the joints increased with disease progression. Moreover, the disease was aggravated by systemic IL-17 gene transfer and attenuated with a soluble IL-17 receptor-Fc treatment [137] or with a neutralizing anti-murine IL-17 monoclonal antibody (mAb) [138]. CIA was also markedly, but not fully, decreased in mice genetically lacking IL-17A, mostly because IL-17A is necessary for the priming of collagen-specific T cells and for IgG2a production [139]. The incomplete inhibition of arthritis in IL-17-deficient CIA mice suggests that other factors contribute to the pathogenesis of CIA. Indeed, CIA in mice deficient for both IL-17 and IL-1 was completely suppressed, showing the role of IL-1 signaling to generate a pathogenic pool of Th17 cells [140]. The interplay between IL-17 and IL-1 was also demonstrated in SCW-induced arthritis where IL-17R deficiency leads to impaired synovial expression of IL-1 and thus to prevention of cartilage destruction [141]. IL-17-producing KRN T cells are present in inflamed joints of K/BxN mice, but Jacobs et al. showed that in this antibody-dependent model, IL-17 only serves as an adjunct amplifier of antibody production [142]. However, Auger et al. showed that IL-17 was not necessary for arthritis development by using a modified model of K/BxN arthritis in mice that express the KRN TCR transgene and the MHC class II allele H-2k (Ag7) which presents endogenous glucose-6-phosphate isomerase (G6PI). In these mice, the deficiency in IL-17A or in Rorγt, the Th17-differentiating transcription factor, did not modify arthritis development [143]. Similarly, PGIA in mice genetically deficient in IL-17 also developed normally [144]. These data must be addressed in the evaluation of the relevance of therapies targeting IL-17 in inflammatory diseases.

Interestingly, a Th17 subset, called Th1/17, producing both IFNγ and IL-17 and requiring IL-23+IL-1 for its differentiation, has been identified in peripheral blood mononuclear cells (PBMC) from early RA patients [145]. Moreover, supernatant from Th1/17 cells strongly activates RA synoviocytes [146], suggesting that this subset could play a determining role in disease development. However, no study of Th1/17 subset in experimental models has been described so far.

IL-23 is the key cytokine promoting Th17 differentiation, and the IL-23-IL-17 axis is thought to have a critical role in RA initiation. IL-23 is composed of two subunits, p40 and p19, and IL-23p19-deficient mice are resistant to CIA, with a correlated Th17, but not Th1 cell, decrease [147]. The deleterious role of IL-23 was also demonstrated in rat CIA, as the disease was attenuated by an anti-IL-23 antibody, with a reduced bone destruction [148]. In another work, active immunization against IL-23p19 improves experimental arthritis [149]. Interestingly, the same study by Yago et al. showed the role of IL-23 in a human system, as IL-23 also induced osteoclastogenesis in cultures of PBMC from healthy donors. These data have to be carefully considered since another study showed that CIA inhibition by IL-23p19 neutralization depended on the stage of treatment, with an anti-arthritic effect when the IL-23p19 antibody was administered before arthritis onset, but no effect on established disease [150]. Moreover, lack of IL-23 does not prevent the onset of joint inflammation but can stop the progression to a destructive synovitis in a model of mBSA-induced arthritis in mice [151]. IL-23 is also a chemoattractant factor since it can induce neutrophil migration in the joint of mice with mBSA-induced arthritis [152]. If numerous studies have been undertaken to evaluate the role of IL-23 in animal models of RA, only a few studies have been published on RA patients. However, biological agents targeting either the p19 subunit of IL-23 or the common p40 subunit of IL-23 and IL-12 are currently used in clinical trials [153].

In chronic inflammatory autoimmune diseases such as RA, a very simple scheme can present the pathology as a result of T cell subset imbalance, positioning Th1 and Th17 on one side, and Th2 and Tregs on the other. Th2 cells mostly secrete IL-4 and IL-13, and analysis of the levels of IFN-γ and IL-4 in the synovial fluid and serum of RA patients indicates that Th1 activity is clearly predominant over Th2 activity. IL-4 protective effect was demonstrated in animal models of RA in numerous studies, for instance by administrating cells engineered to secrete this cytokine or gene therapy viral vectors in mice with CIA [154,155,156,157] or PGIA [158]. Most of these studies used preventive treatments regimens; however, prophylactic treatment with anti-IL-4 mAbs was able to significantly reduce the incidence and clinical scores of CIA [159]. More significantly, arthritis depends critically on IL-4 in the K/BxN model, since it has to be present in the development stage of arthritis, but not the effector one [160]. Pregnancy, which is considered a Th2 condition, is frequently accompanied with an improvement of RA symptoms [161]. Taken together, these data illustrate the fragility of animal models and experimental settings, giving different to opposite results among the studies aiming at showing the role of a given cytokine. Even if the experimental models have helped to understand the T cell subset network in RA, no therapy directly targeting a Th2 cytokine has been undertaken in this disease.

Follicular helper T cells (TFH), a distinct subset of CD4+ T cells specialized in providing help to B cells to differentiate into plasma cells, express high levels of CXC chemokine receptor type 5 (CXCR5), programmed death-1 (PD-1), inducible T cell co-stimulator (ICOS), and the regulator transcription factor B cell lymphoma 6 (Bcl6). Moreover, they are characterized by secretion of high levels of IL-21. In RA patients, IL-21 plasma levels correlate with disease activity and radiological progression, and IL-21-producing TFH cells are increased in the blood and synovial fluid (SF) [162]. These descriptive data have been highlighted by studies in experimental models of RA. For instance, IL-21-deficient mice are resistant to CIA, because B cell—and not T cell—response is altered, thus preventing the production of anti-collagen type II (CII) IgG [163]. These findings support the idea that IL-21 and TFH cells are linked to the development and perpetuation of RA.

More than 20 years ago, a fundamental discovery by Sakaguchi and colleagues led to define Tregs as a subset of CD4+ T lymphocytes expressing high amounts of CD25 and capable of suppressing disease in experimental models of autoimmunity [164]. Foxp3 was defined a few years later as the master transcription factor of Tregs [165]. Although it is not known whether Treg deficiency contributes to the pathogenesis of RA, compromised Treg activity in both human disease and animal models has been demonstrated. The proportion of Tregs in the synovial fluid of RA patients is increased [166], probably as a consequence of increased migration of Tregs to the synovial fluid in response to inflammation. However, this increase is greatly outnumbered by effector T cells [167], resulting in a balance between Tregs and effector T cell shift in this autoimmune microenvironment. Moreover, Tregs are functionally defective in patients with RA, being unable to suppress Th1 responses [168] and B cells [169]. Animal models helped to highlight the role of Tregs in RA. In CIA, Treg cells are found in the joints, synovial fluid, and draining lymph nodes of arthritic mice [170]. Adoptive transfer of CD4+CD25+ cells to arthritic animals delays disease progression and collagen-specific T and B cell responses [170], and CD25+ cell depletion before immunization accelerates the disease. In the G6PI-induced model, depletion of Tregs using anti-CD25 mAb results in a longer disease duration [171]. Likewise, depletion of CD25+ cells in mBSA-induced arthritis results in higher antigen-specific responses and an exacerbated disease which is ameliorated after CD4+CD25+ cell transfer [172]. Treg frequencies are decreased during the initiation phase of TNF-α transgenic mouse arthritis, but a progressive increase occurs during arthritis development, probably as a defense mechanism remaining insufficient to resolve the disease [173]. In this study, anti-TNF treatment led to a restoration of Treg number and functions. In this context, significant interest has been paid to the actions of the pro-inflammatory cytokine TNF to understand the relationship between Tregs and inflammation. Several studies could show that, in RA patients, therapeutic TNF blockade with mAb restores the potency of Treg cell suppression [174] by binding to membrane TNF on monocytes and promoting Treg cell expansion through enhanced TNF-RII signaling [175]. Other targeted therapies such as anti-IL-6 receptor antibody (tocilizumab) also restore Treg number and activation, both in CIA and in RA patients [176].

Recent studies have shown a high degree of plasticity of Treg cells, since they can lose Foxp3 expression and exert effector-like functions under inflammatory conditions [177]. Exposure to IL-6 can for instance downregulate FoxP3 expression and turn them towards a Th17-like cell phenotype [178]. Findings of this type have led to distinguish stable and unstable Treg with distinct FoxP3 epigenetic control. At the gene transcription level, IL-6 can indeed promote DNA methylation in Tregs, thus preventing transcription of the Foxp3 gene [179]. Studies on Foxp3 epigenetic control in experimental arthritis may help to delineate the consequences of Treg instability during chronic inflammation, but these kinds of studies are, at present, lacking. Nevertheless, whether plasticity in human and mouse Tregs is comparable needs to be confirmed.

Angiogenesis

In RA, synovial inflammation and bone and cartilage destruction are sustained by the overproduction of pro-angiogenic factors fostering elevated transendothelial leukocyte infiltration into the inflamed joint [180]. Angiogenesis has also been highlighted in experimental model of RA. Angiogenesis evolves in parallel with joint inflammation during the whole course of CIA [181]. The role of vascular endothelial growth factor (VEGF), the main pro-angiogenic factor, has also been shown in RA and its experimental models. Synovial cells can produce high levels of VEGF [182], and in patients with active RA, an increased circulating VEGF level has been shown, which correlated with disease activity and with radiographic changes [183]. Exogenous VEGF can aggravate CIA, but preventive treatment with the anti-VEGF antibody delays the disease onset, joint swelling, and vascularization in this model [184]. Interestingly, vaccination against VEGF, leading to the production of anti-VEGF polyclonal antibodies, has a significant anti-inflammatory effect in CIA [185]. Hypoxia-inducible factor (HIF)-1α also accumulates in RA joint as a consequence of hypoxia induced by the metabolic demand of the increasing number of recruited leukocytes. Its overexpression enhances the expansion of inflammatory Th1 and Th17 cells [186], thus promoting macrophage activation and tissue destruction. The role of this factor was demonstrated in a K/BxN serum transfer model of arthritis in a conditional knockout of HIF-1α in myeloid cells. In this model, an attenuation of arthritis was shown to be associated with a decrease in myeloid cell homing [187]. Interestingly, some studies show that some approved targeted therapies for RA, including TNF-α and IL-6 inhibitors, may function in part through inhibition of angiogenesis. However, despite the obvious importance of angiogenesis in RA, treatments directly targeting angiogenic factors have been evaluated in preclinical models but could not be translated into successful treatment strategies in RA. This could be due to the complex interaction between pro-angiogenic and pro-inflammatory pathways, and it suggests that the use of combined anti-angiogenic and anti-inflammatory agents may be effective in blocking the RA process.

Environmental factors

There is some evidence that exposure to certain environmental factors increases the risk of developing RA. Studies have identified numerous candidates, including smoking and, more recently, microbiota composition.

Smoke

A significant association was found between smoking and the presence of ACPA in a large number of studies [188], and the association between RA and smoking is stronger in men and in patients positive for ACPAs and expressing the shared HLA epitope (HLA-DRB1) [189, 190]. The first evidence of an association between RA and smoking was suggested in 1987 [191] in RA patients, but experimental models helped to delineate the cellular and molecular mechanisms underlying this association. For instance, PAD enzyme expression was shown to be increased by cigarette smoke exposure with enhanced immune responses to citrullinated collagen and vimentin in HLA-DQ8 or HLA-DR4 transgenic mice [192]. However, smoke exposure has various effects on experimental arthritis, since for instance it aggravates CIA [193] but has no effect on arthritis in TNFΔARE mice [194].

Microbiota

It is now generally assumed that commensal bacteria contribute to immune homeostasis, whereas immune reaction against intestinal flora is a pathological sign. A modified gut and oral microbiome was highlighted in RA patients. Depending on the studies, lower abundance of common commensals [195] and decreased [196] or, in contrast, increased [197] gut microbial diversity were found in RA patients compared to healthy controls. Moreover, a systematic alteration in the gut, dental, and saliva microbiota was shown in a large cohort of RA patients compared to healthy controls [198]. The role of the gut microbiome in arthritis progression has been subject to several studies in experimental models. In particular, a recent report has shown an aggravation of arthritis in RA microbiota-colonized SKG mice [199]. Also interestingly, gut commensal segmented filamentous bacteria can induce strong gut as well as systemic TFH cell responses, leading to the exacerbation of auto-Ab production in K/BxN mice [200].

The underlying mechanistic relationships among gut microbiota, immune system, and RA are still under intense investigations. Multiple potential molecular mechanisms might be involved, for example through Th17 cell differentiation [201]. A break of tolerance due to abnormal citrullination of self-antigens induced by bacteria species at the mucosal level is an alternative hypothesis.

Relevance to treatment targeting

One of the major interests of using animal models is to seek for new therapeutic targets and test new drugs. A new drug is typically a targeted treatment (for instance a mAb); sometimes, this new drug has a likely complex mode of action. In all cases, it is mandatory for a reliable testing to use a relevant model for the proof of concept. Several situations may be described, depending on the degree of previous knowledge of the mode of action of the treatment.

The target is known and a precise relevant model is available

The best example of this opportunity was the development of TNF inhibitors in humans. TNFα was identified as a potential therapeutic target in several steps. Culture of synovial cells revealed abnormal prolongation of TNFα secretion. Ex vivo use of anti-TNFα mAb showed a true TNFα-dependent cascade of production of inflammatory cytokines. As assessed by the leaders of the seminal project themselves, the use of an animal model was determinant to support the relevance of TNFα as a therapeutic target: an amelioration of collagen arthritis was observed with a hamster-anti-mouse TNFα antibody administered after disease onset [202]. Another important confirmation came with another model: mice transgenic for human TNFα were demonstrated as severely involved by a destructive polyarthritis with systemic inflammation and precocious mortality; when treated with infliximab, a chimeric anti-human TNFα mAb, arthritis was significantly improved and mice survived [203]. Clinical trials started in 1992, providing evidence that TNFα blockade in humans—which had previously failed to show any beneficial effect in septic shock—might be beneficial in RA [204]. TNF inhibitors are until now the leaders of the therapeutic market. Even after the clinical approval of the TNFα inhibitors, the use of animal models can still provide information about some major actions of these drugs, as stated above regarding the role of Tregs [173, 174, 205]. The development of anti-TNFα vaccine also used this strategy; when the human vaccine was studied, TTG mice were a relevant model to appreciate its usefulness [206] that was confirmed in phase 2 but not in phase 3 trials to date. By contrast, when interactions between the vaccine-induced polyclonal antibodies and the defenses against infections were studied, the TTG model appeared as not relevant since a persistent secretion of murine TNFα was detected.

The target is known and a general relevant model is available

In many occurrences, the therapeutic target seems evident but animals overexpressing this target (or receiving doses of the protein) do not develop arthritis or any phenotype that should prove a clinical relevance of a potentially active treatment. Examples are IL-1-receptor antagonist (IL-1Ra, developed after engineering as anakinra), CTLA-4 (developed as abatacept), tofacitinib, and other jakinibs. In such cases, the use of a general model of arthritis should be relevant to show a modulatory effect on joint inflammation; these three families of molecules taken as examples were effective in CIA.

The mode of action of the treatment is uncertain

In some cases, namely, in the past, the mode of action of the treatments was poorly known. In most cases, in addition, it is necessary to add some knowledge to a supposed known treatment. Methotrexate, as a classical immunosuppressant, was approved in several rheumatic diseases including RA or spondyloarthritis, independently of any effect on experimental arthritis. Methotrexate is effective in several experimental models, only if high doses (in comparison to humans) are used. Experiments were recently conducted to study the role of CD39-positive Tregs in the action of methotrexate [207]. Gold salts that have been used for 60 years in the last century to treat RA are poorly effective in animal models. Their mode of action, which was never elucidated, is still a matter of debate.

Predictability of an experiment in animal models

We have to claim once more that modeling in animals does not mean reproducing the human disease in animals. Observing an anti-arthritic effect is comparable to drive through cross roads at the green light. It is a positive step, rarely a starting point and never goal by itself. During the course of a pre-clinical development, it is a marker of optimism for further development, a marker for go-no go discussions, a piece of the proof of concept (actually, it represents an animal proof of concept). There are myriads of examples of drugs that were effective in one or several animal models and were either never developed or abandoned for any reason: cost, concurrence, toxicity in animals or humans, poor effect during phase 2–3 development, legal concerns, etc. Many protein kinase p38 inhibitors were particularly effective in murine CIA and failed to demonstrate benefit in phase 1/2 clinical trials, since unanticipated toxicity in some organs was observed at low doses precluding any clinical effect [208, 209]. Another example are the inhibitors of matrix metalloproteinases, which are effective in models of arthritis and do not show a successful development in humans. There are also some drugs that were considered as efficient in a given model and were developed for indications that were not predicted by the model: anakinra, anti-IL-1 mAb, and anti-IL-17 mAb are recent examples. In other cases, one should be careful to predict the uselessness of a drug: we mentioned for instance that the development of rituximab, an anti-CD20 mAb, was the result of serendipity. Monoclonal antibodies directed to the murine CD20 deplete B cells very efficiently, but fail to suppress humoral response to collagen and the development of arthritis; conversely, another anti-B cell mAb, anti-CD79b, inhibits CIA—this inhibition is observed in preventive treatment if mAb is a deglycosylated variant, in late treatment (after arthritis onset) if the fully glycosylated variant is used [210]: this clearly means that the immune regulation of arthritis is different in mice and humans and that the target is not the only major parameter to be considered.

Technical issues

Despite the usefulness of RA models, it is frequently difficult for the novice investigators to establish these models in the laboratory and to acquire the expertise to evaluate the associated pathology.

Spontaneous models, such as TNFα transgenic mice, are of course easy to use, and clinical signs of disease occur usually 5–6 weeks after birth. In contrast, antigen-induced models are certainly the most difficult task, due not only to technical skills but also to environmental factors. The example of collagen-induced arthritis is particularly illustrative. Originally described in DBA/1 (H-2q) strains of mice [211], CIA is induced by immunization with bovine CII (bCII) emulsified in complete Freund’s adjuvant (CFA). The best arthritis incidence is obtained if the emulsion is correctly performed, with a consistency of dense whipped cream and should not disperse quickly when a droplet of emulsion is placed on the surface of water. Collagen preparation must also be adequate. It is a fibrous insoluble protein that must be solubilized and stored in a diluted solution of acetic acid. If soluble CII comes into contact with salts, it may precipitate. The route of injection of bCII/CFA is also crucial. An intradermal immunization offers the best way to get an immune response against bCII. The best injection site is about 1 to 2 cm distal from the base of the tail (in a tissue site and not a vessel) with a noticeable tissue resistance to the injection. If the injection is rapid and easy, it can indicate a subcutaneous or deep tissue injection and can result in a poor incidence of disease. A limit of CIA in DBA/1 mice is that this strain is not currently used to generate gene-knockout mice. CIA was thus developed in C57BL/6 mice (H-2b) with a distinct protocol, using chicken (c) CII. Even if the initial publications announced incidence of arthritis comparable to DBA/1 mice (around 90–100%), several teams experienced difficulty to obtain an incidence to be over 70% in C57BL/6 mice [212]. The induction protocol is slightly different than in DBA/1 mice, including another injection site of the CII/CFA emulsion (at the base of the tail); the doses of cCII may also be adjusted to get the highest incidence and severity of CIA. A crucial factor to a well-controlled CIA experiment in knockout mice is also the use of correct negative controls. It is actually mandatory to use littermate and not commercially available C57BL/6 mice or mice from another breeding, since environmental and genetic factor bias can significantly influence arthritis development and lead to false interpretations.

Evaluation and quantification of arthritis is critical to gain intra-laboratory repeatability and inter-laboratory reproducibility. The identification of an arthritic limb is not difficult, but the quantification of clinical signs is trickier due to variations in severity of the inflammation among limbs and mice. A blind procedure is required, and a subjective scoring system is applied to each limb using a scale of 0–4, with 4 being the most severe inflammation. The subjectivity of this procedure can make findings hard to interpret. Moreover, the global scale varies among laboratories, making it difficult to compare published data. An effort has thus to be undertaken to standardize the quantification procedures at the international level.

Ethical issues

All arthritis models involve pain discomfort, pain, or distress. Regarding ethical issue, it is mandatory to ensure the implementation of the 3Rs: Replacement, Reduction, and Refinement [213].

First, animals should be replaced if an alternative method could lead to the same amount of information. For instance, developments in three-dimensional tissue modeling for in vitro drug evaluation, synoviocyte culture models using cells obtained from human RA, and databases of in vitro, in silico models can be an alternative. If there is no such possibility, the scientific purpose has to be coupled with the choice of an animal model that will guarantee the lowest pain and distress.

Second, studies have to lower at the minimum the number of animals, but with the necessity to conserve enough power analysis to avoid unwanted repetition. For instance, sharing control mice for different contemporary experiments is a possibility.

Third, appropriated techniques have to be set up in order to minimize animal suffering. This means that (i) tools should be developed to improve pain and distress evaluation, (ii) analgesics have to be used to avoid pain and anxiety, and (iii) end point should be defined in protocols to limit suffering of animals. However, analgesia should be used providing that it does not interfere with pathophysiological events of the disease, nor has drug effects. For instance, paracetamol, possessing less anti-inflammatory activity than NSAIDs and COX-2 inhibitors, is potentially a suitable analgesic [214]. Otherwise, a boost is necessary in some models such as CIA. Boost with CFA or LPS should be avoided because of the stressful and severe adverse effects they can induce. Aggression between male mice is also a source of stress and potential injury. The risk can be reduced by establishing groups as earlier as possible and by the use of littermates.

Implementation of the 3Rs is a cornerstone of animal usage and is encouraged by all players involved in medical research: establishments, ethic committees, funding agencies, and journals. International working groups meet periodically to define more accurately animal models of arthritis and to edit guidelines for refined experimental protocols [214]. With this respect, animal models of arthritis will continue to play an important role in preclinical research, even if their predictive efficacy is not always transposable.

Conclusion

Animal models of arthritis are clearly useful for discoveries of novel mechanisms and drugs. It is mandatory to require a high level of rigor and thoroughness to design and conduct the experiments and to interpret the data. It is no more conceivable to perform an initial survey with an animal model to screen hypotheses without a full program. That could mean that these models are better studied in translational centers, where research and development questions are addressed with an integrated perspectivism of disease mechanisms and drug discoveries. Hence, studies in animal models and in human diseases appear as truly complementary.