Applying refinement to the use of mice and rats in rheumatoid arthritis research

There is some debate about the optimum in vivo RA models for particular disease aspects, as well as their translatability to human disease, with different models possessing different mechanistic and clinical features (Vincent et al. 2012). If animal models fail, it may be with respect to clinical predictivity, or to misapplication of the type of validity required, be this ‘face validity’ (similarity to the human disease features of interest), ‘construct validity’ (similar underlying biological mechanisms) or ‘predictive validity’ (whether there is a similar response to clinically effective therapeutic agents) (McGonigle and Ruggeri 2014). For further explanation and discussion of causes of reduced external validity, see van der Worp et al. (2010). A ‘pathogenesis map’ for RA to assist with the decision-making process regarding validity is set out in Vincent et al. (2012).

A ‘harm-benefit assessment’, in which the potential harms to animals (i.e., pain, suffering or distress) are considered against the possible benefits of each project, is commonly used by regulators and ethics committees to make decisions about the justification for animal use (e.g., European Commission 2010; National Research Council 2011). The project should also have realistic objectives that are deliverable in practice.

The Group considered the welfare impact of poly- and monoarthritic models, with respect to severity and the number of joints affected. The consensus was that some of the classical polyarthritic models (e.g., adjuvant- or collagen-induced models) generally result in a more severe arthritis and so should only be used as translational tools if there is strong supporting evidence of a relevant disease mechanism. Low severity (as opposed to hyperalgesia) monoarthritic models, including zymosan-induced and antigen-induced arthritis (Patel et al. 2010; Vincent et al. 2012), possess many relevant disease processes and, being less severe, should be used instead of polyarthritic models wherever possible. In addition, if only one limb is affected then the animal is able to compensate by redistributing weight between the other three limbs.

Harm-benefit assessments may be more complex for novel models that are less well established. Some may be more severe, e.g., the SKG mouse which develops severe arthritis plus extra-articular inflammation (Yoshitomi et al. 2005), while others can be manipulated to have reduced severity with greater clinical relevance (e.g., the KBxN serum transfer model; Montero-Melendez et al. 2011).

Husbandry refinements, including appropriate environmental enrichment, benefit animal welfare and should be provided unless there is sound scientific justification to withhold them. However, it is essential to evaluate any effects on data variability (Mikkelsen et al. 2010) and to allow for these within the experimental design.

Based on the human experience of arthritis, affected animals should benefit from being able to keep warm and comfortable, exercise as appropriate and reach food and water easily. Temperature may be especially important for animals in RA studies, since healthy mice, given an opportunity to select their thermal environment, choose an ambient temperature of 30–31 °C, considerably above the range in most facilities (Gaskill et al. 2009, 2012). This suggests that it is good practice to review ambient temperature levels for rodents in RA studies and provide a sufficient quality and quantity of nesting material. Besides the animal welfare implications, the systemic sympathetic response to cold stress can affect data quality in studies relating to immune function (Karp 2012; Kokolus et al. 2013), and it is worth considering how this might also apply to RA projects.

Regarding enrichment, shifting attention away from pain benefits human patients (Ulrich 1984; Chan et al. 2012; Havey et al. 2014), and distraction from pain also modulates pain perception in animals (Gentle and Tilston 1999; Ford et al. 2008). An appropriately stimulating environment will therefore likely help to improve welfare and reduce pain perception in animals on RA studies.

Defined sources of enrichment items should be used, because contaminants (e.g., dioxin), present in some oils and bleaching agents, can act as confounds (e.g., by affecting Cyp1A1 gene activity; Tischkau and Mukai 2009). Standardisation can be managed in the same way as regulatory toxicology studies, in which in-house Quality Assurance groups set limits of acceptability for different substances in litter, nesting materials and enrichment items. These limits are used to review Certificates of Analysis that accompany such materials. Describing the sources of all materials that are provided will help others to interpret the results and conclusions of publications (Hutchinson et al. 2005).

It has been demonstrated that being caught by the tail is stressful for mice and induces anxiety (Hurst and West 2010; Gouveia and Hurst 2013). Avoiding capture by the tail in RA studies, by catching and restraining using cupped hands, tunnels or Vetbed^®, could thus decrease stress as well as reducing the risk of causing discomfort through involuntary extension/flexion of arthritic joints, or if animals have been injected close to the tail base.

Local adverse effects due to inducers can be minimised by reducing the dose, volume and frequency of administration. The feasibility of this approach depends on the nature of the antigen, its solvent, and the adjuvant, and can be evaluated in pilot titration studies, using scoring systems to monitor the onset and severity of arthritis.

Regarding administration, wide gauge needles cause more pain and increase the risk of ‘tracking’, where a tunnel remains under the skin following withdrawal. The inducer can leak into this, causing irritation and further discomfort or pain. To avoid tracking, withdraw the needle slowly and smoothly, applying slight pressure with the thumb. Mycobacteria for adjuvant and collagen-induced arthritis should be ground very finely to prevent ‘stacking’ in finer needles (Brand et al. 2007) and consequent variability in incidence and response. NB Care must be taken when alternating between strains of Mycobacterium, as M . butericum can induce more severe rat adjuvant disease than M. tuberculosis.

Many protocols use intradermal injection, but this leads to a higher incidence of ulcers, especially if administered close to the tail. In the experience of Group members, subcutaneous injection at single or multiple sites further from the tail (e.g., on the flank) can reduce ulcer incidence while still reliably inducing arthritis. This may be possible in some studies that do not depend on intradermal injection, and a pilot study to evaluate an alternative administration protocol may be justifiable. If there is scientific justification for intradermal injection, sites should be chosen with care, taking into account both the animal’s movements and areas that will be affected during restraint.

Refinement can also be achieved by reducing the duration of experiments, provided this is compatible with the study aims, i.e., all the necessary data can be obtained within the study time. There may also be scientific reasons not to prolong studies, due to the risk that the disease will enter the ‘repair phase’, with associated periostitis or ankylosis that can confound the results.

Boosting is not used in rats, but is sometimes used in mice to support antibody-induced collagen induced arthritis, or to synchronise the collagen-induced arthritis (CIA) model. Boosting commonly takes the form of an intraperitoneal injection of lipopolysaccharide (LPS) or subcutaneous administration of collagen in Freund’s incomplete adjuvant (FIA); Freund’s complete adjuvant (FCA) should not be used due to the challenge response to the adjuvant.

Administering LPS is stressful and induces cytokine release that can cause severe adverse effects including shock, diarrhoea, and malaise. Animals may not drink, and may develop faecal plugs that, coupled with diarrhoea, can be life-threatening if not checked regularly and cleaned away. However, there may be justification for using LPS if it allows study duration to be reduced and/or fewer animals used because the incidence of arthritis is reliably increased. Collagen/FIA boosting should be subcutaneous, and can be refined by restricting sensitisation to one side of the animal and administering the boost at a separate site. The choice of ligand should also take translatability and adverse effects into account; for example, the various TLR ligands may model different aspects of RA more effectively than more commonly used agents such as LPS or FIA, but may have different adverse effects.

Animals experience the most severe pain during the acute or ‘attack’ phase, during which they need close monitoring and additional care, including analgesia if feasible (see Sect. 5). Most therapeutic studies are carried out during this acute phase and it is not usually necessary to extend them into the chronic, resolving phase (see below). The severity of the acute phase should be refined to provide the required statistical power for the primary outcome and not more.

If polyarthritis models are taken into the ‘chronic resolving’ or ‘recovery’ phase (e.g., to study bone remodelling), the impact on the animal must be carefully considered. The consequences of periostitis become more prominent (especially in rat adjuvant arthritis) as aberrant bone outgrowths develop within the paws and along the bone shafts. Although acute inflammation may have receded, periostitis and spondyloarthropathies are painful and add significantly to the lifetime severity.

Joint damage can be so severe in polyarthritis models that resolution to normality is impossible. Low acute-load models should be used to investigate factors influencing resolution, as they allow mechanisms in joint resolution to be seen as well as causing less suffering (Montero-Melendez et al. 2011).

In studies of potential therapeutics, animals in control groups that do not receive the candidate therapeutic agent will develop the most severe form of disease and are of special concern. The requirement, and humane endpoints, for controls should be very carefully considered. In some fields, controls can be avoided by using ‘historic’ controls from the literature or the same institution. Unfortunately, in RA studies this is likely to mislead due to variations between institutions, contemporaneous environmental factors, and the protocols used. However, sharing control groups from different, contemporary experiments within the same institution is valid, and should be encouraged, provided they possess sufficient power.

Inappropriate enrichment can cause aggression in male mice (Marashi et al. 2003), and male DBA/1 and C57BL/6 mice, both of which are used for experimental arthritis, are especially prone to aggression. However, the risk can be reduced by establishing groups early, using littermates, ensuring that animals are not subsequently mixed and selecting appropriately designed refuges. For mice, two refuges or dual entry/exit designs can defuse aggression (T Boden pers. comm.), but if it persists, it may be necessary to remove aggressor(s) since fighting may cause stress, injury and infection—which can all influence arthritis development. Aggression between arthritic rats is not expected and indicates a serious welfare issue that should lead to a review of husbandry and experimental protocols.

Adjuvant arthritis models, such as intradermal/subcutaneous administration of CFA to rats, or pristane to rats and mice, can easily lead to severe outcomes. Severity may be reduced by refining the initiation doses, e.g., reducing the dose of pristane for pristane arthritis, without compromising the arthritis outcomes (Malik et al. 2011).

The severity of CFA arthritis in rats can be compounded by systemic disease that, if uncontrolled, includes liver granuloma (which can affect Cyp enzyme activity), skin disease, splenomegaly, ulceration at the injection site, tail lesions, eye disease and disuse of the hind paws leading to dragging (‘sledging’), and ulceration of the hind feet.

Although there are welfare concerns, there may be scientific benefits to consider as part of a harm-benefit assessment. The ‘structural validity’ of CFA-induced arthritis is different to other models (Patel et al. 2010) with cell mediated immune responses predominating, which may benefit the scientific objectives of some studies. In addition, drug responses can be more easily distinguished (Bolon et al. 2011), and when inbred strains are used responses tend to have low variability, with robust incidence and very predictable disease onset. These points mean that numbers do not need to take variability into account, so can be kept lower than for murine models. Therefore, there may be scientific justification for using CFA, but this should be very closely scrutinised and every opportunity taken to reduce suffering; pilot studies may be advisable before using this model for the first time.

Older publications describe the injection of the inducer into one hind paw, inducing the ‘primary response’, a chronic granulomatous reaction. The ‘secondary response’ of inflammation in the contra-lateral paw reflects the systemic arthritis. This protocol is incapacitating, can lead to severe disease, and should not be used.

Some GA strains require careful monitoring, as the severity of arthritis may be unexpectedly severe or chronic. For example, prostaglandin-D₂ synthase knockout mice have an exaggerated delayed-type hypersensitivity response, which would be predicted to result in severe disease if expressed in collagen arthritis (Trivedi et al. 2006); indeed, PGD₂ antagonism exacerbates disease (Maicas et al. 2012). Where exacerbation of disease is expected in a GA line, models or protocols of reduced severity, such as monoarticular or reduced inducer dose, should be used wherever possible.

Most analgesics affect the immune system in some fashion (Paska et al. 1986; Earl et al. 1994; Dinda et al. 2005; Pulichino et al. 2006), so there can be concern about providing analgesia for fear of introducing a confound. However, there is increasing evidence of effects upon many body systems (including immunity) of unrelieved pain and distress in animals, and how these can influence the experimental outcome (Baumans et al. 1994; Livingston and Chambers 2000; NHMRC 2008; Ren and Dubner 2010). It is also noteworthy that analgesics are rarely excluded from human clinical trials, which are therefore subject to the same confounding influences.

The authors therefore propose that analgesia should be used unless there is sound scientific justification otherwise, e.g., if it can be demonstrated that analgesia would make it difficult to attribute therapeutic effects to the study compound, or if a class of analgesic could affect unpredictably the disease severity in studies using GA animals. For a discussion of decision making regarding pain alleviation, see Carbone (2011).

Pain relief protocols can be designed to minimise potential impact on the scientific objectives. For example, analgesia can be tailored to periods when inflammation is at its peak and likely to be especially painful (Khachigian 2006; McCarthy et al. 2012), with administration in anticipation of (rather than in response to) the pain in order to increase efficacy.

When considering whether to provide analgesia, two essential questions are (i) what effects occur at analgesic doses, and (ii) will these necessarily invalidate experimental outcomes? There are a number of reports of the use of analgesics in arthritis models without negative impacts on the experiment; some examples are set out below.

Gabapentin has been found to be effective in the attenuation of allodynia during the chronic phase of murine K/BxN arthritis (Christianson et al. 2010) and has also been shown not to interfere with the immune response (Van Loo et al. 2006). Gabapentin, the NSAID ketorolac and the TNF receptor antagonist Etanercept (R) have all been reported as effective during the acute phase, whereas gabapentin alone was effective on allodynia in the chronic phase (Christianson et al. 2010). Since NSAIDs are anti-inflammatory and can have DMARD activity (Seed and Burnet pers. comm.), anti-TNF is antirheumatic, and both are only effective in the acute phase, gabapentin could be used.

Buprenorphine, a partial agonist opiate, does not prevent the development of a reliable arthritic response in mice when administered in drinking water (M. Burnet, Synovo, pers. comm., see below), but neither direct comparisons with non-treated mice, nor effects on joint histopathology, have been assessed. Buprenorphine reduced spinal neuronal discharges and reduced allodynia during the acute and chronic phases of mouse CAIA (Bas et al. 2012). However, oral administration of buprenorphine at analgesic doses (2 mg/kg twice daily) in rat SCW arthritis inhibited inflammation and joint erosion (Volker et al. 2000).

Paracetamol (acetaminophen) is potentially a suitable analgesic, since it possesses less anti-inflammatory activity than NSAIDs and COX-2 inhibitors. Paracetamol at 50 mg/kg significantly reduced nociceptive evoked and spontaneous spinal discharges in adjuvant arthritis (McQueen et al. 1991). It also reduced inflammatory hyperalgesia without affecting carrageenan inflammation and central hyperalgesia (Bianchi and Panerai 1996).

Therefore, it is possible to provide analgesia in some RA studies, and some institutions apply analgesia routinely. However, further research is needed into suitable analgesic regimes, including evaluations from the time of induction, or specifically the attack and chronic phases, to assess analgesic efficacy and dissociate these from arthritic outcomes. If in doubt, the veterinarian should be consulted, literature searched and pilot studies conducted if necessary.

Pilot studies could establish whether analgesia might be provided at one or more stages of an RA study without significantly compromising the science. Although the animals in the full study will benefit if it is shown that pain relief is feasible, pilot studies to evaluate this will cause suffering and should undergo a harm-benefit assessment, with full implementation of the three Rs. Ideally, the pilot study should be designed such that data from it could potentially be incorporated into the main study, to avoid using ‘extra’ animals.

The EC has also published a worked example of this approach for Type II CIA in rats (European Commission 2013), and “Appendix” to this document sets out an example hypothetical ‘score sheet’ for mice using some of these indicators. Note that this is a generic example, not suitable for use without adaptation and tailoring to the species, strain and protocol. Some of the indicators and their applications are explained further below.

Swollen paws

Animals should be very gently caught and handled, and paws checked daily, from around day 14 following the initial induction (or from the point of inflammation). The digits and joints should be examined, and if there is swelling it should be noted how high up the limb this is present. Swelling may be measured with (preferably non-spring) callipers or by plethysmometry, as swollen paws are painful (Bolon et al. 2011) (Fig. 1). Swollen paws and digits may also be accompanied by reddening of the skin (Fig. 2).

Open image in new window

Open image in new window

An example approach to visually monitoring clinical scores of the hind paws in pristane adjuvant arthritic rats is outlined below (Table 5; Fig. 3). In this trial, all joints were assessed in order to determine an optimal scoring system relevant to that model, as different joints develop and resolve arthritis at different times.

Table 5

Example scoring scheme for investigating the pattern of paw involvement in pristane arthritis, using DA rats to determine outcomes for the full trial. This can be adapted for use with other models

Arthritis score	Maximum points—hindpaws	Maximum points—forepaws
1 point for each swollen or red digit	5	4
1 point for each swollen knuckle	5	4
1 point for swollen midfoot	1	1
1 points for a swollen ankle/wrist	1	1
Total (×2)	24	20

Open image in new window

In mice, distinguishing individual joints is difficult without handling animals and touching the paws. However, in the authors’ experience an assessment system for mice in which whole paws are scored (Fig. 4) was as robust as a protocol that scored individual digits and joints (see “Appendix”). ‘Global’ scores for each paw, tailored to individual projects, can therefore be used for mice, avoiding handling.

Open image in new window

Infrared thermography (using a video camera) has been suggested to monitor clinical severity, as foot temperature can be correlated with the degree of swelling (Jasemian et al. 2011). Thermographic images can be taken without anaesthesia, restraint or otherwise handling the animals, with welfare advantages if joints are painful.