Thinking bedside at the bench: the NOD mouse model of T1DM

Abstract

Studies over the past 35 years in the nonobese diabetic (NOD) mouse have shown that a number of agents can prevent or even reverse type 1 diabetes mellitus (T1DM); however, these successes have not been replicated in human clinical trials. Although some of these interventions have delayed disease onset or progression in subsets of participants, none have resulted in a complete cure. Even in the most robust responders, the treatments do not permanently preserve insulin secretion or stimulate the proliferation of β cells, as has been observed in mice. The shortfalls of translating NOD mouse studies into the clinic questions the value of using this model in preclinical studies. In this Perspectives, we suggest how immunological and genetic differences between NOD mice and humans might contribute to the differential outcomes and suggest ways in which the mouse model might be modified or applied as a tool to develop treatments and improve understanding of clinical trial outcomes.

Introduction

Type 1 diabetes mellitus (T1DM) is an autoimmune disease that occurs when the immune system destroys β cells of the islets of Langerhans in the pancreas. Pancreatic β cells decline in number and function until they do not produce sufficient levels of insulin to meet metabolic demands. As a result of the inaccessibility of the target tissue and lymphoid organs in humans and the utility of studies of mouse genetics for dissecting mechanisms, studies in rodent models have become guideposts for understanding human T1DM and planning clinical interventions. Before the nonobese diabetic (NOD) mouse model was available, rodent models that shared features with human autoimmune diabetes mellitus included the BioBreeding/Worcester rat, diabetes mellitus induced with lymphocytic choriomeningitis virus and diabetes mellitus induced with multi-dose streptozotocin.^1–3 However, as diabetes mellitus develops spontaneously in NOD mice and because of the range of tools and genetic information available for mouse studies, the NOD mouse has become the model of choice for preclinical studies. Interestingly, the other rodent models have a number of features, such as the phenotype and islet cellular infiltrates, that more closely mimic human disease than the NOD mouse. Over the past 10 years, several authors have reviewed the progress and pitfalls of the NOD model. These studies have emphasized the immunopathological findings in islet cells as well as the lessons that were learned from the model regarding genetic susceptibility.^4–14

The NOD mouse was discovered in 1974 and was subsequently inbred.¹⁵ This model has been a critical tool that has developed our understanding of disease mechanisms and has been used to focus strategies, including the use of broad spectrum immune suppressive regimens attempted in patients (such as ciclosporin,^16–18 azathioprine^19,20 and azathioprine with prednisone²¹) and agents that specifically target cell populations and mediators. Drugs such as metabolic agents, antibodies that target cell surface immune ligands such as CD20, CD3, or co-stimulatory molecules on lymphocytes, as well as cytokines, antigens and others, were first tested in NOD mice;⁴ several were found to partially prevent, delay or reverse T1DM, at least in some patients. In clinical studies, some of these agents (such as anti-CD3 monoclonal antibodies [mAbs],^22–24 anti-CD20 monoclonal antibody [mAb]²⁵ and alefacept²⁶) have had successes in patients with new (within 3 months of diagnosis) or even recent-onset (>3 months from diagnosis) T1DM in attenuating the decline of provoked C-peptide responses, generally with improved HbA_1c levels and reduced requirements for exogenous insulin. Some of these patients have even responded robustly, becoming insulin independent or maintaining their levels of C-peptide responses for >2 years.

However, other clinical trials have either failed or yielded surprising outcomes in light of the NOD studies, including improvement (abatacept [CTLA4-Ig]^27,28), described below, or deterioration (combination therapy with rapamycin and IL-2²⁹) of T1DM. Furthermore, even if the treatment is successful, the results are not permanent. For instance, the C-peptide response declines with time, often at a similar rate in the control and treated groups after a period of stabilization in the latter. Nonetheless, these discrepancies between the mouse and human studies have left investigators with unanswered questions about future directions to pursue a cure for T1DM. In this Perspectives, we discuss preclinical studies in NOD mice and the results of clinical trials with a particular focus on the reasons for the differences between the NOD mouse model and humans with T1DM, as well as the design and interpretations of preclinical studies that might account for failures in clinical translation.

Summary of recent studies

Preclinical (NOD) and clinical studies have reported concordant and discordant results (Table 1). Some of the agents that were effective in prevention and reversal of diabetes mellitus in NOD mice, such as porcine oral³⁰ and parenteral insulin,³¹ GAD65(alum),^32–34 sitagliptin and lansoprazole,³⁵ anti-IL-1,³⁶ IL-1 receptor antagonist agents^37–39 and antithymocyte globulin (ATG),⁴⁰ did not show efficacy in well-designed clinical trials with standard end points that compared provoked C-peptide responses in patients treated with the drug or control individuals given placebo.^41–47 Other agents, such as Fc receptor nonbinding anti-CD3 mAbs^48–52 and anti-CD20 mAb,^53,54 prevented and/or reversed diabetes mellitus in preclinical studies but were only partially effective in clinical trials, delaying onset or disease progression.^{22–24,55–57}

Table 1.

Comparisons of preclinical and clinical studies

Treatment	Preclinical results*	Clinical trial results	Problems in translation
Parental insulin	Prevention of diabetes mellitus in NOD mice31	No prevention of T1DM in high-risk individuals42	Dosing regimens not comparable
Oral insulin	Prevention of T1DM in NOD mice that received porcine insulin30 Recombinant human insulin reduces lymphocytic infiltration96 Single amino acid change at position 30 in the B chain of human insulin prevents tolerance to insulin in NOD mice72	Did not prevent progression to T1DM overall but significant improvement in patients with higher titres of anti-insulin antibodies97	Specificity of oral insulins for conferring tolerance in T1DM was confirmed in a follow-up study with mice that occurred while human trials were already underway72 Only certain forms of insulin induce the production of cytokines72,98
Teplizumab and otelixizumab (FcR non-binding anti-CD3 mAbs)	Prevented diabetes mellitus induced with low-dose streptozotocin in mice49 Reversed diabetes mellitus51 Prevented NOD pups from developing T1DM if injected as neonates48	Treatment maintained increased C-peptide levels and reduced insulin requirements in recent-onset T1DM22,55 Reduced β-cell death22	Effects were not permanent as in mice but some individuals showed robust responses Does the permanent reversal in NOD mice model these patients?
Rituximab (anti-hCD20)	Prevented the onset of T1DM and reversed hyperglycaemia in transgenic NOD mice expressing hCD2053 Induced regulatory B cells53	Improved C-peptide response at 1 but not 2 years25	Transient effect of drug treatment. Disease recurred with recovery of B lymphocytes25 Effects of drug treatment on antigen-specific responses unclear25 Selective effects on autoantibody responses25,88
Abatacept (CTLA4-Ig)	Prevented onset of T1DM in prediabetic NOD mice but not effective when administered after 8 weeks of age62 CTLA4-Ig transgene accelerated T1DM in NOD mice61	Successful delay in decline of C-peptide responses27,28	Kinetics of disease and role of positive and negative T cell co-stimulation for autoimmune responses at different stages? Differences in expression of CD28 on T cells74
Rapamycin and IL-2	Treatment of NOD mice with the combination of rapamycin and IL-2 prevented T1DM incidence63 In treated mice, there was increased FoxP3+ TREG cells63,99–101	Trial discontinued due to poor initial results29	Rapamycin could be toxic for human β cells,85 which might be augmented by IL-229 The increases in NK cells might account for β-cell losses29 The effects of IL-2 on NK cells in NOD mice might not be appreciated because of mutations in pancreatic NK cells in NOD mice75
Antithymocyte globulin	Prevented and delayed the progression of T1DM40 With exendin-4 (glucagon-like peptide 1 receptor agonist), reversed diabetes mellitus in NOD mice82	β-cell function was not preserved47 Relative proportion of TREG cells was not enhanced47	No evidence for establishment of regulatory mechanisms Complement deficiency in NOD mice might explain why induced TREG cells could survive in mice but not in humans66
Alefacept (soluble LFA3-Ig)	Promoted the survival and growth of insulinoma grafts58	No difference in 2 h C-peptide response but a statistically significant increased 4 h C-peptide response26 Decreased number of effector and memory T cells26	Clinical end point chosen for analysis (2 h vs 4 h C-peptide response to a mixed meal) accounted for statistical outcome26 Mice do not have the CD58 gene that binds CD2 but have CD48, which binds CD2 with lower affinity64
Sitagliptin and lansoprazole	Reversed diabetes mellitus by restoring normal glucose levels in mice35 Induced β-cell growth of human islets transplanted into the exocrine pancreatic ducts of mice83 Reduced insulin autoantibodies and results in a cytokine shift in infiltrating leukocytes from IFN-γ to TGF-β102	No effect on loss of C-peptide responses in new-onset patients44	Effect on autoimmunity not clear44 Differences in the proliferative capacity of human and murine β cells

^*Data are from published analyses. Abbreviations; FcR, Fc receptor; mAbs, monoclonal antibodies; NK, natural killer; NOD, nonobese diabetic; T1DM, type 1 diabetes mellitus; T_REG, regulatory T cells.

Alefacept (LFA3-Ig), which targets memory T cells, was originally tested in NOD mice that had received an insulinoma graft that was modified to secrete LFA3-Ig to test whether immunomodulators could improve graft survival. The growth of the tumour cells and survival were improved in NOD mice transplanted with the graft transfected with a transgene encoding LFA3-Ig.⁵⁸ In a clinical trial, alefacept also showed modest effects on C-peptide responses in patients with T1DM.²⁶

Two trials have reported paradoxical findings. Abatacept had shown efficacy in human autoimmune diseases other than T1DM, such as rheumatoid arthritis.^59,60 However, in NOD mice, mouse CTLA4-Ig (expressed from a transgene)⁶¹ or anti-B7.1 mAb, which blocks a CD28 ligand, accelerated the progression of diabetes mellitus in female mice, and anti-B7.2 mAb or human abatacept showed modest protection when given to 6–8 week old mice but not when given to mice >8 weeks of age.⁶² However, patients with new-onset disease treated with abatacept had a statistically significant improvement in provoked C-peptide response compared with patients given placebo.^27,28 Rapamycin administered in combination with IL-2 also reduced the incidence of diabetes mellitus in female NOD mice,⁶³ but caused a transient decline in the C-peptide response in nine patients with recent-onset T1DM and residual insulin production who had been treated with this combination regimen.²⁹

Immunological issues

Mice and humans are separated by 65 million years of evolutionary history⁶⁴ and have differences in their adaptive and innate immune systems that are relevant to the use of mouse models. Interferons have been proposed to be important mediators of an initial innate immune response to islet antigens that occurs in young NOD mice around the time of weaning; however, in mice, interferons do not activate STAT4 or induce a type 1 T helper cell (T_H1) response in adaptive immune cells as they do in humans. NOD mice have additional unique features that distinguish their immune responses from human responses such as their failure to produce the mRNA or protein for the complement inhibitor factor-H related protein-C.⁶⁵ IL-10, which is thought to be an important regulatory cytokine, is only produced during a type 2 T helper cell (T_H2) response in mice but is produced during both T_H1 and T_H2 responses in humans.⁶⁴ These distinctions might account for the different outcomes observed in mice and humans treated with immune agents. For example, unlike in mice,⁶⁶ regulatory T cells (as well as effector T cells) were depleted by thymoglobulin administration in human patients and this reduction might have affected the balance of effectors and regulatory cells.^47,64

Humans and mice have other relevant genetic discrepancies that affect the function of the immune system. Importantly, NOD mice are inbred and 80% of female mice develop diabetes mellitus but only 20% of male mice develop the disease; this sexual dimorphism and inbreeding is not seen in humans.^15,67 T-cell antigen recognition using class II major histocompatibility complexes (MHC) is known to confer susceptibility to T1DM in mice and humans. The binding pocket of the NOD class II allele (H2-A^g7) is similar to the binding pocket of the analogous human MHC molecule (HLA-DQ) but NOD mice do not express the other class II MHC molecule I-E.^9,11 Surprisingly, HLA-DQ8 transgenic NOD mice, which express the human MHC susceptibility gene, develop autoimmune myocarditis.⁶⁸ In addition, the class II human MHC alleles associated with T1DM have a greater degree of polymorphism than the NOD MHC genes. Furthermore, the HLA complex includes several loci that independently confer risk of T1DM. Thus, the total genetic burden of susceptibility alleles (homozygous or heterozygous) differs between individuals.¹⁰

The efficacy of antigen-specific treatments might differ depending on the MHC alleles involved and the way in which peptides bind in the MHC groove.^69–71 Recognition of antigens by autoreactive T cells is generally of low avidity and, therefore, seemingly minor differences in peptide binding between alleles have profound effects on responses to that antigen. For example, a single amino acid difference in the B chain of human insulin seems to account for the ability of porcine, but not human, insulin to prevent the development of diabetes mellitus in NOD mice.⁷² These genetic differences might account for the heterogeneity in responses to oral insulin among at-risk relatives of patients.⁷³ Furthermore, the extended MHC haplotype might affect autoimmune responses, as illustrated by the dominant protective effect of HLA-DR2 alleles on the risk of developing T1DM in humans. Finally, the genetic heterogeneity, sex and environmental factors in humans can affect responses to drugs, but these variations are not well-modelled in mice.

A number of other immunological differences in the immune systems of mice and humans might be relevant. For instance, CD28 is expressed on all CD4⁺ and CD8⁺ T cells in mice, whereas CD28 is expressed on all CD4⁺ but only 50% of CD8⁺ T cells in humans.⁷⁴ This disparity suggests that the effects of abatacept, which blocks CD28 co-stimulation by B7.1 and B7.2 ligands of activated antigen-presenting cells, might differ in mice and humans, particularly for low avidity interactions that occur with autoantigen recognition. NOD mice have impairment of natural killer (NK) cell function that might account for the failure to identify the toxicity of IL-2 treatment in the preclinical studies of the rapamycin and IL-2 combination therapy.^10,29,75 In addition, the insulitis observed in humans with T1DM and in NOD mice are qualitatively and quantitatively different. Even in individuals with new-onset T1DM or those at very high risk of developing T1DM, insulitis can be found in <10% of the islets (versus >50% in NOD mice⁷⁹) and the loss of β-cell mass in the prediabetes period is less than in NOD mice.⁷⁶ The insulitis seen in human disease is characterized by fairly small numbers of CD8⁺ T cells and macrophages, whereas high numbers of B cells and CD4⁺ T cells predominate in NOD mice.^11,77

Disease kinetics

The kinetics of the disease process are important for the timing of when therapies are delivered but also differs between NOD mice and humans. Humans >21 years old at the time of diagnosis have a slower reduction in levels of C-peptide than individuals who are diagnosed at <21 years old.⁷⁸ By contrast, the rate of loss of β cells in NOD mice is rapid;⁷⁹ however, the effects of age on their loss has not been well-studied. The reason for the differences in the rate of progression in humans before or after age 21 years has not been identified but could involve immunological or β-cell differences, such as the diversity of their immune repertoire (expected to be reduced in older patients), the expression of relevant antigens by younger β cells, or even their susceptibility to cell death. Nonetheless, this heterogeneity is very important for the design of clinical trials. Studies in young adults require a greater number of participants than studies in older patients to show a statistically significant clinical response. Clearly, if the disease mechanisms are different in younger and older patients, the effects of immune interventions are also likely to differ. Anti-CD3 mAb is most effective in younger (that is, age <15 years) patients.²⁴ In NOD mice, anti-CD3 mAb shows the greatest efficacy when it is given at the time of peak β-cell death, either in prediabetic NOD mice or in recipients of islet allografts.⁷⁹ However, when anti-CD3 mAb was administered to patients with established autoimmune responses (that is, 4–12 months after diagnosis and particularly those aged >15 years) the effects were less robust than in younger patients with new-onset disease, which suggests a window of treatment opportunity is present in humans that is not defined by observations in mice.²⁴

In addition, it is very difficult to extrapolate from preclinical studies to the clinical setting because no biomarkers clearly define the stage of the disease, as well as for practical reasons. As an example, disease prevention with parenteral insulin in NOD mice was effective when it was initiated at a very young age (4 weeks).³¹ However, in humans, the antigen can usually only be given after the appearance of autoantibodies, when insulin secretion is impaired.⁴² NOD mouse follicular B cells downregulate CD20 when they infiltrate the islets of Langerhans.⁸⁰ The inefficiency of anti-CD20 in late-stage diabetes mellitus might be due to a similar phenomenon if the same process occurs in humans. This finding also suggests that the optimal timing for this therapy is before the appearance of autoantibodies, for example, relatives of patients with T1DM could be given this treatment in the first few years of life.

Target cell issues

A number of preclinical studies have suggested that successful therapies in mouse models, such as rapamycin with IL-2,⁶³ α1-anti-trypsin,⁸¹ anti-lymphocyte serum with exendin-4⁸² and a dipeptidyl peptidase-4 inhibitor with a proton pump inhibitor,⁸³ can cause β-cell regeneration. These findings suggest that T1DM could be cured by both arresting ongoing destruction and permitting spontaneous or enhanced β-cell proliferation. Whether or when β-cell proliferation occurs in humans is still unknown and clinical data offers little evidence to suggest this occurs. For instance, the recovery of β-cell function after disease onset is most consistent with recovery of function of β cells rather than growth of new cells.^79,84 Rapamycin combined with IL-2 can improve β-cell mass in NOD mice,⁶³ but can also be toxic to β cells,⁸⁵ which suggests that either outcome would be possible if this treatment is used in patients.

Of great clinical importance is the notable β-cell function at the time of onset of T1DM and even for long periods of time after disease onset.⁷⁸ The Joslin Medalist study reported that even 50 years after a diagnosis of T1DM, 67.4% of patients had detectable levels of C-peptide.⁸⁶ As residual insulin production is a key factor in determining the clinical outcomes of the disease,⁸⁷ the absence of chronic residual insulin production in the setting of autoimmunity in NOD mice is a shortfall that limits the usefulness of this model in studying the effects of agents at later stages of the disease.

Improving NOD studies

Studies in NOD mice have provided a basic understanding of disease mechanisms and immunological tolerance and have enabled the identification of autoantigens, as well as giving insights into the identity and function of pathogenic T cells and the contribution of genetic polymorphisms that would never have been possible without this model. Furthermore, disease-initiating events can only be studied in the mouse model and preclinical studies have, in fact, accurately predicted the success of a number of clinical studies and identified the successful mechanisms.^{22,25,27,28,48,49,51,53,55,61,62,74,88} Has the model lost its relevance to clinical T1DM or is the problem in the interpretation of the findings? How could the limitations be addressed?

The design of preclinical studies is as critical as the design of clinical studies and studies testing the same regimens should be consistent with respect to key issues such as timing, dosing of agents and sample size. Improving the design of preclinical studies might help ensure confidence in the outcomes of the mouse studies. Preclinical studies are generally performed at a single site and under very controlled housing conditions with sufficient group sizes to achieve statistical significance, but do not capture acquired differences between facilities or even inherent biological variation. Even studies of mice purchased from the same vendor and same lot, and housed under similar conditions, show unexpected results that might be due to differences in the microflora between housing locations and other uncontrolled factors (Herold, K. C. and collaborators, unpublished work). Studies that are reproduced at different times, between sites, with sample sizes larger than the 6–20 mice used in experimental groups in most reports represent a robust test more akin to clinical conditions than the current format of preclinical studies. This strategy would capture variance between sites as well as minimize the idiosyncratic and environmental differences and biological variation. Parallel mechanistic experiments designed to study the targeted pathway would also provide a strong indication that a false positive conclusion had not been reached.

The dosing of agents and timing of administration are important considerations that are often taken for granted, with clinical trials being designed on the basis of practicality or body weight rather than the data from the preclinical studies. The lower dose (per kg) of parenteral insulin given to relatives of patients with T1DM in the Diabetes Prevention Trial versus NOD mice is an example of this phenomenon.⁴² Mice received 220 μg (211 U/kg per dose) of human recombinant insulin at weeks 4, 12, 20 and 28,³¹ whereas humans received recombinant human ultralente insulin twice a day at 0.125 U/kg and participants were hospitalized every 12 months and received 0.015 U/kg/h of parenteral insulin for 4 days.⁴² Rather than body weight, using body surface area has been suggested to be a more reliable approach for transitioning from animal to human doses.⁸⁹ The appropriate timing for interventions in T1DM is challenging, as no absolute measure of disease progression exists that can be compared between mice and humans. However, the time point during disease progression when the efficacy of agents is maximal, reflected by the frequency of antigen-specific T cells in mice or autoreactive T cells and autoantibodies in humans, might help to guide timing of interventions in humans. A critical need (and a limitation of NOD studies) is a model of the deterioration in β-cell function after remission in the new-onset setting. A more direct measure of β-cell death, such as by measurement of unmethylated insulin DNA released from dying β cells, might provide clearer information on the relationship between interventions and the pathological consequences. ⁹⁰

The genetic and environmental heterogeneity of humans undoubtedly accounts for the mixed responses to drugs in patients, but could also contribute to the different outcomes of studies in NOD mice; however, these factors are difficult to model. A direct approach to this problem might involve developing and studying ‘humanized’ mice, that is, mice with human immune cells or even molecules from humans that are involved in immune responses. One example of this approach has been the introduction of the ε chain of the human CD3 molecule into mice to enable the study of the mechanisms of action of anti-CD3 mAb.⁹¹ In this example, signalling through the CD3ε chain by a mAb in clinical testing could be evaluated directly. The use of the CRISPR–Cas9 system for introducing genes into mice might greatly accelerate and expand the development of humanized models by enabling investigators to introduce genetic polymorphisms related to human disease. In addition to introducing single or multiple human genes into mouse immune systems, the responses of the human cells to drugs can be investigated in detail in humanized mice reconstituted with human immune cells.^92–94 Currently, a humanized mouse model of T1DM does not exist, but investigators have described immune phenotypes and pathological responses in mice that have been reconstituted with stem cells and immune effector cells from patients.⁹⁵ Importantly, these models seem to be useful surrogates for understanding the effects of immune modulators that have been used for the treatment of human T1DM, such as anti-CD3 mAb or even induction of autoimmune disease by human T cells with anti-CTLA-4 mAb.^93,94

Conclusions

The rationales for clinical trials require scientific guidance and a small animal model that can be used to address experimental questions and to test the usefulness of therapeutics. Discrepancies between the preclinical studies and the clinical outcomes have highlighted the pitfalls of the NOD model and its use in informing the design of clinical studies. However, the problems are more than the differences between the mouse and human diseases. They include the design and interpretation of the preclinical studies, environmental factors and genetic heterogeneity that affect human disease but cannot be modelled in mice, as well as very practical issues such as drug dosing and the timing of interventions. Human and mouse immune and β cells have important differences, some of which are particular to NOD mice rather than all animal models, which might account for the different outcomes in mice and humans. Nonetheless, the model continues to provide essential guidance to investigators to enable them to develop and test fundamental disease mechanisms such as the role of antigen-specific immune cells, identification of islet antigens and modulation of immune responses by genetic polymorphisms associated with T1DM. In this regard, the NOD mouse model is invaluable— it is the incubator for developing and testing new ideas and for developing new strategies for treatment.