Drug Discovery Today: Disease Models

Editors-in-Chief Jan Tornell AstraZeneca, Sweden Denis Noble University of Oxford, UK

Vol. 1, No. 4 2004

DRUG DISCOVERY

TODAY

DISEASE

MODELS

Autoimmune diseases and inammation

Models of type 1 (autoimmune) diabetes

Alan G. Baxter*, Rowena C. Duckworth
Comparative Genomics Centre, James Cook University, Townsville QLD 4811, Australia

Type 1 (autoimmune) diabetes is caused by T-cell autoreactivity resulting in destruction of the insulinproducing pancreatic islet b cells. Its aetiology involves complex interactions between multiple genetic and environmental factors, its pathogenesis involves interactions between many leukocyte subsets and their associated cytokines and its complications affect nearly every organ in the body. Modelling the disease is, therefore, the subject of much effort and surprising ingenuity. In this article, we review established and emerging models of the disease and give an indication of their use for different types of study. Introduction
Diabetes associated with insulin deciency varies widely in its clinical features. A small proportion of cases is characterised by abrupt onset at birth, or in the rst few months of life, without good evidence of autoimmune responses. A larger group with atypical diabetes develops the disease later in life, often in middle age, and shows variable insulin dependence and many features of type 2 diabetes. This article is limited to the discussion of models of type 1 (autoimmune) diabetes, which is classically characterised by: (1) onset in childhood, puberty or early adulthood; (2) evidence of autoimmune activation (circulating autoantibodies, T cells, or both, with specicity for islets, b cells or their constituents); (3) severe insulin deciency associated with low C peptide levels; and (4) if untreated, the biochemical sequelae associated with the

Section Editor: Daniel Altmann Human Disease Immunogenetics Group, Imperial College, London, UK
Autoimmune diabetes is now considered to be a prototypical example of a T-cell-mediated autoimmune disease. In recent years, there has been enormous progress in the analysis of the genetic and immunological mechanisms underlying this disease. In turn, this is leading to several new therapeutic strategies. These efforts have been helped considerably by the convergence of ndings from clinical studies and rodent models, especially those based on the nonobese diabetic mouse strain. Here, Alan Baxter an expert in clinical and experimental models of diabetes reviews the currently available models.

resulting failure in glucose homeostasis (polyuria, polydipsia, weight loss, ketoacidosis, coma and death).

Type 1 diabetes
Autoimmune diabetes is a disease of complex aetiology, with large numbers of poorly characterised genetic and environmental risk factors and modiers. The pathogenesis of the disease is almost certain to be autoimmune in nature because: (1) it is associated with autoimmune phenomena; (2) it showed evidence of immunological memory in diabetic patients that received human leukocyte antigen (HLA)matched pancreas transplants [1]; and (3) diabetes was inadvertently adoptively transferred by transplantation of non-Tcell-depleted bone marrow [2]. Owing to the relatively inaccessible location of the pancreas and the presence of digestive enzymes in the exocrine tissue of the organ, only limited data are available from biopsies. Similarly, pathological surveys of pancreatic tissue in patients dying of type 1 diabetes are surprisingly scant. As a consequence, the understanding of the mechanisms of b cell destruction is heavily inuenced by in vivo models of the disease.
www.drugdiscoverytoday.com

*Corresponding author: (A.G. Baxter) [email protected] 1740-6757/$ 2004 Elsevier Ltd. All rights reserved. DOI: 10.1016/j.ddmod.2004.11.012

451

Drug Discovery Today: Disease Models | Autoimmune diseases and inammation

Vol. 1, No. 4 2004

Table 1. Comparison summary table

In vitro models Pros Cons Best use of model Simple, inexpensive, scalable Limited in application Studies of pathogenesis In vivo models Generally the best available Complex, difcult to control NOD mouse: studies of aetiology and pathogenesis (especially genetic and environmental interactions), development of prediction strategies and identication of putative therapeutic interventions. BB rat: physiological studies, testing therapeutic devices. Transgenic: studies of pathogenesis, identication of autoantigens NOD mouse available from Jackson laboratoryd; BB rat available from Taconice; TCR transgenic mice available from Kathryn Haskinsb and Pere Santamariac [923] In silico models Flexible Dependant on data training Prediction

How to get access to the models

Insulinoma available from Edward H. Leitera; TCR transgenic mice available from Kathryn Haskinsb and Pere Santamariac [38]

Peptide-binding model available from Vladimir Brusicf; generalized linear modelling tools available from SASg [2427]

References

Abbreviations: BB, BioBreeding; NOD, nonobese diabetic; TCR, T-cell receptor. a http://www.jax.org/staff/edward_leiter.html. b http://www.uchsc.edu/immuno/faculty.html. c http://www.med.ucalgary.ca/webs/irgtraining/home/. d http://www.jax.org/. e http://www.taconic.com/. f http://sdmc.i2r.a-star.edu.sg:8080/$vladimir/. g http://www.sas.com/.

In vitro models
Although tissue and cell culture systems are unable to model the induction phases of the autoimmune response, they play a major role in the investigation of effector mechanisms of b cell destruction. Cultured insulinoma cells (e.g. NIT-1 [3]) and islets from mouse, rat or human [4] pancreata have been exposed to cytokines and immune effector molecules, such as Fas ligand and nitric oxide, to determine their susceptibility to killing or functional modication by these factors [5]. NIT1 insulinoma cells can be obtained by contacting the Jackson laboratory (http://www.jax.org/) (Table 1). An extension of these studies is the co-culture of islets, b cells or antigen-presenting cells pulsed with islet lysates, with various immune cell subsets (usually cloned T cells) from either mouse models or affected patients [6]. These approaches have been used to characterise the immune effector mechanisms of disease [7] and to identify putative autoantigens [8]. Currently, the most convenient source of clonotypic T cells for such studies is T cell receptor (TCR) transgenic mice.

In vivo models NOD mice

The nonobese diabetic (NOD) mouse strain is the most characterised and best-validated model of autoimmune diabetes; it is the gold standard for modelling aetiological, immunological, pathological and genetic aspects of the disease. The most extensively used line is the NOD/Lt line, which is
452
www.drugdiscoverytoday.com

available from the Jackson laboratory. Both male and female mice of this strain spontaneously develop lymphocytic inltrates into their pancreatic islets of Langerhans and, depending on housing conditions, up to 90% of female and 50% of male NOD mice succumb to diabetes caused by specic, T-cell-mediated, destruction of the insulin-producing b cells. As in the human disease, many genetic regions (currently more than two dozen) have been linked to diabetes, and environmental factors such as intercurrent infections seem to play a role in modulating tissue destruction, its biochemical sequelae, or both. Perhaps the most extraordinary aspect of the model is that there is some evidence that the same genetic risk factors that contribute to disease susceptibility in NOD mice also do so in humans. Several mouse diabetes loci have been mapped to locations syntenic to human diabetes susceptibility genes (Idd1, Idd4, Idd5 and Idd21) and, in at least some cases, the homologous gene has been shown to be involved [A (GenBank accession numbers NM_010379 and X54425) and Ctla4 (GenBank accession number NM_009843)] [9]. Indeed, NOD mice have an amino acid substitution at the same position in the gene homologous to the major histocompatibility complex (MHC) class II molecule DQb, to which much of the HLA-linked disease susceptibility in humans has been attributed [10]. Some concerns have been raised about this model. Islet destruction occurs over a much briefer period in NOD mice than in humans; the inltrates are much more pronounced in mice; the autoantibody specicities show some differences

Vol. 1, No. 4 2004

Drug Discovery Today: Disease Models | Autoimmune diseases and inammation

[11]; the diabetes is at the severe end of the clinical spectrum; and, if the diabetes is not treated, death tends to occur from dehydration, rather than ketoacidosis. Recent scepticism over failed clinical trials of candidate therapies identied in the mice [12] is probably unwarranted, because, generally speaking, the trial protocols concerned did not compensate for limitations in clinical translation, such as dose escalation and narrow therapeutic windows, that were clearly identied by the preclinical studies.

Humanised mouse models

A concerted multi-investigator effort is underway to develop transgenic and knockout mouse models of the human immune system for the study of autoimmune diseases and transplantation issues. Various components of human responses are being transgenically introduced into mice, including HLA molecules (optimally in association with targeted deletion of native H2 molecules), TCRs, CD4 and/or CD8, co-stimulator molecules (such as CD80, CD86) and complement inhibitors. Such systems can be used to study thymic selection of autoimmune T cells [20]. In some cases, the aim is to produce a human-like environment to foster the survival or development of human lymphoid cells following adoptive transfer of peripheral blood lymphocytes and/or bone marrow. In these cases, an immunodecient mouse host (e.g. scid or rag/) is used, often on a NOD background, because of a coincidental accumulation of several immune defects in this strain (for a review of these systems, see [21]).

Transgenic mouse models

One aspect of the complexity of the NOD mouse encompasses the processes required to generate a high frequency of autoreactive T cells bearing receptors with sufcient afnity to cause b cell destruction. A transgenic approach has been taken to eliminate this issue by inducing the transgenic expression of a TCR that is specic for native b cell autoantigens {receptor BDC2.5 available from Diane Mathis laboratory (www. hms.harvard.edu/dms/immunology/fac_mathis.html) [13]; receptor BDC6.9 available from Kathryn Haskins laboratory (http://www.uchsc.edu/immuno/faculty.html) [14]; receptor 4.1 [15] and receptor 8.3 available from Pere Santamaria (http://www.med.ucalgary.ca/webs/irgtraining/home/) [15,16]}, by inducing transgenic expression of a non-native antigen on the b cells, for which a high frequency of reacting T cells can be generated {e.g. the nucleoprotein of lymphocytic choriomeningitis virus available from Matthias Von Herraths laboratory (http://www.liai.org/) [17]}, or by transgenic expression of both a TCR and a target antigen {e.g. the INS-HA TCR against inuenza hemagglutinin under control of the rat insulin promoter available from Harald von Boehmer (http:// www.dana-farber.org/res/departments/cancerimmunology/) [18]}. The inherent simplicity of these models can offer distinct advantages over the NOD mouse in the study of accessory factors (such as molecular mimicry, effects of intercurrent infection, accessory molecules and cytokines) in the pathogenesis of disease. Because the TCRs involved are either CD4 or CD8 associated, it is possible to study the roles of these T cell subsets in isolation or in carefully titrated combinations. In several other transgenic mouse models, insulin insufciency is observed following transgenic expression (usually driven by the rat insulin promoter) of proteins in b cells in the absence of any other genetic or immunological manipulation. These particular models are probably artifactual because they do not show evidence of autoimmune causality.

BioBreeding rats
Like NOD mice, BioBreeding diabetes-prone (BBDP) rats spontaneously develop anti-islet autoantibodies and T-celldependent, ketosis-prone autoimmune diabetes. BBDP rats are available from Taconic (http://www.taconic.com/). A major advantage of BBDP rats over NOD mice is their larger size, which facilitates physiological experiments (e.g. draining of lymphatic ducts, intravital monitoring). Disadvantages include the relatively reduced number of immunological reagents (such as monoclonal antibodies and recombinant cytokines), the lack of genetically manipulated (gene-targeted mutant and transgenic) stocks and the presence of a severe T cell lymphopaenia. Comparison with the related, but non-lymphopaenic, BioBreeding diabetes-resistant strain revealed that the lymphopaenia in BBDP rats was largely attributable to a lack of regulatory T cells expressing the ART2 marker. Adoptive transfer of ART2+ cells into BBDP rats prevents the onset of diabetes. An interesting quirk to this model is that diabetes can be induced in BBDR rats by several treatments, including infection with Kilham rat virus and immune activation with polyinosinic-polycytidylic acid [poly(I:C)] (see [22] and references therein).

Streptozotocin-induced diabetes
Streptozotocin is a broad-spectrum antibiotic with diabetogenic properties mediated by direct b cell cytotoxicity [available from Sigma-Aldrich (http://www.sigmaaldrich.com)]. A single dose of 200 mg kg1 results in near-complete elimination of insulin production in a broad range of species, including mice. This model has no major autoimmune features, but is used widely to model the complications of disturbed glucose homeostasis associated with insulin insufciency. Prior to the introduction of the NOD mouse and BioBreeding (BB) rat, a model of diabetes induced by repeated administration
www.drugdiscoverytoday.com

Virally induced mouse models

Some viruses can induce insulin-dependent diabetes in wildtype (i.e. not genetically manipulated) mice. An example is the encephalomyelitis virus variant D, which causes islet destruction by tropism and direct cytotoxic effects. These models do not involve a major autoimmune component (reviewed in [19]).

453

Drug Discovery Today: Disease Models | Autoimmune diseases and inammation

Vol. 1, No. 4 2004

of lower doses (3040 mg kg1) was commonly used, in the belief that the associated islet inltrates indicated an autoimmune origin for the ensuing b cell loss. Although this model has been largely discredited as a model of autoimmunity, it is still occasionally used in species for which other models are not available [23].

(http://sdmc.i2r.a-star.edu.sg:8080/$vladimir/)]. Predictions were validated by testing synthetic peptides for their binding to DR4 and their ability to stimulate T-cell proliferation in humans at risk for diabetes.

Conclusions
The best models of the complex aetiology of type 1 diabetes are, in order, the NOD mouse and the BB rat. These models are also useful for examining aspects of the pathogenesis of the disease, although many of the transgenic models also make important contributions. Diabetes induced by nonautoimmune processes (such as poisons or viral infection) is only really useful for modelling the complications of hypoglycaemia. Most in vitro and in vivo models address much more limited questions and are usually adapted to suit the exact requirements of the investigators. The major exceptions to this generalisation are the ongoing attempts to model whole individuals in silico, which to date have not successfully simulated immune processes. The question of how well these models relate to human disease remains an open one. Generally, the study of type 1 diabetes in humans is complicated by the inaccessibility of the affected tissues and a poor association between pathological changes in peripheral blood and those occurring in the islets. Modelling the disease, therefore, has its greatest value in determining the avenues of prediction and intervention that are most worthy of attention in clinical studies.

In silico models
Computer modelling of autoimmune processes, and indeed immune responses generally, is currently in its infancy [24]. Perhaps the most comprehensive model of the biochemical pathophysiological processes associated with type 1 diabetes published to date is the Archimedes diabetes model [25]. This model consists of a network of interrelated variables (e.g. plasma insulin level, glucose uptake by muscle and glucose production by liver) linked by differential equations describing the nature of the interactions between each variable. The variables used, and the existence of interactions between them, are selected by the investigators on the basis of current knowledge of biological systems. Output variables can be extremely complex, and include severity of symptoms and the presence of vascular complications. The model requires training on clinical data sets to derive each differential equation. It then iteratively generates a pool of virtual individuals with specic clinical and biochemical characteristics, from which subsets can be selected according to clinical trial criteria. The model provides good correlation with the results of data sets obtained from independent clinical studies. As currently congured, Archimedes does not mimic immune processes but, in principle, these too could be incorporated. An alternative strategy is the application of a generalized linear model in which outcomes are dened and explanatory variables (risk factors or causes) that might or might not be relevant to the disease processes are nominated [26]. Training on extensive data sets is needed to generate the linear predictor, an equation that contains the explanatory variables, their weightings and arithmetic terms that combine them in additive or synergistic ways, and the transforming link function which maps the output of the linear predictor to an appropriate distribution. An important advantage of this model is that it is not necessary to determine which explanatory variables are important or to understand pathological processes to model them. A major disadvantage is the large size of the data sets required for training. Generalized linear modelling is now commonly used and implemented in many statistical software packages, including SAS (http:// www.sas.com/). Articial neural networks have been used to model individual components of autoimmune processes. For example, Honeyman et al. [27] created a model to predict peptides that bind to the diabetes-associated MHC class II molecule HLADR4(*0401) to predict T-cell epitopes of the autoantigen tyrosine phosphatase IA-2 [available from Vladimir Brusic
454
www.drugdiscoverytoday.com

References
1 Sutherland, D.E. et al. (1989) Recurrence of disease in pancreas transplants. Diabetes 38 (Suppl. 1), 8587 2 Lampeter, E.F. et al. (1993) Transfer of insulin-dependent diabetes between HLA-identical siblings by bone marrow transplantation. Lancet 341, 12431244 3 Hamaguchi, K. et al. (1991) NIT-1, a pancreatic b-cell line established from a transgenic NOD/Lt mouse. Diabetes 40, 842849 4 Rush, B.T. et al. (2004) Preservation of human pancreatic islet in vivo function after 6-month culture in serum-free media. Transplantation 77, 11471154 5 Augstein, P. et al. (2004) Fas ligand down-regulates cytokine-induced Fas receptor expression on insulinoma (NIT-1), but not islet cells, from autoimmune nonobese diabetic mice. Endocrinology 145, 2747 2752 6 Miller, G.G. et al. (1987) Insulin-specic human T cells. Epitope specicity, major histocompatibility complex restriction, and alloreactivity to a diabetes-associated haplotype. J. Immunol. 139, 36223629 7 Wong, F.S. et al. (1996) CD8 T cell clones from young nonobese diabetic (NOD) islets can transfer rapid onset of diabetes in NOD mice in the absence of CD4 cells. J. Exp. Med. 183, 6776 8 Arden, S.D. et al. (1996) Imogen 38: a novel 38-kD islet mitochondrial autoantigen recognized by T cells from a newly diagnosed type 1 diabetic patient. J. Clin. Invest. 97, 551561 9 Wicker, L.S. et al. (2004) Fine mapping, gene content, comparative sequencing, and expression analyses support Ctla4 and Nramp1 as candidates for Idd5.1 and Idd5.2 in the nonobese diabetic mouse. J. Immunol. 173, 164173 10 Todd, J.A. et al. (1988) A molecular basis for MHC class II-associated autoimmunity. Science 240, 10031009 11 Bonifacio, E. et al. (2001) International Workshop on Lessons From Animal Models for Human Type 1 Diabetes: identication of insulin

Vol. 1, No. 4 2004

Drug Discovery Today: Disease Models | Autoimmune diseases and inammation

12 13 14

15 16

17

18

but not glutamic acid decarboxylase or IA-2 as specic autoantigens of humoral autoimmunity in nonobese diabetic mice. Diabetes 50, 24512458 Couzin, J. (2003) Clinical trials. Diabetes brave new world. Science 300, 18621865 Katz, J.D. et al. (1993) Following a diabetogenic T cell from genesis through pathogenesis. Cell 74, 10891100 Pauza, M.E. et al. (2004) T-cell receptor transgenic response to an endogenous polymorphic autoantigen determines susceptibility to diabetes. Diabetes 53, 978988 Verdaguer, J. et al. (1997) Spontaneous autoimmune diabetes in monoclonal T cell nonobese diabetic mice. J. Exp. Med. 186, 16631676 Verdaguer, J. et al. (1996) Acceleration of spontaneous diabetes in TCR-btransgenic nonobese diabetic mice by b-cell cytotoxic CD8+ T cells expressing identical endogenous TCR-a chains. J. Immunol. 157, 4726 4735 von Herrath, M.G. et al. (1994) How virus induces a rapid or slow onset insulin-dependent diabetes mellitus in a transgenic model. Immunity 1, 231242 Sarukhan, A. et al. (1998) Changes in function of antigen-specic lymphocytes correlating with progression towards diabetes in a transgenic model. EMBO J. 17, 7180

19 Jun, H.S. and Yoon, J.W. (2003) A new look at viruses in type 1 diabetes. Diabetes Metab. Res. Rev. 19, 831 20 Wen, L. et al. (2000) In vivo evidence for the contribution of human histocompatibility leukocyte antigen (HLA)-DQ molecules to the development of diabetes. J. Exp. Med. 191, 97104 21 Taneja, V. and David, C.S. (1998) HLA transgenic mice as humanized mouse models of disease and immunity. J. Clin. Invest. 101, 921926 22 Greiner, D.L. et al. (2001) Translating data from animal models into methods for preventing human autoimmune diabetes mellitus: caveat emptor and primum non nocere. Clin. Immunol. 100, 134143 23 Rolandsson, O. et al. (2002) Streptozotocin induced diabetes in minipig: a case report of a possible model for type 1 diabetes? Autoimmunity 35, 261 264 24 Brusic, V. and Petrovsky, N. (2003) Immunoinformatics the new kid in town. Novartis Found. Symp. 254, 313 25 Eddy, D.M. and Schlessinger, L. (2003) Archimedes: a trial-validated model of diabetes. Diabetes Care 26, 30933101 26 Baxter, A.G. (2001) Modelling the effects of genetic and environmental factors on the risk of autoimmune disease. J. Autoimmun. 16, 331335 27 Honeyman, M.C. et al. (1998) Neural network-based prediction of candidate T-cell epitopes. Nat. Biotechnol. 16, 966969

www.drugdiscoverytoday.com

455