Studies in a variety of animal models, which act as replicas of the major chronic inflammatory diseases that affect humans, have offered many answers to these questions. One of the clearer outcomes is that delivery of autoantigens, administered at different disease stages via a variety of routes, can provide robust, sustained health and protection from inflammatory autoimmune disease. The most appealing element to this approach, termed antigen-specific immunotherapy (ASI), has been that it not only provides an effective means of controlling the autoimmune response via induction or restoration of β-cell–specific tolerance, but that it may achieve these goals without major concerns over safety and certainly without the specter of immune suppression. Yet significant questions remain. Are we doing enough to realize the potential of this sacred cow? How do we move from concept to reality?

In this article, we provide an update on the mechanisms through which ASI is currently thought to operate. We discuss why, despite this body of knowledge, alternative, non-ASI approaches have emerged as the current vogue. We argue that more should be done to counter this trend and realize the potential of ASI, including strategies that combine its strengths with those of other complementary ways forward.

Studies in mice have clearly documented the enhancement of Tregs and the skewing of cytokine responses (immune deviation) after mucosal (oral, nasal) or peripheral (peptide with and without adjuvant, DNA vaccination) insulin, proinsulin, or other autoantigenic peptide administration (8,9). In these experiments, repeated administration of β-cell autoantigens, most notably insulin or its peptides, led to increased interleukin (IL)-4, IL-10, and transforming growth factor (TGF)-β production by insulin-specific polyclonal T-cell populations isolated from the spleen, pancreatic lymph nodes, and, in some cases, the islets. Unfortunately the numbers of autoreactive T-cells in the blood are low and, therefore, little information is available about their frequency and function in peripheral blood, which would be very helpful to guide human biomarker efforts. The antigen-induced or enhanced cell populations can actively suppress effector immune responses as evidenced by adoptive transfer studies. Unless they are enabled by certain chemokines or chemokine receptors or integrins to home to solid organs, the antigen-induced modulating cells are usually found in lymph nodes and spleens following transfer into pre-diabetic recipient mice. Thus, antigenic immunization can endow islet-reactive T-cells with the ability to regulate and suppress deleterious effector responses, which entitles them to be called adaptive regulatory T-cells (aTregs). In some cases, the transcription factor forkhead box P3 (FoxP3), a good marker for naturally occurring Tregs (nTregs), shows increased expression. FoxP3 is of value as a bona fide Treg marker in the mouse, but its utility in humans is much more limited because it is also expressed on recently activated effector T-cells. Thus, successful ASIs that prevent type 1 diabetes in animal models are associated with the induction of cytokines, which can be considered as protective from type 1 diabetes (immune deviation) and are produced by CD4+ T-cells that can function as aTregs. In this context, the difference between immune deviation and Treg induction is mainly a semantic argument. The control of effector responses of various specificities (bystander suppression) likely occurs through modulation of antigen-presenting cells (APCs) resulting in a lack of anti-islet effector T-cell expansion, but not their deletion. New alternative models have also been described recently in which APCs are perhaps incapacitated from their role of antigen presentation to autoreactive effector T-cells through a direct APC-lytic process mediated by Tregs both in vitro in man and in vivo in murine studies (10,11).

Similarly, immune deviation indicative of Treg generation has been documented in humans (for example, the induction of a proinsulin peptide–specific IL-10 response after low dose intradermal administration of the peptide in type 1 diabetic patients) (12), reminiscent of observations in studies of peptide immunotherapy in clinical allergy (13,14). Other examples are the significant increases in GAD-specific γ-interferon, IL-5,-13,-10,-17,-6, tumor necrosis factor-α, FoxP3, and TGF-β mRNA responses as well as autoantibody induction (15) following subcutaneous administration of 20 μg of recombinant human GAD65 adsorbed onto alum (in a classic prime-boost regimen, n = 35/group). In other clinical trials where the limited effects of ASI have been documented, for example after daily dosing of oral insulin (16) or after a proinsulin-expressing DNA vaccine (17), no such clear effects as of yet have been documented and will have to await future biomarker studies.

There are numerous explanations for this evident bias toward evaluation of novel non-ASI reagents at the intervention stage (Table 2).

The picture in relation to ASI and disease acceleration in nonclinical studies is generally reassuring. For example, an extensive analysis of the literature in relation to the nonobese diabetic (NOD) model of spontaneous autoimmune diabetes (in which, for example, approximately 100 published studies since 1996 have involved injection or ingestion of whole or peptide autoantigens, either as simple solutions or in conjunction with powerful adjuvants) has revealed that it is extremely unusual to accelerate disease; in most cases the maneuvers are protective or have no effect. In one reported study, disease was accelerated. In that case, two peptides of the β-cell autoantigen GAD65 were administered intrathymically and caused a mild acceleration of diabetes onset in NOD mice (19). The authors pointed out that immunization with whole GAD65 does not induce CD4 T-cells reactive with either peptide used, implying that these are cryptic epitopes not naturally processed and presented by NOD mouse APCs. The route of administration may also be important in this outcome—our own studies injecting a single dose of one of these cryptic GAD65 epitopes intraperitoneally gave a strong protective effect (20). The use of naturally processed and presented epitopes or whole antigens may be an important safeguard against the danger of priming additional T-cells with cryptic peptides. Other examples of disease exacerbation by autoantigen administration in models of autoimmune demyelination (21) and autoimmune diabetes (22) have also been reported, but these appear to be rare occurrences. In clinical studies, the experience is obviously much less extensive. It is notable that in the study of oral insulin administration to first-degree relatives with insulin autoantibodies, the subgroup analysis of those who did not have confirmed insulin autoantibodies ≥80 nU/ml suggested a trend toward a detrimental effect of the treatment (16), reminding us that we may need to develop carefully argued selection algorithms for trial entry. The examples of disease exacerbation in multiple sclerosis (MS) using altered peptide ligands of the autoantigen myelin basic protein is yet another strong argument for using native peptide sequences or whole antigens (23,24). Apart from these examples, there has been no evidence of disease exacerbation from ASI trials in rheumatoid arthritis (25,26) or more recent studies of MS patients using native sequence peptides (27).

A second safety consideration raised by published nonclinical studies is the risk that antigen injection can result in hypersensitivity, systemic allergy, and anaphylaxis. The most relevant report here was of fatal anaphylaxis in NOD mice after repeated injections of an immunodominant peptide of insulin (residues B9–23). However, relatively large quantities (total >1 mg) and repeated (seven times) dosing of peptide were used in these studies (28). These responses may have been idiosyncratic to this strain and were alleviated by the alteration of the peptide's isoelectric point (29). Such responses have not been seen to date in the setting of autoimmune disease in man (apart from the altered peptide ligand studies in MS discussed above, in which hypersensitivity responses were also observed). It may be that the induction of T helper 2 (T_H2)-like responses requires the achievement of a fine balance between those that may be beneficial (see the comments on immune deviation above) and those that are dangerous.

The final safety issue relates to the induction de novo of an autoimmune process, for example, when an ASI study uses a β-cell autoantigen that is expressed in other tissues as in the case of GAD65, which is expressed in the peripheral and central nervous systems. To date, the nonclinical and clinical experiences suggest that such a complication remains only a theoretical risk; notably there was no induction of neurological disease despite administration of GAD65 with adjuvant in a regime that boosted GAD-specific autoantibody titers to levels more typically seen in patients with stiff-man syndrome (15).

In addition, several other factors need to be resolved. The most pressing issue is the precise dosing regimen. From the studies in various animal models, it is known that too high dosages might not be effective by leading to the deletion of Tregs rather than their augmentation. In murine models, for example, only oral insulin dosages between 0.2 and 2 mg are effective when given twice per week by oral gavage (31). Higher and lower dosages have no strong effect on preventing diabetes, therefore there is a strong need to translate dose regimens used in these animal models to humans as accurately as possible. This has not been achieved to date, at least in part because there is no fully validated formula. However, the currently utilized 7.5-mg dose in the oral insulin study of Diabetes TrialNet, which mirrors that used in the Diabetes Prevention Trial (DPT)-1 (16), is most likely too low comparatively and, based on animal models, less frequent dosing with a higher dose should greatly increase efficacy. This factor could also explain the lack of efficacy in the Finnish nasal insulin diabetes prevention trial (39).

Notwithstanding these fertile new areas, there is a sense that the progress made in getting antigens into the clinic has stalled and requires renewed invigoration. There is nothing intrinsically flawed about oral or nasal antigen administration or antigen injection. They have simply not been fully evaluated in a staged developmental program. As discussed above, much remains to be understood about dose, regimen, and route. We would advocate a return to these questions, addressed in the context of small clinical studies with the emphasis on mechanistic outcomes. A second line approach will be the development of suitable combinations of antigens with immune modulators that have been specifically selected to foster Treg function and expansion while reducing the effector cell load. Such an initiative could be very beneficial in overcoming some of these issues and, most importantly, enabling antigen-specific Treg induction to be effective later during the disease process, for example in individuals at high risk of developing type 1 diabetes or in recently diagnosed patients (35). Animal studies using a combination of anti-CD3 and nasal proinsulin peptide strongly support this concept (36).