Chapter 10

Humoral Autoimmunity

(Updated 9/13/2012)


L. Yu and G.S. Eisenbarth




A little more than 2 decades ago, screening for autoantibodies associated with Type 1A (immune mediated) diabetes was limited to measuring cytoplasmic islet cell autoantibodies (ICAs) and insulin antibodies 1, 2. Currently, multiple sequenced autoantigens have been defined and recombinant autoantibody assays3-7 are available 8, 9  including assays for insulin 10, glutamic acid decarboxylase (GAD) 11, ICA512/IA-2 12-14, I-A2 beta (phogrin) 15, 16, and the islet zinc transporter, ZnT8 autoantibodies 17, 18.   Only autoantibody assays to these proteins are of confirmed utility in diagnosing and predicting Type 1A (autoimmune diabetes) and for which laboratories participating in blinded workshops organized by the Immunology of Diabetes Society submit measurements.  A long list of additional autoantigens with autoantibody assays exist (see below) but in general we believe given lack of implementation one should assume that these other proposed assays lack sensitivity and or specificity, and the proposed targeted antigen is not sufficiently Type 1A diabetes associated to be of diagnostic utility.  In addition to identifying the appropriate target molecules as the basis for Type 1A diabetes autoantibody assays, the assay format is also crucial.  Fluid phase radioassays have historically provided the best sensitivity and specificity but this may be changing.  Standard ELISA assays where one simply binds a given autoantigen to a plate and detects binding of autoantibodies to the plate bound antigen in general have lacked requisite specificity when large numbers of samples are analyzed.  A modified capture ELISA format where autoantibody cross-links a plate bound antigen to the same labeled fluid phase antigen for GAD65, IA-2, and ZnT8 have approached the specificity and sensitivity of fluid phase radioassays.  In addition two additional formats (luciferase based assays19 and electrochemiluminescent20) are being studied.   The electrochemiluminescent assay format has been applied to the insulin autoantibody which is the most difficult for laboratories to implement 20, 21 with a marked increase in sensitivity above that of even radioassays as evaluated in the last two international Immunology of Diabetes workshops (DASP and IASP).

Just amongst the four major islet autoantibody assays, insulin, GAD65, IA-2, and ZnT8 there are major differences.  Insulin autoantibodies are almost always the first autoantibody to appear in children followed from birth (either as an isolated autoantibody or with other autoantibodies)22.  The levels of insulin autoantibodies (and only insulin autoantibodies) correlate with the rate of progression to onset of overt diabetes from the time of first appearance of an islet autoantibody6, 23.  Thus the youngest onset children who must progress rapidly to be onset at a young age, at onset characteristically have very high levels of insulin autoantibodies.  The levels of insulin autoantibodies are not simply related to age, as in general the levels are very low for the whole prediabetic period in young autoantibody positive children who take a long time (e.g. decade) to progress to diabetes(Figure 10.1 below).  A recent report suggests that the prevalence and levels of GAD and IA-2 autoantibodies at diabetes onset have increased between 1985 and 200224.


Figure 10.1.  Child followed from birth in DAISY study till development of overt diabetes.  Autoantibodies were present prior to two years of age with more than a dozen years till onset of hyperglycemia.  Characteristically for children who slowly progress to overt diabetes is the lack or low levels of insulin autoantibodies.


The most specific islet autoantibodies react with IA-2 and ZnT8 18, 25 and for these two autoantibodies there is characteristically a very strong signal to noise.  GAD65 autoantibodies are the most common amongst adult onset patients, but GAD65 autoantibody assays frequently have a 3-5% positivity rate in controls 26 making definitive diagnosis of Type 1A diabetes more difficult if GAD is the only autoantibody present (e.g. diagnosis of Latent Autoimmune Diabetes of Adults) if only GAD65 autoantibodies are present.  Though insulin autoantibodies are present in almost all children followed to diabetes, they are frequently negative at onset of diabetes in teenagers and adults. 

Expression of multiple autoantibodies of insulin, GAD, IA-2 and/or ZnT8 is associated with extreme risk of progressing to overt Type 1A (immune mediated) diabetes25, 27, 28.  Given 20 years of follow up almost all children (whether relatives of patients with Type 1 diabetes or individuals from the general population) who express multiple biochemical autoantibodies progress to diabetes.  If there is only a single autoantibody, which are usually insulin autoantibodies or GAD autoantibodies risk of diabetes is low, often less than 5%25, 29.  As the number of autoantibodies increase the overall risk of diabetes increases29.  The usual explanation for the general rule relating multiple autoantibodies to extreme risk is that epitope spreading with the targeting of different autoantigens directly enhances beta cell destruction.  Though insulin autoantibodies often appear first in prediabetic children followed from birth, the order of autoantibody appearance is very variable.  After initial appearance with either multiple or a single autoantibody relatively rapidly multiple autoantibodies are present which persist till the development of diabetes.  It is doubtful that there is one to one correlation between autoantibodies and T cell targeting of specific islet antigens.  In particular with only the level of insulin autoantibodies related to the rate of progression, subjects with low or negative insulin autoantibodies,  individuals may still be critically targeting insulin at a low level consistent with delayed progression.  Rate of the process needs to be distinguished from risk and it is easiest to do this by only analyzing individuals who have all progressed to overt diabetes with prospective follow up.  A life table mixing individual with and without progression to diabetes reflects both rate and risk23.

An alternative hypothesis for the >=2 autoantibody rule, given that Type 1 diabetes related autoimmunity is relatively rare (e.g. 1/300 in general population developing diabetes) is statistical. By Baye’s theorem if one is measuring four autoantibodies (each assay with 1% false positive rate [which only the best international assays achieve in Immunology of Diabetes Society workshops]) the great majority of positives in general population screening are likely to be “false” positives in terms of development of diabetes.  A simple way to increase testing specificity is to utilize multiple biochemical autoantibody assays with definition of positive as the presence of >=2 of the four autoantibodies (binomial equations).  The combinatorial approach rather than simply asking for all four autoantibodies to be present (four autoantibodies present predicted specificity =(.01)4 or >99.99999%) allows one to preserve sensitivity with a decreased but impressive specificity27.  False positives in the general sense for islet autoantibody results can result from assay problems (e.g. we have observed an individual with autoantibodies reacting only with iodinated insulin which does not exist in nature but I-125 iodinated insulin is utilized for the insulin autoantibody radioassay).  False positives can also result from true autoantibodies which have low positive predictive values (e.g. “biologic” false positives, low affinity insulin autoantibodies20, 22, ELISA assays for insulin autoantibodies not detecting diabetes related autoantibodies10, etc.).  Finally false positives may result from non-antibody serum components that bind or precipitate labeled autoantigens. 


Once an individual is identified with islet autoantibodies, follow up for deterioration of glycemic control is important as it is clear from multiple prospective studies that ketoacidosis with both its attendant morbidity and mortality is preventable30 especially in children less than age 2.

In addition to the highlighted major islet autoantigens, there are many other autoantigens in a variety of stages of characterization, including molecules with characterized sequences but without fully developed assays and proteins with only known molecular weights of 155 kd 31, 32, 52 kd 33, and molecules recognized by T lymphocytes for which autoantibodies have not been described such as chromagranin A  and IGRP 34, 35 and RegII 36. Other molecules have been described but their association with Type 1A diabetes has either not been evaluated in man, has not been substantiated, or assay sensitivities are low, or follow-up studies have not been published. These molecules include carboxypeptidase H,  anti-bovine serum albumin antibodies (anti-BSA) 37, antibodies reacting with ICA69 38-40, anti-insulin receptor antibodies 41, antibodies to heat shock proteins 42-44, anti-topoisomerase II 45 anti-ganglioside 46, lysophospholipids 47and GLIMA38 a membrane glycoprotein 48and antibodies to a series of autoantigens identified by screening islet libraries such as ICA12 49, 50. Finally, a subset of autoantibodies termed anti-islet cell surface antibodies are currently rarely measured as most recent studies have failed to demonstrate disease specificity 51.  Recently described autoantigens include osteopontin 52, importin 53, antibodies reacting with “peri-islet Schwann cells/ GFAP/S100beta (glial fibrillary acidic protein)” surrounding islets 54, densin and filtrin 55 and antibodies reacting with CD38 56.  There are almost certainly additional specificities awaiting discovery.

Table 10.1: Subset of Biochemically Characterized Autoantigens






Higher sensitivity young children/rapid progressors

 GAD (Glutamic acid Decarboxylase)


Higher sensitivity adult onset type 1A



Islet Zinc Transporter



Tyrosine Phosphatase like molecule



Tyrosine Phosphatase like molecule

 Carboxypeptidase H





amphiphilic membrane glycoprotein


 Despite the importance of autoantibodies for disease prediction 28, it is likely that anti-islet autoantibodies do not by themselves cause the beta cell destruction that leads to Type 1A diabetes.  Nevertheless B lymphocytes are important for the development of Type 1 diabetes and results of the TrialNet studies of anti-B cell antibodies (anti-CD20) in new onset diabetes demonstrated slowing of loss of C-peptide.  In the NOD mouse, anti-CD20 antibody treatment decreases development of diabetes 57. In addition, Naji and coworkers have demonstrated the importance of transplacental autoantibodies for progression to diabetes of NOD mice 58.

The most cogent evidence of lack of direct damage by autoantibodies is the lack of diabetes in infants born to antibody positive mothers, or women who developed Type 1A diabetes during pregnancy. Autoantibodies (ICA, anti-insulin, and anti-GAD autoantibodies) are readily detected in the serum of infants of such antibody-positive mothers in the absence of diabetes in the child. It will take careful epidemiological studies to determine if specific autoantibodies (e.g. spontaneous and not insulin therapy induced insulin antibodies) predispose children to develop type 1 diabetes. A report by Ziegler and coworkers indicates that the presence of anti-islet autoantibodies at birth (GAD65 and IA-2 autoantibodies, but not insulin autoantibodies) of offspring of mothers with type 1A diabetes was associated with a decrease 59 in the development of anti-islet autoantibodies for children followed in the BabyDiab study 60. The five year risk of developing anti-islet autoantibodies and diabetes was 1.3% and 1.1% for offspring with newborn anti-islet autoantibodies versus 5.3% and 3% respectively of newborns who were negative for anti-islet autoantibodies. Of note, a child with genetic B cell deficiency and no antibodies progressed to type 1 diabetes 61. On the other hand studies in the NOD mouse indicate that both B lymphocytes and transplacental autoantibodies (insulin autoantibodies detected) greatly increase the development of diabetes 57-59, 61-63.  As illustrated in the Figure 10.2 below, if the mother NOD mouse was unable to make autoantibodies, even if "antibodies" (e.g. transgenic anti-Hen Egg Lysozyme antibody gene) could be produced, the development of diabetes in offspring was greatly decreased.

humoral naji antibodynod transplacental

Figure 10.2 Development of diabetes is greatly reduced if maternal autoantibodies are not present during pregnancy, presumably due to a lack of transplacental autoantibodies. HEL=Transgenic producing antibodies to Hen Egg Lysozyme; KO=IgM knockout; DBA/2 strain of mouse used as foster mother; SCID=Severe Combined Immunodeficient Mother

General Assay Methodology

There are multiple assay formats for the detection of disease associated autoantibodies, and similar to many fields of autoimmunity, the first useful assay for islet autoantibodies utilized indirect immunofluorescence with frozen sections of human pancreas as substrate (the ICA assay-see below) 1.  The other major format for determination of islet autoantibodies consists of variations on the theme of fluid phase assays including recently described luciferase immunoprecipitation assays 19 and electrochemiluminescence assays where the autoantibodies couple a biotinylated autoantigen and fluorophore labeled autoantigen20. In fact the methodology used for the discovery of the radioassay by Berson and Yalow, utilizing insulin antibodies produced by patients treated with insulin, was the basis for the discovery of insulin autoantibodies 2, 64. For most autoimmune disorders and for basic immunologic research, ELISA assays employing plate bound antigen are the standard. Multiple workshops utilizing sera from patients with type 1 diabetes and multiple mouse strains have demonstrated that standard ELISA formats lack both sensitivity and specificity compared to fluid phase radioassay formats. For insulin autoantibodies, the ELISA formats were able to detect high capacity antibodies following insulin immunization, but not the insulin autoantibodies of prediabetic individuals 10. This inability of human insulin autoantibodies to bind to plate bound insulin is specific to human autoantibodies as the insulin autoantibodies of NOD mice can readily be detected in a plate bound ELISA format and we have described a highly sensitive and specific assay for such autoantibodies, equivalent to radioassays 65.  We believe insulin bound to plastic plates obscures the limited key determinant(s) recognized by human anti-insulin autoantibodies.

In addition when one analyzes thousands of individuals with ELISA formats there is always a subset of sera that react with the wells of the plate itself or unique epitopes created by plate bound proteins. A small percentage of such sera (e.g. 1%) in studies such as DAISY (Diabetes Autoimmunity Study of the Young) where individuals are sampled on multiple occasions over time, would result in enough false positives to invalidate efforts to discover environmental, immunogenetic, and diabetes predictive parameters 22, 66. It is possible to develop modified ELISA like assays that utilize fluid phase reactivity of antigen and antibody(see discussion of method below).  In international workshops an assay for GAD65 autoantibodies (company RSR developed assay and distributed by Kronus) has attained a level of specificity and sensitivity equivalent to the radioassays, though in the most recent 2009 DASP workshop multiple laboratories using this modified ELISA had excellent sensitivity but only 95% specificity (Mueller oral communication).  For many applications, 5% false positives would be problematic. 

A disadvantage of the current fluid phase radioassays is that they utilize low levels of radiation (usually 125I, 35S -methionine, and 3H-leucine)(Figure 10.3). They are however often performed in a format in 96-well format that is as convenient as the ELISA format 67-69. It is likely that many of the clinical assays (but not all, e.g. transglutaminase) that utilize ELISA formats for other autoimmune disorders are utilizing assays with compromised specificity and sensitivity, and the assay format development has just not been optimized and directly compared with modern fluid phase radioassays (e.g. Farr assay versus, protein A based assay, versus ELISA assay for anti-DNA autoantibodies) 70.

iaa assay

Figure 10.3    General outline of fluid phase 96-well plate assays for autoantibodies. In above example 125I  -insulin utilized, but assay format identical for GAD65, ICA512 assays.

Most investigators assay GAD65 and IA-2 anti-islet autoantibodies in a high throughput 96-well format  (Figure 10.3)where labeled antigen is incubated with patient sera, and then both are placed in 96-well filtration plates, where a "bead" (e.g. sepharose) with coupled protein A and or protein G is added, and bound from free radioactivity is separated by filtration washing, and then scintillation fluid is added directly to the 96-well filtration plates, and counting is performed on multichannel beta counters able to handle the plates 58. Even the assay for insulin autoantibodies utilizing 125I  insulin is performed in the same manner, detecting with beta counting emission from 125I  (Figure 10.3). A major advance was the simple production of labeled autoantigen by in vitro transcription and translation of cDNAs to produce the label for the fluid phase radioassays. For example, we have utilized a combined GAD65 and ICA512bdc radioassay in which GAD65 is labeled with 3H-leucine and ICA512(IA-2) is labeled with 35S -methionine.    It has been our experience in setting up multiple such autoantibody assays that approximately 2/3 of the time an assay using such in vitro labeled autoantigen works, and if the assay does not work on the first try, it is unlikely to be modified to work. At present we routinely determine GAD65, ICA512bdc, IA-2ic, IA-2 full length, ZnT8, 21-hydroxylase and transglutaminase autoantibodies with this methodology 71-75. Such assays are not useful if post-translational modifications, or particular folding of the protein is essential that is not reproduced in the in vitro production of the antigen. With this in vitro translation and transcription methodology minimal protein preparation is needed and following the kit generation of the labeled product we simply perform size separation to produce the labeled antigen.  Two harmonized NIDDK assay have been created (for GAD65 and IA-2 autoantibodies) which depended in part on using the same standard sera as well as identical reagents. 76

There are a number of modifications proposed and implemented for the determination of defined islet autoantibodies that can detect autoantibodies with various degrees of sensitivity and specificity relative to the best standard fluid phase radioassays 77-80. In particular the GAD65 autoantibody described by Smith and coworkers 79 utilizes a novel ELISA format in which a low concentration of the GAD antigen on the plate captures the autoantibody, and then biotinylated GAD in the fluid phase is added and is captured by the second binding site of the autoantibody, and it is the biotinylated GAD65 that is detected to produce the non-isotopic signal. This assay performed well in the previous Immunology of Diabetes/CDC DASP workshop, but a similar format for IA-2 autoantibodies did not have an equivalent sensitivity to the standard fluid phase radioassays(oral discussion at IDS-Cambridge and in current 2009 DASP specificity for multiple ELISA assays was only approximately 95%).   There is also now a similar ZnT8 commercial assay kit.  Human insulin autoantibodies can be measured with an electrochemiluminescent (ECL) insulin/proinsulin assay20.  A reported ECL insulin autoantibody assay detected antibodies of NOD mice despite binding of insulin to plate81.  I believe it is unlikely for this assay to detect prediabetic insulin autoantibodies in that even though NOD insulin autoantibodies react with plate bound insulin, prediabetic IAA do not