Type 1 Diabetes: Current Advances in High-Throughput Technologies and Computational Biology for Biomarker Studies

Abstract

Biomarkers are essential for the identification of high-risk populations as well as the monitoring of preventive and therapeutic outcomes for type 1 diabetes (T1D). In this chapter, we will discuss the progress made in T1D biomarker discovery using high throughput genomic, transcriptomic, and proteomic technologies collectively called as omic technologies. We also discuss the potential of artificial intelligence and omics data in the early prediction of T1D. Readers will gain an overview of the status of T1D biomarkers based on omic technologies. High throughput omic technologies combined with computational biology offer great opportunities for biomarker discovery. As we move forward, the utilization of a biomarker panel for the prediction and prevention of T1D is needed.

1. Introduction

Type 1 diabetes (T1D) is an autoimmune disease that results from the immune-targeted destruction of insulin-producing β-cells in the pancreas [1, 2]. The disease does not discriminate based on age, sex, or race, making it devastating on a global scale. Public health officials approximated that in 2017, 451 million adults were treated with diabetes across the globe [3]. Estimating that by the year 2045, approximately 693 million patients globally will have diabetes; however, only half will have an official diagnosis [3]. T1D results from complex interactions between genes and the environment leading to autoimmune reactions toward pancreatic islet cells (Figure 1). The unchecked immune reaction reduces the beta-cell mass, thereby causing insulin insufficiency leading to T1D in these genetically susceptible individuals.

Figure 1.

Susceptibility genes and environmental triggers in development and progression of type-1 diabetes. Created using BioRender.com.

Traditional methods for diagnosis include glucose tolerance testing and monitoring of HbA1c levels in at-risk patients. However, these diagnostic techniques are only effective after the onset of diabetes has occurred. Preventative techniques for T1D are currently unavailable due to two main factors: (1) the inability to predict and accurately assess risk for the high-risk population and (2) the etiology of the disease is not well-established [4]. Although higher prevalence is seen in families with established autoimmune diseases, current screening techniques for islet autoantibodies in high-risk first-degree relatives of T1D patients are not very effective, as most cases of T1D occur spontaneously in the general population [5]. In addition, screening large populations poses challenges in both efficacy and funding. Due to these limitations, diabetes research is moving toward personalized prevention strategies, which focus on an individual’s unique genetic and environmental risk factors that lead to the progression of T1D [2].

Biomarkers hold the key to unlocking diabetes preventative strategies and monitoring therapeutic outcomes. Over the past several decades, islet cell autoantibodies (ICA) have become an earlier predictor of T1D [2]. Since then, the discovery of other autoantibodies, such as those against specific islet autoantigens insulin and glutamate dehydrogenase (GAD), have continued to improve assays in their predictive value [2]. However, these islet autoantibodies still face the limitation of appearing in later stages of T1D development [2]. In addition, islet autoantibodies have not been useful for assessing therapeutic outcomes [2]. Early detection and prevention of T1D is crucial for the high-risk population and requires the discovery of new biomarkers. These biomarkers can come from metabolic pathways, DNA, RNA, glycans, lipids, and proteins (Figure 2). This chapter will address current biomarkers, omic technologies, computational biology, and the use of AI in identifying biomarkers in T1D research, and finally major issues with the discovery as well as validation of T1D biomarkers.

Figure 2.

Potential sources of biomarkers for T1D.

2. Current biomarkers in T1D

2.1 Autoantibodies and autoreactive T cells

The immune system develops autoantibodies (AAbs) in order to attack self-proteins. AAbs are used for T1D diagnosis, classification, and prediction of disease development. They do not play a direct role in disease pathogenesis, unlike autoreactive T cells [6]. Instead, AAbs are strong indicators of islet cell destruction [7]. They can be utilized to identify those at increased risk for T1D because they are produced during the natural history of T1D, including the stage before disease diagnosis [8].

There are five commonly tested AAb markers used in the diagnosis of T1D, which include ICA (islet-cell cytoplasmic AAb), GADA (glutamic acid decarboxylase (GAD) AAb), IA-2A (insulinoma 2 (IA-2)-associated AAb), IAA (insulin AAb), and ZNT8A (zinc transporter 8 AAb) (Table 1). At least one of the five autoantibodies is present in >95% of individuals with T1D upon hyperglycemia detection [8]. Patients with multiple autoantibody positivity are at a higher risk of developing T1D, although this risk varies depending on age, antibody type, and metabolic status [12]. However, a subset of patients who do not have any of the above AAbs at diagnosis have been found, indicating either a possibility of insensitive tests or the presence of additional T1D-associated autoantibodies in these patients.

Autoantibody	Prevalence [9]	Risk of T1D
ICA	0.681	3% [10]
IAA	0.424	13.1% [11]
GADA	0.636	12.9% [11]
IA2A	0.714	40.5% [11]
ZnT8A	0.654	55%^*, [11]

Table 1.

Autoantibodies to pancreatic antigens, prevalence rate, and risk of developing T1D in 5 years.

ICA: Intracytoplasmic autoantibody, IAA: Insulin autoantibody, GADA: Glutamate Dehydrogenase-65 autoantibody, IA2A: Insulinoma antigen-2 autoantibody, ZnT8a: Zinc transporter 8 autoantibody. *5 year risk.

Although AAbs appear before T1D diagnosis, they appear relatively late in the autoimmune process. T1D prevention may be more effective before the onset of the active autoimmune response and appearance of AAbs, suggesting that AAbs may not be effective markers to identify high-risk groups for T1D prevention [2]. Overall, AAbs are not useful as biomarkers for therapeutic outcomes and do not track disease progression [13]. Thus, it is crucial to investigate additional biomarkers that can serve as predictors of disease progression.

Autoreactive CD4+ and CD8+ T cells are thought to be the main mediators behind beta cell destruction in T1D [6, 14]. Insulitis, or leukocyte invasion of pancreatic islets, has been observed in mouse models and pancreatic biopsies of T1D patients, with infiltrates, predominately consisting of CD8+ cytotoxic T cells [15]. As autoreactive T cells play a direct role in beta cell destruction, T cell markers have the potential to provide unique insights into the pathogenesis and progression of T1D [6]. Despite their large potential, T cell biomarkers run into many challenges, making them difficult to use routinely. T cell quantification assays are limited by their lack of sensitivity [15, 16], as well as the low frequency of autoreactive T cells in peripheral blood and low avidity for peptide MHC ligands [17]. Avidity for TCR is hoped to increase through engineering the MHC peptide ligand into a multimer in recent studies [18]. Other advancements are being made in T cell biomarker identification, but further validation and standardization of the most promising biomarkers and assays are needed before widespread use [19, 20].

2.2 Genetics

Relatives of patients with T1D are at increased risk of developing T1D compared to the general population [21]. Human leukocyte antigen (HLA) has been shown to be a major genetic determinant in the development of T1D, accounting for approximately 50% of genetic risk [22, 23]. HLA class II alleles DR04, DR03, and homozygous DR04/DR03 genotypes increase the risk of T1D, while DR02 is highly protective against the disease [23]. Currently, genetic risk screening for T1D is performed by HLA-genotyping. However high-risk HLA genotypes are only present in 30–40% of T1D patients. Suggesting that HLA genotyping is too insufficient in sensitivity and specificity to be a useful T1D marker [6, 7]. While the HLA locus encodes for the strongest genetic susceptibility genes for T1D, there are five other non-HLA gene regions also associated with the disease: INS, CTLA-4, PTPN22, SUMO4, IL2RA, and IFIH1 (Table 2) [21, 27, 29].

Gene	OR	95% CI
HLA DR3/DR4	16.59 [23]	13.7–20.1 [23]
Insulin VNTR	2.4 [24]	1.7–3.4 [24]
CTLA4	1.41 [25]	1.31–1.53 [25]
PTPN22	1.83 [26]	1.284–2.596 [26]
SUMO4	1.236 [27]	1.112–1.373 [27]
GRS	AUC 0.73 [28]	0.70–0.77 [28]

Table 2.

Genes associated with T1D and odds ratio.

HLA: human leukocyte antigen, VNTR: variable number of tandem repeats, CTLA4: cytotoxic T-lymphocyte associated protein 4, PTPN22: protein tyrosine phosphatase 22, SUMO: small ubiquitin-like modifier, GRS: genetic risk score, AUC: area under the curve

Platform	Separation	Throughput
Mass-spectrometry	HPLC [46]	↑Proteins ↓Samples
	GC [47]
	CE [48]
Array	96-well [49]	↓Proteins ↑Samples
	Beads [50, 51, 52]
	NAPPA [8, 53]

Table 3.

High throughput proteomic approaches for biomarkers in T1D.

HPLC: high performance liquid chromatography, GC: gas chromatography, CE: capillary electrophoresis, Beads: Luminex bead assay, NAPPA: Nucleic Acid Programmable Protein Arrays.

Individually, non-HLA genes only weakly contribute to the assessment of risk for T1D. However, when used in combination, non-HLA genes may prove to have more predictive value [2]. Among the non-HLA genes, the INS VNTR has the strongest association with T1D [21]. The working theory is polymorphisms in the INS gene may lead to immune tolerance to insulin by changing the amount of insulin mRNA in the thymus during fetal development and childhood [30]. Similar polymorphisms have been also associated with CTLA-4 [31]. In a recent report by Ueda et al., the CT60-A/G single nucleotide polymorphism (SNP) was suggested to produce lower amounts of sCTLA-4 mRNA in T1D patients [32]. However, a recent study found serum sCTLA-4 (protein) levels to be slightly higher in T1D patients. Although contradicting the idea that CT-60 SNP controls the expression of sCTLA-4, sCTLA-4 may still be a risk factor for T1D. These observations propose that sCTLA-4 may contribute to the development of autoimmune diseases, probably through inhibiting the B7-mCTLA-4 interaction and down-regulation of T cell activation [31]. Further characterization of these genes and disease variants requires genotyping of a large number of subjects [2].

3. Omic technologies

Several recent developments in instrumentation and protocols have enabled the evaluation of individual genotypes, gene expression, and protein expression. These developments have led to the creation of high-throughput approaches, collectively known as omics technology, that allow researchers to analyze large-scale measurements of genes, polymorphisms, and proteins in a given sample. In this section, we provide a brief description of the use of individual omics technology that are used in T1D research.

3.1 Genomics

Microarray technology allows researchers to analyze gene expression on a larger scale. The technology utilizes gene expression profiling in a number of disease states including but not limited to: various cancers, SLE, multiple sclerosis, and T1D [7]. One study using microarray analysis identified 116 differentially expressed genes between peripheral blood lymphocyte samples in T1D patients and AAb-negative controls, and many of these genes are involved in important immunoregulatory functions [7]. Microarray analysis has been performed in pancreatic tissue of T1D and healthy controls, with one study by Yip et al., identifying 48 differentially genes in human pancreata [33]. Studies also show the potential for use of miRNA microarrays in identifying biomarkers for the diagnosis of T1D [34, 35, 36, 37] and predicting complications such as ketoacidosis [38]. Differential mRNA and miRNA expression profiles characterized by microarray analysis have helped to provide insight into the pathogenesis of T1D, as many of these genes are involved in cellular functions such as oxidative stress response, DNA repair, inflammation, and apoptosis [39]. After appropriate validation of candidate genes discovered with microarray analysis, multivariate analysis, or computational techniques can be used to identify patients at high risk of developing T1D [40].

3.2 Genome-wide association analysis (GWAS) and genetic risk scores

The polygenic risk score method summarizes multiple genetic risk elements into a single score. There are benefits to establishing successful genetic risk scores. Over the past 20 years, the advancement in technology has allowed for publicly available genetic data. Combine this with the decreased costs in genotyping processes, the T1D genetic risk score (T1D GRS) has the opportunity to demonstrate its applicability for “disease prediction, discrimination, investigation of unusual cohorts, and investigation of biology in large datasets where genetic data are available” [41]. In 2018, a group following a cohort from The Environmental Determinants of Diabetes in the Young (TEDDY) study created a genetic score based on 3 SNPs for HLA class II genotyping and 41 SNPs in other genes. The score identified newborn children, with no family history of T1D, who had a >10% risk for developing pre-symptomatic T1D, a nearly 2-fold higher risk than children identified by high-risk HLA genotypes alone [42].

GWAS previously associated the 3p21.31 locus with T1D [43]. The 3p21.31 locus encodes for many chemokine receptors including the C-C motif chemokine receptor 2 [44]. C-C motif chemokine ligand 2 (CCL2) is a pro-inflammatory chemokine that binds to CCR2 to promote T cell recruitment and macrophage activation. Tran et al analyzed CCL2 levels in the DAISY cohort and found paradoxically decreased CCL2 in T1D patients compared to controls. The proposed mechanism was that variants in the 3p21.31 genetic locus promote the development of T1D by increasing CCR2 expression, causing subsequent pancreatic islet cell destruction while simultaneously depleting the CCL2 pool [44].

The major limitation of genetic risk score development is the genetic heterogeneity among different ethnic groups and populations. This especially proves to be a challenge in identifying T1D susceptibility genes [1]. Most genetic risk score validation utilizes populations of European ancestry. Studies of genetic risk scores in African ancestry populations suggest that an ancestry-specific genetic risk score may improve the prediction of T1D [45].

3.3 Proteomics

As proteins directly carry out cellular functions and disease processes, protein biomarkers have great potential for T1D prediction and monitoring treatment outcomes. The proteomic analysis allows us to discover and quantify these proteins on a large scale. Proteomic techniques can be divided into two main types: mass spectrometry-based and array-based (Table 3) [54].

3.3.1 Mass spectrometry based proteomics

Mass spectrometry (MS) measures the mass charge ratio (m/z) of ions, which is used to characterize biomolecules with extremely high sensitivity and high throughput [55].

Separation techniques such as 2D and 3D polyacrylamide gel electrophoresis and 2D high performance liquid chromatography are often performed before MS to enrich low-abundance proteins [56], Standard 2D polyacrylamide gel electrophoresis (2D PAGE) alone has several limitations for biomarker discovery including low throughput and low resolution, and is thus used in conjugation with MS analysis [7]. The two-dimensional liquid chromatography (2DLC) mass spectrometry platform has been widely used in the discovery of protein candidates that may serve as biomarkers [46]. MS and MS combined techniques have been used to identify a number of biomarkers in T1D.

3.3.2 Array based proteomics

Array-based techniques can be divided into targeted assays that measure a select number of molecules, and Nucleic Acid Programmable Protein Arrays (NAPPA) arrays (Table 4). Targeted array-based techniques such as Enzyme-linked immunoassays (ELISA) and Luminex bead assays involve measurement of selected molecules based on the role of the molecule in a disease process and availability of the assay. These assays are limited by the number of protein markers that can be tested at once, yet a large number of samples can be tested. Array-based assays have been used to discover many candidate cytokines, chemokines, adhesion molecules, and receptors that may play a role in the development of T1D and its complications.

Protein	Technique
Single markers	ELISA [1, 49]
Multimarkers	Bead Array [4, 50, 51, 52, 56, 57, 58, 59, 60]
	NAPPA [8, 53]

Table 4.

Array-based technologies in biomarker studies in T1D.

ELISA: enzyme-linked immunosorbent assay, Array Bead: Luminex bead array, NAPPA: Nucleic Acid Programmable Protein Arrays.

In contrast, NAPPA arrays provide high throughput information on thousands of candidate proteins in a more limited number of samples. NAPPA arrays contain imprinted gene on glass slides and a wheat germ expression system that converts the DNA to RNA to protein. This protein array then detects circulating levels of antibodies against the proteins in the blood of T1D individuals [53]. Recent studies by Bian et al. utilized NAPPA and ELISA to analyze more than 50% of the human proteome in the serum of recent onset T1D patients. They discovered six novel T1D-associated autoantibodies and created a combined AAb panel that had a higher AUC and sensitivity in diagnosis compared to the conventional AAb ZnT8A (Table 5) [8, 61]. More studies are required to confirm the findings utilizing the NAPPA in T1D.

Protein antigen	AUC	Sensitivity at 95% specificity
ZnT8A alone	0.62	38.3
Novel AAb panel (anti-PTPRN2, -MLH1, -PPIL2, and -QRFPR)	0.74	37.5
Novel AAb panel + ZnT8A	0.81	55.2

Table 5.

Novel T1D AAb panel identified by NAPPA array: AUC and sensitivity compared to ZnT8A [8].

AUC: area under the curve, ZnT8A: Zn transporter 8 autoantibody, PTPRN22: protein tyrosine phosphatase receptor 22, MLH1: MutL protein homolog 1, PPIL2: Peptidylprolyl Isomerase Like 2, QRFPR: Pyroglutamylated RFamide Peptide Receptor.

4. Artificial intelligence (AI) use in T1D research

Although various T1D prevention trials have been conducted, response to preventative therapies appears to be limited to only a few individuals. A major factor in this lack of response is the difficulty identifying appropriate high-risk subjects who would benefit from such trials. Personalized prevention strategies based on one’s own risk and etiology may prove to be more efficient than prevention for the whole at-risk population. Therefore, a panel of multiple biomarkers is needed to identify individual risks for potential subjects in T1D prevention trials.

Advances in omic technology have allowed researchers to discover several potential biomarkers. Validation of these biomarkers requires thousands of samples, irrespective of the technique, and thus biomarker development is often hindered by limited availability of biological samples. Recently, T1D repositories such as TEDDY and TrialNet have addressed this problem by providing a large pool of available sample data for omic analysis. However, high computational power is required to analyze numerous biomarkers in these large data sets and identify the optimal markers for a panel.

AI machine learning has been used for a broad range of applications in cancer treatment, including diagnosis and classification of cancer as well as prediction of progression and treatment outcomes [62]. This suggests that AI may help to solve the issue of early prediction of T1D as well. The main goal of AI in T1D biomarker discovery is to combine information on the small differences in serum biomarkers between T1D and healthy patients and utilize this information to predict which patients are at high risk of developing T1D in the future.

Recent studies have utilized various machine-learning techniques to analyze large amounts of omics data (Table 6). Repeated Optimization for Feature Interpretation (ROFI) is one such technique that uses a repeated selection algorithm 500 times to generate important matrices for each feature [65, 67]. These matrices define the importance of a feature as the percentage of times it was selected to be included out of the 500 times the algorithm was run [65, 66]. After the feature importance values have been established, the final model is produced [65]. These studies illustrate the potential utility of machine learning in analyzing large amounts of data from diverse fields of omics data (genomics, proteomics, metabolomics, and lipidomics), clinical risk factors, and environmental factors for early prediction and prevention of T1D. Machine learning also has the potential to improve the prediction of T1D complications such as diabetic nephropathy, retinopathy, and peripheral neuropathy [11, 57]. The goal is that machine learning-derived risk scores can be used to identify T1D patients who would benefit the most from targeted preventative therapies before the development of these complications.

Computational approach	Omics platform	Omics approach	Number of features (post optimization)	Accuracy/ AUC/ Sensitivity	Ref
Model averaging	SELDITOF	Proteomics	146	90.0% sensitivity	[63]
Support Vector Machine	GWAS	Genomics	417	AUC 0.84	[28]
Repeated Optimization for feature interpretation (ROF)	HPLC-MS, Multiplex assay	Proteomics, Immunologic, Genomics, Metabolomics	76	AUC 0.92	[64]
	GC-TOF MS, Illumina ImmunoChip	Metabolomics, Genomics	42	AUC 0.74	[65]
	GC-TOF MS, genetic risk score	Metabolomics, Genomics, Immunologic	16	AUC 0.84	[66]

Table 6.

Computational approaches for biomarkers in T1D.

ROF: Repeated Optimization for Feature Interpretation, number of features presented after post-optimization of the computational method, SELDI-TOF: Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry, GWAS: genome-wide association study, HPLC-MS: high performance liquid chromatography-Mass spectrometry, GC-TOF: gas chromatography time-of-flight mass spectrometry.

5. Conclusion

Currently, individuals at high risk of developing T1D are identified with multiple autoantibodies and HLA genotyping. These existing biomarkers do not fully meet the need for T1D prediction and prevention due to many issues with sensitivity and specificity, as well as the relatively late appearance of autoantibodies. Successful prevention of T1D requires the identification of high-risk populations early in the disease course before the appearance of islet autoantibodies and clinical disease onset. Due to the multifactorial nature of T1D, no single biomarker can provide adequate power to predict the disease. Therefore, research into T1D prevention has to rely on the combination of multiple markers. Advances in high throughput omic technologies such as GWAS and NAPPA arrays have offered new opportunities to discover such biomarkers. In addition, advanced computational techniques including machine learning are being increasingly utilized to analyze numerous biomarkers in large data sets. Despite these advances, there is still an urgent need for new and improved biomarkers for T1D prediction and prevention. Surrogate biomarkers are needed to access the outcomes of preventative therapy trials in their early stages. Due to the long asymptomatic period for diabetes, it is too expensive and time-consuming for clinical trials to wait for the final clinical outcome. The lack of suitable surrogate biomarkers for T1D has severely hampered progress in clinical trials. Newer markers will also need to provide information on response to treatment in existing T1D patients. This information will aid in predicting which patients would benefit from specific therapies.

The simultaneous consideration of genomic, transcriptomic, and proteomic data, using advanced computational techniques will be required for accurate assessment of T1D risk and monitoring of therapeutic outcomes in the future.

Acknowledgments

This work was supported by JDRF awards to SP (2-2011-153, 10-2006-792, and 3-2004-195). PMHT was supported by NIH/NIDDK fellowship (F30DK12146101A1).

Conflict of interest

The authors declare no conflict of interest.