Objective
This project aims to develop a trans-ancestral PRS to successfully predict type 1 diabetes risk across individuals from diverse ancestries. To do this, we will take advantage of genetic data from large collections of European samples and from smaller ancestry-specific cohorts, and ancestry-specific information on the HLA region of the genome. This project will be deployed based on four aims-analytical steps:
1) First, we will seek to improve the non-HLA component in a European type 1 diabetes PRS: The most recent type 1 diabetes genome-wide association study (GWAS) included 13,458 European cases and 20,143 controls, and identified 78 non-HLA type 1 diabetes genes. The most recently-proposed PRS by Sharp et al included only 32 non-HLA genes. Therefore, we will update and improve this score by leveraging this recent GWAS-identified non-HLA gene set. Furthermore, we will validate its performance in an independent European cohort including up to 6,326 type 1 diabetes cases and an equal number of controls. We will make sure to integrate to the PRS HLA information from this cohort, using similar parameters to those used by Sharp et al. Although we do not anticipate that the optimized PRS will overperform the Sharp et al PRS in Europeans, it could lead to a significant improvement in prediction accuracy in non-Europeans. As such, aim 1 serves as a building block for constructing a trans-ancestral PRS in aim 3.
2) Next, we will try to predict multi-ethnic HLA haplotypes: Using a recently developed machine learning algorithm by Naito et al, we will estimate HLA haplotypes in multiethnic cohorts of type 1 diabetes patients from African (n=846), Indian (n=353), Latino (n=491), South-East Asian (n=134), and Chinese (n=~300) ancestry and an equal number of ancestry-matched genotyped controls.These predicted haplotypes will be used in the aim 3 to develop a trans-ancestral PRS.
3) Here we will build trans-ancestral PRS: Specifically, using an approach developed by Weissbrod et al, we will combine the population-specific HLA haplotype estimates from the previous aim with European-estimated SNP effects from step 1, and population-specific SNP estimates using the datasets from aim 2. Performance of these new trans-ancestral PRS will be compared to that of the European PRS from aim 1, using standard metrics for PRS, such as the area under the receiver-operating curve and the R2 metric.
4) Finally, we will validate the newly developed trans-ancestral PRS in external T1D cohorts from two ancestries. Specifically, we will test the trans-ancestral PRS for East-Asians (from aim 3) in a cohort of 1,005 cases and 1,257 controls from Chinese Han descent, and the trans-ancestral PRS from Africans in a cohort of 168 cases and 1,366 controls of African-American ancestry. Similar to the previous aim, we will calculate performance metrics for the trans-ancestral PRS and compare to those obtained using the European PRS from aim 1.
Background Rationale
Type 1 diabetes is a relatively common chronic autoimmune disorder causing destruction of the insulin-producing pancreatic beta cell. Although type 1 diabetes is predominant in Europeans, its prevalence in diverse ethnic groups is rising. For instance, in the United States, individuals of African and Hispanic ancestry exhibit the most rapidly increasing rates of type 1 diabetes. While established therapies do not exist to prevent type 1 diabetes, research mainly focuses on interventions that could preserve pancreatic insulin production in individuals presenting pancreatic islet autoantibodies, which signal high risk. However, clinical diabetes can follow in weeks or even decades later, and timing of onset is difficult to predict. As such, autoantibody-based screening is suboptimal, and yearly autoantibody measurement in the entire population is practically unfeasible. Nevertheless, estimation of an individual's risk to develop type 1 diabetes risk is paramount to identify candidates eligible for trials aiming to preserve the pancreatic beta cell.
The polygenic etiology of type 1 diabetes is well recognized, with >60 genes identified to date in genome-wide association studies (GWAS). Type 1 diabetes is highly heritable, with HLA genes accounting for the largest portion of this heritability. It has been recently shown that combining variable factors (genetic, clinical and antibodies) improves dramatically prediction of type 1 diabetes risk compared to risk stratification algorithms based on autoantibodies or HLA haplotypes alone. Polygenic risk scores (PRS) are scores that summarize information from risk alleles across the genome to predict an individual's risk to develop a specific polygenic disease, and their utility has been also demonstrated in predicting type 1 diabetes. In a 2019 study by Sharp et al, a PRS demonstrated an extremely high accuracy in predicting early-onset type 1 diabetes. PRS are relatively inexpensive prediction tools, since genome-wide genotyping needs to be performed only once in a lifetime, and is affordable (<$50 USD in a research context) -less expensive than some clinical tests. For type 1 diabetes, PRS, combined with auto-antibodies and clinical risk factors, offer the promise of enhanced screening for better prevention trial selection. Unfortunately, the vast majority of available PRS have been developed using European data, and have poor performance to predict disease-risk in non-Europeans.
Recent efforts to develop ancestry-specific type 1 diabetes PRS have yielded promising results. Still, their prediction performance is worse than that obtained in Europeans using European-based PRS, since development of such PRS is limited by the small sample sizes of non-European datasets. Exciting recent methodological advances in PRS construction can lead to important gains in cross-population accurancy. Therefore, now is the time to attempt to develop effective trans-ancestral PRS for type 1 diabetes. The new approaches build on recent findings from large European type 1 diabetes GWAS, and refine the associations found there with population-specific genetic data for validation. Also, the distinct HLA patterns in patients with type 1 diabetes from different ancestries have been carefully studied and can be incorporated to these trans-ancestral PRS. Another so far overlooked parameter that should be considered in genetic prediction of type 1 diabetes is sex.
Therefore, our aim here is to develop trans-ancestral PRS rather than population-specific PRS, taking into account potential sex-specific effects. Feasibility is assured by our access to a large number of DNA samples from 5 different ancestries. Also, our research team has large expertise in genetic risk prediction studies and methods-development for PRS. Finally, by addressing the current limitations in cross-population genetic risk prediction and by fine-tuning for sex, the results of this research will enable a major step in the global application of PRS in type 1 diabetes which can reduce health inequities in research and clinical practice.
Description of Project
Polygenic risk scores (PRS) are disease risk scores that summarize information from genes across the genome of an individual to predict its risk to develop a specific polygenic disease. PRS have yielded promising results in predicting development of type 1 diabetes. In combination with autoantibodies and clinical data, type 1 diabetes PRS can optimize selection of candidates for participation in trials testing preventive therapies. Previously developed PRS for type 1 diabetes have demonstrated a very high performance in predicting type 1 diabetes in Europeans. While development of PRS specific to non-Europeans is feasible, it is limited by restricted availability of samples of non-European individuals with genome-wide genetic data. Unfortunately, since PRS are typically developed using European data, they perform poorly in diverse ancestral populations, due to differences in the genetic architecture of diseases among ancestries. For type 1 diabetes, this is particularly important, since its genetic architecture depends largely on a region of the genome known as human leukocyte antigen (HLA) complex. While HLA genes explain the largest part of type 1 diabetes’ heritability, it has been shown that these genes differ significantly among patients with type 1 diabetes from different ancestries. Restricting application of PRS to Europeans limits their use in research and clinical practice, and this could exacerbate public health inequities. Here, we aim to develop a trans-ancestral PRS for type 1 diabetes, meaning a PRS that can perform almost equally well in different ancestries. To do this, we will apply a machine-learning approach allowing the use of large available European genetic data and incorporating information on ancestry-specific HLA data. We will test this trans-ancestral PRS in patients with type 1 diabetes of African (n=846), Indian (n=353), Latino (n=491), South-East Asian (n=134), or Chinese (n=~300) ancestry, and an equal numbers of controls matched for ancestry. We will compare the performance of the trans-ancestral PRS on these different ancestries to that obtained using a European-based PRS. We will finally test the performance of the trans-ancestral PRS in independent cohorts of individuals with type 1 diabetes of Chinese ancestry (n= 1,005 cases and 1,257 controls) and of African-American ancestry (n=168 cases and 1,366 controls). If our project is successful, it will be a major step in the global application of PRS in type 1 diabetes research, allowing more precise individual risk estimates and better selection of candidates across different ancestral groups for prevention trials.
Anticipated Outcome
We anticipate that our new trans-ancestral PRS will perform better to predict type 1 diabetes in individuals from diverse ancestries, compared to the previously developed European PRS by Sharp et al, and to PRS built specifically for an ancestry (due to the generally small sample sizes of non-European type 1 diabetes cohorts). Also, we anticipate that if effects of sex are observed, incorporating sex in our trans-ancestral type 1 diabetes PRS will further improve its performance in all ancestries, compared to previously developed PRS.
Relevance to T1D
A worldwide effort is currently underway to develop interventions to prevent type 1 diabetes. A striking recent example is anti-CD3 treatment that decreased the 5-year progression risk in cases of presymptomatic type 1 diabetes to 43% from 72%. A necessary condition for applying such interventions is the identification of at-risk individuals to target. Although autoantibodies are highly predictive, yearly surveillance of the entire population starting early in childhood is not logistically feasible. The type 1 diabetes genetic risk score by Sharp et al can identify, among individuals of European ancestry with the 10% highest genetic risk, almost 80% of future type 1 diabetes cases upon whom surveillance programs could be focused. Eventually, effective choice of candidates for preventive therapies will also benefit patients with existing type 1 diabetes, to prevent immune re-destruction after beta cell regeneration therapies. Global application of these principles absolutely requires a powerful and valid trans-ancestral predictor, i.e. an ancestry-informed genetic risk assessment using performant trans-ancestral PRS. With improvements in performance across ancestry, which are fine-tuned for sex, our work is prone to advance the creation of a universal genetic prediction tool for type 1 diabetes, applicable in research and clinical practice.