Objective
Type 1 diabetes (T1D) develops when the body’s own immune system mistakenly attacks and destroys insulin-producing cells in the pancreas. This process usually begins in early childhood and, once started, cannot yet be stopped. By the time standard screening detects autoantibodies—immune markers signaling that the attack has begun—the disease process is already underway. For this reason, most current prevention efforts start too late.
The goal of this project is to move earlier—to find the first biological signals that mark vulnerability before the immune system becomes activated. Our central objective is to determine whether patterns in a baby’s cord blood at birth can predict which children are most likely to develop T1D years later. These early-life signatures could help identify risk before symptoms or autoantibodies appear, opening an entirely new window for prevention, to shift the timeline of intervention.
Our earlier research in Sweden, through the All Babies in Southeast Sweden (ABIS) birth cohort, revealed that children who later developed T1D had distinct protein patterns already detectable in their cord blood at birth. The findings suggests that the path toward T1D may begin before birth, shaped by the prenatal environment. However, these results come from a single population, and it is critical to test whether they hold true in other, more diverse populations.
The current project will test and expand on these findings in a very diverse population. We are partnering with the Carolinas Cord Blood Bank (CCBB), one of the world’s largest and most ethnically diverse cord blood collections. Using advanced proteomic technologies (Olink® Reveal or Explore 3072), we will measure thousands of proteins related to immune and metabolic health to identify early-life signatures associated with later T1D. In the CCBB, 42% of the samples under consideration come from minority racial or ethnic backgrounds. Samples are collected from mothers in urban, suburban, and even rural settings. The breadth of sampling allows us to investigate how early-life protein markers perform across different geographic settings and to better understand risk in rural populations, where health disparities and limited access to preventative care may amplify disease burden, and in diverse background that are historically understudied in medical research.
Our research has three main objectives: (1) First, we will test whether the protein markers discovered in Sweden can predict T1D in a racially and geographically diverse population. This step ensures that the findings are not unique to a single country or genetic background. (2) Second, we will expand discovery to include thousands of additional proteins using next-generation high-throughput assays. This broader approach allows us to uncover new proteins and biological pathways that may be involved in very early disease development. (3) Finally, we will explore whether predictive patterns differ by sex, race, ethnicity, or family history of T1D. Understanding these differences is essential for designing screening tools that are accurate, equitable, and relevant for all children.
To analyze these large and complex datasets, we will use advanced machine learning models that identify the most informative protein combinations while ensuring that results are reproducible and interpretable. This integration of biological discovery with modern data science provides a rigorous framework for uncovering the earliest signals of disease.
Ultimately, our objective is to confirm whether cord blood can serve as a reliable, non-invasive biospecimen for identifying risk before the autoimmune process begins. If successful, this research will lay the groundwork for the world’s first preclinical screening test for T1D—enabling physicians and families to understand a child’s risk at birth and opening the door to preventive strategies that protect pancreatic cells before irreversible damage occurs.
In short, this project transforms how we understand, predict, and ultimately prevent T1D.
Background Rationale
For decades, scientists have sought ways to identify children at risk for T1D before the immune system begins to attack the pancreas. While genetic markers such as HLA variants and islet autoantibodies are useful tools in screening, they only identify risk after the autoimmune process has already begun. What is still missing are reliable preclinical markers that can signal vulnerability earlier — before immune activation — and no standard proteomic or metabolic markers currently exist for this early window.
Our preliminary work in a large Swedish general population study, the All Babies in Southeast Sweden (ABIS) cohort, supports the possibility of detecting early risk signatures at birth. ABIS includes 16,683 infants (representing 78.6% of all children born between 1997–1999 in four counties) followed prospectively from birth. Using Olink Explore 384 proteomics on 146 future T1D cases and 286 matched controls, we identified proteins at birth that distinguished children who would later develop T1D from those who would not. The age at diagnosis ranged from 2 to 24.6 years (mean 12.6, 95% CI [11.7–13.5]). Despite being culturally homogeneous, ABIS remains one of the largest datasets globally for prospective T1D biomarker discovery.
After adjusting for multiple testing, significant proteins clustered in pathways related to antigen presentation, extracellular matrix (ECM) remodeling, and immune regulation. Examples include HLA-DRA, IDS, and CTSC, which were higher in future cases, and TIMP3, ADA, and LSP1, which were higher in controls. Gene set enrichment and network analyses confirmed strong biological coherence, showing these proteins were core components of cytokine signaling and immune-response pathways. Even when accounting for maternal T1D or prenatal factors, several markers remained significant. Machine learning models using a subset of these proteins predicted later T1D with high accuracy (AUC = 0.89 ± 0.02), far outperforming environmental models from the first year of life (AUC 0.43–0.62).
Several of these proteins suggest that inflammatory and metabolic stress signals may already be present at birth. Elevated HLA-DRA indicates increased antigen presentation, possibly priming immune activation early in life. IDS, which enhances glucose-induced insulin secretion in islet cells, was higher in both children who later developed T1D and in adults with long-standing disease, suggesting dysregulated insulin handling. CTSC, a lysosomal protease involved in immune defense and inflammation, was also upregulated and is known to play roles in over 100 human diseases.
Conversely, proteins enriched in controls point to potential protective mechanisms. TIMP3 supports islet structure and limits inflammation, ADA regulates adenosine signaling to prevent overactive immune responses, and LSP1 facilitates neutrophil migration that may prevent damaging accumulation within the pancreas. Proteomic differences also implicate disruptions of glycosaminoglycans (GAGs) and ECM remodeling — key systems for maintaining islet architecture and regulating inflammation. The ECM is an active participant in local immune regulation: under normal conditions it supports β-cell homeostasis, but in inflammation it becomes fragmented and promotes leukocyte infiltration. Enzymes like IDS help degrade heparan sulfate and dermatan sulfate, GAGs essential for ECM integrity and cell signaling. During chronic inflammation, these pathways become dysregulated, creating a positive feedback loop that sustains immune activation and tissue stress.
Together, these findings highlight a balance between immune activation and restraint that may determine β-cell vulnerability or protection from birth onward. These observations led us to hypothesize that inflammatory and stress-response signatures at birth may reflect early immune priming and predict later T1D risk. The ability to detect these signatures non-invasively, from cord blood, could open a new frontier for preclinical T1D risk screening.
Description of Project
Type 1 diabetes (T1D) develops when the immune system mistakenly destroys the insulin-producing cells in the pancreas, leaving the person dependent on lifelong insulin therapy. While genetics play a role, environmental and inflammatory factors early in life—sometimes even before birth—also appear to shape risk. For decades, scientists have sought reliable early warning signs, yet the only current predictive tools rely on detecting autoantibodies, which appear after the autoimmune attack has already begun. What is missing is an approach that detects vulnerability before immune activation takes hold.
Our project aims to fill that gap by exploring how immune and metabolic signals present at birth—captured from cord blood—might identify who is at greatest risk for developing T1D later in life and give additional mechanistic insights into prenatal conditions shaping T1D trajectory at the earliest stages. We hypothesize that patterns of inflammatory and stress-response proteins detectable at birth reflect early immune “priming” and may signal which children are most vulnerable to β-cell injury long before disease onset.
To test this, we will analyze racially and ethnically diverse cord blood samples from the Carolinas Cord Blood Bank (CCBB), one of the largest and most respected repositories of newborn samples in the world. The demographic composition of samples under consideration reflects a diverse representation over their nearly 30 years of sampling from urban, suburban, and rural settings across North Carolina, with 42% minority representation (all races / ethnicities excluding Caucasian). The representation and balance seen enables examination of both sex-specific and population-level discoveries that are inclusive of and broadly applicable to diverse communities and backgrounds affected by diabetes. Using high-throughput proteomics, we will measure up to 3,000 proteins that represent key biological pathways involved in inflammation, metabolism, and immune regulation. We are sufficiently powered to make such discoveries, uncovering proteins with effect sizes (f) as small as 0.27. In addition, machine learning models will be used to identify combinations of proteins that best predict future diabetes risk.
The diversity seen in the CCBB was a primary motivating factor for establishing this collaboration, ensuring that study outcomes are generalizable and equitable across populations. Our preliminary results in the All Babies in Southeast Sweden (ABIS) general population, birth cohort study, have provided promising candidates. We found that children who later developed T1D already had measurable differences in certain proteins at birth—proteins involved in antigen presentation (HLA-DRA), extracellular matrix remodeling (IDS, CTSC), and immune regulation (TIMP3, ADA, PRSS8). These pathways collectively reflect inflammation, cellular stress, and immune activation, suggesting that biological vulnerability to T1D begins before the autoimmune process starts. Our models accurately predicted future T1D cases (AUC = 0.89), supporting the idea that prenatal and perinatal factors leave measurable molecular traces that forecast future disease.
By validating and expanding upon our initial findings but in a new, diverse U.S. population, this project will determine whether these early-life protein markers truly capture universal mechanisms of T1D risk. If successful, the research could transform how we approach diabetes prevention. Instead of reacting after autoimmunity has begun, clinicians could one day identify at-risk infants through a non-invasive cord blood screen, providing a window of opportunity for early interventions that protect β-cells and delay—or even prevent—insulin dependence.
Beyond its scientific impact, this project exemplifies the mission of Breakthrough T1D: turning innovative research into actionable progress for patients and families. By reframing cord blood from routinely discarded medical waste into a rich source of early-life health information, we can begin to identify children at risk years before autoimmunity develops, paving the way for preventive interventions that protect β-cells before irreversible damage occurs. Such a new frontier in disease prevention is scalable, equitable, and hopeful.
Anticipated Outcome
This project represents a major shift in how we think about preventing Type 1 diabetes (T1D). For decades, efforts have focused on detecting the disease only after the immune system has begun attacking the insulin-producing cells of the pancreas. By that time, much of the damage is already done. Our approach moves the timeline of detection back—far earlier—by looking for biological signs of vulnerability that appear even before the immune system starts its attack.
At the heart of this project is a simple but transformative idea: cord blood, usually discarded after birth, may contain valuable biological information that can help identify babies who are at higher risk for developing T1D later in life. Using advanced proteomic technology—capable of measuring thousands of proteins at once—we aim to find patterns that reveal early immune stress or metabolic changes that precede the disease.
Technically, our study combines high-throughput proteomic profiling with modern machine learning (ML). This means we can analyze vast amounts of biological data with precision, identifying which proteins carry predictive power and testing those findings in independent groups to ensure reliability. By using cross-validation and multi-cohort analysis, we minimize bias and avoid the pitfalls that often limit smaller or less diverse studies.
The anticipated outcome is twofold. First, we expect to confirm that certain proteins in cord blood can reliably predict which children are likely to develop T1D years later. Second, by validating these findings in a racially and geographically diverse population, we will demonstrate that this method can be applied broadly and equitably.
If successful, this work could transform the way we identify and ultimately prevent T1D. It will show that a routinely discarded sample—cord blood—can become a powerful, non-invasive screening tool. This would allow healthcare providers to recognize risk much earlier, opening the door to preventive care before the disease process begins.
Beyond individual impact, the broader implication is a paradigm shift—from reacting to disease to anticipating it. This approach could inspire similar strategies for other autoimmune and inflammatory conditions, where early immune priming might also be detected before symptoms appear.
Conceptually, this work shifts the paradigm from reacting to postnatal immune activation to identifying prenatal immune priming and vulnerability and open new avenues for prevention. Technically, it combines high-throughput proteomic profiling with machine learning optimized through cross-validation and multi-cohort design to reduce overfitting and internal bias. Translationally, this approach offers a scalable, equitable, and resourceful screening approach that complements current practices. In short, this project pioneers a new vision for T1D prevention: one rooted in early detection, proactive care, and the reimagining of existing resources to advance population health.
Relevance to T1D
Although genetic screening (such as for HLA genes) and antibody testing can identify children at risk for type 1 diabetes (T1D), these tools typically detect the disease process only after damage has already begun. There are currently no accepted ways to detect T1D before the autoimmune attack starts, and very little research has explored whether proteins or metabolic signals at birth might offer early warning signs. For large-scale population screening and true prevention, we need new approaches that move the timeline of detection much earlier.
Our study directly addresses this unmet need. By analyzing cord blood—collected safely and noninvasively at birth—we aim to validate early proteomic markers that predict future T1D. Cord blood reflects both the baby’s genetic and metabolic state, but also the prenatal environment that shaped it. Our previous research in Sweden found that certain proteins involved in immune regulation, inflammation, and cellular stress are already altered at birth in children who later developed T1D. We now plan to test whether these same signals are seen in a new, more diverse U.S. population to ensure the findings are broadly relevant and generalizable and expand the discovery potential in this new samples by investigating a broader coverage of proteins.
To accomplish this, we will use next-generation proteomics powered by Proximity Extension Assay (PEA) technology—an approach cited in over 2,000 peer-reviewed studies for its precision and reproducibility. The Olink® panels we will use provide extensive coverage of biological pathways relevant to immune and metabolic function and have been benchmarked in large, diverse populations such as the UK Biobank. These panels include many proteins that are established or emerging drug targets, meaning that confirmed associations could inform drug repurposing or support the development of targeted interventions. If new candidate markers are validated, we will explore translation opportunities in collaboration with partners experienced in therapeutic development.
Importantly, our proposed work will leverage a rich racial and ethnic background that is represented in the CCBB, incorporating samples from four major health systems with continuous collections since 1998. These sites collectively serve urban, suburban, and rural areas across North Carolina. This geographic breadth ensures inclusion of mothers and their children from communities that have been historically underrepresented in biomedical research.
This work is significant because it reframes how we think about T1D risk. Instead of searching for disease activity after the immune system has already begun its attack, we focus on the earliest biological clues that may reveal who is most vulnerable. Detecting this risk years before autoimmunity starts could allow for precision prevention—protecting the pancreas and delaying, or even preventing, the onset of T1D.
If successful, this research will lay the foundation for the first preclinical screening test for T1D. Such a test could be integrated into existing newborn screening programs, providing families with early insight into their child’s health and giving clinicians a headstart on prevention. Importantly, by validating markers in a large and diverse cohort, we aim to ensure that future screening tools are equitable and broadly applicable.
Ultimately, this project seeks to make T1D prevention proactive rather than reactive. By turning an existing, routinely collected resource—cord blood—into a predictive tool, we may open a new path toward early detection and healthier futures for children at risk of this lifelong disease.