Abstract Body (Do not enter title and authors here): Intro: Circulating diabetes protein biomarkers have been studied in large human cohorts using measurements from a single timepoint. There are limited studies, however, on how protein changes over time may be associated with T2D, which could reveal novel, dynamic biomarkers of disease.

Hypothesis: The use of repeated plasma samples from the same individual will identify novel circulating proteins modified by IFG and associated with diabetes development over time.

Methods: We measured 2,943 proteins using Olink in 1,923 participants of the Multi-Ethnic Study of Atherosclerosis (MESA, mean age 57.7 years old, 52% women) without diabetes at baseline and with plasma samples from Exam 1 (2000-2002), 5 (2010-2012), and 6 (2016-2018). The effect of IFG on longitudinal protein time trends were modeled as an interaction between time×presence of IFG at Exam 1 (n=244 with IFG, n=1,676 without) using linear mixed effects models adjusted for sex, race/ethnicity, and time-varying values for age, BMI, and eGFR. Proteins with significant interactions were probed for causality using cis-Mendelian Randomization (cis-MR). Longitudinal protein associations with incident diabetes (mean follow-up=21.6 years) were modeled using time-varying Cox models (TV Cox, 447 cases, 1,476 non-cases) adjusted for sex, race/ethnicity, and time-varying age, BMI, eGFR, total cholesterol, triglycerides, HDL-c, and hypertension. Associations were also compared with Cox models using single-timepoint protein values as the dependent variable in the UK Biobank (UKBB). A false discovery (q)<0.05 was used.

Results: We found 189 proteins with time trends that differed based on IFG status (time×IFG interaction term q<0.05). Of these, 54 (28.6%) were associated with incident diabetes in MESA (q<0.05 TV Cox) and the UKBB (q<0.05 Cox). These included SPINK4, a Kazal-type serine protease inhibitor in the intestinal mucosa that had a greater increase over time among individuals with IFG (Fig 1, q=6.5×10^-4). Inverse weighted cis-MR supported SPINK4 having a possible causal role in fasting glucose levels (β=-0.002, q=5.51×10^-10). Circulating SPINK4 was positively associated with incident diabetes in MESA (TV Cox HR 1.25 [1.14-1.37], q=7.12×10^-5) and in UKBB (Cox HR 1.43 [1.36-1.52], q=1.74×10^-34).

Conclusions: We found 189 protein trajectories modified by IFG, 54 of which were additionally associated with diabetes risk. These proteins can reveal dynamic associations that help further elucidate diabetes pathways.