Multi-fluid, multi-omics signatures of insulin resistance and incident type 2 diabetes among Puerto Rican adults

Previous studies have examined the prediction of insulin resistance and type 2 diabetes (T2D) using plasma or saliva omics, but none have combined metabolomics and proteomics from multiple biofluids, such as plasma and saliva. Among Puerto Rican adults, a high-risk population with health disparities, we sought to determine whether adding saliva improves T2D prediction over plasma alone.

In this pilot matched case-control study within the San Juan Overweight and Obese Adults Longitudinal Study (SOALS), we analyzed baseline samples from 40 healthy participants, 20 of whom developed T2D at follow-up (year 3) and 20 age- and sex-matched controls. We profiled 7,595 proteins in plasma and saliva (SomaScan) and 1,051 plasma and 635 saliva metabolites [ultra-high-performance liquid chromatography-tandem mass spectrometry (UHPLC-MS/MS) and gas chromatography-mass spectrometry (GC-MS); Metabolon, Inc.] for analysis. We evaluated nine omics signatures combining biofluid (plasma, saliva, or both) and omics (metabolomics, proteomics, or both). Nested elastic net regression with leave-one-out cross-validation identified insulin resistance signatures, and receiver operating characteristic (ROC) curves [area under the curve (AUC)] assessed their predictive performance for T2D. We used multivariable conditional logistic regression to evaluate associations between omics scores and incident T2D.

The strongest T2D prediction was observed for plasma proteomics and multi-omics, multi-fluid proteomics, and multi-omics signatures (AUCs: 0.80-0.83). Saliva proteomics, metabolomics, and multi-omics, along with plasma metabolomics and multi-fluid metabolomics, exhibited limited prediction (AUCs: 0.51-0.67). Plasma proteomics, multi-omics, and multi-fluid multi-omics were positively associated with T2D [hazard ratios (HRs): 3.00-3.68].

Plasma proteomic signatures provided the strongest T2D prediction. Adding saliva data did not improve predictive performance of plasma data.