Fig. 1: Overview of the study design.

a, In the discovery phase, participants are enroled from the LVS. A total of 1,196 participants who had 7DDR data for carbohydrate intake and metabolomics data with missing rate <30% were included. Elastic net regression was used to select metabolites and build metabolite profile scores correlated to total carbohydrate and carbohydrate intake from different dietary sources. b, In the external replication phase, participants were from the NPAAS-FS. A total of 153 women completed the 14-day feeding study and attended two clinic visits. We mapped metabolites from the LVS with this metabolomics data from the NPAAS-FS. A correlation analysis was employed to examine the relationships between the metabolomic profile scores and carbohydrate intake in replication phase of NPAAS-FS. c, For the cohort analysis of metabolite profiles with incident T2D, participants were selected from NHS, NHSII and HPFS. Blood samples were collected from NHS participants during 1989–1990, from NHSII participants during 1996–1999 and from HPFS participants during 1993–1995. We used Cox regression models to evaluated the prospective associations of carbohydrate associated metabolites and metabolite profile scores with incident T2D over follow-up. All icons adapted from Icons8 (https://icons8.com/icons).