Retinal screening contributes to early detection of diabetic retinopathy and timely treatment. To facilitate the screening process, we develop a deep learning system, named DeepDR, that can detect early-to-late stages of diabetic retinopathy. DeepDR is trained for real-time image quality assessment, lesion detection and grading using 466,247 fundus images from 121,342 patients with diabetes. Evaluation is performed on a local dataset with 200,136 fundus images from 52,004 patients and three external datasets with a total of 209,322 images. The area under the receiver operating characteristic curves for detecting microaneurysms, cotton-wool spots, hard exudates and hemorrhages are 0.901, 0.941, 0.954 and 0.967, respectively. The grading of diabetic retinopathy as mild, moderate, severe and proliferative achieves area under the curves of 0.943, 0.955, 0.960 and 0.972, respectively. In external validations, the area under the curves for grading range from 0.916 to 0.970, which further supports the system is efficient for diabetic retinopathy grading.

Although the pharmacological effects of fibroblast growth factor 21 (FGF21) are well-documented, uncertainty about its role in regulating excessive energy intake remains. Here, we show that FGF21 improves systemic insulin sensitivity by promoting the healthy expansion of subcutaneous adipose tissue (SAT). Serum FGF21 levels positively correlate with the SAT area in insulin-sensitive obese individuals. FGF21 knockout mice (FGF21KO) show less SAT mass and are more insulin-resistant when fed a high-fat diet. Replenishment of recombinant FGF21 to a level equivalent to that in obesity restores SAT mass and reverses insulin resistance in FGF21KO, but not in adipose-specific βklotho knockout mice. Moreover, transplantation of SAT from wild-type to FGF21KO mice improves insulin sensitivity in the recipients. Mechanistically, circulating FGF21 upregulates adiponectin in SAT, accompanied by an increase of M2 macrophage polarization. We propose that elevated levels of endogenous FGF21 in obesity serve as a defense mechanism to protect against systemic insulin resistance.

Diabetic retinopathy (DR) is the main cause of vision loss or blindness in working age adults worldwide. The lack of effective diagnostic biomarkers for DR leads to unsatisfactory curative treatments. To define potential metabolite biomarkers for DR diagnosis, a multiplatform-based metabolomics study is performed. In this study, a total of 905 subjects with diabetes without DR (NDR) and with DR at different clinical stages are recruited. Multiplatform metabolomics methods are used to characterize the serum metabolic profiles and to screen and validate the DR biomarkers. Based on the criteria p < 0.05 and false-discovery rate < 0.05, 348 and 290 metabolites are significantly associated with the pathogenesis of DR and early-stage DR, respectively. The biomarker panel consisting of 12-hydroxyeicosatetraenoic acid (12-HETE) and 2-piperidone exhibited better diagnostic performance than hemoglobin A1c (HbA1c) in differentiating DR from diabetes, with AUCs of 0.946 versus 0.691 and 0.928 versus 0.648 in the discovery and validation sets, respectively. In addition, this panel showed higher sensitivity in early-stage DR detection than HbA1c. In conclusion, this multiplatform-based metabolomics study comprehensively revealed the metabolic dysregulation associated with DR onset and progression. The defined biomarker panel can be used for detection of DR and early-stage DR.