Metabolomics facilitates differential diagnosis in common inherited retinal degenerations by exploring their profiles of serum metabolites

Abstract

The diagnosis of inherited retinal degeneration (IRD) is challenging owing to its phenotypic and genotypic complexity. Clinical information is important before a genetic diagnosis is made. Metabolomics studies the entire picture of bioproducts, which are determined using genetic codes and biological reactions. We demonstrated that the common diagnoses of IRD, including retinitis pigmentosa (RP), cone-rod dystrophy (CRD), Stargardt disease (STGD), and Bietti's crystalline dystrophy (BCD), could be differentiated based on their metabolite heatmaps. Hundreds of metabolites were identified in the volcano plot compared with that of the control group in every IRD except BCD, considered as potential diagnosing markers. The phenotypes of CRD and STGD overlapped but could be differentiated by their metabolomic features with the assistance of a machine learning model with 100% accuracy. Moreover, EYS-, USH2A-associated, and other RP, sharing considerable similar characteristics in clinical findings, could also be diagnosed using the machine learning model with 85.7% accuracy. Further study would be needed to validate the results in an external dataset. By incorporating mass spectrometry and machine learning, a metabolomics-based diagnostic workflow for the clinical and molecular diagnoses of IRD was proposed in our study.

Fig. 1. Flow chart of the present study.

Metabolomic analysis using MS was performed in enrolled IRD cases and the control groups. The heatmap, PLS-DA, and volcano plots were used to reveal the difference in metabolomic profile between IRDs and the control group. By incorporating metabolomic information, a machine learning model was constructed for differentiating specific IRDs that cannot be distinguished by clinical examination. MS, mass spectrometry, IRD, inherited retinal degeneration, PLS-DA, partial least squares-discriminant analysis, BCD, Bietti’s crystalline dystrophy, CRD, cone-rod dystrophy, RP, retinitis pigmentosa, STGD, Stargardt disease; LC-HR-MS/MS, liquid chromatography-high-resolution tandem mass spectrometry, ML, machine learning.

Fig. 2. Metabolomics profile and selected identified metabolites in metabolomics analysis in different IRD groups.

a The heatmap of 40 identified metabolites with the most significant differences among the seven IRD subgroups and control group were selected using ANOVA. The red color represents the upregulated metabolites, while the blue color represents the downregulated metabolites in each enrolled case. For (2b–h), statistical analysis was performed using analysis of variance (ANOVA) with Tukey’s honestly significant difference test (two-sided). Results are indicated by: Nonsignificant; (ns), p > 0.05; *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001. The levels of D-xylonate (b), citronellyl acetate (c), and hexadecanedioic acid (d) were higher in the CD/CRD, STGD, and RP groups than those in the BCD and control groups. The levels of phosphatidylserine (14:1/16:0) (e), phosphatidylcholine (19:1) (f), phosphatidylcholine [16:0/9:0(CHO)] (g), and N-undecanoylglycine (h) were lower in the CD/CRD, STGD, and RP groups than those in the BCD and control groups. Source data are provided as a Source Data file. IRD, inherited retinal degeneration; BCD, Bietti’s crystalline dystrophy, CRD, cone-rod dystrophy, RP, retinitis pigmentosa, STGD, Stargardt disease, ANOVA, analysis of variance, PC, phosphatidylcholine, PS, phosphatidylserine.

Fig. 3. PLS-DA and volcano plots of each IRD subgroup compared with that of the control group.

The significant features highlighted in the volcano plot were defined as having a false discovery rate <0.05 (Benjamini–Hochberg test, two-sided) and fold change >2. The number of metabolite features with significant differences between each IRD subgroup and control group was shown in the parentheses in each volcano plot. No metabolite feature could be identified between the BCD and control group (g). The PLS-DA plot showed that the RP, CD/CRD, and STGD groups could be distinguished from the control group in metabolomic analysis (a, c, e), while the overlapping area in the BCD group was more prominent (g). Moreover, when we further sub-grouped RP by genotype, including EYS, USH2A, ABCA4, and PRPF31, the metabolomic analysis showed more distinguishable results in the PLS-DA plot (b, d, f, h). i Compared between CD/CRD and STGD, 71 features could be identified with significant differences from the volcano plots, and the two groups could be distinguished in the PLS-DA plot. j Compared with RP with the ABCA genotype, STGD showed that only 19 features in the volcano plots could be identified, and the overlapping of the PLS-DA plot was prominent. Despite the significant phenotypic differences, RP and STGD with the same genotype, ABCA4, showed similar metabolomic profiles. Source data are provided as a Source Data file. IRD, inherited retinal degeneration, BCD, Bietti’s crystalline dystrophy, CRD, cone-rod dystrophy, RP, retinitis pigmentosa; STGD, Stargardt disease, PLS-DA, partial least squares-discriminant analysis.

Fig. 4. Performance of the machine-learning LASSO model for IRD subtypes classification.

a Two diagnostic models were established to differentiate (1) CD/CRD, STGD, and control group and (2) RP with EYS, USH2A, and other genotypes, using the machine learning LASSO model. b The sensitivity and specificity were both 100% in the training and validation sets of the cone-predominant disease diagnosis model. c The area-under-curve (AUC) was 1.0 in the three subgroups in the training and validation set. d The diagnostic accuracy was 83.7% in the training set and 85.7% in the validation set of the RP diagnosis model. e The AUC of the RP diagnosis model for the USH2A, EYS, and other genotypes in the training and validation set. Source data are provided as a Source Data file. LASSO, Least Absolute Shrinkage and Selection Operator, IRD, inherited retinal degeneration, CRD, cone-rod dystrophy, RP, retinitis pigmentosa, STGD, Stargardt disease, AUC, area-under-curve.

Fig. 5. Fundus photography and autofluorescence of representative cases and proposed diagnostic flow chart for IRDs.

a Different genotypes of RP cannot be differentiated by fundus examination, which shared common features, including pale disc, vessel attenuation, and pigmentary change, only with different disease severity. Moreover, the fundus appearance of CD/CRD and STGD is also undistinguishable, both with characteristic central maculopathy. However, the fundus appearance of BCD was characterized by crystalline deposition and could be differentiated easily from other IRDs. b we proposed a diagnostic flow chart to facilitate the early diagnosis of IRDs by incorporating clinical information, metabolomics analysis, and genetic diagnosis. IRDs belonging to rod-predominant disease, cone-predominant disease, and crystalline deposition were first determined by fundus examination. By incorporating targeted metabolomics analysis and a machine learning model, we could further differentiate EYS and USH2A genotypes in rod-predominant disease and STGD and CD/CRD in cone-predominant disease. Finally, we could reach a targeted, small panel NGS for the patient’s and family members' genetic diagnosis. IRD, inherited retinal degeneration; BCD, Bietti’s crystalline dystrophy, CD/CRD, cone dystrophy/cone-rod dystrophy, RP, retinitis; STGD, Stargardt disease, NGS, next-generation sequencing.