Using Machine Learning to Identify Metabolomic Signatures of Pediatric Chronic Kidney Disease Etiology.

2022

375-386

2022 Feb

1533-3450

<p><strong>BACKGROUND: </strong>Untargeted plasma metabolomic profiling combined with machine learning (ML) may lead to discovery of metabolic profiles that inform our understanding of pediatric CKD causes. We sought to identify metabolomic signatures in pediatric CKD based on diagnosis: FSGS, obstructive uropathy (OU), aplasia/dysplasia/hypoplasia (A/D/H), and reflux nephropathy (RN).</p>

<p><strong>METHODS: </strong>Untargeted metabolomic quantification (GC-MS/LC-MS, Metabolon) was performed on plasma from 702 Chronic Kidney Disease in Children study participants (: FSGS=63, OU=122, A/D/H=109, and RN=86). Lasso regression was used for feature selection, adjusting for clinical covariates. Four methods were then applied to stratify significance: logistic regression, support vector machine, random forest, and extreme gradient boosting. ML training was performed on 80% total cohort subsets and validated on 20% holdout subsets. Important features were selected based on being significant in at least two of the four modeling approaches. We additionally performed pathway enrichment analysis to identify metabolic subpathways associated with CKD cause.</p>

<p><strong>RESULTS: </strong>ML models were evaluated on holdout subsets with receiver-operator and precision-recall area-under-the-curve, F1 score, and Matthews correlation coefficient. ML models outperformed no-skill prediction. Metabolomic profiles were identified based on cause. FSGS was associated with the sphingomyelin-ceramide axis. FSGS was also associated with individual plasmalogen metabolites and the subpathway. OU was associated with gut microbiome-derived histidine metabolites.</p>

<p><strong>CONCLUSION: </strong>ML models identified metabolomic signatures based on CKD cause. Using ML techniques in conjunction with traditional biostatistics, we demonstrated that sphingomyelin-ceramide and plasmalogen dysmetabolism are associated with FSGS and that gut microbiome-derived histidine metabolites are associated with OU.</p>

10.1681/ASN.2021040538

J Am Soc Nephrol

35017168

Metabolite Biomarkers of CKD Progression in Children.

2021

1178-1189

2021 Aug

1555-905X

<p><strong>BACKGROUND AND OBJECTIVES: </strong>Metabolomics facilitates the discovery of biomarkers and potential therapeutic targets for CKD progression.</p>

<p><strong>DESIGN, SETTING, PARTICIPANTS, &amp; MEASUREMENTS: </strong>We evaluated an untargeted metabolomics quantification of stored plasma samples from 645 Chronic Kidney Disease in Children (CKiD) participants. Metabolites were standardized and logarithmically transformed. Cox proportional hazards regression examined the association between 825 nondrug metabolites and progression to the composite outcome of KRT or 50% reduction of eGFR, adjusting for age, sex, race, body mass index, hypertension, glomerular versus nonglomerular diagnosis, proteinuria, and baseline eGFR. Stratified analyses were performed within subgroups of glomerular/nonglomerular diagnosis and baseline eGFR.</p>

<p><strong>RESULTS: </strong>Baseline characteristics were 391 (61%) male; median age 12 years; median eGFR 54 ml/min per 1.73 m; 448 (69%) nonglomerular diagnosis. Over a median follow-up of 4.8 years, 209 (32%) participants developed the composite outcome. Unique association signals were identified in subgroups of baseline eGFR. Among participants with baseline eGFR ≥60 ml/min per 1.73 m, two-fold higher levels of seven metabolites were significantly associated with higher hazards of KRT/halving of eGFR events: three involved in purine and pyrimidine metabolism (N6-carbamoylthreonyladenosine, hazard ratio, 16; 95% confidence interval, 4 to 60; 5,6-dihydrouridine, hazard ratio, 17; 95% confidence interval, 5 to 55; pseudouridine, hazard ratio, 39; 95% confidence interval, 8 to 200); two amino acids, C-glycosyltryptophan, hazard ratio, 24; 95% confidence interval 6 to 95 and lanthionine, hazard ratio, 3; 95% confidence interval, 2 to 5; the tricarboxylic acid cycle intermediate 2-methylcitrate/homocitrate, hazard ratio, 4; 95% confidence interval, 2 to 7; and gulonate, hazard ratio, 10; 95% confidence interval, 3 to 29. Among those with baseline eGFR &lt;60 ml/min per 1.73 m, a higher level of tetrahydrocortisol sulfate was associated with lower risk of progression (hazard ratio, 0.8; 95% confidence interval, 0.7 to 0.9).</p>

<p><strong>CONCLUSIONS: </strong>Untargeted plasma metabolomic profiling facilitated discovery of novel metabolite associations with CKD progression in children that were independent of established clinical predictors and highlight the role of select biologic pathways.</p>

10.2215/CJN.00220121

Clin J Am Soc Nephrol

34362785

Variability of Two Metabolomic Platforms in CKD.

2018

2018 Dec 20

1555-905X

<p><strong>BACKGROUND AND OBJECTIVES: </strong>Nontargeted metabolomics can measure thousands of low-molecular-weight biochemicals, but important gaps limit its utility for biomarker discovery in CKD. These include the need to characterize technical and intraperson analyte variation, to pool data across platforms, and to outline analyte relationships with eGFR.</p>

<p><strong>DESIGN, SETTING, PARTICIPANTS, &amp; MEASUREMENTS: </strong>Plasma samples from 49 individuals with CKD (eGFR&lt;60 ml/min per 1.73 m and/or ≥1 g proteinuria) were examined from two study visits; 20 samples were repeated as blind replicates. To enable comparison across two nontargeted platforms, samples were profiled at Metabolon and the Broad Institute.</p>

<p><strong>RESULTS: </strong>The Metabolon platform reported 837 known metabolites and 483 unnamed compounds (selected from 44,953 unknown ion features). The Broad Institute platform reported 594 known metabolites and 26,106 unknown ion features. Median coefficients of variation (CVs) across blind replicates were 14.6% (Metabolon) and 6.3% (Broad Institute) for known metabolites, and 18.9% for (Metabolon) unnamed compounds and 24.5% for (Broad Institute) unknown ion features. Median CVs for day-to-day variability were 29.0% (Metabolon) and 24.9% (Broad Institute) for known metabolites, and 41.8% for (Metabolon) unnamed compounds and 40.9% for (Broad Institute) unknown ion features. A total of 381 known metabolites were shared across platforms (median correlation 0.89). Many metabolites were negatively correlated with eGFR at &lt;0.05, including 35.7% (Metabolon) and 18.9% (Broad Institute) of known metabolites.</p>

<p><strong>CONCLUSIONS: </strong>Nontargeted metabolomics quantifies &gt;1000 analytes with low technical CVs, and agreement for overlapping metabolites across two leading platforms is excellent. Many metabolites demonstrate substantial intraperson variation and correlation with eGFR.</p>

10.2215/CJN.07070618

Clin J Am Soc Nephrol

30573658