Metabolomics and Kidney Precision Medicine

The idea that alterations in biologic fluids can denote the health status of an individual goes back to ancient times, and has been a foundation of nephrology practice since its inception. Modern profiling technologies seek to update and extend this paradigm. The systematic analysis of metabolites—small molecules such as sugars, amino acids, organic acids, nucleotides, acylcarnitines, lipids, etc.—has drawn considerable interest in nephrology because of the kidney’s fundamental role in small molecule clearance, and the potential biologic effect of select circulating small molecules (1). To date, metabolomics studies of human kidney disease have generally focused on two questions: (1) What is the relationship between the metabolome and the severity of kidney disease? and (2) Do alterations in the metabolome predict future outcomes? These studies have proliferated rapidly, identifying novel markers that correlate with eGFR, are elevated in ESRD, and are associated with future onset of CKD, CKD progression, and death among patients on dialysis (2).

Largely overlooked has been the question of whether the metabolome can provide insight on the underlying etiology of CKD? In one study, Hao et al. (3) profiled urine from 89 individuals with different types of glomerular disease, highlighting select differences between patients with FSGS compared with individuals with membranous nephropathy, minimal change disease, and IgA nephropathy. This study, however, was limited to approximately 30 abundant metabolites detected by nuclear magnetic resonance spectroscopy, and did not fully adjust for baseline differences in albuminuria and other clinical variables. In the current issue of the Clinical Journal of the American Society of Nephrology, Grams et al. (4) performed metabolomic profiling to differentiate CKD cause using stored blood samples from 589 participants of the landmark Modification of Diet in Renal Disease (MDRD) study (5). The MDRD study was composed of two substudies based on enrollment GFR, with study A enrolling patients with a GFR of 25–55 ml/min per 1.73 m² and study B enrolling patients with a GFR of 13–24 ml/min per 1.73 m². For this analysis, cause of CKD was categorized as polycystic kidney disease (PKD), glomerular disease (over 85% biopsy proven), or other (including interstitial nephritis, vesicoureteral reflux, hypertensive nephropathy, single kidney, and unknown), excluding all patients with diabetic nephropathy and all patients with diabetes mellitus treated with insulin. After discovery in study B and replication in study A, the authors highlight five metabolites that differentiate the three groups of CKD etiology, even after adjustment for demographics, randomization group, GFR, proteinuria, and other factors. When added to routine clinical variables, the panel of five metabolites significantly improved two-way classification models, and there is biologic plausibility on why select metabolites may associate with PKD in particular; for example, the link between 16-hydroxypalmitate and impaired fatty acid metabolism.

The study has several strengths. First, the parent MDRD data set is superb, as individuals were well phenotyped with cause of CKD adjudicated by experts and clinical variables reliably assessed, including direct measurement of GFR using urinary iothalamate clearance. Second, the study utilized a leading nontargeted metabolomics platform that uses liquid chromatography-mass spectrometry to quantitate >1000 metabolites, including >700 of known identity. Confidence in metabolite identifications is high, as they have been corroborated by analysis of purified standards, and the platform has a track record of biomarker discovery in nephrology research (6,7). Third, the authors used a thoughtful statistical approach. They began by performing a principal components analysis to identify 63 principal components that accounted for 90% of the variance across the 687 nondrug, known metabolites. In turn, statistical significance in the discovery analysis was set at 0.05 divided by 63. Significant findings were then tested for replication in an independent data set, this time using a Bonferroni adjustment for the seven metabolites highlighted in discovery.

No single statistical methodology is ideal for all large-scale metabolomics studies, but for blood-based biomarker applications, an approach that acknowledges the intercorrelation among metabolites to set a relatively more permissive filter for discovery (at least relative to a Bonferroni adjustment), followed by independent validation, strikes a reasonable balance in minimizing type 2 and type 1 errors, respectively. An alternative approach to address metabolite intercorrelation not utilized in this particular study is to conduct “pathway analyses” to detect and highlight enrichment from defined biochemical pathways (8). However, this approach may be better suited for analysis of intracellular metabolism than blood-based profiles, as circulating metabolite levels reflect diet, gut microbial activity, and renal clearance as well as endogenous metabolism across different tissues rather than a self-contained and integrated metabolic network. In addition, pathway analyses can be biased toward the analytical strengths of a given metabolomics platform, e.g., if a given method is optimized to measure a select class of metabolites or specific chemical characteristics (such as charge or polarity).

The authors are careful to acknowledge important caveats in the interpretation of their study. Most broadly, this study shows that metabolite profiles of individuals with PKD, an illness that does not typically pose a major diagnostic challenge, are distinct from those of patients with alternative CKD etiologies. The exclusion of individuals with diabetes mellitus treated with insulin is notable, as clinicians may harbor more diagnostic uncertainly in this context; for example, weighing the likelihood of diabetic nephropathy versus some other glomerular, tubulointerstitial, or vascular process. Further, metabolomics studies have already shown that both diabetes and insulin use have major effects on the blood metabolome, complicating the interpretation of metabolite profiles in patients with diabetes and kidney disease (9). Finally, both the MDRD-derived discovery and replication cohorts represent individuals with significant decrements in GFR, as even the “higher GFR” individuals from study A had a mean clearance of only 35 ml/min per 1.73 m². Thus, it is unclear if the study’s findings are generalizable beyond established CKD, and the authors appropriately advocate for similar studies in individuals with preserved GFR.

In addition to enhancing their potential clinical impact, the extension of these findings to normal GFR would reduce the possibility that the metabolites of interest are simply reporting on differences in renal excretion across groups. The five metabolites highlighted in this study were negatively correlated with GFR, and kynurenate and hippurate are established uremic solutes. Levels of these metabolites were highest in the PKD group, and although the differences were not statistically significant, there was a trend for lower GFR among individuals with PKD in both the discovery and replication cohorts. Because small molecule handling by the kidney extends beyond filtration and includes tubular secretion and metabolism, statistical adjustment for GFR does not fully account for how differences in renal function might affect the metabolome. Disentangling the effect of reduced clearance from the potential biologic roles of metabolites in disease pathophysiology is a fundamental challenge in all metabolomics studies of CKD, and ultimately requires a combination of clinical, physiologic, and mechanistic investigation.

In summary, this study provides a proof-of-concept for using metabolomics to differentiate CKD etiologies, utilizing both rigorous methodologies and thoughtful study design. As suggested, this template could be used to examine earlier stages of disease, across a wider range of diagnostic entities to enhance relevance to clinical practice. Measurement of metabolites is attractive because of their breadth and abundance in biofluids, with the potential to circumvent more invasive and costly tests. Proteins are also of interest, and perhaps even more likely than metabolites to provide mechanistic insight or reveal new therapeutic targets, but until recently limitations in throughput have precluded the analysis of large patient cohorts; emerging affinity-based proteomic approaches may help overcome this challenge (10). Similarly, the recognition that exosomes, cell-derived vesicles released into blood and urine, contain mRNA has enabled transcriptomic profiling of biofluids to gain insight on intracellular biology and profiling urinary exosomes may provide a window into intrarenal gene expression (11). Taken together, these analytical approaches across “omics” domains are thematically aligned, seeking to provide more detailed, disease-specific information on each individual patient. With their study, Grams et al. raise the possibility that metabolites may make an important contribution in this pursuit of more personalized, precision medicine.

Disclosures

None.