Limited by the intrinsic risk of kidney biopsies among cirrhosis patients, we applied unbiased techniques — aptamer-based proteomics and metabolomics — to inform the pathophysiologic pathways that are either up or down-regulated by kidney diagnosis and AKI recovery. We highlight in ANCOVA analyses, adjusted for clinical covariates and the false discovery rate, that there were no metabolomic or aptamer-based proteomic differences between those with HRS-AKI and ATN while identifying significant differences between those with and without AKI recovery. Our random forest and Lasso regression analyses demonstrated that metabolomics can improve the diagnostic accuracy of both AKI diagnosis and AKI recovery, and aptamer-based proteomics can improve the diagnostic accuracy of AKI recovery. Our analyses provide novel insight into pathophysiologic pathways, highlighting the metabolomic and proteomic similarities between patients with cirrhosis with HRS-AKI and ATN while also identifying differences between those with and without AKI recovery.
While our principal component analyses revealed metabolomic clusters that distinguished those with HRS-AKI or ATN from those with normal kidney function, CKD, and Pre-Renal AKI, there was substantial overlap between those with HRS-AKI and ATN. We demonstrate this overlap further in our ANCOVA analyses, where when adjusted for clinical covariates and the false discovery and testing for over 1500 metabolites, there were no significant differences. Our random forest and Lasso regression models highlight that metabolites can improve the diagnostic accuracy of clinical variables alone, with metabolites associated with Methionine, Cysteine, SAM and Taurine Metabolism, Glycolysis, Gluconeogenesis, and Pyruvate Metabolism, Xanthine Metabolism, Glutamate Metabolism, and Xenobiotics being most associated with an HRS-AKI diagnosis as compared to an ATN diagnosis — findings that need to be externally validated.
Unlike the metabolomic analyses, our aptamer-based proteomic analysis did not appear to distinguish between any of the kidney function phenotypes. These results were surprising, as this panel included proteins that had previously been associated with both AKI phenotypes and AKI recovery among cirrhosis patients (e.g., neutrophil gelatinase-associated lipocalin [NGAL], kidney injury molecule-1).30–33 One possibility for this discordance was the utilization of serum as opposed to urine for this study. This was an intentional decision, as oliguria is common among patients with HRS-AKI, one which impacts the feasibility of completing these measurements, as well as the future implementation of any findings. That said, even though previous studies have demonstrated that urine NGAL levels are more strongly associated with outcomes than serum NGAL levels, there still appears to be a diagnostic utility to serum NGAL levels.34 A second possibility is the nature of this case-control study. These were sick, decompensated cirrhosis patients who were hospitalized — as such, even those without or with mild kidney dysfunction could have concomitant medical issues that would increase or decrease particular biomarkers (e.g., acute-on-chronic liver failure and NGAL).35 Collectively, each of our analyses—principal component, ANCOVA analyses, and random forests/Lasso regression—demonstrated that aptamer-based proteomics did not significantly identify patients with HRS-AKI as opposed to those with ATN—a finding that suggests that the protein response to HRS-AKI and ATN are similar.
Both metabolomics and aptamer-based proteomics were able to clearly differentiate between those with and without AKI Recovery. In the metabolomic analyses, we identified 217 metabolites that were significantly different between the two groups — metabolites involved in purine metabolism, amino acid metabolism, lipid metabolism, and xenobiotics; these metabolites encompassed several pathophysiologic pathways, demonstrating the comprehensive physiologic disarray of AKI among decompensated cirrhosis patients. The aptamer-based proteomic analyses corroborated these findings. Of the 30 proteins that were significantly different between the two groups, the majority included proteins that were either linked to inflammation (e.g., Tumor necrosis factor ligand superfamily member 15, Fc receptor-like protein 4:Extracellular domain), baseline kidney health (e.g., cystatin C, Beta-2 microglobulin, and Trefoil factor 3), or lipid metabolism (e.g., cholesteryl ester transfer protein). Collectively these analyses, in a simplified conclusion, highlight that AKI recovery is determined by the degree of inflammation or injury and the baseline kidney health.
This study has several limitations. First, our sample size. We accounted for this sample size statistically by adjusting for the false discovery rate using stringent criteria (i.e., q-values < 0.05) and attempted to limit overfitting with more advanced techniques (i.e., random forests through bagging and random subspace methods to reduce the variance, lasso regression by adding a penalty term to shrink non-informative coefficients to zero). However, the possibility of false negatives leading to limited findings, particularly between those with HRS-AKI and ATN, is possible. That said, this represents the first investigation of metabolomics and aptamer-based proteomics among hospitalized cirrhosis patients with AKI, and even with the sample size, significant differences between those with and without AKI recovery were found. Second, AKI recovery is a dynamic process, and for this study, proteomics and metabolomics were completed only at a single time point. Our cohort and biorepository have been expanded. Future work will center on repeated measures and changes in these pathophysiologic processes that dictate AKI recovery. Finally, hospitalized patients with decompensated cirrhosis are complex, which makes the identification of key pathophysiologic pathways difficult, as there are many opportunities for confounding.36–38 We address this in this study through our case-control design and sample size, which allowed for the review of each medical record to confirm the diagnosis and account for a substantial amount of confounding; nevertheless, the possibility of residual confounding remains.
Despite these limitations, our study applies unbiased methodologies (e.g., metabolomics, aptamer-based proteomics) to determine what pathophysiologic differences exist between those with HRS-AKI and ATN and those with and without AKI recovery. For the former, our data suggest that these clinical diagnoses may be more pathophysiologically similar than appreciated in practice — a novel finding that, if validated, would have important implications for clinical practice. For the latter, our data highlight key differences in both the metabolome and the aptamer-based proteome that inform the pathophysiologic mechanisms of AKI recovery. Collectively, we believe these analyses provide a “proof-of-concept” for the utilization of unbiased multi-omics to inform the pathophysiology of AKI among decompensated cirrhosis patients.