**4. Discussion**

Sophisticated interrogation of urine through advanced technologies for kidney diseases is important as urine provides an attractive alternative biospecimen [35] and unlike invasive biopsies, urine metabolite changes can be diagnostic of advanced tissue injury. Additionally, our data sugges<sup>t</sup> that these urine panels can have much greater sensitivity and specificity over measured serum creatinine [35–39]. Molecular perturbations in the kidney have been previously shown to occur much earlier than both histological changes and clinical alterations in kidney function, and previously published studies confirm that urine is an excellent mirror of these intra-graft molecular changes [11,35,40–49]. In this comprehensive study, we had access to a unique resource of over 300 biopsy matched urine samples archived from kidney transplant patients transplanted at multiple transplant centers, mapped with detailed clinical demographics, which enhanced the results obtained from the urine metabolomic studies conducted in this study.

Prior studies, using different assay platforms, have evaluated the urine metabolome for assessing kidney transplant injury [43,50–52]. Wang et al. [50] applied matrix-assisted laser desorption/ionization Fourier transform mass spectrometry (MALDI-FTMS) for studying acute tubular injury, Blydt-Hanson et al. [51] used liquid chromatography tandem mass spectrometry for studying transplant rejection, Dieme et al. [52] used GC-MS to study the metabolic effects of calcineurin inhibitor drugs in kidney transplant patients, Ho et al. [53] used LC-MS for analysis of alloimmune injury, and Suhre et al. [43] applied LC-MS and GC-MS. Most of these studies have larger metabolite diagnostic panels and only few studies have urine samples that are all biopsy-matched, resulting in little overlap across the identified panels to date. Using all biopsy-matched urine samples matched with pathologist blinded biopsy evaluations, GC-MS, and custom informatics analyses using nonlinear, nonparametric machine learning model development, we were able to greatly refine, as well as cross-validate the performance of small panels of urine metabolites for AR and BKVN. These clinical phenotypes are often difficult to distinguish even by biopsy, and pose a clinical challenge for patient management, specifically concerning the decision between immunosuppression augmentation (in AR) or minimization (in BKVN). In addition to selecting the most informative metabolites to provide discriminant diagnostic models for biopsy matched transplant injury categories, we have also tried to better understand the biological mechanisms that some of the metabolites sugges<sup>t</sup> are dysregulated in transplant injury, using panels in this study and other published studies on the urine metabolome.

Many of the metabolites that we identified as correlated with transplant injury have previously been associated with changes in renal physiology. Taurine is a key metabolite that was included in our AR specific panel and was found to have significantly reduced levels in urine during rejection. Taurine plays a role in different physiologic and biologic processes in the kidney as reflected in urinary excretion patterns. Taurine participates in several physiological functions in the kidney [54–56], such as its role in the renal cell cycle and apoptosis. It also functions as an osmolyte during the stress response. Therefore, the changes in urinary taurine levels seen in this study may relate to the higher burden of tissue injury in AR, and its lower levels in the kidney may reflect a failure in protecting the kidney during immune-mediated damage [57]. The low level of taurine in urine during rejection may reflect a combination of decreased production in the kidney and perturbed osmolar reabsorption of taurine in the medulla [58].

Myo-inositol was found to be a significant biomarker in our model and had been previously shown by Dieme et al. [52] and Suhre et al. [43] to be relevant to and increased in transplant rejection. Urinary myo-inositol was the most important metabolite for discriminating between AR, STA, and IFTA phenotypes in our models. Myo-inositol is an osmolyte of the renal medulla that plays an important role in protecting renal cells from hyperosmotic stress [59]. It is enriched under hyperosmotic conditions via the sodium/myo-inositol cotransporter in the thick ascending limb of the loop of Henle [60]. The kidney is the most important organ for myo-inositol metabolism given that there is high expression of its associated enzymes, l-myo-inositol-1-phosphate synthetase, and myo-inositol oxygenase, in the renal parenchyma [61]. Inhibition of myo-inositol transport has been shown to cause acute renal

failure in rats [62]. Furthermore, it has recently been shown through urine metabolomics profiling of humans that increased levels of myo-inositol are significantly associated with kidney disease and inversely proportional to eGFR [63]. It has also been shown to be elevated in the plasma metabolomic profiles of patients with end-stage renal disease [50,61]. These studies sugges<sup>t</sup> an essential role of myo-inositol in renal physiology. Thus, its higher levels in AR may relate to the severity or progression of rejection. In looking at known perturbations in gene expression levels in AR [14] we observed that the sodium-myo-inositol transporter (SMIT), encoded by *SLC5A3*, is located in the thick ascending limb and functions to reabsorb myo-inositol into the renal medullary cells under conditions of hypertonicity [64]. Thus, perturbations of this transporter may be a key mechanism for acute rejection related tissue injury, mediated by hyperosmolar stress [63]. Given that patients with AR may have preserved kidney graft function and stable serum creatinine levels, the utility of this biomarker may be confounded given its correlation with eGFR. Further studies are needed to further validate role of myo-inositol in the progression of rejection.

These are first observations, should be consolidated with more integrative analysis of multi-omic datasets, and could be further aided by spatial metabolomics. Use of metabolomics for kidney disease outcomes generally could complement diagnostics made using other modalities, including gene expression, cell-free DNA, and proteomics [65–69].

This study benefits from the evaluation of the human urine metabolome in different geographic and demographic cohorts, as the samples came from enrolled patients at Stanford and UCSF. The diversity of patient samples supports the robustness and clinical utility of the described metabolite panels. Translation of these biomarker panels to clinical practice can be done by GC-MS/TOF based assays that are readily available in most commercial labs. Despite the fact that we have analyzed urine from a diverse population and using two different immunosuppressive drugs, additional independent validation studies that allow for prospective clinical testing and application of these metabolite panels will be required to validate the performance of these biomarker panels for diagnostic transplant monitoring. This is important because the diversity of the cohort also introduces many factors that are known to affect the metabolic profiles of patients, such as age, gender, BMI, diet, exercise, comorbidities, and even the time of day as a function of the circadian rhythm [70]. In our cohorts, age and gender were significantly different and as would be expected from a pediatric population, they had few comorbidities. A future study consisting of a larger cohort size that collected these additional details would help to delineate the effects of these parameters on the metabolic profiles. Furthermore, we note that storage length is known to affect the metabolites present and detected in urine studies. While our urine samples were stored at −80 ◦C, this may have influenced the distribution of the metabolites in the samples. Future work should be done to compare biobanked samples versus freshly collected samples to delineate any potential differences in the metabolomic profile.

There are a number of additional limitations to this study. We note that certain patient subsets, such as the BKVN arm had a relatively smaller number of patients (*n* = 22), and this warrants further investigation with a larger number of samples to validate the results. We also note that all samples AR samples used were collected at the time of the rejection, rather than before, and thus we could not assess the predictive value of metabolite alterations in transplant rejection. We believe this would be a useful area of future study, as predictive signatures of rejection prior to clinical AR (e.g., picking up subclinical AR) would be valuable to prevent further decline of kidney function [71].

Nevertheless, urinary metabolite profiles provide an exciting opportunity for rapid bedside–to-bench screening for risk assessment, improved immunosuppression titration, and rejection prevention to ultimately improve transplant and patient outcomes. With the increasing number of studies in this space [43,72], a comprehensive metabolomics picture of allograft outcomes can be created and further contribute to the care managemen<sup>t</sup> of transplant patients.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2077-0383/9/8/2341/s1, Table S1: Metabolite selection by VSURF to discriminate BKVN from all other phenotypes, Table S2: Metabolite selection by VSURF to discriminate BKVN from STA, Table S3: Use of all 266 metabolites in VSURF to discriminate BKVN from STA, Figure S1: Volcano plot displaying fold change and significance of metabolites in the urine of acute rejection patients compared to the patients with stable grafts.

**Author Contributions:** Conceptualization, project administration, funding acquisition, supervision, and resources, T.K.S. and M.M.S.; visualization, T.K.S., A.W.S., and J.Y.C.Y.; methodology, software, validation, formal analysis, investigation, data curation, writing—original draft preparation, and writing—review and editing by all authors. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Sarwal Lab startup funds.

**Acknowledgments:** We are grateful for the help from physicians, clinical coordinators, research personnel, patients, and patient families. We are grateful to Maggie Kerwin for proofreading while revising the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
