ABSTRACT
There is a lack of biomarkers for pre-kidney transplant immune risk stratification to avoid over- or under-immunosuppression. Since the circulating lipidome is integrally involved in inflammation, we hypothesized that the lipidome may provide biomarkers that are helpful in the prediction of antibody-mediated rejection. We used mass spectrometry to detect the plasma lipidome in samples collected over 1 year post-kidney transplant from a prospective, observational cohort of adult kidney transplant recipients (KTR), classified in two groups, one with antibody mediated rejection (AMR) and the other with stable graft function (SC). We used linear discriminant analysis to generate predictive models of rejection. A ‘lipid-only’ model generated from samples taken on day of transplant (T1) revealed a seven lipid classifier (lysophosphatidylethanolamine and phosphatidylcholine species) with misclassification rate of 8.9% [AUC = 0.95 (95% CI = 0.84-0.98), R2 = 0.63]. A clinical model [(using donor specific antibody (DSA) and panel reactive antibody (PRA)] was inferior with a misclassification rate of 15.6% [AUC = 0.82 (95% CI = 0.69-0.93), R2 = 0.41]. A combined model using four lipid classifiers and DSA improved the AUC further to 0.98 (95% CI = 0.89-1.0, R2 = 0.83) with a misclassification of only 2.2%. The polyunsaturated phospholipid subspecies that discriminated the two groups were much lower in the AMR group when compared to the SC group. While the lipidomic profile changed significantly among SC patients on serial sampling post-transplant, such changes were not seen in AMR patients. After taking serial lipidomic changes overtime in SC patients in to account, the AMR group still showed sustained decreased levels of specific lipids at the time of AMR. These findings suggest that a lack of anti-inflammatory polyunsaturated phospholipids could identify patients at a higher risk of AMR at the time of transplant.
INTRODUCTION
The complex biochemistry of human biological systems has been operationally separated into a set of large molecular categories. The metabolome, as it is termed, includes four classes of biologically active molecules that consist of proteins and amino acids, carbohydrates and sugars, nucleic acids (both DNA and RNA), and lipids. The full lipid profile that encompasses the complete set of lipid molecules in a human is termed the lipidome. The general term lipid describes a very large, ubiquitous and diverse class of molecules that have a structural and functional role in biological systems. Lipids are an integral structural component of cell membranes, play a significant role in energy storage, are involved in a variety of signaling pathways and intersect in the complex biochemistry of the other classes of compounds in the metabolome(1). Furthermore, by altering the properties of cellular membranes, the lipidome also has the ability to influence membrane mediated events such as enzyme association with membranes required for some catalytic events. Since first characterized in 2002, alterations of the lipidome have been intensely studied in a variety of conditions(2). Distinct lipid profiles have been identified in the normal state and in a variety of pathologic conditions and in response to specific therapeutic interventions(3–7).
Renal allograft transplantation is the treatment of choice for End Stage Renal Disease (ESRD). In the United States, a shortage of suitable organ donors and resultant organs available for transplant, creates a marked supply and demand discrepancy leaving many patients on the waiting list for prolonged periods of time(8). If evidence based risk stratification could occur pre-transplant then more effective and tailored immunosuppressive strategies could be designed to minimize the risk of rejection and infection post-transplant. Current immunosuppression protocols have resulted in a marked decrease in T-cell mediated rejection, at the cost of long term immunosuppression with its resultant adverse effects including susceptibility to opportunistic infections, graft damage, and metabolic complications such as hypertension, diabetes, and lipid abnormalities which predispose to cardiovascular disease(9,10). However, current immunosuppression protocols are not as effective in suppressing antibody mediated rejection (AMR), which is a major cause of graft loss(10).
At the present time standardized immunosuppression protocols rather than individualized immunosuppression is the routine practice for kidney transplantation, because suitable pre-transplant risk stratification biomarkers that can predict future transplant rejection are not available for clinical practice. It was previously thought that donor specific antibodies and the degree of sensitization might serve as stratification tools, but they have been shown to be inadequate predictors of future rejection (11). Thus, there is an unmet need for biomarkers that could allow for better initial risk stratification while enhancing the benefits/risks of immunosuppression therapy for individual patients.
MATERIALS AND METHODS
Patient Selection
The Virginia Commonwealth University Institutional Review Board (IRB) approved this study. Patients were selected from a prospective observational cohort of a single-institution adult kidney transplant center in the United States. The study population consisted of 16 consecutive patients who developed antibody-mediated rejection within 2 years of kidney transplant and 29 stable control (SC) patients who did not develop rejection at any point of post-transplant follow-up. Serial plasma samples were collected and stored at Time 1 (T1 - pre-transplant), Month 6 (T2) and Month 12 (T3) and then yearly for all patient’s post-transplant as part of an IRB approved biobank protocol at our institution. For the AMR group, serum samples were drawn at the time of transplant (T1), at rejection (T2) and at the end of successful therapy (T3).
The SC patients were selected based on the retrospective observation during the period of the study for stable renal function, with no episodes of rejection, with known adherence to the immunosuppressive regimen, and with a sufficient volume of samples at the appropriate time points for lipid research assays. A minimum follow-up of 2 years was mandated to be a candidate for inclusion in the study. Pediatric kidney recipients and multi-organ transplant recipients were excluded.
At our institution all patients received a standardized immunosuppression induction protocol using anti-thymocyte globulin (Thymoglobulin, Genzyme, Cambridge, MA) with a total of 6 mg/kg over four consecutive days beginning in the operating room. Maintenance immunosuppression included a combination of tacrolimus, mycophenolate mofetil and prednisone tapered to 5 mg/day. Highly sensitized patients received 6 sessions of pre-emptive plasmapheresis with intravenous immunoglobulin (IVIG; 100mg/kg) based upon a pre-specified protocol as reported by us previously (12).
Indication biopsies were performed for acute allograft dysfunction defined as a rise in creatinine >20% above baseline, serum creatinine nadir ≥2.0 mg/dL post-transplant; or delayed graft function >21 days post-transplant. Surveillance biopsies were performed in patients with a positive flow-cytometric crossmatch (T or B >100 mean channel shifts) and/or presence of pre-formed donor-specific antibody [DSA; >5000 mean fluorescence intensity (MFI)] at 1 month and 6-months post-transplant. Biopsies were graded based upon the Banff criteria (13). Patients with AMR were treated with 6-9 sessions of plasmapheresis with intravenous immunoglobulin (IVIG; 100 mg/kg) in conjunction with intravenous methylprednisolone 500 mg administered once daily for 3 days. In selected cases, additional drug therapy with rituximab or bortezomib was instituted based upon clinical response.
The details of antibody testing performed at our center have been described previously (14). Briefly, pre-transplant complement-dependent cytotoxicity (CDC) assays and three-color flow-cytometric cross matching (FCXM) were performed for all patients at the time of transplant. Donor-specific antibodies (DSA) were analyzed using the Luminex platform (Immucor Platform, San Diego, CA) with the use of an HLA phenotype panel (Lifematch Class I and Class II ID, Gen-Probe) and a single-antigen panel (Single Antigen Beads, Immucor Platform). Results of bead assays were measured as MFI. For highly sensitized patients an MFI of >5,000 and for de-novo kidney transplant recipients an MFI >10,000 was considered unacceptable for routine transplantation. Calculated Panel Reactive Antibody (cPRA) was determined using the OPTN calculator from the following url: https://optn.transplant.hrsa.gov/resources/allocation-calculators/cpra-calculator/
Lipidomic Analysis
Serial serum samples were stored at −80°C prior to research use. Upon initiation of experiments, samples were prepared for analysis using an HILIC-based UPLC ESI-MS/MS method. 50 μL of plasma was added to 750 μL of MTBE (methyl-tertiary butyl ether), containing 20 μL of SPLASH internal standards (SPLASH LIPIDOMIX Mass Spec Standard - Avanti 330707), and 160 μL of water. After centrifugation for 2 minutes at 12,300 rpm, 350 μL of supernatant was transferred to auto sampler vials and dried under vacuum. Dried extracts were re-suspended using 110 μL of a methanol:toluene (10:1, v/v) mixture containing CUDA (12-[[(cyclohexylamino) carbonyl] amino]-dodecanoic acid) at a final concentration of 50 ng/ml.
Samples were analyzed on a QTRAP 6500+, with Shimadzu Nexera UPLC. Analytes were separated on a Waters BEH HILIC 1.7 μm 2.1×150 mm column (column temperature = 30°C). Mobile phase A: 10 mM ammonium acetate (pH 8) in 95% ACN (acetonitrile). Mobile phase B: 10 mM ammonium acetate (pH 8) in 50% ACN. Gradient (B%) ramps from 0.1 to 20 in 10 mins; rises to 98 at 11 min, keeps for 2 mins, then drops back to 0.1 and maintains for 3 mins.
Statistical Analysis
A comparison t-test analysis (FDR=0.05) was used to select group differences on the day of transplant. Mean values for each lipids class were obtained by sum and average. Linear Discriminant Analysis with regularized correction (RLDA) models for lipids and clinical parameters were created with a stepwise forward method (Fig. 1). Regression performance was estimated with R2, misclassification error and area under the ROC Curve (AUC). Estimates were validated with bootstrap coefficient interval (Fig. 1). Predictors combined model was cross validated with Random Forest method, and the misclassification out-of-bag error (OOB error) was estimated and compared to the RLDA error for validation (Fig. 1). Changes over time were also estimated using the sparse partial least square method and separation of the groups was validated with a permutation test. A t-test was used to compare two time points within a group and for comparing different groups at matched time points. Data was analyzed with JMP Pro 13 and MetaboAnalyst 3.0. The statistical workflow is depicted in Fig. 1.
RESULTS
Demographic comparison of the two groups prior to transplantation is shown in Table 1. Patients in the AMR group were more likely to be female, re-transplants and had a higher degree of sensitization (higher cPRA) and presence of donor specific antibody (higher DSA) at the time of transplant. They were also more likely to have hyperlipidemia. There were no differences noted for age, race, weight, years on dialysis, type of dialysis, delayed graft function, or the presence or absence of diabetes mellitus.
A comparison of phospholipid (PL) classes at T1 revealed relative concentration differences between SC and AMR (Fig. 2). The concentration of phosphatidylcholine (PC) was significantly diminished in AMR, while there was a trend for an increased concentration of lysophosphatidylcholine (LPC). The AMR group also showed a significantly lower concentration of phosphatidylethanolamine (PE), lysophosphatidylethanolamine (LPE), plasmanylethanolamine (PE-O), and plasmenylethanolamine (PE-P). Although not statistically significant, there was also lower concentration of Phosphoglycerol (PG), lysophosphatidylglycerol (LPG), and sphingomyelin (SM). The activity of phospholipase A2 (PLA2) as a signal of increased metabolism was assessed by the ratio of PL to lysophospholipids (LPL). The AMR group showed decreased ratios of PC/LPC and PE/LPE indicating higher activity of PLA2 at T1. PL degradation, evident for PE, was higher in the AMR group compared to the SC group.
Combined lipid and clinical parameters allow for the prediction of rejection on the day of transplant (T1)
Preliminary data demonstrated that there are significant differences in the pre-transplant lipidome between SC and AMR. This led to the hypothesis that the T1 lipidome or some combination of the lipidome and clinical parameters could provide insight into the risk of future transplant rejection, enabling better risk stratification for kidney transplant recipients. To investigate this possibility, a stepwise regularized linear regression was deployed using models of lipids alone, clinical data alone, and a merged lipid and clinical data to test for prediction accuracy (Table 2). The analysis identified seven distinct lipids that discriminated between AMR and SC with 8.9% of the events misclassified [Area under receiver operating characteristic curve (AUC) =0.95 (95%CI=0.84-0.98), R2=0.63 (95%CI=0.4-0.8)]. A clinical model using cPRA and DSA was inferior with 15.6% of the events misclassified, AUC=0.80 (95%CI=0.66-0.90), R2=0.36 (95%CI=0.16-0.57). Still using a stepwise selection approach, a combined model determined with 4 lipids plus DSA further reduced the misclassification events to 2.2% (Fig. 3), and the AUC improved to 0.97 (95% CI=0.88-1.0), R2=0.81 (95%CI=0.49-0.96).
Further comparison of the four lipids predictors of kidney rejection showed that these lipids are significantly decreased in AMR compared to the SC group. In the PC (18:0 /20:4) plot, it is possible to notice the presence of outliers in both groups (Fig. 4A). Random Forest method was used for statistical validation with 500 bootstrap samples, and the mean decrease accuracy test was used estimate the importance of each predictor to the validation model (Fig. 4B). The result revealed that DSA is the more important clinical biomarker of AMR at T1, and together with LPE (16:0) and PC (18:0/20:4) can discriminate AMR with a very low error (2.2%). The statistical validation also revealed that exclusion of LPE (22:6) and LPE (20:4) in the model would have a minimal effect on the misclassification error. Although in the RLDA modeling training, using the entire study population, the addition of these two lipids takes the model estimation from R2=0.75 to R2=0.81.
Serial analyses of the lipidome over the course of one year identify time dependent lipid changes among patients with a favorable transplant outcome, but no differences among graft recipients with non-favorable outcomes
Following the identification of the lipid differences at T1 and their ability to predict graft rejection in association with measured clinical parameters, we wished to investigate how the lipidome changes over time in patients with a favorable transplant outcome (SC). To achieve this end, serial lipid profiles were analyzed from samples collected at Day 0, 6 months and 12 months post-transplant (Fig. 5). A sPLSDA analysis of the data revealed a statistically significant alteration in the metabolic profile at 6 months post-transplant compared to the day of transplant (Fig. 5A). However, for the subsequent times from 6 months to 12 months, there was no significant change in the lipidomic profile. This finding suggests that stabilization of the lipid changes after transplant is associated with the achievement of improved kidney function and possibly a reduced milieu of inflammation (Fig. 5B). The data was subjected to validation using the permutation test (Fig. 5C) and showed a statistically significant metabolic difference (p=0.034) from T1 to 6 months after transplantation.
Further investigation of the lipid differences between T1 and T2 identified 19 lipids that represent the relevant time dependent alterations in the lipidome that had statistically significant elevations at T2 compared to T1 in the SC group. (Fig. 6). A majority of these lipids changes are LPC, with a few PC, one PE-O, two PE-P, and one PG.
Following the identification of the longitudinal lipid trajectory among patients with favorable transplant outcomes, we investigated the trajectory of the lipidome pre-transplant to post transplant one year, among the patients with non-favorable outcomes (AMR) (Fig. 7). sPLSDA analysis of the data reveal that there was no significant alteration in the lipid profile at pre-rejection and post-rejection compared to T1 (Fig. 7a). While a slight change was observed from T1 to post-rejection (Fig. 7B), validation analysis using permutation testing demonstrated this difference to be non-significant (p=0.869) (Fig. 7C). These findings indicate that in contrast to patients with a favorable transplant outcome (SC), patients with non-favorable transplant outcomes (AMR) demonstrated no change in the lipid profile observed pre-transplant over time.
Significant post-transplant lipid differences were observed between Stable Controls vs. those with Antibody-mediated Rejection
As our data revealed that there were significant T1 vs T2 lipid differences between SC, but not in AMR, we further investigated the data to identify the exact differences in the lipidome between SC and AMR at T2. Any differences identified would indicate an alteration in the lipid metabolic environment at the time of rejection that would distinguish AMR from SC. Since there were no significant differences between T2 and T3 for SC group we chose to use SC at T2 (6 months post-transplant) to compare with AMT at T2 (time of AMR). The analysis revealed a panel of 13 lipids that were found to differentiate the two groups at T2 (Fig. 8). As noted previously, these 13 lipids were again comprised of LPE and PC species containing monounsaturated and polyunsaturated fatty acids, except for LPE (16:0). This data further confirms the presence of a sustained lipid metabolic difference between SC and AMR over time that distinguish these two groups of patients.
DISCUSSION
In this first study, we report novel data that the lipidome could be used to identify kidney transplant patients with a higher risk of antibody-mediated rejection at the time of transplant. In addition, for the first time we demonstrate that combining lipidomic and clinical data to create a model merging the presence of donor-specific antibody and lipids (a reduction of each of the four identified lipid biomarkers, one PC and three LPE species) can discriminate AMR with minimal error even at the time of transplant. Statistical validation suggests that DSA, LPE (16:0) and PC (18:0/20:4) are putative biomarkers that should be further tested in a prospective clinical study. These biomarkers could indicate a state of increased inflammation associated with chronic kidney disease and hemodialysis in selected groups of patients compared with others(15).
Modulation of phospholipids (PL) in chronic kidney disease (CKD) is well described in the literature. In a study of CKD among rats, Zhao et al. identified that PC, PE, LPC, LPE and triacyclglycerides (TG) steadily decreased as the pathology progressed over time (16). Braun et al described that the aged kidney from adult wild-type mice expresses significant decreases of PC, PE, PG, SM, phosphatidylserine (PS), and Ceramides, suggesting that change in PL metabolism is associated with CKD (3). Kobayashi et al. reported an elevation of LPE 20:4 in the plasma of adenine-induced CKD rats when comparing with control animals(17). In a human study comparing healthy controls and CKD patients, Reis et al. found that the content of total PC and Ceramides were decreased along with the ratio of LPC/LPE(18). In a study comparing patients with CKD progression compared to control patients, Afshinnia et al. reported that CKD progression was associated with lower Cholesteryl ester (CE), diacylglycerols (DG), PC, plasmenylcholine (PC-P), PE-P, and phosphatidic acid (PA), and elevated PE and monoacylglycerols (MAG)(19). This finding suggests that patients with CKD progression with a decrease of longer acyl chains and polyunsaturated lipids might benefit from the effects of polyunsaturated fatty acid supplementation, as some previous studies have suggested(20,21). In our study, although both groups represent patients who had CKD progression, the SC group had higher PC and LPE than the AMR group and a trend for lower LPC suggesting that subpopulations with varying degrees of inflammatory milieu might exist with the CKD population. This would be consistent with the real-life observation of patients who have varying degrees of risk of rejection.
LPC has being associated with pro-inflammatory effects(22), but there is not much information about the effects of LPE. Some studies suggest that LPE could have a possible protective effect over inflammation. Schober et al. demonstrated that LPE generation from PE oxidation is primarily due to PLA2 activity rather than by hypochlorous acid generated by myeloperoxidase, while LPC can be generated from both processes(23). The dual effect of PLA2 is well known by its pro-inflammatory action in hydrolysis of PC to produce LPC promoting atherogenesis, as well as its anti-inflammatory action in hydrolysis of platelet-activating factor (PAF) and oxidized PLs(24). This suggest that processes that are not directly related to oxidative stress generate LPE in CKD patients. The activity of PLA2 in our study was assessed by the ratio of PL to LPL. The AMR group had a higher PLA2 activity, especially for degradation of PE to produce LPE. The PC/LPC ratio, as an inflammatory marker is also indirectly represented by the increased activity of PLA2 in inflammatory diseases(25,26).
It has been reported that in vitro LPE induces activation of the mitogen-activated protein kinase (MAPK) cascade, an intracellular signal transduction pathway that controls growth, proliferation, differentiation, motility, stress response, and has a survival along with ananti-apoptotic effects(27). Also LPE increases intracellular Ca2+ through a Lysophosphatidic acid (LPA) G-protein-coupled receptor (GPCR)(28). Oral administration of LPE in rats with zymosan A-induced peritonitis demonstrated a vast anti-inflammatory action. In that study LPE-containing polyunsaturated fatty acids administration inhibited plasma leakage by diminishing the formation of LTC4, inhibited the leukocyte extravasation into the peritoneum, decreased formation of potent chemotactic factors such as LTB4 and 12-HETE, lowered IL-1β, IL-6, TNF-α, and augmented IL-10(29).
Our results suggest that the lack of anti-inflammatory protection in patients on the day of transplant is a risk for future rejection. No relevant changes occurred for the AMR group until the onset of rejection, confirming that the metabolic profile at T1 predicting AMR persisted after transplantation. Accordingly, over time comparison of SC and AMR showed that the difference in LPE and PC levels were sustained after 6 months representing the metabolic difference between rejection and non-rejection. The presence of monounsaturated and polyunsaturated fatty acids in PL is also an indication that their low plasma content is a risk factor for kidney health (30). In contrast, the elevation of LPC, PC, PE-O, PE-P, and PG after 6 months in SC group imply that restauration of PL content is the result of successful transplantation. Indeed, some studies have shown that elevation of polyunsaturated fatty acids present a lower risk of developing end-stage renal disease (31), as well as higher survival rates after kidney transplantation(32).
There are some limitations to our study. Demographic comparisons between the SC and AMR groups at T1 revealed that female gender, re-transplant, cPRA, DSA, and hyperlipidemia were statistically more likely to be present in the AMR group. Moreover, we found DSA as the strongest predictor of AMR. These findings are consistent with Dunn et al. who reported that DSA and female gender were risk factors for AMR (33). Thus, the two groups could have been inherently different biochemically. Future larger studies with an increased sample size would be need to confirm this preliminary study. Our finding of hyperlipidemia in AMR group could be linked to the fact that hyperlipidemia is the most common form of dyslipidemia, a common complication in CKD patients, associated with the decline in kidney function, hypertriglyceridemia, low HDL, and low or normal LDL (34).
CONCLUSION
Our study for the first time identifies the pre-transplant, post-transplant, and pre-rejection lipid differences that distinguish kidney transplant patients with favorable transplant outcomes (SC) and a major cause of non-favorable transplant outcomes (AMR). We further demonstrate that unlike SC patients that demonstrate a dynamic longitudinal lipid change, AMR patients maintain a relatively unchanging lipid profile over time with respect to the measured lipids. In addition, we demonstrate for the first time the feasibility of risk stratification of kidney transplant patients on the day of transplant about the possibility of prediction for future AMR. Following prospective validation in a larger cohort, these findings have the potential to alter the current paradigm of pre- and post-transplant monitoring. Treatment of these patients with an evidenced based risk stratification strategy could vastly improve the success of kidney transplantation.
DISCLOSURE
There are no conflicts of interest to report for any of the authors.
ACKNOWLEDGEMENT
Research reported in this publication was supported by research grants from National Institutes of Health under grant numbers HD087198 (to DSW). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This work also received support via a Young Investigator Award from SCIEX for clinical lipidomic research (DSW).