CLINICAL AND PLASMA PROTEOMIC MARKERS CORRELATING WITH CHRONIC KIDNEY DISEASE AFTER LIVER TRANSPLANTATION

Abstract

Chronic kidney disease (CKD) occurs frequently after liver transplantation (LT) and is associated with significant morbidity and mortality. Thus, there is a pressing need to identify characteristics and biomarkers diagnostic of CKD to enable early diagnosis allowing preemptive interventions, as well as mechanistic insights into the progression from kidney injury to irreversible kidney failure. We analyzed 342 patients who had baseline GFR>60 at time of LT and are now >3 years post-LT. Risk factors for post-LT CKD were compared between 3 different groups defined by current GFR: >90 (n=40), 60–90 (n=146) and <60 (n=156) ml/min. Age, cyclosporine use, and pre-LT GFR were independently associated with new onset CKD. A subset (n=64) without viral/immune disease or graft dysfunction underwent multi-analyte plasma proteomic evaluations for correlation with CKD. Plasma proteomic analysis of two independent cohorts, test (n=22) and validation (n=42), identified 10 proteins highly associated with new onset CKD. In conclusion, we have identified clinical characteristics and a unique plasma proteomic signature correlating with new onset CKD after LT. These preliminary results are currently being validated in a prospective, multi-center study to determine if this signature precedes the onset of CKD and resolves with early interventions aimed at preserving kidney function.

INTRODUCTION

Renal injury resulting in chronic kidney disease (CKD) remains a major issue in the success of liver transplantation (LT). The causes of post-LT CKD are multi-factorial, including advanced age, diabetes, hypertension, hepatitis C virus, pre- and intra-operative kidney injury, and calcineurin-inhibitor (CNI)-induced nephrotoxicity. By 12–24 months post-operatively, more than half of liver transplant (LT) recipients have stage III CKD (GFR <60) and up to 10% have stage IV (GFR <30) (1). Regardless of the pathways that lead to CKD, an elevated serum creatinine, the hallmark of renal dysfunction, is a lagging indicator of renal injury in LT recipients. By the time serum creatinine is elevated, there is already significant and largely irreversible damage to native kidney tissue and function. However, serum creatinine is often the marker relied on in isolation since biopsies of native kidneys are rarely performed and other measures of GFR, often creatinine-based, may be inaccurate in this population. Thus, the ability to consider interventions prior to progression of renal disease is impeded by the lack of early, sensitive and minimally-invasive markers of renal injury.

Biomarkers have been proposed that are potentially more sensitive and specific compared to conventional metrics of kidney function. A significant amount of interest has recently been directed toward clinical applications of blood and urine proteomic biomarkers of acute kidney injury in the general population, such as cystatin C (CyC), neutrophil gelatinase-associated lipocalin (NGAL), interleukin-19 (IL-18), α1-microglobulin, β2-microglobulin, trefoil factor 3 (TFF-3), and fatty-acid binding proteins (FABPs), with significantly less focus on markers of chronic kidney disease, early or advanced (2, 3). Preliminary studies have also suggested that some of these biomarkers can be extrapolated to LT recipients (4, 5), while others have reasonably questioned whether these immune-based biomarkers of kidney transplant injury are associated with native kidney dysfunction in the context of LT (6). Therefore, the aims of this study were to identify clinical characteristics in conjunction with the discovery of plasma proteomic markers linked to new onset CKD after LT.

MATERIALS AND METHODS

Patient Population

This study involved a stepwise approach in identifying and characterizing our LT population with and without CKD and subsequently performing proteomic analyses on subsets to determine markers of new onset CKD. First, our LT database was probed for all de novo LT recipients followed at our center for at least three years post-LT. Patients were excluded if they had abnormal renal function (GFR<60) at the time of transplant, were less than 3 years post-transplant, or had received combined liver-kidney or re-transplantation. These patients were consecutively seen in the outpatient liver transplant clinics at Northwestern. Second, clinical characteristics, immunosuppressive therapies, and laboratory values were collected to determine variables associated with the different stages of CKD (GFR >90, 60–90, <60). Third, we consecutively consented all patients from the larger group for proteomic testing who met further refined criteria: CNI monotherapy; no liver dysfunction or history of viral (hepatitis B or C) or autoimmune disease (autoimmune hepatitis, primary biliary cirrhosis, and primary sclerosing cholangitis). This refined subset was specifically chosen to eliminate potential confounders (graft dysfunction, viral/immune disease) and thus select patients only differentiated by the presence or absence of CKD for the final proteomic analysis.

Plasma Proteomic Assays

In the refined test and validation subsets, multi-analyte plasma proteomic panel analyses were performed using a proprietary Luminex Bead technology and assay platform (Rules Based Medicine, Austin, TX) testing two different multi-analyte panels (MAPs). For discovery, we used the Human DiscoveryMAP® v1.0 (189 proteins). To screen for known kidney injury molecules we used the Human KidneyMAP™ v1.0 (13 proteins). Of note, for all GFR estimates, the isotope dilution mass spectrometry (IDMS) reference measurement-modified MDRD equation was used. Informed consent was obtained at all stages and the study was approved by our institutional review board.

Statistical Methods

Categorical and continuous variables were statistically compared using parametric (Chi-squared, T-test) and non-parametric (Fisher’s exact test, Wilcoxon-Mann-Whitney test) tests as appropriate. For correlations between the results of the MAPs and CKD, two separate analyses were performed, using either GFR as a dichotomous measure (< or >59) or as a continuous measure. The advantages of dichotomous measure analyses are that they are the standard used in the field allowing for comparisons and dichotomous metrics are used clinically to define CKD stages in medical records. The advantages of continuous metric analyses are that renal function deteriorates in continuous fashion over time and thus correlations made to a continuous metric are more likely representative of markers in clinical practice. P<0.05 was considered statistically significant.

RESULTS

342 LT recipients with pre-transplant eGFR (calculated by the MDRD equation) of >60ml/min, all on either cyclosporine or tacrolimus, met inclusion/exclusion criteria. They were first categorized into three groups based on their mean eGFR >3 years post-LT (Table 1). Older age, lower pre-LT eGFR, and the use of cyclosporine rather than tacrolimus were all statistically associated with more advanced stages of CKD post-LT (p<0.0001).

TABLE 1.

Demographics of patients based on degree of CKD after LT.

Characteristics	Group 1: eGFR >90ml/min (n=40)^	Group 2: eGFR 60–90ml/min (n=146)^	Group 3: eGFR <60ml/min (n=156) ^	p values
Mean eGFR pre-LT (ml/min)	96.9	90.7	82.3	<0.0001
Mean LT follow-up (yrs)	4.2	4.4	4.7	0.6795
Age (mean)	45.3	48.8	54.9	<0.0001
Gender (female)	23%	25%	35%	0.0984
Race (Caucasian)	65%	74%	80%	0.1177
Weight (kg) mean	77.5	83.3	85.7	0.0959
HTN (%)	23%	27%	37%	0.3292
DM (%)	23%	14%	20%	0.3292
CyA use (%)	5%	6%	22%	<0.0001
FK506 use (%)	95%	94%	78%	0.0005
Recent CyA trough*	87	78	75	0.13
Recent FK506 trough*	4.7	5.6	3.9	0.18

^{^}The three groups are determined by the eGFR at the time of current analysis

^*Trough drug level (ng/ml) measured within a time window of <3 months of the time of current analysis.

A refined subset (n=64, normal graft function, no viral/immune disease) of the 342 recipients was divided into two independent cohorts (22 test, 42 validation). Samples were independently collected for each set at two different times and tested in two separate assay runs over 6 months apart with the plasma proteomic Discovery and Kidney Injury MAPs. Table 2 displays the patient characteristics or both cohorts. Other than renal parameters as expected, the patients in the post-LT GFR<60 group were older (test and validation sets) and more recently transplanted (test set only). Other demographics or clinical parameters, such as the presence of hypertension and diabetes, were not different between the higher vs. lower post-LT GFR groups.

TABLE 2.

Non-immune non-viral subset characteristics

Group	Group #1: CKD (GFR<60)^	Group #2: Control (GFR ≥ 60)^	p values
Test Set	N=14	N=8
eGFR pre-LT (ml/min)	84.7	92.1	0.02
Mean LT follow-up (yrs)	3.5 ± 1.7	6.5 ± 2.5	0.003
Age (years)	60.0 ± 4.8	51.1 ± 6.8	0.01
Sex (male, %male)	7 (50%)	5 (63%)	0.67
CNI (#TAC/CyA)	13/1	6/2	0.5
GFR	46.2±4.3	79.3±7.9	0.0001
BUN	26.4±3.8	15.0±2.8	0.0008
Cr (mg/dL)	1.53±0.15	1.05±0.15	0.001
ALT (U/L)	37.4±7.4	26.9±8.6	0.16
Recent CyA trough*	73	75	0.61
Recent FK506 trough*	4.1	5.2	0.39
Validation Set	N=19	N=23
eGFR pre-LT (ml/min)	83.2	91.6	0.01
Mean LT follow-up (yrs)	6.5 ± 1.7	5.2 ± 1.2	0.3
Age (years)	62.5 ± 3.6	54.3 ± 3.6	0.002
Sex (male, %male)	10 (53%)	17 (74%)	0.2
CNI (#TAC/CyA)	17/2	21/2	1
GFR	43.4±5.6	76.7±7.9	3e-8
BUN	26.3±2.8	16.4±2.5	2e-5
Cr (mg/dL)	1.77±0.60	1.07±0.08	6e-5
ALT (U/L)	30.9±11.6	32.7±9.1	0.5
Recent CyA trough*	73	81	0.16
Recent FK506 trough*	4.0	5.3	0.11

^{^}The two groups are determined by the eGFR at the time of current analysis and plasma collection

^*Trough drug level (ng/ml) measured within a time window of <3 months of the time of current analysis and plasma collection.

Note: For continuous variables, we report mean and 95% confidence interval and for discrete variables we report counts in each class. Listed p-values reflect significances of the differences between the CKD and Control groups. For continuous traits, the p-values were calculated using the Mann-Whitney U test, while for binary traits (Sex and CNI) we used Fisher's exact test.

For the plasma MAPs, two separate analyses were performed. The first analysis studied associations of MAPs with eGFR as a continuous measure. Analysis of continuous eGFR identified 10 proteins (Table 3) significantly associated with eGFR in both cohorts, of which 4 are present in the specific MAP kidney injury panel (cystatin C, α1-microglobulin, β2-microglobulin, TFF-3), and 6 are present only in the MAP discovery panel (fatty acid binding protein, chromogranin A, apolipoprotein CIII, IL-16, CD40, factor VII). For the second analysis, a dichotomous status variable was defined based on eGFR: control (status=0) for eGFR≥60 and CKD (status=1) for eGFR<60. The analysis of dichotomous status identified 9 proteins significantly associated in both test and validation data (Table 4): α1-microglobulin, β2-microglobulin, factor VII, apolipoprotein CIII, apolipoprotein H, chromogranin A, cystatin C, IL-16, and trefoil factor 3. Of note, plasma levels of previously identified acute kidney injury markers KIM-1 and NGAL were not different among the GFR groups.

TABLE 3.

Screening results for eGFR as a continuous measure.

Protein	cor.T	cor.V	Z.T	Z.V	Z.M	p.T	p.V	p.M	q.T	q.V	q.M
CystC	−0.63	−0.6	−3.3	−4.3	−5.3	0.0016	2.90E-05	8.90E-08	0.052	0.0021	6.70E-06
alpha.1micro	−0.64	−0.57	−3.3	−4.1	−5.2	0.0013	7.20E-05	1.80E-07	0.052	0.0021	6.70E-06
FABP	−0.54	−0.59	−2.6	−4.3	−4.9	0.01	3.50E-05	1.20E-06	0.12	0.0021	3.00E-05
Beta.2Microglobulin	−0.66	−0.45	−3.4	−3	−4.6	0.00088	0.0027	4.60E-06	0.052	0.024	8.60E-05
TFF3	−0.55	−0.48	−2.7	−3.2	−4.2	0.0079	0.0014	2.70E-05	0.12	0.015	0.00031
FactorVII	−0.45	−0.54	−2.1	−3.8	−4.2	0.036	2.00E-04	2.90E-05	0.24	0.0044	0.00031
ApolipoproteinH	−0.47	−0.51	−2.3	−3.5	−4.1	0.026	0.00053	4.40E-05	0.2	0.007	0.00038
CgA.ChromograninA	−0.52	−0.47	−2.5	−3.2	−4.1	0.012	0.0015	4.60E-05	0.12	0.015	0.00038
ApolipoproteinCIII	−0.43	−0.53	−2	−3.7	−4	0.047	0.00026	5.30E-05	0.25	0.0044	4.00E-04
CD40	−0.61	−0.31	−3.1	−2	−3.6	0.0025	0.046	0.00031	0.052	0.19	0.0021

Note: We list the 10 proteins whose association with eGFR is significant at the level p<0.05 in both Test and Validation cohorts. In column headings, meaning of the prefixes is as follows: cor, robust correlation; Z, Fisher Z statistic corresponding to the correlation; p, correlation p-value; q, estimate of the local false discovery rate (FDR). The suffix .T refers to the Test cohort, suffix .V refers to the Validation cohort, and suffix .M refers to a combined (meta-) analysis.

TABLE 4.

Screening results for eGFR as a dichotomous measure (< or ≥ 60 ml/min eGFR).

Protein	cor.T	cor.V	Z.T	Z.V	Z.M	p.T	p.V	p.M	q.T	q.V	q.M
alpha-1-microglobulin	0.56	0.47	2.8	3.2	4.2	0.0065	0.0015	2.20E-05	0.12	0.053	0.0018
beta-2-microglobulin	0.59	0.41	3	2.7	4	0.0036	0.0071	5.70E-05	0.12	0.082	0.0022
Factor VII	0.55	0.42	2.7	2.8	3.9	0.0074	0.0054	9.10E-05	0.12	0.074	0.0024
Apolipoprotein CIII	0.57	0.38	2.8	2.5	3.8	0.0058	0.012	0.00016	0.12	0.1	0.0032
Cystatin C	0.54	0.39	2.6	2.6	3.7	0.01	0.0097	0.00023	0.12	0.091	0.0036
TFF3	0.56	0.35	2.7	2.3	3.6	0.007	0.022	0.00035	0.12	0.12	0.0046
Chromogranin A	0.48	0.4	2.3	2.7	3.5	0.022	0.0083	0.00045	0.18	0.086	0.0047
IL 16	0.43	0.42	2	2.8	3.4	0.048	0.0056	0.00072	0.26	0.074	0.0056
Apolipoprotein H	0.54	0.31	2.6	2	3.2	0.01	0.048	0.0012	0.12	0.17	0.0085

Note: We list the 9 proteins whose association with status is significant at the level p<0.05 in both Test and Validation cohorts. In column headings, meaning of the prefixes is as follows: cor, robust correlation; Z, Fisher Z statistic corresponding to the correlation; p, correlation p-value; q, estimate of the local false discovery rate (FDR). The suffix .T refers to the Test cohort, suffix .V refers to the Validation cohort, and suffix .M refers to a combined (meta-) analysis.

We next used the test cohort to train a predictor of eGFR using the 10 associated proteins and age from the continuous analysis, as the associations between the marker proteins and eGFR were overall stronger than those in the dichotomous analysis (Tables 3 and 4). Because of the relatively small size of the test cohort, we used a gene voting predictor based on univariate linear models. While this predictor may not have optimal accuracy, it is more robust as a predictor than, for example, multivariate regression-based predictors. A scatter plot of observed vs. predicted eGFR in the Validation cohort is shown in Figure 1. The predicted eGFR explains 43% of the variance of the observed GFR (Pearson correlation 0.66, p-value 7e-7) The Receiver Operating Curve of the predictor is shown in Figure 1B. Area under ROC (AUC) is 0.78. Because the variable (protein) selection for the predictor took into account the Validation data, the reported accuracy may be higher than if these 10 proteins were used to predict eGFR in an independent data set. To obtain a lower bound on prediction accuracy, we also trained the same predictor on Test data, with variable selection based on Test data only. The "pure-Test" predictor explains 33% of variance in the Validation cohort (p-value 4e-5, AUC 0.74). This proportion of variance explained and AUC can be considered a lower bound on the performance of the predictor in independent data. The protein markers listed in Tables 3 and 4 exhibit negative correlations with eGFR; the higher the eGFR, the lower the marker expression level measured.

Figure 1.

A. Observed (y-axis) vs. predicted (x-axis) eGFR in the validation cohort. The predictor is trained on the test data. The predictor explains 43% of the variance of the observed eGFR. B. Receiver operating curve of the predictor. Area under the ROC is 0.78.

DISCUSSION

From the clinical and biomarker analyses, we have made several observations regarding CKD after LT. We demonstrated that, prior to transplantation, older age, cyclosporine use, and the presence of even mild GFR impairment are associated with the development of late onset CKD (>3 years; Stage 3 or higher) after LT. Next, in a select population without the influence of confounding variables (graft inflammation, viral/autoimmune disease), we have discovered and made an initial validation of a potentially novel plasma proteomic CKD signature. Interestingly, values of well-accepted kidney injury biomarkers KIM-1 and NGAL did not correlate with CKD. While these markers may signify the presence of acute kidney injury and if measured early post-LT could be predictive of future CKD, they do not appear to be useful as markers of the chronic or progressive state. Overall, this multi-analyte signature should now be tested in a prospective clinical study to determine if it has any value as a predictive assay of early onset CKD that might allow for individualized therapeutic strategies.

The development of an elevated creatinine, particularly >1.5 mg/dL at one year, and diminished GFR have significant clinical implications for LT recipients. The relevance and significance of identifying predictive biomarkers for CKD is foreshadowed by the increasing prevalence of older LT recipients and the fact that the liver allocation system favors recipients with pre-LT renal dysfunction (MELD score). Our study demonstrates that these two factors (age, pre-LT eGFR) are highly associated with the development of new onset CKD. Even mildly diminished GFR (82.3 ml/min) at the time of LT was associated with later onset post-LT CKD. Moreover, GFR is likely overestimated in the pre-LT population. As such, these recipient subsets (older age, mildly diminished pre-LT GFR) are ideal patients to prospectively follow with biomarkers that could predict CKD and allow objective testing of early interventions to prevent progression. Finally, the data show that the use of cyclosporine was associated with more CKD than tacrolimus, similar to other reports suggesting more intense systemic vasoconstriction and hypertension with cyclosporine (7). However, we would caution making conclusions based on our work as cyclosporine was used in only a small percentage of our patients, during an earlier era of transplantation and there was no correlation with drug exposure and CKD.

Kidney injury biomarkers have become an attractive measure of acute injury or early renal impairment for use in pharmaceutical drug trials and in the clinical management of patients at higher risk for renal disease. The vast majority of these markers, such as KIM-1 and NGAL, are elevated in acute tubular injury, contrast-induced nephrotoxicity, or peri-operative renal injury (2). However, there have been few reports of a correlation with these molecules and the development of CKD (8). While KIM-1 and NGAL might be elevated in renal allografts affected potentially by both drug-induced and alloimmune/inflammatory injury, our observations indicate that they are not useful markers of chronic native kidney injury in LT recipients. Prospective studies determining the validity of these established acute kidney markers early post-LT (intra-operative or immediately post-operative) in predicting subsequent CKD are needed, as they may only elevate early, decrease after the acute phase, and be no longer useful in the chronic, progressive state that we studied in the present work.

An important question is whether any proteomic signature for CKD in LT recipients is a surrogate marker of decreased glomerular filtration or reflects the underlying physiology of CKD. In our case, both observations may be correct. Blood levels of low molecular weight proteins such as chromogranin A, cystatin c and the microglobulins are known to be GFR-dependent and yet they have been shown (e.g. children undergoing cardiovascular surgery or screening adults with CKD) to be significantly more sensitive markers than creatinine. In contrast, a study of Factor VII levels showed no differences between hemodialysis patients and normal controls and another study concluded it was a biomarker for preeclampsia (9). Levels of apoliprotein A1 were reported as lower in hemodialysis than control patients and this was linked to increased cardiovascular risk. Trefoil Factor 3 is made in the kidney tubules and is markedly reduced in the urine of rats with acute kidney injury, although blood levels have not been studied in any system (5). Early urinary elevations of IL-16 were recently shown to predict delayed graft function in kidney transplants and there is no evidence that it is cleared by the kidney (10). Finally, an observational study of soluble CD40 levels in diabetic nephropathy patients showed no correlation with the decline of GFR (11).

Another question is whether elevations of any of these markers in the first year post-transplant will be predictive of future CKD. We are not making such a claim. The important contribution of this present work is establishing the proof of principle for a biomarker-based approach by identifying a candidate plasma proteomic signature for CKD in LT. A clinical trial with a prospective, serial monitoring study design is required to determine if it is also predictive of CKD. We also acknowledge that if our signature is not predictive of CKD, then its value relative to serum creatinine and eGFR may be minimal unless we prove it is more sensitive for early diagnosis as suggested already for several of our candidates.

Mechanistically, a number of the over-expressed proteins are established markers of kidney injury in other clinical situations, including cystatin-C, α-1 microglobulin, β-2 microglobulin, and TFF-3, but are now validated here in the LT population with CKD (3, 12). Interestingly, other proteins not currently known and not present in the kidney injury MAP panel were also demonstrated to be highly statistically associated with CKD in the test set and confirmed independently in the validation set using the Discovery panel. CD40 may mediate inflammatory signals in renal tubular epithelial cells (13) and factor VII has been found to be elevated, amongst a number of other proteins, in chronic renal insufficiency (14). The associations found between IL-16, chromogranin A and CKD have not been previously established, although we speculate that these are related to the tissue inflammation and repair processes that are naturally linked to progressive kidney injury mechanisms. Apolipoprotein H, otherwise known as β-2-glycoprotein-1, has been previously detected in the urine in patients with various renal diseases and may help distinguish glomerular and tubular renal dysfunction (15). In the blood, apolipoprotein H binds and neutralizes negatively charged phospholipid macromolecules and is thought to protect from inappropriate activation of the coagulation cascade. Apolipoprotein CIII, also higher with inflammation, and fatty acid binding protein (FABP), could be elevated due to reduced renal metabolism of these plasma proteins but they also have been linked to the hyperlipidemia associated with CKD (16, 17). In fact, the association made here of two apolipoprotein family members as markers for CKD in LT is consistent with the recent findings for a third family member, apolipoprotein AI. In the only other biomarker study of CKD in LT, O’Riordan found elevated plasma levels of apolipoprotein AI using SELDI-TOF mass spectrometry though they did not validate their results in an independent cohort and SELDI-TOF is a difficult platform for clinical biomarker work (18). More recently, two papers demonstrated that either missense mutations (19) or single nucleotide polymorphisms in apolipoprotein AI (20) are highly correlated with FSGS and CKD in African-Americans. Note that in these genetic cases, the levels of the protein were low. While there would have to be another genetic mutation in our Caucasian LT patients, it may explain why we do not find the protein elevated in our study. The mechanism of the link to CKD in the African-American population with apolipopotein AI mutations is presently unknown but taken with our data suggests that some unifying mechanism of renal injury linked to apolipoprotein metabolism is involved.

This study has several limitations. Samples were not collected in a prospective fashion. Only one-time ‘snapshot’ samples were performed in the presence of CKD rather than prior to its onset. Thus, we cannot make any conclusions on their possible predictive or prognostic values. As this was a discovery study, it was our intent to identify candidate biomarkers for validation in future prospective studies of serial collections. We also recognize that patient characteristic data were retrospectively collected and only patients seen in our transplant clinics had biomarker assays performed, which are both subject to selection bias. However, we did not pre-select the patients for study and only collected samples during consecutive follow-up visits. Finally, we cannot determine if the proteomic signatures for CKD are specifically related to CNI therapy vs. other factors (age, preexisting renal disease, diabetes, and hypertension) common in this population, although CNI nephrotoxicity is likely the major culprit. Distinguishing these causes would require renal biopsy which is rarely performed after LT (none in our cohorts) and would be difficult to interpret given the likelihood of multiple contributors to renal injury. However, we did exclude patients with viral or autoimmune disease as proteomic signatures may be affected by these etiologies and, importantly, makes inflammatory CKD related to viral disease unlikely.

In summary, we report a candidate plasma proteomic signature for new onset CKD after LT. It should now be validated in larger, serial biomarker studies to determine if it is predictive of early and impending CKD following LT. In addition, therapeutic strategies (i.e. reduction/elimination of CNI therapy, angiotensin inhibitors, anti-fibrotic agents) aimed at protecting renal function in this population could be targeted at a specific population considered at highest risk for CKD based on clinical (age, pre-LT GFR) and protein signature characteristics. As such, we are currently conducting a multi-center study of CKD after LT addressing the issues of biomarker prediction of renal injury and applications for therapeutic interventions.