A molecular signature for delayed graft function

Abstract Chronic kidney disease and associated comorbidities (diabetes, cardiovascular diseases) manifest with an accelerated ageing phenotype, leading ultimately to organ failure and renal replacement therapy. This process can be modulated by epigenetic and environmental factors which promote loss of physiological function and resilience to stress earlier, linking biological age with adverse outcomes post‐transplantation including delayed graft function (DGF). The molecular features underpinning this have yet to be fully elucidated. We have determined a molecular signature for loss of resilience and impaired physiological function, via a synchronous genome, transcriptome and proteome snapshot, using human renal allografts as a source of healthy tissue as an in vivo model of ageing in humans. This comprises 42 specific transcripts, related through IFNγ signalling, which in allografts displaying clinically impaired physiological function (DGF) exhibited a greater magnitude of change in transcriptional amplitude and elevated expression of noncoding RNAs and pseudogenes, consistent with increased allostatic load. This was accompanied by increased DNA methylation within the promoter and intragenic regions of the DGF panel in preperfusion allografts with immediate graft function. Pathway analysis indicated that an inability to sufficiently resolve inflammatory responses was enabled by decreased resilience to stress and resulted in impaired physiological function in biologically older allografts. Cross‐comparison with publically available data sets for renal pathologies identified significant transcriptional commonality for over 20 DGF transcripts. Our data are clinically relevant and important, as they provide a clear molecular signature for the burden of “wear and tear” within the kidney and thus age‐related physiological capability and resilience.


| INTRODUCTION
The changing demographics of age in human society is anticipated to result in a major burden of age-related morbidities, as improvements in health span have failed to match the increase in average global lifespan. Notably, deaths due to chronic kidney disease have increased globally, despite significant decline in other aetiologies (Christopher & Murray, 2015). This is pertinent to researchers investigating ageing, as accelerated cellular and physiological ageing are underlying components of renal dysfunction (Kooman, Kotanko, Schols, Shiels, & Stenvinkel, 2014;McGlynn et al., 2009;Schmitt, Susnik, & Melk, 2015), where systemic differences are layered on top of the dysregulated ageing process and patients show a higher incidence of mortality in comparison to healthy chronologically agematched individuals. As the prevalence of CKD parallels an increased prevalence in type 2 diabetes, obesity and a sedentary lifestyle (Stengel, Tarver-Carr, Powe, Eberhardt, & Brancati, 2003), an allostatic outcome reflecting the "burden of life style" may be present next to the renal dysfunction.
Allostatic load can be defined as a composite indicator of accumulated biological stress over the life course, which predisposes to morbidity in the face of chronic or repeated stress exposure (Rubin, 2016). It is reflective of the biological age of a tissue organ or organism, as it directly impacts on age-related physiological function.
We have previously developed the use of renal allografts as an in vivo model to study healthy tissue ageing in humans, whose physiological function can be tracked longitudinally to demonstrate that allograft biological age is more important than chronological age in prognostication of post-transplant allograft performance (Gingell-Littlejohn et al., 2013;McGuinness et al., 2016).
One testable prediction following this demonstration is that organs with increased biological age should reflect the cumulative burden of "wear and tear" and thus be less resilient to transplant-related stresses and display reduced physiological function as a consequence. Assessing the related changes in molecular biology in these renal allografts is not straightforward (Shiels, McGuinness, Eriksson, Kooman, & Stenvinkel, 2017). To do so, we have used an analysis of a notable clinically relevant allograft phenotype displaying impaired physiological function, termed delayed graft function (DGF), to determine whether organs showing DGF are less resilient than those undergoing immediate graft function (IGF). Additionally, we have determined whether features associated with lack of physiological resilience were reflected in the pretransplant transcriptomes of respective organs.
A higher incidence of DGF has been associated with the use of allografts from older extended criteria donors (ECD,age >60,or >50 with two of the following: a history of high blood pressure, a creatinine ≥1.5, or death resulting from a stroke), donation after cardiac death donors (DCD) and increased allograft biological age (Mallon, Summers, Bradley, & Pettigrew, 2015;McGuinness et al., 2016;Menke, Sollinger, Schamberger, Heemann, & Lutz, 2014;Mundt, Yard, Kramer, Benck, & Schnulle, 2015;Schroppel & Legendre, 2014). The extent to which donor and recipient-related characteristics influence the magnitude of IRI and/or DGF occurrence, beyond accepted clinical risk factors for DGF, remains to be proven (Menke et al., 2014;Mundt et al., 2015;Schroppel & Legendre, 2014), particularly in the context of allograft repair, or regeneration pathways, activated in response to IRI. Increased demand for organ donation, coupled with increasing chronological age and associated comorbidities in the donor population, has necessitated the use of organs that have been previously deemed as marginal for clinical use (Morrissey & Monaco, 2014;Nagaraja et al., 2015).
DGF has also proven refractory to modelling both in vitro and in preclinical model organism studies. This study aimed to identify a human-specific molecular signature associated with DGF and to enable adjustment for the effects of IRI-related molecular changes, in the absence of model systems for analysis of DGF mechanisms, as well as providing direct insight into its manifestation. Additionally, this strategy was designed, to enable the validation of any DGF-associated signature, by comparison with existing publically available renal data sets, in order to elucidate whether there were common underpinning molecular processes in their manifestation and whether these reflected the burden of "wear and tear" in the kidney. We selected a very closely matched clinical cohort based on age, gender, length of ischaemic time and low HLA mismatch.
These were divided into two groups based on recovery of organ function after transplantation, with emphasis on extreme functional differences, namely either DGF or IGF (Table1). An extreme DGF phenotype was defined as (a) the need for dialysis within 7 days of transplantation, with the exception of hyperkalaemia on the first postoperative day, and (b) the failure of serum creatinine to reduce by 50% within the first week, which is indicative of poor recovery of renal function (Figure 1a; Aitken et al., 2015). IGF was defined as reduction in serum creatinine by 50% in less than 3 days post-transplantation. Additionally, we undertook a retrospective analysis of paired allograft biopsies from this cohort, obtained at two time points: preimplantation during the preparatory phase (preperfusion) and after allograft reperfusion when circulation was restored (postperfusion). We have used this approach to test the hypothesis that DGF is a manifestation of organ "wear and tear" (i.e. its allostatic load as a function of its biological age) and that the impaired physiological capability can be defined using a specific set of molecular features, independently of allograft damage acquired during the peritransplantation period.

| RESULTS
2.1 | RNAseq cohort characteristics are related to differences associated with DGF and response to reperfusion injury Differential gene expression associated with perfusion, or DGF status, was assessed alone, or stratified by donor gender (Figure 1b).
This was more pronounced than the pre-/postperfusion or DGF/IGF variance, each of which grouped into distinct gender clusters (Supporting Information Figure S1 in Data S1). The magnitude of change in pre-versus postperfusion biopsies was more pronounced than in a comparison of DGF versus IGF biopsies (Figure1b, Supporting Information Figure S2 in Data S1). Hierarchical clustering of the samples is presented in Figure 1c.  Table S1 in Data S2, Figure 2). DGF outcome and donor gender were related to the DGF-specific signature, with male and female donors forming distinct clusters indicative of a donor gender-driven DGF phenotype (Figure 2a). Further analysis of DGF-specific transcripts revealed that overall expression changes in response to reperfusion occurred along a similar trajectory in both DGF and IGF, but the magnitude of this change was greater for those exhibiting DGF.
This suggests that the degree of response to reperfusion injury is significant in post-transplant outcome (Supporting Information Figure S2 in Data S1). Further analysis of the transcriptome for DGFspecific signatures independent of IRI, but stratified by BioAge, revealed the presence of only 22 DGF-specific targets (Figure 2b).
Additional analyses, involving comparison of the allograft response to IRI (both for DGF and IGF status), using the respective RNAseq data sets, with publically available data sets for other renal pathologies, were undertaken to identify additional transcripts. These were included in further validation testing (Supporting Information Datas S1 and S3). overlapping with processes inherent in ageing, including eIF2 (eukaryotic initiation factor 2) signalling, protein ubiquitination, eukaryotic initiation factor 4 (eIF4) and ribosomal protein S6 kinase beta-1 (p70Sk6) signalling and interleukin-17 (IL-17)-mediated cytokine regulation (Supporting Information Tables S1-S3 in Data S5).

|
These data suggest that DGF outcome may be related to a differential capacity to restore physiological homeostasis following IRI. and an inability to restore transcriptional and physiological processes to function within normal physiological parameters as quickly as organs with IGF.

| Epigenetic status is linked to DGF and perfusion status
The effect of IRI on epigenetic status associated with DGF outcome was analysed in connection with changes in global DNA methylation DGF-specific transcripts revealed differential promoter methyla- were analysed, with only 42 displaying AS events, as indicated by differential exon expression (Figure 3a,b, Supporting Information Data S6). A representative example of an AS event has been illustrated using interferon regulatory factor 1 (IRF1) (Figure 3c,d). This is suggestive of a complex relationship between methylation status and alternative splicing associated with IRI injury that is not observed in organs displaying DGF.

| Validation of DGF-specific transcripts
Transcripts related to DGF, selected based on their ranking by statistical significance of observed expression change and diversity of signalling pathway involvement in relation to publically available data sets, were further validated in 19 paired biopsies from the RNAseq cohort and an independent cohort of 32 pairs of samples. Three samples were excluded from further analysis due to post-transplant complications. Nineteen genes were validated as markers of DGF.
Transcripts were further analysed in relation to donor characteristics, ischaemic time and estimated glomerular filtration rate (eGFR/ MDRD4), as a measure of physiological function, at 3, 6 and 12 months post-transplantation (

| Validation of DGF gene expression at the protein level
The expression of DGF-specific genes at the protein level was undertaken using western blotting in a subset of samples to determine whether changes at the protein level could be related to DGF    Overall, regardless of DGF/IGF status, positivity for γH2AFX appeared to be predominately nuclear in tubular epithelial cells, at least mild and more often in the distal tubules than the proximal tubules. Interestingly, the subcellular localization of the γH2AFX signal within arterial myocytes appeared to associate with donor age more so than with immediate function status after transplantation, in that biopsies from kidneys from younger donors showed more cytoplasmic than nuclear γH2AFX, with older kidneys showing greater nuclear than cytoplasmic positivity in arterial myocytes.

| DISCUSSION
To our knowledge, this work is the first study in human subjects that demonstrates a unique molecular signature for impaired physiological resilience (DGF), encompassing epigenetic and transcriptomic data sets. Critically, at a translational level, it also provides a platform for the development of a universal IRI signature and the ability to relate it to post-transplant outcomes. This is also the first study linking DNA methylation status to reperfusion injury and DGF outcome, in the context of immune system status, overall dysregulation of cellular homeostasis and its consequences for allograft performance.
These data, together with the validation of DGF-associated gene products at the protein level, provide a unique and synchronous genome, transcriptome and proteome snapshot.
Our study provides strong evidence that biological age in combination with physiological stress, resulting from immune system activation and generation of inflammatory responses, plays a major role in DGF occurrence and the physiological manifestations of IRI. From a clinical perspective, this also suggests that these effects are driven by donor characteristics, which may therefore be even more discriminating than reperfusion injury itself. Correspondingly, BioAge in the preperfusion biopsy analyses appeared to be a significant determi-  . Our data are in keeping with such a scenario, with organs displaying DGF also exhibiting a greater change in transcriptional amplitude in DGF signature transcripts following transplantation and requiring a longer period to restore physiological homeostasis, which may be related to deficient proteostasis. These data indicate that allografts exhibiting DGF may therefore be displaying features of allostatic overload at a transcriptional level whose effects are extrapolated across the organ as a whole, resulting in functional impairment (Kooman et al., 2014).
Analysis of transcript expression observed solely in DGF, but not IGF, in relation to perfusion status, indicated a transcript biotype shift (Supporting Information Table S1 in Data S8), including an increase in antisense, pseudogenes, noncoding and coding RNAs, for example immunoglobulin gene (IgV and IgV pseudogenes) and T-cell receptor (TR J) transcripts. These changes in the transcriptome biotypes further support the hypothesis that the response to IRI, both in magnitude and context, are dependent upon donor characteristics and organ response/resilience to stress and may also reflect deregulation of alternative splicing networks and epigenome status overall.
The biotype changes observed may reflect a burst of "transcriptional noise" in DGF allografts in response to IRI, as a direct result of changes in the methylation status of promoters and intragenic regions (Huh, Zeng, Park, & Yi, 2013). These observations are also consistent with the derepression of LINE elements in ageing cells (De Cecco et al., 2013).
Immunological response to "danger signals" may lead to excessive activation of proinflammatory cytokines and chemokines, consequently leading to organ damage over time, as observed in the case of autoimmune diseases, where there is a loss in ability to downregulate/attenuate proinflammatory signalling (de Jesus, Canna, Liu, & Goldbach-Mansky, 2015). This suggests that donor characteristics are important for DGF occurrence and may be linked to organismal/ organ stress levels in relation to the type of organ donation (Bon et al., 2012;Morrissey & Monaco, 2014). Ultimately, allograft quality will be related to organ resilience to stress, and this by itself provides an opportunity for the development of new therapeutic interventions aimed at exploiting this phenomenon. Recent studies focusing on the progression of chronic kidney disease (CKD) have exemplified the importance of epigenetic changes and loss of kidney function (Smyth, McKay, Maxwell, & McKnight, 2014;Wing et al., 2014). Our data have indicated rapid changes in the epigenome during perfusion, suggesting that the effect of IRI on long-term allograft function may be more pronounced than originally anticipated.
The lack of available models for DGF has meant that direct validation of targets and any related mechanism has not been possible.
To mitigate the impact of this shortfall, we have therefore used our DGF transcriptomic signature to identify any commonality with other renal/urological transcriptomic expression profiles derived from publicly available data sets. These included renal interstitial fibrosis, kidney transplant failure and rejection, kidney disease, nephrotic syndrome, cystic disease of kidney and renal tubular disorder (Supporting Information Data S3). Notably, the overlap identified encompassed transcripts involved in immune system activation, both supporting the importance of interaction between lymphoid and nonlymphoid cells in the context of renal function and highlighting the biological plausibility of the DGF-related findings.
Overall, our data suggest that allografts exhibiting DGF present with an impaired ability to restore physiological homeostasis in response to stress, consistent with their biological age and associated allostatic load. This is reflected in changes in epigenome, transcriptome and dysregulation of RNA metabolism. The magnitude of change in transcriptional amplitude in response to physiological stress, along with elevated expression of noncoding RNAs and pseudogenes, raises the possibility that reduction in available cellular resources for activation of damage repair mechanisms slows down physiological and cellular repair processes, resulting in long-term damage to the allograft ( Figure 5).

| Clinical cohort characteristics
Fifty-five paired preperfusion and postperfusion renal biopsies collected from deceased donors were included in this study, and all kidneys were subsequently transplanted with no occurrence of primary nonfunction. Detailed patient characteristics and follow-up markers are described in Table 1. This study and consent procedure was approved by the Regional Ethics Committee of North Glasgow NHS were selected for further analysis (Figure 1a).

| Biopsy processing
Total RNA, genomic DNA and protein were sequentially isolated from the same tissue biopsies using TRI ® Reagent according to the manufacturer's instructions (Invitrogen, UK).

| Alternative splicing analysis
Alternative splicing events were investigated in the RNAseq cohort (top 100 differentially expressed transcripts, pre vs. post) and DGFspecific transcripts identified by RNAseq using DEXSeqv1.16.10 (Anders, Reyes, & Huber, 2012). Differential exon usage (DEU), followed by the application of generalized linear modelling, was used to identify alternative splicing events (FDR < 0.1 was considered significant).

| Whole genome bisulphite sequencing and analysis
Genomic DNA was isolated from the same biopsy, using TRI®Reagent, after separation of RNA into aqueous fraction and further puri- (2 × 150 bp). The raw sequence reads in FASTQ format underwent QC as previously described above. Sample 180b1 was excluded from further analysis as it failed QA and QC.
The human reference genome (GRCh38) was in silico bisulphite converted before aligning trimmed reads with Bismark v0.10.1 and bowtie2 (v2.1.0) as previously described (Krueger & Andrews, 2011;Langmead & Salzberg, 2012). PCR bias was removed by a deduplication step. The methylation content was measured on CpG context sites of DGF-specific targets. The promoter and intragenic regions were extracted from biomart API (BiomaRt; Durinck, Spellman, Birney, & Huber, 2009), and differences within the methylated CpG sites were further analysed using Kruskal-Wallis test. FDR correction for multiple comparison was applied for all analyses. Adjusted pvalue below 0.05 was considered as statistically significant.

| QPCR and data validation
For each individual RT reaction, 150 ng of total RNA from each sample was used and reverse transcription was performed using SuperScript ® II Reverse Transcriptase (# Life Technologies Inc., UK) and then qPCR was performed. Gene expression was analysed using TaqMan®gene expression assays, or custom design assays using Roche UPL (Supporting Information Data S9), which were normalized against HPRT1 and 18S rRNA control primer sets. Taqman® assays, including standards, were performed using the manufacturers recommended qPCR protocols and TaqMan®Master Mix (#4370074, Life Technologies, UK). For UPL probes, primers were used at final concentration 360 nM while probes were used at final concentration of 100 nM.
The comparative threshold cycle method (ΔΔCT) was used to quantify relative gene expression, and the obtained quantification was transformed to exponential value 2 −ΔΔCT . Commercially available RNA was used as a calibrator (#AM7976, Life Technologies, Inc.). Further testing involved Spearman correlations and Kruskal-Wallis test. FDR correction for multiple comparison was applied for all analyses. Adjusted pvalue below 0.05 was considered as statistically significant.

| Western blot
Protein fractions were isolated from the phenol-ethanol fraction after removal of genomic DNA (TRI ® Reagent, Invitrogen, UK). Protein concentration was estimated using DC TM Protein assay (BioRad, UK), and 7.5 µg of total protein was loaded per well. Samples were resolved in