Cytokine absorption during human kidney perfusion reduces delayed graft function–associated inflammatory gene signature

Transplantation is the optimal treatment for most patients with end‐stage kidney disease but organ shortage is a major challenge. Normothermic machine perfusion (NMP) has been used to recondition marginal organs; however, mechanisms by which NMP might benefit organs are not well understood. Using pairs of human kidneys obtained from the same donor, we compared the effect of NMP with that of cold storage on the global kidney transcriptome. We found that cold storage led to a global reduction in gene expression, including inflammatory pathway genes and those required for energy generation processes, such as oxidative phosphorylation (OXPHOS). In contrast, during NMP, there was marked upregulation OXPHOS genes, but also of a number of immune and inflammatory pathway genes. Using biopsies from kidneys undergoing NMP that were subsequently transplanted, we found that higher inflammatory gene expression occurred in organs with prolonged delayed graft function (DGF). Therefore, we used a hemoadsorber (HA) to remove pro‐inflammatory cytokines. This attenuated inflammatory gene expression increased OXPHOS pathway genes and had potentially clinically important effects in reducing the expression of a DGF‐associated gene signature. Together, our data suggest that adsorption of pro‐inflammatory mediators from the perfusate represents a potential intervention which may improve organ viability.


| INTRODUC TI ON
Kidney transplantation represents the optimal treatment for most patients with end-stage kidney disease, with benefits for both quality and quantity of life. 1 Organ shortage is a major challenge, and several strategies have been employed to increase the number of kidneys available, including the use of deceased circulatory death (DCD) donors and extended criteria donors (ECD), both of which are associated with higher rates of delayed graft function (DGF) compared with deceased brainstem death (DBD) donor. 2,3 In DCD kidneys, exposure to warm ischemia during the process of circulatory cessation makes a significant contribution to DGF. DGF occurs due to ischemic tubular cell damage or death, which can stimulate innate immune activation via NLRP3 inflammasome assembly leading to the generation of interleukin (IL)1β and IL18. [4][5][6] Indeed, the presence of inflammatory cytokines in urine has been used as a biomarker of acute kidney injury and DGF. [7][8][9] Normothermic machine perfusion (NMP) allows transplanted organs to be perfused with warm, oxygenated red blood cells, in the absence of circulating immune components, including complement and neutrophils. [10][11][12][13][14] This process has been used to assess marginal organs 15 and to "recondition" organs to facilitate the transplantation of kidneys that were initially declined following offer via national organ allocation service. 16 Our previous experience using NMP provided the rationale for a randomized controlled trial to assess its efficacy in preventing DGF in DCD kidneys, 17 but the mechanisms by which NMP might benefit transplant kidneys are not fully understood. Furthermore, the question of whether additional manipulation of the kidney during NMP, for example by removal of pro-inflammatory cytokines and chemokines from the perfusate, might offer additional benefits in optimizing the organ prior to transplantation has not been addressed in human kidneys, but our study of porcine NMP demonstrated promising results. 18 Here we took an unbiased approach to address these two questions using transcriptomic analysis of human kidney biopsies taken at the start and end of NMP to assess global changes in gene expression. We used pairs of human kidneys from a single donor enabling a comparison of the effect of different interventions in organs with an identical genetic background and were able to assess the impact of those changes on a prediction of graft outcome.

| Paired kidneys are genetically similar and a useful model for assessing interventions
We investigated the transcriptional changes associated with organ preservation and NMP in two independent studies, each with five pairs of kidneys, making a total of 10 pairs/20 human kidneys. Of these, two were from DBD and eight from DCD donors (Table S1, S2 and Figure S1). In these studies, we took cortical biopsies pre-and postintervention (cold storage or NMP) and investigated the transcriptional landscape using RNA sequencing (RNA-Seq). The use of paired kidneys and paired biopsies from the same kidney allowed us to control for biological variation and to regress this confounder out of the analysis ( Figure S2A). We confirmed that kidney pairs start with a common transcriptional landscape (Fig. S2B) and by applying different interventions to each kidney in a pair, we are able to study the effect of an intervention independent of biological variation.

| NMP results in an increase in OXPHOS and inflammatory pathway genes compared with cold storage
To investigate the potential mechanisms by which NMP might impact kidneys, we initially took five kidney pairs (n = 4 DCD donors and n = 1 deceased brainstem death [DBD] donor, Figure S1, Table S1) and performed a time 0 (0 hour) cortical biopsy. At this point, the kidneys were randomized to static cold storage or NMP, as described previously 10 ( Figure 1A). After 2 hours, a second biopsy was taken from both kidneys and RNA-Seq performed. When comparing gene expression between the time 0-and 2-hour biopsies, we found that kidneys placed in static cold storage had no statistically significant change in the expression of any individual gene when corrected for multiple testing ( Figure 1B, left panel). the NIHR, the Department of Health, or NHSBT. transplantation / nephrology, organ perfusion and preservation, translational research / science F I G U R E 1 Kidneys exposed to cold storage show limited changes in gene expression compared with those undergoing NMP. A, Pairs of kidneys were obtained which had been declined for use in transplantation. One kidney was maintained in static cold storage and the other underwent normothermic machine perfusion (NMP). Biopsies were taken from the outer cortex at the start and after 2 hours. B, Volcano plot indicating change in gene expression at 2 hours for the indicated group compared to the start. Red dots indicate differentially expressed genes with an adjusted P value <0.05 and the experimental group is indicated above the plot. C, Gene set enrichment analyses of the differential expressions from B against the hallmarks pathways. Only significant pathways (FDR q value <0.05) are plotted. Red dots indicate positive enrichment and blue negative, the size of the dot is inversely correlated with the FDR q value and the position indicates the normalized enrichment score (NES  a key pathway required to generate ATP. 19 In contrast, OXPHOS was among the pathways significantly upregulated during NMP ( Figure 1C, right panel), with potential benefits for cell viability and the restoration of cellular homeostasis. In addition, a number of pathways involved in immune or inflammatory processes were induced during NMP, with "TNFα signaling via NFkB" demonstrating the largest increase. In keeping with this, TNF as well as IL1B and the neutrophil-recruiting chemokines CXCL8 (IL8) and CXCL2 were among the most upregulated genes in 2-hour NMP biopsies ( Figure 1D). String analysis of the top 50 upregulated genes revealed upregulation of biochemically related genes which were clustered into four major nodes; IL8 and neutrophil-recruiting chemokines, Inflammasome-associated genes, NFkB signaling, and transcriptional regulation ( Figure 1E). Of note, DBD and DCD kidneys were transcriptionally similar at baseline ( Figure S2C), and the gene pathways changing during NMP were similar in DBD and DCD kidneys ( Figure S2D). Together, our analysis demonstrates that during NMP, there is an increased expression of genes that promote the generation of energy, with potentially beneficial effects for the organ, but a simultaneous induction of pro-inflammatory genes, which may be deleterious.

F I G U R E 2
Correlation of transcriptome following NMP with the length of delayed graft function. A, Gene expression was correlated with the length of DGF in biopsies taken following NMP as part of a randomized clinical trial. The expression levels of the 1000 genes with the greatest correlation with outcome have been plotted. B, GSEA was performed for the correlation from A against the hallmarks data set. Only significant pathways (FDR q value <0.05) are plotted. Red dots indicate positive enrichment and blue negative, the size of the dot is inversely correlated with the FDR q value and the position indicates the normalized enrichment score (NES)

| Inflammatory pathway genes in NMP kidneys associated with prolonged delayed graft function
In order to link the transcriptional changes occurring during cold storage and NMP to clinical outcomes, we performed RNA-Seq on biopsies taken from 33 DCD kidneys that had undergone NMP as part of a randomized clinical trial currently assessing its efficacy. 17 Samples were available on a subset of kidneys randomized to the NMP arm of the study that were subsequently transplanted (Table S3, Figure S1). DGF is more common in DCD kidneys and is

| Urine output and renal blood flow during NMP demonstrate differing associations with OXPHOS and inflammatory pathway genes
The quantity of urine produced during NMP is one of a number of parameters included in surgical quality assessment scores used to guide organ utilization decisions, 15 but whether high urine output during NMP truly portends a good prognosis for the kidney and the underlying molecular pathways activated in kidneys with a high urine output is unclear. In a total of 10 kidneys undergoing NMP (NMP only kidneys, Table S1 and S2, Figure S1), we observed a range of urine outputs from 0 to 340 ml over the 2-hour period of perfusion ( Figure 3A left panel). Of note, when we compared the pathways changing during NMP in five kidneys in two independent experiments, we found that similar pathways were induced ( Figure   S3A) confirming the reproducibility of our experimental design and the utility of comparing pathways rather than individual genes. In  Figure S4B), in contrast to urine output ( Figure 3B, Figure S3C).
Together these data suggest that renal blood flow and urine output may not be equivalent indicators of a more viable kidney, but that the former may more faithfully identify a kidney less likely to have prolonged DGF.

| Addition of a hemoadsorber to the perfusion circuit has no effect on perfusion parameters but significantly reduces inflammatory gene expression, including NLRP3 and IL1B
Analyses of kidney perfusate have demonstrated a substantial increase in the concentration of pro-inflammatory cytokines and chemokines during the course of hypothermic and NMP. 20,21 These bioactive molecules recirculate into the kidney, with the potential to induce further inflammation. Given that our analysis of transplanted kidney samples indicated that the induction of TNF-dependent genes in NMP kidneys was associated with DGF ( Figure 2B), we hypothesized that removal of cytokines and chemokines from the perfusion circuit may attenuate inflammatory gene induction, with potential beneficial effects for the kidney. Such an approach has shown some efficacy in patients with systemic inflammatory response syndrome, 22,23 and was associated with increased renal blood flow in porcine kidneys undergoing NMP. 18 To test this in human kidneys, we took an additional five kidney pairs and performed NMP for 4 hours with biopsies taken at 0, 2, and 4 hours ( Figure S1). In each case, a cytosorb hemoadsorber (HA) that removes molecules with a molecular weight of 10-50 kDa was added to the perfusion circuit of one kidney within each pair (NMP+HA) ( Figure 4A). As anticipated, the addition of the HA resulted in lower concentrations of a variety of cytokines in the perfusate ( Figure 4B) but had no effect on renal blood flow, urine output or composition, oxygen consumption, and acid-base homeostasis ( Figure 4C, Table   S4, S5). Thus, over 4 hours of NMP, the HA had no impact on the perfusion parameters currently used clinically to generate quality assessment scores but had a substantial effect on gene expression; Following NMP, 1794 and 4026 genes were upregulated at 2 and 4 hours, respectively, including TNF and IL6 ( Figure 4D,E). but only half this number (n = 898 and n = 2606) were increased when the HA was present ( Figure 4D, Figure S5A and B). The number of genes downregulated was also reduced by the addition of HA ( Figure 4D).
After 4 hours of NMP, 46 genes were significantly upregulated and 181 downregulated with the addition of the HA ( Figure 4F). This attenuated transcriptional response included NLRP3 inflammasome activation-associated genes, such as IL1B, NLRP3, and CASP1 ( Figure 4G, Figure S4B) and some neutrophil-recruiting chemokines ( Figure S5C), previously associated with kidney injury in animal models. 24,25 This demonstrates that soluble mediators released from the kidney recirculate and drive de novo expression of inflammatory genes within the organ, but that this can be alleviated by their removal from the perfusion circuit.

| HA associated with a reduction in a delayed graft function-associated gene signature
Gene set enrichment analysis showed a significant decrease in the "TNFα signaling via NFkB" pathway in NMP+HA kidneys compared with NMP alone (Figure 5A-B). Notably, the presence of the HA not only reduced inflammatory gene expression within the kidney but also increased OXPHOS and fatty acid metabolism pathways, both of which contribute to energy generation ( Figure 5A-B). Therefore, the changes in gene expression pathways occurring with the HA would support the conclusion that its effects are clinically beneficial, since it reduced "TNFA signaling via NFkB" pathway genes and increased "OXPHOS" pathway genes, both of which were associated with a shorter duration of DGF ( Figure 2B).
To further explore the relationship between transcriptional changes in NMP and clinical outcomes, we sought to curate a gene signature present in kidneys with DGF. We identified the top 100 positively (UP) and negatively (DOWN) regulated genes (ranked by log fold change) which correlated with the length of DGF in the 33 clinical trial samples ( Figure 5C, Data S1). We found a significant enrichment of the gene signature associated with increased length of DGF in samples with a higher urine output ( Figure 5D), suggesting that high urine output during NMP identifies kidneys at risk of more prolonged DGF. There was no statistically significant correlation between the DGF signature and the genes associated with higher renal blood flow ( Figure 5E). We next assessed whether and how the addition of the HA to the NMP circuit affected the expression of DGF "UP" and "DOWN" gene signatures. Remarkably, this showed that the expression of the gene signature associated with increased length of DGF was significantly reduced by the addition of the HA, and the signature associated with decreased length of DGF was significantly increased by the addition of the HA ( Figure 5F).
Overall, the transcriptional changes we have identified suggest that NMP has potential benefits over cold storage in terms of its effects on energy generation in the kidney. However, during perfusion, some bioactive molecules are released from the kidney In our analysis of kidneys undergoing NMP in the context of a clinical trial, we correlated gene expression signatures with length of DGF rather than DGF incidence (defined as a need for dialysis within the first week posttransplant). Notably, kidneys with a duration of DGF of 1 day or less were highly transcriptionally similar to those with no DGF, likely reflecting dialysis requirement due to peri-operative-associated hyperkalemia rather than bona fide DGF.
We found that a longer DGF was associated with higher expression of TNFA signaling via NFkB pathway genes and lower expression of genes associated with OXPHOS. These data suggest that kidneys with increased OXPHOS and decreased immune signaling may present better potential as donor kidneys, but this conclusion requires validation in a larger prospective study. We also assessed the molecular processes that correlate with urine output and renal blood during NMP. These parameters have previously been used, along with a number of other measures, to generate a quality assessment score of perfused kidneys. High values of urine output and renal blood flow have been considered to reflect a more viable graft. 15 Our data reveal that pathways correlating with high urine output and high renal blood flow differ, and in fact, these parameters demonstrate polar opposite associations with inflammatory pathways. High urine output was associated with higher expression of immune pathway genes while high renal blood flow was negatively correlated with these pathways. We also found that the DGF gene signature we generated was enriched in kidneys with a high urine output during NMP, suggesting that this parameter may identify kidneys at risk of a longer DGF, in contrast to the current dogma. One potential explanation is that a very high urine output reflects kidneys with more tubular damage that lack the capability to concentrate urine. Thus, there may be a "Goldilocks effect" with regard to urine output, where kidneys with low/little urine output are those with substantial abnormalities in the generation of filtrate, those with a very high urine output have substantial tubular damage precluding urine concentration. This hypothesis would need to be tested in a larger number of kidneys that were subsequently transplanted. Importantly, we showed that the addition of the HA affects the expression of genes that are associated with worse outcomes clinically.
Our DGF gene signature was derived from samples undergoing NMP as part of a clinical trial assessing its efficacy, allowing us to robustly link the changes we observed in our paired kidney studies with clinical endpoints in kidneys undergoing transplant. However, its application in the context of a clinical trial will be needed to definitively prove the utility of its application to kidneys pretransplant. The HA is nonspecific and may remove molecules which are helpful, in addition to those which are detrimental to the organ. We found that the net effect of the HA on the kidney transcriptome appeared beneficial, but it may be that a refinement to specifically remove proven, deleterious mediators may be even more effective.
The kidney pairs we used were from the same individual and were therefore genetically identical and had experienced a similar environment throughout the life of the donor. We confirmed that their time 0 transcriptome was extremely similar (Figure S2 A,B).
However, these biopsies sample a small part of the kidney, and kidney pairs could be asymmetrically affected by pathology, for example, cysts, or small vessel disease. An additional caveat is that all kidneys used in the intervention experiments presented here were declined for transplantation and some represent transplanted organs at the lower end of the quality spectrum. Nevertheless, we observed highly reproducible results when comparing gene pathways in groups of five kidneys. We were also able to demonstrate that kidneys from DCD and DBD had a similar response to perfusion.
By combining this with our curated post-NMP DGF signature, we were able to predict potential clinical benefits. This experimental approach could be employed as a preclinical tool to screen future interventions for potential therapeutic efficacy to enable the rational selection of candidate interventions for clinical trials.
In summary, our study provides the first global transcriptional profile of human kidneys undergoing NMP, resolving the differing molecular pathways that are activated in NMP compared with cold storage, and showing that the deleterious effects of bioactive molecules produced or released from the kidney during NMP can be reversed by the addition of a HA. Furthermore, this intervention reduced the expression of genes associated with prolonged DGF providing a strong mechanistic rationale for applying such an intervention to a future clinical trial. Our data also have implications for perfusion strategies beyond the kidney, including in liver and lung transplantation, suggesting that the removal of bioactive molecules from perfusates should be investigated in these contexts where NMP is increasingly used. Finally, our study highlights the utility of global transcriptional profiling in paired kidneys for assessing novel interventions to perfused organs; transcriptional changes precede changes in protein abundance (traditionally used as biomarkers of kidney injury) and tens of thousands of gene transcripts can be readily measured. Thus, RNA measurement has the potential to provide an early, sensitive readout of cellular function of human organs retrieved for transplantation that can be applied to future studies.

ACK N OWLED G M ENTS
The authors thank all organ donors and their families. The Clatworthy Lab are grateful for the core facilities provided by the dirac.ac.uk).

D I SCLOS U R E
The authors have no conflict of interest to disclose as described by

DATA AVA I L A B I L I T Y S TAT E M E N T
The transcriptional data for the experimental kidneys which supports the findings of this study are available at www.ncbi.nlm.nih.
gov/geo under GSE12 1447. Remaining data will be made available upon request to the corresponding author.