The Future Role of Machine Learning in Clinical Transplantation.

The use of artificial intelligence and machine learning (ML) has revolutionized our daily lives and will soon be instrumental in healthcare delivery. The rise of ML is due to multiple factors: increasing access to massive datasets, exponential increases in processing power, and key algorithmic developments that allow ML models to tackle increasingly challenging questions. Progressively more transplantation research is exploring the potential utility of ML models throughout the patient journey, although this has not yet widely transitioned into the clinical domain. In this review, we explore common approaches used in ML in solid organ clinical transplantation and consider opportunities for ML to help clinicians and patients. We discuss ways in which ML can aid leverage of large complex datasets, generate cutting-edge prediction models, perform clinical image analysis, discover novel markers in molecular data, and fuse datasets to generate novel insights in modern transplantation practice. We focus on key areas in transplantation in which ML is driving progress, explore the future potential roles of ML, and discuss the challenges and limitations of these powerful tools.

Identifying cell-enriched miRNAs in kidney injury and repair.

Small noncoding RNAs, miRNAs (miRNAs), are emerging as important modulators in the pathogenesis of kidney disease, with potential as biomarkers of kidney disease onset, progression, or therapeutic efficacy. Bulk tissue small RNA-sequencing (sRNA-Seq) and microarrays are widely used to identify dysregulated miRNA expression but are limited by the lack of precision regarding the cellular origin of the miRNA. In this study, we performed cell-specific sRNA-Seq on tubular cells, endothelial cells, PDGFR-ß+ cells, and macrophages isolated from injured and repairing kidneys in the murine reversible unilateral ureteric obstruction model. We devised an unbiased bioinformatics pipeline to define the miRNA enrichment within these cell populations, constructing a miRNA catalog of injury and repair. Our analysis revealed that a significant proportion of cell-specific miRNAs in healthy animals were no longer specific following injury. We then applied this knowledge of the relative cell specificity of miRNAs to deconvolute bulk miRNA expression profiles in the renal cortex in murine models and human kidney disease. Finally, we used our data-driven approach to rationally select macrophage-enriched miR-16-5p and miR-18a-5p and demonstrate that they are promising urinary biomarkers of acute kidney injury in renal transplant recipients.

Defining Priorities for Future Research: Results of the UK Kidney Transplant Priority Setting Partnership.

BACKGROUND: It has been suggested that the research priorities of those funding and performing research in transplantation may differ from those of end service users such as patients, carers and healthcare professionals involved in day-to-day care. The Kidney Transplant Priority Setting Partnership (PSP) was established with the aim of involving all stakeholders in prioritising future research in the field. METHODS: The PSP methodology is as outlined by the James Lind Alliance. An initial survey collected unanswered research questions from patients, carers and clinicians. Duplicate and out-of-scope topics were excluded and the existing literature searched to identify topics answered by current evidence. An interim prioritisation survey asked patients and professionals to score the importance of the remaining questions to create a ranked long-list. These were considered at a final consensus workshop using a modified nominal group technique to agree a final top ten. RESULTS: The initial survey identified 497 questions from 183 respondents, covering all aspects of transplantation from assessment through to long-term follow-up. These were grouped into 90 unanswered "indicative" questions. The interim prioritisation survey received 256 responses (34.8% patients/carers, 10.9% donors and 54.3% professionals), resulting in a ranked list of 25 questions that were considered during the final workshop. Participants agreed a top ten priorities for future research that included optimisation of immunosuppression (improved monitoring, choice of regimen, personalisation), prevention of sensitisation and transplanting the sensitised patient, management of antibody-mediated rejection, long-term risks to live donors, methods of organ preservation, induction of tolerance and bioengineering of organs. There was evidence that patient and carer involvement had a significant impact on shaping the final priorities. CONCLUSIONS: The final list of priorities relates to all stages of the transplant process, including access to transplantation, living donation, organ preservation, post-transplant care and management of the failing transplant. This list of priorities will provide an invaluable resource for researchers and funders to direct future activity.

Logistical Factors Influencing Cold Ischemia Times in Deceased Donor Kidney Transplants.

BACKGROUND: Prolonged cold ischemia time (CIT) is associated with a significant risk of short- and long-term graft failure in deceased donor kidney transplants across the world. The aim of this prospective longitudinal study was to determine the importance of logistical factors on CIT. METHOD: Data on 1763 transplants were collected prospectively over 14 months from personnel in 16 transplant centers, 19 histocompatibility and immunogenetics laboratories, transport providers, and National Health Service Blood and Transplant. RESULTS: The overall mean CIT was 13.8 hours, with significant center variation (P < 0.0001). Factors that significantly reduced CIT were donation after circulatory death (P = 0.03), shorter transport time (P = 0.0002), use of virtual crossmatch (XM) (P < 0.0001), and use of donor blood for pretransplant XM (P < 0.0001). The CIT for transplants that went ahead with a virtual XM was 3 hours shorter than those requiring a pretransplant XM (P < 0.0001). There was a mean delay of 3 hours in starting transplants despite organ, recipient, and pretransplant XM result being ready, suggesting that theater access contributes significantly to increased CIT. DISCUSSION: This study identifies logistical factors relating to donor, transport, crossmatching, recipient, and theater that impact significantly on CIT in deceased donor renal transplantation, some of which are modifiable; attention should be focussed on addressing all of these.

Hemin Preconditioning Upregulates Heme Oxygenase-1 in Deceased Donor Renal Transplant Recipients: A Randomized, Controlled, Phase IIB Trial.

BACKGROUND: The enzyme heme oxygenase-1 (HO-1) degrades heme and protects against ischemia-reperfusion injury. Monocytes/macrophages are the major source of HO-1 and higher levels improve renal transplant outcomes. Heme arginate (HA) safely induces HO-1 in humans. METHODS: The Heme Oxygenase-1 in renal Transplantation study was a randomized, placebo-controlled, IIb trial to evaluate HA effect on HO-1 upregulation after deceased donor kidney transplantation. 40 recipients were randomized to either 3 mg kg HA or placebo (0.9% NaCl), given preoperatively (day 0) and again on day 2. Recipient blood and urine were collected daily. Graft biopsies were taken preoperatively and on day 5. Primary outcome was HO-1 upregulation in peripheral blood mononuclear cells (PBMCs). Secondary outcomes were graft HO-1 upregulation and injury, urinary biomarkers, and renal function. RESULTS: The HA upregulated PBMC HO-1 protein more than placebo at 24 hours: HA 11.1 ng/mL versus placebo 0.14 ng/mL (P = < 0.0001). The PBMC HO-1 messenger RNA also increased: HA 2.73-fold versus placebo 1.41-fold (P = 0.02). Heme arginate increased day 5 tissue HO-1 protein immunopositivity compared with placebo: HA 0.21 versus placebo -0.03 (P = 0.02) and % HO-1-positive renal macrophage also increased: HA 50.8 cells per high power field versus placebo 22.3 (P = 0.012). Urinary biomarkers were reduced after HA but not significantly. Histological injury and renal function were similar but the study was not powered for this. Adverse events were equivalent between groups. CONCLUSIONS: The primary outcome was achieved and demonstrated for the first time that HA safely induces HO-1 in transplant recipients. Planned larger studies will determine the impact of HO-1 upregulation on clinical outcomes and evaluate the benefit to patients at risk of ischemia-reperfusion injury.

Heat shock protein 90 inhibition abrogates TLR4-mediated NF-κB activity and reduces renal ischemia-reperfusion injury.

Renal ischemia-reperfusion injury (IRI) is a common cause of acute kidney injury. Toll-like receptor 4 (TLR4) mediates sterile inflammation following renal IRI. Heat shock protein 90 (Hsp90) inhibition is a potential strategy to reduce IRI, and AT13387 is a novel Hsp90 inhibitor with low toxicity. This study assessed if pre-treatment with AT13387 could reduce renal IRI and established if the mechanism of protection involved a reduction in inflammatory signalling. Mice were pre-treated with AT13387 prior to renal IRI. 24 h later, renal function was determined by serum creatinine, kidney damage by tubular necrosis score, renal TLR4 expression by PCR and inflammation by cytokine array. In vitro, human embryonic kidney cells were co-transfected to express TLR4 and a secreted alkaline phosphatase NF-κB reporter. Cells were pre-treated with AT13387 and exposed to endotoxin-free hyaluronan to stimulate sterile TLR4-specific NF-κB inflammatory activation. Following renal IRI, AT13387 significantly reduced serum creatinine, tubular necrosis, TLR4 expression and NF-κB-dependent chemokines. In vitro, AT13387-treatment resulted in breakdown of IκB kinase, which abolished TLR4-mediated NF-κB activation by hyaluronan. AT13387 is a new agent with translational potential that reduces renal IRI. The mechanism of protection may involve breakdown of IκB kinase and repression of TLR4-mediated NF-κB inflammatory activity.

Mouse kidney transplantation: models of allograft rejection.

Rejection of the transplanted kidney in humans is still a major cause of morbidity and mortality. The mouse model of renal transplantation closely replicates both the technical and pathological processes that occur in human renal transplantation. Although mouse models of allogeneic rejection in organs other than the kidney exist, and are more technically feasible, there is evidence that different organs elicit disparate rejection modes and dynamics, for instance the time course of rejection in cardiac and renal allograft differs significantly in certain strain combinations. This model is an attractive tool for many reasons despite its technical challenges. As inbred mouse strain haplotypes are well characterized it is possible to choose donor and recipient combinations to model acute allograft rejection by transplanting across MHC class I and II loci. Conversely by transplanting between strains with similar haplotypes a chronic process can be elicited were the allograft kidney develops interstitial fibrosis and tubular atrophy. We have modified the surgical technique to reduce operating time and improve ease of surgery, however a learning curve still needs to be overcome in order to faithfully replicate the model. This study will provide key points in the surgical procedure and aid the process of establishing this technique.

A comparative study of 2 computer-assisted methods of quantifying brightfield microscopy images.

Immunohistochemistry continues to be a powerful tool for the detection of antigens. There are several commercially available software packages that allow image analysis; however, these can be complex, require relatively high level of computer skills, and can be expensive. We compared 2 commonly available software packages, Adobe Photoshop CS6 and ImageJ, in their ability to quantify percentage positive area after picrosirius red (PSR) staining and 3,3'-diaminobenzidine (DAB) staining. On analysis of DAB-stained B cells in the mouse spleen, with a biotinylated primary rat anti-mouse-B220 antibody, there was no significant difference on converting images from brightfield microscopy to binary images to measure black and white pixels using ImageJ compared with measuring a range of brown pixels with Photoshop (Student t test, P=0.243, correlation r=0.985). When analyzing mouse kidney allografts stained with PSR, Photoshop achieved a greater interquartile range while maintaining a lower 10th percentile value compared with analysis with ImageJ. A lower 10% percentile reflects that Photoshop analysis is better at analyzing tissues with low levels of positive pixels; particularly relevant for control tissues or negative controls, whereas after ImageJ analysis the same images would result in spuriously high levels of positivity. Furthermore comparing the 2 methods by Bland-Altman plot revealed that these 2 methodologies did not agree when measuring images with a higher percentage of positive staining and correlation was poor (r=0.804). We conclude that for computer-assisted analysis of images of DAB-stained tissue there is no difference between using Photoshop or ImageJ. However, for analysis of color images where differentiation into a binary pattern is not easy, such as with PSR, Photoshop is superior at identifying higher levels of positivity while maintaining differentiation of low levels of positive staining.

Macrophage/monocyte depletion by clodronate, but not diphtheria toxin, improves renal ischemia/reperfusion injury in mice.

The role of resident renal mononuclear phagocytes in acute kidney injury is controversial with experimental data suggesting both deleterious and protective functions. To help resolve this, we used mice transgenic for the human diphtheria toxin receptor under the control of the CD11b promoter and treated them with diphtheria toxin, or liposomal clodronate, or both to deplete monocyte/mononuclear phagocytes prior to renal ischemia/reperfusion injury. Although either system effectively depleted circulating monocytes and resident mononuclear phagocytes, depletion was most marked in diphtheria toxin-treated mice. Despite this, diphtheria toxin treatment did not protect from renal ischemia. In contrast, mice treated with clodronate exhibited reduced renal failure and acute tubular necrosis, suggesting key differences between these depletion strategies. Clodronate did not deplete CD206-positive renal macrophages and, unlike diphtheria toxin, left resident CD11c-positive cells unscathed while inducing dramatic apoptosis in hepatic and splenic mononuclear phagocyte populations. Abolition of the protected phenotype by administration of diphtheria toxin to clodronate-treated mice suggested that the protective effect of clodronate resulted from the presence of a cytoprotective intrarenal population of mononuclear phagocytes sensitive to diphtheria toxin-mediated ablation.

Tubular atrophy and interstitial fibrosis after renal transplantation is dependent on galectin-3.

BACKGROUND: Chronic allograft injury (CAI), characterized by interstitial fibrosis and tubular atrophy, leads to a progressive decline in graft function, resulting in the loss of 5% of renal transplants per annum, and eludes specific therapies. Galectin-3 (gal-3) is a ß-galactoside-binding lectin expressed in diverse fibrotic tissue, and mice deficient in gal-3 have reduced fibrosis in kidney, liver, and lung models. The role of gal-3 in CAI is examined in this study. METHODS: We adopted a murine model of CAI, characterized by a single class II mismatch between BM12 donor and C57BL/6 recipient strains. Syngeneic transplants served as controls (C57BL/6). Transplants were then performed between BM12 donors and gal-3 null recipients on a C57BL/6 background. RESULTS: Transplantation of BM12 kidneys into C57BL6 mice was associated with interstitial fibrosis (P<0.0001), tubular atrophy (P<0.0001), and upregulation in gal-3 expression (P=0.002), compared with syngeneic controls. Transplanting BM12 kidneys into gal-3 null mice resulted in significant preservation of tubules (P=0.008) and reduced interstitial fibrosis (P=0.01), with decreased myofibroblast activation (P=0.01) and collagen I expression (P=0.04), compared with wild type controls. The number of infiltrating leukocytes was unaltered by abrogation of gal-3, but reduced expression of YM1 (P=0.0001), a marker of alternative macrophage activation, along with a reduction in the number of circulating CD4-positive T cells (P=0.01), and reduced expression of interleukin-4 (P=0.02) in gal-3 null mice suggest possible mechanisms by which gal-3 may promote renal transplant fibrosis. CONCLUSION: Our results suggest a potential role for gal-3 in CAI, and this represents a potentially exciting therapeutic target.

The induction of macrophage hemeoxygenase-1 is protective during acute kidney injury in aging mice.

Aging is thought to be associated with a higher susceptibility to renal ischemia-reperfusion injury (IRI). To study whether defective induction of hemeoxygenase-1 (HO-1, a protective and anti-inflammatory enzyme) might contribute to this, we found that while 12-month-old mice had similar baseline renal function and HO-1 expression, the induction of HO-1 usually seen in ischemia-reperfusion was reduced. This was also associated with worsened renal function and acute tubular necrosis in the aged compared with young mice. In the older mice, heme arginate (HA) induced HO-1 in the cortex and medulla, significantly improved renal function, and reduced tissue injury. Cellular HO-1 induction in the medulla in response to injury or HA treatment was found to be interstitial rather than epithelial, as evidenced by its colocalization with macrophage markers. In vitro, HA treatment of primary macrophages resulted in marked HO-1 induction without impairment of classical activation pathways. Macrophage depletion, caused by diphtheria toxin treatment of 12-month-old CD11b-DTR transgenic animals, resulted in the loss of interstitial HO-1-positive cells and reversal of the protective phenotype of HA treatment. Thus, failure of HO-1 induction following renal IRI worsens structural and functional injury in older mice and represents a therapeutic target in the elderly. Hence, HO-1-positive renal macrophages mediate HA-induced protection in IRI.

Restorative and rejection-associated lymphangiogenesis after renal transplantation: friend or foe?

The review focuses on lymphangiogenesis as a possible contributor to interstitial fibrosis leading to chronic renal transplant injury, which culminates in the loss of 5% transplants annually. The process of lymphatic reconnection after renal transplantation and the mechanisms and mediators of lymphangiogenesis are explored in the context of new specific lymphatic markers. In addition, potentially exciting research avenues are examined, with the specific aim of determining whether new lymphatic formation is beneficial or detrimental to the transplanted kidney.

Peritubular capillary rarefaction and lymphangiogenesis in chronic allograft failure.

BACKGROUND: Chronic renal allograft failure is a common and multifactorial but incompletely understood process with no effective treatment strategy. METHODS: We used immunohistochemistry to evaluate changes in density and turnover (proliferation) of the microvasculature and lymphatic vessels in endstage human transplant nephrectomies and control tissue derived from macroscopically normal areas of native nephrectomy specimens removed for renal carcinoma. We also examined the expression of angiogenic and lymphangiogenic growth factors in the associated inflammatory infiltrate. RESULTS: Endstage allografts showed reduced microvascular density in cortex and medulla compared with controls (P<0.0001), despite the presence of endothelial cell proliferation. However, the grafts also showed new lymphatic vessels in the tubulointerstitium, not evident in controls, and which appeared to be functional with luminal macrophages. Double labeling studies showed a subpopulation of the graft-infiltrating macrophages to be immunopositive for inducible nitric oxide synthase or vascular endothelial growth factor-C (a lymphatic-specific growth factor). B cells also strongly expressed the inflammatory and angiogenic cytokine vascular endothelial growth factor A. CONCLUSIONS: The present results identify contrasting changes in the microanatomy of vascular and lymphatic beds in endstage renal allografts associated with subpopulations of infiltrating macrophages and B cells that potentially regulate some of these changes. These cells and processes could become a new therapeutic target in chronic allograft failure.