From Preservation to Repair: A Systematic Review of Therapeutic Organ Rehabilitation During Normothermic Ex Vivo Machine Perfusion Authors: Syed Sikandar Raza, Caitlin M. Bocks, Spencer G. Amacher, Kristopher B. Deatrick, Yuliya Tipograf, Alvaro Rojas-Peña Published: 2026 ABSTRACT Background.Normothermic ex vivo machine perfusion (NMP) enables active organ rehabilitation rather than passive preservation. While lung perfusion represents the most mature therapeutic paradigm, kidney, liver, and heart applications are emerging. This review examines therapeutic interventions during NMP across these organs to identify shared mechanisms and translational gaps.Methods.Following PROSPERO registration (CRD420251134637), a PRISMA-compliant search of PubMed, Web of Science, and Embase identified 3336 records. After screening, 155 articles underwent full-text review for inclusion of therapeutic delivery during NMP. Data extraction captured intervention characteristics, perfusion parameters, and outcomes across predefined domains.Results.Twenty-one studies met inclusion criteria: kidney (n = 12, 57%), liver (n = 8, 38%), and heart (n = 1, 5%). Models were predominantly human (n = 13, 62%) or porcine (n = 6, 29%). Interventions included pharmacologic agents (n = 12, 57%), cellular/subcellular therapies (n = 6, 29%), defatting protocols (n = 2, 10%), and circuit modifications (n = 1, 5%). Three mechanistic themes emerged: (1) mitochondrial resuscitation via agents such as AP39 preserved ATP and reduced oxidative stress; (2) immune modulation through CD47 blockade or cytokine hemoadsorption attenuated inflammation; and (3) regenerative approaches using mesenchymal stromal cells to repair structural injuries. Molecular improvements were reported in 95% of studies, but functional improvement in only 38%, revealing an outcome translation gap. Only 19% of studies included posttransplant validation.Conclusions.NMP has enabled effective metabolic and inflammatory reprogramming in kidney and liver perfusion, while cardiac applications lag behind. Standardized reporting, organ-specific functional endpoints, and transplant-linked validation are needed to convert currently rejected organs into viable grafts and expand the effective donor pool. Organ transplantation remains the definitive treatment for end-stage organ disease, yet donor scarcity persists. A largely untapped resource lies in organs deemed unsuitable for transplant. In the United States, discard rates vary substantially: approximately 30% of procured kidneys, 10% of livers, 10% of lungs, and 2% of hearts are discarded annually, with rates exceeding 80% for some extended criteria subgroups, largely because of ischemic injury, donor age, infection, and agonal deterioration.1-3 Normothermic machine ex vivo perfusion (NMP) provides an opportunity for objective functional assessment, targeted repair, or immunologic conditioning of these suboptimal grafts before implantation.4,5 Alternative perfusion strategies, including hypothermic oxygenated perfusion (HOPE) and subnormothermic machine perfusion, have demonstrated efficacy in organ preservation and assessment.6,7 Moreover, NMP supports near-physiologic metabolism and real-time monitoring at normothermia, enabling organ-directed interventions and controlled therapeutic modulation not feasible at lower temperatures. NMP use can be conceptualized across three frameworks: (1) preservation and assessment, where NMP maintains organ viability and enables quality evaluation during organ preservation before transplant; (2) rehabilitation, where the NMP environment itself facilitates recovery from ischemic injury; and (3) therapeutic delivery, where NMP serves as a platform for administering pharmacologic, biologic, or genetic interventions directly to the graft. This review focuses on category 3, though the boundaries are fluid. The therapeutic NMP approach offers 2 key advantages: ex vivo delivery of immune, pharmacologic, and genetic therapies without systemic recipient exposure, and the ability to test intervention effects on injury pathways and function in a controlled environment before committing to transplantation.3,4 Ex vivo lung perfusion has become the prototypical clinical rehabilitation platform, with a large evidence base led by the pioneering work of Keshavjee et al.8,9 In contrast, therapeutic NMP for kidney, liver, and heart remains comparatively heterogeneous, as it is distributed across small preclinical experiments, discarded-organ studies, and limited transplant-linked investigations, often using nonstandardized endpoints. Existing reviews frequently focus on perfusion feasibility, viability testing, or organ-specific preservation outcomes, rather than systematically cataloging interventions delivered during NMP and whether they translate from molecular signals to functional improvement and postperfusion usability.10,11 We conducted a systematic review of therapeutic interventions administered during normothermic ex vivo perfusion of kidney, liver, and heart grafts. Our primary objectives were (1) to classify interventions by mechanistic and operational category, (2) to characterize study models and perfusion protocols, and (3) to evaluate how frequently reported benefits in molecular or metabolic domains translated to functional endpoints and postperfusion outcomes relevant to graft utilization. By identifying convergent strategies (molecular pathways, microvascular obstruction, small-molecule pharmacotherapy to cell- and gene-based therapies) and persistent translational gaps, this review aims to inform the design of future interventional perfusion studies and accelerate the development of a unified framework for the rehabilitation of marginal organs. MATERIALS AND METHODS Study Design and Regulatory Compliance This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) statement.12 The study protocol was prospectively designed to ensure methodological transparency and was registered with the International Prospective Register of Systematic Reviews (PROSPERO: CRD420251134637). Search Architecture and Information Sources A multistage systematic search was performed across 3 major medical databases: PubMed/MEDLINE, Embase (via Elsevier), and Web of Science, spanning from the inception of each database through January 1, 2026. To maximize the retrieval of relevant primary research while filtering out secondary literature, we utilized a combination of Medical Subject Headings (MeSH terms), Emtree terms, and specific Boolean logic targeting the Title and Abstract fields. The finalized search string was optimized for the specific intersection of ex vivo technology and therapeutic bioengineering: (“normothermic perfusion”[Title/Abstract] OR “normothermic machine perfusion”[Title/Abstract] OR “normothermic ex vivo perfusion”[Title/Abstract]) AND (“kidney”[Title/Abstract] OR “renal”[Title/Abstract] OR “liver”[Title/Abstract] OR “hepatic”[Title/Abstract] OR “heart”[Title/Abstract] OR “cardiac”[Title/Abstract]) AND (“intervention”[Title/Abstract] OR “therapeutic”[Title/Abstract] OR “treatment”[Title/Abstract] OR “therapy”[Title/Abstract] OR “stem cell”[Title/Abstract] OR “gene therapy”[Title/Abstract] OR “drug delivery”[Title/Abstract] OR “defatting”[Title/Abstract] OR “regenerative”[Title/Abstract]) AND (Journal Article[pt]) NOT (Review[pt] OR Meta-Analysis[pt] OR “Systematic Review”[pt] OR “Conference Abstract”[pt]). To minimize the risk of missing relevant studies (“grey literature”), we also performed a snowballing manual search by reviewing the bibliographies of all included studies and recent major reviews in the field of machine perfusion. Screening Workflow and Study Selection All identified citations were imported into Zotero (v6.0; Corporation for Digital Scholarship, Vienna, VA) for the initial removal of duplicates. The remaining unique records were uploaded to the Rayyan systematic review platform for screening. Selection followed a strict 2-stage process: Title and Abstract Screening: 2 independent reviewers (S.A. and C.M.) performed a blinded assessment of all records. Discrepancies were flagged by the software, and a third senior reviewer was consulted to resolve conflicts. Title and Abstract Screening: 2 independent reviewers (S.A. and C.M.) performed a blinded assessment of all records. Discrepancies were flagged by the software, and a third senior reviewer was consulted to resolve conflicts. Two reviewers independently screened each record at both stages against prespecified eligibility criteria. Discrepancies were resolved by consensus; if consensus could not be reached, a third reviewer adjudicated. Reasons for full-text exclusion were recorded in Covidence and used to populate the PRISMA flow diagram. Two reviewers independently screened each record at both stages against prespecified eligibility criteria. Discrepancies were resolved by consensus; if consensus could not be reached, a third reviewer adjudicated. Reasons for full-text exclusion were recorded in Covidence and used to populate the PRISMA flow diagram. Full-Text Review: Articles passing the initial screen underwent a comprehensive full-text evaluation. The results of this selection process are summarized in the PRISMA Flow Diagram (Figure 1). Full-Text Review: Articles passing the initial screen underwent a comprehensive full-text evaluation. The results of this selection process are summarized in the PRISMA Flow Diagram (Figure 1). PRISMA 2020 flow diagram. Inclusion and Exclusion Criteria We used the PICO (Population, Intervention, Comparator, Outcome) framework to define eligibility. We included original studies evaluating therapeutic interventions administered during NMP of kidney, liver, or heart grafts. Eligible studies included human organs (eg, discarded or declined for transplantation) and animal organs (preclinical models). To be eligible, studies were required to (1) include a normothermic perfusion phase (reported as normothermic by authors or conducted at near-physiologic temperature), (2) administer an active intervention during perfusion intended to rehabilitate or modify the organ (eg, pharmacologic agents, cellular/subcellular therapies, genetic/vector-based therapies, metabolic/defatting protocols, or circuit/device modifications), and (3) report at least 1 outcome related to organ injury, metabolism, function, or postperfusion usability. We excluded studies that (1) focused on lungs or ex vivo lung perfusion; (2) evaluated perfusion solely as a preservation or assessment strategy without a therapeutic intervention; (3) did not include a normothermic perfusion phase; (4) were in vitro experiments not involving intact organs; or (5) were secondary literature (reviews, systematic reviews, and meta-analyses), editorials, letters, protocols, or conference abstracts without full study data. Studies not available in English were excluded (Figure 1). Data Extraction Two reviewers independently extracted data using a standardized extraction template developed a priori and piloted on a subset of included studies. Extracted variables included study characteristics (first author, year, journal, design), organ and model (organ type, species/human discarded organ versus animal model, donor context where applicable), perfusion protocol (temperature, duration, perfusate composition, circuit parameters when available), intervention details (class, agent or modification, dose/concentration, timing, and delivery route), comparator (if present), and outcomes. Where data were presented only in figures, values were extracted from graphs when feasible; otherwise, outcomes were recorded qualitatively as reported by the authors. When key intervention parameters or outcomes were unclear, attempts were made to clarify from the article text and supplements; contacting authors was not routinely performed. Outcomes and Outcome-domain Classification Given heterogeneity across organs and models, outcomes were synthesized across predefined outcome domains rather than a single-pooled endpoint. Outcomes were classified into the following domains: (1) molecular/inflammatory and endothelial injury markers; (2) metabolic markers (eg, lactate handling, acid–base, oxygen consumption where reported); (3) functional measures during perfusion (eg, bile production/composition, urine output/clearance surrogates, and hemodynamics/contractility parameters); (4) mitochondrial/ATP or bioenergetic metrics; and (5) postperfusion outcomes (eg, reperfusion experiments or transplantation endpoints when performed). For each study and domain, outcomes were coded as significant improvement, improvement without statistical significance, no difference, worsened, or not assessed, based on the authors’ reported statistics. Intervention Classification Interventions were categorized a priori into operational/mechanistic classes: (1) pharmacologic agents; (2) cellular and subcellular therapies (eg, cells and extracellular vesicles); (3) genetic/vector-based therapies; (4) metabolic conditioning/defatting protocols; and (5) circuit/device modifications (eg, filtration, adsorption, oxygenation, or circuit components intended to alter the perfusate environment). Risk of Bias Assessment Risk of bias was assessed independently by 2 reviewers using a simplified framework adapted from established tools. Each study was evaluated across 3 methodological domains: (1) randomization of samples or experimental groups, (2) blinding of outcome assessment, and (3) justification of sample size. Each domain was rated as “Yes,” “No,” or “Not Reported.” An overall risk of bias rating was assigned as low (at least 2 criteria met), moderate (at least 1 criterion met), or high (no criteria met or insufficient reporting). Given the heterogeneity of included studies (human and animal models across multiple organs), a single validated tool was not uniformly applicable. Data Synthesis Because included studies varied substantially in organ type, model (human versus animal), perfusion protocols, intervention classes, and outcome definitions, we planned a priori a narrative synthesis supported by descriptive statistics. We summarized study characteristics and intervention classes by counts and proportions using Microsoft Excel (Microsoft Corporation, Redmond, WA) and mapped outcome-domain reporting and direction of effect using an evidence heatmap. All analyses and figures were generated using Prism 10.0 (GraphPad Software, Boston, MA). RESULTS Study Selection The literature search identified 3336 records across PubMed, Embase, and Web of Science. After removal of 521 duplicates, 2815 titles and abstracts were screened. At the abstract stage, 2660 records were excluded for the following reasons: absence of a therapeutic intervention during perfusion (n = 2300), lung studies (n = 40), in vivo–only therapies (n = 32), and editorials, reviews, or conference abstracts without full text (n = 288). Of the remaining 155 full-text articles assessed for eligibility, 136 were excluded: studies without a clear therapeutic intervention (n = 92), in vivo–only therapies (n = 12), editorials, reviews, or conference abstracts (n = 31), and full text not available (n = 1). Twenty-one studies met inclusion criteria and were included in the qualitative synthesis (Figure 1). Study Characteristics Among the 21 included studies, the kidney was the most frequently studied organ (n = 12, 57.1%),13-24 followed by the liver (n = 8, 38.1%)25-32 and the heart (n = 1, 4.8%).33 Human donor organs declined for transplantation were used in 13 studies (61.9%), porcine models in 6 (28.6%), and canine and rat models in 1 each (4.8%) (Figure 2). The majority of studies used only normothermic machine perfusion (n = 17, 81.0%), while 3 studies (14.3%) utilized a sequential hypothermia-to-normothermia protocol (Table 1). Perfusion durations ranged from 90 min to 12 d, though most fell between 4 and 12 h. Controlled comparisons (intervention versus standard NMP without the agent) were present in 15 studies; only 4 (19.0%) incorporated postperfusion validation with transplantation (Table 2). TABLE 1.Summary of 21 studies including authors, year, organ, species, donor model, intervention, agent, perfusion duration, outcomes, and risk of biasHuman studies (n = 13)AuthorsOrganSpeciesRecovery modelIntervention categoryAgentNMP duration (h)Comments/outcomesRisk of biasArcolino et al13KidneyHumanDCDCellular & Subcellular TherapynKSPC6 h NMPFirst observation of SIX2 re-expression adult human kidneys; upregulation of regenerative markersHighThompson et al23KidneyHumanDCDPharmacological AgentAntagomir oligonucleotide to miR-24-3p6 h NMPNovel microRNA-targeting approach; reduced tubular injury markers and increased target gene expressionHighDiRito et al16KidneyHumanDCD, DBDPharmacological AgenttPA + plasminogen90-min NMPSignificant elimination of microvascular obstruction; more stable vascular resistance and lowered vascular injuryLowFerdinand et al17KidneyHumanDCD, DBDCircuit ComponentCytosorb hemoadsorber (cytokine/chemokine removal)4 h NMPAttenuation of gene signatures associated with DGF; novel device-based approachModerateThompson et al22KidneyHumanDCD, DBDCellular & Subcellular TherapyMAPC7 h NMPReduced injury and inflammatory biomarkers; improved microvascular perfusionLowSousa Da Silva et al32LiverHumanDCD, DBDDefatting ProtocolDefatting protocol (automated insulin/glucagon, L-carnitine, fenofibrate)Brief HOPE + 4–12 d NMPLongest NMP duration reported; protocol optimized for automated metabolic control approachHighRaigani et al30LiverHumanDCDPharmacological AgentEmricasan (pan-caspase inhibitor)12 h NMPSignificant mitigation of innate immune and pro-inflammatory responses and lower levels of apoptotic cell deathModerateHaque et al27LiverHumanDCDPharmacological AgenttPA + plasminogen12 h NMPReduced peribiliary vascular plexus and mural stroma injury without significant bleedingModerateSampaziotis et al31LiverHumanDBD, DCDCellular & Subcellular TherapyCholangiocyte organoidsUp to 100 h NMPProof-of-concept for biliary regeneration using extrahepatic organoidsHighLaing et al29LiverHumanDCD, DBDCellular & Subcellular TherapyMAPC6 h NMPCells distributed throughout liver; proteomic changes detectedHighBoteon et al25LiverHumanDCD, DBDDefatting ProtocolDefatting cocktail (forskolin, GW7647, hypericin, scoparone, visfatin, GW501516, L-carnitine)12 h NMPSignificant reduction in steatosis; enhanced lipid metabolism which aided in functional recovery and reduced injuryLowWatson et al28LiverHumanDCD, DBDPharmacological AgentAlteplase (tPA) + FFPNMP (variable)Two-center RCT; fibrinolytic therapy safe with no increase in postimplant bleeding; optimal protocol: 10 mg alteplase + 2 units FFP in prime for enhanced D-dimer releaseLowMacMillan et al20KidneyHumanDCD, DBDPharmacological AgentFpGalNAc deacetylase + FpGalactosaminidase (blood group A→O conversion enzymes)6 h NMP; 6–24 h HMP~80% blood group A antigen removal in 2 h; ABOe kidneys did not bind anti-A antibodies or activate complement; no treatment-related injuryModerateKawamura et al18KidneyPorcineDCDPharmacological AgentAP39 (mitochondria-targeted H2S donor)5 h NMP14-d transplant follow-up showed improved eGFR and reduced fibrosisModeratevan Leeuwen et al24KidneyPorcineDCDPharmacological AgentTGF-β1 and/or galunisertib (TGF-β receptor inhibitor)24 h HMP + 6 h NMPAnti-fibrotic targeting; short-term transplant data availableLowLohmann et al19KidneyPorcineDCDCellular & Subcellular TherapyAdipose-derived mesenchymal stromal cells (MSC)4 + 14 h oxHMP14-d autotransplant: treatment safe but no functional benefit demonstratedLowPool et al21KidneyPorcineDCDCellular & Subcellular TherapyHuman adipose or bone marrow-derived MSCs2–3 h HMP + 7 h NMPXenogeneic MSCs; compared adipose vs bone marrow sourcesLowBleilevens et al14KidneyPorcineDCDPharmacological AgentVitamin C (ascorbic acid)6 h NMPSimple, low-cost antioxidant intervention; reduced oxidative stressModerateBrasile et al15KidneyCanineDCDPharmacological AgentCobalt protoporphyrin (HO-1 inducer)6 h NMPEarliest study in dataset; demonstrated HO-1 cytoprotection during warm perfusionHighGarcia-Aroz et al26LiverPorcineDCDPharmacological AgentAnti-CD47 monoclonal antibody6 h NMPEvidence of reduced ischemia reperfusion injury via downregulation of CD47 signaling pathway and decreased apoptosisModerateMourouzis et al33HeartRatIsolated OrganPharmacological AgentTriiodothyronine (T3)4 h NMPMicrocirculation protection evident via improved pressures; switch from proapoptotic to prosurvival gene signalingHighABOe, ABO-enzymatically converted; DBD, donation after brain death; DCD, donation after circulatory death; DGF, delayed graft function; FFP, fresh frozen plasma; HMP, hypothermic machine perfusion; HOPE, hypothermic oxygenated machine perfusion; IC, ischemic cholangiopathy; MAPC, multipotent adult progenitor cells; MSC, mesenchymal stromal cells; N, sample size; nKSPC, neonatal kidney stem/progenitor cells; NMP, normothermic machine perfusion; NR, not reported; oxHMP, oxygenated hypothermic machine perfusion; RCT, randomized controlled trial; tPA, tissue plasminogen activator; Tx, transplanted. Summary of 21 studies including authors, year, organ, species, donor model, intervention, agent, perfusion duration, outcomes, and risk of bias ABOe, ABO-enzymatically converted; DBD, donation after brain death; DCD, donation after circulatory death; DGF, delayed graft function; FFP, fresh frozen plasma; HMP, hypothermic machine perfusion; HOPE, hypothermic oxygenated machine perfusion; IC, ischemic cholangiopathy; MAPC, multipotent adult progenitor cells; MSC, mesenchymal stromal cells; N, sample size; nKSPC, neonatal kidney stem/progenitor cells; NMP, normothermic machine perfusion; NR, not reported; oxHMP, oxygenated hypothermic machine perfusion; RCT, randomized controlled trial; tPA, tissue plasminogen activator; Tx, transplanted. TABLE 2.Postperfusion transplant validation studies (n = 4) with transplant model, follow-up, key findingsAuthorsOrganSpeciesInterventionTransplant modelFollow-up/outcomes assessedKey transplant findingsCommentsKawamura et al18KidneyPorcineAP39 (H2S donor)Orthotopic autotransplant3 d follow-up period; outcomes assessed include plasma creatinine and urea nitrogen levels, tubular injury scores, inflammation scores, and FeNaImproved renal function, reduced oxidative stress and fibrosisStrongest preclinical kidney evidence with functional endpointsLohmann et al19KidneyPorcineAdipose-derived MSCsOrthotopic autotransplant14 d follow-up period; outcomes assessed include plasma creatinine levels, glomerular filtration rate, neutrophil gelatinase-associated lipocalin concentrations, kidney damage assessment via histology, any acute kidney injury, and fibrosis-related gene expression at mRNA levelTreatment safe and feasible; no significant functional benefitLargest MSC study; negative result important for fieldFerdinand et al17KidneyHumanHemoadsorber (CytoSorb)AllotransplantThe posttransplant follow-up period was up to the first week posttransplant, focusing on the duration of DGF; outcomes assessed included the duration of DGF and gene expression associated with DGFReduced delayed graft function markersOnly device-based intervention with clinical correlationWatson et al28LiverHumanAlteplase (tPA) + FFPAllotransplant1-y follow-up; outcomes include postimplant blood loss, transfusion requirements, D-dimer release, EAD, MEAF score, posttransplant cholangiopathy, graft and patient survivalNo increase in postimplant bleeding or transfusion; 6 cholangiopathy cases (4 attributed to arterial complications); 12-mo graft survival 83%–100% across groupsLargest fibrinolytic NMP treatment study; pilot safety RCT informing future efficacy trialDGF, delayed graft function; EAD, early allograft dysfunction; FeNa, fractional excretion of sodium; FFP, fresh frozen plasma; MEAF, model for early allograft function; MSCs, mesenchymal stromal cells; RCT, randomized controlled trial; Tx, transplant/transplanted. Postperfusion transplant validation studies (n = 4) with transplant model, follow-up, key findings DGF, delayed graft function; EAD, early allograft dysfunction; FeNa, fractional excretion of sodium; FFP, fresh frozen plasma; MEAF, model for early allograft function; MSCs, mesenchymal stromal cells; RCT, randomized controlled trial; Tx, transplant/transplanted. Overall study composition including organ type, species/model, and perfusion protocol. Intervention Classes Interventions were classified into 4 operational categories (Table 3). Pharmacologic agents constituted the largest class (n = 12, 57.1%), spanning kidney (n = 7), liver (n = 4), and heart (n = 1), and included anti-inflammatory agents, antioxidants, thrombolytics, mitochondria-targeted compounds, and hormonal therapies. Cellular and subcellular therapies comprised the second largest class (n = 6, 28.6%), including mesenchymal stromal cells, multipotent adult progenitor cells, neonatal kidney stem/progenitor cells, and cholangiocyte organoids, and were studied in kidney (n = 4) and liver (n = 2). Defatting protocols were exclusive to the liver (n = 2, 9.5%), employing multiagent lipid metabolism cocktails. A single circuit modification study (n = 1, 4.8%) evaluated cytokine hemoadsorption in kidneys. No studies evaluating genetic or vector-based therapies met inclusion criteria, despite this being a prespecified intervention class. TABLE 3.Intervention category distribution with organ representationIntervention categoryN%Organ studiedPharmacological Agent1257.1Kidney (8), liver (4), heart (1)Cellular & Subcellular Therapy628.6Kidney (4), liver (2)Defatting Protocol29.5Liver (2)Circuit Component14.8Kidney (1) Intervention category distribution with organ representation Outcome Reporting and Evidence Mapping Across Domains Outcome reporting was heterogeneous across studies; therefore, results were synthesized across predefined outcome domains and visualized using an evidence heatmap (Figure 3). Molecular outcomes (inflammatory cytokines, endothelial activation markers, apoptosis, and oxidative stress) were assessed in all 21 studies (100%) and significantly improved in 20 (95.2%). Histological outcomes (tissue injury scores, fibrosis, steatosis grading) were assessed in 14/21 studies (66.7%), with significant improvement reported in 10 (47.6%). Functional outcomes (organ-specific perfusion measures: urine output and creatinine clearance for kidney, bile production and composition for liver, hemodynamics and contractility for heart) were assessed in 19/21 (90.5%), but significant improvement was demonstrated in only 8 (38.1%). Metabolic outcomes (lactate clearance, acid–base status, and oxygen consumption) were assessed in 13/21 (61.9%), with significant improvement in 5 (23.8%). Mitochondrial or bioenergetic endpoints (ATP content, mitochondrial respiration, and membrane potential) were assessed in 6/21 (28.6%), with significant improvement in 3 (14.3%). Only 4 studies (19.0%) included postperfusion validation with reperfusion or transplantation, with significant improvement reported in 3 (14.3%). Heatmap of outcome domains across all 21 studies, grouped by intervention class and organ type, with color coding for significant positive outcome, assessed but not significant, and not assessed. Organ-specific Patterns Kidney studies (n = 12) represented the largest body of evidence with the broadest range of interventions, including pharmacologic agents (n = 7), cellular/subcellular therapies (n = 4), and circuit modification (n = 1). Molecular improvement was nearly universal across kidney studies, but functional outcomes (urine output, creatinine clearance) were significantly improved in a minority. Three kidney studies included posttransplant validation (Table 2): Kawamura et al18 demonstrated improved renal function and reduced fibrosis at 14 d following AP39 treatment, Lohmann et al19 found adipose-derived MSC delivery to be safe but without significant functional benefit at 14 d, and Ferdinand et al17 showed attenuation of delayed graft function gene signatures using cytokine hemoadsorption with short-term clinical correlation. Notably, MacMillan et al20 demonstrated enzymatic conversion of blood group A kidneys to universal group O during NMP using bacterial glycosidases, achieving approximately 80% antigen clearance within 2 h without adverse effects on perfusion parameters. Liver studies (n = 8) encompassed pharmacologic agents (n = 4), cellular/subcellular therapies (n = 2), and defatting protocols (n = 2). Defatting strategies were exclusive to liver and showed some of the most consistent metabolic improvements, including significant reduction in steatosis and enhanced lipid metabolism. Watson et al28 contributed the largest study in this review (n = 80 livers), a 2-center randomized dose-escalation trial of alteplase with fresh frozen plasma during NMP, demonstrating the safety of fibrinolytic therapy with no increase in postimplant bleeding and providing the first clinical dose-finding data for thrombolysis during liver NMP. Heart was represented by a single study,33 which evaluated triiodothyronine (T3) in a rat Langendorff model and demonstrated microcirculatory protection and a shift from proapoptotic to prosurvival signaling. Risk of Bias Risk of bias was variable across the 21 included studies. Using our simplified 3-domain framework (randomization, blinding, and sample size justification), 8 studies (38.1%) were rated as low risk of bias, 6 (28.6%) as moderate, and 7 (33.3%) as high. Among high-risk studies, the most common limitation was the absence of reported randomization and blinding procedures. Human studies were predominantly nonrandomized by design, with inherent susceptibility to confounding from donor heterogeneity. Only a minority of studies reported sample size justification through power analysis. These methodological limitations, combined with the heterogeneity of interventions and outcome definitions, support a narrative synthesis rather than pooled quantitative estimates. DISCUSSION Normothermic machine perfusion (NMP), independent of any additional therapeutic agent, serves as both an assessment and rehabilitation tool. Important work by Hosgood et al34 demonstrated that declined DCD kidneys subjected to NMP-based quality assessment could be successfully transplanted, and Tingle et al35 showed that NMP can serve as a pharmacological testing platform. The therapeutic interventions cataloged in this review are therefore best understood as additions layered upon the inherent rehabilitative capacity of NMP itself. In this systematic review of therapeutic interventions delivered during NMP in kidney, liver, and heart grafts, 3 consistent themes emerge. First, the intervention landscape is broad and mechanistically convergent across organs, with most strategies targeting canonical pathways of ischemia–reperfusion injury (IRI), including inflammatory activation, endothelial dysfunction, microcirculatory impairment, oxidative stress, and metabolic derangements (Table 3). Second, benefits are reported most consistently at the molecular level, whereas improvement in higher-order physiology and postperfusion endpoints is less frequent (Figure 3). Third, validation through transplantation remains uncommon, limiting the ability to conclude that molecular improvements translate into meaningful clinical benefit and reinforcing the need for standardized functional endpoints and postperfusion confirmation (Table 2). Of the 21 studies, 13 used human organs (all declined/discarded) and 8 used animal models (6 porcine, 1 canine, and 1 rat) (Figure 1). Human studies were predominantly proof-of-concept, testing feasibility and molecular effects on organs already rejected for transplant; only Ferdinand et al and Watson et al proceeded to transplantation. In contrast, the preclinical studies (Kawamura et al,18 Lohmann et al,19 and van Leeuwen et al24) were the ones able to complete the transplant loop with autotransplant follow-up, providing the only functional endpoint data postimplantation (Table 2). This divergence is important: human studies provide the most clinically relevant tissue but are limited by donor heterogeneity, lack of transplant endpoints, and nonrandomized design. NMP as a Therapeutic and Research Platform All 21 studies leveraged the normothermic perfusion phase to deliver an active intervention, including pharmacologic agents, cellular products, defatting cocktails, or circuit modifications, directly into the graft vasculature. This organ-confined exposure was exploited in several distinct ways: tPA-based thrombolysis to clear microvascular obstruction (DiRito, Haque), with Watson et al advancing this to a clinical dose-finding trial in 80 donor livers,16,27,28 a mitochondria-targeted H2S donor (AP39) that would be difficult to dose systemically,18 and an extended perfusion window to engraft cholangiocyte organoids onto the biliary epithelium.30 These examples illustrate that NMP is not merely supportive, but, in fact, enables intervention strategies that are impractical or unsafe in the recipient, including high-dose local pharmacologic modulation and cell-based therapies designed to modify the graft before implantation. Evidence Gap: Frequent Molecular Benefit, Inconsistent Functional Translation Across the evidence base, molecular outcomes were reported in nearly all studies and were most likely to demonstrate improvement. However, functional and metabolic improvements were less consistent, and transplant-linked endpoints were infrequently assessed. Bleilevens et al11 showed a 58% reduction in oxidative stress (P = 0.002) but no change in creatinine clearance (P = 0.9); Pool et al21 demonstrated lower LDH and NGAL but no functional improvement. This pattern likely reflects outcome selection favoring molecular readouts, limited statistical power, heterogeneous perfusion protocols, and short evaluation windows insufficient for physiological recovery. A lack of detectable functional improvement may reflect subtherapeutic dosing or inadequate endpoint sensitivity rather than treatment failure, which emphasizes that postperfusion endpoints and recipient outcomes remain the gold standard. Without consistent linkage to graft utilization, the field risks overinterpreting signaling-level improvements. Intervention Classes and Mechanistic Convergence Therapeutic interventions clustered into 4 operational classes that shared mechanistic targets across organs. Pharmacologic agents constituted the largest class (12/21, 57.1%) and targeted inflammation (anti-CD47,26 emricasan30), microvascular obstruction (tPA/plasminogen15,26), oxidative stress (vitamin C,15 AP3919), fibrotic signaling (galunisertib24), and hormonal reconditioning (T333; Tables 1 and 3). Dose–exposure relationships and tissue pharmacokinetics were inconsistently reported, complicating the interpretation of negative results. Notably, MacMillan et al,20 who used bacterial glycosidases to enzymatically convert kidney blood group A antigens to group O during NMP, expanded the pharmacologic category beyond traditional IRI-targeted agents to strategies that modify organ immunogenicity, with implications for overcoming ABO compatibility barriers. Cellular and subcellular therapies (6/21, 28.6%) included MSCs,19,21 MAPCs,22,29 neonatal kidney progenitor cells,13 and cholangiocyte organoids,31 which demonstrated molecular benefits but limited functional improvement. The defatting studies25,32 are notable for demonstrating that hepatic steatosis is a modifiable target during NMP, with Boteon et al25 showing functional recovery alongside lipid clearance. Finally, Ferdinand et al17 employed the only circuit-based modification (CytoSorb hemoadsorption), representing a distinct paradigm shift by altering the perfusate environment rather than delivering a discrete agent. Organ-specific Maturity A notable finding is the imbalance in organ representation, with the kidney and liver comprising the overwhelming majority of included studies. This likely reflects both logistical and biological barriers: cardiac NMP research is technically demanding, standardization of heart-specific functional endpoints (eg, contractility and ventricular loading conditions) is more complex, and translation to postperfusion transplantation is more complicated to implement. The paucity of heart studies within the literature underscores the need for standardized cardiac endpoints during perfusion and for reproducible functional assessments that can serve as bridge endpoints to transplantation outcomes. Standardization Priorities and Implications for Translation The heterogeneity in perfusion parameters and outcome reporting identified in this review (Figure 3) argues for the adoption of a minimum reporting set for therapeutic NMP studies. At minimum, future studies should report donor and injury context (donor type, ischemia times, and injury severity), perfusion protocol details, and intervention specification (agent identity, dose, timing, route, and verification of tissue delivery when feasible). Equally important is the adoption of a core outcome hierarchy linking mechanistic endpoints to functional and postperfusion validation. Molecular and inflammatory markers are valuable for screening candidate therapies, but functional endpoints during perfusion (organ-specific, normalized, and reported at defined timepoints) are necessary for physiological relevance. Finally, postperfusion validation through reperfusion experiments or transplant endpoints should become standard rather than exceptional. Limitations This review has several limitations. First, included studies were heterogeneous in organ type, species, perfusion protocols, intervention characteristics, and outcome definitions, precluding meta-analysis and limiting direct cross-study comparisons. Second, many studies had small sample sizes and incomplete reporting of methodological safeguards; only 8 of 21 were rated low risk of bias, which limits causal inference and may overestimate treatment effects. Third, our risk-of-bias assessment used a simplified framework adapted from established tools rather than a single validated instrument, a pragmatic decision given the heterogeneity of human and animal study designs but 1 that may reduce comparability with other systematic reviews. Finally, by excluding lung studies, we focused on less mature organ systems where the therapeutic NMP literature is most fragmented; while this strengthens the novelty of the review, it limits comparison to the most established ex vivo rehabilitation paradigm. Future Directions and Clinical Roadmap The findings of this review identify several priorities for advancing therapeutic NMP toward clinical translation. First, thrombolytic and fibrinolytic strategies in liver NMP have demonstrated the most mature evidence trajectory, progressing from preclinical proof-of-concept through clinical dose-finding with posttransplant safety data, and may serve as a template for how other therapeutic NMP interventions should be developed. Second, the persistent molecular-to-functional translation gap identified in this review (95% versus 38%) underscores that future studies must be designed with predefined functional and transplant-linked primary endpoints rather than relying on molecular surrogates alone. Standardized core outcome sets, analogous to those developed for clinical transplant trials, should be adopted to enable cross-study comparison and eventual meta-analysis. Third, immunomodulatory approaches are expanding beyond traditional IRI targets to include strategies that modify organ immunogenicity itself, an approach that could fundamentally alter organ allocation by overcoming compatibility barriers. Fourth, emerging gene therapy platforms, including the development of organ-tropic AAV variants via ex vivo machine perfusion capsid selection, suggest that durable genetic modification of donor organs during the perfusion window may become feasible. Fifth, cardiac therapeutic NMP remains a critical evidence gap; the development of standardized working-mode functional assessments and reproducible heart-specific endpoints should be prioritized to enable therapeutic intervention studies in this organ.36 Finally, the therapeutic strategies cataloged in this review lay the groundwork for a future in which marginal organs of varying type and injury profile are directed to centralized rehabilitation facilities, where organ-specific or multimodal interventions are applied during NMP to convert currently unusable grafts into transplantable organs at scale. CONCLUSION This systematic review provides the first cross-organ review of therapeutic interventions delivered during normothermic ex vivo machine perfusion of kidney, liver, and heart grafts. It demonstrates that therapeutic NMP reliably modulates molecular injury pathways across kidney, liver, and heart grafts, yet a persistent translation gap, 95% molecular improvement versus 38% functional improvement and 19% transplant validation, indicates that the field remains in an early proof-of-concept phase for most intervention classes. Closing this gap will require studies designed with predefined functional and transplant-linked primary endpoints, standardized perfusion reporting, and a unified outcome-domain hierarchy that enables cross-study comparison. As the therapeutic toolkit matures—from pharmacologic reconditioning and immunomodulation to cell-based repair and gene therapy—NMP can provide the foundation for systematic rehabilitation of currently declined organs, expanding the transplantable supply across all solid organs. Beyond transplantation, NMP also offers a controlled, near-physiologic human tissue platform to study basic biology and physiology and to test or develop novel therapies in situ before clinical use. ACKNOWLEDGMENTS We wish to acknowledge Dr. Robert H. Bartlett for founding the University of Michigan Extracorporeal Life Support Laboratory and the Michigan ECMO program, for his pioneering contributions to the development of extracorporeal membrane oxygenation, and for his enduring support of translational research in transplantation and ECMO-assisted organ donation.