All animals were euthanised by overdose of anaesthetic according to schedule 1 of the United Kingdom Animals (Scientific Procedures) Act 1986. Use of animals and collection of kidneys for these studies was approved after a full ethical review by Newcastle Universities Animal Welfare and Ethical Review Board and ongoing review via study plan approval (AWERB number 854, study plan 38). Porcine kidneys were retrieved from 60 kg 16-week-old white landrace pigs. These were sedated using intramuscular injection of approximately 5 mL Tiletamine and Zolazepam (Zoletil™, Virbac). Pigs were then euthanised using ear vein injection of 25 mL (Euthatal™, Dopharma Research B.V.). Approximately 500 mL of blood was collected using standard clinical blood bags depleted of CDPA-1 and filled with 10,000 IU heparin and 50 mL saline (Fresenius Kabi). Kidneys were retrieved and kept in the body cavity until 25 min after confirmation of death, then flushed with 1 L of 4°C University of Wisconsin solution with 25,000 IU sodium heparin (Panpharma). This standardised the warm ischaemic time (WIT) to 25 min. Kidneys were kept on ice for 16 h of cold ischaemic time (CIT) prior to initiation of machine perfusion. All porcine kidneys reported here came from different pigs.

We chose a WIT of 25 min followed by CIT of 16 h following a previous series of optimisation experiments which explored a range of ischaemic times (data not published). These ischaemic times provide sufficient ischaemia to mimic the injury seen during NMP of extended criteria human kidneys. This is the same protocol used in our previous studies of therapeutic delivery during porcine NMP [10].

Human kidneys retrieved for transplant but then deemed unsuitable were included. Ethical approval for accepting these kidneys was granted by the national research ethics commission in the United Kingdom, National Research Ethics System (15/SC/0161). We gained approvals for this project from the National Health Services Blood & Transplant’s (NHSBT) Research Innovation and Novel Technologies Advisory Group (RINTAG), who oversee the allocation of such research organs to authorized research groups. In all cases donor families provided generic consent to approved research projects.

NMP was performed using a customised Medtronic pediatric cardiopulmonary bypass system. The renal artery was cannulated, and oxygenated perfusate with red blood cells was perfused at 37°C (continuous flow), aiming for a mean arterial pressure of 75 mmHg. Porcine and human perfusions differed in terms of their duration (6 h and 24 h respectively), source of blood (autologous whole blood versus packed red cells) and perfusate constituents. The increased duration in human versus porcine experiments reflects the growing international research interest in prolonged kidney perfusion and the recent establishment of these extended NMP protocols in our own perfusion laboratories [11]. Full protocols and lists of perfusate constituents for the two protocols is given in Supplementary Tables S1, S2. Total renal blood flow was measured using the Medtronic flow sensor (TX50P flow transducer, Medtronic). Urine production rate was measured via a paediatric nasogastric tube tied into the ureter. Oxygen consumption was calculated from blood gas and flow data, as previously described [12].

Ultrasound was performed with an eL18-4 probe of the Philips EPIQ7 Ultrasound machine with QLab 8.1 software (Philips). The probe (with sterile cover) was placed directly onto the kidney and held stationary to capture a longitudinal view of the kidney.

For CEUS imaging, 1 mL sulphur hexafluoride contrast agent (SonoVue ^®, Bracco) was reconstituted with 4 mL of perfusate in a syringe. The contrast agent was then administered via a three way tap to the arterial limb of the circuit as a rapid bolus. A detailed standard operating procedure for capturing these cine loop recordings is available in Supplementary Table S3. We analysed CEUS scans from 6 h of perfusion (end of perfusion) for porcine kidneys and 2 h of perfusion for human kidneys. Each scan represents a single recording following a single bolus injection of contrast (recording was continued for a minimum of 60 s after contrast was seen reaching the kidney).

The built-in Philips “Contrast” mode was used; this is optimised for capturing microbubble contrast whilst minimising signal from any human tissue. When using the settings detailed in our standard operating procedure (Supplementary Table S3) this resulted in zero or negligible background signal, eliminating the need for normalisation of peak intensity to baseline value.

Raw CEUS DICOM files were imported into QLAB Advanced Quantification Software (Release 15.5 Philips) and analysed using the ROI QApp to get mean intensity for various regions of interest. A 5 × 5 mm square was used for the cortex. A freeform polygon was used to draw further regions of interest around the medulla. This was split equally into outer medulla (closest to the kidney surface) or inner medulla (closest to kidney hilum). Figure 1 displays drawing the regions of interest.

All cineloops were cut so that time zero was the frame that contrast was first seen (in segmental/interlobar arteries). All clips were cut to 30 s total length (selected as this is significantly longer than the time required for all regions to reach peak intensity). The ROI QApp then calculated the mean contrast intensity in the various regions of interest, for each frame of the ultrasound loop. This raw data (mean pixel intensity in decibels on every timestamped frame) for each region of interest was then exported for downstream analysis.

Analysis of CEUS data was performed in R (R Foundation for Statistical Computing, Vienna, Austria) [13]. A curve was plotted to the raw data, using a LOESS smoother. This was performed using the loess() base R function with loess.span set to 0.06 [13]. Data was extracted from this curve to calculate the peak intensity of contrast and time-to-peak intensity.

A TUNEL (TdT-mediated dUTP Nick-End Labelling) assay was performed on 4 µm FFPE sections. The DeadEnd™ Fluorometric TUNEL System (Promega) was carried out according to manufacturer’s instructions. Slides were mounted using VECTASHIELD mounting medium with DAPI (Vector labs). Cells with DNA fragmentation (as a hallmark of apoptosis) and DAPI-stained nuclei were counted using Fiji ImageJ software. Persons performing TUNEL assay and image analysis were blinded to CEUS data.

Neutrophil gelatinase-associated lipocalin (NGAL) concentration in urine was analysed by ELISA (DuoSet Cat: DY1757 for human, Abcam Cat: ab207924 for porcine). This was multiplied by urine production rate to get total nanograms of urinary NGAL per minute.

Martius Scarlet Blue (MSB) staining was used to visualise erythrocytes, red cells, fibrin and collagen. Following dewaxing and rehydration, tissue was stained using a Martius Scarlet Blue Stain Kit (Atom Scientific) according to manufacturer’s instructions. LABKIT, a Fiji ImageJ plugin for segmentation of microscopy images was used to create a pixel classifier [14]. This enabled automatic segmentation of fibrin rich red cell aggregates (representative images in Supplementary Figure S1), which could be used to calculate the percentage of each image containing such aggregates.

To generate a single score for each region normalised values were required; z-scores for peak intensity and time-to-peak were therefore calculated [15]. The z-score for time-to-peak (TTP) could then be subtracted from the peak intensity (PI) z-score, such that the score for each region was calculated as follows: R e g i o n   C E U S   s c o r e = Sample   PI − cohort   mean   PI cohort   standard   deviation   PI − Sample   TTP − cohort   mean   TTP cohort   standard   deviation   TTP

A table of the cohort average and standard deviation for peak and time-to-peak for each region, which were used to calculate these normalised z-scores, is given in Supplementary Table S4. As these are all normalised and on the z-score scale, scores for the three regions were added to generate an overall score for each kidney [15].

The correlation between CEUS metrics and NMP outcomes was assessed using the Pearson correlation coefficient. All statistical analyses were performed in R [13].

As shown in Figures 2A, B, perfusion of the cortex and medulla was correlated with urinary NGAL, an important predictor of kidney quality during NMP [16, 17]. Increasing time-to-peak (indicating worse microvascular perfusion) was associated with higher levels of damage marker NGAL (r = 0.90, P = 0.002 and r = 0.745, P = 0.034 for cortex and outer medulla respectively).

Similarly, improved quality of perfusion in these two regions was associated with a lower proportion of cells with DNA fragmentation (a hallmark of apoptosis) on histology (Figures 2C, D). Representative TUNEL images from kidneys with relatively poor versus good microvascular perfusion are shown in Supplementary Figure S2. There was no significant correlation between any CEUS metric and 6-hour urine flow rate.

In the setting of AKI, shunting of blood through the kidney without parenchymal perfusion has been described as a key pathophysiological factor [4, 5]. We therefore correlated total blood flow through the kidney with parenchymal perfusion at an identical timepoint.

As shown in Figure 3A there was no significant correlation between cortex perfusion and total renal blood flow. However, there was a strong negative correlation between the quality of microvascular perfusion of the medulla and total renal blood flow (Figures 3B, C), indicating that more severe injury leads to shunting through low resistance vessels, with increased total flow but decreased parenchymal perfusion.

One potential contributing factor to the lack of microvascular perfusion is the presence of capillary obstruction by red cell aggregates, which have been reported during NMP previously [18, 19]. We found no correlation between burden of red blood cell microvascular occlusion and microvascular perfusion of the cortex or medulla at 6 h (Supplementary Figures S1, S3). Potentially indicating the role of shunting as opposed to occlusion.

Human kidneys from five deceased donors were included. Donor demographics are provided in Table 1. Three kidneys were rejected due to extra-renal malignancy, one due to presence of glomerulosclerosis on biopsy and CIT, and one due to a significantly calcified aortic patch. None of the donors received normothermic regional perfusion.

The association of CEUS metrics with both tissue DNA fragmentation and urinary NGAL was assessed. Mirroring the porcine results, improved microvascular perfusion was correlated with lower levels of tissue DNA fragmentation, and lower levels of the damage marker NGAL (Supplementary Figures S4, S5 respectively). We also assessed associations between 2-hour CEUS score and oxygen consumption; those with signs of improved cortex and medullary perfusion showed increased oxygen consumption by the kidney (Supplementary Figure S6). There was no significant correlation between CEUS metrics and urine flow rate or renal blood flow at the time of the scan, or at 24 h.

To generate a single CEUS score for each region, which could be combined to give an overall CEUS score for each kidney, z-score normalised peak and time-to-peak values were calculated. A table of the cohort means and standard deviations which were used to calculate these normalised z-scores, is given in Supplementary Table S4.

As shown in Figures 4A–C these region scores have a negative correlation with tissue DNA fragmentation; kidneys with more DNA fragmentation at the start of perfusion displayed worse perfusion of both cortex and medulla at 2 h. The overall score combining the three regions of interest displayed the strongest correlation with the percentage cells with DNA fragmentation on histology (Figure 4D; Pearson r = - 0.937, P = 0.019).

CEUS scores at 2 h were also correlated against 24-h urinary NGAL. Improved CEUS scores at 2 h of NMP were corelated with lower urinary NGAL at the end of perfusion (24 h); Figure 5. This was statistically significant for both outer and inner medulla regions (Pearson r = −0.902, P = 0.036; Pearson r = −0.968, P = 0.007 respectively; Figures 5B, C), and for the overall CEUS score (Pearson r = −0.925, P = 0.024; Figure 5D). Neither CEUS region scores, nor CEUS combined score, showed a significant association with renal blood flow at the time of the scan (Supplementary Figure S7), or the 1-hour “quality assessment score” [2].

Following the analyses described above, we wanted to make these techniques accessible to other groups. We have used the shinyapp framework [20] to create a freely available web application [21]. This allows any time-intensity data to be uploaded, and calculates peak and time-to-peak, with additional options for calculating area under the curve and time-to-peak proportion (Figure 6A). A report is also generated to visually confirm the results (Figure 6B). We have also created an R package “tican,” which is freely available to import from CRAN [22], for those who wish to perform these time-intensity curve analyses using R code.