Metabolic profiling during ex vivo machine perfusion of the human liver

As donor organ shortages persist, functional machine perfusion is under investigation to improve preservation of the donor liver. The transplantation of donation after circulatory death (DCD) livers is limited by poor outcomes, but its application may be expanded by ex vivo repair and assessment of the organ before transplantation. Here we employed subnormothermic (21 °C) machine perfusion of discarded human livers combined with metabolomics to gain insight into metabolic recovery during machine perfusion. Improvements in energetic cofactors and redox shifts were observed, as well as reversal of ischemia-induced alterations in selected pathways, including lactate metabolism and increased TCA cycle intermediates. We next evaluated whether DCD livers with steatotic and severe ischemic injury could be discriminated from ‘transplantable’ DCD livers. Metabolomic profiling was able to cluster livers with similar metabolic patterns based on the degree of injury. Moreover, perfusion parameters combined with differences in metabolic factors suggest variable mechanisms that result in poor energy recovery in injured livers. We conclude that machine perfusion combined with metabolomics has significant potential as a clinical instrument for the assessment of preserved livers.


SUPPLEMENTARY METHODS
Principal Component Analysis (PCA). The metabolomics data for the initial and final biopsies were projected onto two principal components to enable visualization of the overall metabolic shift between pre and post perfusion for the 9 livers. The t=0 h time point and the t=3 h time point biopsies were treated as independent, resulting in 18 observations. An 18X159 matrix N was then constructed with rows as observations and columns for measured metabolites, where every entry N(i,j) was computed as the median peak height for the three triplicate samples. Principal components and 95% confidence band ellipses were computed using Matlab (Mathworks, Natick, MA), using z-scores of each matrix entry relative to other entries in the same column, which prevented metabolites with significantly higher average peak heights from dominating relative contribution towards the principal components.
Heatmap. The metabolomics data are were organized as two 159X9 matrices for each time point, where the rows were denoted by measured metabolites and the columns by each liver, and each entry is the calculated median of the peak intensities from three technical replicates of the same biopsy. Since the peak intensity values range several orders of magnitude, the values in each matrix were replaced by their z-score with respect to the nine entries in each row. In both the pre and post-perfusion data matrices, the minimum calculated z-score was -2.05 and the maximum z-score was 2.65. In this regard, a higher z-score represents a relatively higher abundance of that metabolite in the specific liver compared to other livers at the same time point.
The z-scores were then presented as two heatmaps with metabolite rows and livers in columns with the first heatmap comparing the control livers to warm ischemic livers ( Fig   5A) and the second comparing control livers to the steatotic group ( Fig 5B) at pre-perfusion. The color gradient used to color the entries ranges from green to red, corresponding to relatively low and high abundance of the metabolite respectively. The order of the metabolite rows is determined by the slope of the least-squares line fit through a scatter of z-score versus WIT or z-score versus degree of steatosis (which was just assigned a binary for 0 for no steatosis and 1 for steatosis) in increasing order. For each row, we also compute the correlation coefficient of that scatter and a P-value of whether or not that correlation is significant using the Matlab function corrcoef, which reports the probability of obtaining a correlation as large as reported by random chance when the true correlation is zero. Metabolites for which the correlation P-value was ≤ 0.05 were deemed as significantly correlated to either WIT or steatosis, metabolites with for which the correlation P-value ≤0.10 are presented in Table S2.
Targeted metabolomics -cofactor analysis. Crushed tissue biopsies (averaging ~25 mg) were also analyzed for metabolic cofactors using a targeted MRM (multiple reaction monitoring) analysis on a 3200 QTRAP LC/MS-MS (Triple quadrupole liquid chromatography -mass spectrometry) system (AB Sciex, Foster City, CA). The metabolites were first extracted using 250 µL of a 2/1 (v/v) mixture of methanol/chloroform and were subject to three freeze-thaw cycles, which entailed 30 seconds of rapid freezing in liquid nitrogen, thawing at room temperature, and a 10 second vortex. Ice cold water (200 µL) was then added to each extract and after a 1-minute centrifugation at 15,000 g, the upper phase of the resulting biphasic mixture was transferred to an HPLC autosampler vial for LC/MS analysis. The chromatographic separation conditions and the analytespecific MS parameter optimization routines were the same as that reported in Quinn et al., for which NAD + , NADH, and FAD were quantified(1). In this study, MRM transitions for ATP, ADP, AMP, NADPH, NADP, GSH, and GSSG were also quantified in addition to NAD + , NADH, and FAD. The precursor/product ion transitions for all the compounds are presented in Table S3. The peak area of each MRM transition was correlated to extract concentration based on serial dilutions of pure chemical standards. Pertinent redox ratios were then computed based on the relative concentrations of cofactors calculated to be in the tissue extract.
Untargeted metabolomics analysis. An untargeted profiling of primary metabolites was performed at the West Coast Metabolomics Center (Davis, CA) using GC-TOF-MS (gas chromatography -time of flight -mass spectrometry). Briefly, the homogenized biopsies were subject to solvent-based extraction using a mixture of acetonitrile, isopropanol and water as described in Fiehn et al.
(2). This was followed by a clean-up procedure using acetonitrile/water to remove triglycerides and lipids, after which internal standards were then spiked to the extracted sample. The details of the mass spectrometry data acquisition, including the chromatography conditions and MS operating parameters, have been outlined elsewhere(2). The raw data were provided for each identified metabolite as peak heights normalized to the total ion chromatogram (sum of all peaks) to account for variation in tissue mass and total quantity of extracted metabolites. Peak heights were deemed more precise and reproducible than peak areas for quantification, particularly for low-abundance metabolites. The final raw data were reported as a 159X54 matrix,