ParticipantsThe study was approved by the Swedish Ethical Review Authority (2020-03841, 2021-05087, 2022-06272-02) and conducted in accordance with ethical guidelines of the 2013 Declaration of Helsinki. Informed written consent was obtained from all participants. There was no compensation for participants. Individuals undergoing hepatectomies due to hepatocellular carcinoma or liver metastasis at the Department of Surgery, Linköping University Hospital were recruited consecutively to the study, with age < 18 years as the sole exclusion criteria. Participants were fasted overnight (no solid food) for at least 8 h but received a preoperative liquid carbohydrate loading of 100.8 g in the evening and 50.4 g in the morning (3 h before surgery). Overall, 45% of all donors of material to the study were female. We did not attempt to analyse sex differences due to the small sample size.Quantification of glucose, lactate, triglycerides and insulinIn total, 5 ml of blood was collected in sodium heparin tubes (BD Vacutainer Sodium-Heparin Plasma Tubes, BD Diagnostics). Blood samples were centrifuged at 2,000g for 10 min at room temperature. The plasma fraction was aliquoted and stored in −80 °C until analysis. Quantification of glucose, lactate and triglycerides in the plasma samples and cell culture medium, as well as insulin in cell culture medium, was performed by Clinical Chemistry, Laboratory Medicine at Linköping University Hospital, according to their routine analysis.Synthesis of culture mediumWilliams E medium was synthesized in-house from individual components, according to the formulation given in Supplementary Table 2. For the 13C deep labelling medium, all AAs and glucose were replaced with the following 13C-labelled counterparts (Cambridge Isotope Laboratories): 13C3-alanine (CLM-2184-H), 13C6-glucose (CLM-1396-0), 13C6-arginine (CLM-2265-H), 13C4-asparagine (CLM-8699-H), 13C4-aspartic acid (CLM-1801-H), 13C3-cysteine (CLM-4320-H), 13C5-glutamic acid (CLM-1800-H), 13C5-glutamine (CLM-1822-H), 13C2-glycine (CLM-1017-0), 13C6-histidine:HCl:H2O (CLM-2264-0), 13C6-isoleucine (CLM-2248-H), 13C6-leucine (CLM-2262-H), 13C6-lysine:2HCl (CLM-2247-H), 13C5-methionine (CLM-893-H), 13C9-phenylalanine (CLM-2250-H), 13C5-proline (CLM-2260-H), 13C3-serine (CLM-1574-H), 13C4-threonine (CLM-2261-0), 13C11-tryptophan (CLM-4290-H), 13C9-tyrosine (CLM-2263-H) and 13C5-valine (CLM-2249-H). For the various single isotope tracing experiments, Williams E media were synthesized with 13C6-leucine (CLM-2262-H), 13C6-arginine (CLM-2265-H) or 3-13C-lactate (CLM-1578-PK). Three liver slices were used for all experiments. Unlabelled (12C) Williams E medium (unlabelled nutrients from Merck) was used as control. For cultures containing serum, frozen human serum was obtained from blood banks at Uppsala University Hospital and Karolinska University Hospital, Stockholm, thawed and dialysed in SnakeSkin 10,000 molecular-weight cut-off dialysis tubing (Thermo Fisher Scientific, 88245) to remove small molecules. Williams E medium was then prepared as above but with all concentrations doubled, and mixed at a 50:50 (vol/vol) ratio with dialysed serum to obtain final concentrations of small metabolites equal to those in serum-free medium. Concentrations in dialysed serum were measured by isotope dilution mass spectrometry (Supplementary Table 4) as described below, based on the known concentrations of the 13C-labelled nutrients in the synthesized Williams E medium. All media were supplemented with 1% penicillin–streptomycin (PeSt; Gibco,15140148).Preparation and culture conditions of liver slicesHuman liver tissue and a venous blood sample were obtained from individuals undergoing hepatectomies due to hepatocellular carcinoma or liver metastasis at Linköping University Hospital. Tissue samples (~1 cm3) were taken from healthy liver tissue included in the tumour resection, and immediately placed in ice-cold preservation solution (Storeprotect Plus, Carnamedica) and processed within 2 h. Small tissue pieces (~2 mm3) were cut out, washed in ice-cold PBS (pH 7.4, Gibco, 10010056) and either snap frozen in a dry-ice/ethanol bath and stored at −80 °C until analysis, or placed in 4% paraformaldehyde (Histolab, 2176) for histopathology. The remaining tissue was cut into slices using a vibratome Leica VT1200S (Leica Biosystems; speed: 0.3–0.6 mm s−1, amplitude: 3 mm, step size: 150 or 250 μm) while submerged in ice-cold preservation solution. The resulting tissue slices were punched with a disposable 8-mm biopsy punch (Kai medical, BP-80F) to obtain slices with a consistent size, transferred onto a cell culture insert (hydrophilic PTFE, 0.4 µm, Millicell, Millipore, PICM03050) in six-well plates containing 1.3 ml prewarmed custom-synthesized medium and with or without 50% dialysed human serum and immediately incubated at 37 °C and 5% CO2. After a 2 h preincubation, the cell culture inserts were transferred to new six-well plates containing fresh prewarmed medium and returned to the incubator for an additional 22 h. Fresh or cultured liver slices were washed in ice-cold PBS, transferred to microcentrifuge tubes, snap frozen and stored at −80 °C. Spent culture medium was collected, snap frozen and stored in −80 °C.Culture of primary human hepatocytesPrimary human hepatocytes were prepared from human liver as previously described57 and kept in cryostorage until use. Cells were thawed in DMEM and plated in 24-well plates in attachment medium (DMEM, 5% FBS, 4 μg ml−1 insulin, 1 μM dexamethasone) at 300,000 cells and 500 μl medium per well, and incubated for 3 h at 37 °C and 5% CO2. Medium was then changed to the custom-synthesized labelled (13C) Williams E medium described above (500 μl per well), with the addition of selenium (5 ng ml−1), transferrin (5.5 μg ml−1), insulin (0.58 ng ml−1), dexamethasone (0.1 μM), penicillin (100 U ml−1) and streptomycin (100 μg ml−1). After 24 h of incubation, 400 μl of medium (supernatant) was collected from each well; cells were then washed twice with 500 μl cold PBS and extracted with cold methanol (500 μl per well). Medium and extracts from two wells were pooled for each replicate and used for LC–MS analysis.Rat liver slicesMale Wistar rats (Charles River, Germany) were group-housed in a controlled environment (21 °C, humidity-controlled, reverse 12 h light–dark cycle) with free access to food pellets and tap water for 4 months before euthanasia (deeply anesthetized with isoflurane and decapitated). The liver of one male rat was quickly removed and placed in ice-cold preservation solution and processed in same way as human samples. All procedures were conducted in accordance with the European Union Directive 2010/63/EU, and the protocol was approved by the Ethics Committee for Animal Care and Use at Linköping University.ATP content assayATP content was determined using an enzymatic assay, normalized to the total protein content of each slice. Frozen tissue slices were homogenized in 200 µl ice-cold 70% ethanol (vol/vol) containing 2 mM EDTA (pH 10.9) using a TissueLyser II (Qiagen) with a 5-mm stainless-steel bead for 3 min at 50 Hz. The homogenate was centrifuged for 10 min at 14,000g in 4 °C; the resulting supernatant was used for the ATP assay, and the pellet for protein analysis. The supernatant was diluted 1,000-fold in 0.1 M Tris-HCl containing 2 mM EDTA (pH 7.75). ATP content was measured using a luminescent assay (BioThema, ATP kit SL) in white 96-well plates in a multimode plate reader (Glomax Explorer, Promega) with an external ATP-calibration curve. The pellet was air-dried and reconstituted in 200 µl 1 M sodium hydroxide for 60 min at 50 °C with occasional vortexing. After dilution with water to a concentration of 0.1 M sodium hydroxide, the protein content was determined using a Pierce BCA Protein Assay kit (Thermo Scientific) in clear 96-well plates in a VersaMax microplate reader (Molecular Devices) with an external bovine serum albumin calibration curve.Paraffin sectioning and histologyLiver slices cultured for 24 h and fresh liver tissue from four individuals were fixed in 4% buffered paraformaldehyde solution (Histolab, 2176) at 4 °C for 24 h and 48 h, respectively, washed in PBS (Gibco, 10010056) and stored in 70% ethanol at 4 °C before dehydration. Liver tissue was dehydrated in an automatic tissue processor (Leica TP1020), embedded in paraffin (Leica EG 1150 embedding station), cut at a thickness of 4 µm (Thermo Fisher Microm HM355S) and stained with H&E (Histolab, 01820; Leica ST4020 Small Linear Stainer). Tissue viability and morphology (including ballooning, inflammation, steatosis, fibrosis and infiltration of tumour cells) was evaluated by a specialist pathologist.Albumin, apolipoprotein B and urea assaysMedium from three tissue slices cultured for 24 h (spent medium) and/or plasma samples from five patients were collected and analysed in duplicate in each assay. Secreted albumin was measured using human albumin ELISA (Thermo Fisher Scientific, EHALB, diluted at 1:50 and 1:1,000,000). Apolipoprotein B was quantified by using the human APOB ELISA kit (R&D Systems, DAPB00, undiluted and diluted at 1:1,000). Urea levels were determined by a colorimetric urea assay (Sigma-Aldrich, MAK006, diluted at 1:15 and 1:25). All three assays were performed according to the manufacturer’s instructions. Data were normalized to weight of tissue slice and hours of incubation.TAG assayTAGs secreted from liver slices were measured using a colorimetric assay (Cayman Chemical, 10010303, undiluted) according to the manufacturer’s instructions. Absorbance was measured at 540 nm in a VersaMax microplate reader (Molecular Devices). Williams E medium from cultures of incubated liver slices for 22 h (spent medium) from four individuals with two to three biological replicates were collected and samples were run as duplicates or triplicates.NEFA assayMedium concentrations of non-esterified fatty acids (NEFAs) were measured using a NEFA-HR(2) reagent, an enzymatic colorimetric method assay (Wako Chemicals, Neuss, Germany), according to the manufacturer’s instructions.RNA-seqTotal RNA was extracted from frozen fresh liver tissue pieces and liver slices cultured for 24 h. Briefly, samples were homogenized in 2 ml microtubes (Eppendorf, 0030108078) containing 1 ml Qiazol lysis reagent (Qiagen, 79306) together with a 5-mm stainless-steel bead (Qiagen, 69989) using a TissueLyzer II (Qiagen) for 2 min at 30 Hz. Then, 200 µl chloroform was added to each sample and mixed vigorously by vortexing for 15 s followed by 3 min incubation at room temperature (RT). Samples were centrifuged at 12,000g for 15 min at 4 °C and the upper phase containing RNA was collected. Isolation of total RNA was done on a QIAcube liquid handling system (Qiagen) using RNeasy Mini Kit (Qiagen, 74104) and RNase-Free DNase Set (Qiagen, 79254) according to the manufacturer’s instructions. RNA yield and A260:A280 ratio was determined using a NanoDrop 2000 (Thermo Scientific). RNA-seq for the fresh versus cultured experiment was performed at Vienna Biocenter Core Facilities (Vienna, Austria), and reads were mapped to the human genome version hg38. Mapped reads were summarized into counts per Ensembl gene (ENSG) using subread v.2.0.3 and normalized across samples using the DESeq method58. RNA-seq for the fed versus fasted experiment was performed by Novogene according to standard protocols, and data were analysed using the Novogene bioinformatics pipeline.LC–MSPolar metabolite LC–MS analysisMedia and plasma samples were thawed on ice, and 20 µl of sample was combined with 80 µl of 50:50 HPLC-grade methanol:acetonitrile extraction solvent. Media and plasma samples were then vortexed and left at −80 C° for 30 min to precipitate proteins. To account for variation in tissue mass, tissue samples were weighted and 20 µl of ice-cold extraction solvent made of 40:40:20 HPLC-grade methanol:acetonitrile:water was added per milligram of tissue. Tissue samples were then homogenized with ~300 mg 1 mm zirconium beads using three 10-s cycles at 6,400 Hz, and left at −20 °C for 30 min to precipitate proteins. Samples were centrifuged at 13,000g at 4 °C for 10 min, and supernatants were transferred to LC–MS vials containing 200 µl glass inserts, and stored at −80 °C until analysis. An injection volume of 2 µl was used. The sample run order was chosen to block all known experimental factors.Untargeted LC–MS/MS analysis was performed using a Thermo Q Exactive orbitrap mass spectrometer coupled to a Thermo Vanquish UPLC system. Chromatographic separation of metabolites was achieved using a Millipore (SeQuant) Zic-pHILIC 100 × 2.1 mm 5 µm column maintained at 45 °C, with a 6-min linear gradient starting from a ratio of 90:10 to a ratio of 45:55 ACN:20 mM ammonium bicarbonate at pH 9.6. A Thermo Q Exactive orbitrap mass spectrometer was operated in positive and negative ion modes using a heated electrospray ionization source at a resolution of 30,000, 75 ms ion trap time for MS1 and 15,000 resolution, 50 ms ion trap time for MS2 collection. Data were collected over a mass range of 65–975 m/z, using a sheath gas flow rate of 40 units, auxillary gas flow rate of 20 units, sweep gas flow rate of 2 units, spray voltage of 3.5 kV, capillary inlet temperature of 275 °C, auxillary gas heater temperature of 350 °C and a S-lens RF level of 45. For MS2 collection, MS1 ions were isolated using a 1.0-m/z window and fragmented using a normalized collision energy of 30 eV. Fragmented ions were placed on dynamic exclusion for 5 s before being allowed to be fragmented again.Amphipathic lipid LC–MS analysisAmphipathic lipids (including fatty acids and their derivatives) from plasma and media samples were extracted and analysed as described previously59,60. For extracting bioactive lipids from tissue samples, 20 µl of 80:20 HPLC-grade ethanol:water per milligram of tissue was used. Tissue samples were then homogenized and centrifuged as described above. For each tissue sample, 75 µl of supernatant was transferred to a Axygen V-bottom plate and mixed with 350 µl of water for solid-phase extraction purification and untargeted LC–MS/MS analysis as described previously59,60.LC–MS data analysisThermo.raw format data files from the polar LC–MS analysis were converted to mzML format using the msconvert program from Proteowizard v.3.0.6485, using the vendor-provided peak picking (centroiding) algorithm. Untargeted peak detection was done with the mzMine3 (ref. 61) software, using the ADAP methods62 for chromatogram and peak detection, followed by peak alignment using the JoinAligner module. From the resulting LC–MS peak lists, any peak that was not reproducibly detected in all three replicates of at least one of the experimental groups was removed, resulting in 1,315 LC–MS peaks. Naturally occurring isotopomers, in-source fragments and other mass spectrometry artefacts were removed using the NetID11 R package based on m/z and retention time (co-elution), resulting in 733 LM–CS peaks annotated to putative compounds.Manual review of chromatograms and compound identity was performed for 69 compounds (Supplementary Table 10). Compound identity was verified by m/z and retention time of pure standards, as well as MS2 spectra matching against library spectra from Massbank of North America (MoNA; https://mona.fiehnlab.ucdavis.edu/) using the cosine metric, and also against GNPS63. In addition, metabolite identity was supported by presence of expected mass isotopomers in the 13C-labelled material. Metabolites were identified manually for the amphiphatic lipid data, and also for a few cases not captured by the automated peak detection.MI distributions were computed for each putative metabolite by integrating MS1 peaks at the expected retention time and m/z based on the carbon number, generating a total of 9,005 mass isotopomers per sample. To prevent ‘colliding’ LC–MS peaks from unrelated compounds to be misinterpreted as MI peaks, we removed any MI peak that appeared in unlabelled tissue extracts with an apparent MI fraction > 0.05 above the expected binomial distribution of naturally occurring 13C. MI fractions were corrected for naturally occurring 13C where non-negligible using the binomial transform64. The 13C enrichment e was computed by the formula \(e={\sum }_{i=0}^{n}i{x}_{i}/n\) where \({x}_{i}\) is the fraction of mass isotopomer i.Uptake/release measurementsUptake and release of metabolites by liver tissues was determined over 22 h of culture, following the initial medium change at 2 h. Metabolite concentrations were measured in ‘spent’ medium conditioned by liver slices (\({c}_{\rm{spent}}\)), and in medium incubated without tissue slices over the same time period to control for spontaneous changes (\({c}_{\rm{control}}\)), and the difference \({c}_{\rm{spent}}\mbox{–}{c}_{\rm{control}}\) was taken as an estimate of average uptake/release rates during the 22-h time period. We used three different methods to estimate concentrations. First, for metabolites present in fresh medium, the difference was computed from the known medium concentration \(c\) and fold change in peak area \(f\) as \(\Delta c=c(1\mbox{–}f)\). Second, to verify peak area-derived estimates, we performed absolute quantification in spent medium from 13C tracing experiments by the isotope dilution method, mixing samples at a 1:1 ratio with fresh 12C medium containing metabolites at known concentrations. Consider a metabolite with measured MID vector \(x\) and unknown concentration \({{c}}\) in the spent medium, MID \({x}^{0}\) and known concentration \({c}_{0}\) in the fresh 12C medium, and measured MID \({x}^{\rm{mix}}\) in the 1:1 mixture. It then holds that$$\frac{1}{2}(c+{c}_{0}){x}^{\rm{mix}}=\frac{1}{2}({c\; x}+{c}_{0}{x}^{0})$$from which we solve for the unknown concentration \({{c}}\), assuming \({x}^{0}\) to be binomial \({\rm{Bin}}(n,0.0107)\). Third, for glucose, lactate and triglycerides, concentrations in fresh and spent medium were measured by accredited metabolite assays performed at the clinical chemistry unit of Linköping University Hospital.Model-based 13C flux analysisMFA was performed based on MI fractions measured in 24-h cultured tissue samples and spent medium, assuming metabolic and isotopic steady state. An atom-level model of central liver metabolism was developed iteratively by attempting to fit measured MI fractions and uptake/release fluxes, identifying causes of poor fitting, adjusting the model and repeating until an acceptable fit was found. This analysis assumes isotopic steady state, which is never fully realized in batch cultures; therefore, derived flux estimates should be viewed as approximations. To estimate exchange fluxes between tissue and medium, a pseudo-steady-state model was used where metabolites in spent culture medium were considered as a linear mixture of fresh medium and intracellular metabolites released into the medium. The full model is provided in OpenFLUX format in Supplementary Data 2; schematics in graphML and PDF format are available at https://github.com/Nilsson-Lab-KI/liver-flux-models/, which also tracks the model development history. A more detailed model description is found in Supplementary Note 1.Model simulation was performed using the EMU framework13 implemented in Mathematica v.11 as previously described65. Metabolites present in multiple cellular compartments (cytosol, mitochondria, lysosomes or endoplasmic reticulum) were modelled as linear mixtures. The mean and standard deviation of measured MIDs across replicate slices was used for fitting, with standard deviation estimates less than 0.03 set to 0.03 throughout to account for errors not observable in replicates. In addition to uptake/release measurements as described above, literature data for O2 consumption, redox demand and protein synthesis and degradation rates were included; see Supplementary Data 3 for a complete description.The nonlinear model-fitting problem was formulated with MI fractions and fluxes as free variables66 and solved using the GAMS modelling framework with the CONOPT solver (GAMS Software) as previously described65. Simulations and model solutions were also confirmed independently using the OpenFLUX software67. In each case, the best solution from 10 separate optimizations were taken as the optimal flux vector. Goodness of fit was judged using the \({{{\chi }}}^{2}\) statistic with n = 149 independent measurements (113 MI fractions from 21 metabolites plus 36 uptake/release fluxes) and p = 69 free model parameters (free fluxes and mixture coefficients for compartments), resulting in a one-sided rejection region of \({{{\chi }}}^{2} > 96.6\) (90% quantile). The influence of each individual measurement on flux estimation was assessed from the \({{{\chi }}}^{2}\) residuals (Extended Data Fig. 4b). Flux confidence intervals were computed using the profile likelihood method68, by maximizing or minimizing net flux through each reaction.Carbon flow was computed from the optimal flux vector for each donor as follows. For each substrate \(i\), the enrichment \({e}_{{ij}}\) in each internal metabolite \(j\) was computed by simulating MIDs at the optimal flux vector with substrate i fully labelled and all other substrates unlabelled. For each metabolite \(j\) released from the network with release rate \({r}_{j}\), the carbon flow from substrate i to metabolite \(j\) is then the product \({e}_{{ij}}\,{r}_{j}\). For metabolites that were both taken up and released, we set \({e}_{{ij}}=0\) to discount exchange fluxes.Genome-scale metabolic model analysisThe iHepatocytes 2322 model56 containing metabolic network structure and reaction–gene associations was downloaded in SBML format from https://metabolicatlas.org/. z-scores contrasting the fed and fasted liver slices were computed from the transcriptomics data for each individual and gene, and averaged across the two donors. Pooled z-scores for each reaction in the metabolic network were computed by summing the z-scores over all associated genes, as previously described69. In cases where multiple reactions mapped to the same set of genes, one representative reaction was picked arbitrarily. z-scores greater than three were considered significant.Data handling and statisticsStudent’s two-sided unpaired t-test with unequal variances was used for all P values presented, computed with Mathematica v.11 (Wolfram Research). P values were not corrected for multiplicity. Data distribution was assumed to be normal, but this was not formally tested. Gene Ontology enrichment analysis (Extended Data Fig. 1d) was performed using the MGSA method70 and the resulting model-based posterior probabilities and associated standard errors are shown.Sample size calculation could not be performed since no previous data were available on expected measurement variability in our system. Since analysis of the isotope tracing data from the five donors included here showed good agreement between donors, this sample size was deemed sufficient. Investigators were blinded to donor characteristics during sample allocation, as donor information was not available at the time of allocating donor tissue to specific experiments. Data analysis was not performed blind to the conditions of the experiments. Mass spectrometry data from one donor that was part of an initial pilot experiment were excluded from the final list of five donors due to systematic differences between batches, likely caused by instrument drift. As we did not have sufficient material to perform all assays on tissue from each donor, tissue slices were randomly used for the various assays, as far as available material allowed.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
