Multimodal decoding of human liver regeneration – Nature

Study participants

University of Edinburgh, UK

Local approval for procuring human liver tissue for snRNA-seq, spatial transcriptomics and histological analysis was obtained from the Scotland ‘A’ Research and Ethics Committee (16/SS/0136) and the NRS BioResource and Tissue Governance Unit (study number SR574), following review at the East of Scotland Research Ethics Service (reference 15/ES/0094). Written informed consent was obtained from the participant or a legally authorized representative before enrolment per local regulations. ALF liver tissue was obtained intraoperatively from patients undergoing orthotopic liver transplantation at the Scottish Liver Transplant Unit, Royal Infirmary of Edinburgh. Patient demographics are summarized in Supplementary Table 1 for patients transplanted for APAP-induced ALF and non-A-E ALF. Healthy non-lesional liver tissue was obtained intraoperatively from patients undergoing surgical liver resection for solitary colorectal metastasis at the Hepatobiliary and Pancreatic Unit, Department of Clinical Surgery, Royal Infirmary of Edinburgh. Patients with a known history of chronic liver disease, abnormal liver function tests or those who had received systemic chemotherapy within the past 4 months were excluded from this cohort. For histological assessment of human ALF and chronic liver disease tissue, anonymized unstained formalin-fixed, paraffin-embedded liver tissue sections were provided by the Lothian NRS Human Annotated Bioresource under authority from the East of Scotland Research Ethics Service REC 1 (reference 15/ES/0094).

US Acute Liver Failure Study Group network

The Acute Liver Failure Study Group (ALFSG) consortium of US liver centres was established in 1998 to better define causes and outcomes of acute liver injury and ALF. The study protocol was approved by the local institutional review boards of the participating sites: University of Texas Southwestern Medical Center; Baylor University Medical Center, Dallas, TX; Medical University of South Carolina, Charleston, SC; University of Washington, Seattle, WA; Washington University, St. Louis, MO; University of California, San Francisco, and California Pacific Medical Center, San Francisco, CA; University of Nebraska, Omaha, NE; Mount Sinai Medical Center and Columbia University Medical Center, New York, NY; Mayo Clinic, Rochester, MN; University of Pittsburgh, Pittsburgh, PA; Northwestern University, Chicago, IL; Oregon Health Sciences Center, Portland, OR; University of California, Los Angeles, CA; University of Michigan, Ann Arbor, MI; Yale University, New Haven, CT; University of Alabama, Birmingham, AL; Massachusetts General Hospital, Boston, MA; Duke University, Durham, NC; Mayo Clinic, Scottsdale, AZ; Albert Einstein Medical Center and University of Pennsylvania, Philadelphia, PA; Virginia Commonwealth University, Richmond, VA; University of California, Davis, CA; Mayo Clinic, Jacksonville, FL; University of California, San Diego, CA; The Ohio State University, Columbus, OH; University of Kansas Medical Center, Kansas City, KS; Emory University, Atlanta, GA; University of Alberta, Edmonton, Canada. Written informed consent was obtained from the participant or a legally authorized representative before enrolment per local regulations. Sites obtained portions of fresh explanted liver tissue cut into 1 cm³ pieces, placed into individual cryovials and stored at −80 °C until requested for study. The ALFSG was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK; grant no. U‐01‐58369). The samples used in this study were supplied by the NIDDK Central Repositories. This article does not necessarily reflect the opinions or views of the NIDDK Central Repositories or the NIDDK.

University of Cambridge, UK

Patients were recruited at Addenbrooke’s Hospital, Cambridge, UK with approval from the Health and Social Care Research Ethics Committee A, Office for Research Ethics Committees, Northern Ireland (ORECNI) (16/NI/0196 and 20/NI/0109). Written informed consent was obtained from the participant or a legally authorized representative before enrolment per local regulations. Liver tissue from patients with ALF was derived from explanted livers at the time of transplantation. All tissue samples were snap-frozen in liquid nitrogen and stored at −80 °C in the Human Research Tissue Bank of the Cambridge University Hospitals NHS Foundation Trust.

University of Birmingham, UK

Human liver tissue obtained from the University of Birmingham, UK was obtained under approval by South Birmingham Ethics Committee, Birmingham, UK (reference 06/Q2708/11, 06/Q2702/61), and written informed consent was obtained from the participant or a legally authorized representative before enrolment per local regulations. Liver tissue was acquired from explanted livers from patients undergoing orthotopic liver transplantation at the Queen Elizabeth Hospital, Birmingham. All tissue samples were snap-frozen in liquid nitrogen and stored at −80 °C before being processed and shipped by the Birmingham Human Biomaterials Resource Centre (reference 09/H1010/75; 18-319).

University College, London, UK

Human liver tissue obtained from University College, London, UK was obtained under local ethical approval (London-Hampstead Research Ethics Committee; reference 07/Q0501/50). Written informed consent was obtained from the participant or a legally authorized representative before enrolment per local regulations. Liver tissue (formalin-fixed, paraffin-embedded) was acquired via transjugular liver biopsy from patients presenting with acute, severe liver injury; these patients spontaneously recovered without liver transplantation.

Mice

Mice used for all experiments were 8–12 weeks of age and housed in conventional barrier unit facilities with conventional bedding, 12–12-h light–dark cycle, ambient temperature control (21 °C; humidity 40–60%) and access to food and water ad libitum, under pathogen-free conditions at the University of Edinburgh. Male mice were used for all experiments unless described otherwise in the figure legends. Age- and sex-matched mice were randomly assigned to treatment groups. Blinding to the origin of the tissue samples was not performed. The investigators performing the immunofluorescence staining, single-nucleus RNA-seq and RT-qPCR were different from the investigators collecting tissue. The IVM studies were not blinded. All experimental protocols were approved by the University of Edinburgh Animal Welfare and Ethics Board in accordance with UK Home Office regulations. C57BL/6JCrl mice were obtained from Charles River Laboratories (UK). mTmG (Jax 007676; B6.129(Cg)-Gt(ROSA)26Sor^{tm4(ACTB-tdTomato,-EGFP)Luo}/J)³² and TdTomato (Jax 007914; B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J)³³ reporter mice were obtained from Jackson Laboratories. For the APAP-induced acute liver injury model, mice fasted for 12 h before intraperitoneal injection with 300 mg kg⁻¹ (male) or 350 mg kg⁻¹ (female) of APAP dissolved in sterile PBS as previously described³⁴. Acute and chronic CCl₄-induced liver injury was induced as previously described³⁵. For bile duct ligation mice were anaesthetized with isoflurane and the common bile duct was surgically ligated. Buprenorphine pain relief was administered postoperatively via subcutaneous injection and animals were maintained at 25 °C for the duration of the study. For DDC (3,5-diethoxycarbonyl-1,4-dihydrocollidine) diet-induced cholestatic liver injury, mice were given 0.1% DDC mixed with normal chow (Special Diet Services) for 21 days.

For hepatocyte-specific AAV8-Cre-mediated reporter gene induction, stock AAV8.TBG.PI.Cre.rBG (AAV8-TBG-Cre; a gift from J. M. Wilson (plasmid #107787, Addgene); stored at −80 °C) was thawed on ice, diluted in sterile PBS to achieve a working titre of 2 × 10¹² genetic copies per millilitre and was subsequently stored at −20 °C until usage. On the day of injection, the diluted AAV was thawed and each mouse was injected via the tail vein with 100 μl (2 × 10¹¹ genetic copies per mouse)³⁶. Mice were left for 1 week before APAP-induced acute liver injury. For in vivo hepatocyte-specific knockdown of Anxa2, mice were intravenously injected with 1 × 10¹² genetic copies per millilitre AAV8-GFP-U6-mANXA2-shRNA (shAnxa2) or AAV8-GFP-U6-scrmb-shRNA (shScrmb) and left for 2 weeks before APAP-induced acute liver injury. For administration of BrdU in drinking water, BrdU was dissolved in drinking water at a concentration of 0.8 mg ml⁻¹.

Nuclei isolation for snRNA-seq

Human and mouse livers for snRNA-seq were processed as previously described using the TST method³⁷. Mouse liver nuclei isolation was performed on n = 3 mice per timepoint and nuclei were pooled for snRNA-seq.

Droplet-based snRNA-seq

Single nuclei were processed through the 10X Genomics Chromium Platform using the Chromium Single Cell 3′ Library and Gel Bead Kit v3 (PN-1000075, 10X Genomics) and the Chromium Single Cell B Chip Kit (PN-1000074, 10X Genomics) as per the manufacturer’s protocol, and as previously described⁷. Libraries were sequenced on either an Illumina HiSeq 4000 or NovaSeq 6000.

Spatial transcriptomics

Unfixed liver tissues were embedded in Tissue-Tek (OCT) and snap-frozen. Samples were then cryosectioned (10 µm) and placed on pre-chilled Visium (10X Genomics) tissue optimization slides or Visium spatial gene expression slides. Spot size was 55 µm, with 100 µm between spot centroids. Tissue sections were processed as per the manufacturer’s protocol. On the basis of optimization time course experiments, tissue was permeabilized for 18 min.

Multiplex smFISH

Unfixed snap-frozen liver tissues were cryosectioned (10 µm) onto Resolve Biosciences slides and sent on dry ice to Resolve Biosciences for processing. Gene probes were designed using Resolve Biosciences proprietary design algorithm. Probe details are provided in Supplementary Table 6. Following sample imaging, spot segmentation, pre-processing, and signal segmentation and decoding, final image analysis was performed in R programming language v3.4.1.

Liver IVM

Single-colour tdTomato imaging was performed using a Discovery (Coherent) ‘multiphoton’ laser tuned to 1,050 nm through a ×20 VIS–IR-corrected water immersion objective (NA 1.0) by placing a water drop on top of the coverslip. Dual-colour eGFP–tdTomato imaging was performed using the same setup with the laser tuned to 1,000 nm. Non-descanned GaAsP detectors (GFP NDD filter BP 500–550 nm; tdTomato NDD filter BP 575–610 nm) were used to initially obtain an overview image, after which three peri-central vein fields were selected for timelapse imaging. These three fields were then imaged as z-stacks (40–50 µm) every 10 min for 6 h. Following administration of APAP (350 mg kg⁻¹) mice were anaesthetized with isoflurane (4% induction, 1–1.5% maintenance) in approximately 95% oxygen (0.8 l min⁻¹) produced by an oxygen scavenger (Vettech). The coat above the liver was clipped back, Lacrilube was applied to the eyes, and mice were then placed in a dorsal position on a heated stage on an upright LSM 880 NLO multiphoton microscope (Zeiss). An abdominal incision was made, exposing the surface of the liver; this was then stabilized using a custom coverslip-holding imaging vacuum stabilization armature attached to the stage. The gentlest possible vacuum was applied to the surface in contact with the liver, holding it in place against the coverslip. Mice received subcutaneous fluids every 45 min during imaging. At the end of the imaging session, mice were killed under general anaesthesia by cervical dislocation.

Liver IVM processing and segmentation

Timelapse image analysis and visualization were performed using Imaris 9.7 (Bitplane). The Imaris ‘reference frame’ was first used to correct for xyz drift. To create 4D rendering of the wound, the hepatocyte channel was inverted and smoothed using a Gaussian filter. The Imaris ‘surface’ tool was then used to create a surface corresponding to the wound. Object statistics were then exported to analyse the evolution of the volume of the surface over time (expressed as the percentage of initial volume). To create 4D rendering of individual hepatocytes from the mTmG mice, registered images were first imported in Google drive and the online platform ZerocostDL4mic³⁸ was used to perform Cellpose segmentation³⁹ on the eGFP (hepatocytes) channel. Cellpose-annotated images were processed in ImageJ using the ‘Label to ROI’ plugin⁴⁰ to create eroded region of interests (ROIs) followed by the ‘Mask from ROI’ plugin to produce cell masks. To reduce nonspecific segmentation, the tdTomato signal was thresholded and subtracted from the mask channel using the ‘channel arithmetics’ tool in Imaris. The Imaris ‘surface’ tool was then used to create surfaces corresponding to the hepatocytes. Cell statistics were then exported to analyse morphodynamic parameters according to their relation to the wound (distance to the wound). Cell behaviour was determined using the track length and speed (indicating cell mobility) and the standard deviation of cell sphericity and ellipticity (indicating changes in cell shape over time).

Immunohistochemistry and immunocytochemistry staining

Immunohistochemistry and immunofluorescence staining was performed on formalin-fixed, paraffin-embedded liver tissue sections (4 µm). Slides were deparaffinized and immunofluorescently labelled using a Leica Bond RX_m automated robotic staining system. Antigen retrieval was performed using Bond Epitope Retrieval Solution 1 or 2 for 20 min in conjunction with heat application. Sections were then incubated with primary antibodies diluted in 0.1% Triton-X containing PBS. Sections were stained with DAPI (Sigma) and mounted on glass coverslips in ProLong Gold Antifade mounting medium and stored at 4 °C until time of imaging. For in vitro EdU detection, cells were washed in PBS then fixed for 10 min at room temperature with 4% formaldehyde solution in PBS, and cells were then stained according to the manufacturer’s protocol. TROMA-III was deposited to the DSHB by Kemler, R. (DSHB Hybridoma Product TROMA-III). A full list of antibodies and conditions is included in Supplementary Table 6.

Histology image processing and segmentation

Slides were scanned using a Zeiss Axioscan Z1. Images were processed using Zen Blue software (v2.6). For in vitro immunocytochemistry, wells were imaged using an EVOS FL Auto Imaging System. All image analysis was undertaken in QuPath (v0.3.0)⁴¹ with StarDist nuclei detection extension⁴².

Cell culture

Human immortalized hepatocyte cell line (Huh-7; 300156, Cell Lines Service) was cultured using RPMI 1640 supplemented with 10% FBS and 2 mM l-glutamine. Huh-7 cells were authenticated using STR profiling and mycoplasma tested by the commercial provider. Primary mouse hepatocytes were isolated and cultured as previously described⁴³.

Gene knockdown in hepatocytes

Gene knockdown in Huh7 and primary mouse hepatocytes was performed using siRNA. Cells were plated at 500,000 cells per millilitre (Huh7, Corning Costar; primary mouse hepatocytes, Corning Primaria) followed by serum starvation overnight (in medium without FBS). siRNA duplexes with Lipofectamine RNAiMAX Transfection Reagent were prepared in OptiMEM according to the manufacturer’s recommendations and used at a concentration of 50 nM. Cells were exposed to the duplex for 48 h, in antibiotic-free media containing 2% FBS. Cells were harvested for RNA and RT–qPCR. Gene knockdown efficiency was assessed by RT–qPCR. Cells were treated with control siRNA (1027280, Qiagen), siRNA for ANXA2 (human, Hs_ANXA2_8, SI02632385, Qiagen) or siRNA for Anxa2 (mouse, Mm_Anxa2_3, SI00167496, Qiagen).

Scratch wound assay

The scratch wound assay was performed using the IncuCyte system (Essen Bioscience). Huh7 cells were plated in IncuCyte ImageLock Plates (Essen Bioscience) and treated as above for ANXA2 gene knockdown. The subconfluent monolayer was then wounded using the IncuCyte Woundmaker. To obtain a confluent monolayer of primary mouse hepatocytes for wound assays, cells were plated as previously described⁴⁴, with modifications. In brief, three separate additions of 500,000 cells per millilitre were seeded onto collagen I-coated IncuCyte ImageLock plates at 2-h intervals. Non-adherent hepatocytes were removed between additions with warmed PBS. Cells were then treated as above for gene knockdown before wounding using the IncuCyte Woundmaker. Following wounding, cells were maintained in complete media with the addition of human HGF (100 ng ml⁻¹) and 10% FBS for the duration of the assay. EdU (10 µM) was added to the media 24 h before the end of the assay to assess proliferation. For analysis of wound healing, the scratch wound plugin for the IncuCyte Zoom was used. All experiments were performed as quadruple technical replicates; the number of independent experiments is specified in the figure legends.

In vitro phagocytosis assay

Primary mouse hepatocytes were isolated from uninjured livers as described above and plated into six-well Primaria plates (Corning). Hepatocytes were treated as before for Anxa2 knockdown and hepatocyte death was induced at 48 h post-knockdown using 10 mM APAP. Scrmb-siRNA (control) or Anxa2-siRNA treated dead hepatocytes were then used in an in vitro phagocytosis assay. Bone marrow-derived macrophages (BMDMs) were isolated from mice femurs and differentiated for 7 days in culture with the addition of 10⁴ U ml⁻¹ CSF1. Following differentiation, BMDMs were plated into 24-well plates at 250,000 cells per well and cultured overnight. The next day, dead hepatocytes were cultured with CypHer5e NHS Ester (PA15401) for 10 min at room temperature in the dark, and washed three times in PBS before being applied to BMDMs. CypHer5e-stained dead hepatocytes were then cultured with BMDMs at 37 °C for 1.5 h. Non-ingested hepatocytes were then removed by vigorous washing three times with PBS and residual adherent BMDMs were used for subsequent analysis. For flow cytometric analysis, BMDMs were stained with F4/80 (123141, BioLegend; 30 min at 4 °C, 1:100), and a cell viability stain (DAPI; 1:1,000) was performed immediately before acquiring the samples. Data acquisition was performed on a BD LSR Fortessa flow cytometer (Becton Dickinson) using BD FACS Diva software, and data were analysed using FlowJo 10.9.0 software. The gating strategy is outlined in Supplementary Table 6.

RNA extraction and RT–qPCR

RNA was extracted from primary mouse hepatocytes and Huh-7 cells using the RNeasy Plus Micro Kit, and cDNA synthesis was performed using the QuantiTect Reverse Transcription Kit according to the manufacturer’s protocol (Qiagen). Reactions were performed in triplicate in 384-well plate format. RT–qPCR was performed using PowerUp SYBR Green Master Mix. Primers are detailed in Supplementary Table 6. Samples were amplified on an ABI Quantstudio 5 (Applied Biosystems, Thermo Fisher Scientific). The 2^-ΔΔCt quantification method, using GAPDH/Gapdh for normalization, was used to estimate the amount of target mRNA in samples, and expression was calculated relative to average mRNA expression levels from control samples.

Computational analysis

Four computational datasets were analysed: (1) 72,262 human nuclei from healthy (n = 9), APAP-ALF (n = 10) and NAE-ALF (n = 12) livers; (2) 59,051 mouse nuclei from an APAP-induced acute liver injury timecourse; (3) spatial transcriptomics spots from human liver (n = 3 healthy, n = 2 APAP-ALF and n = 2 NAE-ALF) and mouse liver (n = 1 per timepoint); and (4) multiplex smFISH from human liver (n = 2 healthy and n = 2 APAP-ALF).

snRNA-seq analysis

We aligned to GRCh38 and mm10 (Ensembl 93) reference genomes (modified to allow intronic and exonic feature alignment), and estimated nuclei-containing partitions and unique molecular identifiers (UMIs), using the CellRanger v3.1.0 Single-Cell Software Suite from 10X Genomics. Further analysis was performed in the R programming language v3.4.1.

To enable reproducible analysis, we developed the SeuratPipe R package v1.0.0 (https://doi.org/10.5281/zenodo.7331092), a pipeline building on existing packages. In brief, we performed analysis as follows: we performed per-dataset quality control in the Seurat⁴⁵ R package v4.1.1. We used the Scrublet⁴⁶ Python module v0.2.3 to identify potential doublets and the SoupX⁴⁷ R package v1.5.2 to automatically calculate and correct for background contamination. Finally, we excluded nuclei that expressed fewer than 1,000 genes, or mitochondrial gene content of more than 5% of the total UMI count.

After merging the individual datasets, we normalized feature counts per nuclei by dividing the total UMI count for that nuclei, then multiplying by a scale factor of 10,000 and natural-log transforming. We corrected for sample bias by obtaining principal component embeddings using the Harmony⁴⁸ R package v0.1.0. Furthermore, we downsampled the hepatocyte populations to standardize sample contribution to downstream analysis.

Nucleus clusters were identified using the shared nearest neighbour modularity optimization-based clustering algorithm implemented in Seurat, using Harmony-corrected principal components for the purpose of constructing the shared nearest neighbour graph. We calculated differentially expressed features using a Wilcoxon rank-sum test. To annotate these clusters, we used a curated list of known marker genes per cell lineage in the liver (Supplementary Table 2) to obtain signature scores using the AddModuleScore function in Seurat. Clusters identified as primarily composed of cycling cells were reclustered to split them out into their constituent lineages as above. We then iteratively applied the above workflow for each lineage thus identified, inserting a ‘cleansing’ step in which we removed clusters displaying an abundance of nuclei previously identified as doublets or overexpressing marker genes of other lineages. We generated a hepatocyte migration gene module (Supplementary Table 2) using the top 25 (by avg_log₂FC) differentially expressed features from the human migratory hepatocyte cluster. GO analysis was performed using the clusterProfiler⁴⁹ R package v4.8.3. Liver zonation specificity scores were obtained by first scaling central and portal zonation signature scores to a value between 0 and 1, and subsequently setting zonation score = central score/(central score + portal score). We applied quantile thresholding when plotting features of interest.

To determine any confounding effects of age and sex in the APAP-ALF data, Pearson and point-biserial correlation were performed between the variable of interest and the harmony components.

Diffusion maps and force-directed graphs were generated in Scanpy⁵⁰ Python module v1.9. Cell cycle effects were regressed before the following dimensionality reductions. Diffusion maps were compiled based on a neighbourhood graph recalculated in Scanpy using harmony components. This neighbourhood was subsequently denoised using the diffusion map coordinates and used as input to partition-based graph abstraction alongside associated sublineage annotations. The partition-based graph abstraction was used in turn to initialize calculation of force-directed graphs.

Interactome analysis was performed on human APAP-ALF and mouse APAP-induced liver injury datasets, using CellChat⁵¹ R package v1.6.1 with default parameters. Annotations from each individual lineage were mapped back to a dataset containing all lineages before performing the analysis.

Spatial transcriptomics analysis

We aligned to GRCh38 and mm10 (Ensembl 93) reference genomes using the SpaceRanger v1.0.0 Spatial Gene Expression Software Suite from 10X Genomics. Further analysis was performed in the R programming language v3.4.1.

We performed per-dataset quality control in the Seurat R package v4.1.1. We excluded spots expressing fewer than 800 genes or mitochondrial gene content of more than 20% for both human and mouse samples of the total UMI count. We also manually filtered low-quality spots (those isolated from the main tissue section) using the 10X Genomics Loupe browser (v5.0). Similar to snRNA-seq, we computed gene signature scores for hepatocytes (TTR, TF, HP, CYP2A6, CYP2E1, CYP3A4 and HAL), myofibroblasts (ACTA2, COL1A1, COL1A2 and COL3A1) and cycling cells (genes listed in Seurat cc.genes.updated.2019). We applied quantile thresholding when plotting features of interest.

Gene expression and tissue topography were used to draw spatial trajectories across healthy and APAP-ALF tissues via the SPATA2 (ref. ⁵²) R package v0.1.0. The trajectory modelling functionality of SPATA2 was used to identify central-associated and portal-associated genes and corresponding modules (Supplementary Table 2) whose expression trajectory followed the underlying spatial model.

Multiplex smFISH analysis

Nuclei segmentation and expansion were performed using QuPath to demarcate cells. A gene–cell matrix was then obtained, quality control and normalization was applied, and signature scores in tissue were computed using pre-defined cell populations.

Further statistical analysis

Further statistical analyses were performed using GraphPad Prism. Comparison of changes between two groups was performed using a two-tailed paired Student’s t-test or unpaired Student’s t-test. Comparison of changes between groups was performed using a two-way ANOVA with Sidaks multiple comparison test with a single-pooled variance. Pearson’s correlation coefficient (r) was used to measure the relationship between variables. P < 0.05 was considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.