Hepatic zonation determines tumorigenic potential of mutant β-catenin

Mouse studies

All mouse experiments were performed according to UK Home Office regulations (project licence 70/8646 and PP3908577) following approval by the University of Glasgow Animal Welfare and Ethical Review Body. Mice were housed in a mixture of indvidually ventilated cages and conventional cages in a facility that operates as a specific-pathogen-free facility at constant temperature (19–23 °C) and humidity (55 ± 10%) under a 12-h light–dark cycle with ad libitum access to standard rodent chow and water; bedding material and tunnels were also included in all cages. All mice were genotyped from ear punch biopsies at weaning by a commercial vendor (Transnetyx). To study tumorigenesis in ageing cohorts, mice were monitored daily. Humane end points were determined based on loss of weight and body condition and the appearance of abdominal swelling. Liver cancer end point was generally identified by significant abdominal swelling or progressive weight loss; tumour volume measurements were not used to determine end point. The censoring criteria for ageing liver cancer cohorts: liver tumour-free mice were censored either due to health reasons not related to liver tumours (for example, epidermal wounds and age-associated lymphoma) or when they had passed 500 days post-induction. For genetically engineered mouse models, animals were assigned to groups according to their genotype. Treatment groups were randomly assigned; however, steps were taken during group assignment to avoid separating males in to singly housed cages. Selection of groups also aimed to maintain an equal sex balance. Investigators were blinded during treatment regimens and at sample collection for timepoint experiments. For ageing experiments, it was not possible for investigators to be blind to genotype as this factor needed to be known to maintain the welfare of the experimental cohort. Group sizes were determined on historical experiments where effect size and power were sufficient to detect statical differences.

The genotypes, transgenes and alleles used for this study were as follows: Lgr5^CreER (ref. ³⁹), Gls2^CreER (ref.¹³), Glul^CreER (ref. ¹³), Cyp1a2^CreER (ref. ¹³), Igfbp2^CreER (ref. ⁴⁰), R26^CreER (ref. ⁴¹), Villin^CreER (ref. ⁴²), Ctnnb1^ex3 (ref. ⁴³), Apc^fl/fl (ref. ⁹), Apc^fl/fl-hypomorphic³¹, Rnf43^fl/fl;Znrf3^fl/fl (ref. ⁴⁴), R26^LSL–Myc (ref. ⁴⁵), R26^LSL-tdTomato (ref. ⁴⁶), Braf^V600E (refs. ^47,48), Pten^fl/fl (ref. ⁴⁹), Yap^fl/fl (obtained from the International Mouse Phenotyping Consortium Knockout Mouse Project Repository) and Taz^fl/fl (ref. ⁵⁰).

The Igfbp2-knockout mouse line was generated at the CRUK Scotland Institute by CRISPR gene-editing technology. A CRISPR project was designed to the Igfpb2 gene (ENSMUSG00000039323/GRCm39: CM000994.3). Two CRISPR guides were identified, which cut in the intronic sequence surrounding exon 2 (ENSMUSE00001244511) and were demonstrated to have high cutting efficiency in vitro: TAACTCTGTAGGTGGTAACG, TACTAGCCGCTTGGTGTTGA. Alt-R CRISPR–Cas9 crRNA, Alt-R CRISPR–Cas9 tracrRNA and Alt-R spCas9 nuclease were purchased from Integrated DNA Technologies. To generate the electroporation solution, the crRNA–tracrRNA duplex and spCas9 protein were combined in Optimem (Thermo Fisher Life Technologies). Approximately 7 h after in vitro fertilization, the one-cell stage embryos of 5–6-week-old C57BL/6J mice (Charles River) were introduced into electroporation solution and electroporated using a NEPA21 electroporator (Nepa Gene). The following day, two-cell embryos were transferred into the oviducts of pseudopregnant CD1 females (Charles River). Genotyping of subsequent pups was performed from ear samples by PCR using the primers F: AGACTGGGATGTGGAAGCAG and R: ACTTCTCAGTTCTCCAGGGC followed by Sanger sequencing.

Mice were maintained on a mixed C57BL/6 background. Genetic recombination was induced in both male and female mice, 2–4 months of age, with either an adeno-associated virus expressing Cre under the control of the TBG promoter (AAV8.TBG.Cre; AAV.TBG.PI.Cre.rBG was a gift from J. M. Wilson (Addgene plasmid #107787) to achieve temporal-specific and hepatocyte-specific Cre-mediated recombination of floxed alleles, or tamoxifen to activate Rosa26^CreER (whole body), Villin^CreER (intestinal specific) Lgr5^CreER, Glul^CreER, Cyp1a2^CreER, Igfbp2^CreER and Gls2^CreER (only male mice were used for the Gls2^CreER experiments as female mice did not recombine as efficiently as male mice when treated with tamoxifen). AAV8.TBG.Cre was administered via intravenous (i.v.) tail-vein injections at a volume of 100 μl and a concentration of either 6.4 × 10⁸ GC ml⁻¹ (ref. ⁵¹) or 2 × 10¹¹ GC ml⁻¹. Samples were excluded if there was evidence of a failed injection and impaired genetic recombination. At 6.4 × 10⁸ GC ml⁻¹, the AAV8.TBG.Cre tropism is different between sexes; therefore, these experiments were performed on male mice. An exception to this was in the treatment of Braf^V600E/+;Rnf43^fl/fl;Znrf3^fl/fl mice with dabrafenib and LGK974; here, equal numbers of male and female mice were used per experimental group. Tamoxifen (T5648, Sigma-Aldrich) was administered via intraperitoneal (i.p.) injection at 3-mg, 2-mg or 0.5-mg doses.

The PORCN-O-acyltransferase inhibitor LGK974 (205851, MedKoo Biosciences) was administered twice daily via oral gavage at 5 mg kg⁻¹ in a vehicle composed of 0.5% Tween-80 and 0.5% methylcellulose. Rapamycin (R-5000, LC Laboratories) was administered once daily via i.p. injection at 10 mg kg⁻¹ in a vehicle composed of 5% ethanol, 5% polyethylene glycol 400 and 5% Tween-80 in PBS. The BRAF inhibitor dabrafenib (A-1219, Active Biochem) was administered once daily via oral gavage at 30 mg kg⁻¹ in a vehicle composed of 0.5% hydroxypropyl methylcellulose and 0.1% Tween-80. To assay cell proliferation, the nucleotide analogue BrdU (RPN201, Amersham Biosciences) was administered 2 h before sampling via i.p. injection of 250 μl.

The AAV8.U6.shRNA-mIgfbp2, AAV8.CMV.Igfbp2 and AAV8.TBG.myrAkt were designed and constructed by a commercial vendor VectorBuilder. For the AAV8.U6.shRNA-mIgfbp2, the shRNA targeted the following sequence in the 3′ untranslated region (UTR) of the mouse Igfbp2 gene GAACCTCCCTTGCTTCTGTTA (catalogue number: P230413-1010twp; lot number: 230420AAVN02). Knockdown was confirmed using IGFBP2 IHC on livers treated with AAV8.U6.shRNA-mIgfbp2. The corresponding control AAV8.U6.shRNA-scramble was also purchased from VectorBuilder (catalogue number: AAV8LP(VB010000-0023jze)-C; lot number: 220329AAVJ07). Both AAV.U6.shRNAs were administered via i.v. tail-vein injections at 2 × 10¹¹ GC ml⁻¹. For the AAV8.CMV.Igfbp2 virus, the gene encoding mouse IGFBP2 was packaged in to an AAV8.CMV vector (catalogue number: scAAV8MP(VB241111-1375dkc); lot number: 241122AAVS06). IGFBP2 expression was confirmed using IGFBP2 IHC on livers treated with AAV8.CMV.Igfbp2. For the AAV8.TBG.myrAkt virus, mouse AKT1 with a Src myristoylation sequence at the beginning⁵² was packaged in to an AAV8.TBG vector (catalogue number: AAV8L(VB250225-1147vdk); lot number: 250311AAVG01). For the AAV8.CMV.Igfbp2 and AAV8.TBG.myrAkt experiments, two control AAV8.CMV.eGFP and AAV8.TBG.eGFP viruses were used (catalogue number: scAAV8CP(VB010000-9304aud)-f and AAV8C(VB010000-9287ffw)-b; lot number: 241114AAVP24 and 241012AAVA01).

Human tumour data

TCGA-LIHC data were downloaded from cBioPortal (https://cbioportal-datahub.s3.amazonaws.com/lihc_tcga_pan_can_atlas_2018.tar.gz). For Extended Data Fig. 10, three additional cohorts of HCC were used. The AMC⁵³ and INSERM⁵⁴ data were both downloaded from cBioPortal (https://cbioportal-datahub.s3.amazonaws.com/lihc_amc_prv.tar.gz and https://cbioportal-datahub.s3.amazonaws.com/hcc_inserm_fr_2015.tar.gz). The LINC-JP data were downloaded from the ICGC data portal (https://dcc.icgc.org/releases/release_28/Projects/LINC-JP). Mismatch-repair deficient tumours were identified with SigMA⁵⁵ and removed from the analysis. To stratify HCC samples based on CTNNB1 mutation status, only activating mutations were considered. Activating mutations were defined as missense mutations at hotspots (protein position 32, 33, 34, 35, 36, 37, 41, 45, 335, 383 and 387) and in-frame indels at hotspots (protein position 23–71) according to Chang et al.⁵⁶. Expression data were obtained from the batch-normalized RSEM (RNA-seq by Expectation-Maximization) values from TCGA-LIHC after log₂ transformation with a pseudo-count of 1.

Human tumour data: GSVA analysis

For Extended Data Fig. 2g, GSVA scores were calculated with the R package GSVA (https://bioconductor.org/packages/release/bioc/html/GSVA.html) on log₂-transformed RSEM values after removing lowly expressed genes (mean log₂(RSEM + 1) < 5).

Modelling CTNNB1 mutations

To investigate whether the observed CTNNB1 mutation hotspots are driven by the underlying mutational processes in HCC, we modelled the mutation frequencies using the trinucleotide mutational spectrum. Here only missense single-nucleotide variants were considered. We first calculated the 96-dimensional (six substitution subtypes: C > A, C > G, C > T, T > A, T > C and T > G, each flanked by one of the four types on the 5′ and 3′ sides) trinucleotide mutational spectrum of all HCC samples with at least 50 single-nucleotide variants and took the mean. This mean spectrum represents the exome-wide mutation probability \(P_i^\mathrmexome\) for each mutation type i from 1 to 96 in HCC. We then performed a renormalization using the trinucleotide frequency in the whole exome and in the gene CTNNB1 to obtain \(P_i^CTNNB1\), which represents the local mutation probability at CTNNB1 in HCC. Last, we calculated the predicted missense mutation frequency at each protein position as \(P(\mathrmmissense)=\,\sum _iP(\mathrmmissense_i^CTNNB1\), where \(P(\mathrmmissense|i)\) is the probability of a mutation of type i generating a missense mutation at this particular protein position and can be calculated using the codon table. The predicted mutation frequencies were normalized so that their sum over all protein positions was the same as that of the observed frequencies.

Apoptosis gene signature in HCC

TCGA-LIHC RNA sequencing (RNA-seq) count data and corresponding clinical information were downloaded using recount3 package (v1.6)⁵⁷. TCGA mutational data were downloaded using GenomicDataCommons⁵⁸ R package (v1.12.0). Counts were normalized by applying the variance-stabilizing transformation function from DESeq2 (v1.36)⁵⁹. Single-sample gene set enrichment analysis was performed using R package corto (v1.2)⁶⁰ with the Hallmark gene set⁶¹ ‘Apoptosis’, downloaded using msigdbr (v7.5.1)⁶². Binned MYC expression was divided into two equal bins of the same number of patients. CTNNB1 mutation status was converted to binary: mutated or not. Data were visualized using a combination of ggplot2⁶³ and cowplot⁶⁴ packages in R.

Immunofluorescence and IHC

Livers were collected and fixed in 10% neutral buffered formalin either overnight at room temperature or overnight at 4 °C. Fixed tissue was processed for paraffin embedding, and tissue blocks were cut into 5-µm sections. IHC and immunofluorescence were performed on formalin-fixed paraffin- embedded sections according to standard staining protocols. The primary antibodies used for IHC and immunofluorescence were directed to the following antigens: β-catenin (1:50, 610154, BD Biosciences), glutamine synthetase (1:300, ab73593, Abcam (immunofluorescence); 1:800; HPA007316, Sigma-Aldrich (IHC)), BrdU (1:400, ab6326, Abcam (immunofluorescence); 1:250, 347580, BD Biosciences (IHC)), HNF4α (1:300, PP-H1415-00, Perseus Proteomics), E-cadherin (1:300, 610181, BD Biosciences), Ki67 (1:1,000, 12202, Cell Signaling Technology), IGFBP2 (1:1,000, PA5-81409, Invitrogen), RFP (1:10,000 600-401-379, Rockland), cleaved caspase 3 (1:500, 9661, Cell Signaling Technology), cleaved PARP (1:1,000, ab32064, Abcam), cyclin D1 (1:150, 55506, Cell Signaling Technology), peEF2 (1:100, 2331, Cell Signaling Technology), p4E-BP1 Thr37/46 (1:250, 2855, Cell Signaling Technology), pS6(Ser235/236) (1:75, 4858, Cell Signaling Technology), ribosomal protein pS6(Ser240/244) (1:1,000, 5364, Cell Signaling Technology), SOX9 (1:500, AB5535, Millipore) and MYC (1:800, ab32072, Abcam). Representative brightfield images were acquired using an Olympus BX53 microscope and Olympus cellSens imaging software (v1.7.1). Immunofluorescent images were acquired using the Zeiss LSM 710 confocal microscope and the ZEN Black image acquisition software. Images were further processed using Fiji/ImageJ software (v1.53t)⁶⁵ .

Image analysis

Immunofluorescent images were acquired in up to four fluorescent channels at ×20 magnification on an Opera Phenix high-content imaging system (Perkin Elmer) and subsequently analysed using the Columbus software (v2.9.1.532; Perkin Elmer). Forty images were taken per liver section. DAPI-stained nuclei were identified based on pixel intensity using method ‘B’. Nuclear size (more than 40 μm²) and morphology (roundness of more than 0.6) were then determined. Illumination correction and background normalization were performed using the sliding parabola module. Nuclei were then assigned as positive or negative based on the mean pixel intensity in the corresponding channel in either the nucleus (HNF4α, BrdU) or a 6-μm-thick region surrounding the nucleus (GLUL). The establishment and optimization of analysis algorithms was performed blind. IHC images were acquired using a SCN400F slide scanner (Leica Microsystems) at ×20 magnification. Scanned images were analysed using HALO image analysis software (v2.0.1145; Indica Labs). Liver sections were selected using the manual annotation tool and an image classifier to segment tissue and vasculature. Cell quantification and area quantification algorithms were then used to identify positive cells and staining. For the IGFBP2 IHC analysis, ten circular regions with a radius of 190 μm were used to segment areas around the central vein and portal node for each biological replicate. β-Catenin and RFP scoring was performed manually using images from the SCN400F slide scanner with the Leica Aperio ImageScope software (v12.4.3.5008). For β-catenin and RFP mosaic liver analysis, the following criteria were used: (1) nuclear β-catenin-positive and RFP-positive cells were scored in an average of 29 FOVs at ×10 magnification; (2) when scoring lesions, only clusters of five or more adjacent positive cells were quantified. Igfbp2 scoring in tumours was performed manually using the Olympus BX53 microscope. Tumours containing cells positive for Igfbp2 RNA scope probe were quantified as positive and tumours not containing cells positive for Igfbp2 RNA scope probe were quantified as negative.

RNA in situ hybridization

In situ hybridization for Igfbp2 (405958) and Notum (428988) mRNA (all from Advanced Cell Diagnostics) was performed using RNAscope 2.5 LS Reagent Kit–BROWN (322100, Advanced Cell Diagnostics) on a BOND RX autostainer (Leica Biosystems) according to the manufacturer’s instructions. Negative (dapβ, 312038) and positive (mm-Ppib, 313918) control probes (both from Advanced Cell Diagnostics) were included in each run to ensure staining specificity and RNA integrity.

RNA isolation and sequencing

RNA was isolated from whole-liver tissue using the RNeasy Mini Kit (74104, Qiagen) according to the manufacturer’s instructions. RNA concentrations were determined using a NanoDrop 200c spectrophotometer (Thermo Scientific), and quality was assessed using RNA ScreenTape on an Agilent 2200 Tapestation (Agilent Technologies). A total of 2 μg RNA was purified via poly(A) selection. RNA-seq libraries were generated using an Illumina TruSeq RNA sample prep kit and sequenced on an Illumina NextSeq 500 platform using the NextSeq 500/550 75-cycle High-Output kit (2 × 36 cycles, paired-end reads, single index), with the exception of samples from Rosa26^CreER models for which libraries were prepared using the Illumina TruSeq Stranded mRNA kit before sequencing on an Illumina Novaseq 6000 platform (2 × 150 cycles, paired-end, dual index). Raw sequence quality was assessed using the FastQC algorithm (v0.11.8). Sequences were subsequently trimmed to remove adaptor sequences and low-quality base calls, defined by a Phred score of less than 20, using the Trim Galore tool (v0.6.4). The trimmed sequences were aligned to the mouse genome build GRCm38.98 using HISAT2 (v2.1.0), with raw counts per gene subsequently determined using FeatureCounts (v1.6.4). When comparing across groups, data were normalized as a block using quantile normalization. Differential expression analysis was performed using the R package DESeq2 (v1.22.2), which uses a negative binomial generalized linear model, with significance assessed using a Wald test and Benjamini–Hochberg multiple testing correction. Reactome pathway enrichment was performed using the R package ReactomePA (v1.36.0). Gene set analysis was performed using R packages GSA (v1.03.1) when comparing against WT, and GSVA (v1.40.1) for intergroup comparison of multiple genotypes. Liver zonation gene lists were derived from analysis in Halpern et al.²⁸. Gene expression data across nine layers were filtered to include genes with non-zero expression, and then split into four zonation clusters using Euclidean distance and ‘complete’ hierarchical clustering.

NanoString GeoMX digital spatial profiler

Mouse liver sections from four biological replicates per genotype were analysed using the digital spatial profiling procedure⁶⁶. Formalin-fixed paraffin-embedded tissue sections were treated with 0.1 μg ml⁻¹ proteinase K (AM2546, Thermo Fisher Scientific) followed by heat-mediated epitope retrieval and incubated overnight with RNA oligo probes (mouse whole transcriptome atlas, Nanostring, GeoMx NGS RNA WTA Mm). Morphological markers were then detected using immunofluorescence, an anti-glutamine synthetase antibody (1:300, ab73593, Abcam) conjugated to 594 Alexa fluorescent protein (Fluorescent Protein Labeling Kits, A10239, Thermo Fisher), 647-anti-BrdU antibody (1:300, ab220075, Abcam) and the nuclear stain (SYTO13) were used. Slides were imaged at ×20 magnification using the GeoMx digital spatial profiler with the integrated software suite. Images were then used to select 2,500–300,000 μm² regions of interest (ROIs) on which the instrument focuses UV light (385 nm) to cleave the UV-sensitive probes with the subsequent release of the hybridized barcodes. For the day 4 Ctnnb1^ex3/WT samples, 32 ROIs (8 per replicate) per each condition (GLUL-high, GLUL-low, GLUL-adj., PN-adj. and BrdU^pos) were selected for UV-mediated cleavage and probe collection. For the day 60 Ctnnb1^ex3/WT;R26^LSL-MYC samples, 34 ROIs for single clones, 49 ROIs for lesions and 15 ROIs for normal GLUL^pos central veins were selected with further segmentation to GLUL^pos regions for UV-mediated cleavage and probe collection. Libraries were prepared using GeoMx Seq Code primers (NanoString) and 1× PCR Master Mix (NanoString) and AMPure XP purification. Library quality was checked using an Agilent Bioanalyzer. The libraries were run on an Illumina NovaSeq sequencing system (GeneWiz/Azenta). The FASTQ files from sequenced samples were converted into Digital Count Conversion (DCC) files using the GeoMx NGS pipeline on NanoString’s DND platform. The DCC files were uploaded onto the GeoMx digital spatial profiler analysis suite (NanoString), where they underwent quality control, filtering and Q3 normalization. Normalized GeoMx data were analysed using R base functions and packages described above.

Ribosome profiling

Livers were collected and harvested essentially as previously described⁶⁷. Livers were dissected on surfaces cleaned and treated with RNase Zap to reduce RNase exposure. Livers were rapidly dissected and snap frozen as 5 × 5 × 5 mm fragments in liquid nitrogen. Livers were ground to powder using a mortar and pestle with the CryoGrinder system (OPS Diagnostics) to ensure the samples were kept frozen. Roughly 200 mg tissue aliquots were stored in 1.5-ml tubes at −80 °C until required.

Ground liver tissue was poured directly from a 1.5-ml tube on dry ice into 900 µl ice cold lysis buffer (15 mM Tris-HCl (pH 7.5), 15 mM MgCl₂, 150 mM NaCl, 1% Triton X-100, 0.05% Tween 20, 2% n-dodecyl β-D-maltoside (89903, Thermo Fisher), 0.5 mM DDT, 100 µg ml⁻¹ cycloheximide, 1× cOmplete, Mini, EDTA-free protease inhibitor cocktail (11836170001, Merck), 1× protease inhibitor cocktail (P9599-5ml, Sigma), 5 mM sodium fluoride, 500 U ml⁻¹ RiboLock RNase inhibitor (EO0381, Thermo Fisher Scientific) and 25 U ml⁻¹ Turbo DNase (AM2239, Thermo Fisher Scientific) on ice and immediately mixed together and placed on ice for 10 min, while being inverted every 2 min to maximize the dispersal and exposure of ground tissue to lysis buffer. Lysates were centrifuged at 4 °C for 5 min at 16,000g and 800 µl supernatant pipetted into a fresh 1.5-ml tube on ice. For cytoplasmic RNA sample, RNA was extracted from 40 µl lysate with 1 ml TRIzol, as per the manufacturers’ instructions. Lysate (600 µl) was digested with 1 µl Ambion RNase I cloned, 100 U µl⁻¹ (AM2295, Thermo Fisher Scientific), 1 µl MNase (1:20 diln; 0247S, New England Biolabs) with 3.6 µl 0.25 M CaCl₂, at 22 °C for 15 min at 650 rpm in a thermomixer. The digestion was stopped with 5 µl SUPERaseIn RNase inhibitor (20 U μl⁻¹; AM2696, Thermo Fisher Scientific) and 26 µl 0.5 M EGTA (to stop RNase 1 and MNase, respectively). The samples were loaded onto a 10–50% sucrose gradient, containing 15 mM Tris-HCl (pH 7.5), 15 mM MgCl₂, 150 mM NaCl and 100 µg ml⁻¹ cycloheximide, prepared with a BioComp gradient station and cooled to 4 °C for at least 1 h. Samples were centrifuged in a Beckman XPN-90 Ultracentrifuge with an SW40ti rotor at 38,000 rpm for 2 h at 4 °C. One-ml fractions were collected with a BioComp gradient station and Gilson FC 203B fraction collector. Fractions pertaining to the 80S peak were extracted with acid phenol chloroform, followed by two chloroform washes, and RNA was precipitated with 2 µl glycogen (10901393001, Roche), 300 mM NaOac (pH 5.2) and an equal volume of isopropanol overnight at −20 °C.

RNA was pelleted by centrifugation at 12,000g for 45 min at 4 °C. The supernatant was removed with a pipette, and the pellet was washed twice with 70% ethanol and dissolved in 10 µl RNase-free water. RNA was diluted with 10 µl 2× TBE-urea sample buffer (LC6876, Thermo Fisher Scientific), heated at 80 °C for 90 s, placed immediately on ice, and then loaded onto a pre-run 15% TBE-urea gel (EC68852BOX, Thermo Fisher Scientific) and ran at 200 V for 1 h alongside custom 28-nt (AGCGUGUACUCCGAAGAGGAUCCAACGU) and 34-nt (GCAUUAACGCGAACUCGGCCUACAAUAGUGACGU) RNA markers. The gel was stained with 1× SYBR gold (S11494, Thermo Fisher Scientific) and imaged on a Typhoon FLA 7000. An image was printed to size to allow bands, inclusive of 28-nt and exclusive of 34-nt markers, to be cut from the gel, placed into a 1.5-ml RNA low-binding microcentrifuge tube, and crushed with a RNase-free disposable pestle. RNA was eluted from the crushed gel pieces in 500 µl extraction buffer (300 mM NaOAc pH 5.2, 1 mM EDTA and 0.25% SDS) overnight at 16 °C at 600 rpm in a thermomixer. Gel pieces were filtered out with a Costar Spin-X centrifuge tube filter (0.45 µm; 8163, Scientific Laboratory Supplies) and RNA was precipitated with 2 µl glycogen and 500 µl isopropanol overnight at −20 °C.

Precipitated RNA was again pelleted, washed and dissolved in 13.5 µl RNase-free water, as above, and underwent rRNA depletion as follows. To this, 13.5 µl RNA was added, 5 µl hybridization buffer (10 mM Tris-HCl pH 7.5, 1 mM EDTA and 2 M NaCl), 1 µl RNasin plus ribonuclease inhibitor (N2615, Promega) and 0.5 µl biotinylated DNA oligo pool (Supplementary Table 1; 100 µM total DNA with rRNA_depl_1 and rRNA_depl_2 at a 3:1 molar ratio compared with all other oligos) that had been denatured at 95 °C for 3 min and then placed immediately on ice. This mix was incubated at 68 °C for 10 min at 1,250 rpm in a thermomixer and then allowed to cool slowly to room temperature by turning off the thermomixer. rRNA was depleted with 160 µl Dynabeads MyOne Streptavidin C1 (65001, Thermo Scientific) as per the manufacturer’s instructions and depleted RNA was precipitated with 2 µl glycogen and ethanol.

Precipitated RNA was again pelleted and washed as above and then dissolved in 43 µl RNase-free water and heated at 80 °C for 90 s and immediately placed on ice. RNA then underwent 5′ phosphorylation and 3′ dephosphorylation with 1 µl T4 PNK (M0201S, NEB), 5 µl 10× PNK buffer and 1 µl SUPERaseIn RNase inhibitor (20 U µl⁻¹) at 37 °C for 35 min, with 5 µl 10 mM ATP added for the final 20 min. RNA was extracted with acid phenol–chloroform and precipitated with isopropanol as above.

Purified RNA was quantified on a qubit with the high-sensitivity RNA kit (Q32852) and 5 ng was input into the Bioo Scientific Nextflex small RNA v3 kit (NOVA-5132-06), using the alternative step F bead clean-up, 14 PCR cycles and gel-extraction option.

Cytoplasmic RNA samples were run on an Agilent TapeStation to check RNA integrity. RNA concentration and sample purity were measured on a NanoDrop spectrophotometer. rRNA was depleted from 1 µg cytoplasmic RNA with the RiboCop v2, and sequencing libraries were prepared from the rRNA-depleted RNA with the Corall Total RNA-Seq Library Prep Kit v1 (095.96, Lexogen) using 13 PCR cycles.

Final cytoplasmic RNA and RPF libraries were quantified on an Agilent TapeStation, with a high-sensitivity D1000 ScreenTape (5067-5584, Agilent), and sequenced single-end on an Illumina NextSeq500 instrument with a 75 cycles high-output kit.

Polysome profiles

For undigested polysome profiles, liver tissue was collected, harvested and lysed exactly as for ribosome profiling, except that 600 µl of lysate was loaded straight onto a 10–50% sucrose gradient and centrifuged in a Beckman XPN-90 Ultracentrifuge with a SW40ti rotor at 38,000 rpm for 2 h at 4 °C and run on a BioComp gradient station.

Bioinformatic processing of ribosome profiles

Ribosome footprinting analysis was performed using the Bushell laboratory’s RiboSeq GitHub pipeline (https://github.com/Bushell-lab/Ribo-seq). All versions of scripts used in this publication can be found on Zenodo⁶⁸ (https://zenodo.org/records/17224880). All R scripts were carried out using R (v4.3.1). Basic explanations of what these scripts do is written below.

Raw fastq files were quality control checked with fastQC and have been uploaded to the Gene Expression Omnibus database with the accession number: GSE275864. Cutadapt (v1.18)⁶⁹ was used to remove adaptors, trim 3′ bases with Phred scores < 20 and discard reads fewer than 30 and more than 50 bases after trimming. UMI-tools (v1.0.1)⁷⁰ was used to extract unique molecular indexes (UMIs; 4 nt of random sequence at the start and end of every RPF read) from the reads and appended to the read name. Reads were aligned with BBmap (v38.18), first to remove reads that aligned to either rRNA or tRNA sequences. Non-rRNA or tRNA reads were then aligned to a filtered protein-coding FASTA (see below), containing the most abundant transcript per gene (calculated from cytoplasmic RNA samples). The resulting BAM files were sorted and indexed with samtools (v1.9) and deduplicated using UMI-tools with the directional method. The number of reads with the 5′ end at each position of every transcript was then counted. Protein-coding-aligned read lengths peaked at 33–34 nt and lengths 30–38 were used in this analysis, which showed strong coding sequence (CDS) enrichment and periodicity. The offset for each read length was determined to be 12 nt for read lengths 30–31 nt, 13 nt for 32–35 nt and 14 nt for 36–38 nt. These offsets were applied to collapse all reads into a single frame. Total counts across the entire CDS, excluding the first 20 and last 10 codons, were then summed together and used as input to DESeq2 (ref. ⁵⁹).

The paired cytoplasmic RNA samples were processed as for the RPFs above but with the following exceptions: the UMIs were 12 nt at the start of each read only. No maximum read length was set when trimming reads with Cutadapt. Reads were aligned to the filtered protein-coding transcriptome with Bowtie2 (v2.3.5.1)⁷¹, using the parameters recommended for use with RSEM: –sensitive –dpad 0 –gbar 99999999 –mp 1,1 –np 1 –score-min L,0,-0.1. Both gene-level and isoform-level quantification was performed using RSEM (v1.3.3)⁷². The isoform quantification was used to determine the most abundant transcript per gene (see below), but differential expression was measured at the gene level with DESeq2 (ref. ⁵⁹).

The gencode.vM27.pc_transcripts.fa file was downloaded from https://www.gencodegenes.org/mouse/release_M27.html and filtered to include only transcripts that had been manually annotated by HAVANA and that have a 5′ UTR, a 3′ UTR, a CDS equally divisible by three, an nUG start codon and a stop codon. All PAR_Y transcripts were also removed. The cytoplasmic RNA-seq data were then aligned to this filtered FASTA and the most abundant transcript per gene was determined, based on the mean transcripts per million (TPMs) across all samples from the RSEM output. The RPF reads were then aligned to a FASTA containing only the most abundant transcript for each gene.

DESeq2 (ref. ⁵⁹) was used to test for differential expression. This was carried out separately on either the RPF or cytoplasmic RNA samples to calculate log₂FCs and plot principal component analyses and also to test whether the change in RPFs could be explained by the change in cytoplasmic RNA, as previously described⁷³. TE down groups were mRNAs with adjusted P < 0.1 and log₂FC < 0, TE up groups were mRNAs with adjusted P < 0.1 and log₂FC > 0, no change groups were mRNAs with adjusted P > 0.9 and not significant (NS) mRNAs were those with 0.1 ≥ adjusted P ≤ 0.9.

Gene Ontology over-representation analysis was performed with the R package EnrichR⁷⁴ using the genes in the groups identified as being upregulated or downregulated at the translational level (TE up or TE down).

Statistical analyses

A priori based on historical datasets used to ensure the smallest sample size that could give a significant difference was chosen in accordance with the 3Rs. Statistical analysis was performed with GraphPad Prism (v7.0.4) for Windows (GraphPad Software; www.graphpad.com). Normal distribution of data was determined using the D’Agostino and Pearson omnibus normality test. For non-parametric data, or where the sample size (n) was too small to determine normal distribution, data significance was analysed using a one-tailed or two-tailed Mann–Whitney test; for parametric data, data significance was analysed using a one-tailed or two-tailed unpaired Student’s t-test. Paired data were analysed using a pairwise Wilcoxon test. For comparison between more than two groups, a one-way ANOVA, Kruskal–Wallis or two-way ANOVA was used with either Tukey’s, Dunn’s or Sidak’s multiple comparisons post-hoc tests. To analyse differences in the distribution of data, a Kolmogorov–Smirnov cumulative frequency test was used. Statistical comparisons of survival data were performed using the ‘log-rank’ (Mantel–Cox) test. P ≤ 0.05 was considered significant. For individual value plots, data are displayed as mean ± s.d., unless stated otherwise. Statistical tests and corresponding P values are indicated in the figure legends and figures, respectively. For all histological analysis, the samples were randomized and the researchers were blinded to the genotype or treatment.

Reporting summary

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