Somatic mutations and clonal dynamics in healthy and cirrhotic human liver

Abstract

The most common causes of chronic liver disease are excess alcohol intake, viral hepatitis and non-alcoholic fatty liver disease, with the clinical spectrum ranging in severity from hepatic inflammation to cirrhosis, liver failure or hepatocellular carcinoma (HCC). The genome of HCC exhibits diverse mutational signatures, resulting in recurrent mutations across more than 30 cancer genes^1-7. Stem cells from normal livers have a low mutational burden and limited diversity of signatures⁸, which suggests that the complexity of HCC arises during the progression to chronic liver disease and subsequent malignant transformation. Here, by sequencing whole genomes of 482 microdissections of 100-500 hepatocytes from 5 normal and 9 cirrhotic livers, we show that cirrhotic liver has a higher mutational burden than normal liver. Although rare in normal hepatocytes, structural variants, including chromothripsis, were prominent in cirrhosis. Driver mutations, such as point mutations and structural variants, affected 1-5% of clones. Clonal expansions of millimetres in diameter occurred in cirrhosis, with clones sequestered by the bands of fibrosis that surround regenerative nodules. Some mutational signatures were universal and equally active in both non-malignant hepatocytes and HCCs; some were substantially more active in HCCs than chronic liver disease; and others-arising from exogenous exposures-were present in a subset of patients. The activity of exogenous signatures between adjacent cirrhotic nodules varied by up to tenfold within each patient, as a result of clone-specific and microenvironmental forces. Synchronous HCCs exhibited the same mutational signatures as background cirrhotic liver, but with higher burden. Somatic mutations chronicle the exposures, toxicity, regeneration and clonal structure of liver tissue as it progresses from health to disease.

Extended Data Figure 1. Sensitivity analysis of SNV calls.

(A) Overview schematic of the experimental and analytical approach. (B) Examples of the variant allele fractions (VAFs) of variants from unrelated (top) and related (bottom) microdissection sample pairs from four donors (left to right). X-axis represents the VAF of sample 1 from each pair; Y-axis represents the VAF of sample 2. Each dot represents one variant. Red: variants called in both samples, yellow: variants called in sample 1, blue: variants called in sample 2. (C) Histogram of sensitivities calculated for each sample pair. (D) Heatmap of modelled sensitivity at different values of VAF and coverage. Overlaid dots represent sample pairs used to fit model. (E) Relationship of VAF, sensitivity and coverage according to fitted model of sensitivity. Overlaid dots represent sample pairs used to fit model. (F) Comparison of calculated (x-axis) and fitted (y-axis) sensitivity for each sample pair (n=34 pairs of samples). The R² value quoted is a Pearson’s correlation coefficient. (G) Proportion of hepatocytes that are multinucleated in samples analysed here, estimated by counting 500 cells in each H&E section (n=14 patients). Each point represents the proportion of a patient in the study. The horizontal bars represent the mean for that aetiological group.

Extended Data Figure 2. Copy number and structural variants in chronic liver disease.

(A, B) Genome-wide copy number profiles for two samples. Black points represent read-depth of discrete windows along the chromosome, corrected to show overall copy number. Arm-level and whole chromosome gains and losses are evident. (C-H) Focal copy number changes and structural variants. Black points represent read-depth of discrete windows along the chromosome, corrected to show overall copy number. Lines and arcs represent individual structural variants, coloured by the orientation of the joined ends (purple, tail-to-tail inverted; orange, head-to-head inverted; pale blue, tandem duplication-type orientation; pale green, deletion-type orientation). Events affecting known HCC genes are marked with labelled arrows (panels C, E, F).

Extended Data Figure 3. Events affecting known HCC genes in cohort.

(A) Distribution of somatic point mutations in individual microdissections (x axis) affecting known HCC genes (y axis). The inset to the left shows the frequency of events in individual genes. The inset to the bottom shows the aetiology attributed to the sample, and whether the sample was drawn from non-cancerous hepatocytes (left) or HCC (right). (B) Genomic position of single nucleotide substitutions (SNVs; light blue strip, top) and insertion-deletions (INDELs; dark blue strip, bottom) detected in ALB, the gene encoding albumin. (C) Relationship of gene expression in liver tissue (x axis) and proportion of indels as a fraction of all point mutations (y axis). The grey line represents a Poisson regression model with a significant (two-sided likelihood ratio test; p < 10^-16) coefficient for gene expression as a predictor for the ratio of indels (n=5458 genes included in model). The grey ribbon represents the 99% confidence interval of the parameter estimates.

Extended Data Figure 4. Phylogenetic reconstruction of hepatocyte clones in non-cirrhotic liver samples.

Left column: Heatmap representing the clustering of the variants observed in each microdissection sample (x-axis) of the non-cirrhotic livers. Each cluster (y-axis) contains mutations for which variant allele fractions across samples are very similar. The colour scale of the boxes represents the estimated mean variant allele fraction for that cluster in that sample. Middle column: Phylogenetic trees constructed from the clustering information. Solid lines: nesting is in accordance with the pigeon-hole principle. Dashed lines: nesting is in accordance with the pigeon-hole principle assuming the pool of hepatocytes to be 70% of cells. Dotted lines: nesting is only based on clustering, assigning a clone as nested if its constituent LCMs are a subset of LCMs in the parental clone. Details given in Supplementary Methods. Right column: Representation of clones according to the physical coordinates of the LCM samples, overlaid onto H&E stained sections (top), with Masson’s trichrome and Oil Red-O sections also shown (bottom). Locations of immune/inflammatory cell infiltrates are marked with yellow rings. Sample sizes were for PD36713, n=30 microdissections; PD36714, n=35 microdissections; PD36715, n=26 microdissections; PD36717, n=42 microdissections; PD36718, n=32 microdissections.

Extended Data Figure 5. Phylogenetic reconstruction of hepatocyte clones in alcohol-related cirrhosis.

Analogous to Extended Figure 4, representing the cirrhotic livers of donors PD37105, PD37107, PD37110 and PD37111. The pictures in the right column are of H&E stains on the top, with Masson’s trichrome and a macroscopic photograph of the liver on the bottom, with HCCs indicated by arrows. Locations of immune/inflammatory cell infiltrates are marked with yellow rings. Sample sizes were for PD37105, n=31 microdissections; PD37107, n=41 microdissections; PD37110, n=22 microdissections; PD37111, n=39 microdissections.

Extended Data Figure 6. Phylogenetic reconstruction of hepatocyte clones in non-alcoholic fatty liver disease with cirrhosis.

Analogous to Extended Figure 4, representing the cirrhotic livers of donors PD37113, PD37114, PD37115, PD37116 and PD37118. The pictures in the right column are of H&E stains on the top, with Masson’s trichrome and a macroscopic photograph of the liver on the bottom, with HCCs indicated by arrows. Locations of immune/inflammatory cell infiltrates are marked with yellow rings. Sample sizes were for PD37113, n=37 microdissections; PD37114, n=41 microdissections; PD37115, n=34 microdissections; PD37116, n=43 microdissections; PD37118, n=26 microdissections.

Extended Data Figure 7. Mutation spectrum of individual microdissections

From each donor, we chose 5 clones to represented the heterogeneity in trinucleotide context mutation spectra. The six substitution types are shown in the panel across the top of each clone’s data. Within each panel, the contribution from the trinucleotide context (bases immediately 5’ and 3’ of the mutated base) are shown.

Extended Data Figure 8. Details of mutational signature extractions

(A) Dot plots showing the concordance for signature attributions between the two signature algorithms (n=479 microdissections). Mutational signatures on the y axis were extracted using non-negative matrix factorisation and on the x axis using a Bayesian hierarchical Dirichlet process. Quoted R values are Pearson’s correlation coefficients. (B) Signatures extracted by non-negative matrix factorisation. The six substitution types are shown in the panel across the top of each clone’s data. Within each panel, the contribution from the trinucleotide context (bases immediately 5’ and 3’ of the mutated base) are shown. (C) Signatures extracted by the Bayesian hierarchical Dirichlet process, as for panel B. Where a signature matches one from panel B, it is shown on the same row.

Extended Data Figure 9. Transcription strand bias in mutational patterns

(A) Transcription strand bias of T>C mutations at A[T]D context before and after transcription start sites of highly expressed liver genes. (B) Bar plots representing the numbers of C>A variants on the transcribed and non-transcribed strand. Each hepatocyte clone is represented individually (x-axis). Note the strand bias in the highly mutated clones of PD37111, where the tobacco signature is most active – the strand bias indicates the damaged base is the guanine, as expected for polycyclic aromatic hydrocarbons. (C) Bar plots representing the numbers of T>A variants on the transcribed and non-transcribed strand. Each hepatocyte clone is represented individually (x-axis). Note the strand bias in the highly mutated clones of PD37107, where the aristolochic acid signature is most active – the strand bias indicates the damaged base is the adenine, as expected for polycyclic aromatic hydrocarbons.

Extended Data Figure 10. Mutations in a B lymphocyte clone in a cirrhotic liver

(A) Illustration of a portion of the B-cell receptor (IGH) region on chromosome 14. Shown are the coverage tracks of an LCM sample that does not belong to the lymphocyte lineage (top) and a sample that belongs to the lymphocyte lineage (middle). In the center of the displayed region there is a drop of copy number in the lymphocyte track, indicating a structural rearrangement. The bottom track shows the paired-end reads that contribute to a rearrangement event in the lymphocyte sample, co-localised with the drop in copy number. (B) Application of the pigeonhole principle – if two clusters of heterozygous mutations in regions of diploid copy number are in different cells, then their median variant allele fractions must sum to ≤0.5 (if they sum to >0.5, equivalent to a combined cellular fraction of >1, there must be some cells that carry both sets of mutations – hence one cluster would have a subclonal relationship with the other). Cluster 10 is the cluster with the unique VDJ rearrangement of IGH shown in panel A and the large number of mutations attributed to signature 9. Clearly, samples from clusters 2, 11 and 55 etc have VAFs which, when combined with cluster 10, sum to >0.5. Therefore, they must be subclonal to cluster 10, even though they do show signature 9. (C-H) Representative pairwise decision graphs for clusters of mutations. Median cellular fraction is shown for pairs of clusters across every sample from the patient. Where at least one sample falls above / to the right of the x+y=1 diagonal line, those two clusters must share a nested clonal-subclonal relationship.

Figure 1. Mutational burden observed in non-cancerous hepatocytes.

(A) Burden of SNVs corrected by sensitivity of mutation detection. Each boxplot represents a patient (n=14 patients; 482 microdissections), each dot represents one laser-capture microdissected sample. The grey-to-black intensity of the points reflects the median variant allele fraction (vaf) of mutations in each microdissection. Boxes in the box-and-whisker plots indicate median and interquartile range; whiskers denote range. (B) Burden of insertion-deletion (INDEL) variants (n=14 patients; 482 microdissections). (C) Burden of copy number variants (CNVs) and structural variants (SVs), represented as number of unique events per patient. (D) Chromothripsis involving chromosomes 16 and 21 observed in patient PD37111. Black points represent corrected read-depth along the chromosome. Lines and arcs represent structural variants, coloured by orientation of joined ends (purple, tail-to-tail inverted; orange, head-to-head inverted; pale blue, tandem duplication-type orientation; pale green, deletion-type orientation). (E) Chromothripsis involving chromosomes 1 and 3 observed in patient PD37105. (F) Chromothripsis involving chromosomes 2, 5 and 6 observed in patient PD37105 (in a separate clone to panel E).

Figure 2. Phylogenetic reconstruction of hepatocyte clones.

(A) Phylogenetic tree constructed from clustering of mutations across microdissected samples in a normal patient (PD36715). Lengths of branches (x axis) indicate numbers of mutations assigned to that branch. Solid lines: nesting is in accordance with the pigeon-hole principle. Dashed lines: nesting is in accordance with the pigeon-hole principle assuming hepatocytes represent 70% of cells. Dotted lines: nesting is only based on clustering, assigning a clone as nested if variant allele fractions of constituent microdissections are lower than those in the parental clone. (B) Representation of branches from the phylogenetic tree in panel A according to their physical coordinates, overlaid onto an H+E stained section. Black points represent branches of the tree sharing no mutations with any other samples; coloured points represent branches with shared clonal relationships (n=26 microdissections). (C, D) A second normal liver sample (PD36713; n=30 microdissections). (E, F) Patient with ARLD (PD37105; n=31 microdissections) (G, H) Patient with ARLD (PD37110; n=22 microdissections) (I, J) Patient with NAFLD (PD37114; n=41 microdissections) (K, L) Patient with NAFLD (PD37115; n=34 microdissections) (M, N) Patient with NAFLD (PD37116; 43 microdissections) (O, P) Patient with NAFLD (PD37118; 26 micordissections)

Figure 3. Mutational signatures in normal liver, cirrhotic liver and HCC.

(A) Number of somatic substitutions (SNVs; sensitivity-corrected for non-cancerous samples) and insertion-deletion events (INDELs) in each non-cancer microdissection sample (blue points) and associated synchronous HCC (red diamonds). (B) Stacked bar blot showing estimated proportional contributions of each mutational signature to each phylogenetically defined cluster of somatic substitutions. Data generated using a Bayesian hierarchical Dirichlet process. (C) Stacked bar blot showing proportional contributions of signatures in patients with ARLD. (D) Stacked bar blot showing estimated proportional contributions of signatures in patients with NAFLD. (E) Stacked bar blot showing estimated proportional contributions of signatures to 54 cases of HCC from TCGA. (F) Number of SNVs attributed to prevalent mutation signatures in each non-cancer microdissection sample (blue circles) and synchronous HCCs (red diamonds). Contributions for the TCGA samples are shown on the right. The y-axis is on a logarithmic scale.

Figure 4. The liver as a witness for mutagenic insults occurring throughout life.

(A) Left panel: Phylogenetic tree of clones in patient PD37111, with each branch coloured by the proportion of mutations in that branch assigned to the different mutational signatures. Middle panel: Overlay of the clones represented in (A) onto an H+E stained liver section of patient PD37111 (n=39 microdissections). Colouring of clones is according to the proportion of mutations attributed to Sig. 4, linked to tobacco exposure (blue: low activity of Sig. 4, red: high activity of Sig. 4). Right panel: Representative mutation spectrum for samples with low (top) or high (bottom) burden of Sig. 4. The six substitution types are labelled across the top. Within each substitution type, the contribution from the trinucleotide context are shown as 16 bars. The 16 bars are divided into four sets of four bars, grouped by whether an A, C, G or T respectively is 5’ to the mutated base, and within each group of four by whether A, C, G or T is 3’ to the mutated base. (B) Overlay of mutational signatures onto phylogenetic tree of clones in patient PD37107 (n=41 microdissections). Colouring of clones in the middle panel is according to Sig. 22, linked to the aristolochic acid carcinogen. (C) Overlay of mutational signatures onto phylogenetic tree of clones in patient PD36714 (n=35 microdissections). Colouring of clones in middle panel is according to Sig. 24, linked to the carcinogen aflatoxin-B₁. (D) Overlay of mutational signatures onto phylogenetic tree of clones in patient PD37113 (n=37 microdissections). Cluster 10 has many mutations attributed to Sig. 9, linked to the somatic hypermutation process in B lymphocytes.