A whole-body donor-based approach to study mutational clonality
On average 101 bodies are voluntarily donated to the Ghent University Anatomy and Embryology Unit each year (data 2016–2021; range 82–133). The median age of these whole-body donors is 81 years (range 36–106 years; Fig. 1a) and the majority (53%) deceased between 12 and 36 hours prior to arrival (post-mortem interval; PMI), with 92% of all donors having a PMI below 72 hours (Fig. 1b). They are primarily used for medical and surgical training purposes22,23, but also provide a rich normal tissue resource to study clonal changes in different organs of the human body.
To demonstrate the feasibility of such a post-mortem tissue-based mutational clonality detection platform, we developed a 4-step methodology (Fig. S1a). In a first step, skin and oral tissues were sampled and relevant clinical information (i.e., age, gender, smoking status, oncological history) as well as the PMI was recorded by the responsible physician. The choice for skin tissues for initial evaluation purposes was motivated by their easy accessibility, high expected UV-induced mutation rates (with negative controls from unexposed regions) and knowledge of the expected (positive control) results from earlier studies4. Secondly, 5 mm (diameter) punch biopsies were taken (surface area 19.63 mm2), and the epithelial layer was enzymatically isolated. In a third step DNA was extracted, DNA concentrations and integrity were determined, and samples were sequenced. Importantly, we did not observe any PMI-dependent DNA degradation, as evaluated using gel electrophoresis (eyelid samples in 10 subjects with PMI between 6h and 83h evaluated, Fig. S1b). In a fourth step, somatic mutations were called, followed by an analysis of clone sizes, mutational signatures, and positive selection signals.
We aimed to validate our approach using 2 largely orthogonal approaches: 1) by identifying positive selection of somatic mutations in cancer genes and 2) by demonstrating UV-specific alterations in UV-exposed skin. To achieve this, we sampled skin and oral tissues from 2 different donors: a 94-year-old, male, non-smoking donor (subject PM01, PMI 33h) and a 90-year-old, male donor (subject PM02, PMI 38h) with a smoking history (1 package a day between the age of 18 and 38, as derived from the medical record). To explore the effects of different expected lifetime UV exposure, skin samples were taken at 3 locations: bridge of the nose (high UV exposure), upper eyelid (intermediate UV exposure) and gluteal region (no/low UV exposure). Additional non-UV-exposed oral mucosa samples were taken from the inner buccal region (Fig. 1c).
Somatic mutations in known driver genes are detectable in post-mortem tissues
Deep (1000x) targeted sequenced was performed on 153 genes. Apart from 76 earlier defined skin cancer genes, this gene panel also contained 1) 25 driver genes that have been associated to skin or head and neck cancer, 2) 20 housekeeping control genes and 3) 32 genes that are putatively involved in immune evasion (table S1). After alignment, somatic mutations were called using Shearwater ML, an algorithm that is optimized for the detection of low frequency variants from deep targeted sequencing data. We retrieved 910 somatic mutations with an average variant allele frequency (VAF) of 0.011 and ranging between 0.0023 and 0.11 (Table S1).
The 5 most frequently mutated genes were NOTCH1 (6.6 average mutations per sample; mps), TP53 (3.1 mps), MUC17 (2.5 mps), FAT1 (2.2 mps) and APOB (2.1 mps). Most samples contained multiple mutations in these genes, and remarkably, NOTCH1 mutations were identified in all samples (between 1 and 16 different NOTCH1 mutations per sample). Overall, we identified somatic mutations in 119 different genes, including NOTCH2 (1.6 mps, 6th most frequently mutated gene) and NOTCH3 (0.82 mps), genes that have been previously identified as clonal drivers in healthy skin (Fig. 2)
Most of these mutations were observed in the 101 driver genes (93%, 844 mutations). The majority of the remaining mutations occurred in the 32 immune genes (46 mutations, 5%), while 20 mutations (2%) were observed in the 20 housekeeping control genes (Fig. S2, table S1).
Clonal alterations in epidermal skin and oral mucosa are primarily driven by NOTCH1 and TP53
The large difference between the number of mutations observed in driver versus housekeeping genes suggests positive selection acting on the former. To confirm this, we first focused on the global ratio of nonsynonymous over synonymous mutations. In skin samples, this ratio was higher than the expected ratio, as derived from a random mutation background model (3.68 versus 2.12 respectively; P = 3.7e-08). This higher observed to expected mutation ratio (i.e., global dN/dS = 1.73) suggests that 42% (0.73/1.73) of the identified nonsynonymous mutations have been subjected to positive selection forces (Fig. 3a). Significant selection signals (i.e., dN/dS values above 1) were solely observed in the group of driver genes (dN/dS = 1.78, P = 2.4e-08), and not in the immune genes (dN/dS = 0.81, P = 0.72) nor the housekeeping genes (dN/dS = 0.65, P = 0.81; Fig. 3b). Additionally, signals were stronger for nonsense mutations (dNons/dS = 3.04, P = 3.9e-09) than for missense mutations (dMiss/dS = 1.80, P = 6.3-09) and were also present in the oral samples (dN/dS = 2.13, P = 1.3e-04; Fig. 3a-b, Table S1). When focusing on single genes in skin, we observed significant selection signals (at 10% FDR) in NOTCH1, TP53 and FAT1 (Fig. 3a, Fig. S3).
Because selection pressures are mainly expected on nonsynonymous mutations with high mutational impact, we also compared PolyPhen-2 (PP2) mutational impact scores between observed and expected mutations. Higher PP2 scores were observed for our set of somatic mutations (median PP2 = 0.50) as compared to the expected scores (median PP2 = 0.11; P = 2.3e-09; Fig. 3c). Like the dN/dS approach, this difference was only observed for driver genes (P = 3.5e-07) and not for the other gene sets (Fig. 3d). We confirmed positive selection signals (at 10% FDR) in skin tissues in NOTCH1, TP53 and FAT1 and, additionally, also found significant signals in NOTCH2, CDKN2A, BCORL1 and AJUBA (Fig. 3c, Fig. S3). In oral tissues, positive selection was detected in NOTCH1, FAT1, TP53 and NOTCH2 (Fig. 3c).
Our results confirm earlier reports on somatic driver mutation clonality in healthy skin. Based on the variant allele frequency and biopsy size, we imputed the clone sizes for the genes that were identified to be under positive selection and found 78 clones per cm2 skin (35 NOTCH1, 16 TP53, 15 FAT, 6 NOTCH2, 2 BCORL1, 2 CDKN2A and 2 AJUBA clones; Fig. 4a, b). BCORL1-driven clones had the largest size (median 0.49 mm2) while the smallest clone sizes were found for FAT1 (0.23 mm2; Fig. 4a, c). These clone sizes were higher than described previously4, although the total percentage of skin occupied by the most abundant clones was largely comparable between both studies, with 15.2% of skin cells estimated to contain NOTCH1 mutations, 5.7% TP53 mutations, 4.7% FAT1 mutations and 2.1% NOTCH2 mutations (Fig. 4d). In the oral mucosa, similar clonal densities were observed for the donor with the smoking history, but not for the non-smoking donor, where clones were relatively scarse (Fig. S4).
UV-specific genomic alterations in epidermal skin correlate to the expected amount of lifetime UV exposure
As an additional validation of our approach, we aimed to detect UV-induced genomic changes in post-mortem tissues derived from whole-body donors. The substitutions in the 12 skin samples were predominated by C > T mutations (56%; Fig. 2, Fig. 5a). Contrary to the other 5 substitution types and as expected for UV-induced somatic mutations, these C > T mutations mainly occurred in a dipyrimidine context (88%) and were 24% more prevalent in the coding strand than the template strand (P = 0.054, exact Poisson test), suggesting transcriptional strand bias (Fig. 5a). This mutational pattern is consistent with previous UV exposure, which was further confirmed by the strong predominance (82.4%) of coding strand-specific CC > TT dinucleotide variants (DNVs; P = 3.4e-10; Fig. 5b) and similarity of the samples’ 96 trinucleotide substitution type signature (i.e., substitution type and adjacent base pairs) with the well-known UV-associated single base substitution signature (SBS) 7 (Fig. 5c). This signature was found in all skin samples (mean contribution 39.3% per sample) and was the most prevalent signature in 10/12 skin samples. Interestingly, the smoking-related signature 4 was retrieved in 7/11 samples from subject PM02, which was known to have a smoking history (Fig. 5c).
To exclude that this putative UV signal was biased by the rather limited genomic coverage of our gene panel and/or positive selection processes, we extended our analysis to known UV hotspots in the promoter regions of RPL13A and DPH3. Here, C > T mutations occurring in a CCTTCCGG sequence context have previously been associated with the amount of cumulative UV-exposure due to ETS transcription factor binding24,25. We sequenced these regions using an error-correcting amplicon sequencing protocol (SiMSen-Seq) and detected UV-specific mutations in the UV-exposed eyelid and nose samples, but not in the non-exposed gluteal or oral samples (Fig. 5d, Table S1). Further, the variant allele frequency (VAF) was significantly higher in nose samples than eyelid samples, in line with the higher expected lifetime UV exposure of the former (Fig. 5e). These results confirm that low frequency UV-induced mutations can be accurately detected in post-mortem skin tissues.