Ecological genetics, the study of polymorphism and fitness in natural populations, has been revitalised through the application of next-generation sequencing technology to open up what were previously treated as genetic black boxes4,5. Growing appreciation of the loci and developmental networks that generate adaptive phenotypic variation6 promises to answer fundamental questions about the genetic architecture of adaptation, such as the prevalence of genomic hotspots for adaptation7, the relative contributions of major- and minor-effect mutations8, and the structural nature and mode of action of beneficial mutations9. Characterizing the identity and origin of functional sequence polymorphisms provides the explicit link between the mutation process and natural selection. In this context, while industrial melanism in the peppered moth has retained its appeal as a graphic example of the spread of a novel mutant rendered favourable by a major change in the environment, the crucial piece of the puzzle that has been missing is the molecular identity of the causal mutation(s)10.

A combined linkage and association mapping approach previously localized the carbonaria locus to a <400-kb region orthologous to Bombyx mori chromosome 17 (loci b–d)2. Thirteen genes and two microRNAs occur within this interval, none of which was known to be involved in wing pattern development or melanization. By extending the association mapping approach to a larger population sample and more closely spaced genetic markers (see Methods), we narrowed the carbonaria candidate region to about 100 kb (Fig. 1a). The candidate region resides entirely within the span of one gene — the orthologue of Drosophila cortex (cort), the only known function of which is as a cell-cycle regulator during meiosis11. In B. betularia, cortex consists of eight non-first exons, multiple alternative first exons (of which only two, 1A and 1B, are strongly expressed in developing wing discs), and a very large first intron (Fig. 1b).

Figure 1: The carbonaria candidate region, and the position and structure of the carbonaria mutation. a, Approximately 400-kb candidate region (bounded by marker loci b and d (ref. 2)) indicating gene content and genotyping positions (vertical lines in the continuous grey bar). Intron–exon structure and orientation are illustrated separately for each gene (annotated in GenBank accession KT182637. b, Refined candidate region including candidate polymorphisms (lines on the grey bar). The intron–exon structure of cortex is shown for carbonaria (black moth) and typica (speckled moth), highlighting the presence of a large (22 kb) indel (orange) within the first intron. Exons 1A and 1B are alternative transcription starts followed by the shared exons 2–9. c, The only exclusive carbonaria–typica polymorphism within the candidate region. The structure of the insert, shown in the carbonaria sequence, corresponds to a class II DNA transposon, with direct repeats resulting from target site duplication (black nucleotides) next to inverted repeats (red nucleotides). Typica haplotypes (lower sequence) lack the 4-base target site duplication, the inverted repeats and the core insert sequence. The transposon consists of ~9 kb tandemly repeated two and one-third times (repeat unit (RU)1–RU3), with three short tandem subrepeat units (green dots, SRU1–SRU9) within each repeat unit. Moth images were created from photographs taken by A.E.v.H. Full size image Download PowerPoint slide

The rapid spread of carbonaria gave rise to strong linkage disequilibrium2, such that many sequence variants are associated with the carbonaria phenotype. This poses a challenge for isolating the specific causal variant(s). We reasoned that if the carbonaria mutation arose on an ancestral typica haplotype2, the hitchhiking variants should in principle also be present at some frequency within the typica population, leaving the causal variants as the only ones unique to carbonaria. High-quality contiguous reference sequences were assembled from tiled bacterial artificial chromosome (BAC) and fosmid clones, resulting in one carbonaria and three different typica core haplotypes (see Methods and Extended Data Fig. 1). Alignment of these sequences (Supplementary Data 1) revealed 87 melanization candidate polymorphisms (Fig. 1b and Supplementary Table 1), concentrated within the large first intron of cortex (69–91 kb, depending on haplotype). Eighty-five candidates were eliminated using an increasing number of typica individuals to exclude rare variants. A single nucleotide polymorphism (carbonaria_candidate_25) was eventually excluded on the basis of one individual out of 283 typica, leaving a very large insert (carbonaria_candidate_45) as the only remaining candidate.

The insert was found to be present in 105 out of 110 fully black moths (wild caught in the UK since 2002) and absent in all (283) typica tested (see Methods and Extended Data Fig. 2). Consistent with local carbonaria morph frequencies of 10–30% (ref. 12), 2 out of 105 individuals were homozygous for the carbonaria insert. Five individuals that were morphologically indistinguishable from carbonaria did not possess the carbonaria insert; they do not present any strong haplotype association based on this set of candidate loci but do all differ from the core carbonaria haplotype at many positions. Our interpretation is that these individuals are hetero- or homozygous for the most extreme of the insularia alleles (intermediate phenotypes), which are known to occasionally produce carbonaria-like phenotypes13,14 and segregate as alleles of the carbonaria locus in classical genetics crosses14. Conversely, none of the genotyped insularia morphs (31 individuals, covering the full spectrum of variation from i 1 to i 3 (ref. 14)) contains the carbonaria insert (Extended Data Fig. 2). We conclude that the large insert is the carbonaria mutation.

The carbonaria insert is 21,925 nucleotides long and is composed of a roughly 9-kb essentially non-repetitive sequence (except for approximately 370 nucleotides at the repeat unit junctions) that is tandemly repeated approximately two and one-third times, with only minor differences among the repeats (Fig. 1c). The insert bears the hallmark of a class II (DNA cut-and-paste) transposable element: short inverted repeats (6 bp) and duplication of the (4-bp) target site present in typica haplotypes (Extended Data Fig. 3). We estimate that there are approximately 255 and 60 genomic copies, respectively, of the 9-kb carbonaria transposable element (carb-TE) repeat unit and repeat unit junctions, implying that there are relatively few genomic copies of the complete carb-TE. No nucleotide or translated BLAST hits were found in any relevant database, with the exception of B. betularia RNA-sequencing (RNA-seq) reads (NCBI: SRX371328), indicating that the carb-TE repeat unit is Biston-specific.

To examine patterns of recombination, which provide insight into the evolutionary dynamics of a chromosomal region, we genotyped the same 105 carbonaria and a sub-set of 37 typica, plus 35 insularia, at 119 polymorphic loci within 28 PCR fragments distributed across 200 kb either side of the carb-TE (Fig. 1a). Diploid genotypes were phased, and the resulting haplotypes divided into those with and without the carb-TE. The sequence identity of the ancestral carbonaria haplotype, whose core was known from the BAC and fosmid work, was extended by assigning allelic state at each marker locus to ancestral carbonaria or typica/insularia. Fifty per cent of carb-TE haplotypes had retained the ancestral carbonaria haplotype across the full 400-kb window, and the remainder showed varying degrees of recombination with typica haplotypes on one or both sides of the causal mutation (Fig. 2a). The recent selective sweep15 is reflected by declining linkage disequilibrium between the carbonaria locus and marker loci with increasing genetic distance (Fig. 2b). The tenure of the carb-TE has been transient, having declined from ~99% to less than 5% in its industrial heartland since 1970 (ref. 16). It has nevertheless left a substantial trace of its former abundance in the form of ancestral carbonaria haplotype blocks introgressed into typica and insularia haplotypes, consistent with the simulation-based expectation (Fig. 2c).

Figure 2: Recombination pattern and ageing of the carb-TE mutation. a, Nearest recombination between carbonaria (carb-TE present (orange)) and non-carbonaria (typica and insularia (light grey)) haplotypes (n = 107), 200 kb either side of the carb-TE (at position 0). Dark grey areas indicate boundaries within which recombination occurred. b, Multilocus linkage disequilibrium (r d ) across the same sequence window among carbonaria and non-carbonaria haplotypes. Grey area indicates the widest 99% confidence region, across loci, for the null hypothesis (r d ≈ 0). Red lines represent the simulation-based upper bound under the extreme assumption that all alleles defining the carbonaria haplotype were initially exclusive to it (mean and 90% interval). c, Introgression of the ancestral carbonaria haplotype (black) into non-carbonaria haplotypes (grey; carb-TE absent (n = 144)). Red lines represent the simulation-based expectations (mean and 90% interval). d, Probability density for the age of the carb-TE mutation inferred from the recombination pattern in the carbonaria haplotypes (maximum density at 1819 shown by dotted line; first record of carbonaria in 1848 shown by dashed line). Full size image Download PowerPoint slide

The first reported sighting of the carbonaria form is generally regarded as having occurred in 1848 in Manchester1, although the wording of the record implies that it was rare but not completely unknown at this time. Establishing how long before this date the carbonaria mutation occurred is complicated because it could have existed undetected at a low frequency for hundreds of years (Supplementary Methods). Our approach to this problem was to infer the age of the mutation event independently by considering the erosion of the ancestral carbonaria haplotype due to genetic recombination and mutation. One million simulated time trajectories of the carbonaria phenotype were randomly drawn according to their fit to historical frequency data (Extended Data Fig. 4). Based on these trajectories, recombination patterns were simulated using an empirical estimate of recombination rate and compared to the observed recombination pattern of the carbonaria haplotypes. The probability density for the date of the carb-TE mutation event (Fig. 2d) is highly skewed (median, 1763; interquartile range, 1681–1806) with a maximum likelihood at 1819, a date highly consistent with a detectable frequency being achieved in the mid-1840s.

The position of the carb-TE suggests that its effect on melanization is achieved by altering the expression of cortex through one of several potential mechanisms17 (incorporation of any part of the carb-TE into cortex transcripts has been excluded). Biston cortex is characterized by numerous splice isoforms and alternative first exons; we focus on the population of transcripts initiated by exons 1A and 1B, as the other first exons are absent or only weakly expressed in Biston wing discs, and did not exhibit morph-specific differences (Extended Data Fig. 5). The global pattern of splice isoforms showed neither consistent presence or absence nor crude relative abundance differences among morphs for any developmental stage (Extended Data Figs 6, 7). Cumulative expression across all isoforms (Fig. 3a) increases by an order of magnitude between the sixth larval instar (La6) and day 4 prepupa (Cr4), coinciding with a phase of rapid wing disc morphogenesis (Fig. 3b), and falls back to a low level by day 6 prepupa (Cr6) with no clear difference among morphs (t/t versus c/t, P > 0.5). To exclude interference by potentially non-functional isoforms, we targeted full transcripts only, starting with either 1A or 1B. The abundance of the 1B full transcript shows a consistent trend across several families with different genetic backgrounds (c/c > c/t > t/t) that is most pronounced at Cr4 (Fig. 3c and Extended Data Fig. 8a). The abundance of the 1A-initiated full transcript, which is in general an order of magnitude less than that of the 1B transcript, does not show a significant difference between genotypes (Fig. 3d and Extended Data Fig. 8b).

Figure 3: Relative expression of cortex in developing wings of B. betularia. a, Average expression (across typica and carbonaria morphs) of all cortex splice variants (exons 7–9) relative to the control gene α-Spec in wing discs at different developmental stages (La6, sixth instar larvae; Cr2, day 2 crawler; Pu2, day 2 pupae; PDP, post-diapause pupae). Bars are s.e.m. b, Scaled images (created from photographs taken by I.J.S.) of B. betularia forewings at different stages. c, d, Tukey plots for relative expression of cortex 1B (c) and 1A (d) full transcripts in developing wings of the three carbonaria-locus genotypes (c/c, c/t and t/t) produced within the progeny of a c/t × c/t cross (no data for c/c at Cr2). Genotypes differ significantly for the 1B full transcript (P < 0.001, generalized linear model (GLM)), whereas genotypes do not differ for the 1A full transcript (P > 0.2, GLM). (Note the differing y-axes scales.) Equivalent graphs for the progeny of c/t × t/t crosses (which lack the c/c genotype) are presented in Extended Data Fig. 8. Full size image Download PowerPoint slide

The role of cortex in wing pattern melanization is not obvious. In Drosophila, cortex has been primarily associated with meiosis in ovaries11 (several cortex transcripts are expressed in B.betularia ovaries and testes; Extended Data Fig. 5). The molecular function of cortex is suggested by phylogenetic analysis, which indicates that Biston cortex occurs in a lepidopteran sub-group within an insect-specific clade of a protein family containing the cell-cycle regulators cdc20 and cdh1, encoded by fzy and rap (also known as fzr) in Drosophila (Extended Data Fig. 9b). These proteins help to regulate fundamental cell division processes such as cytokinesis by presenting substrates to, and activating, the anaphase-promoting complex or cyclosome (APC/C), which ubiquitinates cell-cycle proteins, thereby earmarking them for degradation. Substrate recognition occurs by binding to degrons (short linear motifs such as the D box and KEN box). Sequence conservation across lepidopterans and non-lepidopterans reveals a single binding site in cortex (Extended Data Fig. 9c) that probably binds the D box-like18 degron LXEXXXN19. This degron binding capability is predicted for both of the full isoforms (1A (441 amino acids) and 1B (407 amino acids), although 1B apparently lacks the N-terminal C box that is usually required for APC/C binding) but not for the alternative isoforms (Extended Data Table 1). These data demonstrate orthology and are consistent with shared function of cortex between D. melanogaster and B. betularia, although the molecular connection between cell-cycle protein degradation at the APC/C and melanization remains to be determined.

Our results suggest that the carb-TE influences adult melanization pattern by increasing the abundance of cortex, perhaps by altering the course of scale-cell heterochrony, with dominance arising through a threshold effect (the 1B full transcript is more abundant in c/c than c/t). How the carb-TE promotes cortex expression is unknown but the general mechanism is predicted to allow the production of insularia morphs that are putatively controlled by different mutations within cortex. In combination with parallel findings in Heliconius butterflies20, our results support the idea that cortex is a conserved developmental node for generating colour pattern variation in evolutionarily diverse Lepidoptera. However, cortex may not be the only gene in this region involved in patterning, as suggested by recent work on the B. mori mutant Black moth, which has a similar phenotype to B. betularia carbonaria21, although none of the genes implicated is differentially expressed among carbonaria and typica wing discs.

The carb-TE is a spectacular example of an adaptively advantageous transposon22,23,24; its discovery fills a fundamental gap in the peppered moth story and furthers our appreciation of the mechanism underpinning rapid adaptation. A consensus on the general importance of transposable elements for adaptive evolution has yet to emerge3,25. Over longer time frames, phenotypic effects of transposable elements may be obscured by imprecise excision that leaves a minimal trace of the transposable element while retaining the mutant (adaptive) phenotype26. By contrast, we have shown that the carb-TE is young, approximately 200 years (generations) old, during which time it has gone from a single mutation to near fixation (regionally) to near extinction—driven by a pulse of environmental change.