Two male-specific scale types with divergent ultrastructures

The sexual dimorphism of C. eurytheme butterflies manifests on the dorsal surface of their wings, where males show bright UV iridescence across the medial orange region, as well as a thin, continuous marginal black band (Fig 1A). There is no UV-iridescence in wild-type females, and their marginal patterns are wider, jagged at their interface with the orange area, and flecked by yellow spots (Fig 1A). The two sexes thus differ in UV-iridescence as well as in the morphology of the marginal band. We used Scanning Electron Microscopy to further survey the ultrastructure of scale types in C. eurytheme (Fig 1B–1H). Typical scales that are found on either sex or wing surface share a 2-μm distance between their apical ridges (2.0 ± 0.1 μm, N = 125 measured scales), a feature that is relatively constant across melanic scales (e.g., forewing discal spots of both sexes), pterin-pigmented scales (orange, yellow), and pterin-deficient scales (e.g., white scales from Alba female morphs). In contrast, UVI scales are not only characterized by the multilayering of their ridge lamellae [23,29,44], but also by the density of the ridges themselves, with a distance between ridges averaging only 1 μm (1.0 ± 0.1 μm, N = 25 scales) (Fig 1D, 1G). A second type of male-specific scale type, here dubbed the spatulate scales, show a large inter-ridge distance of 4.4 ± 0.5 μm (N = 20 scales) that gives them a corrugated look under light microscopy. These scales have an unusual ultrastructure, with crossribs that resemble soybean pods joining the longitudinal ridges (Fig 1D, 1H), and that overlay a porous inner matrix [45]. It was proposed that these scales play a role in pheromone retention or spreading, due to their male-specificity and the spongy aspect of their ultrastructure [27,46,47]. Spatulate scales occur specifically on the cover scale layer of the male melanic bands, where their wide apical lobes overlay a layer of yellow ground cover scales. In females, the dorsal marginal band is devoid of spatulate scales and shows instead canonical melanic scales as cover scales. In summary (Fig 1I, 1J), the dorsal surface of C. eurytheme wings includes two male-specific scale types that each feature unique ultrastructures—the UVI scale and spatulate scales.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Male-specific scales in the dorsal wings of Colias eurytheme.

A. Dorsal views of male and female butterflies, highlighting the difference in width and patterning of the marginal bands, and the male-specific UV-iridescence (magenta; an UV-photograph was colored in magenta and overlaid on a normal image). Alba morphs are a female-limited phenotype characterized by pterin-pigment deficiency in their scales. B, C. Scanning electron micrographs of representative non-dimorphic scales (here in B, left: female dorsal forewing melanic scales of the discal spot; right: male ventral orange scales), and male-specific scales (here in C: interface between UVI and spatulate scales, dorsal hindwing). D. Mean apical ridge distance in individual scales from melanic discal spots, medial regions, or male marginal bands, highlighting the derived ultrastructure of male-specific UVI and spatulate scales. Each category features N = 25 scales—except for spatulate scales (N = 20). The measurements originate from electron micrographs of dorsal wing surfaces in a total of three individuals: an orange female, an Alba female, and a male (Supplementary Methods in S1 Text and Data 10 at https://osf.io/yjvkc/). E–H. Apical views of the ultrastructure of representative scale types (same scale types as in panels B, C), featuring the longitudinal ridges (horizontal structures) and the transversal microribs (vertical). E: female dorsal cover black scale; F: male ventral cover scale (orange, non-UVI); G: male UVI dorsal cover orange scale; H: male spatulate dorsal cover margin scale. I. Position of male-specific cell types on the dorsal wings. The cell bodies of scale precursors (blue) are transient structures that do not occur in adults. J. Schematic representation of the ultrastructures (viewed as transversal cross-sections) from the four main scale types—canonical (orange and black), UVI, and spatulate. Scale bars; B and C = 50 μm; D = 1 mm; E–H = 2 μm.

https://doi.org/10.1371/journal.pbio.3003233.g001

Doublesex has distinct functions on marginal patterning and UV dichromatism

Gene loss-of-function assays targeting Dsx generate intersexual phenotypes across insects, with secondary sexual characteristics losing their sex-specific differentiation, and sometimes reversing to the state found in the opposite sex [9,48–53]. We thus used CRISPR-targeted mutagenesis to generate G₀ adults carrying Dsx mosaic knock-outs (abbreviated mKO, or “crispants”), using single sgRNAs that targeted either the DNA-binding Domain or the Dimerization Domain shared across Dsx isoforms [6], resulting in deletion alleles at the target sites (S1 Fig). Dsx crispant adults showed mosaic effects on sexually dimorphic features of the dorsal wings (Fig 2, S1 Table). Of note, G₀ crispant wings are mosaics of tissues carrying Dsx-deficient and wild-type cells, and while 33 crispant individuals showed spatially variegated effects as a result of this mosaicism, these experiments yielded consistent effects within each sex (S2, S3 Figs). Dsx crispants from both sexes showed intermediate intersexual states in the aspect of the marginal bands, the melanic patterns at the distal edge of their wings. The marginal bands of female crispants showed a narrowing of the melanic marginal band compared to wild-type females, resulting in a loss of yellow spots. The reverse was observed in the marginal bands from male crispants, with a partial expansion of the melanic band resembling the female state.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. DsxF blocks UV-iridescence in females in a cell-autonomous fashion.

A. Effects of Dsx mosaic knock-outs (mKO), as seen in comparisons of female and male WT individuals with representative Dsx G₀ crispants. The sex of each crispant was determined by concordant genital morphology and genotyping. Phenotypic effects are limited to the dorsal surfaces and most visible in the marginal region. B. UV-photography (320–400 nm) of UV-iridescent dorsal patterns in the same mutants, showing an ectopic gain of UVI scales in females (top). In males (bottom), the feminization of the marginal region triggers a regression of the male UVI pattern distal border. C–C′. Close-up views of the central forewing regions (same individuals as in panels A, B), including overlays of UV photographs (C′: magenta false-color) over visible light images, and highlighting the intermediate states of marginal patterns. In females: gain of UVI scales, regression of the melanic marginal patterns. In males: gain of female yellow spots in males (arrowheads), extension of the melanic marginal patterns (white arrow), and regression of the UVI field (black arrow). D. Cell autonomy of UVI scale gains in female Dsx mKOs, as shown by continuous UVI mutant clones with sharp boundaries (KO). Superimposed views of the same region, taken in visible and ultraviolet light, are shown across each diagonal line. E. RNA interference effects of Dsx siRNA electroporation in male and female forewings. Dotted lines mark the approximate areas that were electroporated. Dsx knockdown results in ectopic UVI (top), and in a feminization of the marginal pattern in males, including with a regression of medial UV iridescence at its distal border. F. High-magnification view of the female wing shown in panel E, at the interface of the treated (ectopic UVI scales) and untreated area. Scale bars: C, E = 1 mm; D = 100 μm.

https://doi.org/10.1371/journal.pbio.3003233.g002

The effects of Dsx mKOs on UV-iridescence indicated two distinct categories of effects between males and females. Female crispants showed a gain of UVI that phenocopies the male state (Fig 2A–2C′). This conversion towards UVI scale types occurred along sharp clone boundaries (Fig 2D), showing that DsxF is necessary for the repression of UVI in cell autonomous fashion. In contrast, while male crispants show a spatial reduction of the UVI field (Fig 2A, 2B), neither these UVI-loss effects showed sharp clonal boundaries that would indicate autonomy, nor did these effects extend into the proximal and central sections of the wing (S2 Fig). This indicates a shift in the positioning of pattern boundaries in the vicinity of the marginal region (Fig 2C–2C′), likely due to non-autonomous effects on morphogenetic signaling in this region. In addition, we introduced short-interfering RNAs (siRNAs) via electroporation to drive RNAi knockdowns of Dsx at the pupal stage [10,54]. UVI scales were unaffected throughout the central domain of the RNAi-treated male dorsal forewings (Fig 2E). Consistent with the mosaic knock-outs, female wings electroporated with Dsx siRNA showed ectopic UV-iridescence in the cover scales (Fig 2E, 2F). To explain these non-autonomous effects in males, we extrapolate that the spatial patterning of the marginal region includes sex-specific inputs on morphogenetic signaling events, that in turn determine its width and sub-division into fields of melanic versus non-UVI yellow scales.

DsxF activates Bab expression in female dorsal cover scales, repressing UVI scale identity

Both Dsx KOs and knock-downs result in ectopic UV iridescence across the female medial region, and conversions to the UVI scale type are restricted to cover scales (Fig 3A, 3B). This cover scale specificity is unlike the effect of Bab KOs [29], which converts both the ground and cover scale layers to a UVI state. Next, we profiled the expression of Dsx between 30%–40% of pupal development—a temporal window where Bab becomes downregulated specifically in male UVI scale cells. To do this, we used a monoclonal antibody that recognizes the Dsx-DBD (DNA-binding domain) shared by both isoforms [55–57]. Because Dsx isoforms are sex-specific in Lepidoptera, for simplicity, we call the detected antigen DsxF or DsxM based on the sex of the dissected pupae. In females, DsxF is detected in both cover and ground scales throughout the entire wing (Fig 3C), like Bab (Fig 3C′). In contrast, DsxM is only expressed in the dorsal cover scales of the wing margin in male wings (Fig 3D–3D′). This absence of DsxM in the medial region explains why its perturbation does not result in autonomous loss of UVI scales in males.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. DsxF medial expression activates Bab in the female dorsal cover scales.

A, B. High-magnification views of central dorsal wings, showing dorsal cover-scale (dcs) specific transformation of non-UVI to UVI scales in female Dsx KOs. Males are unaffected by Dsx mKOs in the central region. Dashed lines indicate small patches of the wing where ground scales (gs) are exposed. C–D′. Immunofluorescent detection of the DsxDBD antigen in the dorsal layer of female and male wings between at 40% of development. The Bab antigen (green) marks both dorsal cover scales (dcs) and ground scales (gs) in females (C′), and only the ground scales in males. Forewings are shown in C′ and D′ insets. E, F. Immunofluorescence of Bab (green) in the control (E, contralateral left wing, horizontally flipped for re-orientation) and treated (F) female forewings following electroporation with Dsx siRNAs, shown at the 40% pupal stage. DAPI DNA stainings (blue) highlight rows of scale nuclei interspersed by smaller epithelial nuclei. Dsx RNAi results in loss of Bab expression in dorsal cover scales (dcs, dotted lines), while maintaining Bab expression in ground scales (gs, arrowheads). Scale bars: A, B = 100 μm; C, D = 1 mm; C′, D′, E, F = 10 μm.

https://doi.org/10.1371/journal.pbio.3003233.g003

Next, we tested the regulatory interaction between DsxF and Bab by examining Bab expression in female wings knocked-down for Dsx. While the contralateral, untreated controls from the same individual show expression of Bab in both dorsal ground and cover scales at the 40% stage, Bab expression is low or undetectable in dorsal cover scales in the wings electroporated with Dsx RNAi, taking an alternating configuration (Fig 3E, 3F). In other words, DsxF is required to maintain Bab expression at high levels in the female dorsal scale, thus feminizing the wing phenotype by preventing UVI scale differentiation.

Dsx controls the sex-dependent identity of margin cover scales

While Dsx is expressed in all the scale cell types of the marginal region in females, it marks dorsal cover scales in the marginal region of males. Marginal dorsal scale cells are sexually dimorphic, with a canonical melanic type in females versus the derived spatulate type in males. Accordingly, Dsx CRISPR KOs showed complete reciprocal transformation of the melanic dorsal cover scales between these marginal/dorsal scale types between sexes, from canonical melanic to spatulate in DsxF mKOs, and vice-versa from spatulate to canonical in DsxM crispants (Figs 4A–4C and S4). Thus, while only DsxF is required for specifying the UV dichromatism, both Dsx isoforms are required to instruct the correct patterning of marginal patterns and the identity of male-specific scales within them.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Doublesex controls the specification of male-specific spatulate scales.

A. Magnified views of the dorsal hindwing marginal bands of mosaic Dsx G₀ crispants. KO clone boundaries are visible with yellow/orange color shifts in the medial region, and in the marginal band, with shifts from canonical black scales to the spatulate scales in females, and vice-versa in males. B, C. Complete, reciprocal shifts in scale composition (B) and melanic scale identities (C) in both female and male Dsx crispants. Scale bars: A–C = 100 μm.

https://doi.org/10.1371/journal.pbio.3003233.g004

The combinatorial logic of dimorphic scale type specification

We summarize below how Dsx controls multiple aspects of sexual dimorphism on the dorsal wing (Fig 5A). The spatial expression of Dsx is sexually dimorphic at the 30%–40% pupal stage: female wings express DsxF in all scales, while males only express DsxM in the dorsal cover scales of the marginal region. Dsx influences the patterning of the marginal band via non-autonomous effects, determining the spatial extent of the melanic band and of non-UVI yellow outlines and spots. The differences in spatial localization of Dsx sexual isoforms likely explain the asymmetric effects of its perturbation in each sex. In the medial region, DsxF is required to feminize dorsal cover scales, by repressing UVI states. In the wing margin, sex-specific Dsx isoforms are required to feminize dorsal cover scales into a canonical melanic type or to masculinize them into the spatulate type.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Dsx and Bab control sexually dimorphic scale fates.

A. Working hypothesis for the effects of Dsx on sexually dimorphic wing traits, including cell non-autonomous patterning effects on marginal patterns, cell-autonomous requirements in the specification of spatulate vs. canonical margin scales, and female-specific repression of UV-iridescence in dorsal cover scales. B–D. Bab RNAi knockdowns on the dorsal forewing phenocopy mosaic KO effects [29]. Bab-expressing cells acquire a UVI state upon Bab perturbation, with the exception of the spatulate scales that are unaffected. This includes canonical melanic scales that acquire UV-iridescence while maintaining melanism (D, cyan arrowheads), resulting in a dark blue iridescent phenotype in the visible spectrum (panel D), as previously described in Bab mosaic KO experiments [29]. Ground scales are transformed to yellow UVI scales in each sex (e.g., in the medial region: C, white arrowheads), except in the female forewing marginal band where they convert from melanic to blue iridescent (combination of melanic and UVI features). Star: inset features yellow vein scales (UV-negative in WT and controls) that acquire a UVI fate upon Bab RNAi, indicating successful knockdown in this area. E, F. Summary of Dsx expression, Bab expression, and perturbation assays in dorsal wing surfaces. Scale bars: B = 5 mm; C, D = 100 μm.

https://doi.org/10.1371/journal.pbio.3003233.g005

Next, we developed a phenomenological model that integrates the effects of Dsx and Bab on male-specific cell type specification. We replicated the effects of Bab mKOs [29] using RNAi knockdowns targeting the dorsal surface of the forewings from each sex, and summarized the observations from both sets of perturbation experiments (Fig 5B–5D). In females, Bab-deficient cells all acquire a UVI-scale morphology including UV-iridescence and high ridge density. Because all the female scale cell precursors express Bab (S5 Fig), this is visible in both cover and ground scales of both the medial and marginal regions. In the marginal region, UVI scales retain a melanic state, and the combination of UVI-structural and melanic features confers them a dark blue iridescence, visible with the naked eye when illuminated with a low angle of incidence, as previously observed in female Bab mKOs [29]. In males, only ground scales express Bab, and accordingly, Bab perturbation results in ground scale-specific conversions to UVI states. These knockdown effects confirm that Bab inhibits UVI trait differentiation and that this occurs during pupal stages. Bab does not appear to control other aspects of sexual dimorphism, other than UVI.

Together, these data suggest a combinatorial logic for male-specific cell type specification (Fig 5A, 5E, 5F). Across the entire dorsal wing of each sex, ground scales express Bab and repress UVI without Dsx input. In the medial region, DsxF feminizes the dorsal cover scales by recruiting the UVI-repressing activity of Bab. In marginal regions, DsxF maintains canonical states, while DsxM is specifically expressed in the dorsal cover scales and masculinizes them into spatulate states.

Overview of cell type diversity in the 40% male pupal hindwing

Our analysis of Dsx shows that male wings differentiate two derived scale types with specialized ultrastructures, the UVI and spatulate scales. Next, we used single-nucleus RNA sequencing (snRNA-seq) to gain further insights into the molecular basis of this diversity of cell types in the male wing. We chose to sequence a hindwing from a C. eurytheme male individual at 40% pupal development, a stage where Bab is consistently repressed in the dorsal cover scales that give rise to the UVI state [29]. Quality control showed low-contamination of mitochondrial reads following filtering at the 3% threshold (S6 Fig), resulting in 2,961 filtered cells. Unsupervised clustering yielded nine robust clusters, as visualized here in UMAP reduced-dimensionality space (Fig 6A). We then used the FindMarkers function of Seurat version 5 to identify differentially expressed features between these clusters (Data 1 at https://osf.io/yjvkc/), and annotate them based on marker gene enrichment and the known function of their orthologs in Drosophila. Two minor clusters, dubbed Misc1 (N = 44 nuclei) and Misc2 (N = 35 nuclei), remain unannotated and will require further work for confident assignment of cell types within them (see Discussion). The seven remaining clusters consist of two epithelial cell types dubbed Wing_epi and Trch_epi, and five clusters related to SOP subtypes—namely one socket cluster (Socket) and 4 scale sub-types (Scale1 to Scale4), as detailed below. Of note, scale cell types showed higher numbers of mapped reads and detected genes compared to non-scale cell clusters (Fig 6B), likely due to differences in ploidy levels. Non-scale cell types also showed a higher level of mitochondrial DNA (mtDNA) contamination relative to scale types (Fig 6C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Single-nucleus transcriptomic profiles of major cell types in the Colias eurytheme male 40% pupal hindwing.

A. UMAP plot representing the 9 clusters resolved across a total of 2,961 cells. B–B′. Violin plots of the number of unique genes (B) and RNA read counts (B′) per nucleus. C–C′. Violin and heatmap plots highlighting differences in mtDNA content across non-scale and scale clusters. D–G. Heatmap plots showing the expression of key marker genes (bold: see main text for citations). Serpent and Rala (arrowheads) in two small unannotated clusters, Stubble (epithelial cells), neuromusculin (nrm, socket cells), rhophilin-2 (wing epithelium), breathless (tracheal epithelium), Notch (non-scales), and shaven (sv, scales). H–H′. Dot plots featuring the expression profiles of 150 differentially expressed genes, chosen among top markers of cluster or cell type identity (Data 1 at https://osf.io/yjvkc/). Panel H includes top markers for non-SOP clusters, socket cells, and scale cells; panel H′ focuses on differentiators between scale sub-types. Dot size reflects the percentage of cells in which gene expression was detected. Color coding allows a relative comparison of gene expression levels within a cluster (horizontal comparisons), but is not proper for vertical comparisons between clusters. Gene names in bold: see main text for references. Feature count matrices and Seurat objects used for the generation of all panels are available at the Open Science Framework Repository [78].

https://doi.org/10.1371/journal.pbio.3003233.g006

The bulk of cells providing structural integrity to the wing surface, to the inner surface of the lacunae (luminal tunnels invaded by trachea), and to the tracheal system are epithelial, diploid cells. Consistent with this, both tracheal (Trch_epi, 140 cells) and wing-related (Wing_epi, 780 cells) epithelial clusters shared top marker genes that are known epithelial cell markers in flies (Fig 6), such as the protease gene Stubble involved wing epithelial remodeling [58], the EGF signaling pathway genes pointed and EGFR [59], the wing cell polarity factors fat and starry night [60,61], and the genes grainyhead, headcase and blistered [59,62–65]. Ultrabithorax is known as an epithelial marker in butterfly pupal hindwings [39] and was enriched in both epithelial clusters. Notch showed epithelial and socket cell signals consistent with our previous study [37]. In addition to their shared epithelial gene expression profile, the two clusters differ by the expression of markers that are hallmarks of tracheal tip cell growth and tracheal branching, including the tracheal progenitor selector gene trachealess, the FGF ligand/receptor pair branchless/breathless, the proteoglycan dally involved in tracheal FGF signaling, and the cytoskeletal factor Zasp52 [66–69]. It is also noteworthy that some Trch_epi markers such as datilografo, APP-like and slow border cells were not expected in a trachea-related tissue based on current knowledge, indicating possible evolutionary divergence with Drosophila. Further investigation will be required to refine the range of cellular identities of the Wing_epi and Trch_epi clusters, for example, to decipher the dynamic complexity of tracheal development, and the differences between wing-membrane and lacunar epithelia in butterflies.

Developmental studies have shown that in butterflies, the SOP lineage has fully differentiated into arranged rows of scale and socket pairs by 30% of pupal development [40]. This differentiation was well resolved in the 40% stage single-nucleus transcriptome, with a clear cluster of 412 socket cells, based on the expression of the markers Su(H), Sox15, and neuromusculin (nrm), previously associated with Drosophila socket specification [70,71]. A total of 1,550 nuclei from scale-building cells were distributed across four distinct clusters that shared the expression of the trichogen master gene shaven (sv) [36,72,73]. Scale-related clusters also shared additional marker genes associated with trichogen development in flies such as pebbled (peb) [74], Osi24 [75], Jupiter [76], and Amun [77], supporting the proposed homology between the shaft of mechanosensory bristles with the scale derivative of butterflies and moths [32,34,37]. The next section focuses on refining the divergence between groups of scale cell precursors.

Transcriptome heterogeneity among scale subtypes in the 40% pupal male hindwing

The preliminary differential expression analysis of whole-wing nuclei suggests that scale cell precursors (sv+/peb+/Osi24+) resolve into at least four subtypes, including the large Scale1 cluster. To delve into the processes of color scale differentiation, we bioinformatically isolated Scale1-2-3-4, re-normalized gene expression counts, and reclustered this subset of nuclei to augment the resolution and separation of scale cell subtypes (Fig 7A). All eight scale clusters expressed canonical scale cell precursor markers such sv/peb/Osi24 (Fig 7B). The Scale1 cluster forms five subclusters numbered Scale1a to 1e for which we have not resolved defined identities at the moment, except for Scale1d, which is marked by Antennapedia (Antp) and the pterin repressor Bar-H1 [25] and seemingly encompasses the white scales of the wing coupling region (Fig 7C, 7D). Scale4 likely corresponds to hairlike scales of the dorsal surface (Fig 7D, 7E), based on the expression of homothorax and apterous-A [79,80], and immunolocalization of Cut (Fig 7F–7J′). Scale1b expresses nubbin and may correspond to a ground scale cell type (Fig 7K, 7L).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Diverging gene expression profiles of male scale cell precursors at 40% development.

A. UMAP plot representing the 8 sub-clusters of scale cell precursors across a total of 1,550 cells. B. Violin plots for scale markers (sv, peb, Osi24) and developmental factors showing differential expression patterns across scale subtype (see text) markers (all scales). C. Violin plot of Antp expression showing enrichment in Scale1d, and Antp immunofluorescence (Antp 4C3 antigen) localization to unpigmented wing coupling scales [81]. D. Violin plots for the unpigmented scales marker Bar-H1 [25], far-posterior region marker mirror [82], and proximal region marker homothorax [80]. E. Violin plots for Scale4 marker genes Cut, previously localized to large wing nuclei and wing margin [36,83], and apterous-A, a marker of dorsal-specific structures [79]. F. Immunofluorescent localization of the Cut 2B10 antigen (gray) shows bright marking of the wing margin, and nuclear signal interspersed in the wing epithelium and in tracheal lumen. G. Cut signal in tracheal lumen likely corresponds to chains of differentiating neurons. H–I′. Large nuclei with Cut signal likely correspond for dorsal wing hair, restricted to the proximal side of the discal crossvein (dotted line). J–J′. Expression of Cut in wing margin scales likely corresponds to the precursors of elongated margin scales. Cyan: staining of the basal wing membrane, here shown using the non-specific signal from animmunofluorescence assay using a guinea pig polyclonal anti-Dve antibody. K. Violin plot of nubbin expression. L. Antigenicity of Nubbin 2D4 revealing repression of Nubbin in dorsal scale cells (dcs) and lower signal in ventral scale cells (vcs). Scale bars: Scale bars: C, G, H, H′, I′, J′ = 100 μm; F = 1 mm; I, L = 20 μm. Feature count matrices and Seurat objects used for the generation of Fig 7A–7E, 7K are available at the Open Science Framework Repository [78].

https://doi.org/10.1371/journal.pbio.3003233.g007

UVI and spatulate scale cells are highly differentiated

To further profile the transcriptomes of the two most divergent clusters, we listed 1,006 genes showing significant differential expression (adjusted p-value < 0.05; minimum log2FC = 1.25; min.pct = 0.25) in comparisons between Scale2, Scale3, and the remaining scale groupings (Data 2 at https://osf.io/yjvkc/). Filtering this list down to 145 most statistically significant genes (adj. p-value < 10^-50) allows a heatmap visualization of the genes that are depleted or enriched in these clusters compared to other scale types (Fig 8A—8A′). Remarkably, both clusters Scale2 and Scale3 share a low expression of the UVI-state repressor Bab, a repressor of the UVI state that is repressed in UVI cells [29]. One of the two clusters thus likely corresponds to UVI scales, and we used fluorescent Hybridization Chain Reaction (HCR) mRNA detection of marker genes to resolve their identity. The snRNA-seq signal for DsxM indicates it is enriched in the Scale2 cluster alongside Arylsulfatase, a more specific marker for this cell population (Fig 8B). Both genes showed strong transcript signals in the dorsal margin areas where spatulate scales are located (Fig 8C–8D). Actin-binding protein 1 (Abp1) is a negative marker of Scale2, and indeed showed a complementary expression to Arysulfatase: while Abp1 is strongly expressed in the orange/UVI area, this signal decreases in the margin and shows a weak marking of alternating scale cells, as expected from an expression in ground scales (Fig 8D). Meanwhile, the Scale3 marker gene Sulfatase1 showed a visible association with the dorsal cover scales of the medial wing area (Fig 8E). Together, these results thus annotate the Scale2 (Bab⁻/Dsx⁺) cluster as the population of spatulate scale cells, and the Scale 3 (Bab⁻/Dsx⁻) as the UVI scale cell precursors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Transcriptional divergence of UVI and spatulate scale cell precursors.

All experiments correspond to male wing tissues at 40% pupal development. A–A′. Heatmap plot of the 145 most-significant differentially expressed genes in clusters Scale2 and Scale3 relative to the remaining scale cell clusters (Data 2-3 at https://osf.io/yjvkc/). The top-left panel (A) shows genes downregulated in Scale2 and Scale3. The remaining panels (A–A′) show genes that are enriched in either or both of these clusters. Bold: see text for details. Blue: see panels C–E for spatial expression. B. Violin plots for Bab and marker genes used for the spatial identification of the Scale2 and Scale3 cluster. C. HCR localization of Dsx mRNA (exons 1-2) in male dorsal wings. D. HCR localization of Arylsulfatase (green) and Abp1 (magenta), respectively, tested as positive and negative markers of the Scale2 cluster, in male dorsal wings. Arylsulfatase is found in the dorsal cover scales of the male marginal region. Arrowheads: ground scale expression of Abp1 in the marginal region, without overlap with Arylsulfatase. E. HCR localization of Sulfatase1 mRNA, tested as a marker of the Scale3 cluster, in the medial region of a male dorsal hindwing. Expression is restricted to alternating scale precursor cells corresponding to the presumptive UVI scales, while ground scale precursor cells (gs; dotted lines) are negative. Scale bars: C-D (top) = 1 mm; C, D insets (bottom) = 100 μm; E = 10 μm. Feature count matrices and Seurat objects used for the generation of Fig 8A, 8B are available at the Open Science Framework Repository [78].

https://doi.org/10.1371/journal.pbio.3003233.g008

In addition, a total of 1,024 genes are differentially expressed between the two Bab⁻ clusters (Data 3 at https://osf.io/yjvkc/), suggesting that while they share the property of being male-specific, UVI and spatulate scale precursors deploy distinct gene expression programs at the 40% pupal stage. For example, the spatulate Scale2 cells are enriched for the ivory lncRNA gene and yellow-c (Fig 8A′), two markers of melanic scales in nymphalids [84–87], and express fz3 and vestigial, which mark the periphery of the wing [88,89]. ImpL2 is the ortholog of BmIMP, a gene required for the male-specific splicing of Dsx in Bombyx [90], and is restricted to the Dsx⁺ spatulate Scale2 cells here, suggesting it may play a similar role in butterflies.

ChIP-seq profiling of Bab genome-wide occupancy identifies potential gene targets

Single-cell analyses revealed the two populations of male-specific scale types are the most divergent, implying that a specific gene expression program specifies their morphology. Of particular interest, Bab may be a repressor of a network of genes involved in specifying the specialized ultrastructures that mediate UV-iridescence. To further explore what genes contribute to this ultrastructural specialization, we used ChIP-seq to profile Bab genome-wide occupancy, using fixed nuclei from C. eurytheme wings sampled at the 40% and 60% stages, with two biological replicates per stage (Fig 9A, 9B, Data 4 at https://osf.io/yjvkc/). Following MEME-ChIP motif discovery, the imputed Bab binding matrix recovered the AT-rich sequence profile previously established for Drosophila Bab1 using DNAse footprinting assays [91], validating the specificity of our ChIP-seq assay. The intersection of predicted Bab occupancy sites at the two stages results in a list of 1,482 candidate genes with at least one stable Bab-binding site in their intragenic interval or immediately adjacent intergenic regions (Fig 9C). Among these, 77 genes are upregulated and 53 are downregulated in the Bab^– clusters Scale2/3 relative to Bab⁺ clusters, thus forming a stringent list of candidate transcriptional targets for Bab (Data 5 at https://osf.io/yjvkc/). It is unclear if Bab, a BTB-domain transcription factor, acts solely as a transcriptional repressor, or if it can also act as a co-transcriptional activator [91–93]. It thus remains uncertain whether the transcriptional targets of Bab should correlate positively or negatively with Bab expression, and we illustrate both possibilities with genes that present Bab binding sites and are either enriched or repressed in clusters Scale2 and Scale3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. ChIP-seq profiling of Bab identifies potential target genes involved in the differentiation of Bab– scale types.

A. MEME-ChIP predicted motif for Bab occupancy as inferred from the Bab ChIP pupal wing dataset (top), resembling a previous binding profile for a Bab fly ohnologue (bottom). B. Summary of the position of all predicted Bab ChIP-seq binding sites across two datasets, at the 40% and 60% pupal stages (Data 4-5 at https://osf.io/yjvkc/). Inner circles: HOMER classification of imputed binding sites relative to gene annotation features. Outer rings: overlap with genes that are among the differentially expressed (DE) genes in Scale2 (spatulate) and/or Scale3 (UVI) relative to other scale types. C. Venn diagram featuring the numbers of genes identified as Bab-bound and differentially expressed (DE) in the Scale2-3 clusters, resulting in an intersection of 77 (up) and 53 (down) DE genes with binding sites identified across two stages. D, E, F, H, J. Genomic intervals featuring the Bab ChIP-seq profiles across replicates and peak calls at each stage, including at the promoters of Scale2-3 markers (bottom, buff color). G–G′, I–I″, K-K′. Violin plots of snRNAseq scale cluster expression for the same marker genes.

https://doi.org/10.1371/journal.pbio.3003233.g009

For example, the Bab locus itself shows multiple Bab ChIP peaks, suggesting it may be auto-regulated by negative and positive feedbacks (Fig 9D). The Dsx locus also contains multiple Bab-binding sites (Fig 9E), suggesting that Bab may be an upstream regulator of Dsx. Conversely, Dsx is an upstream regulator of Bab in flies [18,20] and also binds a putative intronic enhancer of Bab in swallowtail butterflies [94]. Our data thus suggests that Dsx and Bab cross-regulate each other, resulting in feedback that may enforce complex dynamics and switches in their expression during sexual differentiation. The transcription factor genes Msx1 (Msh/Drop ortholog in D. melanogaster), and Msx2 are organized in tandem in butterflies, similarly to Tribolium [95], and show multiple Bab-binding sites. Msx2 is a top marker of Scale3, while Msx1 marks both Bab– clusters Scale2-3, suggesting Bab may act as a repressor at this locus. Thus, not only does Bab provide regulatory feedback on itself and Dsx, it may also regulate other transcription factors, and these findings suggest that the gene regulatory networks that determine scale identity could be more complex than previously appreciated.

Another class of genes appearing as UVI-regulating candidates is the conserved family of Obstructor-secreted proteins, which play roles in chitin modifications and cuticle properties [96–98]. We found that the five members of this family are clustered in tandem in C. eurytheme, and that while Obst-A and Obst-E are not detected in scale cells, Obst-B, Obst-C (gasp) and Obst-D (Peritrophin-A) are all consistently expressed as markers of the UVI cell cluster Scale3 (Fig 9H–9I″). Strikingly, the promoter regions of these three genes show strong Bab ChIP-seq signals, suggesting Bab may repress the expression of Obstructor family genes. Similarly, the chitin deacetylase genes Vermiform and Serpentine occur in tandem, show evidence of Bab-binding, and are enriched in the spatulate scale cluster Scale2 (Fig 9J–9K′).

Overall, these data sketch a broad overview of the potential targets of Bab at both the 40% and 60% stages, leveraging differential expression from both Bab⁻ cells types—the spatulate and UVI scale cell precursors. Presently, we do not have functional evidence that low Bab expression in Scale2 cells is required for spatulate scale specification. In contrast, loss-of-function assays and heterozygous phenotypes of Bab alleles show it is necessary and likely sufficient for UVI scale repression [29]. We thus sought to further refine the set of potential regulatory targets of Bab by focusing on genes that are differentially expressed in the Scale3 UVI cells (UVI-DE). Interestingly, genes with Bab ChIP signals are significantly more likely to be UVI-DE (Fig 10A). Stringent filtering criteria result in a list of 87 UVI-DE genes with 3 or more Bab ChIP binding peaks (Data 6 at https://osf.io/yjvkc/). Within this set, 17 genes have known functions in Drosophila that relate to cytoskeletal dynamics or cuticle formation (Fig 10B, 10C), two key processes that we expect to regulate ridge spacing and multilayering. For example, the loci encoding the nuclear hormone receptor Eip75b and cytoskeletal regulator Multiple wing hair (Mwh) each show more than 50 Bab ChIP sites, and are upregulated in UVI scale precursors, suggesting that Bab directly acts as a transcriptional repressor at these genes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Potential targets of Bab regulation in the UVI scale type.

A. Box and whiskers plot of probability of differential expression, binned by number of called Bab ChIP sites in proximity to each gene (Wilcoxon rank sum tests; *: p < 0.01; ***: p < 1e−15). B. Expression dot plot visualization of selected UVI-DE genes (log2FC > 1.8, adjusted p < 0.01) with at least 3 Bab ChIP peaks and potential roles in ultrastructural specialization (Data 5 at https://osf.io/yjvkc/). C, D. Volcano plots of UVI-DE genes (Data 7-8 at https://osf.io/yjvkc/), colored by number of Bab ChIP sites (C), and by the presence of selective sweeps in a sympatric population of C. eurytheme x C. philodice (D). These two morphospecies hybridize and show genome-wide admixture across autosomes (dendrograms reproduced from [29], resulting in a population polymorphic for UV iridescence (present in C. eurytheme, absent in C. philodice).

https://doi.org/10.1371/journal.pbio.3003233.g010

Differentially expressed genes experiencing selective sweeps

C. eurytheme hybridizes in sympatry with its UV-negative sister species C. philodice in the Eastern US [29]. We reasoned that key downstream loci in the formation of UVI scales could also show signals of selection that are present in C. eurytheme but absent in C. philodice. Selective sweeps were calculated as the μ statistic in RAiSD [99] and intervals called with bedtools, taking a threshold of the top 1% of μ values. 449 sweeps were detected in total, and 63 were found only in C. eurytheme. Of 11,647 unique ChIP binding sites, just 29 were found to be within selective sweeps, and of these, 4 are private to C. eurytheme, one is private to C. philodice, and 24 have undergone a sweep in both species (Data 7 at https://osf.io/yjvkc/). Notably Stubble (Sb), a determinant of bristle actin bundle number in flies [100,101], has two ChIP peaks that have undergone a sweep in C. eurytheme and not C. philodice. Most sweeps do not directly intersect ChIP binding sites, but 313 sweeps near 287 genes are in proximity to both ChIP binding sites and selective sweeps, including 58 UVI-DE genes (Fig 10D, Data 8 at https://osf.io/yjvkc/). Notably, Mwh, Sb, Bab, and ABCG20 (Fig 10B) all have private sweeps in C. eurytheme, along with pdm3, a gene known to be involved in patterning and pigmentation in other butterflies [33], and SCAR, an Arp2/3 interacting protein involved in bristle intracellular patterning in flies [102]. The combination of differential expression, binding by Bab, and signals of selection on this subset of UVI scale-expressed genes provide an initial glance at the candidate genes that fine-tune the specialized ultrastructure of UVI scales.

Antagonistic roles of Dsx isoforms in marginal wing patterning

CRISPR and RNAi phenotypes resulted in concordant phenotypes, and highlight the versatility of Dsx in mediating the sexually dimorphic differentiation of UVI and spatulate scale types, as well as the patterning of marginal melanic bands that are proper to each sex (Fig 5A, 5F).

As Dsx mosaic knock-outs yielded clonal effects with sharp boundaries for these traits, we inferred that Dsx had cell autonomous effects on the control of UVI and spatulate scale cell identities. In contrast, marginal patterning did not show such clonal effects, which could have consisted in splitting effects on the female-like yellow spots, abrupt discontinuities in the position of the orange/black pattern boundary, or in sharp UVI-negative clones in the distal region in males. We propose that this is because Dsx isoforms influence the masculinization or feminization of these marginal patterns in a non-autonomous manner, probably by controlling extracellular signaling and morphogenetic processes that take place in early pupal stages. This dual regulation would be consistent with current models suggesting that DsxF and DsxM isoforms target an identical set of genes, but confer opposite effects on gene expression [18,19]. The Wnt signaling pathway, the transcription factors Spalt and Distal-less, and the melanic scale determinant ivory/miR193 have been implicated the patterning of pierid marginal bands [87,103–107], and could thus be targets of differential regulation by Dsx isoforms.

Likewise, we found that DsxM and DsxF, respectively, masculinize and feminize the state of marginal dorsal cover scales, by promoting or repressing the differentiation of male-specific spatulate scales. Surprisingly, Dsx mKOs induced reciprocal transformations in both directions (Fig 4B, 4C), implying that the canonical-spatulate states are switchable in each sex and under antagonistic regulation from Dsx (Fig 5A). To explain this, we speculate that DsxM and DsxF have opposite effects on the expression of the same target genes—a model that accounts for some of the mode of action of Dsx in other insects [108,109]. In future studies of Colias wing dimorphism, it will be interesting to compare the single-nucleus transcriptomes of both male and female tissues, and to establish the occupancy profile of Dsx in each sex [94].

UV dichromatism is DsxM-independent but requires DsxF-mediated activation of Bab

Bab is required to prevent UV-iridescence in most scales, and is repressed in the orange dorsal cover scales of male C. eurytheme, thus allowing UV-iridescence in a small subset of scales [29]. Previous work on Drosophila sexual dimorphic traits such as the pigmented abdomen, sex combs, and gonadal stem nich, collectively suggest that DsxF and DsxM have an activating and repressing effect on Bab, respectively [20,21,110,111]. Our experiments rule out a repressing role of DsxM on Bab because Dsx perturbation did not induce loss of UV-iridescence in males. In addition, snRNAseq, HCR mRNA detection assays, and immunofluorescence did not detect DsxM in the UVI cell precursors, i.e., the male dorsal cover scales of the medial region. Of note, while we observed a marginal reduction in the UVI field in the vicinity of the peripheral patterns, we extrapolate that this was due to a change in the identity of the marginal patterns rather than a direct effect of DsxM on Bab.

In the medial region, DsxF acts as a likely transcriptional activator that maintains Bab expression at the 40% stage, thus repressing UVI fate. Such regulatory linkage between Dsx and Bab is reminiscent of studies of D. melanogaster, where DsxF activates Bab expression, thereby controlling the sexual dimorphism of abdominal pigmentation [20,111]. At the moment, while these data in distant insect lineages suggest that dimorphic Bab expression often drives sexual differentiation in a variety of tissues, it is unclear if this Dsx inputs on Bab regulation evolved repeatedly or if they derived from ancient, conserved interactions. Interestingly, DsxF binds an intronic cis-regulatory element of Bab during the pupal wing development of Papilio alphenor butterflies [94], but it remains to be determined whether this mode of regulation of Bab is conserved and homologous with C. eurytheme.

The evolution of scale sexual dimorphisms in Colias

Sexually dichromatic butterflies show a general trend where female wings are usually less colorful than the male ones [112], except when colors have an aposematic function. Accordingly, UV-iridescence in pierid butterflies is either male-specific (e.g., in C. eurytheme), or when it is found in females, it is in a much reduced or residual state than in males [29,113]. In other words, sexual differences in UVI likely involved the recruitment of an off-switch in females, rather than an on-switch in males. This is consistent with a two-step model for the evolution of UVI sexual dichromatism in pierids. First, UVI likely originated in both sexes, and this required a loss of Bab expression in the dorsal cover scales of the medial region. Later, reduction of UVI patches evolved in females by recruiting Dsx in the dorsal cover scales, re-activating the expression of Bab. This off-switch mechanism went a further step in Zerene cesonia, a close relative of Colias, where a duplicated copy of Dsx evolved a truncated female-specific transcript that is highly expressed in the female dorsal scales, in a pattern consistent with UVI-repression [114]. We thus postulate that Dsx did not contribute to the origin of the UVI scale type itself, but that its female-specific products were recruited to re-activate Bab in dorsal cover scales and thereby reduce the size of the UVI patch in females. Crucially, this model is consistent with phylogenetic models that place the origin of UVI scales in an early ancestor of Pieridae, followed by secondary reduction of UVI patterns [23].

How does this mechanism fit more broadly into evolutionary thinking? In a decade-long debate about the causes of sexual dichromatism (primarily in birds and butterflies), Arthur Wallace invoked the importance of natural selection to decrease conspicuous traits in females, while Charles Darwin favored the supremacy of sexual selection [115]. Because males and females are often subject to different selective pressures, it is commonly accepted that the two explanations are not mutually exclusive [116,117]. In our evolutionary model, sexual selection remains a key force to explain the origin of UVI and its persistence in males [27,118], as defended by Darwin, while the spatial reduction of this trait in females supports a role for natural selection as championed by Wallace. For example, passerine birds that predate Colias butterflies express UV-sensitive opsins [119], and are likely to perceive male wing strokes as bright flashes of UV-color [22,31], which produce chromatic contrast on green vegetation backgrounds with low UV-emissivity (S7 Fig). Evolutionary losses and reductions of female UVI patterns may thus be linked to predator avoidance, and the recruitment of DsxF in UVI scale repression provides a possible mechanism for this sex-limited effect.

Using snRNAseq to explore wing cell type diversity

The enormous diversity of butterfly wing patterns largely amounts to the ability of these organisms to differentiate various scale types that vary in coloration, and to organize them in space during development. Scales emerge from a single-scale cell precursor during development, and we expect that single-cell approaches to scale differentiation will uncover new insights into how tissues organize complex spatial patterns during development and how new traits evolve. In this study, single-nucleus transcriptomics allowed the detailed profiling of nearly 3,000 cells, about half of which were scale cell precursors. The snRNAseq sample consisted of a single hindwing dissected from a male C. eurytheme pupa, and the pooling of several individuals is unnecessary with this technique and tissue. Subsetting the initial clustering to select for scale clusters, characterized by sv expression [36,37], enables the detailed analysis of scale cell precursor transcriptomes in ways that have not been possible with bulk RNAseq methods (e.g., for mining of top marker genes from differentiated clusters), and generated new hypotheses regarding cell type specification and specialization. Importantly, this study provides only a snapshot of gene expression profiles in a single individual and sex. Future studies of Colias pupal wing development could compare sexes and may benefit from analyzing tissues sampled across a tight developmental series, in order to reveal expression trajectories of differentiating cell populations.

Transcriptomic insights on the differentiation of specialized scale ultrastructures

We found that two scale clusters lack Bab expression, and correspond to male-specific, Dsx-dependent scale types – the UVI and spatulate scales. Each of these types show ultrastructures that diverge from canonical scale types, with extreme variations in the spacing of longitudinal ridges (Fig 1D), and the remarkable multilayering of the UVI scale upper lamellae (Fig 1J) that yields their structural coloration [44,120].

Previous studies have shown that condensed actin filaments pre-pattern the density of ridges in butterfly scales, and then template the formation of iridescent elaborations [39,121–123]. Genes that are differentially expressed in the UVI scales provide a wealth of candidate modulators of actin dynamics including shavenoid, Abp1, singed (Fig 6H′), which are known to modulate actin assembly and ridge patterning in Drosophila bristles [102,124,125]. We found strong evidence of Bab occupancy at the UVI scale marker genes fimbrin and mwh, two key determinants of actin bundling and bristle morphology in flies, indicating they are direct targets of Bab repression (Fig 10B). Similarly, we suggest that Bab-bound DE genes with annotated functions in vesicular trafficking and cuticle formation—such as Osi17, Cpr92F, and the chitin deacetylase ChLD3 [75,126,127]—may be direct targets of Bab regulation that participate in the sculpting of the nanostructures on the scale upper surface, including the stacking of UVI ridge lamellae. This is further strengthened by the fact that a number of these genes also show evidence of selection specifically in C. eurytheme and not its UV-negative sister

species C. philodice (Fig 10D), implying either that regulatory connections are being maintained by purifying selection or that network interactions permissive for UVI scale specialization are under positive selection. While the function of these potential effectors will require further investigation, the application of snRNAseq to mid-pupal stages of butterfly wing development promises to be an exciting avenue of research for understanding how complex exoskeletal ultrastructures form and diversify.

Butterflies

C. eurytheme females were collected from an organic alfalfa field (Buckeystown, MD), and left in small cylinder cages with alfalfa for oviposition. Eggs were collected manually, washed 1′ with Benzalkonium Chloride 5%, rinsed and dried, left to hatch at 25 °C, 60% humidity, 14:10 h day:light cycle and fed with Black Cutworm artificial diet (Frontier Agricultural Sciences) supplemented with dry alfalfa powder, before transfer to Lana woollypod vetch (Vicia villosa) in a greenhouse environment [26]. Fifth-instar wandering larvae were moved to a growth chamber set at 28 °C, 60% humidity, 16:8 h day:light cycle, and their time of pupation recorded. Developmental times are reported as percentages, relative to a mean development time 138 h from pupation to adult emergence at 28 °C, or 170 h at 25 °C.

CRISPR mosaic knock-outs

Eggs were collected on alfalfa shoots and leaves, washed, dried, and placed upward on thin strips of double-sided tape, on the inner side of a 1.25 oz. cup lid, and micro-injected following previously reported procedures [26,29] with an equimolar mixture of Cas9-2xNLS:sgRNA at 500:250 ng/μL. Embryos were left to hatch in cups containing vetch sprouts, and transferred to vetch sprout mats in a greenhouse environment for larval growth.

RNAi electroporation

Dicer-substrate siRNAs (DsiRNAs) were designed against target gene exons (S2 Table), ordered at the 2 nmol scales as Custom DsiRNAs with standard purification (IDT DNA Technologies), resuspended at 100 μM in 1× Bombyx injection buffer (pH 7.2, 0.5 mM NaH₂PO₄, 0.5 mM Na₂HPO₄, 5 mM KCl), and stored as frozen aliquots at −70 °C until use. Electroporation procedures followed a previously described procedure [89,128]. To target the dorsal surface of male wing pupae, the negative electrode was placed in contact with the droplet of saline positioned on the ventral side of the peeled wing, while the positive electrode was in contact with the agarose pad on the dorsal side of the wing, before electroporation with 5 square pulses of 280 ms at 8 V, separated by 100 ms intervals.

Genome annotation

The genome annotation available for Colias croceus (NCBI:PRJEB42949), was mapped to the best available C. eurytheme assembly (NCBI:GCA_907164685.1) using Liftoff [24,129,130]. A total of 12,908 genes had annotations transferred from C. crocea and 7,418 genes were obtained on reciprocal Blast against Flybase protIDs using -blastx with -evalue 1e−50, retaining best hits against subjects.

Live nuclei preparation for single-nucleus transcriptomics

We isolated live nuclei using dounce homogenizers following previously published protocols [37,131]. Briefly, a hindwing was dissected from a single male individual (53 h old pupa at 28 °C, 40% stage) in cold 1× phosphate-buffered saline (PBS) and immediately transferred for sequential douncing in homogenization buffer (250 mM sucrose, 10 μM Tris pH 8.0, 25 mM KCl, 5 mM MgCl₂, 0.1% Triton-X 100, 0.2 U/μL RNasin Plus, 1× Protease Inhibitor, 0.1 mM Dithiothreitol). The homogenized tissue was transferred to a new 1.5 mL tube and spun down in a fixed-rotor centrifuge at 4 °C for 10 min at 1,000g, before resuspending in 500 μL nuclei suspension buffer (PBS, 1% Bovine Serum Albumin, 0.2 U/μL RNasin Plus). The nuclei suspension was filtered using a 40-μm PluriSelect filter and 10 μL were drawn for staining with trypan Blue. Nuclei quality was assessed based on membrane integrity and nuclei were counted using a hemocytometer, yielding a concentration within the target range of 600–900 nuclei/μL. Nuclei suspension was processed for cDNA library preparation with a target output of 3,000 nuclei per sample.

snRNAseq library preparation and sequencing

The droplet-based 10× Genomics Chromium Next GEM Single Cell 3′ Reagent Kit v3.1 with Dual Indexes was used to prepare multiplexed cDNA libraries from nuclei suspensions [37]. Libraries were pooled for PE150 sequencing on a Novaseq S2 flow cell at a target depth of 200,000 reads/nucleus. Raw reads are accessible at the NCBI SRA (NCBI:SRS12272185).

snRNAseq data preprocessing and analysis

BCL files were demultiplexed into fastq files using bcl2fastq. Raw and filtered CellRanger-generated count matrices with the flag --include-introns were used for downstream analysis and are available in an online data repository [78]. Seurat v5.0 was used for preprocessing of Seurat objects, first filtered to remove low-quality Gel Beads-in-emulsion (GEMs) with nCounts < 300, nFeatures < 3,000 and percent.mt > 3%. Library sizes, number of per-cell features and proportion of reads mapped to mitochondrial genome were used as quality control metrics. The sample at 40% pupal development was normalized using SCTransform with an additional regression of nuclei based on percentage of mitochondrial reads (vars.to.regress = “percent.mt”), before performing RunPCA and RunUMAP using top 20 PCs, after which FindNeighbors and FindClusters functions were invoked to generate 2D layouts of graph-based clusters. Clustering resolution of 0.1 was used after examining cluster stability using R/clustree on clustering resolutions at 0.1, 0.2, 0.5, 0.7, and 1.0.

For cluster annotation of recovered nuclei from the 40% sample, we identified differentially expressed genes using FindAllMarkers in Seurat (Wilcoxon Rank Sum test with Bonferroni correction for multiple testing; adjusted p < 0.05). Marker genes were identified based on genes detected in a minimum of 25% of the cells within the cluster and a log2FC >0.25 between the cells in the cluster and all remaining cells. Focusing on scale-building nuclei, the whole object was subset for shaven-expressing nuclei. Preprocessing steps were performed as previously mentioned but using 0.2 resolution for FindClusters. Heatmap was generated using R/ComplexHeatmap [132,133] and genes were ordered via unsupervised hierarchical clustering of 4 k-means groups.

Following cluster identification, we subset the whole object for scale clusters Scale1-4 using expression of canonical markers sv and ss. Normalization with SCTransform was performed on the original, filtered count matrix as mentioned previously, following RunPCA and RunUMAP with top 10 PCs, before FindNeighbors and FindClusters were performed with a clustering resolution of 0.1. FindMarkers comparing each cluster of interest with the rest of the subsetted object adjusted p < 0.05, log2FC > 1.25, min.pct = 0.25) was used to identify marker genes.

Preprocessing and analysis of Chromatin-Immunoprecipitation (ChIP)-sequenced data

ChIP-seq libraries were prepared from live pupal wings (S1 Text), and sequenced as PE42 reads for samples at 40% and PE37 reads for samples at 60%, available on the NCBI SRA, PRJNA1148116). The C. eurytheme genome was indexed using bowtie2 (2.5.3). FastQC was used to assess the quality of fastq files. Trimmed reads were aligned to the genome using bowtie2 (2.5.3), sorted using samtools (1.15.1), filtered, and MarkDuplicates using picard tools (2.26.8) from GATK (4.2.4.0) was invoked to remove duplicate reads. Filtered and uniquely mapping reads were used for peak calling using MACS3, with options q = 0.01, BAMPE, and an effective genome size of 328,651,476 bp. Relaxed peak calling with a p = 0.05 threshold was performed, followed by further statistical correction using the Irreproducible Discovery Rate framework (version 2.0.4.2), which assessed the reproducibility of called peaks between the two replicates based on a fraction of false positive peaks [134,135]. This generated a final list of peaks for each time point at the maximum False Discovery Rate of 0.05 (Data 4 at https://osf.io/yjvkc/). Peak annotation was performed using HOMERv4 annotatePeaks.pl, to assign each peak to a genomic region of a nearby gene [136]. For visualization in IGV, the deeptools (3.5.4) function bamCompare was used to subtract input bam files from each corresponding replicate and derive bigWig files for each sample [137]. To identify conserved motifs within peaks across four samples, fasta files for peak intervals were extracted from IDR output using bedtools (2.28.0) function getfasta, and served as the input for motif discovery using the Combined Fly reference set in MEMECHIP [138]. Finally, gene lists from Data 2 and Data 4 were merged to identify candidate genes regulated by Bab, with differential expression in the Bab⁻ scale clusters and Bab-binding ChIP peaks present in both pupal stages, 40% and 60% development.

Immunohistochemistry and confocal imaging

Antibody stainings were performed as previously described [29]. In short, pupal wings were dissected in cold PBS and fixed in fixative (4% methanol-free paraformaldehyde diluted in PBS, 2 mM egtazic acid) at room temperature for 13−20 min, washed four times in PT (PBS, 0.1% Triton-X 100), blocked in PT-BSA (PT, 0.5% Bovine Serum Albumin), incubated overnight at 4 °C with primary antibody dilutions in PT-BSA, washed in PT, incubated 2 h with secondary antibody dilutions in PT-BSA (1:500 dilutions, following two short centrifugation steps to pellet down particules before incubation with the wings), washed in PT, incubated in 50% glycerol with with 1 μg/mL DAPI (4′,6-diamidino-2-phenylindole), mounted on glass slides with 70% glycerol or SlowFade Gold mountant under a #1.5 thickness coverslip, and sealed with nail varnish before confocal imaging. Antibodies used in this study were a polyclonal, affinity purified anti-Colias Bab [29] rabbit antibody (1:100 dilution); several monoclonal mouse antisera (1:50–1:100; Developmental Studies Hybridoma Bank) targeting DsxDBD [57], Nubbin [139], Cut [140], and Antp [141]; and a polyclonal, affinity purified anti-Dve [142] guinea pig antibody (1:400; kind gift of Mike Perry). Secondary antibodies included conjugated AlexaFluor488 anti-Mouse IgG (Life Technologies, CA) at 1:500 dilution, conjugated AlexaFluor647 anti-Rabbit IgG (Life Technologies, CA) at 1:500 dilution, and conjugated AlexaFluor555 goat anti-guinea pig IgG (Abcam, UK). Stacked acquisitions were obtained on an Olympus FV1200 confocal microscope mounted with PLANAPO 20× and 60× objectives. Immunofluorescent stains of whole pupal wings were imaged with a Zeiss Cell Observer Spinning Disk confocal microscope mounted with a 10× objective (Plan-Apochromat, 0.45 NA) for HCR mRNA stainings, or with an Olympus BX53 epifluorescent microscope mounted with an UPLFLN 4× objective. Stitching correction was performed on spinning disk confocal-acquired stitches using the ZEN software.

Detection of selective sweeps

An admixed population of C. eurytheme and C. philodice (Buckeystown, MD) was previously resequenced, genotype-called, filtered, and analyzed [29]. RAiSD was used to calculate the µ statistic separately on each of C. eurytheme and C. philodice with default parameters [99]. Swept intervals were identified by taking the top 1% of µ for each species and merging the called sites with bedtools merge. Each called interval was then connected to its nearest gene using bedtools closest, and intersected with called ChIP sites with bedtools intersect.