Genomic Resources for the Scuttle Fly Megaselia abdita: A Model Organism for Comparative Developmental Studies in Flies

Abstract

The order Diptera (true flies) holds promise as a model taxon in evolutionary developmental biology due to the inclusion of the model organism, Drosophila melanogaster, and the ability to cost-effectively rear many species in laboratories. One of them, the scuttle fly Megaselia abdita (Phoridae) has been used in evolutionary developmental biology for 30 years and is an excellent phylogenetic intermediate between fruit flies and mosquitoes but remains underdeveloped in genomic resources. Here, we present a de novo chromosome-level assembly and annotation of M. abdita and transcriptomes of 9 embryonic and 4 postembryonic stages. We also compare 9 stage-matched embryonic transcriptomes between M. abdita and D. melanogaster. Our analysis of these resources reveals extensive chromosomal synteny with D. melanogaster, 28 orphan genes with embryo-specific expression including a novel F-box LRR gene in M. abdita, and conserved and diverged features of gene expression dynamics between M. abdita and D. melanogaster. Collectively, our results provide a new reference for studying the diversification of developmental processes in flies.

Summary statement

We report a chromosome-level genome assembly and annotation and transcriptomes for an emerging developmental model organism, the phorid fly Megaselia abdita, which is phylogenetically intermediate between Drosophila and mosquitoes.

Introduction

Comparing related species is a powerful approach to understanding how mechanisms of development evolve and diversify. Their natural diversity and their phylogenetic history can aid in revealing core principles and inform our understanding of evolutionary transitions. Additionally, careful comparisons of developmental mechanisms between multiple species in “model taxa” are critical for determining the directionality of change and identifying mutations responsible for important evolutionary shifts in developmental processes. Such mutations may not necessarily be adaptive and their significance as drivers of evolutionary change might be overlooked, given that complex developmental gene networks can enhance a population’s permissiveness for the passive fixation of mutational variants that open novel paths of adaptive evolution (Kimura and Ohta, 1974; Lynch, 2007a, 2007b).

While it is impractical to adapt a large set of closely related vertebrate model organisms for laboratory studies, insects, in particular Diptera (true flies), offer a cost-effective alternative (Grimaldi and Engel, 2005; Schmidt-Ott and Lynch, 2016; Wiegmann et al., 2011). Flies are particularly appealing for the comparative study of developmental mechanisms because they include a leading model organism in developmental biology, Drosophila melanogaster, and many more species that are relatively easily cultured in laboratories. Developmental biologists have introduced several new dipteran model organisms in recent years, including the humpbacked fly, Megaselia abdita, which has been particularly useful for studying the evolution of developmental mechanisms in dipteran embryos because of its technical advantages and phylogenetic position (Rafiqi et al., 2011). It belongs to the large family Phoridae, also known as scuttle flies (Disney, 1994; Li et al., 2024), and represents a lineage that separated from the Drosophila lineage ca. 145 million years ago at the beginning of the Cyclorrhapha radiation, roughly 100 million years into the dipteran radiation (Grimaldi and Engel, 2005). Developmental biologists started using M. abdita as an experimental system in the 1990s to study the evolution of axial pattern formation (Bullock et al., 2004; Crombach and Jaeger, 2021; Crombach et al., 2016^; Liu et al., 2018^; Rohr et al., 1999^; Schmidt-Ott et al., 1994^; Stauber et al., 1999^; Stauber et al., 2000^; Stauber et al., 2002^; Wotton et al., 2015a^; Wotton et al., 2015b^; Wotton et al., 2015c^; Yoder and Carroll, 2006), and subsequently the evolution of extraembryonic tissue (Caroti et al., 2018; Fraire-Zamora et al., 2018; Horn et al., 2015; Kwan et al., 2016; Rafiqi et al., 2008; Rafiqi et al., 2010; Rafiqi et al., 2012; Schmidt-Ott and Kwan, 2022; Stauber et al., 1999; Wotton, 2014; Wotton et al., 2014), and other aspects of embryo development (Caroti et al., 2015; Dey et al., 2023; Tanaka et al., 2015; Vicoso and Bachtrog, 2015). However, the limited availability of genomic resources in M. abdita (Jimenez-Guri et al., 2013; Vicoso and Bachtrog, 2015) and the Phoridae in general (Feng et al., 2020; Rasmussen and Noor, 2009; Zhong et al., 2016) has limited the potential of this model organism by precluding genome-wide and epigenetic experimental approaches (Figure 1).

Figure 1.

Phylogeny of Diptera with available genome annotations across lineages. The phylogeny is based on Wiegmann et al. (2011) and Jimenez-Guri et al. (2013). The width of each triangle reflects species richness within each group. The number of annotated genomes available for each clade in the genome database at National Center for Biotechnology Information (NCBI) or in the Darwin Tree of Life (DTL) project that have not been uploaded to NCBI Genomes as of Feb. 4th 2025 are indicated. The red arrow highlights the phylogenetic position of Megaselia abdita.

Here we provide a de novo assembled and annotated chromosome-level genome for M. abdita, alongside stage-specific transcriptomes across its life cycle. These resources provide an excellent basis for genome-wide and epigenetic experimental approaches for an understudied but phylogenetically important branch of the Diptera. They also establish synteny relationships with the chromosomes of D. melanogaster, highlight conserved and divergent features of Hox gene clusters, and provide insights into embryonic gene expression dynamics and orphan genes. Collectively, our results will help to establish dipterans as a model taxon to study the evolution of developmental mechanisms from gene regulation to neural networks and behavior.

Results and Discussion

Genome assembly

We generated a de novo reference genome for Megaselia abdita using combined long read and chromatin conformation capture methods obtained from several hundred embryos of a 10-generation inbred line. Long-read, high-fidelity sequences were generated by PacBio (HiFi PacBio reads), and chromatin conformation capture sequences were generated by Dovetail Genomics’ Omni-C method. After removing 33 scaffolds identified as contamination, the initial draft assembly spanned 592.8 megabases (Mb) contained in 89 scaffolds with an N50 of 12.6 Mb. Using HiRise, a software designed to scaffold genome assemblies with proximity ligation data (Putnam et al., 2016), the draft assembly was refined using the Omni-C reads (Figure S1).

The final assembly is 592.8 Mb contained in 15 scaffolds with an N50 of 212.8 Mb (Table 1). The genome size is comparable to a previous estimate of 562.7 Mb based on flow cytometry data (Picard et al. 2012). We estimated heterozygosity at 0% - 0.24% (Figure S2A). We masked 67.9% of the genome, constituting repetitive elements (Figure S2B). The largest three scaffolds correspond to the three chromosomes of M. abdita (Table S1). We aligned the remaining 12 scaffolds to the largest three and found that these scaffolds contain repetitive sequences that cannot be correctly assembled into chromosome-level scaffolds.

Table 1.

M. abdita reference assembly statistics and quality metrics.

Reference Assembly Statistics
Total length (bp)	592,824,975
Number of scaffolds	15
Scaffold N50 (bp)	212,802,314
Scaffold L50	2
# of Ns per 100kb	1.32
GC (%)	29.74%
Masked (%)	67.90%
Eukaroyotic BUSCO’s recovered (%)	99%

Genome annotation

We annotated the reference genome of M. abdita using evidence from RNAseq transcripts and protein sequence databases, and a robust genome annotation pipeline (Figure S3). The process involved mapping RNAseq data and assembling the transcriptome, mapping protein sequences to the reference genome, and generating gene models using various prediction software. We then created consensus gene models using EVidenceModeler, updated the models to include UTRs and alternative isoforms, and filtered out transposable element models before functionally annotating the genes with Eggnog-mapper (Haas et al., 2008; Huerta-Cepas et al., 2019; Cantalapiedra et al., 2021). In the Materials and Methods section, we provide a detailed walkthrough of this pipeline.

M. abdita’s genome contains 11,934 protein-coding genes (Table 2). We assessed the quality of the annotation with Benchmarking Universal Single-Copy Orthologs (BUSCO) (Simão et al., 2015; Manni et al., 2021). BUSCO evaluates annotation completeness by looking for the presence of highly conserved ‘single-copy orthologs’ across specific taxonomic groups. For example, the BUSCO Eukaryota database expects 255 orthologs, while the Diptera database expects 3,285. For the M. abdita genome annotation we found 93% eukaryotic and 88% dipteran ‘complete single copy orthologs’, indicating a high-quality assembly (Table 2).

Table 2.

Genome annotation statistics (left) and quality metrics (right) for M. abdita. The quality metrics include the percentage of complete universal orthologs identified in the annotation across five BUSCO datasets. The number of genes in each lineage-specific BUSCO database is shown in parentheses.

M. abdita Genome Annotation Stasitics		Complete universal orthologs recovered in in M. abdita annotation (BUSCOs)
Number of coding genes	11,934	Eukaryota (n = 255)	93.3%
Number of mrnas	20,560	Metazoa (n = 954)	90.0%
Number of mrnas with 3’ & 5’ UTR	17,607	Insecta (n = 1,367)	91.6%
Mean mRNAs/gene	1.7	Endopterygota (n = 2,124)	90.6%
Mean gene length (bp)	19,478	Diptera (n = 3,285)	88.0%

The genome size of M. abdita is significantly larger than that of D. melanogaster (592.8 Mb vs 139.5 Mb). We also found the mean gene length of protein-coding genes in M. abdita to be significantly longer than that of D. melanogaster, ~19 kilobases (kb) vs ~6.9 kb respectively. However, the lengths of transcripts and exons are remarkably consistent between the two species indicating that M. abdita’s larger genome size is partially attributable to longer introns (Table S2).

Genome browser and genomic analysis tools

To improve accessibility and usability of the genomic data hosted on NCBI, we developed a genome browser ecosystem centered on JBrowse2 (Diesh et al. 2023). This ecosystem is available as a cloud image on NSF’s Jetstream2 platform (image: Megaselia abdita Genome Resources, Hancock et al. 2021^, Boerner et al. 2023), with a portable Docker image currently in development. The browser integrates tools for comprehensive genomic analysis, including BLAST search (SequenceServer2.0, Priyam et al. 2019), CRISPR guide RNA design (modified crisprDesigner, Beeber and Chain 2020), differential gene expression (DGE) analysis via R Shiny (freecount, Brooks et al. 2024), and synteny mapping (ShinySyn, Xiao and Lam 2022). Future updates will include expanded sgRNA profiling, Docker support for deployment on commercial cloud platforms, and continuing optimization of workflows, ensuring this resource remains a powerful and accessible tool for genomic research and education.

Synteny analysis reveals significant collinearity between M. abdita and D. melanogaster genomes, including HOX gene cluster arrangement

Analysis of chromosomal synteny can aid in the identification of orthologs and highlight genomic regions with conserved regulatory potential. To investigate genome-wide synteny between M. abdita and D. melanogaster, we performed a synteny analysis, identifying significant collinearity between their genomes, including the arrangement of the split HOX gene cluster. We compared the synteny and collinearity of M. abdita scaffolds with the D. melanogaster genome and identified 387 collinear blocks encompassing 2,396 genes (Figure 2A). Large portions of D. melanogaster chromosomes are syntenic with single scaffolds in M. abdita. Scaffold 1 of M. abdita largely corresponds to chromosome arm 3L, chromosome 4, and chromosome arm 2R of D. melanogaster, while scaffold 2 aligns with chromosome arm 2L and the distal portion of chromosome arm 3R, and scaffold 3 with the proximal portion of chromosome arm 3R and the X chromosome. The arrangement of the HOX genes in M. abdita mirrored that of D. melanogaster, with distinct Antennapedia and Ultrabithorax complexes separated by 53,306 kb in M. abdita (9,978 kb in D. melanogaster). Both complexes are located on Scaffold 2 of M. abdita’s genome and maintain the same gene order seen in D. melanogaster, including the cuticle gene complex (Figure 2B). Additionally, the genes zerknüllt (zen) and amalgam have undergone duplication in M. abdita (Figure 2B). Zen has experienced many duplications in Diptera (Mulhair and Holland 2024). Since only the ~60 amino acid homeodomain of zen is comparable between species, it is difficult to establish the relatedness of M. abdita’s zen-like to other zen genes and duplications by sequence alone.

Figure 2.

Synteny and Hox gene organization between D. melanogaster and M. abdita.

A) Synteny analysis between D. melanogaster chromosomes and M. abdita scaffolds. Lines represent groups of collinear genes, connecting their positions between D. melanogaster chromosomes and M. abdita chromosome-sized scaffolds. Colors correspond to D. melanogaster chromosomes. The ANT and UBX Hox gene complexes are highlighted by black lines within the chromosomes.

B) Visualization of M. abdita Hox gene clusters. The Antennapedia (top) and Ultrabithorax (bottom) complexes are both located on Scaffold 2. Each blue line indicates the genomic position and length of a gene.

M. abdita’s zen gene is expressed in the serosa and has been characterized in detail (Caroti et al., 2018; Kwan et al., 2016; Rafiqi et al., 2008; Rafiqi et al., 2010; Rafiqi et al., 2012; Stauber et al., 1999). Consistent with these studies, we detected the M. abdita’s zen transcript in embryonic stages 5 – 15, coinciding with the time when the serosa is specified and maintained (Figure S4). In contrast, the newly identified zen-like gene’s expression was only detected at stage 5 (Figure S4). Our findings underscore how a well-annotated genome can facilitate precise comparisons of chromosomal architecture which will enable the identification of both conserved regulatory landscapes and structural variation implicated in phenotypic diversity.

Major transition in transcriptional expression profile during germband retraction

The extent to which gene expression is similar or different between stages and species provides a basis for identifying developmental windows of accelerated change, heterochronic shifts, and evolutionary divergence. To identify conserved features and differences between embryonic stages of M. abdita and D. melanogaster, we performed RNA-seq on single embryos from stages 1, 5, 8, 9, 10, 12, 13, 15, 16, and 17 for both species (Figure S5). We staged embryos based on morphology, corresponding to established staging schemes for each species (Campos-Ortega & Hartenstein, 1997; Wotton et al., 2014). We later excluded the M. abdita stage 16 embryo due to failed sequencing. Despite similar embryo sizes and uniform library preparation and sequencing conditions, M. abdita embryos had roughly twice the number of reads per embryo compared to D. melanogaster (Figure S6A) though the total number of genes with reads assigned was comparable between species with 10,373 in D. melanogaster to 9,999 in M. abdita. The number of genes expressed was similar across embryonic stages between species as well (Figure S6B). The number of reads mapping to genes doubled in M. abdita (8,941 vs. 3,863 reads/gene), which was consistent with its 2x read count. We used normalized read counts (transcripts per million or TPM) to account for the global expression level differences, sample-to-sample variation, and differences in transcript length between genes. Our RNAseq data showed a major shift in transcriptional expression during germband retraction in both M. abdita and D. melanogaster. A multidimensional scaling (MDS) plot, which plotted samples based on the similarity of their top 500 most differentially expressed genes, revealed clustering of embryo transcriptomes before and after germband retraction (Figure 3A).

Figure 3.

MDS clustering and differential gene expression across developmental stages in M. abdita and D. melanogaster.

A) Multidimensional scaling (MDS) plots based on the top 500 most differentially expressed genes from single-embryo RNA-seq samples. The M. abdita data are shown on the left (blue) and the D. melanogaster data on the right (yellow). Points represent individual samples, and their colors correspond to groupings identified by k-means clustering. Arrows indicate key developmental events: germband retraction begins and ends.

B) Number of differentially expressed genes (DEGs) between sequential developmental stages. Comparisons (e.g., “1 vs 5”) indicate the number of DEGs identified between stage 1 and stage 5. Yellow bars represent D. melanogaster, and blue bars represent M. abdita.

To more closely look at the gene networks in both species, we identified the differentially expressed genes (DEGs) between developmental stages. Due to the absence of biological replicates for each embryo, we used a k-mean clustering approach to group samples and calculate a global dispersion estimate (σ = 0.36 for both M. abdita and D. melanogaster). Both species have high dispersion estimates but variance in expression was very similar in both datasets (Figure S7 & S8). Despite the high dispersion in data, we had sufficient power to detect DEGs with fold differences ±3 and p-value < 0.001.

We performed pairwise comparisons of each sequential stage: the zygote, containing maternal-deposited transcripts (Stage 1); cellularization, where maternal to zygotic transition (MZT) of gene expression occurs (Stage 5); gastrulation, germband extension & retraction (Stages 8–12); dorsal closure (Stages 13–16), and the final embryonic stage (stage 17). We detected 1,567 and 2,398 DEGs in M. abdita and D. in melanogaster respectively (Figure 3B). M. abdita exhibited more dynamic gene expression than D. melanogaster between cellularization and gastrulation (stages 5–8), with 418 DEGs in M. abdita compared to 123 in D. melanogaster (Figure 3B). Between stages 8 and 12, corresponding with the period of germband extension after gastrulation and germband retraction, there were markedly fewer DEGs in both species than in the following ~1 hour of developmental time between stages 12 (germband retraction) and 13 (beginning of dorsal closure), when both species showed a strikingly dynamic shift in gene expression (Figure 3B), albeit to a different extent. Between stages 12 and 13, D. melanogaster had 1,463 DEGs compared to M. abdita’s 564 DEGs (Figure 3B), indicating sharper increases and decreases in gene expression with the onset of dorsal closure in D. melanogaster than in M. abdita. The increased turning on and off of genes between stages 12 and 13 in both species also coincides with the phylotypic stage of development when evolutionary divergence between species is most constrained (Sander 1983; Kalinka et al., 2010), suggesting that dynamic changes in gene expression impose evolutionary constraints on gene regulatory networks.

To analyze the expression patterns of DEGs, we clustered them using DEGreports and plotted the expression of each cluster (Figure S9) (Pantano 2024). The clusters confirmed that the most dynamic transcriptional shift occurred during the transitions between stages 12 to 13 for both species (Figure 3B). We then assessed whether the genes expressed in each cluster were significantly enriched for any Biological Process Gene Ontology (BP GO) terms using the Search Tool for the Retrieval of Interacting Genes/Proteins (String) (Szklarczyk et al. 2023) (Figure S9). M. abdita’s clusters 4 and 9 and D. melanogaster’s clusters 6 and 10 were significantly enriched for embryonic, body plan, and systems development terms (Figure 4A & S10A) and contained many well-characterized developmental genes in Drosophila (Figure S10B&C). Genes found in M. abdita cluster 1 and D. melanogaster clusters 2 and 11, which exhibited a pronounced increase in expression at stage 13, were enriched for terms related to nervous system, muscle, and cuticle development (Figure 4B). In addition to these similarities, D. melanogaster’s more extreme expression changes between stages 12 and 13 (Figure 4 A–C) could reflect the fact that Drosophila employs a specialized tissue (amnioserosa) and evolutionarily novel mechanisms for dorsal closure (Schmidt-Ott and Kwan, 2022).

Figure 4.

Expression dynamics of gene clusters and enriched GO terms during embryogenesis in M. abdita and D. melanogaster.

A-C) Expression profiles of differentially expressed gene clusters during embryogenesis and their enriched Biological Process Gene Ontology (BP GO) terms for M. abdita (left - blue) and D. melanogaster (right - yellow). Grey lines show the expression profile of the individual genes within the cluster and bolded lines represent the average expression profile of the entire cluster. DEGreports generated cluster names (e.g., “Cluster 4”) which are arbitrary but retained here for continuity. Key embryonic developmental events: MZT + cellularization, gastrulation + germband extension, and germband retraction + dorsal closure are highlighted in yellow, blue, and red, respectively. Shared enriched BP GO terms for the boxed clusters are listed on the right.

Investigation of ‘orphan’ genes exclusively expressed during embryogenesis in M. abdita reveals novel F-Box LRR gene

As the evolution of new genes within lineages and species is an important mechanism of diversification of developmental mechanisms (Chen et al., 2010; Chen et al. 2013), we searched M. abdita’s genome for ‘orphan’ genes. Orphan genes are either highly diverged from known sequences or represent newly evolved genes (Vakirlis et al., 2020; Xia et al., in press). In M. abdita’s genome annotation, eggnog-mapper was unable to assign orthologs to 1,049 gene models. Further searches using blastn on the coding sequences and blastp on the predicted protein sequences did not reveal any sequence similarity to other Dipterans. Approximately 8.7% of M. abdita’s genes were ‘orphan’ genes.

We found that approximately 10% (109/1,049) of these orphan genes were expressed in embryos, including 28 that were expressed exclusively during embryogenesis (Figure 5A). Most of these genes (24) exhibited sharp expression peaks, with 14 peaking at stage 5 (cellularization), 4 at stage 12 (germband retraction), and smaller groups peaking at stages 13, 15, and 17. Four genes showed broader expression patterns, with two highly expressed during stages 8–10 (germband elongation) and two across stages 5–12 (cellularization to germband retraction).

Figure 5.

Expression and predicted structures of embryonically expressed orphan genes in M. abdita.

A) Expression profiles of orphan genes with expression limited to embryonic stages. E indicates embryonic, L indicates larval, P indicates pupal, and A indicates adult (multiple samples). Highlighted in pink and dark purple are two genes with high-confidence protein structure predictions. All other ‘orphan’ gene expression profiles are shown in black.

B) AlphaFold-predicted protein structure for Scaffold_1_222469999.1571.

C) AlphaFold-predicted protein structure for Scaffold_3_155577901.980. Sequence similarity searches suggest this gene encodes a novel F-box-LRR protein. In the structures, alpha helices are highlighted in blue, and beta sheets are highlighted in yellow.

To further characterize these 28 embryo-specific genes, we examined their open reading frames and protein sequences (Table S3). We classified 21 as likely coding and of those we further classified 9 as likely encoding stable proteins (blue highlighted rows in Table S3). Additional searches of InterProScan’s database revealed no known protein domains or any domains consistent with known transposases (Jones et al., 2014). We then used Alphafold2 to generate protein structure models (Jumper et al., 2021). Two of these genes resulted in confident structures (Table S3, pTM > 0.5). One of them could not be related to any known protein (Figure 5B); however, the second showed significant structural similarity to F-box leucine-rich repeat (LRR) proteins (Figure 5C).

We then compared our F-box LRR orphan protein sequence to known F-box-LRR genes in D. melanogaster but found no obvious ortholog. We identified orthologs to D. melanogaster F-box LRR genes Ppa, Kdm2, Fbxl4, Fbxl7, Fbl6, CG32085, CG9003, and CG8272. In total, we identified 15 F-box LRR genes in M. abdita (8 with clear D. melanogaster orthologs, 6 with orthologs in other dipteran species, and our orphan). One of the dipteran F-box LRR orthologs in M. abdita (Scaffold 3, geneID #3336) had an identical expression profile (stage 5) to the orphan F-box LRR.

F-box LRRs are components of the ‘E3 ubiquitin ligase SCF complex’ which ubiquitinates targeted proteins for later degradation by the cell. Specifically, the F-box domain binds to Skp1 and the LRR domain binds to the target, the molecule slated for ubiquitination and degradation. F-box LRRs are known to ubiquinate important developmental signaling molecules in D. melanogaster including the pair-rule gene paired (prd) which is bound by the F-box LRR protein Partner of paired (Ppa) (Raj et al., 2000). Ppa is unusual in that its expression is patterned rather than uniform as most F-Box LRR genes seem to be in D. melanogaster embryogenesis (Das et al., 2002). Given that in D. melanogaster, the 12 best-known F-box LRR proteins (Skp2, Ppa, Kdm2, FipoQ, Fbxl4, Fbxl7, Fbl6, CG32085, CG13766, CG12402, CG9003, CG8272) are expressed throughout embryogenesis according to both our RNAseq data and ENCODE gene expression data, the stage-restricted expression of M. abdita’s orphan F-box LRR gene and 13 other orphan genes with expression peaking before gastrulation might reflect previously overlooked developmental differences between M. abdita and D. melanogaster at the blastoderm stage.

Conclusions

The scuttle fly M. abdita is an important non-traditional model organism with hitherto very limited genomic resources. We have filled this gap by providing genomic and transcriptomic resources. By assembling a chromosome-level genome and annotating it, we revealed substantial chromosomal synteny with Drosophila melanogaster while uncovering many orphan genes. Additionally, our comparative transcriptome analysis across embryogenesis highlights conserved and divergent regulatory dynamics. The discovery of a novel F-box LRR gene, expressed exclusively during embryogenesis, underscores the potential of M. abdita to reveal new insights into the evolution of developmental gene networks. Ultimately, we hope the addition of these resources will further comparative research of developmental mechanisms in Diptera and continue the development of Diptera as a model taxon more broadly.

Materials and Methods

Generation of inbred M. abdita line

To generate single crosses of M. abdita, we collected pupae at the end of the pupal stage, when pupae darken approximately 1–2 days before eclosion, and transferred them to 35 × 10 mm petri dishes (Fisher Scientific, Cat. No. 50–820-644) until hatching. We monitored the plates every morning and every two hours during the day to collect virgin females. We paired each virgin female with a single male fly and allowed them to mate for two days in 35 × 10 mm petri dishes containing a gel solution made from 2% agar in water. On the third day, we prepared egg-laying vials by boiling 0.8 g of a fish food mixture—composed of 1 part spirulina flakes (Aquatic Eco-Systems Inc., Cat. No. ZSF5) and 2 parts sinking powder (Aquatic Eco-Systems Inc., Cat. No. F1A)—in 10 mL of a 0.8% agar solution (EMD Millipore, Cat. No. 1.01614). We added approximately 1 mL of this solution to 10 × 75 mm culture tubes (Fisher Scientific, Cat. No. 14–961-25) and allowed it to solidify. After cooling, we added 0.1 g of fish food on top using weighing paper, followed by 200 μL of water. We used a cotton swab to compact the food and clean any excess moisture from the sides of the tubes. We then transferred the mating pairs from the agar plates to the culture tubes and plugged the tubes with rayon balls (TIDI, Cat. No. 969162). See supplemental figure S11 for visual diagram of this process. We established multiple single crosses and tracked them using a progressive hierarchical code to identify the lineage. We conducted nine generations of sibling × sibling single-pair matings across three separate parallel lineages (A, B, and C). At generation six, we generated pooled crosses within each lineage to allow for the mixing of potentially lethal recessive alleles that may have accumulated during the single-cross procedure. Following this, we maintained the B lineage, as it exhibited higher overall fertility and health.

1.1 – 1.3. Genome Assembly

1.1. de novo library preparation, sequencing, and assembly

We generated a de novo reference genome for M. abdita with sequencing data from HiFi PacBio reads and Dovetail’s OmniC libraries (Cantata Bio). For HiFi PacBio sequencing, we collected and snap-froze in liquid nitrogen ~600 dechorionated embryos (mostly stages 16 and 17) from a 10-generation inbred line of a previously established laboratory culture of M. abdita Schmitz, 1959 (Schmidt-Ott et al., 1994). We sent these samples to Dovetail Genomics for HiFi PacBio library preparation and sequencing. Library preparation circularizes fragments so they can be read many times to generate a high-fidelity consensus sequence. Sequencing on the SMRT (Single Molecule, Real-Time) nanofluidic chip generates the long reads inherent to PacBio. PacBio generated 25.4 gigabase-pairs reads (42x coverage) from which Dovetail generated a haplotype-resolved draft assembly using the Hifiasm assembler (Hifiasm1 v0.15.4-r347 with default parameters) (Cheng et al. 2021). We used blobtools v1.1.1 to identify possible contamination and removed 33 scaffolds from the assembly (Challis et al., 2020; Laetsch and Blaxter, 2017). After filtering out haplotigs and contig overlaps with purge_dups v1.2.5, 89 scaffolds remained (Guan et al., 2020).

1.2. Omni-C library preparation and sequencing

We collected and snap-froze an additional ~600–700 dechorionated embryos and ~60 young pupae in liquid nitrogen from the same inbred line of M. abdita for Omni-C Library preparation and sequencing. Omni-C is a proprietary technology for long-range proximity ligation and sequencing of genomic libraries which captures spatial information within the genome through chromatin fixation and sequencing. Omni-C differs from other Hi-C preparations in that it digests the chromatin with a sequence-independent endonuclease. This eliminates biases inherent to competitive restriction enzyme-based approaches. Samples were treated with formaldehyde to fix the chromatin and then digested with DNAse I. The resulting ends were then repaired, and biotinylated bridge adapters were ligated to the ends. Subsequent steps involved proximity ligation of adapter-ligated ends, reversal of formaldehyde-induced crosslinks, and DNA purification. Non-internally ligated biotin residues were removed from the purified DNA. Sequencing libraries were prepared with NEBNext Ultra enzymes and Illumina-compatible adapters. Before PCR enrichment, biotin-containing fragments were isolated using streptavidin beads. Sequencing was performed on the Illumina HiSeqX platform.

1.3. Scaffolding assembly with HiRise

Both the Hifiasm draft assembly and Dovetail OmniC sequencing reads were used as input for HiRise, a software tailored for scaffolding genome assemblies with proximity ligation data (Putnam et al., 2016). Based on spatial data from the chromatin conformation capture (Figure S1), HiRise pinpoints regions where contigs are joined incorrectly (misjoins) in the initial assembly and utilizes this spatial information to re-orient contigs and construct larger scaffolds. The OmniC library sequences were aligned to the draft assembly using the Burrows-Wheeler Aligner (Li and Durbin, 2009). HiRise-analyzed read pairs are mapped to the draft scaffolds to develop a genomic distance likelihood model. This model is then used to identify and break potential misjoins, score prospective joins, and execute joins surpassing a set threshold. These scaffolds consist of sequentially arranged contigs separated by gaps. We used QUAST to calculate %GC, N50, L50, and Ns per 100 kbp (Gurevich et al., 2014). We then repeat-masked the genome with RepeatMasker (Smit et al., 2013). To estimate heterozygosity and sequencing error rates, we calculated the frequency spectrum of canonical 21-mers using jellyfish, and input the resulting histograms into GenomeScope (Marçais and Kingsford, 2011; Vurture et al., 2017). Using NUCmer (mummer v3.23 software package), we aligned the non-chromosome size scaffolds (4 – 15) back to the three chromosome-size scaffolds (1 – 3) and found repetitive sequences present in scaffolds 4 – 15 (Kurtz et al., 2004; Marçais et al., 2018).

2.1 – 2.3. Genome Annotation

2.1. Repeat masking the genome

To generate a custom library for repeat masking, we ran RepeatModeler v2.0.4 on the genome to find/model potential repeats (Smit and Hubley, 2008). We then used RepeatMasker’s script fambd.py v0.4.3 to extract Arthropoda records from the dfam database and combined these sequences with the RepeatModeler output to use as the repeat library. We used RepeatMasker v4.1.5 to soft-masked the genome with our custom library of repetitive low-complexity DNA and transposable elements (Smit et al., 2013).

2.2. Data used as evidence of genome features

We downloaded available Illumina RNAseq data for M. abdita from NCBI which included paired end reads from 3 adults and pooled embryos (Table S4). We generated RNAseq data for 9 precisely staged embryos, first and third instar larval stages, and a 1-day-old pupal stage (Table S4). For protein evidence, we downloaded all Dipteran protein sequences from NCBI’s RefSeq database which included 2,122,027 sequences from 617 species (Sayer et al., 2022). We also downloaded Uniprot’s complete protein sequence file (UniProt Release 2023_04) which contains 570,157 sequences from 14,509 species (The UniProt Consortium, 2023).

2.3. Annotation pipeline

We annotated the reference genome of M. abdita using evidence from all RNAseq transcripts (Table S4), protein sequences, and gene prediction software (Figure S3). The choices of software used in this pipeline are based on the methods section of VanKuren’s ‘Draft Papilio alphenor assembly and annotation’ (VanKuren 2023). First, we assembled RNA transcripts using two transcript assemblers: Stringtie v2.2.1 and Trinity v2.15.1 (Pertea et al, 2015; Grabherr et al., 2011; Haas et al., 2013). Trinity performs both genome-guided and de novo assembly; we assembled transcripts with both methods. For Trinity and Stringtie’s genome-guided assemblies, we first mapped reads to the genome with ‘Spliced Transcripts Alignment to a Reference’ software (STAR v2.7.10b) (Dobin et al., 2013). Next, we used the ‘Program to Assemble Spliced Alignments’ (PASA v2.5.3) to identify gene structures from all three assemblies (Haas et al., 2003). We predicted gene models directly from the PASA assemblies using the PASA plug-in TransDecoder v5.7.1 which identifies candidate coding regions from Trinity and StringTie assemblies. To create protein alignments, we used the software Exonerate v2.2.0 which maps protein sequences to the genome (Slater and Birney, 2005). We used the BRAKER3 pipeline (braker.pl v3.0.6) to predict gene models from mapped RNAseq reads and protein data (Gabriel et al., 2023). BRAKER3 relies on the software Augustus and GeneMark to predict gene models (Stanke et al., 2006; Brůna et al., 2020). We also generated our own ab initio gene structure predictions, using GlimmerHMM v3.0.4 (Majoros et al., 2004). First, we collected “hints” for training the ab initio predictors by extracting protein-coding hints from the protein alignments using Augustus’ exonerate2hints function, intron hints from mapped RNA seq reads using Augustus’ bam2hints function, and exon/intron hints from the PASA assemblies using Augustus’ bam2exonhints function. We additionally used the coding predictions from Transdecoder to create training models and further refined those models using the lib.selectTrainingModels function from Funannotate (Palmer and Stajich, 2019). This training data was used to run GlimmerHMM with hints as guidance (Majoros et al., 2004). We then provided the PASA assemblies, mapped protein data, TransDecoder predictions, ab initio predictions, BRAKER3 predictions, and a file weighting each line of evidence to the software Evidence Modeler v2.1.0. Evidence Modeler constructed the consensus gene structures which were updated by PASA to add UTRs and identify alternative transcripts (Haas et al., 2008). We removed gene structures that overlapped with RepeatMasker output using a Funannotate function called “RemoveBadModels.” We then used Funannotate v1.8.1 to identify annotations that match known transposable elements (TEs) and repeat proteins using BLAST and updated the annotation to remove remaining TEs. We used AGAT v1.4.1 to remove genes with an open reading frame (ORF) < 100 amino acids in length and any associated gene structures (Dainat, 2022). We assigned gene names using eggNOG-mapper v2.1.12 which relies on orthology predictions to functionally annotate genes (Cantalapiedra et al., 2021; Huerta-Cepas et al., 2019).

Synteny Analysis

We used MCScanX (primary release) to compare the synteny of M. abdita and D. melanogaster genomes (Wang et al., 2012). MCScanX identifies syntenic blocks based on a score given to each gene pair. We set the match_score = 50 (default), match size = 5 (number of genes required to constitute a syntenic block), gap_pentaly = 0 (no penalty for gaps), and max_gaps = 100. We used the output of this run to generate our synteny map. We used the circlize R package to plot the results (Gu et al., 2014).

Embryo Staging for RNAseq

For precise embryo staging, embryos of the appropriate age were mounted on a microscope slide under halocarbon 27 oil and observed in a Zeiss Axiophot compound microscope equipped with a 10x objective and DIC (differential interference contrast) optics until they reached the desired stage. They were photographed and immediately processed for RNA extraction as previously described (Lott et al., 2014). We collected M. abdita and stage-matched D. melanogaster embryos at stages 1, 5, 8, 9, 10, 12, 13, 15, 16, and 17. Photos of the sequenced embryos can be found in Figure S5.

RNA isolation and Sequencing

We incubated each sample for 5–10 minutes at room temperature in TRIzol. We then froze each sample in TRIzol at −80° C. We extracted total RNA using the TRIzol/Phenol-chloroform protocol detailed in the appendix (Protocol 10) of ‘Functional evolution of a morphogenetic gradient’ by Chun Wai Kwan. The University of Chicago genomics core facility constructed cDNA libraries using the TruSeq kit with PCR (Illumina, CA, USA). The cDNA libraries were barcoded and multiplexed for 100bp paired-end sequencing on one lane of a HiSeq Illumina 2000 sequencer.

Differential Gene Expression Analysis

M. abdita had ~2x the reads for each sample compared to D. melanogaster (Figure S6A). We chose not to down sample M. abdita reads to match D. melanogaster to avoid losing power to detect changes between genes within M. abdita developmental stages. We justified this by looking for any evidence that the higher number of reads skewed the relationship between gene expression and variance, but found that the relationship between mean expression and variance in expression is similar in both species without subsetting the data (Figure S7). Additionally, the average, median, and variance in gene expression (TPM) were very similar across development in both species (Figure S8).

We aligned M. abdita RNAseq reads to the reference genome generated in this publication and D. melanogaster RNAseq reads to D. melanogaster’s genome (Accession: GCF_000001215.4). We used Subread v2.0.5 to align reads (Liao et al., 2013). 90–99% of reads mapped to the genome for each sample. We input the aligned bam files into Subread’s featureCount function which assigns the reads to a genomic feature from the annotation file (gff). At this point, we performed the analysis in Rstudio using the free and open source statistical language R and various R packages including EdgeR, tidyr, dplyr, and ggplot2 (R Core Team, 2023; Robinson et al., 2009; Posit Team, 2024; Wickham et al. 2016; Wickham et al., 2023^; Wickham et al., 2024). We used EdgeR to assess gene expression differences between samples. We filtered out lowly expressed genes and normalized data by library size (TMM normalization) for both species. We calculated counts per million (CPM) to normalize the difference in raw reads between samples and then calculated transcript per million (TPM) to account for differences in gene length. We calculated gene length as the coding sequence length for each gene’s longest isoform. All TPM expression data are plotted as log₂(TPM + 1).

As our RNA-seq samples do not include biological replicates, we estimated the squared coefficient of variation (BCV) using k-means clustering of the samples. We began by calculating a distance matrix for the samples and extracting the first four eigenvectors, which together explained >95% of the variance. To determine the optimal number of clusters (k), we calculated the within-cluster sum of squared errors (WSS) and Silhouette scores for k values ranging from 1 to 8, selecting k = 5 based on these metrics. For M. abdita, clustering with k = 5 grouped the embryonic stages as follows: Group 1 (stage 1), Group 2 (stage 5), Group 3 (stages 8–12), Group 4 (stages 13–15), and Group 5 (stage 17). For D. melanogaster, the clusters were: Group 1 (stage 1), Group 2 (stages 5 and 8), Group 3 (stages 9–12), Group 4 (stages 13–16), and Group 5 (stage 17). Using these groupings, we estimated the dispersion as σ=0.36 for both M. abdita and D. melanogaster. These estimates were applied globally. Differential expression analysis was performed using EdgeR’s exactTest() function, comparing gene expression between pairwise embryonic stages rather than the k-means groups. Genes were considered differentially expressed if they met the criteria: fold change < −3 or > 3 and p-value < 0.001. We then took all differentially expressed genes (DEGs) and input that expression data into DEGreports pattern() function. DEGreports clusters DEGs based on expression profile similarity (Pantano 2024). We used these clusters for gene ontology enrichment analysis described below.

Gene Ontology Enrichment Analysis

We used the STRING database v12.0 (Search Tool for the Retrieval of Interacting Genes/Proteins) to perform enrichment analysis of Biological Process Gene Ontology (GO) terms for our clustered DEG lists (Szklarczyk et al., 2023). STRING compares input gene sets to a reference genome to identify networks of interacting genes and enrichments in biological processes. Since M. abdita is not available in STRING, we first used NCBI’s Blast tool to select the top D. melanogaster ortholog match/hit for each M. abdita gene (Altschul et al., 1990). Using FlyBase’s batch download tool, we retrieved the corresponding FlyBase IDs, which STRING accepts as input (Öztürk-Çolak et al., 2024). STRING performed the enrichment analysis identifying the Biological Process GO terms significantly associated with each developmental stage. STRING measures enrichment based on the strength of enrichment (Log₁₀(observed/expected)), false discovery rate (p-values corrected for multiple testing with Benjamini-Hochberg), and the signal (weighted harmonic mean between observed/expected ratio and −log(FDR).

‘Orphan’ gene identification

We identified genes from M. abdita’s annotation file for which eggNOG-mapper could not assign an ortholog. Next, we assessed the expression of these genes in our RNA-seq data. To identify genes with exclusively embryonic expression, we focused on those with a CPM > 1 in at least one of the nine embryonic stages and a CPM < 1 in pupal, larval, and adult stages. We validated these genes further by analyzing their ORFs using NCBI’s ‘Open reading frame finder’. We used CPC2 to evaluate the nucleic acid sequences to assess coding potential (Kang et al., 2017). Finally, we performed sequence similarity searches using NCBI’s blastn and blastp tools, querying both nucleotide and protein sequences against the entire NCBI database as well as against Dipteran-specific sequences (Altschul et al., 1990; Sayer et al., 2022).

Protein structure prediction and structural similarity search

We used Protparam to assess protein stability, aliphatic index, and hydropathicity of predicted proteins (Gasteiger et al., 2005). We then searched all predicted protein sequences against InterProScan to look for any missed domains, specifically to look for evidence of transposable elements that were not discovered with blast searches (Jones et al., 2014). We used the AlphaFold2.ipynb provided by ColabFold v1.5.5 to predict protein structures (Jumper et al., 2021; Mirdita et al., 2022;). If the predicted protein had a predicted template modeling (pTM) score > 0.5 we then uploaded the structure to FoldSeek’s website and searched the available databases (AlphaFold/Proteome, AlphaFold/Swiss-Prot, AlphaFold/UniProt50, BFMD) for proteins with similar structure (van Kempen et al., 2024; Varadi et al., 2022; Varadi et al., 2024). We used Protein Imager to generate publication-quality images of protein structures (Tomasello et al., 2020).

Funding Statement

Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number R01GM127366 and the University of Chicago. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data and Resource Availability

Genome:

The annotated Megaselia abdita genome can be found in NCBI’s Genome database under BioProject Accession PRJNA1164289.

RNASeq data:

All fastq files containing the raw sequencing reads for each sample have been uploaded to NCBI’s Sequence Read Archive (SRA) and can be found under the BioProject Accession PRJNA1200075. Individual BioSample and SRA accession numbers can be found in supplementary table S4 including those samples that were not generated in this study but used as evidence for the annotation of the genome.

Genome Browser:

Jetstream2 at Indiana University is a resource provider for NSF’s ACCESS program which aims to broaden access to super computing resources at no cost to researchers. To access the Megaselia abdita genome browser and related tools, create an ACCESS ID, then use this ID create an account and login to Jetstream2. You will apply for an ‘allocation’ of credits which can be used on Jetstream2. Detailed instructions on the use of Jetstream2 can be found at https://jetstream-cloud.org/get-started/index.html