DrosOMA: the Drosophila Orthologous Matrix browser

Species selection and annotation sources

Drosophila species with high quality and complete assembled and annotated genomes were selected for inclusion in DrosOMA so as to sample broadly across the genus. Of more than 350 assemblies representing some 140 species at the United States National Center for Biotechnology Information (NCBI), genome annotations were available for 49 species (Sayers et al. 2023). Protein-coding gene annotations for D. melanogaster were sourced from FlyBase (Gramates et al. 2022). All of the source data are available publicly - the accession numbers and version numbers are all given in Table 1. Assessments of completeness performed using Benchmarking Universal Single-Copy Orthologues (BUSCO) (R.M. Waterhouse et al. 2018; Manni et al. 2021) v5.4.0 and sourced from the A³Cat Arthropoda Assembly Assessment Catalogue (Feron and Waterhouse 2022) were used to select only annotated assemblies with Diptera-level BUSCO completeness scores of more than 95%. Filtering to reduce sampling of closely related species resulted in a final set of 36 Drosophila species with high-quality annotated assemblies for orthology delineation, as well as four outgroup mosquito species (Table 1).

Orthology delineation using OMA

All annotated protein-coding genes from the 40 selected species were used as input for delineating orthologous groups for DrosOMA. Briefly, orthology delineation using the OMA Standalone inference algorithm consists of three main stages (Altenhoff et al. 2019, 2021). Firstly, all-against-all Smith-Waterman sequence alignments are computed using the SWPS3 vectorized implementation of the Smith-Waterman local alignment algorithm and significant matches are retained to define homologous proteins (i.e. sequences with a common ancestry). Before inferring orthology, one representative protein per gene is selected. OMA Standalone uses all isoforms for the first all-against-all alignment stage and selects as the reference protein the isoform with the best matches across all species (this can be considered as the most evolutionarily conserved isoform). Secondly, mutually closest homologues between species pairs are identified based on evolutionary distances to infer orthologous pairs (i.e. homologues related through speciation), while accounting for distance inference uncertainties and for potential differential gene losses. Finally, all identified orthologous pairs are clustered using two different approaches to produce catalogues of OMA Groups and Hierarchical Orthologous Groups (HOGs) (Zahn-Zabal et al. 2020). HOGs are defined as sets of genes that descended from a single ancestral gene at a given taxonomic range. These sets correspond to the idea of subfamilies for a given taxonomic range and can contain more than one gene from a species, i.e. inparalogues. OMA Groups on the other hand are sets of orthologues where each gene is orthologous to one another. The history of such sets should correspond to the species phylogeny and hence they are especially useful as markers to reconstruct the species phylogeny. For this dataset the production pipeline of OMA was employed, but the same clustering can also be performed using OMA standalone. In order to build the browsable DrosOMA instance, the OMA orthologues were converted using the oma2hdf command from the pyoma python package into an HDF5 database. CATH domain annotations (Sillitoe et al. 2021) were computed using the cath-tools v0.16.10 package and with the provided hmm models from CATH release 4.2. Protein cross-references were added by matching the sequences against the full UniProtKB and RefSeq databases, requiring exact matches.

Web server virtual machine configuration and setup

The OMA browser instance for DrosOMA was set up and is hosted on a virtual machine using docker containers. The virtual machine requires relatively modest resources, i.e. 2 CPUs clocked at 2.25 GHz each, 8 GB RAM and 25 GB storage. The docker images for the OMA Browser were created from the pyomabrowser repository (https://github.com/DessimozLab/pyomabrowser) following the steps described in https://zoo.cs.ucl.ac.uk/doc/pyomabrowser/setup.html. Before building the docker images, the following aspects of the OMA Browser web interface were adjusted in order to make it a Drosophila-specific instance: We removed all the instances of non-drosophila proteins in the search examples by adjusting the Django templates in oma/templates, oma/test/ and oma_rest/. Similarly, we changed the OMA logo by replacing the corresponding file in/oma/static/image. These customisations are mostly cosmetic changes that will make the service more user friendly, and are not strictly needed for website functionality. Finally, paths, deployment type, and rabbitmq/celery credentials were adjusted and hosts were allowed in for_docker/env.

Species phylogeny reconstruction

The species tree was computed using single-copy orthologues identified during the BUSCO completeness assessments of the genomes of the species selected for inclusion in DrosOMA. The protein sequences of BUSCO genes found in at least 38 of the 40 species were aligned using MUSCLE 3.8.1551 (Edgar 2004) with default settings and subsequently trimmed to retain well-aligned regions using TrimAl (Capella-Gutiérrez et al. 2009) with the “-strictplus” option. The 2,891 alignments were merged to build a 40-species concatenated superalignment (1,581,953 columns; 683,285 distinct patterns; 658,691 parsimony-informative; 180,333 singleton sites; 742,929 constant sites) used as input for phylogeny reconstruction using IQ-TREE 2.2.0-beta (Nguyen et al. 2015) with 1,000 bootstrap samples (options: -msub nuclear -B 1000 -bnni). The molecular species phylogeny was time-calibrated by providing calibration dates for the Diptera root, Culicidae, Drosophilini, willistoni-melanogaster ancestor, and navojoa-albomicans ancestor, from the TimeTree database (Kumar et al. 2022) to the functions makeChronosCalib() and chronos(), from the ape R package (Paradis and Schliep 2019), and plotted using the ggtree R package (Yu 2023).

Implementation

The DrosOMA Drosophila Orthologous Matrix browser implements for users a feature-rich web interface to explore the results of orthology inference amongst complete genomes. The service is implemented with the django framework, a high-level Python web framework that encourages rapid development and clean, pragmatic design.

Operation

The DrosOMA Drosophila Orthologous Matrix browser operates on standard up-to-date web browsers including Google Chrome, Mozilla Firefox, and Apple Safari. The operational setup of an OMA browser instance such as DrosOMA requires a host that runs docker containers orchestrated with docker compose.

Orthologous groups delineated across 36 Drosophila species

Applying OMA orthology delineation to the protein-coding genes from 36 drosophilids and four outgroup mosquito species (see Methods) resulted in the clustering of 93.5% of proteins in OMA Groups and 95.6% in Hierarchical Orthologous Groups (HOGs), with almost 25,000 HOGs at the last common ancestor of all DrosOMA species (Table 2). The OMA Groups are cliques of orthologues based on the orthology graph, meaning that all the components (proteins) of an OMA Group are connected to each other through pairwise orthologous relationships. Although all members of the OMA Groups are orthologous to all other members of the same group, OMA group members are not necessarily 1-to-1 orthologues. The OMA HOGs comprise sets of proteins encoded by genes descended from a common ancestral gene in the last common ancestor of a set of species (i.e. at a specific taxonomic level in the species phylogeny). The “hierarchical” nature of HOGs is due to their being defined with respect to specific clades within the species tree, so HOGs are nested subfamilies with groups delineated for younger radiations being encompassed within larger HOGs defined at older nodes. DrosOMA contains HOGs delineated at the root, three mosquito nodes, and 13 drosophilid nodes including Sophophora, the melanogaster group, and the melanogaster subgroup.

The fully-resolved time-calibrated species phylogeny (see Methods) defines the relationships amongst the 36 Drosophila species and the outgroup mosquitoes over approximately 260 million years of evolution (Figure 1). Analysis of the root-level HOGs shows counts of proteins per species belonging to universal single-copy HOGs (9.8% of HOGs; 17.1% of proteins), universal but variable-copy-number HOGs (19.6% of proteins), non-universal HOGs with outgroup species orthologues (13.7% of proteins), as well as drosophilid-specific HOGs with orthologues from all (16.8% of proteins), the majority (17.9% of proteins), or the minority (7.7% of proteins) of the 36 Drosophila species. This leaves an average of 527±392 proteins per drosophilid species with no identifiable orthologues, i.e. annotated protein-coding genes that, given the set of species under consideration, appear to be species-specific with no traceable common ancestry.

Figure 1. Species phylogeny and orthology classifications across 36 Drosophila and four outgroup species.

The time-calibrated species phylogeny (left) shows the estimated evolutionary relationships amongst the set of 40 species spanning approximately 60 million years since the last common ancestor of the Drosophila genus. The dashed line indicates the Drosophila and Culicidae last common ancestor but for visualisation is not placed according to the timescale. The barchart (right) shows counts of genes per species categorised according to their orthology type based on root-level hierarchical orthologous groups (HOGs). Analysis of the root-level HOGs shows counts of proteins per species belonging to universal single-copy HOGs (Single-copy All), universal but variable-copy-number HOGs (Orthologues All), non-universal HOGs with outgroup species orthologues (Drosophila & Culicidae), mosquito-only orthologues (Culicidae Only), as well as drosophilid-specific HOGs with orthologues from all (Drosophila All), the majority (Drosophila Majority), or the minority (Drosophila Minority) of the 36 Drosophila species. This leaves an average of 527 ±392 proteins per drosophilid species with no identifiable orthologues, i.e. annotated protein-coding genes that, given the set of species under consideration, appear to be species-specific with no traceable common ancestry. Branch lengths are shown in millions of years; all nodes received 100% bootstrap support except * with 95%; D. Drosophila; An. Anopheles; Ae. Aedes; Minority <18 drosophilids; Majority ≥18 drosophilids.

Orthology data exploration using the DrosOMA browser

As DrosOMA uses the same database and interface design and architecture as the OMA browser (Altenhoff et al. 2021), an extensive array of data querying and visualisation options are available to the user. Searches may be performed using gene or protein names, descriptors, or identifiers, or protein sequences, and extensive cross-referencing to public databases allows for searches using identifiers from resources such as UniProt (The UniProt Consortium et al. 2023), RefSeq (O’Leary et al. 2016), EntrezGene (Sayers et al. 2023), Swiss Model (A. Waterhouse et al. 2018), STRING (Szklarczyk et al. 2023), and Bgee (Bastian et al. 2021), in addition to the source FlyBase and NCBI identifiers and annotations (Figure 2A). Search result visualisations are focused on the three main data types, i.e. with views for genomes, groups (Figure 2B), or genes (Figure 2C). Genome-view pages summarise available information per species, e.g. a list of all their genes and of their most closely related species, as well as tools for building pairwise global synteny visualisations. Group-view pages display information about OMA Groups or HOGs, showing filterable lists of member genes with their associated cross-referenced identifiers and cartoon views of protein domain architectures, as well as visualisations of HOG members guided by the species phylogeny. Gene-view pages display information associated with a gene and its protein products, including sequences (protein and cDNA), cross references to other public databases, and available functional annotations in the form of Gene Ontology terms (The Gene Ontology Consortium et al. 2021).

Figure 2. Example orthologous group and gene information views available from the DrosOMA browser.

(A) The simple search entry point for DrosOMA allows for text searches with gene names, descriptors, or identifiers, as well as with protein sequences. (B) Visualising information for Hierarchical Orthologous Groups (HOGs) can be guided by the species phylogeny (left) showing counts of orthologues per species, or as a table (right) with protein identifiers and cartoons showing domain architectures. (C) The gene view page displays available information for genes of interest and their mappings to external databases.

Other useful search, visualisation, and download features are described in the DrosOMA “Explore”, “Tools”, “Download”, and “Help” pages, with several examples and explanations for the general use of the OMA browser elaborated in a dedicated primer (Zahn-Zabal et al. 2020). Examples of these extended features include sequence alignment tools (Figure 3A) and local synteny visualisations (Figure 3B). For both OMA Groups and HOGs, the browser can generate multiple sequence alignments of the member proteins that can further be sorted, filtered, edited, and exported by users, for example, to use as inputs for building gene trees for orthologous groups of interest. Synteny, or how orthologues have maintained or shuffled their genomic arrangements throughout evolution, can be visualised at a local level (e.g. from a context of 9 to 19 orthologues) or at global level (along entire chromosomes for pairs of species), both based on comparing the relative genomic positions of orthologues across the species under consideration.

Figure 3. Example additional analysis views available from the DrosOMA browser.

(A) Multiple sequence alignments of proteins from hierarchical orthologous groups (HOGs) or OMA Groups can be generated, visualised, explored, and downloaded using the DrosOMA Browser. (B) Local gene synteny conservation can be visualised to explore how orthologues have maintained or shuffled their local arrangements in the genomes of each considered species.