Using a modified genomic DNA extraction protocol for filamentous fungi in the genus Trichoderma (Elya and Lee, 2022), we extracted high molecular weight DNA from a in vitro-grown Entomophthora muscae culture inoculated from a single sporulating E. muscae-killed fruit fly. Oxford PromethION sequencing of the resultant genomic library yielded 85.71 Gb of sequenced bases (N50 = 9.54 kb) that passed quality control thresholds (Fig. 1B). These reads were assembled with Flye (v2.8.3) and self-polished using Medaka (v1.2.6) to yield a 1.03 Gb genome consisting of 7,810 contigs (N50 = 301.1 kb). Additional scaffolding using 10x Genomics linked-read data (SRR18312935) ((Bronski et al., 2018) resulted in a final contig count of 7,810 (N50 = 329.6 kb) (Fig. 1B). The assembly was annotated with a custom Funannotate-based pipeline, which included assignment of putative protein functions based on a number of enzyme (e.g., MEROPs (Rawlings et al., 2018), CAZy (Drula et al., 2022)), protein family (e.g., Pfam (Mistry et al., 2021), InterPro (Paysan-Lafosse et al., 2023)) and ontology (e.g., GO (Ashburner et al., 2000; Gene Ontology Consortium et al., 2023), EggNog (Huerta-Cepas et al., 2019), COG (Galperin et al., 2021; Tatusov et al., 2000)) databases as well as prediction algorithms for secretion signals (SignalP (Almagro Armenteros et al., 2019)) and transmembrane domains (see Methods for full details). The final annotation predicted 39,703 genes and 42,657 total transcripts (Fig. 1B).

To glean biological insights from this new genome, we used a comparative approach to place our new genome in the context of other newly-available entomophthoralean data. We collated a dataset consisting of the latest available genomic and transcriptomic data for entomophthoralean fungi, including genomes for the specialist pathogens E. muscae (this study), Entomophaga maimaiga (ARSEF 7190 v1.0) and Massospora cicadina (Stajich JE. et al., 2022), transcriptomes for the specialists Pandora formicae (Małagocka et al., 2015) and Strongwellsea castrans (this study) and genomes for the generalists Zoophthora radicans (Amses et al., 2022), Neoconidiobolus thromboides (Chang et al., 2022) and C. coronatus (NRRL 28638; Fig. 1C).

To assess completeness of these datasets, their corresponding predicted proteomes were analyzed for the presence of Benchmarking Universal Single-Copy Orthologs (BUSCOs) using a fungal reference set (odb10) (Manni et al., 2021). For E. muscae, this analysis detected 81.3% of complete fungal BUSCOs and 5.4% fragmented BUSCOs in the E. muscae proteome; 13.3% of fungal BUSCOs were not detected (Fig. 1D). For the other entomophthoraleans, percentage of complete-copy BUSCOs ranged from 50.3% (M. cicadina) to 87.4% (S. castrans), with a median value of 82.4% across this group (Fig. 1D). Notably only the E. muscae genome had more double-copy (58.2%) than single-copy (23.1%) BUSCOs. The next highest proportions of double-copy BUSCOs were found in the transcriptomic datasets of P. formicae (31.1%) and S. castrans (29.7%) which includes partial transcripts and isoforms in the duplication count. Based on BUSCO scores and phylogenetic positions, we selected three other genomes aside from E. muscae to use as a core set for genomic analysis: Entomophaga maimaiga, Z. radicans and N. thromboides (Fig. 1C).

While the high number of duplicate BUSCOs in the transcriptomes could be accounted for by contaminant host transcripts (each of these transcriptomes was assembled from fungus-killed hosts), this explanation could not account for the high number of duplicate BUSCOs in E. muscae, which was assembled from the DNA of fungal cells cultured in vitro. One possible explanation for many duplicate BUSCOs in E. muscae is that the assembled genome was functionally diploid, not haploid, which has been suggested previously for another E. muscae isolate from houseflies (Musca domestica) (De Fine Licht et al. 2017). To address this, we counted the occurrence of kmers (29-mers or 33-mers) throughout our assembly (Jellyfish v.2.3.0 (Marçais and Kingsford, 2011)) and plotted the coverage of kmers against their observed frequency (GenomeScope v2.0 (Ranallo-Benavidez et al., 2020)). We observed a large peak ∼50x coverage, corresponding to single copy (haploid) kmers and a smaller peak at ∼100x coverage, corresponding to dual copy (diploid) kmers. This small ∼100x peak suggests that our assembly is heterozygous at a low level (0.023%, 0.026% respectively) but does not support the hypothesis that our genome is fully diploid (Fig. 1-S1A,B). We also addressed the possibility that differences in how genomes were annotated (i.e., annotation “pipeline”) could lead to inflated gene counts, owing to differences in the stringency of calling genes. In order to address this, we annotated the genomes for E. maimaga and Z. radicans, which had been annotated with other pipelines, with the pipeline used for E. muscae. While our pipeline did predict more genes than the original annotation pipelines for these assemblies (23,807 vs. 14,701 for E. maimaga; 18,761 vs. 14,479 for Z. radicans), this difference could not fully account for the discrepancy in gene count (Supplemental Table 1). This suggested that differences in annotation approaches are not sufficient to explain the high duplicate gene count in E. muscae.

At 1.03 Gb, E. muscae’s genome is very large compared to the vast majority of other fungal genomes (Fig. 2A). Similarly, genomes of the most closely-related entomophthoralean fungi for which genomes are available (M. cicadina, E. maimaiga and Z. radicans) are also extremely large, each exceeding 500 Mb. However, the impressive sizes of E. muscae and other entomophthoralean genomes do not appear to be a consequence of increased gene number (Fig. 2B). The number of genes predicted in entomophthoralean genomes ranges from 8,867 (N. thromboides) to 14,701 (E. maimaiga) to 39,711 (E. muscae). Gene counts for M. cicadina are also within the low end of this range (7,532), but this number is expected to be an underestimate given the fragmented nature of the assembly. In comparison, fungi with genomes between 10-100 Mb have between ∼3,000-33,000 genes. Thus, as in other fungi whose genome sizes rank within the top 1% (e.g., the mycorrhizal fungi Gigaspora margarita and G. rosea, the rusts Phakospora pachyrhizi, Austropuccinia psidii and Hemileia vastatrix and the bioluminescent mushroom Mycena olivaceomarginata), the predicted number of genes in entomophthoralean fungi is within the range observed for fungi with more typically-sized genomes (<100 Mb).

Unlike gene content, repeat content across entomophthoralean genomes reveals a clear trend: fungi that shared a common ancestor after divergence from Conidiobolus have highly repetitive genomes (Fig. 2C). The genomes of E. muscae, E. maimaiga and M. cicadina consist of ∼90% repeated sequences (90.9%, 90% and 92.4%, respectively); for Z. radicans the proportion of repeated sequences genome-wide is about 20% less, at 71%. Meanwhile, N. thromboides’ genome is predominantly non-repetitive (only 13.5% sequences are repeated). The bulk of this repetitive content consists of transposable elements, which are categorized as Class I or Class II based on their mechanism of action. Class I transposons (retrotransposons) “jump” via transcription to an RNA intermediate that is then reverse transcribed to DNA and integrated in a target genomic site. Class II transposons (DNA transposons) “jump” when they are excised from donor DNA and reinserted into a target genomic site. The majority of repeat content across these four family Entomophthoraceae genomes consists of Class I retrotransposons, specifically Ty3 (formerly called Gypsy (Wei et al., 2022)) long terminal repeat (LTR) retrotransposons. Ty3 elements comprise 39.1%, 43.2%, 56.8% and 38.0% of the E. muscae, E. maimaiga, M. cicadina and Z. radicans genomes, respectively, while only 0.01% of the N. thromboides genome. In addition, the most closely-related species pair in this set, E. muscae and E. maimaiga, also have a sizeable fraction (12% and 13.9%, respectively) of Tc1 mariner DNA (Class II) transposons populating their genomes. M. cicadina is unique among the other entomophthoralean fungi in having a substantial fraction (6.3%) of L1 long, interspersed nuclear elements (LINEs) while effectively lacking any DNA transposons.

Kimura divergence estimates for most populous repeat elements of the E. muscae, E. maimaiga, M. cicadina, Z. radicans and N. thromboides genomes show trends consistent with phylogenetic relationships between these fungi (Fig. 2D). All but N. thromboides have expanded Ty3 and unknown LTR elements with comparable Kimura divergences (∼0-50), indicating similar proliferation of these elements occurred with approximately concurrent timing. M. cicadina uniquely shows a burst of Ty3 LTR expansion around divergence time 20 and an expansion of LINE elements, occurring after Ty3 expansion. E. muscae and E. maimaiga show an increase in TcMar elements more recently than either of the Ty3 and LINE expansions. In addition, E. muscae, E. maimaiga and Z. radicans share an expansion of DNA-MULE-MuDR elements that is not observed in M. cicadina. All five genomes also possess putative repeat elements currently classified as “Unknown” beginning roughly concurrently.

Fungi are known to protect against the proliferation of TEs by mutating cytosine to thymine in repetitive regions via a process called Repeat Induced Point (RIP) mutations (Selker, 2002; van Wyk et al., 2020). This process occurs during meiosis in fungi from Ascomycota and Basidiomycota, where homology between DNA regions directs this transition (van Wyk et al., 2020). We calculated dinucleotide indices to look for evidence of RIP in entomophthoralean fungi, along with Neurospora crassa, an ascomycete fungus in which RIP has been extensively documented (Freitag et al., 2002; Galagan et al., 2003; Lewis et al., 2009; Selker, 2002), and Austropuccinia psidii (Tobias et al., 2021), a basidiomycete rust with a genome of similar size to entomphthoralean fungi (Fig. 2E). Consistent with previous studies, N. crassa showed high levels of RIP (17.4%) (Cambareri et al., 1991). Surprisingly, RIP in Z. radicans occurred at an even higher rate (24.2%) than in N. crassa. We also detected RIP at low levels in the rust Austropuccinia psidii (2.6%), E. muscae (2.4%), M. cicadina (2.7%), and E. maimaga (5.7%). This analysis argues that RIP occurs in Z. radicans and could be occurring at a low level in other entomophthoralean fungi as well.

A cytosine methyltransferase named RIP defective (RID) is required for RIP (Freitag et al., 2002), one of several families of DNA methyltransferases in Fungi (Bewick et al., 2019). RID contains two DNA_methylase (PF00145) domains, so we leveraged this characteristic domain profile to identify proteins with shared domain architecture. This revealed two orthogroups (OG0001715 and OG0003300) containing candidates with two DNA_methylase domains. These orthogroups were aligned with RID protein from Neurospora crassa, and N. crassa RID is sister to the OG0001715 clade in a tree of these proteins, albeit at low (12-14%) sequence identity (Fig. 2F). OG001715 is composed of proteins from Z. radicans, E. maimaiga and E. muscae. Each of these species has two paralogs in this orthogroup, except E. muscae. E. muscae intriguingly has an additional, partial paralog, (DSO57_1016266) of about half the size, which is in tandem with another, complete E. muscae OG0001715 paralog (DSO57_1016267). This analysis suggests that E. muscae, E. maimaiga and Z. radicans contain multiple proteins with similar domain architecture as RID, with some direct evidence for duplication of these methyltransferases.

Many eukaryotic organisms also employ RNA interference (RNAi) to control retrotransposon activity. RNAi occurs when double-stranded RNAs are processed into small interfering RNAs (∼20-30 nucleotides long) that are used to guide degradation (or, in the case of mRNA, repression of translation) of complementary RNA. In fungi, there are three proteins essential for RNAi: 1) RNA-dependent RNA polymerase (RDRP), which processes ssRNA into dsRNA, 2) Dicer, the enzyme that processes dsRNA into siRNA, and 3) Argonaute, a protein that loads siRNA to form an RNA-induced silencing complex (RISC) that identifies target RNAs by complementarity (Nicolás and Garre, 2016). We performed a domain-based analysis to assess if RNAi machinery had been lost in entomophthoralean fungi, which could allow transposable elements to proliferate (Fig. 2G). This analysis suggests that at least one homolog of each of the core components of the fungal RNAi pathway (RDRP, Dicer and Argonaute) is present across our core species, with E. muscae and E. maimaga predicted to have several homologs of each of these genes. Thus, loss of RNAi pathway genes seems unlikely to account for transposable element proliferation in these species.

We next compared the coding regions of the entomophthoralean datasets in order to identify functional trends across these fungi. Since all of these species are insect pathogens, we expected to observe common themes amongst their metabolic and secreted functions related to their utilization of host resources and interactions with the host immune system. First, we performed protein domain analysis, looking broadly at protein families (Pfam), and more specifically at functional domains involved in metabolizing carbohydrates (using CAZy) and proteins (using MEROPS). For these analyses we used all available entomophthoralean datasets (Fig. 1C) except for the M. cicadina genome. M. cicadina was excluded because the fragmented nature of its genome assembly precluded an accurate prediction of this fungus’ proteome, a key requirement for this analysis (Supplemental Table 2).

Pfam families encompass diverse proteins with a broad range of functions (El-Gebali et al., 2019). Enrichment analysis among these domains revealed a unique enrichment in E. muscae of retroviral families (i.e., TE-related), specifically Asp_protease (PF09668), Asp_protease_2 (PF13650), gag-asp_protease (PF13975), RVP_2 (PF08284) and dUTPase (PF00692) domains (Fig. 3A). The domain zf-CCHC (PF00098) was found to be similarly enriched in E. muscae and may represent a retroviral regulatory protein. The largest family that was enriched in E. muscae was Lipase_3 (PF01764; 348 in E. muscae; 5-fold higher than the median across genomes: 66), which was also enriched in E. maimaiga and Z. radicans (132 and 190, respectively).

A much smaller number of Pfam domains were found to be significantly underrepresented in E. muscae (20) compared to the number of domains found to be significantly overrepresented (105) (Fig. 3B). Proportionally more of these underrepresented families were not predicted to be secreted (17/20, compared to 45/105 of overrepresented Pfam families, p=0.00055 per chi-squared test). Three of these not-secreted, underrepresented families in E. muscae were underrepresented by greater than 5-fold: alpha/beta hydrolase fold (abhydrolase_3; PF07859; 6.5-fold lower than median), bifunctional feruloyl and acetyl xylan esterases (BD-FAE; PF20434; 20-fold) and endonuclease/exonuclease/phosphatase (Exo_endo_phos_2; PF14529; 13-fold). Alpha/beta hydrolase folds are found as catalytic domains in many different enzymes, including BD-FAE (PF20434) proteins, which specifically act on complex xylans. The BD-FAE family was significantly underrepresented in E. muscae, E. maimaiga and Z. radicans (1, 2 and 3 BD-FAE-containing proteins, respectively), compared to C. coronatus and N. thromboides (23 and 25, respectively). E. muscae, with only a single predicted BD-FAE-containing protein, was additionally significantly lower than P. formicae and S. castrans, which had 20 and 23, respectively.

We compared the broad Pfam composition of these seven genomes (Fig. 3C). 2,684 Pfam domains were present in all genomes. S. castrans and P. formicae shared a unique overlap of 1,142 domains, each with 170 and 485 unique domains, respectively. E. muscae, by comparison, only had 14 unique Pfam domains. The trend of P. formicae and S. castrans sharing many domains was seen across all domain types. However, the proteomes for these fungi were uniquely generated from transcriptomic data, so we were concerned this pattern may reflect a methodological, not biological, signal. To test this, we recapitulated our Pfam domain analysis incorporating a proteome generated from transcriptomic data collected from E. muscae. The results of this analysis are summarized in an UpSet plot (Fig. 3-S1), in which the E. muscae transcriptomic dataset does not meaningfully group with P. formicae and S. castrans. This suggests that the unique subset shared between P. formicae and S. castrans is likely due to their evolutionary history, not a methodological artifact.

All genomes included in our domain analysis shared 691 MEROPS peptidase domains (Fig. 3C). S. castrans and P. formicae shared 448 MEROPS domains absent in the other genomes, with each having a large number of unique MEROPS domains (173 and 136, respectively). E. muscae by contrast had about a third to a quarter as many unique MEROPS peptidases (37). Among E. muscae MEROPS, 92 were found to be significantly different in comparisons among genomes, with 74 MEROPS domains significantly enriched and 9 domains having significantly fewer proteins (Fig. 3-S2B). A clear trend for E. muscae, E. maimaiga and Z. radicans is enrichment in M16A metallopeptidases (i.e., MER0001214, MER0001218, MER0002283, MER0002345, MER0002423, MER0003386, MER0003823, MER0011096, MER0011744, MER0015259, MER0169735) compared to S. castrans, P. formicae and N. thromboides. Considering proteins containing these 10 M16A domains, E. muscae (62 on average), E. maimaiga (19) and Z. radicans (44) were comprised of approximately 11-fold more proteins (43 vs 4) compared to S. castrans (5), P. formicae (3) and N. thromboides (2). Two metallopeptidases were uniquely enriched in E. muscae: MER0001120, a pro-collagen C-peptidase, and MER0016735, an M23 peptidase. Among serine peptidases, a lipid monooxygenase (MER0031618), two S33 prolinases (MER0036050 and MER0036051) and three kexin/kexin-like (MER0000364, MER0000374 and MER0001604/krp1) domains were found to be significantly enriched. Additionally, E. muscae was uniquely enriched in the monoglyceride lipase MER0045883. Considering cysteine peptidases, E. muscae had a high number of metacaspases across all 10 families, though fewer than the number seen in E. maimaiga (∼50 accessions for each family). Only a handful of domains were found to be significantly underrepresented in E. muscae, including aminopeptidase H11 (MER0002003), cystinyl aminopeptidases (MER0002060) and ERAP2 aminopeptidases (MER0002968) (Fig. 3-S2C).

Beyond these protein domain and enzyme analyses, we were very interested in comparing potential circadian machinery among entomophthoralean fungi as, for many of these pathogens, infection and behavioral manipulation of the host follows a strict daily timing (Dustan, 1924; Elya et al., 2018; Hajek and Harris, 2023; Krasnoff et al., 1995; Milner et al., 1984; Nielsen and Hajek, 2006; Pickford and Riegert, 1964). How this timing is achieved is not understood. One study suggests that timing of death of house flies killed by E. muscae is not dictated by host circadian machinery, instead suggesting that a fungal clock drives the timing (Krasnoff et al., 1995). The circadian clock of Neurospora crassa (Ascomycota) is the most highly studied and best understood circadian clock among fungi (see (Cha et al., 2015) for overview). In N. crassa, two photosensitive proteins, White Collar 1 and 2 (WC-1 and WC-2) form a heterodimeric complex (called the White Collar Complex, WCC). In the presence of light, WCC undergoes a conformational change and becomes a transcriptional activator of light-regulated genes, including the gene frequency (frq) that encodes an intrinsically-disordered protein. FRQ protein forms a heterodimeric complex with its partner, FRQ-interacting RNA Helicase (FRH), and this complex, called FFC (FRQ-FRH Complex), inhibits the activity of WCC. After protein synthesis, FRQ becomes progressively more phosphorylated, eventually leading to its degradation and reduced concentration of FRQ and FFC. When FRQ titers drop below a certain threshold, WCC activity can proceed unimpeded, thus closing an oscillating cycle of transcription and translation with a periodicity of about 24 hours.

To identify potential components of circadian oscillators in our entomophthoralean datasets, we checked for putative homologs of wc-1, wc-2 and frq and domains commonly found in circadian proteins, per-ant-sim or PAS (PAS_3: PF08447) and the structurally-similar domain GAF (GAF: PF01590) (Fig. 3D). All of the analyzed entomophthoralean datasets had at least one copy each of a gene encoding for a protein with the same domains found in N. crassa WC-1, with C. coronatus and P. formicae containing multiple genes (the latter even having two candidate WC-1 genes in tandem, Pfor_14152, Pfor_14153). For wc-2, all but E. muscae, Entomophaga maimaiga and N. thromboides also had at least one candidate gene. None of these species contained a gene encoding for a protein with the hallmark domains of frq (Fig. 3D).

In addition to finding evidence for at least one copy of a wc-1 homolog for all of the species analyzed, we also searched for genes encoding for known light-sensitive domains (rhodopsin family [7tm_1; PF00001], bacteriorhodopsin [Bac_rhodopsin; PF01036], DNA photolyase [DNA_photolyase; PF00875], the FAD-binding domain of DNA photolyase [FAD_binding_7; PF03441] and phytochrome [PHY; PF00360], GPCR rhodopsin 4 [GpcrRhopsn4; PF10192])). We did not detect any bacteriorhodopsins or phytochromes within our datasets, but we found several instances of genes containing the other light-sensitive domains, with E. muscae having far more than the rest (99 genes with rhodopsin family domains as well as 11 photolyases or FAD-binding domains thereof) (Fig. 3D). This analysis suggests that each of these fungi detect light. As light is the predominant cue for setting the phase and period of circadian clocks, this is consistent with the notion that these fungi may also be able to keep time.

We next assessed gene homology among the core genomes in our comparative dataset: E. muscae, Entomophaga maimaga, Z. radicans and N. thromboides (Supplemental Table 2). For this analysis, we used OrthoFinder to identify orthogroups (sets of genes that are descended from a single gene in the last common ancestor of all species within our analysis set, including both orthologs and paralogs, referred to henceforth as OGs) among the pooled genes from this set of genomes (Emms and Kelly, 2015). We reasoned that such an analysis could be a strategy to identify core genes important for an obligate entomopathogenic lifestyle, and conversely, could reveal potentially unique genes in each of the species examined, specifically in our focal species E. muscae. We recovered 17,111 OGs. 6,878 OGs (40.2%) contained representatives in all four species (Fig. 3E), and we refer to these as “core” OGs.

To gain some insight into the function of the core OG genes in E. muscae, we performed a Pfam annotation enrichment analysis. Twelve Pfam domains were significantly overrepresented in these genes (Fig. 3-S3) by factors ranging from 1.52 to 2.35-fold. These domains are found in proteins across a variety of principal cellular functions, spanning DNA replication, transcription, macromolecular assembly and signal transduction and included kinases (PF0069, PF07714), RNA recognition and binding proteins (PF00076), helicases (PF00400, PF00271, PF04851), endonucleases (PF04851), ATPases (PF00004), Beta-propeller (PF00400), phosphatidyl inositol-binding (PF00169), RNA splicing (PF00176), and DNA replication (PF00226). Genes that lacked a Pfam assignment altogether were underrepresented (odds-ratio 0.46), consistent with core OG genes belonging to conserved cellular processes among these four species. Sixteen Pfam domains were significantly underrepresented, ranging from being completely absent to 18.3 – 1.83-fold underrepresented (odds-ratio 0.055 to 0.55, respectively). Protein domains that were completely absent included chitin recognition (PF00187), transferase (PF02458), multicopper oxidases (PF00394 and PF07732) and polysaccharide deacetylase (PF01522). Underrepresented domains tended to be related to metabolism and transcription factors (PF00046 and PF00172). However, underrepresented genes also included the retrotransposon gag protein, which is involved in activity of retrotransposons (such as LTRs), and rhodopsin family genes.

The next largest overlap consisted of 10.8% of OGs that are shared only between the most closely related species in this set, E. muscae and Entomophaga maimaiga (Fig. 3E). OGs shared between E. muscae, Entomophaga maimaiga and Z. radicans, but not the relative outlier N. thromboides totaled 7.5%, with the next largest overlapping set consisting of 2.4% of OGs that are shared between E. muscae, Entomophaga maimaiga and N. thromboides, but not with Z. radicans. Other overlapping sets individually accounted for no more than 0.9% of all OGs. The highest proportion of species-specific OGs (OGs that are populated by genes of a single species) was found for E. muscae (21%), followed by Z. radicans (6.9%), Entomophaga maimaiga (5%) and N. thromboides (2.3%). Accordingly, we observed that E. muscae genes also populate the most OGs overall, though has genes in multi-species OGs no more frequently than Entomophaga maimaiga (Fig. 3F). These observations are consistent with the E. muscae genome containing about three times as many annotated genes as any other fungus considered.

Across these four species, 88% to 94.4% of genes were assigned to OGs (Fig. 3G). With respect to species-specific OGs, E. muscae has the most: 43.5% of E. muscae genes are found in these OGs. The species with the least, N. thromboides, had just 16.3%. For each species, the genes that either failed to cluster with an OG (were not assigned) together with the genes that are found in species-specific OGs comprise a set of genes that are potentially unique to that species. If we consider all of the annotated genes in the E. muscae assembly, E. muscae has the highest percentage of species-specific genes among this species set at 50.9% (Fig. 3G). However, if we filter our annotated genes using a pooled in vivo transcriptomics dataset (Elya et al., 2018), we find that the percentage of species-specific genes for E. muscae falls to 35%, within the range of the other fungal genomes analyzed.

The obligate insect-pathogenic lifestyle of E. muscae prompted us to investigate the presence of canonical entomopathogenic enzymes in the genome. The primary sugar in insect blood (hemolymph) is trehalose (Thompson, 2003), and E. muscae appear to have an expanded group of acid-trehalases compared to other entomopathogenic and non-entomopathogenic Entomophthorales (Fig. 4A). This increase in trehalase enzymes might be driven by the general increase in total number of genes in the E. muscae genome compared to other Entomophthorales. But irrespective of the underlying mechanism, having an expanded repertoire of trehalases likely assists in utilizing host carbohydrate nutrients. The obligate insect-pathogenic lifestyle is also evident when comparing the repertoire of lipases, subtilisin-like serine proteases, trypsins, and chitinases in our focal species versus Zoopagomycota and Ascomycota fungi that are not obligate insect pathogens (Fig. 4B). Sordariomycetes within Ascomycota contains the other major transition to insect-pathogenicity within the kingdom Fungi (Araújo and Hughes, 2016). Based on our comparison of gene numbers, Entomophthorales possess more enzymes suitable for cuticle penetration than Sordariomycetes (Fig. 4B). In contrast, insect-pathogenic fungi within Hypocreales possess a more diverse secondary metabolite biosynthesis machinery as evidenced by the absence of polyketide synthase (PKS) and indole pathways in Entomophthorales (Fig. 4C).

To investigate unique characteristics of the E. muscae proteome, we identified all domains that were either unique to (Fig. 5A) or missing from (Fig. 5B, Fig. S5-1) E. muscae. Among the 41 domains unique to E. muscae (Fig. 5A), the Pfam family Peptidase_A2B (PF12384), a component of Ty3 retrovirus-like elements, was by far the most common (26 proteins). Many Pfam families that were uniquely found in E. muscae were apparently related to retrotransposons. Thioredoxin_4 (PF13462) was the second-most abundant unique Pfam family (6 proteins). Among other E. muscae-specific domain families, two proteins were annotated as containing abhydrolase_8 domains (PF06259), perhaps accommodating for the apparent significant reduction in abhydrolase_3 domains in E. muscae (Fig. 3B). Nearly all MEROPS families that were unique to E. muscae were serine peptidases. A single metallopeptidase (MER0001219; M16A family Axl1 peptidases), threonine peptidase (MER0026205; T3 family) and cysteine peptidase (MER0014097; C19 family) were found uniquely in E. muscae. The most abundant MEROPS family that was unique to E. muscae was a chymotrypsin family (3 proteins; MER0021846). Only three CAZy domains were uniquely found from E. muscae: GH134 (containing endo-β-1,4-mannanases), GH5 (glucanases/cellulases) and PL12 (heparin-sulfate lyase).

In total, 2,237 Pfam domains were absent in the E. muscae genome but observed in other genomes within our core set of entomophthoralean fungi. To assess loss of key domains in E. muscae, we considered domains that were missing from E. muscae, but were significantly enriched in other entomophthoralean fungi (Fig. 5B, Fig. 5-S1). We did not observe any domains across Pfam, MEROPS and CAZy that were both missing from E. muscae and overrepresented across all other entomopthoralean fungi, suggesting that E. muscae has not lost any functions specific to the Entomophthorales. Our domain analysis suggests that E. muscae is unique largely through its protein-domain family expansions, not in its losses of domains. Many domains, of all types in our analysis, that were not enriched (or not present) in E. muscae were enriched only in P. formicae and S. castrans, suggesting this signal is reflective of functionalities unique to those species, rather than functionalities lost in E. muscae.

We also performed an enrichment analysis on the orthogroups that were classified as potentially species-specific for E. muscae. As in our previous OG analysis, we considered genes to be potentially species-specific if they either failed to cluster with an OG in E. maimaiga, Z. radicans or N. thromboides or clustered with OGs comprising only other E. muscae genes. We performed an enrichment analysis of Pfam annotations of genes that met these criteria, comparing the frequency of Pfam occurrence in this set against occurrence within all annotated genes (Fig. 5C). We observed significant overrepresentation of three Pfams among potentially E. muscae-specific genes: insulinases (PF00675) M16 peptidases (PF05193) and genes that lacked any Pfam annotation. That E. muscae may have specific peptidases is consistent with our understanding of the entomopathogenic lifestyle: these enzyme families are both key in host recognition and invasion as well as host resource utilization (Arnesen et al., 2018). Finding genes lacking any Pfam annotation as another enriched group is consistent with the large number of genes observed in E. muscae compared to other species, and suggests that there are novel genes of undescribed/unknown function specific to E. muscae.

Pfams that were underrepresented in the set of potentially E. muscae-specific genes spanned multiple processes, but a few general themes emerged. These include a lack of proteins with cytoskeletal (e.g., ankyrin [PF12796, PF00023, PF13637, PF13606, PF13857], kinesin [PF00225]), cell signaling (e.g., RhoGEF [PF00621], Rab-GTPase [PF00566], proteins tyrosinase and kinase [PF07714, PF00069], thioredoxin [PF00085], pleckstrin homology domain [PF00169]), and transcriptional regulation functions (e.g., zinc fingers [PF13445, PF13923, PF13920, PF00097, PF13639, PF00096], SNF2-related [PF00176], DEAD box helicase [PF00270], RNA recognition motif [PF00076], JmjC [PF02373]). The underrepresentation of these domains in potentially E. muscae-specific genes suggests their functionalities are similar to those in other entomophthoralean species.

We performed a similar enrichment analysis looking at the set of potentially E. muscae-unique genes that are predicted to encode secreted proteins (Fig. 5D). We observed several Pfams that were underrepresented within the predicted secretome, many which encompassed functions that are not expected to serve the fungus in an extracellular context (e.g., domains involved in gene transcription [PF00076, PF00172, PF00270] and ATP production [PF00004]). We also observed a few underrepresented Pfam domains involved in signaling [PF00069, PF07714]). Pfams could be underrepresented in E. muscae-specific secreted proteins either because similar proteins are present across other entomophthoralean fungi considered or because proteins containing these domains are not typically secreted. We suspect the Pfams involved in signaling are underrepresented due to their similarity to proteins in other entomophthoralean fungi, while Pfams involved in intrinsic (i.e., not extracellular) processes are underrepresented due to a both unlikelihood of being secreted and conserved function among species.

Pfams that were significantly overrepresented in the predicted E. muscae-unique secretome included various catabolic enzymes (e.g., phosphoesterase [PF00149], tyrosinase [PF00264], proteases including trypsin [PF00089], subtilase [PF00082], peptidase inhibitor I9 (known constituents of subtilisins) [PF05922], zinc carboxypeptidase [PF00246], insulinase [PF00675], lipase [PF01764], polysaccharide deacetylase [PF01522], glycosyl hydrolases [PF00728, PF02838]) and macromolecular recognition domains (e.g., chitin recognition, ML domain). The observation that catabolic enzymes are overrepresented within the putatively E. muscae-specific predicted secretome is consistent with our current understanding of entomopathogen host interactions (Elya and De Fine Licht, 2021). These enzymes are important during fungal infection, where they help degrade chitin-linked proteins in the cuticle when the fungi force their way inside insects. Both during growth on the cuticle and when proliferating inside the living insects E. muscae use enzymes such as peptidases, glycosyl hydrolases etc. to obtain nutrients that support fungal growth in the hemocoel (Elya and De Fine Licht, 2021). In addition, we saw enrichment of the collagen triple helix repeat domain (PF01391), FAD binding domain (PF01565) and berberine-like proteins (PF08031). As with our analysis of all E. muscae-unique genes, an overrepresentation of genes that were not associated with any Pfam suggests that many E. muscae genes have as yet unknown functions.

DNA sequence data for members of the E. muscae species complex (EMSC) and outgroup taxa within the genus Entomophthora were mined from NCBI GenBank to determine the prevalence of each locus. Novel Sanger sequence data was also generated in support of this analysis, which has been deposited in NCBI (i.e., ARSEF_13514, ARSEF_6918, SoCal_c1 and LTE_c1, See Methods). In parallel, BLASTn searches using reference sequences generated from E. muscae ARSEF 13514 were used to identify closely related sequence matches for both ITS and 28S. Sequences were aligned to assess length and quality, and metadata was compiled to determine which strains had available sequence data and morphological data. Fourteen strains were chosen for inclusion based on the availability of both DNA sequence data and morphological data.

Two methods of phylogenetic inference, maximum likelihood (ML) and Bayesian inference (BI), for the 2-gene concatenated dataset resolved the EMSC as a well-supported monophyletic group that includes E. ferdinandii, E. muscae and E. scatophagae (Fig. 6A). In addition, E. schizophorae and the clade containing E. syrphi and E. aff. grandis were both monophyletic and well-supported by both phylogenetic methods. Within the EMSC, two well-supported clades were observed, referred to hereafter as EMSC Clade 1 and EMSC Clade 2 (Fig. 6A). Clade 1 contained ARSEF 13514 from D. melanogaster (Drosophilidae) and three other strains from Muscidae (Fig. 6B). Clade 2 had host species from three fly families (Anthomyiidae, Muscidae and Scathophagidae) and contained strains from E. ferdinandii and E. scatophagae in addition to strains identified as E. muscae. Strains sequenced for this study from various Drosophila spp. also expanded the known ranges of E. schizophorae and E. grandis (Keller, 2002).

We examined the number of nuclei in, and dimensions of, primary conidia, which are potentially diagnostic morphological characters for these species (Keller, 2002). For the six formally described fly-infecting Entomophthora spp. included in the study, there was considerable overlap in the ranges of size and mean number of nuclei from primary conidia (Fig. 6B, 6D). The ranges of at least four species overlap among the included Entomophthora species. The published sources of data used for this analysis varied in their measures (mean, maximum, minimum, standard deviation and standard error) for number of nuclei and sample sizes. Nuclei data was missing for three of fourteen strains. Number of nuclei was the most commonly reported morphological character, but this measure alone did not resolve species boundaries, except for E. schizophorae.

Primary conidial length and width measurements provided better species-level resolution, but these measurements were only available for four of fourteen strains (Fig. 6C). Reported spore measurement ranges for known species overlaid with those included in this study are not aligned with phylogenetic findings. Strains ARSEF 13514 and ARSEF 6918, which occurred in EMSC Clade 1 and EMSC Clade 2, respectively, overlapped with measurements for E. ferdinandii, but not E. muscae. Strains ARSEF 6716 and KVL14-17, both members of EMSC Clade 1, had overlapping spore measurements with each other and E. muscae, but not E. ferdinandii. These spore measurements further support E. scatophagae as part of the larger EMSC. They also highlight significant overlap between E. grandis and E. syrphi.

We report a new ∼1 Gb assembly of the E. muscae genome obtained using a long-read sequencing approach. Despite the comparative ease of obtaining high molecular weight DNA from other organisms, entomophthoralean fungi are recalcitrant to standard DNA extraction procedures. The reasons for this are not yet clear, but, along with the relatively large sizes of entomophthoralean genomes, they have stymied sequencing and assembly efforts within this fungal order. Our success came from having on-demand access to an E. muscae culture in the lab and to using a modified extraction protocol (See Methods), which could potentially be applied to other entomophthoralean fungi for future genomic work.

While our analysis suggests that this new genome is fairly complete (81.3% complete BUSCOs), we observed a high level of duplicate BUSCOs. Our assembly is decidedly not fully diploid (Fig. S1A,B). Instead, our data are consistent with duplicate BUSCOs arising from heterozygosity in coding regions. If this is the case, we have overestimated the size of the genome by about 20 Mb (2%). The prevailing thinking in the field is that entomophthoralean fungi are haploid (Humber, 2016), but the ploidy of E. muscae is still not fully resolved. Our observation is reminiscent of the functional diploidy previously reported for another E. muscae isolate (De Fine Licht et al., 2017).

Given how little is known about the biology of Entomophthorales, we adopted a comparative approach for our initial analysis of the E. muscae genome, sourcing available transcriptomic and genomic data from species within this fungal order. The inferred phylogenetic relationship between these species, based on conserved protein sequences (Fig. 1C), implied that Conidiobolus is paraphyletic. This is consistent with recent taxonomic revisions of that genus (Gryganskyi et al., 2022; Nie et al., 2020).

The second smallest genome (after Conidiobolus) in our comparative data set, was Z. radicans, which, at 655 Mb, is still >15-fold higher than the mean fungal genome size of all 6,144 genomes considered in Fig. 2A (42.7 Mb). The size of these genomes is primarily driven by a prolific expansion of repeat elements, the most numerous of which are Ty3 retrotransposons (Fig. 2C). Fungal Ty3 elements are common, and appear to have undergone independent expansions across many fungal lineages (Muszewska et al., 2011a). From the consistent patterns of Ty3 sequence abundance and the estimated divergence time of these sequences (Fig. 2D), we speculate that Ty3 expansion was a feature of the last common ancestor of non-Conidiobolus Entomophthorales: this proliferation was not recent.

One possible avenue leading to such a proliferation could be loss of defenses against repeat accumulation. Fungi use repeat induced polymorphism (RIP) and RNA interference (RNAi) as their two main strategies to combat the accumulation of repeated elements. We detected several homologues of the core RNAi pathway components in each of our fungal species: Dicer, Argonaute and RNA-dependent polymerase (Fig. 2G). This analysis suggests that RNAi pathways are intact and potentially function in a similar manner to other fungi within Entomophthorales, and may even indicate an expansion of this defense mechanism to counteract TE proliferation.

While RNAi occurs across fungi, RIP has only been observed in Dikarya. Our analyses suggest that RIP may be active in fungi outside of Dikarya (Fig. 2E, 2F), but there are major caveats to this finding. We detected DNA methyltransferases with shared domain architecture as RID, a key enzyme for RIP function, but this should be validated with deeper analyses and experimental characterization of genes in these orthogroups. Host population dynamics limit opportunities for pathogen sexual recombination temporally and spatially, and entomophthoralean fungi are hypothesized to reproduce almost entirely asexually (Humber, 2016). If this is the case, these cells would not undergo meiosis, which is the cell-cycle stage where RIP occurs in fungi in the subkingdom Dikarya. Parasexuality (i.e., heterokaryons formed via arbitrary fusion of genetically distinct cells) may represent another opportunity for RIP. If parasexuality occurs in entomophthoralean fungi, a suitable moment might be in the protoplast stage, when cells lack cell walls and could meet in insecta after successfully infecting the host. Protoplast production is a necessary step for parasexual fusing in the laboratory setting. However, the existence of parasexuality has been questioned in this group of fungi, as cell-to-cell fusion has only been observed to occur between genetically-identical hyphal bodies prior to resting spore formation (Humber, 2016). Such fusion may still provide an opportunity to silence TEs that accumulate during primarily asexual development. However, parasexuality may also provide an opportunity for TEs to spread among heterokaryotic nuclei, as has been suggested in arbuscular mycorrhizal fungi (Yildirir et al., 2020) and may be particularly amenable for transposons with RNA intermediates (such as Ty3). In support of the presence of RIP in entomophthoralean fungi, all fungi in our set possess genes encoding DNA methylase domains (DNA_methylase: PF00145). If these genes are indeed found to mediate RIP, it would position E. muscae as a model for investigating meiosis-independent RIP.

Rust fungi parasitize plants (e.g., Phakospora pachyrhizi, Astropuccinia psidii; Fig. 2A, 2B) and also have very large genomes as a result of rampant TE proliferation. Recent analysis of three isolates of Phakospora pachyrhizi found that, similar to Entomophthorales, Ty3 elements comprise 43% of the genome (Gupta et al., 2023) and the expansion of these elements was predicted to coincide with radiation of host species. We do not yet know if TEs in P. pachyrhizi and the family Entomophthoraceae expanded via similar mechanisms. The large genomes in these clades are puzzling given that evolution tends to reduce rather than expand parasitic genomes (Wolf and Koonin, 2013). TE expansions may have been triggered by a variety of biotic and abiotic factors and underlie evolutionary choice points (Belyayev, 2014). But what factors would allow an enlarged genome to persist over evolutionary time? We hypothesize that an enlargement of the non-coding genome may set the stage for evolution of novel functionalities, and, in turn, specialization for different hosts and speciation. Rust and entomophthoralean pathogens, which are both marked by Ty3 element proliferation, are known for their extreme degree of host specialization (Sacco and Hajek, 2023). In the absence of frequent sexual recombination, the flexibility permitted by an outsized “genomic canvas” may have been favored by selection and maintained to the present day. Work addressing the basis of TE proliferation across divergent lineages is needed to better understand the evolutionary pressures driving genomic enlargement and maintenance of giant genomes. Considering the observed Ty3 proliferation in Entomophthorales is not recent (Fig. 2D), we can theorize a trajectory to the observed TE-inflation and high degree of host specialization in entomophthoralean fungi: first, a transition away from frequent sexuality to infrequent parasexuality/sexuality (lowering TE defenses), then a proliferation of TEs (generating raw materials for adaptation) and finally a slow, parasexual cessation of TE bursts with selection on genes involved in host-specific infection (strengthening the degree of host specialization).

Analyses of protein domains (Pfam, CAZy and MEROPS) on our core set of entomophthoralean genomes (plus two available transcriptomes: S. castrans and P. formicae) revealed several functions enriched in the predicted E. muscae proteome (Fig. 3A, Fig. 3-S2). Many of the associated proteins are predicted to be secreted and likely metabolize host tissues (lipases, chitinases, insulinases, etc). In addition, trehalase enzymes which act upon the primary sugar of hemolymph (Thompson, 2003), were found to be expanded in E. muscae, providing further evidence that this fly pathogen has evolved to efficiently process host tissue. As more entomophthoralean fungi are sequenced, the enzymatic repertoire of these highly specialized pathogens will provide insights into the nutritional needs of these fungi, which may vary with differences in tissue composition of their hosts.

We also examined genes that may be involved in the ability to sense environmental light cues and maintain circadian time, as these processes are likely involved in fungal manipulation of host behavior (de Bekker et al., 2021; de Bekker and Das, 2022). In N. crassa, the primary fungal model system for circadian biology, two key genes in maintaining circadian rhythms are frequency (frq) and the blue-light sensitive white-collar 1 (wc-1). While we found many genes encoding light-sensitive protein domains, including several wc-1 homologs, we did not find homologs of frq within our entomophthoralean fungi. However, absence of frq does not preclude a functional clock in these fungi; many fungi that lack a frq homolog have clear circadian phenotypes under free-running conditions (e.g., Mucoromycotina fungus Pilobolus sphaerosporus and ascomycetes Aspergillus flavus, Saccharomyces cerevisiae and Cercospora kikuchii), indicating that there are frq-independent mechanisms for keeping time within fungi that have yet to be discovered (Montenegro-Montero et al., 2015; Salichos and Rokas, 2010). In addition, the ascomycete plant pathogen Verticillium dahliae has a homolog of frq that does not cycle over circadian time (Cascant-Lopez et al., 2020). Notably, frq genes have only been observed in Ascomycota (except for the subphylum Saccharomycotina), so it is unsurprising that entomophthoralean fungi would lack homologs of this gene (Montenegro-Montero et al., 2015; Salichos and Rokas, 2010).

As a complementary approach to our domain-based analysis, we also categorized genes from our core genomic set (E. muscae, Entomophaga maimaiga, Z. radicans and N. thromboides) into putative orthologous groups (OGs; Fig. 2E-G). A plurality of OGs contained genes from all four species (40.2%). The next most common species composition of OGs was E. muscae and Entomophaga maimaiga (10.8%), followed by E. muscae, Entomophaga maimaiga and Z. radicans (7.5%). Each species also possessed species-specific OGs, with E. muscae possessing the most. An enrichment analysis of Pfam annotations for E. muscae genes that were assigned to core OGs revealed that many processes that may be host-specific (e.g., chitin recognition, fatty acid, protein and sugar utilization) are underrepresented within the core set (Fig. 3-S3). Conversely, genes predicted to serve in signaling pathways (e.g., kinases, pleckstrin homology domains) and basic cellular metabolism (e.g., transcription, translation and respiration) are overrepresented among E. muscae genes assigned to core OGs (Fig. 3-S3). All told, these results are consistent with specialization in utilizing host tissues as a key driver of species-level differences in these fungi.

We also searched for gene-family-level expansions not just within Entomophthorales (Fig. 4A), but across diverse fungal lineages that include other insect pathogenic species (Fig. 4B). We observed gene family expansions related to host tissue utilization, some unique to Entomophthorales (e.g., lipases, acid trehalases) and others shared across entomopathogens of different phyla (e.g., peptidases). This partial overlap indicates that there are common themes among insect fungal entomopathogens in conquering their hosts as well as divergent strategies within distinct fungal lineages.

Our analysis of predicted secondary metabolites within insect fungal entomopathogens suggests that Entomophthorales may produce non-ribosomal peptides, siderophores and terpenes. A recent report demonstrating the presence of several terpenoid compounds in E. muscae-killed house flies (Naundrup et al., 2022) suggests that this prediction likely missed important classes of metabolites. Current models predict that the quantity and diversity of secondary metabolites is lower in entomophthoraleans than for ascomycete entomopathogens (Fig. 4C), but this conclusion has several caveats. At a first glance, our predictions seem consistent with Entomophthorales following a primary growth, rather than a primary toxin-producing, infection strategy (as proposed for certain generalist Metarhizium species (Kershaw et al., 1999)), and with the long-held dogma that zygomycete fungi produce few secondary metabolites (Bushley and Turgeon, 2010; Voigt et al., 2016).

However, secondary metabolite cluster analysis methods have been developed almost exclusively using data from Ascomycota and Basidiomycota. Thus, the lack of predicted secondary metabolites in Entomophthorales could be due to prediction methods not well suited to function outside of Dikarya. Recent evidence for secondary metabolite gene clusters in Mucoromycota opposes the long-held view that zygomycete fungi do not produce these compounds (Koczyk et al., 2021; Voigt et al., 2016) and suggests that methods to detect metabolites and the enzymes that synthesize them need to be adjusted for non-Dikarya fungi. Indeed, a recent study in Massospora experimentally detected several secondary metabolites that, based on current bioinformatic models of biosynthesis, should not exist in this species (Boyce et al., 2019). In addition, because Zoopagomycota are so uncharacterized with respect to secondary metabolites, it is possible that the apparent lack of standard metabolite classes is accurate, but we failed to identify metabolite clusters that are unique to this phylum and not yet described. We have much to learn about secondary metabolism beyond Dikarya and future efforts should prioritize exploring metabolism of these taxa for the potential discovery of novel metabolic gene cluster architecture or novel metabolite synthesis pathways.

From analysis of Pfam, MEROPS and CAZy domains, we found that E. muscae has a handful of protein domains unique within the Entomophthorales species we analyzed. The most numerous of these was Peptidase A2B, which is a protease family involved in processing retrotransposon Ty3 polyprotein into its component parts (Kirchner and Sandmeyer, 1993). Given the proliferation of Ty3 elements among family Entomophthoraceae fungi in our dataset, we expect that the functional activity of this domain is not restricted to Peptidase A2B. Importantly, we did not find any domains that were missing from E. muscae but enriched in other fungi in our core set (Fig. 5B, Fig. S5C). Together, these observations are consistent with E. muscae having diverged from other entomophthoralean fungi by expansion of existing domain families rather than loss of particular domains.

As an additional approach to identifying unique gene functions in E. muscae, we also looked at orthogroups unique to this species, from either the total proteome or the secreted proteome (Fig. 5C, D). M16 peptidases (domains PF00675 and PF05193) were found to be enriched across all proteins as well as within proteins predicted to be secreted. M16 peptidases are metalloendopeptidases with family members that cleave N-terminal targeting peptides that direct proteins to their destination subcellular compartments (Rawlings et al., 2018). Recent work in the the protozoan Cryptosporidium parvum implicated insulinase-like proteases (which are also members of the M16 family) in early infection (Zhang et al., 2019), perhaps suggesting a role for these proteins in early development of E. muscae.

Based on our domain enrichment analysis, we expect that many of the unique secreted proteins are involved in the metabolism of fly host tissues. Depending on the host range specificity of this study’s E. muscae isolate, these genes could encode proteins for metabolizing or recognizing Drosophila macromolecules specifically. Looking at just secreted proteins, we saw enrichment for a handful of catabolic domains including peptidase inhibitor I9 (PF05922), a constituent of subtilisins (Arnesen et al., 2018; Muszewska et al., 2011b). This observation is consistent with work finding unique subtilisin-like serine proteases in the entomopthoralean fungi E. muscae, C. incongruus and P. formicae and suggests further subspecialization of these enzymes at the species level (Arnesen et al., 2018). We also observed an enrichment for chitin recognition protein (PF00187) in the secreted proteome. Both the fly host and E. muscae produce chitin, so these proteins could function in either host recognition or in masking E. muscae from being recognized by the host immune system upon entry (Cen et al., 2017).

An enrichment for the collagen triple helix repeat domain (PF01391) caught our attention. In the generalist fungal pathogen Metarhizium anisopliae, a collagen-like protein (Mcl1) that is expressed immediately after invasion of the host hemolymph helps fungal cells avoid detection by the host immune system (Wang and St Leger, 2006). In a similar vein, we wonder if E. muscae’s unique collagen triple helix domain may play a role in evading the host immune system.

We also were surprised to observe an enrichment in berberine-like proteins (PF08031) among secreted E. muscae-specific OGs. Berberine-like proteins are involved in the synthesis of isoquinoline alkaloids, a class of compounds that includes the analgesics morphine and codeine, and usually require FAD as a cofactor (Kutchan and Dittrich, 1995). Intriguingly, we also observed an enrichment of FAD binding domains (PF01565). Isoquinolines have been most frequently discovered in plants, but recent work in Aspergillus fumigatus discovered that an orphan metabolic gene cluster containing small NRPS-like genes encodes enzymes capable of producing a novel isoquinoline (Baccile et al., 2016). Isoquinolines have wide-ranging effects (Khan and Suresh Kumar, 2015), and some isoquinolines have been shown to inhibit the innate immune response. Future work leveraging high-sensitivity metabolomics methods will be needed to determine if isoquinolines are produced by E. muscae and, if so, what role they may play in infection and in altering host behavior.

Historically, Entomophthora spp. were identified using a combination of morphology, host species, and/or location (MacLeod et al., 1976; MacLeod and Müller-Kögler, 1973). However, in recent years it has come to light that these features may be unreliable. For example, conidial size of members of the EMSC changes when infecting host flies different from the original source host (Jensen et al., 2006), meaning that this feature is not, on its own, reliable for identifying the fungal species. Such differences in conidial size across insect hosts have also been observed for another closely allied member of the Entomophthoraceae, Massospora levispora (Macias et al., 2020). Another study examining haplotype diversity in E. muscae strains from a single epizootic infecting two fly species in NC, USA uncovered two genetically distinct subpopulations, both with nearly identical primary conidial sizes and numbers of nuclei per conidium (Gryganskyi et al., 2013).

To explore the utility of DNA sequence data (ITS + 28S) for identification purposes in the EMSC, we assessed the congruence of available sequence and morphology data for EMSC isolates. We recovered two well-supported clades (both in the single gene trees and the concatenated tree) that contain strains previously identified as E. muscae, E. ferdinandii, and E. scatophagae (Fig. 6A-C). Using morphological characters, we found that the number of nuclei in the primary conidia contained almost no phylogenetic signal among members of the EMSC based on the strains we assessed (Fig. 6B). Furthermore, the dimensions of the primary conidia were largely overlapping among nearly all Entomophthora species examined, with the exception of E. ferdinandii and E. schizophorae (Fig. 6C). Interestingly, dimensions of primary conidia for our E. muscae strain (ARSEF 13514) overlapped with measurements for E. ferdinandii and outside the reported measurements for E. muscae despite occupying the same well-supported clade as strains whose spore measurements agreed with E. muscae.

Taken together, these results indicate that the morphological characters examined here, which are traditionally used to identify Entomophthorales, are incongruent with molecular data especially within the EMSC. However, this disagreement between morphology and sequence data could reflect 1) the paucity of character data, 2) insufficient taxon sampling that failed to capture the full range of morphological and genetic variation, and/or 3) the presence of misidentified specimens in collections and studies that are potentially masking the presence of morphologically cryptic species (i.,e., if researchers relied on nuclei number and conidia measurements to identify their specimens in the absence of DNA sequence data, then any morphology data they report for the species could include data from multiple species). While ITS and 28S loci may have insufficient phylogenetic signal to fully resolve these species boundaries especially within the EMSC, resolution may be achievable with the addition of other loci. In addition, the evolutionary history of this group may include a combination of introgression and rare sexual recombination events that have softened species boundaries within the EMSC, though data supporting this are currently limited (Gryganskyi et al., 2013). Overall, the EMSC remains monophyletic, but a comprehensive taxonomic revision is needed to define species within the clade.

Many strains of EMSC exist in fungal collections (ARSEF in particular) that have no associated sequence or morphology data. We propose that for each individual specimen, whether novel or from a collection, researchers minimally generate ITS and 28S sequence data and collect morphological data (specifically the length, width, and number of nuclei within primary conidia), so that future research can assess these features in combination to define the species boundaries of this group. We encourage researchers to further deposit voucher specimens, live cultures and/or DNA with fungal collections (e.g., ARSEF) so that future researchers can return to the original sample to examine traits or generate more sequence data if needed. Likewise, historic studies based solely on morphology from herbarium specimens may need to be resampled for both DNA sequencing and morphology.

The work presented here is one of the first steps in exploring Entomophthoralean genomes. We hope that the data and questions posed herein can serve as a springboard for generating and testing new hypotheses about these organisms. Entomophthora muscae has been under scientific scrutiny since its description in 1855, yet, even under the new light of our genomic investigation, this fungus remains “one of the strangest and most interesting apparitions” (Elya and De Fine Licht, 2021) with clear avenues to enrich our understanding of myriad biological phenomena.

E. muscae was isolated for in vitro growth from a single sporulating cadaver in 60 mm sterile petri dishes using the ascending sporulation method (Hajek et al., 2012). All cultures were grown in supplemented Grace’s Insect Media (ThermoFisher Scientific #11605094) with 5% fetal bovine serum added (ThermoFisher Scientific #10437010) in volumes of 20 mL in 25 cm² vented tissue culture flasks (Corning #353014). Cultures were grown without shaking, with flasks laid on the long edge to maximize surface oxygen exchange, in the dark at room temperature (19-21C). DNA was isolated from a log-phase culture via methods of (Elya and Lee, 2022). DNA quantity, quality and size distribution were assessed by Qubit dsDNA HS Assay kit (ThermoFisher Scientific #Q32851), Nanodrop spectrophotometric analysis (ThermoFisher Scientific #ND2000) and DNA Genomic ScreenTape analysis (Agilent TapeStation #5067-5365 and #5067-5366) following manufacturer’s protocols.

3 ug of genomic DNA were used to prepare two Oxford Nanopore DNA libraries, one sheared to 20 kb and one unsheared (45-50 kb) and sequenced sequentially on the Oxford Nanopore PromethION platform by the The Bauer Core Facility at Harvard University (Cambridge, MA) to 82x coverage. Reads are accessible through the NCBI Sequence Read Archive (SRA) via accession number SRR18312934. Nanopore reads that passed filtering were assembled using Flye v2.8.3 (Kolmogorov, 2021; Kolmogorov et al., 2019). The resultant assembly was polished with the same Nanopore reads used for assembly using Medaka v1.2.6 (Oxford Nanopore Technologies Ltd, 2021).

The Flye+Medaka assembly was further scaffolded with 10x genomic sequence library reads from the UCB isolate of E. muscae (Elya et al., 2018) deposited in NCBI SRA via accession number SRR18312935. This was accomplished by demultiplexing with longranger v2.2.2 (10x genomics), reads mapped with bwa-mem2 (Vasimuddin et al., 2019), followed by assembly scaffolding with Tigmint v1.2.2 (Jackman et al., 2018) and ARCS 1.2.1 (Yeo et al., 2018).

This updated E. muscae UCB genome was annotated with Funannotate v1.8.9 (Palmer and Stajich, 2020) employing the default parameters, which automates processing of RNAseq to produce transcripts, gene prediction with evidence from the transcripts and alignments of proteins, and construct consensus gene models from multiple gene prediction evidence, refine gene models for alternative splicing prediction based on the RNA-seq, and produce product function annotation based on protein sequence homology to databases of domains and annotated proteins. To annotate the genome, first repetitive sequences were masked by building a species-specific repeat library with RepeatModeler v2.0.1 (Flynn et al., 2020). These putative repeat families were further screened manually to remove likely protein-coding genes based on Diamond BLASTX v2.0.8 (Buchfink et al., 2021) searched against SwissprotDB v2021_04 (The UniProt Consortium et al., 2020) and removing sequences that matched clear non-repetitive but multicopy gene families. The curated repeat library was combined with the RepBase (Bao et al., 2015) library of fungi repeats with RepeatMasker v4-1-1 (Smit et al., 2013-2022) to softmask the genome before gene prediction to avoid over predicting transposons as host genes.

To train gene predictors and support gene models, RNA-seq data from previously deposited sequence data in NCBI SRA accession ERR1022665 were used as informant data in the annotation process. Briefly, the RNA-seq processing used Trinity v2.11.0 (Haas et al., 2013) in Genome Guided mode which aligned reads to the genome with Hisat2 v2.2.1 (Kim et al., 2019) followed by targeted transcript assembly with the Trinity pipeline on reads clustered to distinct genomic locations. The constructed assemblies total 63,231 which reflect partial transcripts and potential alternative splicing. These sequence assemblies were aligned to the genome with PASA (Haas et al., 2008) which uses the splice-aware aligner GMAP (Wu et al., 2016) and additional software to produce gene models from transcript data. This produced 67,902 gene models. The transcripts with full-length Open Reading Frames as scored by Transdecoder v5.5.0 (Haas et al., 2013) were kept as high quality models for gene prediction training. A total of 2,653 full-length PASA-derived gene models were used as a training set to ab initio predictors SNAP v2013_11_29 (Korf, 2004) and AUGUSTUS v3.3.3 (Stanke et al., 2008). In addition, the tools GeneMark-ES v4.62 (Ter-Hovhannisyan et al., 2008) and GlimmerHMM v3.0.4 (Majoros et al., 2004) were run after these tools performed self-training. Gene models were also predicted by CodingQuarry (Testa et al., 2015) using the transcript alignments as exon hints. Exon evidence was also generated by DIAMOND alignment of SwissprotDB proteins and polished by Exonerate v2.4.0 (Slater and Birney, 2005). The transcript and protein-based hints were provided to GeneMark, SNAP, and AUGUSTUS for evidence-guided gene prediction. EVidenceModeler v1.1.1 (Haas et al., 2008) generated consensus gene models in Funannotate using its default evidence weights. tRNA genes were predicted by tRNAscan-SE v.1.3.1 (Lowe and Eddy, 1997). Putative protein functions were assigned to genes based on sequence similarity to the InterProScan v5.51-85.0 (Blum et al., 2021; Jones et al., 2014), Pfam v35.0 (Mistry et al., 2021), Eggnog v2.1.6-d35afda (Cantalapiedra et al., 2021), dbCAN2 v9.0 (Huang et al., 2018), and MEROPS v12.0 (Rawlings et al., 2018) databases relying on NCBI BLAST v2.9.0+ (Camacho et al., 2009) and HMMer v3.3.2 (Eddy, 2011). Predicted secreted genes and transmembrane domains were annotated with Phobius (Käll et al., 2004) and SignalP v5.0b (Almagro Armenteros et al., 2019). A total of 39,711 gene models comprising 42,665 predicted proteins with alternative splicing isoforms, and 793 tRNAs were predicted. Pipeline for annotation is archived in the github repository https://github.com/zygolife/Entomophthora_muscae_UCB and Zenodo archive (Stajich et al., 2023a)

A comprehensive overview of all of the fungal datasets used in this paper and their associated figures is provided in Supplemental Table 2.

Transcriptomic data for Strongwellsea castrans sensu lato were generated from a cabbage fly (Delia radicum) caught in a cabbage field (farm name: Sørisgård, Latitude: 55.823706, Longitude 12.171149, date: 04/09/2013). The fly contained a clearly visible large hole on the side of the abdomen characteristic of S. castrans infection and was kept alive for a few hours until being snap-frozen in liquid nitrogen. The infected fly was ground to a fine powder in liquid nitrogen before extracting total RNA using a Plant RNEasy Kit (QIAGEN #74904) following manufacturer’s instructions. Total RNA was sequenced using Illumina HiSeq 2000 technology and TRUseq library building by Beijing Genomics Institute (BGI-Europe, Copenhagen, Denmark), and transcriptome sequences were assembled in de novo mode using Trinity v2.11.0. Raw reads are deposited at the European Nucleotide Archive (ENA) with accession number: ERR12023556. Assembled transcripts are available in zenodo archive under Supporting_data folder (Stajich et al., 2023b).

Phylogenetic relationships of species were determined by identifying conserved proteins from the protein translation of predicted genes or transdecoder ORFs from the transcriptome-only sampled species (P. formicae and S. castrans) using the PHYling v1.1 (Stajich, 2023) BUSCO / OrthoDB fungi_odb10 marker set. Briefly, the pipeline searches for conserved, generally single copy proteins via HMMer (Eddy, 2011) searches and builds individual and concatenated protein alignments. The phylogenetic tree was constructed from the concatenated alignments with FastTree v2.1.11 (Price et al., 2009) using the ‘-lg -gamma’ parameters. Comparison of this topology matched previous published relationships of these lineages (eg (Boyce et al., 2019; Wang et al., 2023)) which did not justify further exploration with additional likelihood or Bayesian methods.

Genome completeness was computed with BUSCO v5.2.2 (Manni et al., 2021) using the -m genome mode on the masked assembly employing either fungi_odb10 or eukaryota_odb10 marker sets. The analysis was also performed on the predicted proteins using the -m protein mode.

Data used to generate Fig. 2A and 2B were compiled from NCBI, MycoCosm, the Fungal Genome Size Database (zbi.ee/fungal-genomesize/) and (Mohanta and Bae, 2015) and are available as Supplementary File S1. In some cases, multiple strains of the same species have been sequenced.

To identify repetitive sequences in the entomophthoralean genomes we used RepeatModeler v2.0.1 (Flynn et al., 2020) to develop a de novo library of elements and classify their likely lineage using default parameters and LTR finding with LTRStruct option. The elements were classified by similarity to RepBase v20170127 (Bao et al., 2015) using the RepeatClassifier component of RepeatModeler. The de novo repeat library was combined with all identified fungi repeats in RepBase to produce a composite library to identify repetitive regions in the genomes with RepeatMasker v4.1.1 (Smit et al., 2013-2022) genome masking and transposon exploration. Classification of DNA and LTR elements are made by similarity The repeat landscape plots were generated with the RepeatMasker script createRepeatLandscape.pl using default options.

To test for the presence of Repeat Induced Point Mutation patterns in the genome, a RIP index was calculated for 1kb windows with 500 bp offsets using the composite index (Lewis et al., 2009). Summary statistics were calculated to score the number of windows where the composite RIP index was greater than 0 to summarize the % of the genome RIPped. Scripts and summarized values from each genome are part of the https://github.com/zygolife/Emuscae_Comparative repository and archived in zenodo (Stajich et al., 2023b) in the ‘comparative/RIP’ folder.

Domain analysis for seven entomophthoralean proteomes (E. muscae, Entomophaga maimaiga, Z. radicans, P. formicae, S. castrans, N. thromboides, and C. coronatus) was limited to MEROPS, CAZy and Pfam annotations. These annotations were completed according to a pipeline identifying domains by sequence similarity and motif searches https://github.com/stajichlab/Comparative_pipeline. The identified domain counts per species aggregated as total counts or total number of unique genes with a domain were compiled into a single table for comparison between species.

In RStudio v1.4.1717 (RStudio Team, 2022), a custom R v4.1.0 (R Core Team, 2021) script for enrichment analysis is available in https://github.com/zygolife/Emuscae_Comparative repository and archived (Stajich et al., 2023b), which relies on a number of R packages: tidyverse v1.3.1 (Wickham et al., 2019), fmsb v0.7.2 (Nakazawa, 2022), grid (an R base package), gridExtra v2.3 (Auguie, 2017), ComplexUpset v1.3.3 (Krassowski et al., 2022), UpSetR v1.4.0, ggforestplot v0.1.0 (Scheinin et al., 2020), broom v0.7.12 (Robinson et al., 2023) and viridis v0.6.2 (Garnier et al., 2021). For each domain type, the number of accessions containing each domain was counted. Domains that were present in at least two genomes were used for pairwise enrichment analysis among genomes. For each domain type, enrichment of counts was calculated using fmsb::pairwise.fisher.enrichment with Bonferroni correction and a p-value threshold of 0.01 for significance. For each genome, the number of genes found to be significantly different were counted, and the median count for each domain was used to estimate the fold compared to median and a direction (i.e., up or down) compared to median.

Count tables were used to determine lists of domains present within each genome for set analysis including UpSet plots (ComplexUpset and UpSetR) and analysis of unique/missing genes. A second set analysis was conducted with Pfam domains including an additional E. muscae proteome from RNAseq data (EMU-T) generated from the same strain (NCBI GEO GSE111046).

Putative circadian genes were surveyed using Pfam domain annotations. The following domains were used for this survey: 7tm_1 (PF00001), FRQ (PF09421), GATA (PF00320), PAS_3 (PF08447), PAS_9 (PF13426), PAS (PF00989), Bac_rhodopsin (PF01036), GpcrRhopsn4 (PF10192), DNA_photolyase (PF00875), FAD_binding_7 (PF03441), PHY (PF00360) and GAF (PF01590). These were used to filter single-domain circadian candidates from each genome for our domain analysis. Additionally, we considered the expected domain pattern of known circadian genes to identify curated candidates for FRQ (one FRQ domain), WC1 (one GATA and one PAS_3 and one PAS_9), WC2 (one GATA and one PAS_3 and no PAS_9) and 7tm_1 (rhodopsin; one 7tm_1 domain).

RNAi pathway candidates were similarly surveyed using Pfam domain annotations. Domains included in this survey were: RdRP (PF05183), Dicer_dimer (PF03368), PAZ (PF02170), Piwi (PF02171), Ribonuclease_3 (PF00636) and DEAD (PF00270). Identified candidates containing surveyed domains, but we further employed an expected domain pattern to curate candidates for RNAi pathway proteins: RDRP (containing RdRP domain), Dicer (containing Dicer_dimer), Ago (containing both PAZ and Piwi), Dicer_Alt (containing Ribonuclease_3 and either DEAD or PAZ).

Orthologous genes were identified between the compared entomophthoralean species by first taking the longest peptide for each gene to avoid including alternative spliced isoforms in the analyses. Proteins were clustered with OrthoFinder v2.5.2 (Emms and Kelly, 2015) using Diamond v2.0.6.144 (Buchfink et al., 2021) with the ultra-sensitive parameter.

Gene expression was estimated by aligning a pooled set of 27 in vivo samples (whole female fruit flies that had been exposed to E. muscae 24 to 120 hours previous from NCBI #GSE111046 (Supplemental Table 4) to the annotated assembly using Kallisto (v.0.46.1). Genes with fewer than 5 estimated counts across this pooled set were filtered out of the potentially species-specific gene set (a collection of genes that either failed to cluster in any orthogroup across E. muscae, Entomophaga maimaiga, Z. radicans and N. thromboides or appeared in an orthogroup that was exclusively populated by E. muscae genes).

Orthogroups were used to identify methyltransferases that share RID domain architecture. RID is annotated with multiple Pfam DNA_methylase (PF00145) domains. RID candidate proteins, containing more than one DNA_methylase, were identified with the same approach we used to identify circadian protein candidates. Where necessary, homologs that corresponded with genomes used in orthogroup analysis were identified via BLAST. All proteins within orthogroups that contained RID candidates (OG0001715 and OG0003300) were grouped in Geneious Prime (2021.2.2) and aligned using MAFFT (7.450; BLOSUM62) with the sequence of RID (AAM27408.1) from N. crassa. This alignment was used to produce a protein tree using RAxML (8.2.11; GAMMA BLOSUM62 with 1000 bootstraps). This tree was manually rooted to separate orthogroups, maintaining the clade containing RID and OG0001715. The basal E. muscae protein DSO57_1016266-T1 is apparently partial (140 aa) compared to other E. muscae proteins in OG0001715 (both are >1200 aa).

Protein sequences containing extracellular trehalase enzymes (PF01204) were retrieved from the data and aligned using MAFFT, which was used to calculate a maximum likelihood phylogenetic tree with RAxML using default parameters. In addition, total number of genes and number of genes encoding the PFAM domains Lipases (Lipase_3), Subtilisin-like serine peptidases (Peptidase_S8), Trehalases (Trehalase), Trypsins (Trypsin), and Chitinases (Glycohydro_18) in representative fungal species of Zoopagomycota and Ascomycota were retrieved from JGI Mycocosm and compared to our data of E. maimaiga (EMA), Z. radicans (ZRA), N. thromboides (CTH), C. coronatus (Conco1), and Basidiobolus meristosporus (Basme2finSC).

To examine secondary metabolism potential for the E. muscae genome and its relatives, fungiSMASH v6.0.0 of antiSMASH (Blin et al., 2021) was run as part of the funannotate genome annotation steps with the options “--taxon fungi --genefinding-tool none --fullhmmer --clusterhmmer --cb-general --cassis –asf --cb-subclusters --cb-knownclusters.” The predicted clusters were incorporated into the genome annotation submitted to GenBank.

DNA sequence data for members of the E. muscae species complex and other Entomophthora spp. were mined from NCBI GenBank. Novel Sanger sequence data were also generated in support of this analysis, which has been deposited in NCBI (i.e., ARSEF_13514, ARSEF_6918, SoCal_c1 and LTE_c1). DNA was extracted from individual EMSC-killed cadaver drosophilids collected across California as per (Elya et al., 2018) using a QIAamp DNA Micro Kit (QIAGEN #56304). Extracted DNA was used to template PCR reactions with GoTaq polymerase (Promega #M3001) to amplify ITS (primers emITS-1:TGGTAGAGAATGATGGCTGTTG and emITS-4:GCCTCTATGCCTAATTGCCTTT) and/or LSU regions (primers LR0R-4:GTACCCGCTGAACTTAAGC and LR3-1:GGTCCGTGTTTCAAGAC) (James et al., 2006). PCR reactions were enzymatically cleaned with ExoSAP-IT (ThermoFisher #78201.1.ML) per manufacturer’s instructions and submitted to ELIM (Hayward, CA) for Sanger sequencing with both forward and reverse primers. Consensus sequences were assembled in UGENE (Unipro). Sequences were aligned to assess length and quality, and metadata was compiled to determine which strains had available sequence data and morphological data (Supplemental Table 3). Morphological data were collected by cross-referencing Entomophthora strain IDs with published peer-reviewed literature as well as unpublished/raw data provided by two of this paper’s co-authors. Fourteen strains were chosen for inclusion based on the availability of both DNA sequence data and morphological data.

ITS and 28S sequences were aligned separately using MAFFT (Katoh and Standley, 2013) on the Guidance2 server (http://guidance.tau.ac.il/; (Landan and Graur, 2007; Sela et al., 2015)), and individual residues with Guidance scores <0.5 were masked (1.6% of all residues for ITS, 0.1% for 28S). Overall Guidance scores for each locus were 0.955 and 0.995 for ITS and 28S, respectively (alignments with scores approaching 1.0 have high confidence). Nucleotide substitution models were chosen using corrected Akaike information criterion (AICc) scores in Model Test in MEGA X 10.2.6 (Kumar et al., 2018; Stecher et al., 2020). Alignments of each individual locus, and a concatenated alignment of the two, were used in a maximum likelihood (ML) analysis (RAxML 8.2.12; (Stamatakis, 2014)) and a Bayesian inference (BI) analysis (MrBayes 3.2.5; (Ronquist et al., 2012)), for a total of 6 analyses. In brief, for ML analyses, an appropriate model was chosen, partitions were applied (for each locus in the concatenated analysis only), 1,000 bootstrap replicates were used, and the best-scoring tree was identified and bootstrapped in a single run. For BI analyses, MrBayes was allowed to select a substitution model for each data set, and rates were set based on results from Model Test. One cold chain and three heated chains were used for each of 2 runs, and the first 25% of generations were discarded as burn-in. Each analysis was set for 1 million generations, and no additional generations were needed because the standard deviation of split frequencies fell below 0.01. Finally, run parameters were checked for convergence in Tracer 1.7.1 (Rambaut et al., 2018). Trees were viewed and prepared for publication using FigTree 1.4.4 (Rambaut, 2017) and Inkscape 0.92.2 (https://www.inkscape.org/). All resulting trees & alignments are available in zenodo archive (Stajich et al., 2023b).