Gregarine single-cell transcriptomics reveals differential mitochondrial remodeling and adaptation in apicomplexans

Background Apicomplexa is a diverse phylum comprising unicellular endobiotic animal parasites and contains some of the most well-studied microbial eukaryotes including the devastating human pathogens Plasmodium falciparum and Cryptosporidium hominis. In contrast, data on the invertebrate-infecting gregarines remains sparse and their evolutionary relationship to other apicomplexans remains obscure. Most apicomplexans retain a highly modified plastid, while their mitochondria remain metabolically conserved. Cryptosporidium spp. inhabit an anaerobic host-gut environment and represent the known exception, having completely lost their plastid while retaining an extremely reduced mitochondrion that has lost its genome. Recent advances in single-cell sequencing have enabled the first broad genome-scale explorations of gregarines, providing evidence of differential plastid retention throughout the group. However, little is known about the retention and metabolic capacity of gregarine mitochondria. Results Here, we sequenced transcriptomes from five species of gregarines isolated from cockroaches. We combined these data with those from other apicomplexans, performed detailed phylogenomic analyses, and characterized their mitochondrial metabolism. Our results support the placement of Cryptosporidium as the earliest diverging lineage of apicomplexans, which impacts our interpretation of evolutionary events within the phylum. By mapping in silico predictions of core mitochondrial pathways onto our phylogeny, we identified convergently reduced mitochondria. These data show that the electron transport chain has been independently lost three times across the phylum, twice within gregarines. Conclusions Apicomplexan lineages show variable functional restructuring of mitochondrial metabolism that appears to have been driven by adaptations to parasitism and anaerobiosis. Our findings indicate that apicomplexans are rife with convergent adaptations, with shared features including morphology, energy metabolism, and intracellularity.


Background
The phylum Apicomplexa is a lineage comprised primarily of unicellular endobiotic parasites of animals and includes pathogens of agriculturally important livestock (e.g., Eimeria spp. causing coccidiosis) and humans (e.g., Plasmodium falciparum and Toxoplasma gondii, the causative agents of malaria and toxoplasmosis, respectively) [1][2][3]. Due to public health implications, medically important apicomplexans were an early focus for genome sequencing efforts and remain as some of the most well-studied microbial eukaryotes [1,2,4]. Apicomplexan organelles were identified as possible targets for drug therapy development and, accordingly, have been the subject of detailed molecular characterization [5][6][7][8][9][10][11]. Although most apicomplexans (e.g., Toxoplasma, Babesia, and Plasmodium spp.) retain mitochondria akin to aerobic eukaryotes, they have undergone some restructuring. For example, pyruvate dehydrogenase (PDH) has been replaced by branched chain ketoacid dehydrogenase (BCKDH) and the canonical electron transport chain complex I has been replaced by a single protein that encodes an alternative NADH dehydrogenase (NDH2) [12][13][14]. Cryptosporidium spp. represent the exception, as their adaptation to parasitism and hypoxia within the digestive tract has produced a range of functionally reduced mitochondria, including the complete loss of their mitochondrial genomes [13,15]. Cryptosporidium muris and C. andersoni use a common anaerobic pathway to convert pyruvate into acetyl-CoA which is then shuttled to the tricarboxylic acid (TCA) cycle [13]. At the extreme, the mitochondrion-related organelles (MROs) of C. hominis and C. parvum represent some of the most highly reduced mitochondria known and are no longer capable of ATP generation [13,16]. The observation of such modifications established interest in the apicomplexan mitochondrion as a model for understanding fundamental questions regarding eukaryotic evolution and organellar retention.
Although apicomplexans comprise some of the most well-studied microbial eukaryotes, the gregarines represent an important exception. Genomic-scale data are scarce from gregarines, despite the importance of their phylogenetic placement for interpretations of evolutionary events within the phylum [17,18]. Since most of our knowledge on gregarines is derived from microscopic studies and 18S rRNA sequencing [17][18][19][20][21], little is known about the evolution or metabolic potential of gregarine organelles. Recent technical advances in single-cell methods have reduced the barriers to data generation and enabled broad genome-scale explorations of gregarines. In 2019, Janouškovec et al. [22] and Mathur et al. [23] independently published data demonstrating that a "gregarine-type" morphology (conspicuously large single-celled symbionts which infect the intestine, coelom, or reproductive vesicles of invertebrate hosts and possess an attachment apparatus that contains the apical complex at least in the infective stages) has evolved at least three times, highlighting the importance of these understudied lineages to the accurate interpretation of apicomplexan evolution. The only targeted investigation into the mitochondrial metabolism of gregarines was largely speculative, as it was based on the incomplete and highly fragmented Ascogregarina taiwanensis genome [24]. As close relatives to the MROcontaining Cryptosporidium, gregarines provide an ideal model for exploring mitochondrial functional reduction.
Interestingly, phylogenomic analyses performed by Mathur et al. [23] and Janouškovec et al. [22] produced conflicting topologies for the relationships between gregarines and other apicomplexans. Mathur et al. added data from newly sequenced transcriptomes to a dataset generated by Burki et al. [25], producing a tree that placed gregarines as a sister lineage to Cryptosporidium and the other apicomplexans [23]. Using a different set of taxa and an independently derived phylogenomic dataset, the analysis conducted by Janouškovec et al. [22] supports a clade containing Cryptosporidium and gregarines as sister to other apicomplexans. These conflicting evolutionary histories have significant impacts on our interpretations of the evolutionary events that include patterns of organellar reduction and transitions to intracellularity. Therefore, these relationships must be resolved before meaningful conclusions on organelle and life history evolution can be made.
In this study, we explored the impact of adaptations to the host environment and parasitism on apicomplexan mitochondrial metabolism, with a particular focus on gregarines. We expanded the available genome-scale data for understudied apicomplexans by generating transcriptomes for five gregarines isolated from terrestrial hosts. These data were combined with all publicly available gregarine data to conduct the most taxonomically rich phylogenomic analyses of Apicomplexa to date. We robustly placed Cryptosporidium as the earliest diverging apicomplexan lineage. Within this evolutionary framework, we characterized the previously unknown mitochondrial metabolism of gregarines and provide evidence for multiple independent mitochondrial functional reductions and electron transport chain losses within the Apicomplexa.

Results and discussion
Transcriptomes were generated for gregarines isolated from five species of cockroaches (Additional file 1: Table  S1). Small subunit rRNA (SSU) sequences from the newly sequenced transcriptomes were searched against the NCBI nr database to confirm species identifications of isolated gregarines from different cockroach species. SSU sequences from Blabericola migrator (isolated from Gromphadorhina portentosa) and Protomagalhaensia wolfi (isolated from Nauphoeta cinerea) were each > 99.9% similar to published data for these species. The SSU sequence similarity of gregarines from Polyphaga aegyptiaca (from here on Gregarina sp. Poly) and those from Pseudoderopeltis sp. (from here on Gregarina sp. Pseudo) were each~96% similar to Gregarina blattarum (FJ459741.1). The SSU from gregarines isolated from Gyna caffrorum (from here on Protomagalhaensia sp. Gyna) was~92% similar to the published SSU from Protomagalhaensia wolfi (FJ459758.1).
To explore the evolutionary relationships within the apicomplexans, we reconstructed phylogenies using two different phylogenomic datasets, based on those from previously published studies with conflicting results [22,23]. Mathur et al. [23] used 39 taxa and 198 genes (out of a possible 248 genes [26]), compared to 74 taxa and 246 genes in our version of the dataset (from here forward called dataset A). Janouškovec et al. [22] utilized 50 taxa and 296 genes (based on original dataset of 339 genes [27]), while we have retained 299 genes and 74 taxa (from here forward called dataset B). Overall, our dataset A (246 genes) and dataset B (299 genes) had 27.72% and 32.02% missing data, respectively.
Both datasets recovered identical topologies for the relationships between apicomplexans under maximum likelihood (ML) and Bayesian inference (BI) analyses: Cryptosporidium is the earliest diverging lineage, with gregarines branching sister to a clade containing Agamococcidiorida, Coccidia, Nephromycida, Haemospororida, and Piroplasmorida (from here on referred to as "core" apicomplexans) (  Fig. S6 for individual ML and BI trees). The relationship of gregarines with core apicomplexans was strongly supported by dataset B (99 ML bootstrap (BS), 1.0 BI posterior probability (PP)); however, this clade received lower support from dataset A (81 BS, 0.51 PP). Interestingly, this topology was not recovered by Janouškovec et al. [22] or by Mathur et al. [23] and therefore was the subject of further investigation (see below). The archigregarine Selenidium pygospionis and the blastogregarine Siedleckia nematoides form a clade branching sister to eugregarines. However, with sequence data available for only one member of each group, this branching pattern may change with additional data. Distinct and fully supported clades of eugregarines branched in correlation to their host organisms, habitat, and the localization of infection within the host (Fig. 2). Although Actinocephaloidea is represented by only a single transcriptomic dataset here, the relationship of Actinocephaloidea and Gregarinoidea has been repeatedly supported with the use of more taxa in phylogenetic analyses of 18S [20], complete rRNA operon [28], and phylogenomic datasets [23]. Janouškovec et al. [22] similarly recovered these groups as sister lineages using three individual actinocephaloids merged into one hybrid dataset in their phylogenomic analysis, but also recovered the similar relationship using all three separately, but with each containing a high amount of missing data. Rather than incorporate numerous fragmented actinocephaloid datasets (e.g., the Ascogregarina taiwanensis genome data contains 2.5% complete BUSCO genes) to evaluate this repeatedly established relationship, we selected the highly complete Monocystis agilis dataset (86.5% complete or partial BUSCO genes; Additional file 1: Table S1) to represent Actinocephaloidea. In our analyses, we are still able to confidently reconfirm the relationship of Actinocephaloidea and Gregarinoidea as sister groups. The long branches of many of these lineages, combined with a lack of data from numerous families, indicate the need for an increased sampling effort for gregarines from diverse hosts to test whether these co-evolutionary patterns will hold up. However, based on currently available data, it appears that gregarines have transitioned at least twice between marine and terrestrial hosts. Co-evolution of gregarines and their hosts is exemplified in the monophyly of cockroach-infesting gregarines branching as sister to G. niphandrodes (from the beetle Tenebrio molitor). Furthermore, gregarines originating from Blaberidae cockroaches (B. migrator, P. wolfi, Protomagalhaensia sp. Gyna) form a clade, excluding those from other cockroach families Blattidae and Corydiidae (Gregarina sp. Poly, Gregarina sp. Pseudo). However, more data are needed to draw meaningful conclusions.
To further explore the support and robustness of the three topologies, we performed several additional analyses on each dataset: (1) gene concordance factors (gCF) based on single genes and on the partitioning scheme, (2) approximately unbiased (AU) topology test, (3) random gene resampling, and (4) identification and removal of genes that could strongly influence a particular topology. The gCF scores for all three topologies were below 10% based on individual genes and below 20% based on the partitioning scheme but were highest for the A+G topology with a score of 15.5% (dataset A) and 18.6% (dataset B) based on the partitioning scheme (Additional file 8: Table S2). However, Minh et al. [29] suggest that nodes with low gCF scores should be the focus of further scrutiny. The AU topology test only rejected the G+C topology based on dataset B (P G + C = 0.0313) (Additional file 9: Table S3), further highlighting the difficulty in placing the root of the Apicomplexa. Random gene resampling unsurprisingly showed the highest variability for the smallest datasets (20%) (Fig. 3b)  A+G and A+C topologies were similar across all levels of gene subsampling within dataset A, and support for the A+G topology was dominant in dataset B (Fig. 3b).
Genes that strongly influence a particular topology were identified in each dataset by averaging the absolute difference in gene-wise log-likelihood for each pair of the three possible topologies, according to the protocol described by Shen et al. [30]. After removing outlier genes from dataset A (11 genes), the topology switched to the A+C with 86 BS support (100 PMSF bootstrap replicates) while dataset B (13 genes) topology remained the same with 96 BS support.
To further investigate the differences between datasets A and B, we have identified overlapping and unique gene sets. There are 153 genes shared between the two datasets, with 93 genes being unique to dataset A and 146 Fig. 2 Evolution of apicomplexan mitochondrial metabolism. a Cartoon phylogeny of apicomplexans and their close relatives with Coulson plots representing the presence and absence of the tricarboxylic acid cycle (TCA), electron transport chain complexes (ETC), and other genes important for apicomplexan mitochondrial metabolism. The "#" symbol and lighter coloration for MQO in chrompodellids are to signify that they retain malate dehydrogenase rather than malate:quinone oxidoreductase for their TCA cycle. Genes encoded on the mitochondrial genome are indicated with an asterisk. If mitochondrion genome data are absent for a lineage, the corresponding piece of the Coulson plot is colored gray where the gene is presumed to exist due to retention of other genes in the associated complex. Branch names include taxonomy, host information, localization of parasites within the host (DS, digestive system; RS, reproductive system; C, coelom), and the number of genomes and transcriptomes included in each lineage. Gold stars on nodes indicate transitions to an intracellular lifestyle with the star next to the name for Chrompodellids representing an independent event within the clade giving rise to Piridium sociable. Branches on the phylogeny are colored according to in silico predictions for mitochondrial metabolism that are similarly colored and shown as b representing regular aerobic apicomplexan mitochondria which is depicted in blue and c representing an anaerobic mitochondrial metabolism. The functional role of the ETC complex V (ATP synthase) in Cryptosporidium muris and C. andersoni is shown in gray genes being unique to dataset B (Additional file 10: Table S4). Maximum likelihood trees were reconstructed based on subsets of the shared and unique genes. The trees generated based on the shared genes for both datasets, as well as from genes unique to dataset B, all resulted in the A+G topology (Additional file 11: Table   S5). In contrast, the genes unique to dataset A produced a tree with the A+C topology (Additional file 11: Table  S5) and are likely the cause of lower support for the A+ G topology observed in dataset A.
Since many lineages within the apicomplexans formed relatively long branches, we also tested the possible Effects of removing fast evolving and heterotacheous sites, and random gene subsampling on apicomplexan relationships. a Colored schematic of the three possible relationships between Cryptosporidium spp., gregarines, and core apicomplexans. A+G (yellow) shows monophyly of core apicomplexans and gregarines, A+C (purple) shows monophyly of core apicomplexans and Cryptosporidium spp., and G+C (green) shows monophyly of gregarines and Cryptosporidium spp. b Box-and-whisker plots showing support for randomly sampled subsets of genes from each total dataset (dataset A on the left and dataset B on the right). Non-parametric bootstrap support (n = 100) values are on the y-axis and subsample percentage is shown on the x-axis. The number of individual datasets required to sample every gene with 95% probability of sampling every gene for each percentage of genes being sampled is shown in parentheses. Each bipartition of interest is shown in a different color with A+G, A+C, and G+C corresponding to the colors in a. c Graphs plotting support for bipartitions of interest after the stepwise removal of the 6000 fastest evolving sites (top) and 6000 most heterotachous sites (bottom) until all sites are removed from each dataset (dataset A on the left and dataset B on the right). Nonparametric bootstrap support (n = 100) values are on the y-axis and the number of sites removed, measured in thousands, is shown on the x-axis impact of systematic error (e.g., long branch attraction (LBA) or non-optimal model usage) driving support for any of the topologies. We performed serial fast evolving site removal (FSR) [31,32] and heterotacheous site removal (HSR) [33], with the 6000 fastest or most heterotacheous sites being removed at each step (Fig. 3c). For dataset A, there is a short rise of support for the A+C topology, which then disappears in favor of the A+G topology in both FSR and HSR analyses (Fig. 3c). There is little support for the G+C topology throughout the FSR and HSR analyses of dataset A (Fig. 3c). For dataset B, support for the A+G topology remains > 99 for both FSR and HSR analyses, while the other two topologies are barely supported (Fig. 3c). Given the switch to the A+C topology support in the concatenation of dataset A unique genes, we conducted fast site removal, removing 3000 sites at each step from this dataset as well. In this analysis, after removing 9000 sites, the topology switched from A+C to A+G (Additional file 12: Fig. S7). To further investigate the role of systematic errors contributing to the observed topologies, we ran phylogenomic analyses using the less complex LG model to strengthen a possible artifactual result (compared to the partitioned LG+C60+F+G used in our previous analyses), followed by FSR analysis under this model. Both datasets A and B received higher support for the A+C topology using the LG model. The FSR analyses on both datasets partially rescued the A+G topology that we recovered under the LG+C60+F+G model (Additional file 13: Fig. S8). These findings strongly suggest that the A+C topology recovered by Mathur et al. [23] was likely influenced by a systematic error. It is very likely that the combination of partitioning with mixture models [34], and significantly increased taxon sampling helped to suppress systematic errors in this study. We were unable to assess the driving force behind the topology recovered by Janouškovec et al. [22]; however, we hypothesize that the differences may lie in our increased taxon sampling and use of data partitioning with complex models [34]. It is also possible that the extended taxon sampling improved our resolution of orthologs [35].
The above results show that dataset A cannot strongly distinguish between the A+G and A+C topologies as the topology switches from A+G with bootstrap support 81 to A+C with bootstrap support 86 after the removal of 11 outlier genes (out of 246). Interestingly, these 11 genes do not strictly prefer A+C topology (Additional file 14: Fig. S9). Both FSR and HSR analyses suggest that the support for the A+C topology in dataset A is at least partially driven by systematic errors. The support for the A+C topology declines as fast evolving sites are removed from dataset A (Fig. 3), the LG model analysis of dataset A (Additional file 13: Fig. S8), and in genes unique to dataset A (Additional file 12: Fig. S7).
We conclude that the A+C topology may be an artifact resulting from systematic error; however, we cannot entirely rule out this relationship. The G+C topology was rejected by an AU test for dataset A and received very low support in the absolute majority of our analyses. Based on these results, we consider that A+G topology is most likely correct, given the currently available data. However, it remains possible that improved taxon sampling, especially of novel close relatives of Cryptosporidium, or the development of evolutionary models better suited for resolving deep evolutionary radiations could alter support for this topology.

Implications for apicomplexan evolution
Our topology indicates that evolutionary adaptations to different host environments are ubiquitous throughout the Apicomplexa. Within the gregarines, we observe either two independent transitions from infecting a marine host to a terrestrial host, or a switch from marine hosts to terrestrial, and then back to marine (Fig. 2). Additionally, while the majority of gregarines are described from the digestive system of diverse invertebrate hosts, some Actinocephaloidea reside within the reproductive system of their hosts, requiring adaptation to different host tissues (Fig. 2). Despite the contention between the A+C and A+G relationships observed in our analyses, either topology demonstrates that apicomplexans are prone to transitions between intracellular and extracellular lifestyles. Based on our phylogeny (Fig. 1), we suspect that the apicomplexan ancestor was an extracellular endobiont with intracellularity arising independently on at least three occasions: once in Cryptosporidium, again in the ancestor of Coccidia, and a third time in the shared ancestor of Haemospororida and Piroplasmorida. It is important to note that Cryptosporidium spp. form a parasitophorous vacuole in the plasma membrane of the host, remaining extracytoplasmic [36], unlike more derived apicomplexan parasites like Toxoplasma and Plasmodium which invade the cytoplasm of the host cell [37,38], supporting independent origins of intracellularity. However, alternative scenarios remain possible: the apicomplexan ancestor may have been an intracellular parasite with gregarines and Agamococcidioroida independently becoming extracellular parasites, or a transition to an extracellular lifestyle occurred after Cryptosporidium diverged and then returned again to intracellularity in the ancestor of Eimeriorina. Interestingly, Piridium sociabile, a parasite that infects the subepithelial connective tissue of whelks [39], was recently placed within the chrompodellids [23]. This placement, which is confirmed here, further demonstrates the propensity of apicomplexans and their close relatives to become intracellular parasites.

Apicomplexan mitochondrial metabolism
To explore how parasitism and adaption to an endobiotic lifestyle impacted gregarine mitochondrial functionality, we searched for metabolically important mitochondrial proteins in a database comprised of 53 transcriptomes and genomes from apicomplexans, chrompodellids, and Squirmidea (Additional file 1: Table S1). We identified 146 proteins involved in mitochondrial metabolism present in these lineages (Additional file 15: Table S6, Additional file 16: Fig. S10). When data were available, proteins encoded on the mitochondrial genomes were included in our mitochondrial metabolic reconstruction (Fig. 2). These data revealed striking diversity of metabolic capacity among gregarine lineages and support three independent losses of the electron transport chain within Apicomplexa.

Pyruvate metabolism
In typical aerobic mitochondrion, pyruvate dehydrogenase (PDH) converts pyruvate into acetyl-CoA that is then oxidized into CO 2 in the TCA cycle, while also reducing NAD+ to NADH. NADH is then utilized by the electron transport chain to generate ATP by oxidative phosphorylation. In more reduced anaerobic MROs, the conversion of pyruvate into acetyl-CoA is commonly facilitated by pyruvate:ferredoxin oxidoreductase (PFO), pyruvate:NADP oxidoreductase (PNO), and/or pyruvate formate lyase (PFL) [40]. Acetyl-CoA can then be metabolized to acetate, while generating ATP from ADP by substrate-level phosphorylation. Cryptosporidium hominis and C. parvum both lack most genes involved in the TCA cycle and the ETC and can only produce ATP via substrate-level phosphorylation, for example during glycolysis or metabolism of acetyl-CoA to acetate. Although PDH can be found in many apicomplexan lineages, previous studies demonstrated that apicomplexan PDH is localized to the apicoplast rather than the mitochondrion [41]. Instead, most apicomplexans import pyruvate into the mitochondrion via the mitochondrial pyruvate carrier (MPC) where it is converted to acetyl-CoA by the BCKDH complex which is then shuttled into the TCA cycle (Fig. 2b). In C. muris, the enzyme PNO converts pyruvate to acetyl-CoA in the cytosol which is presumably then shuttled into the mitochondrion via the acetyl-CoA transporter [42,43]. Once in the mitochondrion, acetyl-CoA is utilized by the TCA cycle (Fig. 2c), producing NAD(P)H that is ultimately oxidized by a reduced respiratory chain consisting of NDH2 and alternative oxidase (AOX) [43]. Archigregarines, blastogregarines, and the eugregarine superfamily Actinocephaloidea all maintain the pathway of BCKDH converting pyruvate to acetyl-CoA akin to the aerobically functioning mitochondria in core apicomplexans (Fig. 2b). In contrast, the other three eugregarine lineages lost BCKDH and instead use a PNO-based pyruvate metabolism pathway, akin to MROs described from other anaerobes (Fig. 2c) [44]. PNO is conspicuously missing in the G. niphandrodes data. Since no other enzyme that metabolizes pyruvate has been identified in G. niphandrodes, the absence of PNO may reflect an incomplete nature of this genome.
Substrate-level phosphorylation generating ATP via ADPforming acetyl-CoA synthetase (ACS) has been characterized in anaerobic lineages across the eukaryotic tree of life with functionally reduced mitochondria [44,45]. Investigations into apicomplexan acetyl-CoA synthetase have focused primarily on the AMP-forming ACS found in all wellcharacterized lineages. This AMP-forming ACS has been described to function in fatty acid transport and the production of acetyl-CoA for protein acetylation and fatty acid elongation [14,43,46,47]. Acetyl-CoA synthetase has also been suggested to function in substrate-level phosphorylation in C. muris, but not in any other apicomplexan [13,44]. Here, in addition to the AMP-forming ACS, we identified an ADPforming ACS found in all eugregarines as well as with patchy distribution throughout Apicomplexa (Additional file 17: Fig.  S11). This ADP-forming ACS has been previously shown to have originated in diverse eukaryotic lineages through multiple lateral gene transfer (LGT) events [48][49][50][51] and is known to produce ATP via substrate-level phosphorylation in the cytosol [52,53]. Known exceptions include Mastigamoeba balamuthi (Amoebozoa: Archamoebae) and Spironucleus salmonicida (Metamonada: Fornicata), which both have ACS localized to their MROs [49,50]. In silico predictions of the ADP-forming ACS identified in some apicomplexan lineages suggest it is localized into cytosol, where it can convert acetyl-CoA into acetate while generating ATP [13].
We recovered a complete or near complete TCA cycle for all lineages in this study, with the exception of several Cryptosporidium species (Fig. 2a; Additional file 15: Table  S6). It has been reported that apicomplexans have replaced malate dehydrogenase (MDH) with malate:quinone oxidoreductase (MQO) [13]. We found that Platyproteum, a sister taxon to Apicomplexa and chrompodellids, has both MDH and MQO and Chrompodellids retain solely MDH, suggesting acquisition of MQO by an ancestor of Squirmidea, followed by the retention of MDH in Chrompodellids and MQO in Apicomplexans.

Multiple independent losses of the electron transport chain
One of the most striking findings is the independent loss of multiple ETC complexes in at least three apicomplexan lineages. Functional reduction of Cryptosporidium MROs has been well characterized over the past few decades [13,16,54,55]. Here, we report additional losses in three eugregarine lineages. Although archigregarines and blastogregarines are each represented by only one single-cell transcriptome, we recover a nearly complete electron transport chains for these early branching gregarines, with the few missing genes likely resulting from partial datasets (Fig. 2a). Inferring gene loss from transcriptomic data can be problematic, especially considering that life-stage-specific metabolic fluxes have been characterized for some apicomplexans like Toxoplasma and Plasmodium, which could cause genes to be absent from the transcriptomes [56][57][58]. However, given that gregarine life cycles are much less complex than that of the "core Apicomplexa," and all analyzed datasets are based on sampling of the same life stage (trophozoites), differences between gregarine lineages are very unlikely to result from life-stage-specific metabolic fluxes. Even with these considerations, we observe strong patterns of ETC loss in eugregarines as not a single protein from ETC complexes III-V is recovered from Urosporoidea (n = 5), Porosporoidea (n = 2), or Gregarinoidea (n = 5, plus one genome), which all infect the host digestive system or coelom (Fig. 2a). However, this pattern is rendered non-monophyletic by Actinocephaloidea, which appears to maintain a fully functioning ETC (again, the few unrecovered genes are attributed to partial data). Monocystis agilis (Actinocephaloidea) infects the reproductive system of terrestrial worms and may have greater exposure to oxygen allowing it to generate ATP through oxidative phosphorylation, unlike the digestive tract-infecting eugregarines. Accordingly, we suspect the differential ETC loss in Urosporoidea, Porosporoidea, and Gregarinoidea is likely a result of advanced adaptation to the hypoxic conditions found within the digestive tract of their respective hosts. Furthermore, the lack of cytochrome b, cytochrome c oxidase subunit 1, or cytochrome c oxidase subunit 3 from Urosporoidea, Porosporoidea, and Gregarinoidea sequence data suggests a possible loss of the mitochondrial genome in these lineages. Similar patterns of mitochondrial genome loss and adaptation to anaerobiosis have been observed in Cryptosporidium and other eukaryotic gut endobionts throughout the tree of life [44,45,[59][60][61][62]. Interestingly, the distribution of prokaryotes within the intestinal tract of the polychaete Neanthes glandicincta has shown increased density of aerobic species in the anterior, with anaerobes representing the majority in the posterior of the gut [63]. Accordingly, the retention of a complete ETC capable of oxidative phosphorylation in archigregarines and blastogregarines may be explained by their localization within the host digestive system; however, more data is necessary to test this hypothesis.
In most eukaryotes, ubiquinol (UQ) serves as an electron carrier through the electron transport chain. Curiously, we were unable to identify the majority of the enzymes involved in ubiquinol synthesis in eugregarines (transcriptomes from 13 species and the G. niphandrodes genome), with the exception of COQ3 which is conserved in all groups, and COQ6 which is found only in Actinocephaloidea. In few, but varied exceptions, anaerobes will instead use rhodoquinone (RQ) as an electron carrier [45,[64][65][66]. In aerobic conditions, complex II oxidizes succinate to fumarate, thereby transferring electrons for the reduction of ubiquinone to ubiquinol, while in anaerobic conditions complex II functions in the reverse, regenerating UQ by utilizing electrons from ubiquinol to reduce fumarate to succinate [67]. The lower redox potential of RQ over UQ enables rhodoquinone to transfer electrons via complex II and more favorably reduce fumarate to succinate [67]. However, endogenous production of RQ necessitates a source of UQ [68][69][70]. Utilization of exogenous UQ has been demonstrated in mice [71], and UQ obtained from bacterial prey has been proposed as the source of UQ which is then converted into RQ by the laterally acquired methyltransferase rquA in Pygsuia [66]. We were unable to identify rquA in any apicomplexans in this study negating this as a possibility for eugregarines (data not shown).
The conservation of COQ3 in eugregarines suggests its involvement in ubiquinol (UQ) production, but its synthesis otherwise remains a mystery. It is possible that eugregarines are utilizing COQ3 for alternative functions or that they may have lost the ability to fully synthesize ubiquinol like Cryptosporidium ubiquitum [43] and the presence of COQ3 may represent a final vestige of a fading UQ synthesis pathway. We can think of four possibilities to explain the absence of UQ biosynthesis proteins in eugregarines. (1) Either these proteins have become extremely divergent in eugregarines and were unable to be recovered in our searches; (2) eugregarines have replaced ubiquinol with another Q source, for which the synthesis pathway remains unknown (like RQ in the ciliate Nyctotherus ovalis [64]); (3) eugregarines are utilizing COQ3 to methylate exogenously acquired 3-demethylubiquinol-n for the final step of UQ production; or (4) exogenous Q could satisfy the UQ requirement for the complete ETC retained by Actinocephaloidea as shown in UQ-deficient mice [71], although the mechanism for its acquisition remains unclear [72]. The absence of dihydroorotate dehydrogenase (DHOD) in other eugregarines with a reduced ETC, and the additional loss of glycerol-3phosphate dehydrogenase in Gregarinoidea suggests reduced Q dependency in these lineages (Additional file 15: Table S6). This reduced capacity for ubiquinone biosynthesis likely indicates host dependency for a source of Q in eugregarines. Along with the previously described ETC reductions, these findings support the hypothesis that patterns of reductive mitochondrial evolution in eugregarines were co-mediated by adaptions to parasitism and an anaerobic environment in a host digestive system.
The presence of ATP synthases has been associated with the formation of mitochondrial cristae [73], and a recent study identified 17 ATP synthase proteins that contribute to the curvature of cristae in the Toxoplasma gondii mitochondrion [74]. This study specifically implicated ATPTG11 (called ASAP19 by [75] and in Table  S6) as essential to the structural architecture of T. gondii cristae [74]. Investigations of medically important apicomplexans like T. gondii have characterized a link between bioenergetics and cristae formation [74]. However, gregarines complicate our understanding of these structures despite there being relatively few ultrastructural studies compared to their vast diversity. Cristae have been observed in the mitochondria of some archigregarines and blastogregarines [21,76,77], which are both lineages that retain a complete ETC (Fig. 2). However, other archigregarines appear to lack cristae or present few structures called blistered vesicles that are connected to the inner membrane of the mitochondrion [78]. Surprisingly, members of lineages lacking an ETC in this study (e.g., Urosporoidea and Gregarinoidea) have also been shown to retain cristae [79][80][81]. We were unable to identify ATPGT11/ASAP19 in any gregarine lineage. Other proteins linked to cristae architecture [74] were identified only from archigregarines, blastogregarines, and actinocephaloideans (see ATPTG proteins in Additional file 15: Table S6), all of which retain a functional ETC. The presence of cristae in gregarine mitochondria that lack an ETC (particularly complex V-ATP synthase), as well as the lack of cristae in mitochondria of gregarines with a complete ETC, would be of particular interest in future investigations.

Other mitochondrial pathways
Mitochondrial Fe-S cluster synthesis has been considered the core functional pathway preventing the complete loss of mitochondria/MROs in numerous lineages [54,[82][83][84]. Accordingly, Monocercomonoides exilis (Metamonada: Preaxostyla) is the only eukaryote to have completely lost the organelle and has replaced the mitochondrial ISC pathway with the bacterial SUF system through lateral gene transfer. This laterally acquired SUF system functions in the cytosol in M. exilis [83]. The Pygsuia ISC pathway was similarly replaced by a prokaryotic SUF system that shares a common origin with the fused SUFCB proteins identified in Blastocystis [85]. Additionally, Mastigamoeba balamuthi and Entamoeba histolytica (Amoebozoa: Archamoebae) have both replaced their ISC with a NIF pathway of prokaryotic origin [86][87][88]. However, in apicomplexans, even the highly reduced MROs of C. parvum and C. hominis retain a functional ISC system [54]. Although we did not find the complete suite of genes involved in the ISC pathway in any one gregarine, the overall patterns of gene retention support that this pathway remains functional in all gregarines studied to date. The findings here further support the role of the mitochondrial ISC pathway in the retention of MROs.
Cryptosporidium is known to have reduced metabolic capacity and has lost several pathways for amino acid and fatty acid metabolism [43]. Although we are relying primarily on single-cell transcriptomes, we identified much broader patterns of amino acid metabolism in the gregarine mitochondria compared to those of Cryptosporidium spp. For example, most Cryptosporidium species have lost all genes involved in the glycine cleavage system (GCS) [15]. Only the GCS L protein is retained in C. andersoni, C. muris, Porosporoidea, and Gregarinoidea; however, this protein is known to be widely involved in energy metabolism and maintaining redox balance, including its role as the E3 component in the BCKDH complex [15,89]. In archigregarines, blastogregarines, urosporideans, and actinocephaloideans, we identified a complete GCS pathway (Table S6). Aside from the GCS pathway, we found cystathione beta synthase, alanine aminotransferase, glutamate dehydrogenase, and several other proteins involved in amino acid and fatty acid metabolism are conserved across gregarine lineages.

The role of lateral gene transfer in apicomplexan mitochondrial metabolism
Lateral gene transfer (LGT) has been implicated as an important driver in environmental adaptation with an ever-increasing number of well-supported events occurring in eukaryotes [50,51,66,[90][91][92]. Here, we show that ADP-forming acetyl-CoA synthetase (ACS) has been acquired by the ancestor of chrompodellids and Apicomplexa by LGT from a prokaryotic donor (Additional file 17: Fig. S11). The acquisition of this ADPforming ACS likely facilitates ATP production via substrate-level phosphorylation in the cytosol, providing flexibility in their metabolic capacity under dynamic environmental conditions, as in Giardia intestinalis [45].
We also observe strong support for an independent origin of malic enzyme in the gregarines sequenced for this study. Malic enzyme converts malate to pyruvate which is subsequently converted to acetyl-CoA and can be utilized by the TCA cycle. Seeber et al. [15] proposed that in C. parvum and C. hominis, the NADH generated, in part by malic enzyme, can act as a proton pump and lead to a simplified respiratory chain comprised of NAD(P) transhydrogenase, NDH2, and AOX. Malic enzyme is widely conserved throughout Apicomplexa; however, it appears to have been lost in the ancestor to Gregarinoidea and remains absent in G. niphandroides. Interestingly, the cockroach-infecting gregarines sequenced here have replaced the missing malic enzyme through an LGT from proteobacteria (Additional file 18: Fig. S12).
The ancestor of the Apicomplexa lost the TCA cycle protein MDH, which has been replaced by the acquisition of MQO through an LGT from epsilon-proteobacteria [13,93]. We identified MQO in all apicomplexan lineages except Urospororidea, and Porospororidea, which are represented solely by single-cell transcriptome data, so we consider its absence inconclusive. Both MDH and MQO have been observed in Squirmidea (Platyproteum) and two chrompodellids (Alphamonas and Vitrella). The enzymes may be functioning independently based on cellular conditions as was observed in E. coli [94], or collaboratively as in Corynebacterium glutamicum [95]. Combined, these three LGT events represent important adaptations to mitochondrial metabolism in apicomplexans. Future investigations of understudied apicomplexans will help to pinpoint the origins of adaptive LGTs, reveal patterns of organellar evolution and genomic adaptions to specific host environments, and establish the order of transitions between intracellular and extracellular lifestyles, and those between marine and terrestrial hosts.

Conclusions
Our phylogenomic analyses place Cryptosporidium as a sister lineage to the rest of Apicomplexa, with gregarines and core apicomplexans each forming fully supported clades. However, deeper analysis highlights that the branching pattern observed in the early apicomplexan radiation remains highly contentious, and the topology of gregarines forming sister lineage to a clade of Cryptosporidium and core apicomplexans cannot be fully refuted. Regardless of whether Cryptosporidium or gregarines diverged first, this evolutionary framework suggests that apicomplexans are evolutionarily malleable, with phenotypic adaptations being driven by the host environment. Intracellularity has evolved at least three times independently within the Apicomplexa: once in Cryptosporidium, again in Coccidia, and a third time in the ancestor of Haemospororida and Piroplasmorida. Our phylogeny also supports a propensity for lifestyle transitions specifically within gregarines, with at least two transitions between marine and terrestrial hosts, as well as adaptations to infect different host organs. Additionally, based on our finding that Cryptosporidium represents an evolutionarily distinct lineage from gregarines, it should no longer be considered part of the Gregarinasina or the Gregarinomorphea (as in Adl et al. [96] or Cavalier-Smith [20]).
We characterized mitochondrial metabolism across the Apicomplexa and observed two distinct modes of metabolism within gregarines. Blastogregarines, archigregarines, and Actinocephaloidea possess mitochondria with a fully functioning electron transport chain, akin to core apicomplexans. All other eugregarines appear to lack a mitochondrial genome and have completely lost complexes III-V of the ETC and BCKDH as part of a switch to anaerobic mitochondrial metabolism. Here, pyruvate is converted to acetyl-CoA by PNO or PFL. Acetyl-CoA can then be shuttled to the organelle via acetyl-CoA transporter, utilized by the TCA cycle, and NADH is then oxidized by a reduced respiratory chain consisting of NDH2 and AOX. Alternatively, acetyl-CoA can be metabolized in the cytosol to acetate by acetyl-CoA synthetase, while generating ATP. The mitochondrial or cytosolic localization of PNO/PFL enzymes and acetyl-CoA will help us to fully understand the metabolism in these gregarine lineages. These two types of gregarine mitochondrial metabolism are not restricted to monophyletic groups and the electron transport chain has been reduced at least three times during apicomplexan evolution, twice within gregarines.
We hypothesize that the observed distribution of mitochondrial reduction within gregarines corresponds with their adaptation to an anaerobic environment within their hosts and dependence on the host itself. Actinocephaloidea, blastogregarines, and archigregarines retained seemingly regular mitochondrial functions, while the other eugregarines have twice independently transitioned to functionally reduced mitochondria. The pattern of adaptation corresponds well with the first hypothetical step toward anaerobiosis-loss of complexes III and IV [62]. This has been also observed in many other gut-inhabiting endobionts across the tree of life, including Cryptosporidium [44,45,62]. Moreover, these eugregarines have completely lost complex V and BCKDH. Instead, they possess PNO, which converts pyruvate to acetyl-CoA in anaerobes [44,96], and ADPforming acetyl-CoA synthetase that likely facilitates cytosolic substrate-level phosphorylation, producing acetate as the final metabolic product.
Recent advances in single-cell sequencing have opened the door to genomic characterization of unculturable lineages throughout the tree of life. Unsurprisingly, these studies provide profound insight into the understanding of evolution, ecology, and biochemistry. Gregarines provide a textbook example, and until now, their uncertain placement within Apicomplexa hindered our understanding of evolutionary events within this clade of medically important parasites. Gregarine metabolism has been shaped by adaptions to parasitism and anaerobiosis, and as such, the mitochondrial remodeling observed in distinct gregarine lineages highlights the importance of this group to our understanding of organellar evolution and retention in eukaryotes. Future investigations of under-explored apicomplexans will further help to pinpoint the origins of adaptive LGTs, establish the order of transitions between intra-and extracellular lifestyles, as well as host habitats, and continue to reveal patterns of organellar evolution and genomic adaptions to specific host environments.

Specimen collection, RNA isolation and sequencing, and assembly
Cockroach individuals were obtained from cultures housed in the Department of Zoology, Charles University, Czech Republic. Gromphadorhina portentosa, Gyna caffrorum, and Nauphoeta cinerea originated from laboratory cultures of unknown origin. A colony of Polyphaga aegyptiaca was established in 2012 from individuals collected on Corfu, Greece, and a colony of Pseudoderopeltis sp. was established from individuals collected in 2013 in Ethiopia. Individual gregarine trophozoites were hand-picked from G. caffrorum, P. aegyptiaca, Pseudoderopeltis sp., and N. cinerea on an Olympus IX70 inverted microscope using handdrawn glass pipettes. The gregarines were washed 3 times in phosphate-buffered saline (PBS) and placed in a 0.2-ml tube containing 2 μl of cell lysis buffer for the SmartSeq2 protocol [97], flash frozen, and stored in a − 80°C freezer. Single-cell RNA isolation, cDNA generation, and amplification were completed according to the SmartSeq2 protocol with only 11 amplification cycles [97]. Heavily infected individuals of Gromphadorhina portentosa and Polyphaga aegyptiaca were dissected and the gut contents were emptied into a petri dish with PBS. Under a Meopta DM23 dissecting microscope, 40-60 trophozoites were transferred by pipette to a clean dish of PBS to wash the cells of other gut content and host tissue. Cells were then transferred to a 2-ml tube and total RNA was extracted using the RNAqueous RNA isolation kit (Invitrogen) following the manufacturer protocol. The resulting total RNA from 1 extraction of gregarines from P. aegyptiaca (the first isolation of a gregarine from a Polyphagidae cockroach), 2 extractions of Blabericola migrator isolated from G. portentosa, as well as cDNA independently generated from 3 gregarines from Pseudoderopeltis sp., 2 Gregarina sp. Poly, 2 gregarines isolated from Gyna caffrorum, and 2 Protomagalhaensia wolfi isolated from N. cinerea were sent to Macrogen for library preparation and sequencing. Nextera XT libraries were generated for all single cells and TruSeq stranded mRNA libraries were generated for total RNA. Libraries were sequenced on an Illumina NexSeq producing 30 million 150 bp pairedend reads per library. The resulting reads were checked for quality using FastQC. Low-quality bases and adapters were removed using Trimmomatic [98]. The remaining reads were assembled using rnaSPAdes v3.13 [99], and Transdecoder was used for ORF prediction and translation. Completeness of all transcriptomes generated in this study, along with all public apicomplexan data used for phylogenomic analyses, was assessed using BUSCO with the Apicomplexa gene set (Additional file 1: Table S1) [100]. Gregarine species identifications were confirmed (> 99% identity) or remain ambiguous (< 99% identity, e.g., Protomagalhaensia sp. Gyna) based on BLAST searches of the SSU rRNA sequences recovered from the gregarine transcriptomes and general morphology.

Phylogenomic dataset construction and analyses
The gene sets used in two independently published phylogenomic studies of gregarines were utilized for our analyses [22,23]. All publicly available Apicomplexan genomes/transcriptomes [101][102][103][104][105] were considered for initial ortholog selection in each gene set and run through BUSCO [100] to assess completeness (see Additional file 1: Table S1). Prior to assembling phylogenomic datasets, our assemblies were run through Winston Cleaner [106] to identify and highlight likely cases of cross contamination. To further assist in the identification of contamination and paralogs, transcriptomes and genomes spanning the diversity of the eukaryotic tree of life were included in the initial "unparsed" datasets which are available on Figshare [107]. A BLAST database was constructed from all proteins considered for phylogenomic analyses (Additional file 1: Table S1). A representative seed sequence for each gene was used as a query to identify putative orthologs by BLAST search (e-value < 1e−10). Up to the top five blast hits for each species in the dataset were combined with their respective genes to form a 246-gene dataset that corresponded to the Burki et al. [25] dataset used by Mathur et al. [23] and the 307-gene dataset from Derrelle et al. [27] used by Janouškovec et al. [22]. For each dataset, the individual genes were filtered for sequencing errors and non-homologous sites using PREQ-UAL [108], aligned using MAFFT (--globalpair --maxiterate 1000) [109], alignment uncertainty and errors were filtered using DIVVIER (--partial -mincol 4 -divvygap) [110], and the filtered alignments were trimmed of sites comprised of > 80% gaps using trimAL (-gt 0.2) [111]. Trees were constructed from the trimmed alignments using RAxML (PROTGAMMALG with 100 rapid bootstraps) [112]. Each tree was then carefully manually inspected to identify the correct ortholog for each species and remove paralogous sequences and contaminations. In the case of inparalogs (gene duplications strictly within a single organism), the longest sequence or sequence with the shortest branch length was selected as the ortholog. For midparalogs, only the side of the inferred gene duplication that contained the most data (as the total sum of retained sequence length of selected orthologs) based on the tree was used for ortholog selection, even if this led to a loss of a specific taxon for the particular gene. To clarify, if one side of an inferred duplication had 7 taxa that were each 50% complete, while the other side contained 5 taxa that were all 90% complete, the side with 5 taxa was selected. In the event the total amount of retained data was similar for each side a duplication, the side with greater representation of species was selected (e.g., a side with 9 taxa that average 50% complete would be chosen over a side with 7 taxa that average 65% complete). Once orthologs for all taxa were identified, ortholog-only datasets were filtered and aligned according to the previously described methods and were then trimmed using BMGE (-g 0.3) [113]. Final taxon and gene selection were carried out on these alignments with a threshold of 80% missing data to maximize representation in each dataset and to accommodate for the inclusion of important taxa (e.g., Eimeria falciformis). Our final phylogenomic datasets were comprised of 246 genes (63,201 sites) (corresponding to Mathur et al. [23]) and 299 genes (89,675 sites) (corresponding to Janouškovec et al. [22]) and 74 taxa. These datasets were once again filtered, aligned, and trimmed in the same manner. The resulting alignments were concatenated into a supermatrix (one for each dataset), partitioned using PartitionFinder2 under the LG+G model [114], and subjected to maximum likelihood analysis in IQtree (LG+C60+F+G) [115] with 1000 PMSF bootstrap replicates. Bayesian inference was conducted in Phylobayes [116] under the CAT-GTR model with 2 chains for each dataset run for 7000 generations each (burnin = 20%; maxdiff = 1 and 0.081 for datasets A and B, respectively). The average absolute difference in gene-wise loglikelihood scores, as calculated in IQtree under the LG+ C20+G model, for the A+C, A+G, and G+C topologies was calculated following Shen et al. [30] for each dataset. After removing outlier genes identified in each dataset, maximum likelihood trees were reconstructed in IQtree (LG+C60+F+G) [115] with 100 PMSF bootstrap replicates. Additional datasets comprised of genes unique to dataset A, genes unique to dataset B, and genes shared between the two datasets were also subjected to maximum likelihood analysis in IQtree (LG+C60+F+G). Additionally, maximum likelihood trees for the full datasets were constructed under the LG model were computed using IQtree (LG+F+G) with 1000 ultrafast bootstraps. Single gene datasets including all considered sequences (i.e., including orthologs and paralogs [35];) and ortholog-only datasets are available on figshare [107]. Log files for phylogenomic analyses containing commands and processor information are also available on figshare [117] as recommended by Shen et al. [118].

Removal of fast evolving and heterotacheous sites, and random gene subsampling
For each of our phylogenomic datasets, PhyloFisher scripts [119] were used for fast site removal, heterotacheous site removal, and random gene subsampling. The fastest evolving sites were sequentially removed in 6000 site chunks generating new alternative datasets at each step until all sites were removed (as in [31,32]). Similarly, the most heterotacheous sites [33] were removed in a stepwise fashion, 6000 sites at a time, producing iteratively smaller datasets until no further sites could be removed. Genes from our 74 taxon datasets were randomly subsampled in sets of 20%, 40%, 60%, and 80% of the complete dataset, under the default 95% confidence interval setting as in Brown et al. [31]. The dataset comprised of genes unique to dataset A was similarly subjected to fast site removal, with 3000 sites being removed at each step. All datasets generated by these scripts were then subjected to phylogenetic analysis using IQtree under the LG+C60+F+G model with 100 non-parametric bootstrap replicates generated using PMSF and the fast setting in IQtree. In the case of fast evolving site removal under the LG model, 100 rapid bootstraps were generated using RAxML v8 (PROTCATLGF).

Constrained trees, concordance factors, and topology tests
Both 74 taxon datasets used for the initial phylogenomic analyses, were subjected to constrained maximum likelihood tree reconstruction in IQtree (LG+C60+F+G) using the same partitioning scheme applied in our initial analyses. The overall topologies published in Mathur et al. [23] (i.e., gregarines are sister to other apicomplexans) and Janouškovec et al. [22] (monophyletic gregarines and Cryptosporidium spp. as sister to other apicomplexans) were used as constraints. Gene concordance factors (gCF) for the best ML tree for our initial phylogenomic analysis and the constrained topologies were calculated using IQtree [29]. Additionally, the approximately unbiased (AU) test [120] was conducted on the maximum likelihood tree, the constrained trees, and 100 distinct local topologies saved during the initial ML analysis (-wt option in IQtree) as implemented by IQtree [115].

In silico predictions of mitochondrial metabolism
Genes of interest were identified from previously published studies of anaerobic mitochondria, the published mitochondrial proteome of Tetrahymena thermophila [121], and recent publications on mitochondrial metabolism of Apicomplexa [15,51,61,65,75,85,[122][123][124]. Candidate sequences from our gregarine data and other publicly available genomescale datasets for apicomplexans and their close relatives (Additional file 1: Table S1) were recovered by BLAST using representative seed sequences and collecting hits with e-value <1e−5 were extracted and added to the initial datasets for each gene. When available, datasets published by Rotterova et al. [61,124] were used as starting datasets. Alternatively, datasets for phylogenetic analysis were constructed by using BLAST searches of the seed sequences against nr (max target seqs = 100) to thoroughly sample apicomplexans and other close relatives. These BLAST results were supplemented with sequences selected from gene name search results from the NCBI protein and RefSeq databases to ensure broad eukaryotic and prokaryotic taxonomic sampling and were then clustered using CD-Hit [125] at 75%. Each gene dataset was filtered for sequencing errors and non-homologous sites using PREQUAL, aligned using MAFFT (--localpair --maxiterate 1000), and alignment uncertainties and errors were filtered using DIVVIER (-partial -mincol 4 -divvygap) and single gene trees were constructed using IQtree (LG+C20+F+G) with 100 RAxML rapid bootstraps using constructed under the PROTGAM-MALG4X model. Each alignment/tree has been repeatedly inspected by the eye, and sequences were added/removed as necessary. Ultimately, each tree was manually inspected, enabling precise inference of the presence/absence of proteins involved in mitochondrial metabolism in the studied lineages. Prediction of subcellular localization for all sequences identified as apicomplexan homologs in this study was conducted in TargetP v2 [126] and DeepLoc [127]. Files containing all initially considered sequences, final datasets, final alignments, sequences of all selected Apicomplexan homologs, the resulting trees (raw and colored to show selected apicomplexan homologs), and the subcellular localization prediction results for all mitochondrial genes discussed in this study are available on figshare [128].
Additional file 1: Table S1. Sources, uses, and BUSCO completeness for data used in this study. Additional file 8: Table S2. Gene concordance factors for possible apicomplexan relationships.
Additional file 9: Table S3. Approximately unbiased test scores for possible apicomplexan relationships.
Additional file 10: Table S4. Genes unique and overlapping between phylogenomic datasets A and B.
Additional file 11: Table S5. Maximum likelihood topologies produced by unique and shared genes from datasets A and B.
Additional file 12: Fig. S7. Effects of fast evolving site removal on support for bipartitions of interest in the phylogenomic analyses of genes that are unique to dataset A. Graphs plotting support for bipartitions of interest after the stepwise removal of the 3000 fastest evolving sites until all sites are removed from each dataset (dataset A on the top and dataset B on the bottom). 100 RAxML rapid bootstraps values (PROTCATLGF) are on the y-axis and number of sites removed, measured in thousands, is shown on the x-axis.
Additional file 13: Fig. S8. Effects of fast evolving site removal on support for bipartitions of interest in the phylogenomic analyses of each dataset under the less complex LG model. Graphs plotting support for bipartitions of interest after the stepwise removal of the 6000 fastest evolving sites until all sites are removed from each dataset (dataset A on the top and dataset B on the bottom). 100 RAxML rapid bootstraps values (PROTCATLGF) are on the y-axis and number of sites removed, measured in thousands, is shown on the x-axis.
Additional file 14: Fig. S9. Box-and-whisker plots showing the difference in gene likelihood scores (ΔGLS) dataset A and dataset B. ΔGLS values for genes that plot above the upper whisker are shown with the preferred topology of the gene tree (A + C = core apicomplexans with Cryptosporidium; A + G = core apicomplexans with gregarines; G + C = gregarines with Cryptosporidium).
Additional file 15: Table S6. Presence-absence table for genes involved in apicomplexan mitochondrial metabolism.
Additional file 16: Fig. S10. Maximum likelihood trees for all mitochondrial genes discussed in this study inferred using IQtree (LG+C20+F+G). Support values are calculated from 100 RAxML rapid bootstraps under the PROTGAMMALG model. Sequences identified as apicomplexan homologs are colored in red, sequences in purple are suspected apicomplexan contamination of a host rather than the target organism.
Additional file 17: Fig. S11. Maximum likelihood phylogeny of ADPforming ACS reconstructed in IQtree (LG+C20+F+G) with support values from 100 non-parametric PMSF bootstrap replicates. Prokaryotes are blue, Apicomplexa, Squirmidea and chrompodellids are colored red, and other eukaryotes are shown in black. The topology shows moderately strong support for a laterally transferred ADP-forming ACS gene of prokaryotic origin to the shared ancestor of chrompodellids and apicomplexans.
Additional file 18: Fig. S12. Maximum likelihood phylogeny of malic enzyme reconstructed in IQtree (LG+C20+F+G) with support values from 100 non-parametric PMSF bootstrap replicates. Prokaryotes are blue, Apicomplexa, Squirmidea and chrompodellids are colored red, and other eukaryotes are shown in black. The topology shows strong support for an independent acquisition of malic enzyme by members of the Gregarinoidea through lateral gene transfer from proteobacteria. The malic enzyme sequenced in other eugregarines branches with members of the core apicomplexans, chrompodellids, and Squirmidea.
Authors' contributions EDS, KXT, Ma. K, CL, and SR conceived and designed the study. KXT isolated all gregarines and generated data. EDS, KXT, and Ma. K performed the data analyses. SR identified sampled cells. IC, ZKV, and Mi. K inspected an extensive collection of cockroach cultures for infected species and provided host specimens.. EDS, KXT, and Ma. K wrote the initial manuscript. All coauthors commented and edited the manuscript. All authors read and approved the final manuscript. Availability of data and materials All raw read sequencing data have been deposited into NCBI SRA archive under BioProject PRJNA684345 (SRA accessions SRR13276162-SRR13276175). All nucleotide assemblies for newly generated data are deposited in figshare at https://doi.org/10.6084/m9.figshare.13350524.v1. The predicted protein fasta files used in BLAST databases in this study (see Additional file 1: Table  S1) are deposited in figshare at https://doi.org/10.6084/m9.figshare.13344884. v1. Single gene datasets for all initially considered sequences and final orthologs used in phylogenomic analyses and a list of putative cross contamination sequences for each gene are deposited in figshare at https://doi.org/ 10.6084/m9.figshare.13344227.v1. Log files for all phylogenomic analyses, single gene datasets for all initially considered sequences, and final orthologs used in phylogenomic analyses are deposited in figshare at https://doi.org/ 10.6084/m9.figshare.13927292.v2. Single gene datasets comprised of all initially considered sequences, final datasets, apicomplexan homolog, trees (raw and with apicomplexan homologs colored), and subcellular localizations based on TargetP and DeepLoc for all mitochondrial genes discussed here are deposited in figshare at https://doi.org/10.6084/m9.figshare.13928252.v1 [129].