Chromerid genomes reveal the evolutionary path from photosynthetic algae to obligate intracellular parasites

1The phylum Apicomplexa is comprised of eukaryotic, unicellular, obligate intracellular parasites, infecting a diverse range of hosts from marine invertebrates, amphibians, reptiles, birds to mammals including humans. More than 5000 species have been described to date, and over 1 million apicomplexan species are estimated to exist. Clinically and economically important apicomplexan pathogens, for example, Babesia, Cryptosporidium, Eimeria, Neospora, Theileria, Toxoplasma, and the malaria-causing parasite Plasmodium wreak profound negative impacts on animal and human welfare.

2Despite their diverse host tropism and life cycle strategies, apicomplexans possess several unifying molecular and cellular features, including the abundance of specific classes of nucleic acid-binding proteins with regulatory functions in parasitic processes, extracellular proteins for interactions with the host, an apical complex comprising a system of cytoskeletal elements and secretory organelles, an inner membrane complex (IMC) derived from the alveoli, and a non-photosynthetic secondary plastid, termed the apicoplast. How and when these features arose is unclear, owing to the lack of suitable outgroup species for comparative analyses.

3Chromerids comprise single-celled photosynthetic colpodellids closely associated (and likely symbiotic) with corals. Phylogenetic analysis demonstrates that these algae are closely related to Apicomplexa, confirming the long-standing hypothesis that apicomplexan parasites originated from a free-living, photosynthetic alga. Two known chromerid species, Chromera velia and Vitrella brassicaformis, can be cultivated in the laboratory, and their plastid and mitochondrial genomes have been described. We explored whole nuclear genomes of Chromera and Vitrella to understand how obligate intracellular parasitism has evolved in Apicomplexa.

4A shotgun approach was used to sequence and assemble the Chromera and Vitrella nuclear genome into 5953 and 1064 scaffolds totaling 193.6 and 72.7 million base-pairs (Mb). The disparity in genome size is attributable largely to the presence of transposable elements (TEs) totaling ∼30 Mb in Chromera vs only 1.5 Mb in Vitrella, as the predicted number of protein-coding genes is almost the same at 26,112 and 22,817, respectively. Detailed characterizations of the two genomes and their gene structures are described in Appendix 1 and Supplementary files 1, 2.

5We constructed a phylogenetic tree of 26 species, comprising Chromera, Vitrella, 15 apicomplexans, 2 dinoflagellates, 2 ciliates, 4 stramenopiles, and a green alga. On the phylogenetic tree (Figure 1A), Chromera and Vitrella formed a group closest to the apicomplexan clade, consistent with previous phylogenies. The long branches from their common node are consistent with drastic differences in morphology, life cycle, plastid and mitochondrial genomes between the two chromerids (Figure 1A). Likewise, despite common origins, apicomplexans show extensively diverse lifestyles, including host tropism and invasion phenotypes (Figure 1B).

6We reconstructed the parsimonious gene repertoires for the ancestors of the 26 species, at the nodes of the phylogenetic tree (Figure 2A; Figure 2—figure supplement 1). We note five key nodes on the evolutionary paths to present-day apicomplexans: the alveolate ancestor; the common ancestor of Apicomplexa and chromerids, termed the proto-apicomplexan ancestor; the apicomplexan ancestor; the ancestor of apicomplexan lineages, for example, coccidia and hematozoa; and extant apicomplexans (Figure 2A). Protein-coding genes from the 26 species were clustered by OrthoMCL into groups of homologous genes, hereafter defined as orthogroups. We note that an orthogroup could have homologous genes from different species (putative orthologs) or from the same species (putative paralogs arising from gene duplications). Gains or losses of orthogroups are displayed as green or red sections of a pie on the phylogenetic tree in Figure 2A. Divergence of the proto-apicomplexan ancestor from the alveolate ancestor (Stage I) was accompanied by losses of 1668 and gains of 2197 orthogroups (sum of the two ‘pies’ in Stage I). Transition of the free-living proto-apicomplexan ancestor to the apicomplexan ancestor (Stage II) is accompanied by many gene losses (3862 orthogroups) but few gains (81 orthogroups) (Figure 2A). Divergence of coccidians, for example, Toxoplasma gondii, from the apicomplexan ancestor (Stage III) is characterized by modest changes (537 losses; 414 gains), whereas divergence of hematozoans, for example, Plasmodium spp., is marked by drastic losses (1384 losses; 77 gains) (Figure 2A). Further divergence of apicomplexan taxa beyond Stage III is characterized by modest, lineage-specific gains (Figure 2A). Functional composition of gained genes at various stages will be discussed in later sections. Paucity of gained genes (81 orthogroups) during Stage II indicates that the genome of the free-living ancestor possessed most of the genes that were present in the common ancestor of apicomplexans and survived in their present-day descendants.

7Parasite evolution has been associated with genome reduction across several branches of the tree of life. Examples also exist, however, where parasite genomes are not reduced but expanded, underscoring the fact that the genome reduction process during parasite evolution is not completely understood. We sought to characterize in detail the dynamics of gene loss across apicomplexan evolution, particularly for components of molecular processes that are hallmarks of free-living lifestyle. We performed a systematic analysis of the cellular components involved in: (1) cellular metabolic pathways; (2) the endomembrane trafficking systems, regulating the movement of molecules across intracellular compartments in eukaryotes; and (3) the flagellum, a highly conserved apparatus for motility in aqueous environment.

8The inferred proto-apicomplexan ancestor, like present-day chromerids, possessed complete metabolic pathways for sugar metabolism, assimilation of nitrate and sulfite, and photosynthesis-related functions (Figure 2B, Figure 2—figure supplement 3, Appendix 2, and Supplementary file 3). Unlike in other photosynthetic algae, both Chromera and Vitrella initiate heme synthesis in the mitochondrion using aminolevulinate synthase (C4 pathway), which thus far has been found only in a few eukaryotic heterotrophs, such as Euglena gracilis, dinoflagellates, and apicomplexans (Appendix 2 and Supplementary file 4). Both chromerids and apicomplexans encode modular multi-domain fatty acid synthase I (FASI)/polyketide synthase enzymes and single-domain FASII components (Figure 2—figure supplement 2A,B). Treatment of Chromera with a FASII inhibitor triclosan showed decreased production of long chain fatty acids (Figure 2—figure supplement 2C and Appendix 2), suggesting that Chromera synthesizes short-chain saturated fatty acids using the FASI pathway, which are then elongated using the FASII pathway. This was previously demonstrated in Toxoplasma, an apicomplexan that possesses both FASI and FASII. Likely, the proto-apicomplexan ancestor was a phototrophic alga harboring characteristic metabolic features previously found only in apicomplexan parasites, especially with regard to plastid-associated metabolic functions (see above and other examples in Appendix 2).

9Transition to an apicomplexan ancestor (Stage II) was accompanied by the loss of metabolic processes including photosynthesis and sterol biosynthesis (Figure 2B and Figure 2—figure supplement 3). The apicomplexan ancestor appeared to possess a significant complement of enzymes in various pathways (Figure 2B). The differentiation of apicomplexan lineages (Stage III) was accompanied by further lineage-specific losses: for example, loss of FASI in Plasmodium spp, loss of FASII in Cryptosporidium spp., which has also lost the apicoplast, and loss of enzymes mediating polyamine biosynthesis in all lineages except Plasmodium (Figure 2B and Figure 2—figure supplement 3). These support the notion that enzymes involved in cellular metabolism critical for free-living organisms were not completely lost during the transition to the apicomplexan ancestor, but were further lost during subsequent differentiation and host-adaptation of apicomplexan lineages.

10The proto-apicomplexan had a nearly complete repertoire of the endomembrane trafficking complexes, and much of this repertoire persisted through to the apicomplexan ancestor (Stage II) (Figure 2C, Figure 2—figure supplement 4 and Appendix 3). Differentiation of apicomplexan lineages (Stage III) was accompanied by lineage-specific losses, for example, loss of the Endosomal Sorting Complex Required for Transport II (ESCRTII) in all lineages except in piroplasms, whereas some components were retained across all lineages, such as the retromer complex components and clathrin, both systems implicated in invasion processes (Figure 2C, Figure 2—figure supplement 4 and Appendix 3). These lineage-specific losses have led to diverse, reduced sets of endomembrane trafficking components in present-day apicomplexans. Some of these components that were present in chromerids were absent in specific apicomplexan lineages as well as in dinoflagellates and ciliates, further clarifying that these losses are independent, lineage-specific events rather than ancient, shared events.

11All known components of flagella were present in the proto-apicomplexan ancestor (Figure 2—figure supplement 5A,B). Most of the components were retained in the apicomplexan ancestor (Stage II), but losses occurred as apicomplexan lineages differentiated (Stage III). Components of intraflagellar transport, which are typically essential for assembling flagella, were lost in the other lineages except in coccidians (Figure 2—figure supplement 5A,B). The basal body proteins, which support an organizing center for microtubules, were lost from piroplasms. Some striated fiber assemblin (SFA) proteins, typically associated with basal body rootlets, were maintained in all apicomplexan lineages including piroplasms (Figure 2—figure supplement 5A,B,D); their presence has been hailed as evidence that some flagellar-proteins are repurposed for new functions in apicomplexans (see below).

12In summary, one of the major events during apicomplexan evolution is progressive, continued loss of components important for free-living organisms. While Stage II was accompanied by a massive loss of such components including those implicated in photosynthesis, the apicomplexan ancestor still possessed many proteins, which were lost later during differentiation of lineages with diverse life strategies.

13Evolution of present-day apicomplexan parasites was accompanied not only by gene losses as noted above (Figure 2) but also by gene gains. We sought to determine if genes gained at a particular stage of apicomplexan evolution, as depicted by the gray violin in Figure 3, would be over-represented with those involved in parasitic processes such as intracellular invasion into and egress from host cells. For Plasmodium falciparum and T. gondii, we compiled three classes of protein-coding genes directly or indirectly involved in parasitic processes of apicomplexans based on in silico prediction or information from previous functional studies (‘Materials and methods’). Extracellular proteins are secreted by the apicomplexans for various parasitic processes, for example, some of them are targeted to the host cytoplasm, nucleus, and plasma membrane to modulate parasite–host interactions. Cytoskeletal proteins provide structural support to the cell and also the molecular machinery for motility and intracellular invasion. Proteins with DNA-binding domains (DBDs) or RNA-binding domains (RBDs) can regulate various molecular processes of apicomplexan parasites. Indeed, proteins with AP2 (apiAP2) DBD have been shown to act as genetic control switches for diverse apicomplexan processes.

14Genes encoding extracellular proteins exported into the host environments were over-represented among those gained after Stage III (Figure 3), suggesting that adaptation to specific hosts was accompanied by expansion of extracellular proteins mediating host–parasite interactions. Stage III was accompanied by gains of those encoding DBD proteins, mostly apiAP2 proteins (Figure 3 and Figure 3—figure supplement 1A,B), suggesting extensive regulatory changes mediated by apiAP2 proteins during lineage differentiation. We note that losses of other canonical DBD proteins, for example, proteins with HSF_DNA-bind (Pfam: PF00447) domain during transition to apicomplexan ancestor (Stage II) and proteins with Tub (Pfam: PF01167) domain along the piroplasm lineage, contribute to further dominance of apiAP2 among the DBD proteins (Figure 3—figure supplement 1C). Stage II was accompanied by over-represented gains of various cytoskeletal components, including alveolins, those of the actomyosin complex (e.g., myosins) and glideosome-associated proteins with multiple membrane spans 1 and 3 (GAPM1 and GAPM3), suggesting that the molecular machinery powering gliding motility, which is essential for host cell invasion arose during evolution to apicomplexans (Figure 3, Figure 3—figure supplement 2, and Appendix 4). Gene gains during Stage I were over-represented by proteins with ‘RBD abundant in Apicomplexans’ (RAP, Pfam: PF08373), many of which were conserved as one-to-one orthologs across descending lineages, suggesting development of evolutionarily conserved functions before apicomplexans and chromerids diverged (Figure 3, and Figure 3—figure supplement 3). Chromerid genomes encode many orthologs of apicomplexan cytoskeletal proteins (Appendix 4), including GAPM2, a member of an important protein family for apicomplexan cytoskeletal structure and gliding motility, and the IMC sub-compartment protein family (ISP), implicated in establishing apical polarity and coordinating the unique cell cycle of apicomplexans (Figure 2—figure supplement 5E). These data suggest that some components existed in the free-living proto-apicomplexan ancestor and were subsequently repurposed for parasitic processes of apicomplexans.

15The Chromera and Vitrella genomes encode many proteins that are specific to chromerids yet contain functional domains implicated in molecular processes of apicomplexan parasites. For example, there are chromerid-specific proteins with domain architectures similar to those in apicomplexan extracellular proteins, including those previously implicated in host interactions and described in apicomplexans only (Figure 3—figure supplement 4 and Appendix 5, and Supplementary file 5). Presence of such chromerid proteins implies some commonality in extracellular recognition and cross-species interactions and this correlates well with the presumed associations with the coral holobiont. Importantly, chromerid genomes encode numerous apiAP2 proteins, more abundant than dinoflagellates, suggesting that they have expanded in the proto-apicomplexan ancestor after it split from dinoflagellates (Figure 3—figure supplement 1D). Many of the chromerid apiAP2 proteins belong to putative paralogous clusters, suggesting that their expansion was driven by gene duplication (Figure 3—figure supplement 1D; Appendix 6). Only a small subset of the apiAP2 proteins are shared across apicomplexans, suggesting that the large apiAP2 complement in the proto-apicomplexan ancestor has diversified independently in descending lineages (Figure 3—figure supplement 1A).

16In summary, genes encoding critical components of the parasitic lifestyle of apicomplexans were gained at different stages of apicomplexan evolution, some implying subsequent specialization to particular host niches, but others suggesting early adaptations before committing to parasitic lifestyle. This is evident by chromerid orthologs of many such proteins, for example, RAP proteins and specialized cytoskeletal components. Further, chromerid genomes encode chromerid-specific proteins that are not detected as orthologs of apicomplexan proteins but still have functional domains implicated in parasitic processes in apicomplexans. Together, these data imply that a molecular transition had occurred in free-living ancestors of apicomplexans, providing a foundation for host–parasite interactions and further adaptation.

17Chromera and Vitrella genomes allowed us to reconstruct the gene content of the free-living ancestor of apicomplexans. To infer their putative functions using genome-wide gene expression information, we cultured Chromera under 36 different combinations of temperatures, iron and salt concentrations, and generated their gene expression profiles by RNA-seq. In addition, we have obtained a publicly available growth perturbation data set for P. falciparum. There were 1918 orthogroups shared between the two species. We identified pairs of orthogroups that are co-expressed, that is, showing similar expression patterns across the various conditions, in both species (‘Materials and methods’) (Figure 4—figure supplement 1A). Such an orthogroup pair, that is, those with conserved co-expression between the two species, would include candidate genes that have been co-regulated together during apicomplexan evolution, from the free-living ancestor to present-day parasites due to conserved functions. This approach, successfully utilized by several studies in the past, led to the following two observations in this study.

18Many RAP genes appeared during Stage I and have been conserved across the descending phyla (Figure 3 and Figure 3—figure supplement 3), but their precise cellular roles are unknown. For 11 out of 12 orthogroups with RAP domains, co-expressed orthogroups overlapped significantly (Fisher's exact test, p < 0.05) between P. falciparum and Chromera, suggesting involvement of RAP proteins in cellular processes evolutionarily conserved across apicomplexans and chromerids (Figure 4A). RAP and their co-expressed orthogroups encode proteins with putative mitochondrial import signals more often than expected by chance in Chromera and P. falciparum (Fisher's exact test, p < 0.05) (Figure 4B), and also in other apicomplexans and chromerids (Figure 4—figure supplement 1B). We have randomly chosen three Toxoplasma RAP genes with predicted mitochondrial localization signals (Supplementary file 6) and confirmed experimentally by 3′ endogenous gene-tagging with reporter epitopes that all three are localized to the organelle (Figure 4C). Some of the orthogroups co-expressed with orthogroups containing RAP domains encode protein products predicted to be metabolic enzymes, implying possible involvement of RAPs in mitochondrial metabolism (Figure 4—figure supplement 1C). Consistent with this, the Cryptosporidium lineage that has a highly reduced mitochondrion lacking both the genome and most canonical metabolic pathways is the only apicomplexan group to have also lost its RAP repertoire (Figure 4—figure supplement 1D). Loss of RAPs along with a set of mitochondrial functions in this lineage is consistent with a mitochondrial role for RAPs. We speculate that the free-living proto-apicomplexan ancestor possessed within its mitochondrion a regulatory process mediated by RNA-binding activities of the RAP proteins, which has been retained by the extant apicomplexans and chromerids.

19As discussed earlier, the proto-apicomplexan ancestor appears to have possessed genes implicated in invasion processes of present-day apicomplexans (Figure 3). Among the 1918 orthogroups, we identified 80 orthogroups comprising genes functionally annotated as implicated in invasion processes. The frequency of co-expression amongst them in the free-living Chromera was significantly higher than expected by chance (p < 0.0005), suggesting pre-existing functional relationships before transitioning to parasites (Figure 4D). We identified several modules or groups of co-expressed orthogroups (Figure 4E). In one of the co-expression modules (numbered 1 in Figures 4E), 9 out of 10 orthogroups are co-expressed with a gene encoding SFA (Cvel_872), a key protein for organizing the basal bodies of the flagellar apparatus in algae and the apical complexes in apicomplexans (Figure 4F). We note that SFAs are the only flagellar components found in all apicomplexans tested (Figure 2—figure supplement 5A). Also in this module, for 9 out of 10 orthogroups, their co-expressed orthogroups in Chromera overlapped significantly with those in P. falciparum (Fisher's exact test, p < 0.05), indicating that their regulatory programs have been evolutionarily conserved (Figure 4G). This module include various types of genes implicated in host cell invasion processes of apicomplexans such as genes encoding rhoptry protein ROP9, apical sushi protein ASP, and gliding motility components GAP40 and GAPM2. The apical complex has been postulated to have emerged from the flagellar apparatus and associated cellular transport systems in free-living algae, based on ultrastructural evidence. These results suggest that, in the free-living ancestor, some of the genes implicated in the invasion process of present-day apicomplexans were functionally associated with those implicated in flagellar motility, providing the much-needed genetic evidence for the postulate. We speculate that a group of functionally related proteins associated with the flagellar apparatus was repurposed as a module of the apical complex and became a foundation for the invasion machinery.

20Analysis of Chromera and Vitrella genomes has enabled insights into how apicomplexan parasites have evolved from free-living ancestors. The transition to parasitism was accompanied by massive genomic loss that continued as its descendants became specialized intracellular parasites infecting diverse hosts. The genome of free-living photosynthetic ancestors encodes many component proteins previously assumed to be restricted to the parasitic apicomplexan lineages. Such pre-existing components, including those of what would later become part of the invasion machinery, were co-opted during evolution to facilitate a successful parasitic lifestyle in multiple hosts. The genome of the proto-apicomplexan ancestor served as a molecular blueprint for evolution of the most successful group of eukaryotic parasites known to date.

21Sequencing data have been deposited in the European Bioinformatics Institute under the European Nucleotide Archive (ENA) sample accession number ERP006228 for C. velia and ERP006229 for V. brassicaformis for all DNA- and RNA-seq experiments. The assembly and the annotations were submitted under accession numbers CDMZ01000001-CDMZ01005953 for C. velia and CDMY01000001-CDMY01001064 for V. brassicaformis. Some of the Vitrella DNA-seq experiments were done at Broad Institute and are deposited at Short Read Archive under accession numbers SRX152523 and SRX152525. The annotations and assemblies can be viewed and queried in EupathDB (http://cryptodb.org/cryptodb/).

22Genomic DNA of C. velia CCMP2878 (subsequently referred to as Chromera) and V. brassicaformis CCMP3155 (subsequently referred to as Vitrella) was extracted and then sheared into short fragment size libraries (300–500 base pair (bp)) and large fragment size libraries (3–8 kbp fragments) by focused-ultrasonication (Covaris Inc., Woburn, USA). The last 3–8 kb libraries were prepared following Nextera mate pair protocol, following manufacturer's instructions. We used three different methods to generate the library: the Illumina (Illumina, San Diego, CA) TruSeq DNA protocol LT Sample Prep Kit (catalog no. #FC-121-2001), an amplification-free method (TruSeq DNA PCR-Free LT Sample Preparation Kit catalog no. #FC-121-3001) and the Illumina Nextera Mate Pair Sample Preparation Kit (catalog no. #FC-132-1001). The libraries were sequenced on an Illumina HiSeq2000 platform following the manufacturers standard cluster generation and sequencing protocols. Image analysis, base-calling, and quality filtering were processed by Illumina software.

23For isolation of RNAs, Chromera and Vitrella were grown under standard culture conditions. Total RNA was extracted from the cells using TRIzol. The polyA+ RNA fraction was selected using oligo(dT) beads, and RNA-seq libraries were prepared using TruSeq RNA Sample Prep kit (catalog no. FC-122-1001). Strand-specific RNA-seq libraries were prepared using TruSeq Stranded mRNA LT Sample Prep Kit (catalog no. RS-122-2101) and sequenced as paired-end (2 x 100 bp) reads on a HiSeq2000 platform.

24We performed additional RNA sequencing of Chromera subject to various environmental perturbations, to construct a global gene expression network based on transcriptomes under various perturbation conditions during in vitro growth. Chromera cultures were exposed to a combination of stresses (Figure 4—figure supplement 1C). First, six different media were prepared from the combinations of salt concentration (16.7 g/l, 33.3 g/l, 66.6 g/l) and iron deficiency by chelation. After seeding, the cultures were maintained in the normal temperature and light condition for eleven days. After randomization, the cultures were incubated at 26°C, 37°C, or 14°C for 0 (control), 0.5, or 2 hr. There were two biological replicates of each, in total 66 flasks of the cultures. Then, the cultures were processed with centrifugation at 3500 RPM for 15 min at 4°C to precipitate the cells. Total RNA was extracted from the 66 cultures after the treatments using Norgen RNA Extraction kit based on manufacturer's protocol (Norgen Biotek Corporation, Canada). RNA quality was assessed using Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA). RNA concentration was determined with a Qubit (Invitrogen, Carlsbad, CA). Strand-specific RNA-seq libraries were prepared from extracted high-quality RNAs (RIN ≥8.0 as measured on an Agilent Bioanalyser 2100) using the Illumina TrueSeq LT stranded RNA sample kit according to manufacturer's instructions. Prior to cluster generation, concentration and size of libraries were assayed using the Agilent DNA1000 kit. Libraries from all samples were sequenced as single-end (1 x 50 bp) reads on the Illumina HiSeq 2000. The RNA-seq reads were aligned to the reference genome using tophat (version 2.0.8, default parameters) and cufflinks (version 2-1.0.2, default parameters). The FPKM values were log2 normalized with an offset of 1 and were further corrected for different distributions across the samples using the quantile normalization method.

25For Vitrella, the reads were corrected and assembled followed by several base correction, scaffolding and gap filling steps as briefly described below. As first step, the short insert libraries were corrected with SGA (version 0.9.19). The corrected reads were assembled with velvet (version 1.2.08). Iterating through different parameter settings, we choose a k-mer of 75 bp as the best parameter set. The resulting scaffolds (larger than 1 kb) were further scaffolded with SSPACE using first the Illumina library (insert = 550 bp) and larger insert (1 kb) Illumina library reads. Sequencing gaps were closed with Gapfiller (version 1.1.1) with two iterations, using the bowtie mapping option and PCR-Free libraries. Base pair call errors were corrected in three iterations of ICORN, using the amplification-free library. Furthermore, sequencing gaps were closed, using IMAGE with the amplification-free library. The assembly was quality-controlled using REAPR, breaking the contigs at possible miss-assemblies, using the mate pair libraries. This was followed by another scaffolding step. We systematically removed 620 scaffolds containing 25.65 Mb representing the bacterial contamination. The Vitrella CCMP3155 assembly contains 72.7 Mb (including 931,689 N's) in 1064 scaffolds (ENA accession numbers CDMY01000001-CDMY01001064). The scaffolds were constructed from 4177 contigs.

26For Chromera, the assembly pipeline and the algorithms used were the same as Vitrella, but due to the larger size, higher amount of low-complexity regions, and difficulties in generating high-quality large insert size libraries, additional steps were included to the assembly process. First, the reads of the PCR-Free library were corrected with SGA and then assembled with velvet and using a k-mer of 71 (version August 2011). Next, the contigs were scaffolded, gapfilled, and corrected with ICORN, as described earlier. We mapped the reads of all large insert size libraries using SMALT (ftp://ftp.sanger.ac.uk/pub/resources/software/smalt/). We excluded scaffolds smaller than 1 kb. Different iterations with SSPACE were undertaken and the assembly was quality-checked with REAPR. After scaffolding, gapfiller and IMAGE were run as above, followed by ICORN. The 1725 scaffolds (spanning 16.02 Mb) representing bacterial contamination were removed. The final assembly of Chromera CCMP2878 contains 193.66 Mb (including 582,995 N's) in 5953 scaffolds (ENA accession numbers CDMZ01000001-CDMZ01005953). The scaffolds are constructed from 13,987 contigs.

27We used Augustus (version 2.5.5) for gene prediction. We manually curated 716 and 245 gene models for Chromera and Vitrella, respectively, using BLAST similarity-based approaches, and we also generated automated gene models using Cufflinks from RNA-seq data sets, in order to use them as a ‘training gene model set’ for Augustus prediction. The strand-specific RNA-Seq, mapped with TopHat2, was used as evidence in Augustus for intron evidence.

28In summary, from the Chromera and Vitrella genome, we ab initio predicted 30,478 and 23,503 protein-coding genes, respectively, of which 18,829 and 18,240 were detected as being expressed from RNA-seq evidence as poly A+ transcripts (Supplementary file 1). Excluding putative TEs, 26,112 and 22,817 genes were predicted as protein-coding genes in Chromera and Vitrella. We annotated partial genes, when a gene probably spans more than one scaffold, located at the borders of a scaffold. We demarcated and annotated as pseudo genes if they contain in frame stop codons. We flagged gene models as transposon elements, if they overlap with the predicted TE regions and had no more than three and two intron for Chromera and Vitrella, respectively. To annotate untranslated regions (UTRs) of the predicted protein-coding genes, we used CRAIG with default parameters with mapping of the RNA-Seq data as computed by GSNAP (version 2013-08-19, default parameters). The annotation of both genomes has the ENA accession numbers CDMZ01000001-CDMZ01005953 and CDMY01000001-CDMY01001064 and is also available in EuPathDB.

29The predicted genes were assigned putative functions based on BLASTP (E value <10−6) matches against UNIPROT (version March 2012). The predicted protein products were assigned protein domains using hmmsearch (HMMER 3.1b1, May 2013) for Pfam A v26.0. Statistical threshold defined by the Pfam database was used. We aligned AP2 sequences in apicomplexan species based on PfamA AP2 (PF00847), and built apicomplexan-specific AP2 (apiAP2) hidden Markov model (HMM), and scanned the predicted protein-coding genes for apiAP2 domains; we annotated api-AP2 DNA-binding transcription factor genes with both domain and sequence E values to be less than 10−3. The following Pfam RBDs were used to define RNA-binding proteins: ‘CAT_RBD’, ‘dsRNA_bind’, ‘S1’, ‘DEAD’, ‘KH_1’, ‘KH_2’, ‘KH_3’, ‘KH_4’, ‘KH_5’, ‘RRM_1’, ‘RRM_2’, ‘RRM_3’, ‘RRM_4’, ‘RRM_5’, ‘RRM_6’, ‘SET’, ‘PUF’, and ‘RAP’. The list of DBDs was downloaded from a database of DBDs. Transmembrane domains and signal peptides were assigned with the tools TMHMM 2.0 and signalP 4.0, respectively, with default parameters.

30We collected several categories of genes implicated in parasitic processes in apicomplexans for two archetypal apicomplexan parasites, Toxoplasma and Plasmodium. We primarily obtained annotations from PlasmoDB and ToxoDB. Information for sub-cellular localization of genes is obtained from GeneDB and ApiLoc, a database of published protein sub-cellular localization for apicomplexan species (http://apiloc.biochem.unimelb.edu.au/apiloc/apiloc). Some putative parasite genes were inferred based on orthology by OrthoMCL clustering with closely related species with results from functional studies. We performed exhaustive literature searches to manually curate individual genes, to define rules for in silico searches across the proteomes of this study, and to categorize the identified genes based on their localization and function. The categories of parasite genes are defined as follows.

31The cytoskeleton of an organism provides the necessary structural framework for the maintenance of cell shape and integrity. We compiled two groups of cytoskeletal proteins, IMC associated proteins and actomyosin complex. First, IMC associated proteins, comprises alveolin proteins, a membrane occupation and recognition nexus protein (MORN), which associate with IMC and spindle poles and are indispensible for asexual and sexual development. IMC sub-compartment proteins (ISPs) are critical for establishing apical polarity in the parasite. Second, components of actomyosin motor complex, which powers the characteristic gliding motility, comprises actin, myosin, tubulin, gliding associated proteins (GAPs), aldolase, and various actin-regulatory proteins, which will assist actin in the process of quick polymerization–depolymerization cycles between F-actin and G-actin during this process. Examples of actin-regulatory proteins are Arp2/3 complex and formins (FH2) for nucleation; F-actin capping for filament regulation; coronin for cross-linking/bundling and profilin, CAP, cofilin/ADF and gelsolin for monomer treadmilling.

32Extracellular proteins are defined as parasite proteins, which are localized either on the surface or secreted off the parasite. They are released in a concerted manner to ensure successful adhesion to the surface, entry into the host cell, multiplication, and escape. Extracellular proteins can be categorized as (1) ‘exportome’ are proteins translocated to the host cytoplasm, membranes, and nucleus crossing the boundary membrane parasitophorous vacuole (PV); and (2) ‘others’, which stay on the parasite surface or released from the parasite, but not into the host intracellular space. The exportome genes are released mostly from the parasite's secretary organelles such as rhoptries and dense granules. Some of these genes possess host targeting or also known as the Plasmodium export element (PEXEL). Many PEXEL-negative proteins have been identified too. These genes are sorted and targeted through a specialized structure known as Maurer's cleft formed in the host cytoplasm. These genes are mostly kinases, proteases, and surface molecules, which modulate the host and hijack the host machinery in favor of parasitic growth and host immune evasion. The ‘other’ extracellular proteins consist of surface antigens (e.g., MSPs), SERAs, TRAPs, AMA-1, microneme proteins, ROPs and RONs etc.

33Repeat annotation was done by using the REPET pipeline and LTR finder. The overall pipeline comprises of two steps: de novo detection and classification. In the first step, the scaffolds are split into smaller batches (∼1000 batches of 200 kb each). These genomic fragments were aligned against each other to detect the HSPs (High-scoring pairs) using BLASTER. HSPs are then clustered using a combination of three methods such as GROUPER, RECON, and PILER. Structure-based LTR retrotransposons (RTs) detection tools such as LTRharvest and LTR finder, which are based on 100–1000 bp long terminal repeats with a 1 kb–15 kb separation and target site duplication site at vicinity of 60 bp to the two terminal repeats. These LTRs detected are clustered using BlastClust. Multiple sequence alignment of each cluster was performed using MUSCLE (Edgar, 2004). Each cluster aligned was searched against Repbase using BLASTER and HMMER. A consensus feature was detected for each aligned cluster. Further PASTEC, which is based on the Wicker classification, was used for consensus classification.

34The repeats were annotated as follows. The genomic chunks were randomized and HSPs were detected using BLASTER, CENSOR, and RepeatMasker. These HSPs were filtered and combined. Again, full-length genomic scaffolds were compared to Repbase using MATCHER. Satellite and simple repeats were detected using the mreps, TRF, RMSSR (RepeatMasker). Finally, a long-join procedure was followed to combine the nested repeats. The whole annotation was exported to a genome-browser readable GFF3 file.

35OrthoMCL 2.0 was used with a default inflation parameter (I = 1.5) to generate groups of homologous genes (defined as orthogroups), which could have homologs from different species (putative orthologs) or from the same species (putative paralogs from gene duplications). For some genes of high interest, we manually inspected the alignments of the protein sequences within the orthogroup, which were done with MAFFT. We assigned Pfam domains to an orthogroup if more than half of the genes in an orthogroup were assigned the Pfam domains.

36There are several tools available for a general eukaryotic sub-cellular localization prediction, but they are not applicable to alveolates due to its unique chloroplast membrane arising from secondary endosymbiosis. Therefore, HECTAR, which was developed for the bipartite sub-cellular prediction, was used. There is no stand-alone version of HECTAR, and the online version allows only one sequence at a time. We implemented a modified HECTAR algorithm as a PERL script for batch prediction of the whole proteomes. Each protein sequence was predicted for signal sequence using SignalP 3.0, the signal sequence is cleaved, and the remaining amino acid sequence was used as input for the transit peptide prediction by TargetP. Sequences with both signal peptide and the transit peptide (either chloroplast or mitochondria) are predicted to be in the chloroplast. Sequences without the signal peptide but with the transit peptide (either chloroplast or mitochondria) are predicted to be in mitochondria. Sequences with signal peptide, without transit peptide, and predicted by TargetP to be secretory are classified as secretory proteins.

37For the RAP proteins, we tested the validity of our sub-cellular localization prediction in two ways. First, we compared our in-house algorithm with other published tools: TargetP, MitoProt2, iPSORT, and PredSL (Supplementary file 6, only mitochondrial prediction is shown). We found that our mitochondrial prediction for RAP genes is in concordance with other methods. Second, we experimentally verified mitochondrial localization in T. gondii by 3′ tagging of the endogenous genes with the coding sequence for the hemagglutinin epitope for three RAP proteins that were predicted to target to mitochondria with high probability.

38A statistical environment software R was used for most of the analyses and generating parts of figures. An R package vioplot was used to generate the violin plot. A ward algorithm on the distance matrix based on (1- correlation coefficients) in an R function hclust was used for all hierarchical clustering of gene expression patterns unless noted otherwise.

39We compiled the reference proteomes of 26 alveolate and stramenopile species (Figure 1—source data 2) from public databases such as EupathDB and NCBI Genome database (http://www.ncbi.nlm.nih.gov/genome/).

40We generated a phylogenetic species tree using a data set composed of 101 one-to-one orthologs across the 26 species (see Figure 1—source data 1 for gene IDs). Amino acid sequences were aligned using MAFFT, highly variable sites were edited by trimAL and after manual inspection. The resulting alignment of 33,997 amino acid positions was used to construct trees by a maximum likelihood (ML) method and Bayesian inference. The ML tree was computed using RAxML 8.1.16 by gamma corrected LG4X model. Robustness of the tree was estimated by bootstrap analysis in 1000 replicates. Bayesian tree was constructed by PhyloBayes using two-infinite mixture model CAT-GTR as implemented in PhyloBayes 3.3f. Two independent chains were run until they converged (i.e., maximum observed discrepancy was lower than 0.2), and the effective number of model parameters was at least 100 after the first 1/5 generation was omitted from topology and posterior probability inference. All clades in the tree were supported with posterior probability 1.00 and 100% bootstraps, except for one node, which representing the common ancestor of human Plasmodium spp. was supported by 99% bootstrap.

41We performed the gene gain and loss analysis based on Dollo parsimony using Count software. This approach allows reconstructing gene contents at observed species and at hypothetical ancestors, and gene gains and losses at branching points. The Dollo parsimony strictly prohibits multiple gains of genes. To test for validity of this assumption, we repeated analyses based on parsimony settings allowing multiple gene gains or on a phylogenetic birth-and-death model and reached the same conclusion (Figure 2—figure supplement 1). We have also repeated the analysis using Wagner's parsimony, allowing multiple gains per tree with gain penalty of 2 or greater, and obtained similar results (data not shown). For the analysis of metabolic enzymes, endomembrane trafficking system components, and flagellar apparatus components, the ancestral presence was inferred based on Dollo parsimony from the presence of components in the observed species. For the endomembrane trafficking component analysis, we assumed that the last common ancestor had a complete repertoire of the components.

42We have inferred the evolutionary age of P. falciparum and T. gondii genes as the early node on the phylogenetic tree where the most distant species have genes with significant sequence homology (reciprocal BLASTP E value <10−10 and clustering with OrthoMCL).

43We studied if orthologs of Chromera and P. falciparum show similar gene expression changes to physiologically equivalent growth conditions. Identifying equivalent conditions is difficult as the two species have completely different lifestyles and live in different environments. Instead, we tested if a given gene and its ortholog would show correlated expression patterns with the same set of genes (and orthologs), allowing a way to compare gene expression behavior measured under different conditions. To uncover gene-to-gene co-expression relationships, the organisms from whom transcriptomes are sampled must be exposed to various growth conditions. This approach has been successfully used in other eukaryotes. For Chromera, we generated RNA-seq-based transcriptome under combinations of varying salt concentrations, iron concentrations, and temperature changes, resulting in 36 unique combinations (see ‘Materials and methods’ and Figure 4—figure supplement 1C). For P. falciparum, we obtained previously published microarray-based gene expression data sets of 144 unique conditions from 23 time series, representing stresses from various growth-inhibiting compounds. It has been shown that gene expression data generated using different molecular platforms are reproducible and accurate enough for cross-platform comparisons. Based on each data set, we calculated Spearman correlation coefficients rho between all possible pairs from the 1918 orthogroups shared between Chromera and P. falciparum (1918 × 1918 matrix). We also calculated a 1918 × 1918 weighted adjacency matrix using CLR algorithm as implemented in an R package minet (with parameters of method = ‘clr’, estimator = ‘mi.shrink’, and disc = ‘equalfreq’). Expression level of multiple genes in a given orthogroup was averaged. To rule out any potential systematic biases associated with averaging expression levels of homologous, yet distinct genes, we repeated some of the analyses with 1560 orthogroups that have one-to-one orthologs between the two species and reached the same conclusions (data not shown). A pair of genes (or orthogroup) were determined as co-expressed if the Spearman's correlation coefficient rho is greater than 0.3 and if the value from the weighted adjacency matrix of the network is greater than 0.01. We calculated an odds-ratio to measure the extent of conservation of co-expressed genes: (# of genes co-expressed in both species) × (# of genes co-expressed in none of the species)/([# of genes co-expressed in P. falciparum only] × [# of genes co-expressed in C. velia only]), and Fisher's exact test was used to assess the statistical significance. For calculation of the odds-ratios, co-expression was determined based on correlation coefficient to minimize count granularity in the two-by-two table.