In silico ionomics segregates parasitic from free-living eukaryotes.

Ion transporters are fundamental to life. Due to their ancient origin and conservation in sequence, ion transporters are also particularly well suited for comparative genomics of distantly related species. Here, we perform genome-wide ion transporter profiling as a basis for comparative genomics of eukaryotes. From a given predicted proteome, we identify all bona fide ion channels, ion porters, and ion pumps. Concentrating on unicellular eukaryotes (n = 37), we demonstrate that clustering of species according to their repertoire of ion transporters segregates obligate endoparasites (n = 23) on the one hand, from free-living species and facultative parasites (n = 14) on the other hand. This surprising finding indicates strong convergent evolution of the parasites regarding the acquisition and homeostasis of inorganic ions. Random forest classification identifies transporters of ammonia, plus transporters of iron and other transition metals, as the most informative for distinguishing the obligate parasites. Thus, in silico ionomics further underscores the importance of iron in infection biology and suggests access to host sources of nitrogen and transition metals to be selective forces in the evolution of parasitism. This finding is in agreement with the phenomenon of iron withholding as a primordial antimicrobial strategy of infected mammals.

Introduction

Inorganic ions are essential to life. All cells maintain transmembrane gradients of potassium (K+), sodium (Na+), calcium (Ca2+), and chloride ions (Cl−), and the resulting membrane potential allows electrical signal transduction and drives nutrient uptake. Other ions such as iron (Fe2+), magnesium (Mg2+), copper (Cu2+), and zinc (Zn2+) are important nutrients themselves, functioning as cofactors for metalloproteins and stabilizers of large organic molecules. Polyatomic ions such as sulfate (), nitrate (), phosphate (), or ammonium () may serve as inorganic sources of macronutrients. The majority of essential micronutrients are ions too, for example, cobalt (Co2+), manganese (Mn2+), iodine (I−), and molybdenum (). As many of these ions are harmful at higher concentrations, ion homeostasis is fundamental for cell function. The importance of ion homeostasis is also illustrated by the large number of natural toxins that perturb it, by forming ion-conducting pores into phospholipid bilayers or by interfering with the function of ion transporters.

The term “ionomics” was coined for high-throughput measurements of inorganic nutrients (the ionome) in cells and tissues, usually by atomic emission spectroscopy (Salt et al. 2008). Ionomics quantifies the elemental composition and applied to the screening of reverse genetic mutants, it provided insights into the molecular mechanisms of ion homeostasis in Saccharomyces cerevisiae (Eide et al. 2005) and Arabidopsis thaliana (Baxter et al. 2007). Ionomics combined with forward genetics was used to study natural variation of ion concentrations in plants (Buescher et al. 2010) and mice (Fleet et al. 2011) and their potential association with metabolic disorders in humans (Sun et al. 2012). In particular, ionomics has further highlighted the role of ion transporters as the key players in ion homeostasis. Ionomics enabling the identification and characterization of transporters (Rus et al. 2006; Ciavardelli et al. 2010; Lowry et al. 2012), we reasoned that the genome-wide profiling of ion transporters would in turn permit to draw conclusions about the physiology of the ionomes of different organisms.

The TC system for Transporter Classification (Saier et al. 2006, 2009) recognizes three different classes of ion transporters: ion channels, ion porters (uniporters, symporters, and antiporters), and ion pumps (ATPases). Here, we construct profiles for all known families of ion transporters and use these profiles for “in silico ionomics,” aiming to elucidate convergence as well as divergence in the evolution of the molecular mechanisms of ion homeostasis in eukaryotes.

Materials and Methods

Proteome Files and CEGMA Completeness

Predicted proteomes were obtained from UniProt (www.uniprot.org, last accessed October 1, 2013) and Integr8 (ftp://ftp.ebi.ac.uk/pub/databases/integr8/, last accessed October 1, 2013) and tested for completeness as follows. Profiles were downloaded from the CEGMA database (http://korflab.ucdavis.edu/Datasets/cegma/, last accessed October 1, 2013) and the full set of 458 core eukaryotic proteins (Parra et al. 2007) was run with hmmscan of the HMMer 3.0 package (http://hmmer.janelia.org, last accessed October 1, 2013) against a diverse set of eukaryote reference proteomes: Caenorhabditis elegans, Chlamydomonas reinhardtii, Dictyostelium discoideum, Drosophila melanogaster, Danio rerio, Encephalitozoon cuniculi, Entamoeba histolytica, Giardia lamblia, Homo sapiens, Kluyveromyces lactis, Leishmania major, Mus musculus, Plasmodium falciparum, S. cerevisiae, Schizosaccharomyces pombe, Trypanosoma brucei, T. cruzi, Theileria parva, and Trichomonas vaginalis. The 100 best-scoring profiles returned hits of expectancy (E) values ≤10−50 against all the reference proteomes. These 100 profiles were then used to assess the completeness of additional proteomes (Ascaris suum, A. thaliana, Brugia malayi, Candida albicans, Cryptosporidium hominis, C. muris, C. parvum, Cryptococcus neoformans, Coccmyxa subellipsoidea, Chlorella variabilis, Leishmania braziliensis, L. infantum, L. mexicana, Magnaporthe grisea, Meloidogyne hapla, Micromonas pusilla, Ostreococcus lucimarinus, Oryza sativa, Plasmodium berghei, P. chabaudi, P. knowlesi, P. vivax, P. yoelii, Pediculus humanus, Polysphondylium pallidum, Paramecium tetraurelia, Schistosoma japonicum, Theileria annulata, Trypanosoma congolense, T. vivax, Toxoplasma gondii, Trichinella spiralis, Tetrahymena thermophila), which were only included if they contained a hit for at least 99 of the 100 profiles at a cutoff of E ≤ 10−30.

Ion Transporter Reference Sets and Redundancy Reduction

Lists of known families of channels (TC 1.A, n = 37), porters (TC 2.A, n = 96), and pumps (TC 3.A, n = 51) were obtained from TCDB (http://www.tcdb.org, last accessed October 1, 2013), and all ion transporter families (n = 78) were sorted out by hand. In case of overlap with parental terms (superfamilies), only the nonredundant children (families) were kept (n = 75; supplementary table S1, Supplementary Material online). For each ion transporter family, all the amino acid sequences that had been annotated with the corresponding TCDB accession in the manually curated section of UniProt were retrieved. Each of these sequence sets was redundancy-reduced as follows. A Smith–Waterman local alignment (Smith and Waterman 1981) was performed for all pairs of sequences, and if the resulting score reached 75% or more of the score of the self-alignment of the shorter sequence (i.e., the maximally attainable alignment score), the shorter sequence was removed from the set.

Ion Transportome HMM Library

Before converting the redundancy-reduced sets of ion transporter reference sequences into profiles, predicted ankyrin repeats and cyclic nucleotide binding domains were removed from the sequences. These domains were identified with hmmsearch of the HMMer 3.0 package using the profiles Ank (PF00023) and cNMP_binding (PF00027) from Pfam (http://pfam.sanger.ac.uk/, last accessed October 1, 2013). The parts matching these profiles with E < 10−8 were replaced with letters X in each sequence. Then, a ClustalW multiple alignment (Thompson et al. 1994) was performed for each sequence set and converted into a position-dependent scoring matrix with hmmbuild. The resulting profiles were concatenated to a HMM library for ion transporters. Negative control libraries were constructed by randomly selecting sets of 61 entries from the Pfam-A database (version 26.0: 13,672 entries). All the steps as outlined in figure 1 were carried out with self-made Perl scripts.

F

Overview on the in silico approach for ionomics.

Screening and Clustering of Proteomes

The above profile libraries were used to screen predicted proteomes with hmmscan. When counting the number of hits per profile (fig. 2), every protein in a given proteome was allowed to score only once, that is, with the profile against which it had the highest score. A cutoff E-value of <10−10 was used to call a hit. For clustering (fig. 4), a 65-tuple vector was constructed for each proteome which consisted of the respective best scores to each profile. Hierarchical clustering of these vectors was performed with the R library (R Core Team 2013) Pvclust which implements multiscale bootstrap resampling (n = 10,000) to estimate “approximately unbiased” (au) errors, where P = (100 − au)/100 (Suzuki and Shimodaira 2006). Distance metric (Canberra) and clustering algorithm (McQuitty) were chosen as to maximize the number of species in significant clusters (au ≥ 95).

F

Predicted ion transporters in eukaryotes as percentage of the total proteome (A) or in absolute numbers (B). The data are in supplementary table S2, Supplementary Material online.

F

Hierarchical clustering of ionomic landscapes segregates obligate parasites from eukaryotes with free-living life stages. The tree was produced with pvclust using Canberra distance and McQuitty’s similarity analysis. au are shown in gray, where P = (100 − au)/100.

Random Forests

Decision tree classification was performed using the randomForest (Liaw and Wiener 2002) package for the R statistical language (R Core Team 2013), which implements an ensemble learning method developed by Breiman (2003). A random forest consisting of 5,000 trees was used. At each node of the decision trees, the classification quality before and after splitting the set was quantified by using the Gini coefficient as a measure increasing proportionally to higher inequality of the predictions (i.e., parasite) in a set.

Results

A Profile Library for All Known Ion Transporters

The TCDB transporter database contains more than 600 different families of transmembrane channels, pores, and porters (Saier et al. 2006, 2009). We concentrated on transporters which solely have inorganic ions for substrates; ion-dependent porters of organic nutrients (e.g., the Na+-coupled glucose transporter) were excluded from the present analysis. A total of 75 nonredundant ion transporter families were identified, which subdivided as follows: 35 different kinds of ion channels; 30 ion anti-, sym-, or uniporters; and 10 ATP-dependent ion pumps (supplementary table S1, Supplementary Material online). We then constructed for each ion transporter family a position-dependent scoring profile as outlined in figure 1. Reference protein sequences had been obtained from the manually curated section of the UniProt database (Magrane 2011) and had been redundancy-reduced based on all pairwise alignments (Smith and Waterman 1981). Redundancy reduction minimized bias of the set of reference sequences while preserving diversity. It turned out to be necessary to purge the reference sequences from cyclic nucleotide-binding sites (Shabb and Corbin 1992) and ankyrin repeats (Li et al. 2006) before making the profiles as otherwise, these promiscuous domains returned hundreds of false-positive hits that bore no resemblance to ion channels. The resultant 75 profiles were concatenated to a HMM library (Eddy 2009) for ion transporters which is available from the authors on request.

Genome-Wide Prediction of Ion Transporters

The ion transportome profile library was used to scan predicted proteomes from fully sequenced genomes of the different eukaryote kingdoms, that is, the opisthokonts (fungi, animals), archaeplastida (plants, algae), excavates (kinetoplastids, trichomonas, giardia), chromalveolates (apicomplexa, one ciliate), and amoebozoa (entamoeba, slime molds); rhizaria are still missing from the list of sequenced genomes. Only proteomes of a CEGMA completeness ≥99% were included (fig. 1). The aim was to identify all the different ion transporters from a given species. The numbers of predicted ion transporters per proteome varied greatly between the different species, from more than 500 in C. elegans or A. thaliana to less than 30 in The. parva and E. cuniculi (supplementary table S2, Supplementary Material online). Plants and metazoa generally possessed more ion transporters than unicellular eukaryotes (P < 0.0001, two-tailed Mann–Whitney U test). This held true also when the numbers of ion transporters were normalized by proteome size: Multicellular species devoted a 2-fold greater portion of their proteomes to ion transporters than unicellular organisms (2.14% vs. 1.06%; P < 0.0001, two-tailed Mann–Whitney U test). The fraction of multispanning transmembrane proteins, that is, proteins predicted by Phobius (Käll et al. 2004) to contain two or more transmembrane domains, was the same in both groups (13.1% vs. 12.8%; fig. 2A). Overall, the difference between multicellular and unicellular eukaryotes was most pronounced regarding the predicted numbers of ion channels (fig. 2B). A striking exception was Par. tetraurelia which, with 443 predicted ion channels, possessed the largest number of different ion channel subunits of all the analyzed eukaryotes. This is in agreement with previous reports (Haynes et al. 2003). Paramecium tetraurelia was followed by C. elegans (372 predicted ion channels) and Dan. rerio (337 predicted ion channels; supplementary table S2, Supplementary Material online). Comparing the predicted numbers of hits per proteome also indicated that free-living or facultative parasitic eukaryotes possess more individual ion transporters than obligate endoparasites (fig. 2B and supplementary table S2, Supplementary Material online). To further investigate this phenomenon, we concentrated on unicellular eukaryotes only (to exclude the strong effect arising from the differences between multi- and unicellularity; fig. 2).

Clustering Proteomes According to Ionomic Landscape

The numbers of predicted transporters per proteome being a somewhat crude and arbitrary measure, we used the achieved scores against the ion transporter profiles as a refined and unbiased parameter of a given proteome. Thus, an “ionomic landscape” vector was built for every proteome, consisting of the top scores against the ion transporter profiles of supplementary table S1, Supplementary Material online. Figure 3 depicts these vectors as a heat map where darker shades represent higher scores. The data are shown in supplementary table S3, Supplementary Material online (only 65 different ion transporter profiles were used as 10 appeared to be prokaryote-specific and did not return a hit in any of the analyzed eukaryotes). Although the different unicellular eukaryotes analyzed achieved similar top scores toward the known families of ion pumps (fig. 3, right), the situation was different regarding ion channels and ion porters, which appeared to be generally underrepresented in obligate endoparasites as compared with free-living species or facultative parasites (fig. 3, left and middle). Parasites such as Cryptosporidia, Microsporidia, or Theileria appeared to be devoid of bona fide cation channels. The ionomic landscape vectors were hierarchically clustered in an unbiased way: A selection of distance metrics and clustering algorithms were combined with the program pvclust, and the resulting trees were ranked based on the number of leaves in statistically significant clusters. The best scoring tree is shown in figure 4. Its topology deviates from a phylogenetic tree in several aspects. The microsporidian E. cuniculi does not cluster with the fungi. The free-living amoebozoa D. discoideum and P. pallidum do, whereas Ent. histolytica groups with Tri. vaginalis. The ciliate Par. tetraurelia clusters with the free-living green algae rather than with alveolates (which are all parasitic), and the trypanosomatids are sister to Toxoplasma and the malaria parasites. Strikingly, the first and main division of the ionomic tree of unicellular eukaryotes is into eukaryotes with free-living life stages on one side and obligate endoparasites on the other. This separation was statistically significant as the probability of splitting the 37 analyzed species by chance into the 23 obligate parasites and 14 facultative parasites or free-living species equals (23! × 14! × 2)/37! = 3.3 × 10−10. In addition, we carried out the same procedure of representing and clustering HMMer hits based on sets of 65 randomly chosen profiles from the Pfam database of protein families (Punta et al. 2011); a separation of parasites from free-living species never occurred (data not shown).

F

Ion transporter repertoires of unicellular eukaryotes. The heatmap represents the best HMMer scores achieved by the different proteomes (rows) against the profiles for the different families of ion transporters (columns). Profiles that did not return a hit of score >20 in any of the proteomes are not shown. The data are in supplementary table S3, Supplementary Material online.

Convergent Evolution of Parasites

We concluded that the topology of the tree in figure 4 reflects convergent evolution between obligate endoparasitic eukaryotes, in particular loss of ion channel and ion porter genes. To elucidate which of the ion transporter families contribute the strongest signal for the observed distinction of the parasites, we performed a random forest classification after having assigned to each species an attribute parasite or nonparasite, the latter also comprising facultative parasitic species. The same input vectors were used as for hierarchical clustering (fig. 4). The random forest method generated training and validation sets by random resampling of the input vectors. A total of 5,000 trees were used to determine the impact of each ion transporter family on prediction accuracy regarding parasite status. Gini coefficients served as a measure for inequality (node impurity) of the predictions. Figure 5 depicts the impact of individual transporter families on the sum of all Gini coefficients in the forest. Five families stood out with a mean Gini decrease >1, namely the high-affinity ammonia transporters (AMT; 1.A.11.2), the natural resistance-associated macrophage proteins (NRAMP; 2.A.55), the BOR1-type boron transporters (2.A.31.3), the ZIP family of zinc–iron permeases (2.A.5), and the heavy metal transporter of the ABC-B superfamily (HMT; 3.A.1.210). BOR1 transporters were missing in all obligate endoparasites. ZIP and HMT transporters were present but different, consistently returning lower scores against the respective profiles than the hits from nonparasitic eukaryotes. AMT transporters were absent except in T. cruzi, and NRAMP transporters only occurred in Plasmodium spp. and Tox. gondii (supplementary table S3, Supplementary Material online). Note that only the combined information from the various ion transporter families distinguished the parasites.

F

Random forest analysis measuring the effect of each ion transporter family on the ability to distinguish obligate endoparasites. The typical substrates of the transporters are indicated on the right.

Discussion

Ion transporters are well suited for comparative genomics of evolutionary distant species as 1) most of the known ion transporter families are ancient as reflected by their ubiquitous occurrence in bacteria, archaea, and eukaryotes (Ward et al. 2009); and 2) many ion transporters possess conserved pore loops (MacKinnon 1995) that function as substrate selectivity filters and are readily detectable in silico. Furthermore, ion homeostasis is vital for all cells. Here, we describe a novel approach for comparative genomics and apply it to ion transporters as summarized in figure 1. We have identified 75 nonredundant families of ion transporters from the Transporter Classification Database (supplementary table S1, Supplementary Material online), 65 of which occur in eukaryotes (supplementary table S2, Supplementary Material online). Having constructed hidden Markov model-based profiles for each family, we scanned predicted high-quality proteomes of various eukaryotes and identified all the bona fide ion channels for each species. Although all types of eukaryotes devote around 12% of their proteome to multispanning transmembrane proteins (fig. 2), there are marked differences in the fractions of ion transporters. One major division appeared to be between multicellular (i.e., animals and plants) and unicellular eukaryotes, the former devoting larger fractions of their proteomes to ion channels and ion porters. This can be explained by the expansion of gene families in multicellular but not in unicellular eukaryotes (Wolf and Koonin 2013), with the notable exception of Paramecium. Another, more interesting division across the eukaryotes appeared to be between obligate parasites and nonparasitic or facultative parasitic species, the obligate parasites possessing fewer individual ion transporters (fig. 2). This may be explained by selective gene loss in the obligate parasites (Wolf and Koonin 2013). To further investigate this phenomenon, we concentrated on unicellular eukaryotes, eliminating the dominant effect of multicellularity versus unicellularity (fig. 2).

The analyzed unicellular endoparasites were highly heterogeneous regarding their phylogeny (fungi, trypanosomatids, apicomplexa, amoebozoa, and excavates) as well as habitat of the life stages (intracellular in compartment, intracellular cytosolic, extracellular). However, all the analyzed proteomes were similar in that they exhibited a markedly reduced diversity of ion channels and ion porters, but not ion pumps (fig. 3). There are likely to be unknown families of ion transporters that remain to be discovered, and it is conceivable that such families are overrepresented in the parasites. In any case, using the diversity of the presently known families of ion transporters as a basis for hierarchical clustering, it was possible to segregate obligate endoparasites from free-living and facultative parasitic eukaryotes (i.e., species with a free-living life stage). Such a separation between obligate endoparasitic and free-living species has not been observed in many related analyses performed based on predicted gene products other than ion channels, for example, enzymes of pyrimidine metabolism (Ali et al. 2013), vitamin B (Stoffel et al. 2006), or porphyrin synthesis (Godel et al. 2012). To our knowledge, the only contender category of proteins that would allow a similar distinction is the enzymes of purine de novo synthesis, a pathway which appears to be missing in all obligate endoparasitic eukaryotes (Hassan and Coombs 1988; de Koning et al. 2005). The picture presented by the ion transporters (fig. 3) is less clear-cut as no single protein is absent from all obligate endoparasites while present in the free-living eukaryotes. Nevertheless, the present approach to in silico ionomics demonstrates that there is convergent evolution among unrelated parasites with respect to ion transporters (fig. 4). By applying random forest classification to the data set, we were able to demonstrate that the most distinctive features of the obligate parasites were the lack of predicted ammonia transporters and transporters of iron or other transition metals (fig. 5).

All obligate endoparasites salvage nitrogen-containing nutrients from their hosts such as amino acids, purines, or pyrimidines. These provide them with an ample source of organic nitrogen, rendering the transporters of inorganic nitrogen redundant. Hence, the independent loss of ammonia transporters as a consequence of a metabolic streamlining in parasites—with the notable exception of T. cruzi, which is the only obligate endoparasite among the unicellular eukaryotes that scores high against the ammonium transporter profile (TC 1.A.11.2; supplementary table S3, Supplementary Material online). The observed loss of transporters of divalent cations might have a similar explanation as suggested earlier for the ammonium transporter, the parasites accessing iron from their hosts in other form than free Fe2+ or Fe3+. African trypanosomes express unique, glycosylphosphatidylinositol-anchored receptors on their surface for mammalian transferrin, a glycoprotein that transports Fe3+ in the blood. The heterodimeric receptor is internalized by endocytosis upon binding of host transferrin (Taylor and Kelly 2010). Trichomonas, Toxoplasma, and Entamoeba also obtain iron from mammalian iron-binding proteins such as lactoferrin, transferrin, or ferritin (Lopez-Soto et al. 2009; Horvathova et al. 2012; Ortiz-Estrada et al. 2012). Malaria parasites probably cover their iron requirement from ferric heme, the oxidized end product of hemoglobin degradation. Although the majority of heme molecules polymerize to hemozoin in the food vacuole, some escape to the parasite’s cytosol where they may serve as an iron source after degradation (Ginsburg 1999). For many other parasites, the pathways of iron salvage remain unknown. However, the general importance of iron acquisition for pathogens is illustrated by the phenomenon of iron withholding, a typical defensive response of mammals (Ganz 2009; Weinberg 2009). The peptide hormone hepcidin functions as a master regulator of iron absorption and distribution in mammals, and its expression is critically determined by infection and inflammation (Drakesmith and Prentice 2012).

We conclude that there is strong convergence among obligate endoparasites in the loss of ion transporters (possibly combined with divergence in new strategies for ion uptake) and propose access to organic ammonia and iron to contribute to the selective forces in the evolution of parasitism.

Supplementary Material

Supplementary tables S1–S3 are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/).