The genome of D. purpureum strain DpAX1, an axenic derivative of QSDP1, was sequenced using a whole genome shotgun sequencing approach (see Materials and methods) and assembled into 1,213 contigs arranged into 799 scaffolds with 240 larger than 50 kb (Additional file 1). There were 12,410 genes predicted and annotated using the JGI annotation pipeline (see Materials and methods); these are available from the JGI Genome Portal 13 and from dictyBase 14 . Thirty-three percent of the genes were supported by at least one EST clone and 89% of genes displayed some similarity to a gene in the NCBI non-redundant gene databases (Additional file 1). The genome size, gene count and average gene structure are very similar to those of D. discoideum (Table 1). Moreover, a recent comparative transcriptome analysis of D. purpureum and D. discoideum, using 'RNA-sequence' (RNA-seq), provides evidence for the transcription of 7,619 genes encoding protein orthologs within these species, or approximately 61% of the predicted D. purpureum genes 15 .

The D. purpureum genome contains 1.1 Mb of transposons (3.4%), fewer than in D. discoideum. The largest families of transposons are Gypsy (approximately 400 kb, 35.8% of total transposons), Mariner (approximately 186 kb, 16.7%), MSAT1_Dpu (126 kb, 11.4%), and hAT (105 kb, 9.5%).

The previously sequenced D. discoideum genome showed an unusually high number, length, and density of simple sequence repeats, including triplet repeats that code for amino acid homopolymers 1 . If unopposed by selection, simple sequence repeats can accumulate in genomes because of their high mutation rates and mutation to different repeat numbers that occur by misalignment and slippage during replication 16 . They are often thought of as non-functional 'junk' DNA, though some are known to be functional 17 , and the expansion of some triplet repeats in humans are known to cause disease when the number of repeats exceeds a particular threshold 18 . Despite its considerable evolutionary distance from D. discoideum (see below), D. purpureum also has a considerable density of simple sequence repeats (Figure 1a). Simple sequence repeats comprise 4.4% of the D. purpureum genome, compared to 11% in D. discoideum 1 . There are fewer long repeats that exceed 100 bp in length; 54 in D. purpureum compared to 1,436 in D. discoideum. The lower proportion of simple repeats in the D. purpureum genome and their shorter length may be due to current status of the assembly relative to the D. discoideum genome, since these repeats are difficult to assemble. Dinucleotide repeats, often the most common repeat in other species, are comparatively rare in both dictyostelid genomes (Figure 1b) 1 .

One of the most distinctive characteristics of the D. discoideum genome is the extreme abundance of amino acid homopolymers within coding sequences 1 . As in D. discoideum, simple sequence repeats are common in D. purpureum coding sequences (Figure 1a), particularly those with repeat motifs of three nucleotides or multiples of three (Figure 1b). These types of repeats contribute to many amino acid homopolymers (Figure S1 in Additional file 1), including 2,645 that are longer than expected by chance (>5 to >9 residues, depending on the amino acid; Table S1 in Additional file 1). Though the abundance and density is lower than in D. discoideum, the relative abundance of different amino acids repeats in D. purpureum is very similar, with asparagine and glutamine repeats dominating, followed by serine and threonine (Figure 2a). The correlation between the two species in the densities of different amino acid repeats is 0.997 (Pearson's correlation coefficient, P < 0.001), much higher than either species' correlation with Saccharomyces cerevisiae (0.516 for D. discoideum, and 0.486 for D. purpureum), or with Drosophila melanogaster (0.241 and 0.238). However, the correlations are also high for the densities of amino acid repeats with the A/T-rich protist Plasmodium falciparum (0.917 and 0.923), in agreement with a study showing that A/T content exerts a major influence on which amino acid repeats accumulate and persist within genomes 19 .

Codon usage within these amino acid homopolymers is quite similar to codon usage for the same amino acids outside of repeats, with a pattern quite similar to D. discoideum (Figure S2 in Additional file 1). Again, as in D. discoideum, many amino acid homopolymers contain a single codon, consistent with the relatively recent expansion of those triplet repeats. However, the codon diversity of D. purpureum amino acid repeats is significantly higher than it is for D. discoideum (Figure S3 in Additional file 1), consistent with the D. discoideum repeats being younger, with less time to accumulate changes from the original codon.

The potential function of most amino acid repeats is unknown, but the availability of the D. purpureum genome permits some new tests. If amino acid repeats are generally functionally important, they should tend to be conserved in their position within orthologous proteins. Sixty-four percent of the 2,645 D. purpureum amino acid repeats and 68% of the 11,243 D. discoideum repeats occur in genes that do not have homologs in the other species. Even in those with orthologs, only 19% of D. purpureum repeats and 5% of the D. discoideum repeats appeared to be homologous within global alignments of their respective proteins. The count of homologous repeats would be higher if we included matches where at least one falls below the threshold expectation for non-random homopolymers (for example, a match between 25 asparagines in D. discoideum and 8 in D. purpureum would be excluded as a chance event; P > 0.01; Table S1 in Additional file 1). On the other hand, some could be fortuitous matches forced by a large number of repeated amino acids that are not truly homologous. Inspection of selected sequences shows at least some that appear to be convincing homologs, with strong identity on both sides of the repeat (Figure S4 in Additional file 1). Still, the apparent small fraction of homologous repeats suggests that the very similar patterns of amino acid homopolymer abundance and distribution do not come primarily from conserved ancestral repeats. Instead they may come from some shared physiological properties - perhaps distinctive DNA polymerases or repair enzymes or high AT-content - that generate similar patterns independently.

In addition to the lack of homology for amino acid homopolymers between D. discoideum and D. purpureum, several pieces of evidence suggest that these triplet repeats may be 'junk' that accumulates due to weak selection on proteins that are relatively unimportant for fitness. For genes that have homologs in the two species, those with amino acid repeats in either species have higher non-synonymous substitution rates in the non-repeat regions, as expected if genes with repeats are generally less subject to purifying selection (Figure 2b). Another indicator of the degree of selective constraint on a gene is its expression level, particularly in the single-celled, vegetative stage where the selective pressure is likely to be the greatest. If amino acid repeats accumulate in genes where selective constraints are low, we would predict that they will be more common in genes expressed in the social or developmental stages, as opposed to vegetative stages. Using the recent comparison of the transcriptional profiles of D. discoideum and D. purpureum development by RNA-seq analysis 15 , this prediction is confirmed (Figure S5a,c in Additional file 1). Similarly, we would predict, looking only at RNA-seq reads from the vegetative stage, that genes coding for amino acid repeats would be less abundant and this is also confirmed (Figure S5b,d in Additional file 1). In sum, although a small number of repeats appear to be conserved over long periods of time, most appear to have arisen relatively recently in genes where selection against amino acid changes is weak.

A phylogeny based on small subunit ribosomal RNA gene sequences places D. purpureum and D. discoideum into distinct clades within the most derived of the four groups of social amoebae, the group 4 dictyostelids 7 . Thus, these two species should represent much of the diversity of the group. We constructed a global phylogeny of representative plant, animal, fungal and amoebal species, based on 389 orthologous gene clusters, in order to estimate the divergence of D. purpureum and D. discoideum relative to other eukaryotes (Figure 3). This analysis suggests that the group 4 dictyostelids span a comparable degree of protein sequence divergence as occurs among vertebrate species ranging from the bony fishes to the mammals. Recent comprehensive analyses of orthologous protein clusters from complete predicted proteomes suggests that the rates of protein evolution in the Amoebozoa are comparable to those of the plants and animals 20 . If gene sequence evolution occurs at the same rate in the two groups, these two observations suggest that D. purpureum and D. discoideum shared a common ancestor approximately 400 million years ago.

The initial description of the D. discoideum genome included 18 genes that were proposed to be horizontal gene transfer (HGT) events from bacterial species 1 . After 5 years of refinement of the underlying genome sequence, 16 D. discoideum genes remain potential HGT events. They have not been recognized in the characterized plant, animal or fungal genomes, and each of them is phylogenetically embedded within a bacterial clade. In addition, the thymidylate synthase gene, thyA, has been confirmed as an HGT; it is present only in a minority of the described bacterial species and is structurally unrelated to the canonical eukaryotic thymidylate synthase 21 . To narrow the time frame wherein the HGT events might have occurred, we searched the D. purpureum genome for orthologs to these genes. Each of the proposed D. discoideum HGT genes have an ortholog in the D. purpureum genome (Table 2). This suggests that all 16 of these potential HGT events occurred after the divergence of the Amoebozoa from the plants and animals, but prior to the radiation of the group 4 dictyostelids.

Functional information now exists for 6 of the 16 proposed HGT genes and it is interesting to see how the dictyostelids have utilized these contributions from bacteria. ThyA has completely replaced an essential enzyme in central metabolism 21 . Since it is also present in the amoebozoan slime mold Physarum polycephalum (GenBank accession number [GenBank:AAY87038] 22 ), the change over to the rare bacterial enzyme must have taken place quite early in the radiation of the amoebozoa. The isopentenyl transferase, IptA, produces discadenine, which is a sporulation inducer and spore germination inhibitor 23 . Another gene, pscA, encodes an active penicillin-sensitive peptidase but its function is not known 24 , and Ppk1 is a bacterial type polyphosphate synthase 25 . Colossin A (ColA) appears to be a structural protein of the slug that was fashioned out of hundreds of repeats of a bacterial Cna_B domain 1 . CapA and CapB are two cAMP-binding proteins whose carboxy-terminal half is derived from a subunit of a bacterial tellurium resistance complex 26 . Recently, CapB was identified in a proteomic screen for centrosomal proteins 27 .

Genomes evolve through base substitution and insertion/deletion, and also through rearrangements that alter the order and orientation of genes on chromosomes. Synteny, the nature and extent of conserved gene order between species, serves as an important gauge of the dynamics of genome evolution 28 . To characterize the potential synteny between D. purpureum and D. discoideum, we identified blocks of approximately conserved gene order between their genomes, and compared the number and sizes of these potential conserved syntenic blocks to control genomes in which the gene orders were artificially scrambled. Although the D. purpureum genome is not fully assembled, the current level of contiguity allows for an analysis of conserved gene order on a small scale (approximately 50 kb). Blocks of potential synteny were constructed by single-linkage clustering of D. purpureum genes, where pairs of genes are considered linked if (i) they fall on the same scaffold of the assembly with at most w intervening genes that have D. discoideum orthologs, and (ii) their D. discoideum orthologs all fall on a single chromosome, with no more than w intervening genes that have D. purpureum orthologs. For stretches of perfectly conserved gene order (blocks constructed with w = 0), 4,734 (63%) of the 1:1 ortholog pairs used in the analysis lie in a genomic block of conserved gene order involving at least two genes in each genome. The mean size of such blocks is 2.8 genes in each genome, with the longest perfectly conserved stretch containing 10 genes.

To determine the maximum length scale over which significant conservation of gene order persists, we compared the increase in potential syntenic clusters as a function of an increasing number of intervening genes (w) for D. purpureum versus D. discoideum to the rate obtained for the permutation controls (Figure S6 in Additional file 1). We found that for up to about 15 intervening genes, potential conserved gene clusters grow significantly faster than what is expected for the same two genomes with randomized gene orders, which provides a conservative threshold for identifying blocks of conserved gene order. With this estimate, 76% of orthologous gene pairs participate in a block of approximately conserved gene order, compared to 5.8 ± 0.4% in controls, with a false positive rate, on a gene-by-gene basis, of approximately 7%. The 5,793 genes contained in these blocks, and their positions in the genome, are listed in Additional file 2. This indicates that the majority of orthologs in D. purpureum and D. discoideum are found in small neighborhoods of exactly conserved gene order between the two species, and that these neighborhoods are themselves clustered into larger regions of approximately conserved gene order.

The described catalog of non-coding RNAs (ncRNAs) in the Dictyostelia was long limited to tRNAs, rRNAs, and a handful of experimentally identified short RNAs, all found in D. discoideum (for review, see 29 ). Recent work has expanded this repertoire to include a family of spliceosomal ncRNAs and two classes (class I and class II) of novel ncRNAs 30 31 . The spliceosomal RNAs identified in D. discoideum, U1, U2, U4, U5, and U6, are each characterized by both specific RNA-binding motifs and the ability to fold into characterized secondary structures 30 31 . Using a modified BLAST search (Additional file 1), we have identified a set of D. purpureum spliceosomal homologs that are predicted to fold into the appropriate secondary structures (Table S3a in Additional file 1).

In D. discoideum a 'Dictyostelium upstream sequence element' (DUSE) has been described that sits approximately 63 bp upstream of many ncRNAs, including the class I and II ncRNAs 31 . Identification of the DUSE motif ([AT]CCCA[AT]AA) in D. purpureum revealed that a DUSE also sits upstream of all D. purpureum spliceosomal RNA genes. The DUSE also enriches for a family of putative D. purpureum ncRNAs that are homologous to the two novel classes of D. discoideum ncRNAs. This suggests that the DUSE is not specific to D. discoideum.

Operating under the assumption that the DUSE sits upstream of certain ncRNAs in D. purpureum, we sought to identify novel ncRNAs by focusing on DUSE-enriched 8-bp sequences (see Additional file 1 for methods). Two of the three 8-mers that were found to be highly enriched, CCTTACAG and CTTACAGC, also occur in the novel classes of D. discoideum ncRNAs. These ncRNA gene products are 50 to 60 bp long and have distinct 5' and 3' sequences predicted to form 5-bp stem structures that are conserved within each class (Figure 4). Both classes share a 12-bp 'bulge' sequence, CCTTACAGCCAA, which is immediately 3' to the 5' stem sequence 30 . This 'bulge' sequence is predicted to not bind with any other region of the ncRNA, thus constituting a non-self-binding region (NSBR). The two 8-mers both sit within this NSBR.

To identify putative homologs to the class I and II ncRNAs in D. purpureum, we used the structural characteristics of these ncRNAs to filter all sequences containing the DUSE-enriched 8-mers. Forty members of the class I and II ncRNAs were originally identified in D. discoideum. Some are described as putative, with nine lacking the canonical bulge sequence, and five others lacking an upstream DUSE, or having a degenerate DUSE. The class I ncRNAs have a 5' stem sequence of GTTGA, while two class II ncRNAs have a 5' stem sequence of GCTCG, and all members have a 3' stem sequence complementary to the 5' stem sitting 40 to 70 bp away from the 5' stem 29 .

In our analysis of the masked D. discoideum genome, we identified 46 occurrences of the CTTACAGC 8-mer (Additional file 1). Of these, 26 possess both an upstream DUSE and a 5'/3' stem pair sitting 40 to 70 bp apart, and each corresponds to a previously identified class I or II ncRNA. In the masked D. purpureum genome there are 61 occurrences of the CCTTACAG 8-mer; 26 of these 8-mers have both an upstream DUSE and a 5'/3' stem pair consisting of an identical 5' sequence (GAATT) (Figure 4). These results suggest a class of ncRNAs in D. purpureum similar to the class I and II ncRNAs found in D. discoideum.

The comparative genomics approach to identifying these ncRNAs in D. purpureum lends deeper insight into their function. The 5' and 3' stem sequences have diverged between species, but have done so in a compensatory manner that maintains the predicted 5'/3' structure. The NSBR sequence, however, has remained perfectly conserved between species, and in neither species is it predicted to self-bind. This suggests a functional role for the NSBR beyond self-interaction, possibly as a binding site for another functional element. Initial genomic analysis of the dictyostelids Dictyostelium citrinum and Polysphondylium violaceum also revealed putative ncRNAs with an upstream DUSE, the conserved NSBR sequence, a 5'/3' stem structure, but 5'/3' stem sequences different from those of D. discoideum and D. purpureum (unpublished data).

Of the 12,410 predicted D. purpureum proteins, we identified 7,619 that are likely to be orthologous to D. discoideum proteins using the Inparanoid algorithm, best reciprocal blast hits, and manual curation (Additional file 3). An additional 2,759 predicted proteins are similar to genes in D. discoideum, while 2,001 appear to be unique to D. purpureum (Additional file 4). Thus, at least 84% of the protein-coding genes in D. purpureum share orthologs or paralogs in the D. discoideum genome. The gene product predictions from the D. purpureum genome should be enormously useful for further refinement of the predicted proteome of D. discoideum. Some gene families are completely conserved between D. purpureum and D. discoideum, with clear orthologs for every member of the family, while other families appear to have undergone considerable divergence between the two species (Figure S9 in Additional file 1, and Additional file 4). The differences amongst gene family members should illuminate the physiological differences between these two dictyostelids, whereas the similarities may indicate where the selective pressures, exerted by their common environment, have resulted in stable gene inventories required for survival.

Polyketide synthases (PKSs) are enzymatic production lines for making small molecules by the repeated condensation of malonyl-CoA and other thio-esters of coenzyme A (CoA). A large number of polyketides exist and are probably made for ecological purposes, but they also serve as model natural products for the development of drugs, antibiotics and food additives. Soil amoebae are not commonly regarded as polyketide producers, but they too must face complex ecological challenges, which could be met by polyketide production; competition from other amoebae, infection by bacteria and predation by nematodes, amoebae and fungi. A small number of potential eco-chemicals have been identified from social amoebae 32 33 , but the completed D. discoideum genome sequence revealed a much larger potential 1 34 35 . These PKSs are large, modular proteins of 2,000 to 3,500 amino acids, each having a core of domains for the condensation reaction, together with optional domains for methylation, carbonyl reduction and product release. Two have a unique, 'steely', architecture in which a second PKS - a chalcone synthase - is fused to the carboxyl terminus of a modular PKS 36 . One of these steely proteins makes the precursor of differentiation-inducing factor (DIF)-1, a chlorinated signal molecule for stalk cell differentiation 37 , and the other a pyrone or an olivetol derivative 35 36 38 .

The D. purpureum genome has 50 predicted PKS genes. We constructed phylogenetic trees using the highly conserved ketoacyl synthase and acyl transfer domains of the PKS genes from both species to discern evolutionary relationships (Figure 5a; see Table S6 in Additional file 1 for corresponding genomic loci). The two steely genes within each species are only distantly related to each other but are clearly orthologous between species. This implies that both genes were present in the last common ancestor and that their function has been maintained in both species. There is also a clear ortholog in D. purpureum of the methyltransferase catalyzing the last step of DIF-1 biosynthesis 39 and so D. purpureum is likely to make DIF-1, like D. discoideum, and Dictyostelium mucoroides 40 , another group 4 dictyostelid 7 . Two other clear orthologous pairs of genes are apparent. Dp2 and the very similar Dd1/Dd2 likely encode fatty acid synthases based on their similarity to other fatty acid synthases and their high expression levels. Dp12 and Dd3 are of unknown function, though mutation of Dd3 causes a 'cheater' phenotype, suggesting that it may produce a developmental signal 41 .

In contrast to the four D. purpureum genes described above, most D. purpureum PKS genes do not have obvious orthologs in D. discoideum, indicating species-specific expansions. Given the overall gene conservation between these two species, the divergence of the PKS gene sets is striking. We speculate that this greater evolutionary fluidity reflects different selective pressures placed on the two species, perhaps by different competitor species in their ecological niches, and therefore that most of their polyketides are produced for ecological purposes.

The D. purpureum genome confirms the high potential of social amoebae for polyketide production. The relative paucity of orthologs to D. discoideum PKSs raises the possibility that polyketide production varies substantially from species to species amongst the dictyostelids. As natural products remain the major source of drugs 42 , this diversity suggests that natural products of social amoebae deserve systematic exploration.

The ABC transporters are one of the largest protein superfamilies that are encoded by any genome. In stark contrast to the lineage-specific radiation of the PKS proteins, the complement of ABC transporters has remained remarkably stable since the divergence of D. purpureum and D. discoideum. ABC proteins all have a conserved domain of 200 to 250 amino acids, the ATP-binding cassette, and typically have 12 transmembrane domains. Seven different eukaryotic families have been defined on the basis of sequence homology, domain topology and function. The superfamily has been extensively analyzed in D. discoideum 43 and this allowed a detailed comparison to the predicted D. purpureum ABC superfamily members. Both genomes carry similar numbers of ABC genes overall, but differences in gene number can be observed within groups of closely related genes belonging to the largest families (Tables S7 and S8 in Additional file 1). Only 58 genes can be considered clear orthologs; the remaining genes should be considered paralogs (Figure S10 in Additional file 1). These genes may play partially redundant roles and this might allow their sequences to drift to a point of uncertain orthology.

The Tag subfamily proteins (TagA-D) of the ABC B family have a novel domain structure with a serine protease domain on the amino terminus, a single set of six transmembrane domains, and one ABC domain on the carboxyl terminus. Three of the Tag proteins have defined roles in cell differentiation; TagA is involved in early cell fate determination 44 , TagB is required for pre-stalk cell differentiation 45 , and TagC is expressed in pre-stalk cells and required to process acyl-CoA binding protein into a spore differentiation peptide signal 46 . Interestingly, TagA, B and C are conserved between D. purpureum and D. discoideum, but whereas the TagA orthologs are quite similar, the relationship between the TagB and TagC proteins in the two species is not as clear (they were named based on their gene order within a block of synteny between D. discoideum and D. purpureum).

D. purpureum has a similar complement of protein kinases compared to D. discoideum. Like D. discoideum, D. purpureum does not appear to have receptor tyrosine kinases, or other notable protein kinases such as P70, ATM, and PASK. There are 262 eukaryotic protein kinases and 41 atypical protein kinases, including potential pseudogenes (Table S9 in Additional file 1). This compares to 247 identified eukaryotic protein kinases and 39 atypical protein kinases in D. discoideum 47 .

The 14 D. purpureum histidine kinase genes, and the related acrA gene, each have an unambiguous ortholog in D discoideum (Figure 5b). There is little homology between non-orthologous genes outside of the kinase domain. Thus, the histidine kinases appear to have diverged from a common ancestor before the radiation of the dictyostelids, suggesting that each one of them carries out a distinct and conserved function. The adenylyl cyclase of D. discodeum, AcrA, carries a non-functional histidine kinase domain with mutations in key amino acids that preclude kinase activity 48 . This domain and its variations are well conserved in the D. purpureum AcrA, suggesting that there is a selective advantage to maintaining this non-catalytic domain, probably as a dimerization domain.

The catalytic subunit of cAMP dependent protein kinase (PKA), PkaC, in D. purpureum shows 65% amino acid identity with its D. discoideum ortholog. The homology is highest in the catalytic core and lowest in the low complexity amino-terminal domain, with the exception of the region encompassing the αA amphipathic helix 49 . This helix, which is predicted to interact with a hydrophobic pocket on the catalytic core of the enzyme, is 95% identical in these dictyostelids, which is suggestive of a conserved regulatory function. The regulatory subunit of PKA, PkaR, of D. purpureum and D. discoideum shows 79% amino acid identity and each of them lack the dimerization domain found in metazoa.

GPCRs are found in all eukaryotes and transduce a variety of extracellular signals via heterotrimeric G-proteins and effector proteins inside the cell to elicit physiological responses. GPCRs are characterized by an extracellular domain, an intracellular domain, and a core domain that contains seven transmembrane regions. The GPCRs are subdivided into six major families that, aside from their conserved secondary domain structure, do not share significant sequence similarity. The D. purpureum genome encodes the same families of GPCRs as in D. discoideum, but has a reduced total number, which is mainly due to differences in the numbers of cAMP, family 3 and family 5 receptors (Figure S12 and Table S10 in Additional file 1). There are only two cAMP receptors in the D. purpureum genome, namely orthologs of Dictyostelium carA and carB, but there are no orthologs of carC and carD. In addition, there are 35% fewer family 3 receptors and 40% fewer family 5 receptors. This difference must be due either to an expansion of family 3, 5 and cAR receptors in D. discoideum or to a reduction in the D. purpureum genome. Either D. discoideum has evolved many new functions for GPCRs compared to D. purpureum or else there is more functional overlap amongst the D. discoideum receptors.

The overall comparison of transcription factors in D. discoideum and D. purpureum shows gross conservation both in the total number of genes in each family, and at the protein sequence level (Table S11 in Additional file 1). There are only 11 basic leucine zipper (bZIP) domains in D. purpureum, versus 19 in D. discoideum. Among the 11 bZIPs found in both species are DimA and DimB, which are involved in DIF signaling in D. discoideum, as well as bZIP candidates for CREB and GCN4, which are the most conserved bZIPs among eukaryotes (E. Huang, M. Katoh-Kurasawa and G. Shaulsky; unpublished). There are an equal number of STAT transcription factors in D. purpureum and D. discoideum (four), each with a high degree of protein sequence identity. In the original description of the D. discoideum genome, the paucity of transcription factors was noted 1 . One explanation for the small number of recognized transcription factors was the possibility of new classes of transcription factors that evade conventional detection based on sequence searches. One example is the recently defined CudA nuclear protein that binds in vivo to the promoter of the cotC prespore gene 50 . CudA-related proteins have recently been defined as being specific to the amoebozoa 51 , but there are distantly related proteins in plants 50 .

The D. purpureum repertoire of microfilament system proteins is almost an exact replica of that described in D. discoideum (Table S12 in Additional file 1) 52 . In contrast, the actin-depolymerizing factor (ADF) protein family differs between the Dictyostelium species. A phylogenetic tree of all ADF domains encoded by the genomes of both species shows three major groups (Figure S13 in Additional file 1). The ADF domains present in cofilin, twinfilin and GMF (glia maturation factor) constitute one group. D. purpureum has two genes encoding cofilins, cofA and cofG. Only cofA has a direct ortholog amongst the eight D. discoideum genes. An additional group of ADF domains is present in D. purpureum that includes three proteins, one of which (DPU_G0064410) has no direct ortholog in D. discoideum and another (DPU_G0060306) that is related to two D. discoideum genes (DDB_G0270134 and DDB_G0270132).

A family of proteins where there has been some expansion in D. purpureum is that of the I/LWEQ domain-containing proteins. Besides two talins and a single Sla2/HIP1, D. purpureum harbors three more genes related to hipA encoding only a carboxy-terminal fragment that encompasses the I/LWEQ domain. It is not clear whether these are actually pseudogenes. Similarly, we have found a group of at least eight genes that encode short proteins related to the carboxy-terminal part of HIP1 immediately upstream of the I/LWEQ domain. The extensive family of calponin homology (CH) domain proteins in D. purpureum has two members absent in D. discoideum. One (DPU_G0069574) is related to conventional fimbrins but lacks EF hands and has a weakly conserved fourth CH domain. The other (DPU_G0074288) is a protein with a carboxy-terminal CH domain.

Cytoskeletal remodeling during chemotaxis and phagocytosis is regulated by a considerable number of upstream signaling components. Especially important are those components involved in signaling to and from small GTPases of the Rho family, as recently described in D. discoideum 53 . In general terms the repertoire of genes encoding proteins that participate in Rho signaling is very similar in both dictyostelid species, with some exceptions (Tables S13 and S14 in Additional file 1). The Rho GTPase family itself has diversified considerably in D. purpureum and D. discoideum (Figure S14 in Additional file 1). This family currently comprises 20 rac genes and one pseudogene in D. discoideum and 18 genes in D. purpureum. Most D. discoideum rac genes have a direct ortholog in D. purpureum, but the degree of conservation is variable. There is a second rac1a-related gene, indicating that the ancestral rac1 gene duplicated independently in each organism. There is no ortholog for D. discoideum rac1b, rac1c, racF1, racF2, racI and racM to racO, and the pseudogene racK and, conversely, D. purpureum has five more rac genes without a D. discoideum counterpart (racR to racW), again indicating that the rac family has undergone independent divergence in both species.

Among the Rho regulators D. purpureum appears to have one RhoGAP gene less than D. discoideum. The missing RhoGAP gene is gacII; the corresponding protein consists of a RhoGAP domain followed by a SH3 domain. The protein is very similar to the amino-terminal half of RacGAP1 (xacA gene), suggesting that gacII resulted from a partial duplication of xacA in D. discoideum. Among the Rho effectors, the class PI4P 5 kinases have undergone a notable expansion in D. purpureum (Table S14 in Additional file 1). Additional descriptions of Ras superfamily members can be found in Additional file 1.

Protein N-glycosylation, the most prevalent and highly conserved type of protein glycosylation, is initiated in the rough endoplasmic reticulum of D. discoideum by the transfer of a 14-sugar chain from a lipid-linked precursor 57 that is identical to the yeast and human precursor but distinct from that of many protists 58 . Maturation of the sugar chain leads to a preponderance of high-mannose glycans with bisecting and novel intersecting β-linked GlcNAc, and α3-linked core fucose characteristic of plants and invertebrates, followed by increased α-mannosidase processing during development 59 60 . D. discoideum N-glycans are often rendered anionic by phosphorylation and sulfation 57 59 , in contrast to the typical sialic acid or uronic acid modifications of animal glycans.

A genomic comparison suggests that the N-glycome of D. purpureum will be similar to that of D. discoideum but with some interesting differences (Table S16 in Additional file 1). For example, putative CAZy GT49 β3-GlcNAc transferases, GT10 α3/4-fucosyltransferases, and glycophosphotransferases, expected to mediate peripheral modifications of N-linked and perhaps other glycans, are represented by much smaller gene families in D. purpureum, and low amino acid sequence similarities make ortholog predictions for individual family members less certain. Thus, D. purpureum may exhibit reduced prevalence and diversity of its peripheral glycan modifications.

The most dramatic predicted difference between the two dictyostelid glycomes stems from the apparent absence in D. purpureum of the four-member CAZy GT17 class of GT-like proteins expected to mediate addition of peripheral bisecting and/or intersecting β4-GlcNAc residues. We tested this by performing a matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry glycomic analysis, which confirmed the presence in D. discoideum of N-glycans containing two peripheral GlcNAc residues and/or an α3-linked core fucose, and revealed an apparent absence of these species in D. purpureum (Figure 6). The results suggest that CAZy family GT17 and GT10 sequences present in D. discoideum but absent from D. purpureum encode a novel N-glycan β-GlcNAc transferase and a core α3-fucosyltransferase, respectively, emphasizing the value of comparative genomics for predicting gene functions. Other studies have indicated that N-glycans are dominant contributors to the cell surface glycocalyx, and therefore may strongly influence intra- and inter-specific encounters with other amoebae, and interactions with potential predators, pathogens and prey. Thus, the dramatically different N-glycomes of these species might contribute to, for example, their differential sorting in interspecific mixtures 61 .

A previous inspection of the predicted D. discoideum proteome also indicated the existence of some major classes of biosynthetic enzymes associated with mucin-type O-glycans, O-phosphoglycans, and glycosylphosphatidylinisotol (GPI) anchors 54 , in agreement with biochemical studies 57 . For example, mucin-type O-glycosylation is initiated in the Golgi by a CAZy GT60 polypeptide α-GlcNAc transferase, conserved in both dictyostelids and related to the polypeptide α-GalNAc transferases associated with mucin-type O-glycosylation in animals 62 . Glycophosphorylation of the hydroxyamino acids threonine and serine may be less prevalent in D. purpureum owing to the much smaller size of its glycophosphotransferase-like gene family (Table S16). Although the glycogene comparison suggests a general conservation of these other aspects of the glycome, differences suggest that there may be equally dramatic variations as observed for N-glycosylation.

Whereas glycosylation occurs predominantly in the secretory compartments, formation of the precursors for these pathways generally originates in the cytoplasm, and the cytoplasm is also a site for catabolic deglycosylation. The genome encodes proteins associated with these functions as expected. In addition, like most eukaryotes, D. discoideum encodes a potential nucleo-cytoplasmic Spy-like GT41 O-GlcNAc transferase (OGT or Ser/Thr-βGlcNAc transferase) and D. purpureum encodes two; the physiological function(s) of O-GlcNAc in protists is currently unknown 63 . D. discoideum also possesses a complex cytoplasmic O-glycosylation pathway that modifies hydroxyproline and has an ancient evolutionary relationship with O-glycosylation in the secretory pathway and bacterial glycosylation 64 . The genes of this pathway are highly conserved in D. purpureum, and bioinformatics and biochemical data indicate its partial conservation across at least four major protist phyla. This pathway is devoted to the modification of the E3 ubiquitin ligase subunit Skp1, and is involved in oxygen regulation of development in D. discoideum 65 .

Many glycan functions are mediated in trans via carbohydrate binding domains (CBDs) or lectins. During initial remodeling within the rough endoplasmic reticulum, N-glycans are recognized by lectins in a folding/quality control cycle and, unlike many protists, this pathway appears to be highly conserved between the dictyostelids and animals 66 . D. discoideum encodes numerous cytoplasmically localized lectins, including multiple discoidin, Cup and comitin proteins 67 68 69 , and glycogen-binding proteins involved in metabolic regulation (Tables S18 and S19 in Additional file 1). Except for the latter, the natural glycan ligands in the cytoplasm are unknown. Interestingly, discoidins, like galectins of animals, exit cells via a non-classical process and potentially bind self, prey or predator glycans containing Gal or GalNAc 70 . Discoidin and Cup CBDs appear to be dictyostelid-specific and evolutionarily dynamic, suggesting they serve species-specific functions as suggested for other lineage-specific expansions 71 .

Both genomes encode a few more glycohydrolases (Table S16 in Additional file 1) than glycosyltransferases, with suspected substrates ranging from dietary polysaccharides and glycans of bacterial, yeast and perhaps other prey to endogenous glycans for recycling. Potentially 11 of the glycohydrolase domains are fused to carbohydrate binding modules (CBMs), a subset of CBDs associated with enzymes. As described for some cellulases such as CelA, the CBM may localize the enzyme to the target substrate after secretion, and may also directly promote catalysis 72 . Peptidases may be localized at the cell surface by a similar mechanism. Cellulases are likely to be involved in remodeling of slime sheath cellulose during morphogenesis and spore coat breakdown during germination. D. purpureum and D. discoideum also have a cellulase associated with extracellular digestion in fungi and other cellulose-digesting organisms (CAZy GH7), suggesting a similar role in the social amoebae. Though the number of glycosyltransferases and known glycan binding proteins is 10 to 20% smaller in D. purpureum than D. discoideum, correlating with fewer peripheral modifications, the number of potential glycohydrolases is approximately 10% greater. The latter differences occur in lysozyme-, chitinase-, and alpha-mannosidase-like enzymes, suggesting variation in the spectrum of bacterial and yeast prey between the species.

cAMP controls many aspects of dictyostelid development. As a dynamically secreted chemoattractant it directs the cell movement that causes cells to aggregate, and aggregates to transform into fruiting structures. Secreted cAMP also triggers pre-spore differentiation, up-regulates the expression of aggregation genes and down-regulates stalk gene expression. As an intracellular messenger for other stimuli, cAMP induces spore and stalk encapsulation and maintains spore dormancy 23 46 75 76 .

In D. discoideum, 19 proteins are directly responsible for synthesis, detection and degradation of cAMP and its sister molecule, cGMP, which acts as a signaling intermediate for chemotaxis 77 . To assess whether cyclic nucleotides play similar roles in D. purpureum development, we analyzed conservation and change in all genes that are directly involved in cyclic nucleotide signaling. D. discoideum uses the adenylate cyclases ACA, ACB and ACG and the guanylate cyclases sGC and GCA for synthesis of cAMP and cGMP, respectively 76 78 . All five cyclases are present in D. purpureum inclusive of their functional domain architecture (Figure 7a). The structurally distinct cell surface cAMP receptors (cARs) and intracellular cyclic nucleotide (cNMP) binding domains are the sole targets for cyclic nucleotides in dictyostelids. D. discoideum has four cARs, which are conserved in two other group 4 taxa, D. mucoroides and Dictyostelium rosarium 79 . Nonetheless, only the cAR1 and cAR2 genes were detected in the D. purpureum genome (Figure 7b). No firm conclusions about the absence of cAR3 and cAR4 can yet be drawn, since the assembly of this genome is not fully complete.

The cNMP binding domains are found in the regulatory subunit of PKA (PkaR), the cGMP binding proteins GbpC and GbpD and the phosphodiesterases (PDEs) PdeD and PdeE. PdeD is a cGMP phosphodiesterase that is stimulated by cGMP binding to its cNMP binding domains, while PdeE is a cAMP-stimulated cAMP phosphodiesterase 76 80 . GbpC is a complex multidomain protein in which cGMP binding to its cNMP binding domains sequentially activates the intrinsic RasGEF, Ras/Roc and protein kinase domain, which eventually leads to increased cell polarization. GbpD also contains a RasGEF domain, but no output protein kinase domain. Its cNMP binding domains are not functional and it functions as an antagonist of GbpC in the chemotactic response 81 82 . Genes encoding all five cNMP binding proteins with their complete sets of functional domains are present in the D. purpureum genome (Figure 7c).

In D. discoideum cyclic nucleotides are hydrolyzed by three structurally distinct PDEs 80 . The cAMP PDEs RegA and Pde4 and the cGMP PDE Pde3 harbor a PDE_I type domain with HDc motif that is common to mammalian PDEs. The dual-specificity PDEs PdsA and PDE7 harbor a PDE_II type domain with HSHLDH motif. PdeD and PdeE, which hydrolyze cGMP and cAMP, respectively, carry a related HCHADHDS motif, but are structurally more similar to the lactamase_B protein family. The PDE_I and PDE_III enzymes are fully conserved between D. discoideum and D. purpureum, but the latter species has three instead of two type II PDEs (Figure 7d).

The high level of conservation between D. discoideum and D. purpureum of all adenylate and guanylate cyclases, cNMP binding domains and seven out of eight PDEs, combined with the complete conservation of functional domain architecture of these proteins, is indicative of the central roles of cAMP and cGMP in the control of chemotaxis, morphogenesis and gene regulation in the dictyostelids.

DIF is produced predominantly by pre-spore cells during D. discoideum development and is part of a signaling mechanism that sets the ratio of stalk and spore cells produced in the fruiting body. It both limits the number of pre-spore cells produced and induces differentiation of a subset of pre-stalk cells. DIF is made by a three step biosynthetic pathway, in which a 12-carbon polyketide is assembled by the StlB polyketide synthase, then successively chlorinated by a chlorinating enzyme, and methylated by the DmtA methyltransferase 36 83 84 . Clear stlB and dmtA homologs exist in the D. purpureum genome, as does a homologue of a recently identified FAD-dependent chlorinating enzyme (C Neumann, C Walsh and RR Kay, unpublished). DIF is inactivated by glutathione-dependent dechlorination 85 , and again this enzyme has recently been identified and has a clear homolog in D. purpureum (F Velazquez and RR Kay, unpublished). It thus appears certain that D. purpureum makes and degrades DIF in a similar way to D. discoideum, and presumably utilizes it in a similar way to regulate multicellular development.

Dictyostelids are interesting social organisms because about 20% of the cells in each fruiting body sacrifice themselves to build the stalk. Groups form through aggregation of formerly separate cells, so different clones can aggregate together, and do so in both the lab and in the field 86 . Clones that successfully compete to get into spores, relegating their partners to the sterile stalk, will be more successful. Some degree of conflict is therefore predicted, and the resulting evolution of strategies and counter strategies may drive rapid adaptive evolution, as appears to be true for genes involved in host-parasite conflicts and male-female conflicts 87 . But there is also a second reason to expect that social genes may evolve more rapidly than genes expressed primarily in the solitary stage. If the social stage occurs relatively infrequently, which seems likely but is unknown, then social genes are less scrutinized by selection and could accumulate more changes through genetic drift.

We tested for more rapid evolution using two ways of defining social genes. The first was to examine the set of genes that emerged from a selection for mutants that cheat (make more than their fair share of spores in mixtures) and compare them with all other genes 41 . The two sets do not differ significantly in the probability of having homologs, suggesting that they neither differentially disappear nor differentially evolve beyond the point of clear homology (Figure S18a in Additional file 1). The two sets also do not differ in probability of having paralogs, suggesting that they do not duplicate at different rates (Figure S18b in Additional file 1). Finally, the two sets do not differ for either dN (the rate of non-synonymous change) or conservation score (a measure that declines with both point differences and with non-aligned portions of the sequences) (Figures S18c and S19d in Additional file 1). However, this set of social genes is relatively small, and some will be false positives (of 198 genes identified, 40 were tested for cheating, of which 31 were cheaters).

A larger set of social genes can be identified using RNA-seq reads from the vegetative stage and six social time points (4, 8, 12, 16, 20, and 24 hours after starving) 15 . Using genes with sufficient reads and high reproducibility (Additional file 1), we defined a gene's index of social expression as the average percentage representation in the social-stage libraries over that average plus the percentage representation in vegetative stage (that is, Social expression/Social expression + Vegetative expression).

Using this classification, social genes showed higher rates of change, and manifested fewer orthologs, higher rates of non-synonymous substitution, and lower conservation scores. Genes with orthologs in D. discoideum and D. purpureum have a significantly lower social expression index in D. discoideum than those without orthologs (Figure S19a in Additional file 1; n = 1,739, 1,300, P < 2.2e-16, Mann-Whitney U test). This is driven by significant differences in each time point of the developmental stages (data not shown). An analysis using genes with D. purpureum RNA-seq data following the above criteria gives a similar overall result (Figure S19b in Additional file 1; n = 3,649, 2,102, P < 2.2e-16, Mann-Whitney U test). Thus, genes with more social expression are less likely to have orthologs, indicating more rapid evolution in the gain or loss of genes, or in change of genes beyond the point where they are identifiable as homologs. Homologs that had inparalogs show no significant difference of social indices from those that lacked inparalogs when the social expression is measured with RNA-seq reads from D. discoideum (Figure S19c in Additional file 1; n = 95, 1,644, P = 0.93, Mann-Whitney U test) and D. purpureum (Figure S19d in Additional file 1; n = 137, 3,512, P = 0.24, Mann-Whitney U test), suggesting no significant bias of duplication for the social genes of dictyostelids.

Figure 8a plots conservation score and rates of non-synonymous substitution (dN) as a function of the D. discoideum social expression index. As the percentage of a gene's RNA-seq reads found in social stages increases, there is a significant drop in conservation score and a significant increase in dN, supporting the hypothesis that social genes change more rapidly than vegetative ones. The same significant results hold if we use the RNA-seq reads of D. purpureum genes (Figure 8b). Figure S20 in Additional file 1 shows how the effect is partitioned between different social stages. Previous studies have suggested that individual social genes or small sets of them evolve rapidly because of evolutionary arms races, with conflict driving continuing adaptation and counter-adaptation 88 . This is the first such evidence on a genomic scale. Nonetheless, we cannot rule out the alternative hypotheses that the lower selective scrutiny of social genes might arise if the social stage is not very frequent and not as selectively important as the vegetative stage. Distinguishing these hypotheses further will have to await the more sensitive tests that can be applied to genomes that are more closely related than D. discoideum and D. purpureum.

Dictyostelium has a sexual cycle in which two cells fuse and then engulf many other cells to form a giant macrocyst that undergoes meiosis. However, with the exception of one successful cross 89 , the sexual system has not been available in lab studies of D. discoideum. Although macrocysts are readily formed, there are problems with germination 90 and, when there is germination, there may be no recombinants 91 . Finding the right conditions for sex would add a valuable genetic dimension to D. discoideum studies, but this search would be fruitless if most strains have lost the ability to have sex. If they have lost this ability, we would expect that sex-specific genes would have degraded. We tested this hypothesis using ESTs from gamete-stage libraries made from cells grown in conditions that make them competent for fusion 92 . Figure 8c shows that genes expressed disproportionately in the gamete stage are actually more conserved than other genes, as measured by both dN and conservation score. Provided these truly are sex-specific genes, then it appears that the macrocyst system is functional and not degenerating in D. discoideum. This is supported by an analysis showing that 13 meiosis genes 93 have normal dN and conservation score values compared to other genes (Figure S21 in Additional file 1).