In order to detect genes from all three rodent malaria parasite (RMP) species, the DNA microarray design was based on the 'OligoRankPick' program 18 as this algorithm was shown to have significant improvements over other strategies even when dealing with large fluctuations in GC content and abundant gene duplications. Since the microarray designed here was to include all available information for three RMP species, there are key challenges and modifications to our original program so as to accommodate multiple genomes on a single microarray. Since P. yoelii is the only RMP with a published draft genome 3, we first designed oligonucleotides against the predicted P. yoelii gene models. Next, all possible oligonucleotides of P. yoelii were used to query the homologous regions of the other two RMP species (Figure 1A). Next, four parameters were used to select for the best oligonucleotide, namely: BLAST scores (first and second hit), GC content, sequence complexity and self-annealing scores (Figure 1B). Each score is then transformed into a rank and a weighted rank-sum is calculated for each oligonucleotide with the final oligonucleotide being selected based on the smallest rank-sum value. These oligonucleotides were then selected to be optimal for all three species, followed by oligonucleotides for P. yoelii and P. berghei and then for P. yoelii and P. chabaudi (Figure 1A). For the remaining predicted open reading frames oligonucleotides specific for each species were then designed. A total of 14,736 oligonucleotides were thus obtained and the breakdown is shown in Figure 2.

Since some of the oligonucleotides are designed to be capable of cross hybridizing with more than one species, we have re-sorted all the genes from the three species and matched them with their respective oligonucleotides so that an accurate normalized value can be obtained for analysis. Using this approach, we performed comparative genomic hybridization experiments and first looked at the performance of the species-specific oligonucleotides that were designed to hybridize with their target genome. After normalizing and filtering the data the percentage of successful oligonucleotides for P. yoelii-specific, P. berghei-specific and P. chabaudi-specific oligonucleotides were 91% (7,347), 90% (6,314) and 84% (6,356) respectively. It is not surprising that a low proportion of oligonucleotides were unable to hybridize with their intended targets especially if they were designed in regions with low confidence in sequence quality. A likely reason is the relatively poor genome coverage of the rodent parasite genomes as compared to the more complete P. falciparum genome. Also, there are errors in the current sequence drafts of the rodent malaria parasites due to the propensity of random sequence rearrangements of an AT-rich genome in a sequencing vector.

Using the validated set of oligonucleotides, we identified all the oligonucleotides that hybridized to the DNA of parasites they had not originally been designed against (Figure 2). A positive hybridization signal provides strong evidence that the sequences and therefore the genes represented by these oligonucleotides are also present in the genomes of the other rodent parasite species. This strongly implies that the respective genes that are represented by these oligonucleotides are actually present but are not found in the current database.

Since it has been established that the malaria parasite genomes are well conserved 5, it is conceivable that genes that are missing in the current draft of either of the RMP genome have a well conserved ortholog in one or two of the other species. Using the pan-rodent microarray we wish to investigate this possibility by inspecting comparative genome hybridization (CGH) signals on "cross-species" oligonucleotide microarray elements. For example high signal from P. yoelii gDNA for an oligonucleotide that is not designed to hybridize to P. yoelii implies that this gene (sequence) is present in P. yoelii. In summary, a species where the draft genome currently does not possess a particular gene but now gives a signal on the array suggests the presence of this orthologous gene. Hence, this gene can thus be detected based on homology.

Missing genes in the draft genome could arise due to two scenarios: either the sequence information is missing, or that the sequence is present in the genome but missed by current gene prediction algorithms. Since the oligonucleotides were designed based on predicted open reading frames, a CGH signal constitutes direct experimental evidence for the presence of an orthologous genomic sequence and thus potentially the gene in the RMP genome in which this gene is missing. Based on this approach, we detect 179, 306 and 215 genes missing in the current draft sequences of the P. yoelii, the P. berghei and the P. chabaudi genome, respectively. The majority of these genes code for hypothetical proteins that lack any functional assignment based on their amino acid sequence. For those whose function has been described, genes involved in biosynthesis, protein modifications, kinases and also invasion-related proteins in the case of P. chabaudi and P. berghei have been discovered (Additional file 1: Supplemental Table S1).

A selection of genes that were detected via the microarray data were randomly selected for polymerase chain reaction (PCR) screening (Figure 3) and direct sequencing in order to establish the confidence of this group of newly discovered genes. PCR primer pairs were designed flanking the oligonucleotide sequence to the species where there is known sequence information and these were used to amplify a newly predicted gene in another species where it is not annotated/predicted or where the sequence is absent. All screens were performed on regions where sequence information for the newly discovered orthologous genes is not present (i.e. missing sequence) except for PY00632 and PY03414 where a corresponding P. berghei contig is present but the gene is not predicted. In summary, 7 of the 8 PCR reactions gave a product around the predicted size and the one sample that was PCR negative could be due to sequence polymorphisms at the primer sites. Sequence analysis of the PCR product confirmed that the PCR product did indeed represent the predicted gene (data not shown). Some PCR products also exhibit a change in the predicted size, for example different PCR product sizes of PY00632 and its P. berghei ortholog are expected based on the currently available sequence information. Differences in PCR product size are due to variations in sequence length in the region bounded by the primers. The differences in PCR product size in the PY06972 screen of P. yoelii and P. chabaudi genomes are also due to the same reason. The high congruence of PCR-positive screens showed the power of the array in detecting homologous sequences currently absent from the available genome sequences of the other species. Additional microarray investigations pertaining to the detection and validation of polymorphic genes and differential transcription profiles between related parasite clones have further validated the performance of the oligonucleotide probes (unpublished).

Although a total of 700 novel orthologous rodent parasite genes have been discovered via the array approach, these revisions still leave significant gaps in the genomes of the three rodent parasite species. Therefore in addition to CGH, we also utilize bioinformatics to look at the genome in its entirety so as to probe for open reading frames that have been missed by the automated gene predictions provided by the RMP sequencing projects. The use of bioinformatics tools to query non-coding regions of the rodent malaria parasite genomes can be useful in detecting genes that have been missed by automated gene prediction algorithms. The more complete P. falciparum amino acid sequences were used to query the rodent malaria parasite genomes as a tBLASTn search using an expect value of at least 10^-15 as a cut-off threshold. While this approach does not account for synteny, evidence that regulation of transcription of individual genes occurred independently without any constraints on chromosomal location further supports this approach 1920. In addition, multigene families would be collapsed, as conserved domains would link such members together. Using this approach, 103 orthologous genes were discovered in P. berghei, 93 in P. chabaudi and 286 in P. yoelii. This data also suggests that while the P. yoelii genome is more completely assembled, it is lacking a number of predicted open reading frames. Similarly, PCR screens of these predicted genes were performed in order to obtain additional confidence for the bioinformatics derived dataset. For this confirmation, two groups were selected based on whether a homologous contig for a particular species is present in a region where a gene is not predicted, thereby checking if this gene is truly deleted from this locus or is still present, being either translocated to another locus or missed by the automated contig assembly (Figure 4). On another hand, we have observed another scenario where a newly discovered orthologous gene is not present in the species of interest. While sequence information is absent for these candidate genes, the adjacent genes are present in separate contigs. We thus wanted to screen these gaps or assembly errors that have excluded this gene (Figure 5). The re-annotated P. chabaudi gene orthologous to PFF1480w, PB000730.00.0 and PY03519 was screened twice as the best-fit P. chabaudi contig only contains the 5'-end sequence of this gene while the 3'-end is missing. Based on the random sampling approach (Figure 4), the PFI0535w-PB104921.00.0 pair seems to be exclusively found in P. berghei and not in the other 2 RMP species. For the PF14 0473-PC0001359.02.0 pair, it seems that this gene is also present in P. berghei while the MAL13P1.345-PY04218 pair looks to be also present in P. berghei as well. As for the other three pairs of genes (PFL0595c-PB301230.00.0¬PC000699.01.0; PFF1480w(3'end)-PB000730.00.0-PY03519; PF13_0131¬PC000708.04.0-PY04599) that seem to be missing in either one of the three RMP species, PCR screening suggests otherwise and that they are indeed common to all three RMP species. These results suggest that rodent parasite genes that are seemingly absent on a contig are likely to be present with high confidence in all three rodent malaria parasite species. Almost all of the other panel's screenings (Figure 5) showed that all of those genes were present in the three RMP species barring the P. chabaudi exclusive gene (PFL2450c-PC000344.03.0) that seemed to be present also in P. berghei but seemingly not in P. yoelii. This suggests with high confidence that the missing genes due to incomplete sequence information are present. Overall, it seems that it is highly likely that missing genes due to poor sequencing coverage are present and genes common to two rodent malaria parasite species are very likely to be present in all three species.

With the more complete coverage of the rodent malaria parasite genomes obtained via the analysis here, we now thought to define a common 'core' Plasmodium genome where orthologous genes present in all sequenced malaria parasites was defined. Plasmodium falciparum, being the most completely annotated and researched malaria parasite genome, is used as an index reference species in an attempt to consolidate genes common to as many species as possible. For each P. falciparum gene the orthologs present in the rodent parasite species were matched, and the resulting dataset was further appended with the respective orthologs from another human malaria parasite P. vivax and the simian parasite P. knowlesi as these genomes had been recently sequenced to 10× and 8× 2122 coverage respectively. This approach segregated P. falciparum genes into two groups: firstly, the 'core' genes containing orthologs to the rodent parasite as well as the P. vivax and P. knowlesi sequences (Additional file 2: Supplemental Table S2) and secondly, genes that have no significant alignment with rodent sequences (Additional file 3: Supplemental Table S3). While orthologous genes (with respect to P. falciparum) are highly conserved in regions where housekeeping genes predominate, the telomeric and sub-telomeric regions (i.e. non-syntenic) contain genes involved in antigenic variation that are more divergent 3 and this is consistent with the observation that the group representing the non-orthologous genes is dominated by var, stevor and rifin gene families that are involved in antigenic variation. What is immediately apparent is the high similarity of all six parasite species in the regions that contain housekeeping genes. In total, 4,188 genes were common between P. falciparum and the rodent parasites. Of these, 73 genes, mainly hypothetical genes, were absent in both P. vivax and P. knowlesi. In addition, 50 genes were present in P. vivax but not in P. knowlesi and 79 genes were present in P. knowlesi but not in P. vivax. Without these species-specific genes, we found 3,986 genes that are common to all six species.

An additional 387 P. falciparum genes (conserved in the two primate malaria parasites) appear to be missing in one or two of the rodent malaria parasite species (Table 1) with the vast majority of these representing hypothetical genes. The main exception represents genes conserved between P. falciparum, P. berghei and P. yoelii but missing in P. chabaudi from which approximately 30% have a functional annotation. Of the 387 genes, only a relatively small number (27-40) are exclusively found in only P. falciparum and one of the rodent malaria species with a significantly large number (92-96) being present in two species. While the genes missing in one or two rodent species could potentially represent species specific gene deletions, the fact that a significant number of these (72%-96%) orthologs are also present in the more divergent P. knowlesi and P. vivax genomes are highly suggestive that the respective rodent orthologs that have not yet been identified by either the sequencing project or the analysis reported here could likely be present.

On the other hand we identify 927 P. falciparum genes without a match to any rodent parasite species, 469 were composed of hypothetical genes (Table 2), with the majority of known genes being made up of the var, rif and stevor multigene families that are present at the subtelomeric regions of essentially all P. falciparum chromosomes (Additional file 3: supplemental table S3). Only about 14% of the 927 genes were conserved in P. vivax and/or P. knowlesi, with 117 being present in both species. While all genes exclusive to P. falciparum and P. knowlesi represent hypothetical genes, 6 genes specific to P. falciparum and P. vivax encode proteins involved in parasite invasion; 2 cysteine proteases and 4 MSP7-like genes. Taken together we define 4,188 genes that form a core gene set for a Plasmodium genus. From this set only a small fraction of genes (27 to 96 ~ 0.6%-2.2%) is lost in one or two of the six Plasmodium species. In contrast there are a large number of genes that appear specific to individual Plasmodium species and conserved only in one or two others. These mainly involve gene associated with host parasite interaction and are localized in the subtelomeric regions. However, limited diversification was also observed in genes present in the intrachromosomal region mainly involved in other Plasmodium specific functions such as invasion.

Using a similar approach as in the previous section, we attempt to construct a rodent parasite specific associative table since we can also take advantage of the dataset from the genomic hybridizations. For this purpose, instead of using genes from P. falciparum as an index, the oligonucleotides were chosen as the index and then replaced with the best-hit P. yoelii gene. Next, P. falciparum - P. yoelii, P. berghei - P. yoelii and P. chabaudi - P. yoelii orthologs were appended to the list. This strategy creates a database of genes orthologous to P. yoelii and the reason for choosing this species as the index is due to it being more completely annotated than the other two rodent parasite species. This list was refined by removing matching orthologs from P. falciparum that were already present in the common 'core' set; thereby creating a filtered list that contains only genes that are specific to the rodent malaria parasites. Microarray data and bioinformatics (tBLASTn) to query the P. yoelii amino acid sequences were also employed. In order to reduce the complexity of the dataset, the PIRs, which consist of the largest multigene family and dominate the rodent specific genes, will not be discussed due to their variability of expression and gene expansion amongst the rodent malaria parasites due to its main role in antigenic variation 2324 as well as its copy number being the most abundant in P. yoelii. Consequently, the huge reduction in genes common in all three rodent malaria parasite species is due to the removal of the PIR genes that could also confound relationships based on bioinformatics simply due to multiple matching of common conserved domains (Additional file 4: Supplemental Table S4; Figure 6).

Overall this approach now identified 1,238 genes that are common to all rodent malaria species most of which (1,013) represent hypothetical genes. Importantly only about 8% of these predicted genes have ORF that are shorter than 100 nucleotides making it less likely that these are annotation or prediction errors but indeed represent functional genes. Taking together these gene numbers along with the 4,188 genes found in the 'core' set, the total number of conserved rodent malaria genes is 5,426. In addition it appears that there are 210 genes specific to P. yoelii and P. chabaudi (Additional file 5: Supplemental Table S5), 247 genes specific to P. yoelii and P. berghei (Additional file 6: Supplemental Table S6) and 453 unique to P. yoelii (Additional file 7: Supplemental Table S7) with most of them (> 90%) again representing hypothetical genes. Of the 453 unique P. yoelii genes 45% are less then 100 nucleotide in length, compared to 24% in the P. yoelii and P. berghei group and 17% in the P. yoelii and P. chabaudi group indicating that a number of these genes represent false gene models.

A modification of the 'OligoRankPick' program 18 was employed to design 60-mer probes to the three rodent parasite species P. berghei, P. chabaudi and P. yoelii (Additional file 8: Supplemental Table S8; Additional file 9 - Supplementary Table S9) using sequences deposited in PlasmoDB http://www.plasmodb.org. Cross-species oligonucleotides were selected based on the criteria as described in the results section, i.e. at least 90% homology to the target sequence, less that 37.5% to non-target sequences and capped with a 5% difference tolerance in GC content. All possible oligonucleotides for a particular gene was then ranked and sorted based on target (>90%) and non-target hit (<37.5%), GC content (±5%) and the best Smith-Waterman score (Figure 1). Hence, all possible oligonucleotides were evaluated based on every available gene model and the best sequence was then selected. Oligonucleotides were spotted onto poly-L-lysine-coated microscopic glass slides 15.

Balb/c mice were used as vertebrate hosts for the propagation of P. berghei ANKA, P. chabaudi AS and P. yoelii 17 × 1.1. All procedures were in accordance with the guidelines for the use of experimental animals established by the Institutional Animal Care and Use Committee (IACUC) of the Nanyang Technological University of Singapore. Leukocytes were filtered from whole blood 29 and gDNA extracted using the Easy-DNA™ kit (Invitrogen) according to the manufacturer's protocol. For each parasite line, 3 μg of DNA was mixed with 2.5 μg of random nanomer primer to a volume of 15.25 μl. DNA was denatured at 100°C for 5 min and snap cooled on ice for 5 min. Labeled DNA was generated by adding dNTPs to a final concentration of 1 mM dATP and 500 fM each: dCTP, dGTP, dTTP and 5-(3-aminoallyl)¬2'-deoxyuridine-5'-triophosphate, (aa-dUTP) (Biotium), with 20 units of exo-Klenow Fragment (New England Biolabs) in a total volume of 50 μl and incubated at room temperature for 10 min and then overnight at 37°C. Cy-dye coupling, cleanup, hybridization, washing and slide scanning were performed as described by Bozdech and co-workers 15. Experiments were performed with at least duplicates for each species Subsequently, the data were normalized using the NOMAD microarray database http://ucsf-nomad.sourceforge.net/. For the comparative genome analysis, low quality features and features with a signal level less than two-fold of the background plus two-fold of the background standard deviation was filtered from the initial raw data set.

Verification of genes was performed via PCR thermocycling (Eppendorf) using the following conditions: 1 cycle of 95°C for 2 min; followed by 35 cycles of 95°C for 1 min, 50°C for 1 min and 68°C for 30 s. A final extension was set for 10 min at 68°C and then the tubes were kept at 4°C. For each 20 μl reaction, 2 ng of DNA was used as a template with 10 pmol of the forward and reverse primers (Additional file 10: Supplemental Table S10), 200 μM of dNTPs with 1.5 units of Taq DNA polymerase (Kapa Biosystems).

The sequences of P. falciparum, P. knowlesi and P. vivax were obtained from PlasmoDB http://www.plasmodb.org. For any gene missing an orthologous partner in another species even after complementation with the array data, known orthologs were appended using the resource available at PlasmoDB (version 5.5). Unannotated genes were identified using the tBLASTn search where amino acid sequences from a well-annotated species (PlasmoDB version 5.5) are used to query the genome of another species. A threshold of 10^-15 was used to ensure the stringency of locating matching genes.

Microarray data are publically available at the Centre for Information Biology Gene Expression Database (CIBEX; http://cibex.nig.ac.jp). The Accession number for these data is CBX114.