Here we present an analysis of orphan genes (hypothetical genes with unknown function) in the Streptomyces coelicolor genome, combining gene expression analysis and comparative genomics. The aim is to prioritize orphan genes for further study. In our gene expression studies [1, 2], we frequently encountered genes that showed interesting expression patterns, but had no known function. To identify which of these genes merit in-depth experimental analysis, we developed a strategy for prioritizing protein encoding genes for additional characterization, combining phylogenomic information [3] (i.e. the level of evolutionary conservation of each protein), and gene expression data from a large gene expression time series [1]. We postulate that widely conserved proteins that show a physiologically relevant dynamic expression pattern are the most promising candidates for further experimental study, e.g. using gene overexpression and knock-out or knock-down approaches.

The functional annotation of orphan genes is not only relevant for its basic biological interest, but is also an important help for the improvement of genome-scale metabolic models based on genome annotation. These models in their initial form almost always contain gaps that need to be filled by manual curation or automated gap-filling algorithms that add missing essential metabolic activities to the models [2, 4–7].

During our previous studies of genome-scale metabolic models of Streptomyces coelicolor and its relatives, we regularly had to postulate enzymatic functions that had not been assigned to specific proteins in the organisms [2, 7]. Assigning specific enzyme-coding genes to these orphan metabolic activities is very important for the subsequent analysis and interpretation of the models, and several approaches have been developed to assign sequences to the orphan metabolic activities: they employ, for example, mRNA co-expression analysis [8], phylogenetic profile information [9–11], pattern recognition techniques [12] or comparative genomics [13]. These approaches are organism specific and have mostly been employed for well-studied model organisms such as Escherichia coli and Saccharomyces cerevisiae.

Of the 7825 predicted protein coding genes in the Streptomyces coelicolor genome [14], according to a re-annotation of the genome sequence in 2009, 2688 (34%) are coding for functionally orphan proteins, i.e. proteins that are annotated as "hypothetical protein", "conserved protein", "putative membrane protein" or "putative secreted protein". Of these orphan proteins, 26 are conserved in all and 381 are present in at least half (22/44) of the 44 analyzed complete actinomycete genomes (see Methods section for a complete species list). 683 orphan proteins are present in at least 11 (25%) and 177 are conserved in at least 33 (75%) actinomycete genomes.

Of the 381 generally conserved actinomycete orphan proteins (i.e., those that are present in at least half of the analyzed genomes), 25 are also encoded in all species in a selected set of diverse non-actinomycete bacterial genomes (Bacillus subtilis, Escherichia coli K12, Lactobacillus plantarum WCFS1, Staphylococcus aureus, and Streptococcus pneumonia AP200), and 73 are present in at least three of the representative bacterial genomes (see Additional File 1: Supplementary Table 1.xls).

Of these 73 ultra-highly conserved bacterial orphan genes, 22 also have putative homologues (reciprocal best BLAST hits) in at least half of the species in a representative set of eight non-bacterial genomes (the eukaryotes Caenorhabditis elegans, Arabidopsis thaliana, Plasmodium falciparum, Drosophila melanogaster, Saccharomyces cerevisiae and Homo sapiens, and the archaea Haloterrigena turkmenica and Methanosarcina acetivorans). These proteins are therefore almost universally conserved; however, although there seems to be significant conservation of some orphan proteins, none of them is truly universal, i.e. none has a putative homologue in all of the 58 studied genomes. This is most likely due to the fact that some of the included actinomycete genomes are highly reduced, as a result of the parasitic lifestyle of the organism, and the large phylogenetic distance covered (with the corresponding major differences in basic physiological processes).

To prioritize the orphan proteins for further characterization, we therefore summarized the phylogenomic information (i.e. the level of evolutionary conservation of each protein) in a single "conservation" score, which expresses the degree of conservation across the three domains examined (actinomycetes, non-actinomycete bacteria, non-bacteria). This score was combined with a second measure of expression dynamics across a large gene expression time series studying the metabolic switch caused by phosphate starvation. In this experiment, a fermenter culture of S. coelicolor was grown in phosphate-limited conditions, and gene expression data were obtained at 32 finely spaced time points throughout the duration of the experiment. Phosphate was depleted after about 35 hours, triggering a metabolic switch from primary to secondary metabolism, accompanied by a rapid global reorganization of the transcriptome, involving genes with a wide range of biological functions, from central metabolism and antibiotic biosynthesis, to cellular development and maintenance [1]. The "expression dynamics" score described in the Methods section identifies genes that show a smooth expression trend across (part of) the time series and favors those genes that show a particularly strong (step-like) expression change at one time point. This is intended to allow to focus on genes that are not only passively following the expression change during nutrient depletion but that show evidence for active regulation, which is indicative of a central function in cellular physiology. Based on the p-value of the "expression dynamics" score, we assigned a rank to each gene, and averaged this value with the rank of the "conservation" score. 734 orphan genes are significantly up- or down-regulated with expression dynamic p-values less than 0.00001 (significant after multiple-testing correction).

Based on the gene expression profiles (Figure 1), the candidate genes SCO5746 and SCO1222 are particularly interesting: they show a very strong switch upon phosphate starvation, and their expression increases in stationary phase similar to the expression pattern of the antibiotic biosynthesis gene clusters act and red. All other genes show a decrease of expression along the time course. SCO5746 has a putative uncharacterized homolog in E. coli and contains a domain of the DegT/DnrJ/EryC1/StrS aminotransferase family. The aminotransferase activity was demonstrated for purified StsC protein, which acts as an L-glutamine:scyllo-inosose aminotransferase and catalyses the first amino acid transfer in the biosynthesis of the streptidine subunit of antibiotic streptomycin. It is therefore tempting to speculate that the SCO5746 gene has some role in the biosynthesis of a new antibiotic in S. coelicolor as well, and the same might be the case for the completely uncharacterized SCO1222. The closest putative antibiotic biosynthesis clusters are SCO5799-SCO5801 (siderophore synthetase type) and SCO1206-SCO1208 (chalcone synthetase type; [17]), both of which seem unlikely candidates for interacting with SCO5746 or SCO1222. However, it is possible that these genes contribute to a dispersed biosynthetic pathway, not involving a dense genomic clustering. Of course, they could also be contributing to any other stationary phase process.

Interestingly, we see a strong neighborhood conservation of most of the candidate orphan genes in other Streptomyces species (Figure 2). In some cases, the annotation of the neighbors does suggest at least a broad functional category: for example, SCO1521/1522 might be involved in DNA remodeling during recombination, as their conserved neighbors are a Holliday junction resolvase and DNA helicase (RuvABC complex); and SCO2081 might play a role in cell division, matching its conserved neighbor, the cell division protein ftsZ [18]. However, most of the conserved neighbors are hypothetical proteins themselves and do not seem to immediately identify a putative function for most of the orphan genes; nonetheless, the neighborhood information will be valuable for the design and interpretation of the most efficient experimental perturbations. The dynamic expression pattern of each of the neighborhoods depicted in Figure 2 is shown in Additional File 2: SupplementaryFile1.pdf. This illustrates, e.g., that the expression of SCO1522 shows a very similar expression trend compared with its left and right neighbors (SCO1521 and SCO1523), confirming the relevance of adjacency on the genome for predicting gene functionality.

The prioritization reported in this paper of course depends on implicit assumptions about what constitutes a protein of interest. Here we were a priori interested in any protein that is maintained by purifying selection in a large number of genomes, indicating that it is involved in a generally important physiological process. On the other hand, we assumed that genes that show strong gene expression responses to a major physiological perturbation are more likely to be functionally relevant under the conditions studied. Genes that are not expressed or not finely controlled are more likely to have more specialized functions. This approach does not exclude the identification of housekeeping genes, which may not be directly involved in the physiological process studied in the gene expression analysis, as these genes still tend to show dynamic expression patterns (as evidenced, e.g., by the ribosomal protein genes [1]). The results are, however, affected by the availability of gene expression data sets and will become more informative once other large-scale expression studies, comparable to the one used here, become available.

Our aim was to prioritize protein coding orphan genes (hypothetical proteins with unknown function) for further experimental characterization of their function. We developed an algorithm to detect dynamic switches in a large gene expression time course data set, and assigned an "expression dynamics" score to every orphan gene, arguing that genes that show substantial expression changes corresponding to biologically relevant events would be most interesting to follow up. We also summarized the available evolutionary information in a "conservation" score across a broad range of organisms (many actinomycetes, other bacteria and various non-bacterial species). We combined the "expression dynamics" rank and "conservation" rank to identify a robust list of 30 high priority orphan genes, which are promising candidates for detailed experimental study.