To test whether proteins with similar phylogenetic profiles are functionally linked, we examined the phylogenetic profiles for two proteins that are known to participate in structural complexes, the ribosome protein RL7 and the flagellar structural protein FlgL, as well as a protein known to participate in a metabolic pathway, the histidine biosynthetic protein HIS5. We first identified all other E. coli ORFs with phylogenetic profiles identical to those of these three proteins and then those ORFs with profiles that differ by one bit. The results are shown in Fig. .
Homologs of ribosome protein RL7 are found in 10 of 11 eubacterial genomes and in yeast but not in archaeal genomes. We find that more than half of the E. coli proteins with the RL7 phylogenetic profile or profiles that differ from it by one bit have functions associated with the ribosome (Fig. A). Because none of these proteins have significant amino acid sequence similarity to RL7, the functional relationships to the ribosome—had they not been determined already—could not be inferred by sequence comparisons. This finding supports the idea that proteins with similar profiles are likely to belong to a common group of functionally linked proteins. Several other proteins with these profiles have no assigned function and are listed accordingly as hypothetical. The testable prediction of the clustering of phylogenetic profiles is that these as yet uncharacterized proteins have functions associated with the ribosome.
The comparisons of the phylogenetic profiles of flagellar proteins (Fig. B) further support the idea that proteins with similar profiles are likely to be functionally linked; 10 flagellar proteins share a common profile. Their homologs are found in a subset of five bacterial genomes: those of Aquifex aeolicus, Borrelia burgdorferi, B. subtilis, Helicobacter pylori, and Mycobacterium tuberculosis. Other proteins that appear in neighboring clusters (groups of proteins that share a common profile) include various flagellar proteins and cell-wall maintenance proteins. Flagellar and cell-wall maintenance proteins may be biochemically linked, because flagella are inserted through the cell wall. For example, the lytic murein transglycosylase MltD has a phylogenetic profile that differs by only one bit from that of the flagellar structural protein FlgL. This transglycosylase cuts the cell wall for unknown reasons. Therefore, another prediction is that this enzyme may participate in flagellar assembly.
Fig. A and B includes proteins in structural complexes, whereas Fig. C shows proteins involved in amino acid metabolism. We find that more than half of the proteins with phylogenetic profiles similar (within one bit) to that of the histidine synthesis protein His5 are involved in amino acid metabolism. With the 16 currently available fully sequenced genomes, however, phylogenetic profiles are not able to separate the metabolic pathways of specific amino acids. Instead, because of the limitations of currently available data, a histidine biosynthesis protein seems to have the same profile as a tryptophan, arginine, and cysteine synthesis protein. It is probable that, as more genomes are fully sequenced and the number of entries in phylogenetic profiles is increased, similar but distinct amino acid metabolic pathways will cluster separately in phylogenetic-profile spaces.
The examples included in Fig. show that proteins with phylogenetic profiles similar to a query protein are likely to be functionally linked with it. We next show the converse: that groups of proteins known to be functionally linked often have similar phylogenetic profiles. As shown in Table
1, we chose groups of
E. coli proteins that share a common keyword in their SwissProt (
7) annotation, reflecting well known families of functionally linked proteins. Because homologous proteins coded by the same genome necessarily have similar profiles, they were eliminated from the groups. For each group, we computed the number of protein pairs that are neighbors; neighbors were defined as proteins whose profiles differ by less than 3 bits. For a group of
n proteins there are at most [
n(
n − 1)]/2 possible neighbors.
| Table 1Phylogenetic profiles link protein with similar keywords |
The similarity of the phylogenetic profiles of the proteins that share a common keyword was evaluated by a statistical test; we compared the number of neighbors found in our keyword groups to the average number of neighbors found in a group of the same size but with randomly selected E. coli proteins. We found that, on average, the random sets contain very few neighbors compared with the keyword groups, even though the keyword groups contain only a fraction of all possible neighbor pairs. Thus, proteins that are functionally linked are far more likely to be neighbors in profile space than randomly selected proteins. However, we find only a fraction of all possible neighbors within a group. Therefore, not all functionally linked proteins have similar profiles; they may fall into multiple clusters in profile space. It is interesting to note that hypothetical proteins are also more likely to be neighbors than random proteins, suggesting that many hypothetical proteins are part of uncharacterized pathways or complexes.
A second indication that functionally linked proteins are likely to have similar phylogenetic profiles comes from the analysis of classes of proteins obtained from the EcoCyc library (
Encyclopedia of Escherichia coli Genes and Metabolism, ref.
8). We selected several classes that contain more than 10 members and that represent well known biochemical pathways. The results of our analysis are listed in Table
2. The conclusions that we draw from this analysis are similar to those found with the keyword groups: members of the group are far more likely to have neighboring profiles than members of a randomly selected control group.
| Table 2Phylogenetic profiles link proteins in EcoCycclasses |
Finally, we attempted to determine the ability of our method to predict the function of uncharacterized proteins. We equate the function of a protein with that of its neighbors in phylogenetic-profile space. This equation is accomplished by means of the keyword annotations found in the SwissProt database. To test the efficacy of this method, we compared the keywords of each characterized protein to those of the neighbors in phylogenetic-profile space. All of the neighbors, in this case, were other proteins with identical profiles. We found that on average 18% of the neighbor keywords overlapped the known keywords of the query protein. By comparison, random proteins had only a 4% overlap with the same set of neighbors. We make the rough estimate that, for more than half of E. coli proteins, we can assign the general function correctly by examining the functions of their phylogenetic-profile neighbors. This estimate should also hold true for the ability of phylogenetic profiles to assign functions to uncharacterized proteins.
This article has been 
