Conserved domains (CD) in proteins play a crucial role in protein interactions, DNA binding, enzyme activity, and other important cellular processes. With recently released gene number predictions in the human genome [1,2] being less than many previous predictions, interactions among these domains may prove to be central to proteome complexity. Protein domains are often conserved across many species, and as such, they offer an interesting dataset in how genomes maintain them with relationship to other conserved domains, as well as to proteome size. Many groups have attempted to find, document and annotate these conserved domains. While most groups use a form of Hidden Markov Models [3,4] for profiling, each group approaches the problem in a unique way yielding a wide range of databases that can be used to verify each other.

For this study we used the SMART conserved domain database [5,6,8] to collect data on the number of genes containing each CD in each genome. We restricted our study to the 5 eukaryote genomes sequenced so far, those being H. sapiens, D. melanogaster, A. thaliana, C. elegans, and S. cerevisiae. We confirmed our results using an independent source of similar data called the Proteome Analysis Database [7,9] (abbreviated here as PAD) and checked the sequenced eukaryotic genomes available at the time of writing, those being D. melanogaster, C. elegans, and S. cerevisiae.

We have used this unique opportunity to compare conserved domains across different genomes, and validated the approach by using two separate databases. The findings reveal a close link between numbers of genes with a given CD and the total number of genes in each genome.

Our initial observation was: for many conserved domains, the ratio of the sum genes in genome 1 containing the conserved domain to the total number of predicted genes in genome 1 was proportional to the ratio of sum genes in genome 2 containing the conserved domain to the total number of predicted genes in genome 2. Or:

Given that:

A = sum proteins with given CD in genome 1

B = sum proteins with given CD in genome 2

C = sum predicted genes in genome 1

D = sum predicted genes in genome 2

Then on average:

A/C ≅ B/D     (Relationship 1)

Upon rearranging Relationship 1, it was noted that for many CD's the ratio of the number of genes containing the given CD in each genome accurately reflected the ratio of the total predicted number of genes of each genome. Or:

Given variables in Relationship 1,

Then on average:

A/B ≅ C/D     (Relationship 2)

To normalize the data we used a ratio of the sum genes with a given CD in a genome, to the sum genes with the given CD in all 5 (3 for PAD) genomes. This was used to minimize the effect that the predicted number of genes may be significantly wrong for one of the genomes while the others may be more accurate. Relationship 1 was rewritten to reflect this normalization. This resulted in the relationship:

Given that:

A = sum proteins with given CD in genome 1

E = sum proteins with given CD in 5(3 for PAD) genomes

C = sum predicted genes in genome 1

F = sum predicted genes for all 5(3 for PAD) genomes

Then on average:

A/E ≅ C/F     (Relationship 3)

The sums of CDs in each Relationship 3 ratio range were graphed for each genome, and are displayed in Figure 1 (SMART database) and Figure 2 (Proteome Analysis Database). The average ratio for each genome was calculated and multiplied against the sum predicted genes of all 5 genomes, yielding a number close to the predicted genes in each respective genome (Table 1).

Relationship 2 could be used to predict total genes in a genome given the other variables are reasonably well known, such as from Express Sequence Tag data. More importantly, this raises the question whether conserved domains are maintained in this ratio due to functional constraints or some other unknown reason. The fact that this ratio is maintained fairly well in all 5 eukaryotic genomes attests to its potential importance.

While there is much disagreement on total number of genes for the different genomes, similar gene finding methods were used for each of the 5 published eukaryotic genomes. It can therefore be assumed that ratios of predicted genes between the genomes will remain similar to present ratios as the gene numbers for each genome are further clarified. Likewise, neither SMART nor the Proteome Analysis Database claim to have found all occurrences of each CD in each genome. However, due to similar strategies used for CD finding in different genomes within each database, the ratio of total genes found with a given CD in each genome is likely to remain near constant as gene prediction improves.

An interesting finding from this research was that while ratios for H. sapiens, A. thaliana, and S. cerevisiae corresponded closely to total predicted genes for each organism, both databases gave a ratio that exchanged total predicted gene numbers between D. melanogaster and C. elegans (Figure 1, Figure 2, Table 1). While this exchange cannot be explained presently, it may offer insight into distinctions between the genomes, and genes that remain unidentified.

It has been shown that conserved domains in proteins are maintained in proteome specific ratio for the 5 eukaryotic genomes sequenced so far. The reasons for this ratio are unclear, but it would not be unreasonable to suspect functional interaction of these domains requires they be kept in a specific ratio. Further research will be needed to understand the reasons for, and universality of this ratio in eukaryotic genomes.

For searches against the SMART database, we limited our data to conserved domains occurring at least once in each of the 5 genomes [8]. For the Proteome Analysis Database we restricted our search to those conserved domains listed in the top 200 occurring domains for which there was at least one occurrence in each of the 3 genomes [9]. This strategy of limiting the study to more global CD's was used to increase the chance that the conserved domains were constructed correctly and to increase statistical reliability of the results.

Data gathering was carried out as follows, a perl script was written to submit requests to the SMART database [8] for number of genes with each of 519 CDs in each genome. Information in the Proteome Analysis Database [9] is already in genome specific columns for the top 200 occurring CDs, and, as such, was downloaded directly. The information was parsed and stored for each genome. From the SMART database 211 conserved domains were selected based on the fact that they occurred at least once in each of the 5 genomes (see SMART_CDs for information on these domains). From the Proteome Analysis Database 147 conserved domains were selected based on the fact that they occurred at least once in each of the 3 genomes (see PAD_CDs for information on these domains).

The total number of predicted genes for each genome was as follows: H. sapiens, 35,000 [1,2]; D. melanogaster, 14,100 [10,11]; A. thaliana, 26,000 [11,12,13]; C. elegans, 19,100 [11,14]; S. cerevisiae, 6,300 [11]. This yielded a total of 100,500 genes for all 5 genomes, and a total of 39,500 for D. melanogaster, C. elegans, and S. cerevisiae alone. The number of genes in each of the eukaryotic genomes is an approximate number because the number of genes predicted is always a changing estimate constantly being clarified [13].

1. SMART_CDs is a text, tab delimited file containing all 211 conserved domain names from the SMART database used in this study. For each conserved domain name, the corresponding number of genes containing the CD in each genome is listed.

2. PAD_CDs is a text, tab delimited file containing all 147 InterPro entry numbers for the domains in the Proteome Analysis Database used in this study. For each InterPro entry number, the corresponding number of genes containing the CD in each genome is listed.

Conserved domain (CD); Proteome Analysis Database (PAD) Note: the terms gene and protein, are frequently used interchangeably in this paper as from a conserved domain perspective they are similar.