Sequence Analysis: Strategy, Methods, and Significance Criteria.

For database searches, we adopted an iterative strategy whereby sequences similar to the protein of interest are extracted from the database, and the search procedures are iterated until convergence (12). Additional database searches were performed after partitioning protein sequences into distinct domains. Globular and nonglobular domains were predicted based on sequence complexity, using the seg program with the parameters L(1) = 45, K₂(1) = 3.4, K₂(2) = 3.75 (18). Database search then was performed individually with the sequence of each of the predicted globular domains. For detection of subtle sequence similarities, an important parameter is the size of the database searched; databases of nematode, yeast, and bacterial proteins were used in an attempt to increase the search sensitivity (12).

Protein sequence databases were searched for pairwise similarity using the blastp and blast2 programs (19, 20). Nucleotide sequence databases translated in six reading frames were searched using the tblastn program (19). The blast programs estimate the statistical significance of local alignments produced by database searches using the extreme value distribution statistics for a single alignment segment and the sum statistics for two or more compatible segments. In blast2, the statistics have been modified to include gapped local alignments (20). The values of the probability (P) of obtaining an alignment with a given score by chance computed by blast are reliable as long as regions of biased composition (low complexity) do not contribute to the alignment scores (21). In our analysis, filtering for low-complexity regions was applied twice, first with parameters tuned for the detection of nonglobular domains (see above) and then with the standard parameters optimal for eliminating short, widespread low-complexity segments [L(1) = 12, K₂(1) = 2.2, K₂(2) = 2.5]. With these parameters, a P value of 0.001 or less was considered a strong indication of homology. Alignments in the “twilight zone,” namely those that were detected with the default blast2 parameters but had an associated P value greater than 0.001, were further evaluated using iterative blast2 searches and methods for motif and multiple alignment analysis (12).

The program psiblast (Position-Specific Iterative blast) was used to search the database iteratively, until convergence, with position-specific matrices derived from the original blast2 output and modified after each iteration (S.F. Altschul, A.A. Schaffer, T. Madden, and D.J. Lipman, personal communication). The P values computed by psiblast are comparable to those produced by blast and blast2. Alternatively, conserved sequence blocks were extracted from blast outputs using the cap program (22), or from multiple sequence alignments, using the macaw program (23). The blocks were used for iterative motif searches with the MoST program, with the cutoff set as r = 0.02 or r = 0.01 (22). Hidden Markov models were constructed from multiple alignments and used for iterative database screening with the local version of the Needleman–Wunsch algortihm (24).

Sequence Conservation and Homologs in Model Organisms.

Nearly all of the disease gene products show significant sequence similarity to other proteins in current databases (Fig. ​(Fig.11C and Table ​Table1),1), though for some, sequence conservation involves only short motifs. Most of the disease genes also are represented by highly similar homologs among the expressed sequence tags and in rodents (ref. 31; http://www.ncbi.nlm.nih.gov/Disease_Genes). Only 16 of the 70 proteins contain domains with significant similarity (P < 0.001) to proteins with known three-dimensional structure (ref. 32; http://www.pdb.bnl.gov). Thus in spite of the rapid growth in the number of known structures, homology modeling is not yet applicable to the majority of disease gene products, and sequence similarity search remains the principal route to functional inference (but see examples below, in which structural implications were made possible by iterative database screening).

We specifically assessed the similarity between the disease gene products and their homologs in extensively sequenced model organisms, namely the nematode C. elegans, yeast S. cerevisiae, and bacteria. In each case, we sought to show which homologs are direct counterparts of the given disease gene product with a high level of similarity and the same domain organization (orthologs), and which showed only a distant similarity and/or are similar only to distinct domains or motifs. The criteria for distinguishing orthologs from paralogs have been discussed (33). Briefly, in a comparison of proteins from two species, orthologs are expected to show: (i) significantly higher similarity to one another than to any other sequence from the second species; (ii) significantly higher similarity to one another than to any sequence from phylogenetically more distant organisms; (iii) alignment through the entire length of the proteins, i.e. conserved domain organization. Each of these criteria is critical for inferring the same function for orthologs of functionally characterized genes.

The majority of positionally cloned gene products showed sequence similarity to proteins from each of the model organisms (Fig. ​(Fig.11C); typically, this similarity was significant by the criteria used in blast2 searches (P < 0.001) but in several cases, only conserved motifs were detected (Table ​(Table11 and http://www.ncbi.nlm.nih.gov/Disease_Genes). In all cases when we inferred an orthologous relationship, the similarity between a disease gene product and its apparent ortholog was highly significant (P < 10⁻¹² for C. elegans and yeast orthologs, and P < 10⁻⁴ for bacterial orthologs). There are pronounced differences in the representation of the disease gene set by apparent orthologs in the nematode, on one hand, and in yeast and bacteria on the other hand. Though the worm gene collection is only about 50% complete, 36% of the disease genes appear to have orthologs in C. elegans. It is likely that nearly all conserved domain families already are represented among the nematode sequences, but many individual genes are still missing. Thus, with the completion of the C. elegans genome, the fraction of human positionally cloned genes that have worm orthologs is expected to increase substantially.

By contrast, in yeast, and especially in bacteria, apparent orthologs were found only for a few of the disease genes (Fig. ​(Fig.11C, Table ​Table1,1, and http://www.ncbi.nlm.nih.gov/Disease_Genes). More frequently, the yeast or bacterial homolog contains counterparts to only one domain or a subset of domains of a human protein, and conversely, the proteins from the model organisms may have additional domains of their own (Table ​(Table11 and http://www.ncbi.nlm.nih.gov/Disease_Genes).

Domain rearrangements are well documented for a number of protein families, primarily from vertebrates (34). In our analysis of the positionally cloned genes, which may be representative of a significant subset of human genes, we observed that, at large phylogenetic distances, such rearrangements appear to be a rule rather than an exception. This may have important implications for functional interpretation of the results obtained with homologs of disease genes in model organisms. Indeed, sequence comparisons for 29 human genes that have been isolated by functional complementation of yeast mutations indicated that 25 pairs of the human and yeast genes involved are likely orthologs with a conserved domain architecture; in a few cases, human paralogs of the respective yeast genes are also known but they do not complement the mutations (http://www.ncbi.nlm.nih.gov/Bassett/cerevisiae/). As most of the positionally cloned disease genes do not have yeast orthologs (Fig. ​(Fig.11C), they are unlikely to complement mutations in the yeast homologs. This is even more applicable to bacterial systems.

Protein Functional Categories.

More than one-half of the disease genes encode proteins involved in various forms of cell communication and signaling. These include intracellular regulators such as guanine nucleotide-releasing factors and GTPase activators, and membrane and secreted proteins, such as several receptors and transport system components. Genes involved in genome replication, repair, and transcription are represented by a few transcription regulators and DNA repair proteins, whereas proteins involved in translation are absent (http://www.ncbi.nlm.nih.gov/Disease_Genes).

For 11 proteins, no function has been reported or could be confidently predicted by sequence analysis. In most of the other proteins, even though some domains have known or strongly predicted functions, others remain uncharacterized. Only for a small subset of the disease genes, e.g., glycerol kinase or PAX-6 protein (the product of the gene mutated in aniridia), the functions are understood so thoroughly that further insights through computer analysis are unlikely.

Previously Unknown Motifs and Functional Predictions for Positionally Cloned Disease Gene Products.

In this work, previously undetected, conserved domains and motifs were detected in a number of disease gene products. Some of these findings suggest a new function and/or biochemical activity; other motifs await functional characterization. Table ​Table11 contains the data on those of the disease gene products, for which biologically relevant inferences are possible based on the sequence conservation.

ATP-Binding Domain in a Colon Cancer Gene.

The hereditary nonpolyposis colon cancer gene product (MLH1), which is homologous to the bacterial mismatch repair protein MutL (35), is predicted to contain an ATP-binding domain conserved in the HSP90 family of chaperone proteins, type II DNA topoisomerases, and signal-transducing histidine kinases (Fig. ​(Fig.22A). A position-dependent weight matrix produced by psiblast from the part of the blast2 output that included only highly conserved sequences of MutL homologs retrieved numerous sequences of histidine kinases, type II (ATP-dependent) DNA topoisomerases, and HSP90 from the NR database selectively and with low P values of 10⁻⁴–10⁻⁵. Each of the three blocks with the highest sequence similarity between MutL, HSP90, topoisomerases, and histidine kinase families corresponds to one of the conserved motifs in the histidine kinases (44).

Open in a separate window

Figure 2

Previously undetected conserved domains and motifs in positionally cloned disease gene products. Alignments were constructed using the macaw program. Unique identifiers for each sequence are shown. Distances to the ends of the proteins and distances between the aligned, conserved blocks are shown by numbers. Conserved bulky hydrophobic residues (I, L, M, V, F, Y, W) are indicated by yellow shading and by U in the consensus. Other conserved residues are shown in magenta. Other designations in the consensus: O, small residues (A, G, or S); +, basic residues (K and R); −, acidic residues (D and E). (A) A putative ATP-binding domain in hereditary nonpolyposis colon cancer gene product (HML1), MutL mismatch repair proteins, HSP90 chaperones, type II DNA topoisomerases, and bacterial histidine kinases. The sequences were from the swiss-prot database: MLH1__HUMAN, human MutL homolog, colon cancer susceptibility gene product; PMS1__YEAST and MLH1__YEAST, yeast mismatch repair gene products homologous to MutL; HEXB__STRPN, mismatch repair gene product from Streptococcus pneumoniae; MUTL__ECOLI, E. coli mismatch repair gene mutL product; HTPG__HAEIN, HTPG__BACSU, HS90__THEPA, HS90__CANAL, and HS9A__HUMAN, molecular chaperones of the HSP90 family from Haemophilus influenzae, Bacillus subtilis, Theileria parva, Candida albicans, and human; TOPB__HUMAN, TOP2__DROME, GYRB__HALSQ, PARE__ECOLI, and GYRB__ECOLI, type II topoisomerases from human, Drosophila melanogaster, Haloferax sp., and E. coli; SPHS__SYNP7 and PHOR__BACSU, histidine kinases involved in inducible alkaline phosphatase production from Synechococcus sp. and Bacillus subtilis; PILS__PSEAE, histidine kinase involved in fimbriae biogenesis from Pseudomonas aeruginosa; PHY1__TOBAC, tobacco phytochrome A1 (histidine kinase homolog); ENVZ__ECOLI, osmolarity sensor histidine kinase. Three motifs described in histidine kinases and phenotypes of E. coli envZ mutants (36) are shown below the alignment. Dominant-negative mutations in MutL protein (37) are indicated by gray shading. Asterisks indicate amino acid residues that in E. coli GyrB are in direct contact with ATP; two of such residues are in the spacer between motifs G₁ and G₂ (38). The secondary structure assignments are from the crystal structure of the N-terminal fragment of E. coli GyrB (38); h, α-helix, e, extended conformation (β-sheet), and l, loop. (B) A putative nuclease domain conserved in Werner syndrome gene product (WRNp), bacterial RNase D, and DNA polymerase I. The sequences were from swiss-prot (names with underlines) or from GenBank (National Center for Biotechnology Information accession numbers indicated below). PMSC__HUMAN, human polymyositis and scleroderma autoantigen; RND__HAEIN, RND__ECOLI, and RND/SYNSP (gi1001530), RNase D from H. influenzae, E. coli, and Synechocystis sp.; ORF/SCHPO (gi1256512), uncharacterized ORF product from Schizosaccharomyces pombe; DPO1__ECOLI, DPO1__HAEIN, DPO1/SYNSP, DPO1/TREPA, and DPO1/BPT5, DNA polymerase I from E. coli, H. influenzae, Synechocystis sp., Treponema pallidum, and bacteriophage T5. The structural assignments are from the Klenow fragment structure (39); the designations are as in A. The three aspartates that coordinate the two cations required for the 3′-5′ exonuclease reaction by the Klenow fragment are indicated by asterisks; the two residues directly involved in catalysis (40, 41) are indicated by exclamation marks. ExoI, ExoII, and ExoIII indicate the three motifs that are conserved throughout the nuclease superfamily (42, 43).

The ATP-binding site of E. coli DNA gyrase maps to the N-terminal fragment of the B subunit, and the crystal structure of this fragment complexed with an ATP analog has been determined (38). Each of the conserved motifs in our alignment contains residues that are in direct contact with ATP (Fig. ​(Fig.22A). In the gyrase, the loop in motif G₂ interacts with the phosphates, the conserved Asp in G₁ interacts with the amino group of adenine, and Asn in motif N is involved in Mg²⁺ coordination (ref. 45; Fig. ​Fig.22A). The flexible loop in G₁, which is the most conserved element in the four protein families, is not in direct contact with ATP, and its function will be of further interest.

These findings are compatible with the prediction of an ATP-binding domain in MLH1 and its homologs and suggest that the function of these proteins in DNA repair includes ATP hydrolysis and phosphoryl transfer; furthermore, MLH1 may be autophosphorylated as demonstrated for histidine kinases and HSP90 (46, 47). The recent analysis of random dominant-negative mutants of the E. coli MutL protein (37) showed that most of the mutations concentrated in the three conserved motifs in the predicted ATPase domain (Fig. ​(Fig.22A).

We thank Roman Tatusov and Roland Walker for writing some of the programs used in this work; Stephen Altschul, David Lipman, Tom Madden, and Alejandro Schaffer for providing the unpublished program psiblast, and Victor McKusick for helpful discussions.