Using computational database searches, we have demonstrated previously that no gene sequences could be found for at least 36% of enzyme activities that have been assigned an Enzyme Commission number. Here we present a follow-up literature-based survey involving a statistically significant sample of such "orphan" activities. The survey was intended to determine whether sequences for these enzyme activities are truly unknown, or whether these sequences are absent from the public sequence databases but can be found in the literature.
We demonstrate that for ~80% of sampled orphans, the absence of sequence data is bona fide. Our analyses further substantiate the notion that many of these enzyme activities play biologically important roles.
This survey points toward significant scientific cost of having such a large fraction of characterized enzyme activities disconnected from sequence data. It also suggests that a larger effort, beginning with a comprehensive survey of all putative orphan activities, would resolve nearly 300 artifactual orphans and reconnect a wealth of enzyme research with modern genomics. For these reasons, we propose that a systematic effort to identify the cognate genes of orphan enzymes be undertaken.
After a decade of comprehensive genomic sequencing, more than 500 genomes have been sequenced to completion, mostly prokaryotes. The prodigious rate of new sequence annotation is highlighted by the fact that there were just over 300 genomes available when this study was carried out in late 2004. However, the fraction of genes for which no function can be predicted remains high (30%–50%). In response, proposals have been put forth for the bioinformatics analysis of bacterial genomes to identify genes with high likelihood of scoring true in confirmatory laboratory assays of their respective function [1, 2]. This would increase the field's pool of experimentally characterized proteins, with concomitant increases in the accuracy and coverage of genome annotation. We believe the return on investment of this approach would be particularly high when addressing the problem of orphan activities, that is, enzymatic activities for which no sequence information is available [3, 4].
Decades of detailed enzymology have created a wealth of knowledge about enzymes and their activities. However, crucial aspects of these enzymes are absent from bioinformatics databases with surprising frequency. For example, recent computational analyses of sequence databases demonstrate that at least 36% of enzyme activities that have been assigned an Enzyme Commission (EC) number  appear to be devoid of a gene or protein sequence . Since then, similar analyses have been published, with similar results [4, 6, 7]. The existence of such a large fraction of orphan activities is surprising, given that many of these enzymes have been described decades ago and are often involved in basic cellular functions. Several examples exist of the recent identification of genes involved in important enzymatic functions (reviewed in [1, 2]). Indeed, in our study 44 orphans were found to be present in one or more primary metabolic pathways in a variety of species (described below). Details of many of the orphan enzymes uncovered during this survey point to multiple and significant consequences for the lack of sequence information in areas such as genome annotation, computational pathway prediction, and metabolic engineering. For these reasons, the orphan problem and related issues were highlighted in a recent report of the American Society for Microbiology . In view of the biological richness associated with orphan enzymatic activities (Figure 1, Table 1), we have taken the first steps in creating the foundations of an Enzyme Genomics Initiative .
Here we describe a literature-based survey of presumed orphans intended to further validate and characterize these activities (Figure 2). The confidence of the results of this survey was designed to be within a 5% error margin relative to the universe of orphan activities, based on a randomly selected subset of orphan activities from the Nomenclature Committee of the IUBMB (NC-IUBMB). We have also assessed the practicability of identifying the genes associated with these orphans. As a consequence, the survey captures data from the literature that should facilitate the identification of cognate genes for the orphan activities evaluated. Here, we define the cognate gene for an activity as a gene that has been shown to code for an enzyme that carries out that activity.
The survey confirmed that ~80% of the sampled orphans do not have sequence information associated with them. Consequently, this lack represents a true information deficit. Weaknesses in database integration and a lack of information capture from the literature to databases appear to be largely responsible for most of the artifactual orphans making up the other 20%. Given the importance of these enzymatic activities, we propose that the public sequence databases assign high priority to correcting database entries for artifactual orphans. We further propose that a systematic effort be undertaken to sequence the genes of validated orphans, as this survey demonstrates that primary literature data and database analyses combined with current proteomics and genomic technologies should be adequate to enable the rapid identification of many of these genes.
Most orphan enzymatic activities are bona fide (Table 2). Our survey found that more than 80% of orphans are not due to artifacts such as missing database annotations (primarily failure to capture information from the literature), or lack of database cross-referencing, such as the availability of a sequence in one database not being reflected in a second database. Specifically, a total of 187 orphans out of 228 surveyed activities were validated in at least one of 287 species (species are listed in Table 3 and Table 4, the list of validated orphans is in Table 5). A majority of orphans (54.36%) occurred in Eukaryotes, followed by Eubacteria (39.37%) (Table 6). Within the Eubacteria, genus Pseudomonas was significantly overrepresented (35%) (Table 7). While a systematic determination of the species spectrum of orphan activities was not performed here, we did notice several cases of an orphan activity reported in more than one species, as well as one case of an orphan activity occurring in species from different domains.
Because the eventual isolation of the cognate genes of these activities is greatly facilitated by comprehensive genome sequencing, we determined for what fraction of all validated orphans a full genome sequence is available (Table 8). 43% of Eubacterial species in which orphans occurred were found to have such sequences, available either presently or due shortly. This figure rises to 83% when including the genomes of related species, on the assumption that they might be sufficiently closely related to permit the identification of the cognate gene. For example, at the time of this study the completed genome sequence of Pseudomonas fluorescens was not available, but those of three other Pseudomonas species were.
Oxidoreductases (EC1) and transferases (EC2) were the most frequently represented classes of enzymatic activity for validated orphans (Figure 3). On a per capita basis, oxidoreductases and transferases were overrepresented by ~20%, whereas hydrolases and ligases were underrepresented by 35% and 64%, respectively.
The original publication date for all orphans was broadly distributed around a mean of 1977 (Figure 4), compared to a mean of 1975 for validated orphans.
Causes of artifacts
A comprehensive list of artifactual orphans and the inferred nature of the artifact is available . Although this study was not designed to determine conclusively the causes of artifactuality, incompleteness in database entries appears to be the predominant cause of the artifacts identified here. For example, the DNA sequence associated with reaction 188.8.131.52 is available in the EMBL database, however, the UniProt entry for this enzyme does not list any protein sequence (Table 9). Other representative artifactual orphans are listed in Table 9, along with a description of the cause of the artifact. In a small fraction of cases a clear determination of the species in which the activity was characterized could not be made.
Extent of salvageability
Validated orphans were analyzed to determine whether sufficient information is available from their published characterization that, when combined with other factors, could enable the rapid identification of at least one cognate gene. Overall, we determined that 57 validated orphans (25% of total) might be salvageable (Figure 5A; Table 10), distributed approximately equally across eukaryotes and bacteria. Far more bacterial orphans were judged to have "excellent" or "good" salvageability as compared with eukaryotic orphans: 70% (7+12 out of 27) vs. 48% (5+11 out of 33), respectively (Figure 5B). This discrepancy is primarily due to factors such as the much greater difficulty for purifying an activity from higher eukaryotes, the difficulty of obtaining enough starting protein from lower eukaryotes such as multicellular fungi, and the absence of a comprehensive genome sequence from species such as Bos Taurus and Sus scrofa.
Overall, more than half of the salvageable orphans ranked "good" or "excellent", with oxidoreductases (EC1) and hydrolases (EC2) being overrepresented in that set. All other enzymatic classes were significantly underrepresented (Figure 6).
This survey demonstrates that ~80% of orphan enzymatic activities are bona fide; therefore, we conclude that of the 1,356 putative orphans extant at the time of this study, more than 1,000 are highly likely to constitute true information deficits since their lack of sequence information is not the result of a database error.
The absence of DNA or protein sequences encoding such well-characterized enzymatic activities is particularly consequential because these activities were often identified decades ago, and many have been the focus of significant research activity (Table 1). Without the cognate sequences for these activities, the quality of annotation of all sequenced genomes in terms of both coverage (fraction of genes that can be recognized) and accuracy (fraction of predicted gene functions that are correct) is diminished. Many of these activities may go for years without being sequenced – for example, 1-acylglycerophosphocholine O-acyltransferase (Table 1) was finally purified and sequenced nearly forty years after it was first characterized . Perhaps more troubling is the unknown pool of "false positive" annotations. Phosphogluconate 2-dehydrogenase (Table 1), an orphan at the time of this analysis, has since been assigned to a sequence in the human genome with no experimental evidence linking it to that or any homologous sequence, but apparently instead on the basis of the gene in question already being assigned a similar activity. This kind of "hidden orphan" would have been missed by most orphan analyses, and can be expected to propagate a potentially incorrect assignment to other genomes in the future. Computational metabolic pathway prediction  and metabolic engineering also depend on sequence information and are thus similarly compromised.
Conversely, ~20% of orphans surveyed were observed to be artifacts, such that ~270 orphans out of 1,356 putative orphans examined should be resolvable entirely via literature research and database cleanup. As a result of this process as it was carried out on our sampling of orphans, we have reported 11 artifactual orphan activities to public sequence repositories for correction (see Table 8 for examples).
In addition to validating orphans, the survey was useful in capturing information from the literature to assess their salvageability: more than half of validated orphans were found to be salvageable (Figure 5). Examples of salvageable orphan activities with the traits that make them salvageable are listed in Table 11.
As abundantly noted elsewhere, such database cleansing is essential to maximize the existing research investment and prevent the propagation of mistakes [12–14] (see Table 12 for examples of artifacts that have been resolved). This necessity has not eluded the field of enzymology [3, 4, 15, 16], and the present survey demonstrates the usefulness of correlating biological databases and mining the literature to enhance the value of existing research and facilitate the identification of the remaining orphan-associated genes. Until recently, there were no general repositories of orphan activity data, although some species-specific databases and pages were maintained, such as EchoBase  and a web page listing unidentified E. coli enzymes maintained by the EcoCyc project . Consequently, we updated the MetaCyc  database to identify reactions that have been analyzed by this survey, and annotated them and associated database objects with results such as the validity of their orphan status, links to their cognate protein in the case of artifacts, and the properties of the protein copurifying with the activity in the case of validated orphans. Recently, Lespinet and Labedan created ORENZA , a database dedicated to maintaining an up-to-date listing of all enzyme activities for which no sequences are available in major sequence databases . We are contributing our updated orphan information to ORENZA as well. These data, captured in MetaCyc and ORENZA, should facilitate the work of enzymologists interested in identifying the cognate genes of orphan activities. For instance, the work of Melnick et al.  is an excellent example of the combined application of modern laboratory and bioinformatics techniques that would benefit from the data described here.
Several proposals have been made recently aimed at producing a complete catalog of biochemical activities, biological functions, and their cognate genes [2, 3]. Many of these proposals recommend that such a project begin with prokaryotes because of the general ease of gene cloning from these species [1, 2]. Indeed, our data support this notion, as we find substantially more orphans with a salvageability ranking of "good" and "excellent" in prokaryotes as compared to eukaryotes. The availability of a comprehensive review of the problem achieved by this survey, combined with broad genomic sequencing and powerful computational tools, leads us to conclude that the field is in an excellent position to rectify the information gap associated with the orphan activity phenomenon.
More than one third of enzyme activities with assigned EC numbers are orphan activities, having no associated gene or protein sequence. We carried out a literature-based survey of a representative sample of presumed orphans intended to further validate and characterize these orphan activities. We have also assessed the practicability of identifying the genes associated with these orphans. In doing so, we captured data from the literature that should assist in future identification of cognate genes for the orphan activities we examined.
This survey confirmed that about 80% of sampled orphan activities have no sequence information associated with them, either in databases or in the literature. Weaknesses in database integration and failure to capture information from the literature account for most of the remaining 20%.
This survey points toward the significant scientific cost of having such a large fraction of characterized enzyme activities disconnected from sequence data. It also suggests that a larger effort, beginning with a comprehensive survey of all putative orphan activities, would resolve nearly 300 artifactual orphans and reconnect a wealth of enzyme research with modern genomics. For these reasons, we propose that a systematic effort to identify the cognate genes of orphan enzymes be undertaken.
Literature survey process
This survey was performed from June through August 2004 and relied on enzyme activities described by the NC-IUBMB. This enzyme classification and nomenclature system is hierarchical in nature and is based upon the reaction catalyzed. It assigns specific numerical identifiers, an EC number, to each distinct enzymatic activity. The first digit represents the class of reaction catalyzed (e.g., oxidoreductases are EC1; transferases are EC2). The second digit of the EC number refers to the subclass, which generally contains information about the type of compound or group involved (e.g., an enzyme acting on the CH-OH group of donors, or acting on the aldehyde or oxo group of donors). The third digit defines the sub-subclass, which specifies the nature of the reaction. The fourth digit is a serial number that is used to identify the individual enzyme within a sub-subclass (see  for a description of the classification system).
It is important to bear in mind that distinct proteins catalyzing the same reaction are assigned the same EC number. Since the EC system is based upon the reaction catalyzed, when applied to a protein it describes a biochemical function of this protein. That function can also be shared by several proteins (isozymes) that can be coded by genes in the same or different species.
Presumed orphan EC numbers were identified using the BioWarehouse database system . BioWarehouse  is an integrated database that enables cross-database queries using the structured query language (SQL). SRI's BioWarehouse instance was queried for enzymatic activities with no matching sequences in any major protein sequence databases, including TrEMBL, PIR, SWISS-PROT, CMR, ENZYME, and BioCyc (the selection of these databases is described in ). This query returned an initial list of 1,356 EC numbers that had not been retired or merged at the time of the survey.
This list was randomized and the primary literature associated with a sample of these putative orphans was processed successively according to that random order. The size of the sample necessary to ensure representational accuracy as compared to the total pool of EC numbers was calculated using Equation 1. Approximately 180 orphans are required to achieve better than 95% confidence, given the total number of EC numbers. Since a sample of 228 orphans was ultimately surveyed, the 95% level of significance was exceeded.
Equation 1: sample size estimation
SE is the standard error associated with the survey, and is derived by dividing the sampling error by 1.96, such that for a sampling error of 5% (95% confidence interval), the standard error is 0.0255102. p is the probability that the EC number is a true positive, that is, there is truly no sequence information for that EC number; this value is 0.85 based on data from a preliminary survey. N is the universe of orphan activities. Solving for n provides the sample size.
SE2 = [(p(1-p)/n)] [(N-n)/N]
A comprehensive manual analysis of the literature associated with this sample of 228 orphans drawn from the randomized list was performed as outlined in Figure 2. Various databases (Table 13) were consulted to extract the data elements listed in Table 12. For each selected putative orphan in the sample, the text search engine ExPASy Proteomics Server  was used to search TrEMBL, ENZYME, and IUBMB database records to confirm the absence of sequence data. For each orphan, all protein names, author names, reaction names, substrate names, and product names listed in the IUBMB record for that orphan were used as query arguments.
The primary literature associated with each orphan's entry in the IUBMB database [5, 26] describing the isolation and characterization of the activity was reviewed for the presence of sequence information. In particular, we were alert for the presence of molecular descriptors that might be useful in cloning the associated genes in the papers (described below), particularly Mr, pI, and details of the purification scheme (Table 14). Systematic searches of PubMed were also performed to ascertain whether publications other than those cited by IUBMB might contain relevant sequence and molecular descriptor data. A total of 331 publications (1.45 papers per orphan) were examined for additional molecular descriptors that might be useful in cloning, as described above. Data obtained from these publications were assembled into a database.
All artifactual orphans (orphans for which sequence information was found during the literature review process) were reported promptly to the Swiss Institute of Bioinformatics, the European Bioinformatics Institute, the ORENZA database, and the Nomenclature Committee of the IUBMB for the relevant database entries to be updated.
Data sources and database analyses
The initial searches for presumed orphan activities were performed using BioWarehouse version 3.0 (SRI International) running under Oracle 10G (Oracle Corporation, Redwood Shores, California). BioWarehouse is a bioinformatics data warehousing environment developed under the Bio-SPICE program .
Identification and ranking of salvageable orphans
Salvageable orphans are orphan activities for which it is likely that at least one cognate gene can be identified and confirmed in a practical manner. The extent of this salvageability was determined by ranking validated orphans according to the likelihood and practicality that at least one cognate gene can be identified, and that the gene product can be isolated and demonstrated to catalyze the enzymatic activity in a practical manner.
Orphans were ranked based on data in the original literature, combined with the availability of the complete genome sequences for the species in which an orphan was first elucidated. The principal ranking factors are (1) clear identification of the species involved and its ease of growth; (2) the availability of molecular descriptors, most importantly the molecular mass (Mr), but also the isoelectric point (pI); (3) the types of purification and analytical techniques used in the original literature; and (4) evidence that the protein can be purified with reasonable effort using current techniques, based on factors such as specific activity, purification yield, number of steps involved, and availability of substrate and of alternate purification procedures. "Excellent" and "Good" ratings indicate an activity associated with a sequenced organism, and whose purifications and assays are likely to be straightforward to replicate. "Difficult" activities are those with tricky purifications or complex assays, but a sequenced target organism or sequenced related organism. "Marginal" activities are those for which sequencing is in progress in the target organism, or a related organism. "Poor" activities are those for which no genome sequence is available, or sequencing is in progress in a related organism, and assay or purification conditions are likely to be hard to replicate.
Information about validated orphan activities has been entered into the MetaCyc database . Other data generated by our survey can be found at .
Enzyme Commission (EC):
Nomenclature Committee of the International Union of Biochemistry and Molecular Biology (NC-IUBMB), Structured Query Language (SQL)
Roberts RJ: Identifying protein function–a call for community action. PLoS Biol. 2004, 2 (3): E42-10.1371/journal.pbio.0020042.
Barrett A: Nomenclature Committee of the International Union of Biochemistry and Molecular Biology (NC-IUBMB). Enzyme Nomenclature. Recommendations 1992. Supplement 4: corrections and additions (1997). Eur J Biochem. 1997, 250 (1): 1-6. 10.1111/j.1432-1033.1997.001_1.x.
Chen X, Hyatt BA, Mucenski ML, Mason RJ, Shannon JM: Identification and characterization of a lysophosphatidylcholine acyltransferase in alveolar type II cells. Proc Natl Acad Sci USA. 2006, 103 (31): 11724-11729. 10.1073/pnas.0604946103.
Naumoff D, Xu Y, Glansdorff N, Labedan B: Retrieving sequences of enzymes experimentally characterized but erroneously annotated : the case of the putrescine carbamoyltransferase. BMC Genomics. 2004, 5 (1): 52-10.1186/1471-2164-5-52.
Melnick J, Lis E, Park J-H, Kinsland C, Mori H, Baba T, Perkins J, Schyns G, Vassieva O, Osterman A, Begley TP: Identification of the two missing bacterial genes involved in thiamine salvage: thiamine pyrophosphokinase and thiamine kinase. J Bacteriol. 2004, 186 (11): 3660-3662. 10.1128/JB.186.11.3660-3662.2004.
Nishii W, Ueki T, Miyashita R, Kojima M, Kim Y, Sasaki N, K M-M, Takahashi K: Structural and enzymatic characterization of physarolisin (formerly physaropepsin) proves that it is a unique serine-carboxyl proteinase. Biochem Biophys Res Commun. 2003, 301 (4): 1023-1029. 10.1016/S0006-291X(03)00083-4.
Cheng T, Harvey S, Chen G: Cloning and expression of a gene encoding a bacterial enzyme for decontamination of organophosphorus nerve agents and nucleotide sequence of the enzyme. Appl Environ Microbiol. 1996, 62 (5): 1636-1641.
Barrett A: Nomenclature Committee of the International Union of Biochemistry and Molecular Biology (NC-IUBMB). Enzyme Nomenclature. Recommendations 1992. Supplement 5: corrections and additions (1999). Eur J Biochem. 1999, 250 (1): 1-6.
This work was funded by grant MCB-0438571 from the U.S. National Science Foundation. The BioWarehouse is funded by contract F30602-01-C-0153 from the Defense Advanced Research Projects Agency. This material is based upon work supported by DARPA and the Air Force Research Laboratory under Contract No. F30602-01-C-0153. We gratefully acknowledge Dr. Tadhg P. Begley, Department of Chemistry and Chemical Biology, Cornell University, for help in analyzing biochemical purification protocols; Dr. Ron Caspi, Bioinformatics Research Group, SRI International, for support with the MetaCyc database and analysis of purification protocols; and Dr. Alexander Shearer, Bioinformatics Research Group, SRI International, for assistance with manuscript revision and resubmission.
Authors and Affiliations
Bioinformatics Research Group, Artificial Intelligence Center, SRI International, 333 Ravenswood Ave, Menlo Park, California, 94025-3493, USA
Yannick Pouliot & Peter D Karp
Lane Medical Library and Knowledge Management Center, Information Resources and Technology, Stanford University Medical Center, 300 Pasteur Drive, Stanford, CA, 94305-5123, USA
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.