A Systematic Approach for Discovering Novel, Clinically Relevant Bacteria

We identified 95 isolates from novel taxa that may have clinical relevance.

B road-range PCR amplifi cation and sequencing of the 16S rRNA gene (16S sequencing) is not only widely used as a taxonomic tool but is recognized as an effective reference method for bacterial identifi cation. It has been used to identify novel and emerging pathogens (1)(2)(3)(4) and to defi ne complex microbial communities (5,6). The method has also revolutionized our understanding of microbial diversity (7)(8)(9). In clinical microbiology laboratories, 16S sequencing is useful for classifying microorganisms from pure culture (10,11). Molecular identifi cation is especially valuable for bacteria that are slow growing, biochemically inert or variable, and fastidious, and it has also enhanced our understanding of previously unrecognized, often opportunistic pathogens (1,10,12).
Sequence-based identifi cation relies on limited, yet phylogenetically informative, 16S sequence variation between related bacterial taxa. The entire 16S rRNA gene is ≈1,500 nt long (11); however, sequencing the 5′ third (partial 16S) generally provides suffi cient taxonomic information while limiting costs (10). Partial 16S sequences are compared with reference libraries to determine the species with maximum similarity (10,11). The largest library is the nucleotide database hosted by the National Center for Biotechnology Information (NCBI) (13). Depending on their similarity to reference sequences, unknown isolates can be identifi ed to different taxonomic levels by using interpretive guidelines published by the Clinical and Laboratory Standards Institute (CLSI) (14). For most taxa, sequence identity >99% with a valid reference sequence is required for species-level identifi cation. Although this cutoff is widely used to identify isolates of the same species, a uniform cutoff for defi ning isolates as belonging to separate species is more controversial (1,10,(15)(16)(17). Values of 99.5% to 97.0% have been proposed in the past (12,15,(17)(18)(19)(20)(21)(22), with more recent evidence and recommendations supporting values between 98.7% and 99.0% (10,17,23).
In our laboratory, as in many others, 16S sequencing is performed when morphologic and phenotypic identifi cation is inconclusive or diffi cult or when it is specifi cally requested. By using CLSI guidelines and an NCBI nucleotide-based reference library (24), >90% of these isolates can be identifi ed to the species level. However, clinical isolates belonging to as-yet-undescribed taxa are regularly encountered. Whether they represent emerging pathogens (1) or environmental contaminants is often diffi cult to determine in individual cases. Therefore, we conducted a systematic analysis of large numbers of unidentifi ed strains to screen for novel taxa of potential clinical relevance. We reviewed partial 16S sequences from >26,000 clinical isolates to identify and characterize novel species with possible clinical signifi cance. We identifi ed 673 isolates that belong to as-yet-undescribed species, including 348 isolates of 95 novel taxa that were isolated from multiple patients. Repeated isolation of these undescribed organisms may indicate their clinical relevance and warrant their formal description as species.

Clinical Isolates
From results reported for ≈26,000 clinical isolates identifi ed by 16S rRNA gene sequencing during February 2006-June 2010, we searched for those isolates that could not be identifi ed to the species level by using SmartGene software (24) and CLSI guidelines (14). Phenotypic characteristics were routinely compared with those expected for closely related taxa. Species-level identifi cation might have been unsuccessful for several reasons, including lack of separation between closely related species (which resulted in a report of >1 species), poor sequence quality on multiple attempts, insertions or deletions in multiple nonidentical copies of the 16S rRNA gene (which compromised sequence quality, length, or both), unpublished or unsubstantiated references, or a lack of similar sequences in reference databases. After multiple isolates recovered from the same patients were eliminated, 1,678 (≈6%) isolates were found that had not been identifi ed to the species level. A cutoff of <99% identity with a known species was used to defi ne isolates that may represent novel taxa (17,23). On the basis of provided information, anatomical sites were classifi ed as follows: blood, bones (including bone marrow), central nervous system (brain, cerebrospinal fl uid), eye, gastrointestinal tract (abdomen, gallbladder, stool), genitourinary tract (genitals, placenta, urine), oral cavity/paranasal sinus (including throat), respiratory tract (invasive: bronchoalveolar lavage, bronchial brush/wash, lung; other: sputum, endotracheal aspirate, respiratory specimen), tissue, wound/abscess (including bite wounds, lesion, scraping), other (aspirate, biopsy, body and dialysis fl uids, ear, heart valve, medical devices), or unknown.

Sequence Assembly
Partial 16S rRNA gene sequencing had been performed as reported (25). Original chromatogram fi les were reanalyzed with MicroSeq 500 software (version 2.0; Applied Biosystems, Foster City, CA, USA). Consensus sequences of <400 bp in length were eliminated from further analyses. Remaining sequences with average phred quality scores >35 were included without manual review. Sequences with quality scores <35 were reviewed manually and included only if quality was suffi cient, as determined by visual inspection. Sequences were converted to FASTA format (http://blast.ncbi.nlm.nih.gov/blastcgihelp.shtml) for comparison with reference sequences and submitted to GenBank under accession nos. JQ259197-JQ259857X and JN986812-JN986825. Sequences were annotated with taxonomic information from the best match with specieslevel identifi cation by using CLSI guidelines (14). In brief, isolates with 97% to <99% identity were annotated at the genus level, isolates with 95% to <97% identity were annotated at the family level, and isolates with <95% identity were annotated at the order level. Aerobic actinomycetes (26), members of the family Enterobacteriaceae, and mycobacteria with identities of 95%-99% were annotated at the family level (14).

Comparison to Reference Sequences
NCBI stand-alone-BLASTn version 2.2.23+ with default parameters and internally developed software applications were used to compare sequences to a local copy of the NCBI nucleotide database (13) (downloaded July 2010). Information from 3 matches per isolate was parsed from XML-formatted BLASTn output fi les into a database by using custom python code and biopython libraries (27): 1) top match with valid species-level annotation (e.g., Streptococcus sanguinis); 2) top match with valid genus-level annotation (e.g., Streptococcus sp. oral strain T4-E3); and 3) top BLASTn match irrespective of annotation (e.g., uncultured bacterium). Valid nomenclature was determined by comparing annotations in the GenBank organism fi eld to a list of approved bacterial taxa (28). Values in the following GenBank database fi elds or BLAST XML results were retrieved from each of the 3 matches: organism, taxonomy, associated publication, publication date, alignment length, number of identities, and position in the hit list. Reference sequences with species-level annotation were used, whether they were linked to a publication or not. For each of the 3 matches, the number of ambiguous bases (International Union of Pure and Applied Chemistry codes) and the percent aligned (alignment length as percentage of query length) were calculated. Percent identity was calculated by considering International Union of Pure and Applied Chemistry ambiguity codes as matching any corresponding bases (e.g., Y matched C or T). N symbols were always recorded as mismatches.
Only sequences that had <99% identity with a valid species-level reference were included in subsequent analyses. Since BLASTn uses a local alignment algorithm, resulting alignments may be based on truncated query or match sequences if similarities are low at either end of the sequences. This practice may cause infl ated pairwise sequence identity values. To control for this effect, we also retrieved the 3 matches described above using a minimum alignment length cutoff of 98%, on the basis of the query sequence length. Manual reviews were performed when this fi lter resulted in different best matches. For sequences with percent identity values close to the 99% cutoff and BLASTn alignment length of <100%, pairwise alignments with the best species-level match were analyzed by using MEGA4.1 (29). Percent identity was calculated manually for these isolates on the basis of a full-length alignment of query and match sequences.

Phylogenetic Analysis to Determine Repeatedly Encountered Taxa
Isolates that likely belonged to the same undescribed species were recognized by constructing phylogenetic trees with related isolates in MEGA. Groups of isolates with high sequence identity were specifi ed from phylogenetic trees, and percent identity was calculated from multiple sequence alignments by using MEGA. Isolates that shared >99.0% sequence identity with each other were considered part of the same cluster. For all clusters containing >5 isolates, BLASTn matches were manually reviewed. Phylogenetic trees were constructed by using sequences from clinical isolates in the same cluster and related type strains as identifi ed by the The All-Species Living Tree Project (release 102) (30) and/or List of Prokaryotic Names with Standing in Nomenclature (31).

Clinical Study Isolates
During a 4-year period, 1,678 clinical isolates (≈6%) were not identifi ed to the species level by routine 16S sequence analysis. Reanalysis of these sequences showed that 315 isolates (19%) were unidentifi ed because of inadequate sequence quality; they were excluded from this study. The remaining 1,363 sequences were re-screened by using a current NCBI nucleotide database, and 690 (50.6%) were found to share >99% identity with >1 species-level annotated GenBank reference. The remaining 673 isolates were marked as probable novel taxa and included in this study. Of these 673 isolates, 52 (7.7%) were obtained at the University of Utah Medical Center, and the remaining isolates were referred from hospitals in 41 different US states. Nearly half of the isolates (47.3%) originated from blood cultures. Anatomical sources of the isolates are shown in Figure 1.

Sequence Length and Quality
Most sequences (84%) for the 673 isolates had lengths of 460 to 500 bp, as expected on the basis of the PCR and sequencing primers used (Figure 2, panel A). The median sequence phred quality score for the isolates suspected of representing novel taxa was 45, indicating high-quality sequences ( Figure 2, panel B). One to 18 ambiguous nucleotide positions were observed in 38% of isolates ( Figure 2, panel C), indicating multiple nonidentical copies of the 16S rRNA gene.

Taxonomic Analysis of Novel Taxa Represented by Multiple Clinical Isolates
Overall, 348 isolates (52%) belonged to 95 novel taxa represented by >1 isolate. Cluster sizes ranged from 2 to 15, Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 18  and sequence identities to species-level references ranged from 86.5% to 98.9% (Table 2; online Appendix Table  2, wwwnc.cdc.gov/EID/article/18/3/11-1481-TA2.htm). Clusters within the order Flavobacteriales showed the greatest divergence from known species, with only 92.9% average identity. Not surprisingly, given the preponderance of isolates in this order, the largest number of clusters (n = 45) was identifi ed among the Actinomycetales (online Appendix Table 1). Fourteen clusters with up to 9 members were seen in the family Nocardiaceae, 12 clusters with up to 12 members in Actinomycetaceae, and 9 clusters with up to 10 members in Corynebacteriaceae. Nineteen novel taxa were represented by >5 isolates ( Table 2). Upon manual review, 12 were confi rmed without changes, 2 clusters contained at least 1 isolate with >1% sequence difference in pairwise comparisons, 2 clusters were split because of >1% sequence heterogeneity, and isolates of 3 clusters could be identifi ed to validly described species: Rothia aeria, Cardiobacterium hominis, and Streptomyces thermoviolaceus subsp. thermoviolaceus. One cluster of 12 isolates belonged to a novel genus and species, Kroppenstedtia eburnea, which was described subsequent to our initial analysis (32).

Anatomical Source of Unidentifi ed Isolates
In addition to the frequency with which isolates of novel taxa are encountered in clinical specimens, their importance may also be judged by their anatomical source. Isolates cultured from the following normally sterile sites were considered clinically relevant: cerebrospinal fl uid, pericardial fl uid, synovial fl uid, and tissues (brain, heart valve, or biopsy tissues). A total of 32 isolates were identifi ed from these key sites. A manual analysis showed 3 isolates that were not identifi ed because of short reference sequences and 1 isolate that was subsequently identifi ed as K. eburnea. Of the remaining 28 isolates, 17 (61%) belonged to taxa that were repeatedly encountered. Taxonomic information for all 32 isolates is summarized in Table 3.

Discussion
Broad-range molecular identifi cation methods have facilitated the discovery of novel bacterial species and have resulted in a rapid increase in recognized bacterial taxa (28). The use of these methods in diagnostic laboratories may lead to the detection of bacterial strains that belong to novel species. We reviewed 16S sequencing results for >26,000 clinical isolates in a systematic approach to recognize novel species that may be pathogenic. Their formal description will provide the basis for improvements of sequence databases, antimicrobial susceptibility studies, and epidemiologic surveys to characterize their pathogenicity.
A sequence identity cutoff of <98.7%-99.0% for species discrimination has been shown to correlate with DNA-DNA hybridization results and is recommended for taxonomic purposes (17,23). In this study, 673 isolates showed <99% sequence identity and 535 isolates showed <98.7% sequence identity to any reference sequence with species-level annotation in the NCBI nucleotide database and could thus be considered novel taxa. Comparison of these sequences against the NCBI nucleotide database, the largest reference sequence repository (10,11), which contains 16S sequences for all newly described bacterial species, ensured a robust analysis of possibly novel species. Our algorithm employed 2 quality assurance criteria for reference sequences identifi ed in BLASTn analysis: minimal alignment length of 98% and annotation as a validly described bacterial taxon (28). Because a more stringent manual review of reference sequences, as performed in diagnostic practice (14), was not feasible for this large study, the 673 isolates detected by this algorithm represent a conservative estimate of the total number of novel species encountered.
To ensure that sequence quality was not limiting, we confi rmed that sequences were of expected length ( Figure  2, panel A) and had phred scores showing a median accuracy of >99.99% per base (Figure 2, panel B). It has been recommended that sequences used for bacterial identifi cation should contain <1% ambiguous positions (19), which was the case in 92% of the sequences in our study (Figure 2, panel C). However, ambiguous positions can be seen in bacteria with multiple, nonidentical 16S alleles. We observed up to 18 ambiguous positions in a small number of isolates ( Figure 2, panel C) consistent with whole-genome sequencing data that indicate >19 nucleotide differences in bacteria with multiple rRNA operons (33,34). Although full-length 16S sequencing might have facilitated the identifi cation of some isolates, partial 16S sequencing is considered robust (10) and is an unlikely reason for incomplete identifi cation in most cases.
To determine taxonomic properties of all 673 isolates, we calculated 16S sequence identities to reference sequences with valid species-level (Figure 3, panel A), genus-level ( Figure 3, panel B), or any annotation ( Figure 3, panel C). Consistent with results of previous smaller studies, our results showed that most isolates were gram-positive rods and nonfermenting gram-negative rods (Table 1) (22,35). A total of 294 isolates belonged to the order Actinomycetales, with Actinomyces (n = 71), Corynebacterium (n = 59), and Nocardia (n = 52) being the most common genera. Molecular identifi cation methods have resulted in a dramatic increase in the number of recognized species in these genera, and our results indicate that more species of possible clinical relevance are yet to be described (28). A total of 535 (79.5%) and 225 isolates (33.4%) belonged to novel species even when more conservative cutoffs of 98.7% and 97% identity, respectively, were used (15,23). Of these, 111 isolates (16.5%) represented novel genera at the conservative 95% identity cutoff (10,21).
To determine the isolates most likely to be of clinical importance, we identifi ed novel taxa that were isolated repeatedly or were from normally sterile, clinically relevant anatomical sites. More than half of the unidentifi ed organisms were isolated at least twice, forming clusters that represented 95 novel taxa. Most clusters belonged to the order Actinomycetales (45 clusters, 176 isolates), with 14 clusters (42 isolates) in the genus Nocardia and 12 clusters (52 isolates) in the genus Actinomyces. A total of 19 clusters that contained >5 members were initially identifi ed (total of 156 isolates, Table 2). After manual review, isolates in 2 of these clusters were found to belong to validly described species (Table 2). These species were not identifi ed in the automated analysis due to short reference sequences or because they had a subspecies annotation not covered in the algorithm. The validity of our approach was confi rmed, however, when a novel thermoactinomycete, Kroppenstedtia eburnea (32), was formally described during preparation of this article. The 16S sequence of this organism showed ≈99.5% identity to a large cluster of 12 isolates in our study (Table 2).
While this study only included bacterial strains from clinical specimens (Figure 1), isolates from some anatomical sites (e.g., central nervous system) may be more likely to represent pathogens than others (e.g., upper respiratory tract). When highly stringent criteria are used (e.g., recovery from a normally sterile fl uid or tissue), a minimum of 28 isolates may represent novel pathogens ( Table 3). The presence of multiple isolates for 17 of these novel species further supports their status as potential pathogens. While proving pathogenicity is beyond the scope of this study, our analysis may serve as a sentinel for novel organisms with pathogenic potential and provide a rationale for further studies to defi ne their pathogenicity. During 2001-2007, a total of 215 novel bacterial species and 29 novel genera isolated from clinical samples were formally described (1). Only 100 of these new species were represented by at least 4 isolates, of which Mycobacterium and Nocardia were the most common genera. In contrast to our study, most new species were isolated from nonsterile body sites, such as the oral cavity and gastrointestinal tract, and may thus be commensal or from the environment. Using a proposed minimum of 3 to 5 isolates to describe novel bacterial species (10,36,37), the present study may include up to 46 novel species (<99% identity) and up to 4 novel genera (<95% identity). Alternatively, it has been argued that even a single isolate from a human specimen should be reported to allow for more rapid identifi cation of additional isolates in other laboratories (1,12,22). By this strategy, several hundred novel taxa may be represented in this study. Although our study does not prove that these isolates represent novel species, it provides a framework for screening large numbers of sequences for possible novel taxa that may be of clinical importance. Candidate isolates will require rigorous polyphasic validation, including full 16S rRNA gene sequencing, to confi rm that they are new bacterial species. By providing information on morphologic characteristics, antimicrobial drug susceptibility profi les, virulence factors, and spectrum of disease, future studies will facilitate clinical decision making. Results of our phylogenetic analysis may thus help focus efforts to formally describe novel, clinically relevant species and to improve the diagnostic utility of reference databases. *GNR, gram-negative rods; GPR, gram-positive rods; CSF, cerebrospinal fluid; GPC, gram-positive cocci; GVR, gram-variable rods; GNC, gram-negative cocci; GNCB, gram-negative coccobacilli; Y, isolates belonging to tentative novel taxa represented multiple times in this study. †Results of manual review of BLASTn analysis (23). ‡These pairs of isolates belong to same 3 respective clusters. and a medical director at ARUP Laboratories in Salt Lake City. His research interest includes molecular methods for the diagnosis of infectious diseases and pathogen discovery.