Guidelines for interpretation of 16S rRNA gene sequence-based results for identification of medically important aerobic Gram-positive bacteria

Since its first use three decades ago, 16S rRNA gene sequencing has become one of the indispensable pillars in the polyphasic approach to bacterial classification and novel bacteria discovery (Woo et al., 2008a). In the last decade, the technology has further developed beyond the research realm and matured into a clinical application for bacterial identification. Despite its increasing use in reference laboratories, one of the major limitations to its wider use is the difficulty associated with interpretation of 16S rRNA gene sequence results. The use of 16S rRNA gene sequencing for bacterial identification depends on a significant inter-species difference and a small intra-species difference in 16S rRNA gene sequences. Therefore, one of the major limitations is that when two different bacterial species share almost the same 16S rRNA gene sequence; this technique alone would not be useful for distinguishing them confidently.

Aimed at facilitating 16S rRNA gene sequence interpretation, the MicroSeq 500 16S rRNA gene bacterial identification system (MicroSeq; Perkin-Elmer Applied Biosystems) has been designed, using a 527 bp fragment in the 5' end of the 16S rRNA gene (Cloud et al., 2004; Fontana et al., 2005; Lau et al., 2006; Patel et al., 2000; Tang et al., 1998, 2000; Woo et al., 2003c). Recently, a full 16S rRNA gene sequence (full-MicroSeq) database (version 2.0) has also been generated. However, the MicroSeq databases do not take into account bacterial species sharing similar 16S rRNA gene sequences that would not be confidently identified by 16S rRNA gene sequence analysis. Furthermore, the accuracy of identification is further limited by the bacterial species included in the database, as bacterial species that are not included in the database would never be determined as the identity of an isolate.

In view of these limitations, 3 years ago we started to systematically evaluate the usefulness of full and 527 bp 16S rRNA gene sequencing and the MicroSeq databases for identification of various groups of medically important bacteria. Recently, we have published the results for medically important anaerobes (Woo et al., 2007). In this article, we describe the results of an evaluation of the usefulness of full and 527 bp 16S rRNA gene sequencing and the MicroSeq databases for identification of 409 medically important aerobic Gram-positive bacteria. Practical points when using such technologies for identification of these bacteria are also discussed.

16S rRNA gene sequences of medically important aerobic Gram-positive bacteria. The species of medically important aerobic Gram-positive bacteria included in this study include all the 409 species of the 59 genera of medically important aerobic Gram-positive cocci (GPC) and Gram-positive rods (GPR) listed in the most recent edition of the Manual of Clinical Microbiology (Murray et al., 2007). For each bacterial species, a list of the 16S rRNA gene sequence was retrieved from the GenBank database. In the list, the most representative 16S rRNA gene sequence for each species was chosen for analysis according to the following criteria: (a) strains with good phenotypic characterization, such as the type strains, were preferred to those without good phenotypic characterization; (b) strains isolated from humans were preferred to those isolated from animals or the environment; (c) sequences with fewer undetermined bases were preferred to those with more undetermined bases; and (d) longer sequences, especially those with better coverage of the 5' end, were preferred to shorter sequences and those with poorer coverage of the 5' end.

Comparison of full 16S rRNA gene sequences of medically important aerobic Gram-positive bacteria. The percentage differences of the 16S rRNA gene sequences among the different species in the same group/genus of medically important GPC and GPR were determined by pairwise alignment using CLUSTAL W (Thompson et al., 1994). For sequences with undetermined bases, other 16S rRNA gene sequences of the same species were retrieved and the undetermined bases were manually amended. If there was no other 16S rRNA gene sequence for the same species, the positions of the undetermined bases were deleted in the analysis.

Comparison of 527 bp 16S rRNA gene sequences of medically important aerobic Gram-positive bacteria. The 527 bp sequences that should be amplified by the primers of MicroSeq were extracted from the full 16S rRNA gene sequences. The percentage differences of the resultant partial 16S rRNA gene sequences among the different species in the same group/genus of medically important GPC and GPR were determined by pairwise alignment using CLUSTAL W (Thompson et al., 1994).

Definitions. A bacterium species is defined as confidently identified by 16S rRNA gene sequencing if there is >3 % difference between the 16S rRNA gene sequence of the species and those of other medically important bacteria species. A bacterium species is defined as not confidently identified by 16S rRNA gene sequencing if there is <2 % difference between the 16S rRNA gene sequence of the species and that of one or more medically important aerobic Gram-positive bacterium species. A bacterium species is defined as only doubtfully identified by 16S rRNA gene sequencing if there is 2–3 % difference between the 16S rRNA gene sequence of the species and that of one or more medically important aerobic Gram-positive bacterium species.

The percentage differences of the full and 527 bp 16S rRNA gene sequences among the different groups of medically important aerobic Gram-positive bacteria are shown in Supplementary Table S1(a–k) (available with the online journal).

Full 16S rRNA gene sequencing for identification of medically important aerobic Gram-positive bacteria
None and 1 (2 %) of 46 staphylococci, 8 (20 %) and 9 (22 %) of 41 streptococci, none and 5 (17 %) of 30 enterococci, 32 (65 %) and 7 (14 %) of 49 GPC other than staphylococci, streptococci and enterococci, 2 (3 %) and 7 (9 %) of 74 mycobacteria, 9 (24 %) and 12 (32 %) of 38 corynebacteria, none and 1 (3 %) of 31 nocardia, 16 (44 %) and 4 (11 %) of 36 members of Bacillus and related taxa (Paenibacillus, Brevibacillus, Geobacillus and Virgibacillus), and 24 (38 %) and 11 (17 %) of 64 GPR other than mycobacteria, corynebacteria, nocardia and members of Bacillus and related taxa can be confidently and doubtfully identified by full 16S rRNA gene sequencing, respectively (Table 1, Supplementary Table S2a–k, available with the online journal).

Table 1. Number and percentage of species of major groups of medically important Gram-positive bacteria confidently identified by full 16S rRNA gene sequence, 527 bp 16S rRNA gene sequence and the MicroSeq databases

Full-MicroSeq database for identification of medically important aerobic Gram-positive bacteria
None and 1 (2 %) of 46 staphylococci, 7 (17 %) and 8 (20 %) of 41 streptococci, none and 4 (13 %) of 30 enterococci, 18 (37 %) and 4 (8 %) of 49 GPC other than staphylococci, streptococci and enterococci, 2 (3 %) and 5 (7 %) of 74 mycobacteria, 3 (8 %) and 8 (21 %) of 38 corynebacteria, none of 31 nocardia, 13 (36 %) and 4 (11 %) of 36 members of Bacillus and related taxa (Paenibacillus, Brevibacillus, Geobacillus and Virgibacillus), and 15 (23 %) and 5 (8 %) of 64 GPR other than mycobacteria, corynebacteria, nocardia and members of Bacillus and related taxa can be confidently and doubtfully identified by the full-MicroSeq database, respectively (Table 1, Supplementary Table S2a–k, available with the online journal).

527 bp 16S rRNA gene sequencing for identification of medically important aerobic Gram-positive bacteria
One (2 %) and 6 (13 %) of 46 staphylococci, 19 (46 %) and 5 (12 %) of 41 streptococci, 8 (27 %) and 1 (3 %) of 30 enterococci, 38 (78 %) and 4 (8 %) of 49 GPC other than staphylococci, streptococci and enterococci, 9 (12 %) and 15 (20 %) of 74 mycobacteria, 15 (39 %) and 5 (13 %) of 38 corynebacteria, none and 3 (10 %) of 31 nocardia, 20 (56 %) and 5 (14 %) of 36 members of Bacillus and related taxa (Paenibacillus, Brevibacillus, Geobacillus and Virgibacillus), and 38 (59 %) and 6 (9 %) of 64 GPR other than mycobacteria, corynebacteria, nocardia and members of Bacillus and related taxa can be confidently and doubtfully identified by 527 bp 16S rRNA gene sequencing, respectively (Table 1, Supplementary Table S2 a–k, available with the online journal).

500-MicroSeq database for identification of medically important aerobic Gram-positive bacteria
One (2 %) and 5 (11 %) of 46 staphylococci, 18 (44 %) and 3 (7 %) of 41 streptococci, 7 (23 %) and none of 30 enterococci, 31 (63 %) and 3 (6 %) of 49 GPC other than staphylococci, streptococci and enterococci, 8 (11 %) and 11 (15 %) of 74 mycobacteria, 11 (29 %) and 3 (8 %) of 38 corynebacteria, none and 2 (6 %) of 31 nocardia, 17 (47 %) and 5 (14 %) of 36 members of Bacillus and related taxa (Paenibacillus, Brevibacillus, Geobacillus and Virgibacillus), and 25 (39 %) and 4 (6 %) of 64 GPR other than mycobacteria, corynebacteria, nocardia and members of Bacillus and related taxa can be confidently and doubtfully identified by the 500-MicroSeq database, respectively (Table 1, Supplementary Table S2 a–k available with the online journal).

The major difficulties and controversies in interpreting 16S rRNA gene sequence data relate to the assignment of bacterial species based on similarity search results, as no threshold values are available as for DNA–DNA hybridization (Petti, 2007). As different bacterial species are likely to evolve at different rates, there has been no agreement on a universal cut-off for bacterial species delineation. In 1994, Stackebrandt and Goebel proposed a 97 % similarity level for bacterial speciation using 16S rRNA gene sequence (Stackebrandt & Goebel, 1994). Although in the study by Drancourt et al. (2000) a >99 and >97 % sequence similarity was used as the cut-off for species and genus identification, respectively, the authors indicated that these sharp values were mainly made for practical purposes for interpretation of their large sequence dataset. On the other hand, Janda and Abbott suggested that a minimum of >99 %, and ideally >99.5 %, sequence similarity be used for species identification in their recommended guidelines (Janda & Abbott, 2007). Although different studies have identified groups of bacteria for which 16S rRNA gene sequences are not discriminative enough, no study has attempted to solve this problem in a systematic way. Moreover, the use of public databases such as GenBank can be problematic in that errors in sequence data are not uncommon. For example, the strain with GenBank accession number Z26896 was claimed to be Staphylococcus haemolyticus. However, the sequence has highest nucleotide identities to sequences of 16S rRNA genes of Staphylococcus aureus, rather than other strains of Staphylococcus haemolyticus, and no results of phenotypic characterization were given for this strain. Therefore, the true identity of this strain is difficult to ascertain. In the present study and a previous one (Woo et al., 2007), we used >3, 2–3 and <2 % differences between the most representative 16S rRNA gene sequence of species obtained from GenBank and those of other medically important bacteria to indicate the degree of confidence when using 16S rRNA sequence for identification of medically important bacterial species. When there is 2–3 % difference between the 16S rRNA gene sequence of a bacterial species and a closely related one, the degree of 16S rRNA gene sequence homogeneity in different strains of that species becomes very important in determining whether 16S rRNA gene sequencing is able to identify the bacterium to the species level. More importantly, medically important bacterial species with similar 16S rRNA gene sequences were also identified (Supplementary Table S2 a–k available with the online journal), so that the users can easily note the species that may be potentially confused based on 16S rRNA gene sequence analysis. Further phenotypic tests and/or sequencing of other gene loci can be performed to differentiate these species.

For species identification, 16S rRNA gene sequence results of aerobic Gram-positive bacteria should be interpreted cautiously with basic phenotypic tests results, and additional biochemical tests or sequencing of additional gene loci are often required for definitive identification. Overall, using 3 % base difference as the cut-off, full and 527 bp 16S rRNA gene sequencing can only identify 24 and 40 % of medically important GPC and 21 and 34 % of medically important GPR confidently to the species level. This is apparently different from the results reported in some other studies, in which up to 83 and 92 % of the isolates can be assigned to species level (Mignard & Flandrois, 2006; Bosshard et al., 2006). In some studies, the difference was due to the use of different cut-offs for confident identification; whereas in others, the difference was a result of a larger proportion of bacterial strains, other than Gram-positive bacteria, that can be confidently identified. The higher percentage of bacteria that can be confidently identified by 527 bp 16S rRNA gene sequencing than full 16S rRNA gene sequencing is a result of the non-random distribution of base difference along the 16S rRNA gene. For some genera, such as staphylococci, mycobacteria and nocardia, both full and 527 bp 16S rRNA gene sequencing can identify <13 % of bacteria confidently. Therefore, in order to prevent misinterpretation of 16S rRNA gene sequence results, they should always be interpreted in the context of basic phenotypic test results. Moreover, key phenotypic tests, that can help distinguish between the closest match by 16S rRNA gene sequencing and medically important bacterial species with similar 16S rRNA gene sequences as listed in Supplementary Table S2 (a–k) (available with the online journal), should be performed for differentiation of these possibilities (Woo et al., 2001, 2003d; Lau et al., 2002; Cloud et al., 2005). For example, in our previous study on Tsukamurella conjunctivitis, 16S rRNA gene sequencing was able to identify the various isolates as Tsukamurella to the genus level, but all Tsukamurella species possess >99 % nucleotide identities with each other. Phenotypic tests have to be performed to distinguish among the various Tsukamurella species (Woo et al., 2003d). Alternatively, sequencing of other gene loci can be performed for identification of species in these genera. In fact, another gene locus, groEL, has been shown to be better than 16S rRNA gene for identification of staphylococci and mycobacteria (Kwok et al., 1999; Goh et al., 1996; Ringuet et al., 1999). Other genes, such as tuf, dnaJ and rpoB, and secA1 have also been used for identification of staphylococci, mycobacteria and nocardia, respectively (Heikens et al., 2005; Yamada-Noda et al., 2007; Adékambi et al., 2003; Conville et al., 2006). This conclusion of the necessity for support of 16S rRNA gene sequence results by other identification methodologies, such as phenotypic tests and/or sequencing other gene targets, is in line with the recommendations given by the Clinical and Laboratory Standards Institute, although different cut-offs were used in their recommendations (CLSI, 2008). One has to be particularly cautious if 16S rRNA gene sequencing is used for culture-negative infections, such as infective endocarditis, as no isolate is available for further phenotypic testing (Gauduchon et al., 2003; Kotilainen et al., 2006). Furthermore, some medically important Gram-positive bacteria, such as Streptococcus sinensis, have been reported as a cause of infective endocarditis in Asia and Europe with the oral cavity documented to be the reservoir, but are still not yet included in the Manual of Clinical Microbiology (Woo et al., 2002, 2004b, 2008b; Uçkay et al., 2007; Faibis et al., 2008). One must bear in mind that these bacteria not yet included in the Manual of Clinical Microbiology may also be the identity of an isolate. Particular attention has to be paid if there is discrepancy between 16S rRNA gene sequence and phenotypic test results.

Despite its relatively low confidence for species identification, 16S rRNA gene sequencing is very useful for identification of medically important Gram-positive bacteria to the genus level, especially for some groups of Gram-positive bacteria notoriously difficult to identify by phenotypic methods. Although some genera of Gram-positive bacteria, such as Mycobacterium and Nocardia, possess unique phenotypic characteristics, such as acid-fastness, for their easy identification to the genus level, most streptococcus-like Gram-positive cocci, such as Granulicatella, Abiotrophia, Gemella and Helcococcus, are particularly difficult for genus identification using phenotypic tests. In our experience, most commercial systems failed to identify most of our clinical isolates of Granulicatella, Abiotrophia, Gemella and Helcococcus to the genus level (Woo et al., 2003a, b, 2005). Gram-positive bacterial isolates that possess unusual phenotypic profiles also led to the same problem (Woo et al., 2003e, 2004a). In these cases, 16S rRNA gene sequencing would be extremely useful for genus identification and narrowing down the identification to a few possible species that possess similar 16S rRNA gene sequences.

When using the MicroSeq databases for bacterial identification, additional attention has to be paid to possible medically important bacterial species that are not included in the database, and the accuracies reported for MicroSeq for various groups of Gram-positive bacteria might have been overestimated. The percentage of Gram-positive bacteria estimated to be confidently identified by the MicroSeq databases in this study is low compared to the accuracies reported for selected groups of Gram-positive bacteria in the literature (Patel et al., 2000; Tang et al., 2000; Mellmann et al., 2003). This is mainly because the MicroSeq databases do not take into account bacteria with similar 16S rRNA gene sequences, which can confuse the identity of the bacterium. For example, in two studies that assessed the usefulness of 500-MicroSeq for identification of mycobacteria and nocardia, clinical isolates of Mycobacterium chelonae, Mycobacterium tuberculosis, Mycobacterium marinum, Nocardia asteroides, Nocardia nova and Nocardia farcinica were considered successfully identified by 500-MicroSeq (Patel et al., 2000; Mellmann et al., 2003), but the present study indicated that in fact there are Mycobacterium and Nocardia species with similar 16S rRNA gene sequences. In fact, it is well known that the 16S rRNA gene is probably not the best target for identifying mycobacteria and nocardia, and alternative gene targets, such as dnaJ and secA1, as mentioned above, have been used for their identification (Yamada-Noda et al., 2007; Conville et al., 2006). Therefore, in the present study and our previous one on anaerobes (Woo et al., 2007), supplementary note is given to indicate those bacteria that cannot be confidently identified but are included in the MicroSeq databases, symbolized by ?^c and ?^f in Supplementary Table S2 (a–k) (available with the online journal). In addition to this problem, the MicroSeq databases are also defective in that they do not contain a significant proportion of medically important species. To make our results more user-friendly, supplementary data is also included to indicate whether the inability of the MicroSeq databases to identify a particular bacterium is due to the species not being included in the MicroSeq databases, symbolized by x^a and x^d in Supplementary Table S2 (a–k) (available with the online journal), or high 16S rRNA gene sequence similarity (>98 % nucleotide identity) to a closely related species, symbolized by x^b and x^e in Supplementary Table S2 (a–k) (available with the online journal). To improve the usefulness of the MicroSeq databases, bacterial species that can be confidently identified by 16S rRNA gene sequencing but are not found in the MicroSeq databases should be included (Table 2). These include 15 medically important GPC and 18 medically important GPR that should be confidently identified by full 16S rRNA gene sequencing but are not included in the full-MicroSeq database, and 9 medically important GPC and 21 medically important GPR that should be confidently identified by 527 bp 16S rRNA gene sequencing but are not included in the 500-MicroSeq database (Table 2).

Table 2. Medically important Gram-positive bacteria that can be confidently identified by full or 527 bp 16S rRNA gene sequencing but are not included in the MicroSeq databases

This work is partly supported by the Committee of Research and Conference Grants and University Development Fund, University of Hong Kong.

Footnotes

^,†, Jade L. L. Teng³ †These authors contributed equally to this work.

Tables of sequence comparison and identification data are available as supplementary material with the online version of this paper.