Detection and identification of mycobacteria in clinical specimens is a key issue in the therapy of pulmonary diseases because misidentification can lead to inappropriate treatment. Traditionally, mycobacterial species are identified based on their growth rate, presence or absence of pigmentation, and using biochemical assays of the isolates recovered from specimens. The biochemical assays are time-consuming and labor-intensive, usually taking 1 to 2 months to complete, and assays for non-tuberculous mycobacteria (NTM) species can have poor reproducibility and provide ambiguous results [1,2].

By contrast, molecular identification, notably PCR-restriction enzyme analysis (PRA), is rapid and simple. The hsp65 PRA method, developed by Telenti et al. in 1993, is a popular DNA-based method for mycobacteria identification [3]. Using hsp65 PRA, Wong et al. [4] reported 100% sensitivity and specificity in identifying Mycobacterium tuberculosis complexes but only 74.5% sensitivity in identifying NTM species. This misidentification may occur because of similarities in band sizes that are critical for species discrimination [3]. An additional contributing factor is a lack of knowledge of all existing PRA profiles, especially among species that are very heterogeneous, such as M. gordonae, M. scrofulaceum, and M. terrae complexes. Recently, capillary electrophoresis (CE) with computer analysis [5-9] has provided more precise band discrimination than analysis by the naked eye.

Previously, we developed an algorithm for mycobacterial species identification from acid-fast bacillus (AFB) smear-positive BACTEC tubes by combining the rpoB duplex PRA (DPRA) [10] and the key phenotypic characters of mycobacteria recovered from the tubes [11]. Using rpoB DPRA, we differentiated Mycobacterium tuberculosis complexes (MTC) from NTM with 235 base pair (bp) and 136 bp PCR amplicons in AFB smear-positive BACTEC cultures. The 136 bp rpoB duplex PCR amplicon was further digested with MspI and HaeIII (rpoB DPRA) to divide the NTM species into eight easily distinguishable groups (A–H) as described by Kim et al. [10]. Using two phenotypic characters (growth rate and photoreactivity on pigment production) and two simple biochemical assays (nitrate reduction test and Tween 80 hydrolysis test) [11], the mycobacterial species were identified. However, the sub-culture and biochemical tests for this algorithm took three weeks.

In the present study, we developed a rapid and effective algorithm for identification of mycobacteria by combined rpoB DPRA and hsp65 PRA with CE.

Mycobacterial species identification is generally achieved by standard culture and biochemical methods [12]. Biochemical methods are labor-intensive and time-consuming, and not all are reproducible [2,13]. Recently, PRA and DPRA have been developed for molecular identification of mycobacterial species using different regions of hsp65, 16 S rDNA, 16 S-23 S rDNA spacer, dnaJ, and rpoB as an amplification target [3,14-17]. The most common method is hsp65 PRA, and 74 patterns for 40 species are available in the PRASITE database (http://app.chuv.ch/prasite/index.html).

Previous studies [18,19] have reported that hsp65 PRA is faster and more accurate for species identification than conventional (phenotypic or biochemical) testing. This is because more incorrect and ambiguous results are obtained with conventional methods. The results in our study (Tables 1 and 2) also support this finding. Incorrect and ambiguous results are caused by phenotypic homogeneity among different species and phenotypic variability within species [18]. With by hsp65 PRA, some sub-species, such as M. kansasii, can be identified and rapid-growing mycobacterium can be divided into M. abscessus and M.chelonae, M. fortuitum and M. smegmatis[20], whereas these identifications are difficult with conventional methods [21]. As found in our study (Tables 1 and 2), M. peregrinum was identified as M. fortuitum and M. avium subsp. avium and M. intracellulare were both identified as M. avium complex by the conventional biochemical method.

However, hsp65 PRA limitations have been reported in some articles [22,23]. Failure to identify or incorrect identification of the species may occur because of similarities in band sizes critical for discriminating species, including difficult to distinguish M. tuberculosis complex (M. tuberculosis and M. bovis) [22], and closely related sub-species such as M. avium or M. gordonae, because of sequence heterogeneity [22]. In addition, technical problems can also cause misinterpretation or incorrect identification [23]. Patterns in PRA profiles are complex and difficult to interpret with the naked eye, especially when more detailed sub-types are included [21].

This study combined rpoB DPRA and hsp65 PRA to test both reference strains and clinical respiratory isolates. The mycobacterial identification flow chart (Figure 1) can identify species to the sub-species level, and final species identification can be obtained instantly with concordant results from the two PRA. M. gordonae has a highly variable gene sequence with 10 sub-types in hsp65 PRA, and there are two groups (G and F) in rpoB DPRA. Most M. gordonae is in the G group, but M. gordonae types 3 and 4 by hsp65 PRA are in the F group (Tables 1 and 2).

In addition, there were different rpoB DPRA results (Table 2) for M. simaie type 5 (G group but not E group), M. scrofulaceum type 1 (D group but not H group), and M. intracellulare type 3 (F group but not G group). The identities of all of these isolates were finally confirmed by 16 S rDNA sequencing. Variable numbers of restriction sites for HaeIII in these species may derive from genetic sequence mutation. However, these species are included in the species identification algorithm even though they are uncommon isolates.

Using the mycobacteria identification flow chart (Figure 1) and algorithm (Table 3), M. avium-intracellulare complex (MAC) can be easily divided into M. avium spp. avium and M. intracellulare by both rpoB DPRA and hsp65 PRA. By contrast, this was not possible with the conventional method. Using the results in Table 3, some NTM species with identical or similar hsp65 PRA can be clearly grouped by rpoB DPRA (Table 4). Ambiguous results from hsp65 PRA alone are easier to interpret with combined rpoB DPRA and hsp65 PRA. However, M. intermedium type 1 and M. intracellulare type 3 with identical hsp65 PRA and rpoB DPRA (G group) could not be differentiated further by this species identification algorithm and required 16 S rDNA sequencing for confirmation.

Although 16 S rDNA sequencing is the standard method for mycobacterium species identification, it cannot differentiate some closely related rapid-growing mycobacterium species [24] or slow-growing M. kansasii and M. gastri that had identical 16 S rDNA sequences, but these can be differentiated by hsp65 PRA and rpoB DPRA. There are some reports [6,25] of conflicting results from different methods for mycobacterial species identification, probably caused by a failure of one method to identify all test strains correctly. Combining methods for mycobacterial species identification can improve the accuracy rate, avoid ambiguous results, and save time.

Many CE-based studies [5-9] in PCR-RFLP analysis have investigated improving band size discrimination. In one study by Chang et al. [7], high-resolution CE gave more precise estimates of DNA fragment sizes than analysis by the naked eye, and CE could detect low molecular weight fragments (down to 12 bp). In our study, restriction fragments <50 bp were not selected for analysis to avoid confusion with primer and primer dimer bands. We found that more hsp65 fragment differences than rpoB fragment (data not shown) may explain the size differences with highly variable sequence for species identification but difficult interpretation in hsp65 PRA.

Some sub-types of NTM species are relevant to clinical management, such as the M. kansasii and MAC. M. kansasii type 1 is the most common type associated with human disease [26-28] because of its high pathogenicity. However, M. kansasii types 3–7 are most often isolated from the environment and rarely from humans, and have no significant role in clinical management [26,27]. MAC can be divided into M. avium subsp. avium and M. intracellulare because drug sensitivity test and clinical outcomes are different between these two sub-types [29,30].

It is important to identify NTM to the sub-type level both for epidemiologic data and for differentiating potentially pathogenic sub-types [26,27]. Combined rpoB DPRA and hsp65 PRA with capillary electrophoresis provides precise species identification and overcomes problems associated with discrimination by hsp65 PRA band sizes. This combined method takes around 2–3 days to complete in the laboratory once clinical isolates have been received. However, the identification algorithm has some limitations. First, it could not discriminate M. intermedium type 1 from M. intracellulare type 3, and second, not every hospital laboratory will be equipped with the appropriate equipment for this method.

In conclusion, the novel flow chart and identification algorithm obtained by combined rpoB DPRA and hsp65 PRA with capillary electrophoresis can easily differentiate MTC from NTM and identify mycobacterial species to the sub-type level, which is helpful for clinical management. The results are complementary to 16 S rDNA sequencing, and the effective algorithm provides rapid and accurate mycobacterial species identification.

PRA: Polymerase chain reaction restriction-enzyme analysis; DPRA: Duplex Polymerase chain Reaction restriction-enzyme Analysis; MTC: Mycobacterium Tuberculosis Complexes; NTM: Non-Tuberculous Mycobacteria; CE: Capillary Electrophoresis; AFB: Acid-Fast Bacillus; MAC: M. Avium-intracellulare Complex; LJ: Löwenstein-Jensen.

The authors declare that they have no competing interests.

CCH wrote the manuscript. CSC, JHC, STH participated in the study design, and analysis. GHS and WCH managed the project. JYH, JJL assisted in improving the manuscript. All authors read and approved the final manuscript.