Excluding rrs gene and ITS (non-functional RNA elements and structural ribosomal RNAs), and according to our strategy of genome comparison (Figure 1) most of the genes commonly used for mycobacterial species identification (gyrA, gyrB, hsp65, recA, rpoB, sodA, groEL1, groEL2) code for proteins which present similar conformations in non-mycobacterial studied genomes (Additional file 1). Indeed, protein similarity levels of these genes, in comparison with M. tuberculosis H37Rv genome, were higher than 80% for the other 15 mycobacterial genomes studied (96 ± 2% for gyrA, 94 ± 5% for gyrB, 79 ± 5% for groEL1, 93 ± 4% for groEL2 which is an alternative gene name for hsp65, 99 ± 1% for recA, 96 ± 2% for rpoB, 81 ± 33% for sodA), and also for the 12 non-mycobacterial genomes studied (86 ± 5% for gyrA, 85 ± 5% for gyrB, 89 ± 3% for groEL1, 96 ± 2% for groEL2, 94 ± 3% for recA, 88 ± 4% for rpoB, 69 ± 22% for sodA).

Among the 3989 predicted proteins of M. tuberculosis H37Rv genome (Figure 2A and Additional file 1), about 54.6% (i.e. 2177 proteins) presented protein similarities above 50% with the other studied mycobacterial genomes (n = 15), and only 6.8% of these hypothetical conserved mycobacterial proteins (150 proteins: 150 number in the top of a bar in Figure 2B) displayed similarities less than 50% with the studied non-mycobacterial genomes (n = 12). Consequently, almost half of the M. tuberculosis H37Rv predicted proteins are potentially present in the 12 studied genomes of CNM group members. We chose to decrease the number of candidate proteins by restricting the panel of studied proteins to those exclusively conserved in the mycobacterial genomes, focusing on M. tuberculosis H37Rv proteins with similarity levels between 80% and 100% in comparison with other mycobacterial genomes (n = 15), and less than 50% similarity levels in comparison with genomes (n = 12) of the other CNM group genera. As a result, among the 3989 predicted proteins of M. tuberculosis H37Rv genome (Figure 2A), we selected 11 proteins (11 number in the top of a bar in Figure 2B). Among the 3989 predicted proteins of M. tuberculosis H37Rv proteins (Additional file 1), the selected candidate proteins (Table 1), were the subunits C (locus Rv1305) and A (locus Rv1304) of the ATP synthase, the cyclopropane mycolic acid synthase (CMAS) coded by the cmaA1 gene in M. tuberculosis H37Rv (locus Rv3392c), hypothetical PE or PPE family proteins (loci Rv0285 and Rv3022c), proteins coded by esxG, esxH and esxR genes in M. tuberculosis H37Rv (loci Rv0287, Rv0288, Rv3019c, respectively), and proteins such as a lipoprotein coding by lppM gene (locus Rv2172c), an oxidoreductase (locus Rv0197), and a small secreted protein (locus Rv0236A).

Protein similarities were sorted (Figure 2) according to the strategy of genome comparison (Figure 1). Only proteins presenting more than 80% of similarity with Mycobacterium spp. genomes, and less than 50% of similarity with non-mycobacterial genomes, are shown.

Among the 11 selected mycobacterial proteins, protein alignments revealed that the ATP synthase subunit C (locus Rv1305), the oxidoreductase (locus Rv0197), and the small secreted protein (locus Rv0236A), are the less polymorphous among the 14 NTM species studied (Additional file 2) and even absent in other bacteria genus and thus seemed very promising for primers and probes design. The remaining 8 proteins that were selected, namely ATP synthase subunit A, CMAS coded by the cmaA1 gene, lipoprotein coding by lppM gene, as well as PE, PPE and proteins coded by esx genes esxG, esxH and esxR, were highly conserved in studies MTC species (tuberculosis and bovis) but very polymorphous in the 14 NTM species studied (Additional file 1), which did not allow us to design specific mycobacterial primers and probes, according to the rules of primer and probe design (Additional file 3).

DNA sequence alignment of the oxidoreductase and of the small secreted protein did not allow design of PCR primers with a minimal length of 18 oligonucleotides (Additional file 3). Only the DNA sequence alignment of the ATP synthase subunits C allowed designing a PCR primer pair and a probe. We designed the following primers and probe: forward primer FatpE 5′-CGGYGCCGGTATCGGYGA-3′ (Tm = 62°C), with the probe PatpE 5′-ACSGTGATGAAGAACGGBGTRAA-3′ (Tm = 68°C) which might be hydrolyzed by the reverse primer RatpE 5′-CGAAGACGAACARSGCCAT-3′ (Tm = 59°C, 182 bp).

Based on standard curve comparisons, our results showed reproducible amplification signals with similar Ct values for each genome equivalents of tested mycobacterial strains: M. avium, M. fortuitum, M. intracellulare, M. gordonae, and M. chelonae (Table 2). Detection limit was estimated at about 6 genome equivalents for M. chelonae by real-time PCR reaction by testing repetition of dilution limits (i.e. EC95 value: more than 95% of positive detection for these genome concentration) whereas quantification limits were estimated at about 100 genome equivalents. In the positive collection all 31 mycobacteria species were positively detected by the real-time PCR method. This collection includes NTM species, leprae species and MTC species as tuberculosis and bovis (Table 3). None of the non-mycobacterial environmental strains and none of the CNM collection strains 17 , were detected before the end of the 40 PCR cycles (Table 3). These results indicate a sensibility of 100% (31/31) and a specificity of 100% (0/30).

Ct (cycle threshold) set at 0.02. ND stands for not determined, QL for Quantification Limit, and DL for Detection Limit.

TaqMan® real-time PCR amplification was performed using forward primer FatpE, reverse primer RatpE and probe PatpE in duplicate assays. ND stands for not detected sigmoidal curve.

^aATCC: American Type Culture Collection; CPS: Collection de la Pitié-Salpêtrière, Paris, France; ^T: type strain; CIP: Collection de l′Institut Pasteur, Paris, France; CMR: Collection de Microorganismes de Radomski et al. 2010 17 , Paris, France; DSM: DSMZ-Deutsche Sammlung von Mikroorganismen und Zellkulturen, Braunschweig, Germany; IFM: Culture Collection of the Research Centre for Pathogenic Fungi and Microbial Toxicoses, Chiba, Japan.

In order to compare with culture-based method (Method A) 28 , and evaluate the impact of extraction methods on the quantification process by the new real-time PCR, we used two DNA extraction procedures (Method B and C) on water distribution samples: a commercial kit (Method B) and a published phenol-chloroform extraction (Method C) 29 . DNA extraction from tap water significantly influenced the result of mycobacteria detection by atpE real-time PCR (Figure 3A). Detection levels from DNA extracted by the kit (Method B) were significantly higher (Wilcoxon signed-rank test, n = 90, p = 0.002) than those from DNA extracted by phenol/chloroform procedure (Method C). The percentage of positive samples was significantly higher (Chi-square test, n = 180, df = 1, p = 0.021) when performing the real-time PCR with the DNA extracted by method B (33/90), compared to method C (19/90). In order to evaluate the new real-time PCR method, we compared the levels of mycobacteria detected in water distribution samples with a published culture method called method A 28 . Using the method A, Mycobacterium spp. colonies were obtained from 76% of tap water samples.

Mycobacteria quantification in lake samples by real-time PCR targeting atpE gene, shows a vast diversity of mycobacteria concentration, ranging from 10⁴ to 10⁶ ge/L in water column and neuston samples, and 10⁵ to 10⁶ ge/g DW (dry weight) in sediment samples. Comparison with the previously published methods targeting 16S rRNA 17 shows a high correlation between the results (Figure 3B, Correlation test, n = 30, Rs = 0.571, p = 0.028).

In order to detect M. tuberculosis genes, presenting homologue genes in other mycobacterial genomes, and not presenting homologue genes in non-mycobacteria genomes, we used the MycoHit software version 14.17 (Zipped copy of the files and instructions for this application are available in the Behr Research Lab, https://www.mcgill.ca/molepi/) and performed an alignment search with Stand Alone tblastn algorithm as previously described 27 . Stand Alone tblastn algorithm has been chosen because coding sequences are known to be more conserved in mycobacterial genomes than non-coding sequences, as intergenic regions, insertion sequences, or phage sequences 30 . Genome of M. tuberculosis H37Rv has been used as a reference of the Mycobacterium genus, because it is the most historically described mycobacterial genome 22 . Based on the 3989 predicted proteins from M. tuberculosis H37Rv, corresponding to the query sequences used in order to search for matches in the genomic DNA of other organisms (Figure 1), a matrix of 107703 scores (3989 protein sequences blasted against 12 non-mycobacterial genomes and 15 mycobacterial genomes) was obtained. As previously described 27 and according to NCBI procedures 48 , expected value was set at e^-10. Following sequence comparisons, the MycoHit software allowed to sort scores according to similarity requests which were performed on the one hand toward mycobacterial genomes, and on the other hand toward non-mycobacterial genomes (Figure 1). A protein list of the reference target, which can be downloaded from NCBI web site (http://www.ncbi.nlm.nih.gov), allowed identification of the conserved mycobacterial proteins presenting no homology in non-mycobacterial genomes (Figure 1).

In order to perform comparisons of pathogenic (P) and non-pathogenic (NP) mycobacterial genomes with M. tuberculosis H37Rv genome using MycoHit software, sequences were obtained at NCBI web site (http://www.ncbi.nlm.nih.gov) using the accession numbers: M. abscessus ATCC 19977 (CU458896.1) (P), M. avium 104 (CP000479.1) (P), M. avium subsp. paratuberculosis K10 (AE016958.1) (P), M. bovis subsp. bovis AF2122/97 (BX248333.1) (P), M. gilvum PYR-GCK (CP000656.1) (NP), M. marinum M (CP000854.1) (P), M. smegmatis MC2 155 (CP000480.1) (NP), Mycobacterium sp. JLS (CP000580.1) (NP), Mycobacterium sp. KMS (CP000518.1) (NP), Mycobacterium sp. MCS (CP000384.1) (NP), M. tuberculosis CDC1551 (AE000516.2) (P), M. tuberculosis H37Ra (CP000611.1) (NP), M. tuberculosis H37Rv (AL123456.2) (P), M. tuberculosis KZN 1435 (CP001658.1) (P), M. ulcerans Agy99 (CP000325.1) (P), and M. vanbaalenii PYR-1 (CP000511.1) (P). In order to avoid data lost during genome comparisons performed by MycoHit software, we have chosen to ignore some mycobacterial genomes. Since the number of coding proteins is much lower compared to other mycobacterial species, M. leprae Br4923 (FM211192.1) (P), and M. leprae TN (AL450380.1) (P) were ignored in the analysis (e.g. 1604 coding proteins in M. leprae Br4923 or 1605 coding proteins in M. leprae TN, against 6716 coding proteins in M. smegmatis MC2 155) 22 24 25 26 35 . Genomes of M. bovis BCG Pasteur 1173P2 (AM408590.1) (NP) and M. bovis BCG Tokyo 172 (AP010918.1) (NP) were also not taken into account, because these vicinal genomes present mutations 49 . Moreover, genomes of M. intracellulare ATCC 13950 (ABIN00000000) (P), M. kansasii ATCC 12478 (ACBV00000000) (P) and M. parascrofulaceum BAA-614 (ADNV00000000) (P) were also not used during MycoHit proceedings, because their genomes were still not assembled at the moment we performed the first screening step of our analysis. Nevertheless, the genomes of M. leprae, M. bovis BCG, M. intracellulare, M. kansasii and M. parascrofulaceum were used during alignment of nucleic sequences of the most conserved proteins in mycobacterial genomes.

We selected non-mycobacterial genomes of species from the CNM group using the following accession numbers: Corynebacterium aurimucosum ATCC 700975 (CP001601.1), C. diphtheriae NCTC 13129 (BX248353.1), C. efficiens YS-314 (BA000035.2), C. glutamicum ATCC 13032 (BX927147.1), C. jeikeium K411 (NC_007164), C. kroppenstedtii DSM 44385 (CP001620.1), C. urealyticum DSM 7109 (AM942444.1), Nocardia farcinica IFM 10152 (AP006618.1), Nocardioides sp. JS614 (CP000509.1), Rhodococcus erythropolis PR4 (AP008957.1), R. jostii RHA1 (CP000431.1), and R. opacus B4 (AP011115.1).

In order to check the homology of the selected mycobacterial sequences, the protein and DNA sequences of these selected proteins were aligned using the ClustalW multiple alignment of the BioEdit software 7.0.9.0 with 1000 bootstraps 50 . Primer pair and probe was designed from the best fitted gene sequences (after protein screening and selection) by visual analysis and using the Beacon Designer software version 7.90 (Premier Biosoft International, Palo Alto, Calif.).

Reproducibility, sensitivity and specificity of the new real-time PCR method were estimated using DNA from a previously described microorganism collection, and according to Radomski et al. protocol 17 . Reproducibility, efficiency, limits of detection and quantification of the real-time PCR methods 44 were estimated by quantification of several tenfold dilutions (10 replications of 400, 100, 40, 20, 4, 0.4 and 0.04 genome equivalent (ge) by reaction) of a known quantity of DNA extracted from four strains: M. avium, M. fortuitum, M. intracellulare and M. gordonae (identified from the national French reference laboratory collection). Specificity and sensitivity were estimated against 30 non-mycobacteria (negative) strains and 31 mycobacteria (positive), respectively. The collection contained reference and environmental strains of mycobacteria, as well as, strains of the closely related CNM group, and other non-actinobacteria strains isolated from the environment 17 . Mycobacteria collection included MTC (n = 2) and leprae species (n = 1), as well as species of slow growing NTM (n = 13), and rapid growing NTM (n = 15). TaqMan® real-time PCR were performed in duplicate using an ABI7500 real-time PCR system (Applied Biosystems), a Lifetech 7500 software version 2.0.6 (Applied Biosystems) and TaqMan fast virus 1-STEP Master Mix with 6-carboxy-X-rhodamine (ROX) (Applied Biosystems). The TaqMan® probes were labeled (Eurogentec) with the fluorescent dyes 6-carboxyfluorescein (5′ end) and Black Hole Quencher (3′ end). All reactions were performed in a 25 μl reaction mixture volume (2.5 μl of DNA) with 500 nM of forward primer, 500 nM of reverse primer, 50 nM of probe and 5 mM of MgCl₂. Reverse transcriptase was inactivated immediately (95 °C, 45 s) according to the manufacturer instruction, and real-time PCR consisted in 40 cycles of denaturation (95°C for 3 s), annealing and extension (both steps at 60°C for 30 s). Determinations of cycle threshold were performed by setting the instrument’s threshold line at 0.02 ∆Rn units (fluorescence gain above the baseline divided by the ROX channel signal).

In order to compare the new real-time PCR method to the culture method, 26 tap water distribution points in Paris (France) were sampled between April 2011 and July 2011, corresponding to 90 samples. Briefly, one liter of tap water was sampled in sterile plastic bottle, then centrifuged at 5000 × g for 2h and finally re-suspended in 1 ml of water. Mycobacteria density was estimated by culture (Method A) in all these samples following the procedure previously described by Le Dantec et al. 28 . In parallel, DNA was extracted using two different methods: i) a bacterial DNA extraction kit (QIAamp DNA mini kit, Qiagen) according to the manufacturer recommendations (Method B), and ii) a phenol-chloroform extraction procedure according to Radomski et al. 29 (Method C). Extracted DNA was 10 fold diluted and mycobacteria density was estimated in duplicate using the new real-time PCR method.

Using environmental samples, the new atpE targeting method was also compared a previously described rrs targeting method 17 . More precisely, samples collected from water column, sediment, and neuston of two urbanized lakes (Daumesnil Lake, Paris, France, and Créteil Lake, Créteil, France) were analyzed in triplicate. Water samples (column and neuston) were centrifuged 1 h at 7500 × g, and DNA was extracted using a MagNA Pure System (Roche). Sediment samples were lyophilized and DNA was isolated using FastDNA SPIN kit for Soil according to the manufacturer’s instructions (MP Biomedicals, Santa Ana, CA). Statistical analyses were carried out using R software v. 2.15 51 .