Two hundred and eighty-three adult presumptive TB patients, treatment-naive, under antibiotic treatment or that were previously treated admitted at Laboratorio de Bacteriología de Tuberculosis, Instituto de Tisioneumonología Dr. Raul Vaccarezza, Hospital Muñiz, Ciudad Autónoma de Buenos Aires, Argentina were included in the study. All the participants signed written informed consent forms, previously approved by the Ethics Committee of Instituto Vaccarezza, to participate. The study included adult patients of different nationalities (mostly Argentinean and from neighboring countries) and gender, with different classifications at the time of sample taking: treatment-naive, under treatment, discontinued treatment, relapsed, treatment failure, etc. Only pulmonary samples were analyzed.

Sputum samples were collected in a sterile container and decontaminated using a modified Petroff protocol. Briefly, after addition of one volume of 4% NaOH the sample was incubated at room temperature for 15 min. After incubation, it was centrifuged for 20 min, the supernatant was discarded, and the pellet was neutralized to pH 7 by dropwise addition of 15% H₂SO₄. The pellet was washed and resuspended with distilled water. The pellet was used for routine tests performed in Laboratorio de Bacteriología de la Tuberculosis, including Ziehl Neelsen stain (Weldu et al., 2013) inoculation on Löwenstein Jensen (LJ) media for detection and colony counting of mycobacteria and inoculation on Stonebrink media (LJ supplemented with 2% sodium pyruvate) for detection of M. bovis (Daza Bolaños et al., 2017). Only for samples of patients at high-risk of developing TB disease an aliquot was used for analysis with MGIT 960 (Martin et al., 2009). Finally, the remnant of the pellet was used to test with mCherry_bombϕ.

In this study, we used the second generation fluoromycobacteriophages, carrying an mCherry_bomb gene with optimized expression in mycobacteria (Urdániz et al., 2016). Stocks of mCherry_bombϕ were prepared as previously described (Rondón et al., 2011).

Remnant pellets of processed sputum samples were resuspended in 400 μl of Middlebrook 7H9 broth supplemented with OADC Enrichment (Becton, Dickinson and Company, Sparks, MD, United States) and PANTA (Becton, Dickinson and Company, Sparks, MD, United States) and divided in four 100 μl aliquots. In two aliquots, 4 μg/ml RIF was added and the other two were kept as control. Samples were recovered for 48 or 96 h and infected overnight with 5 μl of a stock of mCherry_bombϕ (10¹⁰–10¹¹ PFU/ml). After that, samples were fixed with 2% paraformaldehyde for 2 h at room temperature, centrifuged and the pellet was resuspended in 10 μl of phosphate buffered saline (PBS) and examined by fluorescence microscopy using 1,000× magnification. The criterion for detection was the presence of at least one fluorescent bacillus per 100 fields in the control samples and for determination of RIF resistance was the same in RIF treated samples. We also compared the results using the criterion of two fluorescent bacilli per 100 field but we did not observed significantly differences between both criteria so we opted for the first one. Detection results using mCherry_bombϕ were compared with growth and colony counting (CFU) in LJ media and determination of RIF susceptibility with the proportion method (Canetti et al., 1963).

A fluorescence microscope (Axiostar Plus; Carl Zeiss) with a 100× objective with oil immersion and phase contrast was used. Fluorescence images were acquired using a Cannon EOS Rebel T3 and the software provided with the camera. For detection of fluorescent mCherry_bomb the 64 HE mPlum shift free (E) filter (EX BP 587/25, BS FT605, EM BP 647/70) (Carl Zeiss) was used. When necessary, image processing was performed using Adobe Photoshop CS2 (Adobe Systems Incorporated), but maintaining identical brightness and contrast settings for all images.

Clinical isolates from sputum samples positive for Mycobacterium spp. and further identified as NTM were inoculated in Middlebrook 7H9 broth supplemented with OADC and grown to exponential phase. At least two clinical isolates identified as M. tuberculosis were also used as controls. Two culture aliquots of 100 μl were incubated with 500 μg/ml of PNB for 24 and 48 h and one aliquot was kept as control. After incubation, samples were centrifuged and the pellet was resuspended in the same media. Samples were infected overnight at 37°C with mCherry_bombϕ at an approximate multiplicity of infection (MOI) of 100. Samples were processed as described above and observed by fluorescence microscopy.

Four M. tuberculosis strains isolated from sputum samples used in this study, that tested positive for Mycobacterium and were RIF^R with mCherry_bombϕ, were further identified as pre-XDR and XDR-TB using the proportion method. To determine the antibiotic resistance profile of these strains using our fluoromycobacteriophage methodology, 200 μl of an exponentially growing culture were preincubated for 24 h with the following antibiotics: isoniazid (0.2 μg/ml), rifampicin (4 μg/ml), streptomycin (2 μg/ml), ofloxacin (10 μg/ml), ethambutol (5 μg/ml), amikacin (1 μg/ml), kanamycin (2.5 μg/ml) and further infected overnight with mCherry_bombϕ at an approximate MOI of 100. After incubation, samples were processed as described above and observed by fluorescence microscopy.

Colonies from LJ media were used for DNA extraction. The protocol for isolation of DNA is described in van Soolingen et al. (1991). Prior to sequencing the integrity of the DNA was evaluated by gel electrophoresis (van Soolingen et al., 1991).

Paired end reads from the four strains analyzed were obtained using a whole-genome shotgun strategy with a MiSeq, Illumina at the Instituto Malbrán, Buenos Aires, Argentina. Sequence reads were checked for quality using FastQC version v0.11.5 (Andrews, 2011) and processed with the PRINSEQ lite version 0.20.4 (Schmieder and Edwards, 2011) for sequence read quality assessment.

All the reads were deposited in GenBank NCBI Bioproject PRJNA438689 with the following accession numbers: 152748 (ID: 8724044, 8724043), 161109 (ID: 8724038, 8724037), 156324 (ID: 8724040, 8724039), and 16138 (ID: 8724042, 8724041).

Subsequently reads were aligned to the H37Rv reference genome (NCBI access number NC_000962.3) using BWA (Li and Durbin, 2010) and variant calling was made with GATK (Mckenna et al., 2010). Resulting variants were annotated with SnpEff (Cingolani et al., 2012) and compared with previously reported drug resistance-associated variants listed in TBDREAM database (Sandgren et al., 2009). Each match was classified using the following criteria: (I) Position match, I(a): Exact match: the variant in the strain is exactly the same that the reported in TBDREAM. I(b) Partial match: the variant in the strain is in the same position that the reported in TBDREAM but present a different nucleotide. (II) Gene match: the strain has a variant in a gene reported in TBDREAM, but it is in different position. (III) No match. Analysis of the genes containing variants was performed with TuberQ database (Radusky et al., 2014). When we were not able to find a gene variant in TBDREAM that could explain the antibiotic resistant phenotype, we manually inspected other genes that have been previously reported in resistant isolates.

On the other hand, a short phylogenetic analysis was performed between the samples mentioned above and samples that were previously reported as M, Ra and Beijing sublineages downloaded from NCBI SRA: ERR760768, ERR776666, and SRR5314268, respectively. Isolates reads were processed using the variant calling pipeline mentioned above. At last, resulting variants from all samples were combined in a single VCF using GATK and a dendrogram was built using VCFKit (Cook and Andersen, 2017) with UPGMA algorithm using Beijing sample as outgroup. Figure 4 was built using ETE Toolkit (Huerta-cepas et al., 2016).

Data analysis for calculation of sensitivity, specificity, and predictive values was performed using the R language (The R Foundation, 2018), and the exact 95% confidence intervals were estimated using the Hmisc library. To compare the performance of mCherry_bombϕ and culture growth in LJ media for detection of Mycobacterium spp. we employed box and whisker plots.

To evaluate the correlation of mCherry_bombϕ based detection and Mycobacterium spp. growth in LJ we applied generalized linear models using the negative binomial distribution. When needed multiple comparisons were performed by fitting a negative binomial models to each pair of categories of growth in LJ media using the MASS (Venables and Ripley, 2002) library for R, and adjusting the p-values with the Bonferroni correction. χ² goodness of fit tests showed data corresponded to the negative binomial (in contrast, when a Poisson model was used this test showed a poor fitting). When needed, incidence rate ratios (IRRs) were calculated as the exponential of the model’s coefficient.

We evaluated the performance of mCherry_bombϕ to detect the presence of Mycobacterium spp. and determination of RIF resistance in the remnant of 283 sputum samples from presumptive TB patients, treatment-naive, previously treated for TB or under antibiotic treatment for TB. The patient population enrolled in this study is described in Table 1. Sputum samples were processed, and part of the pellet was used for routine diagnostic tests and the remnant was used for testing with the fluoromycobacteriophage methodology. The bacteria-containing pellet was recovered in rich media for 48 or 96 h and infected with mCherry_bombϕ in the presence or absence of RIF. Samples were fixed with paraformaldehyde, spun down and 100 fields were examined by fluorescence microscopy. The complete protocol is schematized in Figure 1. A representative image of a processed sputum sample recovered for 48 h and infected with mCherry_bombϕ or mock infected is shown in Figures 2A,B. A sample was considered RIF sensitive if fluorescent bacilli were detected in the control sample and fluorescence was lost in the presence of RIF. On the contrary, if fluorescent bacilli were observed in the presence of RIF, the sample was considered resistant to the drug. The sensitivity, specificity, positive and negative predictive values of the bacteriophage method were calculated in comparison to culture in LJ for detection of Mycobacterium spp., or the proportion method for determination of RIF resistance, respectively. The results are summarized in Tables 2, 3.

For detection of Mycobacterium spp., when the pellet was recovered for 48 h before infection with mCherry_bombϕ, 67.84% (192 samples) were concordant between both methods and 32.16% (91 samples) were not detected by mCherry_bombϕ but cultured positive in LJ media. An increase in the recovery time to 96 h prior infection improved the detection to 94.35% (267 samples) consistent with both methods. Only 6% of the samples were discordant: 15 tested positive by culture but negative with mCherry_bombϕ and 1 was positive by phage and negative by culture (Table 2A).

Overall, the sensitivity of mCherry_bombϕ for detection of Mycobacterium spp. was 51.36% and 91.98% after 48 and 96 h of recovery, respectively, and the specificity was 100% and 98.96% (Table 2B). The high specificity was expected because fluoromycobacteriophages can only infect bacteria of the genera Mycobacterium. This is a notable advantage of the method because fluorescent Mycobacterium bacilli could be easily detected by microscopy even in the presence of other genera contaminants in the sample (Figure 2C).

As shown in Table 2B, in acid- fast bacilli (AFB) smear-positive patients the sensitivity and specificity were 93.41% and 93.33%, respectively, after recovery for 96 h. The sensitivity in the 86 AFB smear- negative samples was only 40% after 96 h, but it is worth noting that from these samples only five were positive by culture and two of them tested positive with mCherry_bombϕ.

Interestingly, for HIV positive patients [that represent 22.61% of the study population and are known to be paucibacillary (Madico et al., 2016; Fan et al., 2017)] the sensitivity and specificity after 96 h reached 97.3% and 100%, respectively.

When the performance of mCherry_bombϕ for detection of Mycobacterium spp. was calculated for treatment-naive patients or patients previously treated for TB, the values after 96 h of recovery were between 94.87 and 100% for sensitivity and 100% for specificity.

When the same analysis was performed for patients that were currently under antibiotic treatment both parameters drop to 89.25% for sensitivity and 96.55% for specificity after a recovery time of 96 h. From the 15 samples that tested positive by culture and negative by phage (Table 2A), 13 corresponded to these groups of patients.

Because fluoromycobacteriophages only infect viable cells that are metabolically active, a likelier explanation for this lower sensitivity is that even after recovery of the sputum sample for 96 h the cells were not in a physiological state to promote mCherry_bomb expression. Alternatively, because we only have access to the remnant of the processed sputum sample, the number of bacteria could be very low in these samples below the level of detection. The reference method used for comparison was culture in LJ medium, which provides all the requirements for optimal growth of mycobacteria, and the time of incubation (30–40 days) allows recovery of bacilli originally present in the sputum sample.

Finally, since we could not discard that Mycobacteria present in those 15 discordant samples were phage resistant, we further prepared liquid cultures from the colonies isolated in LJ media and infected them with mCherry_bombϕ. In all samples we visualized bright red fluorescent cells (data not shown) discarding that were phage resistant bacteria.

Then, we evaluated the performance of mCherry_bombϕ to determine susceptibility to RIF. Out of the 283 samples assessed, 16 were identified as RIF^R according to the reference method (Table 3A).

When we used mCherry_bombϕ to determine susceptibility to RIF after a recovery time of 48 h, five samples were identified as RIF^R with mCherry_bombϕ. In 96.11% of the cases (272 samples) the results for RIF susceptibility were consistent between both methods and in 3.89% (11 samples) were discordant. When recovery time was extended to 96 h, the 16 RIF^R samples tested positive with the fluoromycobacteriophage method; the results were congruent for 99.29% of the cases (281 samples) and only 0.71% (2 samples) were discordant (Table 3A).

Among the samples determined as RIF^R, four isolates were identified a posteriori as XDR-TB and pre-XDR-TB (see below), illustrating the importance of early detection of resistance to this antibiotic.

Altogether, the sensitivity, specificity and predictive values of mCherry_bombϕ to detect rifampicin resistance compared to the proportion method are summarized in Table 3B. While the specificity was 100% after recovery for 48 and 96 h, sensitivity highly improved from 31.25%, at 48 h, to 100% when the sample was recovered for 96 h (Table 3B).

As can be seen from Table 3B, the specificity for detection of rifampicin resistance in all the groups of patients was 100% after recovery for 48 or 96 h.

In AFB smear-positive patients the sensitivity for rifampicin resistance detection reached 100% after recovery for 96 h. The sensitivity in AFB smear- negative samples could not be calculated because none of these samples were RIF^R.

In HIV positive and negative patients 100% sensitivity was obtained after recovery for 96 h. Once again, sensitivity could not be calculated for treatment-naive patients because none of them carried RIF^R strains. For patients under treatment or previously treated for TB, sensitivity highly improved to 100% when cells were recovered for 96 h before infection with mCherry_bombϕ.

Overall, it was clear that an extended recovery time was crucial to increase the sensitivity of the fluoromycobacteriophage methodology to determine susceptibility to this first line antibiotic.

One advantage of phenotypic versus genotypic tests is that can discriminate between viable and non-viable cells. This is a key feature for a test intended to monitor patient’s response to antibiotic treatment. Fluoromycobacteriophages can develop a fluorescent signal only in metabolically active cells. As shown in Figure 3 there is a direct correlation between the number of colonies in LJ and the number of fluorescent bacilli detected by the fluoromycobacteriophage method. When samples were recovered for 48 h before infection, a very low number of fluorescent bacilli (López-Hernández et al., 2016; Sulis et al., 2016) were detected for a colony count between 20 and 100, but this amount increased to 13–27 bacilli per hundred fields when the CFU were more than 100 or uncountable (Figure 3A). The number of bacilli detected by mCherry_bombϕ notably increased when the recovery time was 96 h and even for those samples with poor growth in LJ media (1–19 CFU) we were able to detect an average of 8 bacilli/100 fields with the phage method (Figure 3B) highlighting the good performance of fluoromycobacteriophages in paucibacillary samples when enough time is given for recovery prior to infection.

In order to additionally discriminate between MTBC and NTM using mCherry_bombϕ, we evaluated the effect of PNB during the assay. As stated, mCherry_bombϕ is a derivative of TM4, a broad host range mycobacteriophage (Riska et al., 1997) thus it was not surprising that some of the samples testing positive for Mycobacterium spp. with mCherry_bombϕ were further identified as containing M. kansasii, M. szulgai, M. fortuitum, M. abscessus, and M. avium.

Cultures of M. tuberculosis and atypical mycobacteria were grown until exponential phase and were preincubated with 500 μg/ml of PNB for 24 h prior to infection with mCherry_bombϕ. As shown in Figure 4A, when M. tuberculosis was pretreated with PNB, no fluorescent bacilli were observed. In contrast, infection and expression of mCherrybomb was not affected in the NTM strains tested (Figures 4B–E and data not shown). These observations suggest that PNB in combination with fluoromycobacteriophages can be used as a simple way to distinguish between MTBC and NTM bacteria present in sputum samples.

As we previously mentioned, four clinical isolates identified in this study as RIF^R using mCherry_bombϕ were further reported as XDR or pre-XDR-TB in the lab records.

We evaluated the performance of mCherry_bombϕ for extended DST. Resistance to isoniazid, streptomycin, ofloxacin, ethambutol, amikacin, capreomycin, and kanamycin was assessed using liquid cultures of these clinical isolates and results were compared to the proportion method.

We also performed WGS of the four isolates and we searched for mutations in genes reported to confer resistance to the tested antibiotics. The results obtained for the three methodologies are summarized in Table 4.

For isoniazid and rifampicin we found a perfect correlation between both phenotypic methods and we were able to easily identify mutations that could account for the resistance profile. Genotypic analysis revealed the presence of the katG S315T mutation in three isolates and the other presented a kasA G269S mutation associated to isoniazid resistance (Lee et al., 1999; Galarza et al., 2014). This strain also had an additional mutation in katG (P432L) that has not been previously reported as associated to isoniazid resistance. All RIF^R strains had mutations in rpoB, three corresponded to the well reported S450L mutation and the other was a change from aspartic acid to valine in position 435 (D435V) (Eldholm et al., 2015; Datta et al., 2016).

For streptomycin, we found a good correlation between the fluoromycobacteriophage and the proportion method and we were able to identify the gene variants that could be responsible for the observed phenotype. One of the strains presented a mutation in the 16S rRNA gene, rrs 516C>T (Dobner et al., 1997). The other two strains had a frameshift in gidB that is a well-established determinant of streptomycin resistance (Okamoto et al., 2007). Moreover, the gidB V110 frameshift was previously found in 90% of analyzed isolates of the M strain that caused an outbreak in Argentina in the early 1990s (Eldholm et al., 2015). Streptomycin was introduced in anti TB regimens in Argentina in 1946 (de Kantor and Barrera, 2007) so it is important to include this antibiotic during DST.

Results for ethambutol were more discrepant. Two strains were sensitive by the reference method and resistant by mCherry_bombϕ. The strain 161109 had an embC V981L mutation that could render the strain sensitive or resistant depending of the method use for DST (Sreevatsan et al., 1997). Strain 16138 also had an embB G406A mutation in addition to that but tested sensitive to ethambutol with the proportion method. Antibiotic susceptibility testing was repeated in this strain, with the same outcome. This result was unexpected because strain 156324 presented the same gene variants for embB and embC and was resistant to ethambutol by phage and the proportion method. It has been previously reported that testing of resistance to ethambutol can be inconsistent and can lead to ambiguous results (Madison et al., 2002).

For kanamycin and the other two injectable antibiotics tested, amikacin and capreomycin (data not shown), we got similar results for both phenotypic methods with the exception of isolate 161109 that tested sensitive for kanamycin by the reference method and resistant by the phage method. The sample from this patient was also tested using MGIT and some bacterial growth was observed at day 12 in the presence of kanamycin. This could be indicative of the presence of a mixed population with the Kan^R population present in a very low proportion. It is important to notice that fluoromycobacteriophages can detect individual cells being particularly useful for detection of partially resistant cultures even if resistant cells are underrepresented in the sample. For two of the resistant isolates we identified the rrs1401A>G mutation associated to kanamycin resistance but for strain 152748 and also for isolate 161109 we were not able to find any mutations that could be associated to this phenotype.

All the strains resulted resistant to at least one fluoroquinolone, either with mCherry_bombϕ or the reference method. In all the strains we identified mutations in gyrA previously reported as associated to fluoroquinolone resistance. Additionally, all four isolates presented the following mutations: gyrA E21Q and gyrA G688D but we were not able to find any reports describing these mutations as responsible for resistance to fluoroquinolones.

Finally, we performed a phylogenetic analysis in an attempt to trace these four isolates. For this analysis we constructed a dendrogram based on SNPs between these strains and clinical isolates of two of the largest TB outbreaks in Argentina with available reads at NCBI, ERR760768, and ERR776666, reported as M (of Haarlem family) and Ra (of LAM- Latin American Mediterranean family), respectively (Eldholm et al., 2015; Bigi et al., 2017) (Figure 5). The analysis of the dendrogram shows that 156324 and 16138 cluster in the same group as the M strain. However, strain 152748 was closer to Ra than other analyzed strains. Strain 161109 did not cluster with any of the analyzed strains.