Study setting

This retrospective, cross-sectional study was conducted using routine molecular drug susceptibility testing records maintained at the National Reference Laboratory for Mycobacteria (LRNM; Laboratorio de Referencia Nacional de Micobacterias) of the Peruvian National Institute of Health (INS; Instituto Nacional de Salud). In October 2023, we accessed the complete LRNM-INS database of historical results obtained nationwide between January 1, 2019, and December 31, 2022. The unit of analysis was the Mycobacterium tuberculosis isolate and no clinical data were collected.

During this period, GenoType MTBDRplus v2 first-line LPA (LPA 1L), used to detect resistance to rifampicin and isoniazid, was performed across four nodes nationwide: three local reference laboratories located in Central, Southern, and Eastern Lima, and the LRNM-INS. While the three local laboratories processed samples strictly from their respective jurisdictions, the LRNM-INS maintained comprehensive national coverage, receiving samples from all 24 departments countrywide. This network structure resulted in a shared testing workflow within the Lima Metropolitan Area; while the entirety of the Constitutional Province of Callao and Northern Lima relied solely on the LRNM-INS, cases from Central, Eastern, and Southern Lima were processed both locally by their local laboratories and centrally by the LRNM-INS. In contrast, GenoType MTBDRsl v2 second-line LPA (LPA 2L), used to detect resistance to fluoroquinolones and second-line injectables, was performed exclusively at the LRNM-INS.

As a result of this network architecture, the LRNM-INS registry captured approximately 45% of all LPA 1L tests conducted nationally and 100% of all valid LPA 2L tests. Although the national diagnostic algorithm indicated that LPA 2L was reserved for first-line resistant cases, the centralized LRNM-INS database naturally contains a higher volume of LPA 2L records relative to its own LPA 1L resistant entries. This discrepancy is accounted for by the systematic incorporation of referred data from the three local Lima laboratories, which required only second-line centralized processing at the LRNM-INS (S1 Fig).

Selection and interpretation of LPA strips

All LPA strips archived at the LRNM-INS were initially screened according to predefined inclusion criteria: (i) presence of all internal control bands (conjugate and amplification controls) and the TUB band, confirming the presence of the MTB complex; (ii) banding patterns consistent with the manufacturer’s quality control guidelines (Hain Lifescience GmbH, Nehren, Germany). Duplicate laboratory results of the same sample were subsequently excluded, ensuring that each included result represented a distinct testing event. Genotypic resistance patterns for second-line drugs were determined for levofloxacin, moxifloxacin and amikacin, in accordance with updated WHO guideline [4].

LPA results were classified into three categories: i) resistance not detected, defined as the presence of all wild-type (WT) bands and absence of all mutant (MUT) bands; ii) resistance detected, defined as the presence of one or more MUT bands; and iii) resistance inferred, defined as the absence of one or more WT bands in the absence of corresponding MUT bands. Specific mutations associated with each MUT probe were identified following the manufacturer’s instructions. Mutations conferring resistance to isoniazid and moxifloxacin were further stratified into low- and high-level resistance categories, as recommended by WHO [4]. Both homoresistance (absence of WT bands with presence of the corresponding MUT bands) and heteroresistance (simultaneous presence of WT and corresponding MUT bands) were considered.

Determination of drug-resistant genotypes

Resistance patterns were analyzed to identified drug-resistant genotypes using GeneAlEx v6.503 [12]. Resistant genotypes were defined as unique combinations of resistance-associated mutations in the target genes interrogated by LPAs 1L and 2L, separately. Hybridization patterns from DNA probes were converted into binary matrices indicating the presence or absence of specific bands. First-line (G1) and second-line (G2) resistance genotypes were generated by concatenating banding patterns from resistance-associated loci only. During matrix compilation, the software assigned consecutive nominal identifiers based on initial detection order (e.g., G1-1, G1-2). Genotypes were classified based on their relative frequencies (n) as follows: common (n > 5%), less common (1% ≤ n ≤ 5%), rare (n < 1%; excluding n = 1) and orphan genotypes (n = 1) [13].

Targeted sequencing

A stratified purposeful sampling strategy was employed to select isolates for targeted sequencing, focusing specifically on isolates exhibiting inferred resistance genotypes. From a total universe of 584 available isolates with inferred genotypes, a sample of 170 isolates was selected for sequencing due to reagent availability and sequencing capacity constraints. To ensure representativeness, these 170 isolates were selected using a stratified proportional sampling design that preserved the national database ratio of first-line (485 isolates in the universe) and second-line (99 isolates in the universe) inferred genotypes. This sample size provides a 95% confidence level with a margin of error of approximately 6.3%, assuming maximum variability (p = 0.5), ensuring the representativeness of circulating strains with these inferred resistance genotypes.

Genomic DNA was extracted from cryopreserved MTB isolates using the Presto Mini gDNA Bacteria Kit (Geneaid Biotech, Taiwan), following the manufacturer’s protocol for Gram-positive bacteria. Multiplex PCR of rpoB, inhA, gyrA, gyrB and rrs genes was performed with in-house primers (S1 Table) in a 25 uL reaction (5 uL of 5xTransStart FastPfu Buffer, 2 uL of 2.5 mM of each dNTPs, 5 uL of primer Mix, 0.5 uL of hot-start high-fidelity TransStart FastPfu DNA Polymerase, 7.5 uL molecular biology-grade water, and 5 uL template DNA). Cycling conditions on a Gradient PCR Optima 96 system included: 95°C for 2 minutes; 40 cycles of 95°C for 20 seconds, 65°C for 30 seconds, and 72°C for 80 seconds; with a 5 minutes of final extension at 72°C. Amplicons were sequenced using Next-Generation Sequencing on the Illumina MiSeq platform (2 × 150 bp paired-end reads). Raw sequencing data are available in the NCBI Sequence Read Archive (BioProject PRJNA1265395).

Statistical analysis

Temporal changes in the frequencies of common G1 and G2 genotypes were evaluated using the chi-square (χ²) test implemented in R v4.4.2 [14]. To account for potential bias due to unequal sample sizes across years, absolute genotype counts were normalized using the year with the lowest number of analyzed isolates as reference. A two-sided p-value < 0.05 was considered statistically significant.

Ethic statement

This study was conducted as part of a research project approved by the Institutional Research Ethics Committee of the INS (approval code: OC-028–21). The study was carried out with the authorization of the INS National Reference Laboratory for Mycobacteria, using anonymized data and bacterial isolates collected as part of routine diagnostic testing at the national reference laboratory between 2019 and 2022. No personally identifiable information was collected, and all samples were anonymized using unique study-specific codes. Informed consent was not required due to the retrospective use of de-identified, routinely collected data.

Sample characteristics and molecular resistant profiles

A total of 38,339 valid LPA results were analyzed, including 27,519 from LPA 1L and 10,820 from LPA 2L. Resistance-associated genotypes were detected in 3,726 (13.5%) LPA 1L results and 1,278 (11.8%) LPA 2L results. 90% (3,379/3,726) of the LPA 1L–resistant isolates yielded valid LPA 2L results at the LRNM-INS. However, a substantial proportion of these 3,379 isolates showed fully susceptible LPA 2L patterns, which precluded meaningful genotype-based analyses for second-line resistance. Consequently, genotypes were analyzed independently for LPA 1L and LPA 2L. Geographically, resistant isolates for LPA 1L were reported across all 24 departments of Peru, with 70.5% (2,625/3,726) of the entries concentrated in Lima, Callao, Ica, Ucayali, Ancash, Junín and Loreto. For LPA 2L, Lima, Callao and Ica accounted for 79.3% (1,013/1,278) of resistant isolates, whereas no resistant results were observed in Pasco, Apurímac, or Huancavelica (S2 Table).

Regarding genetic resistance among the LPA 1L cohort, rifampicin resistance was detected in 2,339 isolates, comprising 1,977 MDR and 362 rifampicin mono-resistant [RMR] entries. At nucleotide level, this resistance was dominated by canonical substitutions within the rpoB rifampicin resistance-determining region (RRDR), with S450L emerging as the most prevalent variant followed by D435V. On the other hand, isoniazid resistance was identified in 1,387 isoniazid mono-resistant (HMR) isolates. This resistance was mainly explained by mutations in katG gene and inhA promoter region, with most resistant isolates harboring katG S315T1, inhA c-15t or both. High-level resistance predominated over low-level variants. Overall, MDR-TB profiles accounted for 59% of all isoniazid-resistant and 85% of rifampicin-resistant isolates.

In contrast, within the LPA 2L cohort, 309 isolates showed simultaneous resistance to levofloxacin/moxifloxacin and amikacin (Lfx/Mfx-Am R), 692 exhibited levofloxacin/moxifloxacin resistance only (Lfx/Mfx R) and 277 showed amikacin resistant only (Am R). Levofloxacin/moxifloxacin resistance was mainly driven by gyrA mutations concentrated at codons 94 and 90, whereas gyrB mutations contributed marginally. These variants were associated with both high- and low-level moxifloxacin resistance (low-level resistance being slightly more frequent). Meanwhile, amikacin resistance was overwhelming associated with the non-coding rrs a1401g mutation (S3 Table).

Drug-resistant genotypes

Regarding the first-line cohort, a total of 96 G1 genotypes were identified, displaying a high concentration of resistance within a limited number of molecular patterns. Specifically, 13 dominant genotypes (comprising six common and seven less common) accounted for 88.9% (3,313/3,726) of all first-line resistant isolates, encompassing the principal MDR, HMR and RMR profiles (Table 1). In contrast, the remaining first-line resistant isolates were distributed among 51 rare and 32 orphan genotypes. While these low-frequency genotypes collectively represented slightly over 10% of the cohort, they exhibited substantially greater diversity in resistance patterns and mutational combinations compared to the dominant variants. Notably, while dominant genotypes exclusively carried homoresistant mutations, heteroresistance was restricted to these low-frequency categories (occurring in 19 rare and 12 orphan genotypes) and accounted for 2.8% (103/3726) of isolates. Whitin these heteroresistant configurations, mutations conferring resistance to isoniazid occurred at least twice as frequently as those for rifampicin. Furthermore, double mutations were uncommon and restricted to the rpoB gene, occurring exclusively among rare and orphan genotypes across 10 isolates, seven of which were heteroresistant (S4 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Dominant first-line MTB genotypes identified by line probe assays in Peru.

https://doi.org/10.1371/journal.pone.0352881.t001

A highly consistent architectural pattern was observed within the second-line cohort, where 101 G2 genotypes were identified. Resistance was highly concentrated in 19 dominant genotypes (seven common and twelve less common) accounting for 85.4% (1,092/1,278) of the isolates and capturing the most frequent Lfx/Mfx-Am R, Lfx/Mfx R, and Am R profiles (Table 2). Conversely, the remaining second-line resistant isolates were composed of 35 rare and 47 orphan genotypes, which displayed a markedly higher diversity of mutational combinations than the dominant counterparts. Heteroresistance was confined to three less common, 18 rare, and 26 orphan genotypes, and was detected at least twice as often for levofloxacin or moxifloxacin compared to amikacin. Additionally, advanced mutational complexity, manifested as double and triple mutations, were exclusively detected within the gyrA gene and occurred only among rare and orphan genotypes. This complex genetic signature encompassed 39 isolates, of which 19 were heteroresistant (S5 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Dominant second-line MTB genotypes identified by line probe assays in Peru.

https://doi.org/10.1371/journal.pone.0352881.t002

Mutation patterns of inferred genotypes

A total of 170 samples were successfully sequenced, including 141 from G1 and 29 from G2 genotypes. Across both groups, most inferred genotypes displayed multiple mutations at comparable frequencies, although a subset was dominated by a single, recurrent mutation.

Among G1 genotypes, several patterns were driven by a predominant mutation. The G1-72 genotype was largely explained by the inhA g-17t mutation, whereas G1-24 was exclusively associated with the synonymous rpoB F433F mutation. Other G1 genotypes exhibited greater mutational heterogeneity within the rpoB gene.

Similarly, among G2 genotypes, inferred resistance patterns were frequently dominated by a single mutation. The G2-60 genotype was mainly associated with gyrB T500N, G2-2 with gyrA G88C, and G2-59 with gyrB E501D, indicating consistent mutation-genotype associations despite lower overall frequencies.

A small number of isolates lacked detectable mutations within the sequenced regions. These included isolated samples within G1-72, G1-11 and G2-2, as well as a substantial proportion of G1-48 isolates and all sequenced isolates belonging to G2-65. With one exception, all isolates carried a single mutation with an allelic frequency of at least 98%. The exception was a single G1-29 isolate, which harbored two rpoB mutations at different proportions (H445R [40%] and H445L [60%]), consistent with intraisolate heterogeneity.

Although inferred genotypes were expected to yield resistant profiles distinct from directly detected mutations, a limited number of isolates nonetheless harbored directly detectable resistance mutations, including inhA c-15t, rpoB D435V, and rpoB S450L (Table 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Mutational profiles of first- and second-line drug-resistant MTB genotypes with inferred resistance identified in Peru.

https://doi.org/10.1371/journal.pone.0352881.t003

Spatiotemporal distribution of common genotypes

Regarding temporal trends, the absolute frequencies of common G1 and G2 genotypes remained highly stable across the study period. Rare exceptions to this temporal stability included a significant increase in G1-73 during 2022 (p = 0.002), a transient increased in G1-59 in 2020 (p = 0.037), and a marked decline in both 2020 and 2022 for G2-30 (p = 0.001) (S6 Table). Geographically, the distribution of the common genotypes revealed distinct epidemiological patterns across the country (Fig 1). G1 genotypes exhibited a hyper-endemic core concentrated in the central coastal region (Lima, Callao, and Ica) and the jungle department of Ucayali, where the MDR genotype G1-57 demonstrated a disproportionately high localized burden (n = 162). In contrast, departments in the southern highlands and coast displayed markedly lower sample volumes. Huancavelica and Moquegua exhibited a complete polarization toward HMR genotypes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. National distribution of common drug-resistant MTB genotypes identified by first- and second-line line probe assays.

The heatmap shows the distribution of common genotypes detected by GenoType MTBDRplus v2 and MTBDRsl v2 assays across Peru, operationally grouped for analytical purposes to highlight regional density clustering. Numbers in the cells indicate sample counts, with darker colors showing higher isolates concentration. Resistant profiles were classified by first-line resistance as MDR (multidrug-resistant) and HMR (isoniazid mono-resistant), and by second-line resistance as Lfx/Mfx-Am R (levofloxacin/moxifloxacin and amikacin-resistant), Lfx/Mfx R (levofloxacin/moxifloxacin-resistant), and Am R (amikacin-resistant). G1 and G2 refer to first- and second-line genotypes, respectively.

https://doi.org/10.1371/journal.pone.0352881.g001

A much more restricted and fragmented geographic landscape emerged for G2 genotypes. Resistance was overwhelmingly driven by the amikacin-resistant G2-66 and the three Lfx/Mfx R genotypes (G2-16, G2-35, and G2-52). Outside of Lima, the dispersal of second-line genotypes dropped abruptly, appearing only sporadically as isolated clusters, while remaining entirely undetected across five departments (Tumbes, Pasco, Huancavelica, Moquegua, and Apurimac) (Fig 1).