Acquired linezolid resistance in DR-TB: genotype-phenotype discordance and molecular heterogeneity in a retrospective cohort

Abstract

Background

Linezolid is an important drug for the treatment of drug-resistant tuberculosis (DR-TB). Acquired linezolid resistance threatens treatment efficacy. This study aimed to analyze the risk factors and molecular mechanisms of linezolid-acquired resistance in a clinical cohort.

Methods

We conducted a retrospective study of 61 patients with DR-TB who failed linezolid-containing regimens (2017–2021). Paired baseline and post-treatment Mycobacterium tuberculosis isolates were subjected to linezolid Minimum inhibitory concentration (MIC) testing (Microplate Alamar Blue Assay) and sequencing of rplC, rrl (23 S rRNA), and rplD. Acquired resistance was defined as a ≥ 4-fold increase in MIC or the presence of new resistance-conferring mutation. Risk factors were analyzed using logistic regression.

Results

Acquired linezolid resistance occurred in 31.1% (19/61) of patients. Younger age (< 37.5 years; OR = 1.04/year decrease, p = 0.012) and prolonged therapy (> 14.3 months; OR = 1.07/month, p = 0.004) were the independent risk factors. Sequencing revealed 26 mutation events across the longitudinal isolates; 57.7% (15/26) harbored rplC T460C (Cys154Arg), whereas 38.5% (10/26) had rrl mutations (G2270T, G2814T). A novel rplD A518T (Lys173Ile) mutation emerged concurrently with the rplC T460C mutation. Phenotypic-genotypic discordance occurred in 11.5% (3/26) of resistant isolates, including one strain with MIC elevation (0.25→4.0 mg/L) lacking target mutations. Linezolid resistance correlated with prior resistance to clofazimine (kappa = 0.538, p < 0.001), amikacin, and pasiniazid.

Conclusions

Young age, extended therapy, and ribosomal mutations drive linezolid resistance. Pretreatment susceptibility screening is critical.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-026-12869-8.

Introduction

Tuberculosis (TB) remains a major global health threat, with approximately 10.7 million new active TB infections and 1.23 million deaths reported by 2024 [World Health Organization (WHO) 2025]. Multidrug-resistant/rifampicin-resistant tuberculosis (MDR/RR-TB) accounts for approximately 390,000 cases each year, presenting particular therapeutic challenges due to limited drug options, extended treatment durations (typically 18–24 months), and suboptimal patient adherence, which collectively contribute to success rates of only 68–71% in high-burden settings [1]. Linezolid (LZD), the WHO Group A core drug since 2019 [2], inhibits bacterial protein synthesis by binding to the 23 S rRNA of the 50 S ribosomal subunit and confers potent bactericidal activity with a high genetic barrier to resistance. A meta-analysis showed that treatment success was positively associated with LZD use compared to not using the drug [3], supporting the WHO’s current recommendation of shortened all-oral regimens. These regimens, including BPaL (bedaquiline, pretomanid, linezolid), BPaLM (BPaL + moxifloxacin), and BDLLfxC (bedaquiline, delamanid, linezolid, levofloxacin, clofazimine), have demonstrated high efficacy (e.g., 90%-92% success in pivotal trials of BPaL/BPaLM) in clinical trials [4]. However, acquired LZD resistance is increasingly observed at rates of 9%-15% in MDR-TB strains and up to 60% in extensively drug-resistant TB (XDR-TB) cohorts [5]. This resistance primarily stems from two factors: suboptimal implementation (in addition to failing regimens) or treatment interruption due to toxicity-related adverse events, such as myelosuppression and neuropathy [6, 7].

Molecular studies show 50%-70% of LZD resistance involves ribosomal gene mutations (rplC, rrl, and rplD) [8–10], whereas the remaining cases involve non-ribosomal mechanisms, including efflux pump upregulation and reduced drug permeability [11, 12]. The WHO’s redefinition of XDR-TB, including resistance to bedaquiline or linezolid, underscores the urgency of understanding these mechanisms [13]. However, significant knowledge gaps persist. First, the quantitative genotype-phenotype relationship for LZD remains complex and may vary across genetic backgrounds and clinical settings. Second, identifiable the clinical risk factors, particularly in patients experiencing treatment failure, are not well defined. Third, the global lack of standardized phenotypic susceptibility testing forces empirical prescribing, creating a cycle that hinders the collection of precisely linked clinical and microbiological data needed to resolve the first two gaps.

To address these interconnected challenges, we conducted an integrated analysis of a clinical cohort characterized by confirmed treatment failure—a high-risk scenario critical for studying acquired resistance. This study aimed to identify modifiable risk factors for acquired resistance, elucidate genotype-phenotype correlations across key resistance loci, and ultimately to provide data that can inform clinical decision-making and guide the optimization of LZD use in resource-constrained settings where molecular diagnostics remain unavailable.

Methods

Study design

We retrospectively enrolled patients with drug-resistant tuberculosis (DR-TB) who had failed a linezolid-containing regimen at Beijing Chest Hospital (2017–2021). Eligible patients included those with RR-TB, MDR-TB, pre-extensively drug-resistant tuberculosis (pre-XDR-TB), XDR-TB, or poly-resistant TB, consistent with the clinical use of linezolid for resistant cases. These categories were defined according to the WHO definitions in use during the study period [14, 15], prior to the 2020 revision (see Table 2 footnote for detailed definitions). Failure was defined as persistent or recurrent sputum culture positivity after ≥ 4 months of linezolid treatment. This definition served as a key clinical indicator for regimen failure, prompting the treating physician to modify the therapeutic strategy. It primarily captured patients who either never achieved sustained sputum culture conversion or who experienced microbiological relapse after initial conversion.

Table 2.

Univariable analysis of risk factors for acquired resistance to linezolid

Characteristica	Groups[M (Q1, Q3)b或n (%)]		OR (95%CI)	p value
	Resistant, n = 19	Susceptible, n = 42
Age (year)	34.0 (27.0, 47.0)	48.0 (34.5, 59.5)	0.96 (0.92-1.00)	0.039
Gender/male	16 (84.2)	35 (83.3)	1.07 (0.24–4.67)	0.932
Weight (kg)	57.0 (50.0, 71.0)	60.0 (50.8, 71.0)	1.00 (0.96–1.04)	0.940
Ccr (mL/min)c	118.5 (99.7, 154.2)	111.6 (79.7, 132.6)	1.02 (1.00-1.03)	0.053
Smoking	9 (47.4)	21 (50.0)	0.90 (0.30–2.66)	0.849
Diabetes	6 (31.6)	22 (52.4)	0.42 (0.13–1.31)	0.136
Lung cavity at treatment initiation	19 (100)	36 (85.7)	—	0.999
Previous linezolid exposure	7 (36.8)	6 (14.3)	3.50 (0.98–12.48)	0.053
Number of likely effective drugsd	3.5 (2.5, 4.0)	4 (2.9, 5.0)	0.75 (0.47–1.18)	0.210
Duration of treatment (months)e	20.4 (14.9, 38.6)	12.7 (8.6, 23.6)	1.07 (1.02–1.12)	0.010
Bedaquiline exposuref	1 (5.3)	7 (16.7)	0.28 (0.03–2.44)	0.248
Baseline sputum smear acid-fast staining				0.642
Negative	1 (5.3)	4 (9.5)
1+	4 (21.1)	12 (28.6)
2+	9 (47.4)	16 (38.1)
3+	3 (15.8)	9 (21.4)
4+	2 (10.5)	1 (2.4)
Case category			1.90 (0.20-18.19)	0.580
Initial treatment	1 (5.3)	4 (9.5)
Retreatment	18 (94.7)	38 (90.5)
Diagnosis g				0.851
RR-TB	0 (0)	6 (14.3)
MDR-TB	4 (21.1)	11 (26.2)
pre-XDR-TB	10 (52.6)	14 (33.3)
XDR-TB	4 (21.1)	10 (23.8)
Others	1 (5.3)	1 (2.4)

^a: The listed information indicates the characteristics at baseline

^b: M, median; Q1, upper quartile; Q3, lower quartile

^c: Ccr was calculated using the Cockcroft-Gault formula. ^d: If the drug susceptibility testing indicates sensitivity, it is recorded as 1 for not previously used and 0.5 for use more than 1 month; if the drug susceptibility testing indicates resistance, it is recorded as 0. ^e: Duration of treatment from the time of baseline to collection of the last isolate. ^f: Use of bedaquiline for more than 1 month from baseline to the end of this study. ^g: Molecular susceptibility and phenotypic drug susceptibility, as both are inconsistent, the molecular drug susceptibility shall prevail. OR: Odds ratio. RR-TB (rifampicin resistant tuberculosis): resistance to rifampicin detected using phenotypic or genotypic methods, with or without resistance to other anti-TB drugs. MDR-TB (multidrug-resistant tuberculosis): resistance to at least both isoniazid and rifampicin. pre-XDR-TB (pre-extensive drug-resistant tuberculosis): MDR/RR-TB plus additional resistance to any fluoroquinolones. XDR-TB (extensive drug-resistant tuberculosis, pre-2020 definition): resistance to any fluoroquinolone, and at least one of three second-line injectable drugs (capreomycin, kanamycin and amikacin), in addition to multidrug resistance. Others, poly-resistant tuberculosis, resistance to two or more first-line antituberculosis drugs (except concurrent resistance to isoniazid and rifampicin)

Primary isolates from sputum specimens were confirmed as M. tuberculosis complex by p-nitrobenzoic acid/2-thiophene carboxylic acid hydrazide (PNB/TCH) growth inhibition, subjected to clinical drug susceptibility testing (DST), and cryopreserved at -80 °C. Baseline and final culture-positive isolates were revived from − 80 °C storage, revived at 37 °C for 24 h, and then inoculated onto Lowenstein-Jensen slopes (37 °C/4–6 weeks), excluding non-viable strains. Clinical/demographic data were extracted from electronic medical records. This retrospective study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Beijing Chest Hospital (Protocol no. YJS-2020-012; approved on September 16, 2020). The ethics committee granted a waiver of individual informed consent as the research involved no more than minimal risk, used existing archived bacterial isolates and completely anonymized clinical data, and obtaining consent from all participants was impracticable due to the retrospective nature of the study covering a multi-year period.

Phenotypic antimicrobial susceptibility testing

Linezolid minimum inhibitory concentrations (MICs) against M. tuberculosis isolates were determined using 7H9 broth microdilution with the Alamar Blue assay (MABA). Fresh colonies from 4 to 6 week Lowenstein-Jensen slants were suspended in saline-containing bead-beating tubes, vortexed to homogeneity, adjusted to 1.0 McFarland standard, and diluted 20-fold in OADC-supplemented 7H9 broth. A 100 µL aliquot was added to each well of the 96-well plates, pre-diluted 2-fold with linezolid. After 7–11 days of incubation at 37 °C, 70 µL of the Alamar Blue reagent (20 µL Alamar Blue + 50 µL 5% Tween-80) was added. After 24 h of reincubation, MICs were defined as the lowest concentration that resulted in blue coloration (no growth), with pink indicating growth. Each plate included M. tuberculosis strain H37Rv for quality control. The tested concentrations ranged from 0.03125 to 32.0 mg/L, with a breakpoint of > 1.0 mg/L for resistance [6]. For the purpose of summarizing the distribution and visual presentation in Fig. 2, isolates with MICs of 0.0625 mg/L and 0.125 mg/L were grouped into a single category (< 0.25 mg/L) as they all represented susceptible phenotypes below the clinical resistance breakpoint. A composite “number of likely effective drugs” score was calculated for each regimen based on pre-treatment DST results and prior drug exposure history (see Supplementary Method S1 for detailed scoring criteria).

Fig. 2.

Distribution of linezolid MIC values and associated resistance mutations among 122 Mycobacterium tuberculosis isolates. The grouped bar chart displays the frequency distribution of linezolid MICs. The category “<0.25 mg/L” comprises isolates with MICs of 0.0625 mg/L (n=1) and 0.125 mg/L (n=10). Within each MIC category, bars are color-coded to represent isolates with no detected mutation in target genes (blue) and those harboring mutations in rplC (orange-red), rrl (orange-yellow), or rplD (gray). One strain contained concurrent rplC and rplD mutations. The vertical red broken line is aligned with the center of the 1.0 mg/L category, indicating the critical concentration for linezolid resistance. MIC, minimum inhibitory concentration

DNA extraction, polymerase chain reaction (PCR) and sequencing

Genomic DNA was extracted from fresh M. tuberculosis suspensions by using a previously published crude extraction method [16]. PCR amplification was performed using this DNA as a template with the primers listed in Table S1, under the reaction conditions specified in Table S2. The amplification products were verified using agarose gel electrophoresis (Figure S1). Target bands were excised and purified, and 20 µL of the PCR product was submitted to Beijing Ruibo Xingke Biotechnology Co., Ltd., for Sanger sequencing. The returned sequences were aligned against the M. tuberculosis strain H37Rv reference genome (GenBank accession number NC_000962.3) using DNAMAN v6.0.3.99 for variant analysis. All chromatograms were manually inspected for quality and for the presence of mixed bases (e.g., double peaks) that might suggest heteroresistance.

Statistical analysis and definition of mutation events

Clinical and phenotypic data analysis

Statistical analyses were conducted using SPSS software version 26.0. Continuous variables were tested for normality. Data conforming to a normal distribution were expressed as mean ± standard deviation (mean ± SD), and comparisons between groups were performed using the t-test. For non-normally distributed data, results were presented as median and interquartile range (IQR) [M (Q1, Q3)], and group comparisons were conducted using the Mann–Whitney U test. Categorical variables were compared using the chi-square (χ²) test.

To identify risk factors for acquired linezolid resistance, univariable logistic regression was employed. Considering the limited number of acquired resistance events (n = 19), a formal multivariable logistic regression model was not constructed to avoid model overfitting and instability, as such models typically require a larger number of events per predictor variable. To ensure methodological transparency, we specify that, had the sample size been sufficient, variables with a p < 0.1 in univariable analysis would have been candidates for inclusion in a multivariable model. In such a scenario, multicollinearity among candidate variables would have been assessed using variance inflation factors (VIF), with a VIF < 2 considered to indicate the absence of severe collinearity. Statistical significance was set at p < 0.05.

Definition and genotypic analysis of mutation events

For mechanistic analysis, we enumerated “mutation events”, defined as each instance where a specific resistance-associated mutation was identified in a single isolate at a given time point. This approach captures the full spectrum of genetic evolution under antibiotic pressure, as a single patient could contribute multiple events over time (e.g., emergence of a new mutation, or persistence of an identical mutation in both baseline and follow-up isolates). The total number of mutation events (n = 26) was used as the denominator to calculate the relative distribution of mutation types (e.g., rplC vs. rrl). In contrast, the prevalence of mutations at the isolate level was calculated using the total number of isolates (n = 122) as the denominator, providing an epidemiological perspective.

Results

Distributions of linezolid Mics for clinical isolates

From January 2017 to June 2021, 1,600 patients with DR-TB received treatment at Beijing Chest Hospital. Of these, 272 patients had at least two positive sputum cultures at least four months apart, with available DST results. Seventy-nine patients received linezolid-containing regimens for more than four months, while the remaining patients had culture-positive sputum samples and baseline isolates. After excluding 10 non-viable isolates, 5 contaminated cultures, and 3 failed assays due to issues with DNA amplification or MIC determination, 61 patients with 122 paired Mycobacterium tuberculosis isolates (baseline and final) were included in the final analysis.

Patients were stratified into two groups according to acquired resistance criteria. The acquired resistance group (n = 19) included individuals who demonstrated either a greater than fourfold increase in linezolid MIC from baseline or the emergence of new resistance-associated mutations. The linezolid-susceptible group (n = 42) consisted of patients who did not meet either criterion (Fig. 1).

Fig. 1.

Flow diagram showing numbers of patients and isolates included. The diagram illustrates the stepwise inclusion and exclusion process from initial screening to final analysis, detailing reasons for exclusion at each stage. DST: drug susceptibility test. MIC: minimum inhibitory concentration

MICs of linezolid and sequencing of resistance-associated genes (rplC, rplD, rrl) were determined for all 122 isolates. The measured MICs for linezolid ranged from 0.0625 mg/L to 16.0 mg/L. For clarity of presentation, isolates with MICs of 0.0625 mg/L (n = 1) and 0.125 mg/L (n = 10) were grouped together as “<0.25 mg/L” (total n = 11, 9.0%) as detailed in the Methods. The distribution of MICs is shown in Fig. 2 and summarized as follows: <0.25 mg/L (11 isolates, 9.0%), 0.25 mg/L (30 isolates, 24.6%), 0.5 mg/L (48 isolates, 39.3%), 1.0 mg/L (11 isolates, 9.0%), 2.0 mg/L (5 isolates, 4.1%), 4.0 mg/L (11 isolates, 9.0%), 8.0 mg/L (5 isolates, 4.1%), and 16.0 mg/L (1 isolate, 0.8%).

We detected a total of 26 mutation events across the longitudinal isolates (see patient-wise breakdown in Table 1 and supplementary Table S4): 15 in rplC (predominantly T460C), 10 in rrl (G2270T/G2814T), and 1 in rplD (co-occurring with a rplC mutation). The distribution of these events among wild-type and mutated isolates at different MIC levels is visualized in Fig. 2. Using 1.0 mg/L as the resistance breakpoint, the 26 mutation events were distributed as follows: 23 events occurred within the 22 phenotypically resistant isolates (MIC > 1.0 mg/L), including one isolate (ID 261312) that harbored two concurrent mutations (rplC T460C and rplD A518T). The remaining three mutation events were found in isolates with MICs of 0.25 mg/L (n = 2) and 0.5 mg/L (n = 1). Notably, several isolates harbored mutations without high-level phenotypic resistance. This included isolates from two patients in the susceptible group who maintained low MICs (0.25 or 0.5 mg/L) across timepoints despite having rrl mutations (A2296C or A1843G) at both baseline and follow-up (Supplementary Table S4). In contrast, one isolate (ID 240306, from the resistant group) with an MIC of 4.0 mg/L showed no detectable mutations in the target genes, suggesting alternative resistance mechanisms.

Table 1.

Clinical characteristics and longitudinal mutation profiles of patients with acquired linezolid resistance (n=19)

Participant ID	MIC (mg/L)		Previously exposure to linezolid	Duration of treatment (months)b	rrl mutation	rplC mutation	rplD mutation
	Baseline	Lasta
261312	<0.25	4	No	19.6	WT	T460C	A518T
082626	0.5	4	No	39.4	G2814T	WT	WT
121925	0.25	4	Yes	7.53	WT	T460C	WT
120403	0.5	4	Yes	21.3	WT	T460C	WT
121025	0.5	4	Yes	20.33	WT	T460C	WT
240306	0.25	4	No	16.67	WT	WT	WT
262419	0.5	8	Yes	8.87	WT	T460C	WT
262625	0.5	8	No	40.47	G2270T	WT	WT
2610	0.5	8	Yes	14.87	WT	T460C	WT
232008	0.25	4	No	32.47	WT	T460C	WT
2625	0.25	2	Yes	32.93	WT	T460C	WT
171026	0.5	4	No	10.13	WT	T460C	WT
0704	1	4	No	28.4	G2270T	WT	WT
100623	0.25	8	No	14.47	WT	T460C	WT
1912	0.25	2	Yes	46.07	G2270T	WT	WT
0710	0.5	16	No	43.2	G2814T	WT	WT
230804	0.25	8	No	19.97	WT	T460C	WT
082619	0.25	4	No	20.37	G2270T; G2814T	WT	WT
1203	2	4	No	38.63	WT	G156C→G156C, T460C	WT

Note on data presentation: Each row represents one patient, with paired baseline and last isolate data presented side-by-side for MIC and for the genotype of each target gene (rplC,rplD, rrl). Acquired linezolid resistance was defined as a ≥4-fold increase in MIC or emergence of a new resistance-associated mutation in the last isolate compared to baseline

Genotype notation within each gene column: “WT” indicates wild-type sequence (no mutation detected); “Mutation name (e.g., T460C)” indicates the de novo detection of this mutation in the last isolate, when the baseline was WT; “Arrow, e.g., WT→T460C or T460C→T460C” denotes a change in the mutation status of that specific gene between the two timepoints

Note on mutation events: Patient 261312 developed two distinct mutations (T460C in rplC and A518T in rplD), counted as two mutation events. Patient 1203 had a pre-existing mutation replaced by a new one during therapy, contributing one new mutation event

^a: The last isolate collected before the end of this study

^b: Duration of treatment from the time of baseline to collection of the last isolate

Nucleotide notation: Substitutions are denoted as x#y, where x represents the WT base, # is the position relative to the M. tuberculosis H37Rv reference genome (GenBank NC_000962.3), and y represents the variant base

MIC, minimum inhibitory concentration; WT, wild-type

Mutations associated with phenotypic resistance to linezolid

Among the 19 patients in the acquired resistance group (Table 1), linezolid MIC increased 2- to 32-fold from baseline. Analysis of the 26 mutation events revealed that 20 originated from this group, while 6 events were from the susceptible group (Supplementary Table S4). Considering all 122 isolates, the mutation frequencies were quantified as follows: rplC (12.3%), rplD (0.82%), and rrl (8.2%), with the specific variants detailed in Table S3. All rplC mutations involved T460C substitutions, whereas rrl mutations consisted of the G2270T and G2814T variants.

A single patient developed concurrent mutations in rplD (A518T) and rplC (T460C), resulting in a 32-fold increase in MIC. This dual-mutation strain occurred in a retreated individual, with no prior exposure to linezolid. Another retreatment patient exhibited a baseline rplC G156C synonymous mutation (leucine preservation), with the subsequent emergence of T460C during therapy, accompanied by a 2-fold increase in MIC.

Notably, three cases demonstrated clear phenotype-genotype discordance: two patients (in the susceptible group) maintained subthreshold MICs throughout treatment despite persistent rrl mutations (A2296C/A1843G, Table S4). One strain (ID 240306, in the resistant group) showed 16-fold MIC elevation (0.25→4.0 mg/L) without target gene mutations (Table 1).

Clinical characteristics of the study population

The study enrolled 61 patients with DR-TB. The distribution of drug resistance types at baseline is summarized in Table 2. Patients were subsequently categorized as linezolid-resistant or linezolid-susceptible groups according to acquired resistance criteria. Baseline characteristics assessed comprised age, sex, body weight, creatinine clearance rate (Ccr), smoking status, diabetes mellitus, pulmonary cavitation, prior linezolid exposure, number of effective drugs, treatment duration, concomitant bedaquiline use, sputum smear bacterial load, treatment history (initial vs. retreatment), and diagnostic classification (Table 2).

The linezolid-resistant group (n = 19) was significantly younger (median 34 years, IQR 27.0–47.0) and had a longer treatment duration (median 20.4 months, IQR 14.9–38.6) compared to the susceptible group (median 48 years, IQR 34.5–59.5; median 12.7 months, IQR 8.6–23.6, respectively). Other parameters were similar between groups, including male predominance (84.2% vs. 83.3%), baseline linezolid exposure (36.8% vs. 14.3%), bedaquiline use (5.3% vs. 16.7%), number of effective drugs (median 3.5 vs. 4), and XDR-TB proportion (21.1% vs. 23.8%) (Table 2).

Univariable analysis identified two independent risk factors for linezolid resistance: each 1-year decrease in age increased the risk 1.04-fold [95% confidence interval (CI) 1.01–1.07], and each additional month of treatment increased the risk by 1.07-fold (95% CI 1.02–1.12). Receiver operating characteristic (ROC) analysis (Figure S2) determined optimal predictive thresholds: age < 37.5 years [area under the curve (AUC) = 0.672, 95%CI 0.530–0.815; sensitivity 78.6%, specificity 52.6%] and treatment duration > 14.3 months (AUC = 0.712, 95%CI 0.576–0.849; sensitivity 84.2%, specificity 59.5%). It should be noted that these optimal cut-off values are derived from our cohort and are therefore exploratory; they require external validation in future studies.

Risk factors for acquired linezolid resistance during treatment

Baseline phenotypic drug susceptibility analysis

Baseline drug susceptibility profiles for 16 anti-TB agents were analyzed across all 61 patients stratified by acquired linezolid resistance status. The hierarchical ranking of baseline resistance rates across the 16 tested anti-TB agents (from highest to lowest) is provided in Figure S3. Comparative analysis revealed no statistically significant differences in resistance rates among the 15 drugs between the groups, with kanamycin as the sole exception, demonstrating significant differential resistance (Figure S3).

Relationship between acquired linezolid resistance and other anti-tuberculosis drugs

The phenotypic susceptibility profiles of the final isolate were stratified according to the acquired linezolid resistance status. Resistance to amikacin, clofazimine, and pasiniazid differed significantly between groups (p < 0.05, Fig. 3).

Fig. 3.

Resistance of last strains to other anti-tuberculosis drugs by linezolid resistance status. Rates of resistance to additional drugs on drug susceptibility testing by the broth microdilution method. Drug Abbreviations and Drug Sensitivity Test Resistance Thresholds: LFX: levofloxacin, 2.0 mg/L; MFX: moxifloxacin, 0.5 mg/L; AM: amikacin, 1.0 mg/L; CPM: capreomycin, 2.5 mg/L; KAM: kanamycin, 2.5 mg/L; RFP: rifampicin, 1.0 mg/L; RFT: rifapentine, 0.5 mg/L; RFB: rifabutin, 0.75 mg/L; CLR: clarithromycin, 4.0 mg/L; INH: isoniazid, 0.2 mg/L; EMB: ethambutol, 2.5 mg/L; PAS: p-aminosalicylic acid, 2.0 mg/L; SM: streptomycin, 1.0 mg/L; PTO: protionamide, 10.0 mg/L; CFZ: clofazimine, 2.0 mg/L; pasiniazid, 0.5 mg/L. Rates of resistance to pasiniazid, clofazimine, and amikacin were all statistically significantly different according to linezolid resistance status (p < 0.05). *: p < 0.05; **: p < 0.01

To evaluate potential cross-resistance, a concordance analysis between linezolid and the three agents was performed (Table S5). Susceptibility to clofazimine, amikacin, and pasiniazid was positively associated with linezolid susceptibility, yielding kappa statistics (95% CI) of 0.538 (0.311–0.765), 0.355 (0.104–0.606), and 0.326 (0.154–0.498), respectively. These results indicate moderate concordance between clofazimine and linezolid, but weak associations with amikacin and pasiniazid.

Discussion

Linezolid maintains its position as a WHO-recommended cornerstone in the management of DR-TB. However, the emergence of acquired resistance poses a substantial threat to treatment efficacy. The absence of routine phenotypic and molecular DST for linezolid, coupled with the WHO’s updated definition of XDR-TB, underscores the urgent need to elucidate resistance mechanisms and risk factors. In this retrospective study of 61 patients with DR-TB who experienced treatment failure with linezolid-containing regimens (Beijing Chest Hospital, 2017–2021). The cohort was characterized by a high burden of disease, with most patients having cavitary lesions, retreatment status, and few effective drugs available. However, these factors did not differ significantly between groups (Table 2) and thus were not identified as risk factors specific to acquired resistance in this cohort. Instead, our analysis revealed that 31.1% of patients developed acquired linezolid resistance after a median exposure of 20.4 months—a rate exceeding that reported in most prior studies [17, 18]. Crucially, we identified younger age (< 37.5) and prolonged therapy duration (> 14.3 months) as novel independent risk factors, with each additional month of treatment increasing resistance risk by 7%. This likely reflects our focus on confirmed treatment failures, in which suboptimal drug exposure and accumulated resistance mutations converge.

Although linezolid possesses a high genetic barrier [19, 20], the present data indicate that prolonged or irregular administration within suboptimal regimens facilitates the development of resistance. Initial studies reported rare instances of resistance [21], while subsequent research identified reduced susceptibility, even among linezolid-naïve patients [22], particularly in strains with the Beijing genotype [16]. Notably, subtherapeutic concentrations, despite standard dosing (300–600 mg/day) [23] and regimen inadequacy [6, 7], promote resistance through two primary mechanisms: dose reduction or intermittency due to toxicity, and use within failing regimens.

The molecular mechanisms underlying linezolid resistance remain incompletely understood and involve mutations in the ribosomal targets rrl (encoding the 23 S rRNA peptidyl transferase center), rplC (encoding ribosomal protein L3), and rplD (encoding ribosomal protein L4) [6]. Our cohort, characterized by a 31.1% (19/61) rate of acquired resistance during salvage therapy, reflects the substantial risk of resistance selection within programmatic settings, a finding consistent with the spectrum (33%, 13/39) reported in similar retrospective studies [6]. At the molecular level, our findings strongly align with the prevailing global pattern where rplC mutations are the primary driver. We observed rplC mutations (57.7%, 15/26) dominated by T460C and rrl mutations (38.5%, 10/26) featuring G2270T/G2814T. This aligned with reports indicating rplC mutations (54.9% prevalence, T460C [24]) typically confer moderate MIC elevations (2–32 mg/L; 4–32-fold) [20, 25], whereas rrl mutations (35.4% prevalence [6]) are frequently associated with and can drive higher-level resistance, with MICs often reported in the range of 2–16 mg/L and sometimes reaching 8–16 mg/L [20, 26]. In our cohort, MICs for isolates with rrl mutations spanned from 0.25 to 16 mg/L, overlapping with the range observed for rplC mutants. We observed one strain with dual rplC T460C/rplD A518T mutations, although A518T was not reported in the National Center for Biotechnology Information (NCBI) and did not augment MIC (compared to rplC-only mutants). Other rplD mutations (e.g., G377A [9] and G236A [27]) have been implicated in high-level resistance (> 16 µg/mL). Notably, a synonymous rplC G156C mutation has been detected; however, its role in resistance remains unclear.

Phenotypic-genotypic discordance was observed in several cases, underscoring the complexity of resistance prediction. In the acquired resistance group, one strain (ID 240306) exhibited a significant 16-fold MIC increase without mutations in the canonical target-genes. These parallel reports where a substantial proportion (up to 50%-70%) of linezolid-resistant strains lack ribosomal mutations [16, 28], implicating non-ribosomal mechanisms such as efflux pump upregulation (potentially via Rv0678 mutations) [11, 21, 29, 30] or reduced drug permeability due to cell wall thickening [31, 32]—a mechanism particularly relevant given this patient’s pre-XDR status. Conversely, within the LZD-susceptible group, three patients harbored resistance-associated mutations yet maintained stable MICs and did not meet critieria for acquired resistance (Supplementary Table S4). Two patients with persistent rrl mutations (A2296C and A1843G, respectively) with MICs ≤ 1 mg/L. The A1843G variant is located distal to the peptidyl transferase center (PTC), the primary binding site for linezolid, which may explain its lack of association with high-level resistance. The A2296C variant, while closer to the PTC, has not been functionally characterized in M. tuberculosis; it may represent a lineage-specific polymorphism or confer minimal resistance. The third patient had a persistent rplC T460C mutation with a stable MIC of 2 mg/L, likely representing a pre-existing resistant strain without further evolution under therapy. These observations collectively highlight that the presence of a known resistance mutation does not invariably predict phenotypic resistance or treatment failure [16, 28]. The functional impact of specific variants like A2296C and A1843G requires validation through methods such as site-directed mutagenesis. This discordance necessitates cautious interpretation of genotypic data and reinforces the indispensable role of phenotypic susceptibility testing in guiding clinical decisions.

We also observed associations between acquired linezolid resistance and resistance to amikacin, pasiniazid, and clofazimine, with moderate concordance for clofazimine (kappa = 0.538), and poor concordance for amikacin and pasiniazid (kappa = 0.355 and 0.326, respectively). Although definitive cross-resistance between linezolid and non-oxazolidinone drugs has not been established, some studies noted reduced linezolid susceptibility following fluoroquinolone or kanamycin use [22], while others reported no cross-resistance with kanamycin/gentamicin, but potential overlap with chloramphenicol [19]. This may be related to shared ribosomal targets; aminoglycosides bind to the 30 S subunit, whereas linezolid and chloramphenicol bind to the 50 S subunit. Aminoglycoside resistance mutations can alter ribosomal subunit conformation and reduce linezolid affinity, whereas mutations at distinct ribosomal sites may lead to inconsistent resistance correlations. Additionally, resistance to other anti-TB drugs may upregulate efflux pump genes and decrease the susceptibility to linezolid. For clofazimine, a WHO Group B drug, its primary resistance gene, Rv0678 (encoding a transcriptional repressor of the MmpL5-MmpS5 efflux system), may play a role; mutations in Rv0678 upregulate efflux pump expression, potentially mediating resistance to both clofazimine and linezolid [33]. This is supported by findings in Mycobacterium abscessus, where low-level linezolid resistance has been linked to loss-of-function mutations in MAB_4384 (a repressor of the MmpL5-MmpS5 efflux pump) [34]. Notably, linezolid and clofazimine (WHO Group A and B drugs, respectively) have low resistance rates (0.3%–0.4%) in MDR-TB and XDR-TB populations [35], 16.4% (10/61) of our treatment-failure cohort were resistant to both. This higher prevalence may be influenced by the selective pressure within this failing regimen population and potentially by challenges associated with clofazimine phenotypic susceptibility testing, underscoring the need for further investigation.

Our analysis identified younger age (< 37.5 years) and longer linezolid treatment duration (> 14.3 months) as independent clinical predictors of acquired linezolid resistance. The correlation between longer treatment duration and heightened risk aligns with the established concept that suboptimal regimen adequacy can foster the selection of resistant mutants [6]. A key contribution of our work is the quantification of exploratory thresholds for these variables (age < 37.5 years, treatment duration > 14.3 months) via ROC analysis. It is important to emphasize that these data-driven cut-off values, and the models from which they were derived (AUC 0.672–0.712), require prospective validation in independent, larger cohorts. Furthermore, the modest AUC values indicate that age and treatment duration alone are insufficient for precise individual risk prediction and likely need to be integrated with other clinical, microbiological, or pharmacokinetic markers. The specific mechanisms underlying the association with younger age cannot be eluciated from our retrospective data, underscoring the need for future prospective studies that include data on drug exposure, tolerability, and adherence to clarify this relationship.

It is important to contextualize these findings within the specific study population. Our cohort was intentionally selected to investigate acquired resistance and comprised patients who had failed a linezolid-containing regimen after prolonged exposure (median > 14.3 months) and for whom paired isolates were available. Therefore, the risk factors and resistance mechanisms identified are most applicable to this specific scenario of treatment failure. Extrapolation of these results to all DR-TB patients starting linezolid, particularly those receiving modern, shortened all-oral regimens (e.g., BPaL/BPaLM) with different durations, efficacy profiles, and companion drugs, should be done with caution. The risk of acquired resistance in optimized, shorter-course regimens may differ and requires prospective evaluation.

This study has several limitations. As this was a retrospective analysis based on medical records, there may have been bias in documenting treatment history, particularly drug exposure duration. Strain-preservation issues led to the exclusion of many patients due to failed revival or loss of isolates, and this necessary exclusion could introduce selection bias if the non-viable strains differed systematically from those analyzed, potentially affecting the generalizability of our resistance estimates. Testing only the baseline and final isolates may have missed the development of intermediate resistance. Our genetic analysis targeted the rplC, rplD, and 23 S rRNA (rrl) genes using Sanger sequencing. Manual review of the chromatogram did not reveal evidence of mixed bases at these loci. However, Sanger sequencing has a limited detection threshold for minor alleles (typically > 10–20% frequency). Therefore, the observed phenotype-genotype discordance could be explained by resistance mechanism outside the sequenced regions, by very low-frequency variants below our detection limit, or by non-ribosomal mechanisms. In particular, the isolate (ID 240306) with high-level phenotypic resistance (MIC 4.0 mg/L) in the absence of canonical target-gene mutations represents a key candidate for such whole-genome sequencing analysis. Future studies employing deep next-generation sequencing are needed to fully investigate these possibilities.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Wenqiang Zhou, Wenjie Qi, Yu Lu, and Naihui Chu conceptualized the study and revised the manuscript. Wenqiang Zhou collected clinical data, performed statistical analyses, conducted the experiments, and drafted the manuscript. Qingfeng Wang, Wenjuan Nie, Wenhui Shi, and Yang Yang supervised the project and reviewed the entire manuscript. All the authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Key Clinical Specialist Construction Programs, the High-level Public Health Technology Talent Construction Project (Academic Leader-02-29), the National Key Research and Development Program (2024YFC2311203); Special Project of Guangzhou National Laboratory for 2024- Short Term Treatment of Tuberculosis (GZNL2024A01030); Global Tuberculosis Drug Research and Development Alliance Cooperation Project of China Anti-TB Association (CATA/TBA-20210310HQ7); Beijing Clinical Medicine Development Special Project “Sailing 3.0” (ZLRK202331); and Beijing Municipal Health Commission Research Ward Excellent Clinical Research Program (BRWEP2024W042160109).

Data availability

The datasets analyzed in the current study are available from the corresponding author upon reasonable request, subject to approval by the Ethics Committee of Beijing Chest Hospital and the execution of a formal data-sharing agreement.

Declarations

Ethics approval and consent to participate

This retrospective study was approved by the Institutional Review Board of Beijing Chest Hospital (protocol no. YJS-2020-012; approved on September 16, 2020) and was conducted in accordance with the Declaration of Helsinki. The ethics committee granted a waiver of individual informed consent as the research involved no more than minimal risk and used existing archived bacterial isolates and completely anonymized clinical data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.