Comparative transcriptomic profiling of pyrazinamide resistance in Mycobacterium tuberculosis

Abstract

Aims:

Pyrazinamide (PZA) resistance in Mycobacterium tuberculosis challenges tuberculosis management, primarily through pncA mutations, while broader transcriptional adaptations remain unclear. This study aimed to characterize the transcriptomic adaptations associated with PZA resistance using a PZA-resistant clinical strain carrying a 10-nucleotide deletion in pncA (positions 118–127) that abolishes PZA activation (PZA^R), in comparison with drug susceptible laboratory strain H37Rv under PZA treated (RvT) and untreated (UTRv) conditions.

Methods and results:

Strain specific PZA concentrations was established at 200 μg mL⁻¹ for the PZA-resistant strain (PZA^R) and 12.5 μg mL⁻¹ for the H37Rv strain (RvT). Untreated H37Rv strain was used as a reference for comparison. RNA-sequencing identified 3,413 differentially expressed genes (Padj≤0.05), including 1,428 upregulated and 1,360 downregulated genes, while the remaining 625 genes showed moderate but statistically significant expression changes. Functional enrichment was most pronounced in PZA^R vs RvT comparison, followed by PZA^R vs UTRv, whereas no significant enrichment was observed in RvT vs UTRv, indicating a strong association between the pncA mutation and PZA-responsive transcriptomic profiles. Ribosomal machinery genes (rplC, rplD, and rpsH) were significantly enriched and strongly upregulated in the resistant strain under treatment but only mildly regulated in the laboratory strain. Several anti-TB drug targets (katG, ethA, atpE, panD) were downregulated, while efflux pump genes (Rv1258, Rv3008, Rv3756c) were upregulated, reflecting a coordinated transcriptional response across drug targets. Network analysis identified 19 gene clusters, with prominent modules comprising polyketide synthases, phthiocerol dimycocerosate (PDIM) synthesis genes, fatty acid β-oxidation enzymes and ESAT-6 (ESX) secretion system.

Conclusion:

These findings uncover mutation-associated and PZA-responsive transcriptomic signatures that reveal adaptive pathways involved in tolerance under drug pressure and provide a framework for future functional and therapeutic investigations.

Introduction

Tuberculosis (TB) remains a global health concern in Mycobacterium tuberculosis, with an increasing burden of multidrug-resistant (MDR) and extensively drug-resistant (XDR) TB (Alqaseer et al. 2019). Among the first-line TB regimens, pyrazinamide (PZA) is vital for eradicating semi-dormant bacilli and reducing the duration of TB therapy (Chauhan et al. 2023). However, the MDR-TB cases frequently exhibit PZA resistance (Wang et al. 2023), diminishing its clinical efficacy and requiring further research on the resistance mechanisms. Approximately 70–98% of PZA resistance is attributed to pncA mutations, which encode the pyrazinamidase (PZase) enzyme that converts PZA into pyrazinoic acid (POA), the active bactericidal form (Rajendran et al. 2025). Once converted to POA, it interferes with key cellular functions, including membrane energetics, intracellular pH regulation, and coenzyme A metabolism, leading to reduced bacterial survival under acidic and stress conditions. PZA exposure has also been associated with altered ribosomal activity and activation of stress response pathways, indicating a multi-target mechanism of action (Lamont, Dillon and Baughn 2020). However, increasing evidence indicates that pncA mutations alone do not fully explain PZA resistance, and that multiple alternative mechanisms also contribute to this phenotype (Rajendran and Palaniyandi 2022). Several studies have suggested that broader transcriptional regulators, general stress responses, and metabolic adaptations underlie the multifaceted nature of drug resistance in M. tuberculosis (Thiede et al. 2022; Hailemariam et al. 2025).

Transcriptomic and other omics-based approaches have therefore been increasingly employed to investigate mycobacterial responses under physiologically-relevant antibiotic exposures (Poonawalah et al. 2024). For instance, rifampicin-resistant strains exhibit extensive transcriptional remodelling associated with rpoB mutations and compensatory regulatory mechanisms. (De Knegt et al. 2013). Similarly, capreomycin resistance driven by loss or mutation of the tlyA gene has been linked to altered stress response pathways and protein biosynthesis (Zhao et al. 2019). Distinct transcriptional signatures involving lipid metabolism, efflux mechanisms, and cofactor metabolism have been reported in MDR strains exposed to fluroquinolone and ethionamide (Tang et al. 2020). While, altered expression of Rv3094c and downregulation of type VII secretion system components have been linked to ethionamide metabolism (Wan et al. 2023). Genome-wide transcriptional analysis of streptomycin-resistant strains have further revealed substantial changes in transcriptional regulation and cell wall-related processes (Wu et al. 2024). Consistent with the broader view of drug resistance, prior gene expression and computational studies have implicated a multifactorial PZA resistance landscape involving pncA, rpsA mutations affecting ribosomal interactions with POA, putative efflux mechanisms and metabolic pathways linked to panD mutations. (Sheen et al. 2013; Carter et al. 2024).

Building on this multifactorial view of PZA resistance, a recent study demonstrated the activation of the sigE-dependent cell envelope stress response plays a vital role in regulating PZA susceptibility, implying that sigma factor-mediated stress responses directly influence drug efficacy (Thiede et al. 2022). Mycobacterium tuberculosis encodes 13 sigma factors, more than any other obligate pathogen per megabase of the genome, facilitating highly dynamic transcriptional responses to environmental and chemical stressors. While sigA is essential for viability, alternative sigma factors are selectively activated under specific conditions, including antibiotic exposure (Rodrigue et al. 2006). Antimicrobial treatment induces stress responses in bacteria, and the magnitude of stress imposed by antibiotics has been shown to correlate with their bactericidal efficacy (Kohanski et al. 2007). Consequently, sigma factors are likely to play a key role in mediating both innate and acquired drug resistance by regulating the adaptive transcriptional programs that promote survival under drug-induced stress (Thiede et al. 2022).

Collectively, these studies highlight that drug resistance in M. tuberculosis is dependent on the treatment and characterized by extensive transcriptional remodeling and stress induced adaptative responses. However, the specific transcriptomic consequences of pncA-mediated PZA resistance remain poorly characterized. Understanding how pncA mutations alter gene expression under drug pressure is therefore critical for identifying compensatory and adaptive pathways that support resistance. In this study, we compared the transcriptome profiles of a PZA treated pncA mutant PZA^R strain and a susceptible laboratory strain, H37Rv. The PZA^R strain identified in our previous study belongs to lineage1 and harbors a deletion of 10 nucleotides in pncA gene (position 118–127) which disrupts pyrazinamidase production and confers PZA resistance, providing a clinically relevant model to investigate transcriptional adaptations associated with pncA mediated resistance under drug pressure (Rajendran et al. 2025). Downstream transcriptional analysis including differential gene expression (DEG) analysis, functional enrichment (FE), protein-protein interaction (PPI) network construction and pathway analysis were performed to elucidate the key genetic and regulatory mechanisms underlying PZA resistance and to link transcriptomic alterations to phenotypic drug tolerance.

Materials and methods

Growth and drug treatment of M. tuberculosis strains

Monoclonal cultures of the PZA^R strain and the laboratory strain H37Rv were used as the parental strain (P0) and were initially grown on drug-free 7H11 agar with fetal bovine serum (FBS) instead of oleic acid-albumin-dextrose-catalase (OADC) to support the substantial growth of M. tuberculosis at a lower pH. The medium was maintained at an acidic pH to retain the PZA activity which differs from the conventional pH of 6.8 (Heifets and Sanchez, 2000). These acidic 7H11 conditions were chosen to enable sustained PZA activity during the in vitro exposure.

The cultures were incubated at 37°C for 4 weeks to allow complete growth. After, the cultures were transferred to 7H11 agar containing PZA (Sigma-Aldrich,USA) at concentrations of either 6.25 μg mL⁻¹, 12.5 μg mL⁻¹, 25 μg mL⁻¹, 50 μg mL⁻¹, and 100 μg mL⁻¹ for the laboratory strain H37Rv and 100 μg mL⁻¹ and 200 μg mL⁻¹ for the PZA^R strain. After four weeks of incubation at 37°C, growth was assessed by observing colony formation. The highest concentration of PZA that allowed visible colony growth for each strain was defined as the critical selection concentration and was used for strain selection, rather than minimum inhibitory concentration (MIC) determination. Colonies exhibiting visible growth at or above this concentration were isolated and designated as P1 strains. These strains were further subcultured onto a fresh agar plate containing PZA to prevent the loss of drug activity during prolonged incubation. The subcultured plates were incubated at 37°C for 4 weeks to generate P2 strains. The process was sequentially repeated through P3 and P4 stages with each passage lasting four weeks at 37°C. For transcriptomic analysis, the P4 cultures were standardized to McFarland 0.5, to ensure uniform growth and RNA quality. P4 strains demonstrating sustained growth at their respective critical concentrations were prepared in triplicates for RNA extraction. The PZA-treated resistant clinical strain was designated PZA^R, the PZA-treated H37Rv strain as RvT. The untreated H37Rv transcriptome (UTRv) used for comparative analysis was obtained from a previously published dataset (Li et al. 2020) in which H37Rv was cultured on Lowenstein Jensen medium under standard conditions.

Transcriptome sequencing and analysis

Total RNA was extracted from three biological replicates for each strain with established protocols (Oh et al. 2016). RNA purity and quantity was estimated after the DNase I treatment using an Agilent 2100 Bioanalyzer. Strand specific cDNA libraries were generated using Roche’s KAPA RNA Hyper Prep Kit as per the manufacturer’s instruction. The library quality was checked with an Agilent Bioanalyzer 2100 system and sequencing was performed in Illumina NovaSeq 6000 system. The raw read RNA-seq quality was assessed using FastQC, followed by low-quality bases and adapter sequence removal using trimming and filtering with FastP (Chen 2025). The processed reads were aligned to the reference genome (NC_000962) using HISAT2 v2.2.6 (Kim et al. 2019). The resulting BAM file was visualized using the Interactive Genomics Viewer (IGV) to confirm the presence of the previously identified pncA mutation in the aligned RNA-seq reads.

Comparative transcriptome analysis

We have performed comparative transcriptomic analysis using RvT and PZA^R strains along with untreated control strain (UTRv) obtained from previous study (Li et al. 2020). The preliminary analysis and alignment were uniformly performed for all the datasets. Gene-level raw read counts were generated using feature counts (v 2.0.3) from the Subread package (Liao, Smyth and Shi 2014) and used as input for differential expression analysis with DESeq2 (v1.40.1) (Love, Huber and Anders 2014). DESeq2 was exclusively applied to raw counts for statistical testing and genes with an adjusted p-value (FDR) <0.05 and log₂ fold change >1 and < −1 were considered significantly differentially expressed. Normalized expressions were quantified as Fragments per kilobase of transcript per million mapped reads (FPKM) and were used only for visualization purposes, including the generation of heatmaps. Visualization of DEGs was performed using the enhanced volcano package for volcano plots (https://bioconductor.org/packages/release/bioc/html/EnhancedVolcano) and pheatmap package for clustered heatmaps (https://cran.r-project.org/web/packages/pheatmap) in R. Principal component analysis (PCA) and sample-to-sample distance plots were employed for data exploration and quality assessment.

Functional enrichment analysis

Differentially expressed genes from each pairwise comparison were used as an input gene list for functional enrichment. All genes detected in the RNA-seq dataset were used as the background set. Enrichment analysis were carried out using the DAVID (Database for annotation, visualisation and integrated discovery) Bioinformatics Resources 2021, (https://davidbioinformatics.nih.gov/tools.jsp) with M. tuberculosis H37Rv as the reference (Huang, Sherman and Lempicki 2009a, 2009b). The annotation categories, including Biological Process (BP), Molecular Function (MF), Cellular Component (CC) and KEGG pathways (KP) were analysed. Significance was assessed using the modified Fisher’s exact test and Benjamini-Hochberg FDR correction. Categories with FDR<0.05 were considered significantly enriched.

Protein-protein interaction network interaction

All the DEGs enriched in the INTERPRO domain of DAVID were used as inputs for the STRING database (version 12.0, http://string-db.org) (Szklarczyk et al. 2023). Interactions were retrieved at a high confidence score of 0.7 and medium FDR stringency. Networks were imported into Cytoscape (version 3.10.2) for visualization and analysis (Otasek et al. 2019). Clusters were identified using the MCODE plugin in Cytoscape with the default parameters. The resulting subnetworks were examined for their functional enrichment. Nodes were coloured according to the log₂ fold change values and the edge thickness reflected the combined score.

Results

Characteristics of the strains and determination of critical drug concentration

The critical concentration of PZA for PZA^R was 200 μg mL⁻¹ and 12.5 μg mL⁻¹ for RvT. The strains were grown in triplicates and RNA was extracted from each separately. RNA sequence from drug-free H37Rv strain was retrieved from a publicly available dataset (PRJNA579441), designated as UTRv was used a drug free control for comparative analysis (Li et al. 2020). Details of all the strains used in the study are listed in Table 1. IGV inspection of RNA-seq alignments confirmed a deletion at the pncA locus (2289115 – 2289124) was confirmed in all resistant triplicates, whereas no mutation was detected in treated and untreated H37Rv laboratory strains.

Table 1.

Characteristics of strains used in this study. This table summarizes the strain type, PZA treatment conditions, and pyrazinamide susceptibility profiles included in the transcriptomic analysis.

Strains	Condition	PZA phenotype	Critical PZA concentration (μg mL−1)	Key genotype
PZA treated resistant clinical strain _1	PZAR1	Resistant	200	pncA deletion (118–127)
PZA treated resistant clinical strain _2	PZAR2	Resistant	200	pncA deletion (118–127)
PZA treated resistant clinical strain _3	PZAR3	Resistant	200	pncA deletion (118–127)
PZA treated H37Rv_1	RvT1	Susceptible	12.5	Wild type
PZA treated H37Rv_2	RvT2	Susceptible	12.5	Wild type
PZA treated H37Rv_3	RvT3	Susceptible	12.5	Wild type
Untreated H37Rv_1	UTRv1 (Li et al. 2020)	Susceptible	Not applicable	Wild type
Untreated H37Rv_2	UTRv2 (Li et al. 2020)	Susceptible	Not applicable	Wild type
Untreated H37Rv_3	UTRv3 (Li et al. 2020)	Susceptible	Not applicable	Wild type

Comprehensive transcriptome profiling across the three groups

Detailed sequencing and alignment metrics is given in the Supplementary table 3. The overall read alignment to the reference genome was consistently high for UTRv (98.01%, 97.49% and 97.79%), moderate for RvT (77.16%, 74.35% and 73.29%) and more variable for PZA^R (74.29%, 47.26% and 97.74%) likely reflecting lineage associated sequence divergence. Despite these differences, all the samples yielded sufficient read depth for downstream analysis. Moreover, the PCA of the normalized expression data revealed a clear separation of the three experimental groups along PC1 (75% variance) and PC2 (23% variance) together capturing 98% variability in the dataset (Fig. 1A). Replicates from the same group clustered tightly, indicating high reproducibility, whereas distinct groups were clearly separated reflecting the major transcriptional differences induced by drug treatment. The sample-to-sample Euclidean distance heatmap further supported these distinctions, with shorter distances observed within biological replicates compared to between groups confirming high similarity in transcriptomic patterns within groups and marked divergence between groups (Fig. 1B). The separation across groups confirmed consistent transcriptomic differences suitable for further differential expression and pathway analysis.

Figure 1.

Transcriptomic variations among the experimental groups. (A) Principal Component Analysis of normalized expression data from the UTRv (blue), RvT (green), and PZA^R (red) groups illustrating clear transcriptional separation and consistency among replicates. (B) The pairwise distance between the conditions is depicted in the Euclidean heatmap. Darker shades indicate high similarity while the lighter shades indicate dissimilarity.

Drug induced transcriptional changes in the laboratory strain

To investigate the transcriptional response of the RvT strain upon drug exposure, RNA-seq was used to compare its gene expression profiles with those of the UTRv. Differential expression analysis revealed a total of 2694 DEGs including 788 upregulated genes and 840 downregulated genes upon drug treatment, while 1067 genes were mildly regulated (−1< log₂ foldchange < 1) but yet significant (Padj≤0.05). The overall distribution of DEGs along with the top 10 most upregulated and downregulated genes is illustrated in the volcano plot (Fig. 2A), highlighting a significant increase in the expression of genes such as Rv3229 (desA3) and Rv2947c (pks15), which are involved in lipid biosynthesis, and Rv2591(PE_PGRS44), a member of the PE/PPE family, which is implicated in host-pathogen interactions and immune modulation, while Rv2988c (leuC) encodes a probable 3-isopropylmalate dehydratase, which is important for amino acid biosynthesis. Several genes, such as Rv2625c, Rv2623, Rv1737c, Rv2032, Rv2628, and Rv2627, which are essential for M. tuberculosis survival under hypoxia, nitric oxide, and other conditions that induce dormancy, were significantly downregulated (Chen et al. 2013). Hierarchical clustering of the top 50 DEGs clearly distinguished treated from untreated samples (Fig. 3A) highlighting a substantial transcriptional reprograming in response to drug exposure.

Figure 2.

Volcano plots showing DEGs in three pairwise comparisons, A. RvT vs UTRv, B. PZA^R vs UTRv and C. PZA^R vs RvT. The x-axis shows log₂ fold change and y-axis shows −log₁₀ adjusted p-values. Horizontal and vertical dashed lines represent the thresholds for the adjusted p-value (0.05) and log₂ fold change (±1), respectively. Each point represents an individual gene, with red indicating significantly upregulated genes, blue indicating significantly downregulated genes, and grey indicating non-significant genes. The top 10 DEGs based on statistical significance are labelled.

Figure 3.

Heatmaps of top 25 upregulated and 25 downregulated differentially expressed genes across conditions: Normalized expression values were row wise z-score scaled to highlight relative expression patterns across conditions. Rows represent genes and columns represent sample replicates. The heatmap visualizes direction and consistency of regulation rather than variability among replicates. The colour gradient indicates relative expression levels with red representing higher and blue representing lower expression.

Combined effect of mutation and drug treatment on PZA^R strain

In order to explore the dual effect of the mutation and drug pressure, the gene expression profiles of PZA^R and UTRv were compared. 2706 genes were differentially expressed including 832 upregulated genes and 827 downregulated genes and 1047 mildly regulated but significant genes as depicted in the volcano plot (Fig. 2B). Among the top 10 genes were Rv1483 (fabG1), a β-ketoacyl-ACP reductase involved in mycolic acid biosynthesis; Rv1131 (prpC), involved in propionate detoxification; Rv3323c (moaX) and Rv3324c (moaC3), involved in molybdenum cofactor biosynthesis; and Rv3830c (tetR) transcriptional regulatory proteins were significantly upregulated. This suggests that the PZA^R strain responded to drug treatment by activating pathways related to cell wall maintenance, stress adaptation, and metabolic detoxification. Genes from the DosR regulon, such as Rv2623, Rv2626c, Rv2031c, and Rv1737c, which are known to be essential for dormancy and adaptation to hypoxia and nutrient starvation stress (Chen et al. 2013; Gao et al. 2019) and some conserved hypothetical genes such as Rv2625, Rv2159c, and Rv1738 were significantly downregulated. The heatmap (Fig. 3B) depicts the hierarchical clustering of the top 50 DEGs based on gene-wise normalized expression values, grouping genes and samples according to expression similarity. This analysis enabled the visualization of coordinated transcriptional patterns and sample-level separation between conditions. The PZA-treated resistant strain displayed a distinct transcriptional profile featuring suppression of stress adaptation pathways and activation of metabolic and repair-related genes suggesting a compensatory shift that may mediate PZA resistance in M. tuberculosis.

Distinct transcriptional responses of the PZA^R strain and laboratory strain to drug treatment

The gene expression profiles of PZA^R and RvT were compared to examine differences in transcriptional responses to drug exposure between the resistant and control strains. The comparison revealed a set of 566 upregulated genes, 487 downregulated genes and 1113 mildly regulated genes comprising a total of 2166 DEGs. The overall expression of the DEGs showing top 10 differentially expressed genes and the hierarchical clustering of the top 50 regulated genes are represented in the volcano plot (Fig. 2C) and heatmap (Fig. 3C) respectively. Notably, significant genes such as Rv1483 and Rv1484 (mycolic acid biosynthesis and cell wall maintenance), Rv2036 (stress adaptation), Rv3839 (conserved hypothetical protein), and Rv1038c (ESAT-6 family genes) were highly expressed. Conversely, genes including Rv3487c (lipid metabolism and host interaction), Rv2077c (cell wall/cell process protein), Rv1225 and Rv2161c (putative regulatory proteins), and Rv0842 (efflux pump linked to drug resistance) were downregulated.

Unique and shared DEGs across the three groups

To identify shared and condition-specific transcriptional responses across the three pairwise comparisons, we performed an overlap analysis of DEGs using Venn diagrams. A total of 3413 (Padj≤0.05) DEGs were identified across three comparisons. However, when the overlaps were examined separately for strongly upregulated (log₂FC ≥ 1 & Padj≤0.05) and downregulated (log₂FC ≤ −1 & Padj≤0.05) genes, we observed 1428 DEGs were upregulated and 1360 DEGs were downregulated and the remaining 625 genes showed mild to moderate changes in expression. Analysis of the unique and shared DEGs revealed both specific and overlapping transcriptional responses. PZA^R vs RvT exhibited 522 unique DEGs (294up, 228down) reflecting strain-specific and resistance associated transcriptional differences under PZA pressure, whereas 254 unique DEGs (129up, 125down) were observed in PZA^R vs UTRv representing gene expression alterations driven by the combined effect of PZA exposure and resistant strain background and 607 DEGs (302up, 305down) uniquely in RvT vs UTRv indicating expression changes attributable to drug treatment on the control strain. A core set of 147 DEGs (55up, 92down) was found across all three comparisons, representing a shared transcriptional response to mutations and treatment conditions (Fig. 4A). In line with the known mutation of pncA (Rv2043c) in the PZA^R strain, pncA transcripts were significantly downregulated in PZA^R compared with both RvT (log₂FC=−1.06) and UTRv (log₂FC = −1.19) with Padj≤0.05. No significant differential expression of pncA was observed in RvT vs UTRv comparison serving as an internal biological control, supporting the validity of the differential expression analysis and confirming the transcriptional impact of the truncated pncA at the RNA level.

Figure 4.

Venn diagram showing shared and unique genes across the three groups: The panel A (Total DEGs) of the Venn diagram depicts all genes meeting the initial significance threshold. Panel B and C illustrate the subsets of upregulated and downregulated DEGs filtered based on fold change criteria. Each circle represents the set of DEGs from one comparison. Overlapping regions indicate genes shared between two comparisons, while non overlapping areas represent unique genes specific to each comparison.

Between PZA^R vs RvT and PZA^R vs UTRv, 384 DEGs (217up, 167down) were shared likely reflecting mutation-specific gene expression signatures that persist even in the presence of drug pressure. PZA^R vs UTRv and RvT vs UTRv shared 874 DEGs (431up, 443 down) representing drug-responsive genes common to both the drug treated laboratory strain and resistant strain compared to the untreated control, reflecting conserved core pathways activated by PZA treatment. No genes met the criteria for significant up or down regulation between PZA^R vs RvT and RvT vs UTRv (Fig. 4B and 4C). This indicates that the 481 shared DEGs observed in the total DEGs were either upregulated or downregulated in either one of the comparisons rather in both or mildly regulated in at least one of the comparisons (−1 ≤ log₂FC ≥1) despite being statistically significant (Padj≤0.05). This pattern reflects subtle but consistent transcriptional adjustments rather than strong directional shifts indicating variable regulation across conditions that reflect both common and specific responses.

Biological features in functional enrichment and KEGG pathway

To identify the biological processes and pathways associated with DEGs, functional enrichment and pathway analysis were performed separately for each pairwise comparison using DAVID. Both the analyses revealed a variable set of functionally enriched genes in PZA^R vs RvT, followed by PZA^R vs UTRv, while no enrichment was observed in RvT vs UTRv. Among all the enriched pathways and gene ontology (GO) identified, only one KEGG pathway and two molecular function terms reached statistical significance (FDR<0.05) detected only in PZA^R vs RvT comparison. Specifically, significant enrichment was observed for the KEGG pathway mtu03010: Ribosome pathway (FRD = 0.0065; 50 genes) and the GO molecular function term structural constituent of ribosome - GO:0003735 (FDR = 0.009; 50 genes), while the term rRNA binding – GO:0019843 showed borderline significance (FDR = 0.051; 31 genes). In addition to these, several other KEGG pathways and GO terms including GO:0006412; GO:0001666; GO:0006633; GO:0051607 were enriched at nominal p-values but did not reach the FDR<0.05 threshold. These have been listed in the Supplementary table 1A–1F to provide a complete overview of all enrichment results. To visualize the distribution of the enriched pathways, a bubble plot was generated (Fig. 5), where each bubble represents an enriched pathway, bubble size reflects the number of genes involved and bubble colour indicates the FDR-adjusted significance level.

Figure 5.

Functional enrichment analysis of differentially expressed genes in all three groups. The bubble plot depicts significantly enriched GO terms including (A) Biological Process, (B) Cellular Component and (C) Molecular Function along with (D) KEGG pathways associated with DEGs. Enrichment analysis identifies functional categories that are overrepresented among DEGs. The colour gradient represents the statistical significance (−log₁₀ p-value) of enrichment, while the bubble size corresponds to the number of genes associated with the respective GO term or pathway.

Insights from the protein-protein interaction network

Among the 778 DEGs enriched in INTERPRO domains, 448 protein-coding genes were identified and used to construct a PPI network in STRING (v12.0) at high confidence threshold (> 0.7). The resulting network comprised 445 nodes and 1286 edges, with an average node degree of 5.78 and an average local clustering coefficient of 0.44. The observed PPI network showed significant enrichment (p-value < 1.0e-16), indicating that the proteins are functionally connected rather than randomly associated. MCODE clustering of the PPI network identified 19 distinct clusters, representing densely interconnected protein modules. Cluster 1 was the largest containing 25 genes followed by cluster 2 with 15 genes, cluster 4 with 14 genes, cluster 12 with 12 genes, cluster 3 with 10 genes and cluster 6 with 6 genes. The remaining clusters contained only five or fewer genes. Functional annotations of these clusters revealed enrichment of multiple biological processes, cellular component, molecular functions and KEGG pathways. Some genes were assigned to a single GO term while others were annotated to several terms resulting in overlaps among functional categories (Supplementary table 2A–2C). To visualize the expression trends and interaction confidence, nodes were coloured according to their log₂ fold change values and the edge thickness was scaled according to the combined STRING score (Fig. 6A–6B).

Figure 6.

Protein–protein interaction network and functional gene clusters. A high-confidence PPI network was generated from protein-coding DEGs enriched in INTERPRO domains using STRING v12.0 with a confidence score greater than 0.7, identifying a total of 19 gene clusters. (A) The most densely connected and biologically relevant clusters (Clusters 1–4), representing core functional modules involved in lipid metabolism, ESX secretion systems, and MCE transport operons. (B) Smaller but functionally coherent clusters (Clusters 5–12), representing additional metabolic, regulatory, and stress-associated interactions. Details of the remaining clusters with limited connectivity are provided in Supplementary table 2A–2C. Nodes are coloured according to log₂ fold change, with red indicating upregulated genes and blue indicating downregulated genes, and edge thickness reflects interaction confidence.

DISCUSSION

This study provides comprehensive examination of transcriptomic alterations associated with pncA-mediated PZA resistance in M. tuberculosis. We differentiated the transcriptomic profiles of a PZA^R strain harboring a 10-nucleotide deletion in pncA and the drug-susceptible laboratory strain H37Rv and identified distinct transcriptional signatures that provide insight how the resistant strain adapted to PZA stress. The 10-nucleotide deletion in pncA results in a frameshift, which is predicted to produce a truncated PncA protein, leading to loss of pyrazinamidase activity and PZA resistance (Karmakar et al. 2018). To our knowledge, this is the first study to report a genome wide transcriptome profile of a PZA^R clinical isolate of M. tuberculosis using a high-throughput RNA-sequencing. This analysis reveals a broad transcriptional landscape encompassing coordinated changes in the metabolic, stress response, and ribosomal pathways. Collectively, these data provide a high-resolution view of transcriptional reprogramming in M. tuberculosis under the combined influence of pncA mutation and PZA exposure. In the PZA^R vs UTRv comparison the resistant mutant demonstrated comprehensive remodelling, characterised by increased expression of metabolic and cell envelope biosynthetic pathways accompanied by repression of genes associated with dormancy and virulence. Although differences in growth conditions between datasets must be considered, this expression pattern is consistent with a shift toward a more metabolically active state rather than a drug-induced dormant response potentially reflecting an adaptive strategy that supports survival under sustained drug pressure.

The PZA susceptible RvT strain produced only modest transcriptional shifts, with few genes meeting the significance threshold and no functional categories passing the enrichment cut-offs (RvT vs UTRv). This indicates that short-term drug treatment of the laboratory strain mainly elicits a general stress response without activating specific adaptive pathways. In contrast, the PZA^R strain exhibited a pronounced and distinct transcriptional response relative to both UTRv and RvT strains. Genes upregulated in PZA^R strain were enriched for lipid metabolism, mycolic acid biosynthesis and molybdenum cofactor synthesis, indicating activation of pathways associated with cell envelope maintenance, metabolic adaptation, and detoxification under PZA challenge. Simultaneously, analysis of the PZA^R vs UTRv comparison showed uniform downregulation of all 48 curated DosR regulon genes, along with reduced expression of selected virulence factors, phospholipases, efflux-related genes, and other dormancy-associated genes. This coordinated suppression is consistent with reduced dormancy-associated transcriptional activity and suggests a shift away from dormancy-associated programs in PZA^R strain under drug pressure. Collectively, these transcription patterns align with previous observations from MDR and XDR M. tuberculosis strains, which frequently exhibit remodelling of lipid metabolism, cell wall dynamics, and stress responses during antimicrobial pressure (Chen et al. 2013; Gao et al. 2019; Joshi, Kandari and Bhatnagar 2021; Hailemariam et al. 2025).

Functional enrichment analysis demonstrated a clear enrichment pattern in the PZA^R vs RvT, followed by weaker but related signals in the PZA^R vs UTRv. In contrast, no KEGG pathways or GO terms reached significance in the RvT vs UTRv comparison, indicating that PZA treatment of the laboratory strain alone did not elicit a coordinated functional response. Among the enriched terms in PZA^R vs RvT, the KEGG ribosome pathway (mtu03010) and the GO molecular function structural constituent of the ribosome (GO:0003735) were the most significant, with rRNA binding (GO:0019843) narrowly missed the significant threshold. Together, these results suggest altered regulation of ribosomal and translational components in the resistant strain under drug pressure, consistent with the transcriptional adaptation of protein synthesis-related pathways. Approximately 50 genes associated with the ribosomal machinery were found to be strongly enriched across the mtu03010, GO:0003735, and GO:0019843 categories, representing the core ribosomal protein set. Most of these genes overlapped between ribosome-related terms, while a small number of additional rRNA-binding genes, Rv2364, Rv2925, Rv3462, Rv1165, and Rv2477, were uniquely captured identified under the rRNA-binding category. Several ribosome genes including Rv0682, Rv0700, Rv0701, Rv0702, Rv0704, Rv0708, Rv0715, Rv0718, Rv1037, and Rv1038 were strongly upregulated, in PZA^R compared to both RvT and UTRv, whereas Rv1197 and Rv1198 were downregulated in both comparisons. Other genes, such as Rv0056 and Rv2347 were upregulated, and Rv1642 and Rv3874 were downregulated in PZA^R vs RvT but not in PZA^R vs UTRv. Differences in gene level regulation between these two comparisons reflects the distinct biology contexts captured. PZA^R vs RvT primarily reflects mutation associated transcriptional response under drug treatment, whereas PZA^R vs UTRv, represents the combined effects of mutation and drug pressure resulting in comparatively attenuated regulation of some ribosomal components.

Ribosomal machinery components are established targets for many anti-tuberculosis drugs (Tam et al. 2019; Kumar, Sharma and Kaushal 2021; Zahra et al. 2023). Ribosomal subunits within PPI networks are therefore considered promising targets for novel antibiotic development, owing to fundamental structural differences between bacterial and human ribosomes (Lin et al. 2012). Notably, the ribosomal protein genes Rv0702 (rplD) and Rv0701 (rplC) are jointly inhibited by oxazolidinone-class antibiotics, including linezolid and sutezolid, highlighting the clinical relevance of ribosomal components as drug targets (Zahra et al. 2023). While mutations in rplC and rplD are linked to oxazolidinone resistance, the transcriptional upregulation observed in this study reflects ribosomal adaptation under drug-induced stress rather than reduced susceptibility, as no known resistance-conferring structural mutations were present. In contrast to this adaptive response, M. tuberculosis typically downregulates ribosomal protein genes during dormancy and under stress conditions, such as nutrient starvation, hypoxia, and antibiotic exposure, reflecting an energy-conserving survival strategy (Kundu and Basu 2021). Together, these observations support a transcriptional profile indicative of sustained metabolic activity rather than dormancy associated regulation in PZA^R under drug pressure. Functional enrichment interpretations were confined to KEGG pathways and GO terms meeting FDR significance thresholds, while individual gene-level observations are presented as exploratory analyses.

The PZA^R strain carried a mutation exclusively in the pncA gene, which was confirmed by whole genome sequencing (Rajendran et al. 2025). Whole genome analysis also identified mutations in genes associated with resistance to other anti-tuberculosis drugs, including katG, inhA, rpoB, rpoC, ethA, embA, embB, embC, mmpL5 and ndh, reflecting the broader drug resistance background of the clinical isolate. These additional mutations were neither functionally evaluated nor considered the primary focus of this study. At the transcriptomic level, we observed differential expression of genes associated with multiple anti-TB drugs and their corresponding drug targets, particularly in PZA^R vs UTRv comparison. Notably, the genes associated with PZA (pncA, panD), isoniazid (katG, ahpC), ethionamide (ethA), delamanid (fgd1, ddn), and bedaquiline (atpE, Rv0678, mmpL5, mmpS5) showed reduced expression. These patterns suggests that exposure to drug pressure in a resistant background may be accompanied by broader transcriptional adjustments that extend beyond PZA specific pathways. Although downregulation of ATP synthase-related genes was observed, this finding should not be interpreted as evidence of reduced cellular energy availability. Alternative bio energetic processes including proton motive force-driven transport systems and compensatory metabolic pathways may remain active consistent with previous reports describing metabolic flexibility under drug stress. (Zhang et al. 2014; Li, Plesiat and Nikaido 2015). These studies clearly show that bacteria that develop resistance to one drug may modulate the expression of other drug targets to enhance their survival. Therefore, we hypothesized that the observed downregulation of these drug targets might be part of a broader transcriptional adaptation to drug pressure. In parallel, several efflux pump genes previously implicated in PZA resistance, such as Rv3756c, Rv3008, Rv1258, and Rv0191 as well as efflux genes Rv2936, Rv2937, and Rv1643 associated with MDR-TB were upregulated in PZA^R. While efflux activity was not directly measured, these transcriptional changes are consistent with earlier studies reporting efflux associated responses under anti-microbial stress (AlMatar et al. 2019; Zhang et al. 2017). Collectively, these observations support the interpretation that the PZA^R strain displays a coordinated transcriptional response under PZA exposure, potentially reflecting adaptive stress response common to MDR-TB strains (Che et al. 2021). However, functional validation will be required to determine the extent to which these transcriptional changes contribute directly to phenotypic resistance or cross-resistance.

Protein-protein interaction network analysis revealed a complex interaction landscape comprising 19 distinct clusters, reflecting the multifaceted metabolic and virulence programs of M. tuberculosis. All clusters of genes identified in the PPI network and its functional enrichments were listed in Supplementary table 2A–2C. The majority of the genes were predominantly grouped into five clusters. Cluster 1 integrated polyketide synthase genes, β-ketoacyl synthases, PDIM synthesis genes, and siderophore biosynthesis genes, highlighting the convergence of lipid biosynthesis and iron acquisition pathways that underpin mycobacterial adaptation and persistence (McMahon, Rush and Thomas 2012; Choudhary et al. 2022). Cluster 2 was enriched with fadA and fadE family genes, reflecting the reliance on fatty acid degradation and β-oxidation for energy generation from host-derived lipids (Waddell 2010). Cluster 3 was dominated by ESX family genes, representing the Type VII secretion system (T7SS), which secretes key virulence factors critical for immune modulation and host-pathogen interactions (Shah and Briken 2016; Roy et al. 2020). Cluster 4 represents a network of mce genes, underscoring the role of membrane transporters in lipid metabolism and host interactions, and identifying them as potential therapeutic targets (Zaychikova and Danilenko, 2020). Cluster 12 was dominated by cytochrome P450 (CYP) genes, reflecting a specialized oxidative module involved in diverse redox reactions that are crucial for M. tuberculosis metabolism and adaptation (Ortega et al. 2019). Taken together, the emergence of these distinct yet interconnected modules spanning lipid and iron acquisition, fatty acid degradation, pentose phosphate pathway activity, glycerol metabolism, polysaccharide biosynthesis, and CYP-mediated oxidative processes underscore how M. tuberculosis orchestrates metabolic and structural pathways to maintain its persistence under acidic, drug-challenged conditions. The emergence of these interconnected modules highlights how M. tuberculosis coordinates lipid metabolism, redox homeostasis, and virulence pathways, suggesting that targeting such functional hubs may offer novel therapeutic opportunities. In conclusion, this study provides a genome-wide transcriptomic view of pyrazinamide resistance in a clinically derived PZA-resistant strain and a susceptible H37Rv strain. Transcriptome profiling demonstrated that PZA resistance is associated with a metabolically active transcriptional state under drug pressure, rather than dormancy associated response. The coordinated regulation of metabolic, ribosomal and stress response pathways highlight adaptive transcriptional remodelling in PZA resistant strain. Future studies using isogenic mutants and targeted functional validation will be required to disentangle mutation specific effects and to evaluate therapeutic strategies that exploit active cellular processes in PZA resistant strains.

The study has several limitations. The absence of untreated PZA-susceptible clinical strain which would have enabled a more complete dissection of mutation specific versus PZA treated transcriptional effects. In addition, the strains were exposed to PZA at concentrations determined by their phenotypic susceptibility thresholds rather than a uniform dose, a requirement to maintain bacterial viability that restricts direct dose-matched comparisons. The lack of an isogenic pncA-disrupted H37Rv control limits the mutation-specific interpretation. Comparisons involving the untreated H37Rv reference transcriptome require cautious interpretation, as the external dataset was generated under different culture conditions, including medium composition and pH. Consequently, the PZA^R vs RvT comparison, performed under consistent experimental and analytical conditions, was considered the most reliable internal comparison, whereas analysis involving UTRv were used for supportive interpretation. Moreover, the lineage differences between the PZA^R and RvT strains may influence the comparative expression patterns due to inherent lineage-specific transcriptional attributes, alongside drug-induced effects. Finally, Bulk RNA-sequencing provides population-averaged expression profiles and may not capture transcriptional heterogeneity or low abundance regulatory signals.

Supplementary Material

S1

Supplementary table 1A-1F. Functionally enriched DEGs from multiple annotation categories across three groups. This table summarizes the enrichment results of DEGs for different annotation categories includes COG ontology, Biological Process, Cellular Component, Interpro, Molecular Function, and KEGG pathway comprised in separate sheets. For each category, the columns include the comparison group, enrichment category, term identifier or description, number of genes in the term, p-value, list of genes contributing to the enrichment, fold enrichment, and multiple testing correction values (Bonferroni, Benjamini, and FDR). These results provide an overview of the key functional terms and pathways significantly associated with transcriptional changes under different experimental conditions.

S2

Supplementary table 2A-2C. Gene clusters derived from the protein–protein interaction network and their functional enrichments. This table contains three sheets representing Cluster 1, Cluster 2, and Other Clusters obtained from PPI network analysis. Each sheet lists the genes associated with the respective clusters along with their functional annotations. The table detailed about the clusters, node name, gene ID, protein ID, comparison group, functional category, and the GO or pathway terms that were enriched. This dataset highlights the functional organization of DEGs within key interaction modules.

S3

Supplementary table 3. Sequencing and alignment metrics for all RNA-seq samples are presented. The read alignment rates across the PZA^R, RvT, and UTRv samples, along with the corresponding sequencing depth utilized for subsequent transcriptomic analysis, are summarized.

Impact Statement.

This study presents the first detailed transcriptomic profile of a clinical isolate of Mycobacterium tuberculosis resistant to pyrazinamide, elucidating the relationship between genetic mutations and the overall response to drug pressure in a clinical isolate. Comparative analysis of the PZA^R strain and the susceptible laboratory H37Rv strain revealed significant transcriptional alterations in lipid metabolism, energy production, and ribosomal machinery, alongside a suppressed response in dormancy and virulence-associated genes. These findings indicate that PZA-resistant strains do not enter the dormant stage but maintain active metabolic and repair mechanisms, which is an important feature of survival. The modulation of drug targets and efflux pump gene expression suggests associated transcriptional responses involving pathways shared with other anti-tuberculosis agents. Collectively, this study enhances our understanding of the molecular reprogramming associated with PZA resistance and provides valuable insights that may support the development of new strategies against drug-resistant tuberculosis.

Data availability

RNA sequencing data associated with this study have been deposited in the NCBI BioProject under accession number PRJNA1151825 and sequences are accessible through the following link https://www.ncbi.nlm.nih.gov/sra/?term=PRJNA1151825.