Transcriptional adaptation of Mycobacterium tuberculosis that survives prolonged multi-drug treatment in mice

ABSTRACT To address the ongoing global tuberculosis crisis, there is a need for shorter, more effective treatments. A major reason why tuberculosis requires prolonged treatment is that, following a short initial phase of rapid killing, the residual Mycobacterium tuberculosis withstands drug killing. Because existing methods lack sensitivity to quantify low-abundance mycobacterial RNA in drug-treated animals, cellular adaptations of drug-exposed bacterial phenotypes in vivo remain poorly understood. Here, we used a novel RNA-seq method called SEARCH-TB to elucidate the Mycobacterium tuberculosis transcriptome in mice treated for up to 28 days with standard doses of isoniazid, rifampin, pyrazinamide, and ethambutol. We compared murine results with in vitro SEARCH-TB results during exposure to the same regimen. Treatment suppressed genes associated with growth, transcription, translation, synthesis of rRNA proteins, and immunogenic secretory peptides. Bacteria that survived prolonged treatment appeared to transition from ATP-maximizing respiration toward lower-efficiency pathways and showed modification and recycling of cell wall components, large-scale regulatory reprogramming, and reconfiguration of efflux pump expression. Although the pre-treatment in vivo and in vitro transcriptomes differed profoundly, genes differentially expressed following treatment in vivo and in vitro were similar, with differences likely attributable to immunity and drug pharmacokinetics in mice. These results reveal cellular adaptations of Mycobacterium tuberculosis that withstand prolonged drug exposure in vivo, demonstrating proof of concept that SEARCH-TB is a highly granular pharmacodynamic readout. The surprising finding that differential expression is concordant in vivo and in vitro suggests that insights from transcriptional analyses in vitro may translate to the mouse. IMPORTANCE A major reason that curing tuberculosis requires prolonged treatment is that drug exposure changes bacterial phenotypes. The physiologic adaptations of Mycobacterium tuberculosis that survive drug exposure in vivo have been obscure due to low sensitivity of existing methods in drug-treated animals. Using the novel SEARCH-TB RNA-seq platform, we elucidated Mycobacterium tuberculosis phenotypes in mice treated for with the global standard 4-drug regimen and compared them with the effect of the same regimen in vitro. This first view of the transcriptome of the minority Mycobacterium tuberculosis population that withstands treatment in vivo reveals adaptation of a broad range of cellular processes, including a shift in metabolism and cell wall modification. Surprisingly, the change in gene expression induced by treatment in vivo and in vitro was largely similar. This apparent “portability” from in vitro to the mouse provides important new context for in vitro transcriptional analyses that may support early preclinical drug evaluation.


Efficacy of HRZE
Treatment efficacy was quantified using traditional and novel molecular pharmacody namic markers.In mice, 28 days of HRZE treatment reduced CFU 99.8% (7.56 to 4.87 log 10 /lung).In vitro, 8 days of HRZE exposure reduced CFU 99.993% (7.7 to 3.57 log 10 / mL).A newer biomarker, the RS ratio (44), declined faster than CFU, indicating that interruption of rRNA synthesis occurs more rapidly than killing, both in mice and in vitro (Fig. 3c and d).

Scale of HRZE-induced expression change in mice and in vitro
We began globally by characterizing the scale of expression change induced by HRZE in mice versus in vitro.In mice, HRZE significantly altered expression of 2,049 (57%) and 2,329 (65%) genes at days 14 and 28 relative to pre-treatment control, respectively (Fig. 3e and f).Hierarchical clustering identified genes with similar expression changes over time (Fig. 3g).Most transcriptional changes occurred by day 14 (Fig. 3h, Clusters 2 and 4).

Comparison of transcriptional changes induced by HRZE in mice and in vitro
There were broad similarities and limited but important differences between the effects of HRZE in mice and in vitro.At the latest treatment time points (8 days in vitro and 28 days in mice), differential expression was largely concordant.Overall, 64% of genes had the same significance results in vitro and in the mouse, relative to pre-treatment (Fig. 3i).Only 4% of genes were significantly differentially expressed in opposite directions (colored gold in Fig. 3i and j), indicating differing effects of HRZE in mice and in vitro.The specific processes that differ between mice and in vitro are highlighted throughout the following sections.Notably, fold changes were more extreme in vitro than in mice (Fig. 3j).

Effect of HRZE on the Mtb cellular processes
Since our focus is Mtb phenotypes that withstand prolonged drug exposure, we describe expression changes at the latest time point (28 days in mice or 8 days in vitro) rela tive to control, unless otherwise noted.Other time points can be explored via the Online Analysis Tool (https://microbialmetrics.org/analysis-tools/).After 28 days of HRZE treatment in mice, 36 of 124 functional categories (Table S3) were significantly enriched (Fig. 4a).

Decreased growth and macromolecule synthesis
HRZE decreased expression of genes associated with growth and macromolecule synthesis.Specifically, treatment suppressed ribosomal protein gene expression in mice (adj.P = 2.9 × 10 −15 ) and in vitro (adj.P = 1.0 × 10 −14 ) relative to control (Fig. 4b), consistent with decreased ribosome synthesis, a process fundamentally coupled with bacterial replication (46,47).An exception was the four alternative C-ribosomal protein "remodeling" paralogs lacking the zinc-binding CXXC motif (rpmB1, rpsR2, rpsN2, and rpmG1), which had sustained or significantly increased expression relative to control (Fig. 4b), consistent with ribosomal remodeling by less-active Mtb.Slowing of protein synthesis was suggested by decreased expression of the protein translation and modification category that includes genes responsible for translational initiation, promotion of tRNA binding, elongation, termination, and protein folding (adj P = 0.006 in mice; adj.P = 0.007 in vitro).Transcriptional slowing was suggested by above and below zero represent up-and down-regulation relative to control, respectively.(i) Comparison of differential expression between mouse (day 28) or in vitro (day 8) relative to respective controls.Purple shading indicates genes with concordant fold-change direction and significance between mouse and in vitro experiments.Green shading indicates genes that were significant for either mouse or in vitro experiments but not both.Gold shading indicates genes that were significant for both mouse and in vitro experiments but in opposite directions.Gray shading indicates the genes that were not differentially expressed with HRZE treatment either in mouse or in vitro experiments.(j) Comparison of fold changes between mouse (day 28) or in vitro (day 8) relative to respective controls.
Purple, green, gold, and gray colors have the same meaning as in (i).

Decreased cell wall synthesis but continued remodeling and recycling
To summarize the effect of HRZE on cell wall biosynthesis, we evaluated expression change for the major cell wall constituents: mycolic acids, phthiocerol dimycocerosates (PDIM), peptidoglycan, and trehalose.

Mycolic acids
Slowing of the first step of mycolic acid synthesis was indicated by the decreased expression of Rv2524c (fas), the gene coding for fatty acid synthetase I, in mice (adj. ) and the decreased expression of the fas transcriptional regulator Rv3208 (48) in mice (adj.P = 1.9 × 10 −4 ) and in vitro (adj.P = 1.1 × 10 −11 ).Genes coding for the second step of elongation of acyl-coenzyme A to long-chain fatty acids by fatty acid synthetase II were also decreased in both mice (adj.P = 0.011) and in vitro (adj.P = 0.027) (Fig. 4c).Finally, genes associated with elongation, desaturation, modification, and transport of the mature mycolic acids to the cell wall had decreased expression both in mice and in vitro (Online Analysis Tool).

PDIM
Genes involved in synthesis of PDIM, the outer surface glycolipids important for intracellular survival and virulence, were down-regulated in mice (adj.P = 9.9 × 10 −5 ) and in vitro (adj.P = 0.017).

Peptidoglycan
In contrast to mycolic acids and PDIM, certain genes involved in peptidoglycan synthesis, modification, and recycling had increased expression.Specifically, in HRZE-treated mice, all five peptidoglycan synthesis genes assayed in the division cell wall operon (ftsQ, murC, ftsW, murD, murR, and murE) had significantly increased expression at day 14 and three remained significantly up-regulated at day 28 (Online Analysis Tool).Active recycling was further suggested by significantly increased expression in mice and in vitro of genes coding for the UspABC amino-sugar importer and the DppABCD dipeptide importer that transport peptidoglycan breakdown products (Fig. 4d) (49).

Trehalose
The treY/Z genes that synthesize the essential metabolite and cell wall constituent trehalose from glycogen were significantly up-regulated at all post-treatment time points in mice and in vitro.Two other trehalose synthesis pathways (otsA/B and glgE/treS) were not differentially expressed in mice or in vitro.Remodeling of the trehalose component of the cell envelope was suggested by increased expression of Rv3451 (cut3) in mice (adj.P = 0.002) and in vitro (adj.P = 4.2 × 10 −4 ), which codes for a stress-responsive trehalose dimycolate hydrolase (50).Additionally, HRZE induced significantly increased expression in mice and in vitro of four of the five genes coding for the LpqY/SugABC importer that is specific for the transport of trehalose (51).

Metabolic adaptation
Electron transport and aerobic respiration HRZE suppressed expression of all genes coding for ATP synthetase in mice (adj.P = 7.6 × 10 −4 ) and in vitro (adj.P = 0.003) (Fig. 5a).Oxidative phosphorylation appeared to transition from the primary cytochrome bcc/aa3 supercomplex (down-regulated) to the less-efficient cytochrome bd oxidase (up-regulated) that has been implicated in persistence under environmental and drug stress (54) (Fig. 5b).TCA cycle genes were down-regulated with HRZE treatment in mice (adj.P = 0.001) and in vitro (adj.P = 0.010).
Genes coding for NADH dehydrogenase types I and II and succinate dehydrogenase types I and II were also down-regulated (Online Analysis Tool).By contrast, three of the four fumarate reductase genes were significantly induced by HRZE in mice, and all four were significantly induced by HRZE in vitro (Online Analysis Tool).

Discordant expression of DosR regulon and other stress responses in mice and in vitro
Although heat shock proteins (HSPs) that act as chaperones in protein folding are often described as a stress response, HRZE suppressed expression of HSP genes.A notable discordance between murine and in vitro results is the hypoxia-responsive hspX (noted in red in Fig. 5d) that had the greatest negative fold change of all genes evaluated in mice (48.8-fold decrease, adj.P = 4.8 × 10 −57 ) yet was significantly up-regulated with HRZE treatment in vitro (4.6-fold increase, adj.P = 2.0 × 10 −8 ).Concordant with this result, HRZE repressed DosR expression in mice (adj.P = 0.032) but induced the DosR regulon in vitro (adj.P = 2.2 × 10 −7 ) (Fig. 5e).HRZE was associated with down-regulation of genes of the stringent response, typically induced by nutrient deprivation and other environmental stresses, in mice (adj.P = 2.9 × 10 −15 ) and in vitro (adj.P = 1.6 × 10 −14 ).Similarly, HRZE decreased expression of relA which codes for the stringent response regulator, in mice (adj.P = 2.0 × 10 −6 ) and in vitro (adj.P = 7.0 × 10 −6 ).

Transcriptional and post-transcriptional reprogramming
SEARCH-TB indicated large-scale regulatory reprogramming.For example, HRZE changed most of the 188 transcription factors assayed in mice, with 66 significantly increased and 47 significantly decreased.The regulatory perturbation was even more pronounced in vitro with 87 significantly increased and 55 significantly decreased transcription factors.Of the 12 sigma factor genes included in SEARCH-TB, sigF, sigI, and sigM were significantly up-regulated following HRZE treatment in both mice and in vitro while sigA, sigB, sigD, and sigK were significantly down-regulated with HRZE treatment in both mice and in vitro.HRZE significantly altered expression of sigC and sigL in both mice and in vitro but in discordant directions, suggesting differing regulatory responses in vivo and in vitro.HRZE appeared to activate the post-transcriptional toxin-antitoxin system that modulates the concentration of existing transcripts.Toxin genes were induced in mice (adj.P = 0.011) and in vitro (adj.P = 0.027).The counter-regulatory antitoxins that restrict toxin activity were not categorically altered in mice but were significantly suppressed in vitro (adj.P = 0.037).

Adaptation of efflux pump expression
HRZE altered expression of many efflux pumps, in varying directions (Table S4).As examples, Rv2686c-2688c, which is associated with fluoroquinolone tolerance (55), was up-regulated with HRZE treatment in mice and in vitro (Fig. 5f), but the genes coding for DrrABC, which is associated with daunorubicin tolerance and appears in some clinical drug-resistant strains (55), were down-regulated (Fig. 5g).

Expression of drug targets
Because processes up-regulated during drug exposure may represent survival mecha nisms that could be targeted to eradicate persisting Mtb, we evaluated genes coding for the known targets of 31 established and investigational new drugs (36 genes) (Table S5).These genes were predominantly suppressed in mice (19 down-regulated and three up-regulated) and in vitro (26 down-regulated and two up-regulated).Exceptions to the prevailing down-regulation were increased expression of the genes coding for the Mur ligases B and C that initiate peptidoglycan synthesis in mice (56) and rfe, the gene coding for the phosphoglycosyltransferase WecA that initiates arabinogalactan synthesis with HRZE treatment in mice and in vitro (57).

Differences between Mtb in mice and in vitro at final time points
After 8 days in vitro and 28 days in the mouse, Mtb transcriptomes were more similar than they were prior to HRZE exposure (Fig. 3a).Nonetheless, 1,014 genes (28%) remained differentially expressed between the final murine and in vitro time points evaluated in this study.

DISCUSSION
SEARCH-TB elucidated cellular adaptations of Mtb that withstand long-term HRZE treatment.After 28 days of treatment-which reduced culturable Mtb in mouse lungs by 99.8%-SEARCH-TB indicated broad suppression of cellular activity including slowing of metabolism, reduced synthesis of macromolecules, and reduced secretion of immunemodulating peptides.SEARCH-TB also suggested bacterial adaptation to drug stress, including a shift in electron transport to the alternative, less efficient cytochrome bd oxidase, ribosomal remodeling, cell wall remodeling and recycling, and reprogramming of regulatory and efflux pump activity.Importantly, the effects of HRZE in mice and in vitro were broadly similar with some differences that likely reflect effects of pharmacoki netics (PK), immunity, and in vivo environment.By quantifying extremely low-abundance Mtb transcripts in mice, SEARCH-TB should enable a new era of molecular evaluation of drug effect in vivo.
Differences between mice and in vitro before treatment reflect bacterial adaptation to immunity and the lung environment, consistent with a recent review (37) of the treatment-naïve in vivo Mtb transcriptome, both in terms of broad cellular processes (down-regulation of genes associated with transcription, translation, and metabolism) and specific adaptations (e.g., increased expression of DosR genes, glyoxylate bypass genes).Importantly, the untreated mouse was the starting point for this study.As highlighted below, drug exposure elicited transcriptional adaptations that diverge from well-established transcriptional adaptations to environmental stress (such as pH, hypoxia, and nutrient starvation).
SEARCH-TB indicated that HRZE suppressed Mtb growth and metabolism, consistent with previous in vitro analyses of drug effects (11)(12)(13)(14)(15)(16)(17).There was a general decrease in the expression of genes involved in the synthesis of all major macromolecules.The transition to a less-active bacterial phenotype observed in mice treated with HRZE is similar to the expression change observed in sputa of humans treated with HRZE (8,18).There was also decreased protein and peptide secretion, including Ag85 and the ESX1 Type VII secretion system that exports the immunogenic ESAT-6 and CFP-10 proteins, suggesting that drug stress might alter the pathogen's capacity to adapt to host immunity.Consistent with the current mouse data, our previous study of human sputum showed that treatment with HRZE significantly decreased expression of Ag85 genes and the genes (esxA and esxB) coding for ESAT-6 and CFP-10 (8).
Nonetheless, SEARCH-TB suggested that drug-stressed Mtb is not inert or incapa citated.Increased expression of genes associated with peptidoglycan and trehalose synthesis and recycling suggest ongoing cell wall modification.SEARCH-TB indicated transcriptional and post-transcriptional regulatory reprogramming and altered efflux pump expression.Metabolism and energy generation appeared reconfigured away from high respiratory activity that maximizes ATP generation, toward reduced respiratory efficiency and oxidative phosphorylation activity.
Prior to treatment with HRZE, DosR regulon genes that respond to impaired respiration (58) were expressed on average >10-fold more highly in mouse lung than under aerobic in vitro conditions, consistent with bacterial adaptation to host-derived nitric oxide and hypoxia by intracellular Mtb in the mouse.Strikingly, during treatment with HRZE, expression of the DosR regulon decreased significantly in mice and increased significantly in vitro.The decrease in DosR expression in mice is consistent with our previous observation that HRZE suppressed DosR expression in human sputum (8).We hypothesize that, during the first month of treatment, host immune response to the pathogen is modulated by the >99% reduction in Mtb burden and decreased bacterial secretion of immunogenic peptides.A less-intense inflammatory response with diminished macrophage and neutrophil activation would decrease nitric oxide exposure, reducing the need for Mtb DosR expression.This hypothesis that DosR may serve as an indirect readout of immunity is consistent with our previous observation that the DosR regulon has lower expression in TB patients with AIDS than in immunocompetent patients (29) and with our finding that in vitro, in the absence of immunity, HRZE increased rather than decreased expression of genes in the DosR regulon.
Also notable was decreased expression of the gene coding for isocitrate lyase, the first step of the glyoxylate bypass which is required to establish productive infection in animals (59)(60)(61).Consistent with prior results (59-61), icl1 was expressed more highly in untreated mice than in vitro.icl1 was shown to increase with sublethal exposure to rifampin, INH, or streptomycin in vitro (42).By contrast, we found that lethal doses of HRZE strongly suppressed, rather than induced, expression of icl1 and aceAa which codes for an alternative isocitrate lyase.This is consistent with our previous observation that expression of icl1 and aceAa declined significantly in human sputum during treatment with HRZE (8).This highlights that adaptations to HRZE differ from adaptations to host environments that enable persistent infection in the absence of drug therapy and that the response to exposure to a combination regimen in vivo may differ from single drug exposures in vivo.
Drugs have historically targeted growth-associated cellular processes (62) that SEARCH-TB showed to be down-regulated after HRZE treatment.While transcriptional down-regulation of drug targets does not necessarily mean that a drug will be ineffective, genes such as those coding for Mur ligases and WecA that showed increased expression after 1 month of HRZE in vitro could indicate adaptations that enable Mtb to withstand drug exposure.Similarly, genes coding for the alternative cytochrome bd oxidase that is a proposed drug target (63) were up-regulated at a time when metabo lism was globally suppressed.
A finding with important implications is the concordance of murine and in vitro results.Although Mtb cellular processes were very different in mice and in vitro before treatment, HRZE induced broadly similar changes in vivo and in vitro.This indicates that drug exposure is a stress of sufficient intensity that it overwhelms phenotypic differences that exist due to differences in environmental conditions present in vitro or in a mouse.
One exception to the similarity is that the magnitude of fold change was smaller in mice than in vitro.This likely reflects diminished drug/target engagement (i.e., decreased drug exposure) in a mouse versus in vitro due to in vivo PK.A second exception was the discordance in the effect of HRZE on DosR expression (down in mice, up in vitro) as discussed above.Nonetheless, our observation that differential expression is largely consistent despite baseline differences in environment and phenotype provides new context for interpretation of transcriptome data from in vitro drug exposure experiments.
As a readout of the effect of drugs on Mtb cellular processes, SEARCH-TB could improve precision of pharmacodynamic evaluation, providing greater information than culture-based enumeration of bacterial burden (the existing standard method) (64).We have previously shown that CFU does not capture the entirety of complex drug effects in vivo.For example, TB regimens that have identical effects on CFU in mice can have different long-term relapse outcomes (44,65).Our previous RS ratio studies provided proof of concept that molecular measures of bacterial cellular processes in vivo can distinguish regimens that are indistinguishable based on CFU (44,65).SEARCH-TB advances molecular characterization of drug effects in vivo to a higher level of granular ity.Others have demonstrated the power of Mtb transcriptional readouts of drug effect to predict drug interactions in vitro (66).SEARCH-TB may enable assessment of drug interactions and regimens based on molecular effects on cellular processes in vivo.
This report has several limitations.First, specific and efficient primers could not be designed for ~12% of Mtb transcripts.Second, as discussed in Supplemental Information, between-target variation in the amplification efficiency of SEARCH-TB primers likely affects the rank-order of gene counts, meaning that, for individual genes, a higher absolute count does not necessarily indicate greater expression.However, because amplification is highly repeatable, the modest amplification bias identified does not affect estimation of differential expression between groups.Indeed, we showed that SEARCH-TB provides the same biological interpretation as conventional RNA-seq.Third, transcriptional profiling is inherently unable to resolve the enduring question of whether the sub-population that persists late into treatment results from selection (i.e., elimina tion of easily killed Mtb) or physiological adaptation of extant Mtb.Fourth, there are important differences between in vitro and mouse conditions that we could not control.For example, in vitro exposure conditions employed herein do not recapitulate the dynamic PK drug profiles that occur in mice.It is also noted that pyrazinamide is active in the mouse but is not expected to be active in vitro under the conditions tested herein.The observation that differential expression is similar in vitro and in the mouse despite these differences indicates that the effect of HRZE on the transcriptome is largely independent of the starting bacterial phenotype.Finally, here we evaluated one regimen in one murine model.Next steps include evaluation of individual drugs and diverse regimens in additional animal models.
SEARCH-TB elucidated the effect of prolonged drug treatment on Mtb transcription in animal models, revealing adaptations distinct from those observed under environ mental stress.Mtb that survived 1 month of HRZE treatment appeared substantially less active than prior to treatment but was not inert, with transcriptional changes suggesting adaptation for survival.SEARCH-TB should enable more informative in vivo molecular pharmacodynamic evaluation that accelerates identification of new highly potent regimens.

Design of the SEARCH-TB assay
The SEARCH-TB assay uses an AmpliSeq for Illumina custom pool designed to amplify coding sequences (CDS) of Mtb complex (MTBC) organisms.To assure amplification across diverse lineages, the design used eight MTBC reference genomes (Table S1).Annotations were prepared for Mtb H37Rv (NC_000962) and Mtb Erdman (AP012340.1)as described in the Supplemental Information.To avoid off-target amplification (i.e., of nonMTBC organisms), primers were cross-referenced during design with 12 "exclusion" genomes that included phylogenetically diverse bacteria as well as human and mouse (Table S2).

Evaluation of amplification bias
We tested for amplification bias (i.e., differences in amplification efficiency between primer pairs targeting different Mtb sequences) using replicate human lung RNA samples spiked with Mtb genomic DNA (gDNA) (Supplemental Information).Since all targeted sequences are present as single copies in gDNA, an entirely unbiased assay would hypothetically result in the same copy number for all targets, indicating that all primer pairs amplified with identical efficiency.We defined amplification bias as deviation from this ideal by comparing the observed expression for a gene to the expected expression assuming no amplification bias (Supplemental Information).

Evaluation of repeatability of amplification
To evaluate repeatability, we spiked 1 pg Mtb RNA into 1 ng human lung RNA (Supplemental Information).We compared counts per million values for each gene between technical replicates.We also evaluated repeatability over time by prepping and sequencing 19 of the in vitro and murine samples described below two times with up to a 4-month intervening interval.We quantified batch effect by calculating the difference in expression (normalized with DESeq2's variance stabilizing transformation) (67) of each gene between replicate pairs, then averaging across replicates.We compared the magnitude of the batch effect with the observed treatment effect.

Concordance of SEARCH-TB with conventional RNA-seq
We evaluated whether SEARCH-TB identified the same transcriptional changes as a conventional RNA-seq method without Mtb-targeted amplification (Illumina TrueSeq) after 24 h in vitro isoniazid (INH) exposure (Supplemental Information).We first evaluated RNA from control (N = 4) and INH-treated samples (N = 4) via conventional RNA-seq to serve as a reference standard.We then spiked the same RNA from control and INH-trea ted Mtb into human lung RNA at a ratio of 1:1,000 and sequenced it via SEARCH-TB.After calculating differential expression between control and INH-treated samples separately for conventional RNA-seq and SEARCH-TB using edgeR (68), we compared the significant genes and fold changes identified by the two platforms.

In vitro experiments
Mtb strains H37Rv and Erdman were cultured in vitro using Middlebrook 7H9 broth (Difco) supplemented with 0.085 g/L NaCl, 0.2% glucose, 0.2% glycerol, 0.5% BSA, and 0.05% Tween-80.All culturing was performed at 36.5°C and 5.0% CO 2 .Single-use frozen Mtb aliquots were revived in 7H9 and grown to mid-log phase, then cultures were diluted to OD 600 = 0.05, dispensed in 5.0 mL aliquots into sterile glass tubes (20 × 125 mm) containing sterile stir bars (12 × 4.5 mm), and outgrown for 18 h under rapid agitation (~200 rpm stirring speed using a rotary magnetic tumble stirrer) prior to the initiation of drug exposure.RNA was collected from Mtb H37Rv exposed to INH or Mtb Erdman exposed to HRZE in vitro (Supplemental Information).

Murine drug experiments
All animal procedures were conducted according to relevant national and interna tional guidelines and approved by the Colorado State University Animal Care and Use Committee as described in the Supplemental Information.Briefly, female BALB/c mice, 6 to 8 wk old, were exposed to aerosol (Glas-Col) with Mtb Erdman strain resulting in the deposition of 4.55 ± 0.03 (SEM) log 10 CFU in lungs 1 day following aerosol.After 11 days, five mice were euthanized to serve as the pre-treatment control group.Groups of five mice each were treated with HRZE at standard doses 5 days a week for 14 or 28 days before euthanasia.Lungs were aseptically dissected and flash frozen in liquid nitrogen before processing.

RNA extraction, sequencing, and data preparation
RNA extraction, library preparation, sequencing, and data preparation are detailed in the Supplemental Information.

Statistical analysis of murine and in vitro experiments
Murine and in vitro sequence data were analyzed together using edgeR (68,69) to identify the effect of HRZE treatment and compare gene expression between murine and in vitro experiments.We fit negative binomial generalized linear models to each gene and included terms for murine and in vitro time points (control, day 14 and day 28 in mice; control, day 4 and day 8 in vitro).Likelihood ratio tests were performed to compare expression between murine time points, between in vitro time points, and between murine and in vitro experiments before and at the end of treatment.Genes with Benjamini-Hochberg adjusted P-value (70) less than 0.05 were considered significant.For murine experiments, we used hierarchical clustering to identify groups of genes with similar changes in gene expression over the course of treatment as follows.First, for genes that were differentially expressed between at least two-time points, we calculated the expected expression at each time point using the edgeR models.Then, the expected expression values were hierarchically clustered based on Euclidian distance using Ward's method (71) to find clusters of genes with similar patterns of expression over time.
PCA was performed using data for the 500 most variable genes after normalization with DESeq2's variance stabilizing transformation (67).
Using hypergeometric tests in the hypeR R package (72), we performed functional enrichment for each pairwise combination of murine and in vitro time points to evaluate whether differentially expressed genes were overrepresented in gene categories established by Cole et al. (73) or curated from the literature (Table S3).Enrichment analysis was run twice for each pairwise combination, first using significantly up-regula ted genes and then using significantly down-regulated genes.Gene categories with <8 genes were excluded.Gene categories with Benjamini-Hochberg adjusted P-values (70) less than 0.05 were considered significant.All analyses used R (v4.1.1)(74).

Online analysis tool
Differential expression, functional enrichment, and visualizations can be evaluated interactively using an Online Analysis Tool (https://microbialmetrics.org/analysis-tools/) created using the R package Shiny (75).

FIG 1
FIG 1 Visual summary of methods previously used to quantify the Mtb transcriptome in vivo.Each horizontal arrow represents a distinct combination of enrichment and quantification.Varying enrichment methods are represented via the symbols shown in the key.SEARCH-TB is a unique combination of enrichment (eukaryotic cell lysis + targeted amplification) followed by quantification via RNA-seq that has enabled transcriptome evaluation in mice treated for weeks with a potent combination regimen.Image created with Biorender.com.

FIG 2 4 FIG 3
FIG 2 Evaluation of SEARCH-TB platform.(a and b) Volcano plot showing log 2 fold changes and −log 10 P-values induced by 24 h in vitro INH exposure as quantified by RNA-seq (a) and SEARCH-TB (b).Genes significantly down-and up-regulated with INH exposure relative to control (adj.P < 0.05) are shown in blue and red, respectively.(c) Comparison of differential expression between INH-treated samples and control samples from RNA-seq or SEARCH-TB data.Purple shading indicates genes with concordant fold-change direction and significance between RNA-seq and SEARCH-TB.Green shading indicates genes that were significant in RNA-seq or SEARCH-TB results but not both.Gold shading indicates genes that were significant for both RNA-seq and SEARCH-TB but in opposite directions.Gray shading indicates genes that were not significantly differentially expressed in either RNA-seq or SEARCH-TB.(d) Comparison of INH versus control fold changes from RNA-seq data versus SEARCH-TB data.Purple, green, gold, and gray colors have the same meaning as in (c).

FIG 4
FIG 4 Summary of gene set enrichment and transcriptional changes in biological processes.(a) Gene categories significantly enriched for genes differentially expressed between day 28 and control murine samples.The percentage of genes in each category significantly up-(red) or down-regulated (blue) for each comparison is illustrated.Asterisks indicate statistical significance (adj.P < 0.05).(b) Fold change of ribosomal protein genes in mice on day 28 (left) and in vitro (right) on day 8, relative to control.Red bars indicate the four alternative C-ribosomal protein paralogs.(c and e) Fold-change values in mice at days 14 and 28 (left) and in vitro (right) at days 4 and 8, relative to control for (c) FAS-II and (e) antigen 85 gene sets.(d) Graphical representation of changes in the peptidoglycan synthesis, modification, and recycling pathways in mice on day 28 relative to control.Log 2 fold-change values for genes in the process are indicated by the color of each box and the thick outlines of boxes represent genes that are significantly differentially expressed.Figure adapted from Maitra et al., 2019 (45).Image created with Biorender.com.(f ) Fold change of ESX-1 genes in mice on day 28 (left) and in vitro (right) on day 8, relative to control.Red bar represents esxA and blue bar represents esxB.