PE/PPE proteins contribute to Mycobacterium tuberculosis drug resistance

22 23 The outer membrane (OM) of mycobacteria is a formidable permeability barrier that 24 confers drug tolerance, and whether drugs traverse the OM by mechanisms beyond 25 passive diffusion remains unclear. The proline-glutamic acid (PE) and proline-proline-26 glutamic acid (PPE) proteins of pathogenic mycobacteria include several OM 27 transporters. Here, we tested the role of PE/PPE proteins in Mycobacterium 28 tuberculosis (Mtb) drug transport and resistance. Mutations in multiple pe/ppe genes 29 were strongly associated with drug resistance in a genetic association study, and 30 mutations in ppe42 and ppe51 also conferred increased resistance in vitro. Deletion of 31 a pe/ppe pair transcriptionally responsive to drug exposure, pe25/ppe41, led to elevated 32 resistance to isoniazid (INH) across all major Mtb lineages and accelerated INH 33 resistance emergence in vitro. These data identify a role for Mtb PE/PPE proteins in 34 drug resistance consistent with the PE/PPE transporter paradigm, suggest a broader 35 contribution of this large protein family, and a new factor of Mtb clinical drug resistance.36 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 2

37 38 The outer membrane (OM) of mycobacteria consists of long-chain fatty acids that 39 severely restrict solute diffusion 1. This OM barrier is also highly impermeable to many 40 clinical drugs and at least partially responsible for the high intrinsic drug tolerance of 41 Mycobacterium tuberculosis (Mtb) and other pathogenic mycobacteria 2. While non-42 pathogenic mycobacteria and gram-negative bacteria with a comparable OM express 43 porins to facilitate free diffusion, Mtb and other pathogenic mycobacteria lack canonical 44 porins. As a result, OM permeability of mycobacteria is typically two to three orders of 45 magnitude lower than that of gram-negative bacteria 3 and poses one of the many 46 challenges to effective tuberculosis (TB) treatment. 47 48 A unique feature of the genomes of pathogenic mycobacteria is the presence of two 49 large gene families with mostly unknown function, the pe and ppe genes. Mtb, for 50 example, encodes approximately 100 PE and 69 PPE proteins, which together take up 51 nearly 10% of the genome’s coding capacity 4. Despite their abundance, a unifying or 52 shared function for the PE/PPE proteins has not yet been discovered. Several recent 53 studies now show that some PE/PPE proteins function as specific pores or channels in 54 nutrient transport across the OM 5-8, and pore formation has now directly been shown 55 for one PPE protein 9. This emerging transport function aligns with the relative 56 occurrence of PE/PPE proteins and porins in mycobacteria: PE/PPE proteins are 57 typically present in much larger numbers when porins are absent. 58 59 TB is exceedingly difficult to treat due to several host and bacterial factors. In addition to 60 its impermeable OM, Mtb expresses inner membrane efflux pumps that effectively 61 reduce intracellular drug concentrations 10. Efflux pumps are commonly upregulated in 62 clinical drug-resistant strains and are a major contributor to Mtb drug tolerance and 63 resistance to virtually all drugs currently in use 11. How drugs traverse the impermeable 64 OM of pathogenic mycobacteria in either direction in the absence of porins, however, 65 has remained an open question. 66 67 Here, we show that several PE/PPE proteins affect drug resistance in a way that is 68 consistent with transporter function and are likely facilitating drug uptake across the OM. 69 These data challenge the idea that new TB drugs need to be lipophilic to penetrate the 70 outer membrane and show that the pe/ppe genes contribute to clinical drug resistance. 71 With 169 pe/ppe genes in Mtb, these gene families could widely affect drug 72 susceptibility and be a central contributor to TB drug resistance. 73 74

75 76 Mutations in pe/ppe genes are associated with clinical drug resistance 77 To identify candidate pe/ppe genes that may contribute to clinical drug resistance, we 78 carried out a targeted genetic association study with a collection of 16,891 genome 79 sequences of drug- and multi-drug resistant Mtb. Whole genome sequence data of Mtb 80 was obtained from three sources: 2,659 genome sequences from the Wellcome Trust 81 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 3 Sanger Institute pilot study 12, 3,922 from the NIH-NIAID TB Portals collection 13, and 82 10,310 from the CRyPTIC study 14. We analyzed sequences of the pe and ppe genes 83 along with known resistance loci (aphC, gyrA, inhA, katG, and rpoB) as positive 84 controls. Because of their high sequence variability and complex repetitive sequences, 85 DNA sequence quality and genome assembly across the pe/ppe genes can be poor 86 when using short-read sequencing, which is predominantly used for sequencing of drug 87 resistant Mtb. As a result, many GWAS studies excluded pe/ppe genes from analysis. 88 To assess data quality and possible bias, we analyzed the read depth and mapping 89 quality across pe, ppe, and pe-pgrs genes from the set of nearly 4,000 genomes from 90 the NIH collection. Read depth was generally comparable between pe/ppe and all other 91 genes. Mapping quality for the pe genes was also comparable to all non-pe/ppe genes 92 but was reduced for the ppe genes and more so for the pe-pgrs genes (Supplemental 93 Fig. 1). The pe-pgrs subfamily is also known to harbor large sequence variation 94 independent of drug resistance 15-17 and together with the ppe-mptr subfamily is the 95 main source of errors in Mtb genome sequences 18. This variation and the reduced 96 mapping quality led us to exclude the pe-pgrs genes from further analysis. We 97 considered promoter and gene body mutations for the genetic association analysis. 98 Using the linear mixed model approach in PySEER, some of the most significant 99 associations were found with mutations in ppe42, ppe35, and ppe51 (Fig. 1a). The 100 ppe42 mutation most significantly associated with drug resistance introduces a 101 premature stop codon and was previously noted by the CRyPTIC consortium to be 102 associated with amikacin and kanamycin resistance 14. Mutations in ppe35 have also 103 previously been linked to pyrazinamide resistance 19,20, and ppe51 has experimentally 104 been implicated in resistance to isoniazid (INH), although a transporter or channel role 105 for PE/PPE proteins was not recognized at the time 21. Additional associations with drug 106 resistance that had high odds ratios and low false discovery rate (FDR) were found in 107 ppe3, pe29, and ppe34-37 (Fig. 1a, Table 1). Further filtering to associations with FDR 108 <0.00001, we identified 35 promoter and 443 coding variants associated with resistance 109 to at least one drug within our target genes, in addition to the five positive controls 110 (Supplemental Table 1). 111 ppe42 mutation leads to amikacin resistance 112 To test whether the ppe42 mutation causes amikacin and/or kanamycin resistance, we 113 generated a ppe42 deletion strain by recombineering 22, and complemented this strain 114 with wild-type (WT) and ppe42 with the associated mutation, a stop codon at Tyr290. 115 The complex outer membrane lipids Phthiocerol Dimycocerosates (PDIM) can be lost 116 during in vitro culture and affect OM permeability 8,23. To maintain PDIM, we grew all 117 cultures in 100µM sodium propionate 23. We incubated liquid culture of H37Rv WT and 118 the ppe42 deletion mutant in the presence of drugs for seven days and determined cell 119 viability by spotting on solid medium. The ppe42 deletion strain was more resistant to 120 amikacin, a second-line drug used for treatment of multi-drug resistant tuberculosis 20, 121 but not kanamycin when compared to WT (Fig. 1b). We next plated amikacin-treated 122 strains on solid media for quantitating the difference in amikacin susceptibility by colony 123 forming units (CFU) assay. At amikacin concentrations near the minimal inhibitory 124 concentration (MIC), the ppe42 KO mutant was more resistant, with ~2log higher CFU 125 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 4 than the WT. The ppe42 deletion strain regained some sensitivity to amikacin when 126 complemented with WT ppe42, but not when complemented with the copy of ppe42 that 127 carried the clinical mutation (Fig. 1c, see Supplemental Table 2 for all p-values, CFU 128 data for kanamycin in Supplemental Fig. 2). Since the mutated ppe42 phenocopies the 129 deletion strain, the clinical mutation is likely a loss of function mutation. To test if the 130 increased drug resistance of the ppe42 deletion strain is related to transport across the 131 OM, we next tested if the M. smegmatis porin MspA can functionally complement the 132 ppe42 deletion. MspA has previously been shown to be an OM porin in nonpathogenic 133 mycobacteria that can bypass the outer membrane when introduced into pathogenic 134 mycobacteria 24. To test if this OM porin can revert the phenotype of ppe42 deletion, we 135 expressed mspA in the ppe42 knockout strain. Expression of mspA fully restored 136 amikacin susceptibility in the ppe42 deletion background (Fig. 1d), suggesting that 137 ppe42 is also an OM conduit. These data show that the truncation at Tyr290 in ppe42 138 observed in clinical drug resistance confers drug resistance to amikacin but not 139 kanamycin. 140 ppe51 mutation leads to resistance to multiple drugs 141 To experimentally test the association of ppe51 with drug resistance, we generated a 142 ppe51 deletion strain and tested its susceptibility to drugs. Because the association of a 143 genetic mutation especially with second-line drugs is often confounded by multiple drug-144 resistance, we tested against a panel of ten drugs currently in clinical use. The deletion 145 mutant showed increased susceptibility to amikacin, capreomycin, moxifloxacin, 146 streptomycin, and INH (Fig. 2a). Because deletion of ppe51 conferred resistance to 147 multiple drugs, we explored potential compensatory genetic effects of ppe51 deletion by 148 comparing global transcription in the WT and deletion strains by RNA-seq. Deletion of 149 ppe51 led to multiple significant changes in gene expression compared to WT (Fig. 2b). 150 Notably, transcripts for six inner membrane efflux pump genes (mmr, mmpl5, mmps5, 151 Rv1216c-1218c) were significantly induced (Fig. 2b, c, all differentially expressed genes 152 in Supplemental Table 3). Since these genes are known to efflux TB drugs including 153 INH 11, the effect of ppe51 deletion on drug susceptibility is likely indirect and shows 154 coupling of ppe51 with inner membrane transport. 155 ppe3, ppe35, and ppe36 deletion do not affect drug resistance 156 We tested additional candidates from our association study for effects on drug 157 resistance. We tested a ppe35 deletion strain for resistance to pyrazinamide or any 158 other of ten clinical drugs, but the deletion strain did not alter the sensitivity to drugs. 159 Similarly, deletion of ppe3 and ppe36 did not affect drug resistance to any of the ten 160 tested drugs (Supplemental Fig. 2, 3). These data show the limitations of genetic 161 association studies, which might be particularly challenging for the pe-pgrs and ppe 162 genes due to the poorer mapping quality of sequencing data. These data also show that 163 deletion of ppe genes does not generally affect the OM permeability but only does so in 164 the case of specific ppe genes and in the context of specific drugs. 165 pe25/ppe41 affect susceptibility to INH 166 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 5 167 We next sought to identify additional candidates by an orthogonal approach. 168 Transporters can respond transcriptionally to their substrates, a behavior that has also 169 been reported for several pe/ppe transporters 6,8. To identify additional pe/ppe genes 170 that might affect drug resistance but were not detected by our genetic association study, 171 we mined publicly available RNA-seq data for changes in pe/ppe gene expression that 172 is induced by drugs. Twenty-eight datasets from public repositories captured the 173 transcriptional responses to ten drugs. From these data, we identified the pe/ppe genes 174 that were significantly regulated in response to drug exposure and plotted them against 175 the number of drugs to which they responded (Fig. 3a). Many pe/ppe transcripts 176 responded to drug exposure, and several to different drugs. Generally, more pe/ppe 177 transcripts were down- than upregulated in response to drugs (Fig. 3a). ppe51 was also 178 differentially regulated by several drugs, and pe25/ppe41 showed the most consistent 179 directionality of expression change across different drugs. To test if these candidates 180 have any functional consequence for drug resistance, we generated deletion strains for 181 five and tested their susceptibility with the ten-drug panel by spotting assay. The 182 pe25/ppe41 deletion strain was more resistant to INH than WT. None of the other 183 deletion strains showed altered susceptibility to the drugs tested (Supplemental Fig. 4). 184 To quantitate the effect of pe25/ppe41 deletion on drug resistance, we determined the 185 MIC of the WT and deletion strains by CFU assay. At concentrations of INH 186 corresponding to the MIC of drug-susceptible Mtb (0.225-0.45µM), the pe25/ppe41 187 deletion strain was more resistant than the WT strain, with >10-fold higher CFU and ~2-188 fold increase in the MIC50 (Fig. 3b). Complementation of the pe25/ppe41 deletion strain 189 fully restored the susceptibility to INH (Fig. 3b). 190 To test if the increased drug resistance of the pe25/ppe41 deletion strain is 191 related to transport across the OM, we next tested if the M. smegmatis porin MspA can 192 functionally complement the pe25/ppe41 deletion. MspA restored the susceptibility of 193 the pe25/ppe41 deletion strain to INH to WT levels, suggesting that PE25/PPE41 has a 194 similar function to MspA in transport across the OM (Fig. 3b). The decreased 195 susceptibility to INH upon pe25/ppe41 deletion indicates that the PE25/PPE41 complex 196 imports INH. To test if pe25, ppe41, or both affect INH sensitivity, we complemented the 197 pe25/ppe41 deletion strain with each gene individually under the control of an 198 anhydrotetracycline (ATc)-inducible promoter. Complementation with both but not with 199 pe25 or ppe41 alone restored sensitivity to INH, showing that both are required (Fig. 200 3c). The genetic background of a strain can affect drug resistance phenotypes 25,26. To 201 test the effect of strain background and whether the pe25/ppe41 effect on INH 202 susceptibility is conserved beyond the lineage 4 strain H37Rv, we introduced 203 pe25/ppe41 deletions in representative strains from the major circulating lineages—T46 204 (lineage 1) and HN878 (lineage 2). Both strains showed similar changes in INH 205 resistance to the lineage 4 reference strain H37Rv (Fig3d, e). 206 207 Loss of pe25/ppe41 reduces transcription of INH-responsive genes 208 209 INH is a prodrug that requires activation by KatG in the cytoplasm, and it has well-210 defined transcriptional effects 27. To further test for an INH import function of 211 pe25/ppe41 and to test whether pe25/ppe41 alone can alter the levels of INH in the 212 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 6 cytoplasm, we tested if pe25/ppe41 deletion affects the expression of INH-responsive 213 genes. We chose the three INH-responsive genes iniB, acpM, and kasA, and compared 214 their expression by qRT-PCR in WT and the pe25/ppe41 deletion strain after treatment 215 with 0.2µg/mL INH for 5 hours. INH strongly induced the expression of the three INH-216 responsive genes in WT but significantly less in the pe25/ppe41 deletion strain (Fig. 3f). 217 These data indicate that pe25/ppe41 deletion is sufficient to reduce INH concentrations 218 in the cytoplasm, which is consistent with an importer function. 219 220 pe25/ppe41 mutations accelerate drug resistance 221 222 Although neither pe25 nor ppe41 mutations were significantly associated with drug 223 resistance in our genetic association analysis, anecdotal data also link pe25/ppe41 to 224 clinical drug resistance. The Lisboa cluster of Mtb strains has been associated with INH 225 resistance and high rates of multidrug-resistant and extensively drug-resistant 226 tuberculosis in Portugal 28-30. Although INH resistance in these strains is associated with 227 two canonical inhA mutations 31, strains in the Lisboa3 cluster also carry a truncation of 228 ppe41 after bp112 29. To test if this ppe41 truncation could also contribute to INH 229 resistance, we complemented the H37Rv pe25/ppe41 deletion strain with a WT or 230 truncated copy of ppe41. In contrast to the WT gene, the truncated ppe41 did not 231 restore INH sensitivity of the pe25/ppe41 deletion strain (Fig. 4a), suggesting that it 232 contributes to high-level INH resistance in the Lisboa3 strain and/or promotes the 233 emergence of canonical INH mutations. To quantitate the effect of the truncation, we 234 determined the CFU of the WT strain and a pe25/ppe41 deletion strain complemented 235 with the truncated ppe41. The truncated ppe41 conferred higher resistance, with ~2 236 orders of magnitude higher CFUs (Fig. 4b). To test the idea that pe25/ppe41 deletion 237 can affect the rate of emergence of INH resistance, we plated cultures of WT and the 238 pe25/ppe41 deletion strain on agar plates containing INH at 10-, 25-, 125-, and 500-fold 239 the MIC and quantified the resistant colonies that grew under drug pressure by CFU. 240 The pe25/ppe41 deletion strain developed INH-resistant colonies at an about 3.5-fold 241 higher rate than the WT strain across all tested concentrations (Fig. 4c). Sequencing of 242 resistance mutants showed that resistance in the pe25/ppe41 deletion strain was due to 243 canonical katG mutations typically associated with INH resistance in vitro (Supplemental 244 Fig. 5). These findings indicate that loss of pe/ppe function accelerates the emergence 245 of INH resistance. 246 247

248 249 Mtb and related pathogenic mycobacteria have some of the thickest and most 250 impermeable outer membranes known in microbiology 1,32. The mechanism by which 251 Mtb takes up nutrients through this thick barrier in the absence of canonical porins was 252 a longstanding question until the first Mtb transporters from the PE/PPE protein families 253 were discovered. The PE/PPE proteins are an idiosyncratic protein family with no 254 apparent orthologs outside of mycobacteria, and they are generally more frequent in 255 pathogenic mycobacteria that do not have canonical porins. Indeed, several PE/PPE 256 proteins have now been shown to be nutrient transporters 6,8. These and data showing 257 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 7 direct pore-forming function 9 further support the idea that OM transport might be a 258 common function among the PE/PPE proteins. 259 Drug uptake by Mtb is generally presumed to occur by passive diffusion, an idea 260 that has led to a guiding principle in TB drug development that TB drugs should be 261 lipophilic. However, many effective TB drugs in clinical use are not. The effects of 262 PPE42 on amikacin susceptibility, of PPE51 on several drugs, and of PE25/PPE41 on 263 INH susceptibility are indeed consistent with an importer function, and the presence of 264 proteinaceous OM drug importers would provide a plausible explanation for this 265 contradiction. The effects of the PE/PPE proteins tested here on drug susceptibility are 266 highly specific. Of nine PE/PPE proteins tested, only three showed altered drug 267 susceptibility, two of which only to one drug. These data suggest that PE/PPE deletion 268 does not cause a generalized permeability change in the OM but has specific effects on 269 specific compounds. 270 In addition to mutations in ppe42 and ppe51, loss of function mutations of 271 pe25/ppe41 could be a facile mechanism for INH tolerance and/or resistance, and our 272 data indeed show a higher rate of emergence of INH resistance in the pe25/ppe41 273 deletion strain in vitro, suggesting it could be a steppingstone mutation to high-level 274 drug resistance, for example in some Lisboa strains. The contribution of pe/ppe genes 275 to clinical drug resistance combined with the large number of PE/PPE proteins in Mtb 276 represent a previously unrecognized factor of TB drug resistance and raise the 277 possibility that PE/PPE proteins contribute to drug uptake (and perhaps efflux) more 278 widely. Several observations indeed point to a larger role of PE/PPE proteins in drug 279 susceptibility and resistance. A genome-wide chemical-genetic screen recently identified 280 several pe/ppe genes that conferred increased or decreased susceptibility to drugs in 281 vitro 33. These data are consistent with our findings for pe25/ppe41 and INH but require 282 experimental validation for others since CRISPRi is subject to polar effects, and 283 expression of pe/ppe genes can be linked to expression of inner membrane 284 transporters—as we have shown for ppe51. Also, the large number of pe/ppe genes 285 makes redundancy likely, potentially masking the effects of single gene deletion or 286 knockdown and complicating the identification of PE/PPEs that affect drug susceptibility 287 and resistance. 288 The physiologic substrate(s) and cellular function(s) of PPE42 and PE25/PPE41 289 remain to be identified. Although drugs are arguably accidental substrates of PE/PPE 290 transporters, their cellular substrates might bear chemical resemblance to the drugs 291 they transport. Interestingly, PPE42 contains a predicted serine hydrolase domain. 292 Serine hydrolases play an important role in the resistance to beta-lactams (such as 293 amikacin and kanamycin) through drug inactivation. The predicted AlphaFold structure 294 for the PPE42 hydrolase domain shows a canonical catalytic site organization and a Ser 295 residue in the expected position, suggesting it might be an active enzyme. However, 296 loss of the hydrolase domain in the ppe42 clinical resistance strain would have the 297 opposite effect to known hydrolase-mediated drug resistance mechanisms and not lead 298 to drug inactivation. The molecular function of the hydrolase domain in amikacin 299 resistance needs further study, but it could serve as a binding domain for drug and thus 300 facilitate entry. Our data identify a function for PE/PPE proteins in drug susceptibility 301 that is consistent with their emerging OM transporter function. Although transporter 302 function of PE/PPE proteins has now been shown in several cases, and many 303 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 8 previously observed PE/PPE phenotypes are fully consistent with transporter function 304 7,34, much remains unknown about these most idiosyncratic protein families. For 305 example, the topology and composition of the complexes, which are and which are not 306 pore-like, the way in which pores or channels are formed across the OM, and the 307 biochemical mechanism of transport all remain unknown but will inform drug 308 development, drug targeting, diagnosis of drug resistance, and new strategies to 309 counter drug resistance. 310 311 ACKNOWLEGEMENTS 312 313 This work was supported by National Institutes of Health (NIH) grant R01AI180452 to 314 CG. 315 316

317 Genetic association analysis 318 Sequences were aligned to the MT_H37R7_V3 reference sequence using BWA-MEM 319 35, and the assembled genomes were aligned to the same reference using Minimap2 320 36,37. The 17,047 files were then processed by BCFtools mpileup and BCFtools call 38 to 321 make a joint-called VCF with haploid genotype calls. We considered promoter and gene 322 body mutations for the analysis. A selection of 7,402 SNPs with less than 10% missing 323 data in all three datasets and a minor allele count of at least 5, representing a random 324 10% of genomic regions, were used for estimating relatedness among strains. 325 Relatedness was estimated by multiplying the centered genotype matrix by its 326 transpose. Individual relatedness values were divided by the number of non-missing 327 variants in common between two samples, and negative relatedness values were set to 328 zero. The lineage of each strain was identified based on the set of lineage-specific 329 SNPs packaged with TBprofiler v6.5.0 39. Samples were retained if they had less than 330 10% missing data, showed agreement between assigned lineage and relatedness to 331 other strains, and had antibiotic resistance phenotype data available, leaving 16,891 332 strains. We tested the 104 pe/ppe genes for association with phenotypic drug 333 resistance. We excluded the pe-pgrs subfamily (64 genes) because it is known to have 334 highly variable sequence. We included aphC, gyrA, inhA, katG, and rpoB as controls 335 with known resistance associations. Within these 109 genes, we identified 19,154 336 nucleotide variants after splitting multiallelic variants into individual biallelic variants. 337 Association analysis was performed at the nucleotide level for promoter variants and 338 amino acid level for coding variants. Coding variants were agglomerated by amino acid 339 position and type of variant (substitution, frameshift, in-frame insertion or deletion). Any 340 given promoter or coding variant was used in association analysis with a given drug 341 resistance trait if it had at least five strains with the minor allele and a recorded 342 phenotype, totaling 357 unique promoter variants and 3,418 unique agglomerated 343 coding variants across all 25 drugs. Association testing was conducted with PySeer 40 344 using the linear mixed model approach, with the relatedness matrix as a random effect 345 (similarity matrix), and dataset (Sanger Institute, CRyPTIC, TB Portals) as a fixed 346 covariate. At FDR<0.00001 across all tests, 35 promoter variants and 478 coding 347 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 9 variants were significant for resistance to at least one drug. If the number of susceptible 348 individuals with the minor allele was required to be zero, and the minor allele had to 349 appear in resistant strains of at least two major Mtb lineages, there were six promoter 350 variants and 41 coding variants significant for resistance to at least one drug within our 351 target genes, including the five positive controls and 11 out of the 104 pe/ppe genes. 352 Media and growth conditions 353 Mycobacterium tuberculosis (Mtb) H37Rv, a lineage 4 strain, was used as the parental 354 strain for generating all mutants unless indicated otherwise. Strains were grown in 355 Middlebrook 7H9 medium (Difco), supplemented with 10% (vol/vol) oleic acid-albumin-356 dextrose-catalase (OADC) enrichment (BBL; Becton Dickinson), 0.5% glycerol, and 357 0.05% Tween 80 or 0.05% Tyloxapol. The medium is referred to as “7H9+GO” when 358 supplemented with glycerol and OADC only, “7H9+GOT” with Tween 80, “7H9+GOTy” 359 with Tyloxapol, and “7H9+GOTyP” with the addition of 100 µM sodium propionate to 360 maintain PDIM levels 23. For solid media, 7H10 agar was supplemented with 10% 361 OADC and 0.5% glycerol. Strains harboring antibiotic resistance cassettes were 362 cultured with the appropriate antibiotic: 50 μ g/mL hygromycin, 30 μ g/mL kanamycin, or 363 25 μ g/mL zeocin. All drugs used for the drug screening assay were purchased from 364 Sigma, and stock solutions were prepared in either water or DMSO. 365 Creation of deletion and complemented strains 366 The deletion strains were generated using recombineering, as described previously 22. 367 Initially, 500 bp upstream and downstream of the gene of interest, along with the 368 hygromycin resistance cassette (primers Hyg_F and Hyg_R), were amplified separately 369 (all primers used for each deletion are listed in Supplemental Table 4). The PCR 370 fragments were then Gibson ligated to the 5′ and 3′ ends of the hygromycin resistance 371 cassette to create the recombineering cassette. This linear recombineering cassette 372 was PCR-amplified, purified, and electroporated into the Mtb H37Rv strain carrying the 373 recombineering plasmid pNIT:Etc 41. Hygromycin-resistant colonies were screened, and 374 the positive clones were confirmed by DNA sequencing. The PE25/PPE41 deletion was 375 also generated in Mtb HN878 and T46 using the same recombineering pNIT:Etc strain 376 and cassette used for H37Rv. 377 For the complemented strains, the ppe42, ppe42 Y290*, pe25, ppe41, 378 pe25/ppe41, and pe25/ppe41 (Δ 112) coding regions, as well as the M. smegmatis 379 mspA coding region were amplified by PCR using primers listed in Supplemental Table 380 4 and Gibson cloned into the pDTCF plasmid (Zeocin resistance) with a C-terminal 381 FLAG tag, under the control of an anhydrotetracycline (ATc)-inducible promoter. The 382 resulting plasmids were electroporated into the H37Rv pe25/ppe41 deletion strain, and 383 positive clones were selected by growth in hygromycin and zeocin. 384 Drug screening 385 Drug screening was carried out in 96-well plates. Cultures were grown to log phase in 386 7H9+GOTyP media and incubated with drugs in 2-fold dilutions at a final optical density 387 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 10 (OD) of 0.005. For pyrazinamide, we used medium at pH6. The assay plates were 388 incubated at 37°C for 7 days, and cell viability was measured using either the agar 389 spotting method, Alamar Blue assay, or colony-forming unit (CFU) counting. For the 390 spotting method, cultures were mixed and 3 µL were spotted onto 7H10+GO agar 391 plates, followed by incubation at 37°C for 2-3 weeks. For CFU counting, cultures were 392 diluted and plated onto 7H10 agar plates, which were incubated for 3-4 weeks at 37°C. 393 RNA sequencing 394 H37Rv and the ppe51 deletion strain were grown to an OD600 of 0.8 in 7H9+GOTyP 395 medium. Cells were pelleted and resuspended in buffered water, then incubated for an 396 additional 5 hours. Following incubation, cells were pelleted at 4,000 g for 5 minutes at 397 4°C, resuspended in Trizol, and lysed by bead beating for 30 seconds at 6 m/s for 3 398 cycles, with intermittent cooling on ice. Cell debris was pelleted at 20,000 g for 1 minute, 399 and the supernatant was transferred to a heavy phase-lock gel tube containing 300 µL 400 of chloroform. The tubes were inverted for 2 minutes and centrifuged at 20,000 g for 5 401 minutes. RNA in the aqueous phase was precipitated with 300 µL of isopropanol and 402 300 µL of high-salt solution (0.8 M sodium citrate, 1.2 M sodium chloride). RNA was 403 purified using the QIAGEN RNeasy kit, and ribosomal RNA was depleted using a 404 previously published protocol 42. Briefly, a biotinylated oligo mixture of 23S, 16S, and 5S 405 was incubated with RNA to anneal to rRNA, followed by incubation with streptavidin 406 beads. mRNA was purified from the supernatant using Ampure XP beads. A cDNA 407 library was generated using the NEBNext Ultra II RNA Library Prep Kit, and each 408 replicate was barcoded in the DNA library using the NEBNext Multiplex Oligos for 409 Illumina. Libraries were quantified using the KAPA qPCR quantification kit, pooled, and 410 sequenced at the University of Washington Northwest Genomics Center using the 411 Illumina NextSeq 500 High Output v2 Kit.Read alignment was performed using the 412 Bowtie 2 custom processing pipeline 413 (https://github.com/robertdouglasmorrison/DuffyNGS, 414 https://github.com/robertdouglasmorrison/DuffyTools). Gene expression changes were 415 identified using a combination of five differential expression tools within DuffyTools: 416 round robin, RankProduct, significance analysis of microarrays (SAMs), EdgeR, and 417 DeSeq2. The results from each DE tool were combined using a weighted average of 418 fold change and significance (p-value). Genes with an averaged absolute fold change of 419 more than 2-fold and a p-value <0.01 were considered differentially expressed. 420 qRT-PCR analysis 421 Liquid cultures of Mtb H37Rv WT and pe25/ppe41 deletion strains were grown to early 422 log phase in 7H9+GOTyP media and subsequently treated with 0.2 µg/mL isoniazid for 423 5 hours. RNA was extracted using Trizol, purified, and cDNA was synthesized with 424 SuperScript IV polymerase and random hexamer primers. mRNA expression levels for 425 iniB, acpM, and kasA genes were quantified by qRT-PCR using SybrGreen iTaq 426 chemistry. Expression levels were normalized to sigA mRNA expression, and the 427 relative mRNA levels were calculated using the 2−ΔΔ Cq method and plotted as the ratio of 428 mRNA expression for each strain. 429 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 11 Measuring the in vitro frequency of drug resistance 430 To measure the frequency of drug resistance, Mtb H37Rv WT and pe25/ppe41 deletion 431 strains were grown to an OD600 ~0.6 in 7H9+GOTyP medium. The cultures were 432 diluted to approximately 1 x 108 and 1 x 107 CFU/mL and plated onto non-drug-433 containing 7H10+GO agar plates to determine the total viable cell count, and drug-434 containing 7H10 agar plates supplemented with isoniazid at concentrations 10X–500X 435 above the MIC. Plates were incubated at 37°C for 3-4 weeks. Colonies on drug-free 436 plates represented the total viable cell count, while colonies on drug plates indicated 437 resistant mutants. The frequency of resistance was calculated by dividing the number of 438 colonies on drug-containing plates by the total viable cell count from non-drug plates. 439 FIGURE LEGENDS 440 441 Figure 1. Multiple pe/ppe mutations are associated with clinical drug resistance. 442 a. Genetic associations of pe/ppe mutations and clinical drug resistance are shown as 443 scatterplot with odds ratio and allele frequency. False discovery rate (FDR) is shown 444 according to color scheme (right). Larger symbols indicate the top ten candidates based 445 on FDR, and the respective drug and mutation are indicated. pe-pgrs genes were 446 excluded from the analysis. b. Spotting assay after exposure to drug shows that 447 deletion of ppe42 increases resistance to amikacin, but not kanamycin. Cultures were 448 grown with drug for 7 days before spotting on agar without drug. c. Deletion of ppe42 is 449 partially complemented by WT ppe42 but not the Tyr290 truncation, as shown by CFU 450 assay. d. The MspA porin reverses the resistance of the ppe42 KO to amikacin. Error 451 bars for CFU assays show standard deviation of three biological replicates. P-values for 452 comparisons in c. and d. are given in Supplemental Table 2. Abbreviations: AMI: 453 amikacin. EMB: ethambutol. KAN: kanamycin. LZD: linezolid. We used Gemini and 454 ChatGPT for writing a Python script to generate Fig. 1a. We reviewed and validated all 455 AI outputs before inclusion. 456 457 Figure 2. Deletion of ppe51 increases drug resistance and upregulates inner 458 membrane transporters. a. Spotting of WT and ppe51 KO strain grown in presence of 459 drugs in 2-fold dilution shows increased resistance to multiple drugs. b. RNA-seq 460 analysis shows increased transcription of several known efflux pump genes as a result 461 of ppe51 deletion. c. The efflux pump genes affected by ppe51 deletion. All differentially 462 expressed genes are given in Supplemental Table 3. Abbreviations: AMI: amikacin. 463 CAP: capreomycin. STR: streptomycin. MOX: moxifloxacin. INH: isoniazid. 464 465 Fi gure 3. The PE25/PPE41 complex affects isoniazid susceptibility. a. The 466 transcriptional response of pe/ppe genes to drugs. pe/ppe genes showing 467 transcriptional up- (green) or downregulation (red) in response to drugs in historical 468 RNA-seq data were plotted against the number of drugs they were found to respond to. 469 Deletions of pe and ppe genes tested in this study are labeled. Direction of change is 470 indicated by color. b. A pe25/ppe41 deletion strain shows higher resistance to INH. The 471 heterologous OM porin MspA from M. smegmatis can functionally complement the 472 pe25/ppe41 deletion. Bacteria were enumerated by colony forming units (CFU) assay 473 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 12 and data are shown as the mean ± SEM of two experiments performed in quadruplicate. 474 Statistical significance was determined by multiple comparison t-test. **: p< 0.01, ***: p< 475 0.001. c. Complementation of the pe25/ppe41 deletion restores susceptibility to INH, 476 but not complementation with either pe25 or ppe41 alone, as shown by spotting assay. 477 d. Deletion of pe25/ppe41 also reduces the susceptibility to clinically relevant 478 concentrations of INH in Mtb strains from lineage 1 (T46) and e. lineage 2 (HN878) f. 479 The known INH-responsive genes iniB, acpM, and kasA are induced by INH exposure 480 in the H37Rv WT but less so in the pe25/ppe41 deletion strain, as determined by qRT-481 PCR. Datapoints show the average of two technical replicates for four biological 482 replicates **: p< 0.01, ***: p< 0.001. 483 484 Figure 4. ppe41 truncation contributes to INH resistance. a. The ppe41 deletion at 485 bp112 in the Lisboa3 strain family increases resistance of Mtb to INH, as shown by 486 spotting after growth on drug for 7 days. b. ppe41 truncation increases INH resistance 487 as quantified by CFU assay. Data points represent average of two technical replicates 488 for three biological replicates c. Deletion of pe25/ppe41 increases the rate at which INH 489 resistance emerges. Mtb cultures were plated onto 7H10+GO plates without drug and 490 with INH at concentrations 10x–500x above the MIC. Colonies on drug-free plates 491 represented the total viable cell count, while colonies on plates with drug were counted 492 as resistant mutants. The mutation rate was calculated by dividing the number of 493 colonies on drug-containing plates by the total viable cell count from non-drug plates 494 plates. Each datapoint represents a biological replicate. Error bars are mean ± SD. 495 Statistical significance was determined by multiple comparison t-test. ***: p10 are 499 s hown. 500 501 EXTENDED AND SUPPLEMENTARY DATA 502 503 Supplemental Figure 1: Mapping quality comparison of pe, ppe, and pe-pgrs 504 genes. Within 4,031 protein-coding genes in the M. tuberculosis H37Rv v3 reference 505 genome, 35 were classified as pe, 69 as ppe, 74 as pe-pgrs, and 3,863 as others (x-506 axis). Sequences from the 2,659 strains from the Wellcome Trust Sanger Institute pilot 507 study were aligned to the reference using Bowtie2 and genotyped using GATK 508 HaplotypeCaller, with joint calling performed by GATK GenomicsDBImport and 509 GenotypeGVCFs. With each gene, the mapping quality (MQ) as reported by GATK was 510 averaged across all variants (y-axis). 511 512 Supplemental Figure 2: CFU data for WT, ppe42 deletion, and ppe35 deletion 513 mutants grown in kanamycin and pyrazinamide, respectively. 514 515 Supplemental Figure 3: Negative spotting data for hits from genetic association 516 study. Cultures were grown with drug for 7 days before spotting on agar without drug. 517 518 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 13 Supplemental Figure 4: Negative spotting data for candidates from the RNA-seq 519 analysis. Cultures were grown with drug for 7 days before spotting on agar without 520 drug. 521 522 Supplemental Figure 5: katG mutations in INH-resistant clones of the pe25/ppe41 523 deletion strain. Mutations were identified by DNA sequencing of the katG gene. The 524 corresponding amino acid changes in KatG are shown. Numbers in parentheses 525 indicate multiple occurrences of the mutation. 526 527 Supplemental Table 1: Genetic associations between clinical drug resistance and 528 pe and ppe genes. All associations from our targeted association study with an FDR 529 below 10-5 are listed. 530 531 Supplemental Table 2: p-values for the CFU comparison of ppe42 WT and mutant 532 strains. NS: Not significant. **: p<0.01. ***: p<0.001 533 534 Supplemental Table 3: Differentially expressed genes in the ppe51 deletion strain 535 compared to WT. RNA-seq analysis was done after four days of growth. 536 537 Supplemental Table 4: Primers used for generating knockout strains, 538 complementation strains, and primers used for qRT-PCR. 539 540 541 542

543 544 1 Brennan, P . J. & Nikaido, H. The envelope of mycobacteria. Annu Rev Biochem 545 64, 29-63, doi:10.1146/annurev.bi.64.070195.000333 (1995). 546 2 Dulberger, C. L., Rubin, E. J. & Boutte, C. C. The mycobacterial cell envelope - a 547 moving target. Nat Rev Microbiol 18, 47-59, doi:10.1038/s41579-019-0273-7 548 (2020). 549 3 Nikaido, H. & Jarlier, V. Permeability of the mycobacterial cell wall. Res Microbiol 550 142, 437-443, doi:10.1016/0923-2508(91)90117-s (1991). 551 4 Cole, S. T. et al. Deciphering the biology of Mycobacterium tuberculosis from the 552 complete genome sequence. Nature 393, 537-544, doi:10.1038/31159 (1998). 553 5 Babu Sait, M. R. et al. PPE51 mediates uptake of trehalose across the 554 mycomembrane of Mycobacterium tuberculosis. Sci Rep 12, 2097, 555 doi:10.1038/s41598-022-06109-7 (2022). 556 6 Boradia, V., Frando, A. & Grundner, C. The Mycobacterium tuberculosis 557 PE15/PPE20 complex transports calcium across the outer membrane. PLoS Biol 558 20, e3001906, doi:10.1371/journal.pbio.3001906 (2022). 559 7 Mitra, A., Speer, A., Lin, K., Ehrt, S. & Niederweis, M. PPE Surface Proteins Are 560 Required for Heme Utilization by Mycobacterium tuberculosis. mBio 8, 561 doi:10.1128/mBio.01720-16 (2017). 562 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 14 8 Wang, Q. et al. PE/PPE proteins mediate nutrient transport across the outer 563 membrane of Mycobacterium tuberculosis. Science 367, 1147-1151, 564 doi:10.1126/science.aav5912 (2020). 565 9 Sankey, N. et al. Role of the Mycobacterium tuberculosis ESX-4 Secretion 566 System in Heme Iron Utilization and Pore Formation by PPE Proteins. mSphere 567 8, e0057322, doi:10.1128/msphere.00573-22 (2023). 568 10 Sarathy, J. P., Dartois, V. & Lee, E. J. The role of transport mechanisms in 569 mycobacterium tuberculosis drug resistance and tolerance. Pharmaceuticals 570 (Basel) 5, 1210-1235, doi:10.3390/ph5111210 (2012). 571 11 Laws, M., Jin, P . & Rahman, K. M. Efflux pumps in Mycobacterium tuberculosis 572 and their inhibition to tackle antimicrobial resistance. Trends Microbiol 30, 57-68, 573 doi:10.1016/j.tim.2021.05.001 (2022). 574 12 Walker, T. M. et al. Whole-genome sequencing for prediction of Mycobacterium 575 tuberculosis drug susceptibility and resistance: a retrospective cohort study. 576 Lancet Infect Dis 15, 1193-1202, doi:10.1016/S1473-3099(15)00062-6 (2015). 577 13 Rosenthal, A. et al. The TB Portals: an Open-Access, Web-Based Platform for 578 Global Drug-Resistant-Tuberculosis Data Sharing and Analysis. J Clin Microbiol 579 55, 3267-3282, doi:10.1128/JCM.01013-17 (2017). 580 14 The, C. C. Genome-wide association studies of global Mycobacterium 581 tuberculosis resistance to 13 antimicrobials in 10,228 genomes identify new 582 resistance mechanisms. PLoS Biol 20, e3001755, 583 doi:10.1371/journal.pbio.3001755 (2022). 584 15 Copin, R. et al. Sequence diversity in the pe_pgrs genes of Mycobacterium 585 tuberculosis is independent of human T cell recognition. mBio 5, e00960-00913, 586 doi:10.1128/mBio.00960-13 (2014). 587 16 Gomez-Gonzalez, P. J. et al. Functional genetic variation in pe/ppe genes 588 contributes to diversity in Mycobacterium tuberculosis lineages and potential 589 interactions with the human host. Front Microbiol 14, 1244319, 590 doi :10.3389/fmicb.2023.1244319 (2023). 591 17 McEvoy, C. R. et al. Comparative analysis of Mycobacterium tuberculosis pe and 592 ppe genes reveals high sequence variation and an apparent absence of selective 593 constraints. PLoS One 7, e30593, doi:10.1371/journal.pone.0030593 (2012). 594 18 Marin, M. et al. Benchmarking the empirical accuracy of short-read sequencing 595 across the M. tuberculosis genome. Bioinformatics 38, 1781-1787, 596 doi:10.1093/bioinformatics/btac023 (2022). 597 19 Farhat, M. R. et al. GWAS for quantitative resistance phenotypes in 598 Mycobacterium tuberculosis reveals resistance genes and regulatory regions. 599 Nat Commun 10, 2128, doi:10.1038/s41467-019-10110-6 (2019). 600 20 Organization, W. H. WHO consolidated guidelines on drug-resistant tuberculosis. 601 (2019). 602 21 Martini, M. C. et al. Loss of RNase J leads to multi-drug tolerance and 603 accumulation of highly structured mRNA fragments in Mycobacterium 604 tuberculosis. PLoS Pathog 18, e1010705, doi:10.1371/journal.ppat.1010705 605 (2022). 606 22 van Kessel, J. C. & Hatfull, G. F. Recombineering in Mycobacterium tuberculosis. 607 Nat Methods 4, 147-152, doi:10.1038/nmeth996 (2007). 608 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 15 23 Mulholland, C. V. et al. Propionate prevents loss of the PDIM virulence lipid in 609 Mycobacterium tuberculosis. Nat Microbiol 9, 1607-1618, doi:10.1038/s41564-610 024-01697-8 (2024). 611 24 Mailaender, C. et al. The MspA porin promotes growth and increases antibiotic 612 susceptibility of both Mycobacterium bovis BCG and Mycobacterium 613 tuberculosis. Microbiology (Reading) 150, 853-864, doi:10.1099/mic.0.26902-0 614 (2004). 615 25 Nebenzahl-Guimaraes, H., Jacobson, K. R., Farhat, M. R. & Murray, M. B. 616 Systematic review of allelic exchange experiments aimed at identifying mutations 617 that confer drug resistance in Mycobacterium tuberculosis. J Antimicrob 618 Chemother 69, 331-342, doi:10.1093/jac/dkt358 (2014). 619 26 Safi, H. et al. Evolution of high-level ethambutol-resistant tuberculosis through 620 interacting mutations in decaprenylphosphoryl-beta-D-arabinose biosynthetic and 621 utilization pathway genes. Nat Genet 45, 1190-1197, doi:10.1038/ng.2743 (2013). 622 27 Wilson, M. et al. Exploring drug-induced alterations in gene expression in 623 Mycobacterium tuberculosis by microarray hybridization. Proc Natl Acad Sci U S 624 A 96, 12833-12838, doi:10.1073/pnas.96.22.12833 (1999). 625 28 Perdigao, J. et al. Genetic analysis of extensively drug-resistant Mycobacterium 626 tuberculosis strains in Lisbon, Portugal. J Antimicrob Chemother 65, 224-227, 627 doi:10.1093/jac/dkp452 (2010). 628 29 Perdigao, J. et al. Unraveling Mycobacterium tuberculosis genomic diversity and 629 evolution in Lisbon, Portugal, a highly drug resistant setting. BMC Genomics 15, 630 991, doi:10.1186/1471-2164-15-991 (2014). 631 30 Portugal, I. et al. Outbreak of multiple drug-resistant tuberculosis in Lisbon: 632 detection by restriction fragment length polymorphism analysis. Int J Tuberc Lung 633 Dis 3, 207-213 (1999). 634 31 Machado, D. et al. Hi gh-level resistance to isoniazid and ethionamide in 635 multidrug-resistant Mycobacterium tuberculosis of the Lisboa family is associated 636 with inhA double mutations. J Antimicrob Chemother 68, 1728-1732, 637 doi:10.1093/jac/dkt090 (2013). 638 32 Brennan, P . J. Structure, function, and biogenesis of the cell wall of 639 Mycobacterium tuberculosis. Tuberculosis (Edinb) 83, 91-97, doi:10.1016/s1472-640 9792(02)00089-6 (2003). 641 33 Li, S. et al. CRISPRi chemical genetics and comparative genomics identify genes 642 mediating drug potency in Mycobacterium tuberculosis. Nat Microbiol 7, 766-779, 643 doi:10.1038/s41564-022-01130-y (2022). 644 34 Ates, L. S. et al. Essential Role of the ESX-5 Secretion System in Outer 645 Membrane Permeability of Pathogenic Mycobacteria. PLoS Genet 11, e1005190, 646 doi:10.1371/journal.pgen.1005190 (2015). 647 35 Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler 648 transform. Bioinformatics 25, 1754-1760, doi:10.1093/bioinformatics/btp324 649 (2009). 650 36 Li, H. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics 34, 651 3094-3100, doi:10.1093/bioinformatics/bty191 (2018). 652 37 Li, H. New strategies to improve minimap2 alignment accuracy. Bioinformatics 653 37, 4572-4574, doi:10.1093/bioinformatics/btab705 (2021). 654 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint 16 38 Danecek, P. et al. Twelve years of SAMtools and BCFtools. Gigascience 10, 655 doi:10.1093/gigascience/giab008 (2021). 656 39 Phelan, J. E. et al. Integrating informatics tools and portable sequencing 657 technology for rapid detection of resistance to anti-tuberculous drugs. Genome 658 Med 11, 41, doi:10.1186/s13073-019-0650-x (2019). 659 40 Lees, J. A., Galardini, M., Bentley, S. D., Weiser, J. N. & Corander, J. pyseer: a 660 comprehensive tool for microbial pangenome-wide association studies. 661 Bioinformatics 34, 4310-4312, doi:10.1093/bioinformatics/bty539 (2018). 662 41 Murphy, K. C., Papavinasasundaram, K. & Sassetti, C. M. Mycobacterial 663 recombineering. Methods Mol Biol 1285, 177-199, doi:10.1007/978-1-4939-2450-664 9_10 (2015). 665 42 Culviner, P. H., Guegler, C. K. & Laub, M. T. A Simple, Cost-Effective, and Robust 666

for rRNA Depletion in RNA-Sequencing Studies. mBio 11, 667 doi:10.1128/mBio.00010-20 (2020). 668 669 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint Gene Drug Variant FDR global p-value Allele frequency Odds ratio ppe42 Amikacin Y290* 4.86E-48 1.94E-68 3.69E-03 116.36 ppe35 Ethambutol I539V 1.35E-46 1.30E-56 7.31E-03 37.07 pe29 Amikacin G30R 1.02E-41 2.88E-64 3.69E-03 84.67 ppe42 Kanamycin Y290* 8.74E-40 5.45E-52 3.82E-03 126.06 pe29 Kanamycin G30R 9.02E-35 5.84E-49 3.82E-03 81.76 ppe3 Amikacin G60E 8.64E-33 6.79E-280 1.67E-02 86.47 ›ppe51 Amikacin A222T 1.10E-31 5.87E-56 3.30E-03 74.39 ppe36 Ethionamide W125R 8.77E-31 1.50E-62 1.03E-02 20.87 ppe37 Amikacin D381N 1.65E-30 6.21E-278 1.68E-02 82.34 ppe47 Amikacin E124G 2.63E-30 8.90E-278 1.66E-02 85.79 Table 1 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted September 24, 2025. ; https://doi.org/10.1101/2025.09.24.678375doi: bioRxiv preprint