key: cord-0331431-zhgczajx authors: Brankin, Alice; Malone, Kerri M; Barilar, Ivan; Battaglia, Simone; Borroni, Emanuele; Brandao, Angela Pires; Cabibbe, Andrea Maurizio; Carter, Joshua; Cirillo, Daniela Maria; Claxton, Pauline; Clifton, David A; Cohen, Ted; Coronel, Jorge; Crook, Derrick W; Dreyer, Viola; Earle, Sarah G; Escuyer, Vincent; Ferrazoli, Lucilaine; Fowler, Philip W; Gao, George Fu; Gardy, Jennifer; Gharbia, Saheer; Ghisi, Kelen Teixeira; Ghodousi, Arash; Cruz, Ana Luíza Gibertoni; Grandjean, Louis; Grazian, Clara; Groenheit, Ramona; Guthrie, Jennifer L; He, Wencong; Hoffmann, Harald; Hoosdally, Sarah J; Hunt, Martin; Iqbal, Zamin; Ismail, Nazir Ahmed; Jarrett, Lisa; Joseph, Lavania; Jou, Ruwen; Kambli, Priti; Khot, Rukhsar; Knaggs, Jeff; Koch, Anastasia; Kohlerschmidt, Donna; Kouchaki, Samaneh; Lachapelle, Alexander S; Lalvani, Ajit; Lapierre, Simon Grandjean; Laurenson, Ian F; Letcher, Brice; Lin, Wan-Hsuan; Liu, Chunfa; Liu, Dongxin; Mandal, Ayan; Mansjö, Mikael; Matias, Daniela; Meintjes, Graeme; de Freitas Mendes, Flávia; Merker, Matthias; Mihalic, Marina; Millard, James; Miotto, Paolo; Mistry, Nerges; Moore, David; Musser, Kimberlee A; Ngcamu, Dumisani; Nhung, Hoang Ngoc; Niemann, Stefan; Nilgiriwala, Kayzad Soli; Nimmo, Camus; Okozi, Nana; Oliveira, Rosangela Siqueira; Omar, Shaheed Vally; Paton, Nicholas; Peto, Timothy EA; Pinhata, Juliana Maira Watanabe; Plesnik, Sara; Puyen, Zully M; Rabodoarivelo, Marie Sylvianne; Rakotosamimanana, Niaina; Rancoita, Paola MV; Rathod, Priti; Robinson, Esther; Rodger, Gillian; Rodrigues, Camilla; Rodwell, Timothy C; Roohi, Aysha; Santos-Lazaro, David; Shah, Sanchi; Kohl, Thomas Andreas; Smith, Grace; Solano, Walter; Spitaleri, Andrea; Supply, Philip; Surve, Utkarsha; Tahseen, Sabira; Thuong, Nguyen Thuy Thuong; Thwaites, Guy; Todt, Katharina; Trovato, Alberto; Utpatel, Christian; Van Rie, Annelies; Vijay, Srinivasan; Walker, Timothy M; Sarah Walker, A; Warren, Robin; Werngren, Jim; Wijkander, Maria; Wilkinson, Robert J; Wilson, Daniel J; Wintringer, Penelope; Xiao, Yu-Xin; Yang, Yang; Yanlin, Zhao; Yao, Shen-Yuan; Zhu, Baoli title: A data compendium of Mycobacterium tuberculosis antibiotic resistance date: 2021-09-16 journal: bioRxiv DOI: 10.1101/2021.09.14.460274 sha: f05b5acb6084d1b78f30e548f7c38491b9ec4128 doc_id: 331431 cord_uid: zhgczajx The Comprehensive Resistance Prediction for Tuberculosis: an International Consortium (CRyPTIC) presents here a global collection of 15,211 Mycobacterium tuberculosis clinical isolates, all of which have undergone whole genome sequencing and have had their minimum inhibitory concentrations to 13 antitubercular drugs measured. The isolates represent five major M. tuberculosis lineages originating from 23 countries across four continents. 6,814 isolates were found resistant to at least one drug, including 2,129 samples fully satisfy the clinical definitions of RR/MDR, pre-XDR or XDR. Resistance status to eight antitubercular drugs can be accurately predicted using a genetic mutation catalogue for over 90% of the isolates. Furthermore, we show the presence of suspected resistance conferring mutations for isolates resistant to the newly introduced drugs bedaquiline, clofazimine, delamanid and linezolid. Finally, a case study of rifampicin mono-resistance is presented to showcase how this compendium could be used to advance our genetic understanding of rare resistance phenotypes and evaluate the likely performance of a widely used molecular diagnostic tool. It is hoped that this compendium, the largest M. tuberculosis matched phenotypic and genotypic dataset to date, will facilitate and inspire new research projects for years to come. Tuberculosis (TB) is a curable and preventable disease; 85% of those afflicted 104 can be successfully treated with a six-month regimen. Despite this, TB is the world's 105 top infectious disease killer with 10 million new cases and 1.2 million deaths estimated 106 in 2019 alone [1] . Furthermore, drug resistant TB (DR-TB) is a continual threat; almost 107 half a million people were estimated to have developed resistance to the first-line drug 108 rifampicin (RR-TB), with three quarters of these cases estimated to be multidrug-109 resistant (MDR-TB, resistant to first-line drugs isoniazid and rifampicin) [1] . Worryingly, 110 only 44% of DR-TB cases were officially notified and just over half of these cases were 111 successfully treated (57%) [1] . 112 To address these issues, the World Health Organisation (WHO) is encouraging 114 the development of better, faster and more targeted diagnostic and treatment 115 strategies through its EndTB campaign. Of particular interest is universal drug 116 susceptibility testing (DST). Conventionally, DST relies on lengthy (4 weeks minimum) 117 culture-based methods that require strict biosafety conditions for Mycobacterium 118 tuberculosis. The development of rapid genetics-based assays has decreased 119 diagnostic time to as little as 2 hours through the detection of specific resistance 120 conferring mutations e.g. the Cepheid Xpert® MTB/RIF test [2, 3] . However, assay 121 bias towards specific genic regions can result in misdiagnosis of resistance, the 122 prescription of ineffective treatment regimens and subsequent spread of multi-drug 123 resistant disease, as seen during an outbreak in Eswatini of an MDR strain harbouring 124 an rpoB I491F mutation not detected by the Xpert® MTB/RIF assay [4] [5] [6] . 125 126 Furthermore, detection of rifampicin resistance with the Xpert® MTB/RIF line 127 probe assay (LPA) assay is used to infer MDR-TB as epidemiologically, rifampicin 128 resistance tends to coincide with resistance to isoniazid [7] . While this modus operandi 129 is successful at pragmatically identifying potential MDR cases quickly and effectively, 130 it is not generally true that a single path towards developing MDR or extensively drug Participating collection centres varied in their isolate collection approaches, but 185 the consortium collectively aimed to oversample for M. tuberculosis isolates with drug 186 resistance and multi-drug resistance. 187 188 Plate assay 189 The CRyPTIC consortium designed two variants of the Sensititre MYCOTB 190 plate (Thermo Fisher Scientific Inc., USA) named the "UKMYC5" and "UKMYC6" 191 microtitre plates [10, 11] . These plates contain five to ten doubling dilutions of 13 Pharmaceutical Co., Ltd. and Janssen Pharmaceutica respectively. The UKMYC5 197 plate also contained para-aminosalicylic acid (PAS), however the MICs were not 198 reproducible and hence it was excluded from the UKMYC6 plate design and is not 199 included in any subsequent analysis [10] . 200 A standard operating protocol for sample processing was defined by CRyPTIC 201 as previously described [10, 11] . Clinical samples were sub-cultured using 7H10 agar 202 plates, Lowenstein-Jensen tubes or MGIT tubes. Bacterial cell suspensions (0.5 203 McFarland standard, saline Tween) prepared from (no later than) 14-day old colonies 204 were diluted 100X in 10 ml enriched 7H9 broth prior to plate inoculation. A semi-205 automated Sensititre Autoinoculator (Thermo Fisher, Scientific Inc., USA) was used to 206 inoculate 100 µl prepared cell suspensions (1.5 x 10 5 CFU/ml [5 x 10 4 CFU/ml -5 x 207 10 5 CFU/ml]) into each well of a UKMYC5/6 microdilution plate. All data can be found in and were analysed and visualised 281 using either R or python3 libraries and packages. See for codebase. 282 283 The CRyPTIC dataset comprises isolates from 27 countries collected by 14 286 partner sites worldwide. The compendium presented here contains 15,211 isolates for 287 which both genomic and phenotypic data was collected by 23 of the partner countries 288 across the continents of Asia, Africa, South America and Europe (Fig. 1) . Where the 289 origin of an isolate was not known, the isolate was assumed to originate in the same 290 country as the contributing laboratory (269 isolates in Germany, 17 isolates in India, 6 291 isolates in Peru, 885 isolates in Italy, 510 isolates in South Africa, 357 isolates in 292 Sweden, 208 isolates in Taiwan, 1 isolate in Brazil and 4 isolates in the UK). The 293 largest number of isolates were contributed by India (n = 4,004), Peru (n = 2,691), 294 South Africa (n = 2,155), Vietnam (n = 1,288) and China (n = 1,121). 295 Lineage assignment revealed that all isolates (except 41, 0.3%) belong to the 297 four main M. tuberculosis lineages (L1-L4). Like previous studies, we see a strong 298 association between geolocation and lineage (Pearson's chi-squared test, p < 2.2e-299 16, Fig. S3 ) [22, 23] . The pie-charts in Fig. 1 and supplemental tables show the 300 breakdown of isolate number -v-lineages (Table S2 ) and sub-lineages (Table S3) for 301 each location. The phylogenetic tree in Fig. 2 Regular quality assurance checks detected problems with plate inoculation and 360 reading in two laboratories during collection of the 15,211 isolates. In addition, some 361 plates were not able to be read, usually because of inadequate growth in the positive 362 control wells or extensive contamination. These anomalies were investigated, and 363 samples were removed from the dataset where appropriate, resulting in 2,922 isolates 364 being held out of the final dataset. Further, due to skip wells and other phenomena 365 that prevent an MIC being measured, 11.9% of isolates did not return a phenotype for 366 all 13 drugs on the plate. The compendium therefore has a total of 157,401 MIC 367 measurements for 12,289 isolates (Table 1) MIC readings ("PHENOTYPED") stratified into "high" quality (at least two MIC measurement 372 methods agree), "medium" quality (either Vizion and AMyGDA disagree, or there is no plate 373 picture) or "low" quality (all three MIC measurements methods disagree) phenotype classifications Binary drug resistant/susceptible phenotypes were calculated from the MICs by 380 applying an epidemiological cut-off value (ECOFF); samples with MICs at or below the 381 ECOFF are, by definition, wild-type and hence assumed to be susceptible to the drug 382 in question [11] (Fig. S1 ). Samples with MICs above the ECOFF are therefore 383 assumed to be resistant. 384 385 Unsurprisingly given its size, resistance to each of the 13 drugs is represented 386 within the CRyPTIC compendium (Fig. 3a ). The drugs with the highest percentage of 387 resistance are the first line drugs isoniazid and rifampicin (49.0% and 38.7% 388 respectively). Of the second line drugs, levofloxacin had the highest proportion of 389 resistant isolates in the dataset (17.6%) and amikacin the lowest (7.3%). Reassuringly, 390 a low proportion of isolates were resistant to the NRDs, bedaquiline (0.9%), 391 clofazimine (4.4%), delamanid (1.6%) and linezolid (1.3%). 392 Of the 12,289 isolates which returned a binary phenotype for at least one drug, 394 6,814 (55.4%) were 'resistant' to at least one drug (Fig. 3b) . For the purpose of 395 describing the broader resistance categories present in the dataset, we assumed that 396 all MICs that could not be read had susceptible phenotypes in this section. (Table S1 ) and therefore could be 404 reasonably described as totally drug resistant (TDR). One such isolate belonged to L4 405 and was contributed by South Africa, and the other belonged to L2 with an unknown 406 country of origin. 407 Of the 2,129 resistant isolates that did not fit into either of the MDR or RR 409 categories, more were categorised as isoniazid resistant (73.1%, n = 1,556) than were 410 categorised as susceptible to both isoniazid and rifampicin and resistant to at least 411 one other drug (26.9%, n = 573). 412 413 Whilst many partner laboratories oversampled for resistance, others used a 414 less biased collection protocol (Fig. 3c) . Therefore, the proportions of resistant 415 phenotypes are not necessarily representative of the overall distribution of resistance 416 in the country and are not comparable between countries. Isolates with unknown 417 country of origin were excluded from this analysis. In all countries that contributed 418 more than 100 resistant isolates, each of the five phenotypic resistance categories in 419 Isolates with all possible two-drug resistant combinations were present in this 458 dataset (Fig. 4a) . Resistance to any of the remaining 11 drugs was associated with 459 resistance to both isoniazid and rifampicin. Isoniazid resistance was the most strongly 460 associated bar the following two exceptions, both of which arise because drugs from 461 the same class are being compared: a greater proportion of isolates resistant to 462 rifabutin were also resistant to rifampicin (96.8% had rifabutin -rifampicin co-463 resistance versus 93.3% rifabutin -isoniazid co-resistance) and a greater proportion 464 of isolates resistant to moxifloxacin were associated with levofloxacin resistance 465 (97.6% had moxifloxacin -levofloxacin co-resistance versus 95.0% moxifloxacin -466 isoniazid co-resistance). Of the second line drugs, levofloxacin and moxifloxacin were 467 more commonly seen as a second resistant phenotype than the injectable drugs 468 kanamycin and amikacin. 469 470 Despite this, resistance to both drugs in the aminoglycoside class was common 471 the dataset; 90.4% of amikacin resistant isolates were also resistant to kanamycin 472 although significantly fewer kanamycin resistant isolates were resistant to amikacin 473 (72.0%, p <0.00001). In a similar fashion, a smaller proportion of rifampicin resistant 474 isolates were resistant to rifabutin than rifabutin resistant isolates that were resistant 475 to rifampicin (91.3%, 96.8%, p <0.00001) while a smaller proportion of levofloxacin 476 resistant isolates were resistant to moxifloxacin than moxifloxacin resistant isolates 477 that were resistant to levofloxacin (78.5%, 97.6%, p <0.00001). One possible 478 explanation is that the cut off, in this case the ECOFFs, have been inconsistently 479 defined for the different compounds, however here all ECOFFs have been determined 480 in the same way using a much larger dataset than is typically the case [11] . Differences 481 in drugs of the same class are also well documented by in vitro studies [32] [33] [34] . 482 483 Bedaquiline, clofazimine, delamanid and linezolid were les commonly seen as 484 a second resistance phenotype and the smallest proportional resistance combinations 485 involved the NRDs (e.g. 1.5% of isoniazid resistant isolates were bedaquiline resistant 486 and 1.7% were delamanid resistant). Within the NRDs however, co-occurrence of 487 resistance was proportionally higher; bedaquiline resistance was most seen with 488 clofazimine resistance (52.4%), clofazimine resistance was most seen with 489 bedaquiline resistance (10.6%), delamanid resistance was most seen with clofazimine 490 resistance (26.3%) and linezolid resistance was most seen with clofazimine resistance 491 (34.2%). 492 493 494 We next examined in more detail the correlation structure of phenotypes by 497 conditioning on different phenotypic "backgrounds". (Fig. 4b-f ). We found that a 498 greater proportion of isolates that were susceptible to isoniazid and rifampicin were 499 resistant to second-line drugs than the first line drug ethambutol (15.9%) (Fig. 4b) . The 500 proportion of isolates resistant to clofazimine or levofloxacin was particularly high 501 (32.9% and 24.1%, respectively), and more isolates were resistant to these two drugs 502 than ethambutol in an isoniazid resistant and rifampicin susceptible background but 503 not in MDR/RR isolates ( Fig. 4c-f) . 504 505 MDR/RR isolates (asides from rifabutin) were most commonly resistant to the 506 first line drug ethambutol (46.3%), closely followed by levofloxacin (41.4%), then 507 moxifloxacin (34.1%). A greater proportion of isolates resistant to levofloxacin than 508 moxifloxacin was also seen in all other backgrounds and as expected, the proportion 509 of fluoroquinolone resistance was higher in MDR/RR isolates than non-MDR isolates 510 For isolates with an XDR phenotype, a higher proportion were also resistant to 515 linezolid than bedaquiline (66.7% compared to 44.6%) and 11.3% of XDR isolates 516 were resistant to both bedaquiline and linezolid (Fig. 4f) . XDR isolates were also 517 resistant to the other NRDs, clofazimine (41.3%) and delamanid (18.8%). For isolates 518 with an MDR/RR or pre-XDR background, the most common NRD resistance seen 519 was clofazimine (6.0% and 9.3%, respectively) followed by linezolid (2.3% and 4.7% 520 respectively), bedaquiline (1.8% and 3.2%, respectively), and delamanid (1.7% and 521 2.3%, respectively). In non-MDR/RR backgrounds: clofazimine resistance remained 522 the most common followed by delamanid resistance, linezolid resistance and 523 bedaquiline resistance (Fig. 4b-c) . In order to establish a baseline measure of how well resistance and 541 susceptibility could be predicted based on the state of the art prior to the CRyPTIC 542 project, we constructed a hybrid catalogue of genetic variants associated with 543 resistance to the first-line compounds isoniazid, rifampicin & ethambutol and the 544 second-line compounds levofloxacin, moxifloxacin, amikacin, kanamycin, ethionamide 545 based on existing catalogues [7, 36] . (We note in passing that we specifically did not 546 use the recent WHO catalogue in order to avoid circularity and over-training, as that 547 catalogue was developed (via prior literature, expert rules and a heuristic algorithm) 548 based partially on these isolates [37] . We applied our hybrid catalogue to predict the 549 susceptibility and resistance of all isolates in this compendium to these compounds, 550 and the resulting genetic-based predictions were then compared with the binary 551 phenotypes derived from the MICs (Table 2 ). Since these data were not collected 552 prospectively or randomly, but indeed are enriched for resistance, the calculated error 553 rates are not representative of how well such a method would perform in routine clinical 554 use. We also did not apply the approach used in [7] of refusing to make a prediction if 555 a novel mutation was detected in a known resistance gene, as we simply wanted to 556 measure how well a pre-CRyPTIC catalogue could predict resistance in the 557 compendium. The results were broadly in line with prior measurements on a smaller 558 (independent) set [18] . The hybrid catalogue does not make predictions for rifabutin, 559 linezolid, bedaquiline, delamanid or clofazimine; indeed, this is one of the main aims 560 of the consortium and new catalogues published by CRyPTIC and the WHO will begin 561 to address this shortcoming (Fig. 5) As previously stated, relatively few isolates are resistant to the NRDs, 611 bedaquiline (n = 109), clofazimine (n = 525), delamanid (n = 186) and linezolid (n = 612 156). South Africa contributed the greatest number of isolates resistant to bedaquiline, 613 clofazimine and linezolid (Fig. 5a) , while China and India contributed the most isolates 614 resistant to delamanid. Since the collection protocol differed between laboratories it is 615 not possible to infer any differences in the relative prevalence of resistance to the 616 NRDs in these countries. The results of a survey of all non-synonymous mutations in 617 genes known or suspected to be involved in resistance to these four drugs (e.g. be associated with resistance and coloured blocks denote the presence of a non-synonymous 634 mutation in the relevant gene for a given isolate. Mutations in these genes that are either 635 associated with sensitivity or present in >5% of the collection of isolates as a whole were ignored. Case study on rifampicin mono-resistance 639 Around 1% of TB cases are rifampicin mono-resistant (RMR) and the frequency 640 is increasing [1, 44] . The WHO does not recommend isoniazid for RMR treatment, 641 despite it being effective; this is likely due to the reliance on the Xpert® MTB/RIF assay 642 which cannot distinguish between RMR and MDR. Use of isoniazid could improve 643 treatment outcomes for RMR patients which are currently similar to that of MDR TB, 644 including a higher risk of death compared to drug susceptible infections [45, 46] . Due 645 to its low natural prevalence, RMR has been poorly studied to date but increasingly 646 large clinical TB datasets, such as the one presented here, make its study now 647 feasible. 648 For this case study, we defined RMR as any isolate that was rifampicin resistant 650 and isoniazid susceptible, discounting isolates with no definite phenotype for either 651 isoniazid and/or rifampicin. Of the 4,655 rifampicin resistant isolates in the CRyPTIC 652 database that also had a phenotype for isoniazid, 302 (6.5%) were RMR. These 653 isolates were contributed by South Africa, Peru, India, Italy, China, Pakistan, Brazil, 654 Vietnam, Germany, Nigeria, Turkmenistan and Tajikistan (Fig. 6a) . South Africa and 655 Nigeria contributed a significantly higher number of RMR isolates, as a proportion of 656 rifampicin resistant isolates, than that of the total dataset at 17.5% (p <0.00001) and 657 27.3% (p = 0.00534) respectively. We note that countries may have oversampled for 658 solo resistance, but the higher contribution of RMR isolates from South Africa is 659 consistent with previous studies [44] . 660 RMR has evolved in all four major lineages, with L4 having a significantly higher 662 RMR prevalence than that of the total rifampicin resistant dataset at 9.4% (p = 663 0.00031) and L2 significantly lower RMR prevalence at 4.2% (p = 0.00026). Although 664 these numbers will have been somewhat influenced by differences in sample 665 collection bias at different geographical sites, we would expect L2 to have a lower 666 proportion of RMR isolates due to its association with MDR phenotypes [31, 47] (Fig. 667 6b) . 668 669 A widely used, WHO-endorsed diagnostic tool, the Xpert® MTB/RIF assay, 671 uses a proxy whereby any SNP detected in the "rifampicin-resistance determining 672 region" (RRDR) of rpoB results in a prediction of MDR. However, the suitability of the 673 proxy is dependent upon prevalence of RMR in the population [44] . We tested the 674 reliability of this on the 4,655 rifampicin resistant isolates in our dataset that had a 675 phenotype for isoniazid (Fig. 6c) . 676 Of these isolates, 4,353 (93.5%) were MDR and 302 (6.5%) were RMR. 187 of 678 the MDR isolates had no RRDR mutation and therefore 4.0% of isolates in this study 679 would be predicted as false negative MDR by the Xpert® MTB/RIF assay. 276 of the 680 RMR isolates had a mutation in the RRDR of rpoB and so the Xpert® MTB/RIF assay 681 proxy would incorrectly predict 5.9% of the rifampicin resistant isolates as false 682 positive MDR cases. However, overall, the Xpert® MTB/RIF assay proxy correctly 683 predicts 89.5% of the rifampicin resistant isolates as MDR and 0.6% of the isolates as 684 non-MDR in this dataset, which suggests it is a reasonably successful diagnostic tool 685 with >90% accuracy for MDR classification of rifampicin resistant isolates. As most 686 sites oversampled for resistance, our dataset likely contains a higher prevalence of 687 RMR than the global average and hence the Xpert® MTB/RIF assay is likely to perform 688 better on more representative data. However, the analysis shows how the increasing 689 global levels of RMR TB cases could increase the number of false positive MDR 690 diagnoses by the Xpert® MTB/RIF assay, denying isoniazid treatment to a greater 691 number of patients who would then be moved on to less effective drugs. 692 693 We have analysed our matched phenotypic and genotypic data to examine 696 whether there were any differences in the genetic determinants of rifampicin 697 resistance between RMR and MDR isolates. The proportion of RMR isolates with no 698 rpoB mutation (5.3%, Fig. 6c ) was significantly higher than that of MDR isolates (1.8%, 699 p <0.00001). This suggests that non-target-mediated resistance mechanisms, such as 700 upregulation of rifampicin specific efflux pumps, could be more influential in providing 701 protection against rifampicin in RMR isolates than in MDR isolates. 702 The majority of RMR and MDR isolates contained one or more SNPs in rpoB, 704 with the majority having at least one mutation in the RRDR. To date, several non-705 synonymous RRDR mutations have been found in RMR M. tuberculosis isolates, 706 including the resistance conferring mutations S450L, H445D and D435Y, which are 707 also seen in MDR isolates [48, 49] . For both RMR and MDR isolates in this dataset, 708 the most common rpoB RRDR mutation seen was S450L (63.6% and 41.1% of 709 isolates respectively, Fig. 6d ). Five mutations were present in RMR isolates that were 710 727 728 Figure 6 : Rifampicin mono resistance a) Percentage of rifampicin resistant isolates that 729 are rifampicin mono-resistant (RMR) by country of isolate origin. * indicates RMR proportions that 730 were significantly different from that of the total dataset using a two tailed z-test with 95% 731 confidence. b) RMR prevalence by lineage. * indicates RMR proportions that were significantly 732 different from that of the total dataset using a two tailed z-test with 95% confidence. Hr-TB estimated cases would have received inadequate and unnecessarily longer 773 treatment regimens [1, 50] . Encouragingly, CRyPTIC isolates with a Hr-TB 774 background exhibited relatively low levels of resistance to other antitubercular drugs, 775 including those in the augmented regimen (Fig. 4c) . However, without the tools 776 appropriate to assess and survey this, we will continue to misdiagnose and infectively 777 treat these clinical cases. 778 In 2018, CRyPTIC and the 100,000 Genomes project demonstrated using WGS 780 that genotypic prediction correlates well with culture-based phenotype for first-line 781 drugs, which is reflected in our summary of the genetic catalogue applied to this 782 dataset (Table 3) [7]. While predictions can be made to a high level of sensitivity and 783 specificity, there is still more to learn, as exemplified by the isolates in the compendium 784 that despite being resistant to rifampicin and isoniazid could not be described 785 genetically (Table 2 ). This shortfall, along with the limitations of molecular based 786 diagnostic assays, highlights the need for continual genetic surveillance as drug 787 resistant cases rise and shines a favourable light on a WGS-led approach. 788 789 One of the particular strengths of this compendium lies with the data collated 790 for second-line drugs. A greater proportion of drug resistant isolates had additional 791 resistance to fluoroquinolones than second line injectable drugs (Fig. 4a ). This could 792 be because of more widespread use of fluoroquinolones to treat bacterial infections in 793 general as well as their ease of administration, hence being recommended over 794 injectables for longer MDR treatment regimens [1] . Concerningly, we found that 795 resistance to levofloxacin and moxifloxacin, and kanamycin and amikacin, were more 796 common than resistance to the mycobacterial specific drug ethambutol in an isoniazid 797 and rifampicin susceptible background (Fig. 4b) which suggests a level of pre-existing 798 second line resistance. This concurs with a systematic review that found patients 799 previously prescribed fluoroquinolones were 3 times more likely to have 800 fluoroquinolone resistant TB [51] . Careful stewardship of fluoroquinolones both in TB 801 and other infectious diseases will be paramount for the success of treatment regimens. 802 Despite the variability in sample collection, the observed high proportions of 803 fluoroquinolone resistant MDR/RR isolates from some countries also suggests that 804 MDR treatment regimens could be improved by optimisation on a geographic basis. 805 Further improvement could also be made by the selection of appropriate drugs 807 from each class; for example, the WHO recommends switching from kanamycin to 808 amikacin when treating MDR TB patients [52] . The compendium supports such 809 recommendations as we saw more resistance to kanamycin than amikacin in not just 810 MDR/RR but also all other phenotypic backgrounds. For fluoroquinolones, more 811 isolates were resistant to levofloxacin than moxifloxacin in all phenotypic backgrounds 812 suggesting moxifloxacin may by the most appropriate fluoroquinolone to recommend. 813 However, the amenability of drugs to catalogue-based genetic diagnostics is also an 814 important consideration, and our data suggest levofloxacin resistance could be 815 predicted more reliably than moxifloxacin with fewer false positives predicted (Table 816 2). 817 Testing for fluoroquinolone resistance using molecular diagnostic tests remains 819 limited. However, the limited global data from the past 15 years suggests that the 820 proportion of MDR/RR TB cases resistant to fluoroquinolones sits at around 20%, with 821 these cases primarily found in regions of high MDR-TB burden [1] . While tools such 822 as the Cepheid Xpert® MTB/XDR cartridge, which has just been approved for use by 823 the WHO, will permit both isoniazid and fluoroquinolone testing to be increased, the 824 same pitfalls are to be encountered regarding targeted diagnostic assays [53] . In 825 contrast, the genetic survey in this study demonstrates the potential of WGS for 826 genetic prediction of resistance to second-line drugs and studies within the consortium 827 to investigate this are underway. 828 The CRyPTIC compendium has facilitated the first global survey of resistance 830 to NRDs. Reassuringly, prevalence of resistance to the NRDs was lower than for first-831 and second-line agents in the dataset as a whole (Fig. 3a) , and resistance to the new 832 drugs bedaquiline and delamanid was less common than the repurposed drugs 833 clofazimine and linezolid in an MDR/RR background (Fig. 4c) . However, the presence 834 of higher levels of delamanid and clofazimine resistance than ethambutol resistance 835 in the isoniazid and rifampicin susceptible background does suggest some pre-existing 836 propensity towards NRD resistance (Fig. 4b) . 837 838 Co-resistance between all two-drug combinations of NRDs was seen in isolates 839 in the compendium, the most common being isolates resistant to both bedaquiline and 840 clofazimine. This link is well documented and has been attributed to shared resistance 841 mechanisms such as non-synonymous mutations in rv0678 which were found in both 842 clofazimine and bedaquiline resistant isolates in the compendium [42] (Fig. 5b,c) . 843 Increased clofazimine use could further increase the prevalence of M. tuberculosis 844 isolates with clofazimine and bedaquiline co-resistance, limiting MDR treatment 845 options such as using bedaquiline as the backbone of a shorter MDR regimen [54] . 846 Therefore, proposed usage of clofazimine for other infectious diseases, including 847 COVID-19, should be carefully considered [55] . 848 The WHO recommends not using bedaquiline and delamanid in combination to 850 prevent the development of co-resistance, which could occur relatively quickly [56] ; 851 the rates of spontaneous evolution of delamanid resistance in vitro has been shown 852 to be comparable to that of isoniazid, and likewise bedaquiline resistance arises at a 853 comparable rate to rifampicin resistance [57] . In this compendium, 12.9% bedaquiline 854 resistant isolates were resistant to delamanid and 7.1% delamanid resistant isolates 855 were resistant to bedaquiline. Several scenarios could account for this, including the 856 presence of shared resistance mechanisms. For example, as bedaquiline targets 857 energy metabolism within the cell, changes to cope with energy/nutrient imbalances 858 upon the acquisition of resistance-associated ATPase pump mutations may result in 859 cross resistance to delamanid in a yet unknown or unexplored mechanism [58] . 860 Additionally, off-target mutations in the mmpl3 gene (involved in export), are 861 associated with resistance to delamanid, bedaquiline, clofazimine and linezolid, and 862 were seen in isolates resistant to each of these drugs, although we note this could be 863 artefactual (Fig. 5b- tuberculosis clinical isolates and hope it will lead to a wave of new and inciteful studies 879 that will positively serve the TB community for years to come. CRyPTIC drugs tested. The country of origin is specified using the 3-letter country codes 1013 INH RIF EMB LEV MXF AMI KAN BDQ CFZ DLM LZD ETH RFB S450M and S450Q, however these 711 were seen at low prevalence (less than 2%) of RMR isolates. We found more RMR 712 isolates had His445 mutated than MDR isolates (27.8% of RMR and 9 <0.00001), and mutations to amino acids Ser450 and Asp435 were more prevalent in 714 MDR isolates than RMR isolates (43.7% of RMR and 65.8% of MDR (p <0.00001), 715 and 9.3% of RMR and 15.5% of MDR Relatively little difference in the mutational profiles for rpoB mutations falling 718 outside the RRDR was found; many RMR mutations were also seen in MDR isolates 719 and those found exclusively in RMR were at <2% prevalence (Fig. S2). The most 720 common non-RRDR mutation in both MDR and RMR isolates was a cytosine to 721 thymine mutation 61 bases upstream of the rpoB start codon (10.1% and 8.6% of 722 isolates respectively). The resistance conferring mutations, V695L and V170F, were 723 seen at low proportions with no significant difference between MDR and RMR isolates 724 (V695L was seen in 0.5% MDR isolates and 1.3% of RMR isolates, and V170F was 725 seen References 1056 1. WHO, World Health Organisation Feasibility, diagnostic accuracy, and effectiveness of 1058 decentralised use of the Xpert MTB/RIF test for diagnosis of tuberculosis and 1059 multidrug resistance: a multicentre implementation study Evaluation of the Cepheid Xpert MTB/RIF assay. Expert Rev 1062 Mol Diagn Detection of Drug-Resistant Tuberculosis by Xpert 1064 tuberculosis outbreak clone in Eswatini missed by Xpert 1067 has elevated bedaquiline resistance dated to the pre-treatment era Outbreak of multidrug-resistant tuberculosis in South Africa 1070 undetected by WHO-endorsed commercial tests: an observational study Prediction of Susceptibility to First-Line Tuberculosis Drugs 1073 by DNA Sequencing Rapid, comprehensive, and affordable mycobacterial diagnosis 1075 with whole-genome sequencing: a prospective study Multidrug-resistant tuberculosis drug susceptibility and 1078 molecular diagnostic testing Validating a 14-Drug Microtiter Plate Containing 1080 Bedaquiline and Delamanid for Large-Scale Research Susceptibility Testing of 1081 Mycobacterium tuberculosis Epidemiological cutoff values for a 96-well broth 1083 microdilution plate for high-throughput research antibiotic susceptibility testing of M. 1084 tuberculosis. medRxiv Automated detection of bacterial growth on 96-well plates for 1086 high-throughput drug susceptibility testing of Mycobacterium tuberculosis. 1087 Microbiology (Reading) BashTheBug: a crowd of volunteers reproducibly and accurately 1089 measure the minimum inhibitory concentrations of 13 antitubercular drugs from 1090 photographs of 96-well broth microdilution plates. bioRxiv Trimmomatic: a flexible trimmer for Illumina 1093 sequence data Fast and accurate short read alignment with Burrows-Wheeler 1095 transform The Sequence Alignment/Map format and SAMtools De novo assembly and genotyping of variants using colored de Bruijn 1099 graphs Antibiotic resistance prediction for Mycobacterium tuberculosis from 1101 genome sequence data with Mykrobe Assessment of Mycobacterium tuberculosis transmission in 1103 with whole pathogen genome sequences: an observational 1104 study QuickTree: building huge Neighbour-Joining 1106 trees of protein sequences Using ggtree to Visualize Data on Tree-Like Structures Ecology and evolution of Mycobacterium tuberculosis Consequences of genomic diversity in Mycobacterium 1112 tuberculosis Frequent transmission of the Mycobacterium tuberculosis Beijing 1114 lineage and positive selection for the EsxW Beijing variant in Vietnam Large genomics datasets shed light on the evolution of 1117 the Mycobacterium tuberculosis complex Population structure, biogeography and transmissibility of 1119 Mycobacterium tuberculosis. bioRxiv Reference set of Mycobacterium tuberculosis clinical strains: A tool 1121 for research and product development High degree of fluoroquinolone resistance among pulmonary 1123 tuberculosis patients Treatment outcomes of patients with MDR-TB in Nepal on a 1125 current programmatic standardised regimen: retrospective single-centre study Genome-wide analysis of Mycobacterium tuberculosis 1128 polymorphisms reveals lineage-specific associations with drug resistance. BMC 1129 Genomics Mycobacterium tuberculosis mutation rate estimates from different 1131 lineages predict substantial differences in the emergence of drug-resistant 1132 tuberculosis Rifabutin and rifampin resistance levels and associated rpoB 1134 mutations in clinical isolates of Mycobacterium tuberculosis complex. Diagn 1135 Microbiol Infect Dis Are moxifloxacin and levofloxacin equally effective to treat XDR 1137 tuberculosis? Vitro Susceptibility of Mycobacterium tuberculosis to 1139 Fluoroquinolone resistance in non-multidrug-1142 resistant tuberculosis-a surveillance study in New South Wales, Australia, and a 1143 review of global resistance rates A standardised method for interpreting the association between 1145 mutations and phenotypic drug resistance in Mycobacterium tuberculosis. Eur Respir 1146 J The CRyPTIC Consortium, The Seq&Treat Consortium, Derrick Crook The 2021 WHO catalogue of Mycobacterium 1153 tuberculosis mutations associated with drug resistance: a new global standard for 1154 molecular diagnostics Whole-genome sequencing of rifampicin-resistant Mycobacterium 1156 tuberculosis strains identifies compensatory mutations in RNA polymerase genes Mutations in pepQ Confer Low-Level Resistance to Bedaquiline 1159 and Clofazimine in Mycobacterium tuberculosis In vitro Study of Bedaquiline Resistance in Mycobacterium 1162 tuberculosis Multi-Drug Resistant Clinical Isolates. Front Microbiol Cross-resistance between clofazimine and 1165 bedaquiline through upregulation of MmpL5 in Mycobacterium tuberculosis Population-level emergence of bedaquiline and clofazimine 1168 resistance-associated variants among patients with drug-resistant tuberculosis in 1169 southern Africa: a phenotypic and phylogenetic analysis PA-824 kills nonreplicating Mycobacterium tuberculosis by 1172 intracellular NO release Evolving rifampicin and isoniazid mono-resistance in a high 1174 multidrug-resistant and extensively drug-resistant tuberculosis region: a retrospective 1175 data analysis Treatment outcomes of rifampin-sparing treatment in patients with 1177 pulmonary tuberculosis with rifampin-mono-resistance or rifampin adverse events: A 1178 retrospective cohort analysis Prevalence, Risk Factors, and Treatment Outcomes of Isoniazid-1180 and Rifampicin-Mono-Resistant Pulmonary Tuberculosis in Genome-wide analysis of Mycobacterium tuberculosis 1183 polymorphisms reveals lineage-specific associations with drug resistance. BMC 1184 Genomics Study of the rifampin monoresistance mechanism in Mycobacterium 1186 tuberculosis Prevalence and patterns of rifampicin and isoniazid resistance 1188 conferring mutations in Mycobacterium tuberculosis isolates from Uganda. PLoS 1189 One Prevalence and genetic profiles of isoniazid resistance in 1191 tuberculosis patients: A multicountry analysis of cross-sectional data Protecting the tuberculosis drug pipeline: stating the case for 1194 the rational use of fluoroquinolones Hearing loss with kanamycin treatment for multidrug-resistant 1196 tuberculosis in Bangladesh Update on the use of nucleic acid amplification tests to detect TB 1198 and drug-resistant TB: rapid communication Bedaquiline-based treatment regimen for multidrug-resistant 1200 tuberculosis Clofazimine broadly inhibits coronaviruses including SARS-CoV-2 World Health Organization treatment guidelines for drug-resistant 1204 tuberculosis, 2016 update Rates and mechanisms of resistance development in Mycobacterium 1206 tuberculosis to a novel diarylquinoline ATP synthase inhibitor. Antimicrob Agents 1207 Chemother The CRyPTIC Consortium, Quantitative measurement of antibiotic 1209 resistance in Mycobacterium tuberculosis reveals genetic determinants of resistance 1210 and susceptibility in a target gene approach. 2021, in preparation The CRyPTIC Consortium, Genome-wide association studies 1212 of global Mycobacterium tuberculosis resistance to thirteen antimicrobials in 10,228 1213 genomes. 2021, in preparation The CRyPTIC Consortium, Predicting Susceptibility to First-and 1215 Second-line Tuberculosis Drugs by DNA sequencing and Machine Learning. 2021, in 1216 preparation Deciphering Bedaquiline and Clofazimine Resistance in 1218 Tuberculosis: An Evolutionary Medicine Approach. bioRxiv Program (T32 GM007365). A.L. is supported by the National Institute for Health 926 College London. S.G.L. is supported by the Fonds de Recherche en Santé du Québec.