Genome-Wide Essentiality Analysis of Mycobacterium abscessus by Saturated Transposon Mutagenesis and Deep Sequencing

ABSTRACT Mycobacterium abscessus is an emerging opportunistic human pathogen that naturally resists most major classes of antibiotics, making infections difficult to treat. Thus far, little is known about M. abscessus physiology, pathogenesis, and drug resistance. Genome-wide analyses have comprehensively catalogued genes with essential functions in Mycobacterium tuberculosis and Mycobacterium avium subsp. hominissuis (here, M. avium) but not in M. abscessus. By optimizing transduction conditions, we achieved full saturation of TA insertion sites with Himar1 transposon mutagenesis in the M. abscessus ATCC 19977T genome, as confirmed by deep sequencing prior to essentiality analyses of annotated genes and other genomic features. The overall densities of inserted TA sites (85.7%), unoccupied TA sites (14.3%), and nonpermissive TA sites (8.1%) were similar to results in M. tuberculosis and M. avium. Of the 4,920 annotated genes, 326 were identified as essential, 269 (83%) of which have mutual homology with essential M. tuberculosis genes, while 39 (12%) are homologous to genes that are not essential in M. tuberculosis and M. avium, and 11 (3.4%) only have homologs in M. avium. Interestingly, 7 (2.1%) essential M. abscessus genes have no homologs in either M. tuberculosis or M. avium, two of which were found in phage-like elements. Most essential genes are involved in DNA replication, RNA transcription and translation, and posttranslational events to synthesize important macromolecules. Some essential genes may be involved in M. abscessus pathogenesis and antibiotics response, including certain essential tRNAs and new short open reading frames. Our findings will help to pave the way for better understanding of M. abscessus and benefit development of novel bactericidal drugs against M. abscessus.

M ycobacterium abscessus complex comprises the largest group of rapidly growing nontuberculosis mycobacteria (1). It causes chronic lung infection in individuals with cystic fibrosis (CF) and other structural lung diseases, as well as skin and skin structure infections (1)(2)(3). Treatment of M. abscessus lung infection is difficult. Cure rates are approximately 30% despite long courses of treatment with poorly tolerated regimens (4). The poor efficacy of existing treatments is attributable in part to the high degree of intrinsic resistance to most major classes of antibiotics, including most antituberculosis drugs (2,5,6). Safer, more effective drugs are urgently needed. Genomewide essentiality analyses could help to identify new drug targets for development of novel drugs targeting M. abscessus.
M. abscessus is ubiquitous in the environment, including soil and water (4). Little knowledge exists regarding genome-wide elements that are essential for M. abscessus viability in the inanimate environment or its ability to cause disease. These knowledge gaps result in part from the limited availability and/or use of genetic tools (7)(8)(9). Comprehensive predictions of essential genes and other genetic elements, such as short open reading frames (ORFs), noncoding RNA, and tRNA in Mycobacterium tuberculosis and Mycobacterium avium subsp. hominissuis (here, M. avium), made using saturated transposon (Tn) mutant pools and deep sequencing were recently reported (10,11). Transposon sequencing (Tn-Seq) is a powerful tool to determine the essentiality of genes or other genomic features for growth and survival under experimental conditions (10)(11)(12)(13). Essential and conditionally essential genes may represent ideal targets for novel drugs or important virulence factors to target with interventions in order to better treat or prevent M. abscessus infections.
Here, we optimized conditions for Himar1 Tn mutagenesis of M. abscessus subsp. abscessus strain ATCC 19977 T to generate saturated Tn mutant pools and then prepared fully saturated DNA libraries for deep sequencing. Comprehensive genomic analysis was performed using a Hidden Markov Model (HMM) to predict essentiality of annotated genes and other genomic features for in vitro growth. We also compared essential genes of M. abscessus to those of M. tuberculosis and M. avium to characterize commonalities and differences in essentiality between those pathogens. Our findings provide insights for understanding M. abscessus pathogenesis and pave the way for developing safer, more effective drugs to treat M. abscessus infections.

RESULTS
Optimization of Tn mutagenesis. To generate fully saturated Himar1 Tn mutant pools, we optimized previously described protocols step by step to test the impact of variables, including the multiplicity of infection (MOI) and transduction time and speed, as well as the bacterial density (10,11). Transduction of M. abscessus at an MOI of 20:1 and incubation for 4 h in a 37°C shaker at 180 rpm yielded the highest Tn insertion frequency (Fig. 1A to C). A cell density of 8.2 Â 10 10 CFU/ml obtained by concentrating a culture at early stationary phase yielded a higher transformation frequency (9.3 Â 10 26 ) than a density of 1.5 Â 10 9 CFU/ml under these conditions (Fig. 1D). A total of 10 independent Tn mutant pools were created, each containing 3.4 Â 10 5 to 1 Â 10 6 insertion events. When all 10 Tn mutant pools were combined, 7,454,000 independent insertion events were identified. Spontaneous kanamycin-resistant mutants occurred at a frequency of 1 to 3% among the Tn mutant pools, as observed in M. avium (11). To examine the quality of the pools, randomly selected colonies were subjected to PCR and enzyme digestion to detect the Tn and the Tn-genome junction region using primers listed in Table S1 in the supplemental material. All 40 colonies selected contained the 850-bp DNA fragment of the kanamycin resistance gene of the Himar1 Tn. Ten colonies were further analyzed and confirmed to harbor a single DNA fragment containing the junction of the Tn and a genetic element of M. abscessus ATCC 19977 T , indicating a single Tn insertion event (data not shown).
Saturation and essentiality analysis of TA sites. To achieve full coverage of TA sites on the genome, we performed deep sequencing of Tn DNA libraries prepared in triplicate from each of the 10 independent Tn mutant pools using unique sequencing primers listed in Table S1. The resulting 30 Tn DNA libraries yielded an average of 4.5 million unique Tn-genome junctions (termed "template counts"). The average template count for each TA site is shown in Data Set S1A. Results of statistical analyses of the Tn DNA libraries after deep sequencing are summarized in Table S2. The M. abscessus ATCC 19977 T genome consists of a 5,067,172-bp circular chromosome (14) containing 91,240 TA sites. The average density of Tn insertions into TA sites for each individual Tn DNA library was 65%, but the cumulative density for each Tn pool increased to 74 to 78% after combining results from triplicate DNA libraries (see Table S2). Each pool contained 67,518 to 71,167 unique mutants. The overall insertion density achieved after cumulating the inserted TA sites identified in any of the 30 Tn DNA libraries was 85.7% (78,165 of 91,240 TA sites) ( Fig. 2A). The cumulative insertion density reached a plateau after combining $5 of the 10 Tn mutant pools, indicating full saturation of TA sites available for insertion. Most TA sites (63.8% [58,181/91,240]) were detected in all 10 Tn pools with a mean read count of 229 per TA site, while only small proportions of inserted TA sites (i.e., 1.7 to 4.9%) with low mean read counts (i.e., 1.4 to 21) were observed in only 1 to 9 Tn pools ( Fig. 2B and C). Tn insertions were not detected in the remaining 14.3% (13,075/91,240) of TA sites in any of the 10 pools (Fig. 2B), similar to results observed in M. tuberculosis (15.7%) and M. avium (16.5%) (10,11).
The essentiality of each TA site was defined using an HMM (15,16), a statistical model that considers read counts both at a given TA site and distributed over surrounding sites and dynamically adjusts probability distributions over states using geometric distributions to obtain locally consistent interpretations of essentiality across the genome. As such, TA sites with no insertions in nonessential regions are tolerated because neighboring sites have insertions. In contrast, if a consecutive sequence of TA sites with no insertions is long enough, the most probable state of that sequence will be assigned as essential (16). Read counts were modeled as having geometric distributions conditioned on four different states of essentiality for TA sites: essential (ES), nonessential (NE), and causing either a growth defect (GD) or growth advantage (GA) upon Tn insertion. Parameters for expected read count distributions for each state were set relative to the mean value of nonempty read counts (16). In the M. abscessus genome, 6.8 and 79.6% of TA sites were defined as ES and NE, respectively, for in vitro growth; 10.9 and 2.7% were defined as GA and GD, respectively, when disrupted (see Data Set S1A). The inferred essentiality of TA sites was used to determine the essentiality of individual coding sequences and non-ORF genomic features. The same analysis was performed on the plasmid pMAB23 harbored by M. abscessus ATCC 19977 T .
To better understand why some TA sites had few or no insertions or lower read counts in some number of pools ( Fig. 2B and C), the TA site motif (G/C)GNTANC(G/C) identified in prior M. tuberculosis and M. avium studies as being less permissive to Himar1 insertion (10,11) was investigated in the M. abscessus genome. It was greatly enriched in a set of 6,000 putative nonpermissive TA sites lacking Tn insertions compared to a set of putative permissive sites with the highest 25% of read counts. This nonpermissive motif was identified at 8.1% (7,425/91,240) of TA sites (see Data Set S1A), similar to the frequency in M. tuberculosis (9%) (10). Of TA sites with the nonpermissive motif, 60.9% had no Tn insertion detected in any mutant pool. Furthermore, the probability of the nonpermissive motif appearing at a TA site was inversely proportional to the number of pools in which Tn insertions were detected at that TA site (see Table S3), confirming that the nonpermissive motif was associated with lower Tn insertion frequency and lower read counts in the M. abscessus genome, as in M. tuberculosis and M. avium (10,11). Remarkably, 23% of the 13,075 TA sites without a Tn insertion detected could not be explained by either a prediction of essentiality in the HMM or presence of the nonpermissive motif (Fig. 3).
Essentiality analysis of annotated genes. Of 4920 annotated genes, 326 were identified as ES, 144 as GD, 589 as GA, and 3855 as NE (Table 1). Detailed information regarding essentiality of each gene is shown in Data Set S1B. Only six genes lacked TA loci and therefore could not be assessed by our approach. A homology comparison between proteins encoded by annotated genes of M. abscessus ATCC 19977 T  Table S4), and 41.1% have essential orthologs only in M. tuberculosis (see Table S5), while 3.4% (11/326) have essential orthologs only in M. avium (see Table S5). Interestingly, 12% (39/326) of M. abscessus essential genes are homologous to genes that are not essential in M. tuberculosis (37 genes) or M. avium (2 genes) ( Table 2). For example, MAB_3090c encoding dihydrofolate reductase was defined as essential in M. abscessus, but its M. tuberculosis ortholog Rv2763c is not essential (10). Moreover, 2.1% (7/326) of M. abscessus essential genes have no homology with M. tuberculosis or M. avium genes (Table 3).
A large number of essential genes are involved in DNA replication, RNA transcription and translation, protein folding, cell wall organization and regulation of cell shape. For example, MAB_3869c encodes the ortholog of the DNA-directed RNA polymerase beta chain RpoB in M. tuberculosis (see Table S4), the target of first-line antituberculosis (anti-TB) rifamycin drugs (17). However, this class has limited utility for treatment of M. abscessus infections due to intrinsic resistance (18,19). Another large group of essential genes is associated with biosynthesis and transport of nucleotides, amino acids, fatty acids and cell wall components. All 19 genes encoding tRNA synthetases for transfer of 20 common amino acids are essential ( Table 2 and Table S4). M. abscessus genes responsible for energy support, including MAB_1448-MAB_1453 encoding the ATP synthase operon are also essential, including atpE (MAB_1448), the target of bedaquiline (see Tables S4 and S5), which potently inhibits ATP generation in M. tuberculosis (20) and in M. avium and M. abscessus (21)(22)(23)(24).
Essentiality analysis of genes involved in pathogenesis. Functional analysis and homology comparisons identified 49 M. abscessus genes potentially involved in M. abscessus virulence, some of which were referenced from a previous study by Ripoll et al. (14) (Table 4). Of these, only four are essential for in vitro growth. MAB_1933c encodes glutamine synthetase, type I (GlnA1) ( Table 4), which catalyzes ATP-dependent  (8,550 TA sites) of a previously described nonpermissive motif and the predicted essentiality status. TA sites were predicted by the four-state HMM (ES, essential; GD, growth defect when mutated; GA, growth advantage when mutated; NE, nonessential).
Comprehensive Essentiality Analysis of M. abscessus ® assimilation of ammonia into glutamate to form glutamine in M. tuberculosis (25). This process accounts for 15% of total ATP consumption in Escherichia coli (26,27). MAB_1077 encodes the two-component sensor kinase MprB (Table 4). In M. tuberculosis, MprAB, encoded by Rv0981 and Rv0982, interacts with chaperone protein DnaK (Hsp70) to properly maintain protein homeostasis in the extracytoplasmic compartment in response to cell envelope stress (28). The other two essential genes, MAB_0118c and MAB_3473c (Table 4), encode a probable Mn-dependent superoxide dismutase and an SsrA-binding protein, respectively. The former secreted protein protects mycobacteria from reactive oxygen species, especially in macrophages (29)(30)(31). The latter is responsible for recycling stalled ribosomes and tagging incomplete nascent proteins for degradation during trans-translation (32,33). M. abscessus only contains two type VII secretion (T7S) systems, including ESX-3 and ESX-4 loci, whereas M. tuberculosis harbors five T7S systems (14,34). The essentiality and the homology comparisons for the ESX-3 and ESX-4 genes are shown in Table S6. The M. abscessus genes in these loci were defined as nonessential under our test conditions while, in M. tuberculosis, genes encoding ESX-3 were essential and those encoding ESX-4 were nonessential. The M. abscessus genes were homologous to M. tuberculosis genes except that nonessential MAB_3760 encoding EccE4 is not found in most other interrogated mycobacteria, with the exception of M. chelonae and Mycobacterium immunogenum, thereby potentially distinguishing the M. abscessus ESX-4 system from that of most other mycobacteria (35,36).
Essentiality analysis of genes involved in responses to antibiotics. Table 5 lists 58 M. abscessus genes possibly associated with antibiotic responses (including drug targets, drug-modifying enzymes, drug transporters, and TetR family regulators), identified through the result of comparative genomic and functional analysis. Only 16 a ES, essential; GD, growth defect when mutated; GA, growth advantage when mutated; NE, nonessential; NA, not assessable using our approach due to genomic features without any TA site. genes were classified as essential, virtually all of them being drug targets in M. tuberculosis that have been validated in vivo. Among them, MAB_0019 and MAB_0006 encode DNA gyrase subunits A and B, respectively (Table 5), comprising the sole DNA topoisomerase responsible for DNA decatenation, relaxation, and supercoiling. GyrA is the target of fluoroquinolones such as moxifloxacin in TB treatment (37). The in vivo efficacy of moxifloxacin against M. abscessus remains uncertain despite it showing activity in vitro (38,39). The clinical candidate SPR720 targets GyrB and is being developed for both M. tuberculosis and M. abscessus infections (40)(41)(42). MAB_2722c encodes the NADH-dependent enoyl-(acyl-carrier-protein) reductase that catalyzes the last step in the fatty acid elongation cycle for mycolic acid biosynthesis and MAB_0185c encodes arabinosyltransferase B that is necessary for arabinogalactan biosynthesis (Table 5). These enzymes are targets of the first-line anti-TB drugs isoniazid and ethambutol, respectively (43,44), which are not recommended for treating M. abscessus infections due to intrinsic resistance (45). Among all MmpL family members, only MAB_4508 encoding the trehalose monomycolate flippase MmpL3 is essential in M. abscessus (46)   Table 5). The b-lactamase Bla Mab encoded by MAB_2875 is a major cause of the limited efficacy of b-lactams against M. abscessus (47), although a homology search identified 18 nonessential genes encoding possible b-lactamases worthy of investigation, especially those having no homology with either M. tuberculosis or M. avium (see Table S7). Essentiality analysis of genes involved in peptidoglycan metabolism. Mycolylarabinogalactan-peptidoglycan complex is the key structure for the mycobacterial cell wall that contributes to the intrinsic resistance to a number of antibiotics. Consequently, it is a major target for drug development (48). Table 6 lists genes   associated with peptidoglycan (PG) biosynthesis and remodeling, as previously identified and categorized (49)(50)(51). The essentiality comparisons of these 28 M. abscessus genes with 30 genes in M. tuberculosis H37Rv (10) indicated a few differences. For example, Rv0024 and rpfD orthologs are absent in M. abscessus and MAB_3167c encoding a putative penicillin-binding protein is essential in M. abscessus, but its ortholog Rv2864c is nonessential in M. tuberculosis (Table 6). Similarly, ripA encoding an endopeptidase contains an essential domain in M. tuberculosis, but its ortholog MAB_2728c was classified as nonessential (Table 6). Remarkably, 11 M. abscessus genes were classified as GA genes compared to only one in M. tuberculosis ( Table 6). The GA M. abscessus genes were found in each functional group, and their corresponding orthologs in M. tuberculosis are well characterized regarding their roles in PG biosynthesis (52)(53)(54)(55)(56)(57)(58)(59). Among them, ldt Mab1 and ldt Mab2 were confirmed to encode L,D-transpeptidases in M. abscessus like their orthologs ldt Mt1 and ldt Mt2 , which are targets of cephalosporin and carbapenem b-lactams (55). Essential M. abscessus genes without homologs in M. tuberculosis or M. avium. Most of the 43.6% of M. abscessus genes (2,145/4,920) with no significant similarity to either M. tuberculosis or M. avium are hypothetical genes. The majority were defined as nonessential, whereas 262 and 16 genes conferred a growth advantage and a growth defect, respectively, when interrupted. Four genes without TA sites could not be assessed. Many TetR family regulators, whiB regulators, drug-modifying enzymes, and efflux pumps belonged to this category, some of which are known to be involved in intrinsic drug resistance, such as MAB_0591 (rifampin ADP-ribosyl transferase) (19), MAB_4532c (eis2, Gcn5-related N-acetyltransferase) (60, 61), and MAB_4384 (TetR regulator) (62). Only seven essential M. abscessus genes lacked significant homology with M. tuberculosis or M. avium genes, most of which encode hypothetical proteins (Table 3). However, MAB_3419 encodes NH 3 -dependent NAD 1 synthetase (NadE) (63), which catalyzes the last step in de novo NAD 1 biosynthesis, and is a validated drug target in M. tuberculosis (64)(65)(66). Almost all external genes, including an 81-kb full-length prophage, 3 prophage-like elements, and 17 clusters of horizontally transferred genes from nonmycobacterial organisms described by Ripoll et al. (14), are not homologous to M. tuberculosis or M. avium genes and are dispensable for in vitro growth. However, two essential genes, MAB_0222c (putative DNA-binding protein) and MAB_4828c (unknown function) are located on distinct prophage-like elements ( Table 3). Analysis of peptide sequences from 1,718 M. abscessus complex genomes deposited in GenBank using BLAST revealed homologs of MAB_0222c and MAB_4828c in only 23 and 307 genomes, respectively, and exclusively in M. abscessus subsp. abscessus. In addition, MAB_0222c showed the highest identity of 63% with a DNA-binding protein in M. chelonae, whereas MAB_4828c showed 72 and 75% identities to hypothetical proteins in Mycobacteroides salmoniphilum and Mycobacteroides franklinii, respectively. These strains were isolated from water and have caused fish and, rarely, human infections (67). Further comparative genomic analysis showed that 80% (3,940/4,920) of M. abscessus genes have homologs in M. chelonae type strain CCUG 47445 (68), including 96% (313/326) of essential M. abscessus genes (data not shown). Table S8 lists 13 essential genes without significant homologs in M. chelonae CCUG 47445. Interestingly, the MAB_0222c homolog was not identified in this strain.
Essentiality analysis of short ORFs and non-ORF genomic features. Short ORFs (sORFs) and non-ORF genomic features in M. abscessus ATCC 19977 T were first identified by Miranda-CasoLuengo et al. using RNA-seq/Ribo-seq and proteomics technologies (69). In all, 126 ribosomally protected sORFs were detected over the genome, 80% of which are #50 amino acids in length (69). Table 1 shows the essentiality assignments of sORFs, ncRNAs (noncoding RNAs), tRNAs (transfer RNAs), rRNAs (ribosomal RNAs), 59 UTRs (59 untranslated regions), promoter regions, and Rho-independent terminators. Essential genetic elements are shown in Table 7. Only 5 of 126 sORFs were defined as essential, although 17 without TA sites could not be assessed. Of 36 ncRNAs containing 55 to 404 nucleotides, 4 were classified as essential. Of 47 tRNA genes, 10 were classified as essential, while 2 without TA sites could not be assessed.
Remarkably, 4 tRNA genes conferred a growth advantage when disrupted, and only 2 of 7 singleton tRNA genes (MAB_t5001 [Ile, GAT] and MAB_t5014 [Asp, GTC]) were essential. Interestingly, one or two copies of tRNA genes for transfer of Gln and Thr were essential in M. abscessus, M. tuberculosis, and M. avium (10,11), suggesting that these two amino acids play crucial roles in mycobacterial physiology. As expected, the genes encoding 16S, 23S, and 5S rRNAs were essential in M. abscessus as in M. tuberculosis and M. avium (10,11). 16S and 23S rRNAs are targets of amikacin and macrolides, respectively, which are first-line drugs recommended for treating M. abscessus infections (45). Most predicted promoter regions, 59 UTRs and rho-independent terminators were nonessential. Detailed information on the essentiality analyses of sORFs and non-ORF genomic features is available in Data Set S1C.
Essentiality analysis of the plasmid. M. abscessus ATCC 19977 T contains the 23-kb plasmid pMAB23, which is 99% identical to pMM23 from Mycobacterium marinum ATCC BAA-535 and harbors a putative mercury resistance operon (14,70). All of the 22 annotated coding sequences and intergenic regions are dispensable for in vitro growth except MAB_p16c encoding putative replication protein RepA and the intergenic region upstream of repA (between MAB_p17 and MAB_p16c), which were defined as GD when disrupted (see Data Set S1D).

DISCUSSION
Despite its increasing incidence as a cause of difficult-to-treat opportunistic infections, M. abscessus remains poorly studied, in part because it has been difficult to manipulate genetically. The availability of a comprehensive list of essential genes and other genomic features is a crucial step toward greater understanding of M. abscessus physiology and pathogenesis. To our knowledge, this is the first comprehensive analysis of the essentiality of M. abscessus genetic elements required for in vitro growth using fully saturated Tn mutant pools and deep sequencing.
Use of highly saturated Tn mutant libraries and an HMM increases the predictive accuracy of essentiality analyses, especially for classifying genomic features with fewer TA sites (10,11). HMMs have proven to be a reliable statistical method for estimating different degrees of essentiality across the genome in an unbiased (non-gene-centered) way (10,11,16). However, it has been difficult to achieve high saturation using the Himar1 Tn in M. abscessus. To date, published reports studied M. abscessus Tn mutant pools containing only 6,000 to 8,000 unique Tn mutants (9,36), far from the 91,240 TA sites in the genome of M. abscessus ATCC 19977 T . Through systematic optimization of procedures for Himar1 Tn mutagenesis and use of triplicate Tn DNA libraries per Tn mutant pool, we increased the number of unique TA insertions to 67,518 to 71,167 per pool and achieved full saturation of detectable TA site insertions. The overall proportions of TA sites with or without observed Tn insertions (85.7 and 14.3%, respectively) and with a known nonpermissive motif (8.1%) are similar to those reported in M. tuberculosis and M. avium (10, 11). Our findings also indicate that, in addition to the lethality of Tn insertions into essential genes and the restrictive effects of the nonpermissive motif (10), other unknown factors may restrict Himar1 insertion into certain TA sites. The Himar1 Tn is widely used in mutagenesis studies, but its utility and limitations in diverse mycobacterial genomes is understudied. Our optimization approach may provide a useful roadmap for generating more fully saturated Tn mutant pools in other mycobacteria. The M. abscessus complex is a large and heterogeneous group of species (1, 71, 72) capable of causing opportunistic infections in any organ, but commonly in the lungs and skin and soft tissue (73). In-depth genomic analysis of M. abscessus indicates a nonconservative genome, in which the core genome is limited to 64.15% of the pan-genome, differing from the conservative pathogen M. tuberculosis, whose core genome represents 96.1% of the pan-genome (72). Despite M. abscessus diversity in genome size and content, our findings on the essentiality of genomic elements of M. abscessus ATCC 19977 T will shed light on other M. abscessus complex strains, especially many clinically relevant strains in the United States and Europe, since phylogenomic analyses place this type strain within the predominant clone observed in several global and national studies of clinical isolates (74). Most essential M. abscessus genes defined here are highly homologous to those identified in similar studies of M. tuberculosis and M. avium. These results provide a fundamental basis for utilizing available knowledge and approaches from M. tuberculosis and M. avium studies to promote research to address key knowledge gaps regarding M. abscessus. Our findings also highlight intriguing genomic differences that could be exploited for greater understanding of M. abscessus pathogenesis and development of new tools to treat and prevent M. abscessus infections.
Essential M. abscessus genes sharing significant homology with essential M. tuberculosis genes include validated targets for important anti-TB drugs, such as isoniazid (43), rifampin (17), ethambutol (44), moxifloxacin (37), and bedaquiline (20). However, these drugs are not effective against M. abscessus infections or, in the case of bedaquiline, require further study (21,22,38,45). Thus, drugs developed and optimized against essential M. tuberculosis targets may not be useful against even highly homologous essential targets in M. abscessus due to interspecies differences in target protein structure or the presence or absence of enzymes that activate prodrugs like isoniazid or inactivate drugs, such as rifamycins, or other unique resistance mechanisms, such as efflux transporters (19,47,(60)(61)(62)(75)(76)(77)(78). Thus, developing new anti-M. abscessus drugs against drug targets validated in TB should be an effective approach, but programs focused specifically on M. abscessus are needed to deliver optimized drugs that exploit interspecies differences in structure-activity relationships (SAR) and intrinsic resistance mechanisms. For example, our approach predicted MmpL3 (MAB_4508) to be essential in M. abscessus, as in M. tuberculosis. This flippase required for translocating mycolate precursors to the cell envelope was successfully targeted first in M. tuberculosis by a series of indole-2-carboxamide inhibitors but subsequent evolution of this series and others based on unique SAR delivered compounds with superior in vitro and in vivo activity against M. abscessus (46,(79)(80)(81)(82). Glutamine synthase GlnA1 (MAB_1933c) is predicted to be essential in M. abscessus and may represent a more novel drug target and virulence factor. The attenuation of an M. tuberculosis glnA1 deletion mutant during glutamine auxotrophy and in guinea pigs and mice is encouraging in this regard (83,84), especially since glutamine is not readily available in CF sputum, an important niche for M. abscessus (85). Furthermore, genetic or chemical disruption of GlnA1 increases vulnerability to bedaquiline in M. tuberculosis (27), suggesting that a MAB_1933c inhibitor could synergize with diarylquinolines against M. abscessus.
Genes essential in M. abscessus but not in M. tuberculosis could also be more effectively exploited as drug targets in M. abscessus. MAB_3090c encodes the dihydrofolate reductase (DHFR) DfrA, a conserved enzyme in the folate biosynthesis pathway (86). Although DHFR inhibitors are effective anti-proliferative drug targets for treating a variety of malignancies and autoimmune and infectious diseases (86, 87), they have not proven very effective against M. tuberculosis (88). However, DHFR is only conditionally essential in M. tuberculosis (10) and is not particularly vulnerable because its loss can be compensated by upregulation of a second DHFR enzyme, Rv2671, and ThyX (89,90). Whether the nonessential Rv2671 ortholog MAB_2976 and ThyX (Rv2754c) ortholog MAB_3085c that showed a growth advantage after disruption can compensate for the loss of DfrA in M. abscessus may determine the value of this target in M. abscessus.
Genes essential in M. abscessus that have limited or no homology with genes in M. tuberculosis may represent new and more specific drug targets. Of particular interest is MAB_3419, a putative ammonia-dependent NAD synthetase (NadE) that catalyzes the final step in NAD 1 biosynthesis. NAD 1 is an essential cofactor that mycobacteria synthesize either de novo from aspartate or from nicotinamide/nicotinic acid scavenged from the environment. Both pathways utilize NadE. Interestingly, MAB_3419 is predicted to be a single-domain NAD 1 synthetase that uses ammonia as a nitrogen source, with .90% protein sequence homology with NadE in Mycobacterium chelonae and a few other rapidly growing mycobacteria but limited homology with the M. tuberculosis and Mycobacterium smegmatis enzymes, which are glutamine-dependent NAD1 synthetases comprised of a C-terminal NAD 1 synthetase domain fused with an N-terminal glutaminase domain. NadE is a genetically and chemically validated drug target in M. tuberculosis l (91-93), but the described inhibitors bind to sites not present in MAB_3419, indicating a different chemical route is required to target M. abscessus NadE.
The mycobacterial cell wall is an essential structure for growth and virulence. Comprised of three distinct layers (PG, arabinogalactan, and mycolic acids), it is an attractive target for antimycobacterial antibiotics (48). Unlike in TB, PG synthesis inhibitors, i.e., imipenem and cefoxitin are already first-line drugs for M. abscessus infections. PG requires constant expansion, remodeling and recycling during bacterial growth and division (94). We identified interesting differences in the essentiality of genes associated with PG metabolism between M. abscessus and M. tuberculosis. Although the growth-advantaged phenotypes of Tn insertions in 11 PG-associated M. abscessus genes need to be confirmed, we speculate that M. abscessus has evolved interaction networks that differ from those in M. tuberculosis and may confer a greater ability to compensate for disruption of certain PG-synthesizing enzymes to ensure cell wall integrity and greater adaptability to changing environmental conditions. Approximately 5 to 6% of genes in the M. abscessus genome were likely acquired through horizontal gene transfer from other organisms (14). Among them, only two genes from prophage-like elements (MAB_0222c and MAB_4828c) are defined as essential for in vitro growth. Horizontal gene transfer preferentially occurs between specific groups of organisms that share ancestry or habitat, presumably under evolutionary pressure (95,96). The genes' homology to M. chelonae and to M. salmoniphilum and M. franklinii, respectively, suggests horizontal gene transfer in a shared habitat, perhaps in water (67). Since the genes are found in the more pathogenic M. abscessus subsp. abscessus but not in M. abscessus subsp. massiliense or M. abscessus subsp. bolletii, one wonders whether their acquisition improves pathogenicity in addition to adaptation and survival in the environment.
We evaluated the essentiality of 126 previously identified sORFs (69). Interestingly, essential sORFs also showed significant responses to stress conditions in previous RNAseq studies. For instance, MAB_5003, MAB_5034c, and MAB_5035c are significantly upregulated, while MAB_5050c is downregulated in artificial sputum, and increased expression of MAB_5035c was detected in response to kanamycin (69). MAB_5050c is located at the leader region of essential gene MAB_3798c encoding 30S ribosomal protein S8 (RpsH) that directly binds to 16S rRNA to assemble the 30S subunit of the ribosome for protein synthesis (97). RpsH also plays a critical role in selectively inhibiting synthesis of ribosomal proteins whose genes are in the same spc operon as its own in response to changing environmental conditions (98). Although any functional association between MAB_5050c and RpsH requires further investigation, downregulation of MAB_5050c may play some role in regulating protein synthesis under stress. tRNAs play a central role in protein translation and thus are potential targets for new antibiotics that inhibit attachment of amino acids onto corresponding tRNAs. Most M. abscessus tRNA genes are dispensable for in vitro growth, as in M. tuberculosis and M. avium (10,11). Only two of seven singleton tRNAs were defined as essential, similar to findings in Saccharomyces cerevisiae, in which four of six singleton tRNA genes are essential and most tRNA deletions do not cause altered growth phenotypes in rich medium (99), presumably because they are compensated by members of the same or different anti-codon families in many conditions (99).
We defined essential genes under optimal in vitro growth conditions using a reliable and comprehensive approach. However, further confirmation of essentiality may still be necessary. Genes not required for in vitro growth could become conditionally essential when M. abscessus encounters certain stresses during infection, including in the nutritionally aberrant environments of cystic fibrosis airway mucus, biofilms, in phagocytes, and necrotic pyogranulomatous lesions. For example, although genes comprising the ESX-3 and ESX-4 loci are nonessential under conditions in this study, M. abscessus ESX-3 plays an important role in pathogenesis (100), and M. abscessus ESX-4 genes are required for intracellular survival (36). Nevertheless, differences in essentiality and homology between these genes in M. abscessus and M. tuberculosis may indicate differences in function, which deserves further study. Using saturated M. abscessus Tn mutant libraries to identify new virulence factors and potential drug targets under clinically relevant conditions is under investigation.
Construction of Himar1 Tn mutant pools. A protocol for constructing a saturated Himar1 Tn mutant pool was carefully optimized based on the literatures (11,102). Detailed information was provided in the supplemental material.
Preparation of DNA libraries of Tn mutant pools. DNA libraries of Tn mutant pools were prepared as previously described with modification (11,102). Detailed methods are provided in Text S1 in the supplemental material.
Deep sequencing of Tn insertions and analysis of sequencing data. Thirty Tn mutant DNA libraries representing triplicate samples from each of 10 Tn mutant pools were sequenced on an Illumina HiSeq instrument, collecting 18 (10.4 to 31.5) million 150-bp paired-end reads per sample. The reads were filtered by Trimmomatic v0.39 (103) and processed using TPP in TRANSIT, which counts reads mapping to each TA dinucleotide site against the published M. abscessus ATCC 19977 T (accession number CU458896) genome sequence (after eliminating reads sharing the same template barcode) (15). A Hidden Markov Model (HMM) offered by TRANSIT was used to assign the most probable state of essentiality of the sequence of TA sites based on the read count at the site and the distribution over the surrounding sites (15,16). It parses a genome into contiguous regions belonging to one of four essentiality states-essential (ES), nonessential (NE), growth defect when mutated (GD), or growth advantage when mutated (GA), based on local insertion density and mean value of nonempty read counts at TA sites (with ES being near 0, NE being near the mean, and GD and GA being approximately 1/10 and 5 times the mean, respectively). To search for TA site motifs that are less permissive for Himar1 Tn insertion in the M. abscessus genome, all TA sites without insertions but excluding TA sites from essential gene regions or regulatory regions which may be under selection pressure were selected as a putative nonpermissive set (;6,000 TA sites). The TA site regions with the top 25% read counts of Himar1 Tn insertions were then chosen as a putative permissive set. The nucleotides surrounding TA sites were compared between the two sets.
Ortholog analysis was performed among all annotated genes and the subset of predicted essential genes of M. abscessus ATCC 19977 T , M. tuberculosis H37Rv, and M. avium MAC109 (10, 11) using ortho-venn2 diagram analysis (excluding genes encoding proteins containing less than 20 amino acids) (104). Homology searches were performed between the genomes of M. abscessus ATCC 19977 T and M. chelonae type strain CCUG 47445 (68) using the same method. Other genomic features, including new sORFs and non-ORF genomic elements, were identified based on published information (69). Promoter regions were defined based on a set of 2,653 transcriptional start sites (TSSs) defined previously (69). A region Comprehensive Essentiality Analysis of M. abscessus ® around each TSS (bp 2150 to 170) was used to determine the promoter region as described previously (10). Lastly, Rho-independent transcription terminators were predicted using the ARNold online tool (http://rssf.i2bc.paris-saclay.fr/toolbox/arnold/) (105).
Data availability. The raw data are deposited in NCBI SRA database under BioSample accession numbers SAMN16825978 to SAMN16826007.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. DATA SET S1, XLS file, 9.1 MB. TEXT S1, DOCX file, 0.04 MB.

ACKNOWLEDGMENTS
This study was supported by grants from the National Institutes of Health (R01-AI-141805 to B.N.K. and R21-AI-137814 to E.L.N.). The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. D.R. and E.L.N. designed the study. D.R. performed all experiments, and L.C. conducted all bioinformatics analyses. All authors analyzed and interpreted the results. D.R. drafted the manuscript, and all authors reviewed and edited the draft. All authors approved the final manuscript prior to submission.
We declare there are no competing interests.