Mycobacterium abscessus Strain Morphotype Determines Phage Susceptibility, the Repertoire of Therapeutically Useful Phages, and Phage Resistance

Mycobacterium abscessus infections in cystic fibrosis patients are challenging to treat due to widespread antibiotic resistance. The therapeutic use of lytic bacteriophages presents a new potential strategy, but the great variation among clinical M. abscessus isolates demands determination of phage susceptibility prior to therapy.

clade of closely related strains that includes the ATCC 19977 type strain (8). Two distinct colony morphotypes of M. abscessus are observed, forming smooth (S) or rough (R) colonies on solid medium (9). The S strains generally are thought to be less virulent, due to the prevalence of surface glycopeptidolipids (GPLs) recognized by the host immune system (10)(11)(12). Rough strains are strongly depleted for GPLs, escape immune recognition more efficiently, and are more virulent (13). Smooth-to-rough transition occurs by interruption of GPL synthesis or localization-including mutations in mmpL4, which is required for GPL transport across the inner membrane-and mutation or silencing of mps1, mps2, and gap (9,11,14). Smooth-to-rough transitions are often nonreversible, although temperature-dependent variation has been reported for one variant (11,15,16).
The therapeutic use of bacteriophages may provide an alternative treatment strategy for NTM infections. A three-phage cocktail administered intravenously in a 15year-old CF patient with a disseminated M. abscessus infection following a bilateral lung transplant showed substantial improvement and alleviation of infection (17). Whether phage interventions are useful in other patients with similar infections is unclear because of the extensive genetic variation among M. abscessus strains (18,19).

RESULTS
Collection of Mycobacterium abscessus clinical isolates. Following the successful treatment of a disseminated M. abscessus infection with a bacteriophage cocktail (17), we received 82 M. abscessus strains (designated GD01 to GD111; Table 1) from 78 different patients for characterization of phage susceptibilities; 54 (69%) of these are from within the United States and the others are from 10 different other countries (Table 1). We scored the colony morphotypes of each strain as being either rough (R) or smooth (S) ( Table 1; Fig. 1, Fig. 2A); 48 (58.5%) have R morphotypes (Table 1). One-half of the strains are from patients with cystic fibrosis, and among these a similar proportion of strains (64%) are rough (Table 1). Smooth morphotype strains are less virulent than rough strains in some assays (20), but overall the smooth and rough profiles of these strains is similar to those reported previously (21,22).
The M. abscessus genomes were sequenced either by Illumina technology to give whole-genome sequences (WGS) or to completion by the addition of Nanopore sequencing reads (Table 1; Table S1 in the supplemental material). Phylogenetic analysis of these strains together with the type-strains [M. abscessus subsp. abscessus ATCC 19977 (23), M. abscessus subsp. bolletii BD T (24), and M. abscessus subsp. massiliense GO06 (25)], shows that 62 of the strains are subspecies abscessus, 18 are massiliense, and 2 are bolletii ( Table 1, Fig. 1). The substantial genetic diversity is not surprising (8), and 50% of the strains form a closely related clade that includes the ATCC 19977 type strain (Fig. 1); the prevalence of this clade was noted previously (8). There are two smaller massiliense clades, one of which includes the previously characterized strain GO06 (25), and one that includes GD01, the strain from the previously phage-treated patient (17) (Fig. 1). Both of the subsp. boletii strains have S morphotypes, but R and S morphotypes are distributed throughout the rest of the phylogenetic spectrum (Table  1, Fig. 1, Fig. 2A). All of the strains are CRISPR-free.
Genotypes of smooth and rough colony morphotypes. For four strains (GD43, GD68, GD69, and GD100), both S and R variants were recovered from the same sample and were sequenced to completion ( Table 1). The R variants generally have several differences from the S variants, but these include mutations in GPL synthesis implicated in the rough morphotype (14). For example, rough strains GD43B and GD68A have single base differences in mps1 (MAB_4099) relative to their smooth variants GD43A and GD68B, introducing nonsense and single base insertions, respectively ( Fig. 2B; Table  S2). Although the other R strains do not have a smooth counterpart, over 75% of them have mutations in mps1 and mps2 (MAB_4098), most commonly 1-to 2-bp insertions or deletions. One (GD60) also has a frameshift mutation in mmpL4b (MAB_4115) (Fig. 2B, Table S2). Only three instances of the same mutation in different strains were identified.   Phage susceptibility profiles of M. abscessus clinical isolates. We previously reported that M. abscessus strain GD01 is efficiently infected and killed by the phages Muddy, BPs, and ZoeJ, although these were the only phages identified from a screen of about 100 individual phages isolated on M. smegmatis (17). Few phages have been isolated on M. abscessus strains directly (17), and initial evaluation of several dozen M. smegmatis phages confirmed that many do not infect any of the clinical isolates; we therefore focused on eight of the most promising candidates (Fig. 1, Fig. 2C, Fig. 3): Muddy, BPs, ZoeJ, Itos, Faith1, Fionnbharth, D29, and Elmo (members of clusters/subclusters AB, G1, K1, L2, L2, K4, A2, and A3, respectively) (26,27) or their derivatives, as described below.
The substantial variation in the infection profiles is striking (Fig. 3). Many different susceptibility combinations are observed and there is no evident correlation with the whole-genome phylogenies. Approximately 34% of the strains are not susceptible to any of the phages tested, spanning subspecies and clades (Fig. 3). There are several instances in which closely related strains of different origins have similar profiles, such as GD01 and GD82 from patients in the UK and US, respectively, which differ by only a few dozen nucleotide polymorphisms and have similar susceptibility profiles (Table 1, Fig. 3).
Phages forming plaques on clinical isolates may not necessarily kill them efficiently, a Strains are given a GDXX designation; multiple isolates from the same patient are designated GDXXA, GDXXB, etc. b Strain origin indicates the location of the hospital or laboratory from which the strain was sent. c Indication of whether strain was from a cystic fibrosis patient; Y, yes; N, no; NA, data not available. d Subspecies derived from genome sequencing; a, abscessus; b, bolletii; m, massiliense. e Colony morphotype is shown as R, rough or S, smooth. f Genome sequencing with Illumina whole-genome sequencing (WGS) or Illumina and Nanopore to completion (Com). g Prophages are designated prophiGDXX, with -1, -2-, etc. extensions denoting different prophages in the same strain. h Plasmids are designated pGDXX, with -1, -2-, etc. extensions if there is more than one plasmid in the same strain.
and we therefore determined this using challenge assays to measure bacterial survival ( Fig. 4C and E, Table S3). There is a striking correlation between phage killing efficiency and strain colony morphotype. Of the 48 R strains, 38 (80%) are efficiently killed by at least one phage, 17 of which (35%) are killed by more than one (Fig. 2D). In contrast, none of the 38 S strains are killed, although a subset (21%) is infected by phages Faith1 and/or Itos, but neither kills the strains efficiently ( Fig. 1, Fig. 2D). Surface GPLs on the smooth strains may be responsible for the poor phage infection, but it is unclear why lytic derivatives of Faith1 and Itos (see below) do not efficiently kill the S strains they infect (Fig. 2D, Fig. 3). This poor-killing phenotype is not constrained to the subcluster L2 phages and is also observed in some R and S strains with phages BPs, Muddy, and D29 (Fig. 2D, Fig. 3). Expanding the phage repertoire by genome engineering. Several of the potentially useful phages are temperate and form stable lysogens in M. smegmatis. Muddy is not evidently temperate, and efficiently kills most of the M. abscessus strains it infects. Itos, D29, and Elmo are not temperate, but are naturally lytic derivatives of temperate phages that lack the repressor gene, which is responsible for establishing and maintaining lysogeny. We previously described the engineering of both BPs and ZoeJ to construct lytic derivatives (17,28), and we used similar approaches to construct lytic versions of Fionnbharth (i.e., FionnbharthD45D47) and Faith1 (i.e., Faith1D38-40) (Fig. 4A).
Expanding the phage repertoire using host range mutants. In screening strains, several different phenotypes of M. abscessus phage infection were observed. These include an absence of infection, an efficiency of plaquing (e.o.p.) of one relative to M. smegmatis, and killing of cells only at high phage concentrations (Fig. 2C). Sometimes, individual plaques are observed at high phage titers, arising from phenotypic escape of defense systems, such as restriction or genetic alteration that promotes efficient infection. We previously described a host range mutant (HRM) of phage BPsD33HTH (BPsD33HTH_HRM10) (17), and we successfully isolated another HRM of BPsD33HTH (BPsD33HTH_HRM GD03 ) using M. abscessus GD03 as the host (Fig. 4B). We similarly isolated HRMs of phages D29 and Faith1D38-40 (Fig. 4B).
The lytic subcluster A2 phage D29 (29,30) does not efficiently infect any M. abscessus strain, but an HRM was isolated on M. abscessus GD40, purified, and shown to efficiently infect GD57, GD89, and GD41, in addition to GD40 (Fig. 3, Fig. 4B). D29_HRM GD40 has three mutations relative to its immediate precursor, C12442A, T31726C, and A43657T, but reconstruction of the individual mutants demonstrated that the C12442A mutation conferring a T116N substitution in the capsid protein is responsible for the host range phenotype (Fig. 4B). Capsid mutations associated with phage host range are unusual but have been described to influence adsorption of f X174 (31). The potential therapeutic utility of D29_HRM GD40 is illustrated by its enhanced killing of strain GD40 relative to the D29 parent phage (Fig. 4C).
Faith1 does not efficiently infect M. abscessus GD69A but an HRM was isolated (Faith1D38-40_HRM GD69A ) that does (Fig. 4B), although the HRM does not change its infection of other strains. Faith1D38-40_HRM GD69A has a single mutation at G34936T conferring an early termination codon in gene 43 (E11*), whose function is unknown, but it is highly expressed in early lytic growth of its relative Crossroads (32). Faith1 gp43 is unlikely to be involved in cell surface interactions and may be triggering a host defense system to which the Faith1D38-40_HRM GD69A mutant is able to escape.
M. abscessus phage resistance. Phage-susceptible strains were challenged with phage(s) in liquid cultures and 10 7 cells plated on solid medium to recover survivors. Robust growth was observed for all of the smooth strains (.50% survival) reflecting very poor killing (Table S3). In contrast, 70% of the 75 combinations of phage and rough strains, using a total of 28 strains, yielded either no or very few survivors, which upon retesting were either fully or partially phage sensitive ( Fig. 4D and E; Table S3). Only 12 of the 28 strains (43%) yielded any resistant mutants (Table S3). With the exception of mutant GD19_RM5, which converted to S, all of the resistant mutants retained their R morphotype (Fig. 4F). This is notable because reversion or suppression to smooth could be a simple route to resistance, but evidently is relatively uncommon due to the type of mutations conferring rough morphotypes (Fig. 2B). We note that no resistant strains for GD43B were recovered (Table S3), suggesting that reversion or suppression of the mps1 nonsense mutation occurs below the detection limit of the assay.
Nine phage-resistant strains were sequenced and compared to their phage-  sensitive parents (Table S4). GD19_RM5 (Fig. 4F) contains a wild-type allele of mps1 (MAB_4099), and direct reversion of the 35-bp insertion in mps1 (MAB_4099) has given both the smooth morphotype and resistance to Muddy (Fig. 2B, Fig. 4F). Two mutants (GD17_RM1 and GD22_RM4) have mutations in a type I polyketide synthase (MAB_0939), implicated in synthesis of trehalose polyphleates (33); related proteins are nonessential for M. tuberculosis growth in vitro but are important for virulence (34). Mutants GD22_RM1 and GD22_RM2 have mutations in the C-terminal HRDC domain of UvrD2 (MAB_3511), which is nonessential for M. tuberculosis growth (35). For GD26_RM4 and GD25_RM2, it is unclear which genes are implicated in resistance, either a 28.5-kbp deletion or multiple changes from its parent; however, both include genes implicated in virulence (Table S4). GD22_RM3 has lost plasmid pGD22-1 (36), which could impact phage infection. Mutant GD19_RM3 (Fig. 4F) has a frameshift mutation (4-bp deletion) in rpoZ coding for the RNA polymerase omega subunit, and is the sole sequence difference from GD19; rpoZ is clearly not essential for M. abscessus growth, as reported for M. smegmatis (37). Deletion of rpoZ in M. smegmatis does not alter the GPL profile but it reduces both sliding motility and biofilm formation, and has a notable reduction in short-chain mycolates on the cell surface (37). Similar surface changes in GD19_RM3 may be responsible for the inability of phage Muddy to infect. None of the other strains described here have rpoZ mutations, and this RM characterization illustrates the multitude of mechanisms that alter phage susceptibility.
Isolation and propagation of spontaneously released lytic phages. To expand the suite of phages with therapeutic potential, we searched for phages that are spontaneously released from the M. abscessus strains that form plaques on other strains. A screen of over 1,200 pairwise tests identified nine distinct phages (designated phiGDxx), corresponding to nine different donor and five recipient strains ( Table 2). Each phage was purified on its recipient strain, sequenced, and annotated (Fig. 5, Table 2, Fig. S1 to S5, Fig. S6 to S9) (https://phagesdb.org/documents/categories/15/). All have features typical of temperate phages and their organizations and virion genes are consistent with their siphoviral morphologies (Fig. 5A). None grow on M. smegmatis, and none are closely related to previously described M. smegmatis phages (26,27). Electron microscopy showed all nine phages have siphoviral morphologies (Fig. 5A). The phiGDxx phages are expected to be temperate and form turbid plaques, as they are derived from resident prophages. This is observed for most of the phages, but not for phiGD89-1, which forms clear plaques (Fig. 5B). phiGD89-1 differs from its close relatives phiGD21-1 and phiGD34-2 by several nucleotide differences immediately upstream of the early lytic genes, which may confer the clear-plaque phenotype. phiGD34-2 forms  Genes are shown as boxes above (transcribed rightward) and below (transcribed leftward) each genome; boxes are colored according to the gene phamilies they are assigned (50). Slim bars above genome sequence identify categories of gene functions: green bar, structural genes; red bar, lysis cassette; purple bar, immunity cassette; blue bar, replication genes, orange bar, DNA modification genes; and olive bars, polymorphic toxin genes. Detailed genome maps for all phages are in Fig. S1 to S5 and Fig. S6 to S9 in the supplemental material (https://phagesdb.org/documents/categories/15/). turbid plaques on most strains, but not on all, and the plaques appear clear on strains such as GD35 (Fig. 5B). Each of the nine phages was tested for infection of the GDxx strains (Fig. 3). Phage phiGD89-1 infects about one-third of the R strains, most of which are in the large M. abscessus subsp. abscessus clade (Fig. 3, Fig. 5B). The closely related phiGD21-1 has a similar profile, although some strains, such as GD08, distinguish between the two phages (Fig. 3). Seven strains-four S type (GD34, GD84 GD90, and GD100A) and three R type (GD14, GD23, and GD88)-that are not killed efficiently by other phages are efficiently infected by at least one phiGD phage (Fig. 3), increasing the proportion of efficiently infected R strains to 85%. They also add a potential second phage to 10 other R strains that are only killed by a single phage. This expansion of the phage repertoire is illustrated by the ability of phiGD89-1 to kill GD14, for which no other phages have been identified (Fig. 3, Fig. 5C). The temperate phiGDxx phages will need to be engineered for obligatory lytic growth prior to therapeutic consideration. The phiGDxx genomes vary in length from 40 to 60 kbp and represent five distinct groups based on overall sequence relationships (Fig. 5D, Table 2); detailed genome maps are shown in Fig. S1 to S5 and Fig. S6 to S9 (https://phagesdb.org/documents/categories/15/).
M. abscessus prophages. Using the phiGDxx sequences, we identified the cognate prophages (designated prophiGDxx) in the donor strains and extracted their complete prophage sequences (Table 1, Fig. 6). Genome comparisons showed that phages phiGD21-1 and phiGD57-1 are identical to their cognate prophages, and spontaneous induction involves simple excisive site-specific recombination between the attL and attR junctions. However, six of the phages differ from their prophages by one or more base substitutions and/or insert/deletions, which presumably occurred during induction or subsequent lytic growth (Fig. 6A). Curiously, prophiGD20-1 is 9,977 bp larger than phiGD20-1 and contains an 8,452-bp transposon insertion in an HNH-like gene, which is defined by 25-bp imperfect inverted repeats (IRs) and flanked by a 5-bp target duplication (Fig. 6B). Loss of the transposon-presumably a requirement for lytic growth due to DNA packaging constraints-has occurred by transposase-mediated imprecise excision between the rightmost IR junction and a location within prophiGD21-1 gene 110; excision also removes an Mre-11-like gene which is evidently not required for lytic growth (Fig. 6B). A second closely related prophage (prophiGD15-1; see below) has a similar transposon inserted within the same HNH gene, 25 bp to the left of that in prophiGD20-1. Moreover, two additional prophages, prophiGD41-1 and prophiGD59-1, are identical to prophiGD15-1 with the same transposon insertions. The transposon is also present in M. abscessus subsp. bolletii CCUG, M. abscessus FLAC013, and M. abscessus UC22 (GenBank accession numbers AP014547, CP014955, and CP012044).
Bioinformatic analyses shows that prophages are abundant in the GD strains, as has been reported for other M. abscessus strains (38). These prophages, along with plasmids from these strains, are described in detail elsewhere (36) (Tables S5 and S6) (https://phagesdb.org/documents/categories/15/), and their inclusion in each strain is indicated in Fig. 3. Only 12 strains are prophage-free, and the 122 prophages and nine lytically growing phages (Table 2) are grouped into clusters (MabA to MabQ) according to their sequence relationships (Fig. 6A, Table S5). Because the phage susceptibility profiles do not correlate closely with nucleotide sequence-based phylogeny (Fig. 3), the prophage and plasmid contents are strong candidates for contributing to the phage infection profiles. The abundance and diversity of the M. abscessus mobilome confounds any simple elucidation of their roles. We note, however, that many of the prophages and plasmids code for type II toxin-antitoxin systems which can confer viral defense (36).

DISCUSSION
The phage infection profiles of these M. abscessus strains illuminate both the opportunities and challenges in their therapeutic prospects. Smooth morphotype strains are problematic and not efficiently killed by any of the phages, and a surprisingly high proportion (36%) of strains are smooth, suggesting that they are important pathogens, but perhaps with pathologies distinct from rough strains. The observation that smooth strains with abundant surface GPLs are not efficiently killed by phages is in contrast to the finding that GPLs are required for M. smegmatis infection by phage I3 (39). Through extensive phage screening, genome engineering, host range mutant identifications, and induction of prophages, a relatively small set of phages were identified that can infect and kill 80% of the rough strains. The induced prophages require further engineering to remove genes for lysogeny and potential virulence genes, but are showing genomic differences; phages and prophages are named after the strain they were isolated from. The prophiGD23-1 sequence is incomplete in the WGS assembly and thus cannot be fully compared with phiGD23-1. In phiGD89-1, there are several differences in a region that is well conserved in other MabB phages, including phiGD21-1. Specifically, the sequence 59-TGGACTACGGCTGAGCAGCA-TGCT (coordinates 30,519 to 30,542) replaces 59-TGGGCTACGGCTGAGCAGTAGAACT. This is in the location of a predicted rightward early promoter and operator site, and likely inactivates lysogeny. (B) Comparison of phage and prophage genome segments, illustrating the transposon insertions (gray areas on genome rulers) in prophages prophiGD20-1 and prophiGD15-1. Genome annotations and comparisons are illustrated as in Fig. 5D. The alignment illustrates the transposon insertion (comparing phiGD22-1 and prophiGD15-1), and excision of the transposon and flanking gene in phiGD20-1. The transposons in prophiGD20-1 and prophiGD15-1 are closely related but inserted at targets 25 bp apart. a promising source of new phages, increasing the number of rough strains targeted to about 85%. Until additional phages are identified that expand the repertoire to kill all of the rough strains, screening of individual strains is still required before therapeutic intervention because of the variable and unpredictable phage susceptibility profiles. The variability in phage infection is likely determined in large part by the impressive array of diverse prophages and plasmids in these M. abscessus clinical isolates.
With only a rather limited set of phages available for therapy-and a substantial proportion of rough strains are killed by only a single phage-the frequencies and mechanisms of phage resistance are key concerns. For most M. abscessus phage infections, no stably resistant survivors were isolated, and when resistant mutants were recovered, the mutations suggest there may be a fitness cost for in vivo growth. Furthermore, although reversion to a smooth morphotype is expected to contribute to resistance, this occurs only rarely, reflecting the predominance of mutations leading to rough morphotypes that occur only infrequently. In general, use of a single phage for therapy when antibiotic treatments are no longer available would seem a reasonable strategy. We recognize that some infections contain both rough and smooth M. abscessus variants and can vary by anatomical location and duration (14,15,22), raising the question as to whether phage-mediated killing of the resistant strain alone will provide any therapeutic effect. This is likely to vary considerably among patients, depending on the virulence capacity of the smooth variant.
Challenges in the therapeutic use of phages arise for many other bacterial pathogens other than M. abscessus, including Mycobacterium avium and Burkholderia spp. infections. The strategies deployed here, including exploitation of phage libraries of related hosts, phage genome engineering, host range expansion, and growth of induced prophages, can be applied to these and other pathogens to expand the repertoire of therapeutically useful phages. Although the number of useful phages may remain relatively small for some infections, infrequent phage resistance and tradeoffs for in vivo fitness should improve their therapeutic utility.

MATERIALS AND METHODS
Bacterial strains. M. smegmatis mc 2 155 is a laboratory stock strain and was grown as previously described (40). M. abscessus ATCC 19977 was obtained from the American Type Culture Collection. M. abscessus clinical isolates were received on Lowenstein-Jensen slants and streaked out on Middlebrook 7H10 agar (Difco) supplemented with oleic acid-albumin-dextrose-catalase (OADC) and 1 mM CaCl 2 and were grown for 5 to 7 days at 37°C. Liquid cultures were inoculated from a single colony and grown in Middlebrook 7H9 medium with OADC and 1 mM CaCl 2 for 4 to 5 days at 37°C, with shaking. For plaque assays, M. abscessus cultures were sonicated briefly in a cup-horn sonicator (Q-sonica) as described previously (17). The GD54 strain sequenced and analyzed was GD54H, one of several isolates from the same source. Similarly, the GD35 strain sequenced and analyzed was otherwise designated GD35B. Strains GD43A and B have different numbers of prophages and were therefore analyzed as separate strains. Strain GD43A has six prophages (in clusters MabA, MabB, MabC, MabJ, MabK, and MabL) and GD43B has only four of these (MabA, MabL, MabK, and MabJ). There are no plasmids in either GD43A or GD43B.
Genome sequencing. Bacterial and phage genomes were sequenced by Illumina alone or supplemented with Nanopore reads as described previously (41,42) and details are provided in Table S1. For all strains, Illumina sequencing libraries were prepared from genomic DNA using NEB Ultra II FS kits with dual-indexed barcoding. Libraries were pooled and run on an Illumina MiSeq, yielding 300-base pairedend reads. In some cases, Oxford Nanopore libraries were also constructed from genomic DNA using Rapid Sequencing Barcoding kits, then pooled and run on a MinION device using FLO-MIN106D flowcells. Illumina reads for each strain were trimmed and quality-controlled using Skewer (43). Trimmed Illumina reads were then assembled using Unicycler (44), incorporating Nanopore reads when available.
In the case of complete genomes, assemblies were viewed, stitched, corrected, and finalized using Consed version 29 (45,46). GraphMap (47) was used to align long Nanopore reads to provisional assemblies and resolve repetitive regions. The first base and orientation of each complete circular chromosome was chosen to match those of the ATCC 19977 strain and/or to align with the first base of the dnaA gene. Complete circular plasmids were similarly oriented and cut so that base 1 was the first base of a predicted repA gene.
Of the 122 prophages identified, 80 were either in completely sequenced genomes or were wholly within one contig in WGS assemblies (Table S5), and complete genome sequences were extracted. The other 42 were in multiple contigs (Table S5) and, although complete prophage sequences were not available, sufficiently large segments were available to reveal their relationships to other prophages.
Several of the clusters contained only a single prophage member, although database searches suggested that all of these have relatives in other sequenced M. abscessus genomes.
The very extensive diversity of the plasmid genes is reflected in the finding that when sorted into protein phamilies (as described previously), 58% of the genes are "orphams" without closely related genes in this data set.
Phage infections. Mycobacterial strains were grown and tested for phage susceptibility as described previously (17). Twenty-three mycobacteriophages were used for the initial screening of M. abscessus clinical isolates. These phages were chosen from clusters known to have expanded host range (40) and others were chosen based on screening completed previously on M. abscessus clinical isolates (17). Phage lysates were 10-fold serially diluted and plated on M. smegmatis mc 2 155 and each M. abscessus clinical isolate. The phiGDxx series of phages were isolated by plating clarified lysates of M. abscessus strains on M. abscessus strains, purifying phages from cleared areas on the sensitive strains, and growing to high titer.
Phage-resistant M. abscessus mutants. Survival assays of M. abscessus strains sensitive to phages were set up to isolate resistant mutants. An aliquot of 1 ml of ;1 Â 10 8 CFU/ml of culture was mixed with, or without, one log higher concentration (;1 Â 10 9 PFU/ml) of phage. These were incubated at 37°C with shaking, and 100ml was plated after 2 and 5 days. Plates were grown at 37°C for 5 to 7 days and then photographed. Colonies from these plates were picked, streaked twice, and then grown in liquid culture to be used in top agar overlays for phage-resistance screening. Top agar overlays were spotted with serial dilutions of the phage of interest and incubated at 37°C for 5 to 7 days. phiGDxx phages. Log phase liquid cultures of clinical isolates were grown, and one ml of each culture was centrifuged at 14,000 Â g for 2 min. The supernatant (from each "donor" strain) was saved and used for spotting on top agar overlays of other clinical isolates (the "recipient" strain). These plates were incubated at 37°C for 5 to 7 days. Any clearing on the overlay where a supernatant was spotted was picked, using a sterile pipet tip, into phage buffer (10 mM Tris-HCl [pH 7.5], 10 mM MgSO 4 , and 68 mM NaCl). The phiGDxx phages were purified and then amplified on the receiving strain. DNA was extracted from lysates of phiGDxx phages using a standard phenol-chloroform/EtOH precipitation protocol.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.

ACKNOWLEDGMENTS
We are very grateful to the many colleagues who sent us M. abscessus strains.