The ESX-4 substrates, EsxU and EsxT, modulate Mycobacterium abscessus fitness

ESX type VII secretion systems are complex secretion machineries spanning across the mycobacterial membrane and play an important role in pathogenicity, nutrient uptake and conjugation. We previously reported the role of ESX-4 in modulating Mycobacterium abscessus intracellular survival. The loss of EccB4 was associated with limited secretion of two effector proteins belonging to the WXG-100 family, EsxU and EsxT, and encoded by the esx-4 locus. This prompted us to investigate the function of M. abscessus EsxU and EsxT in vitro and in vivo. Herein, we show that EsxU and EsxT are substrates of ESX-4 and form a stable 1:1 heterodimer that permeabilizes artificial membranes. While expression of esxU and esxT was up-regulated in M. abscessus-infected macrophages, their absence in an esxUT deletion mutant prevented phagosomal membrane disruption while maintaining M. abscessus in an unacidified phagosome. Unexpectedly, the esxUT deletion was associated with a hyper-virulent phenotype, characterised by increased bacterial loads and mortality in mouse and zebrafish infection models. Collectively, these results demonstrate that the presence of EsxU and EsxT dampens survival and persistence of M. abscessus during infection.

Phylogenetic analyses suggest that esx-4 is the ancestral locus from which all other mycobacterial ESX systems have evolved [6]. Until recently, ESX-4 was considered as non-functional in pathogenic mycobacteria, such as M. tuberculosis [8]. Comparatively, deletion of one of the structural components of the M. abscessus T7SS ESX-4, EccB4, resulted in decreased intracellular survival, with impaired phagosomal membrane rupture and phagosomal acidification [17]. However, the underlying molecular mechanism explaining the ex vivo attenuated phenotype of the ΔeccB4 mutant remains to be established. Noteworthingly, ΔeccB4 was associated with reduced secretion of two esx4-encoded proteins, EsxU and EsxT, in the culture supernatant, suggesting a link between decreased intracellular replication of ΔeccB4 and the level of EsxU/EsxT secretion [17].
M. abscessus represents an excellent surrogate to examine the role of ESX-4 effectors with respect to host-pathogen interactions and persistence in the host, involvement in membrane interaction and phagosomal escape and virulence. Critically, deletion of esxUT resulted in enhanced virulence in infected mice and zebrafish, characterised by increased bacterial burdens and mortality. These findings open up the attractive hypothesis that the inability to secrete these two ESX-4 substrates may have led to the emergence of increased virulence in strictly pathogenic mycobacteria.

M. abscessus EsxU and EsxT co-expression and secretion
EsxU and EsxT are encoded by the M. abscessus esx-4 locus [5] (Fig 1A) and belong to the WXG-100 family, similarly to the M. tuberculosis EsxA and EsxB effectors. These proteins are characterised by the presence of the characteristic WXG (tryptophan-X-glycine) motif in EsxU and a secretory signal motif (HxxxD/ExxhxxxH) in the C-terminus of EsxT [18] (Fig 1B). Previous work indicated that the EsxU and EsxT protein levels were slighty reduced in the secretome of M. abscessus ΔeccB4, suggesting that they are, at least partially, secreted through the ESX-4 machinery [17].
Initially, we sought to visualise the localization of EsxU and EsxT on the surface of M. abscessus using the ΔesxUT mutant, which was then individually complemented with GFPtagged EsxT or EsxU (S1A Fig). Probing the crude lysates from the various strains with anti-GFP antibodies revealed a band of the expected size, corresponding to EsxU-GFP or EsxT-GFP proteins (S1B Fig). Using confocal microscopy, we observed dispersed green fluorescent areas for EsxU-GFP, EsxT-GFP, or EsxUT-GFP as well as exclusively detectable cell wall labeling (Fig 1C, white arrows). This contrasts with the green fluorescent signal of GFP alone, which was homogenously distributed along the longitudinal axis of the bacilli (Fig 1C). The intensity of the EsxU-GFP signal was strongly reduced compared with the EsxT-GFP signal, in agreement with the Western blot analysis using anti-GFP antibodies (S1B Fig).
Proteins resulting from subcellular fractionations or concentrated culture filtrates were then probed with specific mouse polyclonal antibodies to reveal the presence of EsxU and EsxT. EsxU-and EsxT-specific bands were mainly found in the culture filtrate and total lysate ( Fig 1D). The low accumulation of EsxU and EsxT subunits in cell wall and plasmic membrane fractions suggests that, unlike their native promoter, the hsp60 promoter, which allows constitutive expression of GFP-tagged versions, may artificially increase their respective association at the bacilli's periphery ( Fig 1C).
As described for other low molecular weight Esx substrates [19,20], EsxU and EsxT were co-purified using M. smegmatis as a heterologous expression system (Fig 1E, left panel). The proteins were identified by mass spectrometry and subsequently analysed by Multiple angle laser light scattering (SEC-MALLS), which showed the presence of an approximately 20-kDa heterodimer in solution ( Fig 1E, right panel). EsxU and EsxT from M. abscessus were also coexpressed in Escherichia coli where EsxT was co-purified with EsxU under native conditions or alone under denaturing conditions after elution at pH 5.9 and 4.5, whereas EsxU was found in the unretained column lysate (Fig 1F). The strong association between EsxU and EsxT was further confirmed by GFP-pull-down experiments performed in M. abscessus ΔesxUT complemented with pVV16-esxUT-GFP and followed by mass spectrometry analysis, with EsxT being physically associated with EsxU (S1C-S1E Fig). This analysis suggests that EsxT may also act in concert with additional protein partners (S1E Fig). Collectively, these data indicate that an EsxU/EsxT heterodimer is expressed in M. abscessus and is translocated across the plasma membrane and secreted into the culture filtrate.

EsxU and EsxT are involved in phagosomal membrane damage in macrophages
In M. tuberculosis, the ESX-1 substrates, EsxA and EsxB, promote intracellular survival and phagosomal escape [21], which prompted us to investigate the functions of EsxU and EsxT in macrophages. To determine the expression levels of esxU and esxT during macrophage infection, qRT-PCR analysis was performed after infection of macrophages, which showed  (Fig 2A). While these results indicate that esxU and esxT are transcribed under intracellular conditions, it may be speculated that the EsxU/EsxT heterodimer plays a role in intracellular replication of M. abscessus. To test this hypothesis, a esxUT deletion mutant (ΔesxUT) was generated (S2A and S3A and S3B Figs) along with the corresponding complemented strains (ΔesxUT::esxUT) (S3C and S3D Fig). As anticipated, this mutant failed to produce EsxU and EsxT (Fig 1D). Macrophages were then infected with the wild-type (WT), ΔesxUT, or ΔesxUT::esxUT strains (Fig 2B). ΔesxUT showed no morphotypic changes on plates or growth defects in broth medium compared with the WT strain (S2B and S2C Fig). Furthermore, there were no major differences in the intracellular growth between the three strains (Figs 2B and S3E). Similar results were obtained in amoebae, a unicellular phagocytic cell type regarded as a potential environmental reservoir, possibly involved in shaping the M. abscessus virulence profile for intracellular adaptation [17,22,23] (S2D Fig). Next, we examined the potential effect of EsxU/EsxT on phagosomal membrane disruption using a FRET-based assay [17]. As shown in Figs 2C and S3F, ΔesxUT and heat-killed WT M. abscessus were unable to damage the phagosomal membrane, unlike the WT or Δesx-UT::esxUT strains. However, ΔesxUT remained present in a non-acidified compartment, as evidenced by confocal microscopy, similar to the WT strain, but to a lesser extent than heat-killed M. abscessus (Fig 2D and 2E). Altogether, these results suggest that the loss of EsxU and EsxT abolishes phagosomal membrane damage but maintains the blockage of phagosomal acidification.

EsxU/EsxT interacts with artificial eukaryotic membranes
To decipher the properties of the EsxU/EsxT heterodimer and its potential interaction with biological membranes, biophysical experiments were performed using dioleoyl-phosphatidylcholine (DOPC) as a model of eukaryotic membranes [24] as well as the EsxU/EsxT complex expressed and purified from M. smegmatis. The heterodimer-membrane interaction was then studied using the NanoSpot approach (Fig 3A). The intensity of a fluorescent dye trapped in cavities underneath a DOPC bilayer was monitored. We observed a decrease in fluorescence upon addition of purified EsxU/EsxT, suggesting a direct interaction with the membrane that allowed the dye to escape (Fig 3B). The decrease in intensity by photo-bleaching of the fluorescence dye is indicated by the red curve. To verify whether the complex formed an organized structure, such as a pore, and did not just perforate the membrane in a single event by complete membrane destruction, a second fluorescent dye (larger dextran conjugate) was added above the membrane. Single-cavity analysis showed an increase in fluorescence of the second dye in several cavities (two cavities represented), indicating dextran influx. The concentration dependence is not linear, demonstrating that the pore formation depends on a critical local aggregation of the proteins on/in the membrane [24]. Taken together, this demonstrates that EsxU/EsxT complexes are capable of forming stable higher-order oligomeric structures in lipid bilayers (Fig 3C).
Representative confocal microscopy images of ΔesxUT strains expressing either GFP, EsxT-GFP, EsxU-GFP or EsxUT-GFP to illustrate protein localization. A series of Z-stacks were collected with a Zeiss LSM880 Airyscan confocal microscope and the images were assembled and reconstructed with Zen blue software. cytoplasm communication within the macrophage, which may in turn affect its pathogenicity. To explore the role of the EsxU/EsxT complex in the virulence of M. abscessus in the presence of a functional immune system, several previously established animal models were used [25,26]. In this context, the zebrafish (Danio rerio) model, which harbours only innate immunity (used as an acute infection model), and the C3HeB/FeJ mouse model, which possesses both innate and adaptive immunity (used as a persistent infection model), were exploited to address the contribution of EsxU/EsxT in M. abscessus pathogenesis. Survival curves in both animal models demonstrated that ΔesxUT confers a hypervirulent phenotype as compared the NanoSpot method. DOPC bilayers were built over cavities filled with a fluorescent dye (green). A second fluorescent dye of known size (red) was added to the top of the membrane prior to addition of the protein complex. Fluorescent signals were continuously measured. (B) Dye influx and efflux using the NanoSpot technique. DOPC bilayers were formed at 37˚C and pH 7.2 on top of cavities filled with Atto488 carboxy fluorescence dye. Fluorescence (black curve) decreased as more protein was added (final EsxU/ EsxT concentration is indicated in blue). Upon addition of Triton X-100, fluorescence decreased abruptly. The red curve represents the trend line of fluorescence loss by photo-bleaching if no protein had been added. Over 300 cavities covered with DOPC membrane are plotted in this graph. (C) 40, 000 MW TexasRed-labelled dextran was added to the media outside the cavities before the EsxU/EsxT complex was supplemented to the well (final concentration: 10 μM). Green dots represent efflux of Atto488 carboxy efflux dye and red dots represent influx of TexasRed dye. Each graph represents one cavity. https://doi.org/10.1371/journal.ppat.1010771.g003

PLOS PATHOGENS
ESX-4 substrates and mycobacterial fitness with WT and complemented strains (Fig 4A and 4F). A significant increase in ΔesxUT bacterial burdens were found in zebrafish embryos ( Fig 4B) as well as in C3HeB/FeJ mice (Fig 4G), presumably explaining the increased mortality observed in both animal models (Fig 4A and  4F). Furthermore, we observed a significantly higher proportion of embryos with granulomas at both 4 and 6 days post-infection (dpi), as well as a significantly increased number of embryos with abscesses at both 4 and 6 dpi in zebrafish infected with ΔesxUT as compared with WT and complemented strains (Fig 4C-4E). Similarly, we observed a significant increase in the number of CFU in the lungs, liver, kidneys and spleen at 14 and 20 dpi, confirming the increased growth of ΔesxUT in vivo ( Fig 4G).
Overall, these findings support the view that deletion of esxUT in M. abscessus confers a hypervirulent phenotype in two complementary animal models.

Discussion
Despite the increasing incidence of M. abscessus infections, little is known regarding the route of transmission of human-acquired infections [27]. Genetic evidence suggests that amoebae may represent an environmental reservoir, but this remains yet to be proven [7]. The amoebae/M. abscessus pairing unraveled important virulence determinants in M. abscessus [22,28], including a functional ESX-4 system [17]. This was subsequently confirmed in M. tuberculosis and M. marinum, by studying secretion of two ESX substrates, EsxE and EsxF [29] and of the CpnT toxin [8,30] and for which M. tuberculosis ESX-4 is essential for their secretion both in vitro and in macrophages [8]. However, M. tuberculosis and M. marinum both lack the essential EccE4 component of the ESX-4 secretory machinery, recently demonstrated to be essential for ESX-3 and ESX-5, which calls into question a functional and/or a stable ESX-4 secretory apparatus in these two species, yet this remains to be further investigated [1,2]. This is supported by a recent study indicating that ESX-4 from M. marinum does not secrete EsxT and EsxU [31]. It is, therefore, possible that the absence of eccE4 in the M. marinum esx-4 locus compromises the pore-forming capacity of ESX-4 and impaired the substrate transport activity. An important aspect of this work relies on the role of ESX-4 in phagosomal membrane permeabilization/disruption [8], a feature previously ascribed to ESX-1, but without incriminating the EsxU or EsxT secreted substrates, encoded by the M. tuberculosis esx-4 locus. As mentioned above, mechanistic insights into the secretion of these molecules by ESX-4 in pathogenic mycobacteria are scarce, and mainly referring to M. marinum [31]. Using molecular and biophysical approaches, we confirmed that both EsxU and EsxT are substrates of M. abscessus ESX-4, forming a 1:1 heterodimer, which is secreted into the culture filtrate. Further biophysical experiments confirmed that the EsxU/EsxT complex is capable of interacting with artificial membranes and forming oligomeric arrangements permeabilizing the bilayer. A possible role in membrane permeabilization is also suggested by the macrophage experiments in which ΔesxUT remained inside the phagosome and was unable to induce phagosome membrane damage, as evidenced by the lack of a FRET signal in the phagosomal disruption FRETsignaling assay. As anticipated, phagosome-cytoplasm communication was restored in the complemented strain. Comparatively, ΔeccB4 also failed to induce phagosomal membrane permeabilization [17]. The ΔeccB4 secretome showed reduced secretion of EsxU and EsxT, confirming the association between EsxU/EsxT with phagosomal membrane permeabilization and the potential communication of M. abscessus with the macrophage cytosolic compartment [17]. Finally, immunofluorescence and immunoblotting analyses emphasized the localization of EsxU and EsxT in the cell wall periphery as well as in the culture filtrate. This localization The phenotypic comparison between the two classes of mutants (ΔeccB4 versus ΔesxUT) revealed a major difference in their intracellular behavior. While ΔesxUT failed to show a defect in intracellular growth compared to the WT strain, presumably because it resides in phagosomes that do not acidify, this contrasts with the phagosomes containing ΔeccB4 that acidify, leading to reduced intracellular survival [17]. The maintenance of M. abscessus within the phagosome, in contact with endocytosis and recycling pathways has been considered advantageous for intracellular multiplication of pathogenic mycobacteria [32]. It allows to escape apoptosis and autophagy, which are events encountered within the cytosol [33][34][35]. Furthermore, differences in phagosomal acidification support the existence of additional bacterial substrate(s), that may still be present in ΔesxUT but not in ΔeccB4. The absence of a functional secretion apparatus in ΔeccB4 may impart the secretion of other compounds than just ESX-4 substrates. Indeed, deletion of eccC4 in M. marinum resulted in elevated secretion of ESX-1 and ESX-5 protein substrates, underscoring the dialogue between ESX-4 and other T7SS [31]. In addition, dysfunctional ESX-4 may also impact the expression and/or biological activity of additional cell wall components, such as lipids. For instance, the synergistic role of phtiocerol dimycocerosates (PDIM) associated with ESX-1-secreted effectors has been demonstrated in M. tuberculosis [36,37]. In M. abscessus, deletion of mmpL8 MAB involved in the transport of another lipid, designated GDND, does not allow the bacteria to damage the phagosomal membrane [38]. These findings demonstrate the fine regulation and interaction between ESX-4 and its substrates in the pathophysiology of M. abscessus within phagocytic cells.
We show that ΔesxUT displays a hypervirulent phenotype in zebrafish and mice, supported by increased bacterial growth and mortality and enhanced granuloma and abscess formation, thus highlighting a central role for EsxU/EsxT in M. abscessus virulence. Given that EsxU/EsxT are largely found at the bacterial periphery and are capable of inducing phagosomal damage, one would predict that the absence of EsxU/EsxT leads to changes in the intracellular host-pathogen interface. This is not surprising, given that the smooth variant of M. abscessus used in this study is characterized by a tight phagosomal apposition surrounding the bacterial surface in macrophages [39]. Moreover, M. abscessus possesses a myriad of complex lipids on the outer surface, including glycopeptidolipids (GPL) and GDND, known to modulate infection outcomes and which are likely unaltered upon esxUT deletion. Interestingly, a hypervirulent phenotype was observed in adult zebrafish following infection with an ESX-5 mutant of M. marinum [40]. Similarly to ΔesxUT, an esx-5 transposon mutant of M. marinum in EspG was accompanied by increased granuloma formation in adult zebrafish, while no growth defect was observed in THP-1 macrophages, as compared to the WT M. marinum. In addition, deletion of a gene encoding a PE-PGRS protein (a substrate of ESX-5 in M. tuberculosis) exhibits a hypervirulent phenotype in mice [41]. These analogies suggest that the ESX-4 substrates, EsxU and EsxT, can exhibit a virulence-associated phenotype similar to some ESX-5 substrates.
Unexpectedly, we did not observe specific antibody production against EsxU and EsxT during infection in either mouse or in human sera (S5 Fig). This result may be related to the early induction of esxU and esxT genes during macrophage infection, and a plausible explanation determined by ANOVA with Tukey's test using GraphPad prism program (B, C and D) and unpaired t test (G); ns, not significant; � P < 0.05, �� P < 0.01, ��� P < 0.001, ���� P < 0.0001. The log-rank (Mantel-Cox) test for Kaplan-Meier survival curves was used to assess survival statistics significance. https://doi.org/10.1371/journal.ppat.1010771.g004

PLOS PATHOGENS
ESX-4 substrates and mycobacterial fitness could rely on a partial or complete loss of EsxU/EsxT production in M. abscessus once it is established in the host, presumably as an adaptive response to increase its level of virulence. Future work will focus on describing the regulatory mechanisms governing esxUT expression under infection conditions. Overall, these observations lead to the tempting hypothesis linking the emergence of strict human and animal mycobacterial pathogens, such as M. tuberculosis, to the loss of ESX-4 substrate secretion resulting from the selection of non-functional ESX-4 systems lacking the essential EccE4 component.

Strains, cells, and culture conditions
M. abscessus CIP104536 T from the collection of Institut Pasteur, regrow from ATCC 19977 T [42] was grown in Middlebrook 7H9 broth (BD Difco) supplemented with 1% of glucose or 0.05% Tween 80 and 10% oleic acid, albumin, dextrose, catalase (OADC enrichment; BD Difco) at 37˚C in the presence of antibiotics, when required. For bacterial selection, media were supplemented either with 25 mg/L of zeocin (Thermo Fisher Scientific) for ΔesxUT, 1 g/L hygromycin (Invivogen) for strains carrying pTEC27 (Addgene, plasmid 30182) and pMVH361 or with 250 mg/L kanamycin when harboring the pMV306 or pVV16 derivatives. J774 cells were maintained at 37˚C in 5% CO 2 using Dulbecco's modified Eagle medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (FBS) (Gibco). J774 cells were allowed to adhere in a 24-well plate at a density of 5x10 4 cells/well for 24 h at 37˚C in 5% CO 2 .

Identification of EsxT protein partners
Briefly, each bacterial pellet recovered from a 400 mL culture was resuspended in 2 mL of PBS containing 1 mM benzamidine. Subsequently, the bacteria were lysed using a bead beater and incubated in lysis buffer (50 mM Tris HCl pH 7.5, 250 mM NaCl, 5 mM EDTA, 1% NP40, 1% Triton, 1% Bridj96 and protease inhibitor [Benzamidine]). The supernatants were then subjected to immunoprecipitation following the instructions provided with the kit (GFP-Trap Agarose Beads from ChromoTek). Proteins bound to the beads were then separated and detected by silver staining. The remaining beads were subjected to trypsin treatment for mass spectrometry analysis.

Mass spectrometry analysis
After denaturation at 100˚C in 5% SDS, 5% β-mercaptoethanol, 1 mM EDTA, 10% glycerol, and 10 mM Tris pH 8 buffer for 3 min, protein samples were fractionated on a 10% acrylamide SDS-PAGE gel. The electrophoretic migration was stopped as soon as the protein sample entered 6 mm into the separating gel. The gel was briefly labeled with Coomassie Blue, and three bands, containing the whole sample, was cut. In gel, digestion of gel slices was performed PLOS PATHOGENS ESX-4 substrates and mycobacterial fitness as previously described [43]. An UltiMate 3000 RSLCnano System (Thermo Fisher Scientific) was used for separation of the protein digests. Peptides were processed and identified as previously described [44].

Proteomic data analysis
MS/MS data were analyzed using search engine Mascot (version 2.4.0, Matrix Science, London, UK). Searches were performed with a tolerance on mass measurement of 10 ppm for precursor and 0.02 Da for-fragment ions, against a composite target-decoy database (181602 � 2 total entries) built with a M. abscessus Uniprot database (taxonomy 561007, march 2020, 4940 entries) fused with the sequences of EsxT-GFP. For each sample, peptides were filtered out according to the cut-off set for protein hits with one or more peptides taller than 9 residues, and a 1% false positive rate.

EsxU and EsxT expression and purification in M. smegmatis and E. coli
esxUT, esxU and esxT were amplified from M. abscessus ATCC 19977 genomic DNA, cloned in the pMyNT vector using the SLiCE cloning method and transformed in M. smegmatis mc 2 155 groEL1ΔC cells. Gene expression was induced with 1% acetamide for 14 hrs at 37˚C. Cell pellets were stored at −80˚C until cells were lysed in a 50 mM Tris pH 8.5, 300 mM NaCl containing protease inhibitors and DNaseI using high-pressure emulsification. Unbroken cells were then removed by centrifugation (40 min at 12,000 g) and proteins were purified using 5 mL HiTrap TALON Crude (GE Healthcare) (5 CV equilibration, Load 100 mL sample, Wash 15 CV, 50 mM imidazole for 5 CV, 200 mM imidazole for 10 CV, 500 mM imidazole for 8 CV). The protein was then dialyzed into 50 mM Tris pH 8.5, 300 mM NaCl overnight in presence of TEV protease and ran through a 5 mL HiTrap TALON Crude (GE Healthcare). The resulting protein was then concentrated and run through an S75 preparative column (GE Healthcare). Concentration was performed using ultrafiltration (AMICON MWCO 3kDa). Protein purity was assessed using sodium dodecyl sulfate polyacrylamide gel electrophoresis.
For expression and purification in E. coli, esxUT, esxU and esxT were also cloned in the plasmid pET6×HN-C (Clontech, CA, USA), under the control of the T7/lac hybrid promoter, for expression in E. coli BL21(DE3) strain. Primers used for cloning are listed in S1 Table. Recombinant proteins were expressed after addition of IPTG at concentration from 0.1 to 10 mM in a 4 hr culture at 28˚C. Proteins were purified either under native or denaturing conditions, according to manufacturer recommendations for Ni-NTA column (Qiagen). The presence of LPS was tested for recombinant proteins at the maximum concentration used for the assay in Fig 3 (80 μM) with the ToxinSensor kit, an endotoxin detection system from Genscript based on the LAL test. Proteins were tested after purification from M. smegmatis and also after passage through an endotoxin removal column (ToxinEraser, Genscript). Furthermore, the 7H9 culture medium was also tested. All values obtained for proteins and medium were out-of-range, i.e. below the lowest value of the standard curve (0.01 EU/mL).

RNA extraction, qRT-PCR
RNA and cDNA were prepared from 5 mL of mid-logarithmic-phase bacterial cultures grown in 7H9. Control experiments without reverse transcriptase were performed for each RNA sample to rule out DNA contamination. The sigA gene was used as an internal control. The sequences of primers used for quantitative real-time PCR (qRT-PCR) are listed in S1 Table. For isolation of bacterial RNA from M. abscessus-infected J774.2 macrophages, RNA were extracted as previously described [28]. Each qRT-PCR was performed with three biological replicates.

ΔesxUT mutant construction and complementation
Disruption of the esxUT gene marked with a zeocin resistance cassette was achieved using the recombineering system as described previously [22,45]. Briefly, the homology arms (1,000 bp upstream and downstream of esxU and esxT) were amplified from the genomic DNA of the CIP104536 (S) strain and the zeocin cassette was amplified from the pMVH361zeo. The amplicons were then cloned into the pUC19 plasmid by using Gibson technology (New England Biolabs). The Gibson product was electroporated into E. coli. The sequence including the homology arms and the zeocin resistance cassette called "allelic exchange sequence" (AES) was purified after cleavage by the restriction enzymes PstI and KpnI. The AES (300 ng) was electroporated into the wild-type strain containing the pJV53, which expresses the Che9c recombinase and a kanamycin resistance gene. Clones were selected on 7H11 supplemented with zeocin (50 mg/L) and kanamycin (250 mg/L) medium and screened by PCR to select those in which both recombination events had occurred. The target gene was then replaced by the zeocin resistance cassette and the pJV53 was removed from the strain. Complementation was performed after amplifying and cloning esxUT into the integrative pMVH361, as described previously [22].
Disruption of the esxUT gene unmarked in M. abscessus was performed as described previously [46]. Briefly, the pUX1-katG suicide vector was used to generate unmarked double deletion mutants in M. abscessus CIP104536 T (S). The left and right arms were amplified, restricted with PacI/MfeI and MfeI/NheI, respectively, and ligated to PacI/NheI-linearized pUX1-katG. Electrocompetent M. abscessus was transformed with pUX1-katG-esxUT to generate ΔesxUT. Selection of bacteria having undergone the first homologous recombination event was done by visual screening of red fluorescent colonies on 7H10 OADC supplemented with 250 μg/mL kanamycin. After subculturing the culture overnight in 7H9 T/OADC in the absence of kanamycin, bacterial suspensions were serially diluted and plated onto 7H10 OADC with 50 μg/mL INH to select for INH-resistant, Kan-sensitive, and non-fluorescent colonies. The DNA junctions were PCR sequenced to confirm the genotype of ΔesxUT.
Complementation was performed after amplifying and cloning esxUT in fusion with an HA-tagging sequence under the control of the hsp60 promoter into the integrative pMV306, as described [46]. All primers used for the construction of the two mutants are listed in S1 Table.

SEC-MALLS experiments
The EsxU/EsxT complex was purified as detailed above and dialyzed in both 20 mM Tris pH 8.0, 300 mM NaCl, 5% glycerol. EsxU/EsxT was injected at 10 mg/mL and experiments were performed at a flow rate of 0.8 mL/min. SEC-MALLS was performed using an Agilent 1260 Infinity II Bio-inert LC system and an S200 5/150 (GE Healthcare). Protein elution was detected by absorbance at 280 nm, and protein concentration was quantified with differential refractometry using an Optilab T-rEX detector (Wyatt). Light scattering data were measured with a miniDAWN TREOS multiangle light scattering detector (Wyatt). Molecular weights were computed from the concentration and light scattering data using the software ASTRA version 7.1.3.15 (Wyatt).

Preparation of mycobacterial subcellular fractions and Western blotting
Briefly, 500 mL of mycobacterial cultures (wild-type smooth strain or ΔesxUT mutant) were grown to an OD 600~0 .3 in 7H9 medium containing 1% glucose at 37˚C with shaking. Cells were collected by a 30-min centrifugation at 10,000 × g and the pellet was washed with PBS containing 1 mM benzamidine and resuspended in 2 mL PBS containing 1 mM benzamidine. Bacteria were lysed in a bead beater and the lysates transferred into 15 mL tubes and subjected to two rounds of sonication for 30 s on ice. Lysates were centrifuged twice at 4,000 × g for 10 min to remove unbroken cells. An aliquot of 200 μl of each supernatant (serving as total lysate sample) was stored at 4˚C. The remaining supernatants were centrifuged at 16,000 × g for 30 min at 4˚C to pellet cell walls. The supernatants were collected and ultra-centrifuged at 200,000 × g for 2 hrs at 4˚C to separate plasma membranes (pellet) from the cytoplasmic fraction. The cell wall and plasma membrane fractions were each washed twice with PBS supplemented with 1 mM benzamidine and resuspended in 200 μl of PBS containing protease inhibitors. The cytoplasmic fractions were ultra-centrifuged one more time for 2 hrs at 200,000 × g to ensure that all residual membranes were removed. The protein concentration in each fraction was determined using the BSA kit and 50 μg of protein fraction was loaded on a gel for Western blotting using specific antibodies. Mouse anti-EsxU and EsxT sera were collected after DNA immunization (In-Cell-Art, Nantes, France), as previously described [47]. The specificity of the mouse serum (dilution, 1/500) was checked using a crude lysate from E. coli expressing the recombinant His-tagged EsxT or EsxU. A peroxidase-conjugated goat antimouse IgG (H+L, Southern Biotech) was then added (dilution, 1/5000) prior to detection using the peroxidase substrate. Mice anti-GroEL1 antibodies and anti-Ag85 antibodies were used at 1/5000 and 1/40 dilutions, respectively.

Cell infection assays and flow cytometry analysis
Infection of J774.2 macrophages and Acanthamoeba castellanii infection were performed, as previously described [17]. After J774.2 infection in 12-well plates, CCF-4 FRET measurements were performed to study phagosomal rupture, as reported earlier [48].

Phagosomal acidification assay
J774.2 cells were distributed into 24-well plates containing glass coverslips and were infected with M. abscessus expressing mCherry for 3 hrs at 37˚C, 5% CO 2 . Infected cells were treated for 1 hr with amikacin (250 μg/mL) to kill extracellular bacteria and Green LysoTracker (dilution,1/2000) (Thermo Fisher Scientific) was added to each well. Colocalization of cell containing red fluorescent M. abscessus and LysoTracker green were counted under the microscope. The Pearson correlation was determined by analyzing more than 100 infected cells from at least 10 random fields with a Leica SP8-X confocal microscope. Pictures analyses and PCC evaluation were done with Icy and Image J software.

Pore spanning lipid bilayer (NanoSpot assay)
Indium tin oxide (ITO)-coated glass slides were used for the electroformation of giant unilamellar vesicles (GUVs), made of dioleoyl-phosphatidylcholine (DOPC, Avanti Polar Lipids, Alabama, USA, Cat # 850375P) as described before [49,50]. For the preparation of lipid bilayers, a Si/SiO 2 chip-based protocol based on the one described in [51], was previously developed and tested in the laboratory. The chips were prepared "in house" using silicon wafer with precut and pre-etched chips (Micromotive GmbH, Mainz, Germany). 250 μl of SEC buffer (300 mM NaCl, 50 mM HEPES, pH 7.5) was added to each well and the chamber was sonicated for 2 min to remove air bubbles trapped in the cavities. 5 μL of Atto488 carboxy fluorescence dye (ATTO-TEC GmbH, North Rhine-Westphalia, Germany, Cat # AD 488-21) was added to each well (final concentration of 5 nM), followed by 50 μL of previously harvested GUVs. Chambers were centrifuged at 400 x g for 10 min at room temperature (Heraeus Megafuge 1.0, Germany) to allow rupture of GUVs and formation of lipid bilayers on top of the cavities. The chamber was placed in a dark camera kept at 37˚C connected to a fluorescence inverted microscope (MORE integrated Life Cell Imaging System, Thermo Fisher Scientific FEI Munich formerly Till Photonics, Planegg, Germany) equipped with a 10×/0.45 objective (planapochromat 420640−9900−000, Zeiss, Jena, Germany), a FITC filterset (ET480_40x/ET535_50m/ T510LPXRXT, Chroma, Rockingham, USA) and LED (transmitted light) and oligochrome (fluorescence) light sources (Thermo Fisher Scientific). 16-bit images were taken with a Andor clara low light imaging interline CCD camera (Andor Technology plc, Belfast, Northern Ireland) using Live Acquisition Software (LAS V. 2.2.2, Thermo Fisher Scientific). 20 μL of EsxU/ EsxT was added to the well to initiate the experiment. Membrane permeabilization was detected in terms of fluorescence dye-efflux. In some experiments, 10 μL of 10 mg/mL Tex-asRed-labeled dextran molecules (Life technologies, Carlsbad, CA, USA) of 40 000 MW were added before the complex.
Image analysis (drift correction, identification of fluorescent areas, and extraction of time series) was done with ImageJ 1.52p (USA). Final data analysis and graphs were done with GraphPad Prism software version 9.0.

Mice infection and bacterial load determination
C3HeB/FeJ mice were infected intravenously (i.v.) in the tail vein with 1 × 10 8 CFU/mouse either with the wild-type or ΔesxUT strains [26]. Bacterial burden in the lungs, liver, spleen and kidneys was evaluated on day 1, 7, 14 and 20. Five mice per group were used, except on day 1 (four mice) and day 20 (two mice due to premature death). Differences between means were analyzed by two-way analysis of variance (ANOVA) and the Tukey's post-test, allowing for multiple comparisons. n.s., non-significant; � , P<0.05; �� , P< 0.01; ��� , P< 0.001. Experiment was performed twice.

Zebrafish experiments
Zebrafish infections were completed as previously described [25,52]. The golden mutant zebrafish line and the Tg(mpeg:mCherry) red fluorescent macrophage line were used as previously described. Embryos at 30 hrs post-fertilisation were dechorionated and anaesthetised in tricaine prior to caudal vein injection with 2-3 nL of M. abscessus strains (~100 CFU/nL) expressing Tdtomato fluorescent protein under the control of pTEC27 plasmid. After infection, embryos were recovered in fresh embryo water and placed into 24-well plates with 2 embryos/well. Embryos were monitored daily, with dead embryos removed from the well daily until 12 days post-infection. At 2, 4 and 6 days post-infection, embryos were anaesthetised in tricaine and placed on 3% methylcellulose (w/v) for fluorescent microscopy using Zeiss Axio Zoom.V16 coupled with an Axiocam 503 mono (Zeiss). Granulomas were quantified based on the bacterial co-localisation with macrophage fluorescent signal. Bacterial burden was quantified in ImageJ using the 'Analyze particles' function to quantify bacterial fluorescence. Abscesses were quantified based on the large uncontrolled extracellular growth of fluorescent bacteria that exceeded 100 μm in diameter. All experiments were performed in at least three independent experiments.

EsxU/EsxT ELISA
A total of 173 sera from CF patients, including 33 patients with M. abscessus positive cultures, and 31 non-CF healthy controls were analyzed. The distribution of the different groups depending on the results of sputum cultures were previously described against the PLC protein [53]. IgG levels were evaluated by indirect ELISA assays using a EsxU/EsxT recombinant protein complex or the PLC protein with sera diluted at 1/400. ELISA assays were developed as previously described [53] with alkaline phosphatase-conjugated goat anti-human IgG (Southern biotechnology, Birmingham, USA) diluted in PBS containing 0.05% (v/v) Tween 20 (PBS-T) and 0.5% BSA (PBS-T-BSA). OD mean ± standard deviation was calculated for each patient group. NTM/Pa represents CF patients without a positive culture for NTM and Pseudomonas aeruginosa. Finally, the HC group represents healthy subjects who are not CF patients. (C) Mean IgG responses (± SD) against M. abscessus PLC in each group, as previously reported [53]. P values were determined by unpaired t test; � P < 0.05, �� P < 0.01, ��� P < 0.001, ���� P < 0.0001. (TIF) S1 Table. Primers used in this study. (DOCX)