ESCRT and autophagy cooperate to repair ESX-1-dependent damage to the Mycobacterium-containing vacuole

Phagocytes capture invader microbes within the bactericidal phagosome. Some pathogens subvert killing by damaging and escaping from this compartment. To prevent and fight bacterial escape, cells contain and repair the membrane damage, or finally eliminate the cytosolic escapees. All eukaryotic cells engage highly conserved mechanisms to ensure integrity of membranes in a multitude of physiological and pathological situations, including the Endosomal Sorting Complex Required for Transport (ESCRT) and autophagy machineries. In Dictyostelium discoideum, recruitment of the ESCRT-III protein Snf7/Chmp4/Vps32 and the ATPase Vps4 to sites of membrane repair relies on the ESCRT-I component Tsg101 and occurs in absence of Ca2+. The ESX-1 dependent membrane perforations produced by the pathogen Mycobacterium marinum separately engage both ESCRT and autophagy. In absence of Tsg101, M. marinum escapes earlier to the cytosol, where it is restricted by xenophagy. We propose that ESCRT has an evolutionary conserved function in containing intracellular pathogens in intact compartments.

The cytosolic bacteria were surrounded by a structure (green arrowheads) with very electron-dense boundaries (pink asterisks). The cytosolic material between the bacteria and this structure or the autophagosome was slightly more electron-dense than the rest of the cytosol (blue asterisks) (D). Section of a cell, showing a disrupted MCV (blue), M. marinum (red) and the dark electron-dense material surrounding the sites of escape (blue asterisks, yellow). (E) 3D reconstruction of the FIB-SEM stack shown in D (see also Movie S1). Abbreviations Vps32 patches always surrounded the MCV, but were not in its lumen (Fig. 2F). During membrane discoideum expressing GFP-Tsg101, GFP-Vps32 or GFP-Vps4 were infected with M. marinum wt or M. marinum ΔRD1, and z-stacks were acquired at 1.5, 5, 8, 24 and 31 hpi. Maximum projections are shown. GFP-Tsg101, GFPVps32 and Vps4-GFP structures (white arrows) appeared in the vicinity of the MCV containing M. marinum wt (red), and to a lesser extent around M. marinum ΔRD1 (red). Scale bars, 5 µm. (B-D) Quantification of GFPTsg101 structures (patches and foci) and GFP-Vps32 and Vps4-GFP structures (patches and rings) in the vicinity of M. marinum wt or M. marinum ΔRD1. Plots show the mean and standard deviation (GFP-Tsg101 1.5, 8, 24,31 hpi N=3, 61≤n≤246; 5 hpi N=2, 56≤n≤167, GFP-Vps32 N=3 32≤n≤145; Vps4-GFP N=3, 64≤n≤233). (E) D. discoideum expressing GFP-Vps32 were infected with M. marinum wt (red) and monitored by time-lapse microscopy every 3 min (see also Movie S2). Maximum projections of the same cell are shown (time indicated in the top right corner). GFP-Vps32 rings formed and appeared to move along the bacterium (arrows). Bottom panels show insets focused on the ring structures (arrows). Scale bars, 10 µm and 1 µm for the insets. (E) Section of a z-stack showing the recruitment of GFP-Vps32 to the vicinity of the MCV (bacteria in red) at 31 hpi. Projections of the xz and yz planes are shown. Scale bar, 10 µm. (G) D. discoideum wt or tsg101-expressing GFP-Vps32 were infected with M. marinum (in red) and z-stacks were acquired at 1.5, 8 and 24 hpi. Maximum projections are shown. GFP-Vps32 was recruited to a lesser extent to the MCV in tsg101-cells. Scale bars, 5 µm. (H) Quantification of GFP-Vps32 structures formed in the vicinity of M. marinum in wt or tsg101-cells. The plot shows the mean and standard deviation (N=3, 76≤n≤154). Two-tailed t-tests were performed.  M. marinum in intact MCVs stained by p80 are rarely ubiquitinated, contrary to bacteria in the 171 cytosol (Fig. 3D). Therefore, we wondered whether Vps32 would be recruited at sites of ubiquitination, 172 together with the main autophagy marker Atg8. Remarkably, all three proteins localized at disrupted MCVs, 173 but the level of colocalisation of Vps32 with ubiquitin ( Fig. 3E) or Atg8 (Fig. 3F) was limited. Instead, 174 GFP-Vps32 seemed to be recruited more proximally to the membrane remnants of the MCV than ubiquitin 175 and Atg8, which predominantly decorated the bacteria poles fully exposed to the cytosol. Besides, at the Only bacteria (blue) that have escaped the MCV (p80, red) showed ubiquitin structures (green). Scale bar, 5 µm. (E-F) D. discoideum expressing GFP-Vps32 were infected with M. marinum (blue) and fixed for immunostaining at 8 and 24 hpi to visualize ubiquitin or Atg8 (red). GFP-Vps32 and ubiquitin or Atg8 were recruited to the same macroscopic region of the MCV, but they did not perfectly colocalise. GFP-Vps32 formed patches devoid of ubiquitin or Atg8 staining (white arrows). Vice versa, ubiquitin and Atg8 appeared in areas where no GFP-Vps32 was observed (yellow arrows). Scale bar, 5 µm and 1µm for the insets. (G) Schematic representation of the Vps32, ubiquitin and Atg8 recruitments at the damaged MCV.

Differential spatial and temporal recruitment of ESCRT and autophagy upon sterile damage 183
Mammalian ESCRT and autophagy machineries localize to damaged membranes for the repair of wounds 184 and removal of terminally incapacitated organelles, respectively 23, 24, 34, 38, 39 . To test whether components of 185 both machineries were also involved in membrane repair in D. discoideum, cells expressing GFP-Tsg101, 186 GFP-Vps32 or Vps4-GFP, as well as GFP-Atg8 were subjected to membrane damaging agents, such as the 187 detergent digitonin or the lysosome-disrupting agent Leu-Leu-O-Me (LLOMe) (Fig. 4). Digitonin inserts 188 first into the sterol-rich plasma membrane and then, upon endocytosis, reaches the endosomes. Consistent Atg8 formed a more continuous ring that became apparent only 5 min later. This spatial appearance and   (Fig. 5C-D), providing a strong evidence that Tsg101 lies upstream of ESCRT-III during 225 membrane repair caused by these types of sterile damage. 226 It has been proposed that the local increase of intracellular Ca 2+ upon membrane damage recruits 227 ESCRT-III to the plasma and lysosomal membranes in HeLa cells and myoblasts 23, 24, 34 . To test whether the discoideum wt, tsg101-and atg1-were subjected to the experimental procedure depicted in G and monitored by time-lapse imaging (see also Movie S6). Before addition of LLOMe, HPTS was quenched in acidic lysosomes, which therefore appeared in red. HPTS dequenching started after 30 sec in tsg101-and atg1-cells and after 1.5 min in wt cells.

To decipher the role of ESCRT during infection, cells lacking Tsg101 or the accessory proteins AlxA and 248
the AlxA interactors Alg2a and Alg2b were infected and examined by EM. In wt cells, alxA-or alg2a-/b-249 mutants, the membrane of the MCV in close vicinity to the bacilli escaping the compartment was even and 250 smooth ( Supplementary Fig. 3A, E-G). However, in the tsg101-mutant, rough and "bubbling" membrane 251 structures were observed ( Supplementary Fig. 3B-C), suggesting cumulating membrane damage. In all 252 cases, escaping bacteria were surrounded by the highly electron-dense material already described in Fig. 1. 253 also colocalized more with Atg8 in the tsg101- (Fig. 6C-D and Supplementary Fig. 4D). Although the 260 percentage of ubiquitinated bacteria in tsg101-cells was close to that observed in the atg1-and atg1-tsg101-261 double mutants (84.6 ± 3.3% and 89.3 ± 7.5%, respectively), the extent of ubiquitin decoration on the 262 bacteria was very different (Fig. 6A-B and Supplementary Fig. 4C). Whereas in cells lacking Tsg101 263 ubiquitin formed foci or patches around M. marinum, in cells devoid of autophagy bacteria were more 264 densely coated with ubiquitin ( Fig. 6A and Supplementary Fig. 4C). This accumulation is probably due to 265 the fact that ubiquitinated bacteria cannot be targeted to autophagic degradation in the atg1-mutant, but 266 autophagy is still functional in the tsg101-mutant. 267 Given that both ESCRT-III and autophagy are involved in the biogenesis of MVBs and 268 autophagosomes, respectively, which rely at least partially on the recognition of ubiquitinated cargoes, we 269 monitored the morphology of endosomes, as well as the levels of ubiquitination, in non-infected ESCRT 270 and autophagy mutants ( Supplementary Fig. 4). In the atg1-and atg1-tsg101-mutants accumulation of 271 high levels of ubiquitinated material was observed, in agreement with the inability of these mutants to 272 degrade it by autophagy. In tsg101-cells, only a minor increase of ubiquitin was observed in endosomal 273 compartments ( Supplementary Fig. 4A-B), as already reported 45 , which does not explain the more frequent 274 and larger ubiquitin decorations around M. marinum in these cells (Fig. 6A). 275 In yeast and mammallian cells devoid of some ESCRT proteins, ubiquitinated cargoes are not 276 properly sorted into MVBs and accumulate on the limiting membrane 46 . Therefore, to confirm that the 277 increase in ubiquitination observed during infection of the tsg101-mutant was due to MCV damage and 278 bacteria access to the host cytosol, and not to failed endocytic cargo sorting, we monitored the colocalization marinum by GFP-Plin was higher in tsg101-and atg1-tsg101-compared to wt cells (Fig. S4E-F), 282 confirming the earlier bacteria escape from the MCV in cells lacking a functional ESCRT machinery. 283 Altogether, these results suggest that both Tsg101 and Atg1 trigger separate membrane repair pathways and 284 restrict M. marinum access to the cytosol during infection. 285 Since we have shown that Tsg101 is not essential for ESCRT-III recruitment to the damaged MCV 286 ( Fig. 2G-H), but has an important role in repairing the MCV and constraining bacteria escape ( Fig. 6 and  287 Supplementary Fig. 4), we wondered whether the accessory proteins AlxA and Alg2a/b, also known to 288 recruit ESCRT-III, were involved in the repair of the MCV. In cells lacking Alg2a/b, the percentage of 289 ubiquitinated M. marinum was comparable to that in its respective parental strain (43.3 ± 15.0% and 50.2 ± 290 11.4% respectively, Supplementary Fig. 5A-B and E) and, similarly, the degree of Atg8 colocalization with 291 the bacteria remained lower in the alg2a-/b-mutant (54.70 ± 9.5%, Supplementary Fig. 5C-D and F). On 292 the contrary, 81.8 ± 9.9% of bacteria were ubiquinated in cells devoid of AlxA, which correlated with a 293 higher but not significant increase of Atg8 recruitment to the bacteria (72.7 ± 13.0%, Supplementary Fig.  294 5A-F). This suggests that AlxA, together with Tsg101 but not Alg2a/b, contributes to the ESCRT-III-295 mediated repair of the MCV. 296 297

Impairment of ESCRT or autophagy has a distinct impact on M. marinum intracellular growth 298
To study how the ESCRT pathway may impact the outcome of M. marinum infection, the ESCRT mutants 299 were infected with luminescent M. marinum 48 and intracellular bacterial growth monitored 49 (Fig. 6E-F and  300 Supplementary Fig. 5G-H). M. marinum luminescence increased around 5-fold in wt D. discoideum in the 301 course of 72 h, reflecting sustained intracellular growth. In the atg1-mutant, since bacteria escape earlier to 302 a cytosol that is devoid of xenophagic defense, M. marinum grew better (Fig. 6F), as already described 20 . suggest that in D. discoideum autophagy and ESCRT-III work in parallel to repair the damaged MCV. It 374 has been shown that ubiquitin serves as an "eat-me" signal that targets cytosolic bacteria to autophagic 375 degradation 2 . Consistent with this, the accumulation of ubiquitin around M. marinum in tsg101-cells 376 correlated with a proportional increase of Atg8 decoration on the bacteria (Fig. 6C-D), and with a decreased 377 bacterial load (Fig. 6E), contrary to what has been described in RAW macrophages infected with the non-378 pathogenic, vacuolar M. smegmatis, in which depletion of Tsg101 led to bacteria hyperproliferation 25 . In 379 addition, only very limited colocalization between GFP-Vps32 and ubiquitin or Atg8 was observed at 380 damaged MCVs (Fig. 3E-F). 381 The cues and signals that recruit ESCRT-III to damage in D. discoideum are still to be identified. 382 The appearance of ESCRT-III components before ubiquitinated material can be detected at the site of 383 Samples were incubated in 2% (v/v) of Milloning buffer and rinsed with distilled water. Then, they were 477 incubated in 0.25% (w/v) uranyl acetate overnight and rinsed with distilled water. Samples were dehydrated 478 using increasing concentrations of ethanol, then in propylene oxide for 10 min and finally embedded in 50% 479 Epon-propylene oxide for 1h, followed by incubation overnight in pure Epon. Samples were embedded in 480 2% agar for subsequent sectioning in an ultramicrotome and placed on TEM grids. Finally, sections were 481 visualized in a Tecnai 20 electron microscope (FEI Company, Eindhoven, The Netherlands). 482 483

Focus Ion Beam Scanning Electron Microscopy (FIB-SEM) 484
Initial sample preparation was performed similarly as for TEM and sent to the Pôle Facultaire de 485 Microscopie Ultrastructurale (University of Geneva). Subsequent contrast enhancement, dehydration and 486 resin embedding was performed as described in 65 Samples were visualized in a Helios DualBeam NanoLab 487