TBK1‐mediated phosphorylation of LC3C and GABARAP‐L2 controls autophagosome shedding by ATG4 protease

Abstract Autophagy is a highly conserved catabolic process through which defective or otherwise harmful cellular components are targeted for degradation via the lysosomal route. Regulatory pathways, involving post‐translational modifications such as phosphorylation, play a critical role in controlling this tightly orchestrated process. Here, we demonstrate that TBK1 regulates autophagy by phosphorylating autophagy modifiers LC3C and GABARAP‐L2 on surface‐exposed serine residues (LC3C S93 and S96; GABARAP‐L2 S87 and S88). This phosphorylation event impedes their binding to the processing enzyme ATG4 by destabilizing the complex. Phosphorylated LC3C/GABARAP‐L2 cannot be removed from liposomes by ATG4 and are thus protected from ATG4‐mediated premature removal from nascent autophagosomes. This ensures a steady coat of lipidated LC3C/GABARAP‐L2 throughout the early steps in autophagosome formation and aids in maintaining a unidirectional flow of the autophagosome to the lysosome. Taken together, we present a new regulatory mechanism of autophagy, which influences the conjugation and de‐conjugation of LC3C and GABARAP‐L2 to autophagosomes by TBK1‐mediated phosphorylation.


131 TBK1 phosphorylates LC3C and GABARAP-L2 in vitro 132
The serine-threonine kinase TBK1 has previously been shown to phosphorylate 133 autophagy receptors such as OPTN and p62 (Pilli et  directly phosphorylated by TBK1 in vitro ( Figure 1A). The S/T phosphorylation 138 sites of LC3-family proteins were identified by mass spectrometry with significant 139 PEP scores ( Figure 1B) and we decided to further investigate the TBK1-mediated 140 phosphorylation sites of LC3C at S93 and S96 and GABARAP-L2 at S87 and S88 141 in detail. The TBK1-mediated phosphorylation sites of LC3C ( Figure 1C) and 142 GABARAP-L2 ( Figure 1D) are topologically equivalent and are present in surface 143 exposed loops (depicted in red). This loop is on the opposite face of the LIR binding 144 pocket indicating that LIR-mediated interactions of LC3C might not be affected di-145 rectly upon phosphorylation. 146 147

TBK1 phosphorylates and binds LC3C and GABARAP-L2 in cells 148
To test if TBK1 also phosphorylates LC3C in cells, HEK293T cells were SILAC 149 labeled and either WT TBK1 (heavy label) or TBK1 kinase dead (K38A; light label) 150 were overexpressed along with GFP-LC3C or with GFP-GABARAP-L2. GFP-pro-151 teins were immunoprecipitated and analyzed by mass spectrometry. Phosphory-152 lation at positions S96 and S93 of LC3C was enhanced in the presence of WT 153 TBK1 (factors 6 and 10 respectively; Figure 2A), as compared to TBK1 K38A. 154 Similarly, the presence of WT TBK1 resulted in enhanced phosphorylation of 155 GABARAP-L2 at S87 and S88 (by factors 13 and 2, respectively; Figure 2A).  fortunately, our efforts to generate phospho-specific antibodies against GABA-157 RAP-L2 S87-PO4 and GABARAP-L2 S87/88-PO4 failed ( Figure S1). To confirm 158 the phosphorylation event directly, we visualized it by using phos-tag TM polyacryla-159 mide gels, where phosphorylated proteins are retained by the phos-tag reagent 160 and appear at a higher molecular weight. Overexpression of WT TBK1, but not 161 TBK1 kinase dead, induced an upward shift and retention of phosphorylated LC3C 162 ( Figure 2B). The ratio of phosphorylated to unphosphorylated LC3C is also higher 163 upon the induction of mitophagy, by the addition of Parkin, an E3 ligase and CCCP, 164 the mitochondrial depolarization agent (compare lanes two and five of phosphory-165 lated upper HA-band from phos-tag gel, Figure 2B). Moreover, the endogenous 166 TBK1 from HeLa or HEK293T cell lysate binds to GST-LC3C and GST-GABARAP-167 L2 ( Figure 2C) indicating direct physical interaction. The binding of TBK1 to LC3C 168 and GABARAP-L2 is independent of its catalytic activity ( Figure 2D) and could be 169 mediated through its C-terminal coiled-coil region (Figure 2D), which is known to 170 bind OPTN (Freischmidt et al., 2015).

172
Phospho-mimetic LC3C impedes ATG4 cleavage and binding 173 To understand the consequences of phosphorylated-LC3C, we looked at its phos-174 pho-sites in more detail. The LC3C phosphorylation sites S93 and S96 are situated 175 on the face opposite to the hydrophobic pocket enabling LIR binding ( Figure 1C) 176 and are therefore less likely to influence the direct binding of LC3C to autophagy 177 receptors or adaptors. However, they are in close proximity to the C-terminal tail 178 of LC3C which is proteolytically processed. ATG4 mediated processing of the 179 LC3C C-terminal tail allows lipid-conjugation and adherence to autophagosomes. 180 To test if phosphorylation of these residues could impair the proteolytic cleavage 181 of the LC3C C-terminal tail by ATG4, an in vitro cleavage assay was performed. 182 Double-tagged LC3C WT, S93/96A or phospho-mimetic LC3C S93/96D were in-183 cubated with ATG4B for indicated times and the C-terminal cleavage of LC3C was 184 monitored by detecting the appearance of truncated His-LC3C protein ( Figure 3A).

185
ATG4B cleaves the entire pool of LC3C WT or S93/96A within 10 minutes, 186 whereas only half of the phospho-mimetic LC3C S93/96D pool is cleaved ( Figure  187 3A). When LC3 proteins are overexpressed in HEK293T cells, they are rapidly 188 processed by endogenous ATG4 proteins. The C-terminal tail of LC3C that is 189 cleaved by ATG4s is considerably larger (21 residues) than that of other LC3 family 190 proteins. Hence, a pro-form of LC3C S93/96D could be visualized by separating 191 cell lysate on a 15% polyacrylamide gel ( Figure 3B).

192
The inability of ATG4 to process phosphorylated LC3C might be distinct during 193 stress conditions. To test this, we induced mitophagy in HEK293T cells by adding 194 CCCP. Upon induction of mitophagy LC3C S93/96D could not be completely pro-195 cessed by ATG4s ( Figure 3C). Similarly, GABARAP-L2 phospho-mimetic (S88D) 196 could not be cleaved by endogenous ATG4s, impairing subsequent lipidation (Fig-197 ure 3D). We reasoned that this inability of ATG4 to process LC3C S93/96D and 198 GABARAP-L2 S87/88D could be due to an impediment in direct protein binding, 199 and therefore tested this by co-expressing GFP-LC3 proteins with either Flag-200 ATG4A or Flag-ATG4B in HEK293T cells and subjected them to GFP-immunopre-201 cipitation ( Figure 3E). Phospho-mimetic mutants LC3C S93/96D and GABARAP-202 L2 S87/88D displayed reduced binding to ATG4A and B. To understand this re-203 duced binding, we modelled the full-length LC3C-ATG4B complex based on the 204 core crystal structure (Satoo et al., 2009) (see Methods). We tested the effect of 205 phosphorylation at both these sites (S93 and S96) by modeling phosphate groups 206 onto serine residues in the LC3C-ATG4B complex and performed molecular dy-207 namics (MD) simulations (up to 1.5 μs). We found that the WT LC3C-ATG4B com-208 plex with and without additional LIR interactions between ATG4B and LC3C re-209 mained stable. The C-terminal tail of LC3C remained bound and strongly anchored 210 to the active site of ATG4B throughout the simulation. In the complex, the phospho-211 sites S93 and S96 of LC3C (red cartoon in Figure 3F) are in close-proximity to the 212 ATG4B interacting surface (grey surface). LC3C S96 forms a hydrogen-bond in-213 teraction with ATG4B E350, and LC3C S93 is close to a network of hydrogen 214 bonds and salt bridges stabilizing. In MD simulations, double phosphorylation of 215 S93 and S96 interfered with these interactions and disrupted the binding interface 216 between ATG4B and LC3C (Movies SM1, SM2). The phosphorylated serine resi-217 dues detached from the ATG4B surface and partially dislodged the LC3C, resulting 218 in partial retraction of the LC3C C-terminal tail from the ATG4B active site. The 219 negative charge introduced by phosphorylation severely weakens complex stabil-220 ity based on calculated binding energies (Table S2), with electrostatic interactions 221 as the dominant factor. Figure S2 shows residue-wise contributions to the binding 222 energy mapped onto the LC3C structure. According to these calculations, phos-223 phorylated S93 and S96 are strongly destabilizing ( Figure S2; red thick cartoon), 224 whereas unphosphorylated S93 and S96 are favorable ( Figure S2; blue thin car-225 toon). The MD simulations and binding energy calculations indicate that phosphor-226 ylation disrupts the LC3C-ATG4B interface and destabilizes the complex.

228
Phosphorylation at S93 and S96 affects LC3C C-terminal tail structure and 229 thereby impedes ATG4-mediated cleavage 230 Based on the simulation results for the LC3C-ATG4B complex, we hypothesized 231 that phosphorylation of unbound LC3C could affect its C-terminal tail structure and 232 prevent binding to the ATG4B active site. In MD simulations (see Methods) of free 233 LC3C, we found that the C-terminal tail of LC3C (126-147) was disordered and 234 highly dynamic (Movie SM3). By contrast, in the phosphorylated variants (S93-235 PO4 LC3C and S96-PO4 LC3), the C-terminal tail adopted more ordered confor-236 mations (Movies SM4, SM5; Figure 4A-B). The phosphoserines formed intramo-237 lecular salt bridges with R134 ( Figures 4A and 4B) that pulled the C-terminal tail 238 of LC3C towards the protein, structuring it locally. In repeated simulations (n = 6 239 each) of unphosphorylated and phosphorylated variants of LC3C ( Figures S3A-240 C), we observed a total of six salt-bridge formation events, indicating that the in-241 tramolecular salt-bridge formation between the phosphoserines and R134 is ro-242 bust. We observed the salt bridge formation on a sub-microsecond time scale (Fig-243 ure 4D and 4E). To confirm this finding and the role of R134, we performed ATG4-244 mediated in vitro cleavage experiments of double tagged LC3C WT, S93/96D, 245 S93/96D R134A, and S93/96D R142A (a control mutation in the C-terminal tail).

246
The LC3C C-terminal cleavage was monitored by the disappearance of its C-ter-247 minal Strep-tag. The mutation of S93/96D delayed the cleavage of the C-terminal 248 tail of LC3C by ATG4B ( Figure 4F). The R134A mutation could partially rescue 249 this phenotype of S93/96D, whereas the other C-terminal tail mutation, R142A, 250 could not ( Figure 4F). The results of the ATG4-mediated cleavage assay are thus 251 consistent with R134-phosphoserine interactions sequestering the LC3C C-termi-252 nal tail and preventing access to ATG4B and subsequent cleavage.

330
On the contrary, we found that GABARAP-L2 S87/88D is not a target of ATG4A or 331 B (Figure 7A,B). 332 In addition, we also tested whether the phosphorylation of GABARAP-L2 has an 333 impact on its de-lipidation from the phagophore by other proteases such as RavZ 334 ( Figure 7C). RavZ is a bacterial effector protein from the intracellular pathogen 335 Legionella pneumophila that interferes with autophagy by directly and irreversibly indicating its effectiveness in circumventing Legionella growth restriction via xe-340 nophagy (when TBK1 is also activated). Likewise, RavZ is also able to cleave 341 LC3C WT and S93/96D mutant from liposomes in vitro ( Figure S6A).

342
Finally, we tested if phosphorylated LC3C or GABARAP-L2 adhered to autopha-343 gosomes are still functional to perform downstream reactions. LC3 family proteins 344 interact with autophagosome receptors such as p62, which link the growing au- Phosphorylation of GABARAP-L2 by TBK1 does not interfere with its ability to bind 354 to PLEKHM1 ( Figure 7E). 355 Hence, TBK1 mediated phosphorylation of GABARAP-L2 and LC3C protects them 356 from premature autophagosome removal by ATG4, but does not interfere with 357 downstream reactions like cargo binding and lysosomal fusion.

360
The autophagy pathway is tightly regulated to ensure proper recycling and disposal 361 of cellular material during nutrient shortage. Here, we present a new regulatory 362 mechanism of autophagy, which influences the conjugation and de-conjugation of 363 LC3C and GABARAP-L2 to autophagosomes. The kinase TBK1 fulfils several 364 roles during selective autophagy. Upon autophagy induction, TBK1 is recruited to 365 the site of autophagosome formation and gets activated by trans-autophosphory- ing and atomistic simulations of the ATG4B-LC3C complex revealed that LC3C 393 phosphorylation impedes binding to ATG4. The weakened binding slows down de-394 lipidation, which ensures that a steady coat of lipidated LC3C/GABARAP-L2 is 395 maintained throughout the early steps in autophagosome formation (Figure 8).

396
The phosphorylation of LC3C/GABARAP-L2 does not impede their binding to au-397 tophagy receptors such as p62 or PLEKHM1, which promotes unhindered down-398 stream steps for, e.g., autophagosome-lysosome fusion (McEwan et al., 2015). 399 Thus, phosphorylation of LC3s aids in maintaining an unperturbed and unidirec-400 tional flow of the autophagosome to the lysosome. 401 At later stages of autophagosome formation, this process could be slowed-down 402 or reversed by either TBK1 dissociation from autophagosomes or diminished cat-403 alytic activity. Alternatively, action of phosphatases could allow de-lipidation prior 404 to autophagosomal-lysosomal fusion, thereby recycling LC3s. 405 406

Materials and Methods 407
Expression constructs 408 Expression constructs of indicated proteins were cloned into indicated vectors us-409 ing PCR or the gateway system. restrained NVT equilibration and NPT equilibration runs for 1000 ps each. Produc-570 tion runs at 310 K and 1 atm were simulated for different times (see Table S1). We 571 used the CHARMM36m force field (Huang et al., 2017), the Nosé-Hoover thermo-572 stat (Nosé, 1984), the Parinello-Rahman barostat (Parrinello & Rahman, 1981), 573 and a time step of 2 fs. For each of the LC3C-systems (Table S1), six replicates 574 were simulated with different initial velocities. We also used the molecular mechan- protein-complex and free proteins are decomposed into a sum of molecular-me-591 chanics, solvent, and configurational entropy contributions, 592 " = !2 MM + !" Solv − 9!: 593 " = !2 bonded + !2 vdW + !2 ele + !" polar + !" non-polar − 9!: 594 The binding energies were evaluated at intervals of 10 ns from the 1000-ns MD 595 trajectories and averaged (see Table S2). Double differences between unphos-596 phorylated and phosphorylated complexes minimize systematic errors caused by 597 possible energy-function inaccuracy. For the dynamic LC3C-ATG4B protein com-598 plexes studied here, these calculated free energy differences point to trends, but 599 should not be interpreted in terms of, say, dissociation constants. 600 601 Liposome and proteoliposome preparation 602 All lipids were purchased and dissolved in chloroform from Avanti Polar Lipids (Al-603 abaster, AL). Liposomes were prepared by combining 55 mol % 1,2-dioleoyl-sn-604 glycero-3-phosphoethanolamine (DOPE), 35 mol % 1-palmitoyl-2-oleoyl-sn-glyc-605 ero-3-phosphocholine (POPC), and 10 mol % bovine liver phosphoinositol (PI). 606 The lipids were dried under nitrogen gas and the lipid film was further dried under 607 vacuum for 1 hour. The lipids were reconstituted in NT350 buffer (350 mM NaCl, 608 20 mM Tris-HCl pH 7.4) and subjected to 7 cycles of flash-freezing in liquid nitro-609 gen and thawing in a 37°C bath. Liposomes were further sonicated immediately 610 prior to the lipidation reaction. by Coomassie blue stain) were mixed with NT350 buffer and kept on ice until ac-636 tivity assays were initiated by the addition of 2 µM (or indicated amounts) of either 637 ATG4A, ATG4B or RavZ. Reactions were incubated at 37°C for 1 hour. Samples 638 were mixed with LDS loading buffer and immediately boiled to stop proteolysis. 639 The reactions were run on a 4-20% SDS-PAGE gel and visualized by Coomassie 640 blue stain and analyzed with Image Lab 6.0 (Biorad). 641 642