Non-canonical activation of the ER stress sensor ATF6 by Legionella pneumophila effectors

Legionella pneumophila secretes toxins into the host cell that induce the non-canonical processing and activation of the ER stress sensor and transcription factor ATF6 via a mechanism that is distinct from the canonical pathway activated by unfolded protein buildup.


Introduction 20
Several intracellular pathogens, including Legionella pneumophila (L.p Once deposited into the cytosol, the effectors target a vast array of host proteins and can influence 25 diverse biological processes which permit the use of L.p. as a tool to uncover novel biological 26 mechanisms. During infection, L.p. uses its effectors to prevent fusion of the Legionella-containing 27 vacuole (LCV) with the host endosomal machinery. Instead these effectors facilitate the 28 remodeling of the LCV into a compartment that supports pathogen replication ( The ER-LCV interactions take on different forms as LCV maturation progresses. L.p. induces 34 tubular ER rearrangements and intercepts ER-derived vesicles destined for the Golgi early in 35 infection (Kotewicz, 2017). However, the mature LCV is studded with ribosomes and reticular ER 36 proteins presenting a vacuolar environment that is substantially different and highlights the 37 complexity of interactions between the ER and LCV. The ER serves as a critical regulatory site 38 for protein and membrane lipid biosynthesis, and imbalances in protein load or membrane lipid 39 perturbations can disrupt many of its vital homeostatic functions (Rapoport, 2007, Halbleib, 2017. 40 The unfolded protein response (UPR) serves as a prominent regulatory pathway in the ER that 41 has been shown to respond to the burden of accumulating unfolded or misfolded proteins in the 42 ER (Ron & Walter, 2007). In mammalian cells, the UPR is coordinated by three ER-localized 43 transmembrane proteins, inositol-requiring protein-1 (IRE1), protein kinase RNA (PKR)-like ER 44 kinase (PERK) and activating transcription factor-6 (ATF6), each of which initiate pathways 45 designed to modulate the cellular response (Cox, Shamu et al., 1993, Lee,Tirasophon et al.,46 2002, Mori, Ma et al., 1993). 47 ATF6 is a type II transmembrane protein that is retained in the ER under normal homeostatic 48 conditions through interactions with the resident chaperone BiP/GRP78 (Shen, Chen et al., 2002). 49 Upon accumulation of unfolded proteins, the ER stress stimulates ATF6 translocation from the 50 ER to the Golgi. At the Golgi, ATF6 is sequentially cleaved first by site-1 protease (S1P) in the 51 luminal domain, then by site-2 protease (S2P) liberating the cytosolic ATF6-N terminal fragment, 52 ATF6(NT) (Shen & Prywes, 2004, Ye, Rawson et al., 2000. Once cleaved, ATF6(NT) is recruited 53 to the nucleus where it binds to cis-acting ER stress response elements (ERSE) in the promoter 54 region of UPR target genes (Kokame, Kato et al., 2001, Yoshida, Okada et al., 2000. ATF6 55 activation is thought to facilitate cytoprotective adaptation to ER stress through the regulation of 56 genes that improve protein folding and processing in the ER. ATF6 has been shown to suppress 57 the UPR induced apoptotic program once the stress is resolved (Wu et al., 2007), highlighting the 58 pro-survival contributions of this signaling network. 59

Studies emphasizing cross talk between the UPR and bacterial infection have revealed an 60
interconnectedness of ER stress sensing and pathogen sensing mechanisms in the cell ( processing, we sought to understand the mechanism by which L.p. modulates the ATF6 pathway. 72 Here, we present evidence of a unique, non-canonical mode of ATF6 activation by L.p. that does 73 not rely on host proteins, that were previously thought to be essential for ATF6 activation. 74 Surprisingly, we discover novel L.p. effectors that play a role in activation of ATF6 during infection. 75

Results 76
Proteasome-dependent degradation pathways are not required for ATF6 loss during L.p. infection 77 Upon infection with wild type L.p. (WT-L.p.), we observed near complete processing of 78 endogenous full length ATF6 ( Figure 1A). The level of processing was similar to that induced by 79 the strong reducing agent and non-specific ER stress inducer, dithiothreitol (DTT) ( Figure 1A). An 80 isogenic strain of L.p. that lacks a functional secretion system (ΔdotA-L.p.) was unable to 81 downregulate ATF6 protein levels, suggesting that one or more bacterial effectors could be 82 responsible for its targeting ( Figure 1A). Processing of ATF6 through regulated intramembrane 83 proteolysis (RIP) by the S1P and S2P proteases has been studied extensively (Okada, 2003 with the proteasome inhibitor MG-132 or control media. ER stress induction using DTT led to rapid 94 ATF6 processing after 1 hour, whereas prolonged exposure to DTT for 3 hours resulted in 95 recovery of ATF6 signal due to autoregulatory feedback from UPR induction ( Figure 1B). In 96 contrast to UPR induction, cells treated with protein synthesis inhibitor cycloheximide (CHX) 97 showed loss of full length ATF6 signal after 3 hours post-treatment ( Figure 1B). While CHX 98 treatment alone resulted in reduced levels of ATF6, pre-treatment with MG-132 stabilized ATF6 99 in the presence of CHX to pre-treatment levels ( Figure 1B), consistent with previously described 100 observations (Haze et al., 1999). We next tested whether proteasomal inhibition could stabilize 101 ATF6 protein levels in cells infected with WT-L.p. or ∆dotA-L.p. strains ( Figure 1C). Similar to

104
infection levels and treatment with MG-132 did not have a significant impact ( Figure 1C). Previous 105 studies have identified ATF6 as an ER-associated degradation (ERAD) substrate that undergoes 106 constitutive degradation mediated by SEL1L (Horimoto et al., 2013). It was shown that ATF6 is a 107 short-lived protein with a half-life less than 2 hours and the stability of ATF6 can be markedly 108 increased by Sel1L knock-out (Horimoto et al., 2013). To test if Sel1L dependent ERAD 109 contributes to loss of ATF6 during infection, we next compared ATF6 processing in HEK293-FcγR 110 cells that were treated with non-targeting or SEL1L-targeting siRNA. SEL1L knock-down led to 111 an increase in ATF6(FL) levels by 1.5-fold in samples not treated with CHX ( Figure 1D). Similarly, 112 while CHX treatment for 2 hours caused a reduction in ATF6(FL) levels in non-targeting siRNA 113 cells, ATF6(FL) levels were again increased by 1.5-fold in SEL1L knock-down cells ( Figure 1D). 114 However, there was no significant increase in ATF6(FL) signal intensity under L.p. infection when 115 normalized to loading controls ( Figure 1D). with WT-L.p. ( Figure 1E). Together, these experiments suggest ATF6 processing during L.p. 123 infection is not a direct result of enhanced proteasomal degradation or a consequence of protein 124 synthesis arrest. 125

L. p. induces ATF6 mediated gene induction 126
We next considered how ATF6 processing affected its distal function as a nuclear transcription 127 factor. To evaluate transcriptional activation, we used RT-qPCR to compare mRNA levels of ATF6 128 target genes, including UPR regulator/ER chaperone BiP (HSPA5), in HEK293-FcγR cells under 129 conditions of ER stress and L.p. infection. UPR induction with DTT increased expression of ER 130 quality control genes BiP and HERPUD1 by greater than 5-fold compared to control (DMSO-131 treated) cells ( Figure 2A). Interestingly, these ATF6-regulated genes were also induced in wild-132 type infected cells by greater than five-fold. When compared to protein synthesis inhibition using 133 CHX, we found CHX did not induce BiP gene expression ( Figure 2B). To gain more insight into 134 ATF6 induction patterns during L.p. infection, we examined the gene activation profile of BiP over 135 the course of infection. Analysis of BiP mRNA expression using RT-qPCR indicated a spike in 136 BiP expression between 4-5 hours post DTT treatment ( Figure 2C). When the expression profile 137 was examined under avirulent ∆dotA-L.p. infections, BiP expression spiked between 1-3 hours ATF6 processing was monitored over the same time period. As shown previously, ATF6 is rapidly 142 cleaved under DTT treatment (30 minutes), but the ATF6(FL) signal recovers after sustained 143 stress due to autoregulatory feedback ( Figure 2D). In comparison, ∆dotA-L.p. infection did not 144 induce significant changes in ATF6(FL) levels over the course of infection ( Figure 2D). Yet, the 145 appearance of a cleavage product at 1 hour correlated with the spike in BiP mRNA expression 146 ( Figure 2D). When monitoring ATF6 processing under wild-type L.p. infection, the ATF6 147 processing profile also correlated with the changes in BiP mRNA expression. ATF6 signal 148 intensity at 1-hour post-infection was approximately 90% of pre-infection levels ( Figure 2E). Within 149 3 hours of infection, the ATF6 signal intensity had decreased by roughly 50%. By 5 hours post-150 infection, BiP mRNA levels peaked and the ATF6 signal intensity dropped below 25% of pre-151 infection levels. In contrast to pharmacological ER stress induction, the loss of ATF6(FL) signal 152 persisted throughout the infection. Taken together, the gene expression changes and ATF6 153 processing analysis demonstrate activation of the ATF6 pathway during L.p. infection. It is 154 possible that L.p. could induce ATF6 downstream gene activation via effector(s) that don't require 155 ATF6 cleavage. We thus determined whether ATF6 itself was required for UPR gene induction 156 during infection. The ATF6 gene was targeted for knockdown in HEK293-FcγR cells using siRNA 157 achieving greater than 80% knockdown efficiency ( Figure 2F). Though ATF6 knockdown nearly 158 ablated BiP induction using DTT, ATF6 knockdown under L.p. infection markedly reduced BiP 159 mRNA induction, suggesting that the endogenous ATF6 cleavage is indeed being utilized to 160 induce downstream gene activation. However, BiP mRNA levels were still elevated nearly 4-fold 161 compared to chemical induction in the knockdown cells ( Figure 2G). The residual BiP induction 162 in WT-L.p. infection could be the result of the remnant ATF6 produced from incomplete 163 knockdown. An alternate scenario is that certain L. Walter, 2016). Treatment of cells with DTT resulted in processing of ATF6 ( Figure 3A) and a 175 greater than 5-fold induction of BiP mRNA ( Figure 3B). In contrast, co-treatment of cells with DTT 176 and Ceapin A7 (+A7) partially protected ATF6 from cleavage ( Figure 3A) and reduced gene 177 activation of BiP by nearly 50%. As shown earlier, WT-L.p. infection alone for 6 hours leads to 178 complete processing of ATF6 and strong induction of BiP mRNA. Yet surprisingly, pre-treatment 179 with Ceapin A7 did not prevent processing of ATF6 by the WT-L.p. strain ( Figure 3A and 3B). 180 Additionally, BiP mRNA persisted at levels similar to untreated infections. This result suggests 181 that ATF6 translocation from the ER to the Golgi is not a pre-requisite step required for its 182 processing during infection. L.p. is known to exploit ER-to-Golgi trafficking and individual effectors 183 have been identified that can disrupt Golgi homeostasis (Mukherjee, Liu et al., 2011). As disrupted 184 homeostasis could result in mis-localization of Golgi proteins, the impact of direct inhibition of S1P 185 on ATF6 activation was tested using the inhibitor PF-429242 (Lebeau, Byun et al., 2018). 186 Inhibition of S1P proteolysis activity using PF-429242 (PF) in the presence of DTT resulted in the 187 appearance of a slower migrating species of ATF6 at a higher molecular weight likely due to 188 extensive glycosylation in the Golgi ( Figure 3A). Further validating S1P inhibition, BiP mRNA 189 induction was reduced by nearly 80% compared to DTT treatment alone ( Figure 3B). Remarkably, 190 S1P inhibition did not alter ATF6 processing in WT-L.p. infected cells, and BiP mRNA was induced 191 to similar levels as in untreated infected cells ( Figure 3A and Figure 3B). The cleavage of ATF6 192 during L.p. infection even in the presence of Ceapin A7 and S1P inhibition are suggestive of an 193 alternative proteolytic mechanism induced by L.p.. Because the S1P protease was not required 194 for L.p. induced ATF6 processing, we next tested whether the canonical cleavage sites were a 195 prerequisite for cleavage. To test this, a construct harboring a point mutation at the ATF6 S1P for Golgi translocation further corroborates our S1P mutants and Ceapin A7 studies that showed 219 that ATF6 processing does not occur at the Golgi during infection. As infection progresses, the 220 LCV is remodeled from a plasma membrane derived vacuole to an ER-like compartment in a 221 process that involves the recruitment of host ER proteins to the LCV and the disruption of ER-to-222 Golgi trafficking (Swanson, 1995, Tinley, 2001. When monitoring ATF6 localization, confocal 223 microscopy revealed substantial recruitment of ATF6 to the LCV membrane at 6 hours post-224 infection with over 80% of LCVs marked positive for ATF6 ( Figure 4E). Whereas a majority of 225 LCVs were marked positive for ATF6 in WT-L.p., the recruitment required a functional Dot/Icm 226 system as the ∆dotA-L.p. infected cells exhibited ATF6 recruitment to approximately 30% of 227 LCVs. These data support the hypothesis that ATF6 is activated through a non-canonical 228 mechanism. 229

ATF6 activation is strain and species specific 230
The data presented here highlights a T4SS-dependent activation strategy requiring the 231 translocation of Legionella effector proteins. Therefore, we sought to identify effector proteins 232 capable of inducing ATF6 activation by identifying Legionella strains that fail to efficiently process 233 ATF6. Genomic analysis of over 30 Legionella strains and species revealed largely non-234 overlapping effector repertoires (Burstein, Amaro et al., 2016); therefore, we sought to use a 235 comparative approach to test for ATF6 processing in different Legionella strains and species. thapsigargin (Tg) treatment produced a greater than 6-fold induction of luciferase activity ( Figure  257 5F, red). Control vectors expressing WT-ATF6 and S1P/S2P-ATF6-null were also evaluated in 258 the luciferase cell line. As reported previously (Ye et al., 2000), overexpression on ATF6 led to an 259 increased basal level of activation producing a 4-fold increase in luciferase activity over control 260 vector, whereas the cleavage deficient ATF6 mutant did not lead to an increase in baseline 261 luciferase activity. This data excludes the possibility that simply overexpressing an ER protein 262 that trigger ATF6 expression. Our assay revealed that expression of 11 of the 17 effectors did not 263 increase luciferase activity as compared to the control vector. While, L.p. Philadelphia str. 264 effectors lpg1948, lpg2523, and lpg2525 stimulated a 2-fold increase in luciferase activity, the 265 effectors lpg0519 and lpg2131 produced a greater than 2-fold increase in luciferase activity when 266 expressed individually ( Figure 5F). Together, these results indicated that multiple effectors 267 possess the capacity to induce the ATF6 pathway. Most of the effectors identified in this screen 268 had little or no known function assigned to them. 269

Lpg0519 localizes to the ER and activates ATF6 270
To experimentally validate the results from the screen, the ATF6 targeting effectors were 271 transiently expressed in HEK293T cells and UPR activation was monitored by immunoblot. Upon 272 Tg treatment, a reduction in endogenous ATF6 was observed by 2 hours ( Figure 6A). By 4 hours 273 of Tg treatment, the level of endogenous ATF6 was restored ( Figure 6A). Downstream of ATF6, 274 UPR induction through Tg treatment also resulted in elevated levels of BiP/GRP78 compared to 275 untreated cells. More broadly, Tg treatment also stimulated PERK pathway activation (a 276 transmembrane UPR sensor at the ER) as shown by elevated ATF4 levels after 4 hours. When 277 compared to Tg treatment, cells transfected with the GFP control vector did not exhibit a reduction 278 in ATF6 levels, and BiP and ATF4 levels were not increased compared to untreated cells ( Figure  279 6A and 6B). However, cells transfected with C-terminally tagged GFP-Lpg0519 exhibited a 280 dramatic reduction in endogenous ATF6 levels ( Figure 6A). Further validating ATF6 activation, 281 BiP levels were also elevated in GFP-Lpg0519 transfected cells ( Figure 6B). However, the levels 282 of ATF4 were not elevated in Lpg0519 transfected cells ( Figure 6A). These results highlight the 283 UPR pathway specificity by Lpg0519 in activating the ATF6 pathway without targeting the UPR 284 more generally. To further characterize Lpg0519, we investigated its subcellular localization in 285 mammalian cells. GFP-Lpg0519 was transiently transfected in U2OS cells, and then examined in 286 live cells using confocal microscopy. Whereas the GFP control vector was uniformly distributed 287 throughout the cell ( Figure 6C), Lpg0519 co-localized with the ER-marker mCherry-ER-3 ( Figure  288 6C). Altogether, these data suggest that Lpg0519 localizes to the ER and has the capacity to 289 specifically induce the cytoprotective branch of UPR (ATF6), without affecting the apoptotic 290 branch (PERK). 291

Discussion 292
The UPR represents a critical node for re-establishing ER homeostasis under perturbations 293 caused by the accumulation of misfolded proteins. Importantly, the three branches of the UPR 294 work synergistically to maximize the response, yet the modalities by which each sensor 295 contributes to homeostatic restoration differ considerably (Ron & Walter, 2007). Of importance, 296 the PERK and IRE1 pathways are also UPR components that integrate signals from pathogen 297 associated molecular patterns that enable cross-talk between the UPR sensors and host cell 298 innate immunity responses to defend again pathogen invasion (Martinon, 2010, Janssens, 299 Pulendran et al., 2014, Smith, 2018). Therefore, the targeted inactivation of IRE1 and PERK 300 mediated enhancement of cytokine production by L.p. likely contributes to pathogen survival. The 301 ATF6 pathway has also been linked as a modulator of pro-inflammatory responses and ATF6 was 302 shown to enhance NF-κB signaling in ER stressed macrophages (Rao, 2014 considered constitutive degradation of ATF6 to be the source for the reduction in protein levels 328 during infection. Yet, we found impaired SEL1L-dependent ERAD and pharmacological inhibition 329 of proteasomal degradation failed to stabilize ATF6 protein levels during infection ( Figure 1D). 330 Though proteasomal degradation pathways are active, our data suggest that degradation of ATF6 331 is not a major contributor to the loss of ATF6 signal during L.p. infection. Furthermore, our data 332 revealed that reduction in ATF6(FL) protein levels correlated with the appearance of an ATF6 333 cleavage fragment during wild type infections ( Figure 2D). We considered the hypothesis that 334 ATF6 is processed to form an active transcription factor capable of increasing the expression of 335 target genes. Indeed, we found ATF6 regulated genes to be expressed under infection and 336 silencing of ATF6 reduced target gene expression directly linking the sensor to downstream 337 transcriptional activation (Figure 2A). Though many UPR genes are highly expressed during 338 infection, the genes are not translated at the rates observed under ER stress (Hempstead, 2015, 339 Treacy-Abarca, 2015). 340 We then determined host and pathogen components required for ATF6 activation under infection. 341 To our surprise, the major host pathway components were dispensable for L.p. mediated 342 activation. Blockage of ATF6 specific ER-to-Golgi translocation and inhibition of S1P/S2P 343 proteolytic activity traditionally restrict ATF6 activation under ER stress, however, neither 344 impacted L.p. induced pathway activation in terms of ATF6 processing and downstream gene 345 activation ( Figure 3A, 3B). Using time-lapse microscopy, we observed L.p. infection stimulated 346 nuclear recruitment of ATF6 without the preceding Golgi translocation observed under ER stress 347 induced activation (Supplemental Movie S2). Furthermore, ATF6 was detected around the LCV 348 at 6 hours post infection suggesting a mechanism for direct recruitment to the pathogen vacuole 349 ( Figure 4E). More so, the capacity for wild type L.p. infection to induce processing of S1P and 350 S2P site null mutants suggests a potentially unique site of cleavage under infection. However, the 351 instability of the ATF6-NT cleavage fragment has so far hindered our efforts to the identify the 352 novel cleavage site. 353 Our observations that ATF6 activation required a functional T4SS and the lack of a dependency 354 for canonical host factors on ATF6 activation suggested one or more L.p. secreted effectors were 355 involved. Using a phylogenetic approach, we identified the L.p. Paris strain lacks the capacity to 356 cleave ATF6. In doing so, we were able to identify several effectors unique to the L.p. Philadelphia 357 strain that, when expressed individually, increased expression of an ATF6 specific reporter 358 ( Figure 5F). Though the L.p. Paris strain lacks effectors shown to cleave ATF6, we do not dismiss 359 the potential for the Paris strain to also contain negative regulators of ATF6 cleavage which could 360 contribute to the observed reduction in ATF6 activation. Taken