The type III secretion system effector EspO of enterohaemorrhagic Escherichia coli inhibits apoptosis through an interaction with HAX-1

Many enteric pathogens employ a type III secretion system (T3SS) to translocate effector proteins directly into the host cell cytoplasm, where they subvert signalling pathways of the intestinal epithelium. Here, we report that the anti-apoptotic regulator HS1-associated protein X1 (HAX-1) is an interaction partner of the T3SS effectors EspO of enterohaemorrhagic Escherichia coli (EHEC) and Citrobacter rodentium, OspE of Shigella flexneri and Osp1STYM of Salmonella enterica serovar Typhimurium. EspO, OspE and Osp1STYM have previously been reported to interact with the focal adhesions protein integrin linked kinase (ILK). We found that EspO localizes both to the focal adhesions (ILK localisation) and mitochondria (HAX-1 localisation), and that increased expression of HAX-1 leads to enhanced mitochondrial localisation of EspO. Ectopic expression of EspO, OspE and Osp1STYM protects cells from apoptosis induced by staurosporine and tunicamycin. Depleting cells of HAX-1 indicates that the anti-apoptotic activity of EspO is HAX-1 dependent. Both HAX-1 and ILK were further confirmed as EspO1-interacting proteins during infection using T3SS-delivered EspO1. Using cell detachment as a proxy for cell death we confirmed that T3SS-delivered EspO1 could inhibit cell death induced during EPEC infection, to a similar extent as the anti-apoptotic effector NleH, or treatment with the pan caspase inhibitor z-VAD. In contrast, in cells lacking HAX-1, EspO1 was no longer able to protect against cell detachment, while NleH1 and z-VAD maintained their protective activity. Therefore, during both infection and ectopic expression EspO protects cells from cell death by interacting with HAX-1. These results suggest that despite the differences between EHEC, C. rodentium, Shigella and S. typhimurium infections, hijacking HAX-1 anti-apoptotic signalling is a common strategy to maintain the viability of infected cells.


Several T3SS effectors expressed by A/E pathogens have been
shown to induce apoptosis (EspH and EspF). However, infection with these pathogens does not lead to cell death as they also employ a number of T3SS effectors to inhibit both intrinsic and extrinsic apoptosis (Wong Fok Lung et al., 2014). NleH1 and NleH2, which are conserved in both EPEC and EHEC, provide protection against a wide range of pro-apoptotic stimuli including staurosporine (STS), tunicamycin (TUN) and brefeldin A (Hemrajani et al., 2010;Wong, Clements, Raymond, Crepin, & Frankel, 2012). NleF inhibits FasLinduced apoptosis by binding caspases-4, -8 and -9 and inhibiting their proteolytic activity (Blasche et al., 2013;Pollock et al., 2017).
Infection of cells with an EPEC E2348/69 mutant (ICC303), missing nleH1, nleH2 and not expressing nleF, is highly cytotoxic resulting in rapid apoptosis leading to cell detachment which can be complemented by plasmids encoding either NleH1 or NleH2 or by the pan caspase inhibitor z-VAD (Hemrajani et al., 2010). The extrinsic apoptotic pathway (FAS ligand-induced) is also blocked by NleB1, an Nacetylglucosaminetransferase that specifically modifies Arg 117 in the death domain of FADD (Pearson et al., 2013;S. Li et al., 2013). EspL blocks caspase-independent necroptosis by cleaving RHIM (receptorinteracting protein homotypic interaction motif) containing adapter proteins  and NleD, a Zn-dependent endopeptidase, inhibits apoptosis by specifically cleaving the activation loop of JNK (Baruch et al., 2011). Therefore, prolonging host cell survival appears to be an important virulence strategy of A/E pathogens.
The gene encoding the effector EspO is duplicated in the genome of EHEC O157:H7 strain EDL933, annotated as espO1 and espO2. In C. rodentium, espO is present as one copy, while espO is a pseudogene in the prototype EPEC strain E2348/69 (Iguchi et al., 2009). Kim and colleagues reported that OspE of Shigella spp., Osp1 STYM of Salmonella enterica serovar Typhimurium and EspO belong to the same effector family and interact with ILK (Kim et al., 2009). A tryptophan residue at position 68 (W68) of OspE, or at position 77 of EspO (W77) is essential for the interaction of OspE and EspO with ILK (Kim et al., 2009).
OspE localizes to focal adhesions (FAs), where it interacts with ILK, inhibits cell detachment and promotes Shigella attachment and cell survival (Kim et al., 2009;Miura et al., 2006). EHEC EspO2, which also binds ILK, can also block FAs disassembly by inhibiting the guanine nucleotide exchange factor (GEF) activity of the EHEC effector EspM2 (Morita-Ishihara et al., 2013). More recently, infection of mice with a C. rodentium mutant lacking EspO resulted in reduced proliferation of intestinal epithelial cells (IEC) and altered IEC signalling, leading to reduced neutrophil recruitment to the bacterial attachment site (Berger et al., 2018). At late stages of infection, a C. rodentium mutant lacking EspO also resulted in higher bacterial burden (Ale et al., 2017).
In this paper, we report a new role for the EspO effector family, which is to inhibit intrinsic apoptosis via its interaction with the antiapoptotic protein HAX-1. In order to identify whether there are additional EspO interaction partners than the previously described ILK, we performed a yeast two-hybrid (Y2H) screen, using EspO1 as bait and human cDNA library as prey. This screen identified HAX-1 as a putative EspO1 interacting protein (19 of 27 yeast clones). We further confirmed the interaction of HAX-1 with EspO1 by direct Y2H and demonstrated that EspO2, EspO CR , OspE1 SF ( Figure 1c) and Osp1 STYM (Figure S1) also interacted with HAX-1. The interaction between EspO1 and HAX-1 was validated by pull down assays using MBP-EspO1 and HEK293 cell lysates (Figure 1d). While a very faint band corresponding to HAX-1 can be observed in the pull down using MBP-lacZ, this band is much stronger in the pull down using MBP-EspO1 confirming an interaction between EspO1 and HAX-1. As the Y2H showed that all family members interacted with HAX-1, we mainly focused our characterisation on EspO1.
3D structure prediction using Phyre2 (Kelley, Mezulis, Yates, Wass, & Sternberg, 2015) revealed that EspO1 does not resemble any known structures in the database. However, the internal (26-62aa) region of EspO1 was predicted with low confidence to adopt an immunoglobulin-like beta-sandwich fold which suggests that EspO1 contains 2 short β-sheets (44-48aa and 51-54aa) connected by 2 positively charged residues (K or R) in the center of the protein (indicated in Figure 1a), flanked by regions of unknown structure at the N and C termini. In order to identify the HAX-1-binding site within EspO1, we generated constructs expressing EspO1 residues 1-45, 1-50, 1-56 and 46-91, which were used as baits in direct Y2H binding assays. The smallest fragment of EspO that allowed growth on selective media indicating full binding capacity spanned residues 46-91 ( Figure 2a). Consistent with this observation residues 1-45 were unable to support growth indicating these are not involved in the interaction with HAX-1. The two positively charged residues (aa 49 and 50) are not involved in the interaction as EspO1 K49/50A still interacts with HAX-1 ( Figure 2a). Most importantly, the conserved tryptophan required for interaction with ILK was not required for the interaction with HAX-1 as EspO1 W77A also interacted with HAX-1 by direct Y2H (Figure 2a). We next mapped the EspO1-binding site within HAX-1 using direct Y2H by generating a series of HAX-1 truncations. HAX-1 is a largely disordered protein that adopts partial folding in lipid membranes (Larsen et al., 2020). HAX-1 contains 2 Bcl-2 homology (BH) domains, BH1 (41-56aa) and BH2 (74-89aa), a PEST sequence (104-117aa) required for proteasomal degradation and a C-terminal region (118-279aa), which mediates interactions with multiple partners ( Figure 2b; Fadeel & Grzybowska, 2009;B. Li et al., 2012). This direct Y2H revealed that, like many HAX-1 interacting partners, EspO binds within the C-terminal (118-279 aa) region of HAX-1 (Figure 2c).

| EspO1 co-localizes with focal adhesions and HAX-1
We used immunofluorescence microscopy (IF) to determine the localisation of ectopically expressed EspO1 in HeLa cells. This revealed that EspO1 localized mainly to FAs (85.5% of cells) or the mitochondria (10.7% of cells), as it co-localized with either phosphorylated Y397 focal adhesion kinase (p-FAK) or the mitochondrial membrane protein TOMM-22 (Figure 3). In a small percentage of cells EspO could be seen in both FAs and mitochondria (3.8%). To study the co-localisation of EspO1 with HAX-1, we ectopically expressed EspO1(FLAG) and HAX-1 (myc). GFP(myc) and GFP(FLAG) were used as controls. IF analysis revealed co-localisation of EspO1 and HAX-1 ( Figure 4a). Interestingly, co-expression of HAX-1 with EspO1 appeared to shift the distribution of EspO1 from predominantly FA-localized to predominantly mitochondria-localized. To verify the dependence of EspO1 localisation on HAX-1 expression, the localisation of ectopically expressed EspO1 (FLAG) was quantified in the presence or absence of ectopically expressed HAX-1(myc) (Figure 4b). When co-expressed with GFP, EspO1 staining closely resembled that of FAs (89.7%), however, when co-expressed with HAX-1, EspO1 staining was observed in structures that resembled mitochondria (70.2%), suggesting that increased expression of one interacting partner may significantly alter EspO1 distribution ( Figure 4b).

| EspO1 has anti-apoptotic activity
As HAX-1 plays a role in inhibiting apoptosis (Fadeel & Grzybowska, 2009), we tested if EspO1 modulates cell death. To F I G U R E 1 EspO and OspE bind HAX-1. (a) Sequence alignment of the EspO family of effectors. EspO1 and EspO2 from EHEC EDL933, EspO CR from Citrobacter rodentium ICC180, OspE1 SF and OspE2 SF from S. flexneri strain M90T and Osp1 STYM from S. typhimurium strain SL1344 were aligned using CLUSTAL Omega. Predicted β-strands at aa 44-48 and 51-54 of EspO1 are underlined, + indicates the positively charged residues at aa 49/50 and * indicates the conserved tryptophan. (b) A phylogenetic tree of the EspO family members shown in (a), constructed using ClustalOmega with neighbour joining using BLOSUM62 and prepared with FigTree. (c) Yeast AH109 co-transformed with pGBKT7-espO homologues and pGADT7-HAX-1 grew on selective medium (QDO), indicating a protein interaction whereas yeast co-transformed with pGBKT7-espO homologues and empty pGADT7 did not. Growth on non-selective media (DDO) indicated a successful plasmid co-transformation. (d) HEK293T whole cell extracts were incubated with purified and immobilised MBP-EspO1, or MBP-LacZ as a negative control. HAX-1 is specifically pulled down with MBP-EspO1 but not MBP-LacZ exclude any effect of ILK interaction EspO1 W77A (which interacts with HAX-1 but not ILK) was included. Cells ectopically expressing EspO1 (FLAG) or EspO1 W77A (FLAG) were treated with STS and the level of apoptosis quantified by DNA fragmentation (TUNEL assay; Figure 5a and Figure S2a), or cleaved caspase-3 staining (Figure 5b and Figure S2b). Cells expressing NleH1(HA) or treated with z-VAD were used as controls. In both assays, EspO1 and EspO1 W77A significantly blocked apoptosis when compared with mock-transfected cells. This was comparable to the anti-apoptotic effect observed with NleH1 ( Figure 5a,b). Similar results were obtained when apoptosis was induced by tunicamycin ( Figure S2d). Importantly, cells expressing OspE1, OspE1 W68A and Osp1 STYM also inhibited STS and tunicamycin-induced apoptosis to a similar level as cells EspO1, EspO1 W77A or NleH1 indicating that the anti-apoptotic function of EspO is conserved ( Figure S2c,d). In contrast, the inhibition of cell death was not observed in control cells expressing GFP ( Figure S2c,d).
These results suggest that EspO1, OspE1 and Osp1 STYM are antiapoptotic effectors and that their anti-apoptotic activity is independent of the tryptophan residue implicated in ILK binding.
In order to investigate whether the anti-apoptotic activity of EspO1 was HAX-1 dependent, we utilized a HAX-1 knocked down HeLa cell line (miHAX-1; Grzybowska et al., 2013). A control cell line, transfected with the same plasmid but without a silencing sequence (miNEG), was used for comparison. Following confirmation that HAX-1 was successfully knocked down by Western blotting (Figure 5c), we ectopically expressed EspO1(FLAG), EspO1W77A (FLAG) and NleH1(HA) in miHAX-1 and miNEG cells; z-VAD was used as an additional control. As previously shown expression of EspO1, EspO1 W77A and NleH1 decreased STS-induced apoptosis in the control miNEG cells compared to MOCK-transfected cells when measured by TUNEL (Figure 5d) or cleaved caspase-3 staining ( Figure 5e).
In contrast, EspO1 and EspO1 W77A lost their ability to protect cells against STS treatment in the HAX-1 knocked down cells by TUNEL and cleaved caspase 3 staining (Figure 5d,e). Importantly, both assays showed that z-VAD and NleH1 were able to equally protect miHAX-1 and miNEG cells from STS-induced apoptosis (Figure 5d,e). These results show that HAX-1 is specifically involved in the anti-apoptotic activity of EspO1 but not NleH1.

| EspO1 interacts with HAX-1 during infection
We next aimed to investigate whether the interacting partners of Side-by-side FLAG and Strep pulldowns were performed using EPEC1 encoding an empty vector (EV). Proteins identified in these samples were designated non-specific and removed from further analysis.
Ninety-five proteins were specifically identified in both FLAG and Strep pulldowns with EspO1 ( Figure 6b and Table S4). The MS/MS parameters for all proteins identified in both controls and pulldowns have been listed in Tables S5-S8. As expected, the bait EspO-1 was F I G U R E 2 Mapping the EspO1-HAX-1 interaction sites. (a) Yeast co-transformed with HAX-1 and the EspO truncations 46-91, 1-56 and 1-50aa grew on selective media (QDO); yeast co-expressing HAX-1 and the EspO truncation 1-45aa or containing the EV (control) did not grow on selective media. Growth on non-selective media (DDO) indicated a successful plasmid co-transformation for all pairs. (b) Schematic representation of HAX-1 with Bcl-2 homology (BH) 1 and 2, and PEST domains indicated. (c) AH109 co-expressing EspO1 and the C-terminal part of HAX-1 (65-279, 100-279 or 118-279aa) grew on selective media (QDO) while yeast expressing the N-terminal region of HAX-1 (1-65, 1-100 or 1-118) did not. Growth on nonselective media (DDO) indicated a successful plasmid cotransformation for all pairs one of the most highly enriched proteins identified with 77 and 333 peptide spectrum matches (PSMs) in the FLAG and Strep pulldowns, respectively (Figure 6c). HAX-1 was also identified in both pull-downs, while ILK was identified as an interacting partner in the Strep but not the FLAG pull-down. The interaction of EspO1 and HAX-1 was further confirmed in a targeted FLAG pull-down by immunoblotting for HAX-1 following infection in HT-29 and HeLa cells ( Figure 6d).

| EspO1 inhibits cell death during infection in a HAX-1-dependent manner
Finally, to investigate whether EspO1 was also protecting against cell death during infection, we utilized a cell detachment assay.
EHEC encodes two EspO paralogs along with the anti-apoptotic effectors NleH1, NleH2 and NleF. Therefore, it is not surprising that EHEC mutants missing either espO1/espO2, nleH1/nleH2 or nleF do not induce cell detachment ( Figure S3). In order to circumvent the problem of creating an EHEC espO1, espO2, nleH1, nleH2, nleF mutant, we used EPEC strain ICC303 (a double nleH1 and nleH2 E2348/69 deletion mutant, which has a polar effect on expression of nleF). ICC303 is highly cytotoxic, causing cell death leading to cell detachment (Hemrajani et al., 2010). The cell detachment phenotype of this strain could be complemented by expression of NleH1 from a plasmid or by z-VAD (Hemrajani et al., 2010). Accordingly, we investigated if EspO1 could also prevent cell detachment following infection of HeLa cells with ICC303 and if so, whether this was dependent on HAX-1. HeLa miNEG and miHAX-1 cells were infected for 90 min with ICC303, or ICC303 expressing EspO1, EspO1 W77A or NleH1 or NleF as controls, before quantification of the remaining attached cells. Treating ICC303-infected cells with z-VAD was used as an additional control. Consistent with previous results, infection of miNEG with ICC303 resulted in significant cell detachment (67%) compared with cells infected with wild type (WT) EPEC (13%; Figure 7). The introduction of EspO1, EspO1 W77A , NleH1 or NleF or z-VAD treatment of ICC303 significantly reduced cell detachment to 15%, 18%, 14% 12% and 16%, respectively ( Figure 7). However, in the absence of HAX-1 (miHAX-1 cells), ICC303 expressing EspO1 and EspO1 W77A were no longer able to prevent cell detachment (59% and 61% cell detachment, respectively), while NleH1, NleF and z-VAD treatment all retained their protective capabilities (15% cell detachment; Figure 7). This suggests that EspO1 inhibits cell detachment during infection in a HAX-1-dependent/ILK independent manner, while NleH1 and NleF operate independently of HAX-1.
This suggests that the EspO effector family is an important T3SS virulence factor amongst pathogenic E. coli as well as Shigella and S. typhimurium.
Using non-targeted pull downs during infection, we confirmed that HAX-1 was part of the EspO1 interactome when EspO is delivered by the T3SS during infection. In addition to HAX-1, 2 other proteins involved in apoptotic signalling were found in the EspO interactome in the STREP pulldown; p53 and BH3 interacting domain death agonist (BID). While these proteins are unlikely to directly interact with EspO they suggest that through an interaction with HAX-1, EspO can modulate apoptotic signalling complexes to alter the cellular response to apoptotic stimuli. Interestingly ILK that was previously reported to interact with the EspO family of effectors (Kim et al., 2009) was identified only in one of the EspO1 pull-downs. This may reflect differing levels of protein abundance between HAX-1 and ILK or a less stable interaction between ILK and EspO1. HAX-1 is a part of the ILK interactome (Dobreva, Fielding, Foster, & Dedhar, 2008), which could suggest that EspO1 and its paralogs target the HAX-1-ILK complex through multiple interactions.
OspE1 interacts with ILK at FAs where it stabilizes cell-matrix adhesion sites and inhibits cell detachment during Shigella infection (Kim et al., 2009). We demonstrated that the ability of EspO1 to pre-  (Zhang et al., 2015). We demonstrate that the S. typhimurium effector Osp1 STYM , which interacts with ILK (Kim et al., 2009), also binds HAX-1 and inhibits STS-induced apoptosis.
HAX-1, which is a ubiquitously expressed anti-apoptotic protein, interacts with numerous cellular partners including XIAP, caspase-3, caspase-9, the mitochondrial serine protease Omi/HtrA and SERCA2 F I G U R E 4 Co-localization of EspO1 with HAX-1. (a) HeLa cells expressing GFP(FLAG) and HAX-1(myc) (top row), EspO1(FLAG) and GFP(myc) (middle row), or EspO1(FLAG) and HAX-1(myc) (bottom row), were immunostained with anti-Flag and anti-Myc antibodies. Anti-FLAG staining is green, anti-myc staining is red and DNA (DAPI) is blue in the merged image. Scale bar = 20 μm. (b) The percentage of cells where EspO1 was distributed to FAs or mitochondria was calculated in cells co-expressing EspO1 with either GFP or HAX-1. When co-expressed with GFP, EspO1 was mainly distributed to FAs; however, when HAX-1 is overexpressed, EspO1 was mainly distributed to mitochondria in the majority of cells (***p < .001). Results are the average of three independent biological repeats, counting 100 cells per condition in the endoplasmic reticulum (ER; Cilenti et al., 2004;Kang et al., 2010;Lee et al., 2008;Shaw & Kirshenbaum, 2006;Vafiadaki et al., 2009). HAX-1 has 8 different splicing variants (Trebinska et al., 2010); only variant I has been extensively studied and is reported to localize at mitochondria (Koontz & Kontrogianni-Konstantopoulos, 2014). In this study, we found that ectopically expressed EspO1 localizes to both FAs and to the mitochondria, but overexpression of HAX-1 shifted the localisation of EspO1 to the mitochondria. SERCA2, which actively pumps Ca 2+ into the ER from the cytosol, is regulated by an interaction with HAX-1 (Vafiadaki et al., 2009). Overexpression of HAX-1 leads to SERCA2 proteasomal degradation, resulting in diminished ER Ca 2+ content and protection of mitochondria from Ca 2+ overload (Vafiadaki et al., 2009). HAX-1 is degraded by Omi/HtrA2 in an early stage of apoptosis (Cilenti et al., 2004). At later stages, SMAC/Diablo is released from the mitochondria and binds XIAP, leading to the liberation of active caspases from XIAP and to XIAP degradation (Srinivasula et al., 2001). The inhibition of Omi/HtrA2, as well as over expression of HAX-1 or XIAP, protect cells from apoptotic stimuli (Hegde et al., 2002;Kang et al., 2010;Trebinska, Högstrand, Grandien, Grzybowska, & Fadeel, 2014). In this paper, we have shown that expression of EspO1 has a similar effect.
We have established that EspO1 exerts its anti-apoptotic effect both during ectopic expression and bacterial infection, in a HAX-1-dependent manner. While EspO and its family members bind both ILK and HAX-1, EspO1 W77A , which cannot bind ILK, is able to protect cells against cell death. Importantly, the anti-apoptotic activities of NleH1 and NleF are HAX-1 independent. Although the exact mechanism through which the EspO family prevents cell death is not known, it is possible that they facilitate the stability of HAX-1 and/or XIAP or modulates the levels of Ca 2+ in the ER to protect cells against intrinsic apoptosis. The fact that the extracellular pathogens EHEC and CR and F I G U R E 5 EspO inhibits STS-induced apoptosis. Cells expressing EspO(FLAG), EspO1 W77A (FLAG) or NleH1(HA) were treated with STS and apoptosis measured by detecting DNA fragmentation (fluoroscein-12-dUTP) (a) or cleaved caspase-3 (b). Ectopic expression of EspO1, EspO1 W77A , NleH1 or z-VAD treatment prevented DNA fragmentation and cleavage of procaspase-3 induced by STS compared with mocktransfected cells. Results are the average of three independent biological repeats, counting 100 cells per condition. (***p < .001). Representative images can be seen in Figure S2a,b. (c) Cell lysates from miHAX-1 and the parental miNEG cells were analyzed by Western blot with HAX-1 and tubulin-control antibodies to demonstrate HAX-1 knockdown. (d) and (e) miNEG or miHAX-1 cells expressing EspO(FLAG), EspO1 W77A (FLAG) or NleH1(HA) were treated with STS and apoptosis measured by detecting DNA fragmentation (fluoroscein-12-dUTP) (d) or cleaved caspase-3 (e). The pan-caspase inhibitor z-VAD was used as apoptotic inhibitor control. The results were obtained from three independent biological repeats, counting 100 cells per condition (*p < .05) the invasive pathogens Shigella and S. typhimurium use T3SS effectors to target HAX-1 signalling suggests that it plays a central role in pathogen host interactions.

| EXPERIMENTAL PROCEDURE
4.1 | Strains, growth conditions and reagents E. coli strains used in this study are listed in Table S1. ICC303, constructed as a double nleH1/nleH2 mutant, was later found to have a polar effect on expression of NleF and therefore is functionally a triple

| Yeast transformation
S. cerevisiae wild-type yeast strain AH109 was grown overnight at 30 C in YPDA broth and then centrifuged at 4000 rpm for 15 min.
The pellet was washed twice with sterile water and resuspended in the transformation mix containing 50% w/v PEG 3350, 36 μl

| Yeast two-hybrid screening
A yeast two-hybrid screen was performed using EHEC EDL933

EspO1 as a bait with the Matchmaker Pretransformed Normalised
HeLa cDNA library (Clontech) according to the manufacturer's instructions. Positive clones were selected by plating on high-stringency media lacking Trp, Leu, His and Ade with X-alpha-gal (quadruple dropout media or QDO). Colonies were re-streaked on QDO and inserts amplified using the AD-LD primers (Clontech). The inserts were sequenced and identified using Blastn and Blastx.

| Transfection and STS induced apoptosis
HeLa cells were seeded at the concentration of 5.5 Â 10 4 cells/well in 24 well-plate containing 13 mm glass cover slips (VWR International).
Immunofluorescence was visualized with a Zeiss Axio Imager immunofluorescence microscope using Zeiss AxioVision software and images were processed using ImageJ. Colocalisation analysis was quantitatively analysed using the Coloc2 function in ImageJ. Pearson's correlation coefficient was calculated for each image with Costes threshold regression. Pearson's r values >.6 were considered to demonstrate strong colocalisation.

| In vitro infection and pulldown
HT-29 cells were seeded at a concentration of 0.3 Â 10 6 cells per well in a six-well plate one day before infection. EPEC-1 EV and EPEC-1 EspO1 strains were primed by diluting the overnight cultures 50Â in non-supplemented Dulbecco's Modified Eagle Medium (DMEM), low glucose and incubating static at 37 C with 5% CO 2 for 3 hr. 0.5 mM IPTG (isopropyl β-d-1-thiogalactopyranoside) was used to induce EspO1 expression from the plasmid 30 min before infection. Infection was carried out at a multiplicity-of-infection (MOI) of 50:1 for 3 hr at 37 C, 5% CO 2 . Post infection, cells were washed three times in PBS followed by incubating the cells in lysis buffer for 20 min (50 mM Tris-HCl, pH 7.4, with 150 mM NaCl, 1 mM EDTA and 1% TRITON X-100) with freshly added protease inhibitors. The lysate was collected and unlysed cells and debris removed by centrifugation at 12,000g. Anti-FLAG M2 affinity resin and MagStrep "type3" XT beads (iba) were used for pulldowns as per manufacturer's instructions.
Binding to beads was performed for 4 hr at 4 C. The beads were directly processed for trypsin digestion as described below or proteins eluted for immunoblotting by addition of non-reduced laemmli buffer at 95 C for 10 min. The samples were analyzed by SDS-PAGE and western blot analysis using anti-HAX-1 and anti-FLAG and HRPconjugated goat anti-rabbit IgG, Fc fragment specific (Jackson Immu-noResearch, 111-035-008).

| On-beads digestion, mass spectrometry and protein identification
Following FLAG or Strep pulldown, beads were washed 3 times and tetraethylammonium bromide (TEAB) buffer was added. This was followed by on-bead protein reduction using 0.5 M tris The mass spectrometry proteomics data has been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the data set identifier PXD023126.

| In vitro ICC303 infection and cell detachment assay
HeLa cells were seeded at the concentration of 7.5 Â 10 4 cell/well in 24 well-plate 48 hr prior to infection. Bacteria were grown in LB overnight and 3 hr before infection, diluted 1:100 in DMEM (no IPTG was added in the primed bacteria). Before infection, the HeLa monolayer was washed three times with PBS and 0.5 ml of primed culture was added to each well with 20 μM z-VAD-fmk when required. The monolayers were incubated for 90 min, washed five times with PBS, then trypsinized and counted to quantify the number of remaining cells.

| Statistical analysis
Data analysis was performed by GraphPad Prism software, using twoway ANOVA, and one-way ANOVA where specified, and Bonferroni post-test. Statistically significant was considered when p value is <.05. and Technology Project, Royal Government of Thailand. This project