Heterologous Display of Chlamydia trachomatis PmpD Passenger at the Surface of Salmonella OMVs

Chlamydia trachomatis is the bacterial pathogen that causes most cases of sexually transmitted diseases annually. To combat the global spread of asymptomatic infection, development of effective (mucosal) vaccines that offer both systemic and local immune responses is considered a high priority. In this study, we explored the expression of C. trachomatis full-length (FL) PmpD, as well as truncated PmpD passenger constructs fused to a “display” autotransporter (AT) hemoglobin protease (HbpD) and studied their inclusion into outer membrane vesicles (OMVs) of Escherichia coli and Salmonella Typhimurium. OMVs are considered safe vaccine vectors well-suited for mucosal delivery. By using E. coli AT HbpD-fusions of chimeric constructs we improved surface display and successfully generated Salmonella OMVs decorated with a secreted and immunogenic PmpD passenger fragment (aa68-629) to 13% of the total protein content. Next, we investigated whether a similar chimeric surface display strategy could be applied to other AT antigens, i.e., secreted fragments of Prn (aa35-350) of Bordetella pertussis and VacA (aa65-377) of Helicobacter pylori. The data provided information on the complexity of heterologous expression of AT antigens at the OMV surface and suggested that optimal expression strategies should be developed on an antigen-to-antigen basis.


Introduction
C. trachomatis is the most common sexually transmitted pathogen worldwide. Its complex, partly intracellular lifecycle and ability to enter a "persistence phase" when meeting harmful conditions hampers the treatment with conventional antibiotics [1]. Importantly, chlamydial infections often remain asymptomatic, which contributes to its transmission in a sexually active population. Consequently, development of a (mucosal) vaccine that offers protection against transmission and/or chlamydial disease is considered a high priority [2]. Research has focused on both whole-cell-based and subunit-based vaccine formulations but has not yet led to a licensed vaccine [3].
C. trachomatis major outer membrane protein (MOMP) is the best studied antigen included in subunit vaccines. It is used in more than half of the published preclinical studies [3] and CTH522, a recombinant non-native MOMP-derivative, has recently entered into clinical phase [4]. The porin is strongly immunogenic and contains various linear and conformational T-and B-cell epitopes [5]. Unfortunately, however, recombinant MOMP in its native conformation is difficult to produce. Other challenges include the partial and serovar-specific immunity it elicits, as well as the variable protection against (re)infection shown in preclinical studies [6].

Reagents and Sera
Restriction enzymes were obtained from New England Biolabs (Ipswitch, MA, Maine). Lumi-light substrate was purchased from Roche (Mannheim, Germany). Phusion High Fidelity DNA polymerase and Pierce Silver stain kit were from Thermo Fischer Scientific (Waltham, MA, USA). Proteinase K and phenylmethanesulfonyl fluoride (PMSF) were from Sigma-Aldrich (Saint Louis, MO, USA). In-Fusion HD plus kit for seamless DNA cloning was obtained from Takara Bio (Mountain View, CA, USA). Coomassie Blue G250 was from Biorad (Oxford, UK). N-terminal amino acid sequence was determined by Edman degradation performed by Alphalyse A/S (Odense, Denmark).

Plasmid Construction
All PCR primers used for plasmid construction (Table S1) and amino acid sequences of the expressed proteins are listed in the supplementary file.
The plasmid pLemo-PmpD (FL) was constructed to express PmpD (FL), which is a fulllength version of C. trachomatis serovar L2 (strain 434/Bu/ATCC VR-902B) PmpD carrying an E. coli Hbp AT signal peptide, 8 residues of its mature, processed part after passage of the cytoplasmic membrane, an HA-tag, and an E. coli Bam recognition motif (RYSF) at its C-terminus [34]. To avoid erroneous disulfide bond formation, twenty-four cysteines encoded in PmpD (FL) were replaced by serines. Two overlapping and E. coli codonoptimized DNA sequences encoding PmpD (FL) were synthesized by Geneart/Thermo Fisher Scientific, assembled and ligated into the SalI and BamHI sites of pLemo [35] using In-Fusion cloning to yield pLemo-PmpD (FL), controlled by the rhaBAD promoter [36].
All other expression plasmids were derivatives of vector pEH3, with constructs under control of an IPTG-inducible lacUV5 promoter [37], of which pEH3-Hbp∆βcleavage has been described previously [23].

Recombinant Protein Expression
Cultures of E. coli BL21(DE3) and BL21(DE3)omp8 strains harboring pLemo-PmpD (FL) were grown to an OD 600 of 0.4-0.5, induced for expression by addition of 8 mM L-rhamnose and then further incubated for 2 h [40]. Cultures of E. coli BL21(DE3) and BL21(DE3)omp8 cells harboring pEH3-derivatives were grown until an OD 600 of 0.4-0.5 and then induced with 0.1 mM IPTG for 2 h for expression of the HbpD-fusions [23]. Cultures of Salmonella SL3261 ∆tolRA ∆msbB cells harboring pEH3-derivatives were first grown to an OD 600 of 1.0-2.0, diluted to an OD 600 of 0.02, after which growth was continued overnight in the presence of 0.1 mM IPTG to induce expression of the HbpD-fusions [24]. Culture samples were subjected to centrifugation (12,000× g, 1 min, RT) to separate the bacterial cells from the medium. The cell pellets were resuspended in PBS and then mixed with SDS-PAGE loading buffer (f.c. 125 mM Tris-HCl, pH 6.8, 4% SDS, 20% glycerol, 0.02% bromophenol blue, 100 mM DTT), heated at 97 • C for 10 min and analyzed on 10% SDS-PAGE gels. Gels were further subjected to Coomassie Blue G250-and silver-staining or immunoblotting. Coomassie-or silver-stained gels were imaged using a Molecular Imager GS800 Calibrated Densitometer (Biorad). The percentage of each recombinant chimera over total OMV proteins was estimated by densitometry using ImageJ (downloaded from https://imagej.net/ij/index.html; accessed on 20 April 2022) [41]. Immunoblotting images were captured using the Amersham Imager 600 (GE Healthcare, MA, USA).

Isolation of Cell Envelopes and Outer Membrane Fractions
Cell envelopes (CEs) were collected using a subcellular fractionation procedure as described previously [40]. In short,~200 OD 600 units of BL21(DE3)omp8 cells expressing PmpD (FL) were resuspended in lysis buffer (100 mM NaCl, 1 mM EDTA, 50 mM Tris HCl, pH 8.0) in the presence of protease inhibitor (Roche, Mannheim, Germany). The suspension was kept in ice for 15 min after which the cells were lysed by two passages through a One-shot cell disruptor (Constant Systems LtD, Daventry, UK) at 1.9 kbar. Whole-cell lysates were centrifuged at 6000× g, 10 min, 4 • C to remove debris and unbroken cells and the resulting supernatant was applied to ultracentrifugation (293,000× g, 60 min, 4 • C) to sediment CEs.
OMs were subsequently purified by sucrose gradient ultracentrifugation using a procedure adapted from Ruiz et al. [42]. Briefly, CEs collected above were resuspended in 8 mL Tris-B buffer (10 mM Tris-HCl, pH 8.0), containing 20% (w/w) sucrose and 1 mM PMSF. The suspension was layered onto a two-step sucrose gradient (top: 4 mL Tris-B buffer containing 40% (w/w) sucrose; bottom: 1 mL Tris-B buffer containing 65% (w/w) sucrose). After ultracentrifugation (87,000× g, 20 h, 4 • C) in a Beckman SW28-Ti rotor with slow acceleration and no braking, a~0.5 mL volume containing OMs was harvested from the 40%/65% interface by puncturing a side of the tube with a syringe needle. The harvested fraction was diluted three times with washing buffer (20 mM Tris-HCl, pH 8.0) and subjected to ultracentrifugation (18,000× g, 30 min, 4 • C) to collect OMs. The pellet was resuspended in TAE-sucrose buffer (50 mM TEA, 250 mM sucrose) and OM material was stored at −20 • C until used.

Proteinase K Accessibility Assay
The proteinase K assay was adapted from a previous study [22]. E. coli or Salmonella OMVs were resuspended in reaction buffer (50 mM Tris-HCl, pH 7.4, 1 mM CaCl 2 ). Where appropriate, OMVs were lysed by incubation with 0.02% (v/v) Triton X-100 for 15 min in ice. Intact and lysed OMVs were incubated for 30 min at 37 • C in the presence of 100 µg/mL proteinase K (ProK). Intact OMVs incubated under identical conditions in reaction buffer without ProK served as untreated control. Reactions were stopped by addition of 4 mM PMSF, after which all samples were precipitated with trichloroacetic acid and analyzed by SDS-PAGE and Coomassie-or silver-staining and immunoblotting.

Expression and Localization of Full-Length PmpD (FL)
In absence of a solved structure of C. trachomatis PmpD (FL), we used Alphafold2 [44] to construct models of PmpD (FL) that guided the construct design ( Figure S1). As expected, the model showed an N-terminal right-handed β-helix and a C-terminal 12-stranded β-barrel, both with a high similarity to solved structures of ATs [10]. Considering the relatively autonomous and conserved mechanism of AT secretion, we initially expressed PmpD (FL) in E. coli in an attempt to obtain OMVs displaying the antigen in its native conformation. C. trachomatis PmpD (FL) was placed under control of the L-rhamnose inducible rhaBAD promoter ( Figure 1A) that allows for tunable expression of recombinant proteins in E.coli [35]. The native signal peptide of PmpD was replaced by the E. coli Hbp signal peptide for optimal engagement of the Sec translocon in the E. coli IM. In addition, twenty-four cysteine residues present in the chlamydial PmpD were replaced by serine residues to avoid disulfide bond formation in the E. coli periplasm, because this is known to hinder AT translocation across the OM [45]. Finally, the four C-terminal residues of PmpD (RLIF) were replaced by those of E. coli AT Hbp (RYSF) to facilitate efficient recognition by the E. coli Bam complex [34]. This motif of four residues is also known as the β-signal and promotes interactions with the Bam complex in a speciesspecific fashion [46,47]. The optimized PmpD (FL) was expressed in E. coli BL21(DE3) and whole-cell lysates from induced and non-induced cells were analyzed by SDS-PAGE and Coomassie staining or immunoblotting using PmpD antiserum ( Figure 1B,C). Induction yielded a~160 kDa product that was readily detectable ( Figure 1B, lane 2 and Figure 1C, lane 2) indicating efficient expression. Moreover, a prominent~120 kDa product (*) was identified on Coomassie-stained gels ( Figure 1B, lane 2), presumably corresponding to a proteolytic product [8].
To characterize the proteolytic product, the~120 kDa band (*) ( Figure 1B, lane 2) was subjected to N-terminal amino acid sequencing [48]. The resulting sequence, GTVNN, corresponded to the first five amino acids of the mature protein, present directly downstream of Hbp signal peptide after it is cleaved from the PmpD (FL) precursor by signal peptidase upon passage of the E. coli IM (see supplementary file). Hence, the proteolytic product represents a~120 kDa N-terminal fragment of the PmpD passenger and resembles a similar-sized, cell-associated proteolytic product found upon expression of native PmpD in C. trachomatis, suggesting similar processing of our PmpD (FL) construct in E. coli [49].  Next, we investigated whether PmpD (FL) is located at the surface of OMVs, which is necessary to enable vesicle-based vaccine formulations. As expression host we used E. coli BL21(DE3)omp8 that lacks four major OMPs (OmpA, OmpF, OmpC, LamB) and, consequently, shows a hypervesiculating phenotype [28]. Expression of PmpD (FL) in BL21(DE3)omp8 revealed similar levels of the~160 and~120 kDa PmpD products in whole cell lysate samples as detected in BL21(DE3) ( Figure 1B, lane 4 and Figure 1C, lane 4). Subsequently, crude CE and OM fractions were isolated from the cells expressing PmpD (FL) and analyzed in parallel to the cell lysates. Coomassie staining ( Figure 1D) and immunoblotting using PmpD antiserum ( Figure 1E) showed that the~120 and~160 kDa species could be recovered from both fractions, consistent with correct biogenesis of PmpD (FL). To monitor recruitment to OMVs, cell-free supernatants of culture samples were subjected to ultracentrifugation and the OMV-containing pellet was analyzed for the presence of PmpD (FL). However, PmpD (FL) could not be detected in this fraction by Coomassie staining ( Figure 1D, lane 4) nor by immunoblotting ( Figure 1E, lane 4). In contrast, the OM-based lipoprotein LpoB was clearly detected, which verifies the OMV isolation procedure and the loading of similar amounts of OM and OMV material, respectively ( Figure 1E, c.f. lanes 3 and 4) [50]. Taken together, PmpD (FL) was successfully targeted to the OM of E. coli but was not included in OMVs.

Display of Antigenic Passengers upon Fusion to Hbp in E. coli
Since PmpD (FL) was not detected in OMVs, we then tried to display the immunogenic passenger part that is secreted in C. trachomatis on the surface of E. coli as part of a chimeric AT construct. Previously, we have used HbpD(∆d1), an engineered version of E. coli AT Hbp, for antigen display [22]. In this construct, antigens are positioned on top of the β-helical core of the Hbp passenger through genetic replacement of its native domain 1 (d1) ( Figure S2A). Given the predicted structural similarities of the Hbp and PmpD passengers, we considered such a fusion a promising display strategy, despite the considerable size and folding complexity of the truncated PmpD ( Figure S2B). To explore the generic applicability of this fusion strategy, we also tested display of truncates of B. pertussis Prn and H. pylori VacA, two alternative AT virulence factors with β-helical structures ( Figure S2B).
In the next step, we tested whether HbpD(∆d1)-PmpD, -Prn and -VacA were included in OMVs. OMVs were isolated from the culture medium of BL21(DE3)omp8 cells and analyzed by SDS-PAGE followed by silver staining ( Figure 3A) and immunoblotting using Hbp antiserum ( Figure 3B). All three fusions appeared to be present in OMVs ( Figure 3A,B,  lanes 1, 4, 7), albeit with a relatively modest density for HbpD(∆d1)-VacA ( Figure 3A,B, lane 7). Of note, due to the use of BL21(DE3)omp8 as the OMV production host, typical porin bands (OmpA/F/C) are absent from the OMV protein profiles ( Figure 3A). To assess surface exposure of the heterologous passenger domains, the OMVs were treated with ProK to digest external proteins. Clearly, the HbpD(∆d1)-fusions were degraded ( Figure 3A,B, lanes 2, 5, 8). As a control for OMV integrity, we analyzed the accessibility of the partially exposed BamA subunit of the Bam complex ( Figure 3C). These control experiments showed that BamA was cleaved by ProK, whereas the intracellular BamB-E components of the complex remained intact ( Figure 3C, lanes 2, 5, 8), unless the OMVs were first solubilized with the detergent Triton X-100 ( Figure 3C, lanes 3, 6, 9) [51,52]. Therefore, display of heterologous β-helical passenger structures fused to HbpD(∆d1) was achieved at E. coli OMV surface, in contrast to what was found for PmpD (FL). In the next step, we tested whether HbpD(∆d1)-PmpD, -Prn and -VacA were cluded in OMVs. OMVs were isolated from the culture medium of BL21(DE3)omp8 c and analyzed by SDS-PAGE followed by silver staining ( Figure 3A) and immunoblot using Hbp antiserum ( Figure 3B). All three fusions appeared to be present in OMVs ( ure 3A,B, lanes 1, 4, 7), albeit with a relatively modest density for HbpD(∆d1)-VacA ( ure 3A,B, lane 7). Of note, due to the use of BL21(DE3)omp8 as the OMV production h typical porin bands (OmpA/F/C) are absent from the OMV protein profiles (Figure 3 To assess surface exposure of the heterologous passenger domains, the OMVs w treated with ProK to digest external proteins. Clearly, the HbpD(∆d1)-fusions were graded ( Figure 3A,B, lanes 2, 5, 8). As a control for OMV integrity, we analyzed the ac sibility of the partially exposed BamA subunit of the Bam complex ( Figure 3C). Th control experiments showed that BamA was cleaved by ProK, whereas the intracellu BamB-E components of the complex remained intact ( Figure 3C, lanes 2, 5, 8), unless OMVs were first solubilized with the detergent Triton X-100 ( Figure 3C, lanes 3, 6

Display of Truncated Passengers upon Fusion to Hbp in Salmonella
Our laboratory prefers to use OMVs from the attenuated and hypervesiculating Salmonella Typhimurium strain SL3261 ∆tolRA ∆msbB as vaccine carrier, since it carries a detoxified version of LPS and was also previously used to display various HbpD(∆d1)fusion constructs at the surface of OMVs [21,26]. We, therefore, tested expression of the HbpD(∆d1)-PmpD, -Prn and -VacA fusions in this strain and used as controls empty vector (pEH3-EV) and expression of Hbp∆βcleavage [23], which displays the full-length Hbp passenger including domain 1. Following our routine production of OMVs [21,26], cells were induced overnight for protein expression with 0.1 mM IPTG, after which cells and OMVs were collected and analyzed (Figure 4). Expression of HbpD(∆d1)-PmpD, -Prn and to a lesser extent -VacA fusions in cells was observed on Coomassie-stained gels, albeit at lower levels than the Hbp∆βcleavage positive control ( Figure 4A). Immunoblotting using Hbp antiserum did confirm the expression of the HbpD(∆d1)-fusion products ( Figure 4B), but also revealed lower running bands (*) that were likely degradation products of the HbpD(∆d1) fusions. Of note, the Coomassie-stained gel hardly displayed these undesirable HbpD(∆d1)-fusion truncates and it remains unclear why they are so prominently detected by immunoblotting (c.f. Figure 4A,B, lanes 3-5).

Display of Truncated Passengers upon Fusion to Hbp in Salmonella
Our laboratory prefers to use OMVs from the attenuated and hypervesiculating Salmonella Typhimurium strain SL3261 ∆tolRA ∆msbB as vaccine carrier, since it carries a detoxified version of LPS and was also previously used to display various HbpD(∆d1)-fu-  VacA in Salmonella and their OMVs. SL3261 ∆tolRA ∆msbB cells were induced for expression of empty vector pEH3 (EV), Hbp∆βcleavage, HbpD(∆d1)-PmpD, HbpD(∆d1)-Prn and HbpD(∆d1)-VacA and whole cells were analyzed by SDS-PAGE followed by Coomassie staining (A) and immunoblotting using antisera against Hbp or DegP (B). Equivalent amounts of OMVs were analyzed by SDS-PAGE followed by Coomassie staining (C) or immunoblotting using Hbp antiserum (D). In the panel (B-D), bands representing proteolytically degraded HbpD(∆d1) were indicated (*). OMVs expressing HbpD(∆d1)-PmpD were used for a proteinase K (ProK) assay. Samples of untreated intact OMVs, intact OMVs treated with ProK, and lysed OMVs, treated with TritonX-100 (TX-100) and incubated with ProK were analyzed by SDS-PAGE followed by Coomassie staining (E) or immunoblotting using PmpD-directed antiserum (F). MW markers (in kDa) are indicated at the left and identified protein bands at the right side of panels.
Although the analyzed amounts of OMV material loaded on the gels were derived from the same amounts of Salmonella cells, as judged from OD 600 densities ( Figure 4C), the OMP protein profiles detected were considerably weaker in HbpD(∆d1)-PmpD, -Prn and -VacA OMVs when compared to Hbp∆βcleavage and EV control OMVs (c.f Figure 4C, lanes 1-2 and lanes [3][4][5]. This could reflect the downregulation of OMP protein expression, but considering the smaller pellets obtained upon ultracentrifugation (data not shown), the impaired vesiculation by cells expressing the heterologous HbpD(∆d1)-fusion constructs appears a more likely explanation. Possibly, this observation is connected to the induction of cellular stress indicated by elevated levels of the periplasmic protease DegP ( Figure 4B, cf. lanes 1-2 and 3-5) [54]. In conclusion, display of truncated PmpD on Salmonella OMVs has been achieved but comes at the cost of decreased efficiency of OMV formation.

Display of Antigenic Passengers upon Seamless Fusion to the β-Helical Stem of Hbp
Progressive stacking of parallel β-strands into a β-helical structure is thought to contribute to the driving force for translocation of AT passengers across the OM [55]. One explanation for the observed suboptimal display of heterologous passengers upon fusion to HbpD(∆d1), may be that β-helix formation is either interrupted or completely disturbed at the fusion point, which is a linker predicted to be unstructured ( Figure S2B). To address this potential incompatibility with efficient folding during secretion, we used an alternative fusion approach that aimed to connect the β-helical PmpD, Prn and VacA passenger segments directly to the β-helical core of Hbp. Alphafold2 was used to test various in silico fusions to identify transition points that are modelled to yield a seamless uninterrupted chimeric β-helix [44]. This resulted in three fusion constructs in which the secreted and antigenic passenger fragments of PmpD (aa68-629), Prn (aa35-350) and VacA (aa65-377) were combined with a display version of HbpD from residue 840 onward (indicated as HbpD(840), see Figure S5A). These seamless fusions were expressed in Salmonella SL3261 ∆tolRA ∆msbB cells. This yielded bands of~90 kDa for HbpD(840)-PmpD and 100 kDa for HbpD(840)-Prn bands on Coomassie-stained gels that were also detected by immunoblotting using Hbp antiserum ( Figure 5A,B, lanes 3-4). However, expression of the~100 kDa HbpD(840)-VacA construct was not detected ( Figure 5A,B, lane 5). Of note, HbpD(840)-PmpD expression in cells appeared more pronounced than the Hbp∆βcleavage control analyzed in parallel ( Figure 5A, lane 2) and, in line with an efficient biogenesis, no upregulation of the periplasmic stress factor DegP was observed ( Figure 5B, lane 3). On the other hand, HbpD(840)-Prn and -VacA expression did result in elevated levels of DegP, indicative of cellular stress due to impaired translocation of the fusions across the CEs ( Figure 5B, lanes 4-5).  2). Importantly, expression of HbpD(840)-PmpD in OMVs was highly efficient and was quantified to constitute~13% of the total OMV protein content ( Figure 5C,D, lane 3). In comparison, the Hbp∆βcleavage control reached a level of~26%, indicating that the presence of the PmpD segment still slightly hampered partitioning of the HbpD(840)-PmpD fusion into OMVs. Yet, considerable improvement over the HbpD(∆d1)-PmpD was achieved when regarding OMV yields and expression ( Figure 4C, lane 3). Finally, display of HbpD(840)-PmpD at the OMV surface was confirmed by its ProK accessibility ( Figure 5E,F, lane 5). Taken together, fusion to HbpD(840) allowed for efficient expression of the truncated PmpD antigen at the OMV surface. However, our results also indicate that we could not apply this approach as a generic fusion for other passenger fragments, such as Prn and VacA.

Discussion
In this work, we applied the HbpD display platform to successfully present a fragment of the C. trachomatis antigen PmpD at the surface of LPS-detoxified Salmonella OMVs. These recombinant OMVs are ready for use as an experimental mucosal vaccine in a mouse model. If effective, the OMV-based vaccine could be combined with other purified antigens such as MOMP to constitute a multivalent vaccine that benefits from the adjuvant properties of OMVs, a formulation strategy that is required for the effectivity of the 4CMenB vaccine (Bexsero, GSK) that protects against infections by N. meningitidis [20]. Importantly, the path towards construction of the recombinant OMVs provided insight in the experimental intricacies of expression of heterologous AT-derived antigens in bacterial OMVs.
Compared to other Gram-negative secretion systems, autotransport is relatively simple, featuring sequential transfer across the IM and OM involving the conserved Sec and Bam complexes, respectively [56,57]. Therefore, secretion or surface display of a specific AT could be feasible in any Gram-negative host, although species-specific targeting signals for the Sec and Bam machinery may influence the efficiency of the secretion process. Indeed, we achieved efficient expression and OM assembly in E. coli of C. trachomatis AT PmpD (FL), in which the C-terminal β-signal was optimized for recognition by the E. coli Bam complex. Expression did not increase in the BL21(DE3)omp8 strain, which lacks multiple highlevel OMPs, indicating that the availability of the E. coli Bam complex is not a bottleneck for PmpD (FL) localization, or some other heterologous OMPs [40,58,59]. Surprisingly, however, PmpD (FL) was not included in OMVs pinching off from the BL21(DE3)omp8 surface ( Figure 1E). This strain is hypervesiculating due to the absence of OmpA that normally provides anchoring of the OM to the periplasmic peptidoglycan layer. While it is known that the partitioning of OMPs into OMVs is not an entirely random process, the total exclusion observed is most likely related to structural features in PmpD (FL) that preclude its recruitment in OMVs. Possibly, a difference in thickness of the hydrophobic core of the C. trachomatis and E.coli OM bi-layers may constitute a mismatch between the relatively extended β-barrel structure in PmpD (FL) [60] and the thinner hydrophobic core of the E. coli OM, while being required for assembly into its native OM. Although little is known about the mechanism of formation of bacterial OMVs, this putative mismatch may cause a local inspissation of the E. coli OM that could interfere with OMV development.
To address this issue, we changed our strategy and fused a large immunogenic fragment of the extracellular passenger of PmpD via a linker to the E. coli AT Hbp, which has been applied before to display antigens on OMVs. The passenger of Hbp is anchored to the surface of OMVs by its cognate β-barrel domain as a result of an inactivated autoproteolytic cleavage site [22]. Despite the relatively large size of the PmpD fragment, we considered this approach promising in view of the conserved β-helical passenger structures shared between PmpD and Hbp ( Figure S2B). Furthermore, we also fused passenger fragments of B. pertussis Prn and H. pylori VacA to examine the general applicability of this approach. The data showed that the truncated PmpD was included and exposed at the surface of BL21(DE3)omp8 OMVs (Figure 3), consistent with the notion that the E. coli Hbp β-barrel anchor being better suited for recruitment to OMVs than the β-barrel of C. trachomatis PmpD. Similar results were obtained for the HbpD(∆d1)-Prn construct, but OMVs carried only low amounts of HbpD(∆d1)-VacA probably due to poor expression levels ( Figure  S4B). Overall, the results suggest that subtle differences in the selected AT fragments or the connection to the HbpD(∆d1)-based platform may strongly influence expression and assembly in the E. coli OM and OMVs. This unexplained complexity is in line with our previous studies using ATs for surface display of heterologous antigens [23,61]. Often, inefficient expression of fusion constructs was observed, concomitant with toxicity and translocation incompetence.
We next expressed the HbpD(∆d1)-fusion constructs in a relevant hypervesiculating and LPS-detoxified Salmonella Typhimurium strain that is used to make vaccine-grade OMVs [21], which resulted in adequate expression levels of the HbpD(∆d1)-fusions. However, inclusion of HbpD(∆d1)-PmpD in Salmonella OMVs was relatively inefficient, and even undetectable for HbpD(∆d1)-Prn and -VacA ( Figure 4C,D). Furthermore, unlike what was observed for BL21(DE3)omp8, expression of these HbpD(∆d1)-fusions in the Salmonella strain lowered the amounts of OMVs produced. This was unexpected, in view of the increased levels of periplasmic protease DegP, which indicates cell envelope stress, which normally induces OMV formation [62]. At present, we have no explanation for these, apparently strain-dependent, effects. In general, the E. coli and Salmonella strains used have a very similar genetic background, although it should be noted that the molecular basis of hypervesiculation of the strains differs. While hypervesiculation in BL21(DE3)omp8 is probably due to the absence of OmpA, the Salmonella strain lacks TolRA thereby disrupting a major link between the IM and OM [22,28].
In any case, expression of the HbpD(∆d1)-fusion constructs in the Salmonella vaccine strain was relatively poor. Possibly, the discontinuation of the β-helical stem structure by inclusion of the flexible linker between HbpD(∆1) and its N-terminally fused heterologous partner disrupts sequential stacking of β-strands at the cell surface. This mechanism is thought to be at least partially responsible for energizing the translocation process [53,60] and point mutations in the C-terminal part of the β-helical stem of Hbp, which disrupt the initiation of stacking, have been shown to negatively influence secretion [63]. This may result in the formation of translocation intermediates that elicit cell envelope stress, consistent with the upregulated levels of the protease DegP observed. In an attempt to provide a smoother structural transition between the fusion partner and the displaying constructs, Alphafold2 prediction was used to select fusions that are modelled to seamlessly stitch the β-strands of the heterologous passenger into the β-helical stem of the Hbp carrier ( Figure S5A). This approach appeared successful for the HbpD(840)-PmpD fusion, as implied by the decreased levels of stress and degradation observed, and the detection of higher levels of PmpD at the surface of OMVs ( Figure 5C,D). Furthermore, the efficiency of OMV formation appeared restored, providing a solid basis to evaluate the safety and immunogenicity of HbpD(840)-PmpD-containing OMVs as a mucosal vaccine in future studies. In marked contrast, the predicted seamless HbpD(840)-Prn and -VacA fusions still caused stress and resulted in lower production of OMVs that also did not contain detectable amounts of these fusions.

Conclusions
In conclusion, while tailoring the HbpD-fusion strategies may improve expression of individual AT antigens at the surface of OMVs, this has to be addressed on a case-to-case basis. In this study, six chimeric constructs were evaluated. Alternatively, a more random fusion approach combined with a screening for improved display could be considered, until a more rational approach is found, based on evidence as to why some fusions are successful and others not. As for non-AT antigens, SpyTag/SpyCatcher coupling is the safest bet for efficient display on OMVs, but this approach comes at the cost of a more complex, hence expensive manufacturing process.