Structural basis for the homotypic fusion of chlamydial inclusions by the SNARE-like protein IncA

Many intracellular bacteria, including Chlamydia, establish a parasitic membrane-bound organelle inside the host cell that is essential for the bacteria’s survival. Chlamydia trachomatis forms inclusions that are decorated with poorly characterized membrane proteins known as Incs. The prototypical Inc, called IncA, enhances Chlamydia pathogenicity by promoting the homotypic fusion of inclusions and shares structural and functional similarity to eukaryotic SNAREs. Here, we present the atomic structure of the cytoplasmic domain of IncA, which reveals a non-canonical four-helix bundle. Structure-based mutagenesis, molecular dynamics simulation, and functional cellular assays identify an intramolecular clamp that is essential for IncA-mediated homotypic membrane fusion during infection.

T he obligate intracellular pathogen Chlamydia trachomatis is the most frequent cause of bacterial sexually transmitted disease and infectious blindness worldwide, yet it is still considered a neglected disease pathogen by the World Health Organization 1 . Chlamydia's life cycle depends on the establishment of a fast-growing parasitic organelle inside the host cell called the "inclusion", the development of which is poorly understood. Inclusion membranes are decorated with 60 transmembrane Inc proteins that are known to directly interact with host cell components and play a critical role in sustaining Chlamydia's life cycle [2][3][4] . Despite their importance, Incs have remained relatively uncharacterized and little is known about their function at the molecular level. The low sequence conservation and minimal similarity that Incs share with other proteins limits the usefulness of conventional bioinformatic tools to predict their structure and function. The best-characterized chlamydial Inc is IncA. It contains two extended 3,4-hydrophobic heptad repeat segments similar to the coiled-coil regions of the eukaryotic "SNAREs" (soluble N-ethylmaleimide-sensitive factor attachment receptors), which are proteins involved in cellular transport and membrane fusion [5][6][7][8][9] . IncA is involved in the homotypic fusion of chlamydial inclusions 10,11 and is the best-characterized bacterial SNARE-like protein to date.
During infection, each Chlamydia bacterium establishes its own inclusion inside a host cell. At high multiplicities of infection (MOI), cells contain multiple inclusions that ultimately fuse together to form one large inclusion per cell. This homotypic fusion event is important for the pathogenicity of C. trachomatis because natural non-fusogenic IncA mutants are replicationdefective in humans and cause significantly milder disease compared with patients infected with normal fusogenic strains [12][13][14] . Measurements of relative chlamydial rRNA quantities in the multiplying organisms have also revealed that non-fusogenic strains grow more slowly than fusogenic strains of Chlamydia, which is supported by a reduced rate of protein synthesis and decreased multiplication efficiency 15 . A direct role for IncA in membrane fusion was demonstrated using microinjection of anti-IncA antibody during infection, which resulted in multiple inclusions that were unable to undergo homotypic fusion 10 . Cells infected with an IncA-deficient strain of C. trachomatis similarly displayed multiple inclusions at a high multiplicity of infection, further establishing that IncA is required for the homotypic fusion of inclusions 11,16 .
IncA localizes to the inclusion membrane where it can potentially interact with host and inclusion-associated proteins.
Yeast two-hybrid analysis demonstrated that IncA can bind itself 10 , while immunoprecipitation also showed that IncA associates in trans when present on opposite membranes in a cell 5,8 . In Chlamydia-infected HeLa cells, the expression of ectopic IncA on the endoplasmic reticulum (ER) membrane impacts inclusion integrity and ER morphology, suggesting that IncA present on the ER interacts homotypically with IncA expressed on the inclusion and induces the fusion of both compartments 5 . While it is well-established that IncA is involved in the homotypic fusion of inclusions, the molecular mechanism of its fusogenic activity remains unknown.
Here, we describe the crystal structure of IncA, which we probed using biophysical, computational, and functional methods. We demonstrate that IncA folds into a stable, non-canonical fourhelix bundle that is maintained as a monomer by intramolecular interactions. We also show that the monomeric conformation of IncA is critical for its activity during membrane fusion. Our work sheds light on a class of bacterial transmembrane proteins that control membrane fusion during infection, which is critical for Chlamydia pathogenicity.

Results
IncA folds into a non-canonical four-helix bundle. IncA is a protein of 273 amino acids composed of a short, cytoplasmic N-terminal moiety (residues 1-34), a bilobed transmembrane domain (residues , and a long cytoplasmic C-terminal domain that ends in a tail with low complexity (residues 247-273) (Fig. 1a). The bacterial SNARE-like domains are found in the cytoplasmic C-terminal domain 5,7 . To better understand the mechanism of IncA-mediated membrane fusion, we generated high-quality crystals of a chymotryptic fragment of IncA spanning residues 87-246 (IncA 87-246 ) ( Supplementary Fig. 1b) and determined a crystal structure of IncA 87-246 to an R work/free of 14.1/ 16.8% at 1.12 Å resolution ( Fig. 1b and Table 1). The electron density for IncA 87-246 , including the six N-terminal histidines of the affinity tag, is exceptionally clear ( Supplementary Fig. 2), consistent with the low B-factor of the crystal structure (~10.8 Å 2 ). We found that IncA 87-246 adopts an asymmetric and slightly blocky conformation, somewhat similar to a four-helix bundle (Fig. 1b). Both in crystal and in solution, IncA 87-246 exists as a monomer ( Supplementary Fig. 3a Fig. 6). IncA 87-246(G144A) crystallized in a triclinic space group with two IncA 87-246(G144A) protomers in the unit cell. Though this crystal form did not diffract as well as the previous one, we were able to collect 88.1% complete data to 1.95 Å resolution (Table 1) and determine an accurate atomic model of  (Supplementary Fig. 7). In both protomers, this region is not implicated in crystal contacts, which suggests IncA 87-246(G144A) may exist in solution in different microstates, also populated in the crystal lattice. Secondary structure superimposition of IncA 87-246 with the reference IncA 87-246(G144A) protomer A reveals the G144A mutation results in significant conformational changes throughout the molecule (RMSD 1.68 Å) (Fig. 2c), especially in the helix H D , which is shifted upwards in IncA 87-246(G144A) . Thus, the G144A mutation in the segmented helix B of IncA plays a global structural role in the architecture of the bundle.
To probe the second region of IncA that has a higherthan-average B-factor (Fig. 2a), we generated a mutant, IncA 87-246(polyA) , where eight residues in H D that make contact with the helices surrounding the H clamp are mutated to alanine, namely L233A, S234A, L237A, T238A, Q240A, I241A, Q244A, and R245A ( Fig. 2d and Supplementary Fig. 1d). Although these mutations result in a loss of 12 intramolecular bonds (Supplementary  Table 1). Interestingly, IncA 87-237 , a shorter deletion construct of IncA that lacks many of the same residues on Helix D mutated in IncA 87-246(polyA) ( Supplementary Fig. 1f), is also predominantly dimeric in solution 8 . Thus, when the intramolecular contacts generated by helix H D are broken, then a higher oligomeric species spontaneously forms, suggesting that the non-canonical conformation of the H clamp may function by locking the helical core of IncA in a monomeric conformation that prevents IncA from self-assembling on the inclusion membrane.
The IncA monomer is highly stable. To determine if the crystallographic structure of IncA 87-246 represents a metastable intermediate, we subjected the 1.12 Å crystallographic model of IncA 87-246 to equilibrium molecular dynamics simulations, along with models of IncA 87-237 and IncA 87-246(G144A) ( Supplementary  Fig. 1). These three systems were subjected to 4 µs of equilibrium sampling to explore conformational flexibility and stability of IncA, and to probe the effects of the hinge, H clamp , and the C-terminus of helix H D on dynamics and structure. Our simulations showed that the structure of IncA 87-246 is highly stable and is unlikely to be a metastable intermediate. This conclusion is supported by the rigid, unchanging conformation of IncA 87-246 throughout the equilibrium sampling at physiological conditions (310 K, 150mM NaCl) and over the entirety of the simulation (Supplementary movie 1). In contrast, IncA 87-237 , which lacks nine additional C-terminal residues, underwent a structure-wide increase in root-mean-square fluctuation (RMSF) compared to both IncA 87-246 and IncA 87-246(G144A) (Fig. 3a), and experienced an eventual conformational change after~3.2 µs of sampling ( Fig. 3b and Supplementary movie 2). This conformation change involved a repositioning of H clamp away from helix H D , toward helix H A , where it ultimately formed H-bonds with residues 94 to 100 of H A , hereinafter referred to as H A(94-100) . Hydrogen bonds were identified using a contact-analysis with a cutoff at 3.2 Å, between hydrogen, oxygen, and nitrogen atoms of the H clamp (res. 165-169) and H A(94-100) . This hydrogen bond analysis was conducted for all three systems (   (Fig. 3a) and the dissimilarity in the position of H D among IncA 87-246(G144A) and IncA 87-246 (Fig. 2c), it is evident that the orientation of helix H D with respect to H clamp is coupled to the overall stability of each construct. Closer proximity between H clamp and H D in IncA 87-246(G144A) prevents the H clamp from forming hydrogen bonds with H A(94-100) resulting in a more stable conformation (Fig. 3c). The H clamp of IncA 87-246 , with a significantly higher RMSF than that of IncA 87-246(G144A) , forms hydrogen bonds with H A(94-100) after 2 μs of sampling. While less-numerous and denoting no clear structural significance compared with those of IncA 87-237 , the lack of hydrogen-bonding in IncA 87-246(G144A) implies that the position of the C-terminus of H D relative to the H clamp directly modulates the behavior of H clamp . This finding suggests that the overall flexibility of the H clamp , potentially its ability to recognize regulatory proteins, is sensitive to the network of hydrogen bonds it can form with adjacent regions of H D and H A .
Markov State models (MSMs) provide information regarding the thermodynamic stability of a protein and are typically used to characterize long timescale dynamic modes 19 , such as folding coordinates. For the trajectories in the present study  Prior to calculating RMSF, each of the three trajectories was aligned by minimizing the root-mean-square deviation (RMSD) of backbone atoms between the equilibrated structure and every subsequent frame. b Three snapshots (t = 0 µs, t = 2.1 µs, t = 3.8 µs) of the observed conformational change during the IncA 87-237 simulation. The two regions of interest, H clamp and H A(94-100) , are colored in cyan. c Bar plot of the number of hydrogen bonds occurring between the H clamp and H A(94-100) over the course of all three simulations. Hydrogen bonds were identified using a contact-analysis between hydrogen, oxygen, and nitrogen atoms within a cutoff of 3.2 Å representing an interval of 100 ps), coupled with an appropriate lag time of 10 ns, resulted in four-and five-state models for IncA 87-237 and IncA 87-246 , respectively. Models were scored and ranked for IncA 87-237 and IncA 87-246 separately, according to the generalized mixed Rayleigh quotient (GMRQ); the latter, a scalar value measuring the fit quality of a given MSM's learned dynamical eigenvectors against its input data (training score) or new data (testing score) 20 . Importantly, the resulting Markov states (shown in Supplementary Fig. 9a) were different for IncA 87-237 and IncA 87-246 . In particular, the resulting MSM of IncA 87-246 describes just four states that are structurally similar beyond fluctuation at the termini and H clamp , in contrast to the five models observed for IncA 87-237 , which demonstrate the construct's flexible conformation observed during simulation. For both of the top-scoring models, the longest implied relaxation time was <200 ns ( Supplementary Fig. 9a), asserting that each of the identified Markov state interconverts on relatively short timescales. The free energy landscape ( Supplementary Fig. 9b) comparing the two MSMs shows the RMSD of each state mapped to its computed free energy (ΔG), illustrating that the sampled states of IncA 87-246 belong to a single free energy basin. This information, coupled with the high structural similarity of each full-length Markov State and associated relaxation times, draws us to conclude that the structure of IncA 87-246 represents a thermodynamic minimum ( Supplementary Fig. 9c).
Simulation of IncA 87-246(G144A) revealed a spike in RMSF at residue 231 (Fig. 3a), a feature which is absent in both the IncA 87-246 and IncA 87-237 RMSF profiles. This result agrees with the structural plasticity of IncA 87-246(G144A) protomers seen in the triclinic crystal form (Table 1), which deviate significantly at the C-terminus of helix H D (Fig. 2c). Interestingly, the relative spike in RMSF of the H clamp region is substantially lower in IncA 87-246(G144A) than the other two structures. The repositioning of H D that results from the G144A mutation (Fig. 2c) and its subsequent effect of lowering the flexibility of the H clamp could possibly explain why IncA 87-246(G144A) was unable to oligomerize in solution during experiments. Altogether, these results indicate that the structure of IncA 87-246 is highly stable and is not likely to spontaneously unfold to engage in homotypic fusion. Evidence for the importance of the C-terminal helix H D in regulating IncA 87-246 stability is illustrated through the simulation of the shorter IncA 87-237 , which underwent a conformational change, and subsequent comparative analysis of the trajectory against IncA 87-246 and IncA 87-246(G144A) . For all three structures, the H clamp is a region of high flexibility, suggesting that it may play a role in the function of IncA. However, when G144 of the hinge is mutated to alanine, the H D -H clamp distance narrows and results in a significantly less flexible H clamp during simulation compared with IncA 87-246 and IncA 87-237 , which may explain experimental results regarding the lack of IncA 87-246(G144A) oligomerization in solution (Supplementary Fig. 3 Supplementary Fig. 1) during Chlamydia infection. We complemented an IncA knock-out (KO) Chlamydia strain 11 with the different IncA mutants. All IncA mutants expressed the intact N-terminal transmembrane domain (residues 1-86) for efficient secretion to the inclusion membrane, as well as a C-terminal FLAG-tag for rapid identification (see Supplementary Table 3). Expression levels for each IncA mutant -FLAG protein were comparable as validated by western blot analysis ( Fig. 4a and Supplementary Fig. 10). HeLa cells were infected with these strains at an MOI of 5, and the homotypic fusion of inclusions was quantified 24 h postinfection (hpi). IncA KO Chlamydia complemented with IncA WT undergoes comparable levels of homotypic fusion as wild-type Chlamydia ( Supplementary Fig. 11) and was used as a positive control, while the IncA KO strain was used as negative control. As shown in Fig. 4b, c, cells infected with Chlamydia expressing IncA 1-246 , which lacks 27 C-terminal residues ( Supplementary Fig. 1), had 23% fewer cells containing single, fused inclusions compared with IncA WT . While this decrease is statistically significant, it demonstrates that the protease-resistant fragment IncA 1-246 remains largely fusioncompetent. When the C-terminus was further shortened, the fusogenicity of IncA continued to drop with the deletion of only three additional residues (IncA 1-243 ) resulting in a drastic 43% inhibition of fusion. When six additional residues were deleted (IncA 1-237 ), we observed~82% inhibition of fusion (Fig. 4c), demonstrating a nearly complete loss of fusogenicity. Since IncA 87-237 is mostly dimeric in solution 8 , this finding suggests that the self-assembly of IncA could impact its fusogenic activity. Altogether, these data reveal that progressive deletion of the C-terminal residues of helix H D in IncA results in an increased loss of fusogenicity, which correlates with a shift towards self-assembly and a loss of monomeric IncA in solution.
To further address the potential impact of IncA oligomerization on homotypic fusion, we complemented the IncA KO Chlamydia strain with IncA 1-246(polyA) , in which the residueforming contacts on helix H D have been mutagenized. We observed that IncA 1-246(polyA) forms oligomers in solution ( Supplementary Fig. 3e, f and Supplementary Table 1). We also observed that IncA 1-246(polyA) was unable to drive homotypic fusion efficiently, and only~33% of infected cells contained single inclusions (Fig. 4b, d). This loss of function was specific for these mutations. Cells infected with Chlamydia expressing another mutated IncA, IncA 1-246(G144A) , in which the hinge was genetically eliminated, were fully fusogenic and mostly displayed a single inclusion (Fig. 4b, d). Glycine 144 is not conserved amongst the fusogenic IncA proteins, further supporting our functional data, which indicate this particular amino-acid does not play a critical role in fusion.
Altogether, these data suggest that the intramolecular contacts generated by helix H D likely maintain IncA in a monomeric fusion-competent state. We observed that the IncA mutants that form oligomers in solution are all non-fusogenic when expressed on the inclusion membrane. Although the oligomerization state of IncA on the membrane is unknown, these results suggest that IncA self-association may lead to its inactivation.
Wild-type IncA rescues fusion of non-fusogenic IncA. To further probe the mechanism controlling IncA function and assess the importance of trans-interactions between IncA present on opposing membranes, we conducted a series of co-infection experiments in which cells were infected with both non-fusogenic IncA mutant complemented-IncA KO C. trachomatis expressing GFP (MOI 5) and wild-type C. trachomatis expressing mCherry (L2 mCherry , MOI 2) (Fig. 5). Wild-type C. trachomatis transformed with mCherry (L2 mCherry ) undergoes the same level of fusion as wild-type Chlamydia (Supplementary Fig. 11). We used the IncA KO Chlamydia strain complemented with IncA 1-246 , and the IncA KO (non-induced) strain as positive and negative controls, respectively. As expected, IncA 1-246 was able to promote fusion with IncA WT , resulting in a mixed inclusion containing both GFP and mCherry Chlamydia (Fig. 5a). Strikingly, the mutants for which homotypic fusion was significantly impaired (IncA 1-237 , IncA 1-243 , and IncA 1-246(polyA) ), were all able to promote "heterotypic" fusion when IncA WT was present on the opposing membrane, thus resulting in complete rescue with almost all of the cells (97%, 96%, 94%, respectively) displaying single red wild-type inclusions that fused with green mutant inclusions (Fig. 5), compared with far fewer of the cells (~18%, 55%, and 35%, respectively) having homotypically fused inclusions (Fig. 4). These results suggest that IncA WT was able to form trans-interactions, likely by disassembling cis-oligomers (see Discussion). Inclusions generated by the IncA KO Chlamydia strain were unable to undergo fusion even with WT inclusions (Fig. 5a, b, non-induced). This observation confirms that IncA needs to be present on both inclusion membranes to promote fusion. Altogether, these data suggest a model whereby IncA WT is able to disassemble cis IncA mutant oligomers to promote IncA WT : IncA mutant fusion.

Discussion
Fusogenic viral proteins 21 and eukaryotic SNAREs [22][23][24] have been extensively studied, and much is known about the key structural determinants and, in some cases, mechanisms of membrane fusion. A common denominator of fusogenic proteins is their intrinsic structural plasticity and ability to undergo dramatic conformational changes upon fusion, typically adopting a metastable conformation in the pre-fusogenic state and a thermodynamically stable structure post-fusion 25,26 . In contrast, the key players and mechanisms that mediate fusion of chlamydial inclusions remain largely unknown. Chlamydial inclusions are extraordinarily challenging to study due to the fragility of the lipid membrane, the poor conservation of Inc proteins, and the lack of an in vitro fusion assay. In addition, the homotypic fusion of chlamydial inclusions presents a formidable topological challenge because the proteins responsible for this event are identical on both membranes 27 . In this paper, we describe the molecular Our work sheds light on three aspects that are important for deciphering the mechanisms of chlamydial inclusion membrane fusion. First, the cytosolic domain of IncA folds into a noncanonical four-helix bundle characterized by a segmented helix H B , a clamp, and a long C-terminal helix H D . This structure differs profoundly from the four-helix bundle structure of eukaryotic SNAREs previously used to describe IncA topology ( Supplementary Fig. 12) but shares similarity with the THATCH domain core of the Huntingtin Interacting Protein 12 that mediates associations between actin and clathrin-coated structures.
Combining biochemical, biophysical, and functional methodologies, we provide evidence that the C-terminal helix H D of IncA makes key intramolecular contacts with H clamp and helix H B , thus locking the IncA bundle into a stable monomeric prefusion conformation. These contacts, in particular, S234, Q240, I241, and Q245, are conserved in all known fusogenic IncA proteins (i.e., CtrIncA expressed by C. trachomatis, CmuIncA expressed by C. muridarum, and CsIncA expressed by C. suis), supporting the idea that this region is important for IncA function. Second, using molecular dynamics simulations, we establish that the crystallographic structure of the IncA cytosolic core is thermodynamically stable and unlikely to undergo major tertiary structural conformational changes spontaneously. Local conformational changes in IncA occur during membrane fusion, but they likely remain confined to regions of the bundle characterized by higher RMSD, such as the H clamp and the C-terminal helix H D . Third, by assessing the fusion of the inclusions in HeLa cells infected with genetically manipulated Chlamydia strains, we establish a direct correlation between the monomeric state of IncA in solution and its fusogenic activity. Deletion of just nine amino acids of the C-terminal helix H D results in the formation of stable IncA dimers in vitro and the drastic loss of fusogenic activity during infection. This inhibition of fusion was nonetheless completely rescued when cells infected by Chlamydia-expressing IncA mutants were co-infected with Chlamydia-expressing IncA WT , suggesting that the full-length bundle exerts a dominant function in trans to restore the fusogenic activity of IncA mutants .
Using these data, we postulate a model in which IncA WT promotes the fusion of inclusion membranes by engaging in homotypic interactions in trans (i.e., trans homodimer in Fig. 6) mediated by the C-terminal helix H D . Such an interaction is unlikely to be a simple swap of the C-terminal helix because molecular dynamic simulations argue against a global unfolding of the IncA four-helix bundle. Instead, this interaction appears limited to a molecular contact between two IncA bundles. The ability of IncA WT to rescue IncA mutants in trans suggests that the association between IncAs is not restricted to the C-terminal helix and may use determinants in the H clamp .
The non-canonical helical structure of IncA is key to solving the topological conundrum of homotypic fusion. In this context, the intramolecular contacts within IncA C-terminal helix H D and H clamp are essential for its fusogenic activity as it allows for homotypic interactions to occur only in trans. By removing the intramolecular interactions through point mutations (IncA 1-246(polyA) ) or truncation (IncA 1-237 ), we show that IncA mutants self-assemble in solution, likely leading to the formation of cis homodimers on the inclusion membrane ( Supplementary Fig. 13). Consequently, the cis homodimers cannot interact with IncA in trans and thus are unable to promote homotypic fusion. This model is supported by a previous analysis of the oligomeric state of IncA during infection, which found that IncA can naturally switch between monomeric and oligomeric states 5,8,10 , but we cannot rule out the existence of an additional host or Chlamydia factor that assists in the fusogenic bundling of IncA. One possibility is that a soluble or membrane-bound regulatory protein could catalyze the formation of the IncA trans-complex, leading to membrane fusion. It is also possible that the unstructured moiety at the C-terminus of IncA helix H D or perhaps an intramolecular interaction between the helical bundle and the N-terminal membrane-embedded hairpin stabilizes the auto-inhibitory state that keeps IncA monomeric and competent for fusion.
In summary, the structure of the cytoplasmic domain of IncA described in this study sheds light on how this unique bacterial protein engages in functional complexes to control membrane fusion. Our studies also provide an invaluable template for homology modeling of Inc-orthologs from other fusogenic and non-fusogenic Chlamydia strains 10,28 , and will help guide us towards a more mechanistic understanding of how mutations and sequence insertions in IncA affect its fusogenic activity. b Quantification of heterotypic fusion between L2 mCherry and IncA mutant inclusions. Data were normalized to IncA WT -FLAG. Graphs display the means of three independent experiments ± the standard deviation. Asterisks (***) denote a p-value < 0.001 (two-tailed student t-test). Source data are provided as a Source Data file Methods Gene cloning and mutagenesis. All primers and plasmids used in this study are detailed in Supplementary Tables 4 and 5, respectively. IncA 87-246 -His 6x (FD578) was constructed by PCR amplification of the region Thr87 to Lys246 using primers FO515/FO516 and full-length IncA as a template. The PCR product was digested with NcoI and XhoI and ligated into pET28a. IncA 87-246(G144A) -His 6x (FD915) was made using the QuikChange mutagenesis kit (Agilent) and the primers FO1067/ FO1068 following the manufacturer instructions. IncA 87-246(polyA) -His 6x (FD985) was constructed by PCR amplification of the region spanning Thr87 to Lys246 using primers FO1116/FO1185 and a synthetic gene containing the alanine mutations (GenScript) as the template. The PCR product was digested with NcoI and XhoI and ligated into pET28a. IncA 1-246 -FLAG (FD930) was constructed by PCR amplification of the region spanning Met1 to Lys246 using primers FO1023/ FO1024 and full-length IncA as the template. The PCR product was digested with NotI and SalI and ligated into pBOMB3-Tet. The pBOMB3-Tet plasmid (FD929) was constructed by removing the tetracycline promoter from pBOMB4-Tet with XhoI and NotI followed by ligation into pBOMB3 (FD923) 29 . IncA 1-243 -FLAG (FD944) was constructed by PCR amplification of the region spanning Met1 to Leu243 using primers FO1023/FO1143 and full-length IncA as template. The PCR product was digested with NotI and SalI and ligated into pBOMB3-Tet. IncA 1-246 (G144A) -FLAG (FD945) was constructed by PCR amplification of the region spanning Met1 to Lys246 using primers FO1023/FO1024, and full-length IncA served as the template. The PCR product was digested with NotI and SalI and ligated into pBOMB3-Tet. Next, the G144A mutation was introduced using the QuikChange mutagenesis kit (Agilent) and the primers FO1067/FO1068 following the manufacturer instructions. IncA 1-246(polyA) -FLAG (FD948) was constructed by PCR amplification of the region spanning Met1 to Lys246 using primers FO818/FO1184 and a synthetic gene containing the alanine mutations (GenScript) as template. The PCR product was digested with NotI and SalI and ligated into pBOMB3-Tet.  , 75 mM, and 100 mM). Proteins were eluted with resuspension buffer containing 400 mM imidazole and concentrated in centrifugal concentrators (3000 Dalton cutoff). Purified proteins were resolved on a HiLoad 16/60 Superdex prep-grade 200 column (GE Healthcare) equilibrated with their indicated buffers without glycerol containing 0.5 mM TCEP. The fractions containing protein were collected and concentrated. The purity of each protein was assessed by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and Coomassie Blue staining. Protein concentrations were determined by measuring the absorbance at 280 nm.
Complementation of IncA KO C. trachomatis with IncA mutants. IncA KO C. trachomatis L2 and the IncA KO complemented with IncA WT -FLAG and IncA (1-237) -FLAG were previously generated 11 . To complement IncA KO Chlamydia with IncA-FLAG mutants, 5 x 10 7 IncA KO elementary bodies and 10 µg of unmethylated DNA were mixed with 100 µl of 2x transformation buffer (20 mM Tris, pH 7.5, 100 mM CaCl 2 ) and brought up to 200 µl with sterile water. Transformations were incubated at room temperature for 35 min and then mixed with 13 ml of cell culture medium (Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% calf serum, 2 mM L-glutamine, non-essential amino acids, 1 mM sodium pyruvate). Two milliliters of transformation was added to each well of a 6-well plate containing confluent Vero cells. Antibiotic selection was initiated at 18 hpi with 100 ng ml −1 chloramphenicol. At 46 hpi, the cells were lysed in sterile water and the lysate was used to infect four T 175 flasks containing confluent Vero cells. The chloramphenicol concentration was increased to 200 ng ml −1 and 1 µg ml −1 cycloheximide was added. At 46 hpi, the cells were lysed in sterile water and the lysate was used to infect two T 175 flasks containing confluent Vero cells. The chloramphenicol concentration was increased to 400 ng ml −1 , and cycloheximide was maintained at 1 µg ml −1 . At 72 hpi, the cell lysate was used to infect one T 175 flask containing confluent Vero cells. The transformations were subsequently harvested every 46 hpi and the lysate was used to infect one T 25 flask containing confluent Vero cells until the flask was 30%-60% infected. The selection medium contained 400 ng ml −1 chloramphenicol and 1 µg ml −1 cycloheximide for all remaining passages. Individual clones were isolated by limited dilution 30 on Vero monolayers for 7-10 days in cell culture medium supplemented with 400 ng ml −1 chloramphenicol and 1 µg ml −1 cycloheximide. To confirm IncA-FLAG expression, cells were infected with clonal isolates for 24 h at 37 o C, fixed, labeled with anti-FLAG antibody and analyzed by immunofluorescence microscopy. IncA-FLAG expression was induced with 10 ng ml −1 anhydrotetracycline at 6 hpi. Selected clones were then purified by density gradient centrifugation 31 . Briefly, infected cells were harvested in K36 buffer (50 mM K 2 HPO 4 , 48.9 mM KH 2 PO 4 , 100 mM KCl, 14.9 mM NaCl, pH 7.0) and lysed by sonication. Lysates were clarified by centrifugation at 500 x g for 15 min at 4°C and the supernatants were layered on top of a 30% renografin (Mallinckrodt) cushion, before being centrifuged at 40,000 x g for 30 min at 4°C. Each pellet was then resuspended in K36 buffer and layered on top of a discontinuous renografin gradient consisting of 54% (bottom layer), 44% (middle layer), and 40% (top layer) renografin solutions. All renografin solutions were prepared in K36 buffer. The gradients were centrifuged at 40,000 x g for 1 h at 4°C. The band at the 54%/44% interface was harvested, diluted in SPG (17.4 mM Na 2 HPO 4 , 2.6 mM NaH 2 PO 4 , 4.9 mM L-glutamic acid, pH 7.4) and pelleted by centrifugation at 40,000 x g for 30 min at 4°C. The pellet was then resuspended in SPG and stored at −80°C. A summary of all the Chlamydia strains used in this study is presented in Supplementary Table 3. Following several washes with TBS-T, the blots were incubated with anti-mouse and anti-goat HRP-conjugated secondary antibodies (Pierce) diluted in TBS-T containing 0.5% low-fat milk powder for 1 h at room temperature. Blots were then washed several times with TBS-T and revealed using SuperSignal West Dura substrate (Pierce).
Homotypic fusion assay. In all experiments, 6 x 10 4 HeLa cells (ATCC) were seeded on glass coverslips 24 h prior to being infected with wild-type C. trachomatis or various C. trachomatis mutants at a MOI of 5 in DMEM containing 10% calf serum, 2 mM L-glutamine, 1 mM sodium pyruvate, non-essential amino acids, and 10 µg ml −1 gentamicin. The infection was synchronized by centrifugation for 1 h at 1000 x g (20°C). The infected cells were then transitioned to 37°C and 5% CO 2  Triton X-100. Next, coverslips were stained with mouse anti-FLAG (Sigma, F180) and goat anti-MOMP (ViroStat, 1621) primary antibodies for 1 h at room temperature (RT). After several washes, the coverslips were incubated with secondary antibodies donkey anti-mouse Alexa Fluor 555 (Invitrogen) and donkey anti-goat Alexa Fluor 488 (Invitrogen) and 5 µg ml −1 Hoechst (Invitrogen) for 1 h at room temperature (RT). The coverslips were washed and mounted using ProLong Diamond AntiFade Mountant (Invitrogen). Single and multiple inclusions were quantified in a total of 250 infected cells for each Chlamydia strain. Images were acquired with a Nikon TiE inverted epi-fluorescence microscope equipped with a 60x oil-immersion objective and Nikon Elements AR software. The percentage of cells with single inclusions was calculated and normalized to either the IncA KO complemented with IncA WT -FLAG or IncA 1-246 -FLAG as described in the figure legend. Statistical significance was measured using a two-tailed student t-test.
Heterotypic fusion (co-infection) assay. HeLa cells were co-infected with mCherry-expressing Chlamydia (L2 mCherry , MOI 2) and IncA mutant -FLAG complemented-IncA KO Chlamydia (MOI 5), which constitutively expresses GFP. The infection was synchronized by centrifugation for 1 h at 1000 x g (20°C). The infected cells were then transitioned to 37°C and 5% CO 2 for the duration of the experiment. IncA mutant -FLAG expression was induced at 5 hpi as described above. DMSO (non-induced), which is effectively the IncA KO, served as negative control. All steps of data indexing, integration, and reduction were carried out using HKL2000 32 , HKL3000 33 , and CCP4 programs 34 . The structure was solved by SAD using an ultra-redundant dataset (i.e., overall redundancy >16) collected in-house. Sixteen iodine sites were located by phenix.autosol 35 and used to calculate an initial set of SAD phases with Figure of Merit (FOM) equal to 0.37 between 15-1.8 Å resolution. An initial electron density map calculated with SAD phases was improved by solvent flattening and histogram matching, auto-built alternating cycles of automated model building using phenix.autobuild 36 and manual rebuilding using COOT 37 . The completed model was then subjected to positional and anisotropic B-factor refinement to the highest resolution available (~1.12 Å) using phenix.refine 38 and subjected to final re-refinement using PDBredo 39 , which yielded the best R work /R free and stereochemistry. The triclinic structure of IncA 1-246(G144A) was solved by molecular replacement using the wildtype IncA 1-246 structure as the search model, as implemented in PHASER 40  Analytical ultracentrifugation sedimentation velocity assay. AUC-SV analysis was carried out in a Beckman XL-A Analytical Ultracentrifuge. IncA samples were dissolved at 75 μM and 150 μM (corresponding to 1.5 and 3 mg ml −1 ) in 20 mM HEPES pH 7.5, 150 mM NaCl, 1 mM DTT and were spun at 45,000 rpm at 4°C. Absorbance values between 285 nm were fit to a continuous sedimentation coefficient (c(s)) distribution model in SEDFIT 45 . Data were visualized and presented using GUSSI (University of Texas Southwestern Medical Center).

Molecular dynamics simulations.
Simulations of IncA 87-246 , IncA 87-237 , and IncA 87-246(G144A) were performed using NAMD2.12 46 and the CHARMM36m protein forcefield 47 . To prepare these structures into models for equilibrium sampling, they were solvated with the TIP3P water model using the Solvate plugin in VMD 48 and each system was charge-neutralized and ionized with NaCl to a concentration of 150 mM using VMD's AutoIonize plugin. Each model was then subjected to 10,000 steps of gradient minimization. After minimization, harmonic restraints were applied to the protein backbone with a force constant of 5 kcal mol −1 during gradual thermalization of the system: the initial temperature was set to 50 K and then incrementally increased by 5 K every 10 ps. Once the systems reached 310 K, the last phase of model preparation was a gradual release of the restrained protein backbone atoms. Ten iterations were employed, removing 0.1 kcal mol −1 of the harmonic restraint's force constant after intervals of 50 ps. Equilibrium simulations of the prepared models were run on the NCSA Blue Waters supercomputer and TACC Stampede2. Full electrostatics were calculated at every other timestep, while non-bonded interactions were calculated at every timestep. The Particle Mesh Ewald (PME) method was used to calculate long-range electrostatic interactions. Short-range non-bonded interactions were calculated within a cut-off of 12 Å. Bonds to hydrogen were constrained using the SHAKE and SETTLE algorithms for protein and solvent, respectively. Temperature regulation was accomplished using the Langevin thermostat method, with a damping coefficient of 1.0 ps −1 . System pressure was maintained at 1.0 bar using the Nose-Hoover Langevin piston barostat. Each equilibrium simulation was run for 4 µs with a timestep of 2 fs. Trajectory analysis was conducted with VMD. Before processing, the three trajectories were aligned by minimizing the RMSD between backbone atoms of the initial equilibrated structure and all subsequent frames.
Markov state model analysis. Markov State Model (MSM) construction and analysis were conducted using the MSMBuilder 49 application. To prepare the IncA 87-246 and IncA 87-237 trajectories, each frame was aligned to the first in its series by RMSD between backbone atom coordinates. Water and ions were removed, and copies were saved every ten frames, with each frame in both of the over 40,000-frame datasets representing a 100 ps interval between samples. Prior to constructing MSMs, the trajectories were clustered by RMSD between atomic coordinates according to the k-medoids algorithm; loop regions were ignored. The initial choice of k-value was determined by a parameter sweep against the resultant within-cluster sum of squares, or inertia, as a metric of fit quality. The selection criterion for optimal k was based on the "Elbow method," where returns on fit quality diminish steeply at k + 1. This pointed to an optimal value of k = 6. Owing to both the limitations of this approach and the sensitivity of the resulting MSMs to this parameter, the range of tested k values was k = {4,8}, followed by sparse sampling up to k = 4000 for both systems. After, MSMs were built from the clustered trajectories, testing differing model lag times using maximum likelihood estimation and (N states −1) dynamical timescales for different input k-clusters. The resulting models of IncA 87-246 and IncA 87-237 were scored based on their generalized matrix Rayleigh quotient 20 (GMRQ) training and testing scores. The optimal lag time (τ) for both IncA 87-246 and IncA 87-237 models was determined to be τ = 10 ns and was used for all reported results. For the MSMs shown in Supplementary Fig. 9, panels (a) and (b): k = 6 clustered input trajectories for IncA 87-237 and IncA 87-246 ; panel c: k = 500 clustered input trajectory for IncA 87-246 only. Free energies were calculated according to ΔG ¼ Àk b T log p, where k b is the Boltzmann constant (kcal mol −1 ) and p is the estimated equilibrium population of a given Markov state.
Thermal stability assay. The thermal stability of IncA 87-246 , IncA 87-246(G144A) , and IncA 87-246(polyA) was measured by recording changes in the ellipticity intensity at 220 nm as a function of temperature. Circular dichroism (CD) spectra were recorded using a Jasco J-810 spectropolarimeter equipped with a Neslab RTE7 refrigerated recirculator available at the Sidney Kimmel Cancer Center X-ray Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The atomic coordinates and structure factors for wild-type IncA 87-246 and IncA 87-246 (G144A) have been deposited in the Protein Data Bank with accession codes 6E7E and 6E6A, respectively. The authors declare that all other relevant data supporting the findings of this study are included in this published article and its Supplementary Information files, or from the corresponding authors upon request.