Mechanism underlying autoinducer recognition in the Vibrio cholerae DPO-VqmA quorum-sensing pathway

Quorum sensing is a bacterial communication process whereby bacteria produce, release, and detect extracellular signaling molecules called autoinducers to coordinate collective behaviors. In the pathogen Vibrio cholerae, the quorum-sensing autoinducer 3,5-dimethyl-pyrazin-2-ol (DPO) binds the receptor and transcription factor VqmA. The DPO-VqmA complex activates transcription of vqmR, encoding the VqmR small RNA, which represses genes required for biofilm formation and virulence factor production. Here, we show that VqmA is soluble and properly folded and activates basal-level transcription of its target vqmR in the absence of DPO. VqmA transcriptional activity is increased in response to increasing concentrations of DPO, allowing VqmA to drive the V. cholerae quorum-sensing transition at high cell densities. We solved the DPO-VqmA crystal structure to 2.0 Å resolution and compared it with existing structures to understand the conformational changes VqmA undergoes upon DNA binding. Analysis of DPO analogs showed that a hydroxyl or carbonyl group at the 2′-position is critical for binding to VqmA. The proposed DPO precursor, a linear molecule, N-alanyl-aminoacetone (Ala-AA), also bound and activated VqmA. Results from site-directed mutagenesis and competitive ligand-binding analyses revealed that DPO and Ala-AA occupy the same binding site. In summary, our structure-function analysis identifies key features required for VqmA activation and DNA binding and establishes that, whereas VqmA binds two different ligands, VqmA does not require a bound ligand for folding or basal transcriptional activity. However, bound ligand is required for maximal activity.

In this report, we characterize ligand-controlled activation of VqmA in V. cholerae. In contrast to canonical LuxR-type QS receptor-transcription factors, which typically require binding to their cognate ligands to fold and activate transcription (13)(14)(15)(16), here, we find that VqmA does not require DPO for folding or basal transcriptional activity but does require binding DPO to drive maximal activity. These same properties are relevant for the phage VP882-encoded VqmA homolog (VqmA Phage ), suggesting that ligand-independent activity is a conserved feature among VqmA proteins. We show that VqmA binds to and is activated by both DPO and the putative DPO intermediate, Ala-AA. We solved the crystal structure of VqmA bound to DPO. Comparison with existing structures (17,18) allowed us to provide insight into the conformational changes VqmA undergoes upon DNA binding. We were unable to obtain the crystal structure of VqmA bound to Ala-AA because Ala-AA spontaneously cyclizes into DPO under our crystallization conditions. Mutagenesis of the VqmA ligand-binding pocket shows that DPO and Ala-AA likely occupy the same site in VqmA, as substitutions in residues that alter DPO binding and activity have similar effects on Ala-AA binding and activity. Characterization of binding by isothermal titration calorimetry (ITC)-based competition experiments and kinetic studies reveals that VqmA prefers DPO over Ala-AA. Assessment of the activities of a panel of DPO analogs revealed that a 2Ј-hydroxyl or -carbonyl group is required for binding to VqmA. Taken together, our results show that VqmA is promiscuous with respect to DPO and Ala-AA. Nonetheless, maximal VqmA activity depends on ligand binding. The ability of VqmA to activate gene expression to different levels in the unbound state and when bound to two different ligands could enable greater flexibility in regulating gene expression than exists in canonical LuxRtype QS receptors, which typically have no activity in the apo-state and show strict specificity for a single partner ligand.

VqmA possesses both ligand-independent and ligand-dependent activity
VqmA is a cytoplasmic transcription factor and QS receptor that binds the AI DPO, a hydroxypyrazine that requires the Tdh enzyme for its production (8). The VqmA receptor is restricted to vibrios and one vibriophage. By contrast, canonical LuxRtype cytoplasmic QS receptor-transcription factors exist in thousands of Gram-negative bacterial species, and they bind AIs that are acyl-homoserine lactones (AHLs) (reviewed in Ref. 1). AHL AIs are produced by LuxI-type AI synthases. LuxRtype receptors typically require binding to their cognate AIs to properly fold, become soluble, and activate transcription (13)(14)(15)(16). Thus, LuxI-type AI synthases and LuxR-type receptors function as obligate pairs, and mutation of the luxI gene confers the identical defect as does mutation of the luxR gene (19). We wondered whether the Tdh-VqmA pair functions in an analogous manner. To probe this question, we constructed a reporter system to monitor activation of transcription by VqmA. We fused the lux (luciferase) operon to the vqmR promoter (PvqmR-lux) and integrated it into the V. cholerae chromosome at an ectopic locus (lacZ). We measured transcriptional activity in different V. cholerae strains and across different conditions. WT V. cholerae harboring the PvqmR-lux reporter exhibited Autoinducer recognition by the VqmA quorum-sensing receptor maximal activity at HCD, which is consistent with our previous report ( Fig. 1A) (8). The ⌬tdh mutant, which is defective for DPO production, elicited ϳ10-fold lower activity (Fig. 1A). Restoration of light production to approximately the WT level occurred in the ⌬tdh mutant following the exogenous addition of 1 M DPO (Fig. 1A). In the ⌬vqmA strain, PvqmR-lux levels were ϳ330-fold lower than in the WT and ϳ23-fold lower than in the ⌬tdh strain to which no DPO was added (Fig. 1A). The exogenous addition of DPO did not induce light production in the ⌬vqmA strain (Fig. 1A). Thus, unlike what is observed for canonical LuxI-LuxR-type systems, the ⌬tdh synthase mutant and the ⌬vqmA receptor mutant have different phenotypes with respect to QS output.
Two possibilities could underpin the differences in the lux phenotypes of the ⌬vqmA and ⌬tdh mutants: either DPO is produced via a Tdh-independent mechanism or VqmA displays basal activity in the absence of its ligand. To investigate the first possibility, we quantified DPO production by WT, ⌬tdh, and ⌬vqmA V. cholerae. Cell-free culture fluids prepared from each strain were supplied to the V. cholerae ⌬tdh PvqmRlux reporter strain, and light production was measured (Fig.  S1A). Activity present in the preparations was compared with that following the addition of known amounts of synthetic DPO. The EC 50 for synthetic DPO is ϳ1 M (8). Using this value, we could estimate that the WT and ⌬vqmA culture fluids contained ϳ1-2 M DPO, whereas the amount of DPO in culture fluid from the ⌬tdh strain was below the 60 nM detection limit of the reporter (Fig. S1A). Thus, if any DPO is produced by a Tdh-independent mechanism(s), it is insufficient to elicit the order-of-magnitude difference in PvqmR activity between the ⌬vqmA and ⌬tdh V. cholerae strains in Fig. 1A. To test the second possibility, that VqmA is expressed and possesses measurable activity in the absence of DPO, we produced VqmA in WT (i.e. tdh ϩ ) and tdh::kan Escherichia coli and assessed its solubility. E. coli, like V. cholerae, produces DPO in a Tdh-dependent manner (8). Fig. 1B shows that comparable amounts of soluble VqmA were obtained from the WT and tdh::kan strains, irrespective of the addition of DPO. To eliminate the possibility that a trace amount of DPO is produced by the tdh::kan strain or that some other ligand stabilizes VqmA in tdh::kan E. coli, we produced hexahistidine-tagged VqmA (His-VqmA) via in vitro translation using a cell-free system. In vitro translated His-VqmA protein was readily produced and soluble (Fig. S1B). Thus, we conclude that VqmA does not depend on a ligand to properly fold.
In V. cholerae, VqmA is produced constitutively, whereas vqmR is expressed predominantly at HCD (8). Given our finding that the solubility of VqmA is unaffected by DPO (Fig. 1B), we hypothesized that the VqmA DNA-binding capability could be enhanced by DPO. To examine VqmA DNA binding as a function of ligand occupancy, we purified VqmA from tdh::kan E. coli that had been grown in the presence or absence of exogenously supplied DPO (100 M). We assessed apo-and holo-VqmA binding to PvqmR DNA using electromobility shift assays (EMSAs). Our results show that, at sufficient protein concentrations, apo-VqmA shifted PvqmR DNA (Fig. 1C), consistent with our finding that VqmA possesses basal activation of PvqmR in the apo-state. We found that 2-8-fold more protein was required to achieve a ϳ50% shift of the PvqmR DNA with apo-VqmA compared with holo-VqmA (Fig. 1C). Thus, holo-VqmA has a higher affinity for PvqmR than apo-VqmA. Furthermore, the addition of DPO to apo-VqmA increased DNA binding to PvqmR in a DPO concentration-dependent manner (Fig. 1D). We conclude that the basal DNA-binding activity of apo-VqmA is enhanced when VqmA is bound to DPO. These results explain why the V. cholerae ⌬tdh mutant exhibits PvqmR-lux expression, whereas the ⌬vqmA mutant does not (Fig. 1A).
Considering the differences in the biochemical properties of the VqmA receptor relative to canonical LuxR-type QS receptors, we wanted to examine whether their control of QS outputs also differed. To do this, we measured the effect of DPO binding to VqmA on PvqmR activation, representing the V. cholerae VqmA pathway, and compared the results with the well-characterized LasI-LasR QS system from Pseudomonas aeruginosa. Briefly, in P. aeruginosa, upon binding 3OC 12 -homoserine lactone (3OC 12 -HSL), LasR activates transcription of many genes, including, germane to our studies, lasB (20). Here, we introduced the relevant P. aeruginosa and V. cholerae QS components into E. coli (Fig. 2). Specifically, we used ⌬tdh E. coli as the host for both reporter systems to eliminate native AI production (E. coli does not produce 3OC 12 -HSL). Into this strain, we introduced a pair of plasmids encoding either arabinose-inducible vqmA and PvqmR-lux or arabinose-inducible lasR and PlasB-lux. Consistent with our finding in ⌬tdh V. cholerae (Fig.  1A), E. coli producing VqmA in the absence of DPO generated 15-fold more PvqmR reporter activity than a strain lacking the vector harboring arabinose-inducible vqmA (Fig. 2). In response to exogenous DPO, PvqmR-driven light production . VqmA exhibits activity in the absence of its cognate AI, and the LasR QS receptor does not. Shown is normalized reporter activity from E. coli harboring the VqmA/PvqmR-lux or the LasR/PlasB-lux system. Left, E. coli carrying one plasmid with PvqmR-lux and a second plasmid with pBAD-vqmA. The strain represented by the bar designated Control lacks the pBAD-vqmA plasmid. Right, E. coli carrying one plasmid with PlasB-lux and a second plasmid with pBAD-lasR. The strain represented by the bar designated Control lacks the pBAD-lasR plasmid. All strains were grown in the presence of 0.075% arabinose and in the absence or presence of the designated AI (20 M DPO or 1 M 3OC 12 -HSL). AIs were added to the control strains. Data are represented as mean Ϯ S.D. (error bars) with n ϭ 3 biological replicates. AU, arbitrary units.

Autoinducer recognition by the VqmA quorum-sensing receptor
increased an additional 3-fold. LasR, by contrast, expressed minimal PlasB-lux production in the absence of 3OC 12 -HSL relative to the no-vector control. In response to 3OC 12 -HSL, a 143-fold increase in PlasB-lux expression occurred (Fig. 2). Taken together, these results show that both VqmA and LasR are highly activated by their respective AIs. However, VqmA activity, in contrast to LasR activity, is not strictly dependent on an AI. In the opposite vein, the relative activation achieved by 3OC 12 -HSL-LasR is almost 2 orders of magnitude greater than that for the DPO-VqmA complex.
We wondered whether the ligand-independent response we observed for V. cholerae VqmA extended to other VqmA proteins, such as that from the phage VP882-encoded VqmA homolog, VqmA Phage . VqmA Phage binds host-produced DPO and, in response, activates expression of the gene, qtip. Qtip promotes host cell lysis (10). Similar to V. cholerae VqmA, VqmA Phage activated Pqtip transcription in the absence of DPO, and activation was enhanced ϳ3-fold when DPO was added (Fig. S1C). VqmA Phage shares ϳ43% identity with V. cholerae VqmA, and the key residues required for ligand and DNA binding are highly conserved (10). These features likely underlie the similar behaviors in the reporter assays. Thus, the VqmA AI-independent function extends to the phage-encoded system and is likely an inherent property of VqmA proteins more generally.

The proposed linear DPO precursor Ala-AA binds to and activates VqmA
As noted, Tdh is required to convert L-threonine to AKB, which spontaneously decarboxylates to aminoacetone. Aminoacetone is presumed to condense with L-alanine to form the putative DPO precursor Ala-AA (8) (Fig. 3A). Consistent with the placement of threonine upstream and Ala-AA downstream of Tdh in this pathway, we previously showed that the exogenous addition of Ala-AA, but not threonine, restores PvqmRlux expression in ⌬tdh V. cholerae to WT levels (8). This finding suggested one of two possibilities. First, in our cell-based assay, the cells convert Ala-AA to DPO, and newly produced DPO drives PvqmR-lux expression; alternatively, Ala-AA could bind to VqmA and induce PvqmR-lux expression without conversion to DPO. To distinguish between these possibilities, we performed EMSAs using apo-VqmA to which we added either threonine, DPO, or Ala-AA (Fig. 3A). As anticipated, provision of L-threonine did not increase VqmA binding relative to when water was added (Fig. 3B). By contrast, inclusion of DPO increased the percentage of PvqmR DNA bound to VqmA ϳ2-5 fold (Fig. 3B). Surprisingly, the consequence of adding Ala-AA was indistinguishable from that of DPO (Fig. 3B) (i.e. Ala-AA also drives VqmA to bind PvqmR). In vitro translated His-VqmA behaved identically with respect to the ligands (Fig.  S2A). Importantly, we excluded the possibility that Ala-AA is Autoinducer recognition by the VqmA quorum-sensing receptor converted to DPO under our EMSA conditions using MS (DPO, 125 m/z; Ala-AA, 145 m/z). In each case, the ligand added, DPO or Ala-AA, was the only ligand detected by MS (Fig. S2B). Taken together, we conclude that both DPO and Ala-AA bind and enhance VqmA DNA binding and transcriptional activity.
To explore whether binding of DPO and Ala-AA to VqmA drives conformational changes and whether they are the same or different, we performed limited proteolysis. Apo-, DPObound, and Ala-AA-bound VqmA were digested with trypsin. The DPO-bound and Ala-AA-bound VqmA proteins were more susceptible to digestion than was apo-VqmA (Fig. 3C). One explanation for these results is that a conformational change occurs upon ligand binding and, moreover, that the change is similar for both ligands. Indeed, in companion experiments, we assessed the binding of DPO and Ala-AA to apo-VqmA using thermal shift analysis. Both ligands caused similar increases (about 6°C) in thermal stability of VqmA (Fig. S2C).
Our data support a model in which both DPO and the linear Ala-AA molecule bind VqmA and effectively activate its transcriptional activity.
The above results show that Ala-AA and DPO have identical activities in both our in vivo cell-based and in vitro proteinbased assays. One caution, however, is that Ala-AA is the presumed DPO precursor, but its in vivo role has not been verified. After DPO was discovered and threonine and alanine were shown to be incorporated into DPO, the prediction of Ala-AA as the DPO precursor was put forward based on common chemical reactions in which amino acids and amino-containing compounds participate (8). Here, we aimed to prove or disprove whether Ala-AA is indeed the biological precursor to DPO. To do this, we conducted isotopic labeling experiments to determine whether Ala-AA cyclizes into the final DPO product. We administered deuterium-labeled Ala-AA (d 3 -Ala-AA) ((S)-2-amino-N-(2-oxopropyl)-3,3,3-d 3 -propanamide, shown in Fig. S3A) to ⌬tdh V. cholerae and analyzed incorporation of the deuterium label into DPO by LC-MS. As controls, we also administered d 3 -Ala-AA to 1) medium alone to eliminate the possibility of spontaneous conversion in the absence of cells; 2) ⌬tdh ⌬vqmA V. cholerae to confirm that binding to VqmA is not required for conversion; and 3) WT V. cholerae to show that the amount of labeled DPO (d 3 -DPO) extracted is identical between the Tdh ϩ and Tdh Ϫ strains to verify that Tdh is not functioning in any additional downstream step in the pathway. We detected d 3 -DPO in extracts at ϳ5% of the total concentration of added d 3 -Ala-AA (Fig. S3B). However, analysis of our synthetic preparations of Ala-AA and d 3 -Ala-AA showed that ϳ5% DPO and d 3 -DPO, respectively, are present (Fig. S3C) as irremovable impurities. The presence of small amounts of DPO demonstrated the facile nature of Ala-AA to undergo cyclization under basic conditions. Thus, in the labeling experiments, the low level of d 3 -DPO present in the extracts can be attributed to the starting material, and it is not a consequence of enzymatic conversion of d 3 -Ala-AA to d 3 -DPO (Fig. S3C). There was no case in which administration of d 3 -Ala-AA to cells resulted in accumulation of recovered DPO with a 3-Da mass shift above the background level. This result suggests that, at least under our conditions, exogenously supplied Ala-AA is not converted into DPO in vivo (Fig. S3B). We suspect that some other compound could be the precursor to DPO, and we are working to discover its identity. Nonetheless, our findings with Ala-AA are curious, as they show that the VqmA receptor binds to and is activated by an apparently nonpathway substrate. We return to ideas concerning the ramifications of these findings under "Discussion."

Structural and biochemical characterization of VqmA
To explore the molecular basis underlying VqmA binding to ligands, we solved the crystal structure of VqmA bound to DPO in the absence of DNA to 2.0 Å using multiwavelength anomalous diffraction (Table S1). The structure of the DPO-VqmA complex bound to DNA was reported recently (17). Additionally, during preparation of this manuscript, the structure of DPO-VqmA without DNA was reported with a similar conformation as in our structure (18). Together, the three structures enabled comparisons to discover the conformational changes VqmA undergoes upon DNA binding and, in the context of our work, to gain insight into how VqmA could bind both DPO and Ala-AA.
The DPO-VqmA complex is a head-to-head dimer with 1:1 protein/ligand stoichiometry. The structure revealed two distinct domains: an N-terminal Per-Arnt-Sim (PAS) ligand-binding domain (LBD) and a C-terminal helix-turn-helix (HTH) DNA-binding domain (DBD) (Fig. 4A). The N-terminal PAS LBD consists of a characteristic five-stranded antiparallel ␤-sheet core, which forms the DPO-binding pocket along with several ␣-helices and loops, and this domain contains the bulk of the buried surface area of the dimerization interface (Fig. 4, B and C, respectively). The PAS domain is connected to the C-terminal HTH DBD by a long flexible loop region. The DBD has a stereotypical HTH tetrahelical fold, which is common in transcription factors, including LuxR-type QS receptors (Fig.  4D). Both the LBD PAS and DBD HTH domains are common across biology. However, according to the DALI server (21), which uses structures from the Protein Data Bank to assess structural homologies based on fold rather than amino acid sequence, protein structures containing a domain composition similar to VqmA have not otherwise been reported. The closest VqmA homologs based on fold either contain the PAS domain as part of a signal transduction protein (i.e. TodS and VraR) or the HTH domain as part of a transcription factor (i.e. CviR and VpsT), but not both (Table S2).
The VqmA HTH region is enriched in positively charged residues. Within this region, we predicted that the residues Lys-185, Lys-193, and Arg-195 are important for DNA binding due to their solvent accessibility and their conservation among VqmA orthologs (Fig. 4D). Indeed, as judged by EMSA analysis, changing any of these residues to alanine eliminated VqmA DNA binding (Fig. S4). This analysis is consistent with the structure of VqmA bound to DNA, which shows that residue Lys-185 makes direct contact with the DNA backbone (17). We compared our structure of DPO-VqmA with a recently reported structure of DPO-VqmA-DNA (17). Alignment based on the LBD PAS domains reveals a conformational change of ϳ30 Å and a rotation of nearly 120°for the DBD (Fig. 4E). Thus, the DBD of the DNA-free VqmA protein adopts an open con-Autoinducer recognition by the VqmA quorum-sensing receptor formation, apparently conducive to DNA binding, whereas DNA-bound VqmA exhibits a closed conformation. We note that our structure is consistent with a recently reported structure of DPO-VqmA ( Fig. 4F) (18), further supporting the above conclusion.
To probe the VqmA ligand binding requirements, we tested a panel of 15 synthesized or commercially available pyrazine, pyrimidine, and pyridine DPO analogs (Fig. 5A). We used the V. cholerae ⌬tdh PvqmR-lux reporter strain and companion ITC analysis to, respectively, assess activation of and binding to VqmA (Fig. 5B and Table S3, respectively). Surprisingly, given the promiscuity VqmA displays with respect to DPO and Ala-AA binding, VqmA is highly selective in terms of cyclic compounds: VqmA responded most avidly to DPO, and the most potent analogs, capable of inducing maximal lux expression, had EC 50 values 6 -8-fold higher than that of DPO (Fig.  5C). Our results indicate that a 2Ј-hydroxyl or 2Ј-carbonyl group is necessary for binding to VqmA (Table S3) and promoting its transcriptional activity (Fig. 5C). Indeed, compound 10, which differs only from DPO by the lack of the 2Ј-hydroxyl group, was inactive and showed no binding to VqmA ( Fig. 5B and Table S3).
Our crystal structure shows that DPO binding is primarily mediated bystacking forces contributed by residues Phe-67 and Phe-99 coupled with hydrogen bonding between the DPO 2Ј-carbonyl group and residue Lys-101 (Fig. 4B). We note that DPO exists as the enol form in solution (8). Here, we built the keto tautomer of DPO in our crystal structure because it best fits the local density. Probably, the difference between the restricted condition in the crystal and the unrestricted environment in solution drives the preference for a particular tau-tomer. To verify our structure and to explore how VqmA could bind to both DPO and to Ala-AA, we conducted site-directed mutagenesis and tested binding and activity of DPO and Ala-AA, using an in vitro ITC assay and our in vivo cell-based reporter assay, respectively. We determined that, for WT apo-VqmA, the K d for DPO is ϳ2.3 M, whereas the K d for Ala-AA is ϳ14 M (Fig. 6A). As a negative control, we performed ITC on a linear Ala-AA derivative called (2S)-2-amino-N-[(2R)-2hydroxypropyl]propanamide (AHP) (Fig. S5A) that did not activate VqmA in our PvqmR reporter assay (Fig. S5B), and no binding was detected by ITC (Fig. S5C). Next, we focused on mutations in the DPO-binding site: K23R, Y36F, F67A, D71N, D85N, H87A, H97A, F99A, K101R, F116A, F67A/F99A, and H87A/H97A (Fig. 4B). ITC results reveal that most of these VqmA variants exhibited approximately WT-binding affinity for both DPO and Ala-AA (Table 1). Exceptions were VqmA F67A, VqmA K101R, and VqmA F67A/F99A, which showed no binding to either molecule, indicating that these residues are crucial for binding to both DPO and Ala-AA (Table 1). Notably, VqmA F99A did not bind Ala-AA but retained residual binding to DPO (ϳ10-fold higher K d relative to WT), suggesting that the contribution of Phe-99 in theinteraction with DPO is less crucial than that of Phe-67. The finding that VqmA K101R was unable to bind DPO coupled with the structure-activity relationship analysis in Fig. 5 suggests that proper hydrogen bonding between Lys-101 and the DPO 2Ј-hydroxyl/carbonyl group is critical for ligand binding and activity. We suspect that introduction of the bulkier Arg at position 101 prohibits formation of the required hydrogen bond at the 2Ј-position of DPO. A hydroxypyrimidine (compound 7, Fig. 5A) reduces the basicity of the oxygen on the 2Ј moiety of the ligand, which would

Autoinducer recognition by the VqmA quorum-sensing receptor
weaken its interaction with Lys-101 relative to DPO. Indeed, compound 7 had an Ͼ800-fold increase in EC 50 and exhibited no detectable binding to VqmA (see Fig. 5C and Table S3). Finally, the VqmA Y36F and H97A alterations made the K d values for DPO and Ala-AA more similar, indicating that these residues are involved in dictating the natural preference VqmA shows for DPO over Ala-AA (Table 1).
The results from the reporter assay mimic those from the ITC assay (Fig. S6A). Fig. S6B shows that all of the VqmA variant proteins were produced at WT levels. Thus, changes in ligand-driven reporter activity stem from alterations in ligandbinding affinity, not changes in protein stability. Our structureguided mutagenesis and biochemical analyses suggest that DPO and Ala-AA occupy the same binding site on VqmA. To date, we have been unable to obtain crystal structures of apo-VqmA or VqmA bound to Ala-AA. Apo-VqmA crystals diffract poorly, and Ala-AA, although stable in solution, readily cyclizes under our crystallization conditions (high salt and alkaline pH), independent of VqmA.

Ala-AA and DPO share the same binding site on VqmA, with DPO being the preferred ligand
Based on the surprising result that both Ala-AA and DPO bind to and activate VqmA, we sought to further characterize their interactions with VqmA to substantiate the same-binding-site hypothesis raised from our site-directed mutagenesis experiments. To garner additional evidence that DPO and Ala-AA bind the same site on VqmA, we performed an ITC-based competition assay. We titrated either DPO or Ala-AA into VqmA that had been prebound to the other molecule and assessed binding. We found that binding of the exogenously added molecule occurred no matter which ligand was prebound ( Fig. 6B, left two panels), suggesting that the ϳ6-fold difference in K d between the two ligands allows each molecule to displace the other. In a conceptually similar experiment, we performed the same titration in the presence of excess of the prebound ligand. Our rationale is that, if Ala-AA and DPO do share the same binding site in VqmA, the presence of an excess of the ligand that had been prebound should inhibit binding of the subsequently added ligand. In this setup, VqmA showed a clear preference for binding to DPO. Specifically, when either an equal concentration or a 10-fold excess of DPO was provided to DPO-bound VqmA, Ala-AA was prevented from replacing prebound DPO (Fig. 6B, middle two panels). In the opposite case, only when there was at least 10-fold excess Ala-AA could Ala-AA prevent DPO from replacing prebound Ala-AA on VqmA (Fig. 6B, right two panels). These data support the hypothesis that DPO and Ala-AA share the same binding site on VqmA and further corroborate that, between the two ligands, DPO is preferred.
The preference VqmA shows for DPO over Ala-AA in the competitive binding experiments is curious, given that the K d values of the two ligands for the apoprotein are only ϳ6-fold different. To better understand VqmA ligand-binding preferences, we investigated the kinetics of DPO and Ala-AA binding

Autoinducer recognition by the VqmA quorum-sensing receptor
using stopped-flow fluorimetry. When excited by 321-nm light, DPO emission occurs at 392 nm, and the fluorescence is quenched upon binding to VqmA (Fig. S7A). Ala-AA is not fluorescent, but quenching of natural VqmA fluorescence ( ex ϭ 280 nm/ em ϭ 340 nm) can be monitored to track Ala-AA binding (Fig. S7B). We found that the k on for DPO and Ala-AA are 7.7 ϫ 10 4 M Ϫ1 ⅐s Ϫ1 and 1.3 ϫ 10 4 M Ϫ1 ⅐s Ϫ1 , respectively, whereas the calculated k off values are 0.27 s Ϫ1 and 0.39 s Ϫ1 , respectively (Fig. 6C). Thus, the association rate of DPO is ϳ6-fold faster than that of Ala-AA, whereas the dissociation rate of DPO is modestly slower, ϳ0.7-fold that of Ala-AA. The calculated K d (k off /k on ) values from the stopped-flow experiments are 3.5 M (DPO) and 30.5 M (Ala-AA). This result is consistent with our ITC data, because the differences are within an order of magnitude, which is well-accepted for these two methods. Moreover, the relative differences between the affinities of the two ligands for VqmA are consistent between the two methods. Taken together, we conclude that, once DPO is bound to VqmA, the complex does not easily dissociate, which prohibits subsequent binding of Ala-AA to the same binding site. In the reverse case, DPO can readily replace Ala-AA.

Discussion
In this study, we characterize ligand-controlled activation of VqmA, a cytoplasmic QS receptor and transcription factor. We show that VqmA is soluble and properly folded and activates

Autoinducer recognition by the VqmA quorum-sensing receptor
basal-level transcription of its target, vqmR, in the absence of its cognate AI, DPO. We show that the transcriptional activity of VqmA is increased in response to increasing concentrations of DPO, allowing VqmA to drive the V. cholerae QS transition to HCD upon DPO accumulation. A previous study reported that the exogenous addition of DPO to purified apo-VqmA does not increase the binding affinity of VqmA for PvqmR (17). We find that this is not the case, as the DPO-bound VqmA protein has ϳ2-8-fold higher affinity for DNA than does apo-VqmA (Fig.  1, C and D). One possible explanation for this discrepancy is that the authors of the previous study purified VqmA from WT E. coli, which is tdh ϩ . As mentioned, WT E. coli makes DPO, and VqmA produced in E. coli binds to it (8). Thus, the authors' putative apo-VqmA could in fact have been holo-VqmA. Consequently, their VqmA was likely saturated with DPO, and the addition of more DPO would not enhance the VqmA DNAbinding properties.
We contrast the behavior of the VqmA receptor to that of a canonical LuxR-type QS AI receptor-transcription factor that is activated only when the ligand is present (Fig. 2). Although we cannot exclude issues arising from using different reporter systems, the dramatic difference in -fold change we observed between the VqmA and LasR receptors in the presence versus absence of their cognate AIs (3-fold for DPO/VqmA and 143fold for 3OC 12 -HSL/LasR) leads us to speculate on the biological consequences of the differences between these systems. In particular, the ability of VqmA to function in both a ligandindependent and ligand-dependent manner enables the VqmA pathway to be active across all cell densities. In contrast, because LasR requires its AI for folding and activity, induction of the LasR pathway likely occurs exclusively under conditions when AI is present. We suggest that VqmA proteins are poised to rapidly respond to DPO when it appears. By contrast, canonical LuxR-type receptors require accumulation of cognate AIs for proper folding and activation, therefore delaying QS activation to later in growth. Last, we note that whereas V. cholerae vqmA is expressed constitutively, vqmA Phage is not (10). Although we do not yet know the natural inducer of vqmA Phage , we suspect that its production is regulated. We propose that the DPO-independent activity of VqmA Phage could allow the phage to overcome the requirement for the host to produce the AI and enable lysis provided that sufficient VqmA Phage is present.
The VqmA crystal structures reveal that the DPO-binding pocket is located in the N-terminal PAS domain. Previous reports compared the VqmA fold with that of LasR, TraR, CviR, and other LuxR-type proteins (17,18), the latter of which, as mentioned, function via simultaneous binding and folding around AHL AIs (13,15,16). Whereas this comparison is certainly informative, it is also worth noting that LuxR-type proteins do not contain PAS domains. PAS domains contain characteristic five-stranded antiparallel ␤-sheet cores, whereas the LBDs of most LuxR-type proteins contain four-stranded antiparallel ␤-sheet cores. The divergence in folds is exemplified by the relatively low z-scores of LuxR-type proteins compared with the VqmA structure (Table S2). We imagine that the LBDs of canonical LuxR-type proteins constrain their selectivity to AHL molecules. By contrast, PAS domains are one of the most highly represented domains among signaling proteins and are capable of recognizing a wide range of ligands and sensory inputs, such as redox, light, gases, and metabolites (22)(23)(24). Within prokaryotic taxa, however, it is unusual for a single PAS domain to bind multiple structurally diverse ligands (23). We are curious to discover what properties conferred by the VqmA PAS domain might enable the receptor to bind both the cyclic DPO and the linear Ala-AA molecules. Finally, despite the prevalence of PAS domains, VqmA represents a unique structure for QS receptors.
Our site-directed mutagenesis and ITC results indicate that DPO and Ala-AA use the same site in VqmA because the identical residues are required for binding and activation by both ligands. Confirmation comes from the competition and kinetic assays. We do note that the enthalpy changes measured in the ITC competition assay (Fig. 6B) are not the same as those that we measured for binding of DPO and Ala-AA to apo-VqmA when there is no competition (Fig. 6A). Although not investigated here, one possibility is that some apo-VqmA remains present when we prebind VqmA to a ligand due to ligand association and disassociation. During the competition ITC, a portion of the ⌬H value could stem from ligand binding to the small fraction of apo-VqmA that is present. A second possibility is that ligand replacement in the binding pocket during the competition involves more than simply ligand exchange (e.g. protein breathing).
It is surprising that VqmA binds to the cyclic DPO and the linear Ala-AA molecule using the same pocket. We speculate that Ala-AA adopts a conformation that resembles DPO when it binds to VqmA. Docking studies based on lowest enthalpies also suggest that Ala-AA and DPO bind to the same site on VqmA. Specifically, AutoDock Vina analysis was applied to the PAS domain of our DPO-VqmA structure (25) (Fig. S8). Multiple possible binding sites with different enthalpies were generated by this program. The predicted binding sites of Ala-AA and DPO with the lowest enthalpies are identical, consistent with the experimental DPO-binding site revealed by the DPO-VqmA structure. Moreover, the prediction suggests that Ala-AA binds in a bent configuration. We note that biases might exist in the output of the analysis because the input protein structure is not apo-VqmA.
As mentioned, Ala-AA was a proposed intermediate in DPO production based on the knowledge that threonine and alanine

Autoinducer recognition by the VqmA quorum-sensing receptor
are both substrates in DPO biosynthesis and the proven structure of DPO (8). However, our present work indicates that Ala-AA is likely not the biological DPO precursor. Nonetheless, it is clear that Ala-AA is able to activate VqmA similarly to DPO. It is not simply a matter of the VqmA receptor displaying ligand promiscuity, as we find VqmA is selective for DPO relative to other pyrazine analogs (Fig. 5). This result suggests that the activity conferred by Ala-AA is not coincidental, but likely the actual precursor to DPO is a molecule similar to Ala-AA. Identification of the biological precursor and exploration of its role in VqmA activation are ongoing.
In addition to VqmA, other examples of proteins capable of binding multiple ligands exist, particularly those involved in metabolism. For example, IclR of E. coli binds pyruvate or glyoxylate, leading to activation or repression, respectively, of the operon regulating glyoxylate metabolism (26,27). This mechanism could enable a single receptor to measure glycolytic flux and efficiently adjust gene expression accordingly. Second, BenM of Acinetobacter binds benzoate and muconate, where muconate is a product of benzoate metabolism (28 -30). Simultaneous binding of both molecules leads to transcriptional activation of the benABCDE operon, which further accelerates catabolism of benzoate to muconate (28 -30). This arrangement enables Acinetobacter to tightly control the expression of genes required for the degradation of benzoate, only activating them when their functions are required. Distinct from both of these cases, VqmA binds to either DPO or Ala-AA, but not both, to induce maximal transcriptional activity. This result is curious to us, given that QS receptors typically maintain signaling fidelity by distinguishing between different ligands (31)(32)(33)(34). This situation appears analogous to that of the V. cholerae AI called CAI-1 ((S)-3-hydroxytridecan-4-one) and its biosynthetic precursor, Ea-CAI-1 (3-aminotridec-2-en-4-one). The study of Wei et al. (35) showed that Ea-CAI-1 is a potent activator of CqsS, but this precursor is likely more sensitive to hydrolysis than CAI-1 and therefore is too unstable to be a satisfactory QS signal. Given that pyrazine molecules are known to be stable compounds (36) and Ala-AA tends to cyclize, we hypothesize that whereas DPO is not an absolute requirement for VqmA-mediated QS, DPO could be the more optimal AI relative to Ala-AA because of its stability and potency.
Finally, we do note that all three V. cholerae QS systems converge to control an overlapping group of targets, including the aphA gene encoding the QS master regulator AphA (6,9) that is expressed at LCD when AI concentrations are below their detection thresholds. The HCD responses to the CAI-1, AI-2 (4,5-dihydroxy-2,3-pentanedione) (37,38), and DPO (8) AIs drive repression of aphA. VqmA, because it does not strictly require an AI for activity, activates some vqmR expression at LCD, leading to some repression of aphA. Thus, VqmA-controlled QS could provide a mechanism to attenuate AphA production, and thus alter expression of its downstream targets, prior to when sufficient levels of CAI-1, AI-2, and DPO have accumulated for the transition to the HCD QS state. Subsequently, DPO-VqmA-mediated QS could reinforce repression of aphA once V. cholerae has committed to the HCD QS pro-gram. Thus, the VqmA QS circuit could be important, at all cell densities, for fine-tuning the expression of QS target genes and, in turn, collective behaviors.

Bacterial strains, plasmids, primers, and reagents
Strains, plasmids, and primers used in this study are listed in Tables S4 -S6, respectively. Unless otherwise indicated, V. cholerae and E. coli were grown aerobically in lysogeny broth at 37°C. M9 medium was supplemented with 0.5% glucose and the indicated amino acids, each at 0.4 mM. Antibiotics were used at the following concentrations: 50 units ml Ϫ1 polymyxin B, 200 g ml Ϫ1 ampicillin, and 100 g ml Ϫ1 kanamycin. Primers were obtained from Integrated DNA Technologies. Sitedirected mutagenesis of vqmA was performed using PCR with the indicated primers and the cloning strategy described in Table S6. The same primer pairs were used for generating mutants in both the E. coli expression vector (JSS-1057) and the V. cholerae arabinose-inducible vector (KPS-0500). Intramolecular reclosure was performed as described (10) using Q5 High-Fidelity DNA Polymerase (New England Biolabs). QuikChange (Agilent) site-directed mutagenesis was performed as described (31) using PfuUltra Polymerase (Agilent). The parent pET15b-vqmA vector (JSS-1057) was constructed by PCR of pET15b-His-vqmA (KPS-0929) with primers JSO-654 ϫ JSO-655, followed by intramolecular reclosure to remove the His tag. The PvqmR-lux::lacZ suicide vector (JSS-0807) was constructed with primers JSO-383 ϫ JSO-384 on BB-Ec0361 (pCN004) to generate a one-sided PacI-digestable pKAS backbone with flanking lacZ sites. The PvqmR-lux insert was generated using primers JSO-358 ϫ JSO-397 on JSS-743 (PvqmR-lux in a pEVS143-derived vector). The insert and vector were digested with PacI (New England Biolabs), ligated using T4 DNA ligase (New England Biolabs), and transformed into E. coli S17-1 pir.

Bioluminescence assays
Assays were performed as described (8) with the following modifications. V. cholerae reporter strains were grown overnight and back-diluted to OD 600 ϭ 0.005 in M9 medium supplemented with threonine and leucine. 150 l of culture was transferred to the wells of a 96-well plate, followed by the addition of 50 l of cell-free culture fluids. Synthetic molecule standards were prepared in M9 medium, and 50 l of these prepa-rations were added to the V. cholerae reporter strains in place of cell-free fluids. AI dose-response assays were performed as described (10) with the following modifications. 198-l aliquots of cell cultures were dispensed into a 96-well plate with 2-l aliquots of serially diluted synthetic compounds. Control wells received 2 l of solvent (water or 10% NaOH).

EMSAs
EMSAs were performed as described (10). If not specified otherwise, the highest concentration of VqmA assessed was 130 nM. 2-Fold serial dilutions of protein were used to generate lower concentrations. Biotinylated PvqmR DNA (540 pM) was used in all EMSAs. The percentage of PvqmR DNA bound was calculated using the gel analyzer tool in ImageJ.

Production of cell-free translated protein
Cell-free translated His-VqmA was produced using the S30 T7 High-Yield Protein Expression System (Promega) according to the manufacturer's protocol with the following amendments. 1.5 g of purified plasmid DNA, encoding His-VqmA under the T7 promoter, was used as the template; 40 units of murine RNase inhibitor (New England Biolabs) was added; five reactions were pooled together; and the reactions were incubated (translated) for up to 1.5 h at 37°C. After translation, the His-VqmA was purified using MagneHis Ni-Particles (Promega).

VqmA/PvqmR, LasR/PlasB, and VqmA Phage /Pqtip reporter assays
For each system being tested, ⌬tdh E. coli harboring one plasmid encoding the arabinose-inducible receptor-transcription factor (VqmA, LasR, or VqmA Phage ) and a second plasmid containing the cognate target promoter (PvqmR, PlasB, or Pqtip, respectively) fused to lux (10, 39) was used. In parallel, a control strain for each system was used that carried the reporter plasmid alone (PvqmR-lux, PlasB-lux, or Pqtip-lux). Each strain was grown overnight and back-diluted 1:500 into fresh lysogeny broth containing 0.075% arabinose and either synthetic AI (20 M DPO or 1 M 3OC 12 -HSL) or an equivalent volume of solvent (water for DPO and DMSO for 3OC 12 -HSL). 200-l aliquots of each culture were dispensed into 96-well plates, and the plates were incubated at 30°C with shaking. OD 600 and bioluminescence were measured at regular intervals for 20 h. The bars in Fig. 2 and Fig. S1C represent the time points with the largest differences in relative light units (RLU) between the AI-receiving and -nonreceiving samples for each system. The same time points were used for the reporter-only strains, which received both arabinose and AI.

DPO extraction from cell-free culture fluids
V. cholerae strains were grown overnight and back-diluted to OD 600 ϭ 0.005 in M9 medium supplemented with leucine. The cultures were grown to OD 600 ϳ1.0, followed by incubation with 100 M synthetic DPO or 100 M synthetic d 3 -Ala-AA for 15 min. DPO, d 3 -DPO, and d 3 -Ala-AA were extracted from cell-free culture fluids by SPE (HyperSep Hypercarb SPE cartridges, Thermo Scientific). Chromatographic separation was achieved with increasing concentrations of acetonitrile (0, 10, 20, 30, 50, and 100%). The relative quantities of DPO, d 3 -DPO, Autoinducer recognition by the VqmA quorum-sensing receptor and d 3 -Ala-AA were determined by LC-MS (described below). Samples were back-diluted 1:10 in water prior to column injection. Standards containing synthetic DPO, Ala-AA, or d 3 -Ala-AA were prepared in 2% acetonitrile.

Ala-AA conversion to DPO
10 M synthetic Ala-AA or 10 M synthetic DPO was incubated with 50 M apo-VqmA in EMSA buffer. Each single component was also diluted into EMSA buffer as a control. Following incubation for 15 min, excess and nonprotein-bound small molecules were removed by size-exclusion chromatography (Zeba), and the resulting protein was heated to 70°C for 10 min. Denatured protein was removed by centrifugation (13,000 ϫ g, 10 min), and supernatants were assessed for Ala-AA or DPO by LC-MS (described below).

LC-MS to detect DPO, Ala-AA, d 3 -DPO, and d 3 -Ala-AA
LC-MS was performed with modifications (10). Samples and standards were loaded onto a C 18 column (1 mm ϫ 100 mm; ACE 3 C18 AQ, Mac-Mod) using a Shimadzu HPLC system and PAL autosampler (20 l/injection) at a flow rate of 70 l min Ϫ1 . Full scan MS data were acquired with an LTQ-Orbitrap XL mass spectrometer (Thermo) at a resolution of 60,000 in profile mode from the m/z range of 110 -200. DPO, Ala-AA, d 3 -DPO, and d 3 -Ala-AA were detected by performing XIC at m/z values of 125.0714 (C 6 H 9 N 2 O), 145.0977 (C 6 H 13 N 2 O 2 ), 128.0903 (C 6 H 6 d 3 N 2 O), and 148.1165 (C 6 H 10 d 3 N 2 O 2 ), respectively, each with a mass accuracy of Ϯ10 ppm. Caffeine (2 M in 50% acetonitrile with 0.1% formic acid) was delivered as a lock mass using an HPLC pump (LC Packing) with a flow splitter and at an effective flow rate of 20 l min Ϫ1 through a tee at the column outlet. Raw files were imported into Skyline version 4.2.0 (40), and peak areas for each molecule were extracted using the small-molecule workflow.

Protein expression and purification
V. cholerae vqmA was cloned into pET15b and overexpressed in ⌬tdh BL21 (DE3) E. coli cells using 1 mM isopropyl 1-thio-␤-D-galactopyranoside at 37°C for 4 h. Cells were pelleted at 3,300 ϫ g for 15 min and resuspended in lysis buffer (20 mM Tris-HCl, pH 8, 150 mM NaCl, 1 mM DTT). Cells were lysed using sonication and subjected to centrifugation at 32,000 ϫ g for 30 min. The supernatant was loaded onto a heparin column (GE Healthcare) and eluted by a linear gradient from buffer A (20 mM Tris-HCl, pH 8, 1 mM DTT) to buffer B (20 mM Tris-HCl, pH 8, 1 M NaCl, 1 mM DTT). Peak fractions were pooled, concentrated, and loaded onto a Superdex-200 size-exclusion column (GE Healthcare) in gel filtration buffer (20 mM Tris-HCl, pH 8, 150 mM NaCl, and 1 mM DTT). Proteins were concentrated to 30 mg/ml, flash-frozen, and stored at Ϫ80°C. All VqmA mutants exhibited production and solubility levels similar to WT VqmA, suggesting that they are correctly folded, although this was not confirmed explicitly.

Crystallization and structure determination
The DPO-VqmA complex was formed by mixing 1 mM purified apo-VqmA with 10-fold excess synthetic DPO prior to crystallization. Crystals were obtained by the hanging drop vapor diffusion method from drops containing a 1:1 mixture of protein/precipitant (0.1 M Tris-HCl, pH 7.3, 16% PEG3350, and 0.2 M sodium malonate) at 22°C. The trays were seeded using cat whiskers (Spark Bassler). Crystals were cryoprotected in a solution containing the precipitant plus 10% (v/v) glycerol. Crystals belonged to the P 2 1 2 1 2 1 space group with unit cell dimensions a ϭ 50.4 Å, b ϭ 84.6 Å, c ϭ 111.6 Å. A data set with heavy atom derivatives was obtained by soaking the crystals in cryoprotectant supplemented with 10 mM KAu(CN) 2 (Hampton Research) for 10 min at RT as reported (41). Crystals were subsequently flashfrozen and stored in liquid N 2 . Diffraction data were collected at the NSLS II (National Synchrotron Light Source) beamlines FMX and AMX. Data were processed using HKL-3000, and experimental phases were determined using single-wavelength anomalous dispersion (42). Experimental phases were calculated using SHELX (43). The data sets were subjected to density modification using PHENIX (44 -46), and the VqmA structure was built manually using Coot (47) over multiple iterations with further refinement in PHENIX. Images were generated using PyMOL.

Thermal shift assay
Thermal shift analyses of VqmA were performed as described (31). In brief, VqmA was diluted to 5 M in reaction buffer (20 mM Tris-HCl, pH 8, 150 mM NaCl) containing either 10 M synthetic DPO, 10 M synthetic Ala-AA, or an equivalent volume of water in an 18-l reaction (final volume). The mixtures were incubated at RT for 15 min. 2 l of 200ϫ SYPRO Orange in reaction buffer were added to the samples immediately before analysis. Samples were subjected to a linear heat gradient of 0.05°C s Ϫ1 , from 25 to 99°C in a Quant Studio 6 Flex System (Applied Biosystems) using the melting curve setting. Fluorescence was measured using the ROX reporter setting.

ITC
A MicroCal PEAQ-ITC (Malvern) instrument was used to measure the binding affinities between VqmA proteins and ligands at 25°C. All synthetic ligands for ITC were prepared in gel filtration buffer (described above). For measurement of K d , 500 M ligand was titrated into 50 M apo-VqmA or mutant VqmA proteins using continuous stirring at 750 rpm. For the ITC-based competition assay, 500 M synthetic Ala-AA was titrated into 50 M DPO-VqmA in the presence of 0, 50, or 500 M synthetic DPO, and in the opposite case, Autoinducer recognition by the VqmA quorum-sensing receptor presence of 0, 50, or 500 M Ala-AA. The initial and subsequent 18 injections were 0.4 and 2 l, respectively. Heat changes during the titration were measured by the PEAQ-ITC Control software (MicroCal). Data were fitted and evaluated by the PEAQ-ITC Analysis software (MicroCal).

Stopped-flow fluorimetry
A SX20 stopped-flow spectrometer (Applied Photophysics) was used for the kinetic study of ligand interactions with VqmA at 25°C. DPO and VqmA have fluorescence spectra of ex ϭ 321 nm/ em ϭ 392 nm and ex ϭ 280 nm/ em ϭ 340 nm, respectively. For DPO binding to VqmA, the k obs was measured by recording the decrease in DPO fluorescence using a 390-nm emission interference filter. The concentration of VqmA was fixed at 2.5 M, whereas DPO was supplied at concentrations from 1 to 20 M. For Ala-AA binding to VqmA, the k obs was measured by recording the decrease in VqmA fluorescence using a 302-nm emission interference filter. The concentration of Ala-AA was fixed at 7.5 M, and VqmA, from 3 to 37.5 M, was supplied. Typically, 5-10 traces were recorded and averaged for each data point. Data were fitted with an exponential decay equation using GraphPad Prism version 6. Formation of complexes follows the equation, k obs ϭ k on ϫ [analyte] ϩ k off . By plotting k obs versus [analyte], k on and k off are obtained from the slope and y-axis intercept, respectively.

Western blot analysis
Western blot analyses probing for FLAG fusions were performed as reported previously (8). V. cholerae cells were collected and diluted 2-fold in Laemmli sample buffer, giving a final concentration of 0.01 OD/l. Following denaturation for 15 min at 95°C, 0.1 OD 600 equivalents were subjected to SDSgel electrophoresis. RNAP␣ was used as loading control. Signals were visualized using an ImageQuant LAS 4000 imager (GE Healthcare).

Docking analysis
Docking analyses were performed as described (25). In brief, AutoDockTools version 1.5.6 was used to generate the pdbqt files, and AutoDock Vina version 1.1.2 was used to assess possible ligand-binding sites in the VqmA PAS domain (25,48). The lowest-enthalpy results are shown in Fig. S8.