High-Throughput Mutagenesis and Cross-Complementation Experiments Reveal Substrate Preference and Critical Residues of the Capsule Transporters in Streptococcus pneumoniae

ABSTRACT MOP (Multidrug/Oligosaccharidyl-lipid/Polysaccharide) family transporters are found in almost all life forms. They are responsible for transporting lipid-linked precursors across the cell membrane to support the synthesis of various glycoconjugates. While significant progress has been made in elucidating their transport mechanism, how these transporters select their substrates remains unclear. Here, we systematically tested the MOP transporters in the Streptococcus pneumoniae capsule pathway for their ability to translocate noncognate capsule precursors. Sequence similarity cannot predict whether these transporters are interchangeable. We showed that subtle changes in the central aqueous cavity of the transporter are sufficient to accommodate a different cargo. These changes can occur naturally, suggesting a potential mechanism of expanding substrate selectivity. A directed evolution experiment was performed to identify gain-of-function variants that translocate a noncognate cargo. Coupled with a high-throughput mutagenesis and sequencing (Mut-seq) experiment, residues that are functionally important for the capsule transporter were revealed. Lastly, we showed that the expression of a flippase that can transport unfinished precursors resulted in an increased susceptibility to bacitracin and mild cell shape defects, which may be a driving force to maintain transporter specificity.

all pneumococcal CPSs are synthesized by the Wzx/Wzy mechanism in which the cps genes are organized into a single operon located between dexB and aliA (Fig. 1). The first gene of the serotype 2 cps locus, cps2A, encodes a LytR/CpsA/Psr (LCP) family protein that presumably conjugates CPS to PG (14)(15)(16). Cps2B, Cps2C, and Cps2D form a tyrosine kinase system that regulates CPS synthesis (17). The downstream genes encode other enzymes such as the polyprenol phosphate phosphoglycosyl transferase (PGT) Cps2E, the flippase Cps2J(Wzx), and the polymerase Cps2H(Wzy) (Fig. 1). The genetic organization is similar for other pneumococcal cps loci. This arrangement presumably facilitates genetic exchanges, which accelerates serotype replacement and the evolution of new serotypes, thereby contributing to the diversity of pneumococcal CPS (18).
MOP (Multidrug/Oligosaccharidyl-lipid/Polysaccharide) superfamily transporters are found in virtually all life forms. Members of this family include the PG precursor (lipid II) flippase MurJ (19,20), the teichoic acid precursor flippase TacF (21), and the capsule flippase Cps2J (22). In general, MOP transporters are specific for their substrates, unless they are overexpressed (23) or mutagenized (24). Although structures of several MOP family transporters have been solved (25)(26)(27)(28)(29), residues responsible for substrate recognition remain unknown. Previously, we attempted to address this question using a directed evolution approach (24). Capsule flippase variants with relaxed substrate specificity were isolated in Escherichia coli by selecting for mutants that can transport PG precursors. Although these mutations provided some mechanistic insights into flippase specificity, the genetic approach was limited by the number of lipid-linked cargos that could be tested. Since the glycan motifs of PG and the colanic acid capsule precursor share little similarity (24), interpreting the results of the study is relatively challenging.
Here, we harnessed the wealth of genetic and structural information on the pneumococcal CPS to revisit this important biological question. CPS flippases from 82 serotypes were tested for their ability to translocate noncognate serotype 2 and 33B precursors. Overall, most CPS flippases are specific in S. pneumoniae. Strikingly, close homologs of Cps2J could not compensate for the Cps2J function, whereas distant homologs in serogroups 10, 33, and 34 could. This is in contrast to the flippases that could complement cps33BJ, for which only close homologs of Cps33BJ could substitute for its function. Site-directed mutagenesis of CpsJ in serogroup 10 revealed residues critical for substrate recognition. To better understand the substrate selectivity of CPS flippases, we isolated Cps23BJ variants that could replace Cps2J. In addition, immutable residues in Cps2J were identified. Lastly, we demonstrated that the expression of a CPS flippase capable of transporting an incomplete precursor resulted in cell shape defects and hypersensitivity to bacitracin, which may explain the selective advantage of substrate selection by MOP transporters.

RESULTS
Cps2J depletion was lethal and resulted in cell shape defects. Genes involved in the late-stage CPS synthesis are thought to be essential for growth (22,30). The prevailing model is that their inactivation stalls CPS synthesis, leading to the accumulation of dead-end intermediates. This process subsequently sequesters Und-P and inhibits PG synthesis. In Gram-negative bacteria, the incomplete precursors are also implicated in the disruption of membrane integrity and cause cell shape defects (31,32). To test whether cps2J is required for growth in pneumococcus, we constructed a strain in which cps2J expression was under the control of a zinc inducible promoter (P Zn ). When Cps2J was depleted, cells became elongated and lysed after ;6 h ( Fig. 2). Our result is consistent with the general observation that premature termination of the Wzx/Wzy pathway is lethal because it disrupts Und-P recycling. Indeed, stopping Und-P flux into the CPS pathway by depleting the initiating PGT cps2E alleviated the lethality caused by the disruption of cps2J (Fig. 1).
Systematic cross-complementation of Cps2J illustrated substrate preferences. Except for serotypes 3 and 37, each pneumococcal cps locus encodes a dedicated CpsJ flippase. We therefore hypothesized that these flippases are specific, similar to other FIG 2 Cps2J is required for growth. (A) Strains NUS0063 (P Zn -cps2J) and NUS0084 (Dcps2J//P Zn -cps2J) were grown in BHI supplemented with ZnCl 2 and MnCl 2 at 37°C in 5% CO 2 . Cells were washed and diluted in BHI and continued to grow at 37°C. Growth was monitored by measuring the OD 600 hourly. Shown are representative growth curves from three biological replicates. (B) The same strains were grown in BHI supplemented with ZnCl 2 and MnCl 2 . Cultures were normalized by their optical densities, serially diluted, and spotted on blood agar with (left) or without (right) ZnCl 2 and MnCl 2 . (C) Cells of the same strains were grown in BHI with or without ZnCl 2 and MnCl 2 . Just before the culture started lysing (;6 h postinoculation), cells were imaged by phase-contrast microscopy, and the images were quantified by MicrobeJ (n = 1969) (D). P values were computed by using the Mann-Whitney U test. Cells exhibiting cell shape defects are highlighted with red arrows. Scale bar, 5 mm.
MOP flippases in Gram-negative bacteria (33)(34)(35). To test this, we aligned the CpsJ variants from 93 serotypes. As expected, CPS flippases are remarkably diverse. They can be classified into 17 sequence clusters (see Fig. S1 in the supplemental material), many of which have less than 40% sequence identity to their closest homolog. Interestingly, there is a strong correlation between the primary sequences of CpsJ and the precursors they transport (see Fig. S1). For example, CpsJ variants in serotypes 35F and 47F, as well as in serotypes 35C and 42, are approximately 97% identical, and their substrates are strikingly similar. In contrast, CpsJ variants in serotypes 7F and 7A are different from those in serotypes 7B and 7C, although they belong to the same serogroup. Consistently, the repeating units of serotype 7F and 7A CPSs could be easily distinguished from serotypes 7B and 7C. Assuming the sequence similarity of CpsJ can be used to predict the chemical structures of their cargos, the repeating units of serotype 38 and serogroup 25, as well as serotypes 7B and 40, are likely to be related.
To test the interchangeability of CPS flippases, we replaced cps2J with the open reading frame of a noncognate cpsJ. Clinical isolates of S. pneumoniae were collected from hospitals and research institutes and served as DNA templates (see Table S1). Since Cps2J is essential, we first introduced an inducible copy of cps2J at an ectopic locus (P Zn -cps2J 1 ) to maintain cell viability. We then engineered the strains such that cps2J at the native locus was replaced by the noncognate cpsJ variants. This strategy maintains the gene dosage and prevents artifacts caused by the overexpression of noncognate flippases, because cpsJ expression was driven by the native capsule promoter (P cps ) and ribosomal binding site (RBS cps ). If the noncognate flippase at the cps locus could complement cps2J and transport serotype 2 cargo, cps2J at the ectopic locus would become dispensable for growth. In total, CpsJ variants from 82 serotypes were tested, and most of them were unable to complement cps2J (see Fig. S1). Six CpsJ variants were able to substitute for the Cps2J function (Cps7AJ, Cps10BJ, Cps33BJ, Cps33CJ, Cps33DJ, and Cps34J) and three of them (Cps20J, Cps29J, and Cps47FJ) could only partially complement cps2J (Fig. 3). Here, we define partial complementation as the scenario where the mutant carrying the cpsJ variant exhibited growth defects when it is the sole copy of CPS flippase in the cell (see below). Unexpectedly, CpsJ variants that could complement cps2J are not the closest homologs of Cps2J. Instead, they cluster around serogroup 33 (see Fig. S1). The results indicate that CpsJ is generally specific to its cognate substrate, and the ability to cross-complement cannot be predicted based on their sequence similarity.
We demonstrated that cell viability faithfully reports the flippase activity of Cps2J. Alternative explanations were eliminated by a series of experiments. First, we excluded the possibility that the CpsJ switch mutants contained suppressor mutations in cps2E (30). Inactivation of the early cps genes (e.g., cps2E) prevents Und-P from entering the CPS pathway and alleviates Und-P sequestration. However, the resulting strain could no longer produce CPS. This is not the case because all CpsJ switch mutants remained encapsulated as judged by immunostaining (i.e., the Quellung reaction) (see Fig. S2). Next, we sought to eliminate the possibility that leaky expression of the P Zn -cps2J cassette might support growth. To test this, the P Zn -cps2J cassette was replaced with a P-erm cassette. The strains remained viable and encapsulated (see Fig. S2), suggesting that these CpsJ variants can indeed complement Cps2J. Finally, inactivation of the noncognate cpsJ allele restored zinc dependence for growth, indicating there is no mutation elsewhere in the genome that can account for the survival of the CpsJ switch mutants (see Fig. S2).
It is plausible that Cps2J-complementing flippases are more stable than the others. As shown in other Wzx flippases (23), overexpression of the noncognate CpsJ may support translocation of unrelated substrates. To exclude this possibility, we fused several CpsJ variants (Cps7AJ, Cps10AJ, Cps10BJ, Cps23FJ, Cps33BJ, and Cps48J) with a FLAG tag (DYKDDDDK) at the C terminus and monitored their levels by immunoblotting. FLAG-tagged Cps2J, Cps7AJ, Cps10BJ, and Cps33BJ are functional because they could complement the cps2J depletion mutant (see Fig. S2). While the amount of Cps2J was the lowest among the CpsJ variants we tested (see Fig. S2), the ability to substitute Cps2J does not seem to correlate with the protein level of the CpsJ variants (see Fig. S2).
As mentioned above, some CpsJ variants could only partially replace Cps2J. When the P Zn -cps2J cassette was deleted in mutants harboring these variants (cps20J, cps29J, and cps47FJ), cells exhibited various degrees of growth and cell shape defects (see Fig. S3). To show that they were indeed defective in substrate transport, we quantified the amount of CPS in strains expressing Cps29J (partial complementation) and Cps34J (full complementation) (see Fig. S3). As expected, cells with Cps29J produced ;40% of the wild-type level of CPS, whereas cells with Cps34J had no significant difference in the amount of CPS (see Fig. S3). Thus, some of these cross-complementing CpsJ variants may be inefficient at transporting serotype 2 substrates. the CpsJ variants that can transport the serotype 2 precursor. Serotypes colored dark green produce CpsJ variants that can fully complement cps2J, whereas serotypes in light green produce variants that partially complement cps2J. The black arrows indicate the serotypes that could not be tested in the study (i.e., serotypes 16A and 35F and serogroup 11). The unit for the scale bar is the number of amino acid differences per site. (B) Cells of strain NUS0650 (rpsL1 Dcps2J::P-sacB-kan-rpsL 1 //P Zn -cps2J) and its derivatives expressing different CpsJ variants were grown in BHI with ZnCl 2 and MnCl 2 . Cultures were diluted and spotted on blood agar as described in the legend of Fig. 2. Plates were incubated at 37°C in 5% CO 2 overnight before being photographed. Shown are the representative images of the spot dilution assays and the rest are illustrated in Fig. S1B. Complementation of cps2J is judged by the formation of smooth colonies on plates without Zn 21 . Cps48J that could not replace Cps2J is shown here as a negative control for comparison.
Cps33BJ could not flip PG and teichoic acid precursors. Since we could not identify any glycan motifs shared by the repeating units of serotype 2 and the cross-complementing serotypes, we hypothesized that the CpsJ variants that could replace Cps2J may have relaxed specificity like Wzk (36). To test this, we investigated whether Cps33BJ could substitute for the lipid II flippase YtgP (a MurJ-like protein) (37) and the teichoic acid precursor flippase TacF (21). We constructed strains in which cps2J was replaced by cps33BJ, tacF, or ytgP. In order to maintain cell viability, cps2J was expressed ectopically with the P Zn promoter. Given cps33BJ could complement cps2J, the Zn 21 inducer was not required for growth. We then deleted ytgP or tacF at the native loci. No transformant was obtained in the strain harboring cps33BJ, compared to the numerous transformants recovered when the recipient cells were the cps2J,.ytgP or cps2J,.tacF mutants (see Table S2A). Unsurprisingly, Cps2J could not substitute for YtgP or TacF, because adding Zn 21 to the medium did not allow deletion of these flippases. Together, our results suggest that cps33BJ cannot flip PG and teichoic acid precursors to the level that restores growth.
Next, we investigated whether Cps2J could reciprocally flip serotype 33B CPS precursor. The native serotype 2 cps locus in strain IU1781 (D39 rpsL1) was replaced with the serotype 33B cps locus, so that the strain produced a serotype 33B instead of a serotype 2 capsule (38). We then engineered the capsule-switch mutant by introducing an ectopic copy of P Zn -cps33BJ, followed by deleting cps33BJ at the native locus (i.e., NUS0490 [rpsL1 CPS33B Dcps33BJ//P Zn -cps33BJ]). The resulting strain now required the Zn 21 inducer for growth. cps2J was then introduced to the cps33B locus, and we tested whether it could complement cps33BJ and restore viability. We did not detect complementation of cps33BJ by cps2J (see Fig. S3), indicating that the cross-complementation event is unlikely caused by the structural similarity between Cps2J and Cps33BJ or their corresponding cargos.
Cps33BJ could be substituted by its close homologs. We investigated whether Cps33BJ could be replaced by other CpsJ variants, similar to the case of Cps2J. First, we employed the capsule-switch mutant that produced a serotype 33B capsule. An ectopic copy of cps33BJ was introduced under the control of a P Zn promoter, followed by replacing cps33BJ at the native locus with a noncognate cpsJ allele (NUS1549 [rpsL1 CPS33B Dcps33BJ::P-sacB-kan-rpsL 1 //P Zn -cps33BJ]). As mentioned above, if the CpsJ variant(s) can replace Cps33BJ, the Zn 21 inducer would no longer be required for growth. We tested representative CpsJ variants from each clade in Fig. S1, as well as the close homologs of Cps33BJ. Unlike Cps2J, none of the distant Cps33BJ homologs we tested were able to compensate for the Cps33BJ function ( Fig. 4). In addition, other than Cps47AJ, CPS flippases that are similar to Cps33BJ could flip Cps33B precursor. These results further support the notion that CPS flippases are generally specific, given CpsJ variants that can substitute for the CPS flippase in another serotype are relatively rare.
A comparison of Cps10AJ and Cps10BJ revealed residues important for transporting the serotype 2 cargo. The phylogenetic tree illustrated in Fig. S1 highlighted two interesting CpsJ homologs that might provide insights into the specificity of CPS flippases. Cps10AJ and Cps10BJ differ by five amino acid residues (see Fig. S4), and yet only Cps10BJ could substitute for Cps2J. Structural modeling by I-TASSER suggested Cps10BJ adopts a V-shape conformation, resembling the "inward-open" state of MurJ (28,29,39) (see Fig. S4). Assuming MurJ and CpsJ function similarly, Cps10BJ is expected to transition between the inward-open and outward-open states during substrate translocation. If so, mutations that promote the conformational change upon substrate binding may allow CpsJ to translocate a noncognate substrate (40). Based on this assumption, residue F109 in Cps10BJ lies in the putative extracellular gate (see Fig. S4), if changed to valine, may destabilize the inward-open conformation and relax substrate specificity (24). Similarly, the asparagine residue at position 222 of TM8 in Cps10AJ likely favors an inward-open conformation compared to a glutamate residue. We hypothesized that residue H192 is not involved in substrate selection because it is too far away from the putative substrate binding pocket (see Fig. S4). To test these speculations, we performed site-directed mutagenesis to pinpoint the change(s) that allowed Cps10BJ to flip the serotype 2 CPS precursor. As predicted, cps10BJ H192Y could still complement cps2J (see Fig. S4). Changes in the remaining four residues to the equivalent of Cps10AJ (T101I, V109F, N222E, and V356A) abolished serotype 2 precursor transport (see Fig. S4). To evaluate the stability of the variant proteins, we fused the C-terminal end of Cps10BJ with a FLAG tag. The constructs remained functional (see Fig. S2), and the only variant that caused a statistically significant change in the protein level is N222E, which is around 60% of the wild-type level of Cps10BJ (see Fig. S4). Next, we tested whether changing the corresponding residues in Cps10AJ would result FIG 4 Cps33BJ could be complemented by its close homologs. (A) Representative CpsJ variants from each clade were tested for their ability to transport the serotype 33B precursor. Serotypes highlighted in green represent CpsJ variants that could complement cps33BJ (in boldface), whereas serotypes in red indicate CpsJ variants that could not. The unit of the scale bar is the number of amino acid differences per site. (B) Strain NUS1549 (rpsL1 CPS33B Dcps33BJ::P-sacB-kan-rpsL 1 //P Zn -cps33BJ) and its derivatives harboring different CpsJ variants were grown, serially diluted, and spotted on blood agar plats with or without ZnCl 2 and MnCl 2 supplement. Plates were incubated overnight at 37°C in 5% O 2 before imaging.
in gain-of-function mutations that allowed cross-complementation of cps2J. None of the single or double mutants of Cps10AJ could transport the serotype 2 CPS substrate (see Fig. S4). With that, every possible combination has been tested because the triple mutants of Cps10AJ would result in proteins identical to the Cps10BJ single mutants. In addition, we showed that the mutated Cps10AJ still retained the ability to transport its own substrate (see Table S2B), indicating that they are functional flippases. Finally, to examine whether cps10BJ could transport structurally distinct substrates, we attempted to delete ytgP and tacF in strain NUS0550 (rpsL1 cps2J,.cps10BJ//P Zn -cps2J). No transformant could be recovered regardless of Zn 21 supplementation (see Table S2C), demonstrating that Cps10BJ could not appreciably transport PG and teichoic acid precursors. In conclusion, the four residues near the proposed central aqueous cavity are involved in the transportation of the noncognate serotype 2 precursor.
Mutagenized Cps23BJ variants could complement cps2J. To investigate why close homologs of Cps2J such as Cps23BJ could not transport serotype 2 substrate, we isolated gain-of-function variants of Cps23BJ that could replace Cps2J (see Fig. S1). Our initial screening strategy was to replace cps2J with a mutagenized cps23BJ allele while maintaining cell viability by ectopically expressed cps2J. However, despite numerous attempts, we could not achieve a sufficiently high transformation efficiency to support this screening strategy. To overcome this problem, we constructed a strain in which cps2E expression was controlled by the P Zn promoter. Inactivation of cps2J was tolerated, unless Zn 21 was added to the medium to initiate CPS synthesis (Fig. 1). With this strain, cps2J at the native locus could be readily replaced by a mutagenized copy of cps23BJ. Approximately 15,000 transformants were obtained, pooled, and selected on Zn 21 supplemented blood plates. Unencapsulated suppressor mutants were discarded by screening ;100 survivors with immunostaining using anti-CPS antibodies. Isolates that maintained serotype 2 capsule production were sequenced (72 in total). Among them, 17 unique Cps23BJ variants were identified (see Data Set S1 in the supplemental material). The amino acid changes clustered around the predicted extracellular region of the transporter (see Fig. S5), strikingly similar to the WzxC variants reported to have expanded specificity (24). To ensure that the causative mutations were located in cps23BJ, we transformed the cps23BJ alleles back to the parent strain (NUS0893 [rpsL1 Dcps2E Dcps2J::P-sacB-kan-rpsL 1 //P Zn -cps2E]), and the resulting strains retained the phenotype (see Fig. S5). For simplicity, the Cps23BJ variants that can replace Cps2J are here collectively referred to as Cps23BJ*.
We noticed that strains harboring Cps23BJ* in place of Cps2J exhibited growth defects and formed smaller colonies (see Fig. S5). Mild cell shape defects were detected under phase-contrast microscopy after cps2E expression was induced (see Fig. S6). This implied that Cps23BJ* variants could not completely replace Cps2J. If so, the amount of CPS produced in these strains should have been reduced. To confirm this, we quantified the CPS in these strains by immunoblotting. As expected, CPS produced in the Cps23BJ* strains were between %12% to %40% of the wild-type level (see Fig. S6). Since strain NUS0267 (rpsL1 Dcps2E//P Zn -cps2E) produced slightly lesser CPS than the wild-type strain, even it is cps2J 1 , we wondered whether Cps23BJ* could support growth when the CPS expression was restored to the wild-type level. We therefore introduced cps23BJ* into the native cps locus of strain NUS0650 (rpsL1 Dcps2J::P-sacBkan-rpsL 1 //P Zn -cps2J) by allelic exchange. Upon cps2J depletion, 11 of the 17 cps23BJ* mutants could no longer support grow (see Fig. S5). We also demonstrated that the cps23BJ* alleles remained capable of transporting the serotype 23B substrate because they complemented cps23BJ in a capsule-switch mutant (see Table S2D). Lastly, to test whether the Cps23BJ* proteins could flip PG and teichoic acid precursors, tacF or ytgP was inactivated in strains harboring cps23BJ* alleles, but no viable transformant could be recovered (see Table S2E). We conclude that Cps23BJ* variants could be isolated, although many of which can only partially complement cps2J and none can compensate for the function of tacF and ytgP.
Essential residues of Cps2J are located in the putative central solvent-exposed cavity. To identify residues required for the Cps2J function, we performed highthroughput mutagenesis and sequencing (Mut-seq) (41). This approach simultaneously assesses the functional consequences of virtually all single nucleotide polymorphisms (SNPs) in cps2J. Briefly, we introduced a PCR mutagenized allele of cps2J into a strain where the cps2E expression was driven by a P Zn promoter (NUS0267 [rpsL1 Dcps2E// P Zn -cps2E]). As mentioned above, when cps2E was not induced, cells grew normally regardless of the cps2J function. Upon Zn 21 addition, only the cells harboring a functional copy of cps2J could survive (Fig. 1). The frequencies of SNPs recovered under these two conditions, which represent the fitness cost of each mutation, were compared. In total, we evaluated 3,802 SNPs that encompassed all 1,410 nucleotides of cps2J. Of these, 60 SNPs had more than a 5-fold decrease in frequency when selected on Zn 21 supplemented plates, indicating that they may have a reduced flippase activity (see Data Set S1). To validate the Mut-seq results, we selected five mutations (G43R, S60L, G342E, G342R, and G346E) and introduced them into the cps2J depletion strain (NUS0650 [rpsL1 Dcps2J::P-sacB-kan-rpsL 1 //DbgaA::P Zn -cps2J]) to test whether these alleles remained functional. Indeed, all of these variants were unable to complement cps2J (Fig. 5). Without the Cps2J structure solved, it is difficult to understand why these changes will lead to a nonfunctional flippase. Nevertheless, based on the structural model, many essential residues identified are buried and may contribute to protein folding. Unlike MurJ (29), in which several solvent-exposed essential residues were identified in the central aqueous cavity, we only found residue S60 in Cps2J at the kink region of TM2 (Fig. 5). This residue is presumably analogous to residue E57 in MurJ, which was shown to be important for the conformational transition from the inwardopen to outward-open state (29,39). In addition, a few solvent-exposed essential residues are detected in the extracellular and intracellular gates, which may stabilize the inward-open and outward-open states, respectively. Mutation of G43 into bulky residues (e.g., G43R) likely causes clashes with the neighboring residues, therefore disrupting the extracellular gate and favoring the outward-open state. In contrast, mutations of G342 and G346 (G342E/G342R and G346E) on the opposite side perhaps disrupt the intracellular gate and favor the inward-open state (Fig. 5). Lastly, mutations of the two glycine residues into charged residues may also exclude substrate binding. Solving the structure of Cps2J in the future will support or disprove these predictions. To examine whether these changes indeed affect protein stability, we introduced the mutations into the cps2J-FLAG construct. When expressed at the native locus, cps2J-FLAG could complement cps2J 1 (Fig. 5) Immunoblotting revealed a 2-fold and 5-fold reduction in protein levels of the G43R and S60L variants, respectively, whereas no noticeable change was detected in the levels of the G342E, G342R, and G346R variants. Together, our Mut-seq results suggest that MOP superfamily flippases share a transport mechanism that likely involves a conformational change at the kink region of the TM2.
Transporting unfinished CPS precursors across the cell membrane resulted in modest cell shape defects and an increased sensitivity to bacitracin. The specificity of MOP transporters could seemingly be relaxed with rather subtle changes in the putative extracellular loops or the central aqueous cavity. This prompted us to investigate the driving force that selects for the substrate selectivity. We hypothesized that the less-specific flippases may accidentally flip incomplete lipid-linked precursors across the cell membrane. Since there is no known mechanism to flip them back into the cytoplasm, the trapped precursors may sequester Und-P and lead to a selective disadvantage. However, mutants expressing WzxC or CpsJ variants with expanded specificity did not exhibit any noticeable growth defect (24). Yet, we have not ruled out the possibility that the variants with growth defects were not recovered with our genetic selection strategy because they were underrepresented. To test whether the expression of a flippase capable of transporting incomplete precursor is detrimental to the cell, we investigated the CPS flippases in serogroup 19. Theoretically, Cps19AJ can compete with the glycosyltransferase WchO in serotypes 19B and 19C by transporting the incomplete precursor ManNAc-Rha-Glc-P-Und to the wrong compartment, before it can be converted to the serotype 19B and 19C repeating units ( Fig. 6; see also Fig. S7). First, isogenic capsule-switch mutants that produced serotype 19A, 19B, and 19C capsules were constructed. Next, we cloned cps19AJ downstream of a strong constitutive promoter P spxB and introduced it into an ectopic locus. The P spxB -cps19AJ cassette was functional because it could complement cps19AJ in the serotype 19A capsule-switch in three different orientations. The N-lobe and C-lobe are colored in cyan and green, respectively. The essential residues are shown as spheres with buried residues colored in red. The exposed essential residues are shown as sticks and labeled with residue numbers colored in magenta. The nearby interacting residues are shown as sticks and labeled with residue numbers colored in black. (B) Ribbon representation of the outward-open model of Cps2J in three different orientations with the same color scheme. D140 engages a salt bridge interaction with K279. S349 makes a hydrogen bond interaction with E65. (C) Cells expressing the indicated Cps2J-FLAG and its corresponding derivatives were grown in BHI with ZnCl 2 and MnCl 2 . Cultures were diluted and spotted on blood agar as described in the legend of Fig. 2. (D) Quantification of the numbers of CpsJ-FLAG variants by immunoblotting. Cells were grown in BHI with ZnCl 2 and MnCl 2 , harvested by centrifugation, and resuspended in 2Â sample buffer. Proteins were separated by SDS-PAGE, transferred to a membrane, and immunoblotted with anti-FLAG antibodies. P values were computed from three biological replicates using the Student's t test.

Specificity Determination of Capsule Transporters
® mutant (NUS1580 [rpsL1 CPS19A Dcps19AJ::P-erm DbgaA::P spxB -cps19AJ]). We then introduced the P spxB -cps19AJ cassette into the serotype 19B and 19C capsule-switch mutants. Overproduction of cps19AJ did not cause any noticeable growth defects in liquid medium, but the strains arguably formed slightly smaller colonies on blood plates (see Fig. S7). The cells also have a mild cell shape defect (Fig. 6). To investigate whether the overproduction of cps19AJ caused a reduction in the Und-P level in serotype 19B and 19C strains, we examined whether the cells are hypersensitive to bacitracin (32). Bacitracin binds to undecaprenyl pyrophosphate (Und-PP) and depletes the Und-P pool. If the cell has a lower Und-P level to begin with, the MIC of bacitracin should be reduced. Indeed, strains harboring serotype 19B and 19C capsule with cps19AJ overexpressed were more sensitive to bacitracin compared to the parent strains (Fig. 6). Finally, to test whether the transporter function of the cps19AJ is required for the bacitracin-sensitive phenotype, inactive variants of cps19AJ were constructed and overexpressed in the serotype 19B and 19C capsule-switch mutants. These variants did  (12,66). The reducing ends are on the right, and the sites of polymerization are indicated by the curved arrows on the left. (B) Cells of strains NUS1525 (CPS19B DbgaA::P-sacB-kan-rpsL 1 ), NUS1527 (CPS19B DbgaA::P spxB -cps19AJ), NUS1568 (CPS19C DbgaA::P-sacB-kan-rpsL 1 ), NUS1584 (CPS19C DbgaA::P spxB -cps19AJ), NUS2066 [CPS19C DbgaA::P spxB -cps19AJ(D64A)], and NUS2067 [CPS19C DbgaA::P spxB -cps19AJ(G71R)] were grown in BHI medium, serially diluted, and spotted on blood plates containing 0.3 mg/ml of bacitracin. Plates were imaged after overnight incubation at 37°C in 5% CO 2 . (C) Cultures of strains NUS0447 (CPS19A DbgaA::P-sacB-kan-rpsL 1 ), NUS1558 (CPS19A DbgaA::P spxB -cps19AJ), NUS1525, NUS1527, NUS1568, NUS1584, NUS2066, and NUS2067 were grown in BHI to an OD 600 of 0.2 and then imaged by phase-contrast microscopy. Cells that exhibited cell shape defects are indicated by red arrows. Scale bar, 2 mm. (D) The areas of the cells in panel C were quantified by using MicrobeJ (n = 4205). n.s., not significant. P values were computed by using the Mann-Whitney U test. not lead to the same degree of increase in the bacitracin sensitivity and cell shape defects, albeit the colonies were smaller on plates with bacitracin (Fig. 6). This result argues that the increased bacitracin sensitivity is not due to the general toxicity caused by the overexpression of membrane proteins. Notably, the inactive Cps19AJ variants were produced at a similar level compared to the wild-type Cps19AJ (see Fig. S7), indicating that the mutant proteins are not destabilized. Together, our results suggest that the expression of a flippase that can transport an incomplete precursor may lead to a reduced Und-P pool, which results in an increased sensitivity to bacitracin.

DISCUSSION
The polymorphic CPS is one of the most important virulence factors in S. pneumoniae. Other than serotypes 3 and 37, all pneumococcal CPSs are synthesized by the Wzx/Wzy pathway (3,4,12). A critical step in this pathway is the translocation of the lipid-linked precursors across the cytoplasmic membrane. This step is thought to be mediated by the widely conserved MOP-family flippase CpsJ (22). Many studies to elucidate the transport mechanism of MOP transporters were carried out using the lipid II flippase MurJ as a prototype (29,39,40,42,43). The prevailing model is that the two pseudosymmetrical bundles of transmembrane helices adopt a V-shaped configuration. To initiate the transport cycle, the lipid-linked precursor at the inner leaflet of the cell membrane interacts with the central hydrophilic cavity of the flippase and destabilizes the inward-open conformation, resulting in the inward-occluded fold. This conformation prevents the substrate from diffusing back into the cytoplasm (39), before the transporter transitions to the outward-facing conformation and releases the cargo to the outer leaflet of the cell membrane. To restore the inward-open fold, an energy source such as a Na 1 or H 1 gradient is presumably required (25,42), although this is still a subject of debate (39). Finally, the inward-open MOP flippase initiates another round of substrate translocation.
Photo-cross-linking and biotin-tagging experiments provided evidence that the substrate transport is coupled with the transition between the inward-open and outward-open conformations (40), although the electron density of the cargo could not be unambiguously assigned in the crystal structure (39). Nevertheless, how the interaction between the precursor and the flippase results in the conformational change is still unclear. Biochemically reconstitution of the translocation event is expected to be challenging, because it does not chemically modify the substrate (20). In addition, the lipid-linked cargos are usually difficult to obtain. We reasoned these problems could be partially circumvented by taking a genetic approach. In the present study, CpsJ variants from various serotypes were introduced into a serotype 2 strain, and we tested whether they could replace Cps2J. The nine cross-complementing CpsJ variants clustered around serogroup 33 (see Fig. S1), suggesting this group of flippases may have relatively relaxed specificity. We also demonstrated that Cps33BJ does not support the translocation of PG and teichoic acid precursor, so it is not totally nonspecific. Unlike Cps2J, it appears that Cps33BJ could only be substituted by its close homologs. We noticed that the precursors of CpsJ variants that complemented cps33BJ function contain galactofuranose at the non-reducing end (12), similar to the serotype 33B CPS precursor. Perhaps the galactofuranose residue is recognized by the flippases. We have yet to fully understand why certain CPS flippases can transport noncognate substrates. Nevertheless, we believe identifying the unexpected cross-complementation events in the CPS pathway may inform future bioinformatics and structural biology efforts to address the longstanding question of MOP-transporter specificity.
We showed that mutations at the extracellular loops of Cps2J could relax its substrate specificity. Similar mutations were identified in an unrelated MOP flippase WzxC (24), and these results support a unifying transport mechanism. While computational docking experiments using the structural models of CpsJ placed the repeating units in the central aqueous cavity (see Fig. S6), we could not clearly identify a region responsible for substrate binding. Work is under way to reveal the substrate binding pocket by solving the CpsJ structures. In addition, Mut-seq results for Cps2J and MurJ (29) did not detect any strictly immutable residue, indicating that the overall fold of these transporters is robust, although they belong to the evolutionarily diverged MVF (mouse virulence factor) or PST (polysaccharide transport) families.
In response to the host adaptive immunity, the capsule of a bacterial pathogen may need to be constantly modified to escape the anti-CPS antibodies. In addition to the variations of glycosyltransferases that alter the repeating units, the new precursors must be able to reach the other side of the cell membrane to complete synthesis. This can be achieved by the overexpression of the flippase involved (33) or the accumulation of mutations at the extracellular/periplasmic gate (24). In this study, we found that changes in the central aqueous cavity can also allow the transporter to flip a noncognate substrate. Since S. pneumoniae is naturally competent, it may acquire flippase genes from the nasopharyngeal microbiome (44,45). We propose the flippases in the clade of serogroup 33 (see Fig. S1) may be evolutionary intermediates that retained some promiscuity until another set of mutations convert them back into the presumably more efficient flippases that dedicate to their substrates (46).
One possible driving force that maintains flippase specificity is that a flippase capable of transporting incomplete precursors across the cell membrane may reduce the amount of free Und-P (Fig. 1). This model is supported by the mild toxicity due to the overexpression of a Cps19AJ (Fig. 6). A similar observation was noted when a promiscuous flippase Wzk was overproduced in Escherichia coli (36). This toxicity was alleviated by reducing the inducer concentration (36). In addition to sequestering Und-P, the incomplete precursor on the cell surface may also act as a competitive inhibitor of the downstream enzymes since they are structural homologs of the native substrates. Nevertheless, the phenotype observed in this study was rather mild. A plausible explanation is that CpsJ may be able to translocate the unfinished precursors back to the cytoplasm (see Fig. S7). Alternatively, CPS enzymes may be organized in a supramolecular complex and prevent diffusion of the incomplete precursors through metabolic channeling (47). These possibilities warrant further research. Perhaps more importantly, the incomplete glycoconjugates displayed on the cell surface may be vulnerable to the host's defense system, such as being recognized by the C-type lectin SIGN-R1 (a murine homolog of DC-SIGN in human) (48). In plants, bacterial flagellin fragments are sensed by the FLS2 receptor, unless they are properly shielded by glycosylation. To "visualize" the flagellin glycopeptides, the host produces glycosidases such as BGAL1 to strip the protective sugar residue at the nonreducing ends (49). As a defense strategy, the MOP flippases in pathogens likely serve as a molecular checkpoint to ensure that the vulnerable glycans remain inside the cytoplasm.

MATERIALS AND METHODS
Bioinformatics analysis. Neighbor-joining phylogenetic trees were generated using the Molecular Evolutionary Genetics Analysis version 7 (MEGA7) software (50). First, CpsJ sequences were retrieved from the National Center for Biotechnology Information (NCBI) website and aligned using the built-in ClustalW program. The evolutionary distance was calculated in the unit of the number of amino acid difference per site using the "p-distance method" and the "15-aa-window" settings of the MEGA7 software. The confidence level of the branch was tested by bootstrapping 500 replicates. Pairwise sequence identities were calculated using Clustal Omega (51). Structural models of CpsJ were generated by I-TASSER (52) and visualized with PyMOL (http://pymol.org). The number of transmembrane helices in the flippases was predicted using TMHMM (53). Structures of the CPS repeating units were downloaded from the bacterial carbohydrate database (http://csdb.glycoscience.ru/), and their PDB files were generated by REStLESS (54). They were docked with the CpsJ structural models using Autodock Vina (55).
Media, culture conditions, and bacterial strains. The strains used in this study are listed in Table S1 in the supplemental material. Clinical isolates of S. pneumoniae expressing 93 types of CPS were collected from the local and international hospitals and research institutes, as indicated in Table S1. Unless otherwise specified, cells were grown in brain heart infusion broth (BHI; Thermo Fisher Scientific) or on tryptic soy agar plates supplemented with 5% (vol/vol) sheep blood (blood plates) (Biomed Diagnostics) at 37°C in 5% CO 2 . Antibiotics were purchased from Sigma-Aldrich and used at final concentrations of 0.3 mg/ml for erythromycin (Erm), 250 mg/ml for kanamycin (Kan), 0.3 mg/ml for bacitracin, and 300 mg/ml for streptomycin (Str). We noticed that bacitracin could not be evenly spread on blood plates, and thus we freshly prepared the plates by mixing bacitracin with the agar solution before pouring it. When indicated, ZnCl 2 and MnCl 2 were added to liquid cultures at final concentrations of 400 and 40 mM, respectively. Mn 2+ was added to the medium to alleviate Zn 21 toxicity (56,57). When blood plates were used, the concentrations of ZnCl 2 and MnCl 2 were adjusted to 500 and 50 mM, respectively.
Strain construction. The primers used for synthesizing the PCR amplicons are listed in Table S1. In general, PCR fragments were synthesized using high-fidelity Phusion DNA polymerase (NEB M0530S) according to the manufacturer's instructions. PCR products were diagnosed by gel electrophoresis and purified using a QIAquick PCR purification kit (Qiagen, catalog no. 28106). Pneumococcal cells were transformed after inducing natural competence, with cassettes assembled by overlap extension PCR or isothermal assembly (58,59). Transformants were selected on blood plates supplemented with the indicated antibiotics. Allelic replacement was performed using the Janus cassette (P-kan-rpsL 1 ) or the Sweet Janus cassette (P-sacB-kan-rpsL 1 ) as described previously (38,60). We could not obtain amplicons of cpsJ from serogroup 11, serotype 16A, and serotype 35F, but the remaining 82 cpsJ variants were introduced into strain NUS0650 (rpsL1 Dcps2J::P-sacBkan-rpsL 1 //DbgaA::P Zn -cps2J) by transforming the cassette harboring the corresponding cpsJ allele. Transformants were selected for Str and sucrose resistance. The resulting strains were validated by PCR using GoTaq DNA polymerase (Promega, M712) and Sanger sequencing.
Microscopy and measurements of growth. Overnight cultures of strain NUS0063 (DbgaA::P Zn -cps2J) and NUS0084 (Dcps2J//DbgaA::P Zn -cps2J) were grown in BHI supplemented with ZnCl 2 and MnCl 2 , as described above. When the optical density at 600 nm (OD 600 ) reached 0.2 to 0.4, cultures were normalized to an OD 600 of 0.2. Cells were pelleted by centrifugation, washed with 1 ml of BHI medium twice to remove residual ZnCl 2 and MnCl 2 , and diluted to an OD 600 of 0.01 in BHI with or without ZnCl 2 /MnCl 2 . The diluted cultures were distributed into a 96-well plate at a final volume of 200 ml. Growth was monitored at 37°C for 10 h using a Tecan microplate reader. OD 600 readings were taken every 10 min immediately after a short purse with 30 s of shaking.
To visualize the cells deprived of cps2J, overnight culture of strain NUS0084 (Dcps2J//DbgaA::P Zn -cps2J) was diluted to an OD 600 of 0.01 and grew for ;6 h at 37°C. Just prior to the onset of cell lysis, the cells were harvested by centrifugation at 3,000 Â g for 5 min at room temperature. The supernatant was removed, and the pellets were resuspended in 50 ml of BHI. Cells were mounted on a glass slide and imaged with an IX81 phase-contrast microscope (Olympus). Micrographs were analyzed and quantitated using MicrobeJ, as described previously (61).
Immunoblotting. Immunoblotting was done essentially as described previously (62). Briefly, cultures were grown to an OD 600 of 0.2 to 0.4. After adjusting the OD 600 of the culture to 0.3, 1 ml of the culture was collected and centrifuged at 16,100 Â g for 1 min at room temperature. Spent medium was discarded and the pellets were washed once with 1Â phosphate-buffer saline (PBS) and resuspended in 200 ml of protoplast buffer (0.5 M sucrose, 20 mM MgCl 2 , 20 mM Tris-HCl [pH 7]). To normalize the samples based on protein concentrations, we measured the amount of protein in the cell suspension using the NI protein assay kit (G-Biosciences, catalog no. 786-005). Alternatively, normalization of the loading volume based on the OD 600 readings of the culture produced nearly identical results. The suspension was treated by adding 4 ml of 10 U/ml mutanolysin (Sigma-Aldrich, catalog no. M9901-10KU) and 15 ml of 10 mg/ml lysozyme (Sigma-Aldrich, 62970-1G-F) to digest the cell wall. Spheroplast formation was confirmed by microscopy, and they were pelleted by centrifugation at 3,000 Â g for 10 min at room temperature. Supernatant contain the digested cell wall were collected. The pellets (spheroplasts) were lysed by resuspension in 200 ml of 1Â PBS. Samples were mixed with an equal volume of 2Â Laemmli sample buffer (Bio-Rad), digested by adding 2 ml of 20 mg/ml proteinase K (Qiagen 19133) and incubation at 50°C for 1 h, before separated to a 4/10% SDS-PAGE gel. The CPS was transferred to a polyvinylidene difluoride membrane and blocked by incubation for 30 min in blocking solution (PBST; 1Â PBS with 0.05% [vol/vol] Tween 20) with 5% (wt/vol) skim milk. To the blot, 50 ml of cell lysate from strain HMS0002 (rpsL1 DcpsE) was added to minimize antibody binding to non-CPS materials (22). After mixing for 2 min, the anti-CPS serum was added at a dilution of 1:5,000, and the blot was incubated overnight at 4°C with shaking. The blot was washed two times with PBST for 5 min and incubated with the antirabbit antibodies conjugated with horseradish peroxidase (HRP; Thermo Fisher Scientific, A16110) at a dilution of 1:10,000 in blocking solution. After incubation for 1 h at room temperature with shaking, the membrane was washed three times with PBST and detected by using enhanced chemiluminescence reagents (Thermo Scientific, catalog no. 34580) according to the manufacturer's protocol.
To detect FLAG-tagged CpsJ (63), strains expressing cpsJ-FLAG variants (cps2J,.cpsJ-FLAG//P Zn -cps2J) were grown in BHI with ZnCl 2 /MnCl 2. Cultures were normalized to an OD 600 of 0.3 and then centrifuged at 16,100 Â g for 1 min at room temperature; next, the pellets were resuspended in 100 ml of 1Â PBS. The cell suspension was mixed in a 1:1 ratio with 2Â Laemmli buffer containing 5% (vol/vol) b-mercaptoethanol and incubated at 50°C for 5 min. Proteins were separated on a 4/12% SDS-PAGE gel, transferred to a polyvinylidene difluoride membrane, and blotted with the anti-FLAG polyclonal antibodies (1:3,000 dilution; Sigma-Aldrich, F7425), followed by anti-rabbit antibodies conjugated to HRP (1:5,000 dilution). Signal detection was done using the chemiluminescence substrate as described above.
Mutagenesis sequencing and isolation of cpsJ variants with an expanded specificity. Amplicons of cps2J and cps23BJ were PCR mutagenized as described before (24,64) except using the oligonucleotides listed in Table S1 and GoTaq DNA polymerase. Briefly, the cps2J or cps23BJ cassettes were amplified for 28 cycles to ensure a low mutation rate. As a negative control, the same amplicons were synthesized using Phusion DNA polymerase. PCR products were purified and concentrated using the Qiagen PCR purification kit. Strain NUS0893 (rpsL1 Dcps2E Dcps2J::P-sacB-kan-rpsL 1 //DbgaA::P Zn -cps2E) was transformed with the mutagenized and control PCR amplicons and selected for Str and sucrose resistance in the absence of ZnCl 2 /MnCl 2 . Transformants were pooled and stored at 280°C or directly used to select for the phenotypes described below.
For Mut-seq, the mutagenized cps2J library was plated on blood plates with or without ZnCl 2 /MnCl 2 . Colonies on the plates were resuspended in BHI and pooled. The suspension was adjusted to an OD 600 of ;0.7 with BHI, and genomic DNA was extracted using a DNeasy Blood & Tissue kit (Qiagen, catalog no. 69506). The cps2J region was amplified by KAPA DNA polymerase (Roche, catalog no. KK2611) using primers P1694 and P1695. The amount of PCR amplicons was quantified using the Qubit dsDNA HS assay kit (Invitrogen), and the cps2J allele was sequenced on a NovaSeq 6000 platform (NovogeneAIT). Data analysis was performed using the CLC workbench software (Qiagen) as described previously (29). In brief, sequencing reads were mapped to the cps2J region with the "length" and "similarity fraction" parameters set to 1. Unmapped reads were collected and aligned again to the cps2J open reading frame, but with the "similarity fraction" reduced to 0.98. Base changes were detected by the variant detection tool in the CLC workbench software and exported to Microsoft Excel for further analysis.
To select for the cps23BJ variants that can transport serotype 2 CPS precursor, cells harboring mutagenized cps23BJ alleles in the NUS0893 (rpsL1 Dcps2E Dcps2J::P-sacB-kan-rpsL 1 //P Zn -cps2E) background were plated on blood plates with ZnCl 2 /MnCl 2 after serial dilutions. Under this condition, cells harboring cps23BJ variants that were unable to flip the serotype 2 CPS precursor could not survive. Indeed, we observed a significant increase (;10-fold) in plating efficiency for cells expressing a mutagenized copy of cps23BJ, compared to the cells expressing cps23BJ 1 . Survivors were screened for capsule production by immunostaining using the antisera specific for serotype 2 CPS (SSI; i.e., Quellung reaction) (65). Encapsulated mutants were stored, and the cps23BJ alleles were sequenced. To ensure the phenotype is linked to the cps23BJ alleles, the variants were PCR-amplified and used to transform the parent strain NUS0893 (rpsL1 Dcps2E Dcps2J::P-sacB-kan-rpsL 1 //DbgaA::P Zn -cps2E). Similarly, the gain-of-function cps23BJ alleles were introduced into strain NUS0650 (rpsL1 Dcps2J::P-sacB-kan-rpsL 1 //DbgaA::P Zn -cps2J) in which the CPS level is similar to that of the wild-type strain.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. DATA SET S1, XLSX file, 0.7 MB.