All authors designed research, analysed data and wrote the paper; S.-Y.P. performed research.
The two-component system PhoP/PhoQ controls a large number of genes responsible for a variety of physiological and virulence functions in Salmonella enterica serovar Typhimurium. Here we describe a mechanism whereby the transcriptional activator PhoP elicits expression of dissimilar gene sets when its cognate sensor PhoQ is activated by different signals in the periplasm. We determine that full transcription of over half of the genes directly activated by PhoP requires the Mg2+ transporter MgtA when the PhoQ inducing signal is low Mg2+, but not when PhoQ is activated by mildly acidic pH or the antimicrobial peptide C18G. MgtA promotes the active (i.e. phosphorylated) form of PhoP by removing Mg2+ from the periplasm, where it functions as a repressing signal for PhoQ. MgtA-dependent expression enhances resistance to the cationic antibiotic polymyxin B. Production of the MgtA protein requires cytoplasmic Mg2+ levels to drop below a certain threshold, thereby creating a two-tiered temporal response among PhoP-dependent genes.
Bacteria generally respond to a stress condition by changing their gene expression programmes (Kato et al., 2007; Kroos, 2007; Battesti et al., 2011). The change in gene expression often requires a sensor that detects a specific chemical or physical cue(s) and alters the levels and/or activity of a cognate regulatory protein. Moreover, it results in modifications of cellular components that help bacteria cope with the stress condition and that eventually lead to adaptation to such a condition. We now identify an unusual behaviour for a bacterial regulatory system by establishing that it promotes expression of different genes depending on both the signal activating the sensor and the extent of time the organism experiences an inducing signal.
The two-component system PhoP/PhoQ controls virulence, Mg2+ homeostasis, and resistance to acidic pH and to antimicrobial peptides in the pathogen Salmonella enterica serovar Typhimurium and other Gram-negative species (Ernst et al., 2001; Groisman, 2001). In Salmonella, the sensor PhoQ responds to low Mg2+ (Garcia Vescovi et al., 1996), mildly acidic pH (Prost et al., 2007) and certain antimicrobial peptides (Bader et al., 2005) in the bacterial periplasm by promoting the active (i.e. phosphorylated) state of the regulator PhoP (i.e. PhoP-P). PhoP-P binds to its target promoters and recruits RNA polymerase, thereby advancing transcription of PhoP-activated genes (Shin and Groisman, 2005; Shin et al., 2006).
Surprisingly, the set of PhoP-activated genes expressed when Salmonella experiences low Mg2+ is not identical to that induced by mildly acidic pH or the antimicrobial peptide C18G (Groisman and Mouslim, 2006). For instance, acidic pH induces transcription of the full-length Fe2+ transporter gene feoB but not of the full-length Mg2+ transporter gene mgtA, and the converse is true when the PhoQ inducing signal is low Mg2+ (Choi et al., 2009). This behaviour reflects: first, that the PhoP-activated regulator RstA is necessary for transcription of feoB but not of mgtA in mildly acidic pH (Choi et al., 2009), and second, that after transcription initiation promoted by PhoP, the 5′ leader region of the mgtA transcript operates as a Mg2+-responding riboswitch that favours transcription elongation into the mgtA coding region when cytoplasmic Mg2+ levels drop below a certain threshold (Cromie et al., 2006; Zhao et al., 2011). By contrast, the feoB transcript lacks such sequences and is not regulated by Mg2+ levels (Choi et al., 2009). Thus, the distinct expression behaviours displayed in acidic pH versus low Mg2+ reflect the participation of an additional transcription factor and cis-acting regulatory sequences in a particular PhoP-dependent transcript, respectively.
Here we report that low Mg2+, mildly acidic pH and an antimicrobial peptide promote similar levels of PhoP-activated mRNAs 2 h post induction, but that, unexpectedly, the levels of a subset of PhoP-activated genes are ∼ 5–35 times higher when examined 4 h post induction if the inducing signal is low Mg2+ but alike to those produced at 2 h when PhoQ is activated by the other PhoQ-inducing signals. We establish that the enhanced expression detected at 4 h requires Mg2+ uptake by the PhoP-activated Mg2+ transporter MgtA, which increases PhoP-P levels by removing Mg2+ from the periplasmic domain of PhoQ. This strategy enables Salmonella to delay full transcription of part of the PhoP regulon until the cytoplasmic conditions resulting in production of the MgtA protein are met. MgtA-dependent gene expression promotes heightened resistance to the antibiotic polymyxin B. Our findings highlight how a regulatory system promotes distinct responses when stimulated by different signals or at different times when stimulated by a given signal.
Distinct temporal expression of PhoP-activated genes in response to mildly acidic pH and low Mg2+
We examined the mRNA levels of the PhoP-activated genes pagC and pagK at 2 h and 4 h after Salmonella was shifted from non-inducing conditions for the PhoQ protein to media containing either mildly acidic pH or low (i.e. 10 μM) Mg2+. When wild-type Salmonella experienced mildly acidic pH, the mRNA levels corresponding to the pagC coding region were similar following incubation for 2 h or 4 h (Fig. 1); and the same was true for the pagK transcript (Fig. 1). Surprisingly, when the inducing signal was low Mg2+ the mRNA levels for both pagC and pagK were 10 times higher at 4 h than at 2 h (Fig. 1). The mRNA levels produced at 2 h were similar whether the inducing signal was mildly acidic pH or low Mg2+ (Fig. 1). These results indicate that different PhoQ inducing signals can elicit distinct responses from PhoP-activated genes depending on the extent of time Salmonella experiences an inducing condition.
The MgtA protein is necessary for full transcription of a subset of PhoP-activated genes when the PhoQ inducing signal is low Mg2+
The difference in expression manifested between 2 h or 4 h of incubation in low Mg2+ (Fig. 1) could reflect the participation of the MgtA protein in transcription of a subset of PhoP-activated genes. This is because: first, the MgtA protein is detected at 4 h but not at 2 h after wild-type Salmonella is switched to media containing 10 μM Mg2+ (Cromie and Groisman, 2010), and second, the Mg2+ transporter MgtA might enhance PhoP-P levels by removing Mg2+ from the periplasm, where it operates as a repressing signal for PhoQ (Garcia Vescovi et al., 1996).
Wild-type and mgtA strains had similar pagC and pagK mRNA levels at 2 h post induction (Fig. 2A) (i.e. ratio of wild-type/mgtA of 1). However, at 4 h the ratio was much higher (Fig. 2A), and this reflected more mRNA in the wild-type strain (Fig. 2A). Wild-type pagC and pagK mRNA levels were restored to the mgtA mutant by a plasmid expressing the wild-type mgtA open reading frame from a heterologous promoter but not by the plasmid vector (Fig. 3). The MgtA-dependent expression is specific to the low Mg2+ signal because the mRNA levels corresponding to the pagC and pagK genes were similar between isogenic wild-type and mgtA strains when Salmonella was incubated with the antimicrobial peptide C18G or at mildly acidic pH (Fig. 2B). Moreover, high expression of the pagC and pagK genes in low Mg2+ is correlated with the presence of the MgtA protein, which was barely detectable at 2 h but highly induced by 4 h (Fig. 2C), in agreement with previous results (Cromie and Groisman, 2010).
To determine whether the MgtA protein is necessary for transcription of members of the PhoP regulon other than pagC and pagK, we examined the mRNA levels of 19 additional genes known to be directly activated by the PhoP protein (Zwir et al., 2012) in isogenic wild-type and mgtA strains following a 4 h incubation in 10 μM Mg2+. The mRNA levels for 11 of these genes were lower in the mgtA mutant than in the wild-type strain (Fig. 2D). Altogether, the most dramatic effect was observed for the pagC, pagD, pagK, pagM and pgtE genes, where the expression difference was > 10-fold. By contrast, the mRNA levels for the remaining eight genes were similar in the two strains (Fig. 2D).
When the phoP and phoQ genes were transcribed from a heterologous (i.e. PhoP-independent) promoter (Shin et al., 2006) (Fig. S1), the pagC and pagK mRNA levels were still higher in an mgtA+ strain than in an mgtA mutant. By contrast, mRNA levels of the MgtA-independent slyB were similar in both strains (Fig. S1). Therefore, the mgtA requirement for normal expression of a subset of PhoP-activated genes does not result from inefficient transcription of the autoregulated phoP gene (Soncini et al., 1995) in the mgtA mutant (Fig. S1). Cumulatively, the data presented in this section indicate that the MgtA protein is necessary for expression of a subset of PhoP-activated genes when low Mg2+ is the signal activating PhoQ.
MgtA's ability to take up Mg2+ is required for expression of PhoP-activated genes
MgtA appears to promote PhoP-dependent gene transcription via its established role as a Mg2+ uptake protein because the mgtA mutant were complemented by a plasmid expressing the wild-type mgtA gene but not by one specifying a Mg2+ transport-defective MgtA variant with an amino acid substitution in a phosphorylation site (Tao et al., 1995), which behaved like the plasmid vector (Fig. 3). Even a plasmid harbouring the Bacillus subtilis mgtE gene, which specifies an unrelated Mg2+ transporter (Smith et al., 1995), partially rescued the mgtA mutant (Fig. 3). (We ascribe the incomplete rescue to suboptimal expression and/or activity of the Gram-positive MgtE protein in the Gram-negative S. enterica.) The Mg2+ transporter-expressing plasmids had modest (i.e. < 2-fold) effects on the slyB mRNA levels (Fig. 3), which are not affected by mutation of the mgtA gene (Fig. 2D).
PhoQ's ability to sense periplasmic Mg2+ is necessary for expression of MgtA-dependent genes
If MgtA exerts its regulatory function by removing Mg2+ from PhoQ, deletion of the mgtA gene should not affect expression of PhoP-activated genes in a strain lacking the phoQ gene. Given that PhoQ is essential to activate the wild-type PhoP protein (Chamnongpol and Groisman, 2000), we used a strain harbouring the phoP* allele, which specifies a PhoP variant that displays enhanced autophosphorylation from acetyl phosphate in vitro and in vivo (Chamnongpol and Groisman, 2000), and thus, can promote transcription of PhoP-activated genes in the absence of PhoQ. As proposed, similar levels of pagC and pagK transcripts were produced by isogenic mgtA strains in a phoP* phoQ background (Fig. S2).
To specifically address whether PhoQ's capacity to sense periplasmic Mg2+ is necessary for MgtA's regulatory function, we compared the pagC and pagK mRNA levels produced by isogenic phoQ and phoQ mgtA strains harbouring a plasmid encoding the wild-type PhoQ protein or a PhoQ variant with two amino acid substitutions in the periplasmic domain that render it blind to inactivation in high Mg2+ (Chamnongpol et al., 2003). The phoQ and phoQ mgtA strains expressing the Mg2+-blind PhoQ displayed similar mRNA levels for the pagC and pagK genes (Fig. 4). By contrast, they were > 20-fold higher in cells expressing the wild-type PhoQ protein (Fig. 4). As expected, the slyB mRNA levels were similar in the isogenic mgtA strains (Fig. 4).
MgtA enhances PhoP-P levels
Mg2+ binds to the periplasmic domain of PhoQ (Garcia Vescovi et al., 1996; Chamnongpol et al., 2003), thereby decreasing PhoQ's ability to autophosphorylate (Castelli et al., 2000) and stimulating its ability to dephosphorylate PhoP-P (Vescovi et al., 1997). To explore whether an MgtA-mediated decrease in periplasmic Mg2+ alters PhoP-P levels in the cytoplasm, we investigated two sets of isogenic mgtA strains expressing our previously described C-terminally HA-tagged PhoP protein from the normal chromosomal location (Shin and Groisman, 2005; Shin et al., 2006). This enabled us to use anti-HA antibodies to follow PhoP-P in vivo.
The levels of PhoP-P were higher in the mgtA+ than in the mgtA mutant when bacteria were examined following 4 h in low Mg2+ media. This was true not only for a strain transcribing the phoP-HA phoQ genes from the wild-type (i.e. PhoP-autoregulated) promoter (Fig. S3), but also for a strain where the phoP-HA phoQ genes were transcribed from a constitutive (i.e. PhoP-independent) promoter (Fig. 5). Total PhoP levels were also higher in the mgtA+ strain than in the mgtA mutant when the phoP-HA phoQ genes were transcribed from the wild-type promoter (Fig. S3), reflecting the effect of positive feedback (Soncini et al., 1995). As expected, total PhoP levels were the same in isogenic mgtA strains when the phoP-HA phoQ genes were transcribed constitutively (Fig. 5). PhoP-P levels were undetectable in all investigated strains when bacteria were examined after only 2 h in low Mg2+ (Fig. 5 and Fig. S3). The latter result likely reflects low sensitivity of the assay because PhoP-activated genes are transcribed after 2 h in low Mg2+ (Fig. 1) and because PhoP-P is the form of PhoP that binds to its target promoters in vivo (Shin and Groisman, 2005). Cumulatively, these results are in agreement with the notion that MgtA exerts its regulatory effects by enhancing PhoP-P levels (Fig. 2).
MgtA promotes time-dependent resistance to polymyxin B
The PhoP-activated pmrD and ugtL genes encode products mediating the chemical modification of negatively charged residues in the lipopolysaccharide (LPS), thereby conferring resistance to the cationic peptide antibiotic polymyxin B (Kox et al., 2000; Shi et al., 2004). Because the mRNA levels corresponding to pmrD and ugtL were approximately fivefold lower in the mgtA mutant than in the wild-type strain (Fig. 2D), we wondered whether an mgtA mutant might display enhanced susceptibility towards polymyxin B. Indeed, survival of wild-type Salmonella was > 40-fold higher than that of the mgtA mutant when bacteria were tested following a 4 h incubation in low (i.e. 10 μM) Mg2+ (Fig. 6). By contrast, wild-type and mgtA Salmonella exhibited similar low survival when grown for only 2 h in low Mg2+ (Fig. 6). These data are consistent with the notion that MgtA is necessary for expression of PhoP-activated LPS modification genes to the levels required for resistance to polymyxin B.
We have now demonstrated that Salmonella utilizes the transporter MgtA to achieve differential expression within the PhoP regulon when the PhoQ inducing condition is low Mg2+ but not when PhoQ is stimulated by mildly acidic pH or an antimicrobial peptide in the presence of high Mg2+ (Fig. 7). The PhoP-activated MgtA protein exerts positive feedback on the PhoP/PhoQ system by removing Mg2+ from the periplasm, where it serves as an inhibitory signal for the sensor PhoQ (Garcia Vescovi et al., 1996). This action enhances the levels of PhoP-P (Fig. 5 and Fig. S3), the active form of the transcription factor PhoP (Shin and Groisman, 2005).
MgtA appears to increase PhoP-P levels by its established role as a transporter that moves Mg2+ away from PhoQ in the periplasm and into the cytoplasm (Tao et al., 1995) because the mgtA mutant were complemented by the wild-type mgtA gene and by mgtE, a gene specifying a heterologous Mg2+ transporter, but not by a gene encoding an MgtA variant that cannot take up Mg2+ (Fig. 3). Moreover, deletion of the mgtA gene did not decrease the levels of PhoP-activated mRNAs in a strain with a PhoQ variant that is blind to periplasmic Mg2+ (Fig. 4). (Although our data are consistent with MgtA promoting a decrease in periplasmic Mg2+, we note that the periplasmic Mg2+ concentration was not examined.) Therefore, the mechanism utilized by MgtA to regulate the PhoP/PhoQ system differs from those described for other transporters, which modulate the enzymatic activities of a sensor or sequester cytoplasmic regulators in the membrane (Tetsch and Jung, 2009).
MgtA controls expression timing within the PhoP regulon
The precise order of gene expression is critical for the construction of complex structures such as the bacterial flagellum (Kalir et al., 2001), during a bacterial cell cycle (McAdams and Shapiro, 2003), and for organismal development (Kunkel et al., 1989). Ordered expression can be accomplished in a variety of ways. For example, in the transcriptional cascade governing flagella production in Salmonella, several regulators control transcription of a master regulator, which in turn promotes expression of yet another regulator responsible for transcription of the flagellin gene (Kutsukake et al., 1990).
MgtA enables distinct expression timing among PhoP-dependent genes by advancing PhoP-P levels (Fig. 5 and Fig. S3). This is the result of two facts: first, PhoP-dependent promoters differ in the number, location, orientation and sequence of the PhoP binding sites (Zwir et al., 2012), suggesting they are activated by different amounts of PhoP-P, and second, transcription elongation into the mgtA coding region requires cytoplasmic Mg2+ levels to drop below the threshold that favours formation of a conformation in the mgtA leader RNA (Cromie et al., 2006) that hinders loading of the transcription termination factor Rho (Hollands et al., 2012). Therefore, a temporal expression profile is created among PhoP-activated genes based on whether transcription of these genes requires MgtA to generate the levels of PhoP-P necessary for expression. The MgtA-dependent temporal response occurs when Salmonella experiences low Mg2+ but not if the PhoQ inducing condition is mildly acidic pH in high Mg2+ (Fig. 1), presumably because high cytoplasmic Mg2+ prevents MgtA synthesis (Cromie et al., 2006; Hollands et al., 2012). This mechanism enables Salmonella to time an expression response by incorporating a cytoplasmic signal that is generated some time after experiencing inducing conditions.
Distinct properties of MgtA-mediated feedback
The positive feedback exercised by the MgtA protein differs from other feedback mechanisms operating on the PhoP/PhoQ system. First and foremost, the MgtA-mediated effect allows Salmonella to integrate cytoplasmic signals into the output of the PhoP/PhoQ system. This is because transcription elongation into the mgtA coding region responds to cytoplasmic Mg2+ (Cromie et al., 2006) and/or proline-charged tRNAPro (Park et al., 2010). Second, MgtA activates the PhoP/PhoQ system by removing an inhibitory signal for PhoQ. This distinguishes MgtA from MgrB, a PhoP-activated membrane peptide proposed to inhibit PhoQ's kinase activity in Escherichia coli by interacting with the extracytoplasmic domain of PhoQ (Lippa and Goulian, 2009). And third, positive transcriptional feedback of PhoP-P on the phoPQ promoter increases the levels of both PhoP and PhoQ (Shin et al., 2006), whereas that mediated by MgtA affects the amount of PhoP-P (Fig. 5 and Fig. S3).
The positive feedback that MgtA exercises on the PhoP/PhoQ system is also unique in that it takes place at a time when other regulatory systems are downregulated due to adaptation to the conditions that trigger their activation in the first place. For instance, adaptation in bacterial chemotaxis takes place in a second timescale (Tu et al., 2008). And expression of genes activated by the regulator PmrA dramatically decreases 2 h after continuous exposure to Fe3+ (Kato et al., 2012), which is an inducing signal for PmrA's cognate sensor PmrB (Wosten et al., 2000). This decrease reflects the activities of PmrA-activated gene products, some of which modify the cell surface, thereby hindering Fe3+ access to PmrB (Kato et al., 2012). By contrast, after 4 h of continuous exposure to low Mg2+, the transporter MgtA removes the inhibitory signal Mg2+ from PhoQ, enabling a dramatic increase in transcription of over half of the PhoP regulon (Fig. 7).
Differential temporal expression of target genes has also been reported for the Bordetella BvgA/BvgS two-component system (Cotter and Jones, 2003; Boulanger et al., 2013). Bvg-A-activated ‘early’ and ‘late’ genes reflect the increase in phosphorylated BvgA (i.e. BvgA-P) protein resulting from positive feedback on the bvgA promoter, as well as the differential affinity of BvgA-P for its promoters. By contrast, the positive feedback on the PhoQ/PhoQ system is mediated by one of its activated gene products.
Different roles of the feedback mechanisms operating on the PhoP/PhoQ system
Our new results have uncovered a second positive feedback mechanism operating on the PhoP/PhoQ system, one that appears to play a different function in Salmonella's lifestyle than that provided by autogenous regulation (Shin et al., 2006). PhoP-P activates transcription from the phoP promoter soon after Salmonella experiences low Mg2+. This activation is needed to generate a surge in the levels of PhoP-P (Shin et al., 2006; Yeo et al., 2012), which is necessary for Salmonella virulence (Shin et al., 2006). By contrast, the MgtA-mediated feedback takes place relatively late as it requires cytosolic Mg2+ levels to drop below a certain threshold (Fig. 2A), and does not appear to be related to virulence because neither mgtA nor the PhoP-regulated genes whose expression is most affected upon mgtA deletion are necessary for Salmonella to cause a lethal infection in mice (Gunn et al., 1995; 1998; Blanc-Potard and Groisman, 1997).
What then is the physiological reason(s) behind Salmonella expressing a fraction of the PhoP regulon only after MgtA is made? One possibility is to help Salmonella cope with low cytoplasmic Mg2+. This is because low cytoplasmic Mg2+ is the condition that triggers production of the MgtA protein. Furthermore, MgtA furthers expression of genes mediating modification of phosphate residues in the lipid A portion of the LPS that are normally neutralized by Mg2+ ions (Fig. 2D) and enhanced polymyxin B resistance (Fig. 6). These LPS modifications might make Mg2+ available for import into the cytoplasm by MgtA to be used for enzymatic reactions that exhibit a strict dependence on Mg2+ (Groisman et al., 2013).
Because the MgtA protein promotes higher PhoP-P levels (Fig. 5 and Fig. S3), our findings suggest that different PhoP-dependent activities require distinct levels of PhoP-P. An analogous situation has been observed in B. subtilis where low levels of the regulator Spo0A promote expression of genes involved in cannibalism and biofilm formation, whereas high Spo0A levels are necessary for expression of genes participating in sporulation (Fujita et al., 2005).
The genes specifying the PhoP/PhoQ two-component system and the MgtA Mg2+ transporter are found in several members of the family Enterobacteriaceae in addition to S. enterica (Perez et al., 2009), suggesting that the MgtA-mediated positive feedback on PhoP/PhoQ may operate in other enteric bacteria. However, MgtA may control dissimilar behaviours in other species because the PhoP-activated genes exhibiting the highest dependence on MgtA in Salmonella (Fig. 2D) exhibit a limited phylogenetic distribution outside the Salmonella genus.
Finally, a mutation that increases transcription of the mgtA gene enhances thermotolerance in Salmonella (O'Connor et al., 2009). Although thermotolerance might be mediated directly by the Mg2+ taken up by the MgtA protein, our findings raise the possibility of MgtA exerting thermoprotection indirectly, by promoting expression of a PhoP-dependent gene(s) (Fig. 2D).
Bacterial strains, plasmids and growth conditions
Bacterial strains and plasmids used in this study are listed in Table S1. Details of strains and plasmid constructions are described in Supplementary materials and methods, and primers are listed in Supplementary Table S2. Bacteria were grown at 37°C in Luria–Bertani (LB) broth or in N-minimal medium, pH 7.7 (Snavely et al., 1991), supplemented with 0.1% casamino acids, 38 mM glycerol, and the indicated concentration of MgCl2. To test the effect of mildly acidic pH on expression of PhoP-dependent genes, N-minimal medium, pH 5.8, with 1 mM MgCl2 was used for bacterial growth. To test the effect of antimicrobial peptide on expression of PhoP-dependent genes, bacteria were grown in N-minimal medium, pH 7.7 with 1 mM MgCl2 and 5 μg ml−1 of C18G.
Determination of transcript levels
Total RNA was extracted using RNeasy® Mini Kit (Qiagen). cDNA was synthesized using SuperScript® VILO™ MasterMix (Life Technologies) following the manufacturer's instructions. Quantification of transcripts was performed by real-time PCR using Fast SYBR Green Master Mix (Applied Biosystems) in an ABI 7500 Sequence Detection System (Applied Biosystems). Data were normalized to the levels of 16S ribosomal RNA rrs gene, previously diluted 100-fold. A list of primers used for qRT-PCR is presented in Supplementary Table S3.
Phos-tag™ Western blot analysis
PhoP and PhoP-P were separated on 10% polyacrylamide gels containing acrylamide-Phos-tag™ ligand (Phos-tag™ Consortium) as previously described (Barbieri and Stock, 2008). Gels were copolymerized with 75 μM Phos-tag™ and 150 μM MnCl2. Whole-cell extracts were prepared as described (Wayne et al., 2012) with a few modifications. The cell extracts, normalized by OD600, were electrophoresed on Phos-tag™ gels with standard running buffer [0.4% (w/v) SDS, 25 mM Tris, 192 mM glycine] at 4°C under 150 V for 2 h, transferred to nitrocellulose membranes, and analysed by immunoblotting with anti-HA (Sigma-Aldrich) for PhoP-HA and anti-RpoD (Neoclone) monoclonal antibodies. The data are representative of three independent experiments, which gave similar results.
Polymyxin B survival assay
Assays were performed as described (Perez and Groisman, 2007) with a few modifications. Details of the assay are described in Supplementary materials and methods.
We thank Kyle Wayne for assistance with the Phos-tag protocol, and Varsha Raghavan, Kerry Hollands, John May, Mauricio H. Pontes, Anastasia Sevostiyanova and Jennifer Aronson for comments on the manuscript. This research was supported, in part, by Grant AI49561 from the National Institutes of Health to E.A.G., who is an investigator of the Howard Hughes Medical Institute.