The FUR-like regulators PerRA and PerRB integrate a complex regulatory network that promotes mammalian host-adaptation and virulence of Leptospira interrogans

Leptospira interrogans, the causative agent of most cases of human leptospirosis, must respond to myriad environmental signals during its free-living and pathogenic lifestyles. Previously, we compared L. interrogans cultivated in vitro and in vivo using a dialysis membrane chamber (DMC) peritoneal implant model. From these studies emerged the importance of genes encoding the Peroxide responsive regulators PerRA and PerRB. First described in in Bacillus subtilis, PerRs are widespread in Gram-negative and -positive bacteria, where regulate the expression of gene products involved in detoxification of reactive oxygen species and virulence. Using perRA and perRB single and double mutants, we establish that L. interrogans requires at least one functional PerR for infectivity and renal colonization in a reservoir host. Our finding that the perRA/B double mutant survives at wild-type levels in DMCs is noteworthy as it demonstrates that the loss of virulence is not due to a metabolic lesion (i.e., metal starvation) but instead reflects dysregulation of virulence-related gene products. Comparative RNA-Seq analyses of perRA, perRB and perRA/B mutants cultivated within DMCs identified 106 genes that are dysregulated in the double mutant, including ligA, ligB and lvrA/B sensory histidine kinases. Decreased expression of LigA and LigB in the perRA/B mutant was not due to loss of LvrAB signaling. The majority of genes in the perRA and perRB single and double mutant DMC regulons were differentially expressed only in vivo, highlighting the importance of host signals for regulating gene expression in L. interrogans. Importantly, the PerRA, PerRB and PerRA/B DMC regulons each contain multiple genes related to environmental sensing and/or transcriptional regulation. Collectively, our data suggest that PerRA and PerRB are part of a complex regulatory network that promotes host adaptation by L. interrogans within mammals. Author Summary Leptospirosis is a neglected tropical disease with a worldwide distribution. Globally, ~1 million cases and ~60,000 deaths are reported each year. The majority of cases of human leptospirosis are associated with Leptospira interrogans. Infection begins when a naïve reservoir (or incidental) host comes into direct or indirect contact with urine from an infected reservoir host. While infection in reservoir hosts, including rats and mice, is generally asymptomatic, incidental hosts, including humans, may develop clinical symptoms ranging from mild flu-like illness to fulminant disease. The gene products required by leptospires for infection remain poorly understood. Herein, we establish that the FUR family regulators PerRA and PerRB function in parallel, contributing to infectivity and renal colonization in mice. By comparative transcriptomics, we identified >100 genes that were dysregulated in the perRA/B double mutant cultivated in rat peritoneal cavities, including the virulence determinants LigA and LigB. Importantly, the PerRA, PerRB and PerRA/B DMC regulons contain multiple genes related to environmental sensing and/or transcriptional regulation. Our data suggest that PerRA and PerRB are part of a complex regulatory network that promotes host adaptation by L. interrogans within mammals.


Introduction
Leptospirosis is a neglected tropical disease with a worldwide distribution [1,2]. Globally, ~1 million cases and ~60,000 deaths are estimated each year [3]. Leptospirosis is now well recognized as a significant public health problem in developing countries and tropical regions [4][5][6]. In poor, urban communities in underdeveloped countries, major outbreaks of leptospirosis often are associated with seasonal flooding [4]. Leptospirosis also is of considerable veterinary importance; leptospirosis in cattle and other ruminants can lead to reduced reproductive fitness and diminished milk production, with substantial economic consequences [7,8].

Leptospirosis is caused by infection with pathogenic spirochetes belonging to the genus
Leptospira [9]. The majority of severe cases of human leptospirosis are associated with L.
interrogans [9]. Infection begins when a naïve reservoir (or incidental) host comes into contact with urine from an infected host, most often via contaminated water or soil [10]. Leptospires gain entry to the host through bruises or abrasions in the skin and/or mucous membranes. Following inoculation, leptospires transition from a saprophytic (free-living) to a parasitic lifestyle by a complex and poorly understood process referred to as 'host adaptation'. Once in the bloodstream, leptospires rapidly disseminate to distal tissues but, in reservoir host, are cleared within several days from all sites except the kidney, where they set up long-term residence in the proximal tubules [11][12][13]. Infected reservoir hosts shed large numbers of leptospires (up to 10 7 /mL) in their urine for weeks to months [9,12,14,15]. While infection in reservoir hosts is generally asymptomatic, incidental hosts, including humans, may develop clinical symptoms ranging from mild flu-like illness to fulminant disease (e.g., Weil's disease and pulmonary hemorrhage syndrome). Even with treatment, mortality for severe leptospirosis ranges between 10-70% [4]. The factors driving disease severity in humans are poorly understood but are thought to include the bacterial serovar and strain, inoculum size, and the host's innate and adaptive immune responses [6,[16][17][18].
In diverse bacteria, iron serves as an essential co-factor for many cellular processes, including energy generation via electron transport, intermediary metabolism and DNA biogenesis [68][69][70][71][72]. For many pathogens, the shift from a high-to low-iron environment is a key environmental signal for induction of expression of virulence genes [69]. Unlike other spirochetes, such as B. burgdorferi and Treponema pallidum, which require iron in trace amounts, if at all [73][74][75][76][77], Leptospira spp. require this metal for growth in vitro and, presumably, in the host [78].
Consequently, leptospires have evolved elaborate mechanisms for iron sensing, scavenging and utilization [79][80][81]. At the same time, leptospires must balance their physiological need for transition metals with the potential damage caused by highly toxic hydroxyl free radicals generated by Fenton chemistry from H2O2 in the presence of ferrous ions [82][83][84]. Many bacteria, including Leptospira spp., encode systems to ameliorate the toxicity of H2O2 and repair damage due to oxidative stress [85,86]. Expression of gene products involved in oxidative stress responses typically are controlled by one of two master regulators -OxyR and PerR. While OxyR acts as a transcriptional activator for gene products (i.e., catalase and superoxide dismutase) related to detoxification of reactive oxygen species (ROS), PerR acts as a repressor and is released from DNA following exposure to peroxide [48,54,87,88]. Both master regulators respond to similar amounts of H2O2 [89,90]; thus, it is unclear why some bacteria have evolved to use PerR while others use OxyR.
A genome-wide survey of L. interrogans identified four putative FUR family regulators (LIMLP04825, LIMLP05620/perRB, LIMLP10155/perRA and LIMLP18690). Prior studies by Lo et al. [91] and Zavala-Alvarado et al. [92], suggest that PerRA functions as a metal-dependent peroxide stress regulator. Consistent with its repressor function in other bacteria, a L. interrogans perRA transposon mutant expresses increased levels of catalase, AhpC and cytochrome cperoxidase and enhanced survival following exposure to peroxide in vitro [91][92][93]. Recently, Zavala-Alvarado et al. [93] demonstrated that expression of perRB also was increased by H2O2 in vitro. Inactivation of perRB increased survival to superoxide but not H2O2 [93]. These data suggest that PerRA and PerRB likely are functionally distinct. Consistent with this notion, Zavala-Alvarado et al. [93,94] saw little overlap between the PerRA and PerRB regulons by RNA-Seq analysis of in vitro-cultivated organisms. Interestingly, while perRA and perRB single mutants are virulent in hamsters [91,94], a perRA/B double mutant was avirulent [93]. Collectively, these data argue that L. interrogans requires at least one functional PerR-like regulator for infection in mammals.
To investigate the molecular basis for the phenotypic differences between PerRA and PerRB single and double mutants and identify putative virulence-related genes dysregulated by the loss of either/both regulators in vivo, we performed comparative RNA-Seq on all three mutant strains cultivated within DMCs. Similar to RNA-Seq data for in vitro-cultivated organisms [93,94], we saw very little overlap between the PerRA and PerRB regulons within mammals.
Interestingly, the PerRB DMC regulon was substantially larger than its in vitro counterpart [93].
Importantly, by RNA-Seq, we identified 90 genes that are dysregulated only in the double mutant cultivated in DMCs. Of particular note, the "double-only" regulon includes at least four virulenceassociated genes, ligA and ligB, encoding Leptospiral Immunoglobulin-like proteins LigA and LigB, and lvrAB, encoding tandem sensory histidine kinases (HKs). Decreased expression of LigA and LigB in the perRA/B double mutant was not due to loss of LvrAB signaling. The perRA/B double mutant DMC regulon also contains 15 additional genes related to environmental sensing and/or gene regulation, including nine putative hybrid HKs and six putative DNA binding proteins; all but two of the 15 were dysregulated only in vivo. Taken together, our data suggest that PerRA and PerRB are part of a complex signaling network that uses mammalian host-specific signals to coordinate the expression of genes required by L. interrogans for adaptation to reservoir and incidental (i.e., human) hosts.

Results
Pathogenic and saprophytic Leptospira spp. encode different FUR-like metalloregulator repertoires. To gain insight into the functions of the four FUR-like regulators encoded by L.
interrogans, we performed a phylogenetic comparison of these proteins against well characterized representative FUR family metalloregulators from Gram-negative and -positive bacteria (Fig 1A).
PerRA and PerRB clustered most closely with PerRs and iron-response regulators (Irrs), while LIMLP18590 and LIMLP04825 clustered with Zur/Nur and Fur/Mur regulators, respectively. We next surveyed the amino acid sequences of the leptospiral FUR-like proteins for conserved regulatory and structural metal binding sites (MBS), which promote DNA binding and folding/dimerization, respectively, in other Fur family regulators [49,50,95]. As noted recently by Zavala-Alvarado et al. [93], PerRA and PerRB contain two PerR canonical amino acid residues (Asn60 and Asn68 in PerRA and PerRB, respectively) involved in peroxide sensitivity and DNA recognition (Asp103 and Asp112 in PerRA and PerRB, respectively) [96,97]. Based on these features and increased expression of perRB upon exposure of L. interrogans to peroxide, LIMLP05620 was named perRB [93]. Interestingly, as shown in Fig 1B, the aspartate of the PerR regulatory MBSs and the asparagine in the PerR DNA-binding helices (DBH) also are conserved in Irr proteins. As noted previously by Kebouchi et al. [92] and Zavala-Alvarado et al. [93,94], both PerRA and PerRB lack the C-terminal conserved CxxC motif(s) used for structural metaldependent dimerization by many, but not all, FUR family regulators; this cysteinate motif also is absent in Irrs. Overall, the PerRA and PerRB DBHs are not highly conserved, raising the possibility that they recognize different upstream sequences. LIMLP04825, on the other hand, contains features conserved across Fur, Mur, Zur and Nur regulators, including a glutamic acid at position 103, one or possibly two CxxC motifs (residues 95-98 and 133-136), and an arginine (Arg60) within its DBH ( Fig 1B). Interestingly, LIMLP18590 contains features of both PerR (Asp at position 103) and Fur/Mur/Zur/Nur (Arg residue within its putative DBH). The regulatory metal binding site(s) for LIMLP18590 most closely resembles that of a Zur (Fig 1B), which includes two putative tetra-coordinated zinc binding sites rather than the single penta-coordinated site used by PerR. However, without additional data regarding the peroxide responsiveness and/or regulatory metal-binding properties of LIMLP04825 or LIMLP18590, it is not possible to discern their function(s). For this reason, we propose designating them as Fur family regulators 1 (Ffr1) and 2 (Ffr2), respectively. We next assessed the conservation of FUR family regulators across pathogenic (P1 and P2) and saprophytic (S1 and S2) leptospiral subclades [98]. Orthologs for PerRA and Ffr1 were identified in all highly pathogenic (P1), some intermediate (P2) and all saprophytic strains (S1 and S2), whereas orthologs for PerRB and Ffr2 were found exclusively in pathogenic strains (Figs 2A-B and S1). Our analyses also identified two additional FUR family regulators, both of which were found only in saprophytic leptospires (Figs 2A-B and S1). The first, designated PerRC, contains features of a canonical PerR (two CxxC motifs, an aspartic acid residue within its regulatory MBS and an asparagine within its putative DBH). The second saprophyte-specific FUR family regulator, designated Ffr3, resembles a Fur/Mur/Nur-like regulator (two CxxC motifs, a glutamic acid residue within its regulatory MBS and an arginine within its putative DBH).
L. interrogans FUR family regulators are expressed at higher or comparable levels in DMCs compared to in vitro. Previously, we reported that expression of perRA in L. interrogans sv.

Inactivation of both perRA and perRB in L. interrogans results in loss of virulence in mice.
Previously, Murray et al. [99] and Zavala-Alvarado et al. [94] independently reported that a L.
interrogans Manilae perRA Tn mutant is virulent in hamsters. More recently, Zavala-Alvarado et al. [93] established that L. interrogans lacking PerRB also retain virulence in hamsters. Zavala-Alvarado and colleagues also generated a double mutant by insertional inactivation of perRA in the perRB Tn mutant; the resulting double mutant (perRA/B) was avirulent in hamsters [93].
Golden Syrian hamsters are exquisitely sensitive to L. interrogans and develop acute, fulminant, disseminated disease at doses as low as 10 1 [17,99,101,102]. Mice, on the other hand, are a natural reservoir for L. interrogans and relatively resistant to infection; at sublethal doses, susceptible mouse strains develop a self-resolving hematogenous dissemination phase (~1 week) followed by chronic, asymptomatic renal colonization marked by shedding large numbers of leptospires in urine [17,[103][104][105][106].
Given the differences in leptospiral disease progression and severity between hamsters and mice, we asked whether PerRA, PerRB, or both are required to establish infection and persistence within a reservoir host model. At the outset, we first established that our wild-type serovar Manilae parent (WT) is virulent in C3H/HeJ mice. Female 10-week old mice (n=5 per group) were infected intraperitoneally with 5 ´ 10 6 , 1 ´ 10 6 , 1 ´ 10 5 and 1 ´ 10 4 leptospires. Mice were monitored daily for signs of disease (i.e., weight loss). Within 6 days, all mice in the 5 x 10 6 group and 3 of 5 mice in the 10 6 group succumbed to infection, while all others survived the entire 42-day experimental time course ( Fig 4A). Based on these virulence studies, the LD50 for the wild-type (WT) parent was @ 7 x 10 5 . Beginning 14 days post-infection (p.i.), surviving mice were monitored weekly for the presence of leptospires in their urine by darkfield microscopy. Urine from all but one (10 4 group) mouse contained large numbers of leptospires at all three time points (14,21, and 35 days p.i.) (Fig 4B). At 42 days p.i., kidneys harvested from all surviving mice infected with the WT parent, including the single urine-negative mouse from the 10 4 group, were culture-positive.
Prior to using the perRA and perRB single and double mutants for murine virulence studies, we first confirmed their genotypes by amplicon sequencing using primers listed in S6 Table and immunoblot and established that loss of one regulator had no obvious effect on expression of the other in the corresponding single mutants (S2 Fig). We next compared infectivity of the WT, perRA, perRB or perRA/B strains in C3H/HeJ mice using a sublethal intraperitoneal inoculum (1 x 10 5 ). All of the mice inoculated with the WT parent and single mutants were infected, shedding comparable numbers of leptospires in their urine at 14-and 21-days p.i. (Fig 4C). In contrast, no leptospires were detected in urine from mice inoculated with the perRA/B double mutant.
Consistent with data from urine, at day 28 p.i., all perRA/B-infected mice were negative for leptospires by both culture and qPCR ( Fig 4D). Lastly, in contrast to mice infected with the WT or single mutant strains, all of which generated robust serological responses against L. interrogans, none of the mice infected with the double mutant seroconverted (Fig 4E).
perRA and perRB single and double mutants grow normally in rat peritoneal cavities. PerR regulators have been linked to a wide range of physiological functions outside of oxidative stress, including metal homeostasis, metabolism and virulence [107][108][109]. In Bacillus subtilis, inactivation of perR leads to increased expression of fur and iron starvation [82]. To examine whether the avirulent phenotype of the perRA/B double mutant could be due to an inability to grow in mammals, we took advantage of our DMC model, whereby leptospires are cultivated for 9-10 days within dialysis membrane chambers implanted in the peritoneal cavity of a rat [41,110].
Originally developed for B. burgdorferi, this model is able to separate genes related to physiological adaption (i.e., nutrient acquisition and metabolism) from those encoding virulence determinants, such as adhesins, motility and immune evasion. However, we saw no significant difference (p>0.05) in the mean number of leptospires for the wild-type (1.75 ´ 10 8 /ml) strain versus each mutant (perRA, 8.5 ´ 10 7 /ml; perRB, 3.57 ´ 10 8 /ml; and perRA/B, 2.70 ´ 10 8 /ml) recovered from DMCs 9 days post-implantation (3 biological replicates per strain). These data demonstrate that the virulence-defect observed with the double mutant is not due to a metabolic lesion (i.e., metal starvation).
Defining the PerRA and PerRB regulons in vivo by comparative RNA-Seq. Prototypical FUR family regulators, including PerR, modulate transcription by binding to DNA via one or more ~19bp inverted repeats ('boxes') located upstream of their target genes [111]. Kebouchi et al. [92] previously identified three potential PerR binding sites upstream of perRA in L. interrogans.
However, searches of the Manilae genome using these sequences, as well as canonical Fur and PerR boxes [111,112], did not identify additional hits [41,44,91]. Therefore, to identify genes controlled by PerRA, PerRB, or both, in response to host signals, we performed comparative RNA-Seq using WT, perRA, perRB and perRA/B strains cultivated in DMCs (3 biological replicates per strain); a summary of the raw Illumina read data is presented in S1 Table. Reads were mapped using EDGE-pro [113] and analyzed for differentially-expressed genes using DESeq2 [114].
Genes expressed at ³3-fold higher/lower levels in the WT versus mutant with a False-discovery rate (FDR)-adjusted-p value (q) £0.05 were considered differentially expressed. Complete RNA-Seq datasets for all comparisons are presented in S2-S4 Tables. Raw read files have been deposited in the NCBI Sequence Read Archive (SRA) database (BioProject accession PRJNA659512).
Overview of the PerRA DMC regulon. The PerRA DMC regulon contained a total of 81 differentially expressed genes; 43 were expressed at higher levels (i.e., upregulated directly or indirectly by PerRA) in the WT parent compared to the perRA mutant, while 38 were expressed at lower levels (i.e., downregulated/repressed directly or indirectly by PerRA) (S2 Table). Notably, the PerRA DMC regulon is substantially larger than its in vitro counterpart (17 genes total but only 14 dysregulated >3-fold), recently reported by Zavala-Alvarado et al. [94]. TonB-dependent transporters (TBDT) for iron typically are repressed by Fur [115]; thus, it was surprising that this system was upregulated by PerRA both in vitro and in DMCs. Interestingly, none of the prototypical oxidative stress-related genes identified by Zavala-Alvarado et al. [94] as being under PerRA control in vitro were dysregulated in DMCs.  Table). The three upregulated genes (LIMLP02515, LIMLP05780 and LIMLP01845) encode putative DNA binding proteins, including a CsoR-like metal sensitive repressor, while the three downregulated genes encode a two-component system (TCS) histidine kinase with four Per-Arnt-Sim (PAS)-type sensor domains (LIMLP10140), a putative DNA binding protein (LIMLP00900) and a putative serine/threonine kinase with GAF domain (LIMLP11575). PAS domains are ubiquitous in bacteria and sense a wide range of ligands, including heme, FAD, fatty acids and divalent metals [116,117]. GAF domains share a similar fold to PAS domains and often regulate the catalytic activity of cyclic nucleotide phosphodiesterases [118]. Of note, none of these putative regulatory factors were dysregulated >3 fold by loss of PerRA in vitro [94].  Table). In contrast, only 30 genes were dysregulated in the perRB mutant in vitro, with only one affected >3-fold [93].
Remarkably, we saw no overlap between the in vitro and DMC PerRB regulons. Overlap between the PerRA and PerRB DMC regulons was limited to genes within the TonB-dependent transporter  Table). Notably, however, the PerRB DMC regulon includes 17 genes (11 upregulated, 6 downregulated) related to signaling and/or gene regulation ( Fig 5B). The 11 upregulated genes include six related to signal transduction, three putative DNA binding proteins (LIMLP16420, LIMLP07225 and LIMLP15900), an ECF-type sigma factor (LIMLP14515) and a putative serine/threonine phosphatase with a PAS-type sensor domain (LIMLP06960) (Fig 5B). The six downregulated signaling genes include two additional putative DNA binding proteins (LIMLP07895 and LIML10055), a second ECF-type sigma factor (LIMLP14940), a putative anti-sigma factor antagonist (LIMLP04815), and an EAL-type phosphodiesterase (LIMLP04775) (Fig 5B). None of these putative regulators were affected in vitro by loss of PerRB [93].  Table 1).

Inactivation of both PerRA and PerRB results in a DMC regulon that differs
LIMLP14200 and LIMLP12135 contain domains found in lipases/esterases (IPR0002489) [119] and alkaline phosphatases and sulfatases (IPR000917), respectively. LIMLP18070 contains an ATP-NAD kinase domain (IPR022504), suggesting a role in maintaining NADP homeostasis and, by extension, NADPH-dependent reductive biosynthetic pathways. LIMLP02545, encoding one of the few gene products in the PerRA/B regulon related to oxidative stress, is a putative methionine sulfoxide reductase, which catalyzes the reversible thioredoxin-dependent oxidation-reduction (repair) of Met-SO to Met [120,121]. Lastly, LIMLP15435 contains a domain found in glyoxalase/bleomycin resistance proteins; in bacteria, glyoxalases are used to detoxify methylglyoxal, a reduced derivative of pyruvate, as part of the glutathione-dependent glyoxalase system [122].  (Table 1). Nine, including lvrAB (discussed below), encode sensory histidine kinases, most of which contain PAS-type sensor domains ( Fig 5C). One of the nine (LIMLP05830) encodes a regulator that contains both PAS and GAF domains ( Fig 5C). The PerRA/B DMC regulon includes six putative DNA binding proteins, four upregulated and two downregulated ( Fig 5C).
Three upregulated genes belong to the TetR (LIMLP02080), Cro/C1-l (LIMLP04315) and CRPlike (LIMLP07225) repressor families, while the fourth (LIMLP00755) encodes a hypothetical protein containing a homeobox winged helix-like domain of unknown function (DUF433). A second TetR-like repressor (LIMLP19135) and a BolA-like regulator (LIMLP08975) were repressed by PerRA/PerRB in DMCs ( Fig 5C). In E. coli, BolA has been linked to a range of adaptive responses, including biofilm formation and entry into stationary phase [123]. All but two of the regulatory proteins in the PerRA/B DMC regulon were dysregulated only in the double mutant; LIMLP06340, encoding a histidine kinase, and LIMLP07225, encoding a CRP-like DNA binding protein, also were upregulated by PerRB alone in DMCs ( Fig 5B).
Although the vast majority (70%) of genes upregulated by PerRA/B encode proteins of unknown function (Fig 6B and Table 1), seven contain conserved domains potentially related to mammalian host adaptation and/or virulence. LIMLP08585 contains a PPM-type phosphatase domain (IPR001932); PPM domains are found in diverse regulatory proteins, including SpoIIE in B. subtilis [124]. LIMLP15425 contains a putative Lambda_Bor-like domain (PF06291), which in E. coli has been associated with increased serum survival [125,126]. LIMLP14585, annotated as a host attachment protein, contains a domain of unknown function (IPR019291) found in virulence-associated proteins from the plant pathogens Agrobacterium tumefaciens and Xanthomonas spp. [127,128]. LIMLP02040 contains a SRPBCC-like domain (cd07812), which forms a deep, hydrophobic ligand binding pocket capable of binding diverse ligands [129,130].
Three hypothetical proteins (LIMLP04635, LIMLP10965 and LIMLP16555) upregulated by PerRA/B are predicted to form b-propeller structures, which are associated with a wide range of functions, including ligand-binding, enzymatic activity, cell signaling, and protein-protein interactions [131]. Interestingly, Thibeaux et al. [132] previously noted that proteins with bpropeller repeats are enriched in highly virulent Leptospira spp. Six upregulated genes encode uncharacterized lipoproteins of unknown function (Table 1).
Eighteen (56%) genes downregulated by PerRA/B in DMCs encode proteins of unknown function ( Table 2). LIMP04970 and LIMLP11660, both predicted to encode lipoproteins, contain domains (pectin lyase-fold/IPR011050 and Ricin B lectin/IPR000772, respectively) potentially involved in binding to and/or cleavage of host-derived carbohydrates. LIMLP04765 contains an alpha/beta hydrolase domain shared by a wide range of hydrolytic enzymes. Lastly, LIMLP01455, encoding an inner membrane protein, contains a DoxX-like domain; in Mycobacterium tuberculosis, DoxX complexes with a thiosulfate sulfurtransferase (SseA) to promote resistance to agents that disrupt thiol homeostasis [133].
Known or putative virulence determinants. The upregulated portion of the PerRA/B regulon contains at least four virulence-associated genes (Table 1). Two, LIMLP15405/ligA and LIMLP15415/ligB, encode the pathogen-specific, multifunctional, Leptospiral Immunoglobulinlike repeat proteins LigA and LigB, respectively [134,135], while LIMLP08490 and LIMLP08485 encode the hybrid histidine kinases LvrA and LvrB, respectively [19]. Although tandemly located on the chromosome, ligA and ligB are not co-transcribed ( Fig 7A). They do, however, have identical upstream regions and respond similarly in vitro to conditions used to mimic the mammalian host milieu (e.g., high osmolality and increased temperature) [21][22][23][24]136]. Three genes located downstream of ligB, all encoding hypothetical proteins, also were upregulated ( Fig   7A). Using antisera against the shared N-terminal repeats (Fig 7A), we compared expression of Given that the upstream regions for ligA and ligB are identical, the molecular basis (e.g., transcriptional, post-transcriptional or both) for the difference between Lig levels in the perA and perRB mutants is unclear.
The downregulated portion of the PerRA/B DMC regulon contains at least one gene potentially related to virulence. LIMLP03665/colA, encoding a collagenase precursor [137], was expressed at ~500-fold lower levels in the WT parent compared to the perRA/B mutant (Table 2).
While collagenase-mediated degradation of host tissues likely enhances dissemination of leptospires during early infection [138], once in the kidneys, repression of colA could help reduce pathogen-mediated damage to renal epithelial cells. Further transcriptional analysis of this gene is needed to establish its expression profile in different tissues over the course of infection.

Loss of LvrAB alone is not responsible for avirulence of the perRA/B double mutant in mice.
As noted above, expression of lvrAB is disrupted only in the perRA/B double mutant (Table 1); similar results were obtained using leptospires grown in vitro [93]. Using LvrA-and LvrB-specific antisera, we confirmed our transcriptomic data at the protein level by immunoblot using whole cell lysates from WT, perRA, perRB and perRA/B mutant strains cultivated in DMCs (Figs 8A and S5A and S5C-D). Previously, Adhikarla et al. [19] reported that inactivation of lvrAB by

Discussion
L. interrogans must sense and respond to diverse signals and threats during the free-living and reservoir host phases of its zoonotic cycle. Not surprisingly, L. interrogans encodes substantially more sensory and regulatory proteins than B. burgdorferi and T. pallidum [139], two pathogenic spirochetes with far more restrictive growth niches. However, the regulatory networks and gene products that sustain L. interrogans in nature remain poorly understood. To gain insight into the transcriptomic changes that leptospires undergo within the host, we previously compared L. interrogans sv. Copenhageni strain Fiocruz L1-130 cultivated in vitro and in mammals using our DMC peritoneal implant model [41,110]. From these studies emerged >100 genes that were differentially expressed in response to host-specific signals, including LIC12034, encoding the peroxide stress response regulator PerRA, which was upregulated 3.83-fold in DMCs. Herein, we confirmed these data using L. interrogans sv. Manilae strain L495 and also established that the three remaining FUR family regulators are transcribed at comparable (perRB) or higher (ffr1 and ffr2) levels in DMCs compared to in vitro. The importance of FUR family regulators for host adaptation was confirmed recently by Zavala-Alvarado et al. [93], who demonstrated that leptospires lacking both PerRA and PerRB are unable to infect hamsters. In our current study, we establish that these regulators also are required for renal colonization of C3H/HeJ mice. In both animal models, loss of virulence was observed only when both PerRA and PerRB were inactivated, suggesting that these regulators may serve redundant or overlapping functions in vivo. Our finding that the perRA/B double mutant survives at wild-type levels in DMCs is particularly noteworthy as it demonstrates that the avirulent phenotype observed for this mutant is not due to a metabolic lesion (i.e., metal starvation) but instead reflects dysregulation of one or more virulence- Despite several attempts, Zavala-Alvarado et al. [93] was unable to restore virulence to the perRA/B double mutant by trans-complementation with perRA or perRB alone. Consequently. we cannot rule out the possibility that the loss of virulence observed with the perRA/B double mutant is due to a spontaneous genetic defect outside of perRA or perRB. However, by comparative genomic sequencing of WT, perRA/B and perRB (the parental background for the perRA/B double mutant) strains, Zavala-Alvarado et al. [93] identified only two differences in the perRA/B strain.
The first was a single nucleotide insertion in LIMLP11570, encoding a putative 3-oxoacyl ACP synthase related fatty acid synthesis. It is important to note that this same insertion also is observed in several L. interrogans isolates from human and animals and, as noted earlier, we saw no difference in the growth of the single and double mutants either in vitro or in DMCs. The second nonsynonymous difference is in LIMLP01895, encoding a putative hybrid histidine kinase; the corresponding polymorphism results in an alanine to valine substitution at amino acid 146; the position of this mutation is within an inter-domain region and, therefore, not likely to affect the protein's putative signal transduction function(s). Nonetheless, further investigation is necessary to establish the extent to which LIMLP01895 contributes to gene regulation and/or loss of virulence in the perRA/B double mutant.
The presence of multiple FUR family regulators in Leptospira spp. was noted previously by Louvel et al. [44], who identified five distinct orthologs between L. interrogans and L. biflexa.
Phylogenetic analyses presented herein identified a sixth FUR family regulator and established that two (PerRA and Ffr1) are conserved within both pathogenic and saprophytic species and two each are unique to either pathogenic (PerRB and Ffr2) or saprophytic subclades (PerRC and Ffr3).
As their designations suggest, based on sequence alignments, three are predicted to function as PerRs. Thus far, only B. licheniformis has been shown to encode multiple PerRs (PerRBL, PerR2 and PerR3), each of which displays a different level of sensitivity to H2O2 (PerR2 > PerRBL > PerR3) [140]; the extent of regulatory overlap between these three PerRs has yet to be determined.
The peroxide responsiveness and/or metal sensing properties of the remaining three leptospiral FUR family regulators cannot be predicted based on sequence alone. Our finding that almost all saprophytic and pathogenic Leptospira spp. encode closely-related PerRA and Ffr1 orthologs, however, implies that these two regulators could function outside of a host (e.g., within soil and/or water). The presence of a single PerR (PerRB) in all but one of the four P2 subclade species examined (L. wolfii) may contribute to the 'intermediate' virulence of these Leptospira spp. compared to highly virulent P1 subclade [6,141].
L. interrogans cultivated in DMCs express increased levels of catalase, AhpC-type peroxiredoxin and cytochrome c peroxidase [41], three enzymes typically associated with detoxification of reactive oxygen species (ROS) in bacteria [85,87,142]. These data also provide strong evidence that L. interrogans is exposed to ROS in vivo [41]. Consistent with this notion, catalase-deficient leptospires are more susceptible to H2O2 in vitro and show reduced virulence in hamsters [143]. Host phagocytic cells, which generate oxygen radicals via a dedicated NADPH oxidase [144,145], are one likely source of exogenously-derived ROS in vivo. Leptospires within renal tubules, a highly oxygenated niche, also would be exposed to elevated levels of oxygen.
Incomplete reduction of oxygen by iron-containing cytochromes is another potential source of endogenous ROS [85,142]. In bacteria, oxidative stress responses often are coordinated by two evolutionarily distinct master regulators --OxyR and PerR. OxyR, the more common of the two, belongs to the LysR family and functions primarily as an activator [146]. In its oxidized state, OxyR activates transcription of genes involved in the detoxification of H2O2 (catalase and AhpC), the prevention or repair of DNA damage (Dps) and/or redox homeostasis (glutathione reductase, thioredoxin) [85]. PerR, first described in B. subtilis [147], typically represses rather than activates many of the same genes as OxyR and is released from DNA by peroxidation [49,54,90]. Although OxyR and PerR regulate transcription by different mechanisms, they react with H2O2 at essentially the same rate constant (10 5 M -1 s -1 ) [89] and orchestrate highly similar responses. L. interrogans does not encode an OxyR homolog but, as noted above, encodes at least two PerR orthologs, PerRA and PerRB. Consistent with PerR functions in other bacteria, as shown here and elsewhere [91,93,94], L. interrogans perRA mutants show enhanced survival following exposure to lethal levels of H2O2 in vitro and increased expression levels of catalase, AhpC and cytochrome c peroxidase in vitro. While inactivation of perRB had no effect on the ability of leptospires to withstand killing by H2O2, the perRB mutant showed increased tolerance to the superoxidegenerating compound paraquat [93]. Moreover, no genes associated with ROS defenses were dysregulated in the perRB mutant in vitro [93]. In DMCs, only cytochrome c peroxidase, AhpC and a glutaredoxin were dysregulated in the double mutant. Interestingly, all three genes were expressed at higher levels in the WT compared to the mutant, suggesting that they are activated rather than repressed by PerRB. Moreover, expression of catalase was not significantly different in the WT vs. perRB or perRA/B DMC comparison.
The above data argue that while PerRA and PerRB may be 'activated' by ROS, the adaptive responses they control likely extend beyond oxidative stress. The prototypical PerR in B. subtilis (PerRBs) can coordinate either Mn 2+ or Fe 2+ .When co-factored with Fe 2+ , DNA binding by PerR:Fe is highly sensitive to H2O2 due to irreversible iron-dependent oxidation of metal-coordinating histidine residues [54,96]. When cofactored with Mn 2+ , however, PerRBs is able to bind DNA but is no longer peroxide sensitive [96]. In this way, PerR functions both as a peroxide responsive regulator and a ratiometric sensor for iron and manganese, altering its transcriptomic output based on intracellular metal availability and/or oxidative stress. Our finding that the PerRA/B DMC regulon contained only three genes related to oxidative stress and no genes related to iron homeostasis raises the possibility that PerRA and PerRB function in vivo may be regulated by metal availability rather than oxidative stress. Moreover, it is possible that both PerR:Fe and PerR:Mn regulate different cohorts of virulence genes, depending on the host milieu. Consistent with differences in the putative DNA-binding helices, we saw very little overlap between the PerRA and PerRB DMC regulons. Seven of the eight genes common to both regulons are located in a single locus encoding a TonB-dependent transporter (TBDT) system. In Gramnegative bacteria, TBDT systems promote the uptake of substrates, such as iron siderophores, heme, vitamin B12, and carbohydrates, that are either poorly transported by non-specific outer membrane porins or are present in the extracellular milieu at low concentration [115,148].
The energy required for substrate transport is provided in the form of proton motive force, which is transduced from the inner to outer membrane by the TonB-ExbB-ExbD complex [115]. L.
interrogans encodes 11 putative TonB-dependent receptors and at least two complete TonB-ExbB- Together, these data suggest that mammalian host signals play a key role in modulating TonBdependent nutrient uptake in L. interrogans and, moreover, that the activity of PerRB is enhanced in vivo.
By comparative RNA-Seq, we identified four distinct PerR regulatory categories in L. interrogans (Fig 9). The first two include genes whose expression is controlled exclusively by a single PerR (PerRA only and PerRB only ). The most straightforward explanation for this category is that PerRA and PerRB recognize different upstream boxes. Given that the perRA and perRB single mutants are fully virulent in hamsters [93] and mice, genes in these two categories either are not  [157]. In Helicobacter pylori and Salmonella enterica sv. Typhimurium, Fur activates expression by binding upstream of target gene and helping to recruit RNA polymerase [158,159]. In a-proteobacteria, Irrs (see below) act as positive and negative transcriptional regulators of genes related to heme homeostasis [59][60][61][62]. BosR, a FUR family regulator in the Lyme disease spirochete B. burgdorferi, activates transcription of the alternative sigma factor rpoS as part of a complex that includes the alternative sigma factor RpoN and the response regulator Rrp2 [160][161][162][163][164][165]. PerRA and/or PerRB also could activate transcription of target genes indirectly via repression of a regulatory small RNA (e.g., RyhB in E. coli) [166] or by preventing the binding of another repressor (i.e., anti-repression) [63,64].
Designation of PerRA and PerRB as peroxide stress regulators in L. interrogans is based largely on in vitro studies showing increased survival of perRA and perRB mutants following exposure to H2O2 and paraquat, respectively [91,93,94]. Several lines of evidence, however, raise the possibility that these gene products function as iron response regulators (Irrs) rather than PerRs.
Based on amino acid sequence alignments, PerRA and PerRB appear to be more closely related to Irrs than PerRs. In a-proteobacteria, Irrs and their regulatory partner, RirA, coordinate the expression of genes involved in heme biosynthesis with iron availability. Similar to FURs, RirA functions as metal-dependent transcriptional repressor but senses iron within Fe-S clusters rather than Fe 2+ . Interestingly, LIMLP06290 (LIC11283), annotated as a hypothetical protein, contains domains consistent with it being a RirA; the contribution of this putative RirA to iron homeostasis in Leptospira spp. has not been examined. At the sequence level, Irrs share a number of features with PerRs, including the presence of Asp and Arg residues in their regulatory metal sites and DNA binding helices, respectively. Irrs and PerRs also are responsive to similar levels of ROS, albeit by a different mechanism, and regulate many of the same effector genes (i.e., catalases and peroxidases) [62,167,168]. In some, but not all cases, irr mutants also show increased survival in vitro under high H2O2 levels [62,169]. As noted above, only B. lichenformis is known to encode multiple PerRs. Numerous bacteria, on the other hand, encode two or more Irrs [167]. Variable affinity of Irrs for their target promoters enables them to modulate gene expression over a wider range of conditions than PerRs [170]. The autoregulatory sequences identified upstream of perRA [92] diverge significantly from canonical PerR and Fur boxes but show strong similarity to "Irrboxes" [171,172]. Moreover, Irrs are known to act as activators as well as repressors [167,171].
Although typically associated with peroxide-sensitive regulation of iron/heme acquisition and utilization, Irrs have been shown to control diverse cellular processes, including virulence.
Moreover, the vast majority of histidine kinases upregulated by PerRA/B in DMCs contain one or more PAS-type sensor domains, which have been shown to function as heme sensors [173]. Given the established importance of heme for survival of L. interrogans in mammals [174][175][176], our findings raise the possibility that heme sensing by PerRA and/or PerRB in mammals could serve as an important initiating event for host adaptation.
Studies presented here and elsewhere [93] demonstrate for the first time that both PerRA and PerRB are required for full transcription of the virulence-related genes ligA and ligB. These pathogen-specific surface lipoproteins have been studied extensively for their contributions to host-pathogen interactions [177,178], virulence [179] and potential use as vaccinogens [180][181][182].
Using a TALE-based transcriptional knockdown approach, Pappas and Picardeau [179] reported that both Ligs are required for virulence in hamsters. As noted earlier, ligA and ligB are not cotranscribed but instead share virtually identical upstream regions and, consequently, are coregulated by the same environmental signals. Matsunaga, Haake and others previously reported that ligA and ligB are upregulated in response to physiological osmolarity (EMJH supplemented with 120 mM sodium chloride) [24] and increased temperature [183]. However, temperature-dependent regulation is mediated by a cis-acting RNA secondary structure that prevents translation at lower temperature and disruption of this cis element had minimal effect on osmoregulation [183]. Eshghi et al. [27] reported that inactivation of lb139 (LIMLP18410), encoding a putative anti-ECF sigma factor, resulted in ~2.5-fold decreased expression of ligB in vitro. However, LIMLP18410 is not in the PerRA/B regulon either in vitro [93] or in DMCs (this study) and,  [184,185] supplemented with 1% rabbit serum at 30°C under static conditions. Mutants were maintained in EMJH under appropriate antibiotic selection (spectinomycin, 40 μg/ml and/or kanamycin, 40 μg/ml). Cultures were harvested at late logarithmic phase (1-5 × 10 8 per ml).
Culture viability (i.e., motility and cell morphology) was evaluated by darkfield microscopy.
Leptospires were enumerated using a Petroff-Hausser counting chamber (Hausser Scientific Co., Horsham, PA). Escherichia coli strains were maintained in Lysogeny broth (LB) or LB agar supplemented with the appropriate antibiotics (ampicillin, 100 μg/ml; spectinomycin, 100 μg/ml; and/or kanamycin, 100 μg/ml). The genotypes of L. interrogans mutants used in these studies were confirmed by PCR and amplicon sequencing using primers listed in S6 Table. Routine DNA manipulation and cloning. Routine cloning was performed using In-

Generation of host-adapted leptospires.
To obtain mammalian host-adapted organisms, L.
interrogans sv. Manilae strain L495 wild-type and mutant strains were cultivated in DMCs as previously described [41,42,110]. Briefly, DMCs were prepared with 9-10 mls of EMJH medium (supplemented with an additional 10% bovine serum albumin to maintain osmotic pressure) at a starting inoculum of 10 4 organisms per ml. Using strict aseptic technique, DMCs were implanted into the peritoneal cavity of an anesthetized female Sprague-Dawley rat. After nine days, animals were euthanized by CO2 narcosis and DMCs harvested. The viability and density of leptospires were evaluated by dark field microscopy using a Petroff-Hausser counting chamber (Hausser Scientific Co., Horsham, PA).
Normalized copy numbers were compared using an unpaired t test with two-tailed p values and 95% confidence interval (Prism v. 6, GraphPad Software).

Quantitation of burdens by qPCR. DNA was extracted from infected kidneys using the Qiagen
DNeasy Blood & Tissue kit according to the manufacturer's recommendations. DNAs were analyzed by quantitative PCR (qPCR) using a TaqMan-based assay for lipL32 [180] in 25 µl reactions performed with SsoAdvanced Universal Probes Super Mix (Bio-Rad). Copy numbers for lipL32 were determined using an internal standard curve for the lipL32 amplicon cloned into pCR2.1 TOPO (Invitrogen). Average values for each strain were compared using an unpaired t test with two-tailed p values and 95% confidence interval (Prism v. 6, GraphPad Software).

RNA sequencing and comparative transcriptomics.
Total RNA was prepared from leptospires cultivated in DMCs using TRIzol Reagent (ThermoFisher) (3 biological replicates per strain) and then treated twice with TURBO DNase (ThermoFisher) followed by purification using RNeasy columns (Qiagen) as previously described [41]. Samples were eluted in RNAse-free water and purified RNA was analyzed using Qubit RNA HS Assay Kit (Thermo) and Agilent TapeStation 4200 (Agilent Technologies, Santa Clara, CA, USA) using the RNA High Sensitivity assay. Only samples with Ribosomal Integrity number (RINe) values >7.5 were used for library preparation.
Stranded libraries were prepared from ribo-depleted RNA using Zymo-Seq RiboFree Total RNA Library Kit according to manufacturer's instructions. Libraries were validated for length and adapter dimer removal using the Agilent TapeStation 4200 D1000 high-sensitivity assay and then quantified and normalized using the double-stranded DNA (dsDNA) high-sensitivity assay for available from https://github.com/najoshi/sickle) and then mapped using EDGE-pro version 1.1.3 [113] using fasta, protein translation table (ptt) and ribosomal/transfer RNA table (rnt) files based on the L. interrogans sv. Manilae strain UP-MMC-NIID LP genome (NZ_CP011931.1, NZ_CP011932.1 and NZ_CP011933.1). Differential expression was determined using DESeq2 [114]. Genes expressed at ≥3-fold higher/lower levels in the mutant compared to the wild-type parent with a False Discovery Rate (FDR)-adjusted p-value (q-value) ≤ 0.05 were considered differentially expressed. Raw read data have been deposited in the NCBI Sequence Read Archive classifications are based on MicroScope, an integrated platform for the annotation of bacterial gene function through genomic, pangenomic and metabolic comparative analysis [188]. Conserved domain searches were performed using Conserved Domain Database (CDD) Search [189], UniProt [190] and InterPro [191]. Candidate lipoproteins were identified based on Setubal et al. [192] and LipoP server [193]. Subcellular localization predictions were performed by BUSCA (Bologna Unified Subcellular Component Annotator) [194]. Multiple sequence alignments were generated by Clustal Omega [195] and MAFFT 7 [196]. Phylogenetic trees were generated using PhyML 3.0 [197] with LG substitution model chosen after an Akaike Information Criterion (AIC) model selection [198]. Tree improvement was done by subtree pruning and regrafting (SPR) method [199] with ten random starting trees. Robustness of branches was assessed by Approximate Likelihood-Ratio Test (aLRT-SH) [200]. The resulting trees were visualized and annotated using Interactive Tree of Life (iTOL, v 4.3) [201]. FUR domain-containing proteins in 26 Leptospira spp. genomes (10, 5, 6 and 5 species from subclade P1, P2, S1 and S2, respectively [98]) were identified using the Leptospira species name as a query in the Ferric-uptake regulator domain entry (IPR002481) in EMBL-EBI InterProScan [202]. Orthologs shared between L. interrogans sv.
Manilae strain L495 and sv. Copenhageni strain Fiocruz L1-130 strains were identified using OrthoVenn 2.0 [203].   3 Localization is based on cumulative data from BUSCA [194], LipoP [193] and SpLip [192]. Abbreviations: CY, cytoplasmic; EC, extracellular; IM, integral inner membrane; OM, outer membrane-embedded; and   but not confirmed by X-ray crystallography to be involved in regulatory or structural metal coordination. Asparagine (N) or arginine (R) residues ( ) in blue, located in DNA binding helix H4, can be used to distinguish between PerR and Fur, respectively [97]. *, the PDB structure for E. coli Fur includes only the DNA binding domain. Numbers on the top correspond to residues positions in L. interrogans PerRA. Phylogenetic analysis of Leptospira spp. FUR family proteins shown in A. Unrooted tree was generated using iTOL [201].  interrogans sv. Manilae strain L495 grown in EMJH at 30°C.