Transcription regulation by CarD in mycobacteria is guided by basal promoter kinetics

Bacterial pathogens like Mycobacterium tuberculosis (Mtb) employ transcription factors to adapt their physiology to the diverse environments within their host. CarD is a conserved bacterial transcription factor that is essential for viability in Mtb. Unlike classical transcription factors that recognize promoters by binding to specific DNA sequence motifs, CarD binds directly to the RNA polymerase to stabilize the open complex intermediate (RPo) during transcription initiation. We previously showed using RNA-sequencing that CarD is capable of both activating and repressing transcription in vivo. However, it is unknown how CarD achieves promoter-specific regulatory outcomes in Mtb despite binding indiscriminate of DNA sequence. We propose a model where CarD’s regulatory outcome depends on the promoter’s basal RPo stability and test this model using in vitro transcription from a panel of promoters with varying levels of RPo stability. We show that CarD directly activates full-length transcript production from the Mtb ribosomal RNA promoter rrnAP3 (AP3) and that the degree of transcription activation by CarD is negatively correlated with RPo stability. Using targeted mutations in the extended −10 and discriminator region of AP3, we show that CarD directly represses transcription from promoters that form relatively stable RPo. DNA supercoiling also influenced RPo stability and affected the direction of CarD regulation, indicating that the outcome of CarD activity can be regulated by factors beyond promoter sequence. Our results provide experimental evidence for how RNA polymerase–binding transcription factors like CarD can exert specific regulatory outcomes based on the kinetic properties of a promoter.

Bacterial pathogens like Mycobacterium tuberculosis (Mtb) employ transcription factors to adapt their physiology to the diverse environments within their host. CarD is a conserved bacterial transcription factor that is essential for viability in Mtb. Unlike classical transcription factors that recognize promoters by binding to specific DNA sequence motifs, CarD binds directly to the RNA polymerase to stabilize the open complex intermediate (RP o ) during transcription initiation. We previously showed using RNA-sequencing that CarD is capable of both activating and repressing transcription in vivo. However, it is unknown how CarD achieves promoter-specific regulatory outcomes in Mtb despite binding indiscriminate of DNA sequence. We propose a model where CarD's regulatory outcome depends on the promoter's basal RP o stability and test this model using in vitro transcription from a panel of promoters with varying levels of RP o stability. We show that CarD directly activates full-length transcript production from the Mtb ribosomal RNA promoter rrnAP3 (AP3) and that the degree of transcription activation by CarD is negatively correlated with RP o stability. Using targeted mutations in the extended −10 and discriminator region of AP3, we show that CarD directly represses transcription from promoters that form relatively stable RP o . DNA supercoiling also influenced RP o stability and affected the direction of CarD regulation, indicating that the outcome of CarD activity can be regulated by factors beyond promoter sequence. Our results provide experimental evidence for how RNA polymerase-binding transcription factors like CarD can exert specific regulatory outcomes based on the kinetic properties of a promoter.
Throughout their life cycle, bacteria must continuously adapt their physiology to respond to and survive in their changing environments. As such, the ability to sense environmental signals and transduce these cues into an appropriate physiological response is important for the virulence of pathogens such as Mycobacterium tuberculosis (Mtb), which face threats from both the host immune system and antibiotic treatment. Regulation of transcription initiation is a major mechanism by which bacteria adapt their gene expression in response to environmental stimuli. Transcription in bacteria is performed by a single RNA polymerase (RNAP) enzyme, which consists of a multisubunit core enzyme that can bind to different sigma factors (σ) to form a holoenzyme and initiate promoter-specific transcription. Mtb devotes a significant fraction of its genome toward encoding numerous transcription factors that can regulate transcription initiation by altering the promoter specificity and recruitment of RNAP (1,2). Classically, transcription factors are recruited to promoters by recognizing and binding a DNA sequence motif, which allows the factor to specifically regulate a subset of the genome. However, bacteria also encode transcription factors that instead localize to promoter regions by binding directly to RNAP (3,4). This class of transcription factors is best exemplified by the stringent response regulators DksA and guanosine (penta)tetraphosphate [(p)ppGpp], which bind to the Escherichia coli RNAP to directly activate or repress transcription from subsets of E. coli promoters (5). These factors exert promoter-specific transcription regulation despite being unable to discriminate promoters at the level of binding. The prevailing hypothesis for the mechanism behind this promoter specificity postulates that these factors can potentiate different outcomes on transcription depending on the underlying initiation kinetics of a promoter (6). Recently, this hypothesis has also been applied to the regulatory mechanisms of other RNAP-binding transcription factors such as CarD (7,8).
CarD is an RNAP-binding transcription regulator that is widely conserved across many eubacteria phyla and essential for viability in mycobacteria (9). CarD associates with transcription initiation complexes by binding directly to the RNAP β subunit through its N-terminal RNAP-interaction domain (RID) (9,10). The CarD C-terminal DNA-binding domain (DBD) also interacts with DNA at the upstream fork of the transcription bubble in a sequence-independent manner (11)(12)(13). Numerous kinetic studies have demonstrated that CarD stabilizes the RNAP-promoter open complex (RP o ) formed by the mycobacterial RNAP during transcription initiation (13)(14)(15)(16)(17). CarD accomplishes this through a twotiered kinetic mechanism in which it binds to RNAPpromoter closed complexes (RP c ) to increase the rate of DNA melting while also slowing the rate of bubble collapse (15). Furthermore, by stabilizing RP o , CarD slows the rate of promoter escape (18), which is a necessary step preceding full-length RNA synthesis. Due to its ability to stabilize RP o in vitro, it was expected that CarD functioned generally as a transcription activator. However, although numerous studies have examined CarD's effect on individual rate constants between transcription initiation intermediates (14,15,18), the composite effect of CarD's kinetic mechanism on full-length RNA production remains unknown. Furthermore, while in vitro studies of CarD have utilized only a handful of promoters, primarily focusing on the Mtb ribosomal RNA promoter rrnAP3 (AP3) (12)(13)(14)(15)(16)(18)(19)(20), chromatin immunoprecipitation sequencing (ChIP-seq) in Mycobacterium smegmatis indicates that CarD colocalizes with the housekeeping sigma factor σ A to promoter regions broadly across the mycobacterial genome (17,21), leaving a gap in our understanding of CarD's activity under different promoter contexts.
To characterize CarD's role in transcription regulation throughout the mycobacterial genome, we previously performed RNA-sequencing (RNA-seq) on a set of Mtb strains expressing mutants of CarD that either impair or enhance its ability to stabilize RP o in vitro (8). We discovered that altering CarD activity in Mtb led to both upregulation and downregulation of numerous protein-encoding transcripts, suggesting that CarD could function as either a transcriptional activator or a transcriptional repressor in different promoter contexts. Prior in vitro studies with Rhodobacter sphaeroides CarD and RNAP have shown that RspCarD activates transcription from promoters lacking a conserved T at the −7 position (22) and represses transcription from its own promoter (23). However, unlike Alphaproteobacteria like R. sphaeroides, which contain a T -7 at fewer than 50% of their promoters, most other bacterial phyla, including Actinobacteria like Mtb, have a T -7 at over 90% of their promoters (22), making it unlikely that the T -7 is a conserved mechanism of CarD promoter specificity. Instead, we previously proposed a model in which the outcome of CarD regulation is dependent on the basal transcription initiation kinetics at a given promoter (7,8). Specifically, at unstable promoters that are rate limited at the step of bubble opening, CarD would facilitate full-length RNA production by stabilizing RP o , while at stable promoters that are rate limited at the step of promoter escape, CarD would make it more difficult for RNAP core enzyme to break contacts with promoter DNA. Herein, we directly test our model using in vitro transcription approaches to explore the relationship between RP o stability and transcription regulation by CarD. We discover that both promoter DNA sequence and DNA topology influence the basal RP o stability of a promoter and the regulatory outcome of CarD on transcription. In addition, we find that in the context of a promoter with high basal RP o stability, CarD can directly repress transcription, marking the first demonstration of direct transcriptional repression by Mtb CarD. This work provides experimental evidence for how RNAP-binding transcription factors like CarD can potentiate multiple regulatory outcomes on transcription through a single kinetic mechanism.

Results
CarD binding correlates with transcriptional regulation but not the direction of regulatory outcome A fundamental feature in our model of CarD mechanism is that the regulatory outcome of CarD on a given mycobacterial promoter is determined based on differences in the basal transcription initiation kinetics of the promoter and not differences in CarD binding. This model is based on comparing ChIP-seq data from M. smegmatis, where CarD is present at almost all RNAP-σ A transcription initiation complexes (17,21), with RNA-seq data from Mtb, where mutation of CarD resulted in both upregulation and downregulation of gene expression (8). However, we cannot yet rule out the alternative hypothesis that CarD's uniform localization pattern in M. smegmatis represents a unidirectional transcription activating mechanism for CarD in M. smegmatis in contrast to the bi-directional regulatory activity in Mtb that is suggested by our RNA-seq data.
To address this gap in our model, we performed an RNAseq experiment in M. smegmatis that could be directly compared to the M. smegmatis ChIP-seq dataset. In our published Mtb RNA-seq experiment (8), we collected RNA from Mtb strains expressing mutant alleles of CarD with either weakened affinity for RNAP (CarD R47E ), predicted weakened affinity for DNA (CarD K125A ), or increased affinity for RNAP (CarD I27F and CarD I27W ). By collecting RNA from Mtb strains with mutations that target different domains of CarD, we were able to dissect how the respective interactions of CarD's functional domains contributed to its role in regulating the Mtb transcriptome. To replicate this experimental design in M. smegmatis, we collected RNA from four strains of M. smegmatis with the native copy of carD deleted and expressing one of four different alleles of Mtb CarD: wildtype (WT) CarD (CarD WT ), CarD R25E (a RID mutant with weakened affinity for RNAP), CarD K125E (a DBD mutant with weakened affinity for DNA), or CarD I27W (a RID mutant with increased affinity to RNAP) as the only carD allele (Table S1). Similar to the CarD mutations used in our Mtb experiment, the CarD mutations that weaken its macromolecular interactions with RNAP or DNA (R25E and K125E) impair CarD's ability to stabilize RP o in vitro (13,15), while the I27W mutation increases its affinity for RNAP and allows CarD to potentiate RP o stabilization at lower concentrations (20). For each strain, we collected RNA from four biological replicates of exponentially growing cells in nutrient replete conditions for sequencing. Two replicates (CarD R25E -1 and CarD K125E -4) were identified as outliers following principal component analysis (PCA) and were discarded from downstream analysis (Fig. S1).
In all three strains with mutations in carD, over 25% of the 6716 coding genes in M. smegmatis Mc 2 155 were significantly differentially expressed (P adj < 0.05) in comparison to the CarD WT strain ( Fig. 1A and Table S1). The number of differentially expressed genes in the CarD R25E (2909 genes) and CarD K125E (2901 genes) M. smegmatis strains is similar to the number of differentially expressed genes in the CarD R47E (2877 genes) and CarD K125A (2690 genes) Mtb strains (8). However, homologous genes between the two species showed little correlation in their transcript expression patterns ( Fig. S2 and Table S2), suggesting that CarD does not simply regulate a subset of homologous genes conserved between Mtb and M. smegmatis. Each of the M. smegmatis CarD mutant strains exhibited a similar number of upregulated genes as downregulated genes (Fig. 1A), following the same pattern as the Mtb CarD mutant strains (8) and suggesting that CarD is capable of potentiating both transcriptional activation and repression in M. smegmatis. Importantly, the strains did not show significant differences in the total amount of RNA per cell (Fig. 1B), suggesting that the transcript abundance differences measured in the CarD mutant strains represent local changes in transcription at specific genes rather than a global decrease in RNA production within the cell that would be expected if CarD functioned strictly as a transcriptional activator.
The transcriptomic relationship between different CarD mutant strains in M. smegmatis was also consistent with the relationships we observed in our Mtb dataset (8). PCA of the RNAseq data illustrated that the M. smegmatis sample replicates clustered tightly with each other based on CarD genotype and samples from CarD mutant strains with impaired RP ostabilizing activity (CarD R25E and CarD K125E ) separated from the strain with enhanced RP o -stabilizing activity (CarD I27W ) along the first principal component (Fig. 1C), demonstrating consistency between replicates from the same genotype and suggesting that altered CarD RP o -stabilizing activity contributes to transcript abundance changes in the mutant bacteria. In addition, the CarD R25E and CarD I27W strains, which encode CarD RID mutants with impaired or enhanced RP o -stabilization in vitro, respectively, displayed largely opposite transcriptomic changes (Fig. S3A), similar to the RID mutants in Mtb (8). In the PCA, the CarD K125E samples separated from all other samples along the second principal component (Fig. 1A) and the direction of transcript abundance changes in the DBD mutant CarD K125E samples correlated poorly with the transcript abundance changes in the RID mutant CarD R25E samples (R 2 = 0.351) (Fig. S3, B and C). This is in contrast to the tight correlation between CarD RID and DBD mutants in Mtb (8) and may suggest that mutations in the DBD and RID have unique effects on CarD's regulatory function in M. smegmatis.
The ChIP-seq dataset shows that CarD is present when RNAP-σ A is also found, supporting a model that CarD is present at the promoters of both upregulated and downregulated genes. However, our data are also compatible with an alternative model in which CarD acts directly as a monotonic transcriptional activator, and genes that appear to be transcriptionally "repressed" by CarD are expressed at lower levels in WT bacteria due to decreased RNAP occupancy at non-CarD-activated promoters. If this alternative model were true, then we would expect to find RNAP-σ A /CarD binding  Initiation kinetics affects CarD activity sites overlapping with transcription start sites (TSSs) ascribed to the transcriptionally "activated" promoters but absent from TSSs ascribed to transcriptionally "repressed" promoters. To examine the overlap between CarD binding sites and CarDregulated transcripts, we used our RNA-seq dataset to identify a list of M. smegmatis genes whose transcript abundance was likely directly responsive to altered CarD-mediated RP o stabilization activity based on having opposite expression patterns in CarD R25E versus CarD I27W . To avoid internal genes within operons, we focused our analysis on 2917 M. smegmatis genes directly downstream of a primary TSS (24) and categorized them into one of four classes (Table S3). TSSs associated with genes that were significantly downregulated (P adj < 0.05) in CarD R25E and significantly upregulated in CarD I27W were classified as 'Activated' by CarD (n = 117), while TSSs associated with genes that were significantly upregulated in CarD R25E and significantly downregulated in CarD I27W were classified as 'Repressed' by CarD (n = 153). TSSs associated with genes that were significantly differentially expressed in both CarD R25E and CarD I27W but in the same direction relative to wildtype were classified as 'Uncategorized' (n = 222) because their expression profile does not reflect the divergent expression pattern expected between CarD mutants with opposing effects on RP o -stabilization in vitro. Lastly, any TSSs that were not significantly differentially expressed in both CarD R25E and CarD I27W were categorized as 'Not Significant' (n = 2425). We reanalyzed our previous ChIP-seq dataset (17,21) and identified 1857 unique CarD binding sites across two biological replicates (Table S3). To avoid broad binding regions that may represent multiple, overlapping CarD binding sites, we focused on 1796 CarD binding sites less than or equal to 1000 base pairs (bp) in width. Of these 1796 CarD binding sites, 1390 sites (77.4%) overlapped with at least one mapped TSS in M. smegmatis and 1129 sites (62.8%) overlapped with a primary TSS associated with a protein-encoding gene (24). We examined the overlap between CarD binding sites and TSSs that were significantly differentially expressed in both Car-D R25E and CarD I27W and found that 57.9% (285/492) of these TSSs were associated with CarD binding (Table 1). Among the differentially expressed genes, 53.0% (62/117) of 'Activated' TSSs and 67.3% (103/153) of 'Repressed' TSSs overlapped with a CarD binding site (Table 1). Thus, CarD binding is associated with transcriptional regulation of M. smegmatis promoters in vivo but is not correlated with the direction of regulation. A similar analysis was performed in the ɑ-proteobacterium Caulobacter crescentus to identify the direct regulon of CdnL (the C. crescentus homolog of CarD) (25). Like our results, CdnL localized to promoter regions of both genes that were upregulated and genes that were downregulated in a ΔcdnL strain, but a vast majority of differentially expressed genes were not associated with CdnL binding, suggesting a broader effect of indirect regulation in C. crescentus. Together, these data support the model that CarD is broadly localized to mycobacterial promoters through its interaction with RNAP but that the regulatory outcome of CarD activity is not determined by occupancy.
CarD directly activates transcription from the Mtb ribosomal RNA promoter rrnAP3 To test our model that the outcome of CarD's RP o stabilizing activity on transcript production depends on the basal promoter kinetics, we used in vitro transcription methods to measure the direct effects of CarD on transcript production. Although several studies have proposed that CarD activates transcription from the Mtb AP3 promoter based on in vitro three-nucleotide transcription assays (16,20) and real-time fluorescence assays (14,15) that report RP o lifetime, full-length transcript production has never been directly measured. To assess full-length RNA production, we performed multiround in vitro transcription assays by incubating recombinantly purified MtbRNAP-σ A holoenzyme with a linear DNA fragment containing the Mtb AP3 promoter (from −39 to +4 with respect to the TSS) driving transcription of a 164 nucleotide RNA product. The addition of a saturating concentration of WT CarD (25:1 M ratio CarD:R-NAP (15)) activated transcription from the AP3 promoter 8fold compared to reactions with no factor added ( Fig. 2A). To investigate how CarD's RP o -stabilizing activity relates to transcriptional activation, we repeated the multiround in vitro transcription assays with CarD mutants impaired in their ability to stabilize RP o in vitro (CarD R25E , CarD R47E , CarD K125A , and CarD K125E ) (13,15). All four of the CarD mutants activated transcription from AP3 compared to reactions with no factor, but the degree of activation by each mutant was reduced compared to WT CarD ( Fig. 2A), suggesting that CarD's RP ostabilizing activity underlies its ability to activate transcription from AP3. In addition, the degree to which each CarD mutant attenuated transcript production correlated with how severe the impact was on the CarD macromolecular interactions with RNAP (10) or DNA (13). In contrast, CarD I27W , which has increased affinity for RNAP and is able to stabilize RP o at lower concentrations than CarD WT (20), activated transcription from AP3 to a greater degree than CarD WT at concentrations below

Initiation kinetics affects CarD activity
where CarD WT is saturating (5:1 M ratio CarD:RNAP) (Fig. 2B), further demonstrating the association between CarD's RP ostabilizing activity and activation of transcript production. Collectively, these results demonstrate that CarD activates fulllength RNA production in vitro from AP3 and this transcription activation is dependent on the RP o -stabilizing activity of CarD.

Additional promoter DNA-RNAP interactions increase basal RP o stability and push CarD toward transcriptional repression
To test the hypothesis that the degree of transcriptional activation by CarD is inversely correlated with the basal RP o stability of a promoter, we explored CarD's direct regulatory effect on transcription from a set of promoters with varying levels of basal RP o stability. Transcription initiation kinetics and RP o lifetime are highly dependent on promoter DNA sequence (26). In the RP o intermediate, the promoter DNA makes multiple sequence-specific contacts with regions of the RNAP holoenzyme to stabilize the transcription bubble (26)(27)(28). The Mtb RNAP-σ A holoenzyme and WT AP3 (AP3 WT ) promoter form a relatively unstable RP o (15,16) that is stabilized by CarD to lead to activation of transcription in vitro (Fig. 2). We, therefore, used the AP3 promoter sequence as a starting point to generate four additional promoter templates (AP3 EcoExt , AP3 MycoExt , AP3 Discr , and AP3 Stable ) with higher levels of basal RP o stability by making targeted sequence mutations that would add or optimize predicted DNA-RNAP interactions in RP o (Fig. 3A). AP3 WT contains near consensus sequence motifs in the −35 and −10 elements (29), which are highly conserved promoter elements that interact with σ region 4 and 2, respectively (30-32), so we did not target these regions in our study. In AP3 EcoExt , we mutated the base at position −14 to a G to introduce a T -15 G -14 motif that represents an extended −10 element that was first identified in E. coli (33). In addition to the classical E. coli-like extended −10 motif, many mycobacterial promoters instead contain a G at position −13 that is associated with promoter strength and RP o formation in DNase I footprinting studies (34). Thus, we also generated AP3 MycoExt , which is mutated to include a G -13 upstream of the −10 element. Both G -14 and G -13 are positioned to interact with a conserved glutamic acid residue in σ A region 3.0 in the mycobacterial RP o (14,34,35). AP3 Discr is mutated to introduce a G -6 GGA -3 motif in the discriminator region immediately downstream of the −10 hexamer that allows for optimal binding with σ A region 1.2 (36)(37)(38). AP3 Stable is mutated to include the mutations made in AP3 EcoExt and AP3 Discr as well as a deletion of a T at position −17 to reduce the length of the spacer region between the −35 and −10 hexamers from 18-bp in AP3 WT to 17-bp. A spacer length of 17 bp allows for optimal interactions of the −35 and −10 hexamers with σ A (39).
To measure the basal RP o stability of RNAP-σ A and the AP3 promoter variants, we performed in vitro three-nucleotide transcription initiation assays (16,20) in the absence of CarD by incubating MtbRNAP-σ A holoenzyme with linear promoter DNA fragments in the presence of a GpU dinucleotide and UTP. In these reactions, the RNAP-σ A holoenzyme can synthesize a three nucleotide 'GUU' RNA transcript but cannot undergo promoter escape, allowing us to assess relative RP o lifetimes by using the amount of three nucleotide product as a proxy. We found that all the promoter variants with additional predicted DNA-RNAP contacts exhibited higher basal levels of RP o stability compared to AP3 WT (Fig. 3B). The most stable variant AP3 Stable displayed 8-fold higher basal RP o stability relative to AP3 WT . To quantify the effect of CarD on RP o stability from these promoter variants, we also performed three-nucleotide transcription assays in the presence of WT  Table S4. The raw gel images directly from the phosphorimager are shown in Fig. S6. AP3, Mtb ribosomal RNA promoter rrnAP3; Mtb, Mycobacterium tuberculosis.
CarD protein (Fig. 3B). On AP3 WT , CarD increased the amount of three nucleotide product by roughly 4-fold over reactions with no factor. As the basal RP o stability of promoter variants increased, the degree of RP o stabilization by CarD decreased to the point that on AP3 Stable , the addition of CarD resulted in no detectable difference in the amount of three nucleotide product.
Having established a set of promoters with different basal RP o stability levels that range over nearly one order of magnitude, we performed multiround in vitro transcription reactions using these AP3 promoter variants in the presence or absence of CarD to investigate the relationship between the basal RP o stability of a promoter and transcriptional regulation by CarD (Fig. 3C). We discovered that across the AP3 promoter variants, basal RP o stability positively correlated with full-length transcript production in the absence of CarD but negatively correlated with transcriptional activation by CarD. Indeed, the two promoters with the highest levels of basal RP o stability (AP3 EcoExt and AP3 Stable ) were transcriptionally repressed by CarD, consistent with the predictions of our model and providing the first in vitro evidence of direct transcription repression by Mtb CarD. Basal RP o stability and CarD regulatory outcome are influenced by discriminator region guanosine + cytosine base pair frequency In addition to forming direct interactions with the polymerase, promoter DNA sequences can also influence RP o stability by affecting the chemical properties of the DNA molecule. For example, guanosine + cytosine base pairs in the discriminator region impose a kinetic barrier to DNA untwisting and unwinding during the formation of the transcription bubble due to their greater base-pairing and base-stacking stability compared to adenosine + thymine base pairs (40,41). Discriminator guanosine + cytosine base pair frequency (G + C%) is inversely correlated with RP o stability (42) and has been shown to be a determinant of transcription control by DksA/(p)ppGpp (5,43). To determine if changing the RP o stability by modifying the G + C% of the discriminator affects the outcome of CarD activity on transcript production, we generated a set of AP3 promoter variants (AP3 Discr1 -AP3 Discr5 ) in which the discriminator region G + C% is titrated from 100% (AP3 Discr1 ) to 16.7% (AP3 Discr5 ) (Fig. 4A). We Initiation kinetics affects CarD activity observed a negative correlation between discriminator G + C% and basal RP o stability as measured by three-nucleotide transcription assays (Fig. 4B). CarD increased three nucleotide RNA production from all promoter variants tested, but the magnitude of RP o stabilization by CarD displayed a negative correlation with basal RP o stability across AP3 variants as the discriminator G + C% was titrated (Fig. 4B).
Discriminator G + C% of the AP3 variants was also negatively correlated with basal transcript production in the multiround transcription assay and the magnitude of transcription activation by CarD decreased as discriminator G + C% decreased (Fig. 4C). On the AP3 variant with the lowest discriminator G + C% and highest basal RP o stability (AP3 Discr5 ), CarD decreased transcript production, further supporting that promoters with high basal RP o stability can be transcriptionally repressed by CarD. Collectively, our experiments show that promoter sequence motifs that increase basal RP o stability decrease the magnitude of transcriptional activation by CarD and can lead to transcriptional repression in the most stable RP o contexts.

Promoter sequences that form more stable RP o are associated with transcription repression by CarD in vitro and in vivo
We show that base substitutions in the spacer region, extended −10 region (Fig. 3), and discriminator (Fig. 4) can affect full-length transcript production and the direction of CarD regulation. To directly examine whether differences in relative RP o stability could explain the outcomes in transcript production and CarD regulation, we performed a linear regression analysis across all of our promoter templates (Fig. 5). For this analysis, the relative RP o stability and relative transcription strength of each promoter variant was normalized to AP3 WT . In the absence of CarD, the rate of full-length transcript production shows a roughly linear positive correlation with the relative RP o stability (Fig. 5A). In contrast, the log 2 ratio of transcript production in multiround transcription reactions ± CarD shows a roughly linear inverse correlation with increasing RP o stability, with the most stable promoter variants (AP3 EcoExt , AP3 Stable , and AP3 Discr5 ) being transcriptionally repressed by CarD (Fig. 5B). The robust relationship across multiple promoter variants suggests that RP o stability is a fundamental determinant of fulllength transcript production and CarD regulatory outcome. Collectively, our experiments illustrate a relationship between RP o stability, transcription strength, and CarD regulation and demonstrate that transcription factors like CarD can discriminate promoters based on their basal kinetic features to potentiate bidirectional outcomes in transcription regulation via a single kinetic mechanism.
Through our in vitro transcription experiments, we have identified multiple promoter sequence motifs associated with high RP o stability in vitro. If our model is generally applicable to transcription from mycobacterial promoters throughout the genome, then we would expect to find an association between DNA sequence motifs associated with RP o stability and transcriptional repression by CarD. To interrogate this prediction, we examined the prevalence of a consensus extended −10 motif (T -15 G -14 N -13 ) and discriminator GC% in promoters that were differentially expressed in our Mtb and M. smegmatis RNA-seq datasets (Table S5). Since all of our in vitro experiments were performed in the context of a MtbRNAP-σ A holoenzyme, we limited our bioinformatic analysis to promoters containing a A -11 NNNT -7 motif representing the consensus σ A −10 element (24,29,44,45), which comprised 90.5% (1609/1778) and 82.5% (2511/3043) of the primary TSSs in Mtb (44) and M. smegmatis (24), respectively. Indeed, in Mtb, promoters that were predicted to be repressed by CarD based on our RNA-seq data were overenriched for extended −10 elements, while promoters predicted to be activated by CarD were underenriched for extended −10 elements relative to the genome-wide proportion of this feature (Fig. S4A). A similar trend was true of the proportion of promoters containing extended −10 elements in M. smegmatis, but the difference in proportions between CarDregulated promoters and the genome-wide distribution was not Initiation kinetics affects CarD activity statistically significant (Fig. S4B). In both species, promoters that were predicted to be repressed by CarD contained significantly more GC-rich discriminator regions than promoters predicted to be activated by CarD (Fig. S4, C and D). The association of stable RP o DNA sequence signatures with genes that are inferred to be repressed by CarD in vivo support that the regulatory mechanisms that we demonstrate in vitro could be relevant to gene expression in vivo.

DNA topology can influence the regulatory outcome of CarD activity
In mycobacteria, CarD transcript levels increase in response to double-stranded DNA breaks and genotoxic stress (9), suggesting that the dynamics of CarD regulation may be important for responding to these environmental cues. DNA breaks in the chromosome can relieve local regions of DNA supercoiling. The supercoiling state of promoters is tightly connected to transcriptional activity in vivo, as positive or negative supercoiling can inhibit or enhance RP o formation, respectively (46,47). Thus, we sought to test the relationship between promoter topology and CarD regulation. We generated a set of templates with identical DNA sequence but varied molecular topology by cloning the AP3 WT promoter into a negatively supercoiled plasmid and incubating the plasmid with either a single-cutting endonuclease to produce a linear "cut" DNA molecule, a nicking endonuclease to produce a circular "nicked" DNA molecular, or with no enzyme to maintain a supercoiled control (mock treated) (Fig. 6A). We performed in vitro three-nucleotide transcription assays using the topologically distinct DNA templates and found that negative supercoiling contributes to a 6-fold increase in basal RP o stability compared to a linear "cut" DNA template containing the same promoter sequence (Fig. 6B). In addition, the "nicked" DNA template exhibited a similar basal RP o stability to the "cut" DNA template, indicating that the higher RP o stability observed in the "mock" template is a result of supercoiling and not the circular shape of the molecule. The addition of CarD decreased the amount of three nucleotide transcript produced with the supercoiled "mock" DNA template. This result could indicate that CarD inhibits progression from RP o toward an initial transcribing complex intermediate (RP itc ) that synthesizes the three nucleotide product quantified in these assays (18). The basal RP o stabilities of the "cut", "nicked", and "mock" AP3 WT DNA templates correlated with the basal transcriptional activity of the promoter, where promoter templates with high basal RP o stability also showed high levels of basal transcript production (Fig. 6C). Furthermore, CarD activated transcription from the "cut" and "nicked" DNA templates but repressed transcription from the supercoiled "mock" DNA template, which has a higher basal RP o stability relative to the "cut" and "nicked" molecules. These data demonstrate a single promoter DNA sequence can exhibit varying levels of basal RP o stability based on DNA supercoiling, and this supercoiling-dependent change in RP o stability can change the regulatory outcome of CarD on transcription. While the DNA sequence of a given promoter is constant within the genome, the topology of the DNA molecule can change over the lifetime of the cell. Thus, our findings reveal an additional layer of complexity in CarD's regulatory mechanism and could help explain how CarD expression in vivo could lead to differential gene expression outcomes in different conditions.

Discussion
CarD is an essential transcriptional regulator in Mtb that affects the expression of over two-thirds of the genome (8) and whose normal function and expression are required for bacterial survival during various stresses and virulence in mice (9,10,13,48). Numerous in vitro studies have shown that CarD stabilizes RP o formed by the housekeeping MtbRNAPσ A holoenzyme (12,15,16), leading to the early model that it functions as a general transcription activator. However, a subsequent RNA-seq study of Mtb strains encoding mutant alleles of CarD revealed a more complex scenario where CarD appears to differentially activate or repress transcription from different promoters (8). In an effort to understand how CarD could affect gene expression in a promoter specific manner, we now provide experimental evidence for a relationship between RP o stability and the outcome of CarD regulation that results in promoter specific effects of CarD activity. We find that the ratio of transcript production in multi-round transcription reactions ± CarD shows a roughly linear inverse correlation with increasing RP o stability, with the most stable promoter variants (AP3 EcoExt , AP3 Stable , and AP3 Discr5 ) being transcriptionally repressed by CarD. CarD's effect on mycobacterial transcription in vivo also reflects the observations from our in vitro experiments where promoters predicted to be repressed by CarD are associated with sequence features that correlate with high RP o stability (extended −10 sequence motif, low GC% discriminator region) and promoters predicted to be activated by CarD are associated with an absence of these features. Collectively, these data support our model in which the specific outcome of CarD-mediated RP o stabilization is dependent on the kinetic properties of a given promoter and not on sequence-specific binding, which could explain the observed differential gene expression effects in CarD mutants in vivo.
Our study deepens our understanding of mycobacterial transcription regulation and demonstrates how RNAP-binding factors like CarD add complexity to this process. The true relationship between promoter sequence and CarD regulation is likely more nuanced than the data presented in this study. Although the AP3 promoter variants we generated were designed to increase or decrease RP o stability in a stepwise manner (i.e., AP3 Stable is a combination of AP3 Discr and AP3 ExoEct ; G + C% is titrated one base at a time in AP3 Discr1 -AP3 Discr5 ), the effects of each mutation are likely more complex. Minor base substitutions in a promoter sequence can result in large-scale allosteric effects on other RNAP-DNA interactions (49) and kinetic steps (50) outside of RP o . Furthermore, our model was built on the idea that CarD uses a single kinetic mechanism, but CarD's effects on specific transcription initiation rate constants may differ between promoters. For example, CarD contains a conserved tryptophan residue in its DBD (W85) that is positioned to interact with a T at the −12 position of the nontemplate DNA strand at the upstream fork of the bubble (12), and it has been hypothesized that this sequence-specific interaction acts as a "wedge" to prevent bubble collapse. In theory, on a promoter lacking T -12 , CarD's inhibitory effect on k collapse may be diminished relative to its effects on the rates of bubble opening and promoter escape, producing a unique kinetic mechanism that is biased toward repression. In our RNA-seq dataset, Mtb promoters that were predicted to be repressed by CarD were significantly enriched for non-T bases at the −12 position (8), lending some in vivo support for the prediction that this DNA sequence context biases CarD toward transcriptional repression.
The specific interaction between CarD W85 and T -12 also raises the possibility that specific promoter DNA sequences could influence CarD's binding preference for partially opened promoter complexes. A recent study showed that CarD has poor binding affinity for RNAP or DNA alone (51), suggesting that CarD may bind to transcription initiation complexes after the initial association between RNAP and DNA. In our ChIP-seq dataset, M. smegmatis promoters associated with a primary TSS and containing a σ A -like −10 element motif (A -11 NNNT -7 ) that were within 100 bp of a CarD binding site were significantly underenriched (hypergeometric test p = 5.66e-06) for promoters lacking a T -12 (22%; 93/429) compared to the genomewide proportion (30%; 765/2511) (Tables S3 and S5). These data support the hypothesis that certain DNA sequence-specific interactions may influence but not determine the association between CarD and transcription complexes at specific mycobacterial promoters.
Another region that we did not study but could affect CarD regulation is the initially transcribed sequence downstream of the TSS, which can affect the kinetics of promoter escape and RNAP pausing (52)(53)(54). Simplistically, CarD represses transcription from certain promoters by overstabilizing RP o and decreasing transcript flux by inhibiting promoter escape (7), leading to an accumulation of abortive transcripts (23). However, this model becomes more complicated when considering a branched pathway of transcription initiation (55), where a fraction of RNAP form moribund complexes that never undergo promoter escape. Kinetic studies using a fluorescent reporter showed that CarD increases the fraction of unescaped RNAP complexes (18), and we show that in some contexts, CarD can inhibit the synthesis of a three nucleotide product from RP o . These data suggest that CarD could affect steps of initial nucleotide incorporation prior to promoter escape and influence the fraction of RNAP complexes undergoing productive versus moribund transcription (54,56).
We also find that this relationship between RP o stability and CarD is not only influenced by promoter sequence, where promoters with identical DNA sequence can be differentially activated or repressed depending on their supercoiling status (Fig. 6). In our experiments, CarD directly activated transcription from the Mtb rRNA promoter AP3 on a linear DNA template but repressed transcription from AP3 on a negatively supercoiled template (Fig. 6), which is the predominant topological state of DNA in bacterial cells (46). On the surface, this seems to contradict CarD's role as a positive regulator of rRNA synthesis in vivo (13,20,25,57). However, one possible explanation may be that CarD is required to maintain efficient transcription of operons downstream of highly transcribed regions, such as the rRNA operon, when they accumulate positive supercoils due to their high transcriptional activity (46). We propose that CarD may function to overcome the topologically self-limiting nature of rRNA transcription to promote rapid bacterial growth.
Beyond its role in specifically regulating rRNA synthesis, CarD also affects the transcription of hundreds of other Mtb genes in vivo (8), which could explain CarD's pleiotropic effects under different stresses. Mtb strains with altered CarD activity are also sensitized to various environmental stresses other than nutrient starvation including oxidative stress, genotoxic stress, and antibiotic treatment (9,10,13,48), but it is still unclear what roles CarD plays under these conditions. The impact of topology and promoter context also implies that CarD may elicit different effects on gene expression in different environments. CarD's ability to interpret the kinetic properties of a promoter add modularity to the mycobacterial transcription response, because while DNA sequence is essentially constant over the lifetime of a bacterial cell, its kinetic properties may be dynamic and responsive to environmental stimuli. In vivo, the supercoiling state of a promoter is constantly changing in response to the translocation of polymerases, the enzymatic action of topoisomerases, and DNA damage caused by antibiotics or other genotoxic stresses (46,58). In addition to DNA supercoiling, other environmental factors such as intracellular NTP concentrations (59) and temperature (60) can influence the kinetic properties of a promoter without affecting DNA sequence. During pathogenesis, Mtb may encounter these environmental stimuli in various combinations. Furthermore, the expression of CarD is itself highly responsive to environmental signals, including nutrient limitation (48) and DNA damage (9). Understanding how transcription factors like CarD interact with these environmental stresses may provide insight into how Mtb responds to the host environment and antibiotic treatment, making this an intriguing direction of future study.
Based on the results of this study, we propose that CarD belongs to a growing class of RNAP-binding transcription factors that include DksA/(p)ppGpp (6,61), TraR and its phage-encoded homologs (62,63), and the σ-subunit interacting transcription factors including the Actinobacteriaspecific protein RbpA (3,(64)(65)(66)(67). Like CarD, these factors coordinate broad transcriptional programs in bacteria (5,14,65), highlighting the expanded regulatory range of these factors compared to classical transcription factors that are limited to promoters containing a specific binding motif. All of these global transcriptional regulators function by modulating the kinetics of transcription initiation, albeit via different mechanisms. Whereas CarD stabilizes RP o , DksA/(p)ppGpp binds RNAP and destabilizes a kinetic intermediate preceding RP o , resulting in transcriptional repression at ribosomal RNA promoters that form unstable RP o and transcriptional activation at promoters of amino acid biosynthesis genes that form relatively stable RP o (6,61,(68)(69)(70). Although they exert opposite effects on initiation kinetics, CarD and DksA/(p)ppGpp share the ability to "read" the kinetic properties of a promoter to exert multiple regulatory outcomes on transcription. This study of CarD's regulatory mechanism demonstrates how kinetic context influences the activity of this class of RNAPbinding transcription factors and reveals another layer in how bacteria coordinate broad gene expression in response to their environment.

Experimental procedures
Bacterial growth and RNA collection All M. smegmatis strains used in this study were derived from mc 2 155 and grown in LB medium supplemented with 0.5% dextrose, 0.5% glycerol, and 0.05% Tween-80 at 37 C. M. smegmatis strains expressing CarD WT , CarD R25E , CarD K125E , or CarD I27W were engineered so that the native copy of carD is deleted, and the respective CarD allele is expressed from a constitutive Pmyc1-tetO promoter integrated into the genome. The construction of these strains has been previously described (13,20). For RNA collection, M. smegmatis cultures were grown to A 600 0.5 to 0.9, pelleted, and lysed in TRIzol reagent (Invitrogen) by bead-beating. RNA was isolated by TRIzol-chloroform extraction followed by isopropanol precipitation and finally resuspended in nucleasefree water (Invitrogen).

RNA sequencing and data analysis
RNA samples were DNase treated using the TURBO DNAfree Kit (Invitrogen) and submitted to the Washington University Genome Technology Access Center for paired-end Illumina sequencing (NovaSeq 6000 XP). Ribosomal RNA was depleted prior to sequencing using the Qiagen FastSelect system. Illumina reads were preprocessed using FastQC, and adapter sequences were removed using trimmomatic (71). Sequencing reads were aligned using HiSat2 (72) to the M. smegmatis mc 2 155 reference genome (assembly ASM1500v1) from the Ensembl database (73). Reads mapping to annotated protein coding regions were quantified using featureCounts (74). Differential expression analysis was performed using DESeq2 (75). Downstream data analysis and visualization was performed using custom R scripts.

In vitro transcription
Promoter fragments used for in vitro transcription were prepared by annealing two complementary single-stranded DNA oligos (IDT) containing the WT or variant AP3 promoter sequence from positions −39 to +4 relative to the transcription start site to create a linear double-stranded DNA fragment that was ligated into the pMSG434 plasmid. Linear DNA templates used for in vitro transcription were prepared PCR amplifying a 437 bp fragment from the pMSG434 plasmid. Plasmid DNA templates for in vitro transcription were constructed by inserting an intrinsic transcription termination sequence (5 0 -TTTAT-3 0 ) into the pMSG434 plasmid 70 bp downstream of the cloned AP3 transcription start site. Negatively supercoiled plasmids were grown in E. coli and then isolated using a QIAGEN Plasmid Midi Kit. To generate cut or nicked plasmid templates, plasmid DNA was incubated with XhoI restriction endonuclease (NEB) at 37 C or Bt.NstNBI nicking endonuclease (NEB) at 55 C for 1 h, respectively. All DNA templates were purified by extracting with buffer-saturated phenol pH >7.4 (Invitrogen) followed by isopropanol precipitation before being used in in vitro transcription reactions. A full list of the primers used to construct the DNA templates can be found in Table S6.

Data availability
Raw RNA-seq data have been deposited in the GEO repository under accession code GSE222815.
Acknowledgments-We thank Drake Jensen, Ana Ruiz Manzano, and Eric Galburt for their helpful discussions and for providing the E. coli protein expression strain for MtbRNAP-σ A purification. We also thank John Errico, Helen Blaine, and Daved Fremont for their help in purification of RNA polymerase.