An Integrated Database of Small RNAs and Their Interplay With Transcriptional Gene Regulatory Networks in Corynebacteria

Small RNAs (sRNAs) are one of the key players in the post-transcriptional regulation of bacterial gene expression. These molecules, together with transcription factors, form regulatory networks and greatly influence the bacterial regulatory landscape. Little is known concerning sRNAs and their influence on the regulatory machinery in the genus Corynebacterium, despite its medical, veterinary and biotechnological importance. Here, we expand corynebacterial regulatory knowledge by integrating sRNAs and their regulatory interactions into the transcriptional regulatory networks of six corynebacterial species, covering four human and animal pathogens, and integrate this data into the CoryneRegNet database. To this end, we predicted sRNAs to regulate 754 genes, including 206 transcription factors, in corynebacterial gene regulatory networks. Amongst them, the sRNA Cd-NCTC13129-sRNA-2 is predicted to directly regulate ydfH, which indirectly regulates 66 genes, including the global regulator glxR in C. diphtheriae. All of the sRNA-enriched regulatory networks of the genus Corynebacterium have been made publicly available in the newest release of CoryneRegNet(www.exbio.wzw.tum.de/coryneregnet/) to aid in providing valuable insights and to guide future experiments.

Due to its importance, both computational and experimental techniques have been developed for identifying sRNAs and their interactions. Experimental methods such as total RNA labeling (Wu et al., 1996), deep sequencing (Sittka et al., 2008;Sharma and Vogel, 2009;Barquist and Vogel, 2015) and co-immunoprecipitation of RNA-binding proteins (Faner and Feig, 2013) have been used to discover novel sRNAs. Other techniques, such as pulse-expression (Massé et al., 2005), MAPS (Lalaouna and Massé, 2015), RIL-seq (Melamed et al., 2016), and GRIL-seq (Han et al., 2016) have been applied to identify sRNA-mRNA interactions. For a comprehensive description see Altuvia (2007), Ahmed et al. (2018), and Diallo and Provost (2020). Computational methods stand out by revealing promising sRNA candidates for further experimental testing without exhaustive wetlab assays (Wright and Georg, 2018). In general, sRNA prediction software can be grouped into three types of methods: de novo, homology-based and experimental-data dependent (Zhang Y. et al., 2017;Backofen et al., 2018). sRNA target prediction software can be divided into two types of methods: local-interaction based and full-hybrid based (Pain et al., 2015). For further explanations and comparisons of these methods see Pain et al. (2015), Zhang Y. et al. (2017), and Backofen et al. (2018).
Both predicted and experimental bacterial sRNAs have been made publicly available in databases such as Rfam (Kalvari et al., 2018) and RNA central (The RNAcentral Consortium, 2019) for several organisms, including bacteria. Likewise, sRNA data for Gram-positive bacteria is available on sRNAdb (Pischimarov et al., 2012). BSRD (Li et al., 2013), sRNATarBase (Wang et al., 2016), sRNAMap (Huang et al., 2009), and RNAInter (Lin et al., 2020) provide sRNA regulatory information for several bacterial species. Despite the influence and importance of these molecules on gene expression, databases integrating sRNA-based and transcriptional regulatory networks are largely missing. To the best of our knowledge, RegulonDB (Santos-Zavaleta et al., 2019), the reference database for Escherichia coli GRNs, is the only one to have done this integration though exclusively for E. coli K12.
In the context of the Corynebacterium genus, CoryneRegNet (Parise et al., 2020) is the reference database for Corynebacterial transcriptional regulatory networks, containing more than 80,000 regulatory interactions but lacking sRNA data. A few Corynebacterial sRNAs can be found in BSRD (Li et al., 2013), Rfam (Kalvari et al., 2018), and RNA central (The RNAcentral Consortium, 2019). For Corynebacterium glutamicium, the model organism for this genus, 805 sRNAs were experimentally identified using deep sequencing and were reported in Mentz et al. (2013). However, there are no experimental or predicted sRNA regulations for the Corynebacterium genus.
Here, we present the first study about the integration of sRNA regulations with transcriptional regulation in corynebacteria. We predicted sRNAs and their targets for six Corynebacterium species of either medical, veterinary or industrial interest, yielding 922 sRNAs and 6,389 sRNA regulatory interactions. This data was integrated into CoryneRegNet 7.5, revealing 754 genes in the GRN to be regulated by both sRNAs and transcription factors and 206 regulatory proteins to be regulated by sRNAs. In a case study of human pathogenic corynebacteria using the CoryneRegNet 7.5 sRNA-enriched database content, we predict the sRNAS Cd-NCTC13129-sRNA-2 and scjk1464.1 to form regulatory cascades with TFs. Cd-NCTC13129-sRNA-2 is predicted to regulate the ydfH homolog, indirectly regulating 66 genes in C. diphtheriae and scjk1464.1 is predicted to regulate mcbR and dtxR, indirectly regulating 35 genes in C. jeikeium. In the animal pathogen C. pseudotuberculosis, the virulence factor fagC is also predicted to be regulated by the sRNA Cp-1002B-sRNA-1. To sum up, the integration of sRNAs and their interactions into the transcriptional regulatory networks in CoryneRegNet provides a more comprehensive view on corynebacterial regulatory mechanisms.

MATERIALS AND METHODS
The CoryneRegNet sRNA integration pipeline consists of seven steps: sRNA collection and prediction, homology detection, alignment, sRNA classification, filter, structure prediction and target prediction. An overview of these steps is shown in Figure 1. We started with compiling a dataset of 805 experimentally verified sRNAs from Mentz et al. (2013) and 70 predicted sRNAs from BSRD (Li et al., 2013). In order to predict novel sRNAs, we used cmsearch (Nawrocki and Eddy, 2013) on the target genomes with no experimental sRNAs publicly available. Details about the sRNA datasets and the genomes used in this analysis are given in Table 1.
Afterward, we identified homologs for every sRNA in the analysis by using GLASSgo (Lott et al., 2018). Homologous sRNAs belonging to the genomes of interest were incorporated into the analysis. For each sRNA in the analysis, we selected its most distant homologs from the same species and from the same genus with ≥80% of similarity. Thus, these sequences were aligned by using clustalo (Sievers et al., 2011). The sRNAs were classified as either functional or non-functional by running RNAz (Gruber et al., 2010) and RNAdetect (Chen et al., 2019) based on the stability and the conservation of the predicted RNA structures as well as on sequence homology. Predicted sRNAs that were classified as non-functional were removed from the analysis. The secondary structure was predicted using RNAalifold (Bernhart et al., 2008) for every sRNA in the analysis. Furthermore, sRNA targets were predicted by running CopraRNA (Wright et al., 2013) with default settings. Adjusted p-values were calculated using the Beijamini-Hochberg correction from the R package stats, method p.adjust (Stats, 2020). Then, we selected the fifteen best-ranked interactions predicted with a p-value < 0.01, as suggested in Wright and Georg (2018). The sRNAs and their targets were integrated into CoryneRegNet (Parise et al., 2020) by updating the front-end and back-end, as well as the database. Finally, we predicted gene ontologies for every gene regulated by sRNAs by running Go Feat (Araujo et al., 2018). A detailed

Database Content
We presented CoryneRegNet 7.5, an updated release of the corynebacterial reference database and analysis platform, now including sRNA networks integrated with the transcriptional regulatory networks of the genus Corynebacterium. A total of 922 sRNAs and 6,389 regulatory interactions for six corynebacterial strains were integrated into our database, as shown in

Website
We updated CoryneRegNet's user interface to present information concerning sRNAs and their targets. Both the regulatory interaction table view and the network view were updated and enriched with corresponding sRNA-related features. The search page now allows the user to (i) search for gene identifiers ( Figure 2B) when querying the database  In the network, green nodes represent activator proteins, red nodes represent repressor proteins, blue nodes represent dual regulators (i.e., that can activate and repress gene expression), orange nodes represent sRNAs and gray nodes represent target genes. The arrows represent the regulatory interactions and their colors represent the same roles as in the nodes.
for mRNA or sRNA genes ( Figure 2A) and (ii) search for a list of genes. Depending on the search choice (Figure 2A), the user will be directed to the gene-centered or sRNA-centered network view, as presented in Figures 2C,D, respectively. sRNAs and their regulatory interactions have been integrated into the network visualization as orange nodes and directed edges. Considering there is no annotation of activation/repression prediction for the sRNA-mRNA interactions, we represent every sRNA regulatory interaction as an orange, directed edge. The complete sRNA-mRNA interactions set of a genome can also be visualized in case no specific gene or sRNA is selected.
In addition, users can now find genes and sRNAs of interest by using the new filtering and sorting features in the table-oriented view, as presented in Supplementary Figures 1A,B, respectively. In the sRNA view, we included filters for: (i) sRNAs regulating transcription factors, (ii) sRNAs regulating genes in the TRN, and (iii) functional sRNAs. Likewise, in the gene view we included filters for: (i) genes encoding regulatory proteins, (ii) genes regulated by regulatory proteins, (iii) genes regulated by sRNAs, and (iv) genes regulated by sRNAs and/or regulatory proteins.
A sample sRNA page is displayed in Figure 3A. It presents essential information of the sRNA of interest such as: type of evidence, position and orientation in the genome, whether or not the sRNA was classified as functional, and the sRNAs' nucleotide sequence. The predicted structure of the selected sRNA is also presented along with its dot plot and alignment graph. The former illustrates the interaction between the nucleotides (Supplementary Figure 2A) and the latter the conservation between the sRNA of interest and its homologous sRNAs (Supplementary Figure 2B). Additionally, the user can visualize the sRNA regulatory interactions in the "Regulates" tab ( Figure 3B). This tab shows information regarding each regulatory interaction predicted by CopraRNA (Wright et al., 2013) of the selected sRNA such as its position, minimum energy, hybridization energy and p-value.
Furthermore, we integrated the sRNA interaction network into the statistics section with three new analyses: (i) quantities of sRNA types (Supplementary Figure 3A), (ii) distribution of sRNAs regulating a gene (Supplementary Figure 3C), and (iii) distribution of co-regulating sRNAs (Supplementary Figure 3B). Finally, we updated the documentation and workflow sections at the website accordingly.

Case Study
We illustrate the utility of the sRNA-enriched CoryneRegNet 7.5 by utilizing the updated filtering features to identify 206 regulatory proteins regulated by sRNAs and 754 genes regulated by both sRNAs and TFs in our six genomes. We selected the genes regulated by both sRNAs and TFs in the following four pathogenic bacteria: C. diphtheriae NCTC 13129, C. jeikeium K411, C. pseudotuberculosis 1002B and C. ulcerans NCTC7910. In addition, we selected gene circuits in these pathogenic bacteria and in the model organism C. glutamicum and presented whether these observations are conserved in C. efficiens. We visualized the regulatory networks of these genes using the list-based network feature in CoryneRegNet 7.5, where we also collected their homologous genes.
In C. glutamicum, we predicted 662 genes to be co-regulated by sRNAs and TFs. Amongst them, we can highlight cg0350, sdhCD, acn, cgtR3, pstA, and the sigma factor sigA, as presented in Figure 4A. The sRNA cgb_1195 potentially co-regulates cg0350 (glxR homolog) together with four transcriptional regulators: cg2544 (ydfH homolog), cg0146 (sucR homolog), sigA, and cg0444 (ramB homolog). Additionally, cg0350 has been reported to regulate itself in this organism. The sRNA is predicted to directly and indirectly regulate the highly regulated genes sdhCD and acn, forming feed forward loop Cg-FF-1 ( Figure 4A). These two genes are also part of the dense overlapping regulon Cg-DOR-1, in which three other sRNAs potentially co-regulate them together with five TFs and sigA. The membrane anchor subunit sdhCD jointly encodes with sdhA and sdhB the succinate dehydrogenase enzyme, a component of the TCA cycle (Polen et al., 2007;Bussmann et al., 2009). The acn gene is also a component of the TCA cycle; it encodes an aconitase enzyme and its inactivation is detrimental to cell growth (Yoon and Woo, 2018). Both the sdhCD and acn genes were found differentially expressed in acetate medium when compared with glucose medium (Bott, 2007). Figure 4B presents the highly regulated pstA as being potentially co-regulated by six sRNAs, two transcription factors and sigA. The sRNA cgb_04174 is predicted to directly and indirectly regulate pstA, forming the feed forward loop Cg-FF-2. In total, pstA is predicted to be directly regulated by six sRNAs and indirectly regulated by eigth sRNAs. This gene is part of the Pst system, which is part of the inorganic orthophosphate (P i ) starvation stimulon in C. glutamicum (Ishige et al., 2003). The transcriptional regulators sigA, cgtR3 and cg0350 are also predicted to be regulated by sRNAs. SigA is the primary sigma factor in C. glutamicum and is potentially regulated by five sRNAs; this regulator is considered responsible for the transcription of the majority of the housekeeping genes in this organism (Oguiza et al., 1996;Schröder and Tauch, 2010). The global regulator cg0350 (glxR homolog) has been reported to be involved in the regulation of 195 genes in C. glutamicum (Freyre-González and Tauch, 2017;Parise et al., 2020) and is potentially regulated by one sRNA. The regulator cgtR3 (phoR) is the master regulator of phosphate metabolism in C. glutamicum and is potentially regulated by two sRNAs (Schröder and Tauch, 2010).
None of the observations mentioned so far is conserved in the other organisms analyzed in this study. Furthermore, mraZ is predicted to be regulated by 22 sRNAs, as presented in Figure 4C. This gene is highly conserved in bacteria and is part of the division cell cluster (dcw) (Eraso et al., 2014). The cleavage of the coding region of its mRNA is required for efficient cell division in C. glutamicum (Maeda et al., 2016). The other genes from the mraZ operon, mraW, and cg2376 (ftsL homolog), are potentially regulated by sRNAs. MraW is potentially regulated by six sRNAs; amongst them, cgb_03605 is also predicted to regulate mraZ. Cg2376 is predicted to be regulated by one sRNA. MraZ homolog genes in C. efficiens, C. jeikeium, and C. pseudotuberculosis are also potentially regulated by 10 sRNAs, two sRNAs and one sRNA, respectively. In C. ulcerans, the mraW homolog is potentially regulated by one sRNA, whereas none of the cg2376 homologs are predicted to be regulated by sRNAs in this study.
In C. diphtheriae NCTC 13129, we predicted 16 genes to be co-regulated by sRNAs and TFs; the regulatory network of these genes can be seen in Figure 5. Amongst them, the sRNA Cd-NCTC13129-sRNA-2 potentially regulates the transcription factor DIP_RS19435 (ydfH homolog), forming a single-input module inside the dense overlapping regulon Cd-DOR-1 (Figure 5). The ydfH homolog is predicted to autoregulate itself and to regulate DIP_RS12895 (glxR homolog). It forms a regulatory cascade where the complete set of genes regulated by glxR may be indirectly regulated by this sRNA, accounting for 66 genes. The complete regulon of ydfH and glxR is presented in Supplementary Figure 4. As presented in the dense overlapping regulon Cd-DOR-1 (Figure 5), the GlxR homolog TF potentially co-regulates four genes with sRNAs: DIP_RS15610 (ispE homolog), gap, odhA and DIP_RS12055. The sRNA Cd-NCTC13129-sRNA-4 potentially regulates both the ispE homolog and DIP_RS14355, a methionine ABC transporter substrate-binding. The latter is also regulated by the TetR/AcrRfamily regulator DIP_RS23775 (mcbR homolog). In C. efficiens, the homologous methionine ABC transporter substrate-binding (CE_RS03295) is also potentially co-regulated by one sRNA (Ce-YS314-sRNA-28) and a TetR/AcrR family TF (CE_RS13790). Also in Cd-DOR-1 (Figure 5), gap and odhA are predicted to be regulated by the same sRNA, scdi510.1, which also co-regulates mdh along with the LuxR family regulator DIP_RS20635 (ramA homolog). Likewise, gap (cg1791) is also predicted to be coregulated by cg0350 (glxR homolog) and the sRNAs scgl2151.1, cgb_23426 and cgb_10355 in C. glutamicum. In general, the genes in Cd-DOR-1 are involved in the TCA cycle and in carbohydrate metabolism.
Also in C. diphtheriae, five other genes are potentially coregulated by both sRNAs and TFs. The hemin-binding protein hmuT (Draganova et al., 2015) is potentially co-regulated by scdi175.1 and dtxR. The sRNA scdi28.1 is predicted to co-regulate the heat-shock protein GroEL2 along with the transcription factor hrcA. In C. efficiens, the GroEL2 homolog (CE_RS12690) is also predicted to be regulated by a sRNA (Ce-YS314-sRNA-3) and a hrcA homolog (CE_RS10870). In C. diphtheriae, Cd-NCTC13129-sRNA1 potentially regulates DIP_RS12535 (pdxS homolog) and pyk, which are also regulated by DIP_RS18315 (gatR homolog) and DIP_RS12530 (pdxR homolog), respectively. In the networks, green nodes represent activator proteins, red nodes represent repressor proteins, blue nodes represent dual regulators (i.e., that can activate and repress gene expression), orange nodes represent sRNAs and gray nodes represent target genes. The arrows represent the regulatory interactions and their colors represent the same roles as the ones in the nodes.
We also observed the DIP_RS18360 gene (hflX homolog) being potentially co-regulated by an XRE family transcriptional regulator and the sRNA scdi1478.1.
In C. jeikeium K411, we predicted twenty genes to be jointly regulated by sRNAs and TFs; the regulatory network of these genes is presented in Figure 6. Amongst these genes we identified two dense overlapping regulons, highlighted as Cj-DOR-1 and Cj-DOR-2. In Cj-DOR-1, the sRNAs scjk260.2, scjk885.1, scjk557.1, scjk1019.1 are predicted to co-regulate five genes (rhtC, fadH, rpfB, cat1, and JK_RS05010) with the global regulator glxR. The gene JK_RS05010 (rpfI homolog) was predicted to have hydrolase activity and is potentially co-regulated by glxR, mtrA FIGURE 5 | Genes regulated by sRNAs and regulatory proteins in C. diphtheriae NCTC 13129. In the network, green nodes represent activator proteins, red nodes represent repressor proteins, blue nodes represent dual regulators (i.e., that can activate and repress gene expression), orange nodes represent sRNAs and gray nodes represent target genes. The arrows represent the regulatory interactions and their colors represent the same roles as the ones in the nodes.
FIGURE 6 | Genes regulated by sRNAs and regulatory proteins in C. jeikeium K411. In the network, green nodes represent activator proteins, red nodes represent repressor proteins, blue nodes represent dual regulators (i.e., that can activate and repress gene expression), orange nodes represent sRNAs and gray nodes represent target genes. The arrows represent the regulatory interactions and their colors represent the same roles as the ones in the nodes. and scjk577.1. The rpfI gene, which encodes a resuscitationpromoting factor interacting protein, is a virulence factor in C. ulcerans (Trost et al., 2011). The deletion of this gene impaired the growth of long-stored cells in C. glutamicum (Hartmann et al., 2004). The other resuscitation-promoting factor, rpfB, is also potentially regulated by mtrA. In C. efficiens, the rpfB homolog is also potentially co-regulated by the sRNA Ce-YS314-sRNA-12, the glxR homolog (CE_RS01675) and the mtrA homolog (CE_RS03955). Also in Cj-DOR-1, metB and metX are potentially co-regulated by metR and one sRNA, these genes are involved in the metabolism of methionine in C. glutamicum (Rückert et al., 2003). In the single-input module Cj-SIM-1, the sRNA Cj-K411-sRNA2 potentially regulates the transcription factor JK_RS05100 (sufR homolog), indirectly regulating the sufBDCS gene cluster and the nif operon (nifU-JK_RS05070). The genes in this circuit are involved in the formation of iron-sulfur clusters in bacteria (Frazzon, 2003;Outten and Wayne Outten, 2015). In C. efficiens, the sufR homolog (CE_RS08375) is also potentially regulated by two sRNAs (scef1290.1 and scef1536.1) and regulates the nif operon (nifU-CE_RS08405) as well as the sufBDCS gene cluster (CE_RS08400, CE_RS08395, CE_RS08390, CE_RS08385).
Cj-DOR-2 ( Figure 6) contains a cluster of 10 sRNAs potentially co-regulating two genes along with the transcription factors TcsR4 and ClgR. When analyzing these sRNAs, we noticed sRNAs scjk2061.1, scjk118.1, scjk463.1, scjk1484.1, scjk1444.1, scjk2091.1, scjk1857.1, scjk1861.1, scjk620.1, and scjk833.1 are identical copies of the same sRNA located in different regions of the genome. The genomic coordinates of these sRNAs are presented in Supplementary Table III. the following regions of the genome: 117083-117197, 462452-462566, 619808-619922, 832580-832694, 1443235-1443349, 1483232-1483346, 1856182-1856296, 1860886-1861000, 2060398-2060.512, 2090313-2090427. The genes potentially regulated by these sRNAs, clpC, and JK_RS07360, encode a Clp ATPase subunit and a hypothetical protein, respectively. In addition to regulating clpC, ClgR is also predicted to co-regulate two other genes with sRNAs, clpP2 and clpX. Both clpC and clpP2 are part of a protein quality control system of the cell along with the other proteolytic subunit clpP1 (Schröder and Tauch, 2010). ClpX is also an ATPase subunit that belongs to the Clp/Hsp100 superfamily, which is involved in stress response, energy metabolism, NADPH synthesis and glucose consumption (Huang et al., 2020). This observation is not conserved amongst the Corynebacterial species analyzed in this manuscript. In Cj-DOR-2, the sRNA scjk1464.1 and tscR4 potentially co-regulate the sensor histidine kinase tcsS4, which belongs to a two-component signal transduction system. These systems are important to bacteria due to their capacity to detect and adapt to changes in the environment (Pao and Saier, 1995). TscR4 is also predicted to regulate the copper chaperone JK_RS07345 alongside the sRNA Cj-K411-sRNA-3. Likewise, this sRNA potentially co-regulates the heat shock protein groES and the flavin-dependent oxidoreductase JK_RS00955, which are also regulated by the hrcA and JK_RS10540 (maR1 homolog), respectively. GroES is involved in the transport of proteins and in the post-translational folding, along with the heat shock protein GroEL (Rinke et al., 1992). In general, genes in Cj-DOR-2 are potentially involved in growth and cell proliferation.
In C. jeikeium, the diphtheria toxin repressor DtxR, regulates many genes associated with iron metabolism and forms the feed forward loop Cj-FFL-1 with the sRNA scjk1464.1 by directly and indirectly regulating rpsH (Figure 6). This sRNA is also predicted to directly regulate the transcription factor mcbR (Supplementary Figure 5). By potentially regulating mcbR and dtxR, scjk1464.1 is predicted to indirectly regulate thirty-five genes. Additionally, two other sRNAs (scjk830.1 and scjk1448.1) are predicted to regulate rpsH. This gene encodes a 30S ribosomal protein that is associated with the small ribosomal subunit and has been considered as a potential drug target in C. diphtheriae (Jamal et al., 2017;Hassan et al., 2018). By analyzing these sRNAs in Rfam, we observed that they do not belong to the same sRNA family. Furthermore, the sRNA scjk1019 is predicted to co-regulate rhtC with glxR and JK_04405 (argR homolog). This gene was used to increase the production of L-threonine in C. glutamicum (Diesveld et al., 2009).
In C. pseudotuberculosis 1002B, four genes were predicted to be co-regulated by sRNAs and TFs; the regulatory network of these genes is presented in Figure 7A. The fagC (Cp1002B_RS00130) gene is potentially regulated by sRNA Cp-1002B-sRNA-1, as well by the diphtheria toxin repressor (dtxR), and is part of the operon fagABC. This operon is an active part of the iron acquisition system and is a known virulence factor in C. pseudotuberculosis (Billington et al., 2002). Likewise, fagC is also potentially regulated by one sRNA (Cu-NCTC7910-sRNA-6) and dtxR (CKV68_RS01925) in C. ulcerans, as shown in Figure 7B. In C. pseudotuberculosis, Cp-1002B-sRNA-1 potentially coregulates the azoR gene along with marR1; this gene encodes a flavin mononucleotide (FMN)-dependent homodimeric azobenzene reductase and is involved in the response of oxidative stress. In C. efficiens, the azoR homolog (CE_RS08755) is also potentially regulated by one sRNA (scef1673.1) and the marR1 homolog (CE_RS06390), whereas in C. glutamicum, the azoR homolog (cg1850) is potentially regulated by three sRNAs (cgb_31975, cgb_30915, and scgl2371.1) and the marR1 homolog (cg1324). In C. pseudotuberculosis (Figure 7A), the gene pfkA (phosphofructokinase) is predicted to be regulated by Cp-1002B-sRNA-2, glxR, and Cp1002B_RS04515 (ramA homolog). This gene is involved in the reduction of the amount of fructose-6-phosphate during the L-serine fermentation process with sucrose as a carbon resource in C. glutamicum (Zhang X. et al., 2017). The PfkA homolog in C. glutamicum is also potentially regulated by sRNAs and TFs, as presented in Figure 4A. Also in C. pseudotuberculosis, Cp-1002B-sRNA-2 also regulates the recX gene along with LexA; both lexA and recX are involved in the bacterial SOS response, acting in DNA damage repair (Pogson et al., 1996;Jochmann et al., 2009;Resende et al., 2011). In C. glutamicum, the recX homolog (cg2140) is also potentially regulated by two sRNAs (cgb_10545 and cgb_17865) and the lexA homolog (cg2114).
In C. ulcerans NCTC7910, we also predicted other 2 genes to be regulated by sRNAs and TFs; the regulatory network of these genes is presented in Figure 7B. The pckG gene, which FIGURE 7 | Genes regulated by sRNAs and regulatory proteins in C. pseudotuberculosis 1002B (A) and in C. ulcerans (B). In the network, green nodes represent activator proteins, red nodes represent repressor proteins, blue nodes represent dual regulators (i.e., that can activate and repress gene expression), orange nodes represent sRNAs and gray nodes represent target genes. The arrows represent the regulatory interactions and its colors represent the same roles as the ones in the nodes. encodes a phosphoenolpyruvate carboxykinase, was predicted to be regulated by one sRNA and three transcription factors (glxR, ramA, and ramB). The transcription factor DnaK is regulated by one sRNA and the transcription factor glnR. Additionally, it regulates the expression of both genes involved in bacterial adhesion and virulence factors in other bacteria (Hanawa et al., 2002;Gomide et al., 2018). These observations are not conserved in the other genomes analyzed in this study.

DISCUSSION
Although several databases on sRNAs and GRNs exist, the integration of these regulatory networks is still a missing point in deciphering gene expression. Several studies have shown the interplay between TFs and sRNAs when regulating gene expression by forming regulatory circuits, as reviewed by Beisel and Storz (2010); Nitzan et al. (2017), Brosse and Guillier (2018). Furthermore, consistency assessments in E. coli (Larsen et al., 2019) and C. glutamicum (Parise et al., 2021) showed that regulation driven by transcription factors is not able to satisfactorily explain gene expression and suggested other layers of regulation to be integrated into the networks in order to model the complexity of gene expression. Our work contributes to expanding the regulatory landscape of two biotechnological and four pathogenic Corynebacterium species by predicting their sRNA regulatory networks and by integrating them into the corresponding GRNs.
Regarding sRNA prediction, we searched for (i) sRNA homologous of the experimentally validated ones from Mentz et al. (2013) using GLASSgo (Lott et al., 2018) and (ii) novel sRNAs belonging to known sRNA families from Rfam (Kalvari et al., 2021) using cmsearch (Nawrocki and Eddy, 2013). The former uses iterative blast search, pairwise identity filtering and graph-based clustering based on secondary structures to find sRNA homologous (Lott et al., 2018). It allows us to search for homologous sRNAs not belonging to a specific sRNA family. Meanwhile, cmsearch allows us to use covariance models to search for novel members of curated sRNA families from Rfam. Cmsearch has been considered the most specific and sensitive sRNA homology tool (Freyhult et al., 2007;Lott et al., 2018) and GLASSgo presented results comparable to cmsearch in a recent benchmark (Lott et al., 2018). RNAz and RNAdetect identify functional sRNA candidates amongst the ones predicted by GLASSgo and cmsearch, yielding strong candidates for further investigation as well as target prediction (Gruber et al., 2010;Backofen et al., 2018;Chen et al., 2019). Regarding the sRNA target prediction, CopraRNA is currently considered the best bacterial sRNA-mRNA interaction prediction software (Pain et al., 2015;Georg et al., 2020). It constructs a combined prediction based on the conservation of sRNA interactions across a given set of organisms, which significantly decreases the false positive rate (Wright et al., 2013;Backofen et al., 2018). In order to maximize the reliability of our regulatory interactions, we selected the most dissimilar sRNA homologs from the same genus and from the same species predicted by GLASSgo with more than 80% of similarity for the sRNA interaction prediction with CopraRNA (Wright et al., 2013). This procedure increases the chances of our regulatory interactions to be true because they will be conserved on a genus-or species-level. This, along with the filtering of the fifteen best-ranked CopraRNA predictions with p-value < 0.01 makes our conservative predictions yielding strong candidates for hypothesis generation and future experimental assay design. Even though these predicted regulatory interactions can either activate or repress the mRNA expression, we provide no functional annotation for them.
By applying our GRN sRNA-enrichment pipeline, we identified TFs, sRNAs and sigma factors jointly forming regulatory circuits in the regulatory networks. We were able to identify feed forward loops, single input modules and dense overlapping regulons. With no information on TFs regulating sRNAs, feedback loops were not possible to be identified for these networks. Furthermore, we presented the occurrences in which the co-regulation by sRNAs and TFs were also observed in other studied organisms. We highlighted genes in regulatory circuits involved in the following pathways: methionine biosynthesis and metabolism of cofactors and vitamins in C. jeikeium; TCA cycle and carbohydrate metabolism in C. diphtheriae; and TCA cycle, phosphate metabolism and cell division in C. glutamicum.
In our gene ontology analysis, ATP-binding is the molecular process with the most amount of genes potentially regulated by sRNAs in all studied organisms. This is not surprising, given the immense importance of ATP for the survival, growth and replication of all living organisms. In bacteria, ATP is associated with virulence factors and can even regulate virulence genes, e.g., the mgtC gene in Salmonella (Klein and Lewinson, 2011;Lee and Groisman, 2012;Mempin et al., 2013). Besides that, the other molecular processes with which most genes are associated are DNA binding and Metal ion binding, showing a probable strong influence of sRNA in these molecular functions. In C. diphtheria NCTC 13129, the sRNA Cd-NCTC13129-sRNA-2 potentially regulates the transcription factor ydfH, which regulates the global regulator glxR. Additionally, it is the regulator with the largest amount of regulations known in the Corynebacterium species. Likewise, in C. jeikeium, the sRNA scjk1464.1 regulates the transcription factors dtxR and mcbR. DtxR is the master regulator of iron metabolism in C. glutamicum (Wennerhold and Bott, 2006;Schröder and Tauch, 2010) and the TetR family regulator mcbR is involved in biofilm formation in E. coli (Zhang et al., 2008). Note that in C. glutamicum cg0350 (glxR homolog) is potentially regulated by the sRNA cgb_1195 and forms a feed forward loop together with this sRNA, sdhCD, and acn.
Amongst the genes potentially regulated by sRNAs, note the virulence factor fagC in C. pseudotuberculosis, the candidate virulence factor rpfI in C. ulcerans and the potential drug target rpsH in C. diphtheriae. We also observed the heat shock protein GroEL and the histidine kinase TcsS4 being regulated by sRNAs in C. jeikeium. While heat shock proteins are essential for bacterial survival and were recently associated with virulence and drug resistance (Neckers and Tatu, 2008), two-component systems are known as regulators of virulence factors and genes related to adhesion, pilus formation and drug resistance (López-Gońi et al., 2002;Matsushita and Janda, 2002;Tiwari et al., 2014). Moreover, the genes related to survival and adaptation in the nif operon and in the suf gene cluster (Stock et al., 1989;Huet et al., 2005) are regulated by the same sRNA and transcription factor in C. jeikeium. Genes of biotechnological interest, such as pfkA in C. pseudotuberculosis, rhtC in C. jeikeium, and pyk in C. diphtheriae, were also pointed out as sRNA targets. These genes are associated with L-threonine production, L-serine fermentation and lactic acid production in C. glutamicum, respectively. These molecules are largely used in the food industry (Diesveld et al., 2009;Chai et al., 2016;Zhang X. et al., 2017). The presented regulations show the potential of sRNAs to regulate genes of medical, veterinary and biotechnological interest in corynebacterial species.

CONCLUSION
We introduce the sRNA regulatory networks integrated with the transcriptional gene regulatory networks of C. glutamicum, C. pseudotuberculosis, C. ulcerans, C. diphtheriae, C. jeikeium, and C. efficiens. This integration allowed us to identify sRNAs and TFs forming generalizable patterns, such as feed forward loops, dense overlapping regulons and single-input modules. It indicates sRNAs and TFs jointly orchestrating the regulation of corynebacterial gene expression, suggesting that sRNAs may have a great impact in modeling the gene expression of important biological processes in corynebacteria. Our results suggest several genes for further experimental investigation in the studied organisms. Amongst them, note the potential regulation of mraZ, which is conserved in four organisms of this study, and of the virulence factor fagC, which is potentially regulated by dtxR and one sRNA in both C. pseudotuberculosis and C. ulcerans. We believe that with CoryneRegNet 7.5, in which we implemented the integrated networks with extended visualization and querying functionality, we move an additional step toward understanding the corynebacterial regulatory mechanisms and provide new starting points to guide future experimental assays to comprehend the regulatory mechanisms underlying pathogenicity, survival, adaptation and amino acid production in the Corynebacterium genus.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.exbio. wzw.tum.de/coryneregnet/processToDownalod.htm.

AUTHOR CONTRIBUTIONS
MP, MR, RB, VA, and JB conceptualized this work. MP and DP developed the software and wrote the manuscript. MP performed the analysis. VA, RK, and JB supervised the work. MR, RB, RK, FA, AP, VA, and JB reviewed the manuscript. All authors contributed to the article and approved the submitted version.