Identification of Hfq-binding RNAs in Caulobacter crescentus

ABSTRACT

Small RNAs are important for post-transcriptional regulation of gene expression, affecting stability and activity of their target mRNAs. The bacterial Sm-like protein Hfq is required to promote pairing between both RNAs when their sequence complementarity is limited. To provide a first global view on the post-transcriptional landscape of the α-proteobacterium Caulobacter crescentus, we have identified the Hfq-binding RNAs employing High-throughput sequencing of RNA isolated by cross-linking immunoprecipitation (HITS-CLIP). A total of 261 RNAs, including 3 unannotated RNAs, were successfully identified and classified according to putative function. Moreover, possible interactions between the identified sRNAs with mRNA targets were postulated through computational target predictions.

Introduction

Regulation of gene expression is essential to bacterial survival, especially for those living in dynamic environments. Besides transcription, RNA molecules are further regulated at the post-transcriptional level, and this process has been extensively investigated for the past few years. Small RNAs (sRNAs) have been revealed as the main post-transcriptional regulators of gene expression, affecting the stability and activity of their target RNAs. Several studies have unveiled a large number of sRNAs in bacterial genomes using transcriptomic approaches, and their involvement in controlling critical cellular processes, such as metabolism, quorum sensing, virulence and stress responses, reviewed in [1,2]. Bacterial sRNAs are heterogeneous in structure and size, and most of them are trans-encoded, exhibiting partial sequence complementarity with their targets, requiring the RNA chaperone Hfq to mediate this interaction [3].

Sequence and structural analyses showed that Hfq harbors a significant similarity with Sm and Sm-like proteins, identified in Archaea and Eukaryotes [4]. Hfq has three RNA binding sites and, after its oligomerization in a homo-hexameric ring-shaped protein, two main RNA binding regions are formed: the proximal and distal faces, leading to binding of U-rich sequences (like sRNAs) to the former, and poly(A) sequences (like mRNAs) to the latter [5]. Hfq selectivity for sRNAs can be explained by recognition of their U-rich 3ʹ end containing a Rho-independent transcription terminator, which presents a longer poly(U) tail than Hfq-independent sRNAs [6,7]. Also, lateral binding sites were shown to be important for sRNA accommodation and stabilization of interaction with the target mRNA [8]. This interaction can lead to alterations on both stability or translation rate of the target mRNA, depending on the region recognized by the sRNA. It can block the ribosome binding site (RBS) [9]; or expose it by unwinding or preventing the formation of an inhibitory hairpin that would mask the RBS [10]. Likewise, mRNAs may be stabilized by blockage of ribonuclease cleavage sites [11] or destabilized in a RNase E-dependent manner, since Hfq can interact directly with this RNA degradosome protein [12].

The inactivation of the hfq gene results in pleiotropic phenotypes relatively conserved in many bacterial taxa: reduced growth rate or morphology alteration [13], lower tolerance to stress [14,15] and reduced virulence in cases of pathogenic bacteria [16,17]. These studies of hfq⁻ strains showed its importance in adaptation to environmental changes and expression of genes related to general stress response.

The α-proteobacterium Caulobacter crescentus is a free-living oligotrophic bacterium with asymmetric cell division, with production of a sessile (stalked) cell and a motile (swarmer) cell [18]. Being well adapted to live in oligotrophic environments and to cope with different stresses, it requires an efficient control of gene expression in response to external signals throughout the cell cycle [19]. To fully understand the regulatory responses of this bacterium, the characterization of different layers of gene expression is required, including post-transcriptional events. Identification of non-coding RNAs (ncRNAs) in C. crescentus is still recent: 199 ncRNAs were already reported [19–21], but little is known about Hfq-mediated regulation.

In a previous work from our group, hfq was shown to belong to the SpdR regulon, along with genes related to RNA metabolism and regulation of transcription and translation [22]. The first characterization of genes regulated by Hfq in C. crescentus was recently obtained with transcriptomic and metabolomic analyses of a Δhfq mutant strain [23], showing its importance in metabolism regulation and how its absence affects cell growth and shape.

Here, we describe for the first time the identification of in vivo Hfq-bound transcripts in C. crescentus. By employing High-throughput sequencing of RNA isolated by crosslinking immunoprecipitation (HITS-CLIP) and bioinformatical methods, we were able to identify and categorize putative RNAs that bind to Hfq and predict some sRNA-mRNA interactions.

Results

Consistent with continuously increasing examples of sRNA-mediated regulation of gene expression in several bacteria and establishment of the importance of Hfq in mediating this regulation in C. crescentus [23], we aimed to determine the sRNAs and putative partner mRNAs that interact with Hfq and find indications of how regulatory networks are coordinated in this important model organism.

In order to identify the RNAs that were bound by Hfq in C. crescentus, we initially constructed a strain carrying a 3xFLAG epitope tag linked to the amino-terminal of Hfq, and this strain presented a similar phenotype relative to the wild-type concerning growth rate and morphology (Fig. S1). This allowed the use of co-immunoprecipitation with an anti-FLAG antibody linked to a matrix as a strategy to isolate RNAs that were crosslinked in vivo to Hfq at exponential phase growth in PYE. The same background strain NA1000 was used as a negative control, to eliminate those RNAs that could bind non-specifically to the anti-FLAG resin. Sequencing data were analyzed to identify peaks of reads that were significantly enriched in the FLAG-Hfq strain. The peaks were mapped to the annotated genome and we could identify peaks corresponding to 258 genes as differentially bound to FLAG-Hfq, which comprise 241 annotated ORFs and 17 ncRNAs (5 tRNAs, 3 rRNAs and 9 sRNAs). Among the mRNAs, 214 had peaks within the coding region, and 27 had peaks in the 5ʹ or 3ʹ untranslated regions (Supplementary Table S1). Interestingly, a hundred and forty-four of the genes found in our analysis belong to previously identified operons [21].

On the other hand, for seven peaks we were not able to assign a corresponding gene in C. crescentus. Four of them were located in the coding region or 3ʹ UTR of two overlapping genes and since the library was not strand-specific we could not assign it to one gene. This was also a limitation for the characterization of 3 additional peaks that were located in intergenic regions outside the boundaries of 200 nucleotides upstream or downstream established for UTRs, as shown in Fig. 1. These three peaks (peak 48, 49 and 127) are 454, 179 and 206 nt long respectively, and seem to correspond to unannotated regions in the NA1000 strain genome. Although we could not assign putative ORFs for peak 48, peak 49 corresponds to an ORF annotated in C. crescentus (locus WP_096053988.1) that encodes a putative protein of 64 amino acids and peak 127 corresponds to part of an ORF annotated in C. crescentus (locus WP_010919371) that encodes a putative protein of 226 amino acids. These results show that there are transcripts in the regions of these two newly annotated ORFs, and that they bind to Hfq.

Figure 1.

Schematic view of peaks with no corresponding ORF in C. crescentus NA1000. Each panel represents the regions on the C. crescentus genome where significant enrichments of reads (peaks) were identified in the FLAG-Hfq strain (indicated FLAG-Hfq, blue) as compared to the parental NA1000 strain (indicated Hfq, red). The numbers refer to the genomic coordinates of the peaks in C. crescentus NA1000 genome. Below each panel, yellow arrows represent annotated CDSs, with the respective CCNA gene number. White arrows indicate putative ORFs.

Moreover, a further analysis of the identified Hfq-bound RNAs was obtained by comparing our data to the transcriptome analysis of the C. crescentus hfq deletion mutant [23]. We found that thirty-two of the Hfq-bound RNAs (30 mRNAs and 2 sRNAs) were also found as differentially expressed by a fold-change ≥ 2 (p-value ≤0.01) in the Δhfq strain (Table S1). This suggests that these transcripts could be subject to direct post-transcriptional regulation by a Hfq-dependent sRNA, leading to stability alterations.

To better understand the biological context of the identified set of genes, a functional categorization was performed and groups based on functional similarity were created (Supplementary Table S2). Functional categories, as well as the number of genes to which a specific function was assigned, are shown in Fig. 2. Categories such as hypothetical proteins, translation, transporters, oxidative phosphorylation and citrate cycle are the most representative.

Figure 2.

Functional categorization of the Hfq-bound RNAs. The functional categories were annotated according to the KEGG database [42,43], based on DAVID annotation [44,45] with manual curation. The numbers indicate the number of genes in each category.

Regarding transcripts’ interaction with Hfq, an in silico target prediction for the co-immunoprecipitated sRNAs found in our analysis was performed using the IntaRNA algorithm, considering the coding region and the 5ʹ and 3ʹ untranslated regions for each target gene. The prediction takes into consideration the secondary structure of the transcripts and the extension of the annealing region to calculate the energy of interaction. Although the interaction with Hfq may change the energy necessary for the interaction to occur, we used the calculated energy of interaction as a scoring parameter.

Eighty-eight of the Hfq-bound mRNAs were predicted as targets of at least one Hfq-bound sRNA, suggesting that they were bound together to Hfq. In 7 cases, the interaction site was located in the intergenic region of an operon. The target mRNAs predicted for each sRNA are shown in Supplementary Table S3 with the boundaries of the interaction sites indicated for both the sRNA and mRNA. To illustrate the results obtained with this analysis, the pairing prediction of two of the Hfq-bound sRNAs, CCNA_R0032 and CCNA_R0088, is shown in Fig. 3. The 66 nt-long CCNA_R0032 is predicted to interact with its target mRNAs by a single region located in the center of the sRNA (nt 17–30, Fig. 3A), while the 140 nt-long CCNA_R0088 is predicted to interact via two independent regions (37–66 and 82–104, Fig. 3C). Interestingly, each region of CCNA_R0088 is predicted to bind to a different subset of mRNAs, and region 37–66 is predicted to be part of a stem-loop structure in the sRNA, which would have to be opened in order to permit interaction with the targets.

Figure 3.

Putative interactions of sRNAs and mRNAs. In silico prediction of sRNA-mRNA interaction was carried out with the IntaRNA package [46]. The secondary structures of the sRNAs were predicted with the RNAfold program, under default settings [48]. a, Prediction of the interaction of CCNA_R0032 with other RNAs in C. crescentus NA1000. The lines correspond to 25 individual RNAs with the best scores, and the blue lines indicate the regions of predicted base pairing. The cartoon shows the predicted secondary structure of CCNA_R0032, and the region of interaction with other RNAs is indicated in red. b, Interaction of CCNA_R0032 with CCNA_03922 mRNA (also identified as Hfq-bound in this work). The cartoon in the left panel shows how the putative pairing occurs, and the pairing sequences of both transcripts are shown in more detail in the right panel. c, Prediction of the interaction of CCNA_R0088 with other RNAs in C. crescentus NA1000. The lines correspond to 25 individual RNAs with the best scores, and the blue lines indicate the regions of predicted base pairing. The cartoon shows the predicted secondary structure of CCNA_R0088, and the two regions of interaction with other RNAs are indicated in blue and red, respectively. d, Interaction of CCNA_R0088 with CCNA_02816 (also identified as Hfq-bound in this work) and CCNA_02822 mRNAs. The cartoons in the left panel show how the putative pairing in each region occurs, and the pairing sequences of the transcripts are shown in more detail in the right panel.

Discussion

Hfq-dependent RNAs have been highly explored in many bacterial species. Lately, techniques have been improved seeking a more specific, accurate and large-scale identification of these RNAs. There are examples employing only co-immunoprecipitation (RIP-seq) or, more recently, in vivo crosslinking (CLIP-seq or HITS-CLIP) followed by deep sequencing [24,25]; also, new methods like GRIL-seq (Global small non-coding RNA target Identification by Ligation and sequencing) and RIL-seq (RNA interaction by ligation and sequencing) allow association of the sRNA and its target-mRNAs by in vivo ligation of both transcripts and sequencing [26,27]. Thus, these constantly evolving approaches have been successfully used to characterize RNAs that interact with Hfq in different organisms.

When characterizing the peaks found in our analyses, stipulation of untranslated regions was important in cases where the highest read enrichment occurred outside the transcript’s CDS. Long 5ʹUTRs (>100 nt) might be important in mRNA translation regulation [21], therefore, we used a limit of 200 nt upstream and adapted this comparison using the TSS when annotated. A differential expression of CDSs within the transcript has also been described in operons, probably generated by diverse mechanisms such as transcriptional polarity [21] and presence of internal promoters, leading to an independent expression of some of the encoded genes [19]. This was also observed in our data: read enrichment presented such variation within each operon that peak calling was irregular and a variable number of genes was identified in each case. Besides the known factors, this differential expression of mRNA fragments could also be a result of RNA degradation that occurs in vivo by ribonucleases as an outcome of the interaction with Hfq.

The hfq mutant strain transcriptomic analysis [23] allowed us to predict the effect of the interaction of Hfq with the co-immunoprecipitated RNAs concerning their stability. The majority of RNAs were not differentially expressed in the hfq mutant but they could still be subject to post-transcriptional regulation by, for instance, modulation of their translation efficiency. Interaction of Hfq with the sigma factor rpoS mRNA, for example, leads to its efficient translation rather than stabilization as demonstrated in S. typhimurium [28] and E. coli [29]; in the latter, translation is positively or negatively regulated by four sRNAs, which are otherwise stabilized or destabilized after binding to Hfq in a RNase E-dependent manner [30].

Also, in the work on the C. crescentus hfq mutant strain, the authors proposed a model of metabolite-dependent regulation of growth and cell shape that relies on maintenance of the levels of some central metabolites, and Hfq was shown to be important in this process [23]. Hence, alterations in the mRNA levels of metabolic genes were observed in the absence of Hfq, leading to morphological defects. Interestingly, one of these genes, identified as vor (CCNA_03280), was bound to Hfq in our analysis, supporting that its regulation is Hfq-dependent. Also, since it presented a 2-fold increase in the Δhfq transcriptome, we can assume that vor is destabilized after direct interaction with Hfq, probably mediated by a sRNA. As assessed on the analyses from Irnov and coworkers [23], the main finding to understanding the growth and cell shape deficiency in the absence of Hfq relies on a crosstalk between the TCA cycle and cell wall biosynthesis, and the imbalance caused in this process by hfq deletion. An example of these metabolic perturbations is the accumulation of α-ketoglutarate, an intermediate metabolite of the TCA (tricarboxylic acid) cycle, by alterations in CoA metabolism. Although the mechanism through which the TCA cycle is affected by Hfq is yet to be elucidated, we could observe many citrate cycle genes in our analysis, suggesting that, besides metabolite-dependent regulation, a direct mRNA translation regulation by Hfq is probable.

In the C. crescentus Δhfq strain transcriptome analysis, COG categories were assigned to the differentially expressed genes and, although a large number of genes had no COG assigned, there was a prevalence of categories similar to the ones obtained in our functional categorization (Fig. 2), such as carbohydrate, amino acid and lipid metabolism, energy production, and proteins involved in translation and transcription [23]. This shows that in C. crescentus Hfq is important for the regulation of genes involved in diverse basal cellular functions.

Melamed and coworkers [27] have used a similar approach, RIL-seq, in order to investigate the RNA pairs that bind to E. coli Hfq. After evaluation of the co-immunoprecipitated transcripts, a prevalence of target-mRNAs encoding proteins involved in ribosome biogenesis, transportation, motility, transcriptional regulation, carbohydrate metabolism and oxidative phosphorylation was observed, similar to the main functional categories found for C. crescentus.

Target prediction for Hfq-bound sRNAs allowed an initial characterization of these post-transcriptional regulators. Considering that most of the ncRNAs identified in C. crescentus so far are not conserved outside of the Caulobacteraceae family [21] and that there is not enough information in the literature to validate and expand our findings, we postulate pairing for two sRNAs based on their secondary structures and interaction sites predicted by IntaRNA (Fig. 3). The predicted binding site in CCNA_R0032 is located in the central portion of the sRNA, which is a single-stranded region preceding a possible Rho transcription terminator. In CCNA_R0088, besides a binding site predicted also in a single-stranded part of the sRNA, another one is predicted as part of a hairpin loop. Considering the chaperone role of Hfq, the opening of the stem-loop structure in CCNA_R0088 could be facilitated by interaction with Hfq, as has been already demonstrated for sRNAs such as RprA and OxyS in E. coli [30,31]. Changes in the secondary structure of RprA in the presence of Hfq appear to expose the interaction site with rpoS mRNA. These alterations were shown to stabilize the sRNA, preventing degradation by blockage of the RNase E cleavage site by Hfq itself or by the acquired conformation, and the interaction with the target mRNA leads to its up-regulation [30]. Moreover, just like we observed for the CCNA_R0088 sRNA, two binding sites were described in the E. coli FnrS sRNA and each one interacts with a different set of target mRNAs [32].

Lately, the ability of Hfq to regulate mRNAs by binding independently to these transcripts has been investigated. It has been demonstrated that Hfq can act as an RNA chaperone and bind to these molecules in the absence of regulatory small RNAs, and, so far data suggests it results in translation repression (reviewed in [33]). This could explain why not all the co-immunoprecipitated mRNAs were predicted as targets for the Hfq-bound sRNAs. Another hypothesis is that our set of Hfq-bound sRNAs is incomplete, either because some sRNAs are too unstable and we could not capture them using the HITS-CLIP method, or because the sRNAs responsible for regulating some of these mRNAs are induced only in specific conditions. Also, the energy of interaction calculated by the IntaRNA algorithm does not necessarily correspond to what happens when Hfq is interacting with both transcripts and altering their conformation; therefore, we could have missed possible sRNA-mRNA duplexes that were predicted with low interaction energy.

Finally, other RNAs that were not identified as Hfq-bound by our analyses could still be regulated by Hfq, since our analysis was limited to a specific growth condition, in order to provide a first global view on the post-transcriptional landscape of this model organism. As we start unraveling the role of Hfq in C. crescentus, its involvement and action mechanism in the response to environmental changes like stationary phase entry and stress conditions, as already described in other organisms, remains unexplored. Further characterization regarding in vivo regulatory mechanisms is needed, for instance, analyses of sRNA-mRNA interaction with Hfq, determination of Hfq binding properties, and whether there are important residues for RNA:Hfq interaction on proximal/distal faces and rim, to allow the classification of C. crescentus sRNAs into class I or class II, as defined by [34] in E. coli. Still, identification of the RNAs that interact with Hfq is of crucial importance to fully understand the extent of its influence in cellular processes.

Materials and methods

Caulobacter crescentus strains were grown in PYE broth [35] at 30°C with agitation. The synchronizable strain C. crescentus NA1000 [36] was used as the wild type parental control. The strain encoding 3xflag-hfq was constructed on the NA1000 background as follows. DNA fragment A was produced by overlap-PCR using PCR products generated with primer sequences AAATCTAAGCTTGGTCCTTCCCTCGGTG with CACCGTCATGGTCTTTGTAGTCCATACGGGGGTTCCCCTCTC and GAGAGGGGAACCCCCGTATGGACTACAAAGACCATGACGGTG with TTTCTTGAATTCTTTATCGTCGTCATCTTTGTAGTCG, respectively. PCR amplification using the primer sequences AAGAAAGAATTCATGTCCGCCGAAAAGAAGCAAAATC with TTTCTTGCTAGCCATAGTCGAGCCTGGCCAG generated DNA fragment B. Triple ligation of the HindIII and EcoRI digested product A, the EcoRI and NheI digested product B and the HindIII and NheI digested vector pNPTS138 resulted in the plasmid pIH90, which was used for the chromosomal exchange of hfq for 3xflag-hfq by double recombination.

The method of HITS-CLIP was performed essentially as described by [37]. C. crescentus strains NA1000 and UJ10248 (carrying an epitope tagged FLAG-Hfq) were grown in 800 ml of PYE medium up to midlog phase at 30ºC with agitation. At this time, formaldehyde (0.5% final concentration) was added to initiate crosslinking and the cultures were incubated at 30°C for 10 min. Crosslinking reactions were quenched by the addition of glycine (pH 7, 0.125 M final concentration). After 5 min at room temperature, cells were harvested by centrifugation (5000 x g, 10 min), washed twice with lysis buffer (20 mM Tris-HCl pH 7.5, 200mM NaCl), and resuspended in 20 ml cold RIPA buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NONIDET, 1% SDS) containing one SigmaFAST^TM Protease Inhibitor Cocktail Tablet, EDTA Free. Cells sonication was performed with a Branson Digital Sonifier in a total of 6 cycles of 15 seconds on/15 seconds off at an output amplitude of 30%, and unbroken cells and debris were removed by centrifugation (16,900 x g, 10 min, 4°C). The Anti-FLAG® M2 Affinity gel (Sigma) (0.15 ml) was washed 3 times with RIPA buffer, followed by incubation with the supernatant for 2h at 4°C, and the precipitated complexes were separated by centrifugation (200 x g, 2 min, 4°C). Crosslinking reversion was executed in 2 ml decrosslinking buffer for 40 min at 70°C, RNA was extracted with TRIZOL (Life Technologies) and the phases were separated by centrifugation (6,738 x g, 10 min, 4°C). The aqueous phase was transferred to new tubes and 0.02 ml of glycogen (15 mg/mL) and 0.02 ml of isopropanol were added. The RNA was precipitated at −80°C for 16h, washed with 80% ethanol and centrifuged. The purified RNAs were resuspended with 0.016 ml DEPC water, concentrated in an Eppendorf Concentrator Plus and pooled in order to reach 1 μg of RNA. Sample pools were digested with DNase I and quantified in a Nanodrop (Thermo Fisher Scientific).

Three RNA samples from the FLAG-Hfq strain (each one constituted by a pool of three independent biological replicas) and two RNA samples from the NA1000 control (from two independent biological replicas) were submitted to high throughput sequencing. The cDNA libraries for each sample were prepared from 1 µg of RNA with the TruSeq RNA Sample preparation V2 kit (Illumina), and the integrity and average size of the cDNAs libraries were verified with the DNA 1000 kit (Bioanalyzer). The libraries were normalized to 4 nM with the Kapa Library Quantification system (Kapa Biosystems) to be sequenced in the MiSeq platform with the MiSeq Reagent V3 system (150 cycles) (Illumina).

Raw input from sequencing was trimmed of adaptor sequences and quality filtered with Trimmomatic 0.36 [38], then the processed reads were aligned with Bowtie 1.2 [39] against C. crescentus NA1000 genome (NC_011916.1). In order to proceed with the peak calling both aligned samples were compressed to binary form with SAMtools 1.3.1 [40]. The MACS2 program [41] was used to determine the peaks of reads for each sequencing, with the default parameters (callpeak), adapting only for C. crescentus genome size (4.04 Mb), as a matter of model construction, as for the band width matching the library average fragment size wet experiment (150 nt). Processing of MACS2 output was determined by in-house perl scripts, and the workflow is as follows. The position of each peak was verified to the annotated NA1000 genome deposited in the NCBI database. In cases when the peak was located outside of an annotated ORF, the untranslated regions were considered within 200 nt upstream or downstream of each ORF. Also, when located in the 5ʹ UTR, the annotated transcription start site (TSS) was considered [19], if available for that gene, and compared with the summits output. This allowed scoring each peak according to the smallest distance from the ORFs boundaries. In addition, the identification of operons done previously [21] was considered for the analysis, to access polar effects on ORF’s targeted by MACS2.

The functional categories were annotated according to the KEGG database [42,43], based on DAVID annotation [44,45] with manual curation. The prediction of sRNA-mRNA interaction was carried out with IntaRNA [46], considering the C. crescentus optimal physiological temperature of 30°C for energy calculations. We used as input gene sequences within 200 nt upstream and downstream the coding region and the output was examined in order to eliminate interactions with adjacent genes. The predictions were confirmed within the range obtained with RIsearch2 [47]. The identified pairs with predicted interaction energy lower than −12 kcal/mol at 30ºC were considered as possible. The secondary structure of the sRNAs was predicted with the RNAfold program, under default settings [48].

The complete RNAseq data set has been deposited in the NCBI´s Sequence Read Archive (SRA) (https://www.ncbi.nlm.nih.gov/sra/) with accession ID: SRP154464.

Funding Statement

This work was supported by São Paulo Research Foundation [FAPESP, grant 2014/04046–8] and Conselho Nacional de Desenvolvimento Científico e Tecnológico [CNPq-Brasil, grants 306558/2013–0 and 307974/2017–0]. During the course of this work, NGA was supported by Undergraduate fellowship from FAPESP [grant 2015/21025–7]; AMV and RAR were supported by Masters fellowship  from FAPESP [grants 2014/13552–4 and 2016/06378–3, respectively]; LGS by a postdoctoral fellowship [grant 2015/1461678] from Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) and MVM was partially supported by CNPq-Brasil.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed here.