Clostridium difficile Transcriptome Analysis Using Pig Ligated Loop Model Reveals Modulation of Pathways Not Modulated In Vitro

Abstract

A pig ligated loop model was used to analyze the in vivo transcriptome response of Clostridium difficile. Bacterial RNA from the loops was retrieved at different times and was used for microarray analysis. Several virulence-associated genes and genes involved in sporulation cascade were differentially expressed (DE). In concordance with observed upregulation of toxin genes in microarray, enzyme-linked immunosorbent assay estimation of total toxin showed high amounts of toxin in the loops. Several genes that were absent in primary annotation of C. difficile 630 but annotated in a secondary annotation were found to be DE. Pathway comparison of DE genes in vitro and in vivo showed that when several pathways were expressed in all conditions, several of the C. difficile pathways were uniquely expressed only in vivo. The pathways observed to be modulated only in this study could be targets of new therapeutic agents against C. difficile infection.

Clostridium difficile, the causative agent of Clostridium difficile–associated disease (CDAD) is a Gram-positive spore-forming anaerobic bacterium with a broad host range [1, 2]. A conservative estimate of the hospital cost associated with this disease in the United States is $3.2 billion annually [3]. A recent increase in CDAD has been attributed to the emergence of a new strain (NAP1/BI) with increased rate of toxin production, sporulation, and antibiotic resistance [4–6]. Comparative genomic analysis of several C. difficile isolates from Europe and North America has revealed very high genome variability in this species, and this genome plasticity has been shown to contribute to the increased virulence [7–10]. In vitro transcriptomic studies using microarrays have been used to delineate C. difficile genes involved in pathogenesis [11]. Recently, we identified a set of C. difficile genes that are significantly modulated during Caco2 cell infection [12]. A better understanding of the genes involved in C. difficile infection (CDI) can aid in the design of new therapeutic agents and better CDI management strategies. In this study, we used a pig ligated loop model to provide a better definition of C. difficile genes involved in CDI and identified cellular pathways that are important in pathogenesis.

MATERIALS AND METHODS

Bacterial Culture

C. difficile 630 strain was grown to mid-log phase as described elsewhere [12]. To remove the secreted toxin from culture, cells were centrifuged at 3000g for 5 min and were resuspended in fresh brain heart infusion medium, and this was repeated twice. All transfers were performed in a Bactron IV anaerobic chamber (Shell Lab).

Establishment of Pig Ligated Loops

All animal experiments were performed with the approval of the Institutional Animal Care and Use Committee. Pigs aged ca. 8 weeks were obtained from Cornell University swine farm and were fed with weaner pellets and water ad libitum. To perturb the natural microbial flora, starting at 6 days after surgery, pigs were administered 50 mg/kg body weight of erythromycin and neomycin thrice daily for 3 days, followed by 3 days of no antibiotic treatment. Pigs were confirmed to be culture negative for C. difficile before surgery by culturing rectal swab specimens in C. difficile selective agar (Becton Dickinson) at 37°C in an anaerobic chamber. Before surgery, pigs were fasted overnight. General anesthesia was induced and maintained with isoflurane. The pigs were positioned in dorsal recumbency, and a 7-cm ventral midline celiotomy was performed. With use of monofilament suture, three 10-cm loops were established with 3-cm spacers between each loop. A total of 1 × 10⁷ C. difficile 630 cells were injected in each loop with use of a 21-gauge needle and syringe. Pigs were kept anesthetized for the duration of the experiment and were sacrificed at 4, 8, or 12 h. Four experimental pigs were used for each time, and pigs injected with sterile brain heart infusion medium served as controls. After the appropriate time loops were excised, 1 mL of the intraluminal fluid was retrieved and the total amount of toxin was estimated using Premier Toxin A&B enzyme-linked immunosorbent assay (ELISA) kit according to the manufacture's protocol (Meridian Bioscience). The reminder of the loop fluid was transferred to twice the volume of RNA protect bacteria reagent (Qiagen) and was used for extraction of bacterial RNA.

Bacterial RNA Extraction and cDNA Synthesis

After pooling the loop contents from the same animal, to eliminate the possibility of host cell and debris contamination, isopycnic percoll gradient centrifugation was performed. In brief, the loop contents were centrifuged at 200 × g for 10 min at 4°C. The supernatant was collected and subjected to further centrifugation at 12,000 × g for 10 min at 4°C. A standard isoosmotic Percoll stock solution was prepared by diluting 90 mL of undiluted Percoll (Pharmacia) with 10 mL of 1.5 M phosphate-buffered saline (PBS). A 60% (vol/vol) solution was made by combining 60 mL from the standard isosmotic Percoll solution with 40 mL of 0.15 M PBS. The pellets were resuspended with PBS and were laid over 10 mL of 60% percoll in PBS. The mixtures were then centrifuged at 12,000 × g for 45 min at 4°C. The upper fraction was discarded while the bacterial fractions were collected, washed with PBS, and subjected to centrifugation at 40,000 × g for 1 h at 4°C. In accordance with the manufacturer's protocol, C. difficile RNA was extracted using RNeasy kit (Qiagen). RNA was then treated with 20 units of Baseline-ZERO DNase (Epicenter Biotechnologies), and was cleaned using RNeasy MinElute Cleanup Kit (Qiagen). Three micorgrams of RNA was used for cDNA synthesis or quantitative reverse-trascription polymerase chain reaction (qRT-PCR), as described elsewhere [12].

cDNA Labeling and Array Hybridization

cDNA synthesized from loop RNA was labeled with Cy3/Cy5, as previously elsewhere [12]. A microarray (GEO platform ID#GPL9481) representing all genes in C. difficile 630 according to annotation provided by the J. Craig Venter Institute (JCVI) was designed using Agilent eArray expression algorithm (https://earray.chem.agilent.com/earray/). In accordance with the manufacturer's protocols, equal amount of labeled cDNA from log phase C. difficile 630 was used as control channel to perform competitive array hybridizations against labeled cDNA from loop RNA from different times (Agilent Technologies).

Microarray Data Analysis and qRT-PCR

Arrays were scanned using an Agilent G2565BA microarray scanner, and data were extracted using Agilent's feature extraction image analysis software, version 10.2. Resulting data files were then loaded to GeneSpring GX, version 7.1 (Agilent Technologies). To allow interpretation of the data, values <0.01 were set to 0.01. The log₂ ratio of the intensity of signal channel (loop RNA) to that of control channel (log phase RNA) was calculated and was then normalized to each slide's median intensity ratio. The default interpretation mode was set to ”log of ratio.” Each time point was then tested separately for statistical differences with use of analysis of variance (ANOVA) at 2% confidence level, in conjunction with Benjamini-Hochberg multiple testing correction. From this list, genes at each time point that changed by at least 1.75-fold were selected as statistically significant. One microgram of total RNA was converted to cDNA, and differential gene expression data were validated using qRT-PCR, as described elsewhere [12]. The list of primer set used for qRT-PCR is given in Table S1.

Cellular Pathway Analysis

C. difficile cellular pathways were reconstructed from C. difficile 630 genome annotation with use of Pathway Tools software [13]. A list of significantly changing C. difficile genes in vitro and genes detected in spores and cell membrane was obtained from published results elsewhere [11, 14, 15]. For comparative analysis, these gene lists and genes changing by 1.75-fold in ligated loops were imported into Pathway Tools and were overlaid on reconstructed cellular pathways with use of Omics Viewer [16].

RESULTS

C. difficile Genome Response to Infection

Microarray analysis of C. difficile transcriptome at 4, 8, and 12 h after infection revealed differential expression of a large number of genes at all time points (Figure 1). Differentially expressed (DE) genes were spanning the whole C. difficile genome, and the maximum number of DE genes was observed at 12 h after infection. Several virulence-associated genes, such as CD2592 (fibronectin binding protein), CD2793 (S layer protein), CD1546, CD1208 (hemolysins), toxin genes, and genes involved in sporulation cascade, were up-regulated as early as 4 h after infection. The highest number of DE genes belonged to transcription, signal transduction, amino acid transportation and/or metabolism, and carbohydrate transportation and/or metabolism. In conjunction with this, several regulatory genes that were part of a 2 component system, ABC transporters, flagellin, and sigma factor, and other key genes, such as codY, were found to be upregulated. Although the primary genome annotation of C. difficile 630 contained only 3804 genes, a recent annotation by JCVI predicted 4021 genes in this genome. Most of these additional genes coded for proteins shorter than 200 amino acids and contained start codons other than ATG. Of interest, several of these genes were found to be in the DE gene list at different infection time points (Table S2). A total of 273 genes showed significant expression difference at all time points (Figure 2A). Of these, 20 genes that were annotated only in JCVI annotation showed expression difference as high as 5-fold (Figure 2B). Most of these genes were annotated as hypothetical proteins and could be candidates for further characterization. The total toxin concentration in loop fluids estimated using ELISA at 4, 8, and 12 h were 44.33 ng/mL, 109.5 ng/mL, and 146.0ng/mL, respectively (Figure 2C). This finding is in concordance with the upregulation of toxin genes observed in the microarray experiments. A complete list of genes with significant expression differences at various time points is given in Table S2.

Figure 1.

Projection of differentially expressed genes at various time points on C. difficile 630 genome. At various time points, those genes showing at least 1.75-fold change at 2% significance were selected. From outside to inside: ring 1, molecular clock of C. difficile genome; rings 2 and 3, cluster of orthologous (COG) gene categories; ring 4, DE genes at 12 h after injection; ring 5, DE genes at 8 h after injection; ring 6, DE genes at 4 h after injection. In rings 4–6, red color indicates up-regulation and green color indicates down-regulation of genes.

Figure 2.

Comparison of DE genes and toxin levels at different time points. A, From list of significantly expressing genes obtained by ANOVA at 2% significance level, total number of genes showing at least 1.75-fold change at different time points are compared as a Venn diagram. B, Fold change pattern of JCVI unique genes that were found to be modulated at all time points. C, Total toxin A/B levels in the loops at different time points was estimated using ELISA.

Modulation of Cellular Pathways

To identify the pathways that are modulated during in vivo conditions, with use of the Pathway tools software, the cellular pathways were reconstructed from the genome annotation of C. difficile 630, and the differentially expressed genes were mapped to the pathways (Table S3). The largest number of DE genes was associated with tRNA charging pathway, super pathways of amino acid biosynthesis, anaerobic respiration, acetyl-CoA fermentation, peptidoglycan biosynthesis, and tetrahydrofolate biosynthesis. The pathways modulated at different time points of infection were then compared with the cellular pathways that were modulated in vitro (Figure 3). Several of the pathways were modulated both in vitro and in vivo. For example, all pathways of cell structure biosynthesis, amino acid biosynthesis, nucleoside and nucleotide biosynthesis, glycolysis, and fermentation pathways were active in all conditions compared. However, differences were observed for pathways of fatty acid degradation; cofactor, prosthetic group, and electron carrier biosynthesis; fatty acid and lipid biosynthesis; carbohydrate degradation; and secondary metabolite biosynthesis. Among the carbohydrate degradation pathways, only glucose degradation was modulated in vitro, whereas in vivo, all pathways except glucose degradation were active. Likewise, when compared with all other conditions, components of the tricarboxylic acid (TCA) cycle were exclusively modulated in vivo. Similarly, among fatty acid degradation pathways, the fatty acid beta-oxidation I pathway was activated only in vivo. Another striking example of a uniquely modulated pathway in vivo was the pathway of phenylethanol biosynthesis (secondary metabolite biosynthesis). However, pathways of aldehyde degradation, aromatic compound degradation, and secondary metabolite degradation were found to be inactive in all conditions compared.

Figure 3.

Comparison of C. difficile pathways modulated in vitro and in vivo: DE genes in vitro and in vivo were mapped to pathways using pathway tools software, and significantly modulated pathways in each condition were compared. Pathways are color coded as follows: blue, pathways predicted in C. difficile 630 but not found expressing in any of the conditions compared; red, pathways modulated in vitro; orange, pathways modulated during Caco2 cell infection; green, pathways modulated at 4 h after injection in loops; pink, pathways modulated at 4 h after injection in loops; brown, pathways modulated at 12 h after injection in loops.

Validation of Microarray Data Using qRT-PCR

The gene expression fold change calculations obtained from microarray results were verified by performing qRT-PCR on 2 sets of 15 DE genes (See Supplementary Table S1). Although the fold change in gene expression detected in the microarray results was less than that of qRT-PCR, the overall fold change was consistent in both techniques, with a correlation coefficient (R²) of 0.859 (Figure 4).

Figure 4.

Validation of microarray data by qRT-PCR. Gene expression changes in ligated loop compared with control, measured by microarray analysis or qRT-PCR. Data are plotted as fold-change differences of microarray data compared with those of qRT-PCR.

DISCUSSION

In recent years, there has been a dramatic increase in the frequency, severity, and refractoriness of CDAD worldwide [17]. Although most of these mortalities are because of the emergence of a hypervirulent strain, the actual mechanisms by which the hypervirulent phenotype is produced is not completely understood. Despite the development of several animal models for CDI, no attempt has been made to understand the modulation of the C. difficile genome during infection [18]. To deduce important genes involved in C. difficile environmental adaptation, transcriptomic analysis of strain 630 has been performed [11]. Recently, we identified a set of genes that were DE expressed during infection of Caco2 cells with C. difficile 630 [12]. In this study, we extend this by analyzing the in vivo transcriptomic response of C. difficile with use of a pig ligated loop model. Strain C. difficile 630 was injected in ligated loops, and bacteria were retrieved at 4, 8, and 12 h after injection. RNA from isolated bacteria was used to perform microarray comparison of transcriptomic differences with bacteria grown in vitro. When genes filtered by 1.75-fold change with P > .02, we find that almost 25% of the genome was modulated after infection of the loops at least at 1 time point (Table S2). Formerly, C. difficile DE genes in vitro after different stress-like heat shock, aerobic exposure, antibiotic treatment, and pH shift have been identified [11]. When DE genes after infection of pig loops were compared with the in vitro DE genes, there was good overlap between the gene lists. In both studies, as expected, the genes involved in transcription and translation machinery were found to be modulated. Likewise, several ABC transporters, components of flagellum, and components of cell membrane were found to be overlapping. However, we found that the components of sporulation cascade and toxin production was activated even 4 h after infection but not in vitro. This finding reflects the complex nature of regulation of toxin production and sporulation cascade that cannot be mimicked by in vitro studies. We also found that virulence factors, such as slpA, fibronection-binding protein that promotes bacterial adhesion to host cell, and hemolysins (CD1546 and CD1208), are upregulated early during infection. It has been proposed that increased rate of sporulation, toxin production, and enhanced ability to colonize the host membrane could be involved in the hypervirulent phenotype of epidemic C. difficile strains [19, 20]. Therefore, the finding that genes involved in sporulation, toxin production, and cell attachment is in agreement with these reports.

The C. difficile genome annotation provided by JCVI contains 200 new genes in addition to those given by the primary annotation. Because our expression array was designed on the basis of the JCVI annotation, we examined whether the new genes identified in the JCVI annotation were expressed during infection. Most of the new genes identified in the JCVI annotation were hypothetical proteins with start codons other than ATG, and some had internal stop codons. We found that several of these genes (NT07CD1855, NT07CD2698, NT07CD0167, NT07CD3532, NT07CD0622, NT07CD0948, NT07CD1483, NT07CD2021, and NT07CD0989) were upregulated >20 times 4 h after infection. Some of these genes (NT07CD1855, NT07CD2698, NT07CD0167, and NT07CD3532) maintained high expression at additional time points. Although the magnitude of fold change was not as high, some DE genes in the JCVI unique genes might be important in CDI. For example, the gene NT07CD2220 (2.003-fold change; P <.001) appears to be a collagen-binding surface protein that may have a role in C. difficile colonization of the host. Another gene, NT07CD0642 (2.986-fold; P = .008), is an aminobenzoyl-gluatamte transporter and could be involved in nutrient harvest. Some of the genes marked as psuedogene (NT07CD0628, NT07CD0299, NT07CD0641, NT07CD0768, NT07CD1127, NT07CD1400, NT07CD1565, NT07CD2060, NT07CD2165, NT07CD2189, NT07CD2497, NT07CD2550, NT07CD3143, NT07CD3463, NT07CD3631, and NT07CD3676) in primary annotation were also found to be expressing in vivo. Although evidence for pseudogene expression in bacteria is sparse, recently, a high level of in vivo expression of several pseudogenes was found in Mycobacterium lepare [21]. Application of next-generation sequencing technologies to analyze the transcriptome has also revealed unknown complexities of bacterial transcriptome organization. When the transcriptome organization of Mycoplasma pneumoniae was analyzed using tiling arrays and direct strand specific sequencing of total RNA, 117 previously undescribed transcripts were detected [22]. In light of these reports, our finding that several of the new genes in the JCVI annotation are upregulated in vivo could be the indication of unknown complexity of C. difficile genome regulation and evolution; this finding needs to be explored further.

We further compared the modulation of C. difficile cellular pathways in vitro and in vivo. A list of significantly changing genes in vitro was obtained from transcriptomic analysis of C. difficile published by our group and elsewhere [11, 12]. Pathways of aminoacyle-tRNA charging, amino acid biosynthesis, carbohydrate biosynthesis, cell structure biosynthesis, and fermentation were modulated in all conditions and appear to be essential for cell survival and pathogenesis. Likewise, the folate biosynthesis pathway was modulated in vitro and in vivo. Although cellular requirement of folates is universal, mammals obtain folate through membrane-associated folate transport proteins [23], and most micro-organisms synthesize it de novo through the folate biosynthetic pathway [23, 24]. Thus, this pathway has been a target for antimicrobial drugs (561 Bermingham,A. 2002). A comparative genomic hybridization comparison of several C. difficile strains showed variation in the folD gene, a key component of the folate biosynthesis pathway [8]. Although the present results indicate that folate inhibitors, such as trimethoprim, might be effective against some strains of C. difficile, strains with variation in folD might be resistant to folate inhibitors. Although most components of cellular fermentation pathways were modulated in all conditions, the TCA cycle pathway was activated only in vivo. In Bacillus subtilis, in the presence of alternate sugars, such as lactose, glutamate from TCA cycle affects the sporulation rate and spore properties [25]. Because our results show that several of the genes involved in sporulation pathways are upregulated during early infection, it might be the indication that alternate sugars might be initiating C. difficile sporulation in vivo. The in vivo impact of alternate sugar metabolism in C. difficile was most striking in the case of carbohydrate degradation pathways. During all time points of infection, when glucose degradation pathways were repressed, degradation pathways of xylose, mannose, and glycogen were upregulated. As known widely, most bacteria from a wide variety of carbon sources use preferred substrates, such as glucose, through carbon catabolite repression (CCR) [26]. Therefore, the activation of alternate sugar degradation pathways of C. difficile in the loops indicates the derepression of CCR in the absence of glucose in the intestinal fluid. This observation supports the fact that derepression of CCR activates several virulence genes in pathogenic Clostridia species [27, 28]. For example, in Clostridium perfringens, glucose represses several virulence-associated processes, such as gliding motility and toxin production [28]. Repression of toxin production in C. difficile in the presence of glucose has been well established [20, 27]. As early as 1987, it was shown that alternate diets can also prevent mice mortality associated with CDI [29]. Because our results revealed the same effect, this will offer a molecular explanation for successful use of alternate diet to ameliorate CDI. These results also suggest that treatments, such as elemental diet therapy, can indeed aggravate CDI. Another interesting observation was that the phenylethanol biosynthesis (secondary metabolite biosynthesis pathway) was uniquely upregulated only in vivo; 2-phenylethanol is produced by phylogenetically diverse bacteria and is often associated with use of alternative nitrogen sources [30]. In Saccharomyces cerevisiae, phenylethanol has been found to be involved in quorum sensing [31]. However, more studies are required to ascertain whether this pathway is involved in energy metabolism or signaling in C. difficile.

In a former study, our group identified several C. difficile DE genes during Caco-2 cell infection (ex vivo condition) [12]. When DE genes in these 2 studies were compared, we found several overlaps and differences in the DE gene expression pattern. In the current study, the number of DE genes even at 4 h was much higher (17% of the genome), compared with the former study (7% of the genome) [12]. We found that, although key virulence-associated genes, such as toxin and hemolysin (CD1546), were upregulated in both studies, many putative virulent factors, such as murF (CD2655), cell wall hydrolase (CD2402), and capsular proteins (CD2769 and CD2770), were not DE in the current study. The ribosomal proteins were upregulated in the Caco-2 cell study but were found to be downregulated in the present study. Downregulation of ribosomal protein genes is an indication of stringent response in bacteria [32]. Apart from the entirely different environmental conditions in both experiments, a major contributing factor in the observed difference is the duration of experiments. In the ex vivo analysis, DE gene expression was monitored at 30, 60, and 120 min, and the shortest period in the present study was longer (4 h). The most likely cause of the observed stringent response in ligated loops, as opposed to Caco-2 cell infection, could be that the bacteria in the ligated loops were subjected to more severe nutritional limitation. Stress-induced stringent response is a well-established bacterial survival mechanism [32]. The influence of differences in bacterial exposure time in both experiments was also apparent in the pattern of cellular pathway modulation. Although several pathways were found to be modulated in both studies, there were several key differences (Figure 3). The TCA cycle was modulated during Caco-2 cell infection and at 4 h ligated loop infection but not at other time points. Carbohydarate use pathways, such as lyxose and fructose degradation and 2-methyl citrate cycle, were modulated only during Caco-2 cell infection. In contrast, the uniquely modulated pathways during ligated loops mostly belonged to pathways of fatty acid and lipid degradation, secondary metabolite biosynthesis, cofactors, prosthetic groups, and electron carrier biosynthesis. Because the experimental conditions in both experiments were vastly different, assigning specific reasons for the observed differences is difficult. However, some of the major influencing factors could be the duration of the experiment, nutritional differences, and stress resulting from host inflammatory response.

A major bottle neck in the antimicrobial drug discovery is the identification of a manageable number of enzymes or pathways that are essential for the bacterium [33]. Because the number of essential in vitro pathways is large, identification of pathways essential for growth in the host has been suggested as a method to reduce the number of compounds that inhibit them [34]. Therefore, the pathways that are found to be modulated only in this study could provide a small subset of pathways that can be used for screening of drugs. In summary, the genes identified to be modulated in this study not only extend the understanding of regulation of C. difficile transcriptome but also could be important targets to develop therapeutic agents against CDI.

Supplementary Data

Supplementary data are available at http://jid.oxfordjournals.org online.

Funding

This work was funded by the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services (N01-AI-30054, Project No. ZC005-06).