Correlation between Clostridium difficile Bacterial Load, Commercial Real-Time PCR Cycle Thresholds, and Results of Diagnostic Tests Based on Enzyme Immunoassay and Cell Culture Cytotoxicity Assay

Abstract

The impact of Clostridium difficile fecal loads on diagnostic test results is poorly understood, but it may have clinical importance. In this study, we investigated the relationship between C. difficile fecal load and the results of four assays: a glutamate dehydrogenase (GDH) enzyme immunoassay (EIA), a toxin A/B antigen EIA (ToxAB), a cell culture cytotoxicity assay (CCA), and PCR targeting the tcdB gene. We also compared the PCR cycle threshold (C_T) with the results of quantitative culture using Spearman's rank correlation coefficient. Finally, we sequenced the genomes of 24 strains with different detection profiles. A total of 203 clinical samples harboring toxigenic C. difficile were analyzed and sorted into one of four groups: 17 PCR⁺ (group 1), 37 PCR⁺ GDH⁺ (group 2), 24 PCR⁺ GDH⁺ CCA⁺ (group 3), and 125 PCR⁺ GDH⁺ ToxAB⁺ (group 4). The overall median fecal load in log₁₀ CFU/g was 6.67 (interquartile range [IQR], 5.57 to 7.54). The median fecal bacterial load of groups 1, 2, 3, and 4 were 4.15 (IQR, 3.00 to 4.98), 5.74 (IQR, 4.75 to 6.16), 6.20 (IQR, 5.23 to 6.80), and 7.08 (IQR, 6.35 to 7.83), respectively. Group 1 samples had lower fecal loads than those from each of the other groups (P < 0.001). Group 2 samples had lower fecal loads than those from groups 3 and 4 (P < 0.001). There was a significant correlation between PCR C_T and fecal loads (ρ = −0.697; P < 0.001). NAP1 strains were associated with the detection of toxins by EIA or CCA (P = 0.041). This study demonstrates an association between C. difficile fecal load and the results of routinely used diagnostic tests.

INTRODUCTION

Clostridium difficile infections (CDI) cause significant morbidity and mortality worldwide (1, 2). Several laboratory methods have been developed to detect C. difficile in stool samples for the diagnosis of CDI in the presence of compatible symptoms. These methods vary significantly in terms of sensitivity, specificity, cost, and turnaround time (3). Widely regarded as the gold standard, toxigenic culture (TC) is sensitive but necessitates substantial expertise and is ill suited for most clinical laboratories (3). For this reason, the cell culture cytotoxicity neutralization assay (CCA), which detects the presence of toxin B (ToxB), is often considered the clinical gold standard, as it is less cumbersome than TC. In contrast, enzyme immunoassays (EIA) detecting toxins A and/or B (ToxAB) are technically simpler but are usually much less sensitive (3, 4). These tests are often combined with an EIA test for glutamate dehydrogenase (GDH), a more sensitive but less specific antigen present in both toxigenic and nontoxigenic C. difficile (5). More recently, PCR assays targeting the ToxA and/or ToxB genes have been commercialized and appear to be highly sensitive and specific (6, 7). Some laboratories also use a combination of tests to improve sensitivity and turnaround time while controlling costs (8–10).

These tests differ in terms of sensitivity, and the factors underlying these differences are poorly understood. For example, studies investigating the effects that the type of C. difficile strain could have on test performance have yielded conflicting results (11, 12). The sensitivity of a diagnostic test may also be influenced by technical variables, such as sample dilution and reagents' freshness (13). Alternatively, the variations could be, at least in part, due to a difference in detection thresholds, with more sensitive tests being able to detect lower fecal bacterial loads. However, few studies have examined this hypothesis (14, 15).

We investigated the relationship between fecal bacterial load and the results of four different diagnostic tests commonly used to diagnose C. difficile: ToxAB, GDH, CCA, and PCR targeting the tcdB gene. To achieve this goal, we compared the results of each of these tests with quantitative culture. As quantitative culture is cumbersome and has a long turnaround time, we also investigated whether amplification threshold (C_T) values of a commercially available PCR can predict C. difficile fecal load and thus become convenient surrogate markers of bacterial load (16). In addition, as previous studies suggest that the C. difficile strain can have an impact on test positivity (11), we used whole-genome sequencing on a subset of samples to investigate the relationship between the C. difficile strain and the laboratory test result.

MATERIALS AND METHODS

We performed a prospective study of stool samples submitted to the clinical microbiology laboratory of the University Institute of Cardiology and Pneumology (IUCPQ), Quebec City, Canada, for C. difficile testing. Between August 2010 and July 2011, all specimens submitted for C. difficile testing were analyzed using a commercially available PCR targeting the ToxB gene tcdB (BD GeneOhm Cdiff, Franklin Lakes, NJ) (17). Positive samples were immediately retested using a three-step algorithm (subsequently referred to as EIA/CCA) (Fig. 1). The first two steps of EIA/CCA consisted of the detection of GDH antigen (C. DIFF Chek-60; Techlab, Blacksburg, VA) and toxins A and B (TOX A/B Quik Check; Techlab) by EIA. Samples yielding a positive result for GDH and ToxAB were considered positive for the presence of C. difficile. Samples with a positive result for GDH but a negative result for ToxAB were immediately retested by CCA. This assay uses the Vero cell line and detects the presence of a cytopathic effect neutralized by C. difficile antitoxin (Bartels Immunodiagnostic Supply, Bellevue, WA). According to institutional policy, only loose or unformed stools were tested for C. difficile. Specimens were transported at room temperature in sterile containers without transport media and kept in the lab at 3 to 6°C until being processed within 24 h of collection. All PCR-positive samples were aliquoted and stored at −80°C for subsequent quantitative culture and toxigenic culture.

Fig 1.

Flow chart of laboratory diagnosis of Clostridium difficile in stool samples by PCR and by the 3-step algorithm. Abbreviations: tcdB, toxin B gene; GDH, glutamate dehydrogenase; EIA, enzyme immunofluorescent assay; ToxAB, C. difficile toxin A or B; TC, toxigenic culture; sp., sample volume; Impossible ID, incorrect labeling preventing identification; CCA, cell culture cytotoxicity assay.

Quantitative culture.

We developed a quantitative culture protocol adapted from previous studies (18–21). Briefly, a 0.15-g aliquot of the sample was transferred to a conical tube and treated by ethanol shock for 30 min in an equal volume of 90% ethanol. The mixture was then diluted in 10 volumes of phosphate-buffered saline (PBS) 1×, homogenized using a vortex mixer for 5 min, and left to sediment for 2 min. The supernatant was used to make serial 10-fold dilutions in PBS 1× from 10⁻¹ to 10⁻⁵. A 100-μl aliquot of each dilution was inoculated in duplicate onto prereduced cycloserine cefoxitin fructose agar supplemented with horse blood and taurocholate (CCFA-HT) (22–24). After 48 h of anaerobic incubation at 37°C, plates were inspected for growth of colonies with morphology characteristic of C. difficile. Plates with 30 to 300 colonies were counted with a detection limit of 3.0 log₁₀ CFU/g. For each dilution, the average of the two duplicate plates was calculated. When two successive dilutions yielded 30 to 300 colonies, the average count of both dilutions was calculated. All colonies that grew on CCFA-HT with characteristic morphology were presumptively considered C. difficile isolates; one colony was subcultured onto blood agar for formal identification. Confirmation of C. difficile was achieved by Gram stain and latex agglutination (Microgen Bioproducts, Ltd., Camberley, United Kingdom). Aliquots of pure cultures of a single formally identified colony for each stool sample were stored in brain heart infusion (BHI) broth (BD Diagnostics, Sparks, MD) plus 10% glycerol at −80°C for subsequent use.

Toxigenic culture.

C. difficile isolated from stool samples positive by PCR, but negative by EIA/CCA, were tested for toxin production using TC. Following formal identification, the isolate recovered from the quantitative culture was inoculated into a BHI broth. After 48 h of incubation, the broth was filtrated on a 0.45-μm filter and the filtrate was kept at −80°C. Toxigenicity of the isolate was confirmed by CCA on the cell-free broth filtrate. Samples positive for C. difficile by both PCR and EIA/CCA and yielding C. difficile on CCFA-HT were classified as toxigenic C. difficile and were not further analyzed. Samples that did not yield C. difficile on CCFA-HT and samples that yielded nontoxigenic C. difficile were excluded.

Real-time PCR and assessment of cycle thresholds.

Amplification of the tcdB gene by the BD GeneOhm Cdiff PCR was performed on a SmartCycler (Cepheid, Sunnyvale, CA) according to the manufacturer's protocol. This assay uses a glass bead lysis kit which can extract DNA from both vegetative cells and spores. Because cycle threshold (C_T) values are not readily available to the user (i.e., the interpretation software presents results as positive, negative, or unresolved), authorization was obtained from the manufacturer to extract the C_T values of every PCR-positive sample. Samples with a C_T value higher than 45 were considered negative per the manufacturer's instructions.

Genome sequencing of selected isolates.

Twenty-four samples—6 from each diagnostic category—were randomly selected for sequencing. A single C. difficile colony was subcultured to perform the sequencing analysis. Genomic DNA was extracted using the PureLink viral RNA/DNA minikit (Invitrogen, Burlington, ON, Canada) and DNA was quantified with PicoGreen (Invitrogen). Libraries were prepared using the Nextera library preparation kit (Illumina, Carlsbad, CA) following the manufacturer's protocol. A total of 20 μl of 2.5 nM diluted DNA was used for library preparation. Libraries were dual indexed during preparation and quantified with PicoGreen prior to sequencing. Twelve libraries were pooled at equal concentrations, and the multiplexed libraries were denatured and diluted to 20 pM following the protocol recommended for sequencing of Nextera libraries on the MiSeq sequencer (Illumina). Sequencing was performed on the MiSeq system using paired-end reads of 150 nucleotides. Genomes were assembled using Ray 2.0.0 (25). Reads were aligned to reference sequence C. difficile CD196 to generate multiple whole-genome sequence alignment. The alignment was loaded in MEGA5 to construct a neighbor-joining phylogenetic tree (26). Strains were classified as NAP1 if they clustered with the NAP1 isolate CD196 in the whole-genome phylogenetic tree.

Statistical analysis.

We compared the bacterial loads and C_T values of four distinct and mutually exclusive diagnostic categories of positive samples (Fig. 1). Group 1 included all samples that yielded a positive result by PCR but negative results by EIA/CCA because of a negative GDH. Group 2 included samples that yielded positive results by both PCR and GDH but negative results by ToxAB and CCA. Group 3 included samples that yielded positive results by PCR, GDH, and CCA but a negative result by ToxAB. Group 4 included samples that yielded positive results by PCR, GDH, and ToxAB. The choice of these four categories was made a priori to reflect diagnostic approaches typically used in clinical laboratories and was based on previous studies suggesting increasing levels of sensitivity between ToxAB, CCA, GDH, and PCR (3). Bacterial loads were expressed in CFU per gram of stool on a logarithmic scale (log₁₀ CFU/g).

As the fecal loads and C_T values did not follow normal distributions, the Mann-Whitney test was used for pairwise comparison. A scatter plot was also generated to investigate the relationship between the bacterial load and the C_T. The Spearman's rank correlation coefficient (ρ) was used to assess the presence of a significant correlation. All tests were two-sided, and differences were considered to be statistically significant at P values of <0.05. All data were analyzed using SPSS version 20.0 statistical software (SPSS, Inc., Chicago, IL).

To evaluate the impact of C. difficile strains on test results, we regrouped the four categories into two, based on the presence or absence of detectable amounts of toxins. Hence, groups 1 and 2 were grouped into “no toxin detected” (i.e., CCA and ToxAB negative), whereas groups 3 and 4 were grouped into “toxin detected” (i.e., CCA or ToxAB positive). We then compared NAP1 strains and non-NAP1 strains in terms of bacterial loads and the presence of toxins using univariate and multivariate logistic regression.

This study was approved by the IUCPQ institutional review board.

Nucleotide sequence accession numbers.

Genome sequences were deposited in the European Nucleotide Archive. The accession number for the data set is PRJEB3957. The individual accession numbers for the samples are ERR277249 to ERR277272.

RESULTS

During the study period, 1,320 consecutive stool specimens were submitted to the microbiology laboratory. Of these, 224 samples with a positive result by PCR were used for evaluation of their bacterial loads. Twenty-one samples were rejected for various reasons, such as incorrect labeling or a missing sample (n = 13), insufficient volume preventing quantitation (n = 7), and lack of cytopathic effect on TC (n = 1). A total of 203 samples were included in the final analysis. There were 17 samples (8.4%) in group 1 (PCR⁺ GDH⁻), 37 (18.2%) in group 2 (PCR⁺ GDH⁺ ToxAB⁻, CCA⁻), 24 (11.8%) in group 3 (PCR⁺ GDH⁺ ToxAB⁻ CCA⁺), and 125 (61.6%) in group 4 (PCR⁺ GDH⁺ ToxAB⁺).

C. difficile fecal loads and correlation with PCR cycle thresholds.

The overall median fecal load (interquartile range [IQR]) of all 203 samples was 6.67 log₁₀ CFU/g (IQR, 5.57 to 7.54 log₁₀ CFU/g). The bacterial loads of samples according to each diagnostic category are shown in Fig. 2. Samples positive by PCR only (group 1) had the lowest bacterial load (median [IQR], 4.15 log₁₀ CFU/g [3.00 to 4.98 log₁₀ CFU/g]). Samples positive by PCR and GDH only (group 2) had higher fecal loads (median [IQR], 5.74 log₁₀ CFU/g [4.75 to 6.16 log₁₀ CFU/g]). Samples positive by PCR, GDH, and CCA (group 3) had a median count of 6.20 log₁₀ CFU/g (IQR, 5.23 to 6.80 log₁₀ CFU/g). Finally, samples positive by ToxAB (group 4) had the highest median bacterial load (median [IQR], 7.08 log₁₀ CFU/g [6.35 to 7.83 log₁₀ CFU/g]). A comparison of the various categories showed that group 1 samples had significantly lower bacterial loads than the samples in the other categories (P < 0.001 for each comparison). Furthermore, group 2 samples had significantly lower bacterial loads than groups 3 and 4 (P < 0.001 for each comparison). However, the bacterial loads were not significantly different between groups 3 and 4 (P = 0.08).

Fig 2.

Box plot showing results of a comparison of the Clostridium difficile bacterial loads of stool samples detected by various laboratory methods. Bacterial loads are expressed in CFU/g of stool and presented on a logarithmic scale. The horizontal line in each box indicates the median, whereas the top and bottom lines represent the 75th and 25th percentiles, respectively. Error bars represent 95% confidence intervals, and the dots represent outliers. Abbreviations: PCR, detection of tcdB gene by PCR; GDH EIA, detection of glutamate dehydrogenase by enzyme immunofluorescent assay; CCA, cell culture cytotoxicity neutralization assay; ToxAB EIA, detection of toxins A and B by enzyme immunofluorescent assay; n/a, not available.

The relationships between the C_T values and each diagnostic test category are presented in Fig. 3. The median (IQR) C_T values of groups 1, 2, 3, and 4 were 39.1 (37.6 to 40.9), 34.5 (30.9 to 37.5), 30.4 (29.3 to 34.0), and 28.4 (27.0 to 30.0), respectively. The differences between each of these groups in terms of the median C_T were statistically significant (P ≤ 0.01 for each comparison).

Fig 3.

Box plot showing results of a comparison of the C. difficile PCR cycle thresholds of stool samples detected by various laboratory methods. The horizontal line in each box indicates the median, whereas the top and bottom lines represent the 75th and 25th percentiles, respectively. Error bars represent 95% confidence intervals, and the dots represent outliers. Abbreviations: PCR, detection of tcdB gene by PCR; GDH EIA, detection of glutamate dehydrogenase by enzyme immunofluorescent assay; CCA, cell culture cytotoxicity neutralization assay; ToxAB EIA, detection of toxins A and B by enzyme immunofluorescent assay; n/a, not available.

The overall median cycle threshold (IQR) of tested samples was 29.8 (IQR, 27.8 to 33.5). The scatter plot (Fig. 4) illustrates the relationship between fecal C. difficile loads and the PCR C_T. There was a significant inverse correlation between PCR C_T values and bacterial loads (Spearman ρ = −0.70; P < 0.001).

Fig 4.

Scatter plot showing the relation between Clostridium difficile fecal load determined by quantitative culture and cycle threshold of a PCR detecting the tcdB gene. Data are presented on a logarithmic scale.

Strain analysis.

Using the Illumina MiSeq sequencer, we sequenced the full genome of 24 selected C. difficile isolates. The mean depth of sequencing was 20×, with 86 million nucleotides sequenced on average per isolate. Assembled genomes were analyzed to identify multilocus sequence type (MLST) groups (27). A whole-genome phylogeny was constructed for these genomes. Some strains showed a high homology to the circulating NAP1 strain (n = 14), while others showed more diversity corresponding to other genotypes (n = 10) (Fig. 5). Strains were classified as NAP1 if they aligned with the CD196 isolate of the NAP1 strain in the whole-genome phylogenetic tree. Two sequences were not included in the phylogeny (E83 and E114) due to low sequencing depth but were still assigned to the NAP1 group and included in the final analysis.

Fig 5.

Whole-genome phylogeny showing alignment to a reference sequence (CD196) of 22 Clostridium difficile clinical isolates obtained from symptomatic patients. Samples were tested for positivity (+) or negativity (−) with four assays: detection of the tcdB gene by PCR (PCR), detection of glutamate dehydrogenase by enzyme immunofluorescent assay (GDH), detection of toxins A and B by enzyme immunofluorescent assay (Tox), and cell culture cytotoxicity assay (CCA). Bacterial loads are represented on a logarithmic scale.

When we assessed the impact of the various strains on the laboratory test results, the fecal bacterial loads were not significantly different between NAP1 and non-NAP1 strains (median, 5.78 versus 4.86 log₁₀ CFU/g, respectively; P = 0.10). In contrast, the presence of toxins was significantly associated with the presence of NAP1 compared with non-NAP1 strains (9/14 [64%] versus 2/10 [20%]; odds ratio [OR], 7.2; 95% confidence interval [CI], 1.08 to 47.96; P = 0.041). However, the association between the NAP1 strains and the detection of toxin was not significant when we controlled for fecal bacterial load (adjusted OR, 5.73; 95% CI, 0.44 to 74.8; P = 0.44).

DISCUSSION

Although fecal bacterial loads can vary significantly between patients with CDI (18), the impact of this variable on test positivity is not fully understood. Our study provides evidence of an association between C. difficile fecal load and the results of different diagnostic tests used routinely by clinical laboratories. Our results are in line with another recently published study showing that ToxAB-positive samples have higher bacterial loads than ToxAB-negative samples (7.0 versus <2 log₁₀ CFU/g) (18).

To our knowledge, this study is the first to evaluate the associations between fecal loads measured by quantitative culture and the results of PCR, GDH, ToxAB, and CCA. Two previous studies have evaluated C. difficile fecal loads using quantitative PCR rather than quantitative culture. In the first study, the relationship between the C. difficile DNA fecal load and the results of ToxAB and CCA was examined in 107 samples, and the results demonstrated that toxin-negative samples had 10¹ to 10⁴ fewer DNA copies than toxin-positive samples (14). In the second study, a difference in sensitivity between two different nucleic acid amplification tests (PCR and loop-mediated isothermal amplification) was also demonstrated to be related to the fecal DNA load (15).

Despite our observation of an association between bacterial load and assay positivity, the exact parameters linking these two variables remain to be elucidated. For example, the association between bacterial load and the results of ToxAB and CCA is probably indirect, as these latter assays detect the presence of toxins rather than the bacteria per se. Samples that are PCR positive but ToxAB and CCA negative might harbor toxigenic C. difficile strains that are not expressing their toxin genes. As the signals regulating toxin expression are not fully understood, additional studies are needed to shed light on this important question (28).

Our study had some limitations. First, it was conducted in a single center using four different diagnostic assays, so our results may not be generalizable to other patient populations and other assays. Some of the toxins may have been degraded by proteases between collection and processing of the specimens. By eliminating vegetative cells, the ethanol shock step may have underestimated the true fecal load by approximately 1.0 log₁₀ CFU/g (29). Because we used GDH as a screening tool, PCR⁺ GDH⁻ samples were not tested for toxin production. Some of these samples may have harbored toxin. We sequenced a limited number of C. difficile strains, so our study may have been underpowered to detect an association between NAP1 strain and test positivity while controlling for bacterial load. Stool samples used for this study were stored at −80°C without the addition of conservation medium; this may have led to underestimation of the true fecal load. Finally, the clinical significance of detecting low quantities of C. difficile without detecting the presence of toxins remains to be determined.

Our study shows that the bacterial load measured by quantitative culture correlates reasonably well with PCR C_T. Other studies support the notion that there is a significant correlation between results obtained by real-time PCR and quantitative culture (18, 30). This suggests that a readily available qualitative real-time PCR that does not require calibration can provide some approximate indication of the C. difficile fecal load. We hypothesize that such a test could be adapted to provide quantitative results in the future.

In addition to standard diagnostic tests, the genomes of 24 C. difficile strains with different detection profiles were sequenced. We observed that 14 of these samples were closely related NAP1 strains, while the 10 non-NAP1 strains had more diverse genome sequences. Although NAP1 strains were associated with the detection of toxin by ToxAB and CCA, we could not demonstrate such an association while controlling for fecal load.

Conclusion.

Our study provides evidence that the reported differences in clinical sensitivities between PCR, EIA, and CCA are due, at least in part, to their relative capacity to detect low fecal bacterial loads. We have also demonstrated that it may be possible to estimate the bacterial load of a stool sample using a commercially available PCR already used in some clinical laboratories. Further research is needed to better understand the clinical impact of these findings.