Comparison of Real-Time PCR Techniques to Cytotoxigenic Culture Methods for Diagnosing Clostridium difficile Infection

Abstract

In the past decade, the incidence of Clostridium difficile infections (CDI) with a more severe course has increased in Europe and North America. Assays that are capable of rapidly diagnosing CDI are essential. Two real-time PCRs (LUMC and LvI) targeting C. difficile toxin genes (tcdB, and tcdA and tcdB, respectively) were compared with the BD GeneOhm PCR (targeting the tcdB gene), using cytotoxigenic culture as a gold standard. In addition, a real-time PCR targeting the tcdC frameshift mutation at position 117 (Δ117 PCR) was evaluated for detecting toxigenic C. difficile and the presence of PCR ribotype 027 in stool samples. In total, 526 diarrheal samples were prospectively collected and included in the study. Compared with those for cytotoxigenic culture, sensitivity, specificity, positive predicted value (PPV), and negative predicted value (NPV) were for PCR LUMC 96.0%, 88.0%, 66.0%, and 98.9%, for PCR LvI 100.0%, 89.4%, 69.7%, and 100.0%, for PCR Δ117 98.0%, 90.7%, 71.9%, and 99.5%, and for PCR BD GeneOhm 88.3%, 96.9%, 86.5%, and 97.4%. Compared to those with feces samples cultured positive for C. difficile type 027, the sensitivity, specificity, PPV, and NPV of the Δ117 PCR were 95.2%, 96.2%, 87.0%, and 98.7%. We conclude that all real-time PCRs can be applied as a first screening test in an algorithm for diagnosing CDI. However, the low PPVs hinder the use of the assays as stand-alone tests. Furthermore, the Δ117 PCR may provide valuable information for minimizing the spread of the epidemic C. difficile PCR ribotype 027.

Clostridium difficile is a major cause of nosocomial diarrhea and pseudomembranous colitis (2). The incidence of Clostridium difficile infection (CDI) has increased in the past decade, which is associated with the emergence of the hypervirulent PCR ribotype (type) 027 (8, 11). The C. difficile enterotoxin A (TcdA) and cytotoxin B (TcdB) are considered major virulence factors, whereas the binary toxin might play a role in virulence through the formation of microtubule-based protrusions, thereby increasing the adherence of the bacteria (9, 15, 20). CDI can also be caused by strains that produce only TcdB (18), but strains producing TcdA only have not been described. Assays for the rapid diagnosis of CDI are important to prevent the spread of C. difficile, in particular for hypervirulent strains like type 027. Conventional diagnostic methods for CDI, such as cytotoxigenic culture (CYTGC), are time-consuming and not available at all routine diagnostic laboratories, whereas the performance of rapid enzyme immunoassays to detect toxins in feces is insufficient (4, 5, 11). Previously, we developed a real-time PCR that detects the presence of the tcdB gene (19). In this study we have improved the performance of this PCR and compared it with the commercially available BD GeneOhm PCR and another in-house-developed real-time PCR assay that detects the presence of the tcdA and tcdB genes (6). In addition, we evaluated a real-time PCR that targets the tcdC gene frameshift mutation at position 117, which can act as a marker for the type 027/NAP1 strain (5, 10).

MATERIALS AND METHODS

CYT and CYTGC.

In total, 526 routine diagnostic diarrheal samples were submitted to the Department of Microbiology at Leeds Teaching Hospitals and tested prospectively by the cytotoxicity assay (CYT) and CYTGC assays as previously described (7). Briefly, all stool samples (less then 48 h old) were stored at 2 to 5°C. Twenty μl of diluted fecal sample (1:5 in phosphate-buffered saline [PBS]) was filtered and added to a monolayer of both Clostridium sordellii antitoxin-protected (Prolab Diagnostics, United Kingdom) and -unprotected Vero cells. A sample was considered toxin positive when cell rounding was observed after 24 or 48 h of incubation. In addition, cultured C. difficile isolates from feces samples that were found negative by the CYT assay were investigated for toxin production using the CYTGC assay (7). Isolates were inoculated into brain heart infusion (BHI) broth. After 48 h of incubation, culture supernatants were added to a monolayer of protected and unprotected Vero cells. Cultured C. difficile isolates that were positive for the CYT assay were considered to be positive for the CYTGC assay.

Culture.

The culture of isolates was performed as previously described (7). In short, following alcohol shock, samples were cultured on Braziers cycloserine-cefoxitin-egg yolk (CCEY) agar (Bioconnections, Wetherby, United Kingdom) supplemented with 5 mg/liter lysozyme (Sigma, United Kingdom) and without egg yolk supplement. Incubation was done in an anaerobic workstation (Don Whitley, United Kingdom) for at least 48 h. Gray-brown colonies with the characteristic horse manure odor were identified as C. difficile. Whenever the identification of an isolate was questionable, the Microgen C. difficile latex agglutination kit (Microgen Bioproducts Ltd., Camberley, United Kingdom) was used to confirm the identity.

DNA extraction.

Fecal samples were stored at 4°C for 1 week and then frozen at −20°C. Specimen preparation and DNA extraction for the BD GeneOhm Cdiff assay were performed according to the manufacturer's protocol. For the other real-time PCRs, fecal samples were pretreated with stool transport and recovery (STAR) buffer (Roche, Penzberg, Germany) according to the manufacturer's protocol. For the LUMC real-time PCR, DNA was extracted on the MagNA Pure (Roche) using the LC DNA isolation kit III (Roche, Penzberg, Germany) according to the manufacturer's protocol. In short, 100 μl supernatant of a STAR buffer- and chloroform-pretreated fecal sample was added to lysis buffer (130 μl) and Prot K (20 μl). This mixture was heated at 65°C for 10 min followed by 95°C for 10 min, after which it was centrifuged for 1 min at 1,000 × g. A total of 200 μl supernatant was used for automated DNA extraction. DNA was eluted in 100 μl elution buffer. The phocine herpes virus (PhHV), which served as an internal control, was added to the lysis buffer. For the LvI real-time PCR, DNA was extracted on the NucliSENS easyMAG (bioMérieux, Boxtel, Netherlands) according to the specific A protocol (bioMérieux, Boxtel, Netherlands). In short, 150 μl of STAR buffer- and chloroform-pretreated feces suspension was added to 2 ml lysis buffer (NucliSens; bioMérieux, Boxtel, Netherlands). After incubation for 10 min at room temperature, the total suspension was transferred to the sample vessel, including 140 μl magnetic silica beads, and used for automated DNA extraction. DNA was eluted in 110 μl elution buffer. The internal control PhHV was added to the lysis buffer.

Real-time PCR.

Amplification of part of the tcdB gene by the BD GeneOhm Cdiff PCR was performed on a Smartcycler (Cepheid, United Kingdom) according to the manufacturer's protocol. Primers and probes that were used for the LUMC real-time PCR and the LvI PCR are described in Table 1. Amplification of the tcdB gene by the LUMC real-time PCR was performed on a CFX detection system (Bio-Rad, Veenendaal, Netherlands) as previously described (19), with some optimizations. The PCR amplification was performed in a 50-μl final volume, containing 25 μl Hotstar mastermix (Qiagen, Venlo, Netherlands), forward and reverse primers at 80 nM each, 3.5 mM MgCl₂, 100 nM tcdB probe, and 10 μl DNA. The PhHV primers described by Niesters (12) were used with a modified probe. The amplification protocol included an enzyme activation step for 15 min at 95°C, followed by 50 cycles of amplification, 94°C for 30 s, 50°C for 30 s, and 72°C for 30 s.

TABLE 1.

Primers and probes used for in-house-developed real-time PCRs

Real-time PCR	Target	Primer or probe	Nucleotide sequence (5′→ 3′)a
LUMC	TcdB	398CLDs	GAAAGTCCAAGTTTACGCTCAAT
		399CLDas	GCTGCACCTAAACTTACACCA
		551CLD-tq-FAM	FAM-ACAGATGCAGCCAAAGTTGTTGAATT-BHQ1
LvI	TcdA	CD-tcdA-F	TTGTATGGATAGGTGGAGAAGTCAGT
		CD-tcdA-R	AATATTATATTCTGCATTAATATCAGCCCAT
		CD-tcdA-MGB1	FAM-ATATTGCTCTTGAATACATAAA-NFQ-MGB
		CD-tcdA-MGB2	FAM-TATTGTTCTTGAATACATAAAAC-NFQ-MGB
	TcdB	CD-tcdB-F1	GAAACAGGATGGACACCAGGTT
		CD-tcdB-F2	AAGAGGATGGACGCCAGGTT
		CD-tcdB-R1	ACGGTCTAACAGTTTTGTGCCA
		CD-tcdB-R2	CTGCCCTTCATAATGATCTCTTATACG
		CD-tcdB-MGB	FAM-AAGAAGCTTAGAAAATG-NFQ-MGB
Δ117 PCR	TcdC	CD-tcdC-F	GCACAAAGGRTATTGCTCTACTGG
		CD-tcdC-R1	AGCTGGTGAGGATATATTGCCAA
		CD-tcdC-R2	CAAGATGGTGAGGATATATTGCCA
		CD-tcdCwt-MGB	FAM-AAACACRCCHAAAATAA-NFQ-MGB
		CD-tcdCmut-MGB	VIC-AAACACRCCAAAATAA-NFQ-MGB
All	PhHV	295PhHVs	GGGCGAATCACAGATTGAATC
		296PhHVas	GCGGTTCCAAACGTACCAA
		531PhHV-tq-CY5	CY5-TTTTTATGTGTCCGCCACCATCTGGATC-BHQ2*
			NED-CGCCACCATCTGGAT-NFQ-MGB**

^aBHQ, Black Hole Quencher; NFQ, nonfluorescent quencher; MGB, minor groove binder; *, used for LUMC PCR; **, used for LVI PCR and Δ117 PCR.

The LvI real-time PCR was designed to target both the tcdA and tcdB genes. Amplification of these genes was performed as a multiplex PCR on an AB 7500 PCR system (Applied Biosystems, Nieuwerkerk a/d IJssel, Netherlands). Each PCR was performed in a 25-μl final volume, containing 1× TaqMan Universal PCR Master Mix (Applied Biosystems, Nieuwerkerk a/d IJssel, Netherlands), forward and reverse primers at 300 nM each, TaqMan MGB probes at 100 nM each, 2.5 μg bovine serum albumin (Roche), and 5 μl DNA extract. For PhHV, the primers described by Niesters (12) were used, whereas the probe was modified to an MGB probe. The amplification protocol included 2 min at 50°C, 10 min at 95°C followed by 40 cycles of amplification; and 94°C for 15 s, 60°C for 1 min.

The Δ117 real-time PCR was designed to target the tcdC gene frameshift mutation at position 117. This assay utilizes two TaqMan MGB probes, a wild-type (WT) probe and a mutant (Δ117 MUT) probe, that both can hybridize with part of the tcdC gene sequence flanking the 1-bp deletion at position 117. Isolates that do not carry the 1-bp deletion will give a stronger signal with the WT probe, while 027/NAP1 isolates will do so with the Δ117 MUT probe. Hence, the threshold cycle (ΔC_T) (C_T WT − C_T Δ117 MUT) for 027/NAP1 strains will be positive, whereas the ΔC_T for other C. difficile ribotypes will be negative, which enables discrimination. The primer/probe set was used in the same multiplex setup as the LvI PCR assay described above, with the primers at 300 nM and the MGB probes (5′-6-carboxyfluorescein [FAM], 5′-VIC, and 3′-nonfluorescent quencher-minor groove binder [NFQ-MGB] [Applied Biosystems]) at 100 nM. Reactions were run on an ABI 7500 with the same amplification protocol as that for the LvI PCR.

PCR ribotyping.

PCR ribotyping was performed at the Department of Microbiology at Leeds Teaching Hospitals following the protocol from the C. difficile Ribotyping Network for England (CDRNE) laboratory (7).

Data analysis.

Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were determined by comparing the real-time PCRs with the CYTGC gold standard, using statistical software PASW 17.0.2. Differences in the sensitivity and specificity between the real-time PCRs were determined by using the McNemar test for correlated proportions. Samples with a C_T value higher then 40 were considered negative. In addition, samples with an internal control C_T value that deviated more than 3 C_T values compared to the internal control C_T value of the negative control were considered inhibited and discarded from the analyses for the LUMC PCR. For the LVI PCR, samples were considered inhibited and discarded from analysis when the C_T value for the internal control exceeded 34.91 cycles (i.e., the mean C_T value for uninhibited specimens had ± 2 standard deviations). The number of inhibited samples for the Δ117 PCR was determined equally to that for the LVI PCR. Furthermore, the Δ117 PCR was compared with PCR ribotyping of CYTGC-positive isolates. The BD GeneOhm Cdiff PCR tests were interpreted according to the manufacturer's protocol. The software on the Smart cycler (Cepheid, United Kingdom) recorded the results of the PCR assay as positive, negative, or unresolved.

RESULTS

Comparing the real-time PCRs with the gold standard CYTGC.

In total, we evaluated 526 diarrheal samples, of which 101 samples (19.2%) were positive in the CYTGC assay. Of 101 positive samples, 13 were derived from CYT assay-negative samples. For the evaluation of the LUMC PCR, 10 samples (1.9%) were excluded from the analysis due to inhibition during the amplification step, whereas 16 samples (3.1%) and 15 samples (2.9%) were inhibited and excluded from the analysis of the LvI PCR and Δ117 PCR, respectively. Five samples (1%) were recorded as unresolved by the BD GeneOhm PCR assay and excluded from the analysis. Sensitivity, specificity, PPV, and NPV for all PCR methods against CYTGC are shown in Table 2.

TABLE 2.

Comparison of four real-time PCR methods with gold standard CYTGC^d

Assay	No. of samples included	No. of samples inhibited	Result	CYTGC assay (no.)		Sensitivity (%) (95% CI)	Specificity (%) (95% CI)	PPV (%)	NPV (%)
				+	−
LUMC PCR	526	10	+	97	50	96.0	88.0	66.0	98.9
			−	4	365	(90.3-98.5)	(84.5-90.7)
LvI PCR	522b	16	+	99	43	100.0	89.4	69.7	100.0
			−	0	364	(96.3-100)	(86.1-92.1)
Δ117 PCRa	522b	15	+	97	38	98.0	90.7	71.9	99.5
			−	2	370	(92.9-99.4)	(87.5-93.1)
BD GeneOhm	512c	N/A	+	83	13	88.3	96.9	86.5	97.4
			−	11	405	(80.3-93.3)	(94.8-98.2)

^aThe lowest C_T value belonging to either the tcdC wild-type or mutant probe was used for the evaluation of the Δ117 PCR as a screening assay.

^bFour samples were not present in the collection.

^cData for 14 samples were not available.

^dFor each PCR method, the numbers of samples included and inhibited are shown. Sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV) of the PCR methods are given as percentages, and the corresponding 95% confidence interval (95% CI) is shown in parentheses.

All stool samples that tested positive for C. difficile by the CYTGC assay were also detected by the LvI PCR. Comparable sensitivities were achieved by the LUMC PCR (96%) and Δ117 PCR (98%), while the sensitivity of the commercial BD GeneOhm PCR was lower (88.3%). The difference in sensitivity between the BD GeneOhm PCR and the three in-house-developed PCRs was significant, with P values of 0.00, 0.01, and 0.04 for the LvI PCR, the Δ117 PCR, and the LUMC PCR, respectively. In contrast, the BD GeneOhm PCR showed higher specificity (96.9%) compared to the LvI PCR (89.4%), the Δ117 PCR (90.7%), and the LUMC PCR (88.0%). The difference in specificity between the BD GeneOhm PCR and all three in-house-developed PCRs was significant, with a P value of 0.00.

Compared to CYTGC, all PCRs had similar NPVs, ranging from 97.4 to 100%. The BD GeneOhm PCR had the highest PPV (86.5%) compared to the LvI PCR (69.7%), the Δ117 PCR (71.9%), and the LUMC PCR (66.0%).

Discrepancy analysis.

Analysis of false-positive results showed an overlap of the numbers of false positives detected by the different PCR methods (Fig. 1 A). In total, 13 false-positive results (14% of the total amount of positives) were found by the BD GeneOhm PCR compared to CYTGC. Of these false positives, 54% (n = 7) were also detected as such by all other PCR methods. Compared to CYTGC, the most false positives (34%) were detected by the LUMC PCR, whereas the LvI PCR and the Δ117 PCR had 30% and 28% false-positive samples, respectively. Fourteen percent of the false positives detected by the LUMC PCR were also detected as such by all other PCR methods, whereas for the LvI PCR and the Δ117 PCR, 16% and 19% of the false positives had a similar test outcome by all other PCR methods.

FIG. 1.

False-positive results (A) and false-negative results (B) detected by real-time PCRs with overlapping samples. All false-positive and false-negative results from each PCR method compared to the CYTGC assay were analyzed for resemblances. Resemblances in false-positive and false-negative results were ordered by PCR method. No false-negative results were found by the LvI real-time PCR.

Figure 1B shows the number of false-negative results detected by the real-time PCRs. No false-negative results were found by the LvI real-time PCR with the CYTGC as the standard. Compared to CYTGC, 11 false-negative results (2.6% of the total amount of negatives) were detected by the BD GeneOhm PCR, whereas the LUMC PCR and the Δ117 PCR had 1.1% and 0.5% false-negative samples, respectively. None of the false-negative samples were detected as such by all three PCR methods; only overlapping results between two PCR methods were found.

Comparing Δ117 PCR samples with PCR-ribotyped CYTGC-positive samples.

Of the 99 CYTGC-positive samples, a total of 21 samples were typed as type 027 by PCR ribotyping (Table 3). The Δ117 PCR was able to confirm 20 of these samples (95%) by detection of the 1-bp deletion at position 117 in the tcdC gene, with a ΔC_T (C_T WT − C_T Δ117 MUT) of +2.9 cycles on average. Compared to PCR-ribotyped CYTGC-positive samples, the sensitivity, specificity, PPV, and NPV values for PCR ribotype 027 samples were 95.2%, 96.2%, 87.0%, and 98.7%. The Δ117 PCR detected 3 samples carrying the Δ117 mutation, which were ribotyped as type 005, type 106, and an unknown type, not type 027.

TABLE 3.

Comparison of the LvI Δ117 PCR with PCR ribotyping^c

Assay	No. of samples includeda	Result	Ribotyping of cultured isolates (no.)		Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
			Ribotype 027	Non-027 ribotype
Δ117 PCR	99	+	20	3b	95.2	96.2	87.0	98.7
		−	1	73

^aOnly CYTGC-positive samples were included.

^bThree CYTGC-positive samples were typed as PCR ribotype 005, ribotype 106, and a rare ribotype (not 027).

^cAll CYTGC-positive samples (n = 99) were analyzed by the Δ117 PCR. Sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV) of the PCR method are given as percentages.

DISCUSSION

Rapid and accurate diagnosis of CDI is essential for patient management and prevention of nosocomial transmission. A main issue in diagnosing CDI is that most conventional tests do not have sufficient performance for applying it as a stand-alone test. Molecular tests are increasingly applied for diagnosing CDI and are also mentioned in a recent published guideline as potentially rapid assays with better performances (4). This study was performed to compare the diagnostic values of three in-house-developed real-time PCRs and a commercially available BD GeneOhm Cdiff assay, using the appropriate gold standard on 526 prospectively collected stool samples. The sensitivity of the in-house-developed PCRs was better than the BD GeneOhm test, in contrast to the specificity. Subsequently, NPVs were similar, whereas the PPV was the highest for the BD GeneOhm test (86.5%). Peterson et al. (13) evaluated a real-time PCR that targeted the tcdB gene of C. difficile and reported a sensitivity of 93.3% and a specificity of 97.4%. This is in line with what has been reported by Sloan et al. (16) on the performance of a real-time PCR, which was directed against the tcdC gene. They reported a sensitivity of 86% and specificity of 97%. The sensitivity and specificity reported by both studies are comparable to what has been found in this study for the BD GeneOhm Cdiff PCR.

Our study had a prevalence of toxigenic C. difficile-positive samples of approximately 20%; repeated samples from positive patients were excluded. The prevalence was high, due to a selection of feces samples with high suspicion of CDI. The PPV is dependent on the prevalence of the disease in the tested population. Several studies report that in the hospital 5% to 10% of the antibiotic-associated diarrhea samples contain C. difficile (1, 4, 14). In the community, the prevalence of CDI is close to 2% (3, 21). Therefore, we calculated PPVs for all PCR methods at prevalences of 2%, 5%, and 10%. The calculated PPVs at 10% and 5% prevalences decreased to 76% and 60% for the BD GeneOhm PCR, 51% and 33% for the LvI PCR, 54% and 36% for the Δ117 PCR, and 47% and 30% for the LUMC PCR. At 2% prevalence, PPVs were 37% (BD GeneOhm), 16% (LvI), 18% (Δ117 PCR), and 14% (LUMC).

The performance of the LUMC real-time PCR reported in this evaluation was better than that reported previously by van den Berg et al. (19). This difference can be explained by the difference in prevalence of CDI-positive samples used in this study (20%) and the previous study (6%). Furthermore, we optimized PCR conditions (PCR mix, modified probe) and reached a detection limit of 10³ CFU/gram stool samples (data not shown), which is improved compared to 10⁵ CFU/gram as reported by van den Berg et al. (19).

When the false-positive PCR results were analyzed, 54% of the false positives detected by the BD GeneOhm PCR were also detected as such by the other PCRs. This suggests that 54% of the false positives contained C. difficile-specific tcdB DNA, since the samples were detected by three PCRs targeting the tcdB gene using different primer sets. In addition, these samples were also found positive by the Δ117 PCR using the tcdC gene as a target. It cannot be excluded that the cultures were false negative due to previous antibiotical treatment, but discrepancies with the CYTGC assay still remain present. We consider this finding an indication of an important lack of the currently available gold standard and think that future clinical studies are necessary to interpret the findings more precisely.

In comparison to CYTGC-positive samples with C. difficile type 027, the Δ117 PCR targeting the tcdC gene 1-bp Δ117 deletion has a high concordance of 95.2%. This high concordance makes the utility of the Δ117 PCR for direct detection of the epidemic strain promising, although further research is needed to determine if other C. difficile types contain the tcdC point mutation at position 117 and, consequently, are detected by this PCR. Furthermore, the performance of this PCR for detection of toxigenic C. difficile indicates that this assay has the potential to diagnose CDI, without prescreening for the toxin genes tcdA and tcdB. The tcdC gene has been recognized as a putative negative regulator of tcdA and tcdB and thereby is indicative of the presence of the pathogenicity locus (17).

A difference between the three in-house-developed real-time PCRs was the percentage of inhibited samples. In total, 3.1% (n = 16) and 2.9% (n = 15) of all samples (n = 522) tested by the LvI PCR and Δ117 PCR were inhibited, respectively, whereas 1.9% (n = 10) of all samples (n = 526) tested by the LUMC PCR were inhibited. The PCRs used different DNA extraction methods and different platforms which might contribute to the observed differences.

Most rapid diagnostic tests do not have sufficient performance to be applied as stand-alone tests. Recently, Planche et al. (14) determined that a test is applicable as a stand-alone test when a sensitivity of at least 90% and a specificity of at least 97% are reached. The three in-house-developed PCRs lack specificity, whereas the BD GeneOhm PCR lacks sensitivity, resulting in too-low PPVs of all PCRs, ranging from 66% to 86.5% at a 20% CDI prevalence. These PPVs decrease substantially when calculating PPVs for CDI prevalences that are more common for a clinical setting (10%) or observed in the community (2%). None of our evaluated real-time PCR methods fulfilled the criteria defined by Planche et al. (14). Therefore, it was concluded that they cannot be applied as stand-alone tests. This finding is in line with what has been found for toxin detection assays and other molecularly based assays by other studies (4, 7, 14). However, due to their high NPVs, all four evaluated PCR methods can be applied as a first negative screening test for CDI in a two-step algorithm. In this algorithm the PCR assay is followed by a second confirmation step to confirm the first positive test result.