A set of 217 samples of feces from patients suffering diarrhea was collected from September 2015 to April 2017 from two healthcare centers in Bogotá, Colombia: Hospital Universitario Mayor—Méderi and Fundación Clínica Shaio. Inclusion criteria were defined in line with the “Clinical Practice Guidelines for Clostridium difficile Infection in Adults” proposed by the Society for Healthcare Epidemiology of America (SHEA) and the Infectious Diseases Society of America (IDSA) (Cohen et al., 2010). This meant including only older patients suffering diarrhea, defined as the passage of 3 or more unformed stools in 24 or fewer consecutive hours. The site of acquiring the infection was classified according to the aforementioned guidelines; the moment of symptom onset was taken as indicator of site of exposure to CDI. A CO patient case was defined as when the episode of diarrhea was presented during the first 48 h following admission to a medical center, while an HCFO patient case was that where diarrhea occurred after the third day following admission.

Considering the lack of a standardized CDI detection strategy in Colombia and the limited conditions and resources available. This study sought to implement a practical scheme with the lowest possible cost, allowing adequate screening of the samples. To this end, a review of the available literature was carried out, including a scheme previously implemented by our group for the processing of stool samples (Sanchez et al., 2017), which was adapted to the requirements of the media manufacturers and commercial kits implemented. In this context, all diarrheic feces samples were collected in sterile recipients with airtight seal (to avoid direct exposure to oxygen) and without transport media (Shin and Lee, 2014). For storage, the samples were placed inside hermetic recipient and stored under refrigeration (2–8°C) until being processed (within the first 72 h following collection). The samples were then transported at the Universidad del Rosario's Microbiology Laboratory (conserving the cold chain), where they were homogenized by mechanical disruption using sterile scrapers, as the initial step for processing them. The manipulation of the sample during this procedure was carried out at high speed in the laminar flow cabin, taking care of not exposing the sample to oxygen for more than 15 s (Brown et al., 2011).

This study was approved by the Universidad del Rosario's Research Ethics' Committee (CEI-UR). This research was considered low risk due to Colombian Ministry of Health resolution 008430/1993 criteria stating that experimental interventions cannot be made regarding research subjects. Data concerning patient identification was treated confidentially, in line with Colombian legal and ethical guidelines and according to that expressed by the latest version of the Declaration of Helsinki (World Medical Association). Hospital informed consent was obtained from HCFO patients. No clinical data regarding CO patients was accessed; this meant that the institution' ethics committee granted permission for the anonymous use of samples for medical research, according to current Colombian regulations concerning ethics.

The CDI detection algorithm described in Figure 1 was applied to the samples immediately after the mechanical disruption process (within the 15 s limit). The sample was extended by depletion directly over the chromID C. difficile agar CDIF (bioMérieux) using a sterile swab. Samples were immediately transferred to an anaerobic jar and incubated under anaerobic conditions (using a GasPak EZ Anaerobe Pouch; Becton Dickinson) for 48 h at 37°C. This culture diagnosis strategy has been reported as being highly sensitive for CDI from symptomatic patients' stools (Eckert et al., 2013). A sample was considered positive by SCM when colonies having the macroscopic morphology described by the manufacturer (gray to black colonies, having an irregular or smooth border) were observed following 48 h incubation. SCM results were confirmed by verification by colony screening (VCS) (Figure 1). Because a possible co-existence of different CD genotypes has been reported in a sample simultaneously (Tanner et al., 2010), from 1 to 7 colonies (depending on the amount of colony-forming units recovered for each sample during the initial step in SCM), were extended on trypticase soy agar (TSA) with 5% sheep blood (Becton Dickinson), and subsequently incubated under aforementioned conditions. VCS was completed by microscopic inspection by routine interpretation by Gram staining, using a smear of the colony on a slide with saline solution. A sample was considered positive by VCS when its morphology revealed typical characteristics expected for CD by detection in culture (gram positive bacillus, occasionally sporulated) (Cohen et al., 2010).

ATCC BAA-1870 (tcdA, tcdB, and cdt presence confirmed by PCR) and ATCC 700057 (toxinotype tcdA-, tcdB- and binary toxin gene cdtB not amplified by PCR) strains were acquired from the American Type Culture Collection (Manassas, VA). The strains were sown on TSA with 5% sheep blood and incubated, according to the aforementioned culture conditions. Cell biomass was recovered in 1X sterile 300 μL phosphate buffered saline (PBS) until a 4 × 10⁷ cell per mL optical density (OD₆₀₀) was reached. The cell suspension was stored at −20°C until further use.

A Stool DNA Isolation Kit (Norgen, Biotek Corporation) was used for extracting DNA from a 300 μL aliquot of the previously homogenized feces samples following the manufacturer's instructions. The aliquot of the sample was obtained just after sowing for SCM, deposited in sealed containers and processed within the maximum time limit established for storage (72 h). The commercial kit used was selected for its high performance for the recovery of DNA from feces, despite the complexity represented by this sample source (Mathay et al., 2015; Sanchez et al., 2017). Extracted DNA was recovered in 100 μL elution buffer included in the kit then stored at −20°C until being used. Regarding control strains, an Ultraclean BloodSpin DNA Isolation kit (MoBio Laboratories) was used for extracting DNA from the cell biomass recovered in 1X PBS, according to the manufacturer's instructions, eluting in the same final volume (100 μL). A NanoDrop2000 (Thermo Scientific NanoDrop Products) was used for spectrophotometric quantification of the DNA extracted from the control strains and then stored at −20°C for later assays.

PCR was used for the molecular detection of CDI; three tests targeting constitutive genes previously reported in the literature were performed independently (Figure 1). Two of these tests were conventional PCR (conv.PCR), one targeting 16S.rRNA (conv.16S) (Naaber et al., 2011) and the other gdh (conv.gdh) (Paltansing et al., 2007). Table 1A lists the sequence of the primers used in each test. Two independent conv.PCR assays were carried out on a Labnet Thermocycler (Labnet International), using GoTaq Green Master Mix (Promega) at 1X final concentration, with 1 μM of each primer and 3 μL of target DNA at 25 μL final reaction volume. The thermal profiles used for the PCRs were standardized from the conditions proposed by the authors for each primer set (Paltansing et al., 2007; Naaber et al., 2011). Test's amplification conditions were verified using 36 ng DNA extracted from ATCC BAA-1870 reference strain as template. The best thermal profiles for each amplification were identified (Table 1B). Serial dilutions from the same stock of DNA were amplified in duplicate in the conditions described for both tests for determining test limit of detection (LoD), defined as the minimum dilution consistently giving a positive result. After verifying the yield of conv.PCR tests, these were carried out with the DNA extracted from the feces samples. 30 ng of DNA extracted from the reference strains were included as positive amplification controls, 30 ng of Clostridium perfringens DNA as control of exclusiveness for clostridia species having the same tropism (gastrointestinal tract) and UltraPure DNase/RNase-free distilled water (Invitrogen) as negative PCR control. PCR products were visualized by horizontal electrophoresis on 2% agarose gels (v/v) stained with SYBR Safe (Invitrogen). Hyperladder V (Bioline) was used as molecular weight pattern.

The third molecular test was also PCR-based, though quantitative (qPCR) (Figure 1); it involved using primers and Taqman probe targeting 16S.rRNA (qPCR.16S) according to previous reports (Kubota et al., 2014), except for the quencher, as TAMRA was replaced by BHQ1 (554 nm maximum absorption), widely recommended for use with the fluorophore FAM (495 nm excitation; 515 nm emission) (Marras, 2006) (Table 1A).

qPCR.16S amplification involved using 1X FastStart Universal Probe Master mix (Roche), 200 nM each primer, 200 nM probe (Kubota et al., 2014), 3 μL sample, at 25 μL final reaction volume. Table 1B lists the thermal profiles. qPCR tests were carried out on a CFX96 Real-Time PCR Detection System (BioRad). Amplification efficiency was evaluated by constructing a standard curve from the DNA extracted from a suspension of the ATCC BAA-1870 strain in 1X 300 μL PBS, having OD₆₀₀ equivalent to 4 × 10⁷ cells per mL. An initial 36 ng concentration was used for 10²–10⁻⁴ serial dilutions (in duplicate); these were then used for the amplification. Linear regression analysis of the standard curve results led to determining the coefficient of correlation (R²), the Y-intercept and the slope (S) of the logarithmic phase of amplification, as reaction efficiency measurement.

After verifying qPCR.16S test efficiency, this was used for all samples (in duplicate) using 3 μL DNA extracted from feces samples as amplification template. The standard curve was included in each of the test's runs for monitoring test efficiency. The amplification controls previously described for conv.PCR were included during the amplifications. Positive samples' threshold cycle (Ct) was used for calculating the initial amount of DNA copies, comparing this to standard curve Ct (absolute quantification).

The toxigenic profiles of samples positive for any CDI molecular detection test were determined by conv.PCR amplification of six molecular markers (Figure 1). Four of these markers fell within PaLoc regions, following the scheme proposed by Griffiths et al. (2010); the two remaining regions encoding binary toxin subunits fell within CdtLoc (Stubbs et al., 2000). The same amplification conditions described for CDI detection tests by conv.PCR were used for amplifying these markers. Table 1A lists the sequences for the primer sets used for each test. The thermal profiles were reported by the authors and are described in Table 1B.

In silico analysis was also used; this involved using sequences from CD strains' complete genomes, downloaded from PATRIC (Pathosystems Resource Integration Center) website's bacterial genomics database (Wattam et al., 2014). The C. difficile 630 (NCBI Taxon ID: 272563, Genome ID: 272563.8) reference genome was used for analyzing the primers used in molecular detection of CDI and for the markers falling within PaLoc (tcdA, tcdB, and tcdC). The C. difficile F548 (NCBI Taxon ID: 1232195, Genome ID: 1232195.4) non-toxigenic strain's complete genome sequence was used for analyzing primers targeting cdd1/cdu1 (indicating the lack of PaLoc) while the C. difficile 2007855 (NCBI Taxon ID: 699033, Genome ID: 699033.6) strain's complete genome sequence was used for analyzing primers targeting cdtA and cdtB, encoding binary toxin subunits. A first in silico analysis phase was aimed at identifying primers' annealing sites on the respective complete genome sequences. A second phase involved a description of the primers' basic characteristics [percent of GC content (%GC) and calculated melting temperature (GC+AT Tm)] and the probability of artifacts occurring during PCR amplification (hairpin loops, dimers, bulge loops and internal loops), both independently as well as in combination for each pair of primers). Gene Runner 3.05 software (http://www.generunner.com) was used for the first two in silico analysis components. The BLAST tool was used during a third phase for identifying potential amplification targets.

Descriptive statistics were used for analyzing the results, calculating frequencies and percentages (along with their corresponding confidence intervals) regarding a positive result from the different tests (as categorical variables) and means with their standard deviations (SD) for infection burden (as continuous variable). The frequency of events of interest was expressed considering the total of samples positive for each test, regarding the total of samples collected, overall (complete set of samples) or by population (HCFO/CO). Infection frequency identified in each population using the different tests was contrasted by comparing means between independent populations. qPCR.16S results were used for evaluating CDI burden means distribution per population (CO and HCFO), reported as 25 and 75 quartile values (RIQ) regarding absolute DNA amount (ng/μL) and on logarithmic scale. The Mann–Whitney U test was used for analyzing the difference between means for comparing CDI burden means according to outcome (HCFO/CO) as the results did not have a normal distribution. Agreement between the tests used for detecting CDI was evaluated by calculating agreement percentages, accompanied by their corresponding Kappa coefficients (κ), standard error (EE), 95% confidence intervals and p-values, as parameter regarding their statistical significance. The tests' Kappa coefficients were compared for determining an overall Kappa coefficient for the tests evaluated here. STATA 11 software was used for analysis, fixing a 0.05 significance level.

In silico analysis of the primer sets used for detecting CDI revealed that those targeting 16S.rRNA (by conv.PCR and qPCR) had 11 recognition sites in the C. difficile 630 reference strain sequence compared to the primer set targeting gdh (conv.gdh) which only had one recognition site. Aligning the 11 theoretical amplification products extracted from this strain's sequence showed that the gene copies within the genome had one change in three of the eight copies giving 99.4% identity in both cases (156 bp amplification products for conv.16S and 155 qPCR.16S). This variable position was not located within conv.16S primer set annealing site; however, it was located within qPCR.16S reverse primer annealing site, related to the degenerate design in this position. Figure 2A shows the primer sets' annealing sites (predicted from the C. difficile 630 reference strain genome); the exact coordinates are given in Supplementary File 1-sheet 1.

In silico analysis also enabled verifying that the primers proposed by Griffiths et al. for describing toxigenic profiles based on the PaLoc sequence (Griffiths et al., 2010) had a single annealing site within the C. difficile 630 strain 20 Kb locus (Figure 2B; Supplementary File 1-sheet 2) and that the region was flanked by cdd1/cdu1, targeted by lok1/3 primers. The lok1/3 primers' annealing site in non-toxigenic strains was then confirmed from C. difficile F548 strain sequence (Figure 2C; Supplementary File 1-sheet 2). The position of the molecular markers targeting the binary toxin (Stubbs et al., 2000) was verified from the C. difficile 2007855 strain sequence (Figure 2D; Supplementary File 1, sheet 2).

Gene Runner oligo analysis showed that the primers used in the conv.16S test sampled most amplification artifacts, dimers able to generate the direct primer being of greater interest for the procedures. These could happen at probable temperatures (6–46°C) during amplification. Even though some artifacts could have been generated in the other primers used in this study, they were predicted at non-probable temperatures during the procedures (< −20°C). Supplementary File 2 gives oligo analysis results.

BLAST search verification of the most probable amplification targets identified that most primers used in CDI detection tests as exclusively matching CD sequences, except for conv.gdh primers which (even though giving better BLAST results with CD sequences) recognized sequences from other species, such as Vulcanisaeta distributa and Arabis alpina by conv.gdh-F and Lentibacillus amyloliquefaciens and Wickerhamomyces ciferrii by conv.gdh-F (query cover was < 90.0% and E ≥ 0.250). BLAST analysis of the primers used for describing toxigenic profiles also revealed recognition with other species (query cover was < 90.0% and E ≥ 1.00). Supplementary File 3 gives BLAST analysis of primer sequences for primers used in CDI detection tests and Supplementary File 4 those used in describing toxigenic profiles.

The CDI detection algorithm was used with the set of samples collected (n: 217) (Figure 1). The populations were classified according to the possible source of infection acquisition, following the parameters described in the methodology (Cohen et al., 2010); 36.4% (n: 79) of the samples were HCFO and 65.6% (n: 138) CO. Regarding HCFO, 75.9% (n: 60) came from intensive care unit (ICU) patients and 24.1% (n: 19) from other services' inpatients. Most CO samples were collected from the emergency service (85.5%; n: 118), others from outpatient consultation (14.5%; n: 20).

Detecting CDI by SCM gave 52.1% overall frequency (n 113: 45.2–58.9 95%CI). Figure 3A lists infection frequency according to infection source (HCFO or CO). A set of 58 samples (49.6%: 40.2–59.0 95%CI) did not pass VCS screening as they did not grow on plates containing TSA and 5% sheep blood or because their morphology was not as expected (gram positive bacillus with possible presence of spores). Only 59 of the samples initially positive by SCM could thus be verified as VCS positive (50.4%: 41.0–59.85 95%CI), i.e., 27.2% (21.4–33.6 95%CI) overall frequency. Figure 3B describes VCS results per population.

Amplifying molecular markers by conv.PCR was as expected when DNA from control strains was used. The times for each cycle during exponential amplification (5–10 s for each phase) were modified for conv.PCR thermal profiles for improving amplification (obtaining single bands at expected height); conv.PCR LoD determination showed that conv.16S test gave 3.6 × 10⁻⁷ amplification products and three orders of magnitude lower than those occurring with conv.gdh test which gave results lower than 3.6 × 10⁻⁴. Supplementary File 5 gives the conv.PCR LoD results.

In spite of in silico verification, tests on DNA extracted from feces samples identified amplified products having a different size to that expected and sometimes multiple bands. Samples having single bands of different amplification sizes were selected to clarify these findings; Sanger sequencing was then used on them, using the same primers. At least three PCR products were sequenced for each marker. Chromatogram analysis (which included BLAST search) identified bands agreeing with CD in most cases (< 0.000 E-values), except for some amplification products matching Faecalibacterium prausnitzii, Prevotella scopos and Burkholderia sp., having better E-values than those found for CD, or even CD results but with E ≥1.00. Supplementary File 6 (conv.16S) and Supplementary File 7 (conv.gdh) show BLAST sequence results. As these bands consistently matched CD, they were considered positive for CDI; all samples had amplification at the heights confirmed by sequencing.

CDI frequency determined by conv.PCR molecular tests was slightly higher for the test targeting 16S, (positive in 45.6% of the samples; n 99: 38.9–52.5 95%CI) compared to the test targeting gdh which was positive for 42.4% of the samples (n 92: 35.7–49.3 95%CI). Figure 3C (16S) and Figure 3D (gdh) show infection frequency for each population by conventional PCR molecular tests. Analysis of CDI frequency per population showed that the in vitro culture approach and conventional PCR targeting 16S gave higher frequency in the HCFO population; however, conv.gdh gave greater frequency regarding positivity for CO, all differences being statistically significant (p < 0.05).

Analyzing real-time PCR results (as qualitative outcome—the presence/absence of infection) revealed considerably lower infection frequency than the other two molecular tests (22.6% of positive samples; n 49: 17.2–28.7 95%CI). Figure 3E gives CDI frequency distribution per population.

Quantitative analysis of qPCR test results led to determining CDI burden according to outcome (HCFO vs. CO) by comparing means. The results showed that most of the data for the group of patients from HCFO was around 0.086 ng/μL (IQR 0.0006-484.89), being slightly lower than for CO (mean bacterial burden 0.187 ng/μL: IQR 0.004- 13.06) (Figure 3F), however, such differences were not statistically significant (p 0.787). Figure 3G gives mean burden results, also represented on logarithmic scale.

A test in conventional format was used as strategy for verifying qPCR.16S results; negative samples having a positive result for some conv.PCR were used for this test. The reaction conditions and thermal profiles described for qPCR.16S were used, but probe volume was replaced by UltraPure DNase/RNase-free distilled water (Invitrogen) and amplification product generation verified by horizontal electrophoresis. Interestingly, verification gave an amplification product at the expected height in most samples.

Analyzing agreement amongst molecular tests used for detecting CDI showed that agreement between tests was limited; conventional PCR targeting the 16S.rRNA molecular marker (conv.16S) gave the highest agreement percentages (statistically significant results) (Figure 4). The best comparison result was con.16S with conv.gdh (the other conventional test) where agreement was 65.4% (p 0.000). The conv.16S test had good agreement, even with in vitro culture results (SCM and VCS). Interestingly, the results of comparing conv.16S to qPCR.16S (in spite of having good agreement: 64.0%) were not statistically significant (p 0.2347), in spite of targeting the same molecular marker. The other comparison having statistically significant results arose from comparing conv.gdh to qPCR.16S (60.8% agreement: p 0.0176). Figure 4 gives test agreement results. Comparing kappa coefficients between all tests (9 comparisons) gave an estimated 0.1272 overall Kappa coefficient (0.086–0.168 95%CI: p 0.0095).

Samples having a positive result by any molecular test (n: 147) were used for amplifying toxin-encoding genes by conventional PCR. A set of 21 samples (14.3%) was toxin typed as negative for all markers evaluated (i.e., by the four molecular markers targeting PaLoc and the two detecting binary toxin-encoding genes).

Amplification results led to identifying similar patterns that in the CDI detection tests; in spite of amplification products being obtained at the expected heights (when control strains' DNA was taken as template and from most samples) some products having different sizes to expected size were also identified. The set of lok1/3 primers (targeting cdd1 and cdu1 genes) designed to amplify a 769 bp fragment only in the absence of PaLoc (as proposed by the authors) (Griffiths et al., 2010) had the most unexpected pattern since amplification size was ≈300 bp in most cases. Sanger sequencing was used for verifying products having unexpected size (analysis including BLAST search); results were similar to what happened for CDI detection tests where most sequences matched CD (< 0.000 E-values) and also matched other species. Supplementary File 8 (tcdA), Supplementary File 9 (tcdB), Supplementary File 10 (tcdC), and Supplementary File 11 (lok1/3) give verifying sequences for the markers used for describing toxigenic profiles. All samples giving amplification products having a size matching those verified by sequencing were considered positive for each marker in the toxigenic profile description scheme.

Describing toxigenic profiles led to identifying circulating strains having toxigenic potential in both study populations (HCFO and CO). The most frequently occurring molecular marker was tcdB, being positive in 53.7% (n 79: 45.3–62.0 95%CI) of samples having a toxinotyping result, followed by tcdA (49.0% frequency; n 72: 40.7–57.3 95%CI). Another interesting finding regarding the test targeting cdd1/cdu1 was that the mostly positive result were in combination with other PCR targeting PaLoc (45 of the 50 positive samples had at least another positive test targeting PaLoc). Frequency for genes encoding binary toxin were differentially found, being cdtA positive in 42.2% (n 62: 34.1–50.6 95%CI) of samples having a toxigenic profile, while cdtB was positive in 23.1% (n 34: 16.6–30.8 95%CI). Figure 5 gives a complete description of the samples' toxigenic profiles. Similar to what happened with the amplification products for both, the CDI detection tests and the PaLoc organization description. Amplicons from different expected sizes were found for the two genes of the binary toxin. The results of the sanger sequencing of these products showed a behavior similar to the previous cases [Supplementary File 12 (cdtA) and Supplementary File 13 (cdtB)], ratifying the limitations of these tests to be applied on stool samples.

The results obtained from the samples were informed to the participating health centers, where they were made available to the treating physicians, who took the corresponding therapeutic measures according to the mandatory health plan of Colombia.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.