|Home | About | Journals | Submit | Contact Us | Français|
Cultivation-based assays combined with PCR or enzyme-linked immunosorbent assay (ELISA)-based methods for finding virulence factors are standard methods for detecting bacterial pathogens in stools; however, with emerging molecular technologies, new methods have become available. The aim of this study was to compare four distinct detection technologies for the identification of pathogens in stools from children under 5 years of age in The Gambia, Mali, Kenya, and Bangladesh. The children were identified, using currently accepted clinical protocols, as either controls or cases with moderate to severe diarrhea. A total of 3,610 stool samples were tested by established clinical culture techniques: 3,179 DNA samples by the Universal Biosensor assay (Ibis Biosciences, Inc.), 1,466 DNA samples by the GoldenGate assay (Illumina), and 1,006 DNA samples by sequencing of 16S rRNA genes. Each method detected different proportions of samples testing positive for each of seven enteric pathogens, enteroaggregative Escherichia coli (EAEC), enterotoxigenic E. coli (ETEC), enteropathogenic E. coli (EPEC), Shigella spp., Campylobacter jejuni, Salmonella enterica, and Aeromonas spp. The comparisons among detection methods included the frequency of positive stool samples and kappa values for making pairwise comparisons. Overall, the standard culture methods detected Shigella spp., EPEC, ETEC, and EAEC in smaller proportions of the samples than either of the methods based on detection of the virulence genes from DNA in whole stools. The GoldenGate method revealed the greatest agreement with the other methods. The agreement among methods was higher in cases than in controls. The new molecular technologies have a high potential for highly sensitive identification of bacterial diarrheal pathogens.
The detection of pathogens in stools has been performed traditionally using techniques developed and applied in high-income countries. These methods, developed over the past century and rigorously tested and approved for use for clinical diagnostic purposes, include the standard clinical culture for identifying pathogens, sometimes after an enrichment step, on a selective or differential medium that inhibits the growth of many bacteria and is permissive to the pathogen of interest. After sufficient growth is attained, the colonies are tested by PCR for the selected genes or agglutinated with antisera specific for the pathogen. Alternatively, enzyme-linked immunosorbent assays (ELISAs) or PCR methods may be used directly on fecal samples. Despite the widespread use of these methods in high-income countries, their adoption for routine use in low-income countries has been very limited (1, 2). The availability and costs of technical expertise, supplies, equipment, and equipment maintenance are factors which have limited the adoption of diagnostic microbiology methods in these settings (3). Simple, inexpensive, and reliable point-of-care diagnostic tests that rapidly identify treatable etiologies of diarrhea are needed in low-income countries. We sought to compare the abilities of four different emerging technologies to identify the presence of enteric pathogens in stools from children in low-income countries.
The new molecular methods have great appeal as the technologies are increasingly robust and affordable and potentially more precise. Three alternative technologies are based upon detection of specific DNA sequences. One technology, the GoldenGate assay (Illumina, San Diego, CA), has been adapted for microbiological detection (4). The assay is based on the specific hybridization of primers to target DNA in the solution and the ligation of two primers to form a product that can be amplified with efficiency equal to that for the other ligated products. Amplified ligation products are hybridized to microbeads and detected in an optical microarray format in a Bead Express station (Illumina). The second technology is the Ibis Universal Biosensor assay (5). It employs a series of PCR primers to amplify selected targets; after amplification, the target DNAs are detected by electrospray ionization-mass spectrometry (ESI-MS) (Ibis, San Diego, CA). The third technology is high-throughput sequencing of fecal community DNA after amplification of 16S rRNA genes (6). Sequences are then processed to identify operational taxonomic units (OTUs). This method can distinguish between genera and some species of bacteria but cannot distinguish among pathotypes of a single species, e.g., enteroaggregative Escherichia coli (EAEC), enterotoxigenic E. coli (ETEC), enteropathogenic E. coli (EPEC), and Shigella spp.
These three new technologies were compared with conventional methods on stool samples collected from case and control children in The Gambia, Mali, Kenya, and Bangladesh during the Global Enterics Multicenter Study (GEMS) funded by the Bill and Melinda Gates Foundation (7, 8). Comparisons were done for seven detectable pathogens, EAEC, ETEC, EPEC, Shigella spp., Campylobacter jejuni, Salmonella enterica, and Aeromonas spp. Our analyses showed that no method is clearly superior, but the GoldenGate method agreed most frequently with the other methods.
Stool samples were collected as part of the GEMS. Children less than 5 years old seeking care for episodes of diarrhea (passage of ≥3 abnormally loose stools within the previous 24 h) at health care facilities serving field site populations were considered. Those who presented within 7 days of the onset of their illness and met the case definition of moderate to severe diarrhea were eligible for enrollment. The episodes meeting the inclusion criteria included exhibition of the most objective signs of dehydration, including sunken eyes, a loss of normal skin turgor, a decision to initiate intravenous hydration or hospitalize the child, or the presence of blood in the stool (dysentery). Within 2 weeks of enrollment of a case, 1 to 3 randomly selected control children without diarrhea, matched to the case by age, gender, and village, were enrolled from the community (8). Stool samples were cultured 7 h on average after defecation. The stool samples were tested for 15 potential diarrheagenic pathogens and their subtypes by culture and subsequent PCR and immunological methods (7). After homogenization of fecal samples by vortexing with 3-mm glass beads (Sigma Life Science, St. Louis, MO), bacterial cell walls were disrupted using a bead beater with 0.1-mm zirconium beads (catalog no. 110791012; BioSpec Products, Bartlesville, OK). The cell slurry was centrifuged, and the supernatant was removed and processed using the Qiagen (Hilden, Germany) QIAamp DNA stool extraction kit. The extracted DNA was precipitated, and the DNA was shipped to the University of Maryland (Baltimore, MD), following the appropriate in-country approvals. DNA was then resuspended, and aliquots were distributed to Ibis and Illumina. For the current study, we utilized specimens from The Gambia, Mali, Kenya, and Bangladesh, chosen based on geographic representation from Asia and sub-Saharan Africa and close collaboration with the sites' principal investigators.
The clinical pathogen identification methods were completed in each of the participating countries (The Gambia, Mali, Kenya, and Bangladesh) and were described in detail by Panchalingam et al. (7). Briefly, the stool specimens were collected in sterile containers and examined within 24 h of passage. The stool specimens were stored at 2 to 8°C while in transit to the laboratory. Conventional bacteriological, immunological, and molecular methods were used to identify bacterial pathogens. Vibrio cholerae, Vibrio parahaemolyticus, Aeromonas spp., C. jejuni, Campylobacter coli, Salmonella spp., Shigella spp., and diarrheagenic E. coli were isolated from the appropriate selective media and identified by standard biochemical tests. The standard protocol was to test three putative lactose-positive and indole-positive E. coli-like colony morphologies selected from the MacConkey plate for diarrheagenic E. coli as previously described (7). The species and subtypes were confirmed by serotyping (for Shigella and Salmonella spp.) with commercially available antisera (Reagensia, Solna, Sweden, and Denka Seiken, Tokyo, Japan) and by PCR tests for both the heat-labile and heat-stable enterotoxin producers of ETEC, typical and atypical EPEC, and EAEC.
Performed at Illumina, the GoldenGate assay consists of allele-specific oligonucleotides (ASOs) or locus-specific oligonucleotides (LSOs) that are hybridized to the target sequence and ligated together, and the ligated products are PCR amplified using primers attached to the 5′ ends of the ASOs or LSOs. The primers were designed using GoldenGate design software (see reference 9 for details in the GoldenGate assay workflow). In addition, a “Laguna” probe was designed to hybridize to the target DNA, specifically avoiding sequences from human and the various target organisms. The probe is attached to streptavidin and, when precipitated, increases the purity of the target DNA. Multiple probes sets were designed for each target organism.
Sixty-nine sets of probes were simultaneously hybridized to the target DNAs following the methods described in the GoldenGate Genotyping with VeraCode Technology instructions (4). After the beads were scored in the array, the background intensity was calculated for each sample by taking the 50th percentile for all probes present, because it is unlikely that more than half of the targets would be present in a sample. After background subtraction from all the probes, the mean intensities and standard deviations were calculated for all the probes. Greater than 3 standard deviations was the threshold for a probe to detect a target. To call a target present, at least half of its probes had to be detected.
Performed at Ibis, this method uses an automated process for aliquoting extracted DNAs, amplifying DNA using PCR primers for the detection of microorganisms of interest, and then subjecting the amplified products to ESI-MS (5). Six pairs of primers targeting 16S and 23S rRNA genes were used to identify microbes broadly (Gram positive, Gram negative, aerobic, and anaerobic), followed by primer pairs to characterize 17 additional genes (ctxA, ctxB, east-1, east-2A, estA1, eltA, eltB, stx1A, stx1B, aggR, aatA, eae, invA, ipaH, ipaA, ipaC, and ipaB) used to identify the pathogen-specific markers. The mass spectrometer measures the amplicons of the PCR product to calculate a molecular weight, and the composition of the nucleotides can be deduced for each amplicon present. The Universal Biosensor assay utilizes a database of sequence base composition of known microorganisms to determine which microorganisms are present. The Universal Biosensor assays tested for broad groups of microorganisms, including Campylobacter spp., Clostridium spp., E. coli, Escherichia spp., nontyphoidal Salmonella spp., Salmonella enterica serovar Typhi, Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, V. cholerae, V. parahaemolyticus, ETEC, EPEC, EAEC, and Yersinia spp. and the genes ctxA, ctxB, eae, east-1, invA, ipaB, ipaC, ipaD, and ipaH.
The DNA from stool samples was amplified using “universal” primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 338R (5′-CATGCTGCCTCCCGTAGGAGT-3′). A single reverse primer and a set of 96 barcoded forward primers were used. The presence of amplified products was confirmed on agarose gels, and aliquots from each of the 96 samples were added in equimolar amounts to a final mix. This was sequenced using the FLX sequencing kit and the 454 FLX sequencing platform (454 Life Sciences, Branford, CT). Individual reads were filtered for quality using custom in-house scripts that perform the following checks suggested by Huse et al. (10): (i) sequences containing any ambiguous bases (N) were removed, (ii) sequences that were shorter than 75 cycles of the 454 instrument were removed (each cycle yields an average of 2.5 bp, depending on the sequence composition), and (iii) sequences for which a barcode could not be identified were removed (10). The remaining sequences were separated into sample-specific sets according to their barcodes, and the barcodes were removed. The conservative OTUs were constructed by pooling together the sequences from all samples and then were clustered using DNACLUST with the default parameters (98% identity clusters) (11). Pooling the samples ensures that the definition of an OTU is consistent across all the samples. To obtain taxonomic identification, a representative sequence from each OTU was aligned to the Ribosomal Database Project (RDP) (release 10.4 [see http://rdp.cme.msu.edu]) using BLASTN with long word length (-W 100) in order to detect only nearly identical sequences. A reference database of 16S rRNA gene sequences from known enteric pathogens was manually curated by extracting full-length 16S rRNA gene sequences from isolated genomes in the RDP (on 1 April 2011) for Clostridium difficile, V. cholerae, V. parahaemolyticus, S. enterica, Salmonella Enteritidis, Salmonella Typhimurium, C. jejuni, C. coli, Helicobacter pylori, and Yersinia enterocolitica. All sequences from each OTU were searched against the full RDP and the enteric pathogen subset using BLAST (best hit with similarity, ≥97%).
The percentages of samples positive for diarrheal pathogens in each method were determined by dividing the number of samples identified as positive by the total number of samples tested. Cohen's kappa was used to describe the agreement between two detection methods. Cohen's kappa is advantageous because it does not require the specification of a gold standard but simply quantifies the agreement between binary outcomes of tests (positive or negative), taking into account agreement occurring by chance (12). The kappa term ranges from −1 to 1 and can be negative if the agreement is less than what would be expected by chance. The following labels were assigned to the corresponding ranges of kappa strength: poor agreement, <0; slight, 0.0 to 0.20; fair, 0.21 to 0.40; moderate, 0.41 to 0.60; substantial, 0.61 to 0.80; and almost perfect, 0.81 to 1.00 (12, 13). All statistical analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC).
The frequency of detection of each pathogen varied from one method to the next (Fig. 1A and andB).B). The comparisons were based on the percent frequency of detection in order to normalize the methods because different numbers of specimens were tested. We tested 3,610 stool samples with the culture method, 3,179 DNA samples with the Universal Biosensor method, 1,466 DNA samples with the GoldenGate method, and 1,006 DNA samples with the 16S rRNA gene survey method (Table 1). Although attempts were made to identify V. cholerae and V. parahaemolyticus on each sample, no method detected more than 10 specimens with these pathogens, so no comparisons were made. Identification of EHEC was attempted with only the Universal Biosensor method, so these results are not presented.
The pathotypes of E. coli and Shigella spp. were differentiated on the basis of virulence genes specific to each subtype. As the 16S rRNA gene survey method does not detect virulence genes, it was not compared to the other methods for these pathogens. As seen in Fig. 1A, the culture-based method detected Shigella spp., EPEC, ETEC, and EAEC in a smaller proportion of the samples than either of the methods (GoldenGate or Universal Biosensor) based on detection of the virulence genes from DNA from whole stools. For Shigella spp., EPEC, and ETEC, the proportions of samples in which the virulence genes were detected were very similar for the Universal Biosensor and GoldenGate methods, while for EAEC, the GoldenGate method detected the virulence genes in a substantially higher proportion of samples.
Two pathogens, C. jejuni and S. enterica, could be detected by all of the four methods (Fig. 1B). Aeromonas spp. were detected by the culture-based method, the 16S rRNA gene survey, and the GoldenGate method, and C. difficile was detected by the 16S rRNA gene survey, the GoldenGate method, and the Universal Biosensor method. The Universal Biosensor method rarely detected C. jejuni and C. difficile, but it identified S. enterica in the highest proportion of cases. As with the pathotypes, the culture-based method consistently detected pathogens in fewer stool samples than did the GoldenGate method. The 16S rRNA gene survey method detected C. jejuni, C. difficile, and S. enterica in proportions similar to those detected by the GoldenGate method.
Shigella spp. and EPEC were detected relatively more consistently among the methods. Figure 2A and andBB shows Venn diagrams that highlight the numbers of Shigella spp. and EPEC identified by each method and the overlap of the methods. Generally, as the number of detection methods that positively identified the pathogen increased, so did the proportion of cases within that group.
The agreement between methods was determined by comparing the positive and negative results for each sample successfully tested by each possible pair of methods. The best agreement exceeding the threshold (0.41) for moderate agreement was observed in four instances; the Universal Biosensor and GoldenGate methods agreed on EPEC, ETEC, and Shigella spp., and the GoldenGate and 16S rRNA gene survey methods agreed on C. jejuni. Fair agreement (kappa values, 0.21 to 0.40) was observed between the GoldenGate and culture methods for five pathogens (EPEC, Shigella spp., C. jejuni, Aeromonas spp., and S. enterica), and between the Universal Biosensor and culture methods for two pathogens (EPEC and Shigella spp.). We found that the agreement between methods was higher in the cases than in the controls in 21 of the 27 possible comparisons (Fig. 3).
To determine whether there were any site-specific biases, kappa values were computed for the data from each individual site (Table 2). After a comparison across all four sites, the GoldenGate method appears to be the one that contributes to relatively high kappa values most frequently. The GoldenGate and Universal Biosensor methods detect pathogenic ETEC, EPEC, EAEC, and Shigella spp. at similar rates and at greater-than-conventional culture rates. The GoldenGate and 16S rRNA gene survey methods detected Aeromonas spp., C. jejuni, Salmonella enterica, and Clostridium spp. at similar rates, greater than the culture-based rates.
The survey detected a total of 97,666 nonunique OTUs. Of these, only 25,834 that were detected in more than 5 samples or had more than 20 sequences in at least 1 sample were included in further analyses. The number of OTUs per sample averaged 465.27 and ranged from 3 to 1,439. For the pathogens C. jejuni, C. difficile, S. enterica, and Aeromonas spp., representative 16S rRNA gene sequences were selected from the Ribosomal Database Project, and OTUs that varied by less than 2% from these sequences were assigned to that species. This was not done for OTUs matching E. coli, as E. coli strains can be either commensal or pathogenic. Seventy OTUs were designated C. jejuni, 8 C. difficile, 3 S. enterica, and 3 Aeromonas spp. C. jejuni was detected most often, being present in 25% of the total number of samples, with C. difficile, S. enterica, and Aeromonas spp. being present in 7%, 2%, and 2%, respectively.
Culture-based methods are commonly used for detecting enteric pathogens, and in our hands, the standard tests detected many pathogens. It is notable, though, that there were numerous samples where the alternative molecular techniques detected the apparent presence of a pathogen despite the absence of that pathogen according to the standard clinical method. However, that same statement could be made for every other method as well; each method failed to detect a specific pathogen in a sample where one or more of the other methods did detect the same pathogen. A priori reasons abound for false positives and false negatives for each of the methods; the potential sources include contamination, PCR failure, limits to detection, genetic variation, and complexity of the composition of the stools (14). Nevertheless, the different methods of detection often detect the same pathogen in the same stool sample, indicating that the methods have some agreement. We interpret these data to indicate that, at this time, there is no gold standard for detecting pathogens. The lack of a gold standard limits the analyses.
When the culture methods identified the target organisms, we were able to confirm that the microbes were truly present. However, a negative result may have been either true, because the organism was not present, or false, because (i) the organism did not grow as a result of inadvertent inhibition by the selective medium used, (ii) the specimen transport conditions compromised the organism, or (iii) laboratory inefficiencies were present. Published results and our own unpublished results revealed that testing a greater number of E. coli isolates (e.g., testing 5 colonies instead of 3 picked from a MacConkey plate) will increase the rates of recovery, thereby confirming the possibility of some false negatives (15). Compared to those of other methods, the culture-based results showed fair agreement with the GoldenGate method on Aeromonas spp., C. jejuni, and S. enterica and with both the Universal Biosensor and GoldenGate methods for EPEC and Shigella spp.
The Universal Biosensor and GoldenGate methods are the two most similar methods. Both methods detected EAEC, EPEC, ETEC, and Shigella spp. at much higher rates than did the culture-based method and had the highest rates of agreement for these pathogens. Each method starts with the specific hybridization of two primers to DNA and detects the same virulence genes, although with distinct primers, but they have limits of detection based on the ability of the primers to hybridize. Any sample with the specific DNA of interest below the detection limit will produce a false negative that may be overcome in the culture-based method, which has the potential to enrich and grow a bacterium from a single CFU. False positives will occur if the virulence genes, usually found on mobile elements, are found in bacteria other than the expected pathogen.
The sequence survey method is based on amplifying the 16S rRNA gene of bacteria using universal primers and then counting the presence of the different 16S rRNA gene sequences. In order to be counted, the bacteria must constitute 1/n or more (where n is the number of sequences counted) of the community. Since n is on average 3,900, the bacteria in question must be ~0.03% of the sample, and those near the cutoff level are also subject to sampling error (for an example, see Fig. 1B in reference 16). The method cannot distinguish among the pathotypes of E. coli and Shigella spp. However, the method did distinguish among Aeromonas spp., C. jejuni, S. enterica, and C. difficile as expected, and for the latter three, the method detected these pathogens at levels that were most similar to those of the GoldenGate method. The two methods revealed moderate agreement for C. jejuni. The Universal Biosensor method for detecting C. jejuni and C. difficile is based on the detection of 16S rRNA genes, and the underestimation of these pathogens is likely the result of the complexity of the stool samples such that even in samples from cases, the sequence-based method yielded extremely large numbers of distinct 16S rRNA gene sequences. The amplification of the whole sample for the sequencing approach raises the question of whether an alternative approach, like complete shotgun sequencing using next-generation sequencing, would detect taxa similar to those detected by the other methods.
The agreement between methods was consistently greater in cases than in controls. Almost certainly this observation can be attributed to the higher concentration of a pathogen in the cases than in the controls and to the fact that the higher the concentration the more likely a pathogen is to be detected. Quantification of the level of pathogens in the stool may improve the ability to accurately diagnose the cause of the diarrhea (17). This may have particular importance as the increased levels of detection (seen in the Golden Gate and Universal Biosensor methods) will mean that multiple pathogens will be detected in an increasing number of stool samples. The proportion of hospitalized patients in Kolkata, India, with multiple pathogens using cultures was approximately 30%, and when multiple pathogens were present, there were often strong associations between them (18, 19). With the GEMS revealing many pathogens in control stool samples, quantitation of pathogens may be an important diagnostic criterion.
In conclusion, our results demonstrate that traditional culture-based methods underestimate the numbers of stool samples in which specific enteric pathogens can be detected compared with those by the alternative molecular methods. Not surprisingly, the alternative methods based on detection of specific unique genes produced the most similar results. Methods based on 16S rRNA gene sequences had distinct limitations. However, in the absence of a gold standard, replication and consistency between methods become important criteria. The GoldenGate method revealed the greatest agreement with the other methods. However, improvements in terms of targeted genes and the ability to quantify the number of pathogens present would enhance the usefulness of the method used for diagnosis.
This work was supported by the Bill and Melinda Gates Foundation (grant 42917).
T.K.M., B.B., and I.L. are employees and shareholders of Illumina, the manufacturer of GoldenGate technology, one of the platforms evaluated in this study. J.H., S.M., J.D., I.Y., L.B., R.R., F.L., R.H., D.J.E., and R.S. are employees of Ibis Biosciences, the manufacturer of Universal Biosensor technology. J.P. has received funding for conference travel and accommodation from Illumina.
M.A., D.A., J.O., B.T., S.P., M.M.L., K.K., U.I., C.E., M.D., M.A., F.A., M.T.A., R.A., S.S., J.B.O., E.O., J.J., E.M., R.O., C.E.O., M.A.H., R.F.B., and D.S. contributed to the collection, culture, and data management; B. Lindsay, M.P., A.W.W., V.M., S.L., J.N.P., B. Liu, R.R., M.D.S., M.U., J.G.M., J.P., O.C.S., and J.P.N. performed 16S rRNA gene data collection and analysis; B.B., I.L., and T.K.M. provided GoldenGate data; J.H., S.M., J.D., I.Y., L.B., R.R., F.L., R.H., D.J.E., and R.S. provided Universal Biosensor data; M.P., A.W.W., V.M., K.K., M.M.L., L.S.M., D.J.E., T.K.M., R.S., M.A.H., R.F.B., J.N.P., J.G.M., D.S., O.C.S., and J.P.N. provided the study design; B. Lindsay, L.S.M., M.P., and O.C.S. performed statistical analyses; and B. Lindsay, M.P., V.M., O.C.S., and J.P.N. wrote the article. We acknowledge the contributions of Eric Mintz, Michele Parsons, and Cheryl Bopp, Division of Foodborne, Waterborne, and Environmental Diseases, U.S. Centers for Disease Control and Prevention, Atlanta, GA.
The use of trade names and commercial sources is for identification only and does not imply endorsement by the Centers for Disease Control and Prevention or the U.S. Department of Health and Human Services. The findings and conclusions in this presentation are those of the authors and do not necessarily represent those of the Centers for Disease Control and Prevention.
Published ahead of print 24 July 2013