Sample collection, processing, antimicrobial susceptibility testing

Stool specimens or rectal swabs were collected into sterile, wide-mouth collection cups and aliquoted into thirds (Cary-Blair transport media, 10% formalin for parasitology, and a vial for freezing at -20C for virology) upon enrollment; previous studies showed no differences in frequency of bacterial isolation between stool samples and rectal swabs [17]. Samples were stripped of all identifiers and were assigned accession numbers before transportation to the WRAIR Microbiology Hub laboratory in Kericho (MHK) within 72 hours of collection. Samples were then plated on primary, selective, and differential media. MacConkey, MacConkey-sorbitol, sheep blood agar, Hektoen enteric agar, thiosulfate-citrate-bile-sucrose agar, cefoperazone-vancomycin-amphotericin agar, and cefsulodin-irgasan-novobiocin agar were the primary media; no specific enrichment step was performed as part of the normal workup. At 24 and 48 hours, colonies were subcultured, Gram stained, and subjected to biochemical testing (indole production, Voges-Proskauer reaction, o-nitrophenyl-Δ-D-galactopyrandoside production) before analysis on Microflex MALDI Biotyper (Bruker Daltonics, Millerica, MA, USA) and MicroScan WalkAway40 (Siemens Healthcare, Sacramento, CA, USA) systems for identification and antibiotic susceptibility testing (AST), respectively. MIC 44 and NC 66 panels were used with LabPro software updated for 2015 CLSI breakpoints [18] and automated interpretation of results. Laboratory personnel performing susceptibility testing were enrolled in External Quality Assurance/Proficiency Testing for both College of American Pathologists (three cycles/year) and United Kingdom National External Quality Assessment Service (monthly). Weekly quality control for AST was performed using recommended ATCC strains [18].

Study sites

Samples were collected from eight Kenyan clinical sites participating in the Walter Reed Army Institute of Research (WRAIR), University of Washington/Kenya Institute of Medical Research Institute (KEMRI) collaborative research group enteric surveillance programs. These surveillance sites serve diverse communities: Mbagathi District Hospital serves a highly urban population near the center of Nairobi. The Eldoret-based clinic at Moi Barracks (MBB1) serves military service members and their families in the Kenyan highlands. Kericho District Hospital, also located in the highlands, serves a relatively rural community of tea pluckers and farmers. Kombewa is similarly considered rural. The remaining sites at the district hospitals of Kisumu, Kisii, Migori, and Homa Bay are located in western Kenya near Lake Victoria and serve both urban and rural populations largely subsistent upon agricultural and fishing economies.

Eligibility criteria

Protocol-trained clinical staff at all sites recruited subjects experiencing acute diarrhea (three or more loose stools within a 24 hour period). The cases were recruited only from outpatient populations, and none were admitted to the hospital. Age-matched asymptomatic controls were recruited from the same sites if the subjects had not experienced acute diarrhea within the previous two week period; when possible, controls were healthy siblings close in age to the index case. Participants experiencing (chronic) diarrhea lasting more than 14 days were excluded. Medical histories were captured for a small subset of samples (n = 13). Both cases and controls provided basic clinical, epidemiological (water source and treatment) and demographic (age, gender, residence) information. Enrollment of all subjects required informed consent and custodial assent for subjects under 18 years of age. No diagnostic or therapeutic decisions were based on any phenotypic or genotypic data generated for this study. Work performed on this study was approved by the KEMRI and WRAIR Institutional Review Boards under KEMRI SSC #1549/WRAIR #1549 and KEMRI SSC #2056/WRAIR #1811.

Detection of resistance determinants

The presence/absence of 238 different AMR genes was determined using the Antimicrobial Resistance Determinant Microarray (ARDM) v.2 as previously described [19,20]. Briefly, this microarray was designed for detection of >200 determinants derived from both Gram-positive and–negative bacteria. Chip content covers genes conferring resistance to 15 categories of antimicrobials (β-lactams, aminoglycosides, macrolides, lincosamides, streptogramins, quaternary amines, ansamycins, diaminopyrimidines, antimicrobial peptides, tetracyclines, phenicols, glycopeptides, platensemycin, fluoroquinolones, sulfonamides); several plasmid-borne multidrug efflux pumps are also represented on the chip. Full chip content information is given in [19]. Following sample processing, hybridization, and washing, the signal associated with each probe was determined electrochemically. An AMR gene was identified as detected when > 50% of its representative probes had signals above the mean signal from the lowest 2,128 probes + 3 standard deviations or when >70% of its probes had signals above either of two less stringent thresholds [20,21]. A limited set of detected AMR and integrase genes were confirmed by PCR and DNA amplicon sequencing (S1 Table).

Statistical analysis

Statistical comparisons between populations were performed using two-tailed student's t-tests (assuming unequal variance). Chi-square tests were used to compare binomial proportions in independent samples (2 × n contingency tables). Linear regression was used to compare the number of genes/isolate with age (H_o: slope = 0, tested by student's t-test).

Sample set characteristics

A total of 90 Klebsiella spp. strains were isolated from participants ranging in age from 4 months to 54 years (median age 57 months). Half of the subjects presented with acute diarrheal illness and half were healthy controls. The majority of isolates came from the Kisii and Kisumu sites (37 [41.1%] and 16 [17.8%] isolates, respectively) (Table 1). Thirty-three of the isolates (36.7%) were non-susceptible to at least three categories of antimicrobials and were considered multidrug resistant (MDR) per Magiorakos [22]. One isolate, MHK02590, was considered extensively drug-resistant (non-susceptible to at least one agent in all but two or fewer antimicrobial categories; Table 2) [22]. As a whole, there were no differences between overall MDR phenotypes (P = 0.940) in the strains isolated from subjects with ADI and asymptomatic controls, nor between genders (P = 0.463). Between 80 and 90% of the tested isolates were susceptible to all β-lactams except ampicillin, to one or more aminoglycosides, and to both of the fluoroquinolones tested. Over half were susceptible to tetracycline, but more than 60% were resistant to sulfamethoxazole-trimethoprim (SXT).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of antimicrobial phenotypic susceptibility for diarrheal and control isolates.

https://doi.org/10.1371/journal.pone.0178880.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Metadata and phenotypic antimicrobial susceptibility for individual isolates.

https://doi.org/10.1371/journal.pone.0178880.t002

A total of 46 AMR genes or gene families covering 11 categories of antimicrobials were identified amongst the 90 isolates using a broad-range microarray (Table 3). PCR was used to verify the presence of a select group of genes detected by microarray, as well as ancillary genes associated with specific combinations of AMR determinants (S1 Table). All but six isolates harbored multiple resistance determinants (Table 4). While there were no differences in MDR phenotype between age quartiles (P = 0.336), a small but significant inverse relationship was observed between the total number of genes per isolate and age (P = 0.029; t-test of linear regression), with isolates from younger subjects harboring a larger number of genes. No significant differences in genes/isolate were observed between diarrheal and control isolates (P = 0.458) or between genders (P = 0.184). The disparate numbers of isolates collected at the various sites (n = 4 to n = 37) precluded any statistically valid site-to-site comparisons. However, sites with highest percentages of MDR phenotype, Mbagathi (3 of 4 isolates) and Kisii (17 of 37 isolates), also harbored the widest overall varieties of resistance determinants (28 and 41 determinants, respectively).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of AMR genes in the tested population.

https://doi.org/10.1371/journal.pone.0178880.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. AMR genes present in individual Kenyan Klebsiella spp. isolates.

https://doi.org/10.1371/journal.pone.0178880.t004

Resistance to β-lactams

The ARDM v.2 content comprises probes for 52 β-lactamase genes, including 12 families of extended-spectrum β-lactamases (ESBLs) and 15 carbapenemases. The ARDM detected bla_SHV, a chromosomal gene presumptively carried in all K. pneumoniae [23], in 63 isolates (70%), while PCR detected bla_SHV in an additional fourteen (S2 Table); 13 of the 90 isolates were negative for bla_SHV by both methods, but this may be due to point mutations within the primer regions (PCR) or regions used for hybridization on the microarray. β-lactamase inhibitors such as clavulanate and sulbactam are typically active against Klebsiella SHV-1 and TEM-1 lactamases, but one-third of the isolates tested here showed resistance to at least one of these inhibitors. While such resistance may arise from hyperproduction of β-SHV lactamases [24], this resistance was highly correlated to the presence of bla_TEM (P<0.0001), suggesting either TEM hyperproduction [25,26] or the possible presence of inhibitor-resistant TEM enzymes. The presence of bla_OXA-1-like genes–most often conferring resistance to clavulanate and sulbactam–can also potentially explain phenotypic inhibitor resistance in two strains (MHK01590, MHK05068), although bla_TEM genes are also present in both. However, strain MHK01305—positive for bla_OXA-1-like, bla_TEM, and bla_CTX-M-1 family genes–is broadly susceptible to almost all tested β-lactams and lactam-inhibitor combinations, suggesting that either none of these genes are expressed or that the encoded gene products are non-functional.

Nine strains were resistant to at least one third or fourth generation cephalosporin (Table 1), with six classified as ESBL producers by the MicroScan. Five of the ESBL-producing isolates were positive for bla_CTX-M-1-group genes (confirmed by PCR, see S1 and S2 Tables). An additional three isolates also carried bla_CTX-M-1-family genes, two of which were resistant to the third and fourth generation cephalosporins tested but negative for ESBL production by Microscan; one of these (MHK04922) also carried ampC/bla_DHA, which can mask the ESBL phenotype [27]. One isolate (NTS01708) was positive for the bla_CTX-M-2-family, which was also confirmed by PCR. The bla_CTX-M-2 amplicon sequence (NCBI Accession no. KX377894) identified this gene as encoding a protein most similar to CTX-M-2 (Toho 1), CTX-M-20, CTX-M-56, CTX-M-75, CTX-M-95, CTX-M-165, and KLUA-9. To our knowledge, this is the first time that a gene from the bla_CTX-M-2-family has been identified within Enterobacteriaceae from East Africa. Interestingly, this bla_CTX-M-2-positive isolate were susceptible to both of the lactam/inhibitor combinations tested and all other tested β-lactams except ampicillin, suggesting that this gene was not transcribed or that the encoded proteins was non-functional. None of the 90 isolates were positive for genes encoding the CTX-M-8 and CTX-M-9 families of ESBLs. The preferential carriage of CTX-M-1-type enzymes over other ESBLs agrees with other studies of this region [28,29].

Only three isolates were phenotypically resistant to either imipenem (one isolate) or meropenem (two isolates). However, none of the 15 carbapenemase genes represented on the ARDM v.2 were detected.

Resistance to aminoglycosides

Isolates were tested for the presence of 44 different aminoglycoside resistance determinants. While only nine of the isolates were resistant to the three aminoglycosides tested, a relatively large number harbored genes commonly associated with aminoglycoside resistance: aac(3)-III (five isolates); aac(6)-Ib family (four isolates); aadA1/A2 family (18 isolates); aad(A4) (one isolate); aphA1 (seven isolates); aph3/str(A) (44 isolates); aph6/str(B) (47 isolates), and rmtB (one isolate). As the microarray cannot detect point mutations, we PCR-amplified and sequenced the aac(6)-Ib genes detected in four isolates to confirm that these alleles were not the aac(6)-Ib-cr variant conferring resistance to quinolones. The presence of aac(3)-III was correlated to phenotypic resistance to gentamicin and tobramycin (P < 0.0001) and aac(6)-Ib family genes to amikacin and tobramycin (P < 0.0001). Not surprisingly, the isolate harboring rmtB, which confers pan-resistance to aminoglycosides, was resistant to all three aminoglycosides.

Resistance to tetracyclines, chloramphenicol

Almost half of the isolates were non-susceptible to tetracycline. Phenotypic resistance was positively correlated to the presence of a tetracycline resistance determinant (P < 0.0005), although 10 isolates harboring resistance genes were phenotypically sensitive. Of the 38 tetracycline resistance genes on the ARDM v.2, only four were detected: tet(A) (18%), tet(D) (12%), tet(B) (10%), and tet(G) (1%).

The ARDM v.2 chip also contains probes directed against 20 chloramphenicol resistance determinants. However, only four were detected in the tested population: cmr (32 isolates); two variants of floR originating from different species (one isolate); cmlA (one isolate); and catA1/cat4 (seven isolates). Phenotypic resistance to chloramphenicol was not assessed.

Resistance to quinolones

A single isolate (MHK02590) was resistant to both ciprofloxacin and levofloxacin, while the remainder were susceptible to one (three isolates) or both quinolones tested (86 isolates). The plasmid-mediated quinolone resistance gene, qnrS, was observed in two isolates, of which one displayed intermediate susceptibility for ciprofloxacin. None of the other plasmid-mediated quinolone resistance genes were detected (norA, qnrA, qepA, aac(6)-Ib-cr). The ARDM is unable to identify mutations in gyrase or helicase genes that confer high-level resistance to quinolones.

Genes conferring resistance to macrolides, lincosamides, streptogramins, and ansamycins

Ansamycins and macrolides, lincosamide, and streptogramin (MLS) antibiotics are not typically considered clinically relevant for treatment of Gram-negative infections. However, some researchers have suggested that commensal Gram-negative organisms may serve as a reservoir of AMR genes that can be transferred to other pathogens and organisms responsible for severe intestinal infections [30,31,32,33]. For this reason, the ARDM v.2 chip content includes ten MLS resistance genes derived from Gram-negative species, in addition to 31 MLS resistance genes derived from Gram-positive species. As expected, none of the isolates tested were positive for any of the Gram-positive-derived MLS resistance determinants, but Escherichia coli-derived genes, mph(A)/mph(K), mac(A), and mac(B), were detected in six, 29, and 25 isolates, respectively. All isolates positive for mac(B) also harbored mac(A). PCR amplification and amplicon sequencing confirmed that the microarray-detected mac(A) and mac(B) sequences are analogous to those derived from E. coli (NCBI accession nos. KX377891 through KX377893), although Klebsiella-derived analogs were also detected. Analogous mac(A) and mac(B) genes derived from Klebsiella spp. are only 70% identical to the E. coli genes and can be discriminated from the E. coli-derived genes by hybridization to the ARDM and amplicon sequencing (S2 Table).

Two isolates were positive for the presence of the rifampicin resistance determinant, arr. The presence of arr and mphA/mphK within stool isolates of K. pneumoniae–while not clinically relevant in itself—may portend the spread of azithromycin or rifaximin resistance, respectively, to other intestinal pathogens, potentially limiting the effectiveness of these drugs for treatment of travelers' diarrhea [34,35].

Resistance to sulfonamides, quaternary amines, streptothricin, and trimethoprim

Sixty percent of the tested isolates were resistant to SXT, a first line agent for treatment of enteric infections in many parts of Africa [33,36]. Phenotypic resistance to SXT was highly correlated to the presence of a sulfonamide or trimethoprim resistance determinant (P << 0.0001). Approximately half of the tested isolates harbored at least one of the 28 trimethoprim resistance genes present on the ARDM: dfrA14 (18 isolates), dfrA1 (nine isolates), dfrA5 (seven isolates), dfrA7 or dfrA8 (5 isolates each), and dfrA12, dfrA13/21/22/23 family, dfrA15, dfrA16, and dfrA17 (three or fewer isolates each). The high rate of dfrA14-positive samples observed here contrasts with other studies showing a much higher proportion of dfrA1 and dfrA7 amongst African intestinal isolates [37,38]. Seven isolates harbored multiple dfrA genes.

Present in 52.2% of the tested isolates, sul2 was the most frequently encountered sulfonamide resistance determinant. Sul1 was detected in 28 isolates, 21 of which also harbored sul2. In agreement with other studies of the region [37,39], sul3 was infrequently encountered (1 isolate).

Twenty-seven of the 28 isolates positive for sul1 also harbored qacEΔ1. Although association of qac genes with phenotypic antiseptic resistance is currently under debate, co-carriage of qacEΔ1 with sul1 within the 3'-conserved sequences of many class 1 integrons is often linked to the presence of other resistance genes, presumptively as gene cassettes within the integrons [40]. The presence of intI1 –indicative of a class 1 integron—was confirmed in all qacEΔ1+/sul1+ isolates. IntI1 was detected in 20 additional strains by PCR, indicating the absence of a full 3'-conserved sequence amongst almost half of the integrons detected here (S2 Table). Carriage of class 1 integrons with alternative structures has previously been documented within Kenya, albeit at lower rates [39]. Similarly, co-carriage of dfrA1, aadA1/A2, and sat2 is often associated with the presence of class 2 integrons. PCR amplification of intI2 confirmed the presence of class 2 integrons in the three isolates harboring all three genes.