Gut microbiota and resistome dynamics in intensive care patients receiving selective digestive tract decontamination

Background Critically ill patients hospitalized in an Intensive Care Unit (ICU) are at increased risk of acquiring potentially life-threatening infections with opportunistic pathogens. The gut microbiota of ICU patients forms an important reservoir for these infectious agents. To suppress gut colonization with opportunistic pathogens, a prophylactic antibiotic regimen, termed ‘Selective decontamination of the digestive tract’ (SDD), may be used. SDD has previously been shown to improve clinical outcome in ICU patients, but the impact of ICU hospitalization and SDD on the gut microbiota remains largely unknown. Here, we characterize the composition of the gut microbiota and its antimicrobial resistance genes (‘the resistome’) of ICU patients during SDD. Results During ICU-stay, 30 fecal samples of ten patients were collected. Additionally, feces were collected from five of these patients after transfer to a medium-care ward and cessation of SDD. As a control group, feces from ten healthy subjects were collected twice, with a one-year interval. Gut microbiota and resistome composition were determined using 16S rRNA phylogenetic profiling and nanolitre-scale quantitative PCRs. The microbiota of the ICU patients differed from the microbiota of healthy subjects and was characterized by low microbial diversity, decreased levels of E. coli and of anaerobic Gram-positive, butyrate-producing bacteria of the Clostridium clusters IV and XIVa, and an increased abundance of Bacteroidetes and enterococci. Four resistance genes (aac(6′)-Ii, ermC, qacA, tetQ), providing resistance to aminoglycosides, macrolides, disinfectants and tetracyclines respectively, were significantly more abundant among ICU patients than in healthy subjects, while a chloramphenicol resistance gene (catA) and a tetracycline resistance gene (tetW) were more abundant in healthy subjects. Conclusions The microbiota and resistome of ICU patients and healthy subjects were noticeably different, but importantly, levels of E. coli remained low during ICU hospitalization, presumably due to SDD therapy. Selection for four antibiotic resistance genes was observed, but none of these are of particular concern as they do not contribute to clinically relevant resistance. Our data support the ecological safety of SDD, at least in settings with low levels of circulating antibiotic resistance.


microbiome among Irritable Bowel Syndrome patients and healthy individuals' (CO-MIC) study at 121
two time-points with a one-year interval between sampling. None of the individuals in this cohort 122 received antibiotics. All included patients and healthy subjects were adults. 123 124

Gut microbiota profiling by HITChip 125
The HITChip is a validated phylogenetic array produced by Agilent Technologies (Palo Alto, CA) and 126 developed at Wageningen University, The Netherlands [21,22]. It contains over 4,800 127 oligonucleotides targeting the V1 and the V6 region of the 16S rRNA gene from 1,132 microbial 128 phylotypes present in the human gut [21]. DNA from fecal samples was isolated as previously 129 described [23]. The full-length 16S rRNA gene was amplified from fecal DNA, and PCR products 130 were further processed and hybridized to the microarrays as described previously [24]. Data 131 analyses were performed using R (www.r-project.org), including the microbiome package 132 (https://github.com/microbiome). Bacterial associations in the different patient groups and healthy 133 subjects were assessed using Principal Component Analysis (PCA) as implemented in CANOCO 5.0 134 [25]. Statistical testing for the differences in microbiota composition between ICU patients and 135 healthy subjects was performed by the non-parametric Mann-Whitney U test. All P-values were 136 corrected for false discovery rate (FDR) by the Benjamini and Hochberg method [26], and corrected 137 P-values (q) below 0.05 were considered significant. 138 139

Quantification of E. coli by qPCR 140
qPCRs for the quantification of E. coli were performed with primers that were previously described 141 [27], using serial dilutions of genomic DNA of E. coli DH5α to generate a standard curve. The quantification of 16S rRNA was performed with primers described in [28]. The PCR conditions were 143 identical to the qPCR conditions for the detection of mcr-1 (described above The primer set used in the qPCR assays covered 81 antimicrobial resistance genes (ARGs) of 14 157 resistance gene classes (Table S1). Primers were designed for the ARGs that are most commonly 158 detected in the gut microbiota of healthy individuals [14,15] and clinically relevant ARGs, including 159 genes encoding extended spectrum β-lactamases (ESBLs), carbapenemases, and proteins involved 160 in vancomycin resistance. We also included 10 genes encoding transposases, and a gene encoding 161 an integrase as representatives of mobile genetic elements [29]. Seven of these genes were detected 162 by qPCR but no significant differences could be observed between patients and healthy subjects 163 (data not shown) and these are not further discussed in our manuscript. Primer design was 164 performed using Primer3 [30] with its standard settings with a product size of 80 -120 bp and a primer melting temperature of 60°C. The universal primers for 16S rRNA genes were previously 166 described by Gloor et al. [28]. Forward and reverse primers were evaluated in silico for cross 167 hybridization using BLAST [31] and were cross-referenced against ResFinder [32] to ensure the 168 correct identity of the targeted genes. All primers that aligned with more than 10 nucleotides at 169 their 3' end to another primer sequence were discarded and redesigned. Additionally, all primer 170 sets were aligned to all resistance genes that were targeted in this PCR analysis to test for cross 171 hybridisation with genes other than the intended target resistance gene. Primers that aligned with 172 more than 10 nucleotides at their 3' end sequence with a non-target resistance gene were discarded 173 and redesigned. A reference sample consisting of pooled fecal DNA from different patients was 174 loaded in a series of 4-fold dilutions and was used for the calculation of primer efficiency. All 175 primers whose efficiency was experimentally determined to be between 80% and 120% were used 176 to determine the normalized abundance of the target genes. The detection limit on the Biomark 177 system was set to a CT value of 20, as recommended by the manufacturer. In addition, to assess 178 primer specificity we performed melt curve analysis using the Fluidigm melting curve analysis 179 software (http://fluidigm-melting-curve-analysis.software.informer.com/). All PCRs were 180 performed in triplicate and sample-primer combinations were only included in the analysis when all 181 triplicate reactions resulted in a CT-value below the detection limit. 182 After completion of the nanolitre-scale qPCR assays, the transferable colistin resistance gene mcr-1 183 was described [33]. To detect and quantify mcr-1, we developed primers (qPCR-mcr1-F: 5'-184 TCGGACTCAAAAGGCGTGAT-3' and qPCR-mcr1-R: 5'-GACATCGCGGCATTCGTTAT-3') for use in a 185 standard qPCR assay. The mcr-1 gene was synthesized based on the sequence described in [33] by 186 Integrated DNA Technologies (Leuven, Belgium) and used as a positive control in our assays. The 187 qPCR was performed using Maxima SYBR Green/ROX qPCR Master Mix (Thermo Scientific, Leusden, 188 The Netherlands) and a StepOnePlus instrument (Applied Biosystems, Nieuwekerk a/d IJssel, The performed with the non-parametric Mann-Whitney U test with a False Discovery Rate (Benjamini 199 Hochberg) < 0.05 to correct for multiple testing.

Microbiota dynamics in ICU patients and healthy subjects 204
Global changes in the gut microbiota of healthy subjects and ICU patients were visualized by 205 Principal Component Analysis (Fig. 1A). The microbiota profiles of healthy subjects clustered 206 together, indicating that they had stable and broadly comparable microbiota profiles, which were 207 clearly distinct from the microbiota profiles of patients during and after ICU stay. These profiles 208 covered a larger area in the PCA plot, indicating that the differences in the microbiota composition 209 of patients were larger than in healthy subjects. Notably, the microbiota of several ICU patients (e.g. 210 #120, #169, #180, #43) was already distinct from the microbiota of the healthy subjects in the first 211 days of ICU hospitalization, indicating that ICU hospitalization has a rapid effect on the microbiota. 212 The diversity of the microbiota, as quantified by Shannon's diversity index, was significantly lower 213 in ICU patients compared to healthy subjects (5.90 ± 0.20 vs 5.19 ± 0.46, respectively; P < 0.001, 214 Student's t-test) . The diversity of the microbiota of ICU patients was highly dynamic (Fig. 1B). 215 Several patients (#108, #163, #164, #165 and #169) experienced a rapid loss of diversity in the 216 first days of ICU stay. In contrast, the diversity of the microbiota was more stable in healthy subjects 217 when comparing samples that were collected one year apart (Fig. 1C). Compared to healthy 218 subjects, the microbiota of patients during ICU hospitalization was characterized by a significantly 219 higher abundance in the taxa Bacteroidetes and Bacilli: Enterococcus and a lower abundance of the 220 taxa Clostridium cluster IV and XIVa (Fig. 1D). 221 We performed quantitative PCRs to accurately determine the abundance of E. coli, one of the 222 primary targets of SDD, in the gut microbiota of patients and healthy subjects (Fig. 2). The 223 abundance of E. coli in samples of ICU patients was lower compared to the healthy subjects (p = 224 0.001; Mann-Whitney U test). Notably, upon cessation of SDD and transfer to a medium-care ward, the abundance of E. coli rebounded in one patient (#105) to levels surpassing those found in healthy 226

individuals. 227
During ICU stay, routine surveillance by conventional microbiological culture was performed on all 228 patients. E. coli could be cultured from six out of 73 rectal swabs that were collected during the 229 patients' ICU-stay. Five E. coli positive rectal swabs, of patients #43, #105, #108, #163, and #169, 230 were collected within one day of ICU admission, while the sixth positive swab (of patient #165) was 231 collected after nine days of ICU-stay. In addition, an E. coli strain from patient #105 with an ESBL-232 producing and tobramycin-resistant phenotype was isolated after ICU discharge, while the patient 233 was in a medium-care ward. The E. coli strains isolated during ICU stay were susceptible to 234 cephalosporins and aminoglycosides. All E. coli strains were susceptible to colistin. 235 236

Resistome dynamics in ICU patients and healthy subjects 237
A total of 46 unique ARGs conferring resistance to 12 different classes of antimicrobials were 238 detected in the DNA isolated from fecal samples of hospitalized patients and healthy subjects (Fig.  239   S2). The number of detected resistance genes per sample ranged between 6 and 38. Eleven 240 resistance genes were detected in >80% of healthy subjects and critically ill patients. This highly 241 prevalent set of resistance genes included tetracycline resistance genes (tetO, tetQ, tetM, tetW), two 242 aminoglycoside resistance genes (aph(3′)-III and an aadE-like gene), the bacteroidal β-lactam 243 resistance gene cblA, and the macrolide resistance gene ermB. 244 Genes associated with major antibiotic resistance threats , including those identified by the Centers 245 for Disease Control, were relatively rare. Genes encoding for extended-spectrum beta-lactamases 246 (ESBLs) were not detected in healthy subjects. In two ICU patient samples (#105C and #108C), 247 however, the ESBL genes blaCTX-M and blaDHA, respectively, could be detected. Sample #105C was collected after ICU discharge and cessation of SDD, while sample #108C was collected after 12 days 249 of ICU hospitalization and SDD treatment. The carbapenemase blaKPC was detected in a single patient 250 (patient #180), but only in the first sample (collected after 5 days in the ICU) and not in the second 251 sample, which was collected after 16 days of ICU hospitalization. No other ESBL-or carbapenemase-252 producing strains were isolated from the patients during ICU hospitalization. Other Enterobacterial 253 β-lactamases were found to be widespread in our resistome analysis. The blaAMPC β-lactamase was 254 present in 37% of samples, with nine of ten patients and eight of ten healthy subjects having 255 detectable levels of blaAMPC at one or more sampling points. The blaTEM β-lactamase was present in 256 26% of samples, corresponding with five of ten patients and four of ten healthy subjects in which 257 this gene was detectable at one or more sampling points. None of the samples were positive for the 258 carbapenemases blaNDM and blaOXA, or the transferable colistin resistance gene mcr-1 (data not 259 shown). Among resistance genes that are associated with Gram-positive pathogens, the 260 staphylococcal methicillin resistance gene mecA was detected in 13 samples from eight of ten 261 patients, but not in samples of healthy subjects. The vancomycin resistance gene vanB was present 262 in 5 samples from three of ten patients and six samples from four of ten healthy subjects. 263 A comparison of the abundance of individual ARGs in samples that were collected during ICU stay, 264 versus samples from healthy subjects, revealed that four ARGs (aac(6')-Ii, ermC, qacA, tetQ) were 265 significantly more abundant in ICU patients, while two ARGs (catA and tetW) were significantly 266 more abundant in healthy individuals (Fig. 3). Although SDD improves survival of ICU-patients, its use remains controversial due to the perceived 319 risk for selection of antibiotic resistance among bacteria that populate the patient gut. In this study, 320 we were not able to include an ICU control group that was not treated with SDD, as this would be a 321 breach of clinical guidelines for ICU-patients in our country. It is notable, however, that we did not 322 find selection for high-risk antibiotic resistance genes (like ESBLs, carbapenemases or vancomycin 323 resistance genes) in SDD-treated patients. The increased abundance of the resistance genes aac(6')-324 Ii, ermC, qacA and tetQ in SDD-treated ICU patients in our study is -in our opinion -of limited 325 concern. The first three resistance genes contribute to resistance in enterococci, either to relatively 326   . Antimicrobial resistance genes present at significantly higher or lower levels in the 527 microbiota of ICU patients, compared to healthy subjects. ARGs that are present at significantly 528 higher (aac(6')-Ii, ermC, qacA, and tetQ) or lower (catA and tetW) abundance in ICU patients, 529 compared to healthy subjects, are shown. Testing for statistically significant differences was 530 performed by the Mann-Whitney U test, with Benjamini-Hochberg correction for multiple testing (* 531 = q < 0.05; ** = q < 0.01). The horizontal line denotes the median value. The detection limit of the 532 qPCR assay is indicated with the dashed line.  43A  169A  164A  120A  163A  165A  180A  157A  169B  157B  120B  105B  164B  169C  163B  43B  108C  163C  163D  157C  120C  180B  165C  120D  120E  157D  120F  157E  105C  108E  120G  164C  169D  H25-