Exposure and resistance

Daptomycin exposure in the 6 months prior to the index sample ranged from 3 to 34 doses, and the most recent dose before the index sample was between 0 and 141 days earlier (see S1 Table). Patients with prior daptomycin treatment had greater proportions of resistant clones, defined as a computed minimum inhibitory concentration (MIC_c—see Methods for a detailed description of this metric) greater than 4 μg/ml (Fig 3). Overall, 50 out of 426 clones were resistant to daptomycin with 94% of these in patients with prior daptomycin exposure. Eight of the 11 patients with resistant clones were from the daptomycin exposed group. Moreover, highly resistant E. faecium clones (MIC_C >8 μg/ml) were only found in the daptomycin-exposed group. Patient mean MIC_C was not associated with enterococcal species diversity (Spearman ⍴ = −0.06, p = 0.79; Fig A in S1 Text), the number of days since the last dose of daptomycin (Spearman ⍴ = −0.07, p = 0.83; S1 Fig), or the number of doses of daptomycin in the previous 6 months (Spearman ⍴ = 0.22, p = 0.50; S1 Fig).

Download:

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

Fig 3. Daptomycin MICs by patient.

Daptomycin resistance of each clone by patient and exposure group. Blue circles are the control patients and red circles are from daptomycin-exposed patients. Each circle is the mean of 2 independent estimates of the MIC_C for a single clone. Clones above the dashed line are daptomycin resistant (due to the continuous nature of the computed MIC (MIC_C), an MIC_C of >4μg/ml is equivalent to the clinical 2-fold MIC cutoff of ≥8μg/ml (see Methods for a more detailed discussion of the MIC_C method)). The underlying data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3000987.g003

We are also interested in how variation in the resistance phenotype is spread across the study population. A Bayesian mixed-effect model was designed to test for the presence of variation in daptomycin resistance at the levels of interest: within patients, between patients, and between groups (daptomycin exposed versus not daptomycin exposed). Using the Deviance Information Criterion (DIC) [24], the best fit model included a random effect for “patient,” and for “clone” nested within patient, such that the distribution for the clone effect differed from one patient to another (Table 1).

Download:

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

Table 1. Summary of Models and DIC analysis results.

https://doi.org/10.1371/journal.pbio.3000987.t001

The fixed effect for daptomycin exposure, M_D, provided support for the hypothesis that resistance is higher in daptomycin-treated patients with 94.5% of the posterior distribution for the fixed effect falling above zero (Fig 4 insert). The mean of M_D is 0.57 on a log₂ scale, which equates to approximately a 50% increase in mean MIC_C.

Download:

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

Fig 4. Posterior distribution of random and fixed effects from the Bayesian mixed effects model.

(Main panel) Posterior distributions for standard deviations of the clones within patients and the error. Blue markers indicate control patients, red indicate daptomycin-treated patients, and black is the error (circle: mean; thick lines: 50% credibility interval; thin lines: 95% credibility interval). The 95% credibility interval of the error falls within the error marker. (Insert) Posterior distribution of fixed effect for prior daptomycin exposure. The underlying data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3000987.g004

Within-patient variation

The variation in MIC_C among clones within patients differed from one patient to another (Table 1; Model 1 is a significantly better fit (as measured by a lower DIC) than Model 2). The posterior distributions for the standard deviations of the clone effects for each patient are summarized in Fig 4. Further, Model A was designed to assess if there was evidence that patients with prior daptomycin exposure had greater between-clone variability in MIC_C. In particular, in Model A, the standard deviation for the “clone” effect for control patients was modeled with a single parameter λ and for the daptomycin patients was λexp(α) (see S2 Text for details). By using this parameterization, we were able to quantify the amount of evidence for daptomycin-treated patients having greater within-patient variability (by determining the proportion of α’s posterior distribution that lies above zero). There is strong evidence for daptomycin-treated patients having greater within-patient variation than control patients with the estimated posterior distribution for α lying essentially above zero (mean = 1.04, 99% credibility interval is 0.677 to 1.798).

Time series

Eight daptomycin patients and 6 control patients had more than 1 swab sample collected in 2016. While these additional swab samples do not allow for a comprehensive analysis of within-patient changes in resistance over time, they make possible a number of interesting case studies of within-patient changes in resistance. Ten E. faecium clones from each of these additional samples were isolated and tested for resistance. For standardization, the first 10 clones from the index samples in the previous analysis were used.

In all 3 patients where a sample was collected before the commencement of daptomycin treatment, the posttreatment sample contains clones that are more resistant than any clones sampled prior to treatment. The MIC_C for the most resistant clone in the sample immediately prior to and posttreatment are: Patient 4 prior = 2.2 μg/ml, post = 4.2 μg/ml; Patient 87 prior = 1.7 μg/ml, post = 2.7 μg/ml; and Patient 150 prior = 0.5 μg/ml, post = 20.0 μg/ml (Fig 5). Patients 4 and 150 also showed an increase in mean MIC_C (Patient 4: prior sample mean = 1.69 μg/ml, index sample mean = 2.86 μg/ml, t = −3.46, df = 18, p = 0.003; Patient 150: prior sample mean = 0.32 μg/ml, index sample mean = 13.08 μg/ml, t = −24.99, df = 18, p < 0.001). Patient 87 had a higher mean in the index sample but this was not statistically different to the pretreatment sample mean (prior sample mean = 1.15 μg/ml, index sample mean = 1.42 μg/ml, t = −1.41, df = 18, p = 0.18).

Download:

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

Fig 5. Within-patient resistance over time.

Each plot shows patient admission periods, drug doses, and resistance of E. faecium clones isolated from screening swabs and blood stream infections for an individual patient. Time 0 is the index sample included in the previous analysis. Patient admission periods are shown as gray blocks, and individual doses of vancomycin (purple), linezolid (yellow), and daptomycin (red) are detailed in the bars at the top of the plot. Circles show daptomycin resistance (MIC_C) for 10 clones per sample. Each circle is the mean of 2 replicates with black circles denoting VR E. faecium and blue circles denoting VS E. faecium. Pink diamonds are isolates from VR E. faecium blood stream infections. The underlying data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3000987.g005

The maintenance of resistance following cessation of daptomycin treatment is highly variable between patients. In patients where a second sample was collected after the index sample (8 out of 12 patients), we see examples of long-term maintenance of resistance in the absence of further daptomycin exposure in our hospital (Patient 86) and rapid loss of resistance (Patients 4 and 150), which in both examples here is coupled with the presence of vancomycin-sensitive E. faecium which was not isolated in the index samples. In 2 patients, there was a slight increase in daptomycin resistance in the sample following the index sample despite no further daptomycin exposure, which was combined with a switch from predominantly vancomycin sensitivity to predominantly vancomycin resistance (Patients 64 and 233, see S2C and S2D Fig). In patients where the index sample contained no daptomycin resistant isolates, little change was observed after treatment ended (Patients 29 and 84, see S2C Fig).

Genome analysis

To gain clearer insight to resistance evolution within patients, 4 patients were chosen that demonstrate the range of temporal pattern of resistance evolution described above. A total of 95 E. faecium isolates from 4 patients were sequenced (at least 5 clones per time point per patient—see Methods for more details). The core genome among these isolates includes 1,804 genes present in 100% of isolates, encoded in 1.68 Mb of sequence. An additional 2,795 accessory genes were annotated. This core genome was used to build a phylogeny (Fig 6). All genomes sequenced for this study fall within the previously defined hospital-associated E. faecium clade A [25].

Download:

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

Fig 6. Phylogeny for E. faecium strains isolated from 4 patients.

Midpoint rooted tree of 89 isolates. Six sequenced isolates were excluded that had core genomes identical to other samples in the alignment. The panel on the right shows the mean daptomycin MIC_C (ug/mL) for each sample. Many isolates were closely related, resulting in very short branch lengths. See Fig 7 for more detailed visualization of phylogenetic structure within hosts. Sequence data is available at NCBI GenBank under the study accession number PRJNA673360.

https://doi.org/10.1371/journal.pbio.3000987.g006

E. faecium isolates were mostly clustered within patients, with 2 patients containing more than 1 cluster of isolates. Patient 150 contained 2 clusters separated by 20,062 SNPs, and Patient 86 contained a single isolate that differed from all the others from that patient by 2,997 SNPs, consistent with at least 2 introductions into these 2 patients. By contrast, the mean number of core genome SNPs between 2 isolates within the remaining clusters was 4, 30, 2, 6, and 14 SNPs for patients 4, 86, 87, Patient 150 cluster 1, and Patient 150 cluster 2, respectively, plausibly consistent with within-patient evolution. This study design cannot rule out many more introductions from a closely related donor, as the specific donor populations are unknown.

Phylogenetic patterns are largely consistent with the diversity of phenotypic trajectories in these 4 patients. In Patient 4, resistant isolates were only observed in the index swab, while later isolates were more sensitive. Concordantly, the resistant isolates from the index swab from Patient 4 were part of a single sub-clade which carried an N13S mutation in cardiolipin synthase A (ClsA), a mutation previously associated with daptomycin resistance [26] (Fig 7A; see Fig A and Table A in S3 Text for more details). The lack of resistance at later points in time reflects the lack of this sub-clade among the later samples. In Patient 86, there was long-lasting resistance and diversity, and concordantly, we see that mutations in ClsA presenting in the index swab persisted in 4 of 6 isolates that were seen to be resistant later (Fig 7B; see Fig B and Table B in S3 Text for more details). Patient 150 was initially colonized with daptomycin- and vancomycin-sensitive E. faecium, then later by daptomycin-sensitive but vancomycin-resistant isolates (Fig 7C and S3 Fig). Sequencing revealed that the blood stream infection was caused by isolates from a second clade that subsequently gained daptomycin resistance in the gastrointestinal isolates. The later reversion to a more daptomycin-susceptible population reflects the reemergence of the initial sequencing type, a subset of which are now vancomycin resistant and carry the vanA containing plasmid of the other clade (S3 Fig; see Fig D and Table D in S3 Text for more details).

Download:

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

Fig 7. Individual phylogenies for E. faecium strains isolated from 4 patients.

A midpoint-rooted phylogeny was created based on the core genome alignment of all isolates using RAxML. Here, subsets of the full tree (Fig 6) are shown, including only the isolates from each patient (i.e., other tips are masked). Scale bars for branch length (mean number of substitutions per site) are included for each tree and insert. Rectangle inserts show detailed structure of indicated groups. Daptomycin MIC_Cs are listed for each tip. For tips representing 2 isolates with identical core genomes, the daptomycin MIC_Cs for both isolates are listed. Asterisks indicate mutations in genes known to be associated with variation in daptomycin susceptibility. For each patient, one assembled genome (two for the 2 clades in Patient 150) was used as a reference genome for variant calling (see Methods). (A) Patient 4 (B) Patient 86 (C) Patient 150 (D) Patient 87. Sequence data is available at NCBI GenBank under the study accession number PRJNA673360.

https://doi.org/10.1371/journal.pbio.3000987.g007

Sequencing of Patient 87 isolates revealed more complicated dynamics than suggested from the phenotypic data (Fig 7D). Just 1 isolate from Patient 87’s index swab had an MIC_C above 4 μg/ml; however, all isolates sequenced from the index swab carried mutations in ClsA (see Fig C and Table C in S3 Text for more details). The resistant isolate and 2 others carried a ClsA R267H mutation previously associated with resistance, but in addition, the 2 sensitive isolates also carried a second mutation in ClsA, suggesting possible loss or reversion. Similarly, the second clade had a ClsA R221Q mutation; however, all isolates were sensitive, raising the question of whether the effects were modified by subsequent mutations.

The genetic analysis of these 4 patients is consistent with dynamic gain and loss of resistance in the off-target population and gives genetic explanations for even subtle resistance changes. However, mutations in known resistance genes do not explain all of the gains and losses of daptomycin resistance, and the sequencing suggests other candidates for further investigation (see Table A–D in S3 Text for a full list of mutations). Finally, the sequencing reveals dynamic gain and loss of vancomycin resistance. In Patient 4, the loss of vancomycin resistance was associated with loss of the entire plasmid containing the vanA cassette, or deletion of all or part of the cassette. In Patient 150, the gain of vancomycin resistance occurred by introduction of a second strain carrying the vanA cassette, or a transfer of the vanA containing plasmid from one strain to another.

Study participants

A cohort study was conducted utilizing perirectal swabs from an infection prevention screening program at Michigan Medicine to determine the impact of intravenous daptomycin treatment on daptomycin resistance in gut populations of Enterococcus faecium. The study was approved by the University of Michigan Institutional Review Board. The initial inclusion criteria (see Fig 1) were all patients with a VRE positive (E. faecium or E. faecalis) swab using VRE Select agar (BioRad, Marnes-la-Croquette, France) in 2016 (n = 618). Patients included in the daptomycin exposure group had at least 3 administered doses of daptomycin (usually equal to 3 days of therapy) in the 6 months prior to the VRE positive swab (14 patients). Three doses was selected as a minimum requirement because treatment of enterococcal infections generally requires at least 3 days of treatment. The control group consisted of patients who had no known daptomycin exposure and at least 6 doses of linezolid (usually equal to 3 days of therapy due to twice daily dosing) in the last 6 months prior to a VRE positive swab (15 patients). For each patient, the first sample to meet the inclusion criteria was defined as the index sample. For each index sample, Enterococcus sp. clones were isolated until there were 20 E. faecium clones per sample. Samples where no E. faecium was isolated after sampling 20 random Enterococcus colonies, only 1 colony was isolated after sampling 40 random Enterococcus colonies, or where no Enterococcus was isolated were excluded from further analysis (Fig 1). The final data set included 12 patients in the daptomycin exposure group and 10 patients in the control group. For a further time-series analysis of these patients, up to 5 prior samples and all subsequent samples available from these patients in 2016 were tested for the presence of E. faecium. Ten random clones per sample were isolated from each of the E. faecium-positive samples. Finally, we collected isolates from all blood stream infection (BSI) in these patients.

Ethics statement

This study was approved by the University of Michigan Institutional Review Board (ID no. HUM00102282), which determined that informed consent was not required as all samples utilized were collected for patient treatment purposes.

Strain isolation

Perirectal swabs were obtained using E-swabs (BD) as part of the hospital VRE surveillance program. The swabs were first tested in clinical microbiology lab by streaking on VRESelect agar (BioRad) per manufacturer's recommendations. The swab was discarded, and the residual media was stored with glycerol (final concentration 20% v/v) at −80°C. To isolate E. faecium, samples were streaked from the sample stored in glycerol onto Enterococcosel agar (BD BBL) in duplicate and incubated up to 72 h at 37°C. The first 10 colonies from each plate (20 colonies per sample) were re-streaked on Enterococcosel agar and incubated for 48 to 72 h at 37°C. One colony from each plate was then streaked on BHI agar (BD BBL) and a vancomycin (30 μg/ml Oxoid) disc was placed on each plate to determine vancomycin resistance. Plates were incubated for 24 h at 37°C. One clone per plate was stored in BHI +20% glycerol at −80°C.

To confirm the species of Enterococcus, a species-specific multiplex PCR was performed using primers for E. faecium, E. faecalis and VanA, VanB, VanC1, and Van C2/3 [44]. Briefly, 11.25 μl PCR Master Mix (iProof HF, BioRad), 50 uM of each primer, and 6.45 μl water (total volume 22.5 μl) per sample were combined. Sample was added as either 2.5 μl of bacteria in BHI glycerol taken from tubes prior to freezing, or 1 colony from a streaked culture of stored bacteria. PCR was performed under the following conditions: 95°C for 4 min; 30 cycles (98°C for 10 s, 55°C for 30 s, 72°C for 30 s); 72°C for 7 min. Gels were run for 1 h at 80 to 100 V on 2% Agarose in TAE buffer with 0.1 μl/ml SybrSafe.

Isolation steps were repeated until 20 E. faecium clones per sample or 10 clones for time-series only samples were isolated. Samples were excluded if more than 40 Enterococcus clones were isolated without any E. faecium (20 clones for time-series only samples), or no Enterococcus was detected after streaking the sample twice and then plating 80 μl of the initial patient sample on Enterococcosel agar (combined approximately 10% of the total sample volume). This sampling method resulted in a data set on species diversity within patients (Fig 2).

Ten patients had Enterococcus blood stream infections (BSI) within 6 months of the index swab sample, and a total of 45 isolates were taken from these patients. Blood samples were cultured in blood bottles and streaked on Chocolate agar in the clinical microbiology lab. Single colonies were streaked on BHI agar 3 times, and then a single colony was stored in BHI +20% glycerol.

MIC testing

MIC testing was performed by broth microdilution (BMD) according to CLSI M7 guidelines [45]; each sample was tested in duplicate and 1 of 4 patient-derived E. faecium strains was included on each run as a positive control. All clones were initially tested on plates containing 2-fold dilutions of daptomycin with final concentrations ranging from 0.125 μg/ml to 16 μg/ml, and the optical density (OD) of each well (600 nm) was determined by plate reader (FLUOstar Omega, BMG Labtech, Ortenberg, Germany). A Hill function was fitted to the OD values for each dilution series, and resistance was measured as the concentration at which this curve crossed a defined cutoff. We set the cutoff at 2 standard deviations above the mean of the negative control wells (see S2 Text). This threshold was chosen because it reflects the automated equivalent of a visual MIC, i.e., the point at which the OD falls below the detectable limit. Thus, we refer to this value as the computed MIC (MIC_C) as it is the minimum antibiotic concentration required to reduce the OD reading to the background level and is reported as a continuous (not 2-fold) value. If the initial concentration range did not contain at least 2 concentrations above and below the cutoff, the assay was repeated on either increased (1 μg/ml to 64 μg/ml) or decreased (0.0625 μg/ml to 4 μg/ml) concentrations as appropriate. Individual assays of clones were also excluded if the Hill curve did not fit the data points well, determined as an MIC_C greater than one 2-fold dilution from the lowest concentration with an OD below the cutoff.

Statistical analysis of MIC_C

We analyzed the log₂ of the MIC_C values using Bayesian mixed effect models [46]. Models included a fixed treatment effect for patients that fulfilled the daptomycin exposure case definition (see above). To quantify the evidence for prior daptomycin exposure increasing MIC_C, we determined the proportion of the posterior for the fixed “treatment” effect that was above zero. The full model included 23 random effects (1 “patient” and 22 “clone” effects) and allowed the distribution of the “clone” effect to depend on patient. We fitted 6 candidate models which considered different combinations of the random effects. In addition to testing different combinations of the random effects from the full model, we also considered models where the distribution of the “clone” effect was identical for all patients, and a model where this distribution depended on treatment group (see Table 1 and S2 Text for details). All random effects are assumed to be normally distributed with zero mean and standard deviations estimated using the MCMC program JAGS [47,48]. Uninformative priors were used (see Table B in S2 Text). We ran each model for 20 × 10^6 iterations with a burn in of 10 ×10^6 steps and a thinning interval of 2 × 10^3. This resulted in 5 × 10^3 parameter samples for each model run. This process was repeated to generate 4 chains, with randomly chosen initial starting values, for each model. Posterior convergence was confirmed in 2 ways: (1) empirical inspection of the estimated posterior distributions (all 4 chains resulted in very similar distributions); and (2) the Gelman–Rubin convergence diagnostic (this statistic was essentially 1 for all parameters, which is consistent with the chains having converged).

The models were then compared using the Deviance Information Criterion (DIC) [24]. The relative fits of the models are summarized using ΔDIC scores which are the differences in DIC between the best model and each alternative model. Although there is no universally agreed threshold for significance of ΔDIC scores, there is precedent for treating ΔDIC scores greater than 10 as providing very little support for the model [49] (similar to rules used for the Akaike Information Criterion (AIC) [50,51]). The smallest ΔDIC score was 84, indicating that Model 1 is clearly the preferred model. To assess how well the data fit the best model (as selected by DIC comparisons), we used the posterior distributions of the selected model to generate synthetic data sets and examined the distributions of a number of different summary statistics (see S2 Text).

Genome sequencing