Research ArticleMICROBIOTA

Bacterial colonization and succession in a newly opened hospital

See allHide authors and affiliations

Science Translational Medicine  24 May 2017:
Vol. 9, Issue 391, eaah6500
DOI: 10.1126/scitranslmed.aah6500
  • Fig. 1. Change in microbial community structure after hospital opening.

    (A) Principal coordinate analysis (PCoA) of all floor samples based on weighted UniFrac distance and colored by whether they were taken before or after the hospital’s opening. (B) PCoA of three nurse station surfaces colored as in (A). (C) Changes in the relative abundance of five key genera after hospital opening. (D) Box plot of changes in the Shannon diversity index of samples after hospital opening.

  • Fig. 2. α- and β-diversity of hospital sample types.

    (A) Average α-diversity of sample types based on Faith’s phylogenetic diversity (x axis) and the Shannon diversity index (y axis). Bars indicate SEM. (B) Heat map of β-diversity relationships between sample types based on the median weighted UniFrac distance between pairwise comparisons. Sample groups are clustered based on similarity in β-diversity patterns, and median distances within individual sample types are highlighted in black along the diagonal.

  • Fig. 3. Heat map of principal coordinate space correlations between sample types.

    Values represent the average correlation between samples taken from the same location and date along the first 10 axes (eigenvectors) of the PCoA plot of all samples, weighted by the variance captured by each axis’ eigenvalues.

  • Fig. 4. Interaction between patient skin and hospital room microbiota.

    (A) Scatter plot of the percent of 16S reads in the “common microbiome” for the eight rooms sampled weekly, with the common microbiome definition on the x axis and the percent of reads in that set on the y axis. Points represent the eight individual rooms, whereas the trend line is a moving average of the data. (B) Microbial similarity between surface types increases with day of stay. Red lines connect the medians of the box plots, and the blue lines are the best-fit linear regressions. ρ is the Spearman rank correlation, and the P value is calculated as the percent of 10,000 test statistics drawn from random permutations of the data set with more negative correlations than the one observed. (C) UniFrac gain between patient room surfaces illustrates directionality of microbial transfer. Values represent the proportional amount of branch length gained by the addition of one sample’s community phylogeny (the “addition sample”) to another’s (the “base sample”). Heat maps are averages across seven patients on date of check-in (top) and after first night (bottom). (D) Change in UniFrac gain dynamics between day 0 and day 1 of hospitalization.

  • Fig. 5. Effect of antibiotic treatment on the diversity of patient skin samples.

    (A) Effect of antibiotic (Abx) treatment on the phylogenetic α-diversity of patient skin samples. Antibiotic status is indicated on the x axis, and violin plots indicate the density of phylogenetic diversity for corresponding skin samples, segregated and colored by route of antibiotic administration. For the “no Abx” or “pre-Abx” values, distributions are segregated by the route of administration for later samples of that patient. The distribution for patients who never took antibiotics during their stay is indicated by a black violin plot with no fill color. (B) PCoA of patient skin samples based on weighted UniFrac distance, split by sample type. Plots at the top are colored by the antibiotic status of the patient, and the plots at the bottom are colored by the route of antibiotic administration if the patient was taking antibiotics at the time of sampling.

  • Fig. 6. Effect of environmental factors on microbial transmission.

    (A) Heat maps of the correlation between environmental factors and the weighted UniFrac distance between samples taken from the same room on the same day. (B) Seasonal changes in the UniFrac distances between the hand and nose samples of nurses working on the same floor. Trend lines are a smoothed moving average of the data. (C) Correlations between environmental factors and the UniFrac distances of hand and nose samples of nurses working on the same floor on the same day. Color scheme is as in (A).

  • Table 1. Effects of seven binary clinical factors on the α- and β-diversity of patient skin and bedrail bacterial communities.

    (A) PERMANOVA analyses of the effects of clinical metadata on observed β-diversity. Each test was based on the weighted UniFrac distance between samples, and significance was assessed through 105 permutations of the randomized data set. (B) Effects of clinical metadata on the α-diversity of skin and bedrail bacterial communities, based on Faith’s phylogenetic diversity index. Significance was assessed through a two-sided nonparametric t test with 105 permutations. For both (A) and (B), significant test results are highlighted in bold, and those that are significant after a Bonferroni correction for multiple comparisons are indicated with an asterisk.


    Embedded Image

Supplementary Materials

  • www.sciencetranslationalmedicine.org/cgi/content/full/9/391/eaah6500/DC1

    Fig. S1. Floor plan of sampling locations.

    Fig. S2. Bipartite OTU network of floor samples.

    Fig. S3. Relative abundances of common genera by surface type.

    Fig. S4. Rarefaction curves demonstrate convergence of α-diversity calculations.

    Fig. S5. Interaction between patient skin and room samples.

    Fig. S6. Differences in community dissimilarity between intra- and interweek samples.

    Fig. S7. Overview of oligotyping data.

    Fig. S8. Predictive accuracy of SourceTracker models using hand samples as source.

    Fig. S9. Predictive accuracy of SourceTracker models using axilla samples as source.

    Fig. S10. Patterns of antimicrobial resistance gene abundance.

    Fig. S11. Coupling between selection and codon usage bias shows differential impact of in situ functional constraints on the same strain of P. acnes.

    Fig. S12. Example PC space correlation calculation.

    Fig. S13. PCoA of all samples.

    Table S1. Summary of the 6523 samples included in the study, grouped by surface type.

    Table S2. Summary of clinical metadata for 49 patients sampled on multiple days.

    Table S3. Predictive accuracy of random forest supervised learning models predicting seven binary clinical factors.

    Table S4. Effect of binary clinical factors on the similarity between sample types taken from the same room on the same day.

    Table S5. Summary of the 65 genome bins assembled from metagenome contigs.

  • Supplementary Material for:

    Bacterial colonization and succession in a newly opened hospital

    Simon Lax, Naseer Sangwan, Daniel Smith, Peter Larsen, Kim M. Handley, Miles Richardson, Kristina Guyton, Monika Krezalek, Benjamin D. Shogan, Jennifer Defazio, Irma Flemming, Baddr Shakhsheer, Stephen Weber, Emily Landon, Sylvia Garcia-Houchins, Jeffrey Siegel, John Alverdy, Rob Knight, Brent Stephens, Jack A. Gilbert*

    *Corresponding author. Email: gilbertjack{at}gmail.com

    Published 24 May 2017, Sci. Transl. Med. 9, eaah6500 (2017)
    DOI: 110.1126/scitranslmed.aah6500

    This PDF file includes:

    • Fig. S1. Floor plan of sampling locations.
    • Fig. S2. Bipartite OTU network of floor samples.
    • Fig. S3. Relative abundances of common genera by surface type.
    • Fig. S4. Rarefaction curves demonstrate convergence of α-diversity calculations.
    • Fig. S5. Interaction between patient skin and room samples.
    • Fig. S6. Differences in community dissimilarity between intra- and interweek samples.
    • Fig. S7. Overview of oligotyping data.
    • Fig. S8. Predictive accuracy of SourceTracker models using hand samples as source.
    • Fig. S9. Predictive accuracy of SourceTracker models using axilla samples as source.
    • Fig. S10. Patterns of antimicrobial resistance gene abundance.
    • Fig. S11. Coupling between selection and codon usage bias shows differential impact of in situ functional constraints on the same strain of P. acnes.
    • Fig. S12. Example PC space correlation calculation.
    • Fig. S13. PCoA of all samples.
    • Table S1. Summary of the 6523 samples included in the study, grouped by surface type.
    • Table S2. Summary of clinical metadata for 49 patients sampled on multiple days.
    • Table S3. Predictive accuracy of random forest supervised learning models predicting seven binary clinical factors.
    • Table S4. Effect of binary clinical factors on the similarity between sample types taken from the same room on the same day.
    • Table S5. Summary of the 65 genome bins assembled from metagenome contigs.

    [Download PDF]

Stay Connected to Science Translational Medicine

Navigate This Article