Research ArticleENVIRONMENTAL STUDIES

Hidden drivers of low-dose pharmaceutical pollutant mixtures revealed by the novel GSA-QHTS screening method

See allHide authors and affiliations

Science Advances  07 Sep 2016:
Vol. 2, no. 9, e1601272
DOI: 10.1126/sciadv.1601272
  • Fig. 1 GSA-QHTS experimental framework.

    The novel approach couples GSA with QHTS (GSA-QHTS) to understand the main effects and interactions of combinations of diverse input factors, such as chemicals, biotic or abiotic factors, etc. Here, GSA-QHTS was applied to study the effects on A. CPB4337 of mixtures of 16 PPCPs (C1 to C16) at environmentally realistic low doses (D1 to D3) and the influence of light intensity, an abiotic factor. The steps of the framework are highlighted: parsimonious GSA sampling (to generate an experimental design template), QHTS (with randomized biological replication), and GSA screening of the importance and interactions of input factors controlling biological response.

  • Fig. 2 Low-dose PPCP mixture experimental design template.

    (A) Discrete dose levels (in nanograms per liter) of each of the 16 PPCPs (C1 to C16) included in the low-dose mixtures. The three discrete levels were the median of means, the mean of maxima, and the maximum of maxima of each PPCP in freshwater. (B) Sampling frequency of the three discrete levels for the 16 PPCPs (C1 to C16) across the 180 mixtures. (C) Mean dose of each PPCP along the 180 low-dose mixtures. (D) Frequency distribution of sum of PPCP doses (in nanograms per liter) within each of the 180 low-dose mixtures.

  • Fig. 3 Exposure to low doses of PPCPs produced significant sublethal effects.

    (A) Notched box plot for relative bioluminescence of A. CPB4337 control observations (n = 880), individual PPCP exposure (Ind) (n = 768), and exposure to mixtures (Mix) (n = 1080). (B) Median relative bioluminescence level of A. CPB4337 exposed to the 180 mixtures with respect to control levels (relative bioluminescence, 1). Mixtures are ranked in decreasing order based on median relative bioluminescence values. Vertical lines are bootstrapped (n = 999) 95% CIs for the median. Horizontal red dashed lines are bootstrapped 95% CIs for control median relative bioluminescence level (n = 999). The most potent mixture, Mix 16, is highlighted. (C and D) Notched box plots sorted by light intensity (L1 and L2) for relative bioluminescence of A. CPB4337 exposed to individual PPCPs (n = 768) and mixtures of PPCPs (n = 1080), respectively. The notches extend to ±1.58 IQR (interquartile range)/n1/2, where no overlapping of notches among boxes offers evidence of statistically significant differences among their medians (71). Statistically significant differences were tested by one-way ANOVA: ***P < 0.001.

  • Fig. 4 The null additive mixture models do not predict the sublethal effects of low-dose PPCP mixtures.

    (A) Individual effects of PPCPs (C1 to C16) on A. CPB4337 relative bioluminescence at 10 mg liter−1. Horizontal lines represent the control’s median, Q25 and Q75 relative bioluminescence. Significance values: *P < 0.05 and ***P < 0.001. (B) Dose-response curves for factors C10 (erythromycin), C11 (ofloxacin), and C14 (venlafaxine). Lines and symbols represent fitted nonlinear models (five-parameter log-logistic models) and experimental data, respectively. Shaded areas represent 95% CIs of model predictions. (C) Modeled mixture dose-response curves for 21 unique C10/C11/C14 ratios present in the sampled 180 PPCP mixtures. The box plot shows the 180 low-dose mixtures’ dose range. (D) Observed versus predicted 1:1 plot (42) for the 180 low-dose mixture effects as predicted by the CA additivity model. Note that only the contributions of C10 (erythromycin), C11 (ofloxacin), and C14 (venlafaxine) are considered by the models. NSE, Nash-Sutcliffe efficiency coefficient with 95% CI in brackets. The inset shows the distribution of observed and predicted luminescence values for the 180 mixtures.

  • Fig. 5 GSA-QHTS is able to characterize global drivers of low-dose mixture sublethal effects.

    (A) GSA schematic representation of input factor distributions along μ*-σ and μ-σ Cartesian planes for linear/additive, mixed, and nonlinear/nonadditive systems. (B and C) GSA results in the μ*-σ and μ-σ Cartesian plots, respectively, for the 17 input factors [16 PPCPs (C1 to C16) and light intensity (L)] analyzed in the present study. Red lines indicate the limits for linear additive systems. For clarity, only important and nonimportant input factors are identified in the figure. Hollow symbols indicate important factors (larger separation from the μ*-σ or μ-σ plane origin). (D) Ranked input factors by importance (μ*) showing the proposed limits (red dashed lines) for important, moderately important, and nonimportant factors.

  • Fig. 6 Ecological scaling-up experiment.

    (A) River benthic microbial community inocula were obtained from a nearby unpolluted stream in the Llémena River (Girona, Spain). Schematic benthic microbial community modified from the study of Egan et al. (72). (B) A set of cobbles was scraped for their microbial communities, which were used to colonize rough glass substrata under laboratory conditions. (C) Experimental microbial communities were exposed to three low-dose PPCP mixtures (Mix 16, Mix 16-4, and Mix 16/10; see Materials and Methods). (D) Selected community-level end points (F0, F, Ymax, Yeff, β-Glu, and Phos; see Materials and Methods) covered both autotrophic and heterotrophic global fitness indicators and were monitored as a function of time.

  • Table 1 ANODEV table summary of GLMMV models.

    GLMMV models were fitted to the response of experimental freshwater benthic communities exposed to selected PPCP low-dose mixtures and sequential analysis of deviance (ANODEV) was performed. MV data used to build each GLMMV include the six community-level metabolic end points measured: F0, the dark-adapted basal fluorescence; F, the light-adapted steady state fluorescence; Ymax, the maximum photosynthetic efficiency of photosystem II (PSII); Yeff, the effective quantum yield of PSII; β-Glu, β-glucosidase; and Phos, alkaline phosphatase. The experiment included two factors: treatment (four levels) and time of exposure (three levels). Treatment levels were as follows: control (n = 5); Mix 16 (n = 3), PPCP mixture 16; Mix 16-4 (n = 3), mixture 16 without the four most important PPCPs from GSA results; and Mix 16/10 (n = 3), mixture 16 diluted 10 times. Time levels were as follows: 24, 36, and 120 hours of exposure for each treatment (n = 4). Therefore, the total number of samples was n = 42. The null hypothesis (H0) for the Wald test is that the reduction in model residual deviance is 0. Res. df, residual degrees of freedom; df diff., degrees of freedom difference added by the sequential inclusion of each term of the model.

    GLMMV modelRes. dfdf diff.WaldP (>Wald)
    Model 1 (full model): treatment × time
    Intercept41
    Treatment3836.590.45
    Time36213.050.86
    Treatment × time30617.160.01**
    Model 2 (control only): time
    Intercept14
    Time12210.030.01**
    Model 3 (Mix 16 only): time
    Intercept8
    Time6215.750.54
    Model 4 (Mix 16-4 only): time
    Intercept8
    Time62162.20.04*
    Model 5 (Mix 16/10 only): time
    Intercept8
    Time6218.180.61

    Significance values: *P < 0.05 and **P < 0.01.

    Supplementary Materials

    • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/2/9/e1601272/DC1

      S1. Description of data sets and guide for data analysis

      S2. Supplementary methods

      S3. Table summary of pharmaceutically active pollutants

      S4. Composition of the 180 mixtures

      table S1.1. Summary of observations by Case and Exp. Group for the GSA.csv data set.

      table S1.2. Summary of the variables included in the GSA.csv data set.

      table S1.3. Summary of observations by Case and Exp. Group for GSA_MIX_MEDIAN data set.

      table S1.4. Summary of the variables included in the GSA_MIX_MEDIAN.csv data set.

      table S1.5. Summary of observations by Exp. Group for ANTIBIOTIC_EC50.csv data set.

      table S1.6. ANTIBIOTIC_EC50.csv data set.

      table S1.7. Summary of observations by Exp. Group for biofilm.csv data set.

      table S1.8. biofilm.csv data set.

      table S1.9. P values of Wald test for significance of time variable of the ANODEV test for each univariate GLMs fitted to each end point (Phos, Gluc, F0, Ymax, F, Yeff) for each treatment with time as explanatory variable.

      table S2.1. Number of observations for calculating each occurrence of statistical descriptors.

      table S2.2. Stability of PPCPs under experimental conditions.

      fig. S1.1. Q-Q plot for normal distribution for control observations (n = 880).

      fig. S1.2. Notched box plot for Lum of control observations by Experiment.

      fig. S1.3. hovPlots for bioluminescence as a function of the Experiment for Control observations.

      fig. S1.4. hovPlots for bioluminescence as a function of the Experiment for Treatment observations.

      fig. S1.5. Notched box plot for Lum of controls (n = 880), individual exposure to PPCPs (n = 768), and exposure to mixtures of PPCPs (n = 1080).

      fig. S1.6. 95% family-wise confidence level for the difference of means according to anova.2_glht model.

      fig. S1.7. Notched box plot for Lum of control observations (n = 880), divided into two aleatory groups C1 (n = 440) and C2 (n = 1080).

      fig. S1.8. Notched box plot for Lum for treatment observations shorted by L level (1, 2).

      fig. S1.9. Boxplots of Lum as a function of PPCPs (C1 to C16).

      fig. S1.10. Histogram (left panel), normal Q-Q plot (central panel) and jack after jackknife plot (right panel) for bootstrapped medians of Lum values (R = 999) for “treatment” = 1.

      fig. S1.11. Bootstrapped (R = 999) medians and 95% CIs for the 180 treatments.

      fig. S1.12. Tolerance level in the estimation of median values for lum for the 180 mixture effects.

      fig. S1.13. EE μ*-σ plot for the 17 studied input factors (16 PPCPs and light intensity).

      fig. S1.14. Ranked input factors by importance (μ*).

      fig. S1.15. EE μ*-σ plot for the 17 studied input factors (16 PPCPs and light intensity).

      fig. S1.16. Diagnostic plots of Anov.1 model.

      fig. S1.17. Diagnostic plots of Anov.4 model.

      fig. S1.18. Dose ranges of PPCPs (in nanograms per liter).

      fig. S1.19. Histogram of the frequency distribution of the total sum of the 16 PPCPs in the 180 mixtures.

      fig. S1.20. Rose plot presenting the relative abundance (counts) of each PPCP (variable) and level (value) of each PPCP in the 180 mixtures.

      fig. S1.21. Rose plot presenting the mean concentration (value) of each PPCP (variable) in the 180 mixtures.

      fig. S1.22. Dose-response data and fitted LL5 drm models (chem.1) to the experimental data.

      fig. S1.23. Schematic representation of the calculation sequence required to predict chemical mixture effects according to CA model.

      fig. S1.24. The 21 dose-response models (LL.5 models) fitted to the in silico–predicted dose-response patterns of the 21 unique combinations of C10, C11, and C14.

      fig. S1.25. Observed versus predicted bioluminescence values of A. CPB4337 to the 180 low-dose mixtures of PPCPs.

      fig. S1.26. FITEVAL report for the predicted versus experimental low-dose mixture effects of PPCPs.

      fig. S1.27. Response of six community-level end points measured along time as a function of treatment on model freshwater benthic microbial communities.

      fig. S1.28. Residual versus fits plot to check the quadratic mean-variance assumption of negative binomial regression (with different metabolic end points coded in different colors).

      References (73109)

    • Supplementary Materials

      This PDF file includes:

      • S1. Description of data sets and guide for data analysis
      • S2. Supplementary methods
      • S3. Table summary of pharmaceutically active pollutants
      • table S1.1. Summary of observations by Case and Exp. Group for the GSA.csv data set.
      • table S1.2. Summary of the variables included in the GSA.csv data set.
      • table S1.3. Summary of observations by Case and Exp. Group for GSA_MIX_MEDIAN data set.
      • table S1.4. Summary of the variables included in the GSA_MIX_MEDIAN.csv data set.
      • table S1.5. Summary of observations by Exp. Group for ANTIBIOTIC_EC50.csv data set.
      • table S1.6. ANTIBIOTIC_EC50.csv data set.
      • table S1.7. Summary of observations by Exp. Group for biofilm.csv data set.
      • table S1.8. biofilm.csv data set.
      • table S1.9. P values of Wald test for significance of time variable of the ANODEV test for each univariate GLMs fitted to each end point (Phos, Gluc, F0, Ymax, F, Yeff) for each treatment with time as explanatory variable.
      • table S2.1. Number of observations for calculating each occurrence of statistical descriptors.
      • table S2.2. Stability of PPCPs under experimental conditions.
      • fig. S1.1. Q-Q plot for normal distribution for control observations (n = 880).
      • fig. S1.2. Notched box plot for Lum of control observations by Experiment.
      • fig. S1.3. hovPlots for bioluminescence as a function of the Experiment for Control observations.
      • fig. S1.4. hovPlots for bioluminescence as a function of the Experiment for Treatment observations.
      • fig. S1.5. Notched box plot for Lum of controls (n = 880), individual exposure to PPCPs (n = 768), and exposure to mixtures of PPCPs (n = 1080).
      • fig. S1.6. 95% family-wise confidence level for the difference of means according to anova.2_glht model.
      • fig. S1.7. Notched box plot for Lum of control observations (n = 880), divided into two aleatory groups C1 (n = 440) and C2 (n = 1080).
      • fig. S1.8. Notched box plot for Lum for treatment observations shorted by L level (1, 2).
      • fig. S1.9. Boxplots of Lum as a function of PPCPs (C1 to C16).
      • fig. S1.10. Histogram (left panel), normal Q-Q plot (central panel) and jack after jackknife plot (right panel) for bootstrapped medians of Lum values (R = 999) for “treatment” = 1.
      • fig. S1.11. Bootstrapped (R = 999) medians and 95% CIs for the 180 treatments.
      • fig. S1.12. Tolerance level in the estimation of median values for lum for the 180 mixture effects.
      • fig. S1.13. EE μ*-σ plot for the 17 studied input factors (16 PPCPs and light intensity).
      • fig. S1.14. Ranked input factors by importance (μ*).
      • fig. S1.15. EE μ*-σ plot for the 17 studied input factors (16 PPCPs and light intensity).
      • fig. S1.16. Diagnostic plots of Anov.1 model.
      • fig. S1.17. Diagnostic plots of Anov.4 model.
      • fig. S1.18. Dose ranges of PPCPs (in nanograms per liter).
      • fig. S1.19. Histogram of the frequency distribution of the total sum of the 16 PPCPs in the 180 mixtures.
      • fig. S1.20. Rose plot presenting the relative abundance (counts) of each PPCP (variable) and level (value) of each PPCP in the 180 mixtures.
      • fig. S1.21. Rose plot presenting the mean concentration (value) of each PPCP (variable) in the 180 mixtures.
      • fig. S1.22. Dose-response data and fitted LL5 drm models (chem.1) to the experimental data according to CA model.
      • fig. S1.23. Schematic representation of the calculation sequence required to predict chemical mixture effects.
      • fig. S1.24. The 21 dose-response models (LL.5 models) fitted to the in silico–predicted dose-response patterns of the 21 unique combinations of C10, C11, and C14.
      • fig. S1.25. Observed versus predicted bioluminescence values of A. CPB4337 to the 180 low-dose mixtures of PPCPs.
      • fig. S1.26. FITEVAL report for the predicted versus experimental low-dose mixture effects of PPCPs.
      • fig. S1.27. Response of six community-level end points measured along time as a function of treatment on model freshwater benthic microbial communities.
      • fig. S1.28. Residual versus fits plot to check the quadratic mean-variance assumption of negative binomial regression (with different metabolic end points coded in different colors).
      • References (73109)

      Download PDF

      Other Supplementary Material for this manuscript includes the following:

      • S4 (Microsoft Excel format). Composition of the 180 mixtures

      Files in this Data Supplement:

    Navigate This Article