Skip to the content

Aquatic risks from human pharmaceuticals—modelling temporal trends of carbamazepine and ciprofloxacin at the global scale

Rik Oldenkamp1,2Arthur H W Beusen3,4 and Mark A J Huijbregts1,3

Published 22 February 2019  © 2019 The Author(s). Published by IOP Publishing Ltd
Environmental Research Letters, Volume 14, Number 3

Abstract

Despite the worldwide presence of pharmaceuticals in the aquatic environment, a comprehensive picture of their aquatic risk (AR) at the global scale has not yet been produced. Here, we present a procedure to estimate ARs of human pharmaceuticals at a freshwater ecoregion level. First, we predicted country- and year-specific per capita consumption with a regression model. Second, we calculated spatially explicit freshwater concentrations via a combination of mass balance models, addressing the pharmaceutical's fate in respectively humans, wastewater treatment plants and the environment. Finally, we divided the freshwater concentrations at the level of individual freshwater ecoregions with the regulatory limit value derived from toxicity tests to come to an ecoregion-specific AR. We applied our procedure to model time-trends (1995–2015) of ARs of carbamazepine and ciprofloxacin, two widely detected and regulatory relevant human use pharmaceuticals. Our analysis of carbamazepine and ciprofloxacin showed that ARs, due to exposure to these human pharmaceuticals, typically increased 10–20 fold over the last 20 years. Risks due to carbamazepine exposure were still typically low for the time period assessed (AR < 0.1), although some more densely populated and/or arid ecoregions showed higher ARs (up to 1.1). Risks for ciprofloxacin were found to be much higher with ARs larger than 1 for 223 out of 449 freshwater ecoregions in 2015. Comparison with measured concentrations in ten river basins showed that carbamazepine concentrations were predicted well. Concentrations of ciprofloxacin, measured in four river basins, were, however, generally underestimated by our model with one to two orders of magnitude. We conclude that our procedure provides a good starting point to evaluate ARs of a wide range of human pharmaceuticals at the global scale.

Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction

Active pharmaceutical ingredients (APIs) are consumed in large quantities (e.g. Van Boeckel et al2014) that are not completely removed during their passage through the human body and wastewater treatment plants (WWTPs). Consequently, their residues can be transported into the environment, and many APIs have been detected in surface waters globally (Hughes et al 2013, Guang et al 2014, aus der Beek et al 2016). Adverse effects on aquatic species have been observed at environmentally relevant concentrations for a wide range of (pharmaceutical classes of) APIs, an extensive overview of which is provided by Brausch et al (2012). In addition, recent studies have shown that chronic exposure to carbamazepine, an anti-epileptic drug, leads to altered feeding behaviour and reduced egg viability in zebrafish (da Silva Santos et al 2018), and lowers the reproductive output in crustaceans (Oropesa et al 2016). Antibiotics are a specific group of APIs of potential ecological concern since all major nutrient cycles in the environment depend on the activity of bacterial communities. Additionally, antibiotics entering WWTPs can affect bacterial communities used for biological degradation and consequently decrease their efficiency to remove other pollutants from the water (e.g. Schmidt et al 2012).

The aquatic risk (AR) of pharmaceuticals has been acknowledged as an emerging environmental problem requiring regulatory and scientific attention (Küster and Adler 2014). Because of technological and practical limitations, monitoring cannot provide a complete overview of the global situation. Therefore, fate models have been applied to estimate concentrations of APIs in several river basins in Europe (e.g. Schowanek and Webb 2002, Rowney et al 2009, Alder et al 2010, Hut et al2013, Johnson et al 2013a, 2013b, Boxall et al 2014), North America (e.g. Anderson et al 2004, Schwab et al 2005), South America (Archundia et al 2018), Asia (Hao et al 2015, Wannaz et al 2018), and Australia (Green et al 2013). These approaches are, however, generally limited in their spatial coverage, and despite the worldwide presence of APIs in the aquatic environment a comprehensive picture of ARs at the global scale has yet to be produced.

Here, we present modelled time-trends (1995–2015) of ARs of two APIs, i.e. carbamazepine and ciprofloxacin, in 449 aquatic ecoregions worldwide (Abell et al 2008). We chose to model carbamazepine and ciprofloxacin, a second-generation broad-spectrum antibiotic, because they are widely detected in surface waters (Hughes et al 2013, aus der Beek et al 2016), and because they were both identified as candidates for the Watch List of the European Water Framework Directive (Carvalho et al 2015). Our model predicts API emissions and aquatic concentrations at a 0.5° by 0.5° grid-cell resolution, and ARs are derived using publicly available predicted no effect concentrations (PNEC).

Methods

Consumption

Global and temporal coverage of API consumption was achieved through development of linear mixed effects models (LMMs) that incorporate socio-economic and demographic predictors expected to influence pharmaceutical consumption (Kookana et al 2014). Our LMMs extend current approaches that extrapolate between countries and years based on population dynamics alone (e.g. Johnson et al 2013b, Oldenkamp et al 2013b), or based on treatment patterns and disease prevalence (Ortiz de García et al 2013).

They were built for each API using a database containing 227 and 567 data points for carbamazepine and ciprofloxacin, respectively, each representing per capita total consumption for a unique combination of country and year, spanning from 1995 to 2015. Data were collected from open literature and online databases. Additionally, data on ciprofloxacin consumption were received from The European Surveillance System—TESSy, released by (European Centre for Disease Control (ECDC). The final database contained data on carbamazepine consumption in 69 countries covering all continents (32 in Europe, 19 in Asia, 8 in South-America, 4 in Central-/North-America, 4 in Africa, and 2 in Oceania), and data on ciprofloxacin consumption in 49 countries covering all continents except for Africa (32 in Europe, 5 in Asia, 7 in South-America, 4 in Central-/North-America, and 1 in Oceania). The supporting information provides a description of the data collection procedure including a list of countries and the years with consumption data available (section SI1).

Based on their supposed correlation with API consumption, predictors selected were the Human Development Index (United Nations Development Programme 2016), the ageing of the population (The World Bank 2016), and the time since first marketing of API of interest. The models were constructed using the 'lme' function of the package 'nlme' (Pinheiro et al 2016) in the R environment (R Development Core Team 2017). They were based on the log-transformed per capita consumption of carbamazepine and ciprofloxacin, incorporating fixed effects from all predictors including their quadratic terms. Random effects were included per country on the regression's intercept and on the slope over the years. Inclusion of these random effects was justified based on the decrease in Akaike Information Criterion score (AIC) compared to the full fixed-effects model. We employed the function 'dredge' of the package 'MuMIn' in the R environment (Barton 2015) to select the most parsimonious models based on their AIC score. Independence of predictors was checked and Variance Inflation Factors were all below 2, well below the threshold of 5 for unacceptable multi-collinearity as proposed by Zuur et al (2009). The model structure of the final LMMs is given by equation (1):

Equation (1)

In which M is the per capita consumption of the API of interest (mg/cap/yr), HDI is the Human Development Index (-), AGE is the ageing of the population (% of population ≥ 65 years), YR_M is the time since first marketing of API of interest (years), and (1 + YR_M) Country) reflects the random effect structure on the regression's intercept and slope over YR_M per country. Coefficients of predictors are represented by β, with subscripts 0 for the intercept, 1 for linear terms, and 2 for quadratic terms. Error term ε represents the residual variance not explained by the model.

Model performance was assessed via their marginal and conditional pseudo-R2 values, calculated according to Nakagawa and Schielzeth (2013) and Johnson (2014). Moreover, the predictive capacity Q2 of the fixed-effects part of the models was computed via a leave-one-cluster-out cross validation (Fang 2011). While the Q2 was smaller than the marginal pseudo-R2 values for both models (0.34 and 0.29 versus 0.27 and 0.15 for carbamazepine and ciprofloxacin, respectively), we considered this decrease sufficiently small to rule out unacceptable overfitting for both models. Table 1 contains the characteristics of the models, including their regression coefficients. Moreover, the supporting information contains partial dependence plots per combination of predictor and API, to aid in model interpretation (figure SI1.1 is available online at stacks.iop.org/ERL/14/034003/mmedia).

Table 1.  Characteristics of linear mixed effects regression models for per capita consumption of carbamazepine (CBZ) and ciprofloxacin (CIP).

  CBZ CIP
Coefficient Value (±SE) Value (±SE)
β0 −2.60 × 101 (±2.57) −1.40 × 101 (±4.31)
βHDI,1 6.98 × 101 (±6.73) 4.50 × 101 (±1.11 × 101)
βHDI,2 −4.00 × 101 (±4.20) −2.65 × 101 (±6.78)
βAGE,1 5.50 × 10−2 (±1.05 × 10−2) −2.71 × 10−1 (±7.15 × 10−2)
βAGE,2 0 8.60 × 10−3 (±2.43 × 10−3)
βYR_M,1 7.95 × 10−2 (±1.66 × 10−2) 2.11 × 10−1 (±1.80 × 10−2)
βYR_M,2 −1.22 × 10−3 (±1.73 × 10−3) −3.87 × 10−3 (±3.95 × 10−4)
Performance Value Value
Marginal pseudo-R2 0.34 0.29
Conditional pseudo-R2 0.99 0.95
pseudo-Q2 0.27 0.15

In our model calculations for API consumption, each unique combination of country and year was preferably assigned a consumption value extracted from our database. If this was not possible, we applied the conditional models to estimate the consumption, provided that our database contained data for the country of interest in at least one other year. For countries not included in the database, we predicted consumption using the fixed-effects part of the models only.

Aquatic concentrations

The hydrological model PCRGLOBWB (PCRaster GLOBal Water Balance model) calculates the global hydrology and water resources at a grid-cell resolution of 0.5° by 0.5° (Van Beek and Bierkens 2008). Here we coupled the discharge from PCRGLOBWB with consecutive spatially explicit emission modules for consumption, human metabolism, WWTP-treatment, and environmental fate of APIs to calculate predicted environmental concentrations (PEC) for the years 1995–2015. Steady-state concentrations in surface water were estimated per spatial grid cell i in PCRGLOBWB via equation (2):

Equation (2)

 

In which Ei is the emission into cell i (kg d−1; see equation (3)), Qi is the mean annual discharge in cell i (m3 d−1), and Vi is the water volume in cell i (m3), to derive a cell-specific downstream PEC (PECi; g l−1). Five loss processes j are accounted for via cell-specific first order rate constants (kj,  i; extra    space d−1), being the three degradation processes biodegradation, photolysis and hydrolysis, and the two intermedia processes sedimentation and volatilization. Individual degradation rate constants were derived from test conditions, and translated into field conditions by accounting for temperature differences, sorption to suspended solids and dissolved organic carbon (e.g. Honti et al2016) and reduced light intensity (Schwarzenbach et al 1993). Local sedimentation and volatilization rate constants were implemented via mass transport velocities between media, similar to Fantke et al(2017) and Margni et al (2004). Detailed information on the derivation of all rate constants can be found in the supporting information (section SI2).

Emissions into surface water in cell i (Ei; kg d−1) were calculated via equation (3):

Equation (3)

 

In which the per capita API consumption (Mi; kg/cap/d) and the local urban population (Purb,i; cap), are combined to derive the total API consumption in the urban part of cell i. PCRGLOBWB considers the non-urban rural population to have on-site treatment (e.g. local septic tanks), from which no emissions to surface water take place. After consumption, APIs are partly absorbed from the intestinal tract and subsequently metabolized. The remaining fraction of the administered dose is eliminated unchanged as parent compound from the body, either egested via faeces (fpc,f; -) or excreted via urine (fpc,u; -). Unchanged API might be transported with wastewater towards a WWTP, depending on the fraction of the urban population connected to wastewater treatment (fconn,i; -). In cells with a WWTP-connectivity below 100%, the fraction excreted via urine for the non-connected part was assumed to be discharged directly into surface water. The fate of APIs during wastewater treatment was estimated using the multimedia model SimpleTreat 4.0 (Struijs 2014). SimpleTreat 4.0 distinguishes between WWTPs that apply primary treatment only and those that combine this with secondary treatment with activated sludge. In equation (3), this is incorporated with Boolean parameter ai indicating the presence or absence of merely primary treatment or primary treatment combined with secondary treatment in cell i, and with the corresponding removal fractions frem,i.

Finally, the chemical inflow into cell i from the upstream cell (Eup,i; kg d−1) is calculated as:

Equation (4)

 

Table SI2.1 contains all parameter values used for the emission and fate calculations of carbamazepine and ciprofloxacin.

Comparison with measured concentrations

Concentration estimations for carbamazepine and ciprofloxacin were compared with measured concentrations included in the global database from the Umwelt Bundesamt, the German national environmental protection agency (aus der Beek et al 2016). From this database, all records were extracted reporting on the presence of carbamazepine or ciprofloxacin in fresh surface waters, provided that the sampling year(s) and sampling location were reported. To enable a meaningful comparison with model predictions, we allocated individual records to hydrologically distinct river basins. Records were excluded if they could not be allocated to a specific river basin based on their reported sampling locations. The original literature sources were consulted for those records representing aggregate average concentrations over multiple sampling points or over multiple years. Where possible, concentrations for unique combinations of sampling location and year were extracted from the original sources. Otherwise, the records were excluded from the analysis. Finally, river basins were only included in the validation exercise if measured concentrations were available for at least five locations in the river basin, with at least 20% of them above their limit of detection. For these river basins, box-whisker plots were constructed with the left-censored data substituted via non-parametric robust regression on order statistics (ROS) using the 'cenboxplot' function of the package 'NADA' in the R environment (Lee 2017). These were compared with box-whisker plots on modelled concentrations for grid cells in the respective river basins. Cells were excluded for which zero concentrations were predicted, because they generally represent upstream areas where measurements are not performed.

Aquatic risks (ARs)

We calculated yearly ARs for carbamazepine and ciprofloxacin over a period of 21 years (1995–2015), for 449 freshwater ecoregions (Abell et al 2008). Ecoregions are distinct geospatial units that represent unique patterns of environmental and ecological variables. As such, ecoregions are considered particularly useful for risk assessment and conservation planning at a larger scale (Schäfer et al 2013, Szöcs et al 2017). The AR for each freshwater ecoregion b (ARb) was calculated as the average risk-quotient over its total water volume (equation (5)).

Equation (5)

 

In which wiεb is the weight of grid cell i in ecoregion b, based on the water volume of the cell. The larger the water volume, the larger the weight of that cell. The PNEC is the PNEC which is 500 ng l−1for carbamazepine (Ferrari et al 2004, Moermond and Smit 2016) and 0.15 ng l−1 for ciprofloxacin (Załęska-Radziwiłł et al 2014).

As an additional measure of risk, we also calculated the percentage of the water volume per ecoregion exceeding the PNEC (equation (6)).

Equation (6)

 

In which diεb is a Boolean parameter indicating whether or not the PEC of grid cell i in ecoregion bexceeds the PNEC.

Results

Consumption

Applying the regression models from 1995–2015, we found that global carbamazepine consumption increased from an estimated 742 tonnes in 1995, to an estimated 1214 tonnes in 2015 (figure 1). Our estimation of global ciprofloxacin consumption shows a much larger increase: by 2015, 2318 tonnes of ciprofloxacin were consumed worldwide, more than seven times the estimated consumption of 298 tonnes in 1995. It should be noted, however, that these numbers represent total tonnages, and are a combined result of increases in per capita consumption as well as in population size. Indeed, the strong increase in total ciprofloxacin consumption in Asia (figure 1(b)), is less pronounced when consumption is expressed on a per capita basis.

Figure 1.

Figure 1. Temporal trends in global consumption of carbamazepine ((a); green) and ciprofloxacin ((b); purple), differentiated between continents.

 Standard image High-resolution image Export PowerPoint slide

Concentrations

Despite its larger consumption volume and its lower degradation in the human body (table SI2.1), concentrations of ciprofloxacin, aggregated at the ecoregion level, were similar to those of carbamazepine (figure 2). Moreover, aquatic concentrations of carbamazepine and ciprofloxacin in the typical (median) ecoregion showed the same temporal trend, increasing at an average rate of 13% and 16% per year, respectively. Interestingly, ecoregions have converged over the years with respect to their carbamazepine concentrations, as indicated by the interquartile range in figure 2(a). This pattern was not found for ciprofloxacin, for which the lower bound of the interquartile range stayed below 10−8 ng l−1 over the whole period 1995–2015 (figure 2(b)).

Figure 2.

Figure 2. Concentrations of carbamazepine ((a); green) and ciprofloxacin ((b); purple) over the period 1995–2015. Dots: individual ecoregions with concentrations over 10−8 ng l−1; solid lines: median ecoregion; dashed lines: interquartile range over all ecoregions. Ciprofloxacin concentrations were always lower than 10−8 ng l−1 in 25% of ecoregions.

 Standard image High-resolution image Export PowerPoint slide

The Sankey diagrams in figure 3 show the global mass flows of both APIs in 2015, with the thickness of respective ribbons reflecting the absolute amounts (tonnes yr−1). Dark shaded ribbons represent transport flows and light shaded ribbons represent actual removal and retention. Similar figures for the years 1995–2014 are included in the supporting information (section SI.3). Figure 2(B) shows that ciprofloxacin has a relatively large in-stream removal. This is the result of its higher removal in surface waters, mainly due to photolysis (table SI2.1), while sedimentation contributes very little to dissipation of ciprofloxacin from surface waters (<1%). In fact, no more than 4% of all ciprofloxacin emitted into fresh surface waters reaches the sea, compared to 79% of all carbamazepine. As a consequence, carbamazepine concentrations vary less throughout the river systems than do concentrations of ciprofloxacin.

Figure 3.

Figure 3. Global mass flows (tonnes) of carbamazepine ((a); green) and ciprofloxacin ((b); purple) in 2015. Dark shaded ribbons represent transports flows; light shaded ribbons represent actual removal and retention. The thickness of respective ribbons represents the absolute amounts (tonnes yr−1).

 Standard image High-resolution image Export PowerPoint slide

Our modelled carbamazepine concentrations show good agreement with measurements in nine river basins throughout Europe and the Mississippi basin in North America, as is shown by the overlap between box-whisker plots of measured versus modelled concentrations (figure 4). In contrast with the ten river basins for carbamazepine, only four river basins were identified where ciprofloxacin was detected in more than 20% of the measurements (figure 4). This comes as no surprise considering that typical detection limits for ciprofloxacin are in the 1–100 ng l−1 range (aus der Beek et al 2016). Two of these basins were located in Asia (Yangtze and Chao Phraya basins), one in Europe (Ebro basin), and one in North-America (Mississippi basin), and in all four of them predicted ciprofloxacin concentrations were generally one to two orders of magnitude lower than the measurements.

Figure 4.

Figure 4. Box-whisker plots comparing measured concentrations (yellow) and modelled concentrations (green: carbamazepine CBZ; purple: ciprofloxacin CIP) in several river basins. Boxes represent interquartile range, with black horizontal line the median concentration; whiskers represent minimum and maximum. Left-censored data (<LOD) were substituted via non-parametric robust regression on order statistics (ROS) using the 'NADA' package for R (Lee, 2017). Red lines represent the largest LODs reported for that river basin. River basins were visualized using the HydroSHEDS drainage network (Lehner et al 2008).

 Standard image High-resolution image Export PowerPoint slide

Aquatic risks

In line with the increasing consumption (figure 1) and concentrations (figure 2), ARs at the ecoregion level increased over time for both APIs (figure 5). The ARs of carbamazepine stayed, however, below 1 for all but one of the 449 ecoregions (figure 5(b)). The ecoregions with relatively large ARs, i.e. 15 ecoregions with a value above 0.1, mainly represented densely populated areas such as western and central Europe, or water-scarce arid regions such as the Arabian and Californian peninsulas. Moreover, throughout the 21 year period the % of water volume exceeding the PNEC in any ecoregion was never larger than 50%.

Figure 5.

Figure 5. Aquatic risks (AR) per ecoregion, expressed as the PEC/PNEC ratios for the total water volume, for carbamazepine (a), (b) and ciprofloxacin (c), (d) in 1995 and 2015. Inset: counts of ecoregions where specific percentage of the water volume exceeds the PNEC.

 Standard image High-resolution image Export PowerPoint slide

In contrast with carbamazepine, the AR of ciprofloxacin exceeded 0.1 in 196 ecoregions and 1 in 112 ecoregions in 1995 (figure 5(c)). These numbers further increased over time, leading to 282 ecoregions with an AR higher than 0.1 and 214 ecoregions with an AR higher than 1 in 2015 (figure 5(d)). In 6 ecoregions, 90% or more of the freshwater volume exceeded the PNEC for ciprofloxacin. These were small and often relatively densely populated ecoregions, such as the coastal region of Israel, the Canary Islands off the African coast, and the area around Rio de Janeiro in Brazil.

Discussion

We showed that it is indeed possible to predict temporal changes in spatially explicit ARs at the global scale for human pharmaceuticals. Our modelling is, however, not without uncertainty. General practice in chemical risk assessment is to follow the consecutive steps of use, emission, environmental fate and effect. Previous research has shown that the first and last of these steps contribute most to the uncertainty in the final risk estimation (Oldenkamp et al 2016). Below, we reflect on the uncertainties encountered during the emission, effect and fate estimations, respectively.

One major hurdle to overcome when performing a sound global environmental risk assessment of APIs, is to reliably estimate their country- and year-specific consumption. Not only are consumption data scarcely and spatially not homogeneously available, consumption also varies substantially between countries and years (Van Boeckel et al 2014, aus der Beek et al 2016, Klein et al 2018). To make optimal use of the available data, we extrapolated per capita consumption to all countries globally using mixed effects regression models (table 1). The models describing carbamazepine and ciprofloxacin consumption were based on a set of readily available predictor variables (equation (1)). The regression coefficients in table 1 and the partial dependence plots in the supporting information (figure SI1.1), indicate that consumption of both APIs increases with HDI, although this trend levels off above HDI values of ~0.8. In addition, their consumption follows traditional trends of initial increase and subsequent decline in the years after introduction. In accordance with Zhang and Geißen (2010), carbamazepine consumption increases with ageing of the population.

Fixed effect predictors explained 29%–34% of the variance in per capita consumption, indicating that extrapolation to countries without any available consumption data is particularly uncertain. Furthermore, the uneven geographical coverage of consumption data might have resulted in additional uncertainties in our estimations for low- and middle-income countries outside of Europe. Nevertheless, our estimation of total global consumption of carbamazepine in 2007, based on the country specific estimates (968 tonnes; figure 1) is in accordance with actual sales data from IMS Health for 96% of the global market that year (942 tonnes; Zhang et al 2008). Moreover, Van Boeckel et al (2014) and Klein et al (2018) report that total antibiotic consumption was stable or had moderately decreased in high-income countries between 2000 and 2010. Klein et al (2018) further estimated an increase in the consumption rate of fluoroquinolones, the subgroup of antibiotics to which ciprofloxacin belongs, in low- and middle-income countries, while it decreased in high-income countries. This supports our estimation of global ciprofloxacin consumption, which levels off towards 2015 (figure 1). It should be pointed out, however, that our estimations of ciprofloxacin consumption in the years after 2010 were largely based on data originating from European countries as provided by ECDC. While this might have introduced a bias in our estimation of global ciprofloxacin consumption, similar temporal trends of levelling-off ciprofloxacin consumption, somewhat delayed compared to the high-income countries, are reported for a variety of other countries, such as Iran (Abdollahiasl et al 2011), Thailand (Apisarnthanarak and Mundy, 2008), Serbia (Veličković-Radovanović et al 2010) and Venezuela (Rivas and Alonso, 2011).

While PNECs should reflect the concentration at which adverse ecological effects do likely not occur (EMA, 2006), they are generally based on a limited number of ecotoxicity data derived under laboratory conditions. To account for uncertainties in the necessary interspecies and lab-to-field extrapolations, PNECs are derived via application of assessment factors to the lowest available ecotoxicity endpoint, with the magnitude of these assessment factors depending on data quality and quantity. This approach has led to a very strict PNEC for ciprofloxacin at 0.15 ng l−1, based on a chronic No Observed Effect Concentration for the sensitive bioluminescent bacterium Vibrio fischeri, combined with an assessment factor of 10 (Załęska-Radziwiłł et al 2014). For comparison, the PNEC for ciprofloxacin when based only on the sensitivity of the standard taxonomic groups of fish, crustaceans, and algae, would be 100 000 times less strict (at 15.6 μg l−1; Załęska-Radziwiłł et al2011). In fact, while our calculations indicate almost half of all ecoregions worldwide to have an unacceptable AR due to chronic exposure to ciprofloxacin, the PNEC based on the three standard taxonomic groups is not exceeded anywhere.

Finally, the comparison of predicted with measured concentrations provides confidence that our model can properly estimate concentrations of carbamazepine, but to a lesser extent ciprofloxacin. The underestimation of measured ciprofloxacin concentrations should at least partially be attributed to analytical limitations and non-representative sampling locations. An additional explanation may be that we did not consider four potential emission sources of ciprofloxacin in our modelling. Non-prescription use of ciprofloxacin might account for 19%–100% of the total consumption of antibiotics outside northern Europe and North America (Morgan et al 2011). Furthermore, veterinary use of ciprofloxacin, while not approved in either the USA (FDA, 2016) or Europe (EMA, 2016), can be widespread in other countries. Indeed, ciprofloxacin has been detected in manure from livestock (i.e. chicken, cow, and pig) originating from several Chinese provinces (Hu et al 2008, Zhao et al 2010). Moreover, ciprofloxacin concentrations can be substantial in biosolids applied to agricultural soils (Verlicchi and Zambello 2015). Biosolids can thus form a significant emission pathway when a considerable part of the excess sludge is disposed of this way (Oldenkamp et al 2013a). For example, 75% of all Spanish excess sludge was applied on agricultural soils in 2012 (Eurostat 2017). However, conservative calculations we performed with the multimedia box model SimpleBox 4.0 (Hollander et al 2016), indicated that transfer of ciprofloxacin from agricultural soils to surface waters was negligible, as ciprofloxacin adsorbs strongly to soils. Finally, we assumed that APIs emitted into rural wastewater do not reach surface waters, resulting in 100% rural retention (figure 3). In reality, however, leakage of pharmaceuticals into ground water from septic tanks and pit latrines may not be negligible (Del Rosario et al 2014, Phillips et al 2015, Schaider et al 2016, Hanamoto et al 2018), showing that this could form an additional route into surface waters that is not yet understood well enough to be accounted for at a global scale.

Photodegradation is the most important environmental degradation pathway for various APIs (Wang and Lin 2014), and the extensive in-stream removal as estimated for ciprofloxacin can be largely contributed to this process (figure 3). The photodegradation rate constant for ciprofloxacin (1.41 × 102 d−1; table SI2.1) is multiple orders of magnitude higher than that of carbamazepine, explaining the difference in environmental removal between the two APIs. We obtained this rate constant as the average of 12 experimentally derived values. Although data were relatively abundant, values ranged from 5.76 d−1 (Belden et al 2007) to 7.85 × 102 d−1 (Babic et al 2013). As a consequence, we might have overestimated the actual environmental photodegradation of ciprofloxacin, providing another possible explanation for the underestimation of measured concentrations.

To further strengthen confidence in the environmental fate calculations, a more extensive validation covering a wider geographical range should be performed, ideally focusing on densely populated and arid ecoregions with relatively high environmental concentrations, as identified in our study.

Conclusions

Time trends of yearly ARs from human pharmaceuticals can be derived with reasonable confidence for all river basins worldwide, by combining spatially explicit chemical fate and effect modelling with predictions of pharmaceutical consumption. To our knowledge, this is the first time such an exercise has been performed for specific APIs at a global scale. Our analysis has shown that ARs due to carbamazepine exposure are typically low, but that risks due to ciprofloxacin exposure resulting from human use are expected to be extensive and widespread. Applying our method to a wider range of (classes of) APIs will provide further insight in the general applicability of our approach.

Acknowledgments

We would like to thank Ana Benítez-López and Luca Santini for their valuable suggestions and help in the development of the consumption models.

Disclaimer ECDC:

'The views and opinions of the authors expressed herein do not necessarily state or reflect those of ECDC. The accuracy of the authors' statistical analysis and the findings they report are not the responsibility of ECDC. ECDC is not responsible for conclusions or opinions drawn from the data provided. ECDC is not responsible for the correctness of the data and for data management, data merging and data collation after provision of the data. ECDC shall not be held liable for improper or incorrect use of the data'.

References

  • Abdollahiasl A, Kebriaeezadeh A, Nikfar S, Farshchi A, Ghiasi G and Abdollahi M 2011 Patterns of antibiotic consumption in Iran during 2000–2009 Int. J. Antimicrobial Agents 37 489–90

    Crossref

  • Abell R et al 2008 Freshwater ecoregions of the world: a new map of biogeographic units for freshwater biodiversity conservation BioScience 58 403–14

    Crossref

  • Alder A C, Schaffner C, Majewsky M, Klasmeier J and Fenner K 2010 Fate of beta-blocker human pharmaceuticals in surface water: comparison of measured and simulated concentrations in the Glatt Valley Watershed, Switzerland Water Res. 44 936–48

    Crossref

  • Anderson P D et al 2004 Screening analysis of human pharmaceutical compounds in US surface waters Environ. Sci. Technol. 38 838–49

    Crossref

  • Apisarnthanarak A and Mundy L M 2008 Correlation of antibiotic use and antimicrobial resistance in Pratumthani, Thailand, 2000 to 2006 Am. J. Infection Control 36 681–2

    Crossref

  • Archundia D, Boithias L, Duwig C, Morel M C, Flores Aviles G and Martins J M F 2018 Environmental fate and ecotoxicological risk of the antibiotic sulfamethoxazole across the Katari catchment (Bolivian Altiplano): application of the GREAT-ER model Sci. Total Environ. 622–623 1046–55

    Crossref

  • aus der Beek T, Weber F-A, Bergmann A, Hickmann S, Ebert I, Hein A and Küster A 2016 Pharmaceuticals in the environment—global occurrences and perspectives Environ. Toxicol. Chem. 35823–35

    Crossref

  • Babic S, Perisa M and Skoric I 2013 Photolytic degradation of norfloxacin, enrofloxacin and ciprofloxacin in various aqueous media Chemosphere 91 1635–42

    Crossref

  • Barton K 2015 MuMIn: Multi-Model Inference (R Package) (version 1.15.6)
  • Belden J B, Maul J D and Lydy M J 2007 Partitioning and photo degradation of ciprofloxacin in aqueous systems in the presence of organic matter Chemosphere 66 1390–5

    Crossref

  • Boxall A B A, Keller V D J, Straub J O, Monteiro S C, Fussell R and Williams R J 2014 Exploiting monitoring data in environmental exposure modelling and risk assessment of pharmaceuticals Environ. Int. 73 176–85

    Crossref

  • Brausch J M, Connors K A, Brooks B W and Rand G M 2012 Human pharmaceuticals in the aquatic environment: a review of recent toxicological studies and considerations for toxicity testing Reviews of Environmental Contamination and Toxicology vol 218 ed D M Whitacre (New York: Springer) pp 1–99
  • Carvalho R N, Ceriani L, Ippolito A and Lettieri T 2015 Development of the first watch list under the environmental quality standards directive Joint Research Centre (Luxembourg: European Commission)
  • da Silva Santos N, Oliveira R, Lisboa C A, Mona e Pinto J, Sousa-Moura D, Camargo N S, Perillo V, Oliveira M, Grisolia C K and Domingues I 2018 Chronic effects of carbamazepine on zebrafish: behavioral, reproductive and biochemical endpoints Ecotoxicol. Environ. Saf. 164 297–304

    Crossref

  • Del Rosario K L, Mitra S, Humphrey C P and O’Driscoll M A 2014 Detection of pharmaceuticals and other personal care products in groundwater beneath and adjacent to onsite wastewater treatment systems in a coastal plain shallow aquifer Sci. Total Environ. 487 216–23

    Crossref

  • EMA 2006 Guideline on the Environmental Risk Assessment of Medicinal Products for Human Use(London: European Medicines Agency)
  • EMA 2016 European Surveillance of Veterinary Antimicrobial Consumption Sales of veterinary antimicrobial agents in 29 European countries in 2014 EMA/61769/2016
  • Eurostat 2017 Sewage sludge production and disposal from urban wastewater (in dry substance (d.s)) [ten00030] (http://ec.europa.eu/eurostat/tgm/refreshTableAction.do?tab=table&plugin=1&pcode=ten00030&language=en) (Accessed: 13 September 2017) European Union
  • Fang Y 2011 Asymptotic equivalence between cross-validations and Akaike information criteria in mixed-effects models J. Data Sci. 9 15–21
  • Fantke P et al 2017 USEtox® 2.0 Documentation (version 1) (https://doi.org/10.11581/DTU:00000011)
  • FDA 2016 Approved Animal Drug Products (Green Book); Section 2—Active Ingredients (Rockville: US Food and Drug Administration)
  • Ferrari B, Mons R, Vollat B, Fraysse B, Paxeus N, Lo Giudice R, Pollio A and Garric J 2004 Environmental risk assessment of six human pharmaceuticals: are the current environmental risk assessment procedures sufficient for the protection of the aquatic environment? Environ. Toxicol. Chem. 23 1344–54

    Crossref

  • Green C, Williams R, Kanda R, Churchley J, He Y, Thomas S, Goonan P, Kumar A and Jobling S 2013 Modeling of steroid estrogen contamination in UK and South Australian rivers predicts modest increases in concentrations in the future Environ. Sci. Technol. 47 7224–32

    Crossref

  • Guang Y, Maohong F and Guangming Z 2014 Emerging contaminants in surface waters in China—a short review Environ. Res. Lett. 9 074018

    IOPscience

  • Hanamoto S, Nakada N, Yamashita N and Tanaka H 2018 Source estimation of pharmaceuticals based on catchment population and in-stream attenuation in Yodo river watershed, Japan Sci. Total Environ. 615 964–71

    Crossref

  • Hao X, Cao Y, Zhang L, Zhang Y and Liu J 2015 Fluoroquinolones in the Wenyu river catchment, China: occurrence simulation and risk assessment Environ. Toxicol. Chem. 34 2764–70

    Crossref

  • Hollander A, Schoorl M and van de Meent D 2016 Simplebox 4.0: improving the model while keeping it simple… Chemosphere 148 99–107

    Crossref

  • Honti M, Hahn S, Hennecke D, Junker T, Shrestha P and Fenner K 2016 Bridging across OECD 308 and 309 Data in Search of a Robust Biotransformation Indicator Environ. Sci. Technol. 50 6865–72

    Crossref

  • Hu X-G, Yi L, Zhou Q-X and Xu L 2008 Determination of thirteen antibiotics residues in manure by solid phase extraction and high performance liquid chromatography Chin. J. Anal. Chem. 36 1162–6

    Crossref

  • Hughes S R, Kay P and Brown L E 2013 Global synthesis and critical evaluation of pharmaceutical data sets collected from river systems Environ. Sci. Technol. 47 661–77

    Crossref

  • Hut R, Van de Giesen N and Houtman C J 2013 Medicinal footprint of the population of the Rhine basin Environ. Res. Lett. 8 044057

    IOPscience

  • Johnson A C, Dumont E, Williams R J, Oldenkamp R, Cisowska I and Sumpter J P 2013a Do concentrations of ethinylestradiol, estradiol, and diclofenac in european rivers exceed proposed eu environmental quality standards? Environ. Sci. Technol. 47 12297–304

    Crossref

  • Johnson A C, Oldenkamp R, Dumont E and Sumpter J P 2013b Predicting concentrations of the cytostatic drugs cyclophosphamide, carboplatin, 5-fluorouracil, and capecitabine throughout the sewage effluents and surface waters of europe Environ. Toxicol. Chem. 32 1954–61

    Crossref

  • Johnson P C D 2014 Extension of Nakagawa & Schielzeth’s R2GLMM to random slopes models Methods Ecol. Evol. 5 944–6

    Crossref

  • Klein E Y, Van Boeckel T P, Martinez E M, Pant S, Gandra S, Levin S A, Goossens H and Laxminarayan R 2018 Global increase and geographic convergence in antibiotic consumption between 2000 and 2015 Proc. Natl Acad. Sci. 115 E3463–E3470

    Crossref

  • Kookana R S, Williams M, Boxall A B A, Larsson D G J, Gaw S, Choi K, Yamamoto H, Thatikonda S, Zhu Y-G and Carriquiriborde P 2014 Potential ecological footprints of active pharmaceutical ingredients: an examination of risk factors in low-, middle- and high-income countries Phil. Trans. R. Soc. B 369

    Crossref

  • Küster A and Adler N 2014 Pharmaceuticals in the environment: scientific evidence of risks and its regulation Phil. Trans. R. Soc. B 369

    Crossref

  • Lee L 2017 NADA: Nondetects and Data Analysis for Environmental Data (R Package) (version 1.6-1)
  • Lehner B, Verdin K and Jarvis A 2008 Eos, Trans. Am. Geophys. Union 89 93–4

    Crossref

  • Margni M, Pennington D W, Bennett D H and Jolliet O 2004 Cyclic exchanges and level of coupling between environmental media: intermedia feedback in multimedia fate models Environ. Sci. Technol.38 5450–7

    Crossref

  • Moermond C T and Smit C E 2016 Derivation of water quality standards for carbamazepine, metoprolol, and metformin and comparison with monitoring data Environ. Toxicol. Chem. 35 882–88

    Crossref

  • Morgan D J, Okeke I N, Laxminarayan R, Perencevich E N and Weisenberg S 2011 Non-prescription antimicrobial use worldwide: a systematic review Lancet Infectious Dis. 11 692–701

    Crossref

  • Nakagawa S and Schielzeth H 2013 A general and simple method for obtaining R2 from generalized linear mixed-effects models Methods Ecol. Evol. 4 133–42

    Crossref

  • Oldenkamp R, Huijbregts M A J, Hollander A, Versporten A, Goossens H and Ragas A M J 2013a Spatially explicit prioritization of human antibiotics and antineoplastics in Europe Environ. Int. 51 13–26

    Crossref

  • Oldenkamp R, Huijbregts M A J, Hollander A, Versporten A, Goossens H and Ragas A M J 2013b Spatially explicit prioritization of human antibiotics and antineoplastics in Europe Environ. Int. 51 13–26

    Crossref

  • Oldenkamp R, Huijbregts M A J and Ragas A M J 2016 The influence of uncertainty and location-specific conditions on the environmental prioritisation of human pharmaceuticals in Europe Environ. Int. 91 301–11

    Crossref

  • Oropesa A L, Floro A M and Palma P 2016 Assessment of the effects of the carbamazepine on the endogenous endocrine system of Daphnia magna Environ. Sci. Pollut. Res. 23 17311–21

    Crossref

  • Ortiz de García S, Pinto Pinto G, García Encina P and Irusta Mata R 2013 Consumption and occurrence of pharmaceutical and personal care products in the aquatic environment in Spain Sci. Total Environ. 444 451–65

    Crossref

  • Phillips P J, Schubert C, Argue D, Fisher I, Furlong E T, Foreman W, Gray J and Chalmers A 2015 Concentrations of hormones, pharmaceuticals and other micropollutants in groundwater affected by septic systems in New England and New York Sci. Total Environ. 512–513 43–54

    Crossref

  • Pinheiro J, Bates D, DebRoy S, Sarkar D, EISPACK authors, Heisterkamp S and Van Willigen B 2016 NLME: Linear and Nonlinear Mixed Effects Models (R Package) (version 3.1-128)
  • R Development Core Team 2017 R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing) (Vienna) (version 3.4.3)
  • Rivas P and Alonso G 2011 Regulations governing the dispensing of medications and their effect on antibiotic consumption in Venezuela Rev. Panamericana Salud Pública 30 592–7

    Crossref

  • Rowney N C, Johnson A C and Williams R J 2009 Cytotoxic drugs in drinking water: a prediction and risk assessment exercise for the thames catchment in the United Kingdom Environ. Toxicol. Chem. 282733–43

    Crossref

  • Schäfer R B, Gerner N, Kefford B J, Rasmussen J J, Beketov M A, de Zwart D, Liess M and von der Ohe P C 2013 How to characterize chemical exposure to predict ecologic effects on aquatic communities? Environ. Sci. Technol. 47 7996–8004

    Crossref

  • Schaider L A, Ackerman J M and Rudel R A 2016 Septic systems as sources of organic wastewater compounds in domestic drinking water wells in a shallow sand and gravel aquifer Sci. Total Environ.547 470–81

    Crossref

  • Schmidt S, Winter J and Gallert C 2012 Long-term effects of antibiotics on the elimination of chemical oxygen demand, nitrification, and viable bacteria in laboratory-scale wastewater treatment plants Arch. Environ. Contamination Toxicol. 63 354–64

    Crossref

  • Schowanek D and Webb S 2002 Exposure simulation for pharmaceuticals in European surface waters with GREAT-ER Toxicol. Lett. 131 39–50

    Crossref

  • Schwab B W, Hayes E P, Fiori J M, Mastrocco F J, Roden N M, Cragin D, Meyerhoff R D, D’Aco V J and Anderson P D 2005 Human pharmaceuticals in US surface waters: a human health risk assessment Regulatory Toxicol. Pharmacol. 42 296–312

    Crossref

  • Schwarzenbach R P, Gschwend P M and Imboden D M 1993 Photochemical transformation reactions Environmental Organic Chemistry (New York: Wiley) pp 436–84
  • Struijs J 2014 SimpleTreat 4.0: a model to predict fate and emission of chemicals in wastewater treatment plants RIVM Report 601353005
  • Szöcs E, Brinke M, Karaoglan B and Schäfer R B 2017 Large scale risks from agricultural pesticides in small streams Environ. Sci. Technol. 51 7378–85

    Crossref

  • The World Bank 2016 Population ages 65 and above (% of total) (World bank staff estimates based on age/sex) (distributions of United Nations population division’s world population prospects) (http://data.worldbank.org)
  • United Nations Development Programme 2016 Human development index (HDI) (The Human Development Report Office) (http://hdr.undp.org/en/data) (Accessed: 5 April 2017)
  • Van Beek L P H and Bierkens M F P 2008 The Global Hydrological Model PCR-GLOBWB: Conceptualization, Parameterization and Verification, Department of Physical Geography, Faculty of Earth Sciences (Utrecht: Utrecht University)
  • Van Boeckel T P, Gandra S, Ashok A, Caudron Q, Grenfell B T, Levin S A and Laxminarayan R 2014 Global antibiotic consumption 2000 to 2010: an analysis of national pharmaceutical sales data Lancet Infectious Dis. 14 742–50

    Crossref

  • Veličković-Radovanović R, Kodela B, Petrović J, Mitić R and Rančić M 2010 Outpatient antibiotic use in primary healthcare of Niš region Acta Fac. Med. Naiss 27 27–32
  • Verlicchi P and Zambello E 2015 Pharmaceuticals and personal care products in untreated and treated sewage sludge: occurrence and environmental risk in the case of application on soil—a critical review Sci. Total Environ. 538 750–67

    Crossref

  • Wang X-H and Lin A Y-C 2014 Is the phototransformation of pharmaceuticals a natural purification process that decreases ecological and human health risks? Environ. Pollut. 186 203–15

    Crossref

  • Wannaz C, Franco A, Kilgallon J, Hodges J and Jolliet O 2018 A global framework to model spatial ecosystems exposure to home and personal care chemicals in Asia Sci. Total Environ. 622–623 410–20

    Crossref

  • Załęska-Radziwiłł M, Affek K and Rybak J 2014 Ecotoxicity of chosen pharmaceuticals in relation to micro-organisms—risk assessment Desalination Water Treat. 52 3908–17

    Crossref

  • Załęska-Radziwiłł M, Łebkowska M, Affek K and Zarzeczna A 2011 Environmental risk assessment of selected pharmaceuticals present in surface waters in relation to animals Arch. Environ. Prot. 37 31–42
  • Zhang Y and Geißen S-U 2010 Prediction of carbamazepine in sewage treatment plant effluents and its implications for control strategies of pharmaceutical aquatic contamination Chemosphere 801345–52

    Crossref

  • Zhang Y, Geißen S-U and Gal C 2008 Carbamazepine and diclofenac: removal in wastewater treatment plants and occurrence in water bodies Chemosphere 73 1151–61

    Crossref

  • Zhao L, Dong Y H and Wang H 2010 Residues of veterinary antibiotics in manures from feedlot livestock in eight provinces of China Sci. Total Environ. 408 1069–75

    Crossref

  • Zuur A F, Ieno E N, Walker N J, Saveliev A A and Smith G M 2009 Violation of Independence—Part I, Mixed Effects Models and Extensions in Ecology with R (New York: Springer) pp 143–60

    Crossref

About the author

Dr. Howard Dryden

Dr. Howard Dryden

Dr. Dryden has unique knowledge combination of biology, chemistry and technology and is the inventor of the activated, bio-resistant filter media AFM®. Dr. Dryden is one of the world`s leading experts in sustainable water treatment.

comments powered by Disqus

Wherever you are and wherever you live, there is no safe haven from the toxic wave of chemical pollution

Dr. Howard Dryden, CSO

Goes Foundation

3/2 Boroughloch Square

Clean Water Wave

The Meadows

Edinburgh

EH8 9NJ

Email. howard@goesfoundation.com