2. 138 C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146
wastewater treatment plants, and if their removal is insufficient,
they may end up in the groundwater and surface water. Antibiotics
reaching the environment also originate from veterinary drugs
used as growth promoters and for treatment of diseases in live
stock (Nonga et al., 2010). This study also found that 65% of the
farmers administered antibiotics without consulting a veterinar-
ian and 100% of the eggs investigated contained antibiotic residues.
Eggs containing antibiotic residues, sold at markets are therefore a
way of involuntary exposure of antibiotics to the general popula-
tion. Manure used as fertiliser, if originated from animals treated
with antibiotics, presents another potential source of contami-
nation of groundwater, surface water and crops (Khachatourians,
1998).
Since 2006 antibiotics have been detected in wastewater at con-
centrations ranging from 0.02 ng/L for sulfamethoxazole (Sul) up
to 2292 ng/L for ciprofloxacin (Cip) (Chang et al., 2010; Gracia-
Lor et al., 2011; Li et al., 2009; Mutiyar and Mittal, 2014).
With the presence of antibiotics in the environment, there is a
greater risk of development of antimicrobial resistance in bacte-
ria. The spread of antibiotic resistance has become a worldwide
problem (Negreanu et al., 2012) especially in Africa where resis-
tance rates have been rising to nearly all known pathogens
within the past 50 years (Vlieghe et al., 2009). Aquatic envi-
ronments may serve as reservoirs for antibiotic resistant genes
(Negreanu et al., 2012). Even antibiotic concentrations below Min-
imal Inhibitory Concentration (MIC) can promote development
of resistance (Gullberg et al., 2011). This suggest that occurrence
of trace amounts of antibiotics in the environment may acceler-
ate development of antibiotic resistant bacteria (Negreanu et al.,
2012).
Antibiotics are a very diverse group of chemicals with very
different physico-chemical properties. Consequently, analysing a
broad range of antibiotics in wastewater is a challenging task,
demanding the availability of sophisticated technology and highly
trained personnel. Alternative methods for evaluating the effective-
ness of wastewater treatment systems such as Waste Stabilisation
Pond (WSP) systems are therefore in high demand, in particu-
lar in low and middle-income countries (LMICs) where resources
are scarce. Modelling the performance of such systems may be
a useful alternative. Such models may predict the concentra-
tion of antibiotic in each sedimentation pond and the removal
efficiency.
To the author’s knowledge, no attempt had been made so far
to model antibiotics from WSP. Therefore, the work presented in
this paper is pioneer work on the topic. The aim of the study was to
present a simple dynamic model using STELLA® (isee Systems) soft-
ware package to describe the most important removal processes of
antibiotics through the WSP system Mafisa in Morogoro, Tanza-
nia. The model is based on measured concentrations in loco of four
antibiotics belonging to four different classes. The antibiotics ana-
lysed were trimethoprim (Trim), metronidazole (Met), Sul and Cip.
Trim is a dihydrofolate reductase inhibitor, Met a nitroimidazole,
Sul a sulphonamide and Cip a quinolone. All four antibiotics are on
the World Health Organisation’s Model List of Essential Medicines
(2013).
We attempted to answer two questions with the developed
model: (1) what is the applicability of the model to determine
the removal efficiency of Trim from the WSP system? (2) Can
the model be applied to other antibiotics? These questions are
answered by calibration of the model based on a set of obser-
vations for Trim, followed by a validation of the model against
another set of observations for Trim and by a validation of the
model against observations for all four antibiotics. The developed
model was afterwards used to assess the relative importance of the
four removal processes (settling, outlet, hydrolysis + photolysis and
biodegradation).
2. Materials and methods
2.1. Location
The Mafisa WSP is located in Morogoro, Tanzania. Morogoro is
a town with approximately 300.000 inhabitants located 200 km
inland from Dar es Salaam. Mafisa is located next to the Morogoro
River in the Northern part of the town, in an area with housing and
farming activities (Fig. 1) and receives wastewater from Morogoro
town. The WSP system consists of two receiving ponds and six sed-
imentation ponds. The ponds have different functions as well as
different dimensions. Pond 1 is an anaerobic sedimentation pond,
pond 2 is a facultative pond, while ponds 3–6 are aerobic stabi-
lization ponds. The dimension, flow rate and pH of the individual
ponds are summarized in Table 1. After the sewage water is guided
through Mafisa, it joins the Morogoro River. During dry season,
the water in the river is low; hence, water from Mafisa is used for
irrigation of the fields, mainly rice fields, surrounding Mafisa and
the river. In the rainy season, the water joins the river immediately
after outlet. Evaluation of the water level was based on a visual
inspection.
2.2. Sampling and analysis
2.2.1. Sampling
Six sampling points were implemented and sampling was con-
ducted in triplets. The sampling points and a schematic overview
of Mafisa are shown in Fig. 1. At each of the sampling points, 2.5 L of
water was collected in glass amber bottles. To prevent any degra-
dation during sample preparation and transport, pH was adjusted
on site to around 3 using hydrochloric acid (HCl) (Carlo-Erba) and
measured using universal pH indicator strips. The samples were
transported to the laboratory where they were filtrated twice. The
first filtration was through a grade 5 filter paper with 20 m par-
ticle retention from Munktell. The second filtration was through
a grade 120H filter paper with 1–2 m particle retention, also
from Munktell. A standard addition method was applied when
analysing the samples, by adding an internal standard (IS) to the
samples (Runnqvist et al., 2010). After filtration the samples were
divided to 3 × 800 mL and spiked with 100 L 2.5 ppm internal
standard mix (IS mix). The IS mix contained ciprofloxacin-d8 (d-
Cip), sulfamethoxazole-d4 (d-Sul) and trimethoprim-d3 (d-Trim).
2.2.2. Sample preparation
Approximately 800 mL of water sample, pH adjusted to 3 and
spiked with 100 L IS, was loaded onto Oasis®HLB 6 cm3 200 mg
(30 m) cartridges from Waters (Milford, MA, USA) using a vac-
uum manifold and pump. The vacuum manifold was a VacMaster
from IST (Sweden) and the pump was from ScanVac (Denmark). The
drop-rate was adjusted to 1.5 mL/min. Prior to loading; cartridges
were pre-conditioned with 2 mL methanol (MeOH) followed by
2 mL distilled water. After loading the water samples, the cartridges
were air-dried using vacuum and stored at −18 ◦C before shipping
to Denmark. During transport the cartridges were stored in a cooler
with a coolant. Upon arrival in Denmark they were stored at −18 ◦C
until use.
Prior to analysis, antibiotics were eluted from the cartridges
with 8 mL mobile phase B (0.01% formic acid in MeOH) after wash-
ing with 2 mL 5% MeOH in water. The eluent was evaporated to
dryness under a gentle stream of nitrogen at 33 ◦C. Nitrogen (99.8%)
was supplied by Air Liquid (Ballerup, Denmark) and the evaporator
was a Dionex SE 500 (CA, USA). Elution and evaporation was done in
12 mL amber tubes. Afterwards, the samples were reconstituted in
100 L mobile phase B and 900 L water. Samples were then trans-
ferred to Eppendorf tubes and centrifuged at 0.4472 RCF for 5 min
3. C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146 139
Fig. 1. Location of the Mafisa WSP in Morogoro, Tanzania. P1–P6 represents the sampling points at the outlet of each pond (Google Earth, 2015).
on a Sigma 113 centrifuge (Sigma, Germany). The supernatant was
transferred to 2 mL LCMS/MS vials for analysis.
2.2.3. Analysis of antibiotics
All samples were analyzed using a liquid chromatography
tandem mass spectrometry (LCMS/MS) system equipped with elec-
trospray interface. The system consisted of a 1200 series high
pressure liquid chromatography instrument (Agilent) equipped
with a degasser, a cooled auto sampler (4 ◦C) and a cooled col-
umn oven coupled to a AB Sciex Qtrap 4500 triple-quadrupole mass
spectrometer detector (Applied Biosystems, Foster city, CA, USA).
The chromatographic separation was performed using a Kinetex
2.6 m biphenyl 100 ˚A 50 × 2.1 mm column with a security guard
column both from Phenomenex. The injection volume was set to
10 L. Separation was performed using a binary gradient consisting
of a mobile phase A and a mobile phase B. Mobile phase A con-
tained 0.1% formic acid in Milli-Q water. Mobile phase B contained
0.1% formic acid in MeOH. The solvents of the mobile phases were
chosen based on Locatelli et al. (2011). The flow rate was set to
250 L/min. The results of the analysis are shown in Fig. 2.
2.2.4. Quality control and assurance
Linear calibration curves were established for each antibiotic on
neat standard dilutions (0.5–100 ng/mL). Absolute method recov-
eries ranged from 95 to 97%. Blank or spiked procedural controls
followed each sample-batch. The LCMS/MS limits of detection
Table 1
The dimensions, dynamics, flow rate (Q) and pH of Mafisa.
Pond
1 2 3 4 5 6
Width (m) 48 59 59 59 59 59
Length (m) 72.2 133 133 133 133 133
Depth (m) 1.6 1.5 1.1 1.1 1.2 1.2
Q (m3
/s) 0.034 0.031 0.031 0.038 0.039 0.027
Volume (m3
) 5614 12,037 8349 8883 9071 9322
Q (m3
/24 h) 2937.6 2678.4 2678.4 3283.2 3369.6 2332.8
Q/V 0.5232 0.2225 0.3208 0.3696 0.3715 0.2502
pH 7.4 7.3 7.6 7.8 7.8 7.8
4. 140 C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146
Fig. 2. Concentrations (ng/L) of trimethoprim (Trim), metronidazole (Met), sulfamethoxazole (Sul) and ciprofloxacin (Cip) measured in the sedimentation ponds of Mafisa.
Sampling was conducted over two periods in December 2013 and March 2014, respectively. For the 1st period the concentrations at the inlet were: Trim = 8480 ng/L,
Met = 108 ng/L, Sul = 148 ng/L and Cip = 3264 ng/L. In the 2nd period the concentrations at the inlet were: Trim = 6840 ng/L, Met = 45 ng/L, Sul = 336 ng/L and Cip = 200 ng/L.
Fig. 3. A conceptual diagram of the removal of antibiotic from the WSP. Left: A schematic overview of the water flow through the WSP. Right: A schematic overview of a
single sedimentation pond, showing the removal processes.
5. C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146 141
Fig. 4. A STELLA®
diagram of the model. HP is the process of removal by hydrolysis
and photolysis, Koc is the soil adsorption coefficient, KHP is the first order rate
constant of hydrolysis and photolysis measured at 30 ◦
C, KBio is the biodegradation
constant expressed as a first order rate constant, Csusp is the amount of suspended
matter, Ksett is the settling constant, and Cinlet is the concentration of the compound
(LOD) and limit of quantification (LOQ) were calculated for each
analyte from the following equations (ICH, 2005):
LOD = 3.3 ×
S
LOQ = 10 ×
S
where is the standard deviation based on a triple injection of
the lowest concentration level in the dynamic linear range of the
calibration curve and s is the slope of the linear regression of the
calibration curve. The LOQ were 1 ng/L for all 4 antibiotics.
2.3. Model development
2.3.1. Conceptual diagram
A schematic overview of the WSP system and a conceptual dia-
gram containing the proposed processes involved in the removal
of antibiotics was constructed (Fig. 3). In each sedimentation pond
the antibiotics can be removed by settling, biodegradation, hydrol-
ysis + photolysis (HP) and outlet. The outlets represent the amount
of antibiotic that is transferred from one pond to the next. Based on
the conceptual diagram a model was constructed in STELLA® (isee
Systems) (Fig. 4). It shows the removal processes as forcing func-
tions and the ponds as state variables (Jørgensen and Fath, 2011a).
When running the model, the time step (DT) was set to 0.5 days.
DT refers to the time intervals between calculations in STELLA®
(Kumar et al., 2011). After approximately 20 days, the concentra-
tions had reached a steady state according to the model. Solutions
to the differential equation were obtained using the fourth-order
Runge–Kutta 4 method when running the model.
2.3.2. Model equations
After setting up a flow chart in STELLA® (Fig. 3), the basic equa-
tions for the state variable and the processes were defined. The
concentration in each of the six ponds (Pondx) at time “t” was
defined in STELLA® as a mass balance differential equation:
Pondx (t) = Pondx (t − dt) + (Inletx − Settlingx − Biodegradationx
−HPx − Outletx) × dt
INTPondx = Cinletx
(1)
where Cinletx is the concentration at the inlet. The process of inflow
to the ponds was defined by Cinletx multiplied by the ratio of flow
rate and volume of water per day (Qv)
Inletx = Cinletx × Qvx (2)
The outflow consists of 4 forcing functions: Settling, Biodegrada-
tion, HP and Outflow. Settling is a process where a compound may
bind to soil, which settles in the pond. The process of a compound
adsorbing to suspended matter that then settles, was described by
the following equation
Settlingx = Kocx × Csuspx × 10−3
× Ksettx × (Pondx) (3)
where Kocx is the soil adsorption coefficient, Csuspx
is the amount of
suspended matter and Ksettx is the settling constant. Biodegradation
is an expression for the amount of antibiotic being removed by
bacteria in the water, and can be found by
Biodegradationx = Kbiox
× Pondx (4)
at the inlet of the pond. Ponds 1–6 are state variables. Settling, Biodegradation, HP,
inlet and outlet are processes. Qv is the ratio of flow rate and volume of water per
day.
6. 142 C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146
Table 2
Parameters used to develop the model. Koc is the soil adsorption coefficient, Ksett is the settling rate, KHP is the first order rate constant of hydrolysis and photolysis measured
at 30 ◦
C, KBio is the biodegradation constant expressed as a first order rate constant and Csusp is the amount of suspended matter.
Symbol Description Unit Source
State variables
Pondx Amount of antibiotic in the water ng/L Measured
Processes
Inlet ng/L × 24 h−1
Outlet ng/L × 24 h−1
Settling Amount of AB removed by settling
Biodegradation Amount of AB removed by biodegradation
Hydrolysis and photolysis Amount of AB removed by hydrolysis and photolysis
Parameters
Koc The rate of AB adsorped to soil (PubChem)
Ksett The rate constant of settling m × 24 h−1
Calibrated
Csusp The concentration of suspended matter in the water mg L Calibrated
KBio The rate constant for biodegradation 24 h−1
(PubChem)
KHP The rate constant for hydrolysis and photolysis 24 h−1
Measured
Cinlet Concentration of AB intering the pond system ng/L Measured
Qv The ratio of flowrate and volume per 24 h 24 h−1
Measured
where Kbiox
is the biodegradation constant expressed as a first order
rate constant. The degradation of antibiotic through hydrolysis and
photolysis combined can be described by
HPx = KHPx × Pondx (5)
where KHPx is the first order rate constant of hydrolysis and photo-
lysis. The amount of antibiotic leaving the pond through the outlet
is defined as
Outletx = Pondx × Qvx (6)
2.3.3. Model parameters
The forcing functions are influenced by several parameters.
These parameters are Koc, KHP, KBio, Csusp, Ksett, Cinlet and Qv. The
parameters used to construct the model are summarized in Table 2.
The compound specific physico-chemical properties of the 4 antibi-
otics are listed in Table 3. The hydrolysis and photolysis rate
constants were found experimentally. Eight test solutions (3 repli-
cates), each with a concentration of 1000 ng/mL, were prepared by
diluting stock solution of the antibiotics in 100 mL Milli-Q water.
Stock solutions of Trim, Cip, Met and Sul were purchased from
Fluka. The test solutions were transferred to Erlenmeyer flasks and
placed in a climate-controlled cabinet. The temperature of the cab-
inet was set to 30 ◦C and the samples were exposed to light in a
12:12 h light:dark regime. 1 mL test solution was transferred to a
LCMS vial for analysis at 0, 1.4; 4.5, 23, 67, 73, 94.5, and 119 h. The
sample preparation procedure is described in Sections 2.2.3 and
2.2.4. Hydrolysis and photolysis follow first order kinetics and can
be described according to
dx
dt
= K1 (a − x) (7)
where K1 is the first order rate constant, a is the initial concentration
and x is the concentration at time t (Florence, 2006). The first order
rate constant for hydrolysis and photolysis combined was found by
Eq. (7). From the first order rate constant, the half live (t1/2) was
found from
t0.5 =
0.693
K1
(8)
Met and Cip are degraded relatively fast by hydroly-
sis/photolysis, with a t1/2 of 11 days. Trim is less degradable with a
t1/2 of 61 days. Sul is only slowly degraded with a t1/2 of approxi-
mately 118 days, which is almost 10 times higher than for Met and
Cip.
2.3.4. Sensitivity analysis
The parameters influence on the state variables was evaluated
by a sensitivity analysis. The analysis was carried out to aid in the
model calibration. Changing the value of the parameters by ±10%
and then running the model obtained the relative change in model
output. The sensitivity (S) was calculated as the relative change in
model output divided by the relative change in the value of the
parameter tested:
S =
CSIMs/CSIMc
P/P
(9)
where CSIMs is the change in concentration in the pond, CSIMc
is the concentration in the pond found by calibration, P is the
change in the value of the parameter and P is the original value
of the parameter. A parameter with a high S greatly influences the
outcome of CSIM.
Table 3
Properties of the 4 antibiotics used in the validation. MW is the molecular weight, Sw is the solubility in water at 25 ◦
C, Koc is the soil adsorption coefficient, log Kow is the
octanol–water partition coefficient, pKa is the dissociation coefficients, KHP is the first order rate constant for hydrolysis and photolysis combined, t1/2 is the half-life of
hydrolysis and photolysis measured at 30 ◦
C and KBio is the biodegradation constant expressed as a first order rate constant. KHP and t1/2 was measured and the rest are from
ChemIDplus (PubChem).
Met Sul Cip Trim
MW (g/mol) 171.15 253.28 331.34 290.32
Sw (mg/L) 11,000 610 30,000 400
Koc 23 72 61,000 73
log Kow −0.02 0.89 0.28 0.91
pKa 2.4 1.6; 5.7 6.1; 8.7 7.1
KHP (d−1
) 0.0604 0.0058 0.0585 0.0112
t1/2 (d) 11 118 11 61
KBio (d−1
): 0 0
Anaerob 0.0093 0.0069
Aerob 0.06 0.0092
7. C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146 143
Inlet
Pond
1
Pond
2
Pond
3
Pond
4
Pond
5
Pond
6
1
10
100
Sampling point
%ofinlet
CSIM
COBS
0% 20% 40% 60% 80% 100%
0%
20%
40%
60%
80%
100%
COBS
CSIM
s = 1.013
R2 = 0.876
noitalerroCnoitarbilaC
Fig. 5. Removal efficiency found by calibration. The values given by the model (CSIM) and measured (COBS) are expressed as % of inlet. The correlation of the calibration is
also shown.
2.3.5. Calibration and validation
The model was calibrated by adjusting Ksett and Csusp in Eq. (3)
so that the model simulated the concentrations of Trim measured
in the 1st sampling period. Ksett and Csusp relate to the environment
in the sedimentation ponds and are not compound specific. Thus,
these parameters are the same for every antibiotic. Validation of
the model was performed using Ksett and Csusp found by calibration
and the compound-specific values listed in Table 3.
2.3.6. Statistical analysis
Both calibration and validation were evaluated by the correla-
tion between the concentration simulated by the model (CSIM) and
the concentration observed by LCMS/MS analysis (COBS). CSIM was
plotted against COBS and fitted to a linear regression described as
Y (t) = a + bX (t)
where X is CSIM, Y is COBS, a is the intercept and b is the slope
(s). A perfect correlation between CSIM and COBS will have a s = 1
and intercept at 0. To test if CSIM and COBS were significantly
different, a paired two-tailed t-test was performed with a confi-
dence interval of 95% using GraphPad Prism version 6.00 for Mac
OS X (GraphPad Software, La Jolla California USA, www.graphpad.
com). The standard deviation (SD) of the differences between CSIM
and COBS was used to evaluate the capacity of the model to sim-
ulate the removal of the antibiotics. Since the model aim to test
the removal capacity of the sedimentation ponds, the results are
expressed as % of inlet, rather than concentration. This provided a
better foundation for comparing different antibiotics.
3. Results
3.1. Model calibration
A calibration of the model, using concentrations of Trim
measured in the 1st sampling period, gave a Ksett = 2 and
Csusp = 4000 mg/L. Fig. 5 shows the result of the calibration. The
figure shows CSIM and COBS as well as a plot of the correlation
between CSIM and COBS. There is a satisfactory correlation as the
slope is 1.013 and the regression coefficient (R2) is 0.876. A two
tailed t-test with a confidence interval = 95% showed that there was
no significant difference between CSIM and COBS for the calibration
(P = 0.4195). The differences had a SD = 13%, giving an acceptable
calibration.
3.2. Validation of the model: Removal of Trim
To validate the model’s capacity to determine the removal effi-
ciency of the WSP system, a validation was performed based on
the concentrations of Trim, measured in the second period (Fig. 2),
using the calibrated values Ksett = 2 and Csusp = 4000 mg/L. In Fig. 6
CSIM and COBS are shown, as well as the correlation. There was a
good correlation as the slope was 1.004 and R2 was 0.887. A two
tailed t-test with a confidence interval = 95% showed that there was
no significant difference between the 2 set of values (P = 0.7819).
The differences between CSIM and COBS had a SD = 1% and conse-
quently the validation was satisfactory.
3.3. Validation of the model: Removal of three other antibiotics
To test the model’s capacity to simulate the removal of other
antibiotics than Trim, three other compounds were used in the
validation. Based on the initial calibration, four validations were
performed with data for Met, Sul, Cip and Trim. For Met (n = 14),
Sul (n = 13) and Cip (n = 13) data from both periods were used, and
for Trim (n = 7) data only from the second period were used in the
validation. Since Ksett and Csusp relate to the environment in the
sedimentation ponds and are not compound specific, the Ksett and
Csusp found in the calibration were used in the validation of the
model for all compounds. Table 4 shows the results of the vali-
dation and the major removal processes. The mean SD was 18%.
Since the SD of the validation is well below 50% it is concluded
that the model simulates reality with an acceptable proximity. Fur-
thermore, the correlations for the validation were acceptable since
the slopes were close to 1. For Sul the major removal processes is
settling and through the outlet. Cip seems only to be removed by
settling in the first pond. Met is mainly removed through the outlet
and Trim is removed through settling and through the outlet.
3.4. Sensitivity analysis
Table 5 shows the sensitivity of the parameters used in the
model. In the first pond Qv, Ksett and Csusp are the most sensitive
parameters. In the last pond the most sensitive parameters are Koc,
Qv, Ksett, and Csusp Fig. 7 shows the most substantial removal pro-
cesses of Trim in every pond. Most of the antibiotic is removed by
outlet and settling in all the ponds. Outlet is the major removal pro-
cess in the first pond, and settling is the major removal process in
the last pond.
8. 144 C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146
Inlet
Pond
1
outlet
Pond
2
outlet
Pond
3
outlet
Pond
4
outlet
Pond
5
outlet
Pond
6
outlet
1
10
100
Sampling point
%ofinlet CSIM
COBS
0% 20% 40% 60% 80% 100%
0%
20%
40%
60%
80%
100%
COBS
CSIM
s = 1.004
R2 = 0.887
noitalerroCnoitadilaV
Fig. 6. Removal efficiency found by validation. The values given by the model (CSIM) and measured (COBS) are expressed as % of inlet. The correlation of the validation is
also shown.
Table 4
Results of the validation, showing the standard deviation (SD) of the differences between CSIM and COBS, the slope (s), the correlation coefficient (R2
) of the linear regression
when CSIM are plotted against COBS and the major removal processes for each antibiotic.
Antibiotic SD (%) s R2
Major removal processes
Met 21 0.8547 0.6718 Settling outlet
Sul 38 0.8266 0.3827 Settling outlet
Cip 1 0.9895 0.999 Settling
Trim 12 0.8831 0.8866 Settling outlet
4. Discussion
The removal patterns of Met, Sul and Trim are similar for CSIM.
For Cip there is a rapid decrease in concentration through pond 1.
Met, Sul and Trim are expected to have high mobility in soil and
not adsorb to suspended matter and sediment because of their Koc
values of 23, 72 and 75, respectively (PubChem). Cip has a Koc of
61,000 (PubChem) and is expected to be immobile in soil and have
a high adsorption to suspended matter and sediment, which may
account for the rapid dissipation. Table 3 shows that the solubility
is higher for Met, Sul and Cip than for Trim, so precipitation in the
water is not relevant for these antibiotics. The pH in the pond water
was 7.3–7.8 (Table 1). A pKa = 7.12 indicates that Trim will partially
exist in the protonated form in the water. Cations adsorb stronger
to suspended matter than neutral compounds, so the cation may
adsorb to suspended matter and sediments. Cip is an amphoteric
compound with pKa of 6.09 and 8.74 (PubChem). At 8.74 < pH > 6.09
Table 5
Sensitivity analysis showing the CSIM with a ±10% variation of the parameter and
the sensitivity (S) of the parameter.
Parameter CSIM (+10%) (ng/L) CSIM (−10%) (ng/L) Sensitivity
Pond 1
Koc 4142 4142 0
KHP 4246 4252 0.0071
KBio 4243 4250 0.0082
Qv 4430 4109 0.3780
Ksett 4100 4409 0.3580
Csusp 4100 4409 0.3580
Pond 6
Koc 90.6 156.5 3
KHP 108.7 111 0.1
KBio 108.9 110.8 0.09
Qv 146.7 81 2.99
Ksett 83.8 145.9 2.82
Csusp 83.8 145.9 2.82
the acid will be primarily dissociated and the nitrogen will be pri-
marily protonated. Thus Cip will have an ionic charge in the pond
water and volatilization from moist soil is not expected. Sul will
partially exist in the anionic form in the pond due to pKa values of
1.6 and 5.7 (PubChem). Anions do not adsorb strongly to suspended
matter containing organic carbon and clay compared to their neu-
tral forms. With pKa = 2.38 (PubChem) Met will exist in the anionic
form in the ponds and anions will absorb weakly to suspended mat-
ter containing organic carbon and clay compared to their neutral
forms. The hydrolysis and photolysis combined are different for the
4 compounds. Met and Cip have a half-live 6 times lower than Trim,
meaning that it is removed 6 times faster. Sul has a t1/2 = 118 days,
over twice the half-live of Trim, meaning that it is removed slower.
Regarding biodegradation, Sul and Cip are essentially not degraded,
Pond
1
Pond
2
Pond
3
Pond
4
Pond
5
Pond
6
0.001
0.01
0.1
1
10
100
Sampling point
%ofinlet
Settling
Outlet
HP
Bio
Fig. 7. Removal processes of Trim. HP = hydrolysis and photolysis,
Bio = biodegradation.
9. C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146 145
while Met and Trim are. Met and Trim have approximately the same
KBio under anaerobic conditions in pond 1, but different KBio under
aerobic conditions in ponds 2–6. Trim is removed by biodegrada-
tion 6.5 times slower than Met under aerobic conditions as seen
by the KBio shown in Table 3. Based on the physico-chemical prop-
erties of the antibiotic, hydrolysis and photolysis combined would
be expected to be a major removal process, but the present study
demonstrates that settling is much more important.
The model was developed under the assumption that settling,
biodegradation, hydrolysis and photolysis are the only removal
processes other than outflow. According to Kummerer (2009),
ozonation is another important removal process in wastewater.
They suggest that sulphonamides are degraded by ozonation.
Although being an important removal process, no ozonation occur
in Mafisa. The study by Kummerer (2009) also indicates that sorp-
tion is important for the removal of Sul and Cip, and that these
compounds are resistant to hydrolysis. According to the model,
some removal by hydrolysis and photolysis occur, but it is below
1% for both compounds at steady state. A study from 2007 (Kim
and Aga, 2007) demonstrate that biodegradation as well as sorp-
tion, are the main removal processes and photolysis is negligible,
which is in accordance with the present model. According to U.S.
National Library of Medicine (Bolton et al., 2008) Sul and Met are
not removed by sorption, biodegradability, hydrolysis or photoly-
sis. According to the model (Table 4) Sul is mainly removed through
the outlet and via settling. Met is mainly removed through the out-
let. Settling, biodegradation, hydrolysis and photolysis remove less
than 10%. Cip undergoes photolysis and sorption but not biodegra-
dation according to the U.S. National Library of Medicine (Bolton
et al., 2008), which is in accordance with the data presented here.
In future studies it may be necessary to examine and determine all
the removal processes of antibiotics from WSP to construct a more
accurate model.
The sensitivity analysis showed that Qv, Ksett and Csusp are the
most sensitive parameters with regard to the concentration in the
first pond. The same parameters are the most sensitive regarding
the concentration in the last pond, but also Koc influence the
concentration. The value for Qv was found by measuring the dimen-
sions and flow rate of the WSP system and Koc was found in the
literature based in measurements. Despite being sensitive parame-
ters, little uncertainty is associated with these parameters. Ksett and
Csusp were found by calibration and associated with reasonably high
uncertainty. Because they are sensitive parameters, efforts should
be made for a more accurate determination, especially if the model
is applied to other WSP systems.
As shown in Fig. 7, outlet is the major removal process in the first
pond, and in the last pond settling is the major removal process.
According to Eq. (3), Koc, Csusp and Ksett are involved in settling, and
it is therefore evident that settling is the major removal process in
the last pond when these parameters are the most sensitive. Outlet
is the major removal process in the first pond. The sensitivity of
Koc is low in the first pond which explains why more antibiotic are
removed by outlet than settling.
A study from 2009 found, that the overall removal efficiency
of a wastewater treatment plant was 68–96% for Sul, 90–92% for
Cip and 49–93% for Trim (Li et al., 2009). The present model gives
a total removal efficiency of 98% for Sul, 100% for Cip and 98.5%
for Trim and removal efficiencies given by the present model
are therefore slightly higher than Li et al. (2009). The reason
for this may be, that the model was developed to describe the
removal processes from a WSP whereas Li et al. (2009) measured
concentrations of antibiotics from a wastewater treatment plant.
Consequently, the dynamics of the two systems may be slightly
different. The correlation coefficients of the validation range
from 0.383 for Sul to 0.999 for Cip. To the author’s knowledge,
no attempts have presently been made to model the removal of
antibiotics from wastewater to this point. Hence no comparison of
the model results can be made. From the SD and the correlation
coefficients in Table 4 it can be concluded that the validation of
Met, Cip and Trim are acceptable. Sul have a SD of 38, which is
acceptable in ecotoxicology (Jørgensen and Fath, 2011b).
A study by Senzia et al. (2002) constructed a model for the
removal and transformation of nitrogen in facultative ponds. They
demonstrated that their model was capable of simulating the nitro-
gen removal and transformation in primary facultative ponds. The
major removal mechanism was sedimentation of organic nitrogen
(9.7%). In Table 4 the major removal processes found in this study
are shown. For all 4 antibiotics, settling is a major removal process,
which is the same as what Senzia et al. (2002) found for nitrogen.
WSPs like Mafisa are the preferred system to treat wastewater
in LMICs such as Tanzania (Mbwele et al., 2003). The present study
demonstrates that these systems are quite effective in removing
antibiotics from wastewater. Thus, WSPs are suitable techniques
in LMICs due to the low costs of building, operating and maintain-
ing these systems. The operation is simple and can be managed by
low skilled personnel. In addition, the tropical climate is favourable
with regard to degradation of antibiotics (Senzia et al., 2002). A
WSP system has a large hydraulic retention time and is thus not
very sensitive to hydraulic and organic shock loads (Senzia et al.,
2002). Furthermore, modelling the removal of antibiotics from a
WSP system appears to be an easy and cost-effective way of eval-
uating the transport and removal of antibiotics from wastewater,
which once again, makes it a good alternative in LMICs to more
advanced monitoring programmes.
Sediment from Mafisa is used as fertilizer. It is dug up and left to
dry. The dry sediment is then spread over the rice fields as fertili-
zers. The vegetable producers also use it as a fertilizer. Studies have
shown that antibiotics can be taken up by lettuce and carrots (Kang
et al., 2013; Tanoue et al., 2012). According to the results presented
in this paper, settling is a major removal process, which means
that the sediment will most likely contain antibiotic residues. Con-
sequently, when used as fertilizer, there is a possibility that the
antibiotics will be transferred from sediment to crops, thereby
increasing the possibility of indirect human exposure to antibiotics.
Further studies should be made to evaluate whether humans are
exposed to antibiotics through this route, and how this possible
exposure to sub-MIC antibiotics affects microbial resistance in the
human intestine.
5. Conclusion
A simple dynamic model was developed using STELLA® describ-
ing the most important removal processes of antibiotics in the WSP
Mafisa. Validation of the model using data obtained during the rainy
season gave a SD = 1%, indicating that the model can efficiently be
applied to determine the removal of Trim. The validation using the
results from all four antibiotics (Trim, Met, Sul, Cip) showed that
the model provided reasonable results (mean SD = 18%). Thus, it
can be concluded that the model can be applied on other antibi-
otics than Trim, by using the characteristic properties of the other
antibiotics. The model was used to assess the relative importance
of the four removal processes and these results can be explained by
the characteristic properties of the antibiotics. New data used for
calibration and validation is, however, warranted in order to imple-
ment the present model in other WSPs. However, the same model
structure can most probably be applied, unless other removal pro-
cesses are introduced. A sensitivity analysis showed that Qv, Ksett
and Csusp were the most sensitive parameters. The value for Qv was
found by measuring the dimensions and flow rate of the WSP sys-
tem and Koc was found in the literature based in measurements.
Despite being sensitive parameters, little uncertainty is associated
with these parameters. Ksett and Csusp were found by calibration
10. 146 C.C. Møller et al. / Ecological Modelling 319 (2016) 137–146
and associated with reasonably high uncertainty. Because they are
sensitive parameters, efforts should be made for a more accurate
determination, especially if the model is applied to other WSP sys-
tems. WSP systems are suitable techniques for effective treatment
of wastewater in LMICs. The cost of building, operating and main-
taining the system are low, and low skilled personnel can operate
the system. As a supplement to WSP, modelling the removal of
antibiotic is an easy and cost-efficient way to evaluate the removal
efficiency of antibiotics.
Acknowledgements
The present study is part of the SaWaFo Safe Water for Food pro-
gram funded by DANIDA grant number 11-058DHI, the Ministry of
Foreign Affairs of Denmark. The authors also wish to acknowledge
the help from the technical staff at the Department of Veteri-
nary Medicine and Public Health, Faculty of Veterinary Medicine,
Sokoine University of Agriculture and for providing access to the
waste stabilisation ponds.
References
Bolton, E.E., Wang, Y., Thiessen, P.A., Bryant, S.H., 2008. Chapter 12 - PubChem: Inte-
grated Platform of Small Molecules and Biological Activities. In: Ralph, A.W.,
David, C.S. (Eds.), Annual Reports in Computational Chemistry, Vol. 4. Elsevier,
pp. 217–241.
Chang, X., Meyer, M.T., Liu, X., Zhao, Q., Chen, H., Chen, J.A., Qiu, Z., Yang, L., Cao, J., Shu,
W., 2010. Determination of antibiotics in sewage from hospitals, nursery and
slaughter house, wastewater treatment plant and source water in Chongqing
region of Three Gorge Reservoir in China. Environ. Pollut. 158, 1444–1450.
Florence, A.T.A.D., 2006. Physicochemical Principles of Pharmacy, fourth ed. The
Pharmaceutical Press, London.
Google Earth (Version 7.1.5.1557). (2015): Google Inc. Retrieved from https://www.
google.com/intl/da/earth/.
Gracia-Lor, E., Sancho, J.V., Hernandez, F., 2011. Multi-class determination of around
50 pharmaceuticals, including 26 antibiotics, in environmental and wastewa-
ter samples by ultra-high performance liquid chromatography–tandem mass
spectrometry. J. Chromatogr. A 1218, 2264–2275.
Gullberg, E., Cao, S., Berg, O.G., Ilback, C., Sandegren, L., Hughes, D., Andersson, D.I.,
2011. Selection of resistant bacteria at very low antibiotic concentrations. PLoS
Pathog. 7, e1002158.
ICH, 2005. International Conference on Harmonisation of technical requirements
for registration of pharmaceuticals for human use. In: Validation of Analytical
Procedures: Text and Methodology Q2(R1). ICH, Geneva.
Jørgensen, S.E., Fath, B.D., 2011a. 2—Concepts of modelling. In: Sven Erik, J., Brian, D.F.
(Eds.), Developments in Environmental Modelling, vol. 23. Elsevier, Amsterdam,
pp. 19–93.
Jørgensen, S.E., Fath, B.D., 2011b. 8—Ecotoxicological models. In: Sven Erik, J., Brian,
D.F. (Eds.), Developments in Environmental Modelling, vol. 23. Elsevier, Amster-
dam, pp. 229–290.
Kang, D.H., Gupta, S., Rosen, C., Fritz, V., Singh, A., Chander, Y., Murray, H., Rohwer, C.,
2013. Antibiotic uptake by vegetable crops from manure-applied soils. J. Agric.
Food Chem. 61, 9992–10001.
Khachatourians, G.G., 1998. Agricultural use of antibiotics and the evolution and
transfer of antibiotic-resistant bacteria. CMAJ 159, 1129–1136.
Kim, S., Aga, D.S., 2007. Potential ecological and human health impacts of antibiotics
and antibiotic-resistant bacteria from wastewater treatment plants. J. Toxicol.
Environ. Health B: Crit. Rev. 10, 559–573.
Kumar, J.L., Wang, Z.Y., Zhao, Y.Q., Babatunde, A.O., Zhao, X.H., Jorgensen, S.E., 2011.
STELLA software as a tool for modelling phosphorus removal in a constructed
wetland employing dewatered alum sludge as main substrate. J. Environ. Sci.
Health A: Tox. Hazard. Subst. Environ. Eng. 46, 751–757.
Kummerer, K., 2009. Antibiotics in the aquatic environment—a review—Part I.
Chemosphere 75, 417–434.
Li, B., Zhang, T., Xu, Z., Fang, H.H., 2009. Rapid analysis of 21 antibiotics of
multiple classes in municipal wastewater using ultra performance liquid
chromatography–tandem mass spectrometry. Anal. Chim. Acta 645, 64–72.
Locatelli, M.A., Sodre, F.F., Jardim, W.F., 2011. Determination of antibiotics in Brazil-
ian surface waters using liquid chromatography—electrospray tandem mass
spectrometry. Arch. Environ. Contam. Toxicol. 60, 385–393.
Mbwele, L., Rubindamayugi, M., Kivaisi, A., Dalhammar, G., 2003. Performance of a
small wastewater stabilisation pond system in tropical climate in Dar es Salaam,
Tanzania. Water Sci. Technol.: J. Int. Assoc. Water Pollut. Res. 48, 187–191.
Mshana, S.E., Matee, M., Rweyemamu, M., 2013. Antimicrobial resistance in human
and animal pathogens in Zambia, Democratic Republic of Congo, Mozambique
and Tanzania: an urgent need of a sustainable surveillance system. Ann. Clin.
Microbiol. Antimicrob. 12, 28.
Mutiyar, P.K., Mittal, A.K., 2014. Occurrences and fate of selected human antibiotics
in influents and effluents of sewage treatment plant and effluent-receiving river
Yamuna in Delhi (India). Environ. Monit. Assess. 186, 541–557.
Negreanu, Y., Pasternak, Z., Jurkevitch, E., Cytryn, E., 2012. Impact of treated
wastewater irrigation on antibiotic resistance in agricultural soils. Environ. Sci.
Technol. 46, 4800–4808.
Nonga, H.E., Simon, C., Karimuribo, E.D., Mdegela, R.H., 2010. Assessment of antimi-
crobial usage and residues in commercial chicken eggs from smallholder
poultry keepers in Morogoro municipality, Tanzania. Zoonoses Public Health
57, 339–344.
Runnqvist, H., Bak, S.A., Hansen, M., Styrishave, B., Halling-Sorensen, B., Bjorklund, E.,
2010. Determination of pharmaceuticals in environmental and biological matri-
ces using pressurised liquid extraction—are we developing sound extraction
methods? J. Chromatogr. A 1217, 2447–2470.
Senzia, M.A., Mayo, A.W., Mbwette, T.S.A., Katima, J.H.Y., Jørgensen, S.E., 2002. Mod-
elling nitrogen transformation and removal in primary facultative ponds. Ecol.
Modell. 154, 207–215.
Tanoue, R., Sato, Y., Motoyama, M., Nakagawa, S., Shinohara, R., Nomiyama, K., 2012.
Plant uptake of pharmaceutical chemicals detected in recycled organic manure
and reclaimed wastewater. J. Agric. Food Chem. 60, 10203–10211.
Vlieghe, E., Phoba, M.F., Tamfun, J.J., Jacobs, J., 2009. Antibiotic resistance among bac-
terial pathogens in Central Africa: a review of the published literature between
1955 and 2008. Int. J. Antimicrob. Agents 34, 295–303.
World Health Organization, 2013. WHO Model Lists of Essential Medicines. World
Health Organization, Geneva.
Zhou, J.L., Maskaoui, K., Lufadeju, A., 2012. Optimization of antibiotic anal-
ysis in water by solid-phase extraction and high performance liquid
chromatography–mass spectrometry/mass spectrometry. Anal. Chim Acta 731,
32–39.
