Modeling antimicrobial fate in the circular food system

Abstract

The livestock sector plays a critical role in the circular food production system, but excessive use of antimicrobials (AMs) in livestock farming can lead to AM residue contamination in human food. CirFSafe, a model framework was developed to predict the fate of five different AMs in a primary circular food production system, comprising mixed farms with arable (maize) and animal (bovine) components. Two bovine exposure scenarios to AMs were simulated: annual constant exposure and a one‐off exposure in the first year of circularity. Over a 5‐year timeframe, model predictions suggest that fertilizing soil with animal manure and feeding animals with maize grown in the same soil are unlikely to cause AM residues in milk or meat exceeding European regulatory limits. Nevertheless, the distinct residual patterns of different AMs across the system underscore the need for precautionary monitoring, particularly for the routine use of flumequine (FLU) and doxycycline (DOX), which exhibits a greater tendency to transfer into food products.

1. INTRODUCTION

Ensuring food safety in circular food systems is critical for the successful application of these sustainable systems. Circular systems reduce the input for food production by reusing and recycling waste‐ or side‐streams. However, such measures could also introduce and recirculate harmful contaminants, such as heavy metals, per‐ and polyfluoroalkyl substances (PFAS), dioxins, mycotoxins, and antimicrobials (AMs), in the environment and food systems (James et al., 2022; Scott & Wu, 2024; van Asselt et al., 2023). AM contamination in circular farming systems has received attention (Gworek et al., 2021; Kuppusamy et al., 2018; Tasho & Cho, 2016; F. Zhao, Yang, et al., 2022), as these compounds are excessively used in livestock production and several are known to be persistent (Berendsen et al., 2018, 2021; Walters et al., 2010; Wu et al., 2012). For instance, the half‐lives (DT50) of doxycycline (DOX) and tetracycline (TC) in soil could reach up to 500 days (Walters et al., 2010). Furthermore, livestock production plays a vital role in circularity since animals can consume and convert waste‐ or side‐streams into valuable food products (Bakshi et al., 2016; Dou, 2021; Van Zanten et al., 2018). Animal manure is commonly used for soil fertilization in circular systems, providing a possible contamination pathway for AMs from livestock to crop farming, and ultimately into human food (Adesokan et al., 2013; Dolliver et al., 2007; Getahun et al., 2023; Kumar et al., 2005). Dietary exposure to AMs could lead to gastrointestinal disorders (Ghimpețeanu et al., 2022), allergies (Getahun et al., 2023; Layada et al., 2016), and teratogenic effects in consumers (Jayalakshmi et al., 2017), as well as contribute to AM resistance (Prajwal et al., 2017).

Despite the growing awareness of the food safety and health risks posed by AMs, studies quantifying residual AMs within circular food production remain limited. Most research focused on investigating the fate of AM in either manure‐soil‐crop systems (Aryal & Reinhold, 2011; Dolliver et al., 2007; Mullen et al., 2019; Prosser et al., 2014) or animal‐manure‐soil systems (Albero et al., 2018; Chee‐Sanford et al., 2009; Filippitzi et al., 2019; Karcı & Balcıoğlu, 2009; Zarfl et al., 2009), whereas circular farming often requires integration of both systems. While Chiţescu et al. (2014) predicted the fate of AM across crop and livestock farming, their predictions focused on one circular production cycle, thus did not account for circular farming over multiple years. Long‐term quantification of AM fate in circular food systems is crucial for assessing potential food safety risks due to AM residual exposure.

The objective of this study is to predict potential long‐term AM presence in the circular food production system that involves both crop and livestock farming. The system is designed to maximize the recycling of the side‐ and waste‐streams, ensuring a closed‐loop mass flow of substances over multiple production cycles. Specifically, maize, dairy cows, and veal calves are included, as maize serves as suitable feed ingredients for these farming animals.

2. MATERIALS AND METHOD

2.1. Scenario designs of AM exposure in circular food production

2.1.1. Circular farming and animal diet systems

Two circular food production systems, with dairy cows (named cow system) and veal calves (named veal calf system), respectively, were designed to predict the fate of AMs in circular food systems during a 5‐year period. We considered closed systems: animals were born and raised at the farm located in the Netherlands, maize was grown at the same farm, and the manure produced at that farm was used as fertilizer. In both systems, the manure excreted daily by the animals was collected and stored before being later applied once per year on arable land at the same farm. The land was fertilized annually before maize planting (Dordas et al., 2008; Montforts, 2006). Maize shoots and grains, which were harvested each year, were processed into compound feed (grain only) and maize silage (grain and shoots) for consumption by the animals at the farm (Figure 1). Both silage‐only diet and the combined diet (silage and maize grains in compound feed) were used in the cow system (CS: cow fed with silage‐only diet, CC: cow fed with combined diet), while only the combined diet was examined in the veal calf system (VC: veal calves fed with combined diet). All animals were assumed to be raised indoors, ensuring that the consumption of weeds and soil was excluded.

FIGURE 1.

Illustration of the circular food production systems and relevant farming compartments.

2.1.2. AM exposure scenarios

Two scenarios were defined, one with annual constant AM exposure and the other with one‐off AM exposure. Both exposure scenarios were used in each of the two livestock systems. In the constant exposure scenario, animals were administered a fixed dosage of AMs based on animal weight and type of compounds. In the one‐off exposure scenario, we assumed a one‐time AM inflow into the circular system in the first year via contaminated manure, which was used as soil fertilizer.

2.1.3. Maize‐based feed preparation and intake

All animals were fed daily with a diet consisting of maize grain and/or silage. The inclusion proportions of maize‐related ingredients were determined based on existing studies (Brugiapaglia et al., 2012; Cao et al., 2008; Ferreira et al., 2023) (Tables S1 & S2). In the Netherlands, maize is typically sown in May and harvested once a year in September (Nonhebel, 1995) or October (García et al., 2023; Gou et al., 2016), allowing for an approximate 165‐day period for AM uptake and translocation. At harvest, the aboveground parts of maize, including grains and shoots, were collected for ensiling, resulting in an estimated 10% loss of dry matter in maize silage (Köhler et al., 2013).

2.1.4. Manure application

No pretreatment of bovine manure is necessary before its application to soil (Berendsen et al., 2018). The maximum storage time of manure at the farm is 1 year, spanning from the date of fertilization of the soil for maize cultivation in the previous year to that in the current year. Consequently, the average storage time of manure, which is continuously excreted by the animals at the farm throughout the year, was estimated to be 182 days. Since both manure application and maize harvest occur once per year, one circular cycle is defined as 1 year. We assume no AM leaching or runoff to groundwater during or after the storage and application of manure. Most of the input values were based on Dutch practices and listed in Tables 2 and 3, with additional data and calculated parameters provided in Tables S1 and S2 for the circular systems.

TABLE 2.

Input parameters of the animal‐manure‐soil and the animal‐food modules.

Abbreviation	Parameters (unit, if applicable)	Dairy cows	Veal calves
DDDA	Daily dosed animals within one livestock sector for a particular year (kg treated animals*kg total animals *day−1)	3.2 β	16.2 β
Manimal	Body weight of a farmed animal (kg*animal−1)	625.2 c	113.0 d
tstorage	Maximum storage time of manure in every circularity (day)	365 α	365 α
tanimal	Lifecycle of farmed animals (day)	>365 α	185 α
NLC	Number of lifecycles of farmed animals per circularity (AN*year−1)	1 α	2 α
Froot	Fraction of root biomass to a total plant (%)	3.7 f	3.7 f
Fshoot	Fraction of shoot biomass to a total plant (%)	37.0 f	37.0 f
Fgrain	Fraction of grain biomass to a total plant (%)	59.3 f	59.3 f
Fsilage	Remaining fraction of maize dry matter after ensiling (%)	90.0 g	90.0 g
DMshoot	Dry matter content of maize shoots (%)	35.6 g	35.6 g
DMgrain	Dry matter content of maize grains (%)	67.5 h	67.5 h
ρdrysoil	Density of dry soil (kg*m−3)	1320 i	1320 i
VWatersoil	Volumetric fraction of water in soil (L*L−1)	0.3 i	0.3 i
Depth	Mixing depth of soil and manure (m)	0.05 j	0.05 j
Factorarea	Conversion factor for the area of agricultural field (m2*ha−1)	10,000 j	10,000 j
NST	Nitrogen fertilization standard in Europe (kg*ha−1)	170 j	170 j
Lipidmeat	Lipid content in animal meat (%)	19.0 k	19.0 k
Lipidmilk	Lipid content in animal milk (%)	4.0 k	n/a
Ratiomeat	Ratio of animal carcasses to live weights (%)	51.0 m	50.0 n
tmilking	The lactation period of animals (day)	224 p	n/a

		Dairy cows—combined diet	Dairy cows—maize silage diet	Veal calves—combined diet
Np	Nitrogen production per farmed animal per day (g nitrogen*day−1*animal−1)	89 q	74 r	11.8 s
Intakeintake	Animal daily feed intake in dry matter (kg dm*day−1)	17.7 q	12.3 r	2.5 s , t
Pgrain	Proportion of maize grain in animal diet (% dm)	22.3 q	0 r	13.6 t
Psilage	Proportion of maize silage in animal diet (% dm)	19.4 q	51.0 r	13.6 t
Ymilk	Yield of milk per day (kg*day−1)	14.3 q	13.1 r	n/a

^α Assumed for this circular farming scenario.

^β The Netherlands Veterinary Medicines Institute (SDa) (2023).

^c Tedde et al. (2021).

^d Bruce and Angelos (2016).

^e Ngapo and Gariépy (2006).

^f Hirte et al. (2018); Karyoti et al. (2018).

^g Köhler et al. (2013).

^h John et al. (2018).

ⁱ Trapp et al. (2023).

^j Chiţescu et al. (2014).

^k Lorber et al. (1994).

^m Campion et al. (2009).

ⁿ Vavrišínová et al. (2019).

^o Correa et al. (2006).

^p Alstrup et al. (2015).

^q Cao et al. (2008).

^r Ferreira et al. (2023).

^s Xiccato et al. (2005).

^t Brugiapaglia et al. (2012).

^u Galassi et al. (2017).

n/a: not applicable.

TABLE 3.

Input parameters of the soil‐crop module.

Abbreviation	Parameters (unit, if applicable)	Values
Soil‐related parameters
OC	Content of organic carbon in soil (g*g−1)	0.02 α
ρdrysoil	Density of dry soil (kg*m−3)	1320 α
VWatersoil	Volumetric fraction of water in soil (L*L−1)	0.3 α
pHsoil	Soil pH	5.0 α
ρwater	Density of soil pore water (kg*L−1)	1 α
Waterspw	Volume fraction of water in soil pore water (L*L−1)	1 α
Isoil	ionic strength in soil pore water (L*mol−1)	0.01 α
Root‐related parameters
Qroot	Transpiration stream in plant roots (L*day−1)	0.35 (Hayat et al., 2020)
Froot	Fraction of root biomass to a total plant (%)	3.7 (Hirte et al., 2018; Karyoti et al., 2018)
Waterroot	Water content of plant roots (L*L−1)	0.89 α
Lipidroot	Lipid content of plant roots (g*g−1)	0.025 α
Proteinroot	Protein content of plant roots (g*g−1)	0.03 α
kroot	Growth rate of root (day−1)	0.22 c
Leaf‐related parameters
RatioQleaf	Ratio of transpiration stream from shoots to roots (L*day−1)	0.9 α
g	Conductivity of shoots (m*day−1)	86.4 α
arealeaf	Area of leaf (m2)	0.5 (Subedi et al., 2006)
ρleaf	Density of leaf (kg*m3)	1000 α
Fleaf	Fraction of leaf biomass to a total plant (%)	37.0 (Hirte et al., 2018; Karyoti et al., 2018)
kleaf	Growth rate of leaf (day−1)	0.22 c
Grain‐related parameters
RatioQgrain	Ratio of transpiration stream to from grains to roots (L*day−1)	0.1 α
Areagrain	Area of grains (m2)	1 α
Ρgrain	Density of grains (kg*L)	1 α
Fgrain	Fraction of grain biomass to a total plant (kg)	59.3 (Hirte et al., 2018; Karyoti et al., 2018)
kgrain	Growth rate of grains (day−1)	0.34 c
Generic plant‐related parameters
t	days of growth (day)	165 (García et al., 2023; Gou et al., 2016; Nonhebel, 1995)
Waterxy	Volume fraction of water in xylem (L*L−1)	1 α
Waterphlo	Volume fraction of water in phloem (L*L−1)	1 α
RatioQPhlo	Ratio of flux in phloem to the total transpiration stream	0.1 α
Lipidxy	Lipid content of xylem (g*g−1)	0 α
Proxy	Protein content of xylem (g*g−1)	0 α
Lipidphlo	Lipid content of phloem (g*g−1)	0 α
Prophlo	Protein content of phloem (g*g−1)	0 α
DMsilage	Dry matter content of maize silage (%)	35.6 (Köhler et al., 2013)
Fsilage	Factor estimating remaining dry matter of maize silage during ensiling (g*g−1)	0.9 (Marsalis et al., 2010)
Density	Planting density of maize per hectare (ha−1)	60,000 c
Cell‐related model
pHcyto	pH in cytosol	7.4 α
pHvac	pH in vacuole	5 α
pHphlo	pH in phloem	8 α
pHxy	pH in xylem	5.5 α
Icell	Ionic strength in plant cells	0.3 α
permroot	Membrane permeability of the root cells towards water	2.2E–09 α
Ksetchnov	Setchenov coefficient for calculation of activity coefficient of neutral molecules (L*mol−1)	0.3 α
F	Faraday constant (C*mol−1)	96485 α
R	Universal gas constant (J*mol−1*K−1)	8.314 α
Temp	Absolute temperature (K)	298 α
UCytoO	Potential of cytosol to outside (V)	−0.12 α
UVacCyto	Potential of vacuole to cytosol (V)	0.02 α
UXyCyto	Potential of xylem to cytosol (V)	0.12 α
UPhloCyto	Potential of phloem to cytosol (V)	0.000001 α
Vcyto	Volume of cytosol (m3)	3.75E–08 α
Vxy	Volume of xylem (m3)	7.54E–09 α
Vphlo	Volume of phloem (m3)	4.11E–09 α
Vvac	Volume of vacuole (m3)	3.38E–07 α
a	Unit conversion factor (L*kg−1)	1.22 α
b	Correction component for differences between lipid and octanol	0.85 β
A	Coefficient in Davies approximation for calculation of activity coefficient of ions	0.5 α

^α Data retrieved from Trapp et al. (2023).

^β Data retrieved from Trapp and Matthies (1995).

^c Estimated based on assumption.

2.2. Compound selection and dissipation in circularity

We selected AMs that are commonly used in Dutch livestock sectors, frequently detected in manure, and has a relatively long half‐lives. These included: tetracycline (TC), DOX, flumequine (FLU), lincomycin (LCN), and trimethoprim (TMP). The investigated compounds are permitted to be used in bovine production in Europe under EU regulation (EU) 37/2010 (DOX is not permitted for milk production). Several studies have reported the detection of residues of these compounds in manure (Berendsen et al., 2015; de la Torre et al., 2012; Huygens et al., 2021; Walters et al., 2010). Sachi et al. (2019) reported the frequent presence of AMs from the TC group in milk. Although the persistence of compounds varies across compounds and matrices, all investigated compounds exhibited long half‐lives in soil (Berendsen et al., 2021; Walters et al., 2010) and manure (Berendsen et al., 2018), indicating potential accumulation during circular food production. For example, the half‐lives of TC, DOX, and FLU in soil were reported to exceed 200 days (Berendsen et al., 2021; Shi et al., 2019; Walters et al., 2010).

The dissipation of AMs in different matrices was described using first‐order dissipation rate constant $k_{diss}$ (Table S3). In plants, AM dissipation was estimated based on the growth rate of plant tissues, as studies investigating the impact of compound metabolism and (bio)chemical decomposition in plants are limited, with a few reporting negligible impacts (Berendsen et al., 2018; Chen et al., 2017; Yang et al., 2020). AM degradation in animals was not considered in the present study since some studies suggested AMs could be poorly adsorbed and metabolized by animals after administration, leading to immediate excretion in urine and feces (Alcock et al., 1999; Kemper et al., 2008). Considering various values of $D{T}_{50}$ were reported in different studies, we averaged the $D{T}_{50}$ obtained from those studies whose experimental conditions (e.g., soil organic carbon, soil pH) were similar to that in our model scenario. Other physiochemical properties of the investigated AMs are listed in Table 1.

TABLE 1.

Input compound‐related parameters.

Parameters	Explanation (unit, if applicable)	Tetracycline (TC)	Doxycycline (DOX)	Flumequine (FLU)	Lincomycin (LCN)	Trimethoprim (TMP)
DCD	Defined course dose of AMs per animal weight (mg*kg−1) (EMA/224954/2016)	130	42	66	62	23
Fexcretion	Fraction of antimicrobials excreted in feces of dairy cows and veal calves (%)	0.80 (Feinman & Matheson, 1978)	0.90 (Rakonjac et al., 2022)	0.05 (Rakonjac et al., 2022)	0.35 (Kuchta & Cessna, 2009)	0.54 (Filippitzi et al., 2019)
pKa	Acid dissociation constant at logarithmic scale	pKa1: 3.3, pKa 2 : 7.7 (chosen value for model input) , pKa3: 9.7, pKa4: 12 (Ge et al., 2018)	pKa1: 3.0, pKa 2 : 8.0 (chosen value for model input) , pKa3: 9.2 (Qiang, Adams, & Surampalli, 2004)	6.35 (Z. Zhao, Liang, et al., 2022)	7.89 (Qiang et al., 2004)	7.2 (Mikes & Trapp, 2010)
log Kow	Logarithm of octanol–water partition coefficient	−1.37 (Naghdi et al., 2018)	−0.02 (Chabilan et al., 2022)	1.11 (Khandal et al., 1991)	0.2 (Kim et al., 2009)	0.91 (Kim et al., 2009; Naghdi et al., 2018)
KHSA	Absorption coefficient to human serum albumin at 298K (L*mol−1)	9.49*10^4 (Anand et al., 2011)	2.73*10^5 (Hu, Chen, & Liu, 2014)	2.37*10^6 (Skyrianou et al., 2010)	0 (no binding) (Keswani et al., 2010)	1.91*10^4 (Deng et al., 2013)
log Kaw	Logarithm of air–water partition coefficient (L water*L−1 air)	−18.59 (ChemSpider, 2024c)	−21.72 (ChemSpider, 2024a)	−11.45 (ChemSpider, 2024d)	−20.91 (ChemSpider, 2024b)	−11.45 (ChemSpider, 2024d)
z	Valency or charge number	−1	−1	−1	+1	+1

The bold face values refers to the pKa values chosen for model input.

2.3. Model framework and respective modules

2.3.1. Model framework overview

To model the circular scenarios described in the previous sections, three existing models were coupled (Figure 2). This model framework is named CirFSafe. We used two models developed by Chiţescu et al. (2014), one predicting the transfer of AMs from animal feed to manure and to manure‐amended soil, and the other one predicting the transfer of AMs from animal feed to animal‐based food products (e.g., milk and meat). These two models are referred to as the animal‐manure‐soil module and the animal‐food module. The third model, developed by Trapp et al. (2023), predicts the uptake of AMs by crops from soil and is referred to as the soil‐crop module. The production cycle of maize is 1 year. To predict the fate of AMs over multiple years, the outputs from year 1 were used as inputs for year 2. This process was repeated for 5 years. The coupling of these models resulted in a deterministic simulation model (framework), which was coded in R‐4.3.3.

FIGURE 2.

Schematic concept of the model framework simulating antimicrobial fate in the circular food system. AM, antimicrobial.

2.3.2. Animal‐manure‐soil module and animal‐food module

In the constant exposure scenario, we assumed that AMs were administered to the animals based on their body weight $M_{animal}$ at a constant dosage during each circular cycle. To calculate the constant dosage of AMs in each cycle, we referred to the AM use in 2022, reported by the Netherlands Veterinary Medicines Institute (sDa) (2023). Lifecycles of farmed animals are various, and AMs are usually used only when animals are sick. Therefore, sDa divided the total treated kilograms of animals by the total kilograms of animals present within a particular livestock sector for a year, deriving a daily dosed animal parameter $DDDA$. The AM dose required per treated animal weight in one treatment $DCD$ is defined by European Medicines Agency according to EMA/224954/2016. By multiplying $DDDA$ and $DCD$, we obtained the constant dosage of each AM throughout a year. We used the average body weight of the animals ready for commercial production or slaughter (Table 2) to calculate the AM consumption per animal in a year/cycle (Equation 1). After the first year, the quantity of AMs administered $Q{d}_{feed}$ via maize‐based feed intake was estimated using the concentration of AMs in maize grain $C_{grain}$ and maize shoot $C_{shoot}$ (Equation 2). The quantity of grains and shoots consumed was calculated based on the biomass fraction of the respective plant tissues to the total plant $F_{grain}$ and $F_{shoot}$. During the ensiling process to produce maize silage, we assumed that 90% of maize dry matter remained $F_{silage}$.

In the one‐off exposure scenario, for both the cow and veal calf systems, the respective one time AM inflow was derived using the concentration of AMs detected in manure samples from Dutch farms (Schmitt et al., 2019). In each of the two systems, the resulting residual concentrations of AMs in the following years (after the one‐time contamination) were predicted across different farm compartments and animal‐based food products.

$$Q_{dose}=DCD\ast DDDA\ast M_{animal,}$$ (1)

$$\begin{array}{ccc}\hfill Q{d}_{feed}& =& \frac{Intak{e}_{total}\ast P_{grain}\ast C_{grain}}{D{M}_{grain}}+\frac{Intak{e}_{total}\ast P_{silage}\ast F_{grain}\ast C_{grain}}{D{M}_{grain}\ast F_{silage}}\hfill \\ & & +\frac{Intak{e}_{total}\ast P_{silage}\ast F_{shoot}\ast C_{shoot}}{D{M}_{shoot}\ast F_{silage}}\hfill \end{array}$$ (2)

(Explanation of variables in Equations 1 and 2 can be found in Table 2).

In the animal‐manure‐soil module, the quantity of AMs excreted in manure was estimated using the total quantity of AM administered from $Q_{dose}$ and $Q{d}_{feed}$ within a circular loop multiplied with excretion factor of AMs $F_{excretion}$ (Equation 3). Since $Q{d}_{feed}$ is the quantity consumed daily from feed, this value is multiplied with the lifecycle of animals $t_{animal}$ and number of animal lifecycles $N_{LC}$ per year. The AM concentration in manure was estimated in nitrogen $C_{nitrogen}$. The first‐order kinetic degradation of the AMs in manure during storage was calculated with the degradation rate constant in manure $k_{manure}$ (Table S3) (Equation 4). Afterward, the residual level of AMs in soil after annual fertilization and maize harvest $C_{soilR}$ was used to estimate the AM concentration in soil for the following year (Equation 5). We assumed that manure application only alters the concentration of AMs present in the soil. The output from the animal‐manure‐soil module is the concentration of AMs in the soil. This output variable served as the linking variable to the soil‐crop module as being its model input (described in Section 2.3.3).

$$Q_{excretion}=(Q_{dose}+Q{d}_{feed}\ast t_{animal}\ast N_{LC})\ast F_{excretion},$$ (3)

$$C_{nitrogen}=\frac{Q_{excretion}\ast e^{-\frac{1}{2}{k}_{manure}\ast t_{storage}}}{t_{storage}\ast N_p\ast 365}$$ (4)

$$C_{soil}=\frac{C_{nitrogen}\ast NST}{({\rho }_{drysoil}+VWate{r}_{soil})\ast Depth\ast Facto{r}_{area}}+C_{soilR}$$ (5)

(Explanation of variables in Equations (3)–(5) can be found in Tables 1, 2, and S1).

In the animal‐food module, the biotransfer factors was determined by the logarithm of the octanol–water partition coefficient $\log K_{ow}$, which were derived from empirical studies according to Chiţescu et al. (2014) (Equation 6). The outputs of this module were the AM concentrations in the meat of veal calves and the milk of cows for each cycle. Based on the original paper, AM concentration in meat (Equation 7) and milk (Equation 8) both remain independent of the length of animal lifecycles.

$$BTF=-0.099\ast \log {K_{ow}}^2+1.07\ast \log K_{ow}-3.56,$$ (6)

$$C_{meat}=(Q_{dose}/365+Q{d}_{feed})\ast BTF\ast Lipi{d}_{meat},$$ (7)

$$C_{milk}=(Q_{dose}/365+Q{d}_{feed})\ast BTF\ast Lipi{d}_{milk}.$$ (8)

All compound‐related parameters used in these two modules are shown in Table 1 and environmental‐related parameters in Table 2.

2.3.3. Soil‐crop module

The soil‐crop module was adapted from the generic model developed by Trapp et al. (2023) for plant uptake of ionizable pharmaceuticals and personal care products (the examined AMs are also ionizable). This model accounts for the translocation of AMs within crops via xylem and phloem, driven by hydrological dynamics. The model assumed equilibrium (steady‐state situation) between soil and soil pore water, between soil pore water and xylem, between xylem and cells of plant tissues (i.e., roots, shoots, and grains), between phloem and cells of plant tissues, and between air and both grains and shoots (Figure 2). All input parameters, except for AM concentration in soil, remained constant. AM concentration in soil could vary annually due to potential residual AMs in soil $C_{soilR}$ after maize harvest. Therefore, we considered the possible maximum residual concentration of AMs in soil (per hectare), assuming that no other outflows of AMs from the system (e.g., leaching to underground water). The total quantity of AMs in maize per hectare $m_{plant}$ was estimated based on the respective AM mass in different maize organs and plant density (Equation 9). The total quantity in soil was estimated using the equation given in the original model (Equation 10). Since the total quantity of AMs in the soil before planting $m_{soil}$ is the sum of the mass of AMs taken up by maize $m_{plant}$ and the residual mass in the soil after harvest $m_{soilR}$, the residual AMs in the soil after harvest can be obtained considering also the first‐order degradation during the period between maize harvest and the next maize planting (Equation 11).

The outputs of this module included the predicted concentration of AMs in each of maize grains, shoots, and roots, as well as the residual AM concentration in the soil after harvest. The concentrations in maize grains and shoots were used as input to the animal‐food module and the animal‐manure‐soil module. The residual AM concentration in the soil was used to calculate the concentration in the soil when the maize is planted at the start of the next maize planting cycle.

$$m_{plant-hectare}=\left(C_{root}\ast M_{root}+C_{shoot}\ast M_{shoot}+C_{grain}\ast M_{grain}\right)\ast Density,$$ (9)

$$m_{soil-hectare}=C_{soil}\ast ({\rho }_{drysoil}+VWate{r}_{soil})\ast Depth\ast Facto{r}_{area,}$$ (10)

$$C_{soilR}=\frac{\left(m_{soil}-m_{plant}\right)\ast e^{\left(-k_{soil}\ast \left(365-t\right)\right)}}{({\rho }_{drysoil}+VWate{r}_{soil})\ast Depth\ast Facto{r}_{area}}$$ (11)

(Explanation of variables in Equations (9), (10), and 11 can be found in Tables S1, 2, and 3).

All input parameters for the soil‐crop module are listed in Table 3. These parameters are categorized into AM‐specific, cell‐related, and compartment‐specific, such as soil‐ and plant‐related parameters. The values of cell‐related parameters were obtained from the original modeling study, whereas several plant‐related parameters, such as maize biomass and growth rate, were adjusted to suit this model scenario.

2.3.4. Model verification

The model verification processes include verification of the mass balance (Supporting Information, Appendix 3), sensitivity analysis (Appendix 3), and comparison of model predictions with the European Food Safety Agency (EFSA) monitoring data. Mass balance check is needed since CirFSafe predicts the repeated transfer of AMs over the years across multiple farming compartments and food products, where the equilibrium between compartments was assumed. Sensitivity analysis was performed to examine the influence of uncertainty of input parameters on the output of the model framework. For the comparison of prediction with real‐life data, we compared our predictive results with EFSA veterinary drug monitoring data from 2022 (EFSA, 2022) and 2023 (EFSA, 2023). We selected data from these two years because our predictions were based on national AM usage data (DDDA per sector for total AM usage) from 2022 in the Nederland (SDa, 2023).

3. RESULTS: PREDICTED CONCENTRATION OF AMS IN CIRCULAR FOOD SYSTEMS

3.1. Constant AM exposure scenario

In the constant exposure scenario, the farmed animals are administered annually with a consistent level of AMs and consume potentially contaminated maize‐based feed. In all three animal‐diet cases, the predicted AM concentrations vary depending on AM types and farming compartments. We found that LYN tends to retain in soil (Figure S1), while TC and DOX transfer in maize shoots (Figure S2). In contrast, the carry‐over of these compounds (i.e., DOX, TC, and LYN) from feed to cow milk and calf meat is lower than that of FLU and TMP (Figure 3), whose predicted concentrations in milk reach 0.046 mg/kg (FLU) and 0.011 mg/kg (TMP), and those in meat reach 0.202 mg/kg (FLU) and 0.047 mg/kg (TMP). The residual levels of all AMs in maize grains are predicted to be negligible (Figure 4).

FIGURE 3.

Predicted annual antimicrobial (AM) residual concentrations in food products under the constant exposure scenario. CC, cow fed with combined diet; CS, cow fed with silage‐only diet; VC, veal calves fed with combined diet.

FIGURE 4.

Predicted annual antimicrobial (AM) residual concentrations in maize grains under the constant exposure scenario. CC, cow fed with combined diet; CS, cow fed with silage‐only diet; VC, veal calves fed with combined diet.

Despite slightly increasing AM concentrations in soil and maize shoots after using contaminated maize‐based feed, the concentrations of AMs in milk and meat are predicted to be constant through the 5‐year farming period. These observations indicate that, in this scenario, the contribution of contaminated feed to AM residues in milk and meat was negligible relative to the contribution of annual AM treatment. This result was further consolidated by comparing residual AM levels in milk from cows that fed different diets, as these levels in food products did not vary a lot over time.

3.2. One‐off AM exposure scenario

In the one‐off exposure scenario, manure contaminated by AMs was used for soil fertilization in the first year and the residual AMs in the three animal‐diet systems are predicted for the following 5 years to examine the residual patterns over time. One year after the initial exposure, without a continuous AM input into the system, different compounds showed different accumulation tendencies in the considered compartments. For example, AMs from the TC group mostly transfer in maize shoots, LYN tends to accumulate in soil, while FLU retains both in the soil and the shoots of maize. The predicted concentrations of AMs in cow milk and calf meat are negligible. In the third year, AM concentrations in all compartments and food products were predicted to be negligible (Figures S3–S6). TMP exposure was not considered, as this AM was not detected in Dutch manure samples.

3.3. Model verifications

The mass balance check under the constant AM exposure scenario was passed. Nevertheless, the model verification for the soil‐crop module failed after the second year in the one‐off exposure scenario, where AM concentration in soil was greatly reduced. The output mass of AMs exceeds the input, indicating AM concentration in grains can be overestimated. This observation shows that the soil‐crop module may only be used to predict the plant uptake from certain AM exposure level.

Sensitivity analysis revealed that, while all considered parameters influence the AM intake from feed, only the ratio of daily dosed animals to total animals $DDDA$ and animal weight $M_{animal}$ account for variation in AM residues in food products (Figure 5). This observation results from, as predicted, the negligible contribution of AMs from animal feed to the total AM administration in every cycle. When examining the quantity of AMs consumed by animals from feed $Q_{feed}$ in the second year, we observe that the output is most sensitive to the storage time of manure $t_{storage}$, followed by the parameter concerning plant density of maize $Density$. Notably, soil pH $p{H}_{soil}$ starts to have more influence on the model output when the pH is increasing, suggesting that the prediction of CirFSafe may vary when applying in acid and alkaline soil.

FIGURE 5.

Sensitivity analysis results with antimicrobial (AM) concentration in milk and AM consumption from feed as respective model outputs.

We compare our predictions with EFSA monitoring data (94,025 bovine samples and 20,927 milk samples for 2022, 110,228 bovine samples and 17,298 milk samples for 2023) (EFSA, 2022, 2023). In both years, each AM investigated in this study exceeded the maximum residual levels (MRLs) in bovine meats once, with noncompliance rates below 0.010%. In milk samples, only one noncompliance case of TC was reported (noncompliance rate of 0.017%). Although absolute measured values are unavailable, this comparison suggests that our predictions align with the occurrence of AMs in bovine and milk products.

4. DISCUSSION AND LIMITATIONS

4.1. Risk evaluation and mitigation of different AMs in circular farming systems

Both exposure scenarios of this study demonstrated that the transfer of AM residues from manure fertilization to food products was negligible, especially compared to that from the routine treatment of AMs in livestock farming. This implies that circular farming does not pose a higher food safety risk related to AM residues than conventional food production systems. Although the predicted concentration of FLU in veal calf meat is comparable with the MRLs of 0.2 mg/kg, as regulated under EU regulation (EU) 37/2010, this observation may result from the potential overprediction. CirFSafe currently does not consider AM run‐off and leaching to the groundwater in soil. Moreover, the AM usage data (DDDA) reflects total AM usage rather than compound‐specific values. Nevertheless, the AM concentrations in Dutch manure samples (median values), which were used as the exposure level in the one‐off scenario, revealed elevated levels of DOX and FLU in manure from veal calf farming in the Netherlands. Therefore, while circular farming may not increase food safety risks related to AM residues in food products, the application of certain AMs still requires continuous monitoring in Dutch livestock farming to avoid residual risks in human food.

Model predictions in the one‐off exposure scenario indicated potential environmental risks due to varying AM residues in the farming compartments in the first two years of the calculated 5‐year circular period. The residual levels of AMs turn negligible after 3 years, AM concentrations in soil, maize shoots, and grains may already reach elevated levels in the first year after exposure. LYN and FLU each are predicted to accumulate a high level in soil, while DOX and TC tend to transfer in maize shoots. The latter finding aligns with the uptake pattern observed by Kang et al. (2013) in a controlled experiment. Large‐scale field samples from China showed significant transfer of TC into both maize grains and shoots (Pan et al., 2014). The variances in these empirical results may be due to the effects of additional environmental factors (Huang et al., 2024), which need to be identified and quantified. For instance, our model excluded the atmospheric deposition of AMs to maize while this could serve as AM inflow into the farming system (Ossola & Farmer, 2024; Zhang et al., 2017). Furthermore, the water dynamics in farmland, such as irrigation, rainfall, and leaking in groundwater, may provide another drive for AM movement within the soil‐crop systems (Abdallat et al., 2022; Burke et al., 2016; Pan et al., 2014). Among all compounds, our predictions showed DOX and FLU may pose a higher food safety and environmental risk compared with other compounds considered. Though CirFSafe did not predict antibiotic resistance genes (ARG) in circular food systems, this could be an interesting direction for model iteration in the future.

To mitigate the risks brought by AM residues in circular food systems, researchers have proposed several measures, among which bio‐based AM remediation shows a great match to the sustainability principle of circularity. Biochar, which refers to carbon‐rich materials derived during pyrolysis (Katiyar et al., 2022), has been found to adsorb AMs and to accelerate its bioremediation efficiently (Jia et al., 2024; Patel et al., 2022; Sarmah et al., 2006; Zeng et al., 2018). Several mechanisms contribute to the adsorption effect of biochar, including π‐π electron doner–acceptor (EDA), hydrophobic, electrostatic interaction, hydrogen bonding, and pore diffusion (Jia et al., 2024). This adsorption increases contact between microorganisms and antibiotics, and enhance electron transport, enzymatic activity, and microbial colonization, thereby improving AM degradation (Chen et al., 2022; Zhang & Wang, 2021). The applications of biochar in soil were also reported to remove AMs from soil with up to 69% of total TC (Yue et al., 2019) and 41.9% of oxytetracycline (coexistence with heavy metals) (Zhang et al., 2022). Biochar remediation is not only broad‐spectrum for multiple AMs but also can serve for soil amelioration during fallow seasons, making this bio‐based solution a strong applicational potential in circular food systems. Another mitigation measure concerns anaerobic and aerobic digestion of manure. The degradation of AMs from different classes responded distinctively to anaerobic digestion (Gurmessa et al., 2020; Yin et al., 2021). For example, Yin et al. (2021) found that maximum 75 mg*kg solid⁻¹ sarafloxacin, difluoxacin, and sulfadiazine could be completely removed during anaerobic digestion, while less than 25 mg*kg solid⁻¹ FLU, enrofloxacin, norfloxacin, and sulfachloropyridazine were removed by the same process. Cha and Carlson (2019) found that aerobic digestion demonstrated better removal efficiency than anaerobic treatment for oxytetracycline, tylosin, sulfamethoxazole, and monensin. The better performance of aerobics was attributed to oxidation, which may accelerate the degradation of AMs (Massé et al., 2014; Sodhi et al., 2021). In addition, temperature during the manure pretreatment will significantly influence the effectiveness of AM removal (Cha & Carlson, 2019; Yu et al., 2019). Lastly, using contaminant‐accumulative plants during the crop rotating could help to concentrate AMs from contaminated soil, allowing more efficient treatment or removal of these contaminants (Chi et al., 2018; Zhang et al., 2013). These findings suggest that potential control points should be located in the treatment of manure and soil to mitigate AM residual risks.

4.2. Model limitations

4.2.1. Biotransfer and excretion factors

The animal‐related modules used two key parameters: biotransfer and excretion factors, to describe the carry‐over of AMs from feed to animal products and excreta. Biotransfer factors, based on empirical observations, simplify the complex mechanism underlying AM carry‐over, though mechanistic models could be developed to improve predictions (Mackay & Fraser, 2000). These models require a better understanding of factors influencing compound carry‐over. Researchers have identified several factors, including compound‐specific properties such as binding affinity to serum proteins (Beyene, 2016), molecular weight (Amutova et al., 2021), and the degree of halogenation, as well as animal‐related factors such as yield of animal products (Dänicke & Brezina, 2013; Huang et al., 2024), age and body condition of animals (Beyene, 2016; Canton et al., 2021; Rychen et al., 2008). These insights suggest that dynamic mechanistic modeling, which has been applied to other contaminants (Hoogenboom et al., 2006, 2011; Van Eijkeren et al., 2006; Wang et al., 2021), may enhance predictive performance for AMs. However, most dynamic models require parameters describing time‐based processes. These parameters are not easily measurable and may in turn hinder model applicability. In addition, compound ionization may play a role, as seen in the soil‐crop module, due to its influence on several factors such as compound binding ability (Beyene, 2016; Naik et al., 2015).

4.2.2. Plant growth and AM phytodegradation

In the soil–crop module, linear plant growth and a constant ratio of transpiration rate between maize organs was assumed, as in the original model. However, linear growth model with a constant growth rate do not reflect the empirical growth patterns, where biomass gain of plants is dependent on current biomass and leaf area (Kawano et al., 2020; Paine et al., 2012; Wardhani & Kusumastuti, 2013). Paine et al. (2012) reviewed alternative models for plant growth simulation, including power law, monomolecular, three‐ and four‐parameter logistic, and Gomertz models. Brunetti et al. (2019) coupled a three‐parameter logistic model to predict plant uptake of pharmaceuticals. In CirFSafe, the application of different plant growth models may affect the parameters such as plant biomass, leaf area, and the dilution of compounds during growth. The uncertainty of these parameters could lead to variation in predicted AM concentrations and residual risks in respective farming compartments and systems, though this may not be evident in the current deterministic model framework.

CirFSafe considered only growth dilution for compound dissipation, as studies on AM degradation and metabolism in plants are scarce. Most studies focused on AM removal in plant‐environment systems through bioaccumulation (Marques et al., 2024; Singh et al., 2019; Stando et al., 2022). Chen et al. (2017) examined the metabolism of sulfamethoxazole and TC in various leafy plants. Although a few studies suggested that (phyto)chemical‐ and microbial‐degradation or metabolism in plants were limited (Chen et al., 2017; Fantke & Juraske, 2013), further empirical research is required to confirm these findings.

4.2.3. Implication of manure application

In addition to contributing to AM accumulation in soil, manure application could alter soil characteristics such as carbon and other nutrient content (Hao et al., 2003; Sleutel et al., 2006; Zhong et al., 2010), pH (Eghball et al., 2004; Zhong et al., 2010), soil density (Dhaliwal et al., 2019), and microbial communities (Zhong et al., 2010). These changes may directly or indirectly affect the AM fate in the circular food systems, particularly over the long term, though CirFSafe has not yet accounted for these effects. Incorporating models that simulate these effects could enhance AM prediction (Grant et al., 2001; Jiang et al., 2018; Ren et al., 2019). Moreover, manure storage conditions such as temperature and moisture, may affect AM degradation (Hammesfahr et al., 2011; Köninger et al., 2021), organic carbon levels (Moral et al., 2005), and microbial dynamics (Oliver et al., 2010), potentially leading to variations in AM contamination in soil.

5. CONCLUSION

This study provides, to our knowledge, the first comprehensive quantification of AM fate over multiple years in a circular food production system involving both crop and livestock farming. The model framework, CirFSafe, interlinked three models that, respectively, analyze AM transfer from animals to manure, from manure to soil and maize, and from maize‐based feed to milk and meat. Under both the continuous and one‐off AM exposure scenarios, the predicted AM concentrations in farming compartments and food products indicate that recycling animal manure and feeding animals with contaminated maize‐based feed may not result in AM residues exceeding European regulatory limits for milk or meat. Instead, precautionary measures and monitoring should focus on routine AM treatments, especially for the use of FLU and TMP, as both substances tend to transfer in the food products. CirFSafe could be further improved by incorporating additional factors related to the biotransfer and excretion of substances in animals, integrating plant growth models that better reflect actual growth patterns, and considering the changes in soil characteristics due to manure fertilization.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

AppendixS1

AppendixS2

AppendixS3

Huang, W. , Focker, M. , & van der Fels‐Klerx, H. J. (2025). Modeling antimicrobial fate in the circular food system. Risk Analysis, 45, 2790–2807. 10.1111/risa.70044

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.