Quantitative Prediction of Sediment–Water Partition Coefficients for Tetracycline Antibiotics in a Typical Karst Wetland

Abstract

The soil–water partition coefficient (Kd) of antibiotics is a critical indicator for assessing their migration potential in the environment. Currently, research on antibiotic Kd values in specific geological settings such as karst wetlands remains relatively limited. This study uniquely integrates partial least squares (PLS) regression with redundancy analysis (RDA), a hybrid approach that effectively handles complex environmental datasets prone to multicollinearity. The results identified Fe³⁺, NO₃⁻, and PO₄³⁻ in water, as well as clay content, organic matter, bulk density, and pH in sediments, as key factors influencing Kd through redundancy analysis. Using PLS, predictive models were developed for the logKd of four antibiotics: tetracycline (TC), doxycycline (DOX), chlortetracycline (CTC), and demeclocycline (DMC). The models demonstrated strong predictability with Q₂cum values of 0.96, 0.93, 0.99, and 0.83, respectively, indicating excellent model convergence. These findings provide important insights into how soil and water physicochemical properties influence the distribution of antibiotics, support the prediction of antibiotic transport and fate, and contribute to the exposure and risk assessment of these emerging contaminants in aquatic ecosystems.

1. Introduction

Antibiotics, widely utilized as drugs, selectively inhibit biological functions. However, significant quantities of these antibiotics remain unabsorbed [1]. Antibiotics are excreted into the environment as proximate forms or metabolites in feces and urine, ultimately leading to worldwide exposure to antibiotics in soil [2], surface water [3], groundwater [4,5], and sediment [6,7]. Despite the low concentration of antibiotics in the environment, their persistent presence can impact the population dynamics and nutrient cycling among bacteria, aquatic organisms, soil biota, and plants within environmental ecosystems. The balance and generation of large numbers of drug-resistant bacteria are disrupted, posing a threat to human health [8,9].

Wetlands, as one of the largest global ecosystems contributing to humans, are becoming infested with antibiotics [10,11]. Karst wetland aquifers are more vulnerable to groundwater contamination due to the unique hydrogeological characteristics of high hydraulic conductivity and short residence time [12]. The concentrations and distribution characteristics of antibiotics in karst areas are still less studied [13,14], including organic pollutants [15,16]. Currently, several preliminary studies on antibiotics in karst groundwater have been conducted, which elucidate sulfamethoxazole, ciprofloxacin, ofloxacin, and trimethoprim as the primary detected compounds [17]. In Guizhou, China, a variety of antibiotics were detected in karst groundwater systems, with detection rates varying from 2.9% to 94.1% across 25 different antibiotics [13]. These findings are consistent with broader research indicating that groundwater contamination by antibiotics is a significant issue.

The distribution of antibiotics is affected by a multitude of factors [18], including discharge loads [19], as well as the characteristics of the antibiotic and the physicochemical properties of the surrounding environment. Adsorption and precipitation constitute significant processes for antibiotics in both water and soil [20]. It has been demonstrated that sulfamethoxazole (SMX) has a more extensive adsorption capacity in smaller particle size sediments due to its larger surface area than larger sediments, because the highest adsorption is associated with the smallest particle size sediment [21]. The exchange of antibiotics between sediment and water is influenced by sediment properties, which can be quantified by their distribution coefficients [22]. A model provides accurate and quantitative estimates of antibiotic distribution coefficients between water and sediment based on the specific properties of the sediment obtained from field investigations, which could provide the characteristics of the antibiotic needed for the antibiotic pollution control [23].

The predicted distribution coefficients of antibiotics between the water column and sediment have been studied [24]. However, most predicted distribution coefficients have been developed considering only soil or sediment physicochemical factors and not water chemistry characteristics [22,25]. Given that sediment and water constitute the primary reservoir of antibiotics, they form an interconnected system within the environment. However, considering that only the factors affecting one compartment overlook the interconnectedness, the mere consideration of physicochemical properties of sediment or water is insufficient for analyzing the distribution pattern of antibiotics within the ecosystem. For example, using 14 soil physicochemical properties, three quantitative models were developed for the water–soil distribution coefficients of sulfamethoxazole (SMZ), oxytetracycline (OTC), and norfloxacin (NFX). These models indicated that metal concentration, pH, and organic matter (OM) were the important factors in the lab [22]. However, field investigations do not fully support the laboratory findings. Because the two models developed to predict the distribution coefficients of NFX and erythromycin (ETM) by the same research group in the intertidal zone of the Yellow River Delta are not identical [25], this discrepancy is due to the spatial and temporal variability of antibiotic concentrations in natural environments, where real equilibrium is rarely established in dynamic water–sediment in the field [25]. In addition, the influence of sediment and water chemistry on the behavior of antibiotics in natural environments is also the main reason for the discrepancy. Therefore, we consider the water–sediment in which antibiotics are located as a system in our study. The influence of water–sediment physicochemical characteristics on antibiotic partitioning behavior is integrated in order to quantitatively predict the sorption behavior of antibiotics in natural water–sediment systems for antibiotic pollution control.

2. Materials and Methods

2.1. Study Area and Sampling

The study area is located in Lingui District, Guilin, China, 110°09′50″~110°14′30″E, 25°05′20″~25°06′45″ N (Figure 1). At an altitude of 150–170 m above sea level, the study area is located in the central subtropical monsoon climate zone, with an average annual temperature of 19.5 °C and an average annual rainfall of 1835.8 mm. The sampling sites are located in the main drainage area of the Huixian Wetland.

A total of four sets of blue hole water and sediment samples, as well as four sets of aquaculture soil and water samples, were collected in October 2020. The Huixian Wetland represents a typical karst wetland ecosystem, where submerged dolines (solution ponds) and subterranean river outlets constitute distinctive groundwater emergence features characteristic of karst hydrology. Given that agricultural cultivation and aquaculture dominate anthropogenic activities within this wetland, our selected sampling sites were strategically chosen to reflect both the unique karst hydrogeological attributes and the influences of human disturbance. Among the samples collected from the blue hole, there were three sets of water and sediment samples, along with one sample from an underground river outlet. For the aquaculture area, the water samples were taken from swine farms, duck houses, chicken houses, and fishponds. The sediment samples were obtained from the respective leachates. The samples were obtained in typical blue holes and agricultural types in the study area. The water sample was collected 5 cm below the water surface, and the soil sample was taken at a depth of 0–20 cm. The information about the sampling sites is illustrated in Table S1 (Supporting Information Table S1).

2.2. Sample Extraction and Instrumental Analysis

Tetracycline antibiotics were examined in water and in matching sediment samples, including tetracycline (TC), demeclocycline (DMC), chlortetracycline (CTC), and doxycycline (DOX) (Table S2). The main reagents and detailed descriptions of pretreatment and extraction processes involved in the study are presented in the Supporting Information (SI). After the pretreatment of the samples, high-performance liquid chromatography tandem mass spectrometry (HPLC-MS/MS, AB SCIEX, Foster City, CA, USA) was used in a multiple reaction monitoring (MRM) model to analyze target antibiotics with a positive ion mode of electric spray ion source. A detailed description of quality assurance and quality control is presented in Table S3.

2.3. Analytical Methods

The study of the antibiotic partitioning between the water and sediment is usually calculated using the partition coefficient K_d, which is calculated as follows:

$${\mathrm{K}}_{\mathrm{d}}={\displaystyle \frac{{\mathrm{C}}_{\mathrm{s}}}{{\mathrm{C}}_{\mathrm{w}}}}$$

where C_s denotes the mass fraction of antibiotics in the sediment (ng·g⁻¹) and C_w denotes the mass concentration of antibiotics in the water (ng·L⁻¹) [27,28]. Antibiotics in natural river systems are not stable in a dynamic equilibrium, but the partition coefficient K_d value is an important way to represent the partitioning behavior of antibiotics when subjected to various environmental factors in the water and sediment [29]. It should be noted that the dynamic and non-equilibrium nature of natural environments implies that field-measured K_d values should be regarded as an apparent distribution coefficient, rather than a strict equilibrium constant. Despite this limitation, K_d remains an extremely useful and widely adopted key parameter as an empirical indicator for comparing the solid–liquid partitioning behavior of contaminants across different systems or conditions.

Partial least squares (PLS) regression, which is a robust statistical method for handling multicollinearity and analyzing multiple dependent variables simultaneously, represents a cutting-edge multivariate statistical data analysis technique that is adept at constructing regression models even when independent variables exhibit high degrees of correlation. The method takes into account the role of the target variable matrix while reducing the dimensionality of the independent variables. It combines compression with regression to provide strong explanatory power for both the independent and dependent variables [30]. In sediment and water, there are numerous factors that influence the value of K_d. Therefore, identifying and filtering out the main controlling factors is pivotal for modeling. In PLS analysis, variable importance in the projection (VIP) is a parameter that shows the importance of a variable in the model. The higher the VIP value is, the more significant the descriptor is. Redundancy analysis (RDA) has received increasing attention in recent years due to its ability to quickly obtain relationships between explanatory and response variables [31]. The main sediment and water physicochemical indicators were firstly ranked for multi-factor contribution using Canono 5.0 software (Microcomputer Power, Ithaca, NY, USA), and those with a contribution greater than 10% were selected as the main influencing factors, which is an empirical value in ecological research [32,33]. Partial least squares (PLS) analysis was carried out using SPSS 26 (IBM, Armonk, NY, USA) to develop a predictive model of the sediment and water and their main contributing factors with each antibiotic K_d.

The main factors involved were the cumulative cross-validated regression coefficient Q² (Q²_cum), standard deviation (SD), correlation coefficient between observed values and fitted values (R), and the significance level (p). These factors were used to assess the robustness and predictive ability of the model. Generally, predictive models with lower SD values are preferred, and Q²_cum > 0.5 is commonly accepted as a satisfactory result in natural environments for modeling [34]. Except for SD, all of these statistics can be directly obtained by PLS analysis. Standard deviation (SD) can be calculated by the following equation:

$$\mathrm{S}\mathrm{D}=\sqrt{{\displaystyle \frac{1}{n-A-1}}\sum _{i=1}^n{[\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{K}\mathrm{d}\left(\mathrm{o}\mathrm{b}\mathrm{s}.\right)\mathrm{i}-\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{K}\mathrm{d}\left(\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{d}.\right)\mathrm{i}]}^2}$$

where n is the number of observations used for model development; log K_d (observed) and log K_d (predicted) represent values observed in the investigation and model-predicted values of distribution coefficients (L·kg⁻¹); and A is the number of PLS principal components, respectively.

3. Results and Discussion

3.1. Sediment and Water Physicochemical Characteristics

The results of geochemical analysis in the water and sediment samples are shown in

Table 1. Characteristic values of the ion concentrations in the water.

Indicator	Min	Max	Aver	Indicator	Min	Max	Aver
Na+	0.84	30.98	5.99	SO42−	4.85	72.55	24.41
Ca2+	35.30	124.90	80.57	NO3−	0.05	58.24	14.23
NH4+	0.02	74.95	10.99	NO2−	0.002	1.10	0.15
Fe3+	0.005	2.26	0.50	PO43−	0.02	3.15	0.61
Cl−	5.34	31.38	11.50	pH	6.91	7.51	7.20

Note: Unit: mg/L; Min: minimum; Max: maximum; Aver: average.

,

Table 2. Characteristic values of the indicators in the sediment.

Indicator	Min	Max	Aver
BD	0.47	1.40	1.07
OM	27.83	85.02	50.30
pH	5.59	7.67	6.61
Clay	5%	13%	8%

Notes: BD: bulk density, unit: g/cm³; OM: organic matter, unit: g/kg; Min: minimum; Max: maximum; Aver: average.

, Table S4 and S5. The highest concentrations of NH₄⁺, Fe³⁺, Na⁺, and Cl⁻ were all observed in G08, which was collected from the swine farm leachate. The highest level of NO₂⁻ was observed in leachate G02, which was collected from the blue hole water. The highest concentration of PO₄³⁻ was detected in G06, which was collected from the duck house wastewater.

In each sediment sample (Table 2 and Table S5), the highest bulk density (BD) is the blue hole sediment (S01), while the highest concentration of OM is the underground river outlet sediment (S08). The highest pH point is the blue hole sediment (S01), and the highest clay content is the swine farm sediment (S05).

3.2. Distribution Characteristics of Tetracycline Antibiotics in Water and Sediment

All four tetracycline antibiotics were detected in the water in this study, with TC detected at 62.5%, varying from 1.90 to 9.00 ng·L⁻¹ with a mean value of 4.34 ng·L⁻¹ (Figure 2). Tetracycline (TC) was detected in three blue hole water samples (G01, G02, and G03), as well as in duck house wastewater (G06) and chicken house wastewater (G07), with the highest concentrations in G02. The detection rate of DMC was 25%, varying from 6.60 to 12.54 ng·L⁻¹, with a mean value of 7.45 ng·L⁻¹. It was mainly detected in blue hole water G01 and G02, with the highest concentration in blue hole water G02. The detection rate of CTC was 25%, varying from 1.20 to 75.60 ng·L⁻¹, with a mean value of 12.48 ng·L⁻¹. It was mainly detected in blue hole water G02, with a maximum concentration in blue hole water. Doxycycline (DOX) was detected in 75% of the samples, with concentrations varying from 2.72 to 26.80 ng·L⁻¹ and a mean value of 9.19 ng·L⁻¹. All the water samples were detected to contain DOX, with the exception of G03 and G04, the outlet of the underground river, with the highest concentration in G06.

Tetracycline (TC) was detected in 88% of the sediments in the study area, with concentrations varying from 1.9 to 8.76 ng·g⁻¹, with a mean value of 7.12 ng·g⁻¹ (Figure 2). Tetracycline (TC) was not only detected in the sediment of underground water outlet S04, but its highest concentrations were also found in the blue hole sediment S02. The detection rate of DMC was 88%, with a variation of 6.6–15.8 ng·g⁻¹ and a mean value of 9.85 ng·g⁻¹. Demeclocycline (DMC) was not only detected in the sediment of the chicken farm, but was also detected in the rest of the sediment, with the highest value in the sediment of S03. The detection rate of CTC was 100%, with concentrations varying from 1.2 to 25.3 ng·g⁻¹, with a mean value of 17.79 ng·g⁻¹. The highest concentration was found in the fishpond water sediment S05.

Overall, the detection rate of TCs was higher in sediment than in the water, e.g., CTC was detected at 100% in the sediment, while its detection rate in the corresponding water was 75%. In addition, there was no obvious correspondence between the highest values detected for each indicator in the water and sediment, e.g., the highest concentration of CTC in the water environment was swine farm wastewater G08. But its detection in the sediment was not the highest. On the contrary, the highest concentration of CTC in the sediment S05 was not detected for CTC in the matching water sample.

3.3. Environmental Impact Factors

3.3.1. Influencing Factors in the Water

A redundancy analysis (RDA) of TCs in sediments with the geochemical characteristics of matching water samples revealed (

Table 3. Explanatory contribution of redundancy analysis for each water quality indicator.

Indicator	Explanation %	Contribution %	Pseudo-F	p
Fe3+	35.7	36.3	5.5	0.024
NO3−	21.2	21.6	4.4	0.016
PO43−	13.3	13.5	3.6	0.044
NO2−	7.4	7.6	2.3	0.09
NH4+	5.1	5.2	1.8	0.17
SO42−	3.7	3.8	1.4	0.32
pH	4.4	4.5	1.9	0.24
Na+	5	5.1	3.5	0.12
Cl−	1.6	1.6	1.2	0.39
Ca2+	0.8	0.8	0.4	0.63

) that the water was explained by 66.1% for RDA1, 21.2% for RDA2, and 87.3% overall (Figure 3a). The contribution of Fe³⁺ was 36.3%, that of NO₃⁻ was 21.6%, and that of PO₄³⁻ was 13.5%, while the contribution of the remaining indicators was less than 10%. In the water samples, DMC and CTC showed positive correlations with NO₂⁻, NO₃⁻, SO₄²⁻, and Ca²⁺, and negative correlations with NH₄⁺, Fe³⁺, Ca²⁺, and pH. Doxycycline (DOX) and TC showed positive correlations with pH, PO₄³⁻, Na⁺, Cl⁻, NH₄⁺, Fe³⁺, and Ca²⁺, and negative correlations with NO₂⁻ and NO₃⁻.

Tetracycline antibiotics contain the largest variety and number of functional groups, and complexation with metal ions is dominated by –OH and –C=O [35]. The complexation of Fe³⁺ with antibiotics significantly enhances their oxidative degradation, as evidenced by advanced oxidation processes and the catalytic role of Fe₃O₄ nanoparticles. The formation of complexes allows the intramolecular transfer of electrons, resulting in more efficient oxidative degradation of antibiotics under mild conditions [36]. In addition, Fe³⁺ may both promote and inhibit the photodegradation of antibiotics. Studies have shown that moderate amounts of Fe³⁺ promote the photodegradation of antibiotics, while excessive amounts inhibit it [37].

Nitrogen (N) pollution is one of the major global pollution issues, and N in the environment not only influences the behavior of antibiotics but is also subject to alterations in its own cycle due to the action of antibiotics [38]. Firstly, nitrate ions (NO₃⁻) exhibit strong light absorption properties, which can inhibit the photolysis of antibiotics by acting as a photomasking agent [39]. This effect has been observed in various environmental contexts, such as in soil and water bodies, where nitrate can reduce the photodegradation of antibiotics by absorbing light and thereby protecting the antibiotics from degradation. Furthermore, antibiotics can also affect nitrification and denitrification in the nitrogen cycle, thereby controlling NO₃⁻ levels in the aquatic environment [40]. Numerous studies have shown that the accumulation of antibiotics in the water affects processes such as nitrification, denitrification, and anaerobic ammonia oxidation to varying degrees [41,42,43]. The effects of sulfonamides, chloramphenicol, tetracyclines, macrolides, and quinolones on the denitrification of estuarine sediments have been investigated, revealing that the addition of various types of antibiotics inhibited denitrification rates and that multiple antibiotics had synergistic inhibition [44].

Phosphorus (P) is an important nutrient for plant and animal growth and can alter the negative charge of sediment and sediment mineral surfaces, thereby affecting the behavior of contaminants in sediment [45]. Research has indicated that phosphorus significantly hinders the adsorption of tetracycline (TCs) onto sediment surfaces. This is attributed to the negative charge of phosphate ions, which competes with TCs for adsorption sites. As the concentration of phosphorus in the water increases, the adsorption of TCs is further inhibited [46]. Phosphorus (P) competes with tetracycline for adsorption, thus reducing the amount of tetracycline adsorbed.

3.3.2. Influencing Factors in the Sediment

The overall interpretation of the RDA in the sediment was 43.8%, with 33.4% for RDA1 and 10.4% for RDA2 (

Table 4. Contribution of sediment physicochemical indicators to the interpretation of redundancy analysis.

Indicator	Explanation %	Contribution %	Pseudo-F	p
Clay	26.2	33.5	2.1	0.094
OM	9.3	11.9	0.7	0.608
BD	14.2	18.1	1.2	0.36
pH	13.9	17.8	1.3	0.384

; Figure 3b). In the sediment, TC and DOX were positively correlated with BD, clay, and pH, and negatively correlated with OM. Chlortetracycline (CTC) was positively correlated with pH and negatively correlated with OM and clay. Demeclocycline (DMC) was positively correlated with clay and BD and negatively correlated with pH.

Due to their large specific surface area and high surface energy, clay-rich sediments exhibit a high adsorption capacity. Furthermore, the abundance of positive and negative charges in clay enhances electrostatic adsorption and fosters bridging bonds [47,48]. Organic matter (OM) is one of the major adsorption-active components of sediments, and its large number of deprotonated functional groups, such as -COO, provides possible adsorption sites for positively charged antibiotic ions [49]. Antibiotics can be adsorbed through hydrogen bonding with polar functional groups in organic matter [50] or by forming complexes with metal ions [51]. The pH of the environment plays a crucial role in determining the charge properties of sediment minerals and the charge state of antibiotics, which in turn influences the adsorption process [52]. The pH of groundwater in the study area is predominantly neutral to slightly alkaline (Table S4). In terms of the pKa values (Table S2) of the four antibiotics, DMC has only one pKa of 4.5, while the other three antibiotics each have three pKa values and exhibit relatively similar properties. Combining the pH of the water and the pKa values of the antibiotics, it can be inferred that DMC primarily exists as DMC- ions, while TC, CTC, and DOX mainly exist as TC^±, CTC^±, and DOX^± ions, respectively. Soil generally carries a negative charge, which causes it to repel the similarly negatively charged DMC- ions, while adsorbing the TC^±, CTC^±, and DOX^± ions. This leads to an increase in the content of TC, CTC, and DOX in the soil, while the content of DMC is relatively reduced. As a result, TC, CTC, and DOX exhibit higher K_d values, while DMC has a relatively smaller one. This explains why, in the RDA, pH shows a negative correlation with DMC but a positive correlation with TC, CTC, and DOX.

Bulk density (BD) represents the ratio of the mass to a specific volume of sediment, encompassing both sediment particles and the pore spaces in between. The smaller the BD, the larger the pore space between the particles, which makes the sediment have more specific surface area for antibiotic adsorption [53].

3.4. Quantitative Prediction

3.4.1. Quantitative Prediction and Fitting Results

By calculating the sediment–water partition coefficients of TCs in each sampling site and using RDA to filter out the main contributing indicators in sediment–water physicochemical properties, the quantitative prediction of TC, DMC, CTC, and DOX with seven environmental sediment (BD, OM, pH, and clay) and water (Fe³⁺, NO₃⁻, and PO₄³⁻) factors was established using PLS analysis. The quantitative prediction, taking into account environmental factors, is detailed in the subsequent equations:

$$\begin{gathered} \mathrm{L}\mathrm{o}\mathrm{g}{K}_{d\left(\mathrm{T}\mathrm{C}\right)}=7.736-0.039\mathrm{\ast }\mathrm{B}\mathrm{D}-0.019\mathrm{\ast }\mathrm{O}\mathrm{M}-0.315\mathrm{\ast }\mathrm{p}\mathrm{H}-0.147\mathrm{\ast }\mathrm{c}\mathrm{l}\mathrm{a}\mathrm{y} \\ -0.028\mathrm{\ast }{\mathrm{F}\mathrm{e}}^{3+}-0.009\mathrm{\ast }{\mathrm{N}\mathrm{O}}_3^{-}-0.213\mathrm{\ast }\mathrm{P}{\mathrm{O}}_4^{3-} \end{gathered}$$

$$\begin{gathered} \mathrm{L}\mathrm{o}\mathrm{g}{K}_{d\left(\mathrm{D}\mathrm{M}\mathrm{C}\right)}=5.221+2.851\mathrm{\ast }\mathrm{B}\mathrm{D}+0.03\mathrm{\ast }\mathrm{O}\mathrm{M}-1.109\mathrm{\ast }\mathrm{p}\mathrm{H}+0.112\mathrm{\ast }\mathrm{c}\mathrm{l}\mathrm{a}\mathrm{y} \\ -0.807\mathrm{\ast }{\mathrm{F}\mathrm{e}}^{3+}+0.005\mathrm{\ast }{\mathrm{N}\mathrm{O}}_3^{-}+0.21\mathrm{\ast }\mathrm{P}{\mathrm{O}}_4^{3-} \end{gathered}$$

$$\begin{gathered} \mathrm{L}\mathrm{o}\mathrm{g}{K}_{d\left(\mathrm{C}\mathrm{T}\mathrm{C}\right)}=7.746-0.405\mathrm{\ast }\mathrm{B}\mathrm{D}-0.028\mathrm{\ast }\mathrm{O}\mathrm{M}+0.039\mathrm{\ast }\mathrm{p}\mathrm{H}-0.211\mathrm{\ast }\mathrm{c}\mathrm{l}\mathrm{a}\mathrm{y} \\ -0.411\mathrm{\ast }{\mathrm{F}\mathrm{e}}^{3+}-0.028\mathrm{\ast }{\mathrm{N}\mathrm{O}}_3^{-}-0.101\mathrm{\ast }\mathrm{P}{\mathrm{O}}_4^{3-} \end{gathered}$$

$$\begin{gathered} \mathrm{L}\mathrm{o}\mathrm{g}{K}_{d\left(\mathrm{D}\mathrm{O}\mathrm{X}\right)}=2.939+1.053\mathrm{\ast }\mathrm{B}\mathrm{D}+0.013\mathrm{\ast }\mathrm{O}\mathrm{M}-0.075\mathrm{\ast }\mathrm{p}\mathrm{H}-0.061\mathrm{\ast }\mathrm{c}\mathrm{l}\mathrm{a}\mathrm{y} \\ -0.344\mathrm{\ast }{\mathrm{F}\mathrm{e}}^{3+}-0.006\mathrm{\ast }{\mathrm{N}\mathrm{O}}_3^{-}-0.243\mathrm{\ast }\mathrm{P}{\mathrm{O}}_4^{3-} \end{gathered}$$

In the TC quantitative predicting equation, clay had the highest VIP score of 1.625, followed by Fe³⁺ at 1.328, which basically controlled the distribution of TC in the water–sediment medium (

Table 5. VIP values of each indicator in the quantitative predicting equations.

Equation	SD a	R2 b	Q2cum c	VIPs
				BD d	OM d	pH d	Clay d	Fe3+ e	NO3− e	PO43− e
TC	0.022	0.99	0.96	0.659	0.949	0.533	1.625	1.382	0.703	0.580
DMC	0.049	0.98	0.93	0.898	0.394	0.862	1.926	0.498	0.378	1.094
CTC	0.040	0.99	0.99	0.099	0.070	0.958	1.372	0.617	1.326	1.430
DOX	0.089	0.95	0.83	0.495	1.122	0.540	0.189	1.995	0.878	0.646

Notes: ^a SD is the standard deviation. ^b R is the correlation coefficient between observed values and fitted values. ^c Q²_cum is the cumulative cross-validated regression coefficient. ^d is the sediment’s physical and chemical indicators. ^e is the water chemistry

). In the CTC model, PO₄³⁻ had the highest VIP score of 1.430, followed by clay and NO₃⁻ at 1.372 and 1.326, respectively, which basically controlled the partitioning of CTC in the sediment. Among all the control indicators, the number of important indicators of water chemistry was slightly larger than that of sediment physicochemical indicators, which also indicated that water chemistry indicators were essential in the quantitative predicting equation of the antibiotic partition coefficient in water and sediment.

3.4.2. Quantitative Predicting Equation Validation

Based on four models, log K_d values were predicted for the sediments in eight sites, and the predicted results are shown in Table S6. The predicted values agree very well with the corresponding observed values due to the low residual values between the observed and predicted values. The maximum differences between the detected and predicted log K_d values for the four antibiotics were 0.04, 0.07, 0.07, and 0.15, respectively. For each of the four quantitative predicting equations, the measured log K_d values were compared with the predicted values. The study demonstrated that the log K_d values for four antibiotics, as predicted by quantitative models, closely aligned with the experimentally measured data, as evidenced by their proximity to the 1:1 line within the 95% confidence interval (Figure 4).

This suggests a high degree of concordance between the predicted and observed binding affinities. It should be noted that the validation of this model was performed using the same dataset as for calibration. While this demonstrates the goodness of fit of the model to the existing data, it may overestimate its predictive accuracy on new data. This limitation is primarily due to the sample size constraints of this study. Therefore, the reported model performance metrics should be regarded as the upper limit of the model’s explanatory power, and its predictive performance in practical applications requires further confirmation through validation with larger-scale independent datasets in the future. Studies have demonstrated the accuracy and feasibility of prediction models, such as those developed for antibiotic distribution in sediments, which can be reliably applied to predict log K_d values in various sedimentary environments.

3.4.3. Quantitative Predicting Equation Interpretation

The highest overall VIP values were obtained for clay among all environmental factors, as shown in Table 5. The observed and predicted log K_d of TCs in karst pool water–sediments are shown in Table S6. The VIP values for clay were in the TC predicting equation (1.625). Values derived from the DMC predicting equation (1.925) and CTC predicting equation (1.372) were much greater than those in the DOX predicting equation (0.189), indicating that the clay content contributed more to the adsorption process in TC, DMC, and CTC. Since the clay in TC and DMC had positive coefficients, it could be concluded that an increasing clay content leads to an increase in K_d values. Clay-rich sediments have a higher adsorption capacity due to their larger specific surface, higher surface energy, and intercalation. In addition, clay is rich in positive and negative charges, which can enhance bridging bonds and electrostatic adsorption [47,48]. The negative coefficients for clay content in CTC and DOX are mainly due to the fact that clay is not the dominant controlling factor in these two models. The other primary controlling factors exert a greater influence than the adsorption effect of clay.

The VIP values for Fe³⁺ in water were also high, with the DOX predicting equation (1.995) and the TC predicting equation (1.382) both having VIP values above 1. The coefficients in the predicting equations were negative, suggesting that an increase in Fe³⁺ concentration in water leads to a decrease in K_d values, due to the significant inhibitory effect of excess Fe³⁺ in water on the oxidative degradation of antibiotics. It was shown that when n(Fe³⁺): n(CTC) > 1, the degradation rate of CTC decreases. This is because the excess Fe³⁺ complexes with the same CTC fraction, while one CTC molecule binds up to two Fe³⁺ ions, at which point the saturated CTC-Fe complex is formed with a more stable spatial structure, and the photodegradation of tetracycline becomes poor [37].

The VIP values in water PO₄³⁻ were also high, with both the CTC predicting equation (1.43) and the DMC model (1.094) having VIP values above 1. All coefficients in the model were negative, except for the DMC coefficient, which was positive. The larger the PO₄³⁻ in the water, the smaller the K_d value, because the phosphate in the water competes with the antibiotic for adsorption, thus inhibiting the adsorption of TCs on the sediment.

Organic matter (OM) is closely related to clay in the sediment and has a strong influence on the sorption behavior of antibiotics [49]. Due to the positive coefficient of organic matter content, the log K_d of DOX is proportional to the sediment OM, and an increase in the OM content of the sediment enhances the adsorption of DOX to the sediment.

The VIP value of NO₃⁻ in water in the CTC predicting equation is higher than 1, at 1.326. The negative coefficient of NO₃⁻ in this model indicates that the K_d value in the system becomes smaller as the concentration of NO₃⁻ in water increases. Firstly, nitrate presence inhibits the photolysis of antibiotics, as reported by Ge et al. (2010) [39]. Secondly, antibiotic presence obstructs denitrification, according to Feng et al. (2025), thereby causing a continuous accumulation of NO₃⁻, which eventually leads to a smaller K_d [40].

The VIP values for the BD and pH indicators in the remaining sediments, although neither exceeded 1, had overall high VIP values for pH (0.533–0.862), showing that they also have a certain degree of importance. The negative pH coefficient in the distribution models for DMC and DOX suggests that these antibiotics preferentially partition into sediments with lower pH levels, and an increase in sediment pH correlates with a reduction in the adsorption of these antibiotics. The positive sign of pH in the TC and CTC models indicates that both antibiotics are often distributed to sediments with high pH. The lower adsorption at high pH may be attributed to electrostatic repulsion of anionic antibiotics and negatively charged sediments at high pH [54,55]. In contrast, under low-pH conditions, antibiotics predominantly exist as cations, and cation exchange emerges as a pivotal mechanism governing their sorption extent [56]. However, this is not conducive to the occurrence of cationic species across the environmentally relevant sediment pH range (4.70–7.80). For the four antibiotics, the adsorption behavior may be a surface complexation mechanism rather than cation exchange [24].

4. Conclusions

Through field investigations and quantitative predictions, our study offers insights into the detection patterns and distribution characteristics of TCs within the sediment and water bodies of the karst wetland.

The results reveal that environmental antibiotic concentrations are influenced by numerous factors. Key physicochemical properties—including Fe³⁺, NO_3⁻, and PO_4³⁻ in aqueous ecosystems, alongside clay content—govern antibiotic behavior in water. Meanwhile, sediment characteristics such as OM, BD, and pH significantly influence antibiotic distribution. This study developed four quantitative predictive equations, offering robust methods to estimate distribution coefficients for antibiotics and structurally analogous novel contaminants like TC, DMC, CTC, and DOX within natural water–sediment systems. Although the selected indicator factors and limited sampling sites do not fully cover the physical and chemical characteristics of the sediment and water, they do contain most of the indicators. Compared with previous studies, the inclusion of water chemistry indicators in this study improves the indicator system for the quantitative predicting equations.