Human impacts on risk of antibiotics in soil differentially depend on land system across scales

doi:10.21203/rs.3.rs-2484516/v1

Download PDF

Article

Human impacts on risk of antibiotics in soil differentially depend on land system across scales

https://doi.org/10.21203/rs.3.rs-2484516/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Antibiotics have been widely used to protect human health and improve food production, however, they persist in soil and undermine ecosystem sustainability. The human impacts on soil susceptibility to antibiotic pollution on various spatial scales are poorly understood. Here, we predicted the geographic characteristics of antibiotic pollution risk and explored the corresponding human impacts on multiple spatial scales in China, a representative of high level of human activity. Given that human activities increase antibiotic pollution risk, land systems play a more fundamental role in risk increase than do population and economic growth. With decreasing spatial scale, land use composition had pronounced contributions to antibiotic pollution risk instead of land management strategy (55.9 ± 13.4% vs. 7.0 ± 1.7%). Furthermore, we identified the regions of concern, and thresholds of the effects of land system intensification on risk increase based on their non-linear changes. The scale-dependent relationships elucidate the sustainability of human–environment systems across spatial scales, which is beneficial to the establishment of strategies and action plans worldwide.

Earth and environmental sciences/Environmental sciences/Environmental impact

Earth and environmental sciences/Environmental social sciences/Sustainability

Since the onset of the Anthropocene, human activities have driven substantial socioeconomic development, however, they have also created significant environmental risks, which must be managed^1–3. As an emerging apparent issues, antibiotics are used worldwide to control microorganisms (e.g. for human disease prevention and animal growth promotion)^{4, 5}. They are emitted into the environment during their manufacture, use, and disposal⁶, thereby posing a significant environmental risk. Among various pollutant sinks, soil can adsorb a large amount of antibiotics, thus, it is a major reservoir of antibiotics in the terrestrial ecosystem⁷. Agricultural, domestic, and industrial discharges, such as through manure fertilisation, wastewater irrigation, and waste landfilling, are significant pathways of antibiotic introduction into the soil⁸. Environmental exposure to antibiotics is increasingly undermining the health of soil ecosystems, such as increasing the prevalence of resistant pathogens in soil, disturbing the structure of microbial communities, and damaging cell metabolism or enzymatic activity^{9, 10}. Given the increase in human demand for antibiotics, pollution risks caused by antibiotic utilisation and emission are expected to increase in the future^{4, 5}, thus, managing antibiotic utilisation and emission is crucial for maintaining soil security and human health¹¹.

Although existing studies are beneficial for policymakers, they have mainly focused on the occurrence and risk assessment of antibiotics in soil from field to regional scales^{12, 13}. In both developed and underdeveloped countries, it is difficult to quantify risk dispersion or conduct risk assessments of antibiotic pollution on a massive scale. As human-made and human-discharged compounds, soil pollution caused by antibiotics is closely related to human activities which can be predictors for modelling approach. Understanding responses of antibiotic pollution to human activities can greatly improve model performance. Several studies have attempted to simulate the spatial characteristics of antibiotics in soil linking to human activities^{14, 15}. However, understanding the causes of antibiotic pollution is still substantially challenging, partially because human impacts on antibiotic pollution can be amplified by interactions between global industrialisation, population growth, and land system change over a vast range of spatial scales. A recent survey found that anthropogenic factors (such as population density, gross domestic product [GDP], and livestock production) were primarily responsible for the high antibiotic resistance in the soil environment¹⁶. However, it is confusing to find that there were no notable correlations between antibiotic concentrations and human population density on a regional scale (e.g., the Yangtze River Delta region of China¹⁷), whereas land use patterns substantially affected antibiotic pollution on the watershed scale¹⁸. How human societies affected antibiotic pollution across scales is key to addressing related environmental risks, however, few efforts are attempted to bridge the scale gaps.

Some important knowledge gaps still exist, and filling those gaps could further elucidate the effects of human activities on antibiotic pollution risk and the related risk reduction strategies. This study aims to address three significant questions by taking China as a case study, which is one of the greatest consumers of antibiotics worldwide and is exposed to antibiotic pollution risk. Specifically, we attempted to (i) provide reliable evidence for antibiotic pollution risk at broad scales using machine learning algorithms, (ii) bridge the scale gaps in the relationships between human activities and antibiotic pollution risk, (iii) reveal the threshold of human impacts on antibiotic pollution risk to manage human societies.

To address these issues, we first predicted the ecological risk associated with antibiotics in Chinese soil using machine learning logarithms. According to the environmental Kuznets curve hypothesis¹⁹, we proposed potential models of human impacts on antibiotic pollution risk on multiple scales (Supplementary Section 1). Using these models, we examined the relationships between human activities (including population, economic, and land systems) and antibiotic pollution risk with downscaling (that is, decline in size or extent of space). After identifying the regions of concern, we explored the thresholds of the land system intensification on antibiotic pollution risk via the moving-window approach²⁰. On the basis of these analyses, we provide guidance for the development of tailored strategies for antibiotic pollution risk control and sustainable soil utilisation.

Predicting the risks of antibiotics in soil. To quantify the risks in each geographic grid cell, we modelled the RQs of each target antibiotic in soil using ensemble random forest (RF) models fed with georeferenced datasets. The differences between the modelled and validated RQs were generally less than 1, demonstrating the good prediction performance of our models (Supplementary Figs. 3 & 4). This procedure allowed us to clarify soil susceptibility to antibiotic pollution. To elucidate the extent of pollution caused by antibiotic mixtures, we calculated the cumulative RQs for the ecosystem in each grid cell considering the additive effects of target antibiotics. The average cumulative RQ of antibiotics in the Chinese soil environment was 6.2 ± 2.1 (Fig. 1a). The cumulative RQ was particularly high in central and eastern China, which suffer considerable exposure to antibiotics. The relative uncertainty (ratio of standard deviation to mean) across the ensemble model was ± 26% of the predicted data on average, but areas of substantial uncertainty existed. Huang–Huai–Hai and the Northeast Plains, which are the main grain-production areas accounting for more than one fifth of the agricultural output of China, exhibited the highest uncertainties (Fig. 1b). This suggests that anthropogenic activities possibly caused a wide range of antibiotic pollution risks in soil. An RQ > 1 indicates that antibiotics pose serious risks to the ecosystem (Supplementary Fig. 3). Our model estimated that ~ 11.4% of the land was contaminated by more than one antibiotic compound. Among the target antibiotics, ofloxacin accounted for the most pronounced risk, with an RQ of 0.06–1.71, followed by tetracycline (RQ = 0.05 to 1.70, Fig. 1c).

Scale effects and human impacts on risks of soil antibiotics. We found that the risk levels of antibiotics in soil differed among land use types, but the relationships were dependent on the geographic scale (Figs. 1d–g, and Supplementary Fig. 5). The risk levels of quinolone antibiotics in soil from built-up areas were consistent across all scales. However, tetracycline antibiotics exhibited pronounced risks in agricultural soil on a larger scale, while quinolones exhibited higher risks on a smaller scale. Forestlands exhibited the lowest antibiotic risk levels, as they are least disturbed by human activities. The results suggest that human activities are associated with antibiotic pollution risks on multiple scales.

We then calculated a human impact index to examine the scale-dependent relationships between human activity intensity and antibiotic pollution risks (Figs. 2a–d). First, we divided the areas into high- and low-impacted groups, according to the Hu Huanyong line (population and economic demarcation line in China, and the east and west of the line are characterised by high and low human impact levels respectively^{21, 22}, Supplementary Fig. 6). The relationships between human activity intensity and antibiotic pollution risks varied with the high- and low-impacted area. In the high-impacted areas, antibiotic pollution risk was strongly dependent on human activity intensity, as verified by their strong positive correlations on different scales (R² = 0.41–0.53, p < 0.05). However, human activities did not increase antibiotic pollution risks across all scales, and particularly not in low-impacted areas (that is, west of the Hu Huanyong line). The low-impacted areas featured a relatively low human influences on antibiotic pollution risk at the larger (level 1) scales (R² = 0.14, p > 0.05, Fig. 2a).

In addition, we evaluated scale effects by changing the scale extent (that is, the proportion of areas selected along the human-impacted gradient, Methods). We evaluated the effects of segment exclusion from the lowest or highest ends of the human-impacted gradient on the relationships between human impacts and antibiotic pollution risk. At the large watershed level, the change in scale extent had divergent effects, suggesting that the spatial extent strongly modulated human impacts on antibiotic pollution risk (Supplementary Fig. 7). Moreover, management (particularly manure application) increased the risk levels in most cases (16 of 20 analyses, Fig. 2e). However, with declining watershed level, all anthropogenic factors generally showed N-shaped or linearly increasing relationships with antibiotic pollution risk at both the lowest and highest ends of the human-impacted gradient (that is, human activities tended to promote antibiotic pollution risk on the small watershed scale). Generally, land use (mainly arable land and built-up land) was positively correlated with antibiotic pollution risk and showed more pronounced effects on the small watershed scale (Figs. 2f–h, and Supplementary Fig. 7). The results indicate that the scale (both grain size and extent) tended to influence human impacts on antibiotic pollution at the large watershed level. Land use and management could increase the likelihood of antibiotic pollution risk to the soil at a smaller watershed level and a broader spatial extent.

Capturing shifts in land systems, such as land use composition and management practices, provides a more sensitive predictor of antibiotic pollution risk. Given the complementarity between land use and management, they showed opposite scale-dependent contributions to the RQs of antibiotics (Figs. 2i–l). With downscaling, the contribution of land use composition increased from 9.0 ± 9.8% to 55.9 ± 13.4%, while that of land management declined from 17.2 ± 8.5% to 7.0 ± 1.7%. The scale dependence suggests that antibiotic pollution risk resulting from land management generally dominated at larger scales, while land use-related pollution risk dominated at smaller watershed levels.

Risk reduction through land systems. Changes in the GDP–RQ relationship with land use intensity also showed that a tradeoff existed between economic growth and antibiotic pollution risk (Fig. 3). The area between the GDP and RQ curves indicates the potential to improve land use intensity. According to the tradeoff between GDP and RQ, the threshold and improvement potential of land use intensification were lower on the smaller scale (Supplementary Fig. 8). The results demonstrate that land use intensity could be important for explaining antibiotic pollution risk. To provide a synthesized insight of risk control, we integrated the indicators of antibiotic pollution risk and land use and management to identify regions of concern, i.e., those that may need sustainable land systems (Figs. 4a and b). In the maps, concern level 5 corresponds to the regions of high risk and high land use intensity. Compared with land management, the consideration of land use composition resulted in larger level 5 regions of concern.

To manage land systems, their effects on antibiotic pollution risk were evaluated on four scales via the moving-window approach. The partial effects of land systems on antibiotic pollution risks changed significantly with the indicators of land use composition and management (Fig. 4c). On the larger watershed scale, the partial effects of arable land, manure application, and irrigation first decreased and then increased to a higher value with land system intensification. In contrast, the effects of built-up land showed a monotonous positive correlation with antibiotic pollution risk. The changes in the effects of land use composition and management on the small watershed scale were consistent (they first increased and then decreased with increasing land use intensification) and opposite to the changes on the large scale. Thus, the thresholds for the arable land percentage, built-up land percentage, manure application rate, and irrigated area proportion on watershed level 4 could be identified: 22.9%, 1.9%, 424.8 kg N/km²/yr, and 34.2%, respectively. Moreover, we observed that effects of land use composition and management were depended on watershed type (Supplementary Fig. 9), and their effects were generally significant in agriculture-dominated watersheds. Our findings suggests that the adapted management to the watershed type should be tailored.

During the Anthropocene era, human demands and land use intensity have increased. Throughout history, humans have modified the soil ecosystem to obtain life-sustaining resources. This has recently resulted in negative environmental effects, such as antibiotic pollution, which must be managed to protect soil health²³. Our findings obtained using an ensemble random forest model fill the gaps in information on antibiotic pollution in China and expand previous regional-scale risk assessments^{15, 24} (which were limited by data availability and spatial scale). This study first investigated soil susceptibility to antibiotic pollution on a large scale. The results revealed that antibiotic pollution poses a widespread risk to the soil ecosystem of China, which has similar geographic patterns with socioeconomic status (which increase from west to east) partially substantiated by a national-scale survey²⁵. Our prediction reveals that ofloxacin poses a higher risk to the soil ecosystem than other target antibiotics, which is similar with its risk priority by by ranking the prevalence, persistence, bioaccumulation, and toxicity ²⁶. Our findings reinforce the need for a more detailed risk assessment of antibiotic pollution to control and monitor soil susceptibility.

Here, we quantitatively investigated human impacts on soil susceptibility to antibiotic pollution across a broad range of scales. In areas with both high and low levels of human activity intensity, the antibiotic pollution risk was mainly originated from anthropogenic activities across all scales. Despite the importance of soil and climatic conditions, human activities are the dominant predictor of antibiotic pollution risk⁷. However, on the large watershed scale, human activities do not necessarily expose the soil ecosystem to high antibiotic pollution risk in low-impacted areas, which are characterised by limited human-dominated landscapes and low precipitation and temperature. In these areas, while human activities certainly explain part of the variability in soil susceptibility to antibiotic pollution, the variability in Earth’s surface processes, such as surface runoff, soil erosion, and leaching, plays an appreciable role in explaining the observed patterns^{27, 28}.

Among various anthropogenic factors, land systems, which are terrestrial social–ecological systems interacted by human and environmental sub-systems through land use and management²⁹, have been highlighted to affect antibiotic pollution risk on multiple spatial scales^{7, 13}. Since the intensification of land system, land use composition and management intensity are the predominant factors affecting antibiotic pollution risk across spatial scales instead of population and economic systems. The weak correlations between population density and antibiotic pollution in soil were also demonstrated as previous studies shown^{17, 30}. Compared with population migration and industrial restructuring, land systems are more sensitive to large-scale policy programmes³¹. Land system optimisation may offer a sustainable alternative for antibiotic pollution control. In the current study, land use and management practices directly determined antibiotic emission into the soil ecosystem. For example, the expansion of arable lands, which are treated with antibiotic-containing fertiliser, could result in antibiotic pollution to more land^{7, 32}. It can be expected that the watersheds consisting of higher proportion of arable land suffer more severe antibiotic pollution than watersheds with higher proportion of natural land. The domestic and industrial discharges from built-up land also expose soil to a high risk of antibiotic pollution⁹. Thus, reducing the area proportion of arable and built-up land can mitigate antibiotic pollution.

Interestingly, we find that the effects of land use and management were influenced by spatial scale in this study. Land management, rather than land use composition, primarily explained antibiotic pollution risk on a large spatial scale. Despite the expansion of human-dominated landscapes, natural land (including forest and grassland) still accounts for most of the Earth’s land surface (for example, built-up lands only account for less than 0.7% of the total land area on Earth³³). From a broad-scale viewpoint, the variability of land use change is generally lower than the variability of management among watersheds, and it is the latter that causes the differences in antibiotic pollution risk.

Intensification of land system is now one of the main agendas benefiting socioeconomic development across and within human societies, and thus is typically associated with benefit–risk tradeoffs³⁴. In our study, the tradeoff between economic growth and antibiotic pollution risk shows that the high priority afforded to efforts of land use intensification is necessary to achieve the expected likelihood of win–win outcomes, that is, economic development and ecological risk control. However, it is often not clear whether these twin benefits are achieved across different scales. We attempted to account for the scale effects, as scale changes can completely shift statistical supports. Our analysis revealed that the likelihood of achieving win–win outcomes on the small watershed scale is lower than that on a large scale; thus, policymakers, farmers, and other actors should pay more attention to small-scale cases. Overall, the findings of this study and others³¹ suggest that land system intensification is an urgent issue limiting sustainable development and more efforts should be conducted.

The predictions of this study indicate that land use and land management play significant roles in antibiotic pollution risk. Understanding the effects of land use and management on antibiotic pollution risk in soil ecosystems is crucial for land system management. On a larger scale, the effects of land use and management on antibiotic pollution risk increased with land use intensification; however, the effects were inverted U-shaped on the smaller scale. The nonlinear changes in land use/management effects partially coincided with the findings of previous studies from distinct perspectives^{14, 18}. The intensification of land use and management elevated antibiotic discharge into the soil and thus increased the risk level. However, the increasing trend is expected to flatten (or even decrease) after a turning point, because of the limitations of the soil-carrying capacity and loading capacity³⁵. The threshold for antibiotic persistence in soil exists because the soil-carrying capacity of antibiotics is a function of clay content, organic matter, and acidity^{7, 36}. Moreover, land use intensification increases the need for further financial inputs (such as manure fertilisers), which are either not available or, more often, not affordable (i.e., not cost-effective), and the threshold for loading of antibiotics likely exists when input–output ratio peaks³⁴. The above-mentioned studies explained the observed decoupling between land use intensification and antibiotic pollution risk at high human-impacted levels. Our analysis provides insights into the consequences of land use intensification to enable sustainable land policy planning and management.

Our study goes beyond previous work on quantitative assessment of human impacts on antibiotic pollution in soil across multiple spatial scales. However, the following uncertainties and limitations need to be further considered. (i) The antibiotic pollution risk presented in this study may be overestimated, mainly because the risk assessment model assumed maximum exposure of non-target organisms to represent the worst scenario; nonetheless, the estimated risk can be a good reference for regulatory agencies in making critical decisions. (ii) We assumed the non-existence of target antibiotics that were not reported; however, these antibiotics might be present but below the detection limit of the adopted methodology³⁷. (iii) Although this study solely focused on ecological security, the effect of antibiotics on human health is also an important aspect that requires comprehensive assessment^{38, 39}. Several studies have modelled antibiotic concentrations in soil using regression techniques, although with limited modelling accuracy^{14, 40}. Additional biophysical models and available data should be developed to accurately predict the risks associated with antibiotics in multiple environmental compartments and investigate the mechanisms underlying the social–ecological system. Nevertheless, our findings might enable the generalization to other regions or countries. Complex human activities often result in heterogeneous land use and management strategies. Sustainable socioeconomic development is required for antibiotic pollution risk control. To achieve sustainability goals, innovative and effective policies, implementation strategies, and stakeholder actions should be established to transition towards sustainable soil use and sustainable living with low exposure to antibiotic pollution.

Overview. We collected data on the measured environmental concentrations (MECs) of antibiotics in Chinese soil. A large number of antibiotics widely occurred in soil, and nine frequently detected antibiotics with wide distribution and high toxicity were selected: tetracycline, chlortetracycline, oxytetracycline, doxycycline, ofloxacin, norfloxacin, ciprofloxacin, enrofloxacin, and lomefloxacin. We also collected data on anthropogenic factors, climatic factors, and soil and vegetation characteristics that could explain the variability in the risks of antibiotic pollution as predictors. Next, we scaled up the risks associated with target antibiotics in soil using a machine learning model (RF algorithm)⁴¹. We combined a linear model, generalised additive model, and moving-window approach to describe the relationships between human activities and antibiotic pollution risks on four scales. All modelling and analyses were conducted using R software 4.2.1.

Data collection. To determine soil susceptibility to antibiotic pollution, a comprehensive MEC dataset was established. We compiled the publicly available dataset of Zhang et al.⁴², which records antibiotic occurrence and distribution in Chinese soil. This dataset contains georeferenced occurrences for our target antibiotics. Moreover, we searched Google Scholar and the China National Knowledge Infrastructure database (http://www.cnki.net/) for field investigations of antibiotics in Chinese soil and gathered 376 studies published from 2000 to 2020. The search terms were ‘antibiotics’ or ‘pharmaceutical and personal care products’ and ‘soil’ or ‘solid matrices’. We recorded the average concentration of target antibiotics from the literature, considering land use type (urban green space, cropland, grassland, and forestland) and experiment location (coordinates) following the standards of Zhang et al.⁴². Concentrations that were ‘not detected’ or ‘below detection limit’ were regarded as zero values. The MECs were converted into standardised units (µg/kg or ng/g). When the data were presented in figures, mean values and standard error were extracted using GetData 2.26 (www.getdata-graph-digitizer.com). The antibiotic measurements in the dataset correspond to different soil depths, from 5 to 40 cm, with an average depth of ~ 20 cm. When scaling up the susceptibility of soil to antibiotic pollution, we considered a fixed depth value of 0–20 cm.

To expand the existing data, we also collected soil samples from Yunnan and Zhejiang provinces and measured the concentrations of target antibiotics. The soil samples were transported to the laboratory, freeze-dried, and stored at − 20°C until further analysis. The extraction and analysis of antibiotics are detailed in Supplementary Methods (Supplementary Section 2). The measured antibiotic data were combined with the dataset. Our dataset contained the records of antibiotics in most regions of China, which comprehensively depicts the MECs of antibiotics in soil (Supplementary Table 1).

Risk assessment model. To assess the ecological risks of the target antibiotics, the predicted no-effect concentrations (PNECs) were first estimated via the assessment factor approach⁴³. PNECs were determined as the ratio of the median effective concentration (EC50) or median lethal concentration (LC50) to the assessment factor (AF = 1000), or the ratio of the chronic no-observed-effect concentration (NOEC) to the assessment factor (AF = 100); that is, PNEC = EC50 (or LC50)/1000 or PNEC = NOEC/100. Owing to the scarcity of available data, the toxicity data of the relevant terrestrial organisms (e.g., vegetables, crops, earthworms, and bacteria) were collected from the literature⁴⁴. Then, the species-specific RQs of the antibiotics were calculated as the ratio of MEC to PNEC (that is, RQ = MEC/PNEC). For some antibiotics, the PNECs of soil (PNEC_soil) were estimated from their values in water (PNEC_water), which were corrected by the solid–water partition coefficient (K_d) via the equilibrium partitioning method (that is, PNEC_soil = PNEC_water × K_d). To consider the worst scenarios, we defined the risk levels of antibiotic pollution as the highest RQ values. We determined the cumulative risks in soil as the sum of the highest RQs of each target antibiotic.

Identification of candidate predictor variables. The predictor variables were determined from spatially explicit datasets (Supplementary Table 2). Six anthropogenic factors were collected, namely land use type, livestock density (cattle, pigs, and chickens), human population density, GDP, chemical fertilisation (nitrogen and phosphorus), and pesticide use, which are linked to socioeconomic development and agricultural activities. Climatic information on the annual average temperature and precipitation was sourced from the CRUTS V4 database⁴⁵. Soil properties, which substantially influence the fates and behaviours of antibiotics in soil, were collected as important predictor variables; they included clay content, organic carbon content, bulk density, soil thickness (distance to bedrock), and saturated hydraulic conductivity. We selected groundwater table depth and normalised difference vegetation index as indicators of leaching and plant uptake capacity. Moreover, we obtained terrain information from the GTOPO30 database⁴⁶, as the terrain can influence antibiotic movement on land. To ensure compatibility across datasets and variables, we processed all data at a spatial resolution of 1 km. The data were extracted according to the experimental site location. The anthropogenic and climatic factors and soil and vegetation characteristics determined the amount of antibiotic residue in soil and the soil susceptibility to antibiotic pollution.

Scaling-up approach and output maps. An RF model including all of the predictor variables⁴¹ was fitted to the log-transformed (log₁₀) RQs of each antibiotic using R packages randomForest⁴⁷ and Caret⁴⁸. All predictor variables were standardised using the Z-score. We constrained the RF models by setting the maximum number of allowed trees to 1000. Since the RQ data were modelled, cross-validation was important. The dataset was randomly divided into training and validation sets, corresponding to 70% and 30% of the data, respectively. Finally, the data-trained RF model was applied to the gridded data of predictors to estimate the risk levels of nine target antibiotics on a larger scale. the ecosystems not covered by soil (including bare land and water,) were excluded from the analysis based on the land-cover map. An ensemble model for our final predictions and uncertainty analysis was developed via the Monte Carlo approach. Our final model was an ensemble of 500 RF models (that is, the model was separately trained 500 times; each time, a different set of data was selected from our dataset) to generate a distribution of model error on random data⁴⁹. The distribution density of differences between measured and predicted data was determined for each target antibiotic to assess the accuracy of the RF models. We derived the models’ mean prediction across the ensemble models and determined the uncertainties associated with the predictions from the standard deviation. Maps of the distribution of soil susceptibility to antibiotic pollution, expressed as average RQs, and the corresponding uncertainty maps were derived using the data-trained RF models at a resolution of 1 km.

Finally, we integrated the antibiotic pollution risk, land use composition, and management indicators to identify the regions of concern, where human activities resulted in critical antibiotic pollution. We standardised the land use composition and management indicators via the minimum–maximum method and calculated land use intensity by summing the standardised scores. Antibiotic pollution risk was scaled to 0–1, and then multiplied by 4 (the maximum score of land use intensity) to maintain the comparability. The concern level was calculated as the sum of antibiotic pollution risk and land use intensity, and it was further divided into five levels via the natural break method.

Scale effects and human impacts on risk of antibiotics. The primary goal of the analysis was to identify the effects of spatial scale on the relationships between human activities and antibiotic pollution risks. We evaluated the scale effects by generating datasets after changing the grain size (that is, the resolution at which data are generated at a given extent), and then resampled the datasets after changing the scale extent (that is, the proportion of sampled data at a given resolution). First, to explore the scale effects, four watershed levels (sizes 3, 6, 9, and 12) were selected from the HydroSHEDS dataset⁵⁰ to represent the changes in spatial grain (Supplementary Fig. 5); here, these levels are renamed levels 1, 2, 3, and 4, respectively. Watershed level 1 represented a broad-scale grain, while level 4 was a small watershed scale.

The land system can be manipulated by technologies and governance systems across spatial scales and were more sensitive in comparison with other human activities²⁹. Thus, we identified areas highly subject to human disturbance using the geographically gridded maps of the land system, which includes land use patterns and management practices. We summarised the mean risks of antibiotics at four watershed levels and the mean land use composition and mean management intensity. Land use composition was reclassified into the area proportions of arable, built-up, and natural (mainly consisting of grassland and forestland) lands. Land management intensity was indicated by the manure nitrogen application rate and irrigated area proportion (Supplementary Table 2), because manure application and irrigation are generally considered the main pathways of antibiotics into soil⁷. Finally, we integrated the antibiotic pollution risk and land use composition and management indicators to identify the effects of scale on the interaction between soil antibiotic pollution and land system.

We generated a human impact index to characterize the human pressures at four watershed levels using the framework developed by Venter et al.⁵¹ with minor modification, which contained seven anthropogenic variables including arable land, pasture land, built-up land, manure nitrogen application rate, irrigated area proportion, GDP, and population density. For comparison across variables, we placed each variable on a 0–10 scale. The grid cells of built-up land were assigned a score of 10, which represents the pressures from the built environment, and the cells of arable land and pasture land, which represent the agricultural environment, were assigned pressure scores of 7 and 4, respectively. Other land use types were scored zero. The grid cells were divided into 10 quantiles with scores 1 to 10, corresponding to increasing manure nitrogen application intensity and irrigated area proportion. Cells with no manure application and irrigation were assigned zero, and GDP standardisation followed the same procedure. Grid cells with a high human population density (> 1000 people/km²) were assigned a score of 10, and other cells with a lower density were assigned a lower log-scaled score⁵¹:

$$Score=3.333 \times \text{log}(Population density+1)$$

Furthermore, we summed the scores of these variables in each grid cell to estimate the cumulative human impacts. For any grid cell, the human impact index can range from 0 to 50. The data were divided into two sets (high- and low- human impacted groups) according to the Hu Huanyong line²¹. We separately fitted the risks of antibiotics to the human impact index using a linear regression model. The relationships between RQs and the seven adopted human activity indices were assessed using multiple linear regression models.

Second, to avoid arbitrary model choices, we conducted an additional 20 runs by selecting different data to explore the effects of spatial extent. We ranked the data according to increasing human impact index. Reduction in scale extent was through the omission of 0–90% of segments (in intervals of 10%) from the lower and upper limits of the ranked data (Supplementary Fig. 7c); that is, 20 datasets were generated for linear model fitting. Through this approach, data with different ranges of human impacts and scale extents were generated. Then, the best models were selected via a procedure based on the corrected Akaike information criterion (AICc; ΔAICc < 5). Model averaging was conducted to calculate the coefficients of predictors and test their significance. This procedure was performed using the dredge function in the R package MuMIn⁵². Moreover, we characterised the contributions of land use and management to antibiotic-related risks. For doing so, we expressed the contributions of predictors as the percentage of variance explained, according to the absolute values of the ratio of the standardised regression coefficients to the sum of all standardised regression coefficients from all predictors.

Using a moving-window approach, we analysed the effects of land systems on antibiotic-related risks²⁰. We generated a range of window sizes and calculated the relative range of RQs (the ratio of the range of RQs under a given window size to the maximum range under all window sizes). When the relative range exceeded 60%, we fixed the window size. For each window, we calculated the mean RQ and four variables (area proportions of arable land and built-up land, manure application rate, and irrigated area proportion). The partial Pearson correlation coefficients were also calculated. Then, we fitted RQs and partial correlation coefficients against the four variables using a generalised additive model to analyse the effect of the variables on the relationships between human impacts and risks of antibiotics in soil. All of the above-mentioned procedures were repeated on four watershed levels to explore the scale effects.

Acknowledgements

This research was funded by the National Natural Science Foundation of China (42201121), the Young Elite Scientists Sponsorship Program by the China Association for Science and Technology (2021QNRC001), the Youth Innovation Promotion Association, Chinese Academy of Sciences (2018057), and China Postdoctoral Science Foundation (2021M702775).

Jørgensen, P.S., et al. Antibiotic and pesticide susceptibility and the Anthropocene operating space. Nat. Sustain. 1, 632–641 (2018)
Tang, F.H.M., Lenzen, M., McBratney, A. & Maggi, F. Risk of pesticide pollution at the global scale. Nat. Geosci. 14, 206–210 (2021)
Hou, D., et al. Metal contamination and bioremediation of agricultural soils for food safety and sustainability. Nat. Rev. Earth Env. 1, 366–381 (2020)
Van Boeckel, T.P., et al. Global trends in antimicrobial use in food animals. Proc. Natl. Acad. Sci. USA 112, 5649–5654 (2015)
Klein, E.Y., et al. Global increase and geographic convergence in antibiotic consumption between 2000 and 2015. Proc. Natl. Acad. Sci. USA 115, 3463–3470 (2018)
Wilkinson, J.L., et al. Pharmaceutical pollution of the world’s rivers. Proc. Natl. Acad. Sci. USA 119, e2113947119 (2022)
Wu, J., et al. Antibiotics and antibiotic resistance genes in agricultural soils: A systematic analysis. Crit. Rev. Environ. Sci. Technol., 10.1080/10643389.2022.2094693, 1–18 (2022)
Ghirardini, A. & Verlicchi, P. A review of selected microcontaminants and microorganisms in land runoff and tile drainage in treated sludge-amended soils. Sci. Total Environ. 655, 939–957 (2019)
Zhu, Y.-G., et al. Soil biota, antimicrobial resistance and planetary health. Environ. Int. 131, 105059 (2019)
Van Boeckel, T.P., et al. Global trends in antimicrobial resistance in animals in low- and middle-income countries. Science 365, eaaw1944 (2019)
Lehmann, J., Bossio, D.A., Kögel-Knabner, I. & Rillig, M.C. The concept and future prospects of soil health. Nat. Rev. Earth Env. 1, 544–553 (2020)
Bueno, I., et al. Identifying the spatiotemporal vulnerability of soils to antimicrobial contamination through land application of animal manure in Minnesota, United States. Sci. Total Environ. 832, 155050 (2022)
Camotti Bastos, M., et al. Antibiotics and microbial resistance in Brazilian soils under manure application. Land Degrad. Dev. 29, 2472–2484 (2018)
Liu, X., Zhang, J., Gbadegesin, L.A. & He, Y. Modelling approaches for linking the residual concentrations of antibiotics in soil with antibiotic properties and land-use types in the largest urban agglomerations in China: A review. Sci. Total Environ. 838, 156141 (2022)
Chen, H., Jing, L., Teng, Y. & Wang, J. Multimedia fate modeling and risk assessment of antibiotics in a water-scarce megacity. J. Hazard. Mater. 348, 75–83 (2018)
Zheng, D., et al. Global biogeography and projection of soil antibiotic resistance genes. Sci. Adv. 8, eabq8015 (2022)
Sun, J., Zeng, Q., Tsang, D.C.W., Zhu, L.Z. & Li, X.D. Antibiotics in the agricultural soils from the Yangtze River Delta, China. Chemosphere 189, 301–308 (2017)
Zhao, F., et al. An innovative modeling approach of linking land use patterns with soil antibiotic contamination in peri-urban areas. Environ. Int. 134, 105327 (2020)
Sarkodie, S.A. & Strezov, V. A review on Environmental Kuznets Curve hypothesis using bibliometric and meta-analysis. Sci. Total Environ. 649, 128–145 (2019)
Wu, X., et al. Decoupling of SDGs followed by re-coupling as sustainable development progresses. Nat. Sustain. 5, 452–459 (2022)
Hu, H. The distribution of population in China, with statistics and maps. Acta Geograph. Sin., 33–74 (in Chinese) (1935)
Chen, M., Gong, Y., Li, Y., Lu, D. & Zhang, H. Population distribution and urbanization on both sides of the Hu Huanyong Line: Answering the Premier’s question. J. Geogr. Sci. 26, 1593–1610 (2016)
Banwart, S.A., Nikolaidis, N.P., Zhu, Y.-G., Peacock, C.L. & Sparks, D.L. Soil functions: Connecting Earth's critical zone. Annu. Rev. Earth Planet. Sci. 47, 333–359 (2019)
Chen, Q., et al. A method to study antibiotic emission and fate for data-scarce rural catchments. Environ. Int. 127, 514–521 (2019)
Zeng, Q., Sun, J. & Zhu, L. Occurrence and distribution of antibiotics and resistance genes in greenhouse and open-field agricultural soils in China. Chemosphere 224, 900–909 (2019)
Huang, F., Chen, L., Zhang, C., Liu, F. & Li, H. Prioritization of antibiotic contaminants in China based on decennial national screening data and their persistence, bioaccumulation and toxicity. Sci. Total Environ. 806, 150636 (2022)
Hall, M.C., et al. Influence of setback distance on antibiotics and antibiotic resistance genes in runoff and soil following the land application of swine manure slurry. Environ. Sci. Technol. 54, 4800–4809 (2020)
Davis, J.G., Truman, C.C., Kim, S.C., Ascough, J.C. & Carlson, K. Antibiotic transport via runoff and soil loss. J. Environ. Qual. 35, 2250–2260 (2006)
Meyfroidt, P., et al. Ten facts about land systems for sustainability. Proc. Natl. Acad. Sci. USA 119, e2109217118 (2022)
Yi, X., Lin, C., Ong, E.J.L., Wang, M. & Zhou, Z. Occurrence and distribution of trace levels of antibiotics in surface waters and soils driven by non-point source pollution and anthropogenic pressure. Chemosphere 216, 213–223 (2019)
Bryan, B.A., et al. China’s response to a national land-system sustainability emergency. Nature 559, 193–204 (2018)
Lee, C., An, J., Lee, Y.S., Choi, K. & Kim, J.Y. Uncertainty-based concentration estimation of chlortetracycline antibiotics in swine farms and risk probability assessment for agricultural application of manure. J. Hazard. Mater. 402, 123763 (2021)
Gao, J. & O’Neill, B.C. Mapping global urban land for the 21st century with data-driven simulations and Shared Socioeconomic Pathways. Nat. Commun. 11, 2302 (2020)
Rasmussen, L.V., et al. Social-ecological outcomes of agricultural intensification. Nat. Sustain. 1, 275–282 (2018)
Kosnik, M.B., Hauschild, M.Z. & Fantke, P. Toward assessing absolute environmental sustainability of chemical pollution. Environ. Sci. Technol. 56, 4776–4787 (2022)
Riaz, L., et al. Fluoroquinolones (FQs) in the environment: A review on their abundance, sorption and toxicity in soil. Chemosphere 191, 704–720 (2018)
Huang, L., Bradshaw, K., Grosskleg, J. & Siciliano, S.D. Assessing space, time and remediation contribution to soil pollutant variation near the detection limit using hurdle models to account for large proportion of non-detectable results. Environ. Sci. Technol. 53, 6824–6833 (2019)
Chen, J., Ying, G.-G. & Deng, W.-J. Antibiotic residues in food: Extraction, analysis, and human health concerns. J. Agric. Food Chem. 67, 7569–7586 (2019)
Steffan, J.J., Brevik, E.C., Burgess, L.C. & Cerdà, A. The effect of soil on human health: an overview. Eur. J. Soil Sci. 69, 159–171 (2018)
Duarte, D.J., Oldenkamp, R. & Ragas, A.M.J. Modelling environmental antibiotic-resistance gene abundance: A meta-analysis. Sci. Total Environ. 659, 335–341 (2019)
Breiman, L. Random forests. Machine learning 45, 5–32 (2001)
Zhang, Q., et al. A dataset of distribution of antibiotic occurrence in solid environmental matrices in China. Sci. Data 9, 276 (2022)
Ashfaq, M., et al. Chapter 21 - Environmental risk assessment of antibiotics and AMR/ARGs. In: Antibiotics and Antimicrobial Resistance Genes in the Environment (ed^(eds Hashmi, M.Z.). Elsevier (2020).
Zhao, F., et al. Soil contamination with antibiotics in a typical peri-urban area in eastern China: Seasonal variation, risk assessment, and microbial responses. J. Environ. Sci. 79, 200–212 (2019)
Harris, I., Osborn, T.J., Jones, P. & Lister, D. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci. Data 7, 109 (2020)
LDAAC. Global 30 Arc-Second Elevation Data Set GTOPO30. Land Process Distributed Active Archive Center. (2004).
Liaw, A. Classification and Regression by randomForest. R News 2, 18–22 (2002)
Kuhn, M. Building Predictive Models in R Using the caret Package. J. Stat. Softw. 28, 1–26 (2008)
Haaf, D., Six, J. & Doetterl, S. Global patterns of geo-ecological controls on the response of soil respiration to warming. Nat. Clim. Change 11, 623–627 (2021)
Lehner, B., Verdin, K. & Jarvis, A. New Global Hydrography Derived From Spaceborne Elevation Data. Eos, Transactions American Geophysical Union 89, 93–94 (2008)
Venter, O., et al. Global terrestrial Human Footprint maps for 1993 and 2009. Sci. Data 3, 160067 (2016)
Barton, K. MuMIn: Multi-model Inference. available at https://CRAN.R-project.org/package=MuMIn, (2017)

There is NO Competing Interest.

SupplementaryMaterials.pdf

Download PDF

Version 1

posted

You are reading this latest preprint version

Human impacts on risk of antibiotics in soil differentially depend on land system across scales

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1