A Spectrum of Preferential Flow Alters Solute Mobility in the Critical Zone

Preferential ow reduces water residence times and allows rapid transport of pollutants such as organic contaminants. Conventional understanding of solute transport implies that preferential ow reduces the inuence of soil matrix-solute interactions; however, this assumption lacks robust validation in the eld. To better understand how physicochemical properties affect solute transport across a range of preferential ow conditions, we applied deuterium-labeled rainfall to eld plots containing manure spiked with eight common antibiotics with a range of anity for the soil. We then quantied preferential ow and solute transport using 48 soil pore water samplers spread along a hillslope. Based on >700 measurements, our data showed that solute transport to lysimeters was similar – regardless of antibiotic anity for soil – when preferential ow represented less than 15% of the total water ow. When preferential ow exceeded 15%, however, concentrations were higher for compounds with relatively low anity for soil. These results suggest that bypassing water ow can select for compounds that are more easily released from the soil matrix, thus providing fundamental insight into how ow heterogeneity affects pollutant mobility in soils. Moreover, because these data do not fully align with existing solute transport theory, they may be useful for building improved process-based transport models. manure was applied to eld plots (200 x 150 cm) on the soil surface or injected to a depth of 10 cm (n = 3 plots per application method). After 7 days of rainfall suppression, we applied rainfall (7 cm h − 1 ) to these plots, plus additional 3 plots without treated manure to serve as controls for ow and transport. Rainfall simulations were labeled with deuterium to facilitate preferential ow quantication. Monitoring soil pore water isotope signatures and antibiotic concentrations in suction lysimeters across time (1 h before, 30 min into, and 1 h) and space (multiple locations and depths of 30 90 cm) allowed us to quantify solute transport with > 700 estimates of preferential ow. See Methods and Supporting information for more experimental details.

minimal, more homogenous ow through the soil matrix dominates, favoring transport of compounds with low a nity to soil, such as those with a low sorption coe cient (K d ). [30][31][32] The conventional assumption holds that the in uence of solute-matrix a nity decreases as ow becomes proportionally more preferential [33][34][35][36] such that the kinetics of rapid ow restrict sorption 37 or enhance desorption. 38 However, studies to date have mostly compared known preferential transport of solutes to more homogenous ow -either modeled based on advection-dispersion processes 39,40 or focused on speci c conditions such as frozen soils 41 -but have never directly assessed how organic contaminants of varying chemical properties become mobilized along a spectrum of ow heterogeneity. As a result, the conditions necessary to dampen versus amplify the in uence of compound physiochemical properties on solute transport are not well understood. The primary objectives of this study were to 1) quantify transport of eight veterinary antibiotics under different preferential ow conditions and 2) determine if preferential ow can eliminate the in uence of solute-soil a nity on transport of these solutes. This analysis is necessary to provide a fundamental understanding of how preferential ow alters contaminant mobility and build process-based transport models needed to manage water quality and thwart water resource degradation.
Here we explore the in uence of solute-matrix a nity across a range of preferential ow by applying simulated rainfall to eld plots containing manure spiked with eight common veterinary antibiotics (listed by decreasing relative a nity to the soil matrix): erythromycin (ERY), tylosin (TYL), tetracycline (TC), pirlimycin (PLY), chlortetracycline (CTC), oxytetracycline (OCT), sulfadimethazine (SDM), and sulfamethazine (SMZ). Veterinary antibiotics were chosen for 1) their environmental ubiquity, as in the U.S. up to 52 million kg per year are applied to soils in manure 42,43 and for 2) their wide range of a nity to soils. 44 Antibiotic-spiked manure was applied to eld plots (200 x 150 cm) on the soil surface or injected to a depth of 10 cm (n = 3 plots per application method). After 7 days of rainfall suppression, we applied rainfall (7 cm h − 1 ) to these plots, plus additional 3 plots without treated manure to serve as controls for ow and transport. Rainfall simulations were labeled with deuterium to facilitate preferential ow quanti cation. Monitoring soil pore water isotope signatures and antibiotic concentrations in suction lysimeters across time (1 h before, 30 min into, and 1 h) and space (multiple locations and depths of 30 and 90 cm) allowed us to quantify solute transport with > 700 estimates of preferential ow. See Methods and Supporting information for more experimental details.
We de ned antibiotic movement in terms of change in concentration, ΔC, from samples collected 0.5 h into and 1 h after rainfall versus pre-event (background) values from the same lysimeter. We deemed ΔC to be zero whenever veterinary antibiotic concentrations decreased from background or were nondetectable. At the same time, we considered ow to be partitioned into two distinct hydrological domains assuming faster advection through preferential pathways (e.g., root channels and macropores) versus slower ow through the soil matrix via combined advection and dispersion mechanisms. Following the conceptual framework provided by Stumpp, et al. 45 , the fractional contribution of preferential ow was calculated by where sampled deuterium concentrations, D t (t), were used in a two-member mixing model that separated rainfall moving through preferential ow paths, D PF (t), from pre-event soil matrix water, D MF (t) (see Methods for full derivation). We estimate that the average of 7 cm of rainfall in ltrated in this experiment (Table S1) would have replaced ~ 20 cm of storage via pure advection. This calculation suggests that a homogenous wetting front would not have reached our most shallow pore-water samplers (30 cm) and that the sampled water was derived from some combination of rainwater bypassing the soil matrix and pre-event matrix storage. As a result, we consider the mixing model to be suitable for quantifying preferential ow during the simulated rainfall experiment.
A total of 153 of the 768 measurements (20%) resulted in zero or negative f PF values, which we considered to represent entirely matrix-derived water (f PF = 0) in subsequent analyses. Though event water was applied at a constant rainfall intensity (7 cm h − 1 ) and in ltrated in similar rates between plots (Table S1), simulated rainfall produced nearly three orders of magnitude of variation in preferential ow (f PF from 0.0017 to 0.6; Figure S1). The range of positive ΔC values extended nearly four orders of magnitude, from 0.006 to 3.85 µg L − 1 ( Figure S2), with probability of detection highest in the low range of preferential ow ( Figure S3). These numerous point estimates of preferential ow in space (i.e., different lysimeter depths and random positions) and time (i.e., during and after rainfall) enabled analysis of solute mobility under a spectrum of ow heterogeneity.
A frequency analysis of samples with detectable changes in antibiotic concentration (ΔC > 0) showed clustering in three distinct ranges of preferential ow: 0 < f PF ≤ 0.15, 0.15 < f PF ≤ 0.35, and 0.35 < f PF ≤ 0.61 (Fig. 1a). In general, solutes with high relative a nity for the soil matrix (e.g., TYL, TC) were most frequently detected under low preferential ow conditions (i.e., f PF ≤ 0.15). The relatively low-a nity sulfonamides (SDM and SMZ) had a more uniform distribution across the range of preferential ow.
While the high-a nity solutes ERY and PLY had similar distributions as the sulfonamides, it should be noted that they were detected less often (N = 11 for ERY and 19 for PLY versus N = 22 for SDM and 35 for SMZ). Further, the sulfonamides were always detected under high preferential ow conditions (e.g., f PF > 0.4), whereas ERY and PLY continued to have non-detects (i.e., ΔC = 0) in that range ( Figure S4). Altogether, these results suggest that bypass ow preferentially mobilizes some solutes over others, with relative a nity to the matrix acting as an important factor in this process.
After binning data into the three preferential ow ranges (0 < f PF ≤ 0.15, 0.15 < f PF ≤ 0.35, and 0.35 < f PF ≤ 0.61), the compounds with the greatest contrast in relative a nity to soil (e.g., the high-a nity macrolides TYL and ERY versus the low-a nity sulfonamides SDM and SMZ) were similar in ΔC for the low range of preferential ow, but diverged with increasing bypass ow (Fig. 1b). For example, when f PF was < 0.15, TYL and SMZ had nearly identical ΔC values. However, when preferential ow exceeded 0.35, ΔC was more than an order of magnitude higher for SDM compared to ERY. This nding suggests that the in uence of solute-matrix a nity on transport was weakest when bypass ow was minimal.
Here we note that the antibiotic PLY, which had a relatively moderate a nity for soil, produced the highest ΔC in drainage in the high range of f PF (Figs. 1b). However, a previous antibiotic transport study conducted in the same eld site reported PLY as being highly mobile with 50x more PLY transported in runoff compared to the sulfonamide SMZ 46 . This result suggests that PLY sorption to the A p soil sample used for K d determination may not have been representative of the entire eld, or else that our ranking was accurate and the high apparent mobility of PLY seen in eld-runoff studies 46,47 re ects the in uence of other controls on transport (such as colloidal transport, as discussed in the Supporting Information).
Some numerical simulations 33,48 and one recent column study 41 have also suggested that solutes with moderate a nity for soil may be most susceptible to preferential ow. Though the underlying mechanisms are not yet clear, we speculate that these compounds may have high enough K d to be sorbed throughout the soil medium, yet soluble enough to be partitioned or displaced into local bypass ow. Thus, when f PF approached ~ 0.5 (i.e., roughly equal matrix and preferential contributions to ow), compounds with moderate relative a nity for soil could be selected in higher proportions relative to other antibiotics.
Solute transport to our lysimeters also appears to have been most susceptible to preferential ow (i.e., ΔC/f PF was highest) when f PF was < 0.15 (Fig. 2). The in uence of preferential ow on the magnitude of ΔC was therefore dampened when solute-matrix a nity became more in uential, indicating a fundamental shift in solute and ow partitioning. For example, in a situation where ΔC linearly increased across the range of f PF values (red dashed line t to raw data in Fig. 2), the antibiotic detection in drainage would respond similarly (i.e., nearly constant ΔC/f PF ) across the spectrum of preferential ow.
Instead ΔC was an order of magnitude more susceptible to preferential ow at the lowest versus highest f PF values. These data thus further illustrate that solute responses to bypass ow differ along the spectrum of preferential ow, with low f PF conditions causing non-selective transport and high f PF conditions causing selective transport at the length scale of the lysimeters.
These differences in transport behavior can be explained by both the amount of preferential ow and the ability of this bypass ow to access antibiotics. For example, 7 days of rainfall suppression would likely have been su cient time for compounds to diffuse into the soil matrix 13,50−53 and for sorption equilibrium to occur. 46,54 Consequently, solute transport in plots spiked with antibiotics was nearly identical to control plots ( Figure S5), suggesting that these compounds may be stored in the soil matrix from previous applications. Therefore, when drainage water was less preferential (i.e., f PF < 0.15), the likelihood of sampling all compounds was higher ( Fig. 1a and S3) as most of the drainage water originated in the matrix. In ltrating water may have mixed with a greater volume of pre-event storage before triggering preferential ow events with trace levels of antibiotics, allowing for compounds strongly sorbed to the soil matrix (e.g., high relative a nity) and compounds weakly bound to macropore walls (e.g., low relative a nity) to be transported in similar proportions. In contrast, higher proportions of preferential ow would have excluded ow through the matrix where much of the compounds resided, 52,55−58 causing the fast preferential ow domain to become more distinct from the slow matrix ow domain [59][60][61] and in ltrating water to select for compounds with a higher a nity for the aqueous phase.
We additionally note that initiation of macropore ow often requires contributions from the soil matrix, 22 with the potential to dilute or displace the tracer signal in preferential ow paths. 24,62 This process can lead to underestimations of event water contributions to preferential ow. 24,25,62 Our method may not have distinguished these preferential ow scenarios from matrix water; rather, our analyses were intentionally focused on preferential ow paths that originated at or near the soil surface. Under the assumption that antibiotics were near the surface at the time of rainfall (max manure injection depth of 10 cm), our f PF calculations would have detected fast-owing event water contributions with the greatest potential to rapidly transport these solutes to depth. Altogether, our f PF estimates should provide a useful representation of ow heterogeneity and identify source contribution of water and solutes in drainage.
Further, we encourage the use of alternative preferential ow detection methods, 5 under similar experimental conditions, to determine the relevance of this range of detected preferential ow-and its bearing on relative solute transport-in other heterogeneous systems.
In this study we treated preferential ow as an explanatory variable, which revealed that conventional transport phenomena may hinge on the degree of ow heterogeneity. This distinction appears to be unprecedented in previous studies, in part, because none have considered how the magnitude of preferential ow alters the in uence of solute-matrix a nity in soils. As a result, these ndings contradict conventional understanding of solute transport, where the in uence of compound properties was thought to be signi cantly reduced with bypass ow. 33,63,64 To further explore this result, we used the conventional dual permeability model framework of Gerke and Van Genuchten 65 with the HYDRUS 1D 66 numerical platform to simulate analogous conditions to our experimental design (See Supplemental Information for details). Modeling results clearly predict that the difference between solutes of high and low relative a nity would decrease as the fraction of preferential ow increased (See Figure S6, and Tables S2, S3, and S4). Rather, our data indicated that when preferential ow intensi ed, ΔC in drainage became more in uenced by the physiochemical interactions with the medium rather than just the medium itself (Fig. 3). Alternatively, if rainfall was applied much closer to the time of manure application one would expect the opposite trend: solute-matrix interactions might govern solute transport at low values of f PF but become less important as bypass ow mobilizes compounds with high and low relative a nity to soil equally. Using a similar experimental design, Le, et al. 46 detected comparable losses to runoff for four antibiotics of varying mobility when rainfall occurred just 2 hours after manure application, yet losses differed by an order of magnitude when manure was undisturbed for 3 days. More directly, the timing of precipitation appears to be an important factor controlling compound behavior in the presence of preferential ow, due to sorption kinetics and physical partitioning of the compounds below-ground. We expect that this selective transport phenomenon will be observable in many other systems, as many farmers select periods without forecasted rain for manure application and organic contaminants in soil often reach sorption equilibrium within a few days. 67,68 Implications It has long been known that solute diffusion and sorption equilibrium within the soil matrix may limit subsequent transport through preferential ow paths when rainfall timing is lagged relative to chemical application, 52,53,56,58 with strongly sorbing substances most often affected. 57 Similarly, preferential ow has been shown to non-selectively transport a range of compounds for decades. 13,34,39,40,69 However, by simulating a range of preferential ow, we identi ed conditions necessary to dampen versus amplify the in uence of compound physiochemical properties on solute transport, thus providing novel insight into subsurface water and solute partitioning. Our results suggest that under eld-relevant scenarios the in uence of solute-chemical properties appear damped below ~15% preferential ow, but ampli ed at higher contributions of event water. Mechanistically, this means that fast ow paths may preferentially select for solutes with low matrix a nity. Practically, this means that soil and solute physicochemical properties may become more, not less, in uential as the magnitude of preferential ow increases. Given the ubiquity of preferential ow observations, 8,9,19,70,71 it is important to develop better strategies for retaining mobile chemicals within the soil pro le.
Here we recognize that this study only encompassed one particular set of eld conditions. For example, by simulating just one storm with a constant rainfall intensity and timing we may have missed the opportunity to study precipitation-driven transport shortly after manure application, where the in uence of compound properties is likely greatest. 56,57 Simulating rainfall 7 days after manure application may have, rather, increased the likelihood of selective transport through macropores. 57 Additionally, our suction lysimeters would not have necessarily intercepted all potential preferential ow paths. However, we posit that these results can be considered more broadly applicable to instances of non-equilibrium ow and transport because 1) our storm produced preferential ow estimates spanning 3 orders of magnitude, 2) the relative a nity of our eight antibiotics differed by up to two orders of magnitude, and 3) we collected >700 lysimeter measurements through time and space. In this way, we were able to detect both selective and non-selective transport. At the same time, our study quanti ed the response of antibiotics with wideranging chemical properties under varying levels of preferential ow, thereby emphasizing both the novelty and potential for broader transferability of these results.
Altogether, these results suggest that it may be necessary to re-evaluate common assumptions of solute transport under preferential ow conditions. Speci cally, our ndings indicate that: 1) the a nity between a solute and the soil matrix has little bearing on large-scale contaminant transport under conditions of low preferential ow, and 2) traditional reactive transport models (e.g., single domain ow and single sorption site) can better describe solute movement as the proportion of ow moving preferentially increases.

Field Study Site and Preparation
The eld experiment was conducted in the spring of 2018 on a no-till agricultural eld in Whitethorne, Virginia. The eld had a 9 to 11% slope and was underlain by two loam-textured soil series: Braddock and Unison (Typic Hapludults) with moderate soil structure. Soil physiochemical and hydraulic properties methods are described in Table S3. A total of nine randomly placed rainfall simulation plots were installed in the eld. Each plot consisted of a 200 x 150 cm steel frame inserted 10 cm into the soil surface, with adjacent 40 cm x 200 cm buffer strips maintained outside of the frame for installation of soil pore-water samplers. A steel pan was tted to each frame, sealed for runoff collection, and piped down-gradient to a container for storage and quanti cation ( Figure S7). Weed growth was then suppressed in all plots with glyphosate. Plots were differentiated into two treatments whereby manure was homogenously broadcasted on the soil surface (surface application; n = 3 plots) or injected belowground into two 5 cm wide x 10 cm deep slits placed perpendicular to the slope and spanning the width of the plot frame and buffer strip (subsurface injection; n = 3 plots). The three remaining plots were used as controls by avoiding manure application and input of antibiotics. However, we detected some residual antibiotics in the control plots, which likely remained from manure applications in previous years ( Figure  S8). The presence of residue antibiotics in the control plots provided us with the opportunity to assess the mobility of compounds under short (up to 7 d) and long term (e.g., greater than 6 months) equilibration with the soil matrix.
Prior to manure application, we installed a series of suction lysimeters (200 kPa ceramic cups; Soil Moisture Equipment Corp., Santa Barbara, CA) in the plot buffer strips to sample veterinary antibiotic transport in the subsurface. These buffer strips received the same amount of rainfall and manure treatment yet were located outside of the metal plot frames. Soil pore water samples were withdrawn from two randomly positioned lysimeters in both the Bt1 (30 cm) and Bt2 horizons (90 cm) to detect vertical movement of antibiotics in surface application plots, with two probes per depth making four probes per plot ( Figure S7 and S9). This same installation scheme was also adopted for control plots. In subsurface injection plots a series of nested (30 and 90 cm probes) lysimeters were installed both within and 25 cm down-gradient of the injection slit to detect vertical and lateral transport of antibiotics, with two probes per depth resulting in eight probes per plot.
A liquid slurry of dairy manure (5% solid content) was spiked with eight commonly used antibiotics: two macrolides, erythromycin and tylosin; two sulfonamides, sulfamethazine and sulfadimethoxine; three tetracyclines, oxytetracycline, chlortetracycline, and tetracycline; and one lincosamide, pirlimycin. A target concentration of 500 µg L -1 was used for all antibiotics. The spiked dairy manure slurry was then applied to each plot 7 days prior to rainfall simulations at a rate of 56 Mg wet mass ha -1 . 46 If natural rainfall occurred within the 7-day equilibration period, the plots were covered with plastic tarps to prevent unintentional water input to the plots. 46

Assessment of Antibiotic Relative A nity to Soil
We performed a simple solute partitioning test in the laboratory to determine the linear sorption coe cient, K d , for each of the eight antibiotics with soil from the eld site (details in the Supplemental Information). Under this framework, each antibiotic was assigned a rank from 1 (highest a nity) to 8 (lowest a nity) based on measured K d values (Table S5). The antibiotics ERY and TYL were not detectable in the supernatant during the test, so were given respective rankings of 1 and 2. Additionally, we used the USEPA's BIOWIN model from the EPI (estimation program interface) Suite tool 72 to estimate dissipation half-lives of each compound in soil following the methods described in Chen, et al. 43 . Using these half-lives we projected that less than 10% of the originally applied antibiotic mass would have degraded during our 7-day experiments, so we therefore assumed that decay played a minor role during transport.

Field Rainfall Simulations and Water Sampling
After the 7-day equilibration period, we conducted rainfall simulations using deuterium-labeled well water to trace mobile in ltrating water and detect preferential ow contributions to pore water signature. The rainfall simulator (240 cm x 300 cm) followed the original design of Humphry, et al. 73 , which has been adopted as standard protocol for the national research project for simulated rainfall-surface runoff studies 74 because it provides constant droplet size and velocity between locations and studies. We conducted the rainfall simulations with the SERA-17 standard intensity of 7 cm h -1 , with rainfall continuing on each plot until the collection containers received 30 min of continuous runoff. Rainwater was isotopically labeled using a Dealglad venturi injector (9.0 x 5.5 x 5.5 cm; Shandong Jiujin Plastic Products Co., Shandong, China) tted to the sprinkler inlet. This system dispensed an enriched deuterium solution into the well water at a desired ratio of ~ 4:100 (deuterium-spiked water: well water). Discrete pore water samples were taken from all lysimeters by applying 60 kPa of suction for 10 min, with samples collected 1 h before the simulation, 0.5 h into the simulation, and 1 hour after the simulation ( Figure S9). All liquid samples were analyzed for 2 H via cavity ring down spectroscopy (Model L1102-i, Picarro, Santa Clara, CA) and for all eight antibiotics via HPLC MS/MS, as detailed in the Supporting Information. To understand how these preferential ow estimates affected the transport of our eight veterinary antibiotics with a spectrum of relative a nity for soil, we quanti ed the change in concentration from lysimeter samples collected before versus during and after simulation (ΔC) as a function of f PF .
Here we note some potential constraints of using suction cup sampling to represent soil pore water. For example, suction lysimeters often have a smaller volume-of-in uence compared to alternative pore water samplers, [75][76][77] reducing the likelihood of intercepting every preferential ow path below the plots. Suction cups can also have biased representation of water in larger-more "mobile"-pores. [78][79][80] We note that, though matrix and macropore waters can resist mixing during extreme rainfall, 81 complete mixing between pores can occur within days. [81][82][83] Thus, point measurements from our samplers (after 7 days of rainfall exclusion and equilibration) likely yielded representative samples of pre-event water from the matrix and labeled event water from mobile water in maropores, while capturing a wide range of stable isotope signatures.

Preferential Flow Analysis
We considered ow to be partitioned into two distinct hydrological domains: faster advection through preferential pathways (e.g., root channels and macropores) and slower ow through the soil matrix via combined advection and dispersion mechanisms. Following the conceptual framework provided by Stumpp,et al. 45 the isotope mass balance can be described as: Component. Assuming that preferential ow pathways translate to rainfall inputs during each sampling period, we consider the rainfall isotope signal to be equivalent to the preferential ow signal in the outlet 45,84 : We also note that mass transfer between the slow owing matrix water is implicitly considered in the mixing model. For example, we can consider a scenario where event water in ltrates into the soil matrix and spills into a preferential ow path yielding an f PF value of 0.50. Because event water reached the outlet before the wetting front it must have required preferential transport and thus 50% of the total water out ow is deemed preferential; with the remainder derived from pre-event matrix water.

Reactive Transport and Experimental Perspective
By applying labeled rainfall simulations to a heterogeneous no-till soil containing manure spiked with 8 antibiotics, we were able to quantify the amount of preferential ow in lysimeter drainage and assess the in uence of compound properties on solute transport using the compounds' wide range of relative a nity to the matrix. Additionally, because 1) the mass of antibiotics applied in manure was consistent between compounds and manure treatment (surface application versus subsurface injection), 2) estimated halflives (described above) suggest that degradation was a minimal over the 7-day equilibration period, and 3) we were not concerned with metabolites of these antibiotics, our analysis did not require the explicit use of reactive transport models.

Statistical Analysis and Data Processing
We used a two-way analysis of covariance (ANCOVA) to statistically compare the slope of lines tted to log-transformed ΔC as function of preferential ow (f PF > 0 and ΔC > 0) by manure application treatment (i.e., subsurface injection, surface application, and control plots with only levels of antibiotics) and lysimeter depth (30 cm vs 90 cm). Speci cally, this allowed us to identify the statistical signi cance of manure treatment on ΔC across the observed range of the covariate (preferential estimates) while also testing for the interaction of depth on this relationship. The log-transformed data were found to meet ANOVA assumptions of normality (via normal-quantile plots) and homogeneity of variances (via Fligner's test). We used R version 3.5.2 49 to conduct all statistical analyses with α = 0.05. We found no signi cant difference between the slope of lines tted to ΔC data across the range of f PF for all treatments, and no signi cant in uence of lysimeter depth on this relationship (see Supplemental Information and Figure  S10). Thus, we compiled all data together for subsequent analyses without separating treatment or depth.

Declarations Supporting Information
This document contains information about soil characteristics, analytical procedures, eld plot schematics, numerical simulations, and additional resources related background detection of antibiotics in soil.   Table S2): red indicates the compound with the lowest a nity (SMZ) and black indicates the compound with highest a nity (ERY) to soil. Lines track SMZ and ERY. R v3.5.2 was used to plot this gure.49 Figure 2 Solute susceptibility to preferential ow (ΔC /fPF) across the detected range of preferential ow. The red dashed line depicts a linear t to raw (not log-transformed) data using all antibiotics (ΔC = 0 and fPF = 0 excluded). The linear ts indicate a possible condition where antibiotics have constant susceptibility to leaching regardless of the amount preferential ow. Note that the y-axis in the inset gure has a logarithmic scale. R v3.5.2 was used to plot this gure.49 Different subsurface partitioning scenarios of solutes (dots) with high (black) and low (red) relative a nity to soil. Hypothetical solute concentration pro les (C vs x) are expressed at arbitrary locations spanning macropores surrounding a portion of the soil matrix. The top panel illustrates how both solutes would behave if rainfall simulations were conducted on the same day as antibiotic-spiked manure was applied. Compounds would have limited time to in ltrate into the soil matrix and come into sorption equilibrium, and high amounts of bypass ow through macropores could sample both compounds