The Llobregat MAR system is located 15 km inland from the Mediterranean Sea, close to Barcelona city (Fig. 1). The aquifer thickness in the vicinity is 10–15 m, with alternating sands and gravels. Non-continuous fine-grained sediments are widely present . The distance from the bottom of the pond to the water table oscillated from 9 m (July 2014) to 7 m (March 2015) in the period under study.

Water entering is diverted from upstream of the river to a pre-sedimentation basin. After 2–4 days of residence time, it is diverted to an infiltration basin of 6500 m2. The infiltration capacity has been estimated at 1 m3 m-2 d-1 on average, and the local transmissivity of the aquifer is estimated as 14 000 m2 d-1 (unpublished).

In 2011, a reactive barrier was installed at the bottom of the infiltration basin to increase the organic load of the infiltration water, thus promoting biological processes through the soil and the vadose zone. The barrier was composed of organic compost (50 % in volume) mixed with sand and gravel. Small amounts of clay and iron oxides were added to foster adsorption and ion exchange. Previous studies demonstrated that the reactive barrier enhanced the removal of some emergent contaminants, such as sulfamethoxazole or caffeine . More information about the site and the performance of the reactive barrier can be found in .

There are six piezometers distributed in a 500 m transect across the study area (Fig. 1). P1, P3, P2, P5 and P10 are fully screened. P8 is a multilevel piezometer drilled at three different depths. Water from piezometer P1 represents background conditions (not affected by recharge).

Two recharge situations were compared to evaluate the effect of MAR on groundwater chemical signature. After 6 months of continuous recharge operation, a sampling campaign took place in July 2014 (wet campaign). Samples were collected from surface water in both basins and in the existing piezometers at different depths (from -5 to 3 m a.s.l.; see diamonds in Fig. 1). The second sampling campaign was performed in March 2015 after recharge had been discontinued for 4 months. In this case, groundwater was also sampled.

Water was analyzed for cations, anions (Cl-, NO3-, SO4-2, HCO3-), dissolved organic carbon (DOC) and total organic carbon (TOC). Analytical procedures are widely described in the Supplement.

In both campaigns, temperature and electrical-conductivity vertical profiles were mapped along the transect from data obtained at 50 cm intervals (MPS-D8, SEBA Hydrometrie).

Water samples were extracted from the pre-sedimentation and infiltration basins at three locations (entrance, middle and end) during recharge conditions (from now on, wet scenario). On the contrary, three soil samples were extracted at the same locations in the infiltration basin under non-recharge conditions (termed dry scenario). Soil samples were obtained from around 10 to 50 cm in depth. The sampling procedure for soil was done taking into account recommendations, especially regarding the variability of microbial communities along a field transect. Soil samples were taken by means of cores, individually disassembled and kept in a sterile bag. Groundwater samples were taken from -5 to 3 m a.s.l. depending on the piezometers (10 samples for wet scenario and 7 for dry), at the same location as samples for the hydrochemical characterization. All soil and groundwater samples were taken in duplicate, kept in sterile conditions and preserved in the dark at -4 ∘C until being taken to the laboratory for molecular analyses.

Protocols for molecular analyses of liquid and soil samples are thoroughly described in the Supplement.

Once the main microbial communities were characterized, three diversity indices were calculated. The first one is richness (S), defined as the proportional number of microbial species present in a sample, i.e., equal to the total number of bands; the other two, Shannon (H) and evenness (E), were calculated for each sample as follows: H=∑i=1Spiln(pi),E=HHmax, where Hmax equals lnS and pi is the relative intensity of each band of the sample. Values reported correspond to the average of the two replicas. The evenness index is proportional to the equitable distribution of bands in the gel. Shannon is an entropy index, reaching maximum (Hmax) for the most equally distributed band patterns and with the higher value of richness.

To complement the soil microbial community's characterization, particle size measurements of soil samples were taken according to the ASTM guidelines. The soil was sampled in the pre-sedimentation basin and at the entrance, middle and end of the infiltration basin. Soil sampling was performed close to the location where samples were taken for microbial analyses.

To study the impact of MAR on microbial community structure, recharge water from pre-sedimentation and infiltration basins, as well as soils, were analyzed. Figures 8 and 9 show the relative abundance of bacterial phylotypes at the taxonomical level of classes for surface water and soil samples. The results are displayed according to the distance to the recharge basin inlet.

Microbial richness in soil samples was controlled by water content. Non-recharge conditions had a primarily negative effect on the populations of Dehalococcoidia, Acidobacteria and Chlorobia, but favored the presence of Nitrospira, Cytophagia and Actinobacteria (Fig. 8). Shannon and evenness indices demonstrated that soils were more diverse under wet conditions than under dry ones (Table 1).

For surface water samples (Fig. 9), there was a decreasing gradient in community complexity along the ponds. Acidobacteria, Betaproteobacteria and Cyanobacteria were the main phylotypes present.

Groundwater is a quite stable aquatic environment as it is shown by the low variability in the hydrochemical data (Table S1). One could expect that microbial communities in groundwater should also display low variability that could be reflected in the diversity indices. In this way, piezometer P1, which is unaffected by recharge, showed stable diversity indices regardless of the sampling campaign. However, disturbances produced by recharge, evidenced by temperature and conductivity gradients below the pond (Figs. 2 and 3), favor the highest Shannon values for the moderately disturbed samples. As a result, the average of diversity indices remains constant between both scenarios but with a higher standard deviation during the wet scenario. These two facts combined suggest that perturbations caused by recharge influence much more the composition of microbial communities in groundwater than the natural variability of the background aquifer water between scenarios.

MAR is a passive treatment technique that can provide simultaneously oxic and anoxic conditions . This has wide implications for the potential biological removal of selected emerging contaminants, as each micropollutant is most efficiently removed under specific redox conditions . Some can even be degraded by co-metabolism, involving different redox states in the process . In this sense, MAR is an efficient remediation system. In addition, many sequenced phylotypes, such as Nitrospira sp., Stenotrophomonas sp. and Methylobacterium sp., have been associated with degradation capabilities . In short, the MAR microbial ecosystem studied in this work presents many more phylotypes than previous studies reported in groundwater systems , and thus MAR can be considered an efficient remediation system.

We further tested the intermediate disturbance hypothesis (IDH) for microbial communities in groundwater (Fig. 6) related to MAR activities. IDH was originally proposed for tropical rain forests and coral reefs and supports the idea that small perturbations create new access to resources for species which have overlapping niches, allowing their coexistence. This mechanism, known as a competition–colonization trade-off, can explain IDH in local communities, leading to an increase in diversity. However, when the degree of disturbance rises, only eurytolerant populations can survive and grow. Thus, an inverse correlation between diversity and the degree of disturbance (reflected in the temperature and conductivity profiles) was expected (see Table 1). Such correlations have also been reported in recharge wells and snowmelt-influenced aquifers .

In the Llobregat MAR system, the initial diversity in the microbial community increased with perturbation caused by recharge (Fig. 6), with maximum diversity associated with Type IV water, and was lowest for the most disturbed water (Type II). In ecological terms, Type III and Type IV waters represent different proportions of perturbation.

In the most altered groundwater zone (represented by P8 samples), Betaproteobacteria grew above 50 % of the relative abundance (Fig. 7). The main phylogenetic affiliation of this phylotype at the genus level is Vogesella. Strains of this genus are able to catabolize monosaccharides under aerobic conditions, but not under low-oxygen conditions. Furthermore, all Vogesella strains are denitrifiers . Indeed, P8, located below the pond, receives oxygen-rich water during the recharge process, driven by fingering in the vadose zone. Although dissolved oxygen was not measured in the present study, data from other campaigns confirm this behavior for oxygen in P8 samples (data not shown). Moreover, nitrate concentration in the surface water was low (Table S1), and thus most denitrification is expected to occur under the pond. recently confirmed that nitrate was consumed via the denitrification pathway under the infiltration pond in the Llobregat MAR system, supporting the idea that Vogesella sp. could be one of the genera involved in nitrate consumption. Likely, depending on the oxygen content, Vogesella sp. will adapt its metabolic function in favor of aerobic oxidation of organic matter or by means of denitrification, thus becoming a good indicator of highly disturbed MAR environments.

For Type III and Type IV waters, Dehalogenimonas sp., within the Dehalococcoidia class, is characteristic of medium-disturbance groundwater (Figs. 4 and 7). Dehalogenimonas sp. has been studied in recent years because some strains are associated with dechlorination in contaminated sites. This genus is strictly anaerobic and mesophilic, and some species can reductively dehalogenate polychlorinated aliphatic alkanes . As a result, recharge creates reducing conditions, likely indicating the existence of microzones or microsites , defined as local anoxic areas that coexist with fast-traveling oxygen-rich paths. Thus, microbial analysis can be used to unmask the apparent mishap of water samples that are oxic and display some typical anaerobic species. Moreover, some species of Dehalogenimonas can dechlorinate some trichloroethane isomers , a pollutant reported in the Llobregat lower valley at levels as high as 300 µg L-1 , thus opening the door for the development of enhanced remediation activities.

The evenness index is an indicator of the equity of a community and can be quite informative for observing perturbations to microbial communities. In the wet scenario, the lowest values of E were obtained for the samples most affected by recharge (Table 1), indicating that some species developed into predominant members of the microbial assemblage. Groundwater samples displayed the highest evenness values in the area less affected by recharge and in the dry scenario. In the latter case, values indicate the recovery of microbial communities from the disruption caused by recharge. In fact, P8(1) samples in the dry scenario were not fully consistent with this conceptual model, with a low evenness index and very low nitrate concentration. Furthermore, the presence of Methylotenera mobilis (Betaproteobacteria class) in both P8 sampling points was more than 40 %, on average, of the relative abundance. Methylotenera mobilis is a methylotroph species with denitrification abilities . These results suggest that P8 denitrification processes occur below the basin even when it is empty, indicating that 4 months is not enough time to revert back to natural conditions at this sampling point. This assumption is consistent with nitrate isotopic data presented in and is also in agreement with the study of in which biomass decay acted as an endogenous carbon source for respiration once the input carbon was reduced, maintaining denitrification rates.

We analyzed the heterogeneity of the microbial community structure in soil and surface water in terms of the distance to the infiltration basin entry point (Fig. 9 and Table 1). Patterns in surface water microbial composition are linked to sequential sedimentation processes as revealed by granulometric analyses of soil samples (Table S3). The result was that surface water became poorer in terms of the presence of microbial communities between the pre-sedimentation basin and the end of the infiltration basin. The main reason could be the decrease in solids suspended throughout the system due to the sequential decantation of particles and their attached biomass. Furthermore, surface water displays relatively higher values of Cyanobacteria and Acidobacteria classes compared to groundwater. Cyanobacteria constitute the largest, most diverse, and most widely distributed group of photosynthetic prokaryotes, which are capable of conducting nitrogen fixation . However, members of the phylum Acidobacteria are physiologically diverse and ubiquitous in soils, degrade a wide range of carbon sources (from substances with a wide range of complexity), and are capable of reducing nitrates and nitrites . This heterogeneous effect with distance to the entry point of the basin is also observed in the diversity indices, which lose diversity with distance and are inversely correlated to the proportion of fine particles. Similar behavior for richness correlated to soil texture was reported elsewhere .

Differences in the microbial communities in soils between the two basins were concentrated in the large organic matter content provided by the reactive barrier present in the latter, being a source for the growth of bacterial communities and enhanced diversity under recharge conditions. The role of the humidity on microbial diversity is also significant, as was previously reported in horizontal subsurface constructed wetlands . Furthermore, phylotype distribution changes among scenarios. Whereas Dehalococcoidia and Chlorobia classes appear in wet soils, Nitrospira, Cytophagia and Actinobacteria are favored under dry conditions.

The role of the reactive layer at the infiltration pond could be extrapolated as a system fed with a considerable organic carbon load. Laboratory experiments and constructed wetlands demonstrate that concentration of microbial activity and TOC degradation is concentrated in the first centimeters of the filter material in response to oxygen concentration vertical distribution. Although rapid oxygen depletion and consequent denitrification conditions have been evidenced in lab-scale MAR experiments , this effect may not happen rapidly under real infiltration conditions, where entrapped gas or fingering processes may provide higher oxygen concentrations than in lower-dimension systems (e.g., columns). Lab experiments are doubtlessly useful to elucidate the behavior of microbial communities under controlled conditions. However, sometimes it may be difficult to transfer conclusions obtained from lab samples to real sites.

Field studies are indeed realistic, but their transferability to other areas becomes challenging. Following an ecological argument, this study evidences that the intermediate disturbance hypothesis has been accomplished in the Llobregat MAR site. Therefore, we could expect the same behavior in other impacted areas under similar recharge conditions. Despite the novelty and transferability of this study being quite clear, future work is needed to keep evaluating the relationship between different ecological, microbial, hydrochemical and physical variables in different sites worldwide. An improvement of this multidisciplinary understanding of processes could be achieved by combining statistical techniques with process-based models. This would allow the direct extension of the results from one experiment to other sites.