Responses of microbial activity in hyporheic pore water to biogeochemical changes in a drying headwater stream

1. Microbial heterotrophic activity is a major driver of nutrient and organic matter processing in the hyporheic zone of headwater streams. Additionally, the hypor‐ heic zone might provide refuge for microbes when surface flow ceases during drought events. 2. We investigated chemical (organic and inorganic nutrients)


| INTRODUC TI ON
From a hydrological perspective, the hyporheic zone is defined as the sediments, where stream water enters the subsurface and returns to the stream channel at the reach scale (Wondzell & Gooseff, 2013). From an ecological perspective, the hyporheic zone is a biogeochemical hot spot with high turnover rates of organic matter and inorganic nutrients, because most of microbial heterotrophic activity occurs in this compartment (Boulton, Findlay, Marmonier, Stanley, & Valett, 1998;Danczak et al., 2016;McClain et al., 2003).
For instance, hyporheic denitrification can comprise up to twice the rate of whole stream (benthic and pelagic) denitrification (Harvey, Böhlke, Voytek, Scott, & Tobias, 2013) and the biggest portion of dissolved organic matter (DOM) is metabolised there (Naegeli & Uehlinger, 1997;Rasilo, Hutchins, Ruiz-González, & del Giorgio, 2017). Principal drivers of the rates and types of metabolic processes occurring in the hyporheic zone are hydrological exchange and discharge, because these drivers determine the length and water residence time of hyporheic flow paths (Boulton, Datry, Kasahara, Mutz, & Stanford, 2010). Similarly, various studies report differences in dissolved oxygen concentration and nitrogen removal between upand downwelling locations, with downwelling sites as hot spots of nitrification (Edwardson, Bowden, Dahm, & Morrice, 2003;Storey, Williams, & Fulthorpe, 2004;Triska, Duff, & Avanzino, 1990). The physicochemical gradients along hyporheic flow paths are steep and, at the same time, disturbances at the surface are buffered (Krause et al., 2011). Hence, the hyporheic zone of intermittent streams might serve as a refuge for microbes, when surface flow ceases in the course of annual summer droughts (Romaní et al., 2013). In the light of today's pressures on fluvial ecosystems that enclose water abstraction and changes in precipitation regime due to global warming (Pachauri et al., 2014), it is imperative to understand hydrological and biogeochemical constraints on microbial activity across the surface-hyporheic interface.
Microbial activity has been found to be related to DOM content of the sediment (Eiler, Langenheder, Bertilsson, & Tranvik, 2003;Fischer, Sachse, Steinberg, & Pusch, 2002) whereby the source and lability of DOM play a key role for microbial metabolism (Chafiq, Gibert, & Claret, 1999;Hall & Tank, 2003). Leachates from microbial assemblages provides an autochthonous DOM source that is assumed to be even more labile for microbes (Anesio, Granéli, Aiken, Kieber, & Mopper, 2005). Allochthonous DOM is typically considered resistant to microbial metabolism. However, in highly oligotrophic systems the quantity of allochthonous DOM has shown to drive shifts in community composition (Eiler et al., 2003). By contrast, autochthonous DOM has been found to affect community composition only transiently (Wagner et al., 2014). Additionally, the mixture of fresh labile and accumulated refractory DOM that naturally occurs in aquatic ecosystems could accelerate bacterial growth and bacterial DOM removal (Farjalla et al., 2009).
So far, the responses of microbial activity to specific environmental conditions, such as desiccation, oxygen distribution or temperature, have been mainly investigated under laboratory conditions in sand filtration columns or in mesocosm experiments with streambed sediment Freixa, Rubol, et al., 2016;Liu et al., 2017;Perujo, Sanchez-Vila, Proia, & Romaní, 2017). While these experiments target only one or two of the physical or chemical factors that potentially drive microbial activity, we expected that the microbial activity in streams is driven by a whole set of environmental conditions (Arce, del Mar Sánchez-Montoya, Rosario Vidal-Abarca, Suárez, & Gómez, 2014;Zhu & Dittrich, 2016). These environmental conditions are even more heterogeneous during the drying period of an intermittent stream (Febria, Beddoes, Fulthorpe, & Williams, 2012). This set of microhabitats at the surface and subsurface is subjected to rapid biogeochemical changes that determine microbial activity (Lake, 2003;Vervier, Dobson, & Pinay, 1993). For instance, the DOM composition and inorganic nutrient availability undergo severe changes during the drying of an intermittent stream (Gómez, García, Vidal-Abarca, & Suárez, 2009;Vázquez, Amalfitano, Fazi, & Butturini, 2011). Based on prior research carried out at the intermittent study site, we had selected a reach where the hyporheic zone was confined by impermeable bed rock and, therefore, impeding groundwater mixing (Harjung, Sabater, & Butturini, 2017).
This situation excluded groundwater as a potential end-member and offered the possibility to investigate the hyporheic zone as a biogeochemical reactor driving dissolved organic carbon (DOC) and inorganic nutrient retention and release, as well as DOM quality changes.
In the present study, we linked chemical changes to the bacterial cell concentration, live-dead ratio and extracellular enzyme activities. We expected that the microbes will survive and might even proliferate in the hyporheic zone, as the hyporheic zone represents a refuge when surface flow ceases. The living bacteria accumulated at the hyporheic pore water may maintain key functions for nutrient and carbon cycling that are reduced at the drying surface. We focused on microbial activity in the interstitial pore water rather than attached to the sediment, because in the water, the responses to environmental changes will be observed immediately (Febria et al., 2012). The two specific objectives were: (1) to compare microbial activity and chemical parameters in surface and hyporheic pore water; and (2) to explore temporal variability of microbial activity along hyporheic flow paths and its relationship to increasing water residence times and changes in nutrient and energy availability. Our hypothesis was that DOM transformation is limited by availability of oxygen and nutrients when water progressively vanishes during the drought period. We further hypothesised that as surface flow ceases, microbiota survive in the hyporheic zone, where moisture permits maintenance of the microbial activity.

| Study site and sampling strategy
The Fuirosos stream is an intermittent headwater stream in the north-east of the Iberian Peninsula, with an underlying catchment geology comprised of leucogranite and granodiorit, draining off into the Tordera river (Vazquez et al., 2013). The mean annual temperatures range from 3°C in January to 24°C in August (Butturini, Bernal, Sabater, & Sabater, 2002), and the average annual mean precipitation is exceeded by the average annual potential evapotranspiration (Medici et al., 2008). Baseflow ranges from 0 to 20 L/s (Butturini et al., 2003). The studied reach, located at the valley bottom of the Fuirosos catchment between 160 and 165 m above sea level (latitude 41° 42′23″-28″, longitude 2° 34′81″-86″) was chosen because the exposure to the surface of the bedrock interrupts the hyporheic zone that is composed of alluvial gravel (2-5 cm) with sand and fine sand fractions (for more information see Artigas, Romaní, Gaudes, Muñoz, & Sabater, 2009). The exposed bedrock channel is 63 m long, flanked by shallow sandy sediments and then covered again forming a channel of alluvial sediments of approximately 1-2 m depth.
This sediment channel is supplied by surface water and during rainy periods by hillslope runoff but is cut off from groundwater supply by the bedrock formation. The hyporheic connectivity is restricted to surface flow by the uplift of the bedrock acting as a natural barrier.
Due to the impermeability of the bedrock channel surface water is still captured in small pools (5-7 m 3 ) and in the hyporheic zone, even when there is no surface flow present above the alluvial sediments.
The riparian vegetation along the study reach is not dense and hardly causing light limitation.
In a previous study, electrical conductivity and stable water isotopes were used to separate the hydrological phases and investigate the hydrological connectivity between the sampling locations (Harjung et al., 2017). We performed five samplings between June and September 2014 to follow the drying period. The first sampling (19 June) represented the pre-drought phase with a discharge of 6 L/s and the second sampling (3 July) exemplified the contraction phase with a discharge of 1.4 L/s. The contraction phase started when the water level in the pool dropped and showed a continuous decrease (Figure 1). The approximated water residence times for these two phases were estimated from salt slug injections, taking the time that electrical conductivity needed to recover to background values at the measuring point downstream. This tail of the salt tracer curve is indicative for the water volume having passed through the hyporheic zone. It was not possible to achieve a similar estimation for the later samplings. However, the absence of diurnal temperature cycles in the hyporheic zone downstream indicated that water residence time exceeded 24 hr. The reason for this is that the longer the water travels through the hyporheic zone, the more the diurnal temperature cycles of the surface water will be dampened. The third sampling on 17 July, with a discharge of 0.07 L/s, represented the transition phase from contraction to fragmentation. The fourth sampling (28 July with a discharge of 0.05 L/s) was performed during the fragmentation phase, which is characterised by the absence of surface water at the permeable streambed upstream and the hydrological disconnection of the hyporheic zone up-and downstream the bedrock. Finally, the fifth sampling (3 September with a discharge of 0 L/s) was performed during the dry phase, when the surface water was absent along the whole reach, with the exception of the pools in the impermeable zones. The interstitial pore water of the hyporheic zone was pumped with a peristaltic pump from PVC tubes that were perforated over 30 cm on the bottom and were installed at a depth of 50 cm in the hyporheic zone. The sampling locations included an upstream well (HZ up ) two lateral wells (HZ lat1 , HZ lat2 ) and two wells downstream of the pool (HZ inf , HZ dw , Figure 1). The well F I G U R E 1 Scheme of stream reach, where the top panel is an overview of the complete sampled reach together with the legend for the subsurface type on the right and the individual panels on the bottom represent the magnification of geomorphologic situation of the sampling locations. The coloured and differently dashed lines indicate the approximate water level of each phase. Panels a and d show the longitudinal view, while panels b and c present cross-sectional views [Colour figure can be viewed at wileyonlinelibrary.com] HZ up was located upstream of the bedrock and was characterised by downwelling during flow conditions. The two wells at the lateral of the main channel were located in the sandy sediments flanking the bedrock channel. HZ lat1 was located at the riparian zone next to the bedrock structure on the left side receiving water from the hillslope, and HZ lat2 was located on the right lateral receiving mainly surface water from the bedrock channel. The most downstream of the pools mentioned previously served as a reference for the water entering the hyporheic zone channel. HZ inf refers to the well directly downstream of the pool, where the water of the pool infiltrated into the hyporheic zone. The well HZ dw was located 25 m downstream of the pool and was subjected to upwelling.
Dissolved oxygen was measured using an YSI 20 Pro oxygen sensor probe inside the wells immediately after pumping and was deployed in the pool and HZ dw to obtain continuous dissolved oxygen and temperature measurements. Surface and interstitial water samples for microbiological analyses were either incubated in the field for the measurement of extracellular enzyme activities, or placed in sterile vials and fixed for bacterial cell concentration and viability taking three replicates at each location (see below). Both surface and interstitial pore water samples for chemical analyses were filtered with ashed GF/F filters 0.7μm nominal pore size, and then electrical conductivity (WTW Cond 3310 Conductivity meter) and pH (Thermo Scientific Orion Star A121 pH meter) were measured. Samples for laboratory analysis were collected in pre-washed (MilliQ) polyethylene bottles (one sample per location and sampling).
The samples for inorganic nutrient concentration and DOM optical properties were filtered through 0.2μm nylon filters, and the samples for DOC analysis were acidified with 10% HCl. All samples were stored at 4°C temperature in the dark and analysed within a week.
After a 1-hr incubation, glycine buffer (4 ml, pH 10.4) was added at each sample in order to stop the reaction and maximise MUF and AMC fluorescence. Samples were kept cold and transported to the laboratory for fluorescence readings. Aliquots (350 μl) of each sample were placed into wells of a 96-well black plate (Greiner bio-one). Fluorescence was measured at an excitation/emission (ex/em) wavelength of 365/455 (MUF fluorescence) and 364/445 (AMC fluorescence) in a fluorimeter plate reader (Tecan, infinite M200 Pro). To determine extracellular enzyme activities, MUF and AMC standards were prepared and measured for their fluorescence. Results are given in nmol MUF ml −1 hr −1 or nmol AMC ml −1 hr −1 .
To calculate bacterial cell concentration, an internal standard (10 μl of beads solution 10 6 beads/ml, Fisher, 1.0 μm) was added to each sample. Bacterial cell concentration was measured by flow cytometry (FACSCalibur, Becton-Dickinson). Results are reported as bacteria cells/ml. Bacterial viability was determined by microscope counting. Water samples (2 ml) were stained (3 μl) using the Live/Dead BacLight Bacterial Viability Kit, which includes Syto 9 and propidium iodide stains. Samples were incubated for 15 min in the dark. Samples were filtrated (black polycarbonate filter, 0.2 μm), and each filter was prepared for counting in an epifluorescence microscope (Nikon E 600 at 1,000× magnification). Syto 9 penetrates all bacterial membranes and stains the cells fluorescent green, while propidium iodide only penetrates cells with damaged membranes, and the combination of the two stains produces red fluorescing cells (Boulos, Prévost, Barbeau, Coallier, & Desjardins, 1999). Results are presented as the ratio of the counted live cells (in green) to the counted dead ones (in red; live-dead ratio).

| Nutrient analysis
All water samples were analysed for ammonium (NH 4 ), nitrate (NO 3 ), soluble reactive phosphorus (SRP), and DOC concentrations, as well as DOM optical properties. Dissolved organic carbon concentrations were measured with the high-temperature catalytic oxidation method (Shimadzu TOC analyser). NO 3 was measured using the cadmium reduction method (Keeney & Nelson, 1982) with a Technicon Autoanalyzer. SRP and NH 4 were measured using a Shimadzu UV-2401 UV-Vis spectrophotometer using the molybdate method of Murphy and Riley (1962) for SRP and the salicylate method described by Reardon (1966) for NH 4 + . The samples for DOM optical properties were analysed at room temperature. Absorbance measurements were conducted using a 1-cm path length cell with the same spectrophotometer over a wavelength range of 200-800 nm. Fluorescence was measured with a Shimadzu RF-5301PC spectrofluorometer over (ex/em) wavelengths of 240-420 nm and 280-690 nm respectively using a 1-cm path length cell.

| DOM quality indices
All DOM data treatment was performed with MATLAB (version R2015b, MathWorks). The raw fluorescence data were divided by a correction file obtained for the lamp in use following the protocol of (Gardecki & Maroncelli, 1998) and then normalised to the Raman area to account for lamp decay over time (Lawaetz & Stedmon, 2008). Absorbance data of the respective samples were used to correct for the inner-filter effect (Lakowicz, 2006). MilliQ water blanks were subtracted to remove Raman scattering (Goletz et al., 2011).
Excitation-emission matrices were generated over (ex/em) wave-  Ohno, 2002). Specific UV absorbance was computed as absorbance measured at 254 nm, normalised to the DOC concentration and reported in units of L mg -1 m -1 . Higher specific UV absorbance at 254 nm has been found to correspond to higher aromaticity of DOM compounds (Weishaar et al., 2003). Spectral slope ratio was calculated as described by Helms et al. (2008), attributing an increase in this ratio to irradiation. The absorbance ratio E 2 E 3 , which is the ratio of absorbance at wavelength 250-365 nm, increases as molecular weight decreases (Peacock et al., 2014).

| Data analysis
For testing our first hypothesis regarding the differences between sampling locations, we used the following statistical tools. To assess if the location significantly affected water chemistry and microbial activity, a resemblance matrix based on the normalised Euclidean distance was calculated for a one-way analysis of similarity (ANOSIM), which calculates a global R statistic that assesses the differences in variability between groups compared to the variability within the group and checks for the significance of R using permutation tests. The ANOSIM was tested for the biogeochemical and the microbiological dataset separately and were tested for groups (1) sampling locations and (2) surface versus hyporheic. A canonical analysis of principal coordinates (CAP) was performed to visualise and classify the locations. CAP is a constrained ordination tool that discriminates locations defined a priori and determines the level of misclassification among sampling locations (Anderson & Willis, 2012). These features enabled us to determine which sampling locations were undergoing most changes and therefore reached highest misclassifications and which sampling locations showed stable conditions over time. Appropriate axis (m) was chosen by minimising the p-value from the permutation test based upon the trace statistic and maximising the leave-oneout allocation success, as suggested by Ratkowsky (2016). This approach tests how good the locations were discriminated using CAP. Furthermore, to quantify the effect of each variable to potential differences among locations, Spearman correlations were calculated for all variables and the CAP axes. Only the variables with a correlation coefficient |r| > 0.4 were considered. For both analyses, we used the PRIMER 6 + PERMANOVA (v. 6.1.11) computer program (Primer-E Ltd., Plymouth, UK) (Anderson, Gorley, & Clarke, 2008). The values were log(x + 1)-transformed to achieve normality. Additionally, one-way ANOVA with Tukey's post hoc analysis was used to evaluate differences between sampling locations at a significance level of p < 0.01.
Concentration differences along hyporheic flow paths were tested with a paired Student t test at a significance level of p < 0.05 for differences between pool and HZ inf , as well as HZ inf and HZ dw .
To investigate if there is a correlation between increases/decreases along hyporheic flowpaths with drought, we calculated Spearman rank correlations between the percentage increase/decrease and the discharge. A redundancy analysis (RDA) using the R-package vegan (Oksanen et al., 2013) was performed, whereby the environmental biogeochemical conditions were explanatory variables for the microbiological variables. Pretreatment consisted in removing redundant environmental variables by checking for collinearity and Hellinger transformation of microbiological variables (Legendre & Gallagher, 2001). The function envfit was used to evaluate which biogeochemical variables were significantly correlated with the first two RDA axes (p < 0.05).

| Physico-chemical characteristics and microbial activities across the surface-hyporheic pore water interface
Water physics and chemistry were characterised by electrical conductivity among 200-400 μS/cm, a pH between 6.5 and 7.2, 2-8 mg/L of DOC, and low nutrient content (Table 1). With regard to the microbiological parameters, the surface waters were characterised by higher β-glucosidase and leucine aminopeptidase, while the interstitial pore waters exhibited a higher live-dead ratio (see ANOVA results in Table 1).
Significant differences between locations were found for the chemical variables (Table 1, ANOSIM R = 0.394, p < 0.001) but surface water samples were not separated from the hyporheic pore water samples (ANOSIM R = 0.053, p = 0.216 for factor surfacehyporheic, Figure 2). With regard to the microbiological activities, differences among locations and among surface water and the interstitial pore water samples were significant (ANOSIM R = 0.272, p < 0.001 for factor location; ANOSIM R = 0.412, p < 0.001 for factor surface-hyporheic, Figure 2).

| Changes in water chemistry and microbial activity along hyporheic flow paths
We investigated the temporal variability of the hyporheic zone with drying by plotting the biogeochemical and microbiological parameters for the pool, HZ inf (10 m after the pool) and HZ dw (upwelling location 25 m after the pool) for pre-drought, contraction, transition, and fragmentation separately ( Figure 3) and compared increases or decreases with discharge data of the sampling date. For the dry phase, the water level in HZ inf was already below the well bottom.
As shown by the one-way ANOVA (Table 1), the live-dead ratio was higher in the hyporheic zone than in the pool. With lower discharge and longer water residence time, the live-dead ratio decreased to half of the initial values in HZ inf but achieved values of around 0.7 in HZ dw (significant increase along hyporheic flowpaths with lower discharge, Table 2). Extracellular enzyme activities were TA B L E 1 Mean ± standard deviation of all sampling locations (column) for all five samplings, when the location was not dry

Variable HZ up (n = 5) HZ lat1 (n = 5) HZ lat2 (n = 5) HZ inf (n = 4) HZ dw (n = 5) Pool (n = 5) Stream (n = 7)
Bacteria*, 10 6 cells/ ml 5.9 b ± 8.9 1.1 ab ± 0.3 1.3 ab ± 1.0 0.4 a ± 0.5 1.1 ab ± 0.7 2.8 ab ± 2.9 0.6 ab ± 0.1 Asterisks next to the variables highlight significant differences between sampling locations detected with the one-way ANOVA (p < 0.01), and letters next to the means indicate the results of Tukey's post hoc analysis, where different letters represent significant differences between the locations. The highest value for each variable is marked in bold. The explanation of each variable can be found in the Methods.
DOC, dissolved organic carbon; SRP, soluble reactive phosphorus generally lower in HZ inf compared to the pool, but phosphatase and β-glucosidase recovered along the hyporheic flow paths. Generally, these extracellular enzyme activities decreased in the hyporheic zone with longer water residence times and during fragmentation they were nearly undetectable for both hyporheic zone locations (significant for phosphatase, Table 2, Figure 3). Only leucine aminopeptidase showed maximum values in HZ dw during the transition phase, which was when water residence time already exceeded a day. The value of 0.84 nmol L −1 hr −1 measured this day represents three times the mean value in HZ dw over the whole sampling period.
During the transition phase, a peak of bacterial cell concentration of 2.3 × 10 6 cells ml −1 , which is twice the mean bacterial cell concentration of HZ dw , was also detected.
The change in concentrations between the pool and HZ dw revealed that the hyporheic zone acted as a sink for dissolved oxygen and both as a sink and a source, for DOC and inorganic nutrients.
During contraction, anoxic conditions were detected at night time in HZ dw , but still the dissolved oxygen concentration would recover to values above 0.5 mg/L during the day (data not shown). When water residence time exceeded 1 day, the dissolved oxygen was completely consumed in HZ dw at any time of the day. In general, DOC was retained in the hyporheic zone, only interrupted by DOC pulses from HZ inf during the transition phase and from HZ dw during fragmentation. Conversely, SRP and NO 3 concentrations always increased when surface water entered the hyporheic zone at HZ inf . This increase between pool and HZ inf was higher with longer water residence time and lower discharge (significant for SRP, concentrations were lowest at HZ inf and increased along hyporheic flow paths with a peak in HZ dw during the transition phase.
Three optical indices represent how DOM quality changed between the pool and HZ dw : E 2 E 3 increased along hyporheic flow paths, but the magnitude of this increase was not associated with water residence time. The fluorescence index did not show any change between HZ inf and the pool, but clearly increased between HZ inf and HZ dw . The humification index showed a similar pattern as the inorganic nutrients, whereby this index increased between the pool and HZ inf , but decreased between HZ inf and HZ dw . In line with the pattern of the inorganic nutrients, the increase/decrease sequence of humification index showed higher magnitudes with higher water residence time.

| Chemical constraints on microbial activity
We explored the relationships between environmental variables and microbial activity with an RDA (Figure 4). The first two axes were significant (p < 0.001), whereby the first axis explained 16% and the second axis 16% of the variation. Out of the 14 environmental variables, six showed a high correlation with the first two RDA axes, namely dissolved oxygen, NO 3 , SRP, pH, NH 4 , and E 2 E 3 ratio (all p < 0.05). The E 2 E 3 absorbance ratio loaded negatively on the first axis and plotted together with the live-dead ratio. This was in contrast to dissolved oxygen concentration, which loaded positively on the first axis and was related to leucine aminopeptidase and β-glucosidase. The second axis separated SRP, NO 3 , and pH loading positive, from E 2 E 3 ratio and NH 4 that were loading negatively.
The E 2 E 3 ratio and NH 4 concentration were related to bacterial cell concentration. Negative values indicate decrease and positive values indicate increase. The outcome of the Spearman rank correlation test is presented with Spearman rho and p-values testing the correlation between percentage increases and decreases. A negative rho indicates an increase with drying, hence lower discharge, and a positive rho indicates a decrease with drying. If numbers are bold, there was a significant difference between the two locations along all four samplings (paired Student t test significant at p < 0.05). Rows on the bottom show temperature and oxygen ranges of sampling dates continuously measured at HZ dw and the discharge measured at the outlet of the pool on the same sampling date. The values from pre-drought sampling refer to a date 8 days after the actual sampling date due to battery failure of the same day. DOC, dissolved organic carbon; SRP, soluble reactive phosphorus towards lower SRP and NO 3 concentrations and lower pH, but higher NH 4 concentration. This shift was related to a lower live-dead ratio and lower phosphatase, but higher bacterial cell concentration.

| Microbial activity and biogeochemistry across the surface-hyporheic interface in an intermittent system
We hypothesised that we would find a different nutrient status and DOM availability in the surface waters versus the interstitial pore waters. However, the surface locations did not separate from the hyporheic locations based on these characteristics but based on differences in extracellular enzyme activities and living bacteria abundance. The water in the hyporheic zone of the main channel originated exclusively from the surface water during the study period, but of- Conversely, the microbiological variables changed immediately once the water has entered the hyporheic zone, since all interstitial pore water samples showed a clear division from the surface waters ( Figure 2). Hence, we suggest a delay between the change in microbial activity that occurs immediately once the water enters the hyporheic zone and the chemical response observed along the hyporheic flow paths.
Very few studies have compared extracellular enzyme activities and bacterial cell concentration in surface and interstitial pore water of natural stream ecosystems (e.g., Ann, 2015;Romaní, Vázquez, & Butturini, 2006). The extracellular enzyme activities β-glucosidase and leucine aminopeptidase from Fuirosos stream surface water in spring season during well-established baseflow connection were twice the maximum values of this study (Ylla et al., 2010). Only right before streambed desiccation, when nearby groundwater did not supply the stream anymore, were the values similarly low.

| Nutrient retention and DOM mineralisation in the hyporheic zone
Drying and its consequences, such as disconnection of flow paths laterally and longitudinally, as well as longer water residence times (Fisher, Grimm, Martí, & Gómez, 1998;Harvey, Conklin, & Koelsch, 2003) provoked a decrease in SRP and NO 3 concentrations that is also observed in other studies from intermittent streams (Bernal & Sabater, 2012;Martí, Grimm, & Fisher, 1997;von Schiller et al., 2008). Additionally, in intermittent streams, the absence of surface flow can involve the diffusion of oxygen into the pore space of the hyporheic zone that triggers ammonia oxidation (Merbt et al., 2016).
Both nitrification preceding denitrification in the hyporheic zone increase with water residence times (Zarnetske, Haggerty, Wondzell, & Baker, 2011a). We observed NO 3 removal of up to 0.21 mg-N/L during fragmentation between HZ inf and HZ dw when water residence time exceeded a day.
Labile DOC can additionally enhance hyporheic denitrification, with rates of up to 0.37 mg-N/L in 10 hr (Zarnetske, Haggerty, Wondzell, & Baker, 2011b), which is higher but still of a similar magnitude as found in this study. Conversely, DOM infiltrating with the surface water showed a high humification degree in HZ inf , in line with high nitrification rates reported when recalcitrant DOM is abundant (Strauss & Lamberti, 2002). However, the DOM quality was improving along hyporheic flow paths, as shown by decreasing humification, but increasing fluorescence index (autochthonous) and E 2 E 3 (smaller molecules) values. The fluorescence index values showed similar dynamics as βglucosidase, an extracellular enzyme activity that is reported to correlate with autochthonous DOM availability (Freixa, Ejarque, et al., 2016;Proia et al., 2016) and higher pCO 2 levels (Grossart, Allgaier, Passow, & Riebesell, 2006). It should be mentioned that humic-like DOM can also be preferentially sorbed to the hyporheic sediments, although we suggest that this might have played a minor role due to the coarse nature of the sediment body of this study (low silt content, no clay fractions in the catchment (Juhna, Klavins, & Eglite, 2003)).
Furthermore, the dynamics of β-glucosidase, a drop at HZ inf , and an increase in HZ dw , are consistent with the DOM retention and the pCO 2 release that occurred in this hyporheic zone reach (data in Table S2). DOM derived from surface water can fuel hyporheic metabolism (Clinton, Edwards, & Findlay, 2010), but DOM can also be derived from particulate organic matter, extracellular enzymes or released during anabolic processes in the hyporheic zone (Burrows et al., 2017;Stegen et al., 2016). Fischer et al. (2002) suggested that the cycling of these autochthonous DOM fractions might be faster and consequently their contribution to bacterial metabolism would be higher than inferred from the apparent DOM retention.
Moreover, bacterial cell concentration increased in all sampling locations with drying that is in contrast to continuous flow column experiments, where bacterial cell concentration rapidly decreased with depth (Perujo et al., 2017). Conversely, a reported increase in bacterial cell concentration in the interstitial pore water of a sediment desiccation laboratory experiment (Pohlon, Fandino, & Marxsen, 2013), suggests that microbial activity in intermittent streams follows a different distribution pattern of microbial activity than their perennial counterparts. Furthermore, the results of this laboratory experiment indicated a community shift towards bacteria that are more efficient in carbon and nutrient acquisition (Pohlon et al., 2013). This described community shift could also explain the decrease in nutrient concentrations and apparent DOM molecular size.

| The hyporheic zone as a refuge
The hyporheic zone appeared to have acted as a refuge for the bacteria during the drying period of the stream as shown by higher live-dead ratio of subsurface waters compared to the surface water. The sediment desiccation of intermittent rivers is reported to drastically reduce living bacteria when experimentally tested in mesocosms . However, we suggest that natural streams and rivers often maintain deeper, watersaturated layers of the hyporheic zone as a refuge for microbial activity during drought as suggested by Romaní et al. (2013). We found significantly higher live-dead ratios in the interstitial pore water compared to surface water that points towards this survival mechanism of bacteria and is further evidenced by the increase of the live-dead ratio along the hyporheic flow path between HZ inf and HZ dw . Similarly, the bacterial community in the hyporheic zone (at 10 cm depth) of Fuirosos showed higher resistance to drying than the upper layers of sandy sediment cores (Timoner, Borrego, Acuña, & Sabater, 2014). Furthermore, Ann (2015) reports higher extracellular enzyme activities in interstitial water of the Tordera River with absence of surface flow from sediment depths similar to those of this study (−30 to −50 cm). Concerning the resistance mechanisms of microbial metabolism during drought, there is a paucity of studies reporting live-dead ratios and bacterial cell concentration in interstitial pore water. This would be particularly of interest, as prokaryotes are able to travel greater distances and deeper into the hyporheic zone with the water flow, because of their ubiquitous nature (Febria et al., 2012;Peralta-Maraver, Reiss, & Robertson, 2018;Romaní et al., 2017). In this sense, prokaryotes might have a survival advantage compared to larger microbes that makes them the essential actors of nutrient retention and DOM mineralisation in intermittent streams as well as other environments with fluctuating water levels (Goldman et al., 2017;Stegen et al., 2016). Hence, there might be an underestimation of bacterial survival of droughts and hot spots in the hyporheic zone that can show enhanced carbon turnover rates Ylla et al., 2010). This idea is further corroborated by the much lower microbial activity that we found in the interstitial pore water in October when the fluvial continuum had re-established (data shown in Table S1). Still, we have to acknowledge the fact that the number of observations in this study is small and further studies are needed to fully understand the hyporheic zone as a refuge for microbial activity. For example, microbial metabolism was shown to be affected by the hydraulic conductivity of the hyporheic zone (Mendoza-Lera & Datry, 2017). Further, we suggest comparing intermittent streams with previously perennial streams or to test different water depths to understand how deep the prokaryotes can migrate. Conclusively, our results emphasise the importance of the hyporheic zone during drought: not just that remaining watersaturated locations are biogeochemical hot spots during drought that act as DOM and nutrient sinks, but they can also provide a refuge for microbial activity that can explain enhanced DOM and nutrient turnover rates upon rewetting.

ACK N OWLED G M ENTS
We thank M. A. Gallegos and P. Rodrigo for field and laboratory assistance. We are grateful for the helpful comments of four anonymous reviewers that improved the manuscript. This study was funded by European Union's Seventh Framework Marie Curie training project INTERFACES (grant agreement no 607150) and was partly supported by the project GLOBAQUA (603629-ENV-6.2.1).

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.