Nitrate and Nitrite Variability at the Seafloor of an Oxygen Minimum Zone Revealed by a Novel Microfluidic In-Situ Chemical Sensor

Microfluidics, or lab-on-a-chip (LOC) is a promising technology that allows the development of miniaturized chemical sensors. In contrast to the surging interest in biomedical sciences, the utilization of LOC sensors in aquatic sciences is still in infancy but a wider use of such sensors could mitigate the undersampling problem of ocean biogeochemical processes. Here we describe the first underwater test of a novel LOC sensor to obtain in situ calibrated time-series (up to 40 h) of nitrate+nitrite (ΣNOx) and nitrite on the seafloor of the Mauritanian oxygen minimum zone, offshore Western Africa. Initial tests showed that the sensor successfully reproduced water column (160 m) nutrient profiles. Lander deployments at 50, 100 and 170 m depth indicated that the biogeochemical variability was high over the Mauritanian shelf: The 50 m site had the lowest ΣNOx concentration, with 15.2 to 23.4 μM (median=18.3 μM); while at the 100 site ΣNOx varied between 21.0 and 30.1 μM over 40 hours (median = 25.1μM). The 170 m site had the highest median ΣNOx level (25.8 μM) with less variability (22.8 to 27.7 μM). At the 50 m site, nitrite concentration decreased fivefold from 1 to 0.2 μM in just 30 hours accompanied by decreasing oxygen and increasing nitrate concentrations. Taken together with the time series of oxygen, temperature, pressure and current velocities, we propose that the episodic intrusion of deeper waters via cross-shelf transport leads to intrusion of nitrate-rich, but oxygen-poor waters to shallower locations, with consequences for benthic nitrogen cycling. This first validation of an LOC sensor at elevated water depths revealed that when deployed for longer periods and as a part of a sensor network, LOC technology has the potential to contribute to the understanding of the benthic biogeochemical dynamics.


Introduction
In situ, high frequency observations are crucial to uncover the temporal and spatial complexity in the aquatic environments [1]. Considerable progress has been made with the increasing use of in situ sensors on a variety of platforms; however these devices mostly address physical parameters such as temperature, light or pressure [2,3]. On the other hand, with the notable exception of oxygen [4,5], pH [6] and nitrate at low (μM) resolution (see below), routine calibrated and accurate measurements of chemical parameters still depend on spatially and temporally limited sampling schemes and subsequent analyses onboard research vessels or in shorebased laboratories. Hence, the biogeochemical variability of oceanic habitats is so far largely unknown but is ultimately important to understand the productivity, elemental cycles and ocean response to a changing global climate.
As a limiting nutrient of biological productivity, nitrate (NO 3 -) and nitrite (NO 2 -) are among those parameters for which several sensor technologies already exist, such as a UVspectrophotometry based system reported by Johnson et al. [7] (also see [8,9]), and the colorimetric system by Le Bris et al. [10] (for reviews see [11][12][13]). Some nitrate sensors have become commercially available (such as Satlantic ISUS, Systea WIZ), and recent applications of these sensors uncovered the high temporal variability in NO 3 -in the open ocean [8]. The temporal variability can be even higher in estuaries, where current can be fast and freshwater inputs are significant [14]. Despite these improvements, the capacity of many nutrient sensors to perform long term, stable measurements at elevated water depths may be limited due to shallow depth range, large size, large power consumption and/or high detection limits. The bottom waters of oceanic oxygen minimum zones (OMZ) are one of those environments for which there is almost no information on the short-term temporal nutrient variations. Here, the availability of NO 3 -/NO 2 drives the anaerobic degradation of organic matter in the water column and the sediment surface [15]. Moreover, these electron acceptors are involved in the anaerobic oxidation of NH 4 + , which is released in high amounts from the seafloor under anoxic conditions [16,17]. In organic-rich sediments NO 3 -/NO 2 can be further involved in the oxidation of sulfides [18]. Besides consumption and production, physical transport processes such as currents, submesoscale eddies or passage of internal waves can also drive the variability of these nutrients over continental shelves below OMZs [19].
To accurately constrain the variability of NO 3 -/NO 2 at the seafloor of an OMZ, we need pressure-insensitive, robust sensors with the capability of in situ calibration. The in situ calibration is particularly important for an application at a fixed point at the seafloor, where there is potential for variability over short time scales. In this regard, microfluidics, or 'lab-on-a-chip' (LOC) is a promising technology that meets several demands needed for autonomous measurements at elevated depth such as low weight, low energy demand, and low volume consumption of reagents and samples [20,21] and ease of integration into underwater platforms. Having been mostly developed for medical and pharmaceutical research [22], LOC technology is now being transferred to aquatic environmental research as well [23][24][25][26][27]. Newly developed LOC devices can measure not only NO 3 and NO 2 -, [23,26,27] but also PO 4 3- [28] and dissolved Fe and Mn [29]. Among these devices, the NO 3 -/ NO 2 -LOC system is approaching maturity for time-series applications but, prior to the work reported here, its suitability for the in situ deployment at elevated depth has not been demonstrated. Using a new generation LOC sensor that was capable of in situ calibration, we report the first in situ time series (up to 40 h) of NO 3 -+ NO 2 -(SNO x ) and NO 2 at the seafloor (max. 170 m depth) in an upwelling region offshore Western Africa. We combine these data with time series of oxygen, temperature and bottom currents to demonstrate that the LOC technology provides a previously unavailable window into the variability of nutrient concentrations in the bottom waters of a low-oxygen coastal ocean.

Study Area
The study site was near 18°N in the Mauritanian upwelling region which is part of the Canary eastern boundary upwelling system that roughly extends between 43°N to 10°N [30,31] While the core of the main OMZ is found at about 400 m depth, a secondary oxygen minimum is situated below the surface mixed layer and above 200 m depth [32]. The ventilation of the waters above the continental margin occurs primarily through the Mauritania Current [33] in the near-surface layers and the Poleward Undercurrent below [34]. Both currents transport relatively oxygen-rich South Atlantic Central Water, which is supplied by the eastward flowing North Equatorial Countercurrent and North Equatorial Undercurrent, northward into the upwelling region. A physical-biogeochemical measurement program was performed during the R/V Meteor cruise M107 from May 28 to July 3, 2014 during the termination period of the upwelling season as a part of a collaborative research center Climate-Biogeochemistry Interactions in the Tropical Ocean (SFB754) and the Helmholtz-Alliance ROBEX (Robotic Exploration of Extreme Environments). The observational program included benthic lander deployments, conductivity-temperature-depth-oxygen (CTD/O 2 ) profiling paired with water sampling and mooring deployments measuring currents and hydrography along a transect at 18°N.

Lander deployments at the Bottom Boundary Layer
A GEOMAR benthic lander (thereafter as the Lander) was used as a platform to make autonomous in situ measurements [35]. The Lander was deployed on the sea floor at depths of 50, 100 and 170 m ( Table 1). These depths were chosen as bottom water oxygen time series displayed high variability during a previous cruise [36] (R/V MS Merian leg 17/4) and benthic nitrogen turnover rates were high due to hypoxia [17]. At all locations, the LOC sensor (Ocean Technology and Engineering Group, National Oceanography Centre Southampton) was attached to the lander and at the 50 and 100 m sites this LOC system was complemented with a CTD probe (RBR, Ottawa, Canada) and an O 2 optode (Aandrea, Bergen, Norway). For the deployment at the 50 m site an additional LOC sensor measuring NO 2 was added. Prior to the Lander

Application of the Lab-on-Chip Nitrate/Nitrite Sensor
The LOC SNO x sensor was previously described in detail by Beaton et al. [23,26]. That paper as well as Ogilvie et al. [37] and Floquet et al. [38] described various aspects of the lab-on-achip approach for nutrients. In short, the chip used in the sensor was made up of PMMA and harbored precision-milled microchannels (< 300 μm), mixers and optical components such as LEDs (525 nm) and photodiodes. An integrated syringe pump, valves and electronics complemented the chip and are encased in mineral oil-filled housing (PVC, 12 cm diameter, 30 cm height) with an internally fitted pressure-compensating bladder. The system ran autonomously storing the data in a memory card and was powered by an external battery. The unit weighed 1.1 kg in water without and 1.5 kg with an internal battery. The sensor uses attached standards and blank for regular in situ calibrations along with sample (0.45 μm filtered) measurements, using the colorimetric Griess assay for NO 2 detection [39]. The addition of an off-chip Cu-activated Cd column enabled NO 3 -+NO 2 -(SNO x ) detection through the reduction of NO 3 to NO 2 and subsequent on-chip analysis with a detection limit of 20 nM, an accuracy of 0.6% and a precision of 7 nm at low concentrations or 0.5% at high concentrations [23]. The deployment involved externally attached gas impermeable Flexboy bags (150 mL, Sartorius, UK) that contained two standard solutions (for either SNO x or NO 2 -, as needed), Griess reagent, artificial seawater blank and imidazole buffer. The preparation of the standards and reagents followed previous protocols [40, 23, and 26]. Waste was collected in a 500 mL Flexboy bag. For example, for a deployment to measure SNO x , the sensor started to perform two sets of calibrations (artificial seawater blank and 11.3 and 33.9 μM standards for NO 3 -) followed by a repeating sequence involving the measurement of blank, sample, standard solution and sample again. Each of these steps included 6 flushing cycles to avoid any carryover. The final flush cycle was followed by a 100-second waiting stage that enabled color development and the photometric measurement at 525 nm. A set of flushes and the waiting stage altogether took about 7 minutes. The sensor recorded optical absorbance every 1 second and the average of the last 3 reading of the waiting stage was used for calculations. Raw data were processed in R 3.0.2 [41]. With this scheme, we obtained one blank-corrected sample measurement every 14 minutes and one standard solution measurement every 28 minutes during deployment. For NO 2 measurements, the off-chip Cd column was replaced with a short tube. The sensor operated the same as SNO x sensor, with the only exception that the standard solutions of 0.5 and 2 μM NO 2 were used.

Results and Discussion
First tests of the LOC analyzer at elevated depth Two mid-depth O 2 minima, sharp nitracline (increasing nitrate concentration over a relatively small depth range in the water column) and a subsurface NO 2 peak were persistent features in different CTD casts at the Mauritanian Shelf (one example shown in Fig 1A). Before the Lander deployments, we assessed the quality of the LOC sensor to reproduce vertical nutrient gradients by comparing it to concentrations obtained from water samples. It was therefore attached to the CTD frame during a profile to a depth of 160 m (CTD #16, S1 Fig). The sensor started to perform two sets of calibrations while the CTD was being lowered to 160 m. During the upcast, the rosette was stopped for 14 minutes at 7 different depths to allow for the in situ analysis of the blank (or a standard) and a sample. In the final minute of this period, a Niskin bottle sample was taken. Overall, the SNO x distribution obtained from the LOC agreed well with the distribution of SNO x obtained from shipboard autoanalyzer (AA) water samples ( Fig 1B). All profiles showed that SNO x was depleted at the surface with a nitracline located at about 20-40 m depth, below which SNO x increased to about 30 μM. The location of the nitracline coincided with a NO 2 concentration peak of 0.6 μM at 30 m, which was also the upper boundary of the O 2 minimum (29 μM after 60 m) extending down to 160 m.

Analytical Uncertainty Assessment
In addition to the submerged LOC tests during hydrocasts, we have performed additional comparisons (LOC measurement also shipboard) on bottle samples from another hydrocast at 1000 m (CTD#2). The combined scatter plots showing autoanalyzer (AA) and LOC measurements during CTD#2 and 16 (Fig 2) indicate that the results from the two approaches are highly correlated (R 2 > 0.99). The differences at high (>5 μM) concentrations were comparable to the estimated uncertainties of the two approaches, which were calculated from two times the standard deviations of successive calibrations (on board for autoanalyzer and in situ for LOC). This analysis yielded a value of ±0.1 μM for the autoanalyzer and ±0.4-1 μM for the LOC analyzer (summarized in Table 2). Considering these inherent analytical uncertainties with the differences in sample handling by different operators, we can assert that the LOC nutrient sensor was capable of delivering in situ measurements at elevated depth with an analytical performance only slightly less precise than the established techniques.  In addition to the water sample comparison to validate sensor performance, it was also verified that the sensor performed reproducible in situ calibrations throughout all deployments, essential to obtain reliable time series data. An excerpt of raw photodiode readings featuring in situ blank, standard and sample measurements during the deployment at 170 m is shown in Fig 3. During all deployments no significant drift in the in situ calibrations was observed, however a random variability was present between successive calibrations from which sensor uncertainties were estimated ( Table 2). The one blank-corrected sample per 14-minute measurement frequency used in this work should be considered as a minimum because we have aimed to fully utilize the in situ calibration feature and we had to use relatively high number of 6 flushing cycles due to high NO 3 concentrations. Although we have not investigated a full optimization, we think that the frequency can increase to 1 measurement per 10 minutes with less frequent blank/standard analyses. Using a smaller number of flushing cycles can further increase the sample analysis frequency for low concentration samples (<5 μM). Use of newly emerging microfluidic architectures such as multiplexed stop flow [42] can enable multiple simultaneous measurements on the same chip, increasing the frequency to one sample per minute or more.

Nitrate and nitrite dynamics in the bottom boundary layer
The CTD casts only provided a snapshot of a dynamic system while the time series for SNO x from the bottom waters at 50, 100 and 170 m (Fig 4)  The median values of the time series measurements increased with depth, consistent with the trend suggested by the water column profiles. While the long-term (hours) variability was remarkable (discussed in detail in the next section), short-term variability on time scales less than hour was also high in certain periods. To compare the inherent sensor precision with natural variability, we have also plotted a 'moving uncertainty', calculated from two times the moving (n = 5) standard deviations of the successive in situ measurements of a standard solution (Fig 4 and Table 2). This comparison showed that most of the short-term variation less than a micromolar could be attributed to sensor response. However at certain periods there were marked deviations from the expected uncertainty, such as those in the 170 m time series (Fig 4C). The most likely physical explanation for the short-term SNO x variability is the propagation of bores and non-linear internal wave trains, observed at the upper continental slope and the shelf off Mauritania at 18°N [19]. They are generated by interaction of barotropic and low-mode baroclinic tides with topography [43] and propagate onshore. Individual nonlinear internal waves within the bores exhibited periods of 10 to 15 minutes and associated vertical velocities exceeded 0.15 m s -1 while on the continental slope they were particularly pronounced between 75 and 200m depth [19]. During their passage, they vertically displace water over a depth range of more than 70m. Thus we hypothesize that the short-term variability was due to the downward displacement of water having lower SNO x concentrations from above. Non-linear internal waves have been observed on continental slopes in many continental slopes in the world ocean [44]. We suggest that they cause elevated high-frequency variability of near-bottom SNO x concentrations and other solutes having vertical concentration gradients in the lower 50m to 100m of the water column, which may affect biogeochemical processes in the sediments and within the bottom boundary layer. At 50m, a location near the subsurface NO 2 peak (Fig 1) When LOC measurements are appended to the results of the discrete bottom water measurements (Fig 5, inset), a more complete picture emerges where the starting NO 2 poor, NO 3 rich waters were replaced by waters higher in NO 3 and lower in NO 2 through the middle of series.
At the end of this period, the bottom water chemistry returns to the starting composition, suggesting a~3 day cycle.  Table 1 for dates). The gray line in each series denotes a 12-point moving average. Also plotted as gray shading is 'moving uncertainity' corresponding to concentrations calculated from two times the moving (n = 5) standard deviations of the successive in situ measurements of a standard solution (

Drivers of the long-term (hours) benthic biogeochemical variability
In order to account for the observed variability of nutrients over several hours, we also recorded physical parameters and oxygen at the 50 and 100m sites (Fig 6). In both sites, the tides did not explain any of the nutrient variability. Cross correlation matrices of the series (S2 Fig) revealed that the 50 m site had more correlated pairs than the 100 m site. Considering this difference, and the fact that our dataset is more complete for the 50 m site, we focus our discussion on the 50 m site. At the 50 m series, the NO 2 -, which correlated inversely with NO 3 -(Spearman Rank Correlation, r = -0.59, p<0.01), also had a significant inverse correlation with temperature (T) (r = -0.52, p<0.01) and positive correlation with O 2 (r = 0.58, p<0.01). Nitrate had weaker, but still significant correlations (in opposite sign to that of NO 2 -) with these parameters. Taken together, these findings suggest that during the course of the 50m time series NO 3 rich NO 2 --poor cooler waters replaced the original warmer, O 2 -rich, NO 2 rich waters. Therefore the overall trend was towards increasing temperatures, decreasing NO 2 and O 2 while increasing NO 3 - (Fig 4 and Fig 6a).
What drives these non-tidal variations in the chemical and physical parameters? To attempt an explanation we examined the velocity profiles from an acoustic Doppler current profiler (ADCP) that was deployed on the sea floor at the depth of 50 m, close to the Lander deployment site (Table 1). Over a 15-day period, the alongshore and across-shore currents fluctuated The correlation between near-bottom onshore flow variability and alongshore flow variability in the water column can be interpreted in terms of bottom-boundary layer Ekman dynamics. Due to bottom friction and the Coriolis force, a northward alongshore flow above the bottom boundary layer will generate an offshore flow component in the bottom boundary layer, which is in agreement with velocity time series (Fig 7). Thus, during the period of elevated northward flow on the shelf, the intrusion of deeper waters was blocked and water masses from the shallower shelf regions that were low in NO 3 and high in NO 2 were advected offshore. When the northward current ceased, deeper waters started to move on shore, which resulted in higher NO 3 -, lower NO 2 and lower O 2 concentrations (Fig 6). While the northward flow on the shelf can generally be attributed to the Mauritanian current during this period [32], the elevated northward velocities between June 22 and June 26 represent an intermittent feature that is probably due to local wind forcing. One biogeochemical implication is that these fluctuating currents may lead to periodic pumping of deep waters at the 50 m site. At this shallow site this might be an important mechanism for nutrient transport to support primary production as well as for the supply of electron acceptors to drive denitrification and the

Conclusions and Future Perspectives
We have successfully deployed a new prototype microfluidic lab-on-chip (LOC) NO 3 -/ NO 2 sensor in the bottom waters of a productive coastal ocean. To our knowledge this is the first report of time series from a LOC NO 3 -/ NO 2 sensor in a deep underwater setting. We found that the analytical performance of the sensor at elevated depth was comparable to a conventional autoanalyzer. Coupled to time series measurements of oxygen and currents, the LOC measurements recorded large variations in nutrients, probably linked to cross-shelf water transport occurring in relation to the dynamics of the alongshore flow on the Mauritanian shelf. Relative variability of NO 2 in the bottom waters was larger and this could affect the magnitude and direction of benthic biogeochemical fluxes in this low-oxygen coastal ocean.
Although we have not tested the LOC sensor in situ at depths beyond 170 m, our results show that the microfluidic sensor approach is not compromised by elevated pressures and therefore has a high potential for deep-water applications. New developments in sensor technology (such as miniaturization and new microfluidic architectures) are expected to decrease the size of these devices, allow the introduction of new chemical parameters, increase data acquisition frequency and enhance the applicability of the sensor towards Lagrangian (moving) platforms. These developments will not only be beneficial to Earth observation but also to space and planetary exploration-indeed this motivation was the starting point of our sensor testing work under the framework of Helmholtz Alliance ROBEX (Robotic Exploration of Extreme Environments) [45]. Particularly, having robust sensors at high technological maturity for underwater applications could contribute to future missions to active icy moons such as the Jovian moon Europa and Enceladus in the Saturnian system [46]. As the potential habitability of these moons is receiving increasing attention; the detection of biosignatures, dissolved nutrients or energy-yielding chemicals in the subsurface oceans [47] of these icy satellites would considerably benefit from in-situ analysis. In order to support this vision, and more specifically to increase the technological readiness of the new sensors, we need to continue to bring new generation chemical sensors to extreme environments such as deep sea, assess their capabilities and limitations and eventually implement sensor networks that will pave the way for more extensive testing and the much-needed highly resolved in situ biogeochemical datasets.