In situ quantification of ultra‐low O2 concentrations in oxygen minimum zones: Application of novel optodes

Conventional sensors for the quantification of O2 availability in aquatic environments typically have limits of detection (LOD) of  > 1 μmol L−1 and do not have sufficient resolution to reliably measure concentrations in strongly O2 depleted environments. We present a novel trace optical sensor based on the palladium(II)‐benzoporphyrin luminophore, immobilized in a perfluorinated matrix with high O2 permeability. The trace sensor has a detection limit of ∼5 nmol L−1 with a dynamic range extending up to ∼2 μmol L−1. The sensor demonstrates a response time < 30 s and a small, predictable, and fully reversible response to hydrostatic pressure and temperature. The sensor showed excellent stability for continuously measurements during depth profiling in Oxygen Minimum Zones (OMZ). The novel sensor was deployed in situ using a Trace Oxygen Profiler instrument (TOP) equipped with two additional O2 optical sensors, with higher dynamic range, allowing, when combined, measurements of O2 concentration from ∼5 nmol L−1 to 1000 μmol L−1 with a single instrument. The TOP instrument was deployed in the OMZ regions of the Eastern Tropical North Pacific (ETNP) and Bay of Bengal (BoB). The measurements demonstrated that O2 concentrations in the ETNP generally were below the LOD of the trace sensor, but that large sub‐micromolar O2 intrusions, spanning 60–80 m with maximum O2 concentrations above 50 nmol L−1, could be observed in the OMZ core. The O2 concentrations in the BoB were high compared to the ETNP and rarely decreased below 50 nmol L−1, but demonstrated tremendous small‐scale variability.

The methods and sensors typically used to quantify O 2 concentrations in aquatic environments are based on standard Winkler titrations (Grasshoff 1983;Winkler 1888), electrochemical methods (e.g., Revsbech 1989;Martini et al. 2007), and only more recently, the use of optical sensors (Tengberg et al. 2006;Chipman et al. 2012;Kirf et al. 2014b). The majority of these techniques, however, have LOD's of > 1 lmol L 21 .
While such measurements are adequate for many investigations, they do not have sufficient resolution to reliably measure O 2 concentrations in strongly O 2 depleted environments, such as marine oxygen minimum zones (OMZ's) where O 2 concentrations can fall below 10 nmol L 21 (Revsbech et al. 2009;Thamdrup et al. 2012;Tiano et al. 2014;Ganesh et al. 2015).
To date, in situ submicromole O 2 concentrations in OMZ's have only reliably been measured using electrochemical STOX sensors (Switchable Trace OXygen) (Revsbech et al. 2009(Revsbech et al. , 2011. The high sensitivity of the STOX sensor is achieved by a switchable front guard. The front guard electrochemical removes all O 2 at the tip of the sensor, thus allowing in situ calibration of the zero signals, minimizing uncertainties caused by, for example, fluctuating temperature, electronic drift, and zero offsets. For typical STOX sensor measurements, the front guard is switched on/off in a cyclic pattern, with a period of 30-75 s (Revsbech et al. 2011), representing the time to obtain a single measurement point. With STOX sensors it is generally possible to detect and quantify O 2 concentrations of $10 nmol L 21 (Revsbech et al. 2009(Revsbech et al. , 2011. With this high sensitivity, STOX sensors have substantially improved our understanding of O 2 dynamics and concentrations within OMZ regions (Kalvelage et al. 2011;Dalsgaard et al. 2012;Thamdrup et al. 2012;Ulloa et al. 2012;Tiano et al. 2014;Ganesh et al. 2015). However, the production and use of these sensors require highly specialized expertise and very few laboratories mastered their production (Revsbech et al. 2011). Furthermore, the commercially available STOX sensors are costly. These constraints have limited our advancement in resolving and understanding how the availability and dynamics of O 2 regulate key biogeochemical processes in OMZ's (Ulloa et al. 2012) and other low-oxygen environments.
O 2 optodes offer an attractive alternative to STOX methodology for trace O 2 sensing. O 2 optodes with high sensitivity (LOD $1-50 nmol L 21 ) are relatively easy to manufacture, and they typically feature a nonlinear response that significantly enhances the resolution and the signal to noise ratio at low O 2 concentrations (Lehner et al. 2015). While trace O 2 optodes have been applied to laboratory-based measurements (Nestler et al. 2007;Holtappels et al. 2014;Ganesh et al. 2015;Lehner et al. 2015), in situ-based systems have not had sufficient sensitivity for measurements in OMZ regions. Furthermore, such sensors have not been adequately tested for the influence of changing hydrostatic pressure on signal response, limiting their application for in situ deployments. To our knowledge, in situ trace optical sensors (LOD $10 nmol L 21 ) have so far only been applied in relatively shallow limnic environments (Kirf et al. 2014a,b).
The sensing material in optodes is traditionally a twocomponent system, composed of an indicator and a polymer matrix. While several luminescent O 2 indicators exist, only a few are frequently used for monitoring O 2 in aquatic environments (Quaranta et al. 2012). The function of the matrix is to physically entrap the indicator, to eliminate undesired cross-talk from ionic species and to offer the possibility of tuning the sensitivity of the sensors by using matrices with different permeabilities (Draxler et al. 1995;Koren et al. 2013). For all O 2 optodes, the dynamic range spans about three orders of magnitude. Accordingly, the dynamic range of a "standard" optical O 2 sensor is typically $0.5-1000 lmol L 21 , whereas available trace sensors with a LOD of $ 50 nmol L 21 typically have a dynamic range up to $60 lmol L 21 . Therefore, given the low upper range of highly sensitive optodes, it is necessary to use a combination multiple sensor to span the O 2 gradient typical of OMZ regions, where O 2 can range from > 200 lmol L 21 to < 10 nmol L 21 in the OMZ core.
Here, we present and evaluate a novel trace O 2 sensor design that combines a long-lived palladium(II)-fluorinated benzoporphyrin luminophore, immobilized in a highly O 2 -permeable perfluorinated polymer, allowing detection and quantifications of ultra-low O 2 concentrations of down to $5 nmol L 21 . The optical trace range sensor has been deployed together with mid-range and standard range O 2 optodes, integrated in a stand-alone trace oxygen profiler (TOP) instrument. The TOP instrument was used to resolve O 2 availability and dynamics within the OMZ of the Eastern Tropical North Pacific (ETNP) and the Bay of Bengal (BoB). The optode results are compared and discussed in relation to pressure and temperature effects and with respect to concurrent measurements with STOX sensors.
Easily manufactured optical sensors allow detailed in situ measurements of O 2 concentration in a range from $5 nmol L 21 to 500 lmol L 21 by a single instrument-an ability that will markedly advance our present understanding of biogeochemical processing in OMZ's and other highly O 2 -depleted aquatic environments.

Optical sensors principle and design
Luminescent-based optical sensors are based on the ability of O 2 to act as a dynamic quencher, reducing the luminescent intensity and the lifetime of the O 2 indicator (Kautsky 1939;Klimant et al. 1995), allowing quantification of the O 2 partial pressure. Two of the most commonly used indicator classes are ruthenium(II) polypyridyl complexes and platinum(II) porphyrins (Quaranta et al. 2012), the latter being increasingly popular due to higher brightness and longer luminescent lifetime. Platinum(II) porphyrin indicators have been more broadly applied during the last decade (Tengberg et al. 2006;Borisov et al. 2008;Larsen et al. 2011), allowing the design of sensors with higher O 2 sensitivity.
We designed three sensors (standard-, mid-and tracerange) that in combination allow measurements of O 2 concentration from $1000 lmol L 21 to $5 nmol L 21 by using a combination of two different luminescent O 2 indicators and two different polymers. The standard-range sensor is based on a platinum(II) tetra(4-fluorophenyl)tetrabenzoporpyhrin indicator (Pt-TPTBPF)  immobilized in polystyrene (PS). The mid-range sensor is based on a palladium(II) tetra(4-fluorophenyl)tetrabenzoporpyhrin (Pd-TPTBPF) (Borisov et al. 2008) immobilized into PS. Finally, the trace-range sensor is based on the same palladium(II)tetrabenzoporpyhrin indicator as used for the mid-range, modified to allow immobilization in Teflon V R AF1600 (AF). Hereafter, we refer to the standard-, mid-and trace-range as Pt-PS, Pd-PS, and Pd-AF, respectively (Table 1).
The Pt(II) indicators typically have a luminescent lifetime $40-70 ls in the absence of O 2 (Oguri et al. 2006;Wang and Wolfbeis 2014). In comparison, the Pd(II) indicators have a much longer lifetime ($350-1000 ls) under the same conditions (Quaranta et al. 2012;Wang and Wolfbeis 2014). The significantly longer lifetime of the Pd(II) indicators makes them ideally suited for highly sensitive O 2 sensors. This is especially true when the Pd(II) indicators are immobilized into highly gas-permeable matrices such as AF, where the O 2 permeability is $136 times higher than that of PS (Arcella et al. 2003;Bikson et al. 2004). An additional benefit of AF is that it has a very high chemical stability. This stability prevents chemical reaction of photosensitized singlet O 2 with the polymer backbone resulting in consumption of O 2 and therefore inaccurate results (Koren et al. 2013).
Compared to the commonly used Pt(II) and Pd(II) porphyrins (Lee and Okura 1997;Amao and Okura 2009), the indicators used in this study rely on tetrabenzoporphyrins. The tetrabenzoporpyhrin indicators feature a strong absorption in the region from nm 600 to 630 nm and an emission in the near infrared (NIR) range (Borisov et al. 2008). The red excitation light and NIR emission have the advantages of low interference with ambient background light and autofluorescence from the excitation light.
The tetrabenzoporphyrin indicators used in this study are directly compatible with the commercially available FiresStingO2-mini optical oxygen meters (Pyro science, GmbH). The FireSting excites the indicators with a 620 nm LED and measures the luminescent lifetime of the indicators using the phase modulation method (Holst et al. 1995;Schm€ alzlin et al. 2005) eliminating the interference by ambient light.

Sensing chemistry and sensor construction feedthrough construction
The indicator dyes were synthesized as described in Borisov et al. (2008Borisov et al. ( , 2009). The Pt-and Pd-TPTBPF indicators show good solubility in PS. However, in order to ensure compatibility of the O 2 indicator with the perfluorinated AF matrix and to avoid dye aggregation, the palladium(II)-benzoporpyhrin was modified with one perfluorooctyl chain via radical perfluoroalkylation, according to (Li et al. 2010), using perfluorooctyl iodide as a reagent and copper as a catalyst in dimethylsulfoxide. Analogously, the Pt(II) complex with one perfluorooctyl chain immobilized in AF was used for investigation of pressure effects (see later).
For all types of sensors, 1% (wt/wt) of the indicator with respect to the polymer was used. For the standard and midrange sensors, the cocktail was a 10% (wt/wt) PS solution in chloroform. For the trace sensors a 5% (wt/wt) solution of AF in octafluorotoluene was used. The sensing "cocktail" was coated onto the tip of an Ø1 mm glass fiber by dipping the tip of the fiber in the sensing cocktail. The tip of the glass fiber had been rounded with a high-temperature flame. Prior to dip coating, the fibers for the PS-based sensors were treated with trimethylchlorosilane, and the AF based sensors were treated with fluoroalkylmethylchlorosilane (Aquaphobe V R CF, Gelest) to ensure good adhesion of the polymers to the glass fiber. The $50 mm long fibers were inserted into a simplex fiber optic connector (Avago Technologies, HFBR-4501Z) and fixed with cyanoacrylate adhesive. The sensor assembly was mated with a specially designed and now commercially available optical feedthrough (Subport, PyroScience, GmbH) (Fig. 1). The feedthrough is based on an Ø3 mm sapphire glass rod acting as a direct coupling between the fiber and the FireSting oxygen meter on the low-pressure side. The feedthrough has a UNF 7/16-20 thread and can penetrate bulkheads up to 30 mm. The feedthrough is designed to withstand a hydrostatic pressure of 600 bar. The sensor assembly is pressed against the sapphire glass rod using a stainless steel dome nut with a 1.5 mm hole at the center (Fig. 1). The end of the feedthrough is inserted into the oxygen meter and fixed using a set screw.

Sensor calibration
Sensor calibration was performed using a digital gas mixer (SensorSense BV) by mixing N 2 (99.999%) with either atmospheric air or 0.1% O 2 in N 2. With the gas mixer and the available gasses, we could obtain O 2 concentrations between $40 nmol L 21 and $280 lmol L 21 with a minimum increase in O 2 of $20 nmol L 21 (208C, Salinity 0), based on the minimum increase in flow and the gasses used for the calibration The reported accuracy of the gas mixer output is 6 1% of given flow rate.
All gas lines were stainless steel or Vitron V R . However, due to O 2 contamination in the source gases, it was not possible to reproducibly maintain O 2 levels below $40 nmol L 21 within the gas mixer and gas lines. Absolute zero was defined from a solution containing 0.5% wt/vol sodium dithionite. Calibration was performed in a 0.5 L stainless steel cylinder filled with MilliQ water. The cylinder was sealed with a stainless steel lid, fitted with the optical feedthrough. The outflow from the gas mixer flushed the water in the cylinder via a G18 stainless steel needle inserted into the cylinder through an Ø2 mm hole in the lid. Typically, a calibration curve was generated from 6 to 10 points across the dynamic range of the respective sensors. Generation of a single calibration curve typically took 3-4 h, independent of the sensor type. Calibration was generally repeated three times.
All calibrations were performed at a fixed temperature by placing the cylinder in a thermostat-controlled cooling bath . During the calibration, the temperature did not fluctuate by more than 6 0.28C. Calibration curves were fitted using the modified Stern-Volmer equation Glud et al. 1996): Where s0 is the luminescent lifetime in the absence of O 2 , s is the luminescent lifetime in the presence of O 2 at a given partial pressure (p), K SV is the Stern-Volmer quenching constant representing the sensitivity, and a is the nonquenchable fraction of the luminescent signal (Glud et al. 1996;Larsen et al. 2011). For in situ measurements, s0 and K SV were adjusted for changes in hydrostatic pressure and temperature as described in the following section.
To allow for two-point calibration during in situ measurements, we determined a fixed a value to each sensors type as derived from multipoint calibrations of the three different sensor types at different temperatures. For the Pt-PS, Pd-PS, and Pd-AF we derived a fixed a values of 0.10, 0.15, and 0.20, respectively. A total of 8-10 sensors of each type, were used to determine an appropriate a value.
For all sensor types, the two-point fit of Eq. 1 provided R 2 values above 0.999 (Fig. 2), when fitted to the multipoint calibrations and the differences in K SV were typically no more than 1%, as compared to multipoint curve fitting with a fixed a value. The good fit across the temperature range also suggests that at the given conditions application of different a values would be negligible. We assume that a is unaffected by changes in hydrostatic pressure (see later).
Partial pressure was converted to concentrations using the solubility coefficients of (Garcia and Gordon 1992) and their corrected Eq. 8 at standard pressure (1013.25 mbar). The glass rod is inserted into a simplex fiber optic connector. A dome nut (b) is used to press the optical connector to the optical feedthrough (c). On the drawing the feedthrough is mounted through the lid of a small pressure vessel (d). The opposite end of the feedthrough is directly attached to the optical oxygen meter (e) on the low pressure side.

Sensor temperature and pressure compensation
In order to compensate for variations in temperature and hydrostatic pressure during in situ measurements, we determined the temperature and pressure effect on s0 and K SV for each of the three sensor types, applying the same calibration procedure as above. For each sensor type, we established calibration curves for three individual sensors using three to four different temperatures from 58C to 208C. The respective changes in K SV and s0 were calculated from the O 2 calibration curves obtained at these different temperatures.
The effect of hydrostatic pressure was investigated by pressurizing the water inside the calibration cylinder (Supporting Information Fig. S1). A pressure sensor with an analogue output (Sick, PBS 6041724) and a multi-fluid hand pump (Enerpac, MP-1000) was connected to the lid through a T-piece connection. From a reservoir connected to the hand pump, water was pumped into the cylinder, increasing the internal pressure in steps of ! 2 bar. The output from the pressure sensors was recorded with a 16-bit D/A converter (Measuring computing, USB-1608FS). Before the cylinder was pressurized it was filled with water via an overflow, assuring that no air was entrapped. During measurements, a glass coated rotating magnet ensured fully homogeneous conditions inside the pressure cylinder.
For each sensor type, three to six different sensors were tested in a pressure range from 1 to 400 bar. All pressure tests were performed at 208C. The change in s0 at increasing pressure was measured in sodium dithionite amended water (0.5% wt/vol).
As the change in K SV at increasing pressure is primarily related to the properties of the matrix material (Glud et al. 1999), we carried out pressure tests for the PS and the AF matrix. For both matrices we used Pt(II) benzoporphyrin as the indicator. For both matrices we used the Pt(II) benzoporphyrin as the indicator. We chose to immobilize the perfluorooctyl modified version of Pt(II) benzoporphyrin into AF, enabling us to carry out the pressure investigations at O 2 levels higher that would otherwise have been possible with the Pd-AF. This was necessary as it proved impossible to reliably fill and maintain O 2 levels inside the cylinder below $2% air saturation, due to O 2 contamination/ingress from the pressure pump.
For the Pt-PS sensors the pressure response was also evaluated at 15 different O 2 levels, ranging from $5% to 100% air sat. For the Pt-AF sensor ten levels between 2% and 12% air saturation were used.

In situ instrumentation electronics
For in situ measurements, we designed a stand-alone instrument for TOP. The instrument contained four channels for optode measurements and two channels for STOX measurements. Data acquisition and hardware control was performed using a single board computer (SBC; fit-PC2i, CompuLab, 2MHz CPU, 2GB RAM), equipped with a 60 GB flash drive. The SBC was installed with a Windows 7 V C operating system, and controlled via a Windows V C remote desktop through a SubConn V R Ethernet bulkhead connector using a laptop or via a wireless connection. A remote desktop was used to operate all SBC programs and initiate data acquisition. Before deployment, the Ethernet communication cable was removed and the instrument operated fully autonomously. After profiling, the data acquisition was stopped and the recorded data were downloaded.
Data from four FireStingO2-Mini oxygen meters were recorded at 1 Hz using the PyroScience Oxygen logger software running on the SBC. The signals from two STOX electrodes were amplified by two custom-made picoamplifiers. The sensor front guard switching was controlled by a custom-made cyclic switch operating with a 60 or 30 s on/ off cycle, depending on deployment. The signals from the STOX amplifiers and the front guard switching were recorded at 120 Hz using a 16-bit A/D converter (DT9816, Data Translation).
During initial configuration as used in the ETNP, pressure, and temperature were recorded by a Conductivity-Temperature-Depth profiler (CTD, Sea-Bird 911plus) deployed together with the TOP instrument. Subsequently, the timestamps of the instruments were aligned. In the ETNP the TOP instrument was mounted horizontally with the sensors pointing downwards. In the BoB, the instrument was mounted vertically underneath the CTD rosette. For both cruises, due to the placement of the instrument with the sensors closest to the bottom of the rosette, the down-cast CTD profiles had the best water exchange around the sensors. We, therefore, only present in situ depth profiles from the down-casts.
For deployments in the OMZ of the Bay of Bengal (see later) the instrument was upgraded with internally mounted pressure sensors (ECO ST1, Sensor Technik Sirnach AG) and temperature sensor (Pt-100, 4-wire, TS100, Sensor Technik Sirnach). The 4-wire signal from the TS100 sensor was recorded directly using one of the FireSting oxygen meters. The signal from the pressure sensor and two externally mounted sensors (conductivity (7-pole, AMT Analysenmesstechnik GmbH) and photosynthetically active radiation (PAR; QSP-2350, Biospherical instrument Inc.) was recorded using an A/D converter (DT9818, Data Translation) along with the signal from the two STOX sensors. The electronics were housed in a titanium cylinder (Ø; 180 mm, length; 760 mm) depth rated for 600 bar. The internal battery package (4 3 4500 mAh) ensured an operating time of > 12 h. The cylinder was fitted with a pressure relief valve to prevent potential pressure buildup inside the cylinder due to temperature increase (Dual O-Ring/SAE Port, DeepSea Power and Light).

In situ measurements
The TOP instrument was successfully deployed at eight stations in the ETNP as part of the Oxygen Minimum Zone Microbial Biogeochemistry Expedition (OMZoMBIE 1) cruise onboard the R/V New Horizon off the coast of Manzanillo, Mexico, in June 2013. Here, we present data from two stations: Sta. 2 (ETNP2; 18855.218 0 N, 108847.971 0 W); Sta. 6 (ETNP6; 18855.145 0 N, 104853.275 0 W). In addition, the upgraded instrument was deployed in January 2014 in the BoB at six stations during a cruise onboard R/V Sagar Kanya from Paradip to Chennai. Here, we present data from one station: Sta. 6 (BoB6, 15899.666 0 N, 87800.264 0 W).
The CTD data was sampled at 25 Hz, and subsequently down-sampled to 1 Hz. During the up-cast the CTD was positioned at discrete depths for 3-5 min to complete several STOX measuring cycles. The optical sensors were calibrated immediately before each deployment using a two-point calibration. The Pt-PS sensor was calibrated in a 0.5% (wt/wt) sodium dithionite solution and in 100% air saturated water. The Pd-PS and Pd-AF sensors were likewise zero calibrated in 0.5% (wt/wt) sodium dithionite. However, the second calibration point was obtained from the Pt-PS in situ measurements. The data from the Pt-PS, Pd-PS, and Pd-AF are presented with no smoothing of filtering. The maximum continuously deployment time for the sensors during the CTD casts were $2 h.
During deployments in the ETNP and BoB the TOPinstrument was mounted on the CTD (Sea-Bird 911plus) equipped with a Sea-Bird Electronics 43 oxygen sensor (SBE43) and Seapoint or WETlabs chlorophyll fluorometer for the ETNP and BoB cruise, respectively.

Sensor performance
Calibration of the three different sensors showed a K SV of 0.017, 0.169, and 2.18 (lmol L 21 ) 21 at 208C for the Pt-PS, Pd-PS, and Pd-AF sensors, respectively (Table 1). These values demonstrate that the sensitivity of the Pd-AF sensor is approximately 100 and 10 times higher than for the Pt-PS and Pd-PS sensors, respectively. The upper limit for the dynamic range of the Pt-PS sensor is $1000 lmol L 21 at salinity 358C and 208C, which is comparable to most commercially available electrochemical and luminescence-based O 2 sensors for oceanographic research (Martini et al. 2007;Tengberg et al. 2006). The upper limits for the dynamic ranges of the Pd-PS and Pd-AF sensors were $50 lmol L 21 and 2 lmol L 21 , respectively (Table 1).
Due to the nonlinear response of the sensors (Fig. 2) the resolution is highest at low O 2 concentrations. Based on the noise of the s0 (three times the standard deviation), the LOD for the sensors is estimated to be approximately 200 nmol L 21 , 50 nmol L 21 , and 5 nmol L 21 for the Pt-PS, Pd-PS, and Pd-AF sensors, respectively (Table 1).
The benzoporphyrin indicators used in this study demonstrate excellent photostability, brightness, and temporal stability (Borisov et al. 2008;Koren et al. 2013). The stability was demonstrated by continuous (1 Hz) measurements in a sodium dithionite solution for 24 h. The change in luminescent lifetime of the three sensors was less than 0.18 3 10 21 , 1.5 3 10 22 , and 7.1 3 10 22 & h 21 for the Pt-PS, Pd-PS, and Pd-AF sensors, respectively; for all sensors this is below the respective LOD. The signal intensities change by 0.31 3 10 22 , 2.7 3 10 22 , 6.8 3 10 22 & h 21 , respectively. The somewhat larger decrease in the signal for the intensity-based signal vs. the lifetime-based signal demonstrates the advantage of luminescent lifetime as compared to absolute intensity measurements. However, for both measurements, the stability was excellent.
The potential drift in K SV for the Pd-AF sensors was investigated by repeating a multipoint calibration three times during 12 h at 158C. Fitting the data points from the three calibration curves with a two-point calibration, the calculated mean K SV was 2.124 (lmol L 21 ) 21 SD 6 0.0102 and did not show any time-dependent drift. The K SV calculate based on the multipoint calibration (with fixed a) did typically not deviate by more than 1%, as compared to the two-point calibration.
For measurements at 500 lmol L 21 with the Pd-AF sensors, an increase in the K SV by 1% will induce an error of less than $5 nmol L 21 . Thus, using a two-point calibration compared to a multipoint will not adversely affect the quality of the O 2 measurements. Calculating the O 2 concentration for each for the three curves using the mean K SV demonstrated a good accuracy as compared with gas mixer output. The difference between the mean calculated O 2 concentration for the calibration curves and the gas mixer output was lower than $12 nmol L 21 at concentrations below 1 lmol L 21 . The standard error of the mean for the three curves was generally below 10 nmol L 21 , with values > 5 nmol L 21 at concentrations below 500 nmol L 21 . The performance of the Pt-PS and Pd-PS sensors demonstrates the same degree of accuracy/precision and stability, thus all presented sensors can be used for typical CTD profiling with no need for drift correction. The response time for optical sensors depends on the thickness of the sensing layer as this controls O 2 diffusion into the sensing layer. Therefore, bright indicators allow for the construction of fast sensors, as even thin sensing layers have a sufficiently high luminescence for detection by the oxygen meter. The bright Pt-PS sensor enabled typical t 90 response times of $5 s, whereas the Pd-PS and Pd-AF had response times of approximately 20-30 s (Table 1). It should be noted that the response time represents the time for a 90% signal change and that the sensors will respond dynamically when profiling through an O 2 gradient. The response time of the Pd-PS and Pd-AF sensor can be increased by applying thinner sensing layers.

Effect of temperature and hydrostatic pressure on sensor response
The signal of all luminescent indicators is affected by changes in temperature due to thermal quenching (Quaranta et al. 2012) and correction of the sensor signal is essential for in situ measurements with changing temperature. For optical O 2 sensors calibrated by the modified Stern-Volmer equation (Eq. 1) the effect of temperature is accounted for by independent temperature correction of s0 and K SV (Rickelt et al. 2013).
For the three different sensors both s0 and K SV showed a linear temperature dependence (R 2 > 0.99) in the range of 5-208C ( Fig. 3; Table 1). The decrease in s0 ranged between 20.078% C 21 and 20.255% C 21 , with the Pt-PS sensor demonstrating the lowest influence, followed by the Pd-AF and Pd-PS sensors, respectively. The decrease of s0 with rising temperature is related to thermal quenching of the phosphorescence of indicators rather than to changes in physicochemical properties of the immobilization matrix Rickelt et al. 2013;Mcneil and D'asaro 2014). This is consistent with the fact that the two Pd(II)-porphyrin-based sensors had similar s0 temperature coefficients that were approximately three times higher than the Pt(II)-dye based sensor (Table 1). The absolute values align with previously published values for temperature coefficients of these indicators (Borisov et al 2008). The temperature coefficient for K SV ranged from 0.62%8C 21 to 0.96%8C 21 with the Pd-AF sensor demonstrating the lowest sensitivity, followed by the Pt-PS and Pd-PS sensors, respectively (Table 1). The increase in K SV at increasing temperature is primarily due to the increasing permeability of the immobilization matrix . Therefore, the temperature coefficients of K SV of the two PS-based sensors were similar (Table 1).
The effects of hydrostatic pressure on the performance of optical sensors have not been reported in great detail, and to our knowledge only a few studies have previously investigated pressure effects on O 2 optodes, with two lab investigations using optodes with PS as the matrix material (Glud et al. 1999;Tengberg et al. 2006) and two studies on the Aanderaa optodes response (Uchida et al. 2008;Mcneil and D'asaro 2014;Bittig et al. 2015). However, since hydrostatic pressure effects depend on both the mechanical, physicalchemical design of a sensor, O 2 solubility, and optical feedthrough, the hydrostatic effect needs to be evaluated for each specific design. For all sensors we observed a decrease in s0 with increasing pressure. Both PS-based sensors demonstrated a linear decrease in s0 and signal intensity as a function of increasing pressure in the range from 0 to 400 bar (Fig. 3). The decrease in s0 amounted to 0.22% and 0.24% pr. 100 bar increase, for the Pt-PS and Pd-PS sensor, respectively. For measurements near the sensors LOD an increase in pressure from 0 to 400 bar would equate to an apparent increase in O 2 concentration of $0.7 lmol L 21 and $0.1 lmol L 21 (58C, salinity 35) for the Pt-PS and Pd-PS sensors, respectively.
The sensor based on AF did not show the same linear decrease in s0 at increasing pressure. Instead, we consistently observed a small initial increase in s0 and signal intensity as the pressure was increased to $10 bar. However, this increase was subsequently followed by a pressure dependent linear decrease (Fig. 3f). The initial increase in s0 corresponded to a $1% increase and corresponds to an apparent underestimation in O 2 concentration of $4 nmol L 21 (at 58C, Salinity 35). The linear decrease in s0 above $10 bar corresponds to 0.95% pr. 100 bar (Fig. 3f), which is somewhat higher than for the PS-based sensors. For measurements near the sensors LOD an increase in pressure from 0 to 400 bar would correspond to an increase in apparent O 2 concentration of $15 nmol L 21 at (58C, Salinity 35). We found that the initial increase in s0 of the Pd-AF sensor could be eliminated via coating the outer tip with an additional layer of AF containing TiO 2 (10% wt/wt). The additional coating with TiO 2 scattering layer significantly increases the signal intensity of the sensor as compared to the uncoated sensor. Increasing the signal intensity makes the sensor less affected by signal loss Fig. 4. Effect of hydrostatic pressure on s and KSV for the PS and AF matrices with immobilized Pt(II) tetrabenzoporpyhrin. (a) and (b) example of changes in s at $70% air sat., and $5% air sat., for PS matrix and AF matrix, respectively. (c) and (d) calculated effect of changes in hydrostatic pressure on K SV for the PS matrix and AF matrix. The change in K SV calculated for the PS matrix was applied to pressure compensates the Pt-PS and Pd-PS sensors and the effect on the AF matrix to compensate the Pd-AF sensor.
in the optical pathway due to the increased hydrostatic pressure.
As for temperature, the effect of hydrostatic pressure on K SV is predominantly determined by the matrix polymer (Glud et al. 1999). We therefore only determined changes in K SV for the PS and AF matrices using the Pt-TPTBPF and its perflurooctyl-modified version, respectively. For both matrices, the change in K SV was determined based on the change in s as function of pressure at ten to 15 different O 2 levels. For the PS matrix we observed a nonlinear increase in s at increasing pressure. This trend was observed across the entire concentration range (Fig. 4a). Likewise, the K SV also followed a nonlinear trend, with decreasing K SV at increasing pressure (Fig. 4), corresponding to a decrease of approximately 3.21% pr. 100 bar increase, assuming a linear trend (Table 1).
For the AF matrix, s decreased linearly with increasing pressure across the entire concentration range (Fig. 4b). Correspondingly, the K SV increased linearly with increasing hydrostatic pressure, corresponding to a change of 2.18% pr. 100 bar (Table 1). The apparent change in O 2 concentration due to pressure effects on the K SV value for the Pd-AF sensor at 50, 200, and 400 bar amounts to an overestimation of $35 nmol L 21 , 200 nmol L 21 , and 400 nmol L 21 when exposed to 2 lmol L 21 . None of the sensors exhibited hysteresis during repeated pressure cycles or during compression vs. de-compression.
The effect of hydrostatic pressure on the sensor matrix is a complex interaction between several factors and a detailed evaluation of the respective components is outside the scope of this study. We note, however, the opposing pressure effects on K SV for the PS and AF matrices, which underpin the need for careful evaluation of pressure effects on different sensor types prior to in situ application.
In summary, hydrostatic pressure had a small and reproducible influence on the sensor responses. However, similar to the temperature cross-talk, the predicable behavior allowed for effective pressure compensation and high precision measurements at changing hydrostatic pressure. Furthermore, the response at increasing hydrostatic pressure for the PS-based sensor agree with the previous findings of ( Glud et al. 1999;Tengberg et al. 2006;Uchida et al. 2008;Mcneil and D'asaro 2014) for sensors with comparable sensitivity.

In situ OMZ measurements
The calibrated O 2 signals of the SBE43 and the Pt-PS sensors correlated extremely well in all deployments (Fig. 5). However, for all deployments the SBE43 was found to have a zero-offset ranging from $0.61 lmol L 21 in BoB to $2.2 lmol L 21 in the ETNP. Such zero-offset has previously been observed for commercially available oceanographic O 2 sensors, such as the SBE43 (Revsbech et al. 2009;Thamdrup et al. 2012;Ganesh et al. 2015).
The down-cast O 2 distribution at station ETNP2 as obtained with the SBE43 exhibited a steep decline in the O 2 concentration from the maximum concentration of $220 lmol L 21 in the chlorophyll a (Chl a) maximum at $55 m depth, reaching the zero-offset ($2.2 lmol L 21 ) at $160 m (Fig. 6a). Below 160 m, the concentration from the SBE43 remained relatively stable, increasing slightly with depth to reach 2.8 lmol L 21 at 800 m (Fig. 6a). At $780 m a small increase in O 2 concentration of $0.8 lmol L 21 above the zero-offset was observed. Apart from the small increase at 780 m, the O 2 profile from the SBE43 did not show a distinct structure and the mean O 2 concentration from 160 to 800 m, as resolved by the SBE43 sensor, was 2.25 lmol L 21 (SD 6 0.20 lmol L 21 ).
In comparison, the concentration measured with the standard range Pt-PS sensor generally showed levels around the LOD (0.2 lmol L 21 ) from $230 to 660 m. But below 600 m the O 2 concentration gradually increased to a maximum of 0.59 lmol L 21 at 770 m. The Pt-PS data demonstrate that a well-calibrated luminescence-based sensor can provide improved O 2 resolution, compared to the standard O 2 sensor, such as the SBE43 sensor.
The Pd-AF trace sensor resolved significant spatial variation in sub-micromolar O 2 concentrations in the depth interval from 125 to 800 m (Fig. 6b). At 125 m, the concentration was 2.12 lmol L 21 , but dropped to $25 nmol L 21 at 138 m, overlapping with the position of the secondary Chl a The comparison shows an excellent agreement between the sensors, but also show a zero-offset of 0.61 lmol L 21 for the SBE43 sensor. It should be noted that the SBE43 offset could be largely eliminated if the sensor was recalibrated. maximum (Fig. 6a). The concentration increased to 2.15 lmol L 21 at 146 m before gradually decreasing to concentrations below the LOD (5 nmol L 21 ) of the sensor at 213 m. From 213 to 390 m, the measured concentration was generally below the sensor LOD. However, numerous microstructures with "spikes" of 2-3 nmol L 21 above the "background" concentration could be observed. At 390-490 m O 2 concentrations increased to 7-11 nmol L 21 and the frequency of the spikes became more regular. Below 490 m, larger O 2 intrusions could be observed. These intrusions spanned 60-80 m with maximum O 2 concentrations above 50 nmol L 21 . At 777 m O 2 concentration increased to 780 nmol L 21 . Looking closely at the Pt-PS profile in Fig. 6a, the contours of this rather large O 2 intrusion can be observed as an O 2 increase to 0.82 lmol L 21 , while this increase is barely visible in the SBE43 data. Two casts were performed at this station, both confirming the existence of the large sub-micromolar O 2 intrusions in the core of the OMZ.

Trace O 2 levels as resolved by optodes and STOX sensors
During the BoB cruise the TOP instrument was deployed with two Pd-PS and two STOX sensors. Generally, the O 2 concentration in the BoB rarely reached consistent stable concentrations below $50 nmol L 21 (Bristow et al. unpubl.) and extensive O 2 dynamics at concentrations < 3 lmol L 21 could be observed (Fig. 7). The measured concentrations from the two Pd-PS sensors demonstrated an excellent agreement with a mean difference between the two sensors below the sensors LOD. The dynamics and concentrations measured with the two optical sensors agreed well with the measurements from the two STOX sensors (Fig. 7). Generally, the agreement between concentrations measured by the STOX and Pd-PS sensors was better that 2%, in the range from the sensors LOD up to 20 lmol L 21 .

Discussion
It is apparent that standard range Pt-PS and SBE43 sensors do not have sufficient sensitivity for resolving O 2 concentrations and dynamics within the core of the OMZ as revealed by the highly sensitive Pd-AF sensor. Relying solely on the SBE43 data from the ETNP6, one could only infer that the O 2 concentration was below $2 lmol L 21 , in the depth range from $130 m to 800 m at Sta. 2, but otherwise uncertain of the true O 2 concentrations and of any small scale dynamics occurring.
However, measurements with the trace Pd-AF sensor demonstrate that water masses from approximately 225 m to 300 m had O 2 levels at the LOD of the sensor ($5 nmol L 21 ) and that the remaining water column had O 2 concentrations in the core of the OMZ ($550 m) as high as 73 nmol L 21 , due to large-scale intrusions spanning 60-80 m. Closer inspection CTD data demonstrated that the O 2 intrusions in all cased coincided with increases in the sigma-theta value, suggesting that the OMZ core is composed of several water masses with different O 2 concentrations.
To obtain high precision in situ measurements it is important that sensor signals are effectively compensated for changes in temperature and hydrostatic pressure. As demonstrated by the lab-based temperature and pressure tests, for measurements in OMZ regions with typical pressure and temperature changes of $100 bar and 208C during CTD casts, the decrease in temperature would have the greatest effect on the sensor signal. However, such effects are highly predictable. Of the three sensors presented here, the Pd-PS (mid-range) sensor is most affected by changes in temperature (Table 1), as is clearly observed in the luminescent lifetime (s) in situ data (Fig. 8).
During the deployment in the ETNP, both the temperature and pressure changed as the CTD was raised stepwise from 700 m to 350 m in waters with an O 2 concentration below the LOD of the Pd-PS sensor ($50 nmol L 21 ). It can be observed that the measured in situ s decreased $4 ls from $460 ls to 456 ls as the temperature increased by $48C and the pressure decreased by $35 bar. Under the given pressure and temperature variations, a change in luminescent lifetime of $4 ls would correspond to an apparent change in O 2 concentration of $80 nmol L 21 . The change is, however, fully predictable, as observed by the s0(t,p) representing the calculated s0 lifetime at the given in situ temperature and pressure. We assume that the influence of temperature on the hydrostatic pressure effect is insignificant, and therefore we only conducted pressure testing at one temperature. The excellent agreement between the measured and calculated s, indeed suggests that potential temperature effects on the pressure compensation coefficients can be ignored under the given in situ conditions in the ETNP and BoB. A potential cross effect of temperature on the pressure compensation cannot be fully excluded and should, if relevant for the measurements, be further investigated.
The excellent agreement between the Pt-PS and SBE43 (Fig. 5) and the Pd-PS and STOX data (Fig. 7) also clearly validate the procedure for compensating sensor signals for pressure and temperature interference. That such compensation The STOX sensors were independently calibrated against the SBE43 sensor and operated out of phase to increase the temporal/spatial resolution. For the entire cast both Pd-PS sensors show excellent agreement and generally also a good agreement with the two STOX sensors. It can be observed on the inset that the O 2 concentration from 220 m to 300 m never consistently reach values below 50 nmol L 21 . Comparing the Pd-PS and STOX measurements, it can be observed that the relative slow measuring time of the STOX sensors is inadequate for quantifying the extensive small scale O 2 dynamics typically observed in BoB. is important is exemplified by the data in Fig. 5. If no compensation had been applied to the Pt-PS sensor signals, the sensor would have provided an O 2 concentration that underestimated the actual concentration by $20 lmol L 21 at 1400 m depth.
The pressure and temperature compensation coefficients for different sensors of the same production batches generally demonstrated a good agreement, e.g., for six Pt-PS sensors from two different batches the mean change in s0 was found to be 20.237 (SD 6 0.13 3 10 22 ) and 20.235 (SD 0.96 3 10 22 ), respectively. This suggests that the coefficients obtained for a single sensor can be applied to a batch of sensors and presumably the constants are universal for a given sensor type. This is further supported by the fact that in situ data obtained with the Pt-PS sensors in ETNP and BoB both showed an excellent agreement with the respective SBE43 sensor, even though the Pt-PS sensors for the two cruises were from different batches and the pressure/temperature compensation data were acquired from a third batch of sensors. However, we recommend that the temperature and hydrostatic responses of the different sensor types and batches are checked regularly to quantify batch variation and potential effects of polymer aging. Likewise, should the a value for each sensor type be assessed regularly, if the sensor are to be calibrated with a two-point calibration.
Correct calibration is essential for optimal sensor performance. As for the STOX sensors, the trace O 2 optode can be calibrated using a two-point calibration routine. One point can be recorded in sodium dithionate-amended water and the second point can be obtained from either deck calibration in a solution of known concentration or from in situ intercalibrations against other calibrated sensors (Revsbech et al. 2009). However, for such two-point calibrations the value of a must be determined from a multipoint calibration. As previously mentioned, the value of a can be determined for the specific sensor type and thus is in not necessary to determine a for individual sensors, as the a value is primarily related to the immobilization matrix.
Most available O 2 sensors (such as the SBE43), however, do not have the accuracy to allow direct in situ two-point calibrations of the trace sensors. In such cases, sensors with different dynamic ranges would have to be used, as was done here. Alternatively, the sensors can be calibrated using a ship-based two-point calibration. The two-point calibration should be performed with sodium dithionite and a premixed gas mix, e.g., 0.5% air saturation. Achieving the 0.5% air saturation with the compresses gas would require the use of a calibration vessel with minimized exposure to ambient airsuch as shown on Supporting Information Fig. SI1-but otherwise the procedure would be identical to a calibration of a normal range sensor. Such calibration will in our opinion provide a reliable and relatively easy way to perform calibration of the trace sensor, provided that the data subsequently are correctly compensated for changes in temperature and pressure. We recommend that such two-point calibrations are performed as frequent as possible, ideally before each deployment. As noted previously, determination of the K SV from a two-point calibration demonstrated that the K SV did not typically deviate by more than 1%, when compared to the K SV calculated from a calibration using eight or more points. For measurements at 500 lmol L 21 with the Pd-AF sensors an increase in the K SV by 1% will induce an error of less than $5 nmol L 21 . Thus, using a two-point calibration compared to a multipoint will not adversely affect the quality of the O 2 measurements.
However, pre-deployment ship-based multipoint calibration at fixed temperature is the preferred calibration approach. Such calibrations can ideally be performed with a gas mixer to obtain a multipoint calibration in the desired concentration range. A digital gas mixer similar to the one used in the presented study can be acquired for $$5000.
The Teflon AF1600 TM matrix used for the trace sensors in this study represents an almost ideal polymer for high sensitivity sensors, due to its high O 2 permeability and good stability. However, to improve the stability of the trace sensors sensing "cocktail," we have investigated the use of the perfluorinated polymer Hyflon AD60 as an alternative to the trace sensor. The AD60 demonstrates better dye solubility and very low temperature dependence (Lehner et al. 2015). were recorded during an up-cast where the O 2 concentration was below the LOD of the sensor. The CTD was stopped at $5 min intervals at eight depths from 700 m to 350 m. s represents the measured in situ luminescent lifetime of the sensor. s0(t,p) represents s0, obtained from predeployment zero calibration, compensated for in situ temperature and pressure (p). From the measurements it is clearly observed how the luminescent lifetime of the Pd-PS sensor is affected by temperature (and to a less extent pressure), but this effect can effectively compensated for, as seen by the excellent agreement between the measured s and the calculated s0(t,p).
However, with a permeability $7 times lower than the AF1600 (Arcella et al. 2003), Hyflon AD60 has a somewhat higher LOD of $10 nmol L 21 . Furthermore, initial pressure tests demonstrated that the AD60 matrix has very low dependence on hydrostatic pressure, lower than that of the AF and PS. The low temperature and pressure dependence of the matrix could potentially provide robust midrange O 2 sensors by immobilizing the Pt(II) porphyrin indicator in the AD60 matrix.
As for other aquatic sensors, biofouling could compromise sensor performance. This is especially true for trace in situ O 2 measurements, as O 2 consumption within any biofilm will significantly affect the measurements. To some extent the biofouling can be reduced by coating the sensing chemistry with a layer of antifouling material (Navarro-Villoslada et al. 2001) or by the direct addition of silver nanoparticles to the sensor matrix (Wencel et al. 2010), thereby significantly enhancing the operational lifetime of the sensor. In addition, sensors can be mechanically cleaned using a water jet (Tengberg et al. 2006) or through the application of sensor wipers (Delauney et al. 2010). However, for short term profiling in OMZ's, biofouling is not a major problem. Furthermore, biofilm formation on the perfluorinated materials (AF and Hyflon) is likely to be lower that for the standard polymers, such as PS.
With the TOP instrument and with the highly sensitive Pd-AF trace O 2 sensors, we have for the first time demonstrated the application of optodes for quantification of ultra-low O 2 concentrations, even at high hydrostatic pressure. With the trace O 2 sensors we have also for the first time documented high resolution measurements of sub-micromolar O 2 intrusions in the core of OMZ's. The extent and mechanism of such O 2 intrusion events in OMZs, however, remains to be explored. Doing so is critical, as variations in O 2 availability in the sub-micromolar range have been shown to have significant effects on rates of key microbial biochemical transformations (e.g., denitrification, anaerobic ammonium oxidation) in OMZs .
Our data demonstrate that improved insights into O 2 concentrations and dynamics in low-oxygen environments can be gained using trace O 2 sensors, as compared to traditional O 2 sensing technologies. Our observations in OMZs also highlight the need for additional trace sensor measurements to better understand the relationships between sub-micromolar O 2 dynamics in OMZ's and O 2 -sensitive microbial processes.

Comments and recommendations
Compared to STOX sensors, the main advantages of the trace optodes is the simple production procedure combined with comparable or better response times and LOD. Calibration of the trace sensors is relatively simple and can be done using standard gas mixers or using pre-mixed gases. The described optodes show a predictable and reversible interference from temperature and hydrostatic pressure that can be effectively accounted for. One of the drawbacks of the optical sensors is the limited dynamic range of the specific sensors. Therefore, several sensors are required to fully cover the concentration range encountered in OMZ's. However, in most instances, commercially available O 2 sensors that are integrated in standard CTD instruments will adequately resolve O 2 concentrations down to $1-2 lmol L 21 . We have, therefore, also developed a one-channel mini-TOP instrument that can be equipped with a single trace O 2 sensor that can be directly interfaced to a free analogue input channel of standard CTD's. Alternatively, measurements can be stored using the internal data logger on the mini-TOP. We envision that such mini-TOP instruments could be added to Argo floats, gliders or AUV's (Autonomous underwater vehicles) to provide sub-micromolar O 2 measurements in a wide range of O 2 -depleted environments. However, for longterm deployments on e.g., Argo floats the sensors drift and stability needs to be better characterized.
In this study, we have used the Pt(II)-and Pd(II) tetrabenzoporphyrins indicators. However, these indicators can be substituted with the commercially available Pt(II) and Pd(II) complexes (5,10,15,20-tetrakis-(2,3,4,5,6-pentafluorphenyl)porphyrin). Both indicators show good solubility in PS and Hyflon AD 60 matrices without further chemical modification (Lehner et al. 2015). One advantage of using the pentafluorophenyl-substituted porphyrins indicators is the longer luminescent lifetime when immobilized in the same polymers compared to the benzoporphyrins and hence an increased sensitivity. With such sensors and low noise readout devices (Lehner et al. 2015) a LOD of $0.1 nmol L 21 become possible. However, even with low-cost color ratiometric methods (Stich et al. 2009;Larsen et al. 2011;Schutting et al. 2014;Santner et al. 2015), a LOD of $0.5 nmol L 21 can be achieved. It should also be noted that optical feedthroughs for standard fiber optic cables rated for operation at full ocean depths are available from a range of manufacturers. Thus, multiple options exist for designing in situ sensors and instruments for trace O 2 sensing.
Laboratory experiments suggest that aerobic respiration can persist even at O 2 concentrations down to 3 nmol L 21 (Stolper et al. 2010). Thus, the LOD of 5 nmol L 21 of the trace sensors presented here may still be too high to fully document relevant O 2 dynamics in OMZ's. However, by using optical indicators with even longer luminescent lifetimes than Pd(II), such as boron-and aluminum chelates, sensors with LOD in the low pmol L 21 range could be realized (Lehner et al. 2014).