Lab-on-chip analyser for the in situ determination of dissolved manganese in seawater

A spectrophotometric approach for quantification of dissolved manganese (DMn) with 1-(2-pyridylazo)-2-naphthol (PAN) has been adapted for in situ application in coastal and estuarine waters. The analyser uses a submersible microfluidic lab-on-chip device, with low power (~ 1.5 W) and reagent consumption (63 µL per sample). Laboratory characterization showed an absorption coefficient of 40,838 ± 1127 L⋅mol−1⋅cm−1 and a detection limit of 27 nM, determined for a 34.6 mm long optical detection cell. Laboratory tests showed that long-term stability of the PAN reagent was achieved by addition of 4% v/v of a non-ionic surfactant (Triton-X100). To suppress iron (Fe) interferences with the PAN reagent, the Fe(III) masking agents deferoxamine mesylate (DFO-B) or disodium 4,5-dihydroxy-1,3-benzenedisulfonate (Tiron) were added and their Fe masking efficiencies were investigated. The analyser was tested during a deployment over several weeks in Kiel Fjord (Germany), with successful acquisition of 215 in situ data points. The time series was in good agreement with DMn concentrations determined from discretely collected samples analysed via inductively coupled plasma mass spectrometry (ICP-MS), exhibiting a mean accuracy of 87% over the full deployment duration (with an accuracy of > 99% for certain periods) and clear correlations to key hydrographic parameters.

tection of Mn was characterized with regard to its sensitivity and selectivity. Calibrations were obtained using both a double beam spectrophotometer (Shimadzu UV-1800) and the LoC device (Fig. 1).
Absorbance values obtained with the spectrophotometer (blue) and the medium channel of the LoC analyser (black) exhibited a linear relationship over the concentration range of 0-1 µM and 0-3 µM Mn(II), respectively. In contrast, absorbance values measured with the long channel of the LoC analyser (red) became non-linear above 0.5 µM Mn(II) (exponential fit shown in Fig. 1) due to too high absorbance values and therefore deviation from Beer-Lambert law and also exhibited a lower precision when measurements were repeated (e.g. for a 250 nM Mn(II) standard solution: R.S.D. Long channel = ± 5.4%, R.S.D. Medium channel = ± 2.8%; n = 20; error bars shown in Fig. 1). Therefore, the medium measurement channel was chosen for further experiments and in situ quantification of Mn(II). Linear slopes obtained with a benchtop spectrophotometer and the medium channel of the LoC analyser were distinct by a factor of ~ 3, with slopes of (4.503 ± 0.102) × 10 -4 L⋅nmol -1 (R 2 = 0.997) and (1.413 ± 0.039) × 10 -4 L⋅nmol −1 (R 2 = 0.991), respectively (Fig. 1), scaling with the different lengths of the optical paths, with a ~ 3 times longer path for the spectrophotometer (10 cm) than for the LoC analyser (3.46 cm). Normalized with respect to the length of the optical path, an absorption coefficient of 45,030 ± 1020 L·mol −1 ·cm −1 at 562 nm was obtained for the benchtop spectrophotometer, which is in good agreement with reported values of 44,000 L·mol −1 ·cm −128 and 46,000 L·mol −1 ·cm −126 . For the medium measurement channel of the LoC analyser a somewhat lower absorption coefficient of 40,838 ± 1127 L·mol −1 ·cm −1 was obtained, which can be attributed to the difference between peak emission wavelength of 575 nm of the LED and the absorption maximum of 562 nm of the Mn(PAN) 2 complex. However, the reported value here is still significantly higher than those reported for other in situ Mn analysers using the PAN method, such as for the METIS analyser with 8000 L·mol −1 ·cm −124 and the SCANNER analyser with ~ 20,000 L·mol −1 ·cm −1 (estimated from calibration curve in Chin et al. (1992) 30 ). The enhanced sensitivity of the LoC analyser here might be attributed to the high sample to reagent ratio of 8.8:1. In the METIS and SCANNER analysers the samples were mixed with reagent in a 1:1 ratio and a 5:1 ratio, respectively, resulting in reduced sensitivity. All presented values do not include any correction/compensation of the dilution factor, as the different dilutions are a characteristic feature of each individual device which determines the sensitivity. Additionally, the employed optical detection system (e.g. the LED emission spectrum and its overlap with the optical absorption peak of the Mn(PAN) 2 complex) has an impact on the sensitivity. The detection limit (LOD) for the medium channel of the LoC device was 27 nM Mn(II), defined as three times the standard deviation of the analysis of a Mn(II) blank solution (n = 13) divided by the slope of the linear fit, and was within the LOD range reported for SCANNER (between 15 and 48 nM) but lower than the LOD of METIS (77 nM). The LoC analyser is therefore suitable for quantification of DMn concentrations in many freshwater  30 ) was stable for 4 weeks. Visual examination showed precipitation of orange coloured PAN crystals when stored beyond 4 weeks. As any kind of particle formation needs to be avoided when using microfluidic technology, because of a high vulnerability to clogging in the narrow channels (e.g. 300 µm diameter), it was necessary to adapt the reagent composition with respect to the surfactant. Therefore, two different batches of PAN reagent were prepared; (a) with 2% m/v of the ionic surfactant SDS 24,33 which facilitates faster reaction rates compared to Triton-X100 but lower molar absorbances 34 and (b) with an increased concentration of 4% v/v of Triton-X100 33 in order to provide a higher number of non-ionic surfactant molecules for micelle formation. Both reagents showed good stability as no particle formation was observable upon visual examination even after several months of storage, which is an essential requirement for long-term remote deployments. The response of the LoC analyser using reagents containing 2% v/v and 4% v/v Triton-X100 as well as 2% m/v SDS was tested (Fig. 2).
Processed absorbance values from the medium measurement channel were obtained for 15 consecutive analysis of a 500 nM Mn(II) standard solution with each reagent (Fig. 2A). For both reagents containing Triton-X100, absorbance values showed little variation (R.S.D. = ± 2.9%) indicating stable sensitivity over time. In case of the SDS surfactant, the absorbance decreased over time, resulting in a loss of sensitivity. Figure 2B indicates that the reduced sensitivity over time of PAN reagent prepared with SDS was linked to a decrease in the overall PD output, whereas a stable signal was obtained using PAN with Triton-X100. This may be caused by the SDS mediated formation of a light absorbing coating on the optical windows which separated the flow path from LEDs and PDs. Consistent with this explanation, rigorous flushing with 0.1 M sodium hydroxide solution brought the PD output back to the initial value. Sodium dodecyl sulphate forms a precipitation at temperatures < 15 °C, inhibiting its use in field-deployable systems for temperate and polar waters. Clinton-Bailey et al. (2017) 35 suggested therefore the use of the dispersant polyvinylpyrrolidone (PVP) in microfluidic devices. However, PAN was not soluble in aqueous solutions containing PVP, even after stirring for one day at 80 °C. Therefore, the use of a PAN reagent containing 4% v/v Triton-X100 was chosen as optimum reagent solution for environmentally relevant conditions, ensuing stable sensitivity and no observable precipitation of PAN crystals over a period of several months.
Iron interference with PAN Mn analysis. The metals iron, zinc, nickel, copper and cobalt show significant interferences with the PAN method at a wavelength of 562 nm, when prevalent in a free ionic form at equimolar con- Calibration graphs for the detection of Mn using the PAN method (0.8 mM PAN in 4% v/v Triton-X100) acquired with the LoC analyser (long channel: 91.6 mm; medium channel: 34.6 mm; absorbance values were processed against a sample blank without the addition of PAN reagent) and a benchtop spectrophotometer (10 cm path length). Absorbance spectra from the spectrophotometer were measured against a reagent blank and processed at the absorption maximum of 562 nm.  Fig. 3) featured a broad absorption band at wavelengths > 650 nm (with an absorption maximum at 768 nm) as well as an increased absorbance at 562 nm compared to the 1 µM Mn(II) standard solution (black, Fig. 3). The absorbance at 562 nm of the mixed Mn(II) and Fe(III) standard was equivalent to a processed Mn(II) concentration of 1.2 µM, and thus an Mn overestimation of 20%, which is in agreement with the value reported by Chin et al. (1992) 30 . With the use of DFO-B as Fe masking agent, the absorption spectrum of the mixed Mn(II) and Fe(III) standard (dashed line, Fig. 3) featured no absorption band at wavelength > 650 nm indicating efficient Fe(III) masking capability. However, the absorbance at 562 nm decreased compared to the 1 µM Mn(II) standard, resulting in a Mn(II) underestimation of ~ 15%. It appeared that the DFO-B removes part of the Mn(II) from the PAN accessible pool via DFO-B promoted oxidation to Mn(III) and subsequent stabilization of the higher oxidation state 38 . An  www.nature.com/scientificreports/ underestimation of in situ determined DMn concentrations using the PAN method with DFO-B compared to ICP-MS measurements of discrete samples was also found for the METIS analyser 24 . Whilst this was ascribed to internal hardware issues rather than to DFO-B related issues, it may have been a combination of both based on our findings. It was further observed that the Fe masking strength of DFO-B decreased over time, when stored as mixed PAN/DFO-B reagent at room temperature (Fig. 4). Three days after the preparation of the mixed PAN/DFO-B reagent, any Fe(III) interference was effectively suppressed, also when Fe(III) was present in large excess (up to 20 times of the Mn(II) concentration) in a 500 nM Mn(II) standard solution, for reagents stored at room temperature (black) and at 7 °C (red). The measured absorbance values were comparable to a standard solution without Fe(III) addition. Increased absorbance values were observed for the analysis of standards containing 5 µM and 10 µM Fe(III) after storage for 8 days at room temperature, and degradation of DFO-B led to observable Fe interferences beyond 8 days even at equimolar concentrations of Mn(II) and Fe(III) (500 nM of each). Storage at lower temperatures impeded the degradation of DFO-B and up to 5 µM Fe(III) was effectively masked even 15 days after reagent preparation.
Similarly to DFO-B, the catechol type Fe complexing agent Tiron showed effective Fe masking capability. No absorption band was observable at wavelength > 650 nm (Fig. 3, dotted line). However, a lower absorbance at 562 nm was found for the 1 µM Mn(II) + 1 µM Fe(III) standard solution when applying Tiron compared to the 1 µM Mn(II) standard analysed without Tiron. This suggests that Tiron also complexes Mn(II), resulting in an underestimation of Mn(II).
These findings suggest that DFO-B and Tiron have drawbacks as masking agent for in situ determination of Mn(II) using the PAN method; because of (1) a potential underestimation of Mn(II) and (2) the limited life time of DFO-B. Degradation might be a particular problem for remote in situ deployments over extended periods of weeks to months. However, in most natural waters, such as the Kiel Fjord, the metals which are interfering with the PAN method occur at low concentrations compared to Mn(II) (see next section). Furthermore, these trace metals are naturally prevalent in seawater as strongly bound organic complexes 39,40 , which limits their availability to form a complex with PAN reagent over short equilibration time periods 30,41 . With this critical evaluation we justify the application of a measurement protocol without any Fe masking agent at low Fe(III) concentrations as this would be linked to an underestimation of DMn. Therefore, we deployed the LoC analyser in an estuary, Kiel Fjord, without addition of a masking agent to the PAN reagent and evaluated the outcomes through LoC measurement validation against ICP-MS.
In situ time series in Kiel Fjord. Hydrographic setting. The performance of the DMn in situ analyser was evaluated during a deployment in the Kiel Fjord from October 22 to November 17, 2018. Kiel Fjord forms part of the southwestern Baltic Sea, and is subject to anthropogenic perturbations due to shipping and shipbuilding activities and discharges from a population of ca. 250,000 in the surrounding area. The main sources of freshwater input include rainwater run-off from the city of Kiel, the Schwentine River which is located at the eastern shore of the inner Kiel Fjord, and the Kiel-Canal which represents one of the busiest artificial waterways worldwide, located at the western shore of the fjord. The Kiel Fjord has a mean depth of ~ 13 m, with water level changes of up to ± 1 m caused by winds and pressure gradients over the Baltic Sea. During the period of the deployment, a storm flood approached Kiel Bay in two phases with a water level rise of up to 0.7 m on October 27 and up to 1.0 m on October 29 due to strong northerly winds. This hydrological extreme event caused reductions in salinity of 1.5 and in water temperature of 1.5 °C within less than 12 h (Fig. 5A). Before and after the storm flood, salinity and temperature were almost constant with a mean salinity of 21.5 ± 0.1 and a mean temperature of 13.5 ± 0.3 °C before the flood (October 22-27), and a mean salinity of 20.4 ± 0.1 and a mean temperature of 10.7 ± 0.1 °C after the flood (November 07 to 16). A minor response of pH and oxygen concentration to flood Performance evaluation of the analyser. During the period October 29 to November 7 the system underperformed, producing unreliable data for both the on-board standard solutions and the natural water samples, www.nature.com/scientificreports/ with negative calibration slopes and a relative change of up to 70% between adjacent time points, respectively. This period was therefore excluded from the time series shown in Fig. 5. Replacement of the filter attached to the sample inlet and a careful cleaning procedure of the microfluidic manifold with detergent and de-ionized water resolved the issues and therefore we assume that the fault was caused by a clogged filter membrane and/or microfluidic channels as a consequence of high suspended material loads in the water column during the storm flood event.
Reliable absorbance values for natural water samples mixed with the PAN reagent, with a relative change of less than 35% between adjacent time points, were obtained by the analyser for the periods October 22 to 29 and November 7 to 17. The analysis of on-board blank and standards for the purpose of in situ calibration produced highly variable calibration values, likely due to problems with their supply towards the microfluidic manifold. Slopes ranging from − 1.307 × 10 -4 L·nmol −1 to 4.607 × 10 -4 L·nmol −1 (mean: (1.955 ± 0.751) × 10 -4 L· nmol −1 ) and intercepts in the range from 0.011 to 0.123 (mean 0.038 ± 0.020) were obtained in situ. Therefore, absorbance values of the natural water sample were processed using laboratory calibration of the analyser with standards prepared at a salinity of 18 (i.e. calibration curve slope = 1.104 × 10 -4 L·nmol −1 /intercept = 0.046). Due to a temperature offset of ca. 10 °C between laboratory calibration (conducted at 20 °C) and in situ temperatures, calibrations were conducted at 10 °C, 15 °C, 20 °C and 25 °C using a benchtop spectrophotometer equipped with a temperature-controlled cuvette holder. No change of sensitivity was observed when different temperatures were applied. DMn concentrations of 215 in situ measurements and 47 discrete samples analysed via ICP-MS as validation method are presented in Fig. 5B. Considering the data points of both time series with matching time stamps (n = 47) a mean accuracy of 87% was achieved for in situ measurements compared to ICP-MS samples, resulting in a strong significant correlation with a Spearman correlation coefficient of 0.873 (p < 0.01, n = 47; see Figs. 5C and 6). Particularly during the period October 24 to 27 DMn concentrations determined with the two methods were in close agreement with a high accuracy of > 99%. Between November 8 and 13, a mean offset of < 6% between the in situ determined concentrations compared to discrete samples was observed. A comparison to ICP-MS determined DMn concentration is more robust than comparison to samples treated with PAN reagent and analysed via a benchtop spectrophotometer. This verifies not only that the analyser measurements were reproducible, but also that sensor derived measurements were not subject to interferences or low recoveries-which it would not be possible to determine directly from benchtop measurements using the PAN method. In prior work using an earlier version of the analyser it was verified that the sensor derived measurements were approximately the same as benchtop measurements 31 , but as noted this only demonstrates the correct mechanical functioning of the sensor and not the accuracy of DMn concentrations.
Considering all data points (Fig. 5B), DMn concentrations determined in the discrete samples were in between 282 and 710 nM, with a mean DMn concentration of 430 ± 104 nM (n = 52). Using the in situ LoC analyser, a minimum of 196 nM and a maximum of 990 nM were observed, with a mean DMn concentration of 435 ± 163 nM (n = 215), in good overall agreement with the value obtained from discrete samples. However, as www.nature.com/scientificreports/ shown in Fig. 5B, there were several periods where in situ determined DMn concentrations were either lower or higher compared to discrete samples. This was also indicated by the divergent extreme values of both time series. For example, distinct overestimation of 18 ± 12% on average was observed from October 22 to 24, whereas the period November 12 to 16 was characterized by 17 ± 14% on average lower in situ DMn concentrations compared to those of discretely collected samples. The discrepancies between the two time series might be attributed to three facts: (1) analysis of slightly different waters due to the used deployment/sampling setup, (2) presence of Mn(III) species in the dissolved phase which was detectable with ICP-MS but not spectrophotometrically with the PAN method and (3) a lower temporal resolution of manually collected samples which may have missed a concentration peak observed by the analyser. As to (1), the sample inlet of the LoC analyser was orientated horizontally at a water depth of ca. 2.0 m. Prior to each in situ analysis, 2.8 mL of the fjord water were aspirated for flushing of the channels. In contrast, the inlet of the hose used for collecting discrete samples was orientated vertically towards the seabed at the same depth as the analyser's inlet, but 1.5 m laterally apart. Approximately 10 L of fjord water were pumped at a flow rate of ca. 1.5 L⋅min −1 through the system prior to sample collection to ensure a careful flushing of the hose and the attached filter cartridge. Due to the different orientation of the water inlet as well as a 3000 times higher water throughput for flushing purposes, it is possible that water from greater depths was aspirated with the setup for the discrete sample collection. Differences in DMn concentrations depending on the sampling depth may arise because of stratification of the water column, a decrease of light driven redox processes with depth (photoenhanced reduction of high-valent Mn to the dissolved Mn(II) fraction 42 ) and/or external DMn input to the upper layers e.g. through wet deposition.
As to (2), in contrast to the speciation of Fe in seawater, where the presence of organic ligands plays an important role, there is only limited evidence for organic complexation of Mn in seawater 43 . However, a recent study suggested the presence of organically complexed and therefore stabilized Mn(III) species by humic ligands in the DMn pool of estuarine waters 44 . Speciation studies showed that 17.6% of the DMn pool occurred as complexed Mn(III) species at 3 m depth, with increasing relative concentrations with depth in the St. Lawrence Estuary. Critically, those species are within the ICP-MS detectable DMn pool, whereas the spectrophotometrically detectable DMn pool, using the PAN method, accounts for only soluble Mn(II). Thus, the presence of dissolved Mn(III) species may have caused lower spectrophotometrically determined in situ DMn concentrations compared to those from discrete samples, e.g. from November 13 onwards (Fig. 5B).
As to (3), the variability in DMn concentrations between October 27 and 29 highlighted the need for temporally well resolved data, achievable only with in situ measurements. The observed dynamic nature of DMn concentration, with an increase of about 400 nM DMn during the night of October 27 to 28, was considered to be caused by the extreme weather conditions as the storm flood that approached the Kiel Bay was linked to an inflow of fresh water (drop of salinity of 1.5 PSU), a water level rise of up to 1 m a.s.l. and wind speeds of up to 13 m⋅s −1 . This fast dynamic environment could not be resolved with discrete samples as, especially under those extreme weather conditions, sample collection was inconvenient and hazardous.
Overestimation due to the cross-sensitivity of the PAN reagent to other ions was negligible as all interfering ions occurred at low dissolved concentrations compared to DMn, with mean values of 90.8 nM Fe, 47.1 nM Zn, 15.0 nM Cu, 6.1 nM Ni and 0.7 nM Co, quantified from discrete samples via ICP-MS. Additionally, no statistically significant evidence of a correlation between the time series of these elements and an enhanced DMn fraction was found.
According to a Spearman rank correlation test, DMn concentrations showed significant relationships with a range of other parameters ( Fig. 6; for time series see Fig. 5A,B). For example, in situ DMn concentrations anticorrelated significantly with oxygen and pH, with correlation coefficients of − 0.57 and − 0.71. The transformation between Mn(IV) oxides and dissolved Mn(II) species can be described as follows: Oxidation, and therefore a decrease of dissolved Mn(II) species, is favoured at increased oxygen concentrations and high pH values. In contrast, reduction of Mn(VI) oxides towards Mn(II) takes place under oxygen depletion and low pH values, typically in anoxic sediments with subsequent diffusion into overlying waters 45 . In natural waters, the reduction pathway is sun-light mediated, which is responsible for elevated DMn concentrations in the euphotic zone 6 . Furthermore, organic matter can mediate the (photo)-chemical reduction of Mn oxides to soluble Mn(II) 6 , which would explain the correlation between observed in situ determined DMn and DOC concentrations (r = 0.40).
A positive correlation with a Spearman correlation coefficient of 0.53 between DMn and salinity was found. This may be related to the storm flood, as we would generally expect an increase of DMn with decreasing salinity because freshwater, e.g. through riverine input, represents a major source of DMn species. When considering the time phases before and after the flood event individually, anti-correlations between DMn and salinity were found with correlation coefficients of − 0.27 and − 0.30, respectively. A strong statistical correlation was found between DMn and phosphate concentration (r = 0.82), suggesting that the marine biogeochemical pathways of both species are, together with oxygen, connected. In oxygenated waters, DMn species are oxidized towards Mn(IV) oxides (Eq. (1)). Mn(IV) oxides act as important adsorbents of phosphate, which leads to a removal of phosphate from the dissolved phase 46 . Upon reduction of particulate Mn in sediments, Mn(II) and phosphate are remobilized 45 .
This study demonstrated that LoC technology forms a powerful tool for in situ quantification of DMn species in a wide range of environments with respect to the linear detection range, e.g. in coastal waters and estuaries, in benthic boundary waters and in the vicinity of hydrothermal vents. Compared to previously reported in situ analysers which used the PAN method for the quantification of DMn in natural waters (e.g. SCANNER 30 , METIS 24  www.nature.com/scientificreports/ a prolonged operational lifetime due to minimized fluid consumption and a higher sensitivity in terms of molar absorptivity was shown. Miniaturization led to an easily deployable device when compared to a modular system such as the SCANNER. Furthermore, the chemical recipe of the PAN reagent was adapted in order to improve its long-term stability and the effect of masking agents on the measurements was critically evaluated. The second generation LoC presented herein is a significantly updated version to that described by Milani et al. (2015) 31 , and features smaller microfluidic channels (160 × 300 µm vs 300 × 400/600 µm), lower average power consumption (1.5 W vs 3.8 W), improved electronics and software (Windows-based graphical user interface), and onboard data processing in order to acquire reliable and well resolved in situ DMn data over long periods most preferably remotely and unattended. This was demonstrated in the presented study with a deployment over 17 days, while the first generation LoC analyser as well as the METIS analyser were deployed on CTD casts where physical inspection and troubleshooting of the systems were facilitated between casts. The time series highlights the ability of LoC analysers to produce temporally well resolved measurements of trace metals, achievable only with in situ systems. This enables the unravelling of biogeochemical questions in remote areas, in dynamic areas with highly variable DMn concentrations over short time scales, or during hydrologically extreme events.

Methods and materials
Preparation of standard and reagent solutions. A thorough cleaning procedure was applied to all glass and plastic ware for standard and reagent preparation prior to their use. Soaking in a ~ 2% v/v acidic detergent bath (Citranox, Sigma-Aldrich) was followed by a 1.2 M HCl bath (reagent grade, Carl-Roth) for > 24 h. Glass and plastic ware was rinsed thoroughly with de-ionized water (MilliQ, 18.2 MΩcm, Merck Millipore) after each treatment and stored in plastic bags until required. A working stock solution containing 100 µM Mn(II) was prepared on a weekly basis using 546 µL of a 1000 mg⋅L −1 Mn standard (1000 ppm Manganese for ICP, Inorganic Ventures) diluted to 100 mL with de-ionized water. Standard Mn(II) solutions were obtained by further dilution of the 100 µM Mn(II) working stock with de-ionized water. All Mn(II) solutions were stored at room temperature in opaque low-density polyethylene (LDPE) bottles (Nalgene).
The preparation of the PAN reagent followed, with some adaptations, the method reported by Chin et al. (1992) 30 . 0.05 g of 1-(2-Pyridylazo)-2-naphthol (PAN) (general purpose grade, Fisher Scientific) and 5 mL Triton-X100 (laboratory grade, Sigma-Aldrich) were added to approximately 50 mL of de-ionized water and stirred at 80 °C for at least 12 h until the PAN was dissolved completely. The warm orange coloured mixture was then poured into 100 mL of a 0.1 M borate buffer (pH ~ 10) and made up to 250 mL with de-ionized water, giving final PAN and Triton-X100 concentrations of 0.8 mM and 2% v/v (equivalent to 33 mM), respectively. For method evaluation, further PAN reagents were prepared containing 4% v/v Triton-X100 by the addition of 10 mL Triton-X100 or 2% m/v of the ionic surfactant sodium dodecyl sulphate (SDS) by the addition of 5 g SDS (ultra-pure, Carl Roth). No heat was required for the dispersion made with SDS to solubilize PAN. Borate buffer contained 0.618 g H 3 BO 3 (99.99%, trace metal basis, Acros Organics) and 0.4 g NaOH (98.5%, Acros Organics) made up to 100 mL with de-ionized water, giving a concentration of 0.1 M for both boric acid and sodium hydroxide. As an interference of Fe(III) ions was reported for the PAN method 30 , two different Fe specific complexing/masking agents were tested: DFO-B and disodium 4,5-dihydroxy-1,3-benzenedisulfonate (Tiron). A 1.5 mM DFO-B solution was prepared by dissolving 5 mg of deferoxamine mesylate salt (95%, Acros Organics) in 5 mL de-ionized water. This solution was kept refrigerated when not in use. A 50 mM Tiron stock solution was prepared by dissolving 0.785 g of Tiron (Acros Organics) in 50 mL de-ionized water.

Spectrophotometric benchtop experiments.
For benchtop experiments using the PAN method, Mn(II) standard solutions and PAN reagent were mixed using a volumetric ratio of 9:1. For Fe(III) interference studies, the prepared Mn(II) standards were spiked with an acidified 100 µM FeCl 3 solution (≥ 98%, Carl Roth) and a Fe(III) masking agent if required. Absorbance spectra between 400 and 800 nm were acquired after 5 min with a double beam Shimadzu UV-1800 spectrophotometer using 10 cm quartz cuvettes. Absorbance was processed at the peak maximum of 562 nm, and also at 768 nm for the Fe(III) interference experiments.
Lab-on-chip analyser. The PAN method was adapted for its application in a microfluidic LoC analyser.
The analyser hardware (Fig. 7) was identical to a second generation device characterized for the detection of Fe in coastal waters using the Ferrozine (FZ) method 29 as the Mn(PAN) 2 complex features the same absorbance maximum at a wavelength of 562 nm as the Fe(FZ) 3 complex. A first generation device was previously reported in Milani et al. (2015) 31 .
Briefly, the second generation analyser comprised a microfluidic chip (119 mm in diameter and 24 mm in thickness; Fig. 7B) made from tinted Poly(methylmethacrylate) (PMMA) 47 , with milled microfluidic channels of the dimensions 160 × 300 µm, forming the end-cap of a watertight polyvinylchloride (PVC) housing (Fig. 7A). A custom-built syringe pump unit including a stepper motor, two barrels for reagent supply (3.28 mm ID) and one barrel for sample and standard solutions (9.71 mm ID) was mounted onto the microfluidic chip for sample and reagent withdrawal from the reservoirs and injection into the microfluidic channels. Micro-inert solenoid valves (LFNA1250125H, The Lee Company) provided full fluidic control via individual actuation. Sample and standard solutions were mixed on-chip with the PAN reagent in a volumetric ratio of 8.8:1, defined by the volume of the barrels as the plungers of the pumping unit are moving simultaneously at the same speed. Here a waiting period of 15 min under stopped flow condition was sufficient to provide complete mixing and full colour development, leading to a maximum sample throughput of ~ 3 samples per hour including the flushing routine. After chemical reaction and full colour development the absorbance was measured in three individual optical channels of different length (91.6 mm, 34.6 mm and 2.5 mm) using LEDs with a peak wavelength of 575 nm Deployment and discrete sampling. The capability of the LoC analyser for DMn measurements under environmental conditions was tested during a field campaign conducted in October/November 2018 in the inner Kiel Fjord (Germany). The analyser was deployed together with other LoC devices (for in situ analysis of Fe(II), DFe, and pH) and a SeapHOx unit (SeaFET pH sensor plus SBE 37-SMP-ODO MicroCAT CTD + DO sensor) for continuous acquisition of hydrographic parameters (pH, temperature, salinity and dissolved oxygen). All instruments were mounted on a frame (Fig. 8) which was lowered from a pontoon to a water depth of 2 m. The Mn analyser was equipped for the deployment with a blank solution (South Atlantic seawater diluted with de-ionized water), two Mn(II) standard solutions and PAN reagent. Blank and standard solutions were prepared at a salinity of 18 using South Atlantic seawater (with initial DMn concentrations below the analyser's detection limit) in order to closely match the sample matrix of fjord water 41 . The analysis of a sample, which was withdrawn through a 0.22 µm membrane filter (Millipore, polyethersulfone (PES)) from a PTFE tubing with an inlet at 2.0 m depth, was undertaken every 90 min. Prior to each sample measurement, a calibration procedure was conducted using the blank solution followed by analysis of 150 nM and 300 nM Mn(II) standards (for a detailed overview of the measurement routine see Table S1 in the Supplementary Information). In order to validate the DMn concentrations measured by the in situ analyser, discrete samples for trace metal analysis were collected