Metal-organic frameworks combined with mid-infrared spectroscopy for the trace analysis of phosphates in water

Detecting traces of phosphates in water is critical for monitoring water quality in aquatic ecosystems. Mid-infrared techniques are effective for label-free and quick measurements in at-line or in-line applications, but sensitivity is limited due to high water absorption. To combat this, preconcentration schemes combined with evanescent field spectroscopy can be used to enrich the analyte in the probed volume and thereby increasing sensitivity. Metal-organic frameworks (MOFs) are versatile materials with defined porosity and tuneable chemistry, which make them ideal for selective adsorption of target molecules. In this study, an NH 2 -MIL-88B (Fe) ( = Fe 3 O(NH 2 -BDC) 3 ) MOF-based enrichment layer was prepared on a diamond attenuated total reflection (ATR) crystal for in situ Fourier transform infrared (FTIR) spectroscopic measurements of ortho-phosphates in water. A workflow for single-use enrichment layers was established, and an automated flow system was used to apply aqueous phosphate solutions. Using internal referencing and compensating for variation in MOF film depositions, phosphate analysis reproducibility increased from 74% to 94%. The Langmuir adsorption model was used to derive a limit of detection (LOD) of 0.18 mg L -1 phosphorus in water. Overall, this work demonstrates the effectiveness of NH 2 -MIL-88B(Fe) MOFs as enrichment layers for aqueous phosphate sensing. Our results provide a promising avenue for the development of sensitive and selective sensors for environmental and biomedical applications.


Introduction
Phosphorus is an essential element for the development and survival of life.Besides the use as fertilisers in agriculture, phosphates are also prominently used in detergents, accounting for a worldwide phosphate production of 220 million tons in 2021 [1].The excessive use of phosphates led to an accumulation of nutrients in aquatic ecosystems around the world.This resulted in the eutrophication of several aquatic ecosystems, creating algal and cyanobacterial mats, reduced oxygen concentrations in affected waters and, as a consequence, a severe loss of biodiversity [2,3].As preventive measures, the US environmental protection agency and the European environmental agency recommend a maximal phosphorus concentration between 0.05 and 0.1 mg L -1 in flowing waters.In recent years, point source water pollution due to industrial sewage has been reduced by strict regulations and mitigation schemes, making non-point sources such as agricultural runoff, construction sites or other seasonal events the main contributors to the continuing problem of phosphate eutrophication.As these non-point sources are difficult to monitor, constant surveillance of nutrient concentration is essential in vulnerable waters in order to preserve the quality of drinking water and aquatic ecosystems [4][5][6].To date, the standard method to determine trace amounts of phosphates is based on a colorimetric assay, which detects the presence of the blue phosphomolybdate complex by means of spectrophotometry [7].This method allows for an application range between 0.5 mg L -1 and 5 mg L -1 and very accurate measurements [8], but involves an off-line multi-step sample preparation, additional chemical reagents (yielding toxic chemical waste) and generally either narrow detection ranges or high limits of detection [9].
Fourier transform infrared (FTIR) spectroscopy is a widely used technique for analysing liquid samples in the food and beverage industry, such as soft-drinks, milk, wine, beer, and edible oils, as well as in biotechnological applications for process monitoring and control [10][11][12][13][14][15].This spectroscopic technique provides molecular-specific information by probing characteristic vibrational transitions.FTIR spectroscopy has further shown promise for the surveillance of both natural and process water streams, for instance in the detection of bacteria in eutrophic lakes and the analysis of oil in water content in process water streams [16,17].
The analysis of concentrations greater 0.1 g L -1 of phosphate in nonsweetened soft-drinks and biotechnology has also been demonstrated using FTIR spectroscopy, based on specific IR absorption bands of phosphates in the mid-IR region between 1200 and 1000 cm -1 [18,19].Using this approach, fast and direct quantification can be achieved without requiring the use of toxic reagents such as phosphomolybdate, which is needed for spectrophotometric phosphate analysis.However, the high water absorption in the characteristic fingerprint region, thus, also in the spectral region of the phosphate bands, limits the sensitivity of IR spectroscopic techniques in aqueous samples.To combat this, preconcentration schemes can be employed, boosting the achievable limits of detection (LODs) from the high mg mL -1 range to the sub-mg L -1 range [20,21], enabling analysis of trace amounts of phosphates in flowing waters.
For evanescent field sensing, this can be realised by enrichment coatings, which help in accumulation the analyte in the volume probed by the evanescent field.Early works used polymer-coated ATR elements or fibres to extract and preconcentrate hydrophobic analytes such as chlorinated hydrocarbons present in water [22].However, their response time is long due to slow diffusion of the analyte into the polymer enrichment films.The use of porous oxides such as silica and zirconia improves on this concept by providing a faster sensor response based on adsorption.The introduction of surface functionalities such as hydrophobic groups or ion exchange sites enables further fine-tuning of the layer towards respective analytes such as organic contaminants, volatile organic compounds, or nitrates [23][24][25][26][27][28].
The strong absorption bands of silica and zirconia overlap with the phosphate bands around 1100 cm -1 , which impede their application as a phosphate enrichment layer.Polymers lack mechanical stability due to swelling in water and show long response times, although almost full recovery after phosphate adsorption into polymer-based anion-exchange membranes has been reported [29,30].Therefore, selecting a material for phosphate enrichment requires a compromise between stability, affinity, and capability to regenerate, and new materials beyond polymers and oxides need to be considered.
In metal-organic frameworks (MOFs), organic linkers coordinate to single metal ions or metal-oxide clusters (secondary building units, SBUs), yielding crystalline materials with high specific surface areas and tuneable porosity and functionality.The chemical affinity of MOFs can be modified by linker and cluster design towards various target molecules including CO 2 , VOCs and BTX [31][32][33][34].This makes MOFs attractive materials for the adsorptive removal of pollutants [35,36] and adsorption-based sensing schemes [37,38].Most MOF-based sensors rely on changes in the capacity, fluorescence or luminescence of the MOF or hosted species such as quantum nanodots upon interaction with the analyte [39][40][41][42].Thereby, phosphates can be detected at concentrations as low as 0.005 mg L − 1 .However, additional chemical substrates and an incubation time of 40 min are required and the sensing material cannot be regenerated [43].
In this contribution, we combine the concept of preconcentration in the evanescent field in ATR-FTIR spectroscopy with the high adsorption capacity of MOFs (Fig. 1) for direct liquid-phase phosphate sensing without the need of additional substrates.The employed NH 2 -MIL-88B (Fe) (Fig. 1-A) is a member of the water-stable iron(III) aminoterephthalate family, which has shown high phosphate affinity [44].The MOF was synthesised according to literature [45] and characterised with X-ray diffraction and FTIR spectroscopy.
We aimed for a highly reproducible and fast method of preparing single-use powder-based enrichment layers.For this, we spin-coated an NH 2 -MIL-88B(Fe) suspension onto a diamond ATR crystal, yielding MOF covered ATR crystals, which are further denoted as 'MOF films'.The sensing performance of MOF films was investigated and compared to an enrichment film made of non-porous iron oxide powder with similar particle size to prove the advantageous properties of the highly porous materials.These results demonstrate the high potential of MOFs for mid-IR liquid phase sensing.

Synthesis of NH 2 -MIL-88B(Fe)
The synthesis of NH 2 -MIL-88B(Fe) was performed following the report of Bauer et al. [45]: 2.025 g FeCl 3 .6H 2 O was dissolved in 25 mL of DMF.To this solution, a solution of 0.675 g NH 2 -H 2 BDC in 20 mL of DMF was added.The reaction mixture was poured into a Teflon lined autoclave and heated for 24 h at 115 • C.After cooling down, the brown precipitate was separated from the solvent via centrifugation for 3 min at 4000 rpm (Rotofix 32 A, Hettich, Germany).The as-synthesised MOF was then washed four times using DMF, twice with deionised water, and twice with EtOH.After the last step, the MOF was dispersed in EtOH and poured into a crystallising dish.After evaporation of the EtOH at ambient conditions, the brown powder was dried at 90 • C for 24 h to remove any remaining trace of solvent.

Characterisation of NH 2 -MIL-88B(Fe)
X-ray diffraction patterns were collected with an X'Pert PRO MPD diffractometer, equipped with a X'Celerator line scan detector and the Data Collector software (all from PANalytical, Netherlands).The measurements were conducted in Bragg-Brentano geometry, with the Cu anode operating at 45 kV and 40 mA.The diffractograms were recorded between 5 • and 25 • 2θ with a step size of 0.01 • and an integration time of 360 s/step in continuous scan mode.Simulation of the theoretical diffractogram was done by Rietveld refinement using data [45] from the Cambridge Structural Database (CSD).Rietveld refinement was performed using HighScore Plus [46] (PANalytical, Netherlands).FTIR-ATR measurements for the characterisation of NH 2 -MIL-88B(Fe) were performed using a Tensor 37 FTIR spectrometer (Bruker Optics, Germany) equipped with a Platinum ATR unit (Diamond, single-bounce, Bruker Optics, Germany) and a N 2 -cooled MCT detector.Spectra were recorded with 4 cm -1 resolution, averaging 32 scans per spectrum.

Preparation of enrichment film
The coating suspension was prepared by dispersing the dried MOF or Fe 2 O 3 MP in IPA, resulting in a mixture with 3.5 wt% solid.The liquid was continuously stirred using a magnetic stir bar to maintain the homogeneity of the suspension.The enrichment film was then spin-coated onto the ATR element at a spinner velocity of 1150 rpm and subsequently annealed for 15 min at 110 • C.

Optical setup for FTIR spectroscopy
MOF films were coated onto a diamond ATR crystal (20 ×10 x 0.5 mm 3 , facets with an angle of 55 • (Diamond Materials, Germany).This geometry resulted in 14 active bounces (N) at a total effective pathlength (d e,tot = d e N) of 18.1 µm at 1100 cm -1 (n dia = 2.4, n sample = n water = 1.33) and a depth of penetration (d p ) of 1 µm [47].The measurement setup comprised a custom built ATR mount and an aluminium flow cell previously reported by Freitag et al. [48] and Baumgartner et al. [23], which allowed the integration into the sample compartment of a Vertex 70 v FTIR spectrometer (Bruker, Germany).Spectra were collected with OPUS 8.1 software using a spectral resolution of 4 cm -1 , 32 scans per spectrum (double sided, forward-backward acquisition mode).During all measurements, the sample compartment was flushed with dry air.The noise floor of the sensor was determined by evaluating the RMS noise of 100% lines, measured in a stopped flow environment of the water background, yielding an RMS noise of 1.85 × 10 -4 A.U. between 1000 and 1200 cm -1 .

Data processing and spectra evaluation
The evaluation of the spectra collected during the sensing experiments was done using an in-house MATLAB R2021a script.Spectra were baseline corrected by a two-point linear algorithm between 1195 cm -1 and 920 cm -1 .The band heights were then determined at the peak maximum of the baseline corrected spectra situated at 1025 cm -1 .

Liquid handling and automation
The automated sequential injection analysis (SIA) system is schematically shown in Fig. 2. It is modified from the SIA system previously reported by Freitag et al. [48], and consisted of a 10-port selection valve (VICI, Switzerland), as well as a Cavro XC syringe pump (Tecan, Switzerland) equipped with a 500 µL glass syringe (Tecan, Switzerland).The components were connected with PTFE tubing (VICI, Switzerland, O.D. = 1/16 in., I.D. = 0.75 mm).Each concentration of the sample phosphate solutions (ortho-phosphate, pH = 5.15-5.75,5-75 mg L -1 ) was fed to a separate port on the selection valve.The SIA setup and the FTIR spectrometer were controlled by a LabVIEW VI using a server-client program structure [49].
The SIA sequence used for the automated sensor experiments is shown in Fig. 3 (characterisation of SIA system in ESI).It consisted of first measuring a phosphate standard followed by analysis of the actual sample.The whole sequence comprised three parts: 1) flushing the flow cell, followed by a 9 min conditioning step of the MOF in the aqueous sample medium (a-c in Fig. 3), 2) measurement of the standard with a concentration of 10 mg L -1 of ortho-phosphate in water (d-e in Fig. 3) and 3) the actual measurement of the sample at the end of each measurement cycle (f-g in Fig. 3).
Furthermore, intermittent washing steps using deionised water were employed after each measurement step.This procedure allowed to acquire the data needed to account for slightly different sensitivities of the individually prepared sensing films.Each measurement step (both standard and sample) consisted of a background measurement (denoted with asterisks in Fig. 3), followed by the injection of 300 µL orthophosphate solution at a speed of 100 µL s -1 and 7 sample channel measurements with 10 s between each of them, equating to a total duration of 63 s.

Evaluation of sensor performance
The sensor performance was evaluated by executing the measurement sequence as shown in Fig. 3 in triplicate for each concentration (5 mg L -1 , 10 mg L -1 , 25 mg L -1 , 50 mg L -1 , and 75 mg L -1 ), with new films for each measurement.Each film was prepared following the procedure described in Section 2.3, while the ATR crystal was cleaned between each loop by wiping the film off with a clean room wipe, followed by ultrasonication for 3 min in EtOH.All sensor experiments were performed at 22 • C and standard pressure.
To assess the influence of interfering ions, 25 mg L -1 of sulphate and nitrate were added to phosphate samples of the aforementioned concentrations.These measurements were carried out in a manner consistent with the methodology described in 2.6, with two replicates for each sample.The standard measurements for these samples containing the interfering ions were also done with a 10 mg L -1 pure phosphate standard to ensure comparability with the pure phosphate measurements.

Characterisation of NH 2 -MIL-88B(Fe)
ATR-FTIR spectroscopy and powder X-ray diffraction of the assynthesised NH 2 -MIL-88B(Fe) confirmed the MOF structure and presence of the incorporated amino-moiety.The ATR-FTIR spectra of the assynthesised NH 2 -MIL-88B(Fe) and the commercially available Fe 2 O 3 MPs are shown in Fig. 4-A.Bands corresponding to the aminoterephthalate linker were assigned as the asymmetric and symmetric NH 2 stretching vibrations at 3460 cm -1 and 3329 cm -1 , two intense asymmetric and symmetric C--O bands at 1574 cm -1 and 1379 cm -1 [50], a differently coordinated C--O band at 1425 cm -1 [51], the skeletal C--C vibration band at 1493 cm -1 [50], the C-N vibration mode at 1254 cm -1 [50], and the aromatic C-H bending vibration band at 766 cm -1 [52].The band at 517 cm -1 corresponds to O-Fe-O vibration seen in the ATR-FTIR spectrum of the Fe 2 O 3 MP and confirms the formation of the coordination network of the MOF [50,53].
The powder X-ray diffractogram of NH 2 -MIL-88B(Fe) as well as the simulated diffractogram obtained by Rietveld refinement of the powder refraction data using the literature CIF file 647646 [45] are shown in Fig. 4-B

Evaluation of sensing performance
The phosphate enrichment experiments were performed using the automated SIA sequence described in Section 2.6 with five different concentrations (5 mg L -1 , 10 mg L -1 , 25 mg L -1 , 50 mg L -1 , and 75 mg L - 1 ) for both the MOF and the Fe 2 O 3 MP films (with higher concentrations added due to lower enrichment), respectively.The enrichment was monitored using the absorbance maximum of the phosphate band at 1025 cm -1 .
The sensor response was determined first to find the right compromise between response time and sensitivity (Fig. 5-A).As can be seen in Fig. 5-B, it took 9 min to reach 90% of the equilibrium signal for a phosphate solution with a concentration of 25 mg L -1 , while 43% can be Fig. 2. Schematic depiction of the automated sequential injection analysis system.Fig. 3. SIA sequence for a sample automated sensor experiment with a standard concentration of 10 mg L -1 and a sample concentration of 25 mg L -1 .A phosphate concentration of 0 mg L -1 at a positive flow rate signifies flushing the flow cell with deionised water.The timing of background measurements is denoted by asterisks.achieved in 1 min.In order to present a fast yet sensitive method of sensing phosphates, an enrichment time of 1 min (63 s, as 3 s are needed for sample injection) was chosen.This enrichment time can, however, be extended in order to increase sensitivity at the cost of response time.A sample phosphate enrichment profile for this enrichment time is depicted in Fig. 5-B* .
Recorded phosphate spectra with an enrichment time of 63 s used for calibration as well as a 100% line as noise reference are shown in Fig. 6.The broad absorption band between 1160 cm -1 and 950 cm -1 seen in the phosphate spectra can be explained by overlapping P-O vibrations similar to other oxidic adsorbents [56].
The band heights obtained after 63 s of enrichment (Section 2.6 for description of the measurement steps) were normalised by the band heights obtained from the initially recorded standard spectrum.The normalised data were used to establish the calibration function, which is given in Fig. 7.The unavoidable difference between the spin-coated enrichment films, which were prepared for each sample measurement, required such normalisation, as already mentioned in Section 2.6.By employing the normalisation, an increase of the repeatability from 74% to 94% could be observed (see ESI for unreferenced data and discussion of errors).
The better quality of the fit (see ESI) of the Langmuir adsorption model (1) over the multi-layer Freundlich model further suggests the expected single-site adsorption of phosphate into the MOF [23].As a result, the Langmuir model (1) was used as the calibration function.In this equation, q e (unitless) denotes the relative band height for the adsorbed analyte referenced with the band height for the standard concentration, with q m (unitless) being the maximum relative band height associated to the adsorbed analyte, and K L (L mg -1 ) being the Langmuir constant.
The calibration function fitted with the Langmuir equation shown in Fig. 7 revealed a theoretical maximum relative band height q m of 7.55 and a K L of 0.0183.Using K L and q m and the noise floor of 1.85 × 10 -4 A. U. (as mentioned in Section 2.4) allowed the calculation of a theoretical limit of detection (LOD) for the sensor.The LOD was determined as three times the relative band height above the referenced noise floor (2) [57,   58].The noise floor was referenced with the mean band height (s rel = 2.61 × 10 -2 ) of the phosphate standard.This equates to a theoretical LOD of 0.57 mg L -1 phosphate in water (equivalent to 0.18 mg L -1 phosphorus in water).The limit of quantification (LOQ, defined as nine times the relative band height above the referenced noise floor) equates to 1.72 mg L -1 phosphate in water (equivalent to 0.55 mg L -1 phosphorus in water) [58].
Identical experiments were performed using Fe 2 O 3 MP films.The lower specific surface area of Fe 2 O 3 MP compared to the MOF did not allow for significant phosphate loading, making the referencing and, thus, a reproducible sensing assessment, impossible (see ESI).This finding justifies the use of the highly porous NH 2 -MIL-88B(Fe) for phosphate enrichment.

Phosphate sensing with interfering ions
In order to account for sample solutions closer to environmental applications and to simulate the presence of interfering ions in natural waters, experiments in the presence of sulphate and nitrate ions with a fixed concentration of 25 mg L -1 each were performed.Comparing the relative band intensities with and without interfering ions (Fig. 8) shows a negligible influence of the interfering ions at lower phosphate concentrations (c < 10 mg L -1 ), as the measured values are within or close to the margin of errors at the respective concentrations.At larger concentrations, however, the calibration curve flattens significantly compared to the pure phosphate samples, impeding the quantification of unknown phosphate concentrations in the presence of sulphate and nitrate.This can most probably be attributed to competitive adsorption effects, which increase with total ion concentrations.Similar effects have been observed for phosphates and sulphates on goethite, albeit with lower interferences [59], and can be attributed to the lack of equivalent surface sites for the adsorption above a certain threshold concentration.

Conclusion
In this work, we introduced a phosphate sensor comprising a MOF as enrichment film combined with liquid-phase, ATR-FTIR spectroscopy.The MOF NH 2 -MIL-88B(Fe) was chosen as enrichment material as it offers a good compromise between stability, affinity, and enrichment capability.The successful synthesis of NH 2 -MIL-88B(Fe) was confirmed through FTIR spectroscopy, X-ray diffraction, N 2 sorption, and scanning electron microscopy.We established a facile and reproducible procedure for spin-coating NH 2 -MIL-88B(Fe) films onto a diamond ATR.The MOF powder films acted as enrichment layer, effectively preconcentrating phosphate from water within the volume probed by the evanescent wave of the ATR crystal.The sensing performance of the system was evaluated using an automated SIA system coupled with an FTIR spectrometer, measuring concentrations ranging between 5 mg L -1 and 75 mg L -1 with a measurement time of 1 min.By referencing the sample measurement with an internal reference point at a concentration of 10 mg L -1 , we showed that measurement errors due to variations in the film application could be compensated, improving the repeatability from 74% to 94%.The data from a concentration series was described with the Langmuir adsorption model, which also served as the sensor's calibration function.An LOD of 0.18 mg L -1 phosphorus in water was determined.Adsorption studies in the presence of interfering ions showed a good sensing performance below a phosphate concentration of 10 mg L - 1 , with more interferences at higher phosphate concentrations.
Regarding speed, sensitivity, and reproducibility, the use of MOFbased enrichment films represents a good compromise for phosphate sensing.Further, the combination of MOFs with ATR spectroscopy is not limited to phosphates and paves the way for porous enrichment layers with tuneable affinities towards various pollutants.To enhance the performance of the enrichment layers, future work will focus on tuning the material to allow for full regeneration.Additionally, further studies will be performed to better understanding the competitive adsorption of different ions in MOFs.

Fig. 1 .
Fig. 1.A: Structure of NH 2 -MIL-88B(Fe), SBUs shown in light yellow, aminoterephthalate linker shown in green.B: Schematic of the measurement principle: Phosphate (dark yellow and red) is enriched from the aqueous phase into the MOF film, which covers the evanescent field (light red) of the infrared beam (dark red) passing through the diamond ATR crystal (grey), leading to specific absorption in the mid-IR region.

Fig. 5 .
Fig. 5. A: Development of the spectrum of the enriched phosphate until equilibrium is reached.B: Normalised peak area over enrichment time.B* : Phosphate enrichment profile for a phosphate concentration of 75 mg L -1 as a function of time.

Fig. 6 .
Fig.6.Spectra of the enriched sample phosphate (75 mg L -1 , dark blue), the standard phosphate (10 mg L -1 , light blue), both as used for sensor calibration, and the 100% RMS line (black) for noise evaluation.

Fig. 7 .
Fig. 7. Referenced phosphate band heights as function of the applied concentration.The band height was referenced to the standard concentration of 10 mg L -1 .Error bars show the standard deviation for each concentration (n = 3 for each respective concentration).Data points were fitted with the Langmuir equation (solid line) with a 95% confidence of the fit (dashed line).

Fig. 8 .
Fig. 8. Relative heights of the phosphate bands of the pure phosphate samples referenced to the standard concentration (10 mg L -1 ) as function of the applied concentration (blue).Relative heights of the phosphate bands of samples with added interfering ions (25 mg L -1 , orange).Error bars for the samples with interfering ions show the standard deviation for each concentration (n = 2).