Rapid Discovery and Structure-Property Relationships of Metal-Ion Fluorescent Sensors via Macroarray Synthesis

A macroarray immobilisation of fluorophores on filter papers for sensing metal ions by in-situ reductive amination and carbodiimide coupling is reported herein. Chemometric approaches resulted in a rapid discovery of sensors that can synergistically discriminate up to 12 metal ions with great prediction accuracies. Covalently bound on paper, sensoring scaffolds that were synthesised from the macroarray format can readily be adopted as practical paper-based sensors with great reusability and sensitivity, achieving the limit of detection at low nanomolar level with some repeating spotting. Lastly, the discovered scaffolds were also confirmed to be functional as unbound molecules, thus paving the way for more diverse applications.

investigate the essential core of binding and found that a mere glycine unit was sufficient to achieve clear turn-on fluorescence sensing with Zn 2+ ion 33 .
In the current study, we aimed to synthesise an expansion set of quinoline probes from the aforementioned essential core by "reversing the discovery paradigm 34 " and thus anticipated to discover some new fluorescent probes in a systematic and rapid manner. Notably, while solid-phase synthesis on polymer beads is a well-established field, our current study was made possible by the use of a carefully-planned macroarray synthesis 35 . This type of technique, originated from the peptide-only synthesis (SPOT synthesis) 36,37 , is a synthesis strategy for small molecules performed on cellulose support. With the spatially-addressable nature, macroarray synthesis allows for the rapid synthesis of multiple compounds on one paper sheet without complex deconvolution. In addition, as similar to other solid-phase techniques, macroarray synthesis was relatively rapid due to the fact that unreacted molecules can be easily washed away without any purification steps. With these unique advantages, previous works have showcased various applications [38][39][40][41] that were directly benefitted from this rapid, low-cost, and reliable method.
Importantly, our current work highlights the first example of how macroarray synthesis can be used to both get access to diverse molecular scaffolds and to perform metal ion sensing on-support without extra cleavage or post-synthesis steps. This accelerated the discovery of sensing molecules even further, and resulted in a set of compounds that, with an aid of chemometrics, can complementarily sense various metal ions, a finding of which would otherwise be improbable or tedious to attain with traditional synthesis. Last but not least, we also demonstrated the utilisation of a Zn 2+ -selective scaffold as a ready-to-use paper-based sensor with great sensitivity and reusability. This was all made possible thanks to the permanent binding nature of the compound to the cellulose support.

Results and Discussion
The synthesis, after some optimisations and/or consideration based on previous other works, can be summarised as follows (Fig. 1). The first step was the treatment of cellulose with NaIO 4 and LiCl 42,43 . This reaction resulted in the conversion of diol at position 2 and 3 of the glucose repeating unit into the dialdehyde 1. Thereafter, 1 was subject to reductive amination with a variety of amino acids (reagent X, see Supplementary Fig. S1), followed by NaBH 4 treatment to reduce unreacted surface-bound aldehydes to alcohols. This library X consists of readily-available amino acids along with some variations to also probe other effects such as chirality or steric effect due to extended carbon skeletons. The resulting modified surfaces 2 with free carboxyl groups then underwent carbodiimide-mediated couplings with various aminoquinolines and an aminonaphthalimide (reagent Y, Supplementary Fig. S1). After washing unbound materials off, 3 can be directly incubated with a solution of a metal ion of interest. Fluorescence response of each surface-bound compound with a metal ion can be easily probed by subjecting the paper sheet with direct illumination from a handheld UV lamp or a transilluminator, thereby revealing active compounds by comparing fluorescence intensity and colour hue with the controls. Also, while the exemplary synthesis in Scheme 1 utilised the spotting method for synthetic steps with the whole-sheet immersion in the sensing step, this overall process can be swapped. This may be more suitable and efficient in certain cases, e.g., the screening for metal ion partners (by spotting) within a single sheet containing only one sensing scaffold of interest. An example of the exact protocol for the sensing step can be found in the method section. Notably, all synthesis steps could be performed at ambient condition, and the finished cellulose sheets could be kept under a desiccator at ambient temperature for up to at least 2 weeks without deterioration in sensing performance ( Supplementary Fig. S2). This reflected a fairly high versatile and durable nature of the whole methodology.
In this proof-of-concept experiment, we commenced with the screening of the X counterpart while keeping the Y component to be 8AQ to allow some comparison with our previous study 33 . The evaluation was performed by observing any fluorescence change when incubated with metal ions including Li + , Na + , K + , Mg 2+ , Ca 2+ , Ba 2+ , Al 3+ , Cr 3+ , Fe 2+ , Co 2+ , Ni 2+ , Cu 2+ , Zn 2+ , Ag + , Cd 2+ , Pb 2+ , and Hg 2+ . After obtaining a preliminary www.nature.com/scientificreports www.nature.com/scientificreports/ result ( Supplementary Fig. S3), the X components Gly and Asn were selected for the Y-component screening. After compiling obtained results, a set of interesting fluorescent compounds was revealed (Fig. 2). Interestingly, these scaffolds, generally weak or non-fluorescent in water, showed obvious increase in fluorescence upon binding to certain metal ions such as Zn 2+ , Al 3+ , and Cr 3+ , while some metal ions, e.g., Hg 2+ , can further decrease the fluorescence intensity of certain fluorophores, likely due to the inherent quenching attributes of Hg 2+ such as enhanced spin-orbit coupling 44,45 . More importantly, the responses of these scaffolds were dependent on specific chemical features present in each scaffold. For example, while the modified surface Gly.8AQ exhibited strong fluorescence upon binding with Zn 2+ (believed to be due to a combination of reasons such as the prevention of non-radiative relaxations, e.g., photoinduced electron transfer from the nitrogen of the -CH 2 -NH-moiety to the quinoline ring, and the restriction of bond rotation), trivalent Al 3+ and Cr 3+ seemed to give better fluorescence increase with Gly.3AQ and Gly.6AQ, whose aminoquinoline scaffolds are merely regioisomers to that of Gly.8AQ. Asn.8AQ, on the other hand, showed strong preference towards both Zn 2+ and Cd 2+ , the latter of which was never discovered as a metal ion capable of turning on fluorescence of related compounds before. In essence, the ability to obtain such useful information in a relatively short amount of time clearly underscores the fact that macroarray synthesis is indeed an efficient method in revealing new kinds of fluorescent sensors.
To capitalise on the observed signals, chemometric approaches were used to glean more information. First, all fluorescence signals were captured into a digital camera, which were then processed by the imaging software ImageJ. The percentage of red, green, and blue intensities were calculated, and the resulting data were used to create a multidimensional data set with 180 samples (17 metal ions and a blank ×10 replicates) and 12 variances (3 sets of %RGB ×4 fluorophores). Thereafter, the data were analysed by Linear Discriminant Analysis (LDA) and the resulting score plots for all metal ions are shown in Fig. 3, with each group representing the sensing capability of the combination of four sensors (Asn.8AQ, Gly.3AQ, Gly.6AQ, Gly.8AQ) on each analyte. The LDA score plot showed that most tested ions including Ba 2+ , Al 3+ , Cr 3+ , Fe 2+ , Co 2+ , Ni 2+ , Cu 2+ , Zn 2+ , Ag + , Cd 2+ , Pb 2+ , Hg 2+ can be clearly discriminated by this combined sensing system. Only a handful of group IA and IIA metal ions (Li + , Na + , K + , Mg 2+ , Ca 2+ ) could not be distinguished in this system -this is likely due to the well-documented properties of these metal ions such as the harder acid characters and the lack of π-bonding formation capabilities. In addition, LDA with the Jackknife (Leave-One-Out) approach was performed to investigate the predictive ability of the sensing array on the ions. The Jackknife classification yielded 72.35% accuracy in placing all ions into the sensing system (Supplementary Table S1), while the successful prediction was raised up to 93.64% with the exclusion of the aforementioned group IA and IIA metal ions (Supplementary Table S2). This is in good agreement with the LDA score plots.
Apart from the multiple detections of metal ions by chemometric approaches, a ready-to-use, paper-based device for detecting a single metal ion with naked-eye observation was also fabricated and evaluated for its sensitivity. Specifically, Gly.8AQ was in-situ synthesised on paper support as usual, followed by wax printing to generate confined channels for sample dropping. Thereafter, several concentrations of Zn 2+ ion were dropped to each channel containing covalently bound Gly.8AQ, followed by a data collection process (Fig. 4A,B, numerical data in Supplementary Table S3). The digitised data were then used to create a calibration curve, which then revealed that Gly.8AQ gave a good linear response to the Zn 2+ concentrations between 0-1 μM. At higher concentrations, the responses reached a plateau. Interestingly, the limit of detection (LOD), obtained from the calibration curve, was found to be about 200 nM (13 ppb), which is comparable to several works using more complicated methods www.nature.com/scientificreports www.nature.com/scientificreports/ in determining Zn 2+ ion 46,47 . In addition, the maximum concentration tested, which gave very clear change by naked eyes, was still well below the minimum tolerable limit in bottled waters based on the Code of Federal Regulations at 5 ppm 48 .
This cellulose-bound sensor also offers two additional benefits for sensing application. First, repeating steps of spotting and drying the analyte solution on the detection zone should increase the amount of the analyte in contact with the sensing unit without washing it away. This was found to be a practical way to boost the sensitivity of the sensor by least 10 times (LOD = 13 nM, Fig. 4C,D) when the photographic image was analysed after 10 cycles of spotting and drying steps. The proof of principle here should also be applicable to improve sensitivity of other cellulose-bound sensors or any dryable surface-bound sensors. Second, a reusability test was conducted to exploit the surface-bound nature of the sensor. By simply incubating the Zn 2+ -bound sensor with EDTA, a chelator for Zn 2+ , followed by aqueous washing, the sensor was free and ready to sense Zn 2+ again. Using this washing process, the cellulose-bound Gly.8AQ can be easily reused at least 10 times (Supplementary Fig. S4). In addition, the same concept was applied to two other prominent examples including Asn.8AQ with Zn 2+ and Cd 2+ ( Supplementary Fig. S5), where the determination of LOD could be done in a similar manner. Taken together, this level of performance is considered to be superior to previous paper-based sensors for Zn 2+ (Table 1). Clearly, all of them showed higher LOD values/working concentrations while only few studies reported the reusability of their sensors. Hence, the whole process very well demonstrated the utility of the macroarray synthesis in expediting the discovery of compounds with great practicality in sensing applications. To further confirm the success of the synthesis, and to indirectly characterise the immobilised molecules on support, we synthesised selected compounds containing the suggested scaffolds from the results in Fig. 2 (see the method section) 33 , fully characterised them by NMR spectroscopy and MS, and studied their fluorescence and UV-vis properties. For instance, 2-amino-N-(quinolin-8-yl)acetamide (surface-free version of Gly.8AQ) did show preferential increase in fluorescence upon binding to Zn 2+ (Fig. 5). The fluorescence titration ( Supplementary  Fig. S6) also confirmed the gradual decrease of the fluorescence at the emission maximum around 420 nm with concomitant increase in the fluorescence intensity at 500 nm under the higher concentrations of Zn 2+ . UV-vis titration ( Supplementary Fig. S7), on the other hand, showed that the absorption maximum around 300 nm was decreased in absorbance with a slight increase in the region around 330-400 nm when the concentrations of Zn 2+ was increased. On the contrary, it was shown that the fluorescence of 2-amino-N-(quinolin-6-yl)acetamide (surface-free version of Gly.6AQ) was increased upon incubation with Cr 3+ . A similar fluorescence titration ( Supplementary Fig. S8) revealed that there was a single emission maximum (450 nm) in this case, whose fluorescence intensity was increased in a direct relationship with the increasing Cr 3+ concentrations. UV-vis titration ( Supplementary Fig. S9) of this sensor-metal ion pair appeared to be slightly complicated, with a general trend being that the absorption around 270-320 nm was significantly increased with higher Cr 3+ concentrations. This behaviour was also found to be similar for the interaction between the same ligand and Al 3+ ion (Supplementary Figs S10 and S11). Interestingly, although surface-free Gly.8AQ could give much higher fluorescence intensity when bound to Zn 2+ than did all other cases, the turn-on ratio (the fluorescence ratio of Cr 3+ -Gly.6AQ complex over free Gly.6AQ) was slightly higher than that of Gly.8AQ with Zn 2+ (ca. 25 folds vs 22 folds) due to much lower background in free Gly.6AQ. This indicated that both can function well as sensors for their respective metal ion partners, and suggested that the synthetic strategy used in this study was an effective one to rapidly evaluate fluorescence properties of a variety of molecular scaffolds.
Last but not least, titration experiments via MS and NMR were also conducted to further confirm the binding interactions between selected pairs of ligands and metal ions. For MS titrations, as shown in Supplementary  Fig. S12, surface-free Gly.8AQ exhibited both 1:1 and 2:1 binding modes (ligand:metal ion) when Zn 2+ ion was sub-stoichiometric amounts, and 1:1 binding seemed to be dominant when the molar ratio was around 1:1. On the other hand, Gly.6AQ interacted with Cr 3+ in a more complicated manner, showing only 3:2 binding mode (ligand:metal ion) once the molar ratio reached 1:1 (Supplementary Fig. S13). For NMR titration, the set of Gly.8AQ and Zn 2+ showed several apparent shifts, with the most notable one being the methylene protons www.nature.com/scientificreports www.nature.com/scientificreports/ (starting at around 3.5 ppm) when more metal ion was added into the solution (Supplementary Fig. S14). All of these evidences clearly pointed out that there were indeed some significant bindings between the ligands and the metal ions used in this work.
In conclusion, the macroarray synthesis approach was implemented herein to rapidly generate a variety of fluorophores, which were readily amenable to direct screenings to discover metal ion sensors. Using chemometric approaches, combined fluorophore systems could predict several metal ions with high prediction accuracies, while the fabrication of a single fluorophore to detect Zn 2+ gave a practical paper-based sensor with great sensitivity and reusability. Lastly, this information could be reversed to yield typical solution-based sensors, hence demonstrating that the synthesis methodology is also a very useful tool for rapid discovery of even larger and more diverse libraries of fluorescent sensors.

Methods
General information. All reagents were purchased from commercial sources and used without further purification. All metal ions used in this study were in the form of nitrate salts except Fe 2+ , where the acetate salt was used instead. Paper used in this study was Whatman ™ Grade 1 Qualitative Filter Paper, which was used without further treatment. Solutions were made with Milli-Q water from ultrapure water systems with a Millipak 40 filter unit (0.22 μm, Millipore, USA). NMR spectra were acquired on a Bruker NMR spectrometer at 400 MHz ( 1 H) and 100 MHz ( 13 C). The high resolution mass spectra were obtained from an electrospray ionisation MS (micro-TOF, Bruker Datlonics). Absorption spectra were measured by using Varian Cary 50 UV-vis spectrophotometer. Fluorescence spectra were recorded on a Varian Cary Eclipse spectrofluorometer. Mass titration spectra were obtained from an electrospray ionisation MS (Thermo Scientific TSQ Quantum EMR triple quadrupole MS) with the following parameters: positive mode; spray voltage at 3000 V; capillary temperature at 300 °C; tube lens offset at 84 V).
Generation of aldehyde functional groups on paper. Paper sheets were washed with a series of solvents (DMF, 10 min, 3x, then MeOH, 10 min, 3x) prior to chemical functionalization. Thereafter, paper sheets were immersed into a solution of NaIO 4 (10 mM) and LiCl (100 mM) premixed in 10 mL H 2 O at 55 °C for 1 h. This was followed by washing with Milli-Q water (10 min, 3x), MeOH (10 min, 3x), and air-dried.  Carbodiimide coupling with reagent Y. A solution of each reagent Y (aminoquinolines or an aminonaphthalimide, 1 mmol) with HOBt (154 mg, 1 mmol) was prepared by adding 1 mL of DMF into pre-weighed reagents. The resulting solution was mixed with DIC (126 μL, 1 mmol) and the new mixture was immediately used for spotting (2 μL) onto desired areas on carboxyl-terminated papers (surface 2 in the main article). The sheet was incubated at room temperature for 5 min, and the spotting was repeated two more times. The process ended by washing the sheets with HEPES buffer pH 5 solution (10 min, 1x), HEPES buffer pH 7 solution (10 min, 1x), DMF (10 min, 5x), and then MeOH (10 min, 3x). After air drying, the paper sheets were ready for metal ion sensing.
Final finishing with wax printing. If desired, finished sensors can be patterned with hydrophobic wax as shown in our studies. This was done by a Xerox ColorQube 8870 wax printer to create circular patterns (diameter at 7 mm) on paper (which was designed by Adobe Photoshop). Heating at 110 °C for 30 sec on a hotplate was then performed to melt the wax and create the final hydrophobic patterns (diameter of around 5 mm).

Probing fluorescence response with various metal ions. First option -immersion method.
Each finished paper sheet was immersed in a solution of a metal ion (10 μM in HEPES buffer pH 7). The solution was incubated for 30 min. The sheet was air-dried and exposed to UV light. A handheld UV lamp was used to illuminate directly on the paper for fluorescence observation.
Second option -spotting method. Each desired area on a finished paper sheet was spotted (2 µL) with a solution of a metal ion (10 μM in HEPES buffer pH 7). The sheet was incubated for 10 min. The sheet was air-dried and exposed to UV light. A handheld UV lamp was used to illuminate directly on the paper for fluorescence observation.  www.nature.com/scientificreports www.nature.com/scientificreports/ Chemometrics. ImageJ software was used to extract all red, green, and blue color intensity from each reaction data points. These RGB values were then used to calculate for percentage of each color by using the formula: %R = (R/(R + G + B) × 100. The same can be done for %G and %B.

Digital conversion and quantitative analysis of fluorescence signals.
Linear discriminant analysis (LDA) was used with a data set consisting of 180 samples (17 metal ions and blank ×10 replicates) and 12 variances (3 sets of %RGB ×4 fluorophores). The calculation and plotting were performed using in-house codes based on MATLAB R2018a. Linear discriminant analysis (LDA) for the Jackknife classification provided percent classification accuracies. This is represented in the tables below (Tables S1 and S2). synthesis of representative fluorophores. Synthesis of (2-amino-N-(quinolin-8-yl)acetamide) (unbound Gly.8AQ). A mixture of 8-aminoquinoline (500 mg, 3.47 mmol), DMAP (21.18 mg, 0.17 mmol) and Boc-Gly-OH (1.22 g, 6.94 mmol) in dichloromethane (30 mL) was stirred in ice bath (~0 °C) for 30 minutes, followed by an addition of EDC·HCl (1.33 g, 6.94 mmol). The reaction mixture was stirred at 0 °C for 2 hours and stirred overnight at room temperature. The crude was extracted with NH 4 Cl (aq). The organic layer was dried over MgSO 4 , filtered and concentrated under vacuum. The crude product was purified on a column chromatography using ethyl acetate: hexane (2:3) as an eluent to afford (2-Boc-amino-N-(quinolin-8-yl)acetamide) as a white solid (910 mg, 3 mmol, 87%). This compound was stirred overnight with trifluoroacetic acid (0.46 mL, 6 mmol) in dichloromethane 20 mL at room temperature. The crude was extracted with saturated NaHCO 3 . The organic layer was dried over MgSO 4 , filtered and concentrated under vacuum. The crude product was purified on a column chromatography using ethyl acetate as an eluent to afford (2-amino-N-(quinolin-8-yl)acetamide) as a yellow solid (free base form, 437 mg, 2.2 mmol, 72%). 1 13 C NMR can also be found in Supplementary Fig. S15.

Data Availability
All data that are related to this study but not presented herein are available from the corresponding author upon request.