Harnessing Selectivity and Sensitivity in Ion Sensing via Supramolecular Recognition: A 3D Hybrid Gold Nanoparticle Network Chemiresistor

The monitoring of K+ in saliva, blood, urine, or sweat represents a future powerful alternative diagnostic tool to prevent various diseases. However, several K+ sensors are unable to meet the requirements for the development of point‐of‐care (POC) sensors. To tackle this grand‐challenge, the fabrication of chemiresistors (CRs) based on 3D networks of Au nanoparticles covalently bridged by ad‐hoc supramolecular receptors for K+, namely dithiomethylene dibenzo‐18‐crown‐6 ether is reported here. A multi‐technique characterization allows optimizing a new protocol for fabricating high‐performing CRs for real‐time monitoring of K+ in complex aqueous environments. The sensor shows exceptional figures of merit: i) linear sensitivity in the 10–3 to 10–6 m concentration range; ii) high selectivity to K+ in presence of interfering cations (Na+, Ca2+, and Mg2+); iii) high shelf‐life stability (>45 days); iv) reversibility of K+ binding and release; v) successful device integration into microfluidic systems for real‐time monitoring; vi) fast response and recovery times (<18 s), and v) K+ detection in artificial saliva. All these characteristics make the supramolecular CRs a potential tool for future applications as POC devices, especially for health monitoring where the determination of K+ in saliva is pivotal for the early diagnosis of diseases.

In such devices, specific organic molecules are employed in order to form a robust network of NPs, whose electrical features can be modulated by the analyte physical absorption and release. [20] Wohljen and Snow developed the first NPbased CR for vapor sensing. [11a] They fabricated a device featuring a film of octanethiolcapped Au NPs on interdigitated electrodes (IDEs) that was exposed to various volatile compounds such as toluene, tetrachloroethene, 1propanol, and water vapor. Variations of the electrical resistance of the device were ascribed to the inter action of the volatile molecules with the NP network, which undergoes swelling, producing structural changes. [11a] The same rationale has been successfully demonstrated by using specific CRs operating in liquid, whose geometrical layout is designed to minimize electrical contributions arising from the electrolytic solution. [21] As a result, Raguse and coworkers reported the first CR based on a thin film of Au NPs capped with 1hexanethiol monolayer inkjetprinted onto a microelec trode. [21] This sensor featured very interesting properties such as fast response time (<3 min) and high sensitivity (down to the 0.1 ppm range) toward toluene, dichloromethane, and ethanol molecules dissolved in 1 m KCl solutions. [21] Since then, many efforts have been devoted to the optimization of the sensing performance of CRs. The supramolecular interaction between analytes and receptors [22] followed by alterations of the elec tron transport through metal NP networks [23] acting as signal transduction, are two fundamental phenomena that make CRs extremely sensitive device for chemical sensing.
To the best of our knowledge, few examples of CRs based on metal NPs for ion sensing (viz. Cu 2+ , [24] and Zn 2+ , Cd 2+ , and CH 3 Hg + ) have been published so far. [25] However, these devices rely on the strong complex formation between the receptor and the cations of interest operating in dry state. Such a technology satisfies the demand for singleuse (i.e., disposable) sensors, but it cannot provide a sound option for continuous, realtime monitoring of the analyte of interest. To fill this gap, we report on the fabrication of metal NPbased CRs capable to perform realtime and selective sensing of K + in water and in complex matrices, such as artificial saliva, driven by supramolecular interactions. [26] Crown ethers, which are wellknown macro cycles able to complex ions via reversible hostguest interac tions, [27] bearing two peripheral thiol groups were devised to conceive a 3D covalentlylinked network of Au NPs. Here, dibenzo18crown6 (DTDB18C6) molecules act as both linkers to build up the 3D Au NPs networks and supramolecular K + receptors. [28] The network fabrication is achieved by an itera tive deposition of Au NPs and DTDB18C6 molecules through a covalent layerbylayer (LbL) assembly technique onto a glass substrate featuring IDEs on top. We built up a standardized deposition protocol of these two components in order to pre cisely control the network assembly by evaluating important parameters such as the Au NPs size, adhesive layers, and depo sition kinetics. The CR performance was studied and optimized in terms of number of deposition steps (DSs) for network for mation, NPs size, as well as the IDEs layout. The selectivity of the sensing platform toward K + was analyzed in the pres ence of the most common interfering metal cations present in bodily fluids (Na + , Ca 2+ , and Mg 2+ ). We have also performed the monitoring of K + in water by means of a microfluidic system aiming for the development of a PoC technology able to provide realtime responses. Finally, we challenged the sensing perfor mance of our CR in artificial saliva in order to verify its effective sensitivity and selectivity in complex media. The successful K + detection in saliva endows such a CR the required characteris tics to be used as the sensing element in future technologies for alternative health monitoring and early diagnostics of diseases.

Chemiresistor Fabrication and Operation
The assembly of the 3D covalent network for the fabrication of CRs relies on two key components: i) citratestabilized Au NPs [29] (Figures S1 and S2, Supporting Information), and ii) a dithiomethylene dibenzo18crown6 ether, DTDB18C6 (Scheme S1, Figures S3-S7, Supporting Information). The former endows the sensing platform with excellent elec trical features for signal transduction, [11b] whereas the latter guarantees the covalent linkage of adjacent Au NPs and the supramolecular recognition of K + via wellknown host−guest interactions. [27][28] The fabrication of the CRs has been the subject of a deep investigation in order to optimize its overall sensing perfor mance. The final protocol consisted of 4 steps: i) substrate cleaning, ii) deposition of an adhesive layer for Au NP attach ment, iii) Au NPs deposition, and iv) DTDB18C6 grafting. The first two steps have been adapted from previous proto cols. [30] Aiming at a standardized deposition of Au NPs the choice of the adhesive layer is pivotal, as its chemical fea tures must provide both strong adhesion of Au NPs on the substrate surface and precise control over their spacing for molecular bridging with DTDB18C6 linkers. As mentioned in the Experimental Section, four approaches have been explored such as no adhesive layer, dithiolated tetra(ethylene glycol) (TEG), PDDA, and APTES ( Figure S8, Supporting Informa tion). No adhesion ( Figure S8a, Supporting Information), partial adhesion ( Figure S8b, Supporting Information), and uncontrolled grafting ( Figure S8c, Supporting Information) of Au NPs have been observed respectively by making use of the first three approaches. The best option turned out to be APTES ( Figure S8d, Supporting Information), which allowed www.afm-journal.de www.advancedsciencenews.com

(3 of 11)
© 2020 The Authors. Advanced Functional Materials published by Wiley-VCH GmbH a controlled deposition of Au NPs over the entire chip that is composed of a glass surface and Au IDEs. Although such a protocol guarantees a high control upon the first deposited layer of Au NPs, two further aspects have been investigated in order to reach the best sensing performance, namely the UV/Ozone surface treatment during substrate cleaning and the purification or not of the Au NPs ( Figure S9, Supporting Information). The former turned out to be detrimental for K + sensing because it creates negative charges on the sub strate capable to interfere with the supramolecular interaction exerted by DTDB18C6 ( Figure S9a, Supporting Information). The latter filters out the excess Na + cations and citrate anions in the Au NP solution. We observed that no purification leads to higher controlled substrate coverage, as the high ionic strength of the Au NP solution is responsible to keep the NPs closer with respect to a purified solution in bidistilled water ( Figure S9b, Supporting Information). Thus, the optimized conditions rely on APTES as an adhesive layer, no UV/ozone substrate treatment, and no previous purification of Au NPs.
The 3D Au NPsDTDB18C6 network was assembled onto glass chips by the LbL approach (Figure 1a) that can be sum marized as (i) APTES adhesive layer deposition onto glass substrate via immersion into a 1% v/v aqueous dispersion of APTES, (ii) Au NPs covalently attachment via immersion of the modified substrate into an aqueous dispersion of 0.5 mm Au NPs, (iii) DTDB18C6 macrocycles covalently grafted on the surface of Au NPs via immersion in a 1 mm DTDB18C6 Ndimethylformamide (DMF) solution. The iteration of step ii and iii defines the number of DSs for the network formation. Hereafter, one DS means the ensemble of step ii and step iii.
Concerning the CR operation and signal transduction, electrical impedance spectroscopy (EIS) has been chosen to ensure high sensitivity and avoid electrophoretic effects exerted by a standard DC bias, insofar reducing electrical drift and enhancing the reversibility of the sensor response. [21,31] Imped ance measurements provide many parameters that can be used as the output signal in a sensing device, such as the impedance module (Z), imaginary (Z′′), and real (Z′) parts of impedance, capacitance (C), and impedance phase (φ). [21,32] Our investiga tion is mainly focused on capacitive changes, where the CR's output signal (C p ) is obtained from equation (see details in the Supporting Information). (1) where ν is the frequency of the AC stimulus. Our rationale is similar to the one introduced by Raguse et al. who employed singlepair electrodes to measure the impedance response of functionalized Au NPs to the presence of organic analytes in water. [21] Here, we use an IDEs layout to maximize the sensor active area, allowing us to exploit the capacitance changes in the network due to K + intake and supramolecular recognition by the macrocyclic host. For this purpose, we ini tially identified the most appropriate CR operational frequencies by comparing the response of the bare and the networkcoated IDEs in a series of K + solutions with concentration spanning from 10 -6 to 10 -1 m ( Figure S11, Supporting Information). From the capacitance (C p )frequency response of the networkcoated IDE, we observed that 0.1 Hz is the best frequency, where C p increases by 443% in 10 -3 m K + , distinctively from C p increase in water (283%) and the response of the bare chip (Figure 1b,c). Indeed, at low frequencies (<100 Hz) charged species in solu tion have enough time to follow the alternating electric field to interact noncovalently with the supramolecular receptors in the Au NP network, contributing to the overall capacitance increase of the device. [32] At high frequencies (10 5 Hz), K + ions are unable to follow the electric field oscillations, and the C p corresponds to the geometric capacitance of the IDEs having the solution as the dielectric medium. [32] Other, intermediate frequencies were not able to provide undistinguishable output C p responses for the K + detection at varying concentrations, as a result of the recogni tion mechanism (Figure S11, Supporting Information). At such intermediate frequencies, the bare chips respond nonselectively to the increase of the K + concentration, very likely due to a cor responding increase of the electrolyte conductivity.

Optimization of the CR Sensing Capabilities
The CR's sensing core is the 3D covalent network assembled onto the IDEs. The proper functioning of this platform relies on the accurate control of the deposited Au NPs and their bridging by DTDB18C6 macrocycles. Thus, to optimize the CR performance, two keyparameters have been assessed: i) the assembly kinetics of the network individual components (viz. Au NPs and DTDB18C6), and ii) the number of DSs.
The kinetics has been characterized by monitoring the tem poral evolution of C p at 0.1 Hz for both of the individual net work components (Figure 2a). In particular, it takes 40 min for the DTDB18C6 molecules to displace completely the citrate anions of the solvation shell of Au NPs immobilized during the first DS ( Figure S12, Supporting Information). This process is pivotal for achieving reliable CR responses as citrate traces can act as unspecific sites capable to electrostatically compete against the supramolecular recognition of K + by the crown ether receptors, in addition to responding to several cationic interferents. The replacement of the negativelycharged citrate by DTDB18C6 results in a decrease of C p , likely ascribed to a decrease of the overall charges in the Au NPs immobilized in the network. The subsequent deposition of DTDB18C6 showed faster kinetics (≈15 min) with respect to the initial one. On the other hand, Au NPs showed slower kinetics for any DS. Although a plateau is not reached, we defined 45 min as Conversely to the DTDB18C6 attachment, the grafting of Au NPs resulted in a clear increase of C p , which can be ascribed to the formation of a 3D network endowed with larger active area and volume, along with the contribution of excess charges from stabilizing citrate molecules.
Concerning the number of DSs, we monitored the electrical resistance of the networks in dry state for DSs varying from 1 to 11. The measured CR's resistance (R) and corresponding DSs can be grouped in three regions: i) few DSs (>10 GΩ), ii) 5-6 DSs (2-5 GΩ), and 11 DSs (<1 GΩ), as shown in Figure 2b. Such a less resistive trend is consistent with the improved surface coverage by the 3D covalent network onto the IDEs, as revealed by scanning electron microscopy (SEM) images ( Figure S13, Supporting Information). For DSs equal to 1, 3, 6, and 11, we found surface coverages of 29.60%, 36.49%, 47.79%, and 85.89%, respectively. The higher the coverage, the more parallel paths for charge transport are created within the network, thereby decreasing the overall electrical resistance. The increase of the number of DSs is followed by a coherent increase of C p , both in pure water and in a K + 10 -3 m solution, due to the formation of a 3D network endowed with larger area and volume ( Figure 2c).
In order to evaluate the CR sensing capabilities, networks formed by 6 and 11 DSs have been tested ( Figure 2d). As previously discussed, networks composed of only a few DSs possess poor substrate coverage, which compromises the ulti mate CR sensing performance. Both tested networks (viz. 6 and 11 DSs) exhibited similar sensitivity (S) toward K + within the mmμm concentration range (i.e., 10 -3 -10 -6 m), as shown in Figure 2d. As the increase of the DSs does not reflect in a substantial gain of sensitivity or detectable K + concentra tion range, further sensing experiments were performed with CRs bearing networks formed by 6 DSs rather than 11 that are more time consuming (10 versus 15 h for their respective fabrication). Within this context, other parameters governing the CR performance were also evaluated, including the IDEs geometry (i.e., electrodes spacing of 10, 20, and 30 μm) and Au NP size (namely, ∅ equal to 12.5, 24, and 40 nm), as shown in Figures S14a,b, Supporting Information, respec tively. Minor changes have been observed while varying such a wide combination of parameters, hence IDEs bearing electrodes with 30 μm spacing along with 12.5 nm Au NPs have been selected as the final CR configuration for optimal sensing performance.

Device Sensing Performance
Several relevant aspects related to the CR performance have been investigated, such as the best EISrelated output signal, pHdependancy and reversibility, device stability, shelflife, and also the underlying sensing mechanism. Such investigations have been accomplished by impedance spectroscopy, optical spectroscopy, SEM, and Xray photoelectron spectroscopy (XPS).
From the different parameters obtained by EIS measure ments, such as the impedance modulus (Z), the imped ance imaginary (Z′′) and real parts (Z′), conductance (G p ), impedance phase (φ), capacitance (C p ), etc., we verified that the lowfrequency C p is the one presenting the most sensitive and straightforward CR signal translation of the selective detec tion of K + at different concentrations (Figures S15 and S16, Supporting Information). Lowfrequency C p changes are related to the ability of cations to follow the oscillations of the electric field, [32] thereby interacting with the supramolecular receptors within the network. From the C p output signal, a sensitivity of 0.075 pK -1 featuring a linear response within a K + concen tration range of 10 -3 to 10 -6 m with a correlation coefficient of R 2 = 0.9845 (Figure 3a).
For sensing applications in real bodily fluids (e.g., saliva, sweat, blood, to name a few), it is important to obtain a stable CR response upon media subjected to pH variations. Figure 3b shows that the CR response (C p ) varies to a maximum of 10% for measurements carried out in K + 10 -3 m for pH spanning from 3.5 to 11. This indicates that our CRs do not need any support of a buffer solution to give a reliable response, and practical applications can rely on a simple calibration procedure according to the medium's pH.
The stability and reversibility of the electrical signal are other two important aspects of real sensing applications. Here, we have monitored the continuous CR output signal in 10 -3 m K + for more than 1 h. We found the maximum relative standard deviation (RSD) of 1.8% ( Figure S17, Supporting Information). Furthermore, the device was exposed to water and 10 -3 K + repeatably in order to define the degree of signal reversibility related to cation recognition ( Figure 3c). The data were acquired after 1 min of the CR immersion in the solution of interest and then rinsed by bidistilled water. From Figure 3c, one can observe the complete reversibility of the CR output signal. Such an important outcome arises from the properties of our 3D covalent network based on Au NP and DTDB18C6 where K + detection is endowed by noncovalent hostguest interac tions. This hints that this technology overcomes the usual limi tation of singleshot sensors. [7,9a] Shelflife is another prerequisite for the development of robust sensing technologies. Aiming at assessing the stability of our CRs, we compared the response of devices stored in air and water. We monitored the CR resistance in dry state and C p in water (Figure S18, Supporting Information). As previously mentioned, dithiolated DTDB18C6 molecules are handled in an O 2 free condition for the network assembly to avoid the formation of polymeric disulfides. [33] For this reason, CRs stored in air showed a marked instability when compared to those kept in water, which can be ascribed to the different oxygen content in the two environments (viz. dissolved oxygen in air and in water conrresponds to 21% and 1%, respectively). This rationale is supported by XPS data ( Figure S19, Supporting Information), which reveal a dramatic increase of the physically adsorbed sulfur with respect to covalently bonded one for devices stored in air. This is accompanied by the appearance of a new peak at 169 eV related to the SO 4 2bond and hence to the oxi dized sulfur. On the contrary, the stability of Au NPs is not affected by the different types of storage, as also verified by XPS ( Figure S19a,c, Supporting Information). Within 45 days of storage in water (data not shown), the CR resistance and capacitance varied less than 1.4% each. As aforementioned, the sensing mechanism is expected to be governed by the reversible adsorption/desorption of K + occurring with the supramolecular recognition moieties of the Au NPsDTDB18C6 network. Such an interaction could modify the network's structure (e.g., via swelling), changing its electrical properties. Thus, variations of the CR output signal would be ascribed to such structural modifications rather than the recognition action by the receptor molecules. To investigate a possible swelling, we evaluated the optical properties of the Au NPsDTDB18C6 network during the hostguest pairing by immersing the device in pure water and in K + 10 -3 m solution ( Figure S20, Supporting Information). The optical extinction spectrum of the 3D covalent network features a plasmon band at 516 nm related to the dipole resonance of individual Au NPs, as well as a broad band at 850 nm due to the plasmon coupling between NPs in close proximity. [34] The network exposure to K + does not affect such optical features (i.e., no wavelength shift), suggesting that swelling does not occur during the sensing event. This indicates a pure electronic effect resulting from the supramolecular interaction between K + and the 3D cova lent network as the main mechanism at the basis of the sensor operation.

Selectivity
Selectivity is one of the crucial figures of merit of a chem ical sensor, hence the CR response toward Na + , Ca 2+ , and Mg 2+ interferents at relevant concentrations has been tested. Figure 4a,b illustrate the sensor response and respective sen sitivities toward the different tested cations. The recorded device sensitivity is at least one order of magnitude higher for K + when compared to other tested cations. Our sensing experi ments have been conceived by conditioning the device into the cation solution of interest for one min before performing the EIS measurements. Given the comparable, but slightly lower, affinity of crown ethers toward Na + , [35] the CR response to the presence of Na + was registered for longer times to rule out the possibility of recognition events occurring at slower rates ( Figure S21, Supporting Information). No capacitive changes have been recorded on the timescale of ≈1 h, suggesting that no slow recognition processes compensating the lower affinity constant of DTDB18C6 toward Na + occur. Provided that one min conditioning time is needed for reliable measurements ( Figure S21, Supporting Information), the outcome of these experiments not only proves the selectivity of our CRs, but it also validates our hypothesis for which these POC sensors can afford a fastreliable response for practical applications like PoC sensing technologies.
The sensing mechanism of our CRs relies on the chemical affinity of the DTDB18C6 receptor toward K + as a result of the match between the cation radius and the diameter of the crown ether cavity. To exclude that other noncovalent interac tions govern the CR sensitivity and selectivity-such as any role played by Au NPs or peculiar characteristics of the chip as a result of its fabrication-we performed further investiga tions with reference systems. Devices bearing as molecular receptors 1,10decanedithiol and dithiolated TEG ( Figure S22, Supporting Information), whose contour length along the main  axis is similar to that of the DTDB18C6 receptor (≈16 Å), were thus engineered. The former linker lacks oxygen atoms in its structure, whereas the latter molecule (despite having oxygen atoms) does not possess a cagelike structure acting as a cation scavenger. The response of CRs based on net works formed by these two receptors, in addition to the bare device (i.e., in absence of active material inbetween the IDEs), have been compared to the performance described by devices having networks bearing DTDB18C6 receptors (Figure 4c,d). Bare devices do not show any selectivity toward K + , ruling out any unspecific interaction related to the chip characteris tics (e.g., surface hydrophilicity) on the device response. CRs having 1,10decanedithiol linkers showed a trend similar to the bare device, thereby highlighting the crucial role exerted by the molecule's oxygen atoms as a cation interaction site. Devices containing dithiolated TEG molecules showed some response toward K + , however, a nonlinear output signal as concentration increases were observed, in stark contrast to the linear response exhibited by DTDB18C6 linkers over a broad K + concentration. Such a lack of linearity prevents us to determine the sensitivity of CRs made of dithiolated TEG receptors. More importantly these devices were found to respond similarly to other cations (viz. Na + , Ca 2+ , and Mg 2+ ), displaying very limited selectivity ( Figure S23, Supporting Information). Slight differences in the intensity of the signal output though were found for mono and bivalent cations. These experiments not only point out the importance of having oxygen atoms in the molecular receptor for interactions with metal cations, but also the presence of the crown ether supramolecular cage structure and the 3D arrange ment of such receptors in order to attain high cation selectivity and sensitivity by means of electrical transduction ( Figure S24, Supporting Information).

Continuous Sensing under Analyte Flow
Realtime monitoring of analytes continuously varying their concentration is a decisive characteristic in several chemical sensing applications. [36] Aiming to reach this highdemanding goal, a microfluidic cell has been designed to deliver small amounts (0.4 mL min −1 ) of analyte on demand to the sensor surface. For this purpose, a new IDE layout has been designed to be accommodated in such a fluidic platform (Figure 5a and Figure S25, Supporting Information). In addition to the possi bility to perform realtime measurements, different technolog ical drawbacks can be avoided by using a microfluidic platform, for example, no exposition of the device to adventitious contam inations from the environment, no solution evaporation during longterm analysis, drastic reduction of the sample volume, and standardization of the sample handling.
Realtime analyte monitoring has been performed in a way to verify both the CR sensitivity and the reversibility of its output signal (C p ). This means that two specular ramps of K + concentra tions have been explored (i.e., increase from 10 -6 to 10 -3 m, and  from 10 -3 to 10 -6 m, with a pace of one decade) inter spersed by a rinsing step in water before each K + concentration (Figure 5b). This experiment lasted 6000 s by using a constant solution flow equal to 0.4 mL min −1 ; allowing us to record 10 con secutive measurements at each analyte concentration. Regarding the kinetics of complexation/release, the noncovalent interac tion between K + and receptor is faster than the minimum C p acquisition time (viz. 18 s). Upon water rinse, the device showed a perfect recovery of its basal signal, which hints to comparable kinetics for complexation and release. Additionally, no electrical degradation (RSD lower than 1.5% for any experimental point) or significant signal drift have been observed in the time scale of such experiments. Furthermore, the reversibility of K + detec tion has been successfully proved, with a sensitivity of (105 ± 8) × 10 -3 pK -1 and (101 ± 8) × 10 -3 pK -1 for both forward and backward ramps of K + concentrations, respectively.

K + Monitoring in Artificial Saliva
One of the main applications for POC K + sensors refers to the development of novel, noninvasive technologies for health monitoring. [37] For this purpose, the devised sensor should be able of detecting K + in complex physiological media, for example, saliva-one of the most accessible bodily fluids, along with sweat-where variations of K + concentration can indicate the presence or the early stage of a disease in course, such as cystic fibrosis. [38] Here, we have tested the performance of our CRs in artificial saliva as a proofofconcept K + sensor for the development of future, noninvasive POC technologies for health monitoring.
Since the physiological content of K + in saliva varies between 3 and 20 mm, the CR calibration curve has been obtained by preparing K + spiked artificial saliva featuring a concentration between 0.5 and 50 mm (Figure 6) followed by a 1:100 in dilu tion in Milli Q water. The CR output signal has been recorded in static mode, namely by manually introducing every spiked aliquot of saliva onto the device surface. The linear regression of the device response yields a sensitivity equal to (76 ± 1) 10 -3 pK -1 which is coherent to the previously reported benchmark test in aqueous K + solution. The CR performance (viz. sensitivity) is not jeopardized by the presence of interfering analytes such as Na + , Mg 2+ , Ca 2+ , or any other compound present in artificial saliva, confirming its efficiency as a selective K + sensor in com plex media. We attribute the successful K + detection in both aqueous solution and saliva to the finely tuned characteristics of 3D networks based on Au NPsupramolecular receptors which provide reliable, fast, and recoverable signals that are ideal for CRs acting as transducers in POC sensing technologies.

Conclusion
In summary, we built up a novel and robust protocol for the assembly of 3D networks based on Au nanoparticles and dithi omethylene dibenzo18crown6 ether supramolecular receptors a selective hosting material for CR sensing devices. A multiscale characterization allowed us to optimize the deposition method, the material's characteristics, and the device architecture Figure 5. a) Picture of the microfluidic system, b) CR response under a flow of K + at different concentrations. The K + concentration was initially increased from 10 -6 to 10 -3 m and then decreased from 10 -3 to 10 -6 m, interspersed by a rising step in water before each K + concentration, c) calibration curve for the CR response with K + concentration increasing and decreasing (red curves). The CR response in water is also shown (blue curve). including the Au NP size, the adhesive layer, molecular receptor, the kinetics of the LbL assembly, and the IDEs geometry. Such finetuning made it possible to boost the CR performance. Our unprecedented technology showed outstanding figures of merits such as i) a robust shelflife (viz. at least 45 days), ii) high reversibility of the sensing signal, iii) full compatibility with flow measurements for realtime K + monitoring, iv) high sensitivity of (75 ± 1) × 10 -3 pK -1 (linear range between 10 -3 and 10 -6 m), and excellent selectivity (≈13 times higher sensitivity) toward K + in the presence of diverse relevant interfering cations (Na + , Ca 2+ , and Mg 2+ ), v) effective capability to detect K + cations in arti ficial saliva. Our findings provide unambiguous evidence that CRs bearing 3D networks precisely engineered with Au NPs and supramolecular receptors are ideal platforms for future develop ments of POC technologies for health monitoring, where the detection of K + in saliva can be used for the noninvasive, early diagnostics of diseases.
Synthesis of 24 and 40 nm Au NPs: The reaction mixture was cooled down to 90 °C after the synthesis of the 12.5 nm Au NPs in the same reaction vessel. Afterward, a solution of 25 mm HAuCl 4(aq) was added (1 mL). The reaction lasted 30 min, and it has been performed twice. Then, 55 mL was extracted from the reaction mixture, and subsequently, 53 mL of water and 2 mL of 60 mm sodium citrate (aqueous solution) were added. The resulting solution was then used as Au seeds for the following growing step. This process was repeated three or seven times to yield 24 or 40 nm Au NPs respectively.
Electrode Fabrication: Test patterns containing four pairs of Au/Ti (100 nm/20 nm thick) IDEs were produced by optical lithography on 15 mm × 15 mm glass substrates by Metrohm DropSens (Oviedo, Spain). Every chip had a total number of 36 fingers. The electrodes' width and their spacing were 10, 20, or 30 μm. The ratio between width and spacing was 1:1. Bottom contact IDEs on Si/SiO 2 (Fraunhofer Institute IPMS, resistivity ρ Si ≈ 0.001 Ω•cm, oxide thickness t ox = 90 nm) with 20 μm channel length were patterned by photolithography (AZ1505 photoresist and MIF726 developer, Micro Chemicals) using a Microtech LW405B laser writer. 5 nm Cr and 40 nm thick Au were thermally evaporated with Plassys MEB 300 following a lift-off process.
Fabrication of a Au NP Monolayer: Glass test patterns were copiously rinsed in acetone and isopropanol for cleaning, followed by 20 min treatment by UV/ozone (NovaScan, Digital UV/Ozone System). Afterward, the chips were immersed in 2 mL of the adhesive layer for a given time, rinsed, blown dry in N 2 , and subsequently immersed in 2 mL of the previously purified Au NPs aqueous solution for 3 h. To remove the excess of reagents, 12.5 nm Au NPs were centrifuged at 8228 g for 30 min and redispersed in the same volume of water. Three different adhesive layers were evaluated: 1) APTES aqueous solution (1% v/v) for 30 min, 2) PDDA aqueous solution (1 mg mL −1 , 0.5 M NaCl) for 15 min, and 3) dithiolated TEG solution in EtOH (5 mm) for 48 h. The effect of the UV/ ozone treatment and the purification of Au NPs were also studied. Finally, a PDMS reservoir was fabricated (silicon elastomer: curing agent 10:1, curing time 48 h at RT in vacuum) and attached to the test pattern surface for the sensing experiments ( Figure S11m, Supporting Information). The volume of the PDMS reservoir was ≈100-150 μL.
LbL Fabrication of Au NP-Networks: A glass chip containing a monolayer of Au NPs was placed inside a 5 mL vial, capped with a septum, and purged for 5 min with N 2 . Dithiomethylene dibenzo-18-crown-6 ether solution (1 mm) was prepared inside a glovebox in anhydrous DMF and in a septum capped vial. Afterward, 100 μL of DTDB-18C6 solution was injected in the vial that contains the glass chip and let react for 20 min. Then, the septum was removed, and the substrate was immediately rinsed with water, blown dry in N 2 , and 100 μL aqueous Au NPs solution was placed in the PDMS reservoir for 45 min. Later, the glass chip was rinsed with water, dried by an N 2 stream, and the following deposition of DTDB-18C6 was repeated. The alternated depositions of Au NPs and DTDB-18C6, which hereafter is denoted together as 1 DS, were repeated up to eleven times. Importantly, the last element of the depositions was always DTDB-18C6 molecules in order to remove all the citrate molecules from the surface of the Au NPs. Experiments using test patterns with different channel lengths (10, 20, and 30 μm) and with Au NPs of different sizes (12.5, 24, and 40 nm Au NPs) were performed in order to optimize the sensing performance of the CRs.
As control experiments, different molecular linkers for the formation of Au NP networks have been tested, such as 1,10 undecanedithiol, and dithiolated TEG, whose backbones share the same molecular length of the DTDB-18C6. Networks and related devices were assembled following the optimized protocol employed for DTDB-18C8 receptors.
Electrical Measurements: The devices were electrically characterized by DC and AC electrical measurements. DC measurements were performed measuring the two-terminal device resistance at a constant bias of 100 mV. All DC electrical measurements were carried out in a probe station in ambient conditions by using a Keithley 2612B Source Meter unit. AC measurements were performed by means of EIS employing a Metrohm Autolab PGSTAT204 potentiostat/galvanostat. The two-terminal EIS was recorded from 10 −1 to 10 5 Hz, swept from high to low frequencies, with a sine-wave voltage signal amplitude of 50 mV (root-mean-square, RMS). The device complex impedance (Z*) was converted into capacitance (C p ) using a simple parallel resistor-capacitor (RC) circuit (SI). Here, the measured C p is expressed as the areal capacitance by taking into account the IDE dimensions.
Electrical Detection of K+ Ions in Static Mode: All the electrical characterizations were performed in wet state. 100 μL of KCl solutions (concentrations ranging from 10 -1 to 10 -9 m) were added into the PDMS reservoir placed on the surface of the devices (bare device, Au NP networks interconnected with DTDB-18C6, 1, 10 undecanedithiol or dithiolated TEG). The capacitance was recorded after 1 min of conditioning time. Control experiments with other cations, using NaCl, CaCl 2 , and MgCl 2 , were performed in an analogous way.
Microfluidic Cell Fabrication: The microfluidic was designed ( Figure S25a, Supporting Information) to satisfy the geometrical requirements of the silicon chip ( Figure S25b, Supporting Information) and the probestation setup, where the electrical measurements have been performed ( Figure S25c, Supporting Information). The microfluidic cell was manufactured by a Stratasys Objet30 3D printer using VeroClear resin and has an internal volume of ≈12 μL. Aiming at getting a perfect sealing onto the silicon chip, the microfluidic cell was conceived to have a lower part which acts as a sample holder, where the silicon chip can be placed firmly ( Figure S25c, Supporting Information). The upper part is clamped to the lower one by two metallic screws, which allow controlling the pressure exerted onto the chip. The sealing is guaranteed by the presence of an o-ring upon the area where the IDEs are placed. The self-standing chip can be easily connected to the electrical probes and the tubes of the peristaltic pump ( Figure S25c, Supporting Information). This set-up has been used for the real-time monitoring of K + cations by impedance measurements.
Electrical Detection of K + Ions under Flow: The Si/SiO 2 chip was placed inside the microfluidic cell. A constant flow of KCl (0.4 mL min -1 of flow rate) was applied by using a Hei-FLOW-peristaltic pump and a Tygon tube (inner diameter 0.8 mm, wall thickness 1.6 mm). The capacitance response of the device to aqueous KCl solutions (concentrations ranging from 10 -6 to 10 -3 m) and water was recorded continuously for 1 h at 0.1 Hz.
Electrical Detection of K+ Ions in Artificial Saliva: 500 mL of K + -free artificial saliva were prepared using NaCl (1.02 g), NaH 2 PO 4 (0.17g), CaCl 2 2H 2 O (0.148 g), MgCl 2 6H 2 O (0.025 g), sodium carboxymethyl cellulose (5 g), and sorbitol (15 g). The pH was adjusted to 6.75 with NaOH. KCl solutions (concentrations ranging from 0.5 to 50 mm) were spiked in the saliva solutions. A dilution of 1:100 was performed by using Milli Q water and tested electrically as previously described.
Characterization: Thin layer chromatography was conducted on precoated aluminum sheets with 0.20 mm Merk Millipore Silica gel 60 with fluorescent indicator F254.
Melting Point: Melting point was measured on a Gallenkamp apparatus in open capillary tubes and has not been corrected.
NMR: NMR spectra were recorded on a Bruker Fourier 300 MHz spectrometer equipped with a dual ( 13 C, 1 H) probe. 1 H spectra were obtained at 300 MHz, 13 C spectra were obtained at 75 MHz NMR. All spectra were obtained at RT Chemical shifts were reported in ppm according to tetramethylsilane using the solvent residual signal as an internal reference (CDCl 3 : δ H = 7.26 ppm, δ C = 77.16 ppm). Coupling constants (J) were given in Hz. Resonance multiplicity was described as m (multiplet), d (doublet), and t (triplet). 13 C spectra were acquired with a complete proton decoupling. The attached proton test experiment was used to determine CH multiplicities in carbon spectra and the peaks assigned as singlet (s) for quaternary and doublet (d) for tertiary carbon atoms. Infrared spectrum (IR) was recorded on a Shimadzu IR Affinity 1S FTIR spectrometer in ATR mode with a diamond mono-crystal.
Mass Spectrometry: HRMS was performed on a Waters LCT HR TOF mass spectrometer in the positive ion mode. The Visible-NIR extinction spectra of the Au NPs solutions and devices were recorded by using a JASCO V670 UV-vis-NIR spectrophotometer. Optical images were acquired with an Olympus BX51 optical microscope. SEM imaging was conducted using FEI Quanta 250 FEG instrument operated in high vacuum mode (pressure in 10 −4 Pa range), with accelerating voltages of 30 kV for the incident beam. ImageJ was used to calculate the Au NPs size and the coverage of the substrates from SEM images. XPS analyses were carried out by using a Thermo Scientific K-alpha X-ray photoelectron spectrometer equipped with an aluminium X-ray source (energy 1.4866 keV). The spectra were recorded at a pressure of 10 −8 mbar in the main chamber. The X-ray spot size was settled at 250 μm. Survey spectra were recorded as a result of 10 scans with a pass energy of 200.00 eV and a step size of 1 eV; high-resolution spectra were an average of 10 scans with a pass energy of 50.00 eV and a step size of 0.1 eV. Analysis of the XPS data was carried out using Avantage software. All the presented XPS spectra and subsequent data analysis were performed following the subtraction of the Shirley background from the region of interest. The Au4f and S2p XPS spectra were fitted using components with Gaussian/Lorentzian lineshapes and without full-width-half-maximum values constrained. Neither the energy nor the relative area of the components was constrained.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.