A multi-ion electrophoretic pump for simultaneous on-chip delivery of H⁺, Na⁺, and Cl⁻

Bioelectronic devices aim to bridge the gap between biological system and electronics.^1,2 Signal transmission in biological system mainly relies on ion fluxes and the movement of biomolecules, such as neurotransmitters, while conventional electronic devices control the movement of electrons and holes.^3–5 Many tools have been developed to control or monitor the activities of ions and biomolecules with electronic or optical signals for the purpose of diagnostics and therapeutics.^6–8 Iontronic devices generate, store, and transmit signals via modulating the flow and concentration of ions and small molecules.^9,10 In analogy to electronics, examples of iontronic components include organic electrochemical transistors (OECTs),¹¹ organic electronic ion pumps (OEIPs),¹² ionic diodes,¹³ transistors,¹⁴ memristors,¹⁵ and bipolar membrane junctions.¹⁶ By manipulating ions and biomolecules directly, iontronic devices extend the domain from electrically excitable cells to all cell types, such as stem cells to guide differentiation^17,18 and skin cells for improved wound healing.^19,20 Furthermore, iontronic devices can target specific biological processes by using an ion or biomolecule of choice. These include H⁺ for enzyme activities, gene expression, and neuronal function,²¹ Cl⁻ for cancer and birth defects,²² and gamma-aminobutyric acid (GABA) for suppression of epilepsy.²³ 

Among these iontronic devices, OEIPs are used to deliver charged ions and biomolecules with high spatiotemporal resolution and dosage precision.¹² In recent years, OEIPs have been reported to deliver H⁺, K⁺, Ca⁺⁺, and GABA for triggering cell polarization status in vitro,¹² controlling epileptiform activity in brain slice models,²³ affecting sensory function in vivo,²⁴ suppressing pain sensation in awake animals,²⁵ and modulating plant physiology.²⁶ Some physiological processes need more than one species cooperation at the same time, such as the fact that the early embryonic face is patterned by H⁺ and K⁺ ion gradients.^27,28 Jonsson et al. reported a novel ion pump design that controls multiple delivery points with high temporal resolution.²⁹ To continue on this path, here, we demonstrate the first multiple-ion pump that can selectively deliver multiple ionic species simultaneously, including both anions and cations. This ion pump is composed of four independent reservoir electrolytes connected to one target electrolyte, in which 36 microelectrodes are independently controlled with 50 μm resolution. Here, we demonstrate the delivery of three major ions, H⁺, Na⁺, and Cl⁻. As a proof of concept, we also demonstrate closed-loop control of ionic concentration using a machine-learning algorithm.¹² Thus, this multi-ion pump not only enables controlling complicated physiological processes that need more than one ionic species but also provides a highly efficient and customized toolbox for fundamental biological research with an intelligent manner.

Figure 1 shows the top view of the multi-ion pump, which comprises four independent reservoirs (R1, R2, R3, and R4) connected to one target. Each reservoir has one reference electrode and is connected to nine independent microelectrodes (3 × 3 array) in the target through an independent ion channel. Figure 1(b) shows the mapping of the 36 microelectrodes in the target: every four microelectrodes form an array that is capable of delivering different ionic species. Here, we used three reservoirs to deliver H⁺, Na⁺, and Cl⁻ and named each individual type as H⁺ pump, Na⁺ pump, and Cl⁻ pump, respectively [Fig. 1(b)]. Nanoparticles were coated on the microelectrodes to increase the surface area and electrode capacitance. The microelectrodes of the H⁺ pump were coated with Pd NPs, which are selective to H⁺ by forming PdH_x.³⁰ AgCl nanoparticles were coated on microelectrodes of the Cl⁻ pump because of its selectivity to Cl⁻.³¹ For all the other general ionic species, platinum nanoparticles (Pt NPs) were deposited on the microelectrodes to increase the surface area and the capacitance for general delivery, such as Na⁺. The reference electrodes in the reservoir and auxiliary electrode were also AgCl NP coated, which work as an electron-to-ion transducer.³¹ On top of the microelectrodes, we deposited four independent ion channels that connected reservoirs and the target, which are ion conducting materials that allow ions to move following an external electric field.³² Polymers with negative charges are selective to the cation and are used as ion channels for cation delivery, such as poly(2-acrylamido-2-methyl-1-propanesulfonic acid), while polymers with positive charges are selective to anion and are used as ion channels for anion delivery, such as poly(diallyldimethylammonium chloride).³³ Here, we chose Polyvinyl Alcohol: Polystyrene sulfonate (PVA:PSS) and PVA:Chitosan as ion channels. These ion channels are spin-coated and patterned by photolithography into two layers separated by parylene. Then, a SU8 photoresist was used to build the microfluidic channels to provide solutions for the reservoirs and target [Fig. 1(c)]. The microfluidic channels were sealed by microfluidic tape. More details about the device fabrication can be found in Sec. IV and Fig. S1. Figure 1(c) represents the side view of the multi-ion pump, showing the multilayer structure and the electrical circuits with H⁺ reservoir as the representative.

We demonstrated that we can control the ion concentration of three ions: H⁺, Na⁺, and Cl⁻ (Fig. 2). In Fig. 2(b), first, V_1R = 1.6 V, applied between the R1 in the H⁺ reservoir and H⁺ microelectrodes, delivers H⁺ to the H⁺ channel. Then, V₁ = 1.6 V delivers H⁺ from the microelectrodes into the target and increases [H⁺] in the area close to the microelectrodes as monitored by 5-(and-6)-Carboxy SNARF^™-1. Next, we reversed V₁ to be −1.6 V to absorb H⁺ back into the H⁺ channel, inducing a [H⁺] decrease in the target. Figure 2(b) shows the [H⁺] change following V₁, indicating a quick and precise [H⁺] control by the voltage. [Na⁺] control was achieved by the same protocol as that of [H⁺] with 0.1M NaCl in the reservoir and CoroNa as the fluorescence indicator [Fig. 2(c)] because they are all cations. For [Cl⁻], 0.1M NaCl was in the reservoir electrolyte and MQAE was used as a fluorescent indicator in the target electrolyte to monitor [Cl⁻] in real time. Here, V_4R = −1.8 V between R4 in the Cl⁻ reservoir and Cl⁻ microelectrodes delivers Cl⁻ into the Cl⁻ channel. Then, V₄ = −1.8 V delivers Cl⁻ to the target, increasing [Cl⁻] of the area close to the microelectrode as measured. Next, we reversed the polarity of V₄ = 1.8 V to absorb Cl⁻ and reduce [Cl⁻] in the target [Fig. 2(d)]. In addition to the temporal resolution, the high spatial resolution control of ion pumps is another critical advantage for ion/biomolecule delivery. In Fig. 2(e), when we applied voltage on one working microelectrode (highlighted in yellow square), the fluorescence intensity in the adjacent area (black, red, blue, and pink rectangular square) was measured to show the ion concentration gradient. Following the protocols that change [H⁺], [Na⁺], and [Cl⁻], we measured and plotted their fluorescence intensity in different areas [Figs. 2(f)–2(h)]. The results show that while the monitored area gets further from the working microelectrodes, the ion concentration change by the voltage is diminished, indicating a precise control of spatial resolution.

Each microelectrode was designed to function independently. In addition, the demonstrated spatial ion fluxes also control by changing [H⁺] in different directions with two microelectrodes. In Fig. S2, we applied 1.6 and −1.6 V on two microelectrodes, and their [H⁺] changes are in the opposite way, indicating a precise dynamic control. We also presented a video that shows the control of each H⁺ microelectrode in sequence in the supplementary material.

To apply ion delivery to biological processes is important to be able to closely control ion concentration toward a specific target value.^31,34,35 Traditional control methods are difficult for biological systems due to their complex dynamics and sensitivity to environmental changes.³⁶ Therefore, integrating the machine-learning-based closed-loop control with the versatile multi-ion pump platform could introduce a powerful toolbox to further manipulate complex biological processes. Here, we successfully demonstrated the automated closed-loop control of ion fluxes by machine learning using the multi-ion pump. Figure 3(a) shows the general architecture of the implemented online machine-learning-based controller designed for the multi-ion pump system. By the user choice, the ML-controller uses one of the four sub-controllers³⁴ for controlling the predefined ions. The blue line represents the reference signal, and the red line represents the system output (i.e., fluorescence intensity). In Figs. 3(b)–3(d), we set different reference signals for different ions with blue lines. Then, we applied voltage through the controller to the corresponding microelectrodes and made the output to track the reference signals. The results show that we were able to control the ion fluxes of H⁺, Na⁺, and Cl⁻ following different patterns. In addition, the output of the controller applied to the multi-ion pump and the tracking error of the system are provided in Fig. S3. The tracking performance of the multi-ion pump shows a promising response to the controller commands.

In summary, we presented the first bioelectronic device capable of electrophoretic delivering multiple ionic species simultaneously to date, including both cations and anions. The delivery sites are composed of 36 independent 50 μm microelectrodes that are able to achieve an output of different ions (H⁺, Na⁺, and Cl⁻) with high spatiotemporal resolution. Additionally, we demonstrated machine-learning-based closed-loop control to maintain a specific set point of ion concentration for precisely affective biological processes. The control of multiple ionic species opens the chance to manipulate complicated physiological processes, such as cell polarization, that we have already manipulated with H⁺¹² and Cl⁻³¹_.

A multiple-ion pump platform was fabricated on a 4 in. borosilicate glass wafer.

• Step 1: Au contacts and traces were patterned by a positive photoresist (S1813; Micro-Chem Corp.) and deposited by e-beam evaporation (10 nm Ti, 100 nm Au). Acetone and IPA are used for liftoff.

• Step 2: Subsequently, a S1813 photoresist is again used to selectively expose the Au contacts for nanoparticle electrodeposition(see Sec IV B).

• Step 3: After electrodeposition, a 1.5 µm thick insulating layer of parylene-C was deposited (Specialty Coating Systems Labcoter 2 system) in the presence of an A174 adhesion promoter. The parylene was etched by an oxygen plasma with the regions over the electrodes and contact pads exposed and the rest protected by SPR220-4.5 or SPR220-7 (Micro-Chem Corp).

• Step 4: Prior to the deposition of ion channels, (3-glycidyloxypropyl) trimethoxysilane (GOPS) was deposited on the wafer to promote adhesion of the polymer. 5% GOPS was dispersed in ethanol and spin coated at 1000 rpm for 30 s and then baked at 110 °C for 5 min.

• A blend of 8 wt. % polyvinyl alcohol (PVA) with 2 wt. % polystyrene sulfonic acid (4:1 weight ratio) was thoroughly mixed by using a microwave and hotplate. The PVA:PSS solution was filtered by using a cellulose ester (MCE) syringe filter with 0.8 μm pore size and spin-coated on top of the wafer at 1500 rpm for 30 s and baked at 120 C for 2 h, yielding a film thickness of 2 μm. A positive photoresist Dow SPR220-4.5 was spin-coated following the protocols of the manufacturer. The PVA:PSS film was etched with an oxygen plasma with the desired pattern defined with the SPR220-4.5 photoresist.

  Steps 5 and 6: A second 1.5 µm coating of parylene was then deposited with the same protocol as above to insulate and protect the PVA:PSS film by only exposing the 36 microelectrodes.

• Step 7: To promote adhesion between the parylene and the next polymer layer, PVA:Chitosan, GOPS was again deposited using the aforementioned process prior to PVA:Chitosan patterning. Chitosan is dissolved in 1% acetic acid and mixed with 10 wt. % PVA (1:2 weight ratio) thoroughly with the help of a microwave and hotplate. The PVA:Chitosan solution was filtered by using a cellulose ester (MCE) syringe filter with 0.8 μm pore size and spin-coated on top of the wafer at 1000 rpm for 5 s with 500 rpm/s ramp and then 4500 rmp for 30 s with 1500 rpm/s and baked at 80 C for 2 h, yielding a film thickness of 2 μm. The PVA:Chitosan film was etched with an oxygen plasma with the desired pattern defined with a SPR220-4.5 photoresist.

• Step 8: A third 1.5 µm coating of parylene was deposited with the same protocol as above to insulate and protect all the polymers from the subsequent SU8 deposition.

• Step 9: To promote adhesion between the parylene and SU8 photoresist, GOPS was again deposited using the aforementioned process prior to SU8 patterning. SU8 3025 was spun onto the wafer at 500 rpm for 5 s with 100 rpm/s ramp, 1000 rpm for 30 s with 300 rpm/s ramp, and 3000 rpm for 1 s with 3000 rpm/s ramp. The patterned 70 μm high SU8 photoresist formed the sidewalls of microfluidic channels for reservoir and target chambers.

• Step 10: The third parylene insulation layer protecting the polymers was etched to expose the electrode contacts in the reservoir and target channels using the same process as the previous parylene etch.

• Step 11: Finally, devices were diced from the wafers prior to sealing the microfluidics with single-sided microfluidic transparent diagnostic tape (3M 9964). Features in the tape layer were punched out with 1 and 2 mm diameter biopsy punches for exposing imaging area and fluidic inlets. The tapes were then aligned to SU8 features on the device and pressed to seal by hand. PDMS was also punched with the 1.5 mm diameter biopsy punches to provide support from the fluidic inlets.

Electrodeposition is accomplished with a three-electrode configuration at room temperature with an Ag/AgCl pellet as a reference electrode and a platinum wire coil as a counter-electrode using an Autolab potentiostat. The following procedure yields the most repeatable and stable results among several plating procedures tested.

1 wt. % PdNO₃ solution was diluted from 10 wt. % PdNO₃ and used to electroplate the nanoparticles by applying a DC voltage of −0.3 V for 5 s. During this process, 10 μC charges are measured in the circuit, indicating that a single microelectrode with Pd NP coating theoretically has the capability of converting 10 μC electrons to H⁺ and vice versa.

H₂PtCl₆ was dissolved in DI-water by 1:1 and used to electroplate the nanoparticles by applying a DC voltage of −0.06 V for 8 s. Here, we characterized the double layer capacitance by cyclic voltammetry and calculated the delivered charges to be 0.5 μC. It also shows that the ion-to-electron transducers (Pd/PdHx, Ag/AgCl) are much more efficient in ion delivery.

10 mM AgNO₃ in 0.1 M KNO₃ was used to deposit Ag nanoparticles followed by chlorinating Ag nanoparticles into AgCl nanoparticles. For microelectrodes, −0.2 V was applied for 5 s. During this process, 3.1 μC charges are measured in the circuit, indicating that a single microelectrode with the coating of Ag/AlCl NPs theoretically has the capability of converting 3.1 μC electrons to Cl⁻ and vice versa.

For a reference electrode and auxiliary electrode, −0.3 V was applied for 50 s. Then, we oxidize approximately Ag NPs to AgCl NPs by applying a constant anodic current (100 μA) for 10 s on Ag NPs in 50 mM KCl solution at room temperature. We observe a clear color change from silver white to dark gray, indicating the formation of AgCl NPs.

The modular multi-channel potentiostat is a unique piece of equipment that can operate multiple electrochemical devices or a single electrochemical device with more than one working electrode. The modular aspect allows the multi-channel potentiostat to scale from 8 to 64 channels by adding more stackable boards. The stackable board provides eight channels of circuits with the output range of ±4 V and the input range of ±1.65 μA. In addition, the modular multi-channel potentiostat also offers an external control mode where it allows for interfacing with external software, such as a machine-learning control algorithm. More detailed information of the multi-channel potentiostat can be found here.³⁷ 

We used microscope-based real-time imaging over the microelectrodes to monitor the ion concentration change. We used 50 μM 5-(and-6)-Carboxy SNARF-1 (SNARF, ThermoFisher) dispensed in the 0.1M Tris buffer as a fluorescent indicator for H⁺, 50 μM CoroNa™ Green (CoroNa, ThermoFisher) dispensed in the 0.1 M Tris buffer for Na⁺, and 100 μM [N-(ethoxycarbonylmethyl)-6-methoxyquinolinium bromide] (MQAE, ThermoFisher) dispensed in the 0.1M Tris buffer for Cl⁻. The fluorescent probe solution was flowed into the target chamber via a sealed microfluidic channel. Thereafter, the device was monitored by using a BZ-X710 fluorescence microscope with 10× Nikon objective. Different filters are selected by the ion we are interested to image: TxRed (ex: 560/40 nm, em: 630/75 nm) for SNARF, GFP for CoroNa (ex: 502/30 nm, em: 520/36 nm, and DAPI (ex: 377/50 nm, em: 447/60 nm) for MQAE. Imaging data were collected every 2 s in real time. Data were analyzed using ImageJ software.

A machine-learning-based controller consists of four sub-controllers, one for each ion, performing in real-time and updating their parameters online. The overall schematic of the ML-based closed-loop control of the multi-ion pump is demonstrated in Fig. S4 of the supplementary material. The ML-controller utilizes one of its four sub-controllers, one at a time. Each controller is based on the real-time adaptive machine-learning-based control methodology developed by the authors.³⁴ More detailed information on the ML-based control algorithm could be found here.³⁴ 

See the supplementary material for Figs. S1–S4 and the supplementary text.

The authors would like to thank Dr. John Selberg for fruitful discussions regarding fabrication. This research is sponsored by the Defense Advanced Research Projects Agency (DARPA) through Cooperative Agreement D20AC00003 awarded by the U.S. Department of the Interior (DOI), Interior Business Center. The content of the information does not necessarily reflect the position or the policy of the Government, and no official endorsement should be inferred. Microfabrication was performed using equipment sponsored by the W. M. Keck Center for Nanoscale Optofluidics, the California Institute for Quantitative Biosciences (QB3), and the Army Research Office award under Grant No. W911NF-17-1-0460.

The authors have no conflicts to disclose.

Ethics approval is not required.

The data that support the findings of this study are available from the corresponding author upon reasonable request.