Intracellular recordings of action potentials by an extracellular nanoscale field-effect transistor.

The ability to make electrical measurements inside cells has led to many important advances in electrophysiology. The patch clamp technique, in which a glass micropipette filled with electrolyte is inserted into a cell, offers both high signal-to-noise ratio and temporal resolution. Ideally, the micropipette should be as small as possible to increase the spatial resolution and reduce the invasiveness of the measurement, but the overall performance of the technique depends on the impedance of the interface between the micropipette and the cell interior, which limits how small the micropipette can be. Techniques that involve inserting metal or carbon microelectrodes into cells are subject to similar constraints. Field-effect transistors (FETs) can also record electric potentials inside cells, and because their performance does not depend on impedance, they can be made much smaller than micropipettes and microelectrodes. Moreover, FET arrays are better suited for multiplexed measurements. Previously, we have demonstrated FET-based intracellular recording with kinked nanowire structures, but the kink configuration and device design places limits on the probe size and the potential for multiplexing. Here, we report a new approach in which a SiO2 nanotube is synthetically integrated on top of a nanoscale FET. This nanotube penetrates the cell membrane, bringing the cell cytosol into contact with the FET, which is then able to record the intracellular transmembrane potential. Simulations show that the bandwidth of this branched intracellular nanotube FET (BIT-FET) is high enough for it to record fast action potentials even when the nanotube diameter is decreased to 3 nm, a length scale well below that accessible with other methods. Studies of cardiomyocyte cells demonstrate that when phospholipid-modified BIT-FETs are brought close to cells, the nanotubes can spontaneously penetrate the cell membrane to allow the full-amplitude intracellular action potential to be recorded, thus showing that a stable and tight seal forms between the nanotube and cell membrane. We also show that multiple BIT-FETs can record multiplexed intracellular signals from both single cells and networks of cells.

Our BIT-FET (Fig. 1a) is a combination of a Si nanowire (SiNW) FET detector and an electrically-insulating SiO2 nanotube that connects the FET to the intracellular fluid (the cytosol). When there is a change in transmembrane potential, V, such as during an action potential, the varying potential of the cytosol inside the nanotube yields a change in the SiNW FET conductance, (G), in a manner equivalent to a time-varying potential applied to a gate electrode in a traditional FET. The polarity of the change in G will be inverted with respect to the varying V for a p-type FET under constant source/drain (S/D) bias as used in our studies (right panel, Fig. 1a). This BIT-FET design uses the tip of a size-controllable nanotube to interface with cells, which represents the smallest ultimate probe size (enabled by the use of the FET sensor). Because the nanotube is built on top of the planar FET, BIT-FET arrays can fully exploit the high density of planar nanoFETs in contrast to previous work[10].

Figure 1

Schematics and structural characterization of the BIT-FET

a, schematics showing (left) a cell coupled to a BIT-FET and (right) the variation in device conductance, (G), during an action potential, where S, D, V, and t correspond to source electrode, drain electrode, membrane potential, and time, respectively. The SiO2 nanotube connects the cytosol (orange) to the p-type SiNW FET, and together with the SiO2 passivation excludes the extracellular medium (light blue) from the active device channel. The schematic structures on the membrane represent different ion channels, and are not scaled to the true size of the BIT-FET. b, SEM image of a GeNW branch on a SiNW with an orientation near the surface normal. Inset, SEM image of an Au nanodot on a SiNW prior to GeNW growth. c, SEM image of a GeNW/SiNW heterostructure coated with ALD SiO2. Magnified images of the top and bottom are shown in Fig. S2. d, SEM image of a final nanotube on a SiNW. Insets, magnified images of the top and bottom of the nanotube. Scale bars, 100 nm in inset of b, 200 nm in all others.

The BIT-FET devices are prepared in a sequence of growth and fabrication steps that enable control of key individual device parameters and the density of multiple devices as shown schematically in Figure S1. Ge nanowire (GeNW) branches grown on top of SiNWs from Au-nanocluster catalyzed vapor-liquid-solid mechanism[13] were used as sacrificial templates for the nanotubes (see Supplementary Methods). Representative scanning electron microscopy (SEM) images (Fig. 1b) show both an Au nanodot (inset) and the resulting GeNW branch ‘standing-up’ on the SiNW and oriented nearly normal to the substrate surface. After defining the S/D contacts on each side of selected GeNWs (Fig. S1d), a conformal, controlled-thickness SiO2 layer deposited by atomic layer deposition (ALD)[14] provides the nanotube wall and S/D passivation (see Supplementary Methods). Representative SEM images of the resulting structure show clearly this conformal SiO2 shell and the GeNW core (Figs. 1c and S2).

Finally, the BIT-FET devices fabrication is completed by two etching steps (Figs. S1f-i). The first one is to selectively remove the topmost part SiO2 shell by buffered hydrofluoric acid (BHF) to expose the GeNW core, and the second one is to etch away the GeNW and then leave a hollow SiO2 nanotube on SiNW. SEM images of the resulting branched SiO2 nanotube on the SiNW structure (Fig. 1d) confirm the open end of the nanotube. Comparison of images before and after GeNW etching (Fig. 1c vs. 1d; Fig. S2 vs. insets of Fig. 1d) further shows that the nanotube structure is open to the SiNW surface as evidenced by bright to dark image contrast change associated with removal of the Ge. In addition, these images demonstrate that the SiO2 shell is tapered at the tip: for 50 nm GeNW and 50 nm ALD SiO2 used here, the very tip of the nanotube is about 55 nm in outer diameter and increases to a maximum of 150 nm about 2.2 μm away from the tip. This tapering effect results from the isotropic etching of BHF (see Supplementary Methods) and we believe that it is especially attractive for decreasing the size of the probes.

We have characterized the electrical properties of the BIT-FETs and several control devices in solution to elucidate the behavior of this new device architecture. A SEM image (Fig. 2a) shows a representative two-FET structure, where a BIT-FET and conventional FET with similar channel length were fabricated with a common S electrode on the same SiNW. In both devices, the SiNW and electrodes exposed to solution are passivated with about 50 nm ALD SiO2 as described above. Measurements of conductance (G) for both devices as a function of water-gate voltage (V) prior to etching the GeNW core of the BIT-FET (Fig. 2b) show very little change, with sensitivity of ca. -170 nS/V. Significantly, measurements made on the same devices after removal of the GeNW core to yield an open nanotube structure (Fig. 2c), demonstrate a large increase in the sensitivity of the BIT-FET to -4530 nS/V, while the control SiNW FET shows no change. Taken together these results validate that BIT-FET devices respond selectively and with high-sensitivity to the solution inside vs. outside the nanotubes, and thus meet the requirements for intracellular recording outlined schematically in Fig. 1a. The difference in sensitivity of the BIT-FET devices to solution inside vs. outside the nanotubes originates primarily from the gate capacitance difference[11,12]. Specifically, Ge over-coating on the SiNW may lead to a larger contact area between the SiNW and the internal solution of the nanotube (the active FET area) than defined by the nanotube inner diameter, which can increase this sensitivity difference (see Supplementary Methods).

Figure 2

Water gate characterization and bandwidth analysis

a, SEM image of a BIT-FET device (S-D1) and control device (S-D2). b and c, Water gate, V, responses prior to and after GeNW etching, respectively. Blue, (S-D1); red, (S-D2). d, pulsed V with 0.1 ms rise/fall time, 1 ms duration and 100 mV amplitude (upper), and the corresponding conductance change of a BIT-FET device (black, lower). The red trace is the pure field-effect response after removing the capacitive signals of the passivated metal electrodes (see Supplementary Methods). e, baseline to plateau conductance change of the same BIT-FET device as in d versus pulse rise/fall time. The change was measured as an average over data 0.2-0.5 ms after the start of the pulse. Pulse amplitude was kept at 100 mV, and duration was ten times the rise/fall time in all measurements. f, Calculated bandwidth of the BIT-FET device versus nanotube inner diameter (ALD SiO2 thickness was the same as the nanotube inner diameter, and the nanotube length was fixed at 1.5 μm). The black and red symbols correspond to upper and lower limits, respectively (see Supplementary Methods). Inset, calculated change of the potential at the SiNW FET surface V (normalized with the step change V of potential at the nanotube opening) versus time.

We have also characterized the temporal response of BIT-FET devices to assess their capability for recording fast cellular processes. A pulsed V with 0.1 ms rise/fall time, 1 ms duration and 100 mV amplitude was applied to approximate an action potential. The conductance exhibits a peak (dip) coincident with the 0.1 ms rise (fall) of the pulse, and a plateau step down during the constant 100 mV portion of the pulse (Fig. 2d). For pulsed V measurement with different rise/fall times from 0.1 to 50 ms, the conductance change associated with the baseline to plateau is independent of the pulse rise time (Fig. 2e), and moreover, this change is consistent with the device sensitivity determined from quasi-static measurements (e.g., Fig. 2c).

The peak and dip features in the pulsed V results correspond to the expected capacitive charging[15] of the passivated metal electrodes and are not intrinsic to the BIT-FET. Specifically, a control pulsed V measurement made on a SiNW FET without a nanotube branch showed the same peak (dip) features associated with the rapid rise (fall) of the V pulse (Fig. S4a). These capacitive features can be readily removed from the BIT-FET and control devices data to yield the pure FET response (red curves, Fig. 2d and S4a, see Supplementary Methods), and in the case of the BIT-FET, it demonstrates clearly that the conductance change follows the 0.1 ms V pulse rise/fall without detectable delay. The results shown in Fig. 2d and e demonstrate that the BIT-FET can faithfully record potential changes with at least a 0.1 ms time resolution. Indeed, our modeling (below) shows the temporal resolution, which is beyond our measurement capabilities, should be much better than this value. We also note these capacitive features are not expected in cellular measurements because (i) metal electrodes are only coupled to extracellular media, where the potential changes are quite small[16] and (ii) these changes will be localized on the size of a cell, which is much smaller than the electrode area exposed to solution (~cm2) in the pulsed V experiments here.

In addition, we have modeled the BIT-FET device to estimate the bandwidth, which is beyond our current measurement limit, and also investigated the bandwidth dependence on nanotube diameter. The signal transduction in the BIT-FET device can be readily solved by the classical transmission line model[15]. In our analysis (Fig. S4b, see Supplementary Methods), we determine the change of the potential at the SiNW FET surface, V, as a function of time following a step change of transmembrane potential at the nanotube opening to V. For a typical nanotube (inner diameter = 50 nm; ALD SiO2 thickness = 50 nm; length = 1.5 μm), the calculated response (inset, Fig. 2f) yields a bandwidth of ca. 1.2 MHz. This represents an upper limit assuming the active FET area and relevant device capacitance C (Fig. S4b) is defined only by the nanotube inner diameter, and could be reduced to 0.2 MHz if we assume the entire SiNW surface is active due to Ge over-coating (see Supplementary Methods). A summary of results (Fig. 2f) shows that the BIT-FET can achieve a bandwidth ≥ 6 kHz, which is sufficient for recording a rapid neuronal action potential[1,2], for nanotube inner diameters as small as 3 nm (fixed length = 1.5 μm). The high bandwidth determined for the BIT-FET devices results in large part from the small device capacitance, despite the increasingly large solution resistance within the nanotube with decreasing inner diameter (see Supplementary Methods). The small diameters accessible with the BIT-FET suggest that it could be minimally-invasive and capable of probing the smallest cellular structures, including neuron dendrites and dendritic spines, which are difficult using conventional electrical-based techniques[17,18].

We investigated the capability of the BIT-FET to record intracellular signals using spontaneously beating embryonic chicken cardiomyocyte cells, which were cultured on thin pieces of polydimethylsiloxane (PDMS) as described previously[16]. After modifying the devices with phospholipids[10] to facilitate the internalization of nanotubes into cells, the PDMS/cell sheet was manipulated to put a cell into gentle contact with the nanotube of a BIT-FET under standard electrophysiology microscope (see Supplementary Methods). Approximately 45s after gentle contact was made and in the absence of applied force to the cell substrate, the recorded data showed a dramatic change (Fig. 3a). Before the transition, the signal exhibits a relatively flat baseline with small biphasic peaks (5~8 mV amplitude; ~1 ms duration) with ca. 1 Hz frequency (e.g., Figs. 3b, c). These peaks are coincident with cell beating and consistent with extracellular recording reported previously[16]. Then the baseline shifts ca. -35 mV and new peaks with 75-100 mV amplitude and ~200 ms duration are observed (Fig. 3a). The recorded conductance data yields inverted peaks for the p-type SiNW FETs used here, although the calibrated potentials are consistent with standard peak polarity and shape of intracellular action potentials. These peaks (e.g., Fig. 3d) have the shape and features characteristic of an intracellular action potential of cardiomyocyte cells[10,19,20], including fast depolarization at the beginning of the peak, plateau region, fast repolarization, and hyperpolarization and return to baseline. The signal transition from extra- to intracellular indicated the penetration of the cell by the nanotube. The baseline shift is similar with that measured recently using kinked-nanowire probes[10], but smaller than the standard resting potential for cardiomyocytes[19,20]. Our reproducible and stable recording of full-amplitude action potentials, which is a central result of our work, suggests that this baseline difference is not due to poor sealing during nanotube internalization. We propose that the discrepancy in resting potentials here could be attributed to a stronger suspension effect introduced by the intracellular polyelectrolytes at the junction[21,22] due to an order of magnitude smaller size of SiO2 nanotube opening than a typical patch clamp pipette, although more detailed studies will be required to quantitatively understand the origin of this effect. Although the nanotube diameter routinely used in our intracellular recording studies, 50 nm inner diameter and 55 nm tip outer diameter, is larger than the smallest achievable for BIT-FETs (Fig. 2f), it is still much smaller than the size of typical glass micropipettes[1,2] and metal microelectrodes[3,4,7] used for intracellular studies.

Figure 3

Intracellular action potential recording

a, a representative trace reflects the transition from extra- to intra- cellular recording. b, magnified trace of the part in the black dashed rectangle in a. c, magnified trace of the peak in the blue dashed rectangle in b. The stars in b and c mark the position of extracellular spikes. d, magnified trace of the peak in red dashed rectangle in a. e, the trace corresponding to the second entry of the nanotube around the same position on the cell. The potential was calibrated using the sensitivity values measured on phospholipid-modified devices by quasi-static V measurement (e.g. blue trace in Fig. 2) and pulsed V measurement with 0.1 ms pulse rise/fall time (same for Fig. 4). The sensitivity obtained from these two measurements is same as discussed before.

The change from extracellular to intracellular signal without external force applied to cell suggests the spontaneous penetration of cell membrane by the nanotube versus mechanical insertion. We speculate that lipid fusion[23,24] may play an important role in this penetration similar to our previous observations[10], and also that the small nanotube size is likely beneficial for this lipid fusion process and the formation of a tight seal. There are several attractive consequences of the spontaneous penetration. First, this typically leads to full-amplitude action potential recording (e.g., Fig. 3) without circuitry to compensate for probe-membrane leakage, thus suggesting tight sealing between the nanotube and cell membrane. Indeed, control experiments carried out without phospholipids modification of the BIT-FETs required external forces to achieve the transition to intracellular action potential signals, and the smaller amplitude of these signals, 10-30 mV, suggests leakage at the nanotube-membrane interface[3]. Second, we find that spontaneous penetration occurs in the same way for a broad range of nanotube orientation (i.e., within 30° of the surface normal), which contrasts mechanical insertion. Third, we believe that the tight nanotube-membrane seal and the very small nanotube internal volume, ca. 3 aL, help to preserve cell viability and a stable signal over time. In general, we find that termination of signal recording by the BIT-FET is due to random separation of the nanotube from the motion of the beating cardiomyocyte cell and not cell death or degradation of the nanotube/cell membrane interface, where the latter normally occur during recording with glass micropipettes[1,2]. In addition and unlike a glass micropipette, when the BIT-FET nanotube is separated from a cell (on purpose as shown below or by the beating motion), the nanotube can re-penetrate into the same cell multiple times at approximately the same position without affecting the cell or recorded signal (see below). Last, we note that the total recording time from multiple penetrations at a given position on a cell with the BIT-FET can exceed an hour.

We also find that the BIT-FET devices are robust and reusable. Specifically, following retraction of the cell substrate from the device, which results in return of the conductance to the extracellular baseline, subsequent gentle contact of the nanotube to the same cell without changing position leads to the development of stable intracellular action potential signals again (Fig. 3e). We have repeated the gentle contact/intracellular recording/retraction cycle up to five times with the same BIT-FET nanotube near the same position on the cell without observable change in the beating frequency and action potential features. A SEM image of the BIT-FET device following these repeated cycles (Fig. S5) shows that the nanotube remains intact with some residue on the upper outer surface. In addition, we did not see evidence for the blockage of the nanotube during these cycles, which we attribute to spontaneous penetration versus suction or mechanical insertion. Furthermore, we note that devices can even be reused after being dried. Taken together these results demonstrate the reliability and robustness of the BIT-FETs and strongly indicate that this is a minimally-invasive intracellular recording technique.

A unique feature of the BIT-FET design is the straightforward fabrication of multiple, independent devices to enable multiplexed recording from single cells through cell networks. For example, we have readily aligned two phospholipid-modified BIT-FET devices separated by about 20 μm with a single, beating cardiomyocyte cell (Fig. 4a). Following gentle contact, conductance versus time measurements made simultaneously from both devices (Fig. 4b) show that device-1 first bridged the cell membrane to yield clear intracellular signals, and ~10s later we observed the development of intracellular peaks from device-2. Subsequently, intracellular signals were recorded from both devices (e.g. Fig. 4c). We can glean several important points from these data. First, the sequential nature by which the intracellular signals develop, in the absence of an applied force, strongly supports the suggestion above that penetration of the phospholipid-modified nanotubes is a spontaneous biomimetic process that does not adversely affect the cell rather than mechanical insertion. Second, the intracellular peaks recorded simultaneously by devices-1 and -2, with full amplitude of 75-100 mV, and stable cell beating over time are consistent with the tight seal being established between the cell membrane and the nanotubes in both devices. In addition, we have also demonstrated that multiplexed measurements with BIT-FETs can be extended to cell networks (Figs. 4d & e), where we record intracellular action potentials simultaneously from different sites in a monolayer of beating cardiomyocyte cells. In the future, we note that this BIT-FET design is implementable on high density integrated planar nanoFETs, either large arrays of nanowire FETs[25] or conventional top-down nanoFET arrays[26], to enable multiplexed recording at a far higher density than demonstrated in these initial studies.

Figure 4

Multiplexed intracellular recording

a, Differential interference contrast microscopy (DIC) image of two BIT-FET devices (position marked with dots) coupled to a single cardiomyocyte cell, with cell boundary marked by the yellow dashed line. b, The simultaneously recorded traces from the two devices in a, corresponding to the transition from extra- to intra- cellular signal. The transition happened in sequential manner. The break mark labels the ~1 s discontinuity between the two adjacent traces. c, representative trace of stable intracellular action potentials recorded ~120 s after the internalization of the devices in a. d, DIC image of three BIT-FET devices coupled to a layer of beating cardiomycyte cell network (from a different PDMS/cell sample than in a). e, representative traces recorded simultaneously from the devices shown in d. The three devices exhibit intracellular action potential signals from different cells in the cell network. We note that devices used in a and d have different sensitivities (and are also different from the one used in Fig. 3). These differences are primarily due to variations in Ge over-coating during GeNW growth (see Supplementary Information). The potential was calibrated using the sensitivity values measured for each individual device, and all devices yield corresponding intracellular action potential values with full amplitude of 75-100 mV (independent of this conductance/sensitivity variation).

Additional work remains to improve further the BIT-FET based intracellular measurement technique. The signal-to-noise ratio is still lower than that of glass micropipette. And implementing the capability for cell stimulation in addition to recording is also important for intracellular study. However, we believe the advantages of the BIT-FET demonstrated in this work, including the capability to realize sub-5 nm size probes, the formation of tight nanotube-cell membrane seals, and the potential for large-scale high-density multiplexed recording, make it an attractive new measurement tool to extend substantially the scope of fundamental and applied electrophysiology studies to regimes hard to access by current methods.

Methods Summary

SiNWs were synthesized using Au-nanocluster catalysed VLS growth described previously[16]. After dispersing SiNWs on Si3N4 surface of silicon wafers, Au nanodots were defined on the top surfaces of the SiNWs by electron-beam lithography (EBL) and metal evaporation, and GeNW branches were grown from these nanodots by another Au-catalysed VLS step. Source and drain metal contacts were defined by EBL and metal evaporation on each side of selected GeNW branches on the corresponding SiNW backbones, and then a conformal and uniform SiO2 layer was then deposited on the entire chip by ALD. Photoresist was spin-coated with thickness smaller than the selected GeNW branches height, and then BHF was used to remove the SiO2 from the exposed tips of the GeNW/SiO2 core/shell structure. Following photoresist lift-off, H2O2 was used to etch the GeNW cores and yield the final BIT-FET devices. Conductance versus V measurements were carried out in 1×PBS buffer using an Ag/AgCl electrode. Electrical recordings from embryonic chicken cardiomyocytes were carried out using methods published previously[10,16], with cells cultured on thin PDMS films and device chips modified with lipid layers. A glass micropipette was used to control the relative position between the cell and the nanotube(s) of BIT-FET device(s), and Ag/AgCl reference electrodes were used to fix the extracellular solution potential. The BIT-FET bandwidth as a function of nanotube ID was determined from simulations of the time-dependent change in potential at the SiNW FET surface, V, following a potential step change at the open nanotube end.