High-resolution imaging of cellular dopamine efflux using a fluorescent nanosensor array

Significance

Cells communicate between themselves using waves of chemical concentration that change in both direction and time. Existing methods are not able to capture this spatial and temporal dependence at the necessary resolution to study the influence of cellular morphology. This study utilizes arrays of fluorescent nanosensors based on single-walled carbon nanotubes placed under and around neuroprogenitor cells to label-free image the efflux of the neurotransmitter dopamine. We use the spatiotemporal resolution of this approach to resolve where on the cell surface dopamine is released and how cell morphology affects the location of release sites.

Abstract

Intercellular communication via chemical signaling proceeds with both spatial and temporal components, but analytical tools, such as microfabricated electrodes, have been limited to just a few probes per cell. In this work, we use a nonphotobleaching fluorescent nanosensor array based on single-walled carbon nanotubes (SWCNTs) rendered selective to dopamine to study its release from PC12 neuroprogenitor cells at a resolution exceeding 20,000 sensors per cell. This allows the spatial and temporal dynamics of dopamine release, following K⁺ stimulation, to be measured at exceedingly high resolution. We observe localized, unlabeled release sites of dopamine spanning 100 ms to seconds that correlate with protrusions but not predominately the positive curvature associated with the tips of cellular protrusions as intuitively expected. The results illustrate how directionality of chemical signaling is shaped by membrane morphology, and highlight the advantages of nanosensor arrays that can provide high spatial and temporal resolution of chemical signaling.

Chemical signaling between cells is a hallmark of life, with one key example being neurotransmitter release from neurons to transmit and process information (1). However, unlike electrical potential, the measurement of chemical signaling between cells and within cellular networks is not as well developed, despite its importance to cellular biology. Neurotransmission via neurotransmitters, paracrine signaling between immune cells, or quorum sensing between bacteria in biofilms (2, 3) rely on spatial and temporal precision in chemical signaling. Analytical tools are not yet available to resolve these dynamics. Recently, the groups of Lindau and Ewing used 4–16 microelectrodes per array per chromaffin cell to estimate catecholamine release direction and dynamics (4–6). Nanosensors have several advantages for the study of cell signaling, as we have highlighted (7, 8). Their nanoscale size means that they can form dense arrays, increasing the number of probes per cell. In this work, we construct a fluorescent nanosensor array placed in the vicinity of PC12 neuroprogenitor cells at a density of ∼20,000 sensors per cell, with better than 100-ms temporal resolution to study dopamine release with much higher resolution.

The platform uses near-infrared (nIR) fluorescent single-walled carbon nanotubes (SWCNTs) rendered chemically selective to dopamine. nIR fluorescent single SWCNTs are versatile building blocks for biosensors and demonstrate detection limits down to single molecules (9–13). We selected the neurotransmitter dopamine for this study because of its central role for reward control and learning and because it is one of the neurotransmitters for which volume transmission is heavily discussed (14, 15). The spatially dependent concentration profile of dopamine plays an especially important role when neurotransmitters spill over from synapses and freely diffuse in all directions (volume transmission) through neural tissue (14, 16, 17). By placing such dopamine-sensitive nanosensors in arrays beneath dopaminergic cells we are able to image dopamine efflux, identify hotspots on the cell surface, and map the directionality of dopamine efflux (Fig. 1A). This platform has advantages over protein-based fluorescent sensors and semisynthetic protein–fluorophore conjugates on the cell surface in that it has higher sensitivity, time resolution, and lacks photobleaching (18–21). Electrochemical techniques, although the most advanced for measurement of redox-active neurotransmitters in vitro and in vivo, provide poor spatial resolution (4, 6) to date. Indirect (optical) methods can label cellular components that are involved in exocytosis or load vesicles with fluorescent dyes but cannot directly detect the molecules of interest or their dynamics in the extracellular space (22–24).

Fig. 1.

Chemical imaging using nanosensor arrays. (A) Fluorescent carbon nanotubes are rendered sensitive to dopamine by noncovalently attaching specific DNA sequences to them. They are immobilized onto a glass substrate, and dopamine-releasing PC12 cells are cultivated on top. When cells are stimulated they release dopamine and the SWCNT fluorescence changes. (B) Fluorescence intensity trace of a single (GA)₁₅-ssDNA/SWCNT imaged on a surface while adding dopamine (10 µM). (C) Proposed sensing mechanism: Dopamine pulls phosphate groups to the SWCNT surface, which removes quenching sites and increases SWCNT fluorescence quantum yield (MD simulations). (D) PC12 cells are cultivated on top of a substrate coated with nanosensors. In an image of this substrate every pixel corresponds to a region containing one or more nanosensors and serves as a reporter of local dopamine concentration. Each pixel of an nIR-movie produces a trace that contains information about the dopamine signal. A fitting procedure of every trace yields amplitude, width, and time point of the signal. Those values can be represented in false-color images, and their spatial composition can be further analyzed. (E) Bright-field image of a PC12 cell adhering on top of a nanosensor array and corresponding image of signal amplitudes. (F) Examples for pixel traces from regions under and around the cell shown in E.

Results and Discussion

Previously we found that the corona phase around SWCNTs can be engineered to recognize certain small analytes––a phenomenon we termed Corona Phase Molecular Recognition (CoPhMoRe) (7, 25, 26). Specifically, DNA-wrapped SWCNTs were found to increase their nIR fluorescence in the presence of catecholamines (13). Here, we synthesized and characterized different DNA/SWCNT complexes and identified the best candidates for dopamine detection (SI Appendix, Fig. S1). We measured the fluorescence responses of a panel of 10 DNA-wrapped (6, 5)-chirality enriched SWCNTs versus a panel of 15 molecules (SI Appendix, Figs. S2–S4) including dopamine, dopamine homologs, and potential interfering compounds present in biological environments (SI Appendix, Fig. S2). Our results indicate that (GA)₁₅-ssDNA/SWCNT complexes provide the most promising selectivity and sensitivity profile for cell experiments (Fig. 1B and SI Appendix, Figs. S4–S6). These dopamine nanosensors were excited with a laser at 560 nm in a standard fluorescence microscope equipped with an nIR-camera that detects the nIR emission of (6, 5)-SWCNTs at ∼980 nm. We performed molecular dynamics (MD) simulations to gain a deeper understanding of molecular recognition of these sensors. Our results show that dopamine interacts via its two hydroxy groups with phosphate groups of the DNA backbone (Fig. 1C). This interaction pulls the phosphate group closer to the surface of the SWCNTs and changes the local potential (SI Appendix, Figs. S21–S26). These findings rationalize previous phenomenological theories of molecular recognition in the corona phase of SWCNTs (27, 28). Furthermore, we found that naked SWCNTs do not respond to dopamine at all (SI Appendix, Fig. S8 and S10) and the least responsive DNA sequences showed the highest (starting) fluorescence counts (SI Appendix, Figs. S9–S11). This result suggests that movement of the phosphate groups removes (pre)-quenching sites for excitons and that dopamine’s ability to pull these groups to the SWCNT surface results in an increase of SWCNT fluorescence.

With respect to the envisioned application (Fig. 1A) we prepared homogeneously coated nanosensor surfaces and treated every pixel of the nIR image as a local sensor. Whereas nanosensors are randomly adsorbed on the surface, it would be possible to pattern them more regularly as it was shown for other nanoparticles (29–31). The SWCNT sensors were densely coated on the surface to maximize the signal/noise ratio (SI Appendix, Figs. S12–S14). Ensemble calibration curves (SI Appendix, Fig. S5) indicate that we can detect dopamine concentrations down to 100 nM (SI Appendix, Fig. S5). However, this is the mean response of a large population of sensors. When we analyzed individual nanosensors in the single-molecule regime (100 pM) we observed stochasticity indicative of binding and unbinding events of single dopamine molecules (SI Appendix, Fig. S7). To ensure that our sensors were capable of detecting dopamine in the presence of cells, we cultivated dopamine-releasing pheochromocytoma cells (PC12) on top of nanosensor arrays that were additionally coated with collagen to increase cell adhesion (SI Appendix, Figs. S13–S16). Collagen has been known to increase PC12 cell adhesion and differentiation, whereas nanotubes affect neurite outgrowth (32–35). Furthermore, the elasticity of the substrate affects PC12 cell morphology (36). PC12 cells on our sensor/collagen-coated surfaces attained a morphology similar to PC12 cells cultured without sensors. Similar scenarios in the literature are nanotube networks that were glycosylated to enhance biocompatibility and have been used for amperometric studies of dopamine release (37). This procedure increased adhesion and viability of PC12 cells on SWCNT networks. We focused our investigation mostly on isolated PC12 cells and not clusters to minimize cross-talk from different cells.

Labeling with the fluorescent false neurotransmitter FFN551 showed that PC12 cells on these surfaces contain vesicles with dopamine transporters, indicating dopamine-containing vesicles (SI Appendix, Fig. S16) (24, 38). After 1 d of cultivation at 37 °C in full culture medium, the medium was exchanged to PBS buffer and the response to dopamine and homologs (10 µM) was tested. The results (SI Appendix, Fig. S6) show that the nanosensor arrays are still functional and resist biofouling. Catecholamine homologs such as epinephrine, norepinephrine, and ascorbic acid could interfere with this array (SI Appendix, Fig. S6). Therefore, the sensor array reports about the total changes in the concentration of all catecholamines (ascorbic acid concentrations should be negligible/constant). However, PC12 cells are thought to mainly release dopamine even though pharmaceuticals and hypoxia can affect the ratio between the different catecholamine neurotransmitters (32, 39, 40). Consequently, sensor responses are expected to be attributed mainly to dopamine and to a lesser extent other catecholamines.

The results from Fig. 1 indicate that we can use surface-immobilized nanosensors to optically detect dopamine in the presence of PC12 cells. Consequently, we recorded nIR movies (100 ms per image, 640 × 512-pixel resolution, 212 nm/pixel) of PC12 cells adhered on collagen-coated nanosensor arrays and stimulated the cells with potassium buffer to observe dopamine release. To increase the signal to noise ratio, we divided the field of view into squares of 4 × 4 pixels and developed a fitting algorithm for the normalized intensity traces of the pixel groups. This algorithm fits the curve $f\left(t\right)=a+bt+c{t}^2+x{e}^{-\left({(t-y)}^2/2z^2\right)}$ to the intensity traces (Gaussian function plus background correction; Fig. 1 D–F).

Consequently we get 20,480 “reporter” pixels per fitted image (pixel size ∼850 × 850 nm). Therefore, we expect ∼1,700 reporter pixels under a (round) cell (d = 40 µm), >180 reporter pixels in a 2-µm zone around the cell, and ∼150 reporter pixels under the cell contour (pixel size = 850 × 850 nm). The fitting process reduced the 3D movie to a 2D image of fitting parameters, which can then be further analyzed (Fig. 1D). The different fitting parameters represent how the local reporter pixel responds to dopamine and the parameters account for amplitude (x), width (y), and time point (z) of the event. Fig. 1F shows typical traces from different locations under the cell, from the cell border, and other regions [1–5, shown in the bright-field (BF) image of Fig. 1E].

We performed independent analyses to verify that (i) temporal and (ii) spatial patterns are related to the stimulation time point and the location of cells. To verify the temporal correlation in raw data traces (Fig. 2A), we defined three periods (before, during, and after stimulation) and searched for maxima (in each period) that would indicate a dopamine-induced fluorescence increase. The exact stimulation time point (±1 s) was known from the time point when K⁺ was added on top of the cells. The nonrandom distribution during the stimulation period proves that the observed sensor responses are related to stimulation (Fig. 2 A and B).

Fig. 2.

Spatial and temporal correlation between sensor response and cells. (A) Temporal correlation between stimulation event and sensor responses (single experiment). (B) Large amplitudes are more likely in time intervals with cell stimulation. Two-sided t test of independent experiments (n = 4). (C) Spatial correlation between sensor responses (normalized amplitude change) and parts of the cell. The black/white image on the left indicates the regions/pixels (in white) that were analyzed. The histograms show that the sensor responses are related to the cell and are not randomly distributed.

Histograms of the normalized sensor signals from different regions of interest related to the cell (shown in white in Fig. 2C, Left) prove that the signals are not randomly distributed throughout the image. Specifically, there is a clear spatial correlation between responses (normalized amplitude change) and regions associated with the cell (e.g., cell border, cell body). The time resolution of our approach (100 ms) is not yet fine enough to observe a propagating wave of dopamine released by the cell given that the diffusion length scale during image acquisition $x=\sqrt{2Dt}$ is ∼12 µm [D = 7.63 × 10⁻⁶cm²/s (14)], comparable to the cell diameter. However, the technique is able to localize release events and their duration as shown below.

Fig. 3A shows the fitted response of the sensors at different time points from the contour of a single cell (shown in Fig. 3B, Left). The 3D plots show the sensors turn on after cell stimulation, which indicates release of dopamine. Furthermore, the responses decay as dopamine unbinds and diffuses away. Fig. 3 B–E shows bright-field images of four different cells and corresponding 3D plots of the sensor response magnitudes along the cell contour (line width = 850 nm). Here, in contrast to the time-resolved Fig. 3A, the maximum responses of the whole experiment are shown (fit parameter x). The height and color of the 3D plots indicate the magnitude of sensor responses indicative of higher transient dopamine concentrations (Fig. 1D). For clarity we only show responses along this contour (full response plots are shown in SI Appendix, Fig. S17 and a time-resolved full hotspot analysis in SI Appendix, Fig. S18).

Fig. 3.

Dopamine release profiles along the cell border. (A) Three-dimensional plots of fitted sensor responses along the cell border at different time points relative to the stimulation at t₀ (the corresponding cell is shown in B). Height and color indicate the relative fluorescence change normalized to the maximum fluorescence change of the experiment. Other areas are shown in blue for clarity. (B–E, Left) Bright-field image of PC12 cells adhering on top of collagen-coated dopamine nanosensor arrays (contour shown in light blue). (Right) Three-dimensional plot of maximum sensor responses (of all frames) along cell contour. Height and color indicate magnitude of the dopamine response, which correlates with dopamine concentration. Other areas are shown in blue for clarity.

The contour is also the most interesting region because efflux parallel to the nanosensor array is centered there with dopamine concentrations highest at exocytosis sites. The response profiles show heterogeneity along the contour, which is expected from multiple dopamine release sites. Furthermore the shape of the cell affects the distribution of release sites. Fig. 3 demonstrates the superior spatial resolution of our sensing strategy compared with existing technology. For example, the groups of Lindau and Ewing used microelectrode arrays with up to 4–16 electrodes (per cell) to “image” catecholamine release from chromaffine cells (4–6). In contrast, our approach contains 349, 242, 192, and 206 sensor pixels just along the contour of the four cells shown in Fig. 3 B–E and even more (>20,000) in the vicinity of the cell. Therefore, with nanosensor arrays there are around two orders of magnitude more sensors positioned on the contour of a single cell and more than three orders of magnitude more sensors in the area close to the cell. Compared with amperometric methods, nanosensor arrays currently exhibit reduced temporal resolution but amperometric methods cannot distinguish molecules of similar redox potential. Therefore, due to the higher chemical selectivity it is more appropriate to compare nanosensor arrays with cyclic voltammetry methods, which also have a time resolution of ∼100 ms (20). An advantage of amperometric methods is their ability to quantify (count) the number of molecules that encounter the electrode. Fluorescent sensors are more complex in terms of their signal transduction mechanism but are also capable of detecting single molecules (SI Appendix, Fig. S7) and could in principle be calibrated.

In a next step we further analyzed data such as those shown in Fig. 3 to answer longstanding questions about dopamine efflux from PC12 cells. If we focus on responses >3× of the mean response (along the cell contour) we can pinpoint to locations that we refer to as hotspots (>3× mean response) and are distinct from locations close to the noise level. Fig. 4 A–C shows line profiles of the sensor response along the cell contour. Hotspots are maxima in these line plots. They are also shown in the cell contour in Fig. 4 A–C as blue dashed circles. Arrows in both plots indicate locations on the cell surface and the line plots show the responses in a clockwise fashion. It is striking that not all cells show hotspots, which means that not every stimulation finally leads to dopamine release. This behavior is known for PC12 cells and ∼50% of stimulation events do not lead to dopamine release as measured with electrochemical methods (33).

Fig. 4.

Hotspots and directionality of dopamine release. (A–C) Bright-field (BF) images of PC12 cells cultivated and stimulated atop nanosensor arrays. Responses were analyzed along the contour of the cell (width ∼850 nm) and show hotspots (defined as 3× the mean response). Blue circles indicate hotspots in the contour picture. Arrows point at specific locations of the cell and corresponding locations in the response profile along the cell contour. Directionality plots show angular distribution of hotspots. Length of arrows is proportional to the magnitude of the sensor response. (D) Anisotropy of dopamine release is governed by the distribution of protrusions/cell surface. The histogram shows the angular distribution of contour pixels (cell in A). Hotspots are shown in black and reveal that they are correlated with protrusions (shown in grey). (E) Hotspots are not randomly distributed but colocalize with protrusions (n = 8). Errors are SEMs. (F) Hotspot magnitudes (from n = 8 cells) versus local curvature. Hotspots are more often found in regions of negative curvature (64% vs. 36%) but due to cell morphology there is also more negative curvature per cell (see SI Appendix, Fig. S20 for normalized histograms).

Our results indicate 2–17 hotspots per cell (contour). This number is in agreement with electrochemical results from Zerby and Ewing that indicated ∼28 release events of PC12 cells per stimulation (41). Hotspots have been discussed in the past but the lack of spatial resolution of electrode-based methods made it impossible to directly pinpoint release sites on single cells (42). It is also known that neurons do not necessarily form well-defined synapses and therefore it is not known a priori where cells release signaling molecules (43).

Interestingly, hotspots are not accumulated at the tips of protrusions (see also Fig. 3). The local curvature of the cell outline was calculated and is color-coded in the cell contour plots in Fig. 4. Our data (Fig. 4F) indicate that more hotspots are found in regions of negative curvature (64%) compared with regions with positive curvature (36%).

However, 65% of the pixels along the cell contours (n = 8) have a curvature <0. The reason for this ratio is that cells with protrusions have many slightly concave segments and fewer strongly convex segments (tips). If the number of hotspots at a given curvature is normalized to the probability of this curvature (in all cells), there is no preference for either negative or positive curvature such as tips of protrusions (SI Appendix, Fig. S20). However, the nanosensor approach generates single-cell data about hotspot distribution (Figs. 3 and 4 A–C) and averaging might bury information.

Previously, it was shown that nondifferentiated PC12 cells release dopamine from the cell body, whereas differentiated PC12 cells release it mainly from structures called varicosities. However, our data indicate that they are distributed along cell protrusions with similar probabilities for regions of positive and negative curvature (41). Curvature and membrane tension are known to be a driving force of exocytosis and therefore spatially resolved images such as in Fig. 4 could provide new insights into this process (44).

Our analysis focused on the cell border because it is the most relevant region for studying signaling to other cells. Nevertheless, when all sensors under the cell are analyzed, hotspots under the cell appear (SI Appendix, Figs. S17 and S18). Total internal reflection fluorescence microscopy studies of labeled vesicles demonstrated exocytosis from the bottom of chromaffine cells (45). However, these methods naturally focus on the parts of the cell close to the substrate (i.e., cell bottom). Other studies with different techniques reported exocytosis from the whole cell surface (46, 47). Our data (Fig. 4 and SI Appendix, Fig. S17) show that there is release from regions close to the cell border but also from the cell bottom. It is possible that the functionalization of the substrate and cell adhesion affect the distribution of release sites. Furthermore, our sensors directly image the released molecule whereas these other optical methods image the membrane fusion process.

We further analyzed directionality of release. Fig. 4 A–C (Right) shows the signals from the cell perimeter region relative to center of mass of the cell and the arrow lengths indicate signal amplitudes. These plots reveal anisotropy of release—a measure not accessible with other methods. Anisotropy of cells translates into an anisotropy of dopamine release.

The most important cell features that govern anisotropy are protrusions. Hotspots are associated with protrusions (Fig. 4 D and E)––not necessarily at the tips but as well in regions where the protrusions begin (corresponding to negative curvature). To calculate a correlation, angular histograms of all pixels and hotspot pixels (e.g., Fig. 4D and SI Appendix, Fig. S19) were multiplied and integrated (n = 8). The results were compared with a completely random distribution of hotspots. Fig. 4E shows that hotspots are correlated with protrusions and are not randomly distributed. An anisotropic distribution of hotspots should cause anisotropic dopamine concentration profiles. Cells could therefore shape their signal by increasing or decreasing the number of protrusions into certain directions. Obviously primary neurons are anisotropic and the release profile is also predetermined by the geometry and location of synapses. But, PC12 cells do not form classical synapses and their morphology resembles tissue cells. Such cells could use this strategy to adjust and fine-tune their chemical signaling. The distribution of hotspots around cells is most likely dependent on cell type or may even change over time for a certain cell. Such differences can be precisely investigated using our nanosensor array approach.

In summary, we present a method for spatiotemporal chemical imaging of dopamine release from cells. The spatial resolution of nanosensor arrays allows us to map hotspots of dopamine release on cell surfaces and assess anisotropy of dopamine release. We found that dopamine efflux happens more frequently at protrusions but not preferentially in regions of positive curvature such as the tips. Furthermore, we show how cell protrusions and anisotropy are correlated with dopamine release and suggest that this is a possible mechanism for cells to shape a chemical signal.

Materials and Methods

Materials and Preparation of Polymer-Wrapped Carbon Nanotubes.

Chemicals were purchased from Sigma-Aldrich or IDT if not noted otherwise. To wrap SWCNTs, 2 mg of the 30-mer nucleic acid was dissolved in 1 mL of a 0.1 M NaCl solution containing 1 mg (6, 5)-enriched SWCNTs (SG65 from SouthWest Nano Technologies, distributed by Sigma-Aldrich). The solution was tip sonicated for 10 min (3-mm tip diameter, 40% amplitude, Cole Parmer) in an ice bath. Then samples were centrifuged twice for 90 min at 16,100 g and the supernatant was collected. All experiments were performed in PBS at pH 7.4. For solution-based experiments the stock solution was diluted to obtain a final absorption of 0.0075 at the (6, 5)-SWCNT S₂₂-peak maximum at ∼570 nm.

Spectroscopy of SWCNT/Polymer Complexes.

nIR-spectra were collected with a fluorescence spectrometer equipped with a 785-nm photodiode laser (450 mW), an Acton SP2500 spectrometer (Princeton Instruments), and a 1D-InGaAs OMA V detector (Princeton Experiments). For all solution-based experiments the SWCNT/nucleic acid complexes were diluted in PBS to the final concentration and 198 µl of this solution was placed in a 96-well plate. Then 2 µl of the analyte (10 mM or less) was added to yield a final concentration of 100 µM or less. All experiments were done in triplicate and repeated at least twice.

Imaging of Single SWCNTs.

SWCNTs were imaged with an inverted microscope (Zeiss, AxioObserver.Z1) equipped with different objectives [Zeiss α-Plan-APOCHROMAT 100×/1.46 Oil DIC (UV) VIS-IR and Zeiss Plan-Apochromat 63×/1.40 Oil DIC] and an appropriate filter set (Dichroic 620 nm longpass, 900 nm longpass emission filter, Chroma). For image acquisition we used either a 2D OMA V InGaAs array (Princeton Instruments) with 320 × 256 pixels or a NIRvana 640 camera (Princeton Instruments) with 640 × 512 pixels. SWCNTs were excited with a 561-nm (100 mW–1 W) fiber laser (MPB VFL series). The laser was expanded by two lenses (Thorlabs) into the back focal plane of the microscope. The typical laser excitation power was 500 mW.

Cell Experiments.

PC12 cells (ATCC CRL-1721.1) were purchased from American Type Culture Collection (ATCC) and cultivated according to the supplier’s protocol. In short, medium consisting of F-12K medium, 12.5% horse serum, and 2.5% FBS (all from ATCC) was used. Cells attached to cell culture flasks (Corning CellBIND) and were passaged when they reached ∼70% confluence. Cells were incubated at 37 °C and 5% CO₂.

Sensor surfaces were prepared by the following procedure. First, glass substrates (#1, ThermoScientific) were functionalized with (3-aminopropyl)triethoxysilane (APTES) in ethanol (1% APTES, 1% water, 1 h). Then DNA-wrapped SWCNTs [(GA)₁₅ ssDNA if not stated otherwise] were diluted 1:100 in PBS [corresponding to a final absorption of 0.0075 at the (6, 5)-SWCNT S₂₂-peak maximum at around 570 nm] and incubated on top of those surfaces to let them adsorb (1 h). Afterward these surfaces were rinsed 3× with PBS. Then they were incubated with a collagen solution (Collagen, Type I, C3867, Sigma) overnight at 4 °C and rinsed with PBS. For most experiments the glass substrates were mounted from the beginning in a flow chamber (ibidi sticky-Slide VI 0.4).

Finally, PC12 cells were sparsely cultivated in full medium on these surfaces to avoid clustering (at least 4 h). Cell adhesion was verified by light microscopy and the full medium was exchanged to PBS supplemented with calcium and magnesium (Sigma). Cells in the flow chamber and all buffers were kept at 37 °C before the experiment. Cell experiments were performed without an incubation chamber but given the time scale of the experiment (<10 min) the temperature was close to 37 °C. Cells were stimulated by adding 110 mM potassium chloride solution to the inlet of the flow chamber (final concentration ∼55 mM for the complete flow chamber volume).

Data Analysis and Fitting Algorithm.

Movies collected during cell experiments were processed by a fitting algorithm to identify regions where sensors responded to dopamine release. During each frame, the image of 640 × 512 pixels was separated into 4 × 4-pixel groups over which mean intensity values were calculated, resulting in a time-dependent intensity trace for each of the 20,480 groups. Each intensity trace was fit to the following equation:

$$I=a+b\mathrm{\cdot }t+c\mathrm{\cdot }{t}^2+x\mathrm{\cdot }{e}^{\left(\left(-{\left(t-y\right)}^2\right)/2z^2\right)}.$$

The functions formed by parameters a, b, and c represent the possible drift in baseline intensity caused by defocusing of the microscope during the experiment. The parameters x, y, and z represent the amplitude, peak center, and SD, respectively, of a Gaussian function used to fit a region of sensor response. A script was used to process the movies and perform the fittings. The fitting procedure provided images of fitting parameters.

The curvature of the cell contour was calculated by using a three-point approximation (48). For each pixel of the contour the mean (curvature) of the local pixel curvature plus three adjacent pixels was calculated.