- Email: [email protected]
An apparatus for recording synaptic potentials from neuronal cultures using voltage-sensitive fluorescent dyes
Journal of Neuroscience Methods, 38 (1991) 93-105
93
© 1991 Elsevier Science Publishers B.V. 0165-0270/91/$03.50 NSM 01241
An apparatus for recording synaptic potentials from neuronal cultures using voltage-sensitive fluorescent dyes Chi-Bin Chien and J e r o m e Pine Department of Physics, California Institute of Technology, Pasadena, CA 91125 (U.S.A.) (Received 21 January 1991) (Accepted 4 March 1991)
Key words: Voltage-sensitive dye recording; Optical recording; Synaptic potentials; Rat superior cervical ganglion; Microcultures; In vitro networks Voltage-sensitive dyes offer the promise of noninvasive multiceli recording of electrical activity, and should therefore be useful for studying the synaptic interactions of small networks of cultured neurons. We have designed and built a system for recording from microcultures of 1-15 neurons from the rat superior cervical ganglion (SCG), using voltage-sensitive fluorescent dyes of the styryl class. The apparatus comprises a standard inverted epifluorescence microscope; a mercury arc lamp with an optical feedback regulator; a 256-pixel fiber-optic camera with individual photodiode detectors and very low-noise amplifiers; and a personal computer-based data acquisition system. Its dark noise and illumination fluctuations are low enough that at typical fluorescence levels for these cells, it is limited by shot noise (the inherent physical limit of detection). Recording from SCG neurons, the signal-to-noise ratio is high enough to see large subthreshold synaptic potentials without signal averaging. This apparatus should be useful for studying long-term synaptic plasticity in cultures of vertebrate neurons, and several of its features should apply to optical recording from other preparations.
Introduction
This paper, along with a related one (Chien and Pine, 1991), describes the application of voltage-sensitive dye recording to microcultures of neurons from the rat superior cervical ganglion (SCG). Microcultures - isolated networks of a few neurons, grown in cell culture - are good preparations for studying synaptic interactions (Furshpan et al., 1976; Kaczmarek et al., 1979; Ready and Nicholls, 1979) and properties of biological neural networks (Parsons et al., 1989; Kleinfeld et al., 1990; Syed et al., 1990) because
Correspondence: Dr. C.-B. Chien, Dept. of Biology B022, University of California at San Diego, La Jolla, CA 920930322, U.S.A. Tel.: (619) 534-2526; Fax: (619) 534-8866.
they have few cells, often have a high proportion of synaptic connections, and are easily manipulated. Cell culture also makes it possible to follow development of synaptic connections over long times (Potter et al., 1986). Dye recording (for reviews, see Salzberg, 1983; Cohen and Lesher, 1986; Grinvald et al., 1988) is particularly appropriate for microcultures of vertebrate neurons because it can give a complete record of the activity of every neuron in the network and because it is relatively noninvasive, allowing repeated measurements from the same culture. (Using intracellular electrodes, in contrast, it is difficult to record from more than two or three cells at once, and even harder to repeatedly penetrate the same cell.) Our eventual goal is to combine dye recording with extracellular stimulation using integrated multielectrode dishes (Re-
94 gehr et al., 1989), in order to measure all the synaptic strengths in a microculture without any intracellular penetrations. Dye recording is technically demanding for any preparation; the main challenge is to obtain an adequate signal-to-noise ratio. Our apparatus is optimized for SCG neurons, which we chose because they have previously been grown in microculture, where they form fast cholinergic synapses with each other (Higgins et al., 1984; Furshpan et al., 1986). As with all vertebrate neurons, these cells are relatively small (20-30 txm in diameter) and so give very low detected fluorescence. This apparatus can operate at the shot-noise limit, even at these low light levels, because of several straightforward innovations, including an optical feedback regulator circuit for stabilizing the mercury arc lamp, and very quiet preamplifiers for the detector array. Using the fluorescent styryl dye RH423, a close analog of RH421 (Grinvald et al., 1983), we can record from all the neurons in an SCG microculture at a signal-to-noise ratio sufficient to see large postsynaptic potentials without signal averaging. This recording setup can be used with other cultures of vertebrate neurons (mass cultures as well as microcultures), and its key features may be adapted to other systems. This paper deals only with the construction and performance of the apparatus itself, emphasizing design decisions and technical features that will be useful for other systems. Results of dye recording experiments on SCG microcultures are described in Chien and Pine (1991). Parts of this work have been described in abstract form (Chien et al., 1987), and an earlier version of the optical feedback regulator circuit has been mentioned in a review (Cohen and Lesher, 1986). Further details of the apparatus and its design are available (Chien, 1990).
System overview The apparatus (Fig. 1) is built around an Olympus IMT-2 inverted microscope, with an incandescent lamp for phase contrast and standard attachments for epifluorescence: mercury arc
mercury I ( ~ c u l t u r e ~ arclampshu'tel ~
pus,M 2
22;o'12;'i
sideport (tofiber-opticcamera) Fig.1.Schematicdrawingofthedye-recordingapparatus.
lamp, epifluorescence illuminator, and filter cube. An electromechanical shutter controls the fluorescence illumination; it is mechanically isolated by a heavy support and two layers of foam rubber. Dye signals are detected by a 256-pixel fiber-optic camera detector (not shown here; see Fig. 4) that fits into the side port of the microscope; a 35-mm camera takes photographs; and a video camera provides an image for aligning the culture with the pixels of the detector. For electrical and vibrational isolation, the apparatus sits inside a Faraday cage and on top of a heavy marble table supported by four inner tubes. In designing a dye-recording system, photons are at a premium. The system should maximize the signal-to-noise ratio, while at the same time minimizing bleaching and phototoxicity. This requires (1) efficient optics, with a high illumination intensity and a high efficiency for collecting fluoresced photons (the latter is especially important); (2) a dye that gives bright fluorescence and good voltage sensitivity; and (3) low instrumental noise, stable illumination and quiet photodetectors. Optical efficiency is achieved by using a good commercial microscope and an objective lens with high numerical aperture (NA). The NA determines the illumination intensity and the collection efficiency for fluorescence, which are both roughly proportional to the square of the NA (see
95 Grinvald et al., 1983; however, the exact proportionality is obtained only under certain ideal conditions). In general, maximizing NA requires high objective magnification (M). (Strictly speaking, this is only true for conventional microscope optics: one could, in theory, design a microscope with a super-large-diameter objective that would give high NA at low M.) However, M cannot be increased without bound, since the field of view is limited to the tube diameter (about 20 mm) divided by M. In order to encompass a typical microculture, 750 /zm in diameter, we use a dry 20 x objective (Fluor 20 Ph3DL, Nikon); its NA is 0.75. For small microcultures we sometimes use a 40 x lens (Fluor 40 Ph3DL, Nikon, NA 0.85). To reduce bleaching and phototoxicity, it is important to illuminate the specimen only when actually recording. The electromechanical shutter (No. 1 Synchro, Ilex Optical Co., Rochester, NY) turns illumination on and off; it opens and closes in a few milliseconds and is usually set for flashes lasting 100-250 ms. The sensitivity of dye recording depends greatly on the particular dye used. With SCG cultures, we found the best results when staining with micromolar concentrations of the styryl dye RH423 (a kind gift of Dr. Amiram Grinvald) for 5-10 min; this yielded a typical detected fluorescence photocurrent of 1 nA per cell, and a typical dye sensitivity of 1% fluorescence change per 100 mV change in membrane potential. Culture methods, staining procedures, and the dye's structure are described.in detail elsewhere (Chien and Pine, 1991). For fluorescence, we use the Olympus " G " filter cube with an extra LP515 excitation filter. This combination passes exciting light from 520550 nm (the excitation is dominated by the 546 nm mercury line) and emitted light longer than 610 nm. Using these filters with the standard IMT-2 epi-illuminator and the 20 x ,/NA 0.75 objective, the illumination intensity at the specimen is 10 W / c m 2. With this illumination, the detected fluorescence from a typical stained SCG neuron is 6 x 109 photons/s, which gives the 1 nA photocurrent quoted above. Photodynamic damage at this intensity is tolerable: SCG neurons illuminated for at least 100 flashes of 100 ms
each do not show detectable deterioration of their action potentials. The sources of noise in a dye-recording system can be classified by the power of their dependence on the detected intensity I. Illumination noise is proportional to I, and is usually due to intensity fluctuations in the light source. Shot noise, proportional to x/I, is due to statistical fluctuations in the detected light. Dark noise, independent of I, is due to noise in the detection system's photodetectors or amplifiers. Since these noise sources are uncorrelated, they combine in quadrature: (itotal) 2 = (/shot) 2 + (iill) 2 + (idark) 2 (A lowercase i denotes a root-mean-square [rms] noise amplitude.) What we refer to as illumination noise has also been called extraneous noise (Cohen and Lesher, 1986) and technical noise (Parsons et al., 1989). For a good discussion of noise in dye recording, see the review by Cohen and Lesher (1986). A standard procedure in noise design is to identify the noise source that is hardest to eliminate, and then attempt to bring all other sources below this level. Shot noise is a fundamental physical limit for photodetection: it comes from the quantal nature of light, and cannot be eliminated. Therefore, when /ill and /dark are less than about half of/shot, the signal-to-noise ratio will be optimal; further reduction will not significantly decrease the total noise. The present apparatus is nearly shot-noise limited: /ill and /dark are somewhat less than /shot, though not always as low as ishot/2. The shot noise in a photocurrent I is given by isho, = ~/(2q/B)
(1)
where q = 1.6 x 10 -19 C is the charge of the electron and B is the detection bandwidth (Horowitz and Hill, 1989). B must be at least 300 Hz to see action potentials (about 1 ms long) with minimal distortion. The rms shot noise in I = 1 nA is then 0.3 pA, or 0.03%; with RH423 dye signals of 1%/100 mV, this corresponds to a 3 mV change in membrane potential. Therefore, in shot-noise-limited recordings at these levels of
96
fluorescence, it should be possible to see large synaptic potentials in a single trial. (Noise may be reduced further with judicious signal averaging.) The next three sections describe the key parts of the apparatus: the optical-feedback regulator (which reduces illumination noise), the fiber-optic camera detector (which has low dark noise), and the PC-based data acquisition system.
4 5 A from power supply
I
(
~VV~-.-
Harnamats~
s,os,.o,
i ,0o~.'2 t~') ,J v'-~
2N6388
Feedback-regulated light source
7 7 ,,,1~ ~
~
°~'° !~
, .... >.-
-
1'- --- +1>~
~
]-"--$-<,~F356
r ~""
I
047 ,uF ~'-
T :.~EG.. /
~
i
<"
_
/
Detecting a reasonably large fluorescence from small cells requires a bright light source; we find a commercial mercury arc lamp (HBO 100W/2, Osram) to be suitable. Mercury arcs, though bright, are notoriously unstable (Fig. 3a,c), so we use an optical-feedback regulator circuit to reduce illumination noise. A photodiode mounted in the side of the lamp housing (see Fig. 1) monitors the lamp intensity, and the regulator circuit modulates the current supplied to the arc so as to keep this intensity steady.
Apparatus The arc lamp is driven by a standard power supply (Olympus BH2-RFL-T2) with an integral radiofrequency (RF) igniter; it typically operates at 4.5 A and 22 V. We originally used a currentregulated DC supply with a spark igniter (as in Grinvald et al., 1982), which had lower linefrequency ripple, but the spark igniter had a distressing tendency to damage other electronics. The Olympus unit's RF igniter does not have this problem, and its power supply's line-frequency ripple is suppressed by the feedback regulator. Fig. 2 shows the regulator circuit. The AC component of the photocurrent from the lamphouse photodiode is amplified and used to drive a power transistor that shunts the lamp. When a fluctuation increases the lamp intensity, the transistor shunts more current and reduces the intensity: negative feedback. The standing current through the transistor determines the maximum output swing of the circuit. During ignition, a switch is used to disconnect the regulator and avoid shortening the ignitor. The R E G output is proportional to the lamphouse intensity; the op-
Fig. 2. Optical-feedbackregulator circuit. The output of the lamphouse photodiode (Hamamatsu S1087-01) is amplified and drives a shunt regulator: when the lamp intensity increases, the 2N6388 Darlington power transistor in parallel with the lamp shunts more current. The x 230 amplifier is a combination of LF356 op-amps with a gain of 230. The circuit's passband is 3-300 Hz and its gain can be adjusted from zero to full (using the 1 kg2 potentiometer) in order to check the effects of regulation. The REG output provides a monitor of intensityfluctuations.
eration of the regulator can be checked by watching R E G while varying the regulator's gain from zero to full, using the 1-kO potentiometer. The bandwidth of the regulator is 3-300 Hz (3 dB down frequencies). Some care has been taken to avoid unnecessary phase shifts; "motorboating" (oscillations at very low frequencies) was originally quite troublesome. Those fluctuations falling within this bandwidth should be suppressed by a factor equal to the closed-loop gain. The DC photocurrent of the lamphouse photodiode is typically 0.7 mA. Making the approximation that this photocurrent is proportional to the lamp intensity, which is in turn proportional to the lamp current, the closed-loop gain is [(0.7 mA)/(4.5 A)][(7.7 kO)(230X15)/(15 g2)] -- 280.
Performance The mercury arc lamp has two sorts of noise, fast and slow. The fast fluctuations are mostly line-frequency ripple; with the Olympus power supply, their amplitude is 1.5% peak-to-peak, 0.5% rms. This ripple can obscure action potentials ( ~ 1 ms timescale) or fast PSPs ( ~ 10 ms). The slow fluctuations, presumably due to arc
97
a) unregulated
b) regulated
'amp house
10'%
stage ~J~j~JW~AJ~,J~/,J/I~%
~ . ~
~ J ~"~ I 0.1%
20 ms
c) unregulated
lamp- --".-J'~k_ 11% house t/'-"-stage --,~k._q~/---'--- ]1% I
d) regulated
I'% ..~.~"-'~
II %
I
ls Fig. 3. Regulation of lamp fluctuations, measured at the lamphouse and at the specimen stage. The regulator circuit's gain was zero for the unregulated traces and full for the regulated traces, a and b show fast fluctuations (filtered with passband 0.16-320 Hz), mostly at the line frequency; c and d show slow fluctuations (passband 0.16-5.3 Hz). In each pair, the lamphouse and stage fluctuations were measured simultaneously.
shot noise of 0.03% (see above System Overview), but it could presumably be reduced to 0.002% by replacing the lamphouse photodiode with an intensity monitor that directly sampled the light going to the specimen. Fig. 3c and d show the regulation of slow fluctuations. These are also anisotropic: note the differences between the lamphouse and stage traces in Fig. 3c. Though the slow fluctuations seen at the lamphouse photodiode are heavily suppressed by the regulator, those at the stage are still present. (Again, better placement of the intensity monitor would improve regulation.) These slow fluctuations are inconvenient, but can be dealt with in software by rejecting those data sets that show signs of an arc wander: a sudden shift or slope change in the dye signal's baseline, simultaneous across pixels.
Optical detector and electronics wander (movement of the mercury arc), consist of occasional spikes in intensity that rise in tens of milliseconds and decay in hundreds of milliseconds, with amplitudes ranging from below 0.1% to above 3%. Bulbs go through noisy and quiet periods that last for minutes, changing unpredictably from quiet to noisy and back. (They seem to become noisier with age.) During a typical quiet period, spikes exceeding 0.1% occur at 0.l/s; during a typical noisy period, 0.1% spikes occur at 2/s, with half exceeding 0.5%. These spikes are too slow to obscure action potentials, but cause a sloping baseline in the dye signal that can make it impossible to quantitate small potential changes. Comparison of Fig. 3a and b shows the regulation of fast fluctuations. The rms noise measured by the lamphouse photodiode (located at right angles to the output light path of the lamp) drops from 0.5% to 0.002%, nearly the 280-fold improvement predicted above. However, since the lamp's fluctuations are not isotropic, it is important to measure the illumination intensity at the stage where the specimen is. A photodiode put in place of the specimen measures a smaller improvement: the rms noise drops from 0.6% to 0.02%. This noise is already below the typical
The optical detector, shown schematically in Fig. 4a, has two parts: a primary photodetector array, and amplifying electronics. The primary detector is a fiber-optic camera (Fig. 4c), which comprises a bundle of 256 optical fibers connected to 256 discrete photodiodes. A real image of the culture is focussed onto the face of the bundle (Fig. 4b); each fiber picks up the light from one pixel and conducts its light to its own photodiode. The photodiode transduces this light into a photocurrent, which is carried by a coaxial cable to the amplifying electronics. There the photocurrent is converted to a voltage by a lownoise transimpedance amplifier, multiplexed with signals from other pixels, and sent to the data acquisition system for digitization. Fiber-optic camera
The fiber bundle of the fiber-optic camera fits within the tube diameter of the IMT-2 (roughly 20 mm). The number of fibers matches the geometry of SCG microcultures. A typical cell is 30 /xm in diameter, roughly 1/20th the diameter of a culture. In order to have at least one pixel per cell, the square plastic fibers (Poly-Optical, Santa Ana, CA) are arranged in an octagonal bundle 18
98 fiber bundle (end view)
multiplexer
fiber-optic camera
preamplifiers
a
coaxial cables
currents, while dye-recording amplifiers have more stringent signal-to-noise requirements and must be produced by the hundreds. A fundamental limit for the noise of a transimpedance amplifier is the Johnson noise (noise due to thermal fluctuations) in its feedback resistor Rro. All amplifier noise sources are equivalent to either an input c u r r e n t /in or an input voltage vm; for instance, the Johnson noise in Rrb is equivalent to a current noise with rms amplitude
V/(4kTB/Reo) (2) where kT is the product of Boltzmann's constant /Job. . . . =
2,
b Fig. 4. a: schematic view of the 256-pixel optical detector. Only eight of the fibers and diodes within the fiber-optic camera are shown, b: face view of the fiber bundle. Scale in cm. c: oblique view of the fiber-optic camera.
fibers across. The average spacing is 0.9 mm, though nonuniformities in the fibers make the array slightly irregular. With a 20 × objective, the detector has an 800 ~m field of view with 45/zm pixels. The dead space between fibers (due to cladding and adhesive) is about 20%. We have tested the fibers' transmission with a light-emitting diode source and found a variation of 20% (SD) between fibers; three fibers with very low transmittance were probably nicked or poorly joined to their photodiodes. The $1087-01 (Hamamatsu) photodiodes were chosen for low noise, good quantum efficiency, and moderate capacitance.
Preamplifier design The preamplifier circuit (Fig. 5) is designed to minimize noise, given the desired bandwidth (300 Hz) and the expected range of photocurrents (0.2-2.0 nA). It has much in common with patch-clamp preamplifiers (Sigworth, 1983): both are very low-noise, low-current transimpedance amplifiers. The main differences are that patchclamp amplifiers are faster and amplify smaller
and the absolute temperature of the resistor (Horowitz and Hill, 1989). The relative contributions of ii. and el. depend on the input load Zin. The total amplifier noise referred to the input, /amp, is given by (iamp) 2 =
(ii.) 2 + ( v i . / Z i . ) z
If the amplifier is sufficiently quiet, i Job.son will dominate /in, which in turn will dominate /amp, and the amplifier's noise will be proportional to 1/((Reb) • The larger the feedback resistor, the quieter the amplifier may be; however, if Reo is too large the amplifier will saturate at very low light levels and stray capacitances will become a problem. This preamplifier comprises an OP27 operational amplifier preceded by a low-noise input stage. The op-amp in the second stage has moderately low noise, and provides very high gain; the Rfb(5 6 G~.,~) .
+11,
I I I
[
510 ~2 5%
51~ 5%
muhiple~:cl
L
Fig. 5. Preamplifier circuit for the optical detector. A JFET input stage and an OP27 form a high-input impedance inverting amplifier. Together with the feedback resistor Rt~, they form a low-noise, low-input-current transimpedance amplifier. Two RC filters prevent aliasing.
99
JFET (junction field-effect transistor) input stage has very low noise and provides moderate amplification, so that noise from the second stage is unimportant. The bipolar transistors (2N3704) set the drain-source voltage of the FETs (2SK240GR, Toshiba), and prevent second-stage noise from coupling back to the amplifier input via the FET gate-drain capacitance. We use feedback resistors in the range 5-6 GO (Victoreen), large enough that a very bright cell (2 nA photocurrent) will just saturate the amplifiers. A quantitative analysis of the circuit (details omitted) shows that /Johnson should dominate i~n, while the FETs' voltage noise should dominate vi,. Zi, is determined by the capacitances of the photodiode, coaxial cable, and FET, which sum to 300 pF. At the working frequencies of the amplifier, Zi, is large enough that /amp should be well approximated by the Johnson noise of Rn,. For a resistance of 6 GO and a bandwidth of 300 Hz, theory predicts /amp = 28 fm rms.
Performance In practice, the measured dark noise of the preamplifiers is somewhat higher than that predicted from Johnson noise, for several reasons: most have bandwidths greater than 300 Hz, there is line-frequency noise, and some amplifiers have excess noise of undetermined origin. A power spectrum of the quietest amplifier (not shown) shows it to be right at the Johnson noise limit, but the amplifiers' average dark noise is 78 + 13 f A r m s (mean + SD). This deviation from theory is not too distressing, because this dark noise is still well below shot noise at typical fluorescence levels; for instance, at I = 1 nA it is only 0.008%, against a shot noise of 0.03%. A useful figure of merit is /as, the photocurrent at which amplifier noise is equal to shot noise. For an amplifier limited by the Johnson noise of its feedback resistor, combining Eqns. 1 and 2 yields a simple rule: Rfb Ias = 50 mV. If our preamplifiers were Johnson-noise limited they would have Ia, = 10 pA, but because of excess noise they have I~s = 50 pA on average. For photocurrents greater than 50 pA, shot noise will be greater than dark noise. The multigigohm resistors used here are very
susceptible to small stray capacitances, which affect their frequency responses. Capacitances in parallel with the resistors cause the preamplifier bandwidths to vary from 150 to 650 Hz (they are mostly around 400 Hz); capacitances to ground cause gain peaking in some amplifiers; and capacitances between certain amplifiers cause crosstalk. Fortunately, these effects are all small and do not introduce spurious signals, though they can alter signal sizes somewhat. (For quantitative measurements, signals can be checked for contributions from crosstalk, and corrected in software for their amplifiers' bandwidth and gain peaking.)
Data acquisition hardware and software
Hardware The data acquisition system (Fig. 6) is built around an 8 MHz IBM AT personal computer (PC) and a TransEra 7000-MDAS modular data acquisition system (TransEra, Provo, UT; now manufactured by Acurex, Mountain View, CA). The preamplifiers are arranged in groups of 16 on 16 printed-circuit cards. Each card has two CD4051 multiplexer chips that combine its preamplifier outputs into a card output, which is sent to a controller card; two more CD4051s on the controller card combine the card outputs into the signal that is sent to the MDAS for digitization. The MDAS has a 12-bit analog-to-digital converter (ADC) and its own 68000 microprocessor, and talks to the PC over an IEEE-488 bus. It is an elegant device because it relieves the PC of addresses
shutter frigger
camera
IEEE-488 bus
Fig. 6. Data acquisition system. An I B M A T personal computer controls an autonomous data acquisition system, which sends pixel addresses to the multiplexers and digitizes their
output. The PC also acquires images from the video camera.
100 direct responsibility for data acquisition: the PC need only send it simple ASCII commands, whereupon it triggers the shutter, digitizes several thousand fluorescence values, and returns the data to the PC. The MDAS uses custom software written by TransEra to address the multiplexers and acquire data simultaneously. This subroutine can sample data at up to 90 kHz, i.e., 45 pixels at 2 kHz/pixel. Even allowing 2 or 3 pixels per cell for cells that fall on pixel boundaries, this is enough to record from all the cells in a large microculture (at least 15 cells). Because the analog-to-digital converter is bipolar, only 11 of its bits are useful; however, its dynamic range is extended by a power-of-2 programmable amplifier. The rms digitization noise varies between 0.014% and 0.028%, depending on the exact light level; it is reduced by offline digital filtering (Fig. 8e).
Software The data acquisition and analysis software for the PC consists of four parts: an alignment program OPTIK, a data acquisition program MULTI; a graphing and analysis program DISPLAY; and about 20 small analysis programs that average, filter, bleach-subtract, Fourier-transform, etc. These are variously written in Microsoft QuickBASIC ® and Lattice C ®, with a few subroutines in Assembler for speed. The source code, about 800 kbytes long, is available from the authors on floppy disk. When an experiment is first set up, O P T I K uses a frame-grabber board to acquire a video image of the culture, align it with a stored representation of the pixel array, and choose pixels for recording. M U L T I then runs the experiment: it takes electrophysiological and fluorescence data, displays them on-screen in real time, and does simple data analysis such as signal averaging. Data files are stored on magnetic disks and analyzed afterwards using DISPLAY and the suite of analysis programs. At present, MULTI is arbitrarily limited to recording from 16 channels at once. The MDAS can record up to 30 000 points from each channel, while available memory in the AT is limited to about 200000 points total from all channels.
Fig. 7 shows an example of the pixel alignment procedure for a 5-cell microculture. The phasecontrast photograph a shows the 5 cell bodies, the dense mat of processes, and the thick fascicle that girdles the culture. The video image b shows the culture, with a superimposed pattern of squares showing pixel locations. Dye signals were recorded from the pixels shown as white squares, while the dark squares (not all easily visible here) are unselected pixels. (The unevenness of the squares r e f e c t s the irregularities in the optical fiber bundle.) The phase-contrast photo serves as a record of the detailed anatomy of the culture, while the stored video image merely records the positions of the pixels used. Tracing over the video image yields the culture scheme c, used for easy reference during data analysis. Fig. 8 shows the steps used to extract a dye signal from the raw fluorescence data; they are a simplified version of the procedure described by Grinvald et al. (1982). An SCG neuron in mass culture was stained with RH423, then penetrated intracellularly and stimulated while recording its fluorescence. Trace f shows the intracellular action potential. In the raw fluorescence trace a, the dye signal is dwarfed by the initial step from 0 to 0.86 nA (the cell's resting fluorescence) when the shutter opens. Trace b is an amplified version of the signal, calibrated as a fractional change in fluorescence; it has been inverted so that an upward deflection corresponds to a depolarization. The action potential is clearly visible above the noise (the digitization noise is quite obvious). The rising baseline is an artifact due to photobleaching of the dye. To remove this bleach artifact, a pure bleach trace is recorded without stimulation (c, jagged trace). A quadratic curve fitted to this trace (c, smooth trace) is subtracted from b to yield the bleach-subtracted trace d. Finally, filtering the bleach-subtracted trace at 300 Hz yields e, which is very similar to the intracellular recording f. The peak-to-peak noise in Fig. 8e is about 0.07%; since the dye sensitivity for this cell was 0.92%/100 mV, this translates to 7.5 mV. This signal-to-noise ratio is sufficient to see a large EPSP (it might be desirable to average a few trials). Fig. 9 shows an example of an optically-re-
101 shutter opens a) raw
05 n
b) x-100 c) bleach & fitted bleach
AF/F [ 0-5% d) b]eachsubtracted e) filtered
50 mV
-
-
f) intracellular
20 ms Fig. 8. Dye signal from an SCG neuron, showing standard data reduction. This data is from a single, unaveraged trial, a: raw fluorescence signal; the dye signal is a small decrease at the limit of visibility. The arrow indicates the shutter opening. b: amplified, inverted fluorescence signal (gain = - 100). c: fluorescence signal with no stimulus, showing the bleach artifact only. The jagged trace is the actual bleach; the smooth trace is a quadratic fit. d: bleach-subtracted trace (b minus the fit from c). e: bleach-subtracted trace, filtered digitally at 300 Hz. f: intracellutar recording. The cell was stimulated through a bridge circuit with a square current pulse. Scale bars: 0.5 nA for a, 0.5% for b, c, d, e (arrow points towards increasing fluorescence); 50 mV for f.
ol
corded PSP, with simultaneous intracellular r e c o r d i n g s f o r c o m p a r i s o n . I n a five-cell m i c r o c u l t u r e , cells 1 a n d 3 w e r e p e n e t r a t e d i n t r a c e l l u larly w h i l e all five cells w e r e m o n i t o r e d o p t i c a l l y ; s t i m u l a t i o n o f cell 1 r e s u l t e d in a l a r g e P S P in cell 3 w i t h t w o c o m p o n e n t s , t h e first a p p a r e n t l y a m o n o s y n a p t i c c o n n e c t i o n f r o m cell 1, a n d t h e s e c o n d a d i s y n a p t i c c o n n e c t i o n t h r o u g h cell 2
I
I
100 pm
Fig. 7. Example of pixel alignment for a 5-cell microculture, a: phase-contrast. The bright object in the center is a piece of debris, b: stored video alignment image. White squares are selected pixels; dark squares are unused pixels, c: culture schematic, made by tracing b. Scale bar = 100 p.m for all 3 panels.
102
a (av 3 trials)
intra
b (single trial)
J
/
50 mV
optical
50 m s
Fig. 9. An excitatorypostsynapticpotential in an SCG neuron, measured intracellularly(intra) and with dye recording from a pixel over the cell body (optical). The detected photocurrent from this pixel was 0.77 nA. a: subthreshold PSP (average of 3 trials), b: suprathreshold PSP, with resulting action potential (single trial). Scale bars apply to both panels.
(data not shown). Fig. 9a shows an average of three trials in which the compound PSP was subthreshold; Fig. 9b shows a single trial in which it was suprathreshold. The signal-to-noise ratio of the dye signal is clearly sufficient to detect the PSP and make a quantitative estimate of its amplitude relative to that of the action potential.
Discussion Dye recording from cultured vertebrate neurons is difficult because the cells are small and so give little fluorescence. With the 10 W / c m a illumination and NA 0.75 objective used here, detected fluorescences from SCG neurons are only 0.2-2 nA; these are difficult to increase. There are 3 ways to try: a higher NA objective, more intense illumination, or heavier staining. The NA could be improved somewhat by switching to a 32 x oil-immersion objective (if a compatible lens could be found); the field of view would still be tolerable, and the detected fluorescence would be roughly doubled. Brighter illumination would probably require using a laser light source, and
seems likely to cause an undesirable level of phototoxicity. The cells are already very heavily stained: at a rough estimate (Chien and Pine, 1991), there is 1 dye molecule for every 100 phospholipids in the cell membrane. In view of these difficulties, it will be difficult to increase detected fluorescences from this preparation by a large factor. Given these fluorescence levels and a 1 % / 1 0 0 mV voltage sensitivity (the best that we have been able to achieve after strenuous attempts at optimization), it is just possible to detect synaptic potentials optically. Taking 1 nA as a typical detected photocurrent and a bandwidth of 300 Hz, shot noise is 0.03% rms. Our apparatus achieves its design goal of being shot-noise limited: amplifier dark noise is 0.008%, digitization noise is 0.014-0.028%, and fast illumination noise is 0.02%. (Trials are sometimes spoiled by slow illumination fluctuations, but these can be rejected.) These calculations predict that large synaptic potentials should be visible above shot noise, and Fig. 9 confirms this. Below, we discuss the parts of the apparatus in turn, comparing them to previous setups and mentioning possible improvements. Light source It is possible to compensate for illumination fluctuations in software by monitoring the illumination intensity and using it to correct the detected fluorescence signal, but it is preferable to use a regulator to eliminate these fluctuations at the source. With the Olympus power supply, the optical feedback regulator somewhat reduces slow fluctuations due to arc wander, and reduces linefrequency ripple at the specimen from 0.5% to 0.02% rms. This is significantly lower than the 0.06% rms we found when using a DC current supply (Grinvald et al., 1982) without the feedback regulator. The fluctuations are not spatially uniform: both slow and fast fluctuations differ between the lamphouse photodiode and the stage. In fact, the fluctuations at the lamphouse diode are roughly tenfold lower than those at the stage, suggesting that regulation could be improved 10fold by replacing the photodiode mounted in the lamphouse with another that directly sampled the
103 light going to the stage, using a partially reflecting mirror. Even with a properly placed intensity monitor, the present regulator can only compensate for fluctuations that are spatially uniform across the specimen. Fluctuations due to movement of the mercury arc (arc wander) are often nonuniform: using the fiber-optic camera to record reflected illumination light from across the specimen, we found (data not shown) that roughly one-third of the fluctuations were clearly nonuniform. It may be possible to eliminate spatial nonuniformity by passing the illumination through an optical-fiber scrambler. Alternatively, nonuniform illumination could be corrected by a dual-wavelength ratioing scheme (Montana et al., 1989).
Optical detector Previous multipixel detectors for dye recording (e.g., Grinvald et al., 1981) have used monolithic photodiode arrays, though Tank and Ahmed (1985) built a fiber-optic camera similar to ours. When we built our detector, the largest commercially available photodiode array was the Centronic MD-144, with 12 x 12 pixels. The main reason for using the fiber-optic camera instead was our desire for more pixels. Discrete photodiodes also have other advantages: the wide variety available lets one pick diodes with especially good properties (low noise and capacitance), and bad or damaged diodes may be replaced. It is relatively straightforward to scale up the fiber-optic camera to a larger number of pixels, and the cost per pixel should be constant, so a fiber-optic camera is a reasonable choice for a large detector. However, the technology for large monolithic detectors is likely to improve rapidly; Matsumoto and Ichikawa (1990) have recently reported using a 128 x 128 pixel array. The dark noise of our preamplifiers is such that our experiments are shot-noise-limited for photocurrents greater than 50 pA. Though preamplifier designs for dye recording have improved (e.g., Grinvald et al., 1988, section IV.B.2; Bonhoeffer and Staiger, 1988), we are not aware of published circuit details or noise data by which to compare them to our design. The amplifiers are in fact quieter than is necessary for our
experiments: 50 pA is too dim to record synaptic potentials (the shot noise is too large). However, they would be useful for experiments that only need to detect action potentials, where it would be advantageous to operate at low levels of illumination in order to reduce phototoxicity. Two improvements would be useful for the next generation of amplifiers: reducing the feedback resistor to 1 GO, and improving the printed-circuit layout. Smaller feedback resistors would ameliorate stray-capacitance problems and (because of the present excess noise) should not appreciably increase amplifier noise. A combination of lower Rro with better layout should eliminate crosstalk and improve amplifier frequency responses. The low noise and gigohm feedback resistors of our preamplifiers are specifically designed for the low light levels of fluorescence, and they cannot be used for recording with absorption dyes, as they would be saturated by the much higher light levels used there. Simpler amplifiers with megohm feedback resistors would suffice for absorption recording.
Data acquisition Most dye-recording systems have used PDP-11 minicomputers for data acquisition (Cohen and Lesher, 1986, section 5.7; Grinvald et al., 1988, sections IV.B and IV.C), with software written in Assembler language. The rapid advances in the power of personal computers made it possible for us to use an 8 MHz IBM PC-AT, which has the advantages of cheapness and ease of programming. Our system only has a d]ta acquisition rate of 90000 samples/s, but since the cell bodies in SCG microcultures are sparse, about 30 pixels (2 pixels for each of 15 cells) suffice to record from all the cells in a culture, so this rate is sufficient. Much faster personal computers have become available since this system was built, and it is clear that a personal computer-based system could record from hundreds of pixels at a time.
Applications This apparatus was designed to detect subthreshold synaptic potentials with dye recording, and its sensitivity is sufficient for this purpose.
104 U n f o r t u n a t e l y , for S C G m i c r o c u l t u r e s the optic a l l y - r e c o r d e d P S P shown in Fig. 9 is t h e exception, not t h e rule. A s d e s c r i b e d in d e t a i l elsew h e r e ( C h i e n a n d Pine, 1991), S C G m i c r o c u l tures a r e a p o o r p r e p a r a t i o n for such e x p e r i m e n t s b e c a u s e o f t h e i r e x t r e m e l y c o m p l e x axonal arbors: dye signals from PSPs in t h e cell b o d i e s a r e usually o b s c u r e d by signals f r o m action p o t e n t i a l s in axons t h a t pass t h r o u g h the s a m e pixel. ( T h e e x a m p l e shown a b o v e is o n e of two cases w h e r e we w e r e a b l e to r e c o r d a c l e a r P S P u n o b s c u r e d by b a c k g r o u n d signals.) W e h o p e to resolve this p r o b l e m by switching to a d i f f e r e n t c u l t u r e system w h e r e axons a r e s h o r t e r a n d cell b o d i e s a r e spatially isolated. In such a system, c o m b i n i n g dye r e c o r d i n g with ext r a c e l l u l a r s t i m u l a t i o n by m u l t i e l e c t r o d e dishes ( R e g e h r et al., 1989), it should b e possible to m e a s u r e synaptic plasticity in m i c r o c u l t u r e s over days a n d weeks, a n d to study t h e effects o f c h r o n i c s t i m u l a t i o n at the single-cell level. T h e p r e s e n t system s h o u l d be widely a p p l i c a ble to f l u o r e s c e n c e dye r e c o r d i n g f r o m c u l t u r e s of v e r t e b r a t e n e u r o n s , r e c o r d i n g e i t h e r f r o m multiple cells, or from m u l t i p l e sites on the p r o c e s s e s of an individual cell. T h e p r e p a r a t i o n s for which the system is useful a r e mostly d e t e r m i n e d by its spatial r e s o l u t i o n (18 pixels across, e a c h pixel 45 ~ m with a 20 x objective, 2 2 / ~ m with 40 x ) a n d the f l u o r e s c e n c e levels for which it is d e s i g n e d ( 0 . 2 - 2 . 0 n A d e t e c t e d p h o t o c u r r e n t ) . Its key comp o n e n t s - the o p t i c a l f e e d b a c k r e g u l a t o r , fiberoptic c a m e r a , a n d J F E T p r e a m p l i f i e r design should be g e n e r a l l y useful for optical r e c o r d i n g systems.
Acknowledgements W e w o u l d like to t h a n k A m i r a m G r i n v a l d for his gift o f dyes a n d for g e n e r o u s l y supplying his e x p e r t i s e d u r i n g t h e initial p h a s e s o f this project; A n d r e w H s u a n d M i c h a e l W a l s h for t e c h n i c a l help in b u i l d i n g t h e optical d e t e c t o r ; a n d L a r r y C o h e n a n d J i a n - Y o u n g W u for useful c o m m e n t s on t h e m a n u s c r i p t . This w o r k was s u p p o r t e d by g r a n t s f r o m t h e N I H ( N S 2 2 4 5 0 - 0 2 ) , t h e System Development Foundation, and Sperry-Univac.
References Bonhoeffer, T. and Staiger, V. (1988) Optical recording with single cell resolution from monolayered slice cultures of rat hippocampus, Neurosci. Lett., 92: 259-264. Chien, C.-B., Crank, W.D. and Pine, J. (1987) Noninvasive techniques for measurement and long-term monitoring of synaptic connectivity in microcultures of sympathetic neurons. Soc. Neurosci. Abstr., 13: 1426, abstr. #393.15. Chien, C.-B. (1990) Voltage-sensitive dye recording from networks of cultured neurons. Ph.D. thesis, California Institute of Technology. Chien, C.-B. and Pine, J. (1991) Voltage-sensitive dye recording of action potentials and synaptic potentials from sympathetic microcultures. Biophys. J., in press. Cohen, L.B. and Lesher, S. (1986) Optical monitoring of membrane potential: Methods of multisite optical measurement. In DeWeer, P., and Salzberg, B.M. (Eds.), Optical Methods in Cell Physiology, Wiley (Interscience), New York, NY, pp. 71-99. Furshpan, E.J., MacLeish, P.R., O'Lague, P.H. and Potter, D.D. (1976) Chemical transmission between rat sympathetic neurons and cardiac myocytes developing in microcultures: evidence for cholinergic, adrenergic and dualfunction neurons. Proc. Natl. Acad. Sci. USA, 73: 42254229. Furshpan, E.J., Landis, S.C., Matsumoto, S.G. and Potter, D.D. (1986) Synaptic functions in rat sympathetic neurons in microcultures. I. Secretion of norepinephrine and acetylcholine. J. Neurosci., 6: 1061-1079. Grinvald, A., Cohen, L.B., Lesher, S. and Boyle, M.B. (1981) Simultaneous optical monitoring of activity of many neurons in invertebrate ganglia using a 124-element photodiode array. J. Neurophysiol. 45: 829-840. Grinvald, A., Hildesheim, R., Farber, I.C. and Anglister, L. (1982) Improved fluorescent probes for the measurement of rapid changes in membrane potential. Biophys. J., 39: 301-308. Grinvald, A., Fine, A., Farber, I.C. and Hildesheim, R. (1983) Fluorescence monitoring of electrical responses from small neurons and their processes. Biophys. J., 42: 195-198. Grinvatd, A., Frostig, R.D., Lieke, E. and Hildesheim, R. (1988) Optical imaging of neuronal activity. Physiol. Rev., 68: 1285-1366. Higgins, D., Iacovitti, L. and Burton, H. (1984) Fetal rat sympathetic neurons maintained in a serum-free medium retain induced cholinergic characteristics. Dev. Brain Res., 14: 71-82. Horowitz, P. and Hill, W. (1989) The Art of Electronics, 2nd edn., Cambridge University Press, New York. Kaczmarek, L.K., Finbow, M., Revel, J.P. and Strumwasser, F. (1979) The morphology and coupling of Aplysia bag cells within the abdominal ganglion and in cell culture. J. Neurobiol., 10: 535-550. Kleinfeld, D., Raccuia-Behling, F. and Chiel, H.J. (1990) Circuits constructed from identified Aplysia neurons exhibit multiple patterns of persistent activity. Biophys. J., 57: 697-715.
105 Matsumoto, G. and Ichikawa, M. (1990) Optical system for real-time imaging of electrical activity with a 128x128 photopixel array. Soc. Neurosci. Abstr., 16: 490, abstr. #212.1. Montana, V., Farkas, D.L. and Loew, L.M. (1989) Dual-wavelength ratiometric fluorescence measurements of membrane potential. Biochemistry, 28: 4536-4539. Parsons, T.D., Kleinfeld, D., Raccuia-Behling, F. and Salzberg, B.M. (1989) Optical recording of the electrical activity of synaptically interacting Aplysia neurons in culture using potentiometric probes. Biophys. J., 56: 213-221. Potter, D.D., Landis, S.C., Matsumoto, S.G. and Furshpan, E.J. (1986) Synaptic functions in rat sympathetic neurons in microcultures. II. Adrenergic/cholinergic dual status and plasticity. J. Neurosci., 6: 1080-1098. Ready, D.F. and Nicholls, J. (1979) Identified neurones isolated from leech CNS make selective connections in culture. Nature, 281: 67-69.
Regehr, W.G., Pine, J., Cohan, C.S., Mischke, M.D. and Tank, D.W. (1989) Sealing cultured neurons to embedded dish electrodes facilitates long-term stimulation and recording. J. Neurosci. Methods, 30: 91-106. Salzberg, B.M. (1983) Optical recording of electrical activity in neurons using molecular probes. In J. Barker and J. McKelvy (Eds.), Current Methods in Cellular Neurobiology, Wiley, New York. pp. 139-187. Sigworth, F.J. (1983) Electronic design of the patch clamp. In B. Sakmann and E. Neher (Eds.), Single-Channel Recording, Plenum, New York, pp. 3-35. Syed, N.I., Bulloch, A.G.M. and Lukowiak, K. (1990) In vitro reconstruction of the respiratory central pattern generator of the mollusk Lymnaea. Science, 250: 282-285. Tank, D.W. and Ahmed, Z. (1985) Multiple-site monitoring of activity in cultured neurons. Biophys. J., 47: 476a (abstr.).
93
© 1991 Elsevier Science Publishers B.V. 0165-0270/91/$03.50 NSM 01241
An apparatus for recording synaptic potentials from neuronal cultures using voltage-sensitive fluorescent dyes Chi-Bin Chien and J e r o m e Pine Department of Physics, California Institute of Technology, Pasadena, CA 91125 (U.S.A.) (Received 21 January 1991) (Accepted 4 March 1991)
Key words: Voltage-sensitive dye recording; Optical recording; Synaptic potentials; Rat superior cervical ganglion; Microcultures; In vitro networks Voltage-sensitive dyes offer the promise of noninvasive multiceli recording of electrical activity, and should therefore be useful for studying the synaptic interactions of small networks of cultured neurons. We have designed and built a system for recording from microcultures of 1-15 neurons from the rat superior cervical ganglion (SCG), using voltage-sensitive fluorescent dyes of the styryl class. The apparatus comprises a standard inverted epifluorescence microscope; a mercury arc lamp with an optical feedback regulator; a 256-pixel fiber-optic camera with individual photodiode detectors and very low-noise amplifiers; and a personal computer-based data acquisition system. Its dark noise and illumination fluctuations are low enough that at typical fluorescence levels for these cells, it is limited by shot noise (the inherent physical limit of detection). Recording from SCG neurons, the signal-to-noise ratio is high enough to see large subthreshold synaptic potentials without signal averaging. This apparatus should be useful for studying long-term synaptic plasticity in cultures of vertebrate neurons, and several of its features should apply to optical recording from other preparations.
Introduction
This paper, along with a related one (Chien and Pine, 1991), describes the application of voltage-sensitive dye recording to microcultures of neurons from the rat superior cervical ganglion (SCG). Microcultures - isolated networks of a few neurons, grown in cell culture - are good preparations for studying synaptic interactions (Furshpan et al., 1976; Kaczmarek et al., 1979; Ready and Nicholls, 1979) and properties of biological neural networks (Parsons et al., 1989; Kleinfeld et al., 1990; Syed et al., 1990) because
Correspondence: Dr. C.-B. Chien, Dept. of Biology B022, University of California at San Diego, La Jolla, CA 920930322, U.S.A. Tel.: (619) 534-2526; Fax: (619) 534-8866.
they have few cells, often have a high proportion of synaptic connections, and are easily manipulated. Cell culture also makes it possible to follow development of synaptic connections over long times (Potter et al., 1986). Dye recording (for reviews, see Salzberg, 1983; Cohen and Lesher, 1986; Grinvald et al., 1988) is particularly appropriate for microcultures of vertebrate neurons because it can give a complete record of the activity of every neuron in the network and because it is relatively noninvasive, allowing repeated measurements from the same culture. (Using intracellular electrodes, in contrast, it is difficult to record from more than two or three cells at once, and even harder to repeatedly penetrate the same cell.) Our eventual goal is to combine dye recording with extracellular stimulation using integrated multielectrode dishes (Re-
94 gehr et al., 1989), in order to measure all the synaptic strengths in a microculture without any intracellular penetrations. Dye recording is technically demanding for any preparation; the main challenge is to obtain an adequate signal-to-noise ratio. Our apparatus is optimized for SCG neurons, which we chose because they have previously been grown in microculture, where they form fast cholinergic synapses with each other (Higgins et al., 1984; Furshpan et al., 1986). As with all vertebrate neurons, these cells are relatively small (20-30 txm in diameter) and so give very low detected fluorescence. This apparatus can operate at the shot-noise limit, even at these low light levels, because of several straightforward innovations, including an optical feedback regulator circuit for stabilizing the mercury arc lamp, and very quiet preamplifiers for the detector array. Using the fluorescent styryl dye RH423, a close analog of RH421 (Grinvald et al., 1983), we can record from all the neurons in an SCG microculture at a signal-to-noise ratio sufficient to see large postsynaptic potentials without signal averaging. This recording setup can be used with other cultures of vertebrate neurons (mass cultures as well as microcultures), and its key features may be adapted to other systems. This paper deals only with the construction and performance of the apparatus itself, emphasizing design decisions and technical features that will be useful for other systems. Results of dye recording experiments on SCG microcultures are described in Chien and Pine (1991). Parts of this work have been described in abstract form (Chien et al., 1987), and an earlier version of the optical feedback regulator circuit has been mentioned in a review (Cohen and Lesher, 1986). Further details of the apparatus and its design are available (Chien, 1990).
System overview The apparatus (Fig. 1) is built around an Olympus IMT-2 inverted microscope, with an incandescent lamp for phase contrast and standard attachments for epifluorescence: mercury arc
mercury I ( ~ c u l t u r e ~ arclampshu'tel ~
pus,M 2
22;o'12;'i
sideport (tofiber-opticcamera) Fig.1.Schematicdrawingofthedye-recordingapparatus.
lamp, epifluorescence illuminator, and filter cube. An electromechanical shutter controls the fluorescence illumination; it is mechanically isolated by a heavy support and two layers of foam rubber. Dye signals are detected by a 256-pixel fiber-optic camera detector (not shown here; see Fig. 4) that fits into the side port of the microscope; a 35-mm camera takes photographs; and a video camera provides an image for aligning the culture with the pixels of the detector. For electrical and vibrational isolation, the apparatus sits inside a Faraday cage and on top of a heavy marble table supported by four inner tubes. In designing a dye-recording system, photons are at a premium. The system should maximize the signal-to-noise ratio, while at the same time minimizing bleaching and phototoxicity. This requires (1) efficient optics, with a high illumination intensity and a high efficiency for collecting fluoresced photons (the latter is especially important); (2) a dye that gives bright fluorescence and good voltage sensitivity; and (3) low instrumental noise, stable illumination and quiet photodetectors. Optical efficiency is achieved by using a good commercial microscope and an objective lens with high numerical aperture (NA). The NA determines the illumination intensity and the collection efficiency for fluorescence, which are both roughly proportional to the square of the NA (see
95 Grinvald et al., 1983; however, the exact proportionality is obtained only under certain ideal conditions). In general, maximizing NA requires high objective magnification (M). (Strictly speaking, this is only true for conventional microscope optics: one could, in theory, design a microscope with a super-large-diameter objective that would give high NA at low M.) However, M cannot be increased without bound, since the field of view is limited to the tube diameter (about 20 mm) divided by M. In order to encompass a typical microculture, 750 /zm in diameter, we use a dry 20 x objective (Fluor 20 Ph3DL, Nikon); its NA is 0.75. For small microcultures we sometimes use a 40 x lens (Fluor 40 Ph3DL, Nikon, NA 0.85). To reduce bleaching and phototoxicity, it is important to illuminate the specimen only when actually recording. The electromechanical shutter (No. 1 Synchro, Ilex Optical Co., Rochester, NY) turns illumination on and off; it opens and closes in a few milliseconds and is usually set for flashes lasting 100-250 ms. The sensitivity of dye recording depends greatly on the particular dye used. With SCG cultures, we found the best results when staining with micromolar concentrations of the styryl dye RH423 (a kind gift of Dr. Amiram Grinvald) for 5-10 min; this yielded a typical detected fluorescence photocurrent of 1 nA per cell, and a typical dye sensitivity of 1% fluorescence change per 100 mV change in membrane potential. Culture methods, staining procedures, and the dye's structure are described.in detail elsewhere (Chien and Pine, 1991). For fluorescence, we use the Olympus " G " filter cube with an extra LP515 excitation filter. This combination passes exciting light from 520550 nm (the excitation is dominated by the 546 nm mercury line) and emitted light longer than 610 nm. Using these filters with the standard IMT-2 epi-illuminator and the 20 x ,/NA 0.75 objective, the illumination intensity at the specimen is 10 W / c m 2. With this illumination, the detected fluorescence from a typical stained SCG neuron is 6 x 109 photons/s, which gives the 1 nA photocurrent quoted above. Photodynamic damage at this intensity is tolerable: SCG neurons illuminated for at least 100 flashes of 100 ms
each do not show detectable deterioration of their action potentials. The sources of noise in a dye-recording system can be classified by the power of their dependence on the detected intensity I. Illumination noise is proportional to I, and is usually due to intensity fluctuations in the light source. Shot noise, proportional to x/I, is due to statistical fluctuations in the detected light. Dark noise, independent of I, is due to noise in the detection system's photodetectors or amplifiers. Since these noise sources are uncorrelated, they combine in quadrature: (itotal) 2 = (/shot) 2 + (iill) 2 + (idark) 2 (A lowercase i denotes a root-mean-square [rms] noise amplitude.) What we refer to as illumination noise has also been called extraneous noise (Cohen and Lesher, 1986) and technical noise (Parsons et al., 1989). For a good discussion of noise in dye recording, see the review by Cohen and Lesher (1986). A standard procedure in noise design is to identify the noise source that is hardest to eliminate, and then attempt to bring all other sources below this level. Shot noise is a fundamental physical limit for photodetection: it comes from the quantal nature of light, and cannot be eliminated. Therefore, when /ill and /dark are less than about half of/shot, the signal-to-noise ratio will be optimal; further reduction will not significantly decrease the total noise. The present apparatus is nearly shot-noise limited: /ill and /dark are somewhat less than /shot, though not always as low as ishot/2. The shot noise in a photocurrent I is given by isho, = ~/(2q/B)
(1)
where q = 1.6 x 10 -19 C is the charge of the electron and B is the detection bandwidth (Horowitz and Hill, 1989). B must be at least 300 Hz to see action potentials (about 1 ms long) with minimal distortion. The rms shot noise in I = 1 nA is then 0.3 pA, or 0.03%; with RH423 dye signals of 1%/100 mV, this corresponds to a 3 mV change in membrane potential. Therefore, in shot-noise-limited recordings at these levels of
96
fluorescence, it should be possible to see large synaptic potentials in a single trial. (Noise may be reduced further with judicious signal averaging.) The next three sections describe the key parts of the apparatus: the optical-feedback regulator (which reduces illumination noise), the fiber-optic camera detector (which has low dark noise), and the PC-based data acquisition system.
4 5 A from power supply
I
(
~VV~-.-
Harnamats~
s,os,.o,
i ,0o~.'2 t~') ,J v'-~
2N6388
Feedback-regulated light source
7 7 ,,,1~ ~
~
°~'° !~
, .... >.-
-
1'- --- +1>~
~
]-"--$-<,~F356
r ~""
I
047 ,uF ~'-
T :.~EG.. /
~
i
<"
_
/
Detecting a reasonably large fluorescence from small cells requires a bright light source; we find a commercial mercury arc lamp (HBO 100W/2, Osram) to be suitable. Mercury arcs, though bright, are notoriously unstable (Fig. 3a,c), so we use an optical-feedback regulator circuit to reduce illumination noise. A photodiode mounted in the side of the lamp housing (see Fig. 1) monitors the lamp intensity, and the regulator circuit modulates the current supplied to the arc so as to keep this intensity steady.
Apparatus The arc lamp is driven by a standard power supply (Olympus BH2-RFL-T2) with an integral radiofrequency (RF) igniter; it typically operates at 4.5 A and 22 V. We originally used a currentregulated DC supply with a spark igniter (as in Grinvald et al., 1982), which had lower linefrequency ripple, but the spark igniter had a distressing tendency to damage other electronics. The Olympus unit's RF igniter does not have this problem, and its power supply's line-frequency ripple is suppressed by the feedback regulator. Fig. 2 shows the regulator circuit. The AC component of the photocurrent from the lamphouse photodiode is amplified and used to drive a power transistor that shunts the lamp. When a fluctuation increases the lamp intensity, the transistor shunts more current and reduces the intensity: negative feedback. The standing current through the transistor determines the maximum output swing of the circuit. During ignition, a switch is used to disconnect the regulator and avoid shortening the ignitor. The R E G output is proportional to the lamphouse intensity; the op-
Fig. 2. Optical-feedbackregulator circuit. The output of the lamphouse photodiode (Hamamatsu S1087-01) is amplified and drives a shunt regulator: when the lamp intensity increases, the 2N6388 Darlington power transistor in parallel with the lamp shunts more current. The x 230 amplifier is a combination of LF356 op-amps with a gain of 230. The circuit's passband is 3-300 Hz and its gain can be adjusted from zero to full (using the 1 kg2 potentiometer) in order to check the effects of regulation. The REG output provides a monitor of intensityfluctuations.
eration of the regulator can be checked by watching R E G while varying the regulator's gain from zero to full, using the 1-kO potentiometer. The bandwidth of the regulator is 3-300 Hz (3 dB down frequencies). Some care has been taken to avoid unnecessary phase shifts; "motorboating" (oscillations at very low frequencies) was originally quite troublesome. Those fluctuations falling within this bandwidth should be suppressed by a factor equal to the closed-loop gain. The DC photocurrent of the lamphouse photodiode is typically 0.7 mA. Making the approximation that this photocurrent is proportional to the lamp intensity, which is in turn proportional to the lamp current, the closed-loop gain is [(0.7 mA)/(4.5 A)][(7.7 kO)(230X15)/(15 g2)] -- 280.
Performance The mercury arc lamp has two sorts of noise, fast and slow. The fast fluctuations are mostly line-frequency ripple; with the Olympus power supply, their amplitude is 1.5% peak-to-peak, 0.5% rms. This ripple can obscure action potentials ( ~ 1 ms timescale) or fast PSPs ( ~ 10 ms). The slow fluctuations, presumably due to arc
97
a) unregulated
b) regulated
'amp house
10'%
stage ~J~j~JW~AJ~,J~/,J/I~%
~ . ~
~ J ~"~ I 0.1%
20 ms
c) unregulated
lamp- --".-J'~k_ 11% house t/'-"-stage --,~k._q~/---'--- ]1% I
d) regulated
I'% ..~.~"-'~
II %
I
ls Fig. 3. Regulation of lamp fluctuations, measured at the lamphouse and at the specimen stage. The regulator circuit's gain was zero for the unregulated traces and full for the regulated traces, a and b show fast fluctuations (filtered with passband 0.16-320 Hz), mostly at the line frequency; c and d show slow fluctuations (passband 0.16-5.3 Hz). In each pair, the lamphouse and stage fluctuations were measured simultaneously.
shot noise of 0.03% (see above System Overview), but it could presumably be reduced to 0.002% by replacing the lamphouse photodiode with an intensity monitor that directly sampled the light going to the specimen. Fig. 3c and d show the regulation of slow fluctuations. These are also anisotropic: note the differences between the lamphouse and stage traces in Fig. 3c. Though the slow fluctuations seen at the lamphouse photodiode are heavily suppressed by the regulator, those at the stage are still present. (Again, better placement of the intensity monitor would improve regulation.) These slow fluctuations are inconvenient, but can be dealt with in software by rejecting those data sets that show signs of an arc wander: a sudden shift or slope change in the dye signal's baseline, simultaneous across pixels.
Optical detector and electronics wander (movement of the mercury arc), consist of occasional spikes in intensity that rise in tens of milliseconds and decay in hundreds of milliseconds, with amplitudes ranging from below 0.1% to above 3%. Bulbs go through noisy and quiet periods that last for minutes, changing unpredictably from quiet to noisy and back. (They seem to become noisier with age.) During a typical quiet period, spikes exceeding 0.1% occur at 0.l/s; during a typical noisy period, 0.1% spikes occur at 2/s, with half exceeding 0.5%. These spikes are too slow to obscure action potentials, but cause a sloping baseline in the dye signal that can make it impossible to quantitate small potential changes. Comparison of Fig. 3a and b shows the regulation of fast fluctuations. The rms noise measured by the lamphouse photodiode (located at right angles to the output light path of the lamp) drops from 0.5% to 0.002%, nearly the 280-fold improvement predicted above. However, since the lamp's fluctuations are not isotropic, it is important to measure the illumination intensity at the stage where the specimen is. A photodiode put in place of the specimen measures a smaller improvement: the rms noise drops from 0.6% to 0.02%. This noise is already below the typical
The optical detector, shown schematically in Fig. 4a, has two parts: a primary photodetector array, and amplifying electronics. The primary detector is a fiber-optic camera (Fig. 4c), which comprises a bundle of 256 optical fibers connected to 256 discrete photodiodes. A real image of the culture is focussed onto the face of the bundle (Fig. 4b); each fiber picks up the light from one pixel and conducts its light to its own photodiode. The photodiode transduces this light into a photocurrent, which is carried by a coaxial cable to the amplifying electronics. There the photocurrent is converted to a voltage by a lownoise transimpedance amplifier, multiplexed with signals from other pixels, and sent to the data acquisition system for digitization. Fiber-optic camera
The fiber bundle of the fiber-optic camera fits within the tube diameter of the IMT-2 (roughly 20 mm). The number of fibers matches the geometry of SCG microcultures. A typical cell is 30 /xm in diameter, roughly 1/20th the diameter of a culture. In order to have at least one pixel per cell, the square plastic fibers (Poly-Optical, Santa Ana, CA) are arranged in an octagonal bundle 18
98 fiber bundle (end view)
multiplexer
fiber-optic camera
preamplifiers
a
coaxial cables
currents, while dye-recording amplifiers have more stringent signal-to-noise requirements and must be produced by the hundreds. A fundamental limit for the noise of a transimpedance amplifier is the Johnson noise (noise due to thermal fluctuations) in its feedback resistor Rro. All amplifier noise sources are equivalent to either an input c u r r e n t /in or an input voltage vm; for instance, the Johnson noise in Rrb is equivalent to a current noise with rms amplitude
V/(4kTB/Reo) (2) where kT is the product of Boltzmann's constant /Job. . . . =
2,
b Fig. 4. a: schematic view of the 256-pixel optical detector. Only eight of the fibers and diodes within the fiber-optic camera are shown, b: face view of the fiber bundle. Scale in cm. c: oblique view of the fiber-optic camera.
fibers across. The average spacing is 0.9 mm, though nonuniformities in the fibers make the array slightly irregular. With a 20 × objective, the detector has an 800 ~m field of view with 45/zm pixels. The dead space between fibers (due to cladding and adhesive) is about 20%. We have tested the fibers' transmission with a light-emitting diode source and found a variation of 20% (SD) between fibers; three fibers with very low transmittance were probably nicked or poorly joined to their photodiodes. The $1087-01 (Hamamatsu) photodiodes were chosen for low noise, good quantum efficiency, and moderate capacitance.
Preamplifier design The preamplifier circuit (Fig. 5) is designed to minimize noise, given the desired bandwidth (300 Hz) and the expected range of photocurrents (0.2-2.0 nA). It has much in common with patch-clamp preamplifiers (Sigworth, 1983): both are very low-noise, low-current transimpedance amplifiers. The main differences are that patchclamp amplifiers are faster and amplify smaller
and the absolute temperature of the resistor (Horowitz and Hill, 1989). The relative contributions of ii. and el. depend on the input load Zin. The total amplifier noise referred to the input, /amp, is given by (iamp) 2 =
(ii.) 2 + ( v i . / Z i . ) z
If the amplifier is sufficiently quiet, i Job.son will dominate /in, which in turn will dominate /amp, and the amplifier's noise will be proportional to 1/((Reb) • The larger the feedback resistor, the quieter the amplifier may be; however, if Reo is too large the amplifier will saturate at very low light levels and stray capacitances will become a problem. This preamplifier comprises an OP27 operational amplifier preceded by a low-noise input stage. The op-amp in the second stage has moderately low noise, and provides very high gain; the Rfb(5 6 G~.,~) .
+11,
I I I
[
510 ~2 5%
51~ 5%
muhiple~:cl
L
Fig. 5. Preamplifier circuit for the optical detector. A JFET input stage and an OP27 form a high-input impedance inverting amplifier. Together with the feedback resistor Rt~, they form a low-noise, low-input-current transimpedance amplifier. Two RC filters prevent aliasing.
99
JFET (junction field-effect transistor) input stage has very low noise and provides moderate amplification, so that noise from the second stage is unimportant. The bipolar transistors (2N3704) set the drain-source voltage of the FETs (2SK240GR, Toshiba), and prevent second-stage noise from coupling back to the amplifier input via the FET gate-drain capacitance. We use feedback resistors in the range 5-6 GO (Victoreen), large enough that a very bright cell (2 nA photocurrent) will just saturate the amplifiers. A quantitative analysis of the circuit (details omitted) shows that /Johnson should dominate i~n, while the FETs' voltage noise should dominate vi,. Zi, is determined by the capacitances of the photodiode, coaxial cable, and FET, which sum to 300 pF. At the working frequencies of the amplifier, Zi, is large enough that /amp should be well approximated by the Johnson noise of Rn,. For a resistance of 6 GO and a bandwidth of 300 Hz, theory predicts /amp = 28 fm rms.
Performance In practice, the measured dark noise of the preamplifiers is somewhat higher than that predicted from Johnson noise, for several reasons: most have bandwidths greater than 300 Hz, there is line-frequency noise, and some amplifiers have excess noise of undetermined origin. A power spectrum of the quietest amplifier (not shown) shows it to be right at the Johnson noise limit, but the amplifiers' average dark noise is 78 + 13 f A r m s (mean + SD). This deviation from theory is not too distressing, because this dark noise is still well below shot noise at typical fluorescence levels; for instance, at I = 1 nA it is only 0.008%, against a shot noise of 0.03%. A useful figure of merit is /as, the photocurrent at which amplifier noise is equal to shot noise. For an amplifier limited by the Johnson noise of its feedback resistor, combining Eqns. 1 and 2 yields a simple rule: Rfb Ias = 50 mV. If our preamplifiers were Johnson-noise limited they would have Ia, = 10 pA, but because of excess noise they have I~s = 50 pA on average. For photocurrents greater than 50 pA, shot noise will be greater than dark noise. The multigigohm resistors used here are very
susceptible to small stray capacitances, which affect their frequency responses. Capacitances in parallel with the resistors cause the preamplifier bandwidths to vary from 150 to 650 Hz (they are mostly around 400 Hz); capacitances to ground cause gain peaking in some amplifiers; and capacitances between certain amplifiers cause crosstalk. Fortunately, these effects are all small and do not introduce spurious signals, though they can alter signal sizes somewhat. (For quantitative measurements, signals can be checked for contributions from crosstalk, and corrected in software for their amplifiers' bandwidth and gain peaking.)
Data acquisition hardware and software
Hardware The data acquisition system (Fig. 6) is built around an 8 MHz IBM AT personal computer (PC) and a TransEra 7000-MDAS modular data acquisition system (TransEra, Provo, UT; now manufactured by Acurex, Mountain View, CA). The preamplifiers are arranged in groups of 16 on 16 printed-circuit cards. Each card has two CD4051 multiplexer chips that combine its preamplifier outputs into a card output, which is sent to a controller card; two more CD4051s on the controller card combine the card outputs into the signal that is sent to the MDAS for digitization. The MDAS has a 12-bit analog-to-digital converter (ADC) and its own 68000 microprocessor, and talks to the PC over an IEEE-488 bus. It is an elegant device because it relieves the PC of addresses
shutter frigger
camera
IEEE-488 bus
Fig. 6. Data acquisition system. An I B M A T personal computer controls an autonomous data acquisition system, which sends pixel addresses to the multiplexers and digitizes their
output. The PC also acquires images from the video camera.
100 direct responsibility for data acquisition: the PC need only send it simple ASCII commands, whereupon it triggers the shutter, digitizes several thousand fluorescence values, and returns the data to the PC. The MDAS uses custom software written by TransEra to address the multiplexers and acquire data simultaneously. This subroutine can sample data at up to 90 kHz, i.e., 45 pixels at 2 kHz/pixel. Even allowing 2 or 3 pixels per cell for cells that fall on pixel boundaries, this is enough to record from all the cells in a large microculture (at least 15 cells). Because the analog-to-digital converter is bipolar, only 11 of its bits are useful; however, its dynamic range is extended by a power-of-2 programmable amplifier. The rms digitization noise varies between 0.014% and 0.028%, depending on the exact light level; it is reduced by offline digital filtering (Fig. 8e).
Software The data acquisition and analysis software for the PC consists of four parts: an alignment program OPTIK, a data acquisition program MULTI; a graphing and analysis program DISPLAY; and about 20 small analysis programs that average, filter, bleach-subtract, Fourier-transform, etc. These are variously written in Microsoft QuickBASIC ® and Lattice C ®, with a few subroutines in Assembler for speed. The source code, about 800 kbytes long, is available from the authors on floppy disk. When an experiment is first set up, O P T I K uses a frame-grabber board to acquire a video image of the culture, align it with a stored representation of the pixel array, and choose pixels for recording. M U L T I then runs the experiment: it takes electrophysiological and fluorescence data, displays them on-screen in real time, and does simple data analysis such as signal averaging. Data files are stored on magnetic disks and analyzed afterwards using DISPLAY and the suite of analysis programs. At present, MULTI is arbitrarily limited to recording from 16 channels at once. The MDAS can record up to 30 000 points from each channel, while available memory in the AT is limited to about 200000 points total from all channels.
Fig. 7 shows an example of the pixel alignment procedure for a 5-cell microculture. The phasecontrast photograph a shows the 5 cell bodies, the dense mat of processes, and the thick fascicle that girdles the culture. The video image b shows the culture, with a superimposed pattern of squares showing pixel locations. Dye signals were recorded from the pixels shown as white squares, while the dark squares (not all easily visible here) are unselected pixels. (The unevenness of the squares r e f e c t s the irregularities in the optical fiber bundle.) The phase-contrast photo serves as a record of the detailed anatomy of the culture, while the stored video image merely records the positions of the pixels used. Tracing over the video image yields the culture scheme c, used for easy reference during data analysis. Fig. 8 shows the steps used to extract a dye signal from the raw fluorescence data; they are a simplified version of the procedure described by Grinvald et al. (1982). An SCG neuron in mass culture was stained with RH423, then penetrated intracellularly and stimulated while recording its fluorescence. Trace f shows the intracellular action potential. In the raw fluorescence trace a, the dye signal is dwarfed by the initial step from 0 to 0.86 nA (the cell's resting fluorescence) when the shutter opens. Trace b is an amplified version of the signal, calibrated as a fractional change in fluorescence; it has been inverted so that an upward deflection corresponds to a depolarization. The action potential is clearly visible above the noise (the digitization noise is quite obvious). The rising baseline is an artifact due to photobleaching of the dye. To remove this bleach artifact, a pure bleach trace is recorded without stimulation (c, jagged trace). A quadratic curve fitted to this trace (c, smooth trace) is subtracted from b to yield the bleach-subtracted trace d. Finally, filtering the bleach-subtracted trace at 300 Hz yields e, which is very similar to the intracellular recording f. The peak-to-peak noise in Fig. 8e is about 0.07%; since the dye sensitivity for this cell was 0.92%/100 mV, this translates to 7.5 mV. This signal-to-noise ratio is sufficient to see a large EPSP (it might be desirable to average a few trials). Fig. 9 shows an example of an optically-re-
101 shutter opens a) raw
05 n
b) x-100 c) bleach & fitted bleach
AF/F [ 0-5% d) b]eachsubtracted e) filtered
50 mV
-
-
f) intracellular
20 ms Fig. 8. Dye signal from an SCG neuron, showing standard data reduction. This data is from a single, unaveraged trial, a: raw fluorescence signal; the dye signal is a small decrease at the limit of visibility. The arrow indicates the shutter opening. b: amplified, inverted fluorescence signal (gain = - 100). c: fluorescence signal with no stimulus, showing the bleach artifact only. The jagged trace is the actual bleach; the smooth trace is a quadratic fit. d: bleach-subtracted trace (b minus the fit from c). e: bleach-subtracted trace, filtered digitally at 300 Hz. f: intracellutar recording. The cell was stimulated through a bridge circuit with a square current pulse. Scale bars: 0.5 nA for a, 0.5% for b, c, d, e (arrow points towards increasing fluorescence); 50 mV for f.
ol
corded PSP, with simultaneous intracellular r e c o r d i n g s f o r c o m p a r i s o n . I n a five-cell m i c r o c u l t u r e , cells 1 a n d 3 w e r e p e n e t r a t e d i n t r a c e l l u larly w h i l e all five cells w e r e m o n i t o r e d o p t i c a l l y ; s t i m u l a t i o n o f cell 1 r e s u l t e d in a l a r g e P S P in cell 3 w i t h t w o c o m p o n e n t s , t h e first a p p a r e n t l y a m o n o s y n a p t i c c o n n e c t i o n f r o m cell 1, a n d t h e s e c o n d a d i s y n a p t i c c o n n e c t i o n t h r o u g h cell 2
I
I
100 pm
Fig. 7. Example of pixel alignment for a 5-cell microculture, a: phase-contrast. The bright object in the center is a piece of debris, b: stored video alignment image. White squares are selected pixels; dark squares are unused pixels, c: culture schematic, made by tracing b. Scale bar = 100 p.m for all 3 panels.
102
a (av 3 trials)
intra
b (single trial)
J
/
50 mV
optical
50 m s
Fig. 9. An excitatorypostsynapticpotential in an SCG neuron, measured intracellularly(intra) and with dye recording from a pixel over the cell body (optical). The detected photocurrent from this pixel was 0.77 nA. a: subthreshold PSP (average of 3 trials), b: suprathreshold PSP, with resulting action potential (single trial). Scale bars apply to both panels.
(data not shown). Fig. 9a shows an average of three trials in which the compound PSP was subthreshold; Fig. 9b shows a single trial in which it was suprathreshold. The signal-to-noise ratio of the dye signal is clearly sufficient to detect the PSP and make a quantitative estimate of its amplitude relative to that of the action potential.
Discussion Dye recording from cultured vertebrate neurons is difficult because the cells are small and so give little fluorescence. With the 10 W / c m a illumination and NA 0.75 objective used here, detected fluorescences from SCG neurons are only 0.2-2 nA; these are difficult to increase. There are 3 ways to try: a higher NA objective, more intense illumination, or heavier staining. The NA could be improved somewhat by switching to a 32 x oil-immersion objective (if a compatible lens could be found); the field of view would still be tolerable, and the detected fluorescence would be roughly doubled. Brighter illumination would probably require using a laser light source, and
seems likely to cause an undesirable level of phototoxicity. The cells are already very heavily stained: at a rough estimate (Chien and Pine, 1991), there is 1 dye molecule for every 100 phospholipids in the cell membrane. In view of these difficulties, it will be difficult to increase detected fluorescences from this preparation by a large factor. Given these fluorescence levels and a 1 % / 1 0 0 mV voltage sensitivity (the best that we have been able to achieve after strenuous attempts at optimization), it is just possible to detect synaptic potentials optically. Taking 1 nA as a typical detected photocurrent and a bandwidth of 300 Hz, shot noise is 0.03% rms. Our apparatus achieves its design goal of being shot-noise limited: amplifier dark noise is 0.008%, digitization noise is 0.014-0.028%, and fast illumination noise is 0.02%. (Trials are sometimes spoiled by slow illumination fluctuations, but these can be rejected.) These calculations predict that large synaptic potentials should be visible above shot noise, and Fig. 9 confirms this. Below, we discuss the parts of the apparatus in turn, comparing them to previous setups and mentioning possible improvements. Light source It is possible to compensate for illumination fluctuations in software by monitoring the illumination intensity and using it to correct the detected fluorescence signal, but it is preferable to use a regulator to eliminate these fluctuations at the source. With the Olympus power supply, the optical feedback regulator somewhat reduces slow fluctuations due to arc wander, and reduces linefrequency ripple at the specimen from 0.5% to 0.02% rms. This is significantly lower than the 0.06% rms we found when using a DC current supply (Grinvald et al., 1982) without the feedback regulator. The fluctuations are not spatially uniform: both slow and fast fluctuations differ between the lamphouse photodiode and the stage. In fact, the fluctuations at the lamphouse diode are roughly tenfold lower than those at the stage, suggesting that regulation could be improved 10fold by replacing the photodiode mounted in the lamphouse with another that directly sampled the
103 light going to the stage, using a partially reflecting mirror. Even with a properly placed intensity monitor, the present regulator can only compensate for fluctuations that are spatially uniform across the specimen. Fluctuations due to movement of the mercury arc (arc wander) are often nonuniform: using the fiber-optic camera to record reflected illumination light from across the specimen, we found (data not shown) that roughly one-third of the fluctuations were clearly nonuniform. It may be possible to eliminate spatial nonuniformity by passing the illumination through an optical-fiber scrambler. Alternatively, nonuniform illumination could be corrected by a dual-wavelength ratioing scheme (Montana et al., 1989).
Optical detector Previous multipixel detectors for dye recording (e.g., Grinvald et al., 1981) have used monolithic photodiode arrays, though Tank and Ahmed (1985) built a fiber-optic camera similar to ours. When we built our detector, the largest commercially available photodiode array was the Centronic MD-144, with 12 x 12 pixels. The main reason for using the fiber-optic camera instead was our desire for more pixels. Discrete photodiodes also have other advantages: the wide variety available lets one pick diodes with especially good properties (low noise and capacitance), and bad or damaged diodes may be replaced. It is relatively straightforward to scale up the fiber-optic camera to a larger number of pixels, and the cost per pixel should be constant, so a fiber-optic camera is a reasonable choice for a large detector. However, the technology for large monolithic detectors is likely to improve rapidly; Matsumoto and Ichikawa (1990) have recently reported using a 128 x 128 pixel array. The dark noise of our preamplifiers is such that our experiments are shot-noise-limited for photocurrents greater than 50 pA. Though preamplifier designs for dye recording have improved (e.g., Grinvald et al., 1988, section IV.B.2; Bonhoeffer and Staiger, 1988), we are not aware of published circuit details or noise data by which to compare them to our design. The amplifiers are in fact quieter than is necessary for our
experiments: 50 pA is too dim to record synaptic potentials (the shot noise is too large). However, they would be useful for experiments that only need to detect action potentials, where it would be advantageous to operate at low levels of illumination in order to reduce phototoxicity. Two improvements would be useful for the next generation of amplifiers: reducing the feedback resistor to 1 GO, and improving the printed-circuit layout. Smaller feedback resistors would ameliorate stray-capacitance problems and (because of the present excess noise) should not appreciably increase amplifier noise. A combination of lower Rro with better layout should eliminate crosstalk and improve amplifier frequency responses. The low noise and gigohm feedback resistors of our preamplifiers are specifically designed for the low light levels of fluorescence, and they cannot be used for recording with absorption dyes, as they would be saturated by the much higher light levels used there. Simpler amplifiers with megohm feedback resistors would suffice for absorption recording.
Data acquisition Most dye-recording systems have used PDP-11 minicomputers for data acquisition (Cohen and Lesher, 1986, section 5.7; Grinvald et al., 1988, sections IV.B and IV.C), with software written in Assembler language. The rapid advances in the power of personal computers made it possible for us to use an 8 MHz IBM PC-AT, which has the advantages of cheapness and ease of programming. Our system only has a d]ta acquisition rate of 90000 samples/s, but since the cell bodies in SCG microcultures are sparse, about 30 pixels (2 pixels for each of 15 cells) suffice to record from all the cells in a culture, so this rate is sufficient. Much faster personal computers have become available since this system was built, and it is clear that a personal computer-based system could record from hundreds of pixels at a time.
Applications This apparatus was designed to detect subthreshold synaptic potentials with dye recording, and its sensitivity is sufficient for this purpose.
104 U n f o r t u n a t e l y , for S C G m i c r o c u l t u r e s the optic a l l y - r e c o r d e d P S P shown in Fig. 9 is t h e exception, not t h e rule. A s d e s c r i b e d in d e t a i l elsew h e r e ( C h i e n a n d Pine, 1991), S C G m i c r o c u l tures a r e a p o o r p r e p a r a t i o n for such e x p e r i m e n t s b e c a u s e o f t h e i r e x t r e m e l y c o m p l e x axonal arbors: dye signals from PSPs in t h e cell b o d i e s a r e usually o b s c u r e d by signals f r o m action p o t e n t i a l s in axons t h a t pass t h r o u g h the s a m e pixel. ( T h e e x a m p l e shown a b o v e is o n e of two cases w h e r e we w e r e a b l e to r e c o r d a c l e a r P S P u n o b s c u r e d by b a c k g r o u n d signals.) W e h o p e to resolve this p r o b l e m by switching to a d i f f e r e n t c u l t u r e system w h e r e axons a r e s h o r t e r a n d cell b o d i e s a r e spatially isolated. In such a system, c o m b i n i n g dye r e c o r d i n g with ext r a c e l l u l a r s t i m u l a t i o n by m u l t i e l e c t r o d e dishes ( R e g e h r et al., 1989), it should b e possible to m e a s u r e synaptic plasticity in m i c r o c u l t u r e s over days a n d weeks, a n d to study t h e effects o f c h r o n i c s t i m u l a t i o n at the single-cell level. T h e p r e s e n t system s h o u l d be widely a p p l i c a ble to f l u o r e s c e n c e dye r e c o r d i n g f r o m c u l t u r e s of v e r t e b r a t e n e u r o n s , r e c o r d i n g e i t h e r f r o m multiple cells, or from m u l t i p l e sites on the p r o c e s s e s of an individual cell. T h e p r e p a r a t i o n s for which the system is useful a r e mostly d e t e r m i n e d by its spatial r e s o l u t i o n (18 pixels across, e a c h pixel 45 ~ m with a 20 x objective, 2 2 / ~ m with 40 x ) a n d the f l u o r e s c e n c e levels for which it is d e s i g n e d ( 0 . 2 - 2 . 0 n A d e t e c t e d p h o t o c u r r e n t ) . Its key comp o n e n t s - the o p t i c a l f e e d b a c k r e g u l a t o r , fiberoptic c a m e r a , a n d J F E T p r e a m p l i f i e r design should be g e n e r a l l y useful for optical r e c o r d i n g systems.
Acknowledgements W e w o u l d like to t h a n k A m i r a m G r i n v a l d for his gift o f dyes a n d for g e n e r o u s l y supplying his e x p e r t i s e d u r i n g t h e initial p h a s e s o f this project; A n d r e w H s u a n d M i c h a e l W a l s h for t e c h n i c a l help in b u i l d i n g t h e optical d e t e c t o r ; a n d L a r r y C o h e n a n d J i a n - Y o u n g W u for useful c o m m e n t s on t h e m a n u s c r i p t . This w o r k was s u p p o r t e d by g r a n t s f r o m t h e N I H ( N S 2 2 4 5 0 - 0 2 ) , t h e System Development Foundation, and Sperry-Univac.
References Bonhoeffer, T. and Staiger, V. (1988) Optical recording with single cell resolution from monolayered slice cultures of rat hippocampus, Neurosci. Lett., 92: 259-264. Chien, C.-B., Crank, W.D. and Pine, J. (1987) Noninvasive techniques for measurement and long-term monitoring of synaptic connectivity in microcultures of sympathetic neurons. Soc. Neurosci. Abstr., 13: 1426, abstr. #393.15. Chien, C.-B. (1990) Voltage-sensitive dye recording from networks of cultured neurons. Ph.D. thesis, California Institute of Technology. Chien, C.-B. and Pine, J. (1991) Voltage-sensitive dye recording of action potentials and synaptic potentials from sympathetic microcultures. Biophys. J., in press. Cohen, L.B. and Lesher, S. (1986) Optical monitoring of membrane potential: Methods of multisite optical measurement. In DeWeer, P., and Salzberg, B.M. (Eds.), Optical Methods in Cell Physiology, Wiley (Interscience), New York, NY, pp. 71-99. Furshpan, E.J., MacLeish, P.R., O'Lague, P.H. and Potter, D.D. (1976) Chemical transmission between rat sympathetic neurons and cardiac myocytes developing in microcultures: evidence for cholinergic, adrenergic and dualfunction neurons. Proc. Natl. Acad. Sci. USA, 73: 42254229. Furshpan, E.J., Landis, S.C., Matsumoto, S.G. and Potter, D.D. (1986) Synaptic functions in rat sympathetic neurons in microcultures. I. Secretion of norepinephrine and acetylcholine. J. Neurosci., 6: 1061-1079. Grinvald, A., Cohen, L.B., Lesher, S. and Boyle, M.B. (1981) Simultaneous optical monitoring of activity of many neurons in invertebrate ganglia using a 124-element photodiode array. J. Neurophysiol. 45: 829-840. Grinvald, A., Hildesheim, R., Farber, I.C. and Anglister, L. (1982) Improved fluorescent probes for the measurement of rapid changes in membrane potential. Biophys. J., 39: 301-308. Grinvald, A., Fine, A., Farber, I.C. and Hildesheim, R. (1983) Fluorescence monitoring of electrical responses from small neurons and their processes. Biophys. J., 42: 195-198. Grinvatd, A., Frostig, R.D., Lieke, E. and Hildesheim, R. (1988) Optical imaging of neuronal activity. Physiol. Rev., 68: 1285-1366. Higgins, D., Iacovitti, L. and Burton, H. (1984) Fetal rat sympathetic neurons maintained in a serum-free medium retain induced cholinergic characteristics. Dev. Brain Res., 14: 71-82. Horowitz, P. and Hill, W. (1989) The Art of Electronics, 2nd edn., Cambridge University Press, New York. Kaczmarek, L.K., Finbow, M., Revel, J.P. and Strumwasser, F. (1979) The morphology and coupling of Aplysia bag cells within the abdominal ganglion and in cell culture. J. Neurobiol., 10: 535-550. Kleinfeld, D., Raccuia-Behling, F. and Chiel, H.J. (1990) Circuits constructed from identified Aplysia neurons exhibit multiple patterns of persistent activity. Biophys. J., 57: 697-715.
105 Matsumoto, G. and Ichikawa, M. (1990) Optical system for real-time imaging of electrical activity with a 128x128 photopixel array. Soc. Neurosci. Abstr., 16: 490, abstr. #212.1. Montana, V., Farkas, D.L. and Loew, L.M. (1989) Dual-wavelength ratiometric fluorescence measurements of membrane potential. Biochemistry, 28: 4536-4539. Parsons, T.D., Kleinfeld, D., Raccuia-Behling, F. and Salzberg, B.M. (1989) Optical recording of the electrical activity of synaptically interacting Aplysia neurons in culture using potentiometric probes. Biophys. J., 56: 213-221. Potter, D.D., Landis, S.C., Matsumoto, S.G. and Furshpan, E.J. (1986) Synaptic functions in rat sympathetic neurons in microcultures. II. Adrenergic/cholinergic dual status and plasticity. J. Neurosci., 6: 1080-1098. Ready, D.F. and Nicholls, J. (1979) Identified neurones isolated from leech CNS make selective connections in culture. Nature, 281: 67-69.
Regehr, W.G., Pine, J., Cohan, C.S., Mischke, M.D. and Tank, D.W. (1989) Sealing cultured neurons to embedded dish electrodes facilitates long-term stimulation and recording. J. Neurosci. Methods, 30: 91-106. Salzberg, B.M. (1983) Optical recording of electrical activity in neurons using molecular probes. In J. Barker and J. McKelvy (Eds.), Current Methods in Cellular Neurobiology, Wiley, New York. pp. 139-187. Sigworth, F.J. (1983) Electronic design of the patch clamp. In B. Sakmann and E. Neher (Eds.), Single-Channel Recording, Plenum, New York, pp. 3-35. Syed, N.I., Bulloch, A.G.M. and Lukowiak, K. (1990) In vitro reconstruction of the respiratory central pattern generator of the mollusk Lymnaea. Science, 250: 282-285. Tank, D.W. and Ahmed, Z. (1985) Multiple-site monitoring of activity in cultured neurons. Biophys. J., 47: 476a (abstr.).