The patch clamp

Before 1976, the ion channel was an inference. The Hodgkin–Huxley equations described membrane conductances that behaved as if discrete gates were opening and closing, but nobody had watched one do it. The patch clamp closed that gap, and it did so by solving an instrumentation problem rather than a biological one.

What it is

A fire-polished glass pipette, tip on the order of a micrometre, is pressed against a cell membrane and gentle suction applied. If the glass and the lipid make intimate contact, the resistance of the leak path around the rim climbs into the gigaohm range — the “gigaseal.” Everything that makes the technique work follows from that number.

With the patch electrically isolated, a feedback amplifier holds the membrane at a commanded voltage and reports the current required to do it. When a single channel in that patch opens, the current step is a few picoamps.

Why it matters

Picoamps are not a difficult measurement in isolation. Picoamps in the presence of a leak path are another matter, and that is the whole story of why this took until 1976.

The seal resistance sets the thermal noise floor of the measurement:

$$i_n = \sqrt{\frac{4 k T B}{R_{\text{seal}}}}$$

Here $k$ is Boltzmann’s constant, $T$ the absolute temperature and $B$ the measurement bandwidth. At room temperature over a 1 kHz bandwidth, a 100 MΩ seal contributes roughly half a picoamp RMS — the same order as the signal, which leaves only the largest and slowest events visible. Push $R_{\text{seal}}$ to 10 GΩ and the noise falls by a factor of ten, comfortably below a single-channel event. The gigaseal was not a convenience; it was the difference between an experiment and a method.

This is worth dwelling on because it is the pattern that recurs across this whole direction of the site. The biology did not change in 1976. The noise floor did.

1976, without a gigaseal

By the early 1970s the channel had a predicted size. Katz and Miledi, and then, more precisely, Anderson and Stevens, had read it out of the noise in the frog endplate’s response to acetylcholine: if the fluctuations were the sum of independent unit events, each was a conductance of a few tens of picosiemens lasting a few milliseconds — a few picoamps. Steps like that had been seen, but only in artificial lipid bilayers doped with antibiotics or proteins. In living cells, as Neher later put it, the background noise was about a hundred times too high.

He also put the requirement as a number. Resolving 1 pA with 10% accuracy at 1 kHz needs a signal source of about 2 GΩ or more, and the large cells that conventional microelectrodes required had input resistances between 100 kΩ and 50 MΩ. The source had to be made smaller — a patch of membrane a few micrometres across, isolated by the pipette — and the seal around it then became the leak path.

In 1976 that seal was tens of megohms. Neher put the early ones at 10–20 MΩ, two orders of magnitude short of what was wanted, and even the later ones were of the order of 50 MΩ. By the arithmetic above that should barely have worked, and it worked because everything else was arranged in its favour. The frog muscle was denervated, which makes the fibre put acetylcholine receptors all over its surface rather than only at the endplate, and treated with enzymes — Sakmann’s experience from Katz’s laboratory in London — so that glass could reach bare membrane. The pipette held suberyldicholine, an acetylcholine analogue that keeps the channel open longer. The fibre was cooled to 8 °C and held at −120 mV, both of which lengthen openings, and the large voltage enlarges the current too. Larger, slower events survive the low-pass filtering that brings noise down. The pipette only listened: the fibre’s potential was set by a conventional two-microelectrode voltage clamp. The methods paper that followed in 1978, with Steinbach, called getting the necessary resolution “a delicate compromise.”

What came out were rectangular pulses of a few picoamps that switched on and off abruptly, at random — the all-or-none, stochastic gating that had been postulated, at about the size the noise analysis had predicted. Neher later described it as the first time anyone could watch a biological macromolecule change conformation in situ and in real time. It was also plainly marginal. The background still hid smaller and briefer channels, and many channels sat under the rim of the pipette, where their currents were only partly collected.

The gigaseal, around 1980

In the years that followed, attempts to improve the seal — cleaning the membrane, coating the pipette, reversing the charge on the glass — had little success. Then, in Neher’s account, the group noticed by chance that slight suction on the pipette made the seal jump by more than two orders of magnitude, into the gigaohm range, and that it did so reproducibly provided everything was clean: a fresh pipette for each approach, filtered solutions.

The noise fell by about an order of magnitude and recordings could open out towards 10 kHz, but the gigaseal changed more than the noise floor. With the leak all but gone, the pipette could clamp the patch itself, without microelectrodes, which is what let Sigworth and Neher step the voltage and record single sodium channels in 1980. And the seal was mechanically strong. Pull the pipette away and the patch came with it, free of the cell — found at about the same time by Horn and Patlak and by Hamill and Sakmann. Rupture the patch instead and the pipette had low-resistance access to the whole cell. Hamill, Marty, Neher, Sakmann and Sigworth described the lot in 1981, with amplifiers that Fred Sigworth, who brought engineering experience to the group, had improved to match.

How it works

Four configurations became standard, and each trades access against integrity:

• Cell-attached — the pipette sits on an intact cell. Channels are observed in their native environment, but the cell’s interior potential is not under your control.
• Whole-cell — rupture the patch and the pipette interior becomes continuous with the cytoplasm. You now command the whole cell’s membrane potential, at the cost of slowly dialysing its contents.
• Inside-out and outside-out — excise the patch entirely, exposing one face to the bath. Either face can be presented to a drug at a known concentration.

The 1981 paper that catalogued these is arguably more responsible for the technique’s spread than the 1976 result, because it turned a difficult demonstration into a protocol other labs could follow.

Of the four, whole-cell is the one that took over. The leak around a gigaseal is small enough that cells as small as red blood cells tolerate it, so cells that would not survive repeated impalement by conventional microelectrodes became recordable, many of them mammalian. Neher counted the shift in the Journal of Physiology: in the first half of 1981, 5 of 14 voltage-clamp studies used mammalian cells; the first issue of 1991 had 10, none on invertebrates, all by whole-cell or single-channel recording. Two methods papers in 1989, from Sakmann’s group and from Kriegstein’s, took it into brain slices, and it has since reached the intact brain, where two-photon imaging can steer a pipette onto one genetically labelled neuron (Margrie and colleagues, 2003).

Whole-cell has costs of its own. It replaces the cell’s interior with the pipette solution: small ions exchange in seconds, second messengers within a minute, small proteins over several minutes, and whatever depended on them runs down. Perforating the patch rather than rupturing it, so that it passes only small ions, fixed much of that (Lindau and Fernandez, 1986; Horn and Marty, 1988): in Horn and Marty’s cells, with nystatin in the pipette, a response that washed out within about five minutes was undiminished after an hour. And in a neuron the pipette clamps the soma: the cable properties that attenuate synaptic signals on their way in also attenuate the command on its way out, so currents from distant synapses are measured shrunken and distorted — more so the farther out, the faster and the larger they are (Spruston and colleagues, 1993).

Where the rest of the noise lives

A gigaseal does not make the measurement quiet; it stops the seal being the loudest thing in it. In 1981 the background was still several times what the seal and patch alone would allow, and the excess came in roughly equal parts from the pipette and from the current-to-voltage converter.

The converter reads the pipette current as the voltage across a feedback resistor, and that resistor obeys the same equation as the seal — so to stay out of the way it has to be comparably enormous, 10 GΩ in the 1981 design. A 10 GΩ resistor with 0.1 pF of stray capacitance across it is itself a low-pass filter with a 1 ms time constant, a corner near 160 Hz, so the circuit corrected the response afterwards to reach 10 kHz. Later designs replaced the resistor with a capacitor — integrate the current, then differentiate — which removes its thermal noise at the cost of periodic resets.

Above a few hundred hertz the spectrum rises, and capacitance is the reason. The feedback loop impresses the amplifier’s own input voltage noise on the pipette, and that voltage drives current through every capacitance at the input — pipette wall, holder, the amplifier’s own input, around ten picofarads in all in 1981:

$$S_i(f) = (2 \pi f C_{\text{in}})^2 \, S_v$$

Here $S_i$ is the resulting current-noise spectral density, $S_v$ the amplifier’s voltage-noise density and $C_{\text{in}}$ the total input capacitance. Because $S_i$ rises as $f^2$, the RMS noise from this term grows roughly as the bandwidth to the power 3/2, against the square root for a resistor, so resolution in amplitude and in time trade steeply. In the 1981 set-up a 1 pA opening had to last at least 0.15 ms to be detected; a 0.2 pA one, at least 2.2 ms.

The pipette adds rising noise of its own, from a film of solution creeping up the outside of the glass and from losses in the glass wall. Hence the rituals of the bench: Sylgard on the shank, which stops the film and thickens the wall; a shallow bath; thick-walled glass; and eventually quartz, whose dielectric loss is about a tenth of the best glasses’. With all of that and a capacitive-feedback amplifier, Levis and Rae routinely reached about 100 fA RMS at 5 kHz — roughly the thermal noise of a single 10 GΩ resistor over the same band.

Single ion channels became observable

What neuroscience got was not a better number but a different kind of object. Before, a single channel was a quotient: a peak conductance divided by a count of toxin-binding sites, or a current variance divided by its mean. After, it was a trace — closed, open, closed — with an amplitude you could histogram and dwell times you could fit.

The sodium channel is the clearest case. In a voltage-clamped nerve preparation the sodium conductance rises and falls smoothly, because it is the sum of ten thousand to ten million channels. In Sigworth and Neher’s patches on cultured rat muscle, a step of voltage opened a channel briefly, or not at all, at about 18 pS, and only an average over hundreds of sweeps rebuilt a smooth transient. The conductance in the Hodgkin–Huxley model is an ensemble average of all-or-none events — the discrete gates its equations had implied, seen one at a time.

Being observable also made the model checkable, and parts of it failed. Single-channel records from mammalian cells showed that delayed first openings make an important contribution to what looks like inactivation in the macroscopic current, and that an open channel inactivates faster than that current decays (Aldrich, Corey and Stevens, 1983; Aldrich and Stevens, 1987). Fitted to such records, the Hodgkin–Huxley scheme, with its independent gates, was rejected outright (Horn and Vandenberg, 1984), as gating-current measurements tying inactivation to activation had foreshadowed. Whether inactivation has much voltage dependence of its own was disputed between the groups.

The same quiet front end measured more than channels. Capacitance is proportional to membrane area, so in whole-cell mode the fusion of a single secretory vesicle shows up as a capacitance step; Neher and Marty saw such steps in adrenal chromaffin cells in 1982.

In 1991 Neher and Sakmann shared the Nobel Prize in Physiology or Medicine “for their discoveries concerning the function of single ion channels in cells.”

What it cost

The gigaseal is a mechanical accident that nobody fully controls. Success rates depend on pipette geometry, cell health, and the operator’s hands, which makes patch clamping famously an acquired skill rather than a procedure. Automated systems now handle much of it, but the throughput ceiling this imposed shaped decades of electrophysiology — you studied one cell at a time, carefully, because that was what the physics allowed.

That constraint is exactly the one microelectrode arrays attacked from the opposite direction: give up the single-channel resolution, and record from hundreds of cells at once instead.

Automation came at the same constraint while keeping the seal. One form removes the pipette: a hole about a micrometre across in a flat chip, onto which a cell from suspension is pulled by suction. Fertig, Blick and Behrends recorded whole-cell currents through such a hole in quartz in 2002, and Klemic and colleagues at Yale reported one moulded in silicone the same year. Industrial instruments now record from hundreds of cells in parallel, mostly in drug discovery, where screening new compounds against cardiac ion channels has become a standard safety step. The other form keeps the pipette and replaces the hands: a robot that lowers it into a living brain and recognises a neuron by the sequence of impedance changes at the tip as it approaches (Kodandaramaiah and colleagues, 2012).