What Should Systems Neuroscience Do Next? Voltage Imaging

The firing and the wiring at the same time

Mark Humphries
Jun 3 · 10 min read
Credit: Pixabay

The best thing about being a neuroscientist is that neuroscience never stands still. Barely a week passes without some new major result, a sparkling technological breakthrough, a provoking theoretical idea. And the sheer complexity of brains means the questions available are practically infinite. So even if your specific corners of brain research have briefly slowed their breathless pace, there is always more to learn. Always new questions to tackle. Indeed, there are whole regions of the mammalian brain whose mysteries have barely been probed, and which will no doubt turn out to be crucial for our understanding. My money’s on the zona incerta, the globus pallidus, and the septum. Exciting times.

The worst thing about being a neuroscientist is that neuroscience never stands still. Barely a week passes without some new result, breakthrough, or theory that you don’t have time to read; that ends up filed for later, destined never to be opened; or to be skimmed and not assimilated. And the sheer complexity of brains means the questions available are practically infinite. So even if you’re lucky enough that your specific corners of brain research have briefly slowed their breathless pace, there is always more to learn. Always new questions to tackle. Indeed, there are whole regions of the mammalian brain whose mysteries have barely been probed, and which will no doubt turn out to be crucial for our understanding. Worrying times.

Paradoxically, this best and worst of all possible worlds in mind research is created by mindless churn. Of doing whatever can or could be done next. Not what should be done next. So, hubristically, I thought I’d plant my feet against the torrent and take a stab at separating the should from the could. A series of occasional pieces that set out to answer the question: what should systems neuroscience do next?

In this first piece, we start with the very definition of systems neuroscience. It is at heart the study of the activity of multiple individual neurons, of the messages they are sending. Everything we see, do, or think in the moment is through neurons sending spikes to each other. So a clear priority for systems neuroscience is to make the best recordings of the output of the most neurons, and with as much metadata about those neurons — where they are, how many there are, what type they are — as possible.

We have two mainstream ways of recording the output of individual neurons: insert electrodes to record spikes, or image calcium fluxes in the neurons’ bodies as a proxy for spikes. Both have unique strengths, both are constantly evolving in the white-heat of technology (and cash), but both have problems that are solved by the other. So our first “should”: we should find a recording method that combines the strengths of both. The great news is that we already know the answer. The answer is voltage imaging.

And if we get it solved, voltage imaging comes with a massive bonus prize, something neither electrodes nor calcium imaging can buy us: live connectomics.


Don’t get me wrong, the current state of the art in recording technology is jaw dropping.

Electrodes are going from strength to strength. As lovingly documented by Ian Stevenson, in the past few decades we’ve seen an exponential increase in the number of neurons we can record simultaneously with electrodes. A ceaseless acceleration, currently doubling about every six years. The shaft du jour is the NeuroPixels probe, a marvel of engineering, with a staggeringly high yield of spikes from a few hundred individual neurons, across a shaft so long that it can record from the cortex through to the depths of the rodent brain at the same time.

But. But no matter how amazing the shaft, the shaft is still blind. We do not know which neurons we record from. Nor can we see anything other than spikes. And recording blind is biased: by definition, we cannot see neurons that send few to no spikes on a device that records only spikes.

Calcium imaging solves the blindness problem. Here we insert fluorescent molecules into the bodies of many neurons, molecules that bind calcium. Calcium in the body of a neuron increases with each spike. So the logic of calcium imaging in systems neuroscience is simple: increased calcium means increased fluorescence means more spikes. We simply have to point a video camera or equivalent at the neurons to record the changes in fluorescence, and we have a way of recording the output of many tens to tens of thousands of neurons at the same time. Better even than this massive yield of neurons, we can see what we record from, which types of neuron(s) are active or are not active, and where they are.

But. But changes in calcium are slow compared to the time-scales over which spikes are produced. So rapid sequences of spikes can be at best obscured and at worst totally invisible to calcium imaging. And the link between spikes and calcium is nonlinear: there is not the same increase in calcium for each spike in a sequence. Drawing the line from changes in fluorescence to the spiking output of a neuron is then anything but straightforward.

What we want is both: we want imaging of spikes. Then we can combine knowing the spikes from each neuron (like electrodes) with knowing what and where each neuron is (like calcium imaging).

And we’ve had it for decades. So what’s the catch?


When I say decades, I mean decades. The first major report of using voltage imaging for a single neuron was in 1973. [Salzberg, B. M., H. V. Davila, and L. B. Cohen. 1973. Optical recording of impulses in individual neurones of an invertebrate central nervous system. Nature 246: 508–509.] As I understand it, there are some significant roadblocks.

The mainstay of voltage imaging was (and is) dyes, molecules that bind to the membrane of a neuron and fluoresce in proportion to the voltage across that membrane. Like all dyes, these have issues (as do dyes used for calcium imaging). We lack control over which neurons take up the dye, and the dyes can be taken up unevenly across the neurons in the same piece of tissue.

The biggest problem is that changes to the fluorescence of voltage-sensitive molecules is absolutely tiny, akin to detecting a gnat’s heartbeat in a wind tunnel. Their signal to noise ratio is dire. The number of photons emitted by the molecules within a patch of membrane are barely different to the background noise; one estimate is that the combined optical signal from one neuron has a maximum excursion of 0.0001 to 0.003 percent over the background. Which also means the change in photon emission at the neuron’s body is barely different from surrounding tissue. Up to now, solving the signal-to-noise problem has meant collecting lots of photons to distinguish signal from noise.

One way to get lots of photons is look at really large neurons. Voltage imaging works beautifully in the kinds of invertebrates that have giant neurons. And used this way, voltage imaging has already addressed fundamental questions in systems neuroscience. For example, in the mid 1990s seminal papers from Wu et al established the existence of multifunctional circuits — a single population of neurons whose dynamics underpinned two different behaviours. Ten years later, Briggmann and Kristan showed that the low-dimensional activity of a neuron population encoded the decision to crawl or swim in a leech. Another ten years on (mid-2010s), and my collaborator Bill Frost’s lab records every single spike from up to 200 simultaneous voltage imaged neurons in sea-slugs. We’ve had some fun with these data, like working out exactly what kind of attractor the low-dimensional activity is sitting on. Who knows what another ten years might bring?

Another way to get lots of photons is to look at larger bits of brain than single neurons. Like whole patches of cortex at once, where each separate train of fluorescence is the sum over hundreds or thousands of neurons. In the mid 90s, Arieli, Grinvald and colleagues combined this wide-field voltage imaging of cortex with electrode recordings of a single neuron within that field, to get some idea of what was happening when that neuron fired. In the process, they showed that both the spontaneous and stimulus-evoked activity in visual cortex was remarkably similar, a keystone finding for modern “predictive processing” theories of what cortical activity is.

These are great. But we want to record many individual neurons in mammals. Plus, we’ve already rehearsed that dyes have issues. So the priority for voltage imaging is to solve the signal-to-noise problem, by finding molecules that emit a much higher fraction of photons. And to solve that problem using genetically encoded sensors, for excellent take up across neurons, in defined populations, and that will allow for clever manipulations I’m not smart enough to think about but other people are. The priority is imaging of spikes in multiple neurons by genetically encoded voltage sensors.


In a sudden flurry of activity these issues are being resolved.

The last few years has witnessed a spate of new genetically encoded voltage sensors, each solving some problems. Like increasing how long the sensor is stable, before it bleaches out and becomes useless. Better signal-to-noise. Or getting the photons emitted in a different part of the spectrum, so imaging the sensor’s fluorescence could be combined with the light sources needed for optogenetics. And Mark Schnitzer’s lab recently introduced an approach to do fibre optic recordings of voltage sensors, allowing deep brain imaging of the joint voltage of a group of neurons.

And now we have a major breakthrough, a glimpse of the full potential, and one reached simultaneously by multiple groups, as these things often are. In a Nature paper this month from Adam Cohen’s group, a pre-print led by Xue Han and Ed Boyden, and a pre-print from a colossal team at Janelia Farm, we now have proof of principle that we can do everything we want: image the spikes from multiple, genetically tagged neurons at the same time, in deep tissue, in behaving animals. Sure, compared to the scale of the cutting edge electrode recordings or calcium imaging, the numbers here are tiny, a mere handful of neurons (though one recording of a few dozen is tantalisingly on display in the Janelia Farm pre-print). But the key breakthrough is the sensitivity of the sensors themselves.

All teams announced staggeringly high signal-to-noise ratios. High enough that we can get to see not just spikes, the big jumps in voltage, but the membrane voltage itself, and in multiple neurons at the same time. And that’s the bonus prize.


Imaging of spikes alone is really cool: combining the strengths of electrodes with calcium imaging. That’s enough of a reason to prioritise voltage imaging above all. But my real agenda here is that we want to image the membrane voltage itself. Because with that, we have the best possible data for working out the connections between the neurons we record from.

Inferring the connections between neurons from their spikes has been a cottage industry in neuroscience for a couple of decades. The basic idea is that if we record neurons {1,2,3,…,N} and neuron 1 is connected to neuron 2, then the spikes from neuron 1 should have a detectable effect on when neuron 2 sends a spike. The problem is, no matter how fancy the maths, the problem is horribly under-determined. Spikes are really sparse, and never alone cause other spikes. After all, in cortex, a spike occurs when a neuron integrates tens to hundreds of inputs, causing its membrane voltage to hit a tipping point (to a first approximation). And that’s without accounting for the uncountable confounds of other inputs to the neurons.

But, spikes cause small changes in their target neuron’s membrane voltage, the post-synaptic potentials. A spike from neuron 1 will likely cause a flicker of voltage in neuron 2. Sure, this is not a 1:1 relationship; but the co-occurrence of spikes and flickers will be far more dense than spikes alone. And they come with critical timing information, for while a spike in neuron 2 could come a few milliseconds or a few hundred milliseconds after a spike in neuron 1, the voltage flicker in neuron 2 must come at about the same delay each time. Being able to record the voltage flickers caused by the spikes arriving at each neuron will thus give us a lot of information we can use to work out the connections between neurons. Will give us predictable, time-locked causes and effects in pairs of connected neurons. Exactly the same kind of information we get from the delicate, difficult process of patching electrodes onto two, three, or four neuron, that we already use as a gold standard to work out functional connections between neurons, but scaled up by orders of magnitude.

And that’s the bonus prize: voltage imaging with a high signal-to-noise gives us not just spikes, but the flickers of voltage caused by spikes. Voltage imaging could give us not just the fires, but also the wires.

Here’s then what we should strive for: voltage imaging of the membrane potentials of thousands (or more) neurons at once. How? By combining all the above recent advances in voltage sensors with the recording techniques being endlessly optimised by calcium imaging studies.

Then imagine this. One day we will be able to record the voltage of thousands of neurons at the same time, in a behaving animal. That data will give us the firing of thousands of neurons in cortex, or hippocampus, or striatum, or amygdala. Neurons of known location, type, and number. And it will also give us their membrane voltages. With those data we could reconstruct the wiring between the neurons in situ, do “live connectomics”. The wiring of not just one example circuit from one animal, painstakingly constructed over years of deep, hard work under the microscope. But the wiring of what you’re recording from, right now. In simpler circuits, and in simpler animals, that could give us a complete wiring map. The live connectome. Seems worth a shot.

Want more? Follow us at The Spike

Twitter: @markdhumphries

The Spike

The science of the brain, from the scientists of the brain

Mark Humphries

Written by

Uses his brain to understand brains. Is that possible? Neuroscience: https://humphries-lab.org

The Spike

The Spike

The science of the brain, from the scientists of the brain

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade