Ever since Luigi Galvani applied electric shocks to dead frogs’ legs and watched them twitch in the 1780s, electrodes have been an essential part of neuroscience.
The foundations of systems neuroscience were built by experimenters inserting electrodes into the brains of animals and listening to the chatter of neuronal firing. In 1959, David Hubel developed tungsten microelectrodes that provided excellent resolution of spikes coming from individual neurons. Then he and Torsten Wiesel went on to describe how visual cortical neurons fired in response to what the eyes saw.
In the 1970s, John O’Keefe implanted electrodes in the hippocampus of freely moving rats and, noting that certain neurons fired only when the rat was in specific places, he started to describe the brain’s intrinsic navigational system. Later, Edvard Moser and May Britt Moser showed that the adjacent entorhinal cortex also contained neurons that fired in spatially determined ways, this time in grid-like patterns superimposed on the animal’s environment.
These two Nobel Prize-winning bodies of work and many other investigations showed the power of in vivo electrophysiology to reveal how information is encoded by the brain, and how neuronal activity could be elegantly linked to sensory inputs, cognitive processes or motor outputs of the nervous system.
But today’s systems neuroscientists face a choice. While the animals they study engage in some perceptual, memory-dependent, decision-making or other task, how should they monitor neural activity? Should they use electrodes? Or should they introduce into their neurons of choice a molecule that will signal when these cells are active and image activity?
Neuronal imaging emerged as a major technique in the 1990s, with the development of calcium indicator dyes, but surged in popularity for in vivo neuroscience after the development, in 2001, of GCaMP – a fusion of three proteins that fluoresces on calcium binding and which can be genetically introduced into specific cell types.
At first, many physiologists were suspicious of calcium imaging. Imaging and electrophysiology were often presented as competing to be the best way to tackle the same experimental questions. Georg Keller – who uses two-photon microscopy to image visual cortical activity at the Friedrich Miescher Institute of Biomedical Research in Basle, Switzerland – recalls how, five to ten years ago, people reviewing his papers regularly asked him to confirm all his findings using electrophysiology.
But now, he says, thinking has changed and those requests have stopped. “The methods are complementary – it’s not that one is better or worse than the other,” Keller says. “It’s become clear you adapt the method to your question.”
Edvard Moser concurs. His Nobel Prize-winning work was done using electrodes, but now he also images neurons at the Norwegian University of Science and Technology in Trondheim, Norway. “The two go together,” Moser says, “Electrophysiology has strong temporal resolution, imaging has strong spatial resolution, so we use them both, to address slightly different questions.”
These differences in temporal and spatial resolution are part of a list of differences between the two techniques, all of which need to be considered when deciding which one to use.
The Numbers Game
Last year, Aman Saleem, of University College London, published a paper describing how an animal’s perception of its own position in space affected the way its visual cortex represented what it saw. The initial experiments were done using electrophysiology, and later ones with calcium imaging. Asked why he switched, Saleem replies simply: “More cells!”
In one experiment using electrophysiology, four mice provided usable recordings from 226 neurons. In an imaging experiment, four mice yielded data from 8,610 visual cortical neurons, of which 4,958 had reliable visual responses. Crucially, the activity recorded using each technique gave the same conclusions.
The ability to image from hundreds of neurons at once is the most widely cited advantage of calcium imaging. This is particularly true of wide-field two-photon imaging. Gathering data from as many as cells as possible simultaneously has many advantages. Most pertinently, neurons in most brain regions have heterogenous firing properties – for example, less than 10% of entorhinal cortical neurons are grid cells and only a small fraction of visual cortical neurons are the types of “feature detectors” that Hubel and Wiesel’s work was based around. Therefore, to get an unbiased sampling of a brain region’s encoding properties – and to increase the chances of finding neurons doing interesting things – it is good to sample many cells.
Indeed, as neuroscientists bid to understand how coding properties emerge from populations of neurons, they are looking increasingly at interactions between many simultaneously recorded neurons to describe network properties. Additionally, in studies aimed at understanding how activity in separate brain regions is coordinated, small samples of neurons from two regions may not reveal correlated activity, when larger samples will.
Observing many neurons using calcium imaging is also particularly useful when a brain region is only sparsely active, says Mazen Kheirbek of University of California San Francisco (UCSF). In the dentate gyrus – the hippocampal subregion that much of Kheirbek’s research focuses on – less than 5% of neurons are active at any given time. This means firing neurons are hard to locate when sample sizes are small.
Saleem does note however, that electrophysiology is striking back in this regard. Having been a discipline that studied neurons one-by-one for decades, the emergence of tetrodes (four wires bound together) then multisite silicon probes in the 1990s and 2000s allowed tens of neurons to be recorded simultaneously. But now recently developed high-density multisite electrodes are recording from hundreds of neurons at once.
For example, Neuropixelsprobes have 1000 recording sites and were shown, in 2017, to be capable of recording from 500 cells distributed across five separate brain regions at once. More recently, Loren Frank’s group at UCSF described flexible polyimide probes, with 1,024 recording sites, that move with the motion of the brain in freely moving rodents. These were chronically implanted and shown to be able to stably record from hundreds of neurons for weeks.
Mazen Kheirbek’s lab at the University of California San Francisco uses calcium imaging to observe cells in the dentate gyrus – a hippocampal region where less than 5% of neurons are active at any given time.
What’s being recorded from?
In GCaMP-based imaging, the gene for the calcium indicator is heterologously expressed in the neuronal subtype of choice and those cells make their own activity sensor. Or GCaMP is expressed in many different types of neurons, and other fluorescent proteins are expressed in specific cell types to individually identify them. Either way, experimenters are able to unambiguously identify the type or types of neuron they are recording from. Moser, Kheirbek and Keller all cite this as a huge advantage of the technique.
By contrast, electrophysiologists can identify the brain region they are recording from, but beyond this, all they have to identify neurons by is an electrical waveform. Some neurons, such as fast spiking interneurons, generate instantly recognisable spikes, but in general there is a degree of uncertainty as to which neurons exactly have been studied. One way around this is expressing, in genetically defined cell types, light-sensitive proteins that can be activated during a recording to confirm a cell’s identity by a functional response.
But then calcium imaging can also be used to examine cellular compartments. It’s frequently been combined with high-resolution two-photon imaging to examine signals in dendrites. And Keller is using it to study axonal signals. “With imaging this is trivial,” he says. A GCaMP-carrying virus is injected in one area, then after GCaMP has been synthesised and trafficked to the axon terminals, the signals those terminals deliver to a distant region of interest can be imaged.
Finally, once a field of neurons has been visually defined using imaging it is straightforward to return to that population, identify the previously imaged cells and, thereby, accumulate data from the same neurons for days or weeks, following how these neurons change properties with learning for example. With most electrophysiological techniques, however, slight movements of the electrodes relative to the neurons they’ve recorded from cause the electrical read-outs to change form. This leads to a lack of confidence in saying that the same neurons are being recorded from over time, albeit the new high-density flexible probes developed by Frank and colleagues may overcome this.
What’s being recorded?
Electrodes record the direct consequences of ions flowing back and forth across neuronal membranes. In electrical recordings, the researcher can detect individual action potentials, plus the local field potential – a larger, typically oscillatory, signal reflecting the average activity of the local population of neurons.
In contrast, calcium imaging charts only a proxy of neural activity – the multifactorial rise and fall in intracellular calcium resulting from neuronal firing. Calcium imaging is, therefore, excellent for indicating which neurons are active during a given task, with an approximate but smoothed indication of their exact activity.
This issue – of how exactly intracellular calcium levels relate to a neuron’s electrical activity – that made so many physiologists guarded about imaging when it first emerged. Calcium was the ion of choice to track, because although fluxes of potassium and sodium do most of the bioelectrical legwork, intracellular calcium concentration increases up to hundred-fold upon action potential firing, giving an excellent signal to noise ratio.
However, the relationship between intracellular calcium concentration and the number of spikes a neuron has fired is typically nonlinear and can differ markedly between cell types.. Also, calcium levels rise and fall very sluggishly compared to the actual membrane potential changes. “The temporal resolution of calcium imaging cannot at all match that of electrophysiological recordings,” says Moser.
Moser explains that because calcium signals are cumulative there’s a bias in most recordings towards detecting bursts of spikes and highly active neurons. And because calcium signals are slow – rising and falling over 10s or 100s of milliseconds – they cannot give precise information about the timing of spiking in neurons relative to one another, or relative to the local field potential.
“For some questions these things matter and for others not, and one may be happy with calcium imaging as a proxy of the neural spike activity,” Moser says. As examples where electrophysiology is preferable, he cites work showing how different hippocampal place fields become sequentially active in roving animals relative to the timing of the local theta (8-12 Hz) oscillation in the field potential and the way hippocampal neurons replay activity in very brief sharp wave ripples.
For Kheirbek’s behavioural research, timescale is a big factor. The animals he studies may maintain a working memory trace for only a second or two during a decision-making task. Calcium signals representing activity over 100s of milliseconds or a second don’t really convey useful information of how neurons are dynamically responding during such brief time windows.
A further consideration when using calcium indicators is that they are calcium binding molecules that buffer free cytosolic calcium. This means they potentially disrupt calcium-based intracellular signalling.
What can the animal do?
At present, the high-quality imaging achieved by two-photon microscopes is, for most labs, only achievable in head-fixed animals. This limits what behaviours the animals can engage in, although there is an array of possible tasks available and placing animals in virtual reality simulators can also help.
To do calcium imaging in freely moving animals, researchers are increasingly using miniature head-mounted endoscopes, which, despite a lower image quality than two-photon imaging, still provide excellent data from many neurons.
Electrophysiological techniques have long been honed to work as head-mounted technologies, allowing freely moving animals to be studied. This is very much true of the new, flexible, in-dwelling polyimide probes with 1,024 recording sites. In contrast, whilst free-movement recording remains possible using Neuropixels probes, these probes appear better suited to recording in head-fixed animals .
Getting in there
Electrophysiology requires inserting electrodes into the brain, whilst imaging in any location bar the most superficial layers of the cortex requires inserting either fibre optics, lenses or windows into the brain. How much damage these procedures do remains a fairly neglected topic of discussion.
However, based on the number of neurons theoretically present in the volume of tissue from which an extracellular electrode records, electrophysiologists have found they typically only detect signals from a fraction of that number. This suggests a good number of neurons are either silent or, quite likely, damaged or killed.
The damage done to gain optical access to a brain region can be frankly larger, but the imaged neurons are usually 100s of microns from the lens and therefore likely to be spared direct physical trauma. These issues become more pronounced when the area of interest is deeper in the brain or multiple brain regions need to be accessed.
Making sense of the data
Both imaging and electrophysiology require careful analysis of the data generated to ensure high quality results. The central issue is ensuring that what the experimenter says is an individual neuron is indeed one neuron. If two adjacent neurons generate similar action potential waveforms, the firing of that pair of neurons maybe erroneously attributed to one cell. Likewise, if one region of an imaged field of view contains cytoplasm from two neurons, a similar error can occur in imaging.
Consequently, while automated data analysis algorithms are constantly being refined, a semi-manual quality control step is the most time-consuming aspect of both techniques. Long recognised as the analysis bottleneck for physiology, Kheirbek says that going through a data file from 40-minute imaging session, accepting and rejecting regions of interest identified by the algorithm, can take a full day.
Neuroscience technology never stands still. Advances in microscopy continue apace, both in terms of the hardware used to acquire images and the molecules generating those signals . Moser, for instance, is excited about using miniaturised, head mounted two-photon microscopes. Additionally, new fluorophores operating using longer wavelength light, for instance, promise the ability to image deeper into the brain without the need to remove overlying tissue.
Another innovation poised to have a major impact on neuronal imaging is the development of voltage sensitive dyes. Long known to be theoretically plausible, voltage indicators need to be sensitive and respond quickly to small voltage changes and to be expressed at high enough levels to give robust signals. These issues are looking increasingly to have been resolved and voltage imaging may provide many of the established benefits of calcium imaging while also giving a direct, real-time read out of neurons’ membrane potential changes. “People are very interested in using them,” says Kheirbek, “it’s only going to be a matter of time.”
Similarly, novel electrophysiological methods continue to emerge. In addition to the newer high-density multi-site arrays, there are other exciting techniques coming to the fore, such as neural dust, where tiny implantable particles use ultrasonic signals to transmit information about neuronal activity.
For now, different researchers maintain their own preferences. Keller embraces imaging’s versatility– having characterised the activity of visual cortical neurons for years using imaging, he’s now imaging the signals that arrive down axons arriving into the visual cortex from other regions–. Moser, whilst remaining excited about what head-mounted two-photon microscopes can bring to his study of the navigational system, says of electrophysiology, “In many ways I still prefer it, you get very high temporal resolution and you record every single spike.”
As Saleem charts how that navigational system interacts with the visual system, he is happy to have head-mounted imaging endoscopes running in parallel to in vivo electrophysiology. And while Kheirbek’s imaging rig continues to be used for about 18 hours a day probing ever deeper the function of the hippocampus, it will soon run alongside a new Neuropixels rig.
Rather than the two techniques being in opposition, contemporary neuroscience is blessed with the luxury of choosing the best tool for the job.