On April 2nd, 2013 President Obama announced the BRAIN initiative, a project aimed at understanding brain activities at a level of detail that is unprecedented. A lot has been said and written about the initiative but some people feel in the dark as to what are its precise aims. Although announcing the project before the aims were set in concrete might have raised some worries1,2,3,12, I personally see it as a fresh and open way to proceed. It gives a chance for everyone to express their opinion. Sebastian Seung’s Twitter comment illustrates this very well:
@emckiernan13 @eperlste Not earmarked…you should join the debate by making constructive arguments about how to spend the money. SebastianSeung, 2 Apr 2013.
For this post I will leave aside the political aspect – I think that as a French Canadian my view on American politics is of limited interest. For the curious, I will simply say that I believe it is a very exciting project. It might end up having just as much impact as the big science projects that the USA have been leading during the last century including the exploration of space and the sequencing of the human genome.
What I would like to do is regroup the information that is currently available on the scientific aspect and technical challenges that await the BRAIN project and present them to you in the simplest terms possible. I think it is important for me to specify that I have no particular access to the scientists on that project and the rest of this text is not “insider” information – it is based on published references that are listed at the end of the post.
The BRAIN project proponents argue that it is time for a “large-scale effort in neuroscience to create and apply a new generation of tools to enable the functional mapping and control of neural activity in brains with cellular and millisecond resolution”4.
This means recording many neurons in the brain and having high precision in terms of time and space. Millisecond resolution means knowing the state of the neuron a thousand times per second. It is also expected that the techniques will provide high spatial resolution. Cellular resolution means that neurons are being recorded one by one (see illustration on the left), not as an ensemble as is the case in brain imaging. These two levels of precision are rather common in neurophysiology. We know how to insert tiny electrodes in the brain to record or stimulate a couple of neurons – the technique is already being applied to cure some diseases. What would be novel here is that instead of recording 1, 2 or 3 neurons, the BRAIN project would develop tools allowing us to record from thousands to millions of neurons simultaneously. A million neurons is approximately what would be needed to fully record the brain of the zebrafish4.
How will neuroscientists be able to record a million neurons if the current widespread techniques only allow recording a small number of them ? The answer is that we do not know exactly yet – the goal of the project is to develop the technologies – but some possibilities come to mind and are being raised by the proponents of the BRAIN project. I will thus review those possibilities and point out to a specific challenge, pro and con for each of these techniques. This should provide you with an overview of why the BRAIN project constitutes an advancement and what kind of technological progress we can expect from it. This is in no way an extensive list but a (hopefully) reader-friendly description of the potential avenues that have been discussed.
1. The molecular ticker tape.
The BRAIN project’s main goal is to develop technologies to record brain activities. One of the challenges that has been discussed and that neuroscientists might try to tackle is to create a molecular ticker tape. Developing this technique can be considered a “moon shot”. We do not know if it will work, we have no idea how practical it will be, but if developed correctly the payoff will be huge. In principle it is a very realistic way to record neural activities, but no one has applied it yet. It consists in using a molecule that is expressed in biological cells: the DNA polymerase (DNAP), or similar molecules that produce long strings of other molecules.
Those small molecular machines are part of our normal biological arsenal. They are in charge of copying our DNA. The DNA is the molecule which contains the code to construct all proteins of our body. The polymerase makes operations on the DNA at a very small scale (we cannot see those using a traditional microscope). The polymerase is essential for creating the copies of DNA that are transferred to daughter cells when cells divide during development. Thanks to the polymerase and many other enzymes, the DNA that is present at first in the egg is copied to every cell of the embryo and later on to every cell of the body of the animal as it develops. DNA strings pop out from one side of the polymerase a little bit like a long sheet of paper would come out of a printer.
The idea of using polymerases such as this one to record neural activities has been discussed by many, including George Church and Konrad P. Kording5,6. Since the enzyme is already working well to write information on DNA sequences, scientists think they can “harness” it to write the neural activity of a neuron onto a DNA string. This is due to a fortunate “defect” in this molecule; they tend to make errors – not copying DNA perfectly. In particular, the more calcium there is in the cell, the more likely the polymerase makes errors in copying the DNA. There is no particular reason for that, it just doesn’t make perfect copies. Luckily, when neurons are active, it is usually followed with an increase of calcium in their cell body and dendrites. Thus one can imagine a DNA polymerase that would generate a dummy string of DNA in every neuron of the brain and would make more errors when there is calcium in the neuron. The DNA would be like a micro-document, kept in a region of the cell where it probably would not interfere with normal cell functions. Recovering the DNA dummies of each neurons would allow us to have a “recording” of the moments when the polymerase was making no error (silent neurons) and when it was making errors (active neurons). This would literally provide us with a history of the neural activity of specific cells.
It seems that strings of DNA would not get too long and that we might be able to get data from extended periods. Alivisatos and colleagues made a rough estimation of about 7 days of neural activity that could be encoded in a tiny 5-μm-diameter pack of synthetic DNA7.
The capability of DNA for dense information storage is quite remarkable. In principle, a 5-μm-diameter synthetic cell could hold at least 6 billion base pairs of DNA, which could encode 7 days of spiking data at 100 Hz with 100-fold redundancy.
There is a great post at the Nucleus Ambiguous blog about the molecular ticker tape method.
1. Create the custom version of the DNA polymerase and show that it works.
2. Study alternatives to the calcium-induced errors.
There are many reasons why the calcium-induced errors technique that is for now the most developed approach (although not completely developed yet) might not be the best way to record neural activity. First, the calcium enters relatively slowly in the neuron when they get excited. Second the concentrations of calcium, when neurons are excited, vary depending of where you are in the cell.
One possibility to overcome these problems would be to stick the DNA polymerase to another molecule that would confer additional properties. For instance, the sodium channel is a molecule present in almost all neurons and it detects small changes in voltage in neurons already. Perhaps we could rely on this sodium channel hitting on the polymerase a little bit like a hammer on a nail. There is actually a number of channels like the sodium channel that can be thought of as alternatives to improve the technique but none of them, including the sodium channel, have been tested or are known to work in any way. I have no doubt that these possibilities are currently being considered in the laboratories interested in developing such techniques.
3. Solve the problem of which cell is being recorded.
When extracting the “recordings” from the brain, we will end up with millions of DNA strings but it will be impossible to know from which neuron they come. To be valid the molecular ticker tape approach will require imprinting every DNA string with some sort of bar code, which could allow us to know from which neuron the recording comes. It is still unknown what will be made to address this issue but some of the possibilities include generating a specific random code for each neuron8. This idea requires more development to become practical. There is also a series of molecules that are more or less concentrated in certain spots of the brain such as the molecule called Sonic hedgehog. Detecting those molecules and printing the information on the DNA string might be one way to know where the recording comes from.
– This would be the best technique that we can think of to record all neurons in a brain.
– It is not developed yet.
2. Optical imaging.
The technique of optical imaging is already widely used to record neural activities. I myself have been using it. It consists in inserting a molecule that reacts to light into the neurons. Calcium imaging is one example in which the molecule inserted in the neuron changes the color it emits based on the concentration of calcium in the cell. The more the neuron is excited, the more it appears as “bright” under the microscope. This allows filming the part of the brain of interest and tracking the activity of dozens, up to hundreds or thousands of neurons at the same time.
1. To achieve the degree of spatial resolution and the number of neurons that the BRAIN project is targeting, the method will have to be improved if it is intended to record neurons in behaving mammals. For now the technique is mostly being used on neurons maintained in artificial conditions and we do not know if the thickness of the brain will be an insurmountable obstacle to record many neurons in animals. In animals with very transparent and small brains like zebrafish, it might still be possible. The big challenge is that we need to send light to the neurons and then receive the light emitted from the neurons.
– Realistic method that has been proven to work already.
– Difficult to apply in the deep parts of the brain and is generally more useful on neurons maintained in artificial conditions, rarely in behaving animals or humans.
3. Silicon-based nanoprobes.
With progresses in microelectronics, one could expect that it might become possible to simply create such small electrical probes that it would be possible to insert thousands of them seemlessly in the brain, thus providing many thousands of recording sites7. These techniques are already being used – with about a hundred working recording sites on a probe being available. For now they have been used in the development of brain-machine interfaces, which are prosthetics that patients can use to control artifical limbs with their brain9,10.
1. Make them smaller with more recording sites.
– Realistic method that has been proven to work already.
– The technique is invasive and requires adding a piece of electronic in the brain.
In the most recent text published by the proponents of the BRAIN project4, they also mention that the activity of neurons in the brain will not only need to be recorded but that technologies should also be further developed to alter neural activity. There are already optical techniques that allow researchers to send ray of lights to brain regions in which the neurons have been equipped with light receptors – this technique can be used to shut down or excite some neurons11. This is essential to researchers to assess whether the neurons are actively involved in some specific behavior.
There are also many techniques that are appearing to map connections between neurons, some of which rely on contributions from the general public such as EyeWire. Although these techniques show us the connections without revealing the functions, it is likely that they could be used in combinations with the previously discussed techniques to map the connections between the recorded neurons.
One aspect that has been much less discussed is what kind of behavior will be studied. It’s one thing to say we get neural recordings of millions of neurons, it’s another thing to decide what we want subjects to do. Will the project include early attempts at applying those methods to understand perception, motor control, decision making? It seems that for now the precise experiments that will be run is left open.
Finally one interesting question left is how the data will be made accessible to the wide neuroscientific community and even the general public4. The authors mention that the huge amount of data coming from such high numbers of neurons might require analysis by a wide community. It remains to be determined what means will be taken to make this data accessible but I think it would be very nice if members of the general public could participate in digging through those numerous recordings!
I hope I have clarified what the aims of the project were, as far as what has been discussed in public scientific forums up to now. Keep in mind that we will not have a first official document from the committee being consulted for this project before the end of the year so everything could change.
1. Christopher Chabris (2013) How Much BAM for the Buck, and Other Thoughts on the Brain Activity Map Project. http://http://blog.chabris.com.
2. John Markoff (2013) Obama Seeking to Boost Study of Human Brain. New York Times.
3. Jeanne Garbarino (2013) A 3 Billion Dollar Mistake: Why the American government should think twice about a Brain Activity Map (BAM). http://http://incubator.rockefeller.edu.
4. A. Paul Alivisatos, Miyoung Chun, George M. Church, Karl Deisseroth, John P. Donoghue, Ralph J. Greenspan, Paul L. McEuen, Michael L. Roukes, Terrence J. Sejnowski, Paul S. Weiss, Rafael Yuste (2013) The Brain Activity Map. Science 339:1284-1285.
5. Bradley Michael Zamft, Adam H. Marblestone, Konrad Kording, Daniel Schmidt, Daniel Martin-Alarcon, Keith Tyo, Edward S. Boyden, George Church (2012) Measuring Cation Dependent DNA Polymerase Fidelity Landscapes by Deep Sequencing . PLoS ONE 7:e43876.
6. Konrad P. Kording (2011) Of Toasters and Molecular Ticker Tapes. PLoS Computational Biology 7:e1002291.
7. A. Paul Alivisatos, Miyoung Chun, George M. Church, Ralph J. Greenspan, Michael L. Roukes, Rafael Yuste (2012) The Brain Activity Map Project and the Challenge of Functional Connectomics. 74:970–974.
8. Anthony M. Zador, Joshua Dubnau, Hassana K. Oyibo, Huiqing Zhan, Gang Cao, Ian D. Peikon (2013) Sequencing the Connectome. PLoS Biology 10:e1001411.
9. Hochberg LR, Bacher D, Jarosiewicz B, Masse NY, Simeral JD, Vogel J, Haddadin S, Liu J, Cash SS, van der Smagt P, Donoghue JP (2012) Reach and grasp by people with tetraplegia using a neurally controlled robotic arm. Nature 485:372-5.
10. Pais-Vieira M, Lebedev MA, Wiest MC, Nicolelis MA (2013) Simultaneous top-down modulation of the primary somatosensory cortex and thalamic nuclei during active tactile discrimination. Journal of Neuroscience 33:4076-93.
11. Kim TI, McCall JG, Jung YH, Huang X, Siuda ER, Li Y, Song J, Song YM, Pao HA, Kim RH, Lu C, Lee SD, Song IS, Shin G, Al-Hasani R, Kim S, Tan MP, Huang Y, Omenetto FG, Rogers JA, Bruchas MR. (2013) Injectable, cellular-scale optoelectronics with applications for wireless optogenetics. Science 340:211-6.
12. Scicurious (2013) The BRAIN Initiative: BAM or BUST?. Scientific American Blogs.