This week over 150 neuroscientists were invited to meet in Arlington, Virginia to discuss the finer points of President Obama’s recently announced BRAIN Initative. Rather than discuss funding particulars, each participant was given the chance to broadly declare what they thought needed to be done in neuroscience. At least 75 of the participants initially responded to a request for a short white paper outlining the major obstacles currently impeding neuroscience research. A live webcast of some of the key talks was available, although many of the smaller workshops were held in private. Fortunately, updates regarding the content discussed at these workshops was posted live to twitter under the handle @openconnectome. This precipitated lively discussion, primarily under the hashtags #nsfBRAINmtg or #braini, and provided a way for a larger audience to be involved.
The working title of this inaugural NSF meeting was Physical and Mathematical Principles of Brain Structure and Function. In actuality, there was little discussion of all that, and for good reason—no such principles have been shown to exist. Even more concerning, only a few principles have ever even been proposed. Simplistic scaling laws dealing with connectivity, particularly within sensory systems or the cortex, have been suggested in the past. Generally they seek to account for only one or two structural parameters at a time, like for example, axon diameter and branching order. Typically, the chosen parameters are only considered in the context of optimizing a single physical variable, like for example, electrotonic function. While these efforts are a start, they usually do not garner much attention from the larger neuroscience community.
The early days of neuroscience were marked with the assertion of many principles and laws. They served well to focus ideas, but over time, they lost much of their original perceived generality. For example, concepts like one transmitter type per neuron, and no new neurons in adult brains later proved to have significant exceptions. The early breakthrough days in neuroscience have now given way to a grant system that stifles imagination, and by its competitiveness, encourages fraud. Many of the speakers at the BRAIN Initiative meeting have called for new tools and theories, but in most cases, they have offered only little has been offered. Instead of expanding the range acceptable pursuits, their vision appears to have imploded inward with calls for increased rigor, statistical power, diversity of animal models, experimental falsifiability, and most of all, data, on an increasingly limited range of ideas.
A lot of talk was given to the resolution at which connectivity, and activity maps should be detailed. Similar points were made for the need to develop electrode arrays of higher density and durability to more accurately record function. The ample discussion of an ideal animal model was punctuated by the notable advances made this year in whole brain recordings from Zebrafish, and also from large scale connectivity mapping now possible in small mammals with the new CLARITY transparent brain techniques. The general lack of agreement and clear path forward as to which organisms among many are ideal here was noted by representatives from several funding bodies who spoke at the meeting. Highlighting points made earlier in a talk by George Whitesides, they stressed the need to come to forward with a concrete plan that is comprehensible not only to the funding organizations, but the larger public as well.
Many discussions focused on brain mechanisms, like for example, how many neurons might contribute to a particular function. One participate, David Kleinfeld, called for a study of how many neurons are involved in communication at different scales. He also stressed the importance of looking at basic systems involving feedback, such as the brain stem and spinal cord, and their dynamic interaction with muscle. Michael Stryker observed that the goal should not be recording from the most neurons, and storing the most data, but rather finding the right neurons.
While it was not explicitly stated, a lot of the talk begged the conclusion that the answers to the questions we have will not be answered with animal studies. Knowing what a neuron does is itself an ill-posed question. In worms and flies, where the inputs and outputs of single neurons can be mapped to static sensory and motor functions in the real world, we might know what that neuron does. However in larger, human brains, we can ask an even better question—what does the neuron feel like? In most cases that answer will likely be, nothing.
If however, in a given human brain, a single neuron critically poised within that brain’s structural hierarchy can be stimulated to observable effect, some measure of its function has been gained. That effect might be a simple itch or twitch. Less plausibly perhaps it could be seeing a picture of a face undergo a change, sensing fear, or even imagining your grandmother. If that turns out not to be possible for most single neurons, we already know that we can find some minimal group of neurons where stimulation has uniquely perceivable effects.
While understanding the brain on different scales is important, the most rewarding endeavors likely exist where functionality can be correlated across those scales. Behavior at the scale of the organism within a given environment is readily observable. At the next scale down, the behavior of neurons witnessed by its spikes and structural alterations, is only observable now in part. Below the scale of the neuron, the mitochondria and other organelles move with a purpose and relation to activity of the neuron that has only been imagined, but is experimentally addressable.
Several speakers also mentioned the idea of a neural code. Spikes are a convenient metric for assessing brain activity, and we should seek to correlate their occurrence with behaviors on various scales mentioned above. They are a universal and non-local currency, among others in the brain, that inflates rapidly with stimulation and arousal. Unfortunately, the most logical conclusion for us must be that there is no code for spikes. Anyone attempting to observe and record a code for one neuron would probably find that it has, in short order, become unrecognizable, particularly in the context of the next. There are however constraints on spikes, and on neurons, and while considerable mention of the word was made at the meeting none were detailed in depth.
To formulate constraints on a system, at a level we don’t understand, we might look at constraints on other systems that we have some knowledge about. Neurons are neither wholly like ants, nor tress, but share some aspects of both. Similarly brains are neither like ant colonies, or forests, but shares some features in common. The most obvious constraint that comes to mind, and applies to these systems at every level, is energy. A subtle refinement of that is the concept of entropy generation. One key idea is that entropy generation at different scales, while proceeding according to as yet determined laws, need not necessarily maximize entropy at each point in time, but rather along paths through time.
A voice heard throughout the conference was that of Bill Bialek who diffusely observed that attempts to apply the laws of statistical mechanics to aspects of brain functions are not very productive because the brain is not at an equilibrium state. That would have been a good sentence to begin the conference perhaps rather than end it. Hopefully, the next NSF meeting will be a little more transparent to the public than the first. A more thorough webcast, with uploading to a media channel would be desirable to many who like to participate, as would a path for two-way communication on the issues. Mention should also be made of the efforts of a few neuroscientists peripheral to the BRAIN Initiative that have been maintaining important blog discussions, and metablog publication lists to track the progress made over last few months. This morning, NIH announced a new website has just been set up to provide additional public feedback.