The Mind-Body Disconnection

Our brains and bodies are linked. Anatomically, the spinal cord and nerves provide conduits for information to travel between the two. Sensory information about the environment and current state of our bodies is relayed from receptors up to the brain; information and “commands” are transferred from the brain down to our muscles to generate movements. Psychologically, our state of mind can have profound effects on our physical well-being. Stress has a myriad of manifestations in our bodies, from difficulty sleeping to a racing heartbeat and difficulty breathing. Yet more mysterious are things like the placebo effect, where a false perception of treatment can actually yield physical improvements in health, which highlight the complex interaction between our brains and bodies.

But how hard-wired is this link? Can we dissociate the activity of our brains from our actions and physical selves? One clear example of this disconnect is motor imagery. Researchers have found that imagining movements involves the same regions of the brain that would be used to execute those actions [See Jeannerod and Frak Current Opinions in Neurobiology 1999 for a review]. Yet, we can easily think about movements without so much as a muscle twitch. Clearly, there is some amount of flexibility in our brain-body links—a toggle switch of sorts.

The disconnect between neural activity and our bodies begs the question of volitional control of the brain. Can we ‘disconnect’ the actions of our neurons and the muscles they normally control at will? Is it possible to make movements without activating the neurons once associated with that action? If so, what are the limits and resolution of our control? Perhaps we can only control large regions of the brain and in very simple ways; maybe we have precise control of individual neurons and can orchestrate complex activity patterns.

In 1971, Eberhard Fetz and Dom Finocchio addressed some of these questions with a unique experimental paradigm often credited as one of the first brain-machine interfaces. Their Science paper (“Operant Conditioning of Specific Patterns of Neural and Muscle Activity”) was among the first to highlight the incredible flexibility of the motor cortex. The researchers recorded the firing rate of individual neurons in the motor cortex along with EMG activity from several muscles on the animal’s arm. The monkey was given visual feedback of a weighted sum of these signals via a simple level-meter. Their experiments then used operant condition to link different combinations of activations with reward. That is, when the monkey achieved an activity level above a set threshold (i.e. moved the meter all the way to the top), he was given a bit of juice as a treat. By manipulating the weights in their summed average, the researchers could shift the goal of the task. For instance, with zero weights on muscle activity and a positive weight on the neural activity, the monkey would get rewarded for increasing the firing rate of the neuron with no restrictions on muscle activity. Negative weights on the EMG signals, however, require him to increase his neural activity without contracting his muscles.

Fetz and Finocchio explored how well monkeys could volitionally control neural and muscle activity using this set up. They tested 4 different reinforcement conditions: 1) increased neural activity with no restrictions on EMG, 2) activating single muscles (while suppressing all others) without restricting neural activity, 3) increased neural activity with suppressed EMG and 4) contracting single muscles while suppressing neural activity. The researchers found that the monkeys could, with time, learn all of these associations. They were able to successfully control their neural and muscle activity to obtain rewards. What’s more, in conditions 3 and 4, the monkeys learned to dissociate the two.

As might be expected, the disconnect conditions (3 and 4) were harder for the animals to learn than the individual muscle or neuron control conditions (1 and 2). Indeed, the researchers were only able to demonstrate muscle control without neural activity in one monkey that showed extremely adept control abilities. But that this dissociation could occur at all is an impressive demonstration of our central nervous system’s flexibility. More amazing still is that before the conditioning training, these neurons’ firing was often correlated with the contraction of one or more of the recorded muscles. The monkeys were able to overcome these preexisting connections. Not only could they control individual neurons, they could disconnect their brains and bodies.

This work (and a collection of other Fetz papers in the early 1970’s that further explore this paradigm) adds new complexity to neuroscience efforts to understand the activity and computations of motor cortex. Much of motor cortex research focuses on simultaneously recording movement-related variables (EMG signals, joint positions, etc.) and firing rates of neurons. A temporal correlation between the two is often interpreted as a causal link—the neuron is thought to encode activation of a particular muscle or joint location. The evident flexibility between cortical activity and the periphery, however, makes such conclusions less concrete. As Fetz and Finocchio point out, correlations are “necessary but never sufficient evidence for a causal relation.”

Fetz’s demonstrations of neural malleability also raise profound questions in the field of brain-machine interfaces (BMIs). The ultimate goal of BMIs is to use neural activity to control an external device like a prosthetic arm (decoding), and to relay sensory information back to the brain via neural stimulation (encoding). Current research in the decoding-half of BMI has taken a very biomimetic approach—experimenters use the existing correlations between neural activity and movement variables to create decoding algorithms. They try to tap-in to the motor cortex’s control scheme. If successful, these approaches would allow users to move a prosthetic arm with the same neural control they would have used for their own limb. But, as we’ll be exploring in later posts, deciphering the neural activity of motor cortex is anything but simple. Developing accurate decoding algorithms that capture the intricate detail of our motor control requires significant advances in our understanding of motor cortex.

Given that our brains are inherently flexible, should BMI decoding take advantage of this instead? We are able to control individual neurons, independent of their original motor functionality. Rather than BMIs attempting to decode natural neural activity into movements, perhaps the relationship should be switched—our brains could learn to control a neuroprosthesis. Indeed, there’s growing evidence that the brain is adapting to the decoders used in BMI demonstrations. For example, the relationship between neuron firing and movement direction often shifts between natural limb movement and neural control of a BMI [see Carmena et alPLoS Biology 2003]. The decoding algorithms are based on existing neural circuitry, yet the brain appears to shift its activity significantly in order to achieve the task.

Perhaps these shifts are simply compensating for inaccuracies in neural decoding. After all, almost any BMI researcher will admit the inadequacies of current decoding methods. Or, maybe it is easier for the brain to learn some new, artificial relation between movement and neural activity. My intuition is that the ideal BMI approach will probably lie somewhere in between the completely artificial and biomimetic. Further research at both extremes is still needed to answer these lingering questions. The incredible flexibility of our brains adds yet another level of complexity to the motor cortex and brain-machine interfaces.

[Fetz Journal of Physiology 2007 is an excellent review of neural volitional control research and its applications to brain-machine interfaces. I highly recommend it to those interested in exploring this topic further.]