The Cost of Cortical Computation

Abstract

Electrophysiological recordings show that individual neurons in cortex are strongly activated when engaged in appropriate tasks, but they tell us little about how many neurons might be engaged by a task, which is important to know if we are to understand how cortex encodes information. For human cortex, I estimate the cost of individual spikes, then, from the known energy consumption of cortex, I establish how many neurons can be active concurrently. The cost of a single spike is high, and this severely limits, possibly to fewer than 1%, the number of neurons that can be substantially active concurrently. The high cost of spikes requires the brain not only to use representational codes that rely on very few active neurons, but also to allocate its energy resources flexibly among cortical regions according to task demand. The latter constraint explains the investment in local control of hemodynamics, exploited by functional magnetic resonance imaging, and the need for mechanisms of selective attention.

Modeling the hemodynamic response to brain activation

“…Experimental studies comparing electrophysiological measurements with BOLD and CBF changes have found that the hemodynamic responses correlate better with local mean field potential, rather then local spiking rates, suggesting that the hemodynamic response is dominantly driven by input synaptic activity rather than output spiking activity (Lauritzen, 2001;Lauritzen and Gold, 2003;Logothetis, 2002;Logothetis et al, 2001). Theoretical analyses of the energy budget for neuronal signaling provide some support for this picture as well (Attwell and Laughlin, 2001;Lennie, 2003). The primary expenditure of energy is required to restore the ion gradients degraded during neural activation.…”

How networks communicate: propagation patterns in spontaneous brain activity

“…At the systems level, this paradigm shift was prompted by two major findings. First, although the human brain represents only 2% of total body mass, its intrinsic activity consumes 20% of the body's energy, most of which is used to support ongoing neuronal signalling ( [5][6][7][8][9], but see also [10]). Task-related increases in neuronal metabolism are generally small (less than 5%) when compared with this large intrinsic energy consumption (for a recent review, see [9]).…”

Cortical energy demands of signaling and nonsignaling components in brain are conserved across mammalian species and activity levels

“…13 C MRS findings in the human cortex have been generally consistent with the rat results (6). However, there remain questions as to how well the energy costs of specific subcellular processes needed to support synaptic transmission and conduction are conserved over different activity levels and/or across species.Recent bottom-up energy budgets for gray matter in the mammalian brain have attempted to understand the energetic costs of neuronal and glial electrical and neurotransmission events occurring in the neuropil (7,8) by calculating the ATP used per neuron for signaling (P s ) and nonsignaling (P ns ) events. In the awake cortex, the total ATP used per unit cortical volume per unit time (E tot ; in units of ATP/s per centimeter 3 ) was determined by multiplying the P s (in units of ATP/neuron per spike) and P ns (in units of ATP/neuron per second) parameters with cellular densities (η) and average cortical firing rates (<f>) to give signaling (Es) and nonsignaling (Ens) components,…”