- 3D reconstruction of the monkey brain showing the activated (V1) and deactivated regions during ES. Area V2 and much of the extrastriate cortex show BOLD decreases as a result of V1 input loss.
The fMRI signal predominantly reflects regional perisynaptic activity, i.e. the classical events of synaptic transmission with its respective population excitatory or inhibitory postsynaptic potentials, as well as a number of integrative processes, including somatic and dendritic spikes with their ensuing afterpotentials, and voltage-dependent membrane oscillations [1, 4, and 5]. To some extent these processes underlie different components of the comprehensive neural signal. Specifically, dendro-somatic integrative processes determine the power modulation of various low-frequency bands (<150Hz) of the field potentials (LFPs), while the population spiking of stimulus/task-selective neurons (multiple-unit activity, MUA) is captured by higher frequency bands (>1kHz) or by the activity of isolated units, although these are of little help in trying to understand the changes in the fMRI signals that typically reflect mass action.
Because LFP and MUA correlate with each other, the details of the relationship between the fMRI signal and the underlying neural events are studied best in cases of a response dissociation between the two. Such dissociations can be induced by certain types of visual stimulation, by pharmacological manipulations , or by electrical microstimulation. The first type was reported in the primary visual cortex of both anesthetized and alert animals  and most recently in area V5 (MT). BOLD responses could be elicited in the absence of any MUA activity. A similar dissociation was found during electrical stimulation (ES). We recently combined ES with fMRI  to study in vivo connectivity. The technique was also used to study ES-induced signal propagation in cortex. Stimulation of dLGN induced strong BOLD activation in V1, whose thalamocortical synapses (input and local processing) are driven by the electrical pulses despite the fact that the V1 output is entirely suppressed.
Input-output dissociations like the ones just described are not rare phenomena that can only be induced experimentally. When cortical areas are “feedforward” stimulated with simple sensory stimuli then field potentials and spiking are correlated, barring certain stimulation conditions like those described above. Activations due to cognitive processing, on the other hand, might be dominated by top-down and neuromodulatory signals that could increase the excitation-inhibition balance of cortical microcircuits without necessarily causing a concomitant increase in the spiking of task-selective neurons. The effects of neuromodulation on large populations can be studied in a number of ways, including investigation of spontaneous cortical activity. Finally, the study of functional neurovascular coupling may profit greatly from attempts to calibrate the BOLD signal, in order for the latter to reflect as closely as possible the actual energy metabolism.