Print

---

Dear Rik,

Since Christoph and Hanneke have already addressed most of your questions, I will focus on one aspect, i.e. whether events can be used to define modulatory inputs in DCM.

It is true that we normally think of modulatory effects on connection strength as being somewhat extended in time.  This may be because one of the most well-known (and probably most important) mechanisms for this is the action of neuromodulatory transmitters such as dopamine; these modulatory mechanisms can have a very quick onset (e.g. milliseconds), but are often enduring in time (e.g. seconds) which may result from volume transmission and other anatomical and biochemical constraints.  

Nevertheless, there are also various mechanisms that can mediate very brief and transient changes in coupling.  These include, for example, changes in postsynaptic dendritic excitability due to previous synaptic inputs or due to back-propagating action potentials; such mechanisms alter the opening probability of voltage-sensitive ion channels, e.g. at glutamatergic synapses, and thus postsynaptic responses to presynaptic inputs.  They are also relevant for synaptic “gating” that is often observed in neural systems (for a more in-depth discussion, you could consult my paper on nonlinear DCM).  

Therefore, it is conceptually perfectly reasonable to ask whether short-lived inputs (4 seconds in your example) or even single events lead to transient changes in effective connectivity.  However, whether you have sufficient statistical sensitivity to detect such changes depends entirely on the temporal structure of your inputs.  First, any modulatory input will only have a discernable effect if it acts during a time period in which activity is conveyed between regions.  (Note that the effect that a single driving input has on remote regions can be quite protracted in time, depending on regional decay rates and the resulting shape of the neuronal kernels.)  Second, your statistical efficiency of detecting such brief transient modulations is constrained by exactly the same thing as in event-related fMRI analyses, i.e. the shape of the transfer function.  In other words, if the temporal pattern of your modulatory inputs leads to neuronal signal variations (in the target area) that live in high frequency bands, you are unlikely to detect any such effect because they will be washed out in the BOLD response.  If, however, events are clustered or grouped appropriately, the statistical efficiency of detecting any effects increases.  This is analogous to considerations of statistical efficiency in conventional event-related fMRI.  The only difference is that in conventional event-related fMRI analyses the events specify transient episodes of neuronal activity directly, whereas when used as modulatory inputs in DCM (B matrix), they model changes in postsynaptic responses (of the target area) whose magnitudes depend on presynaptic activity (in the source area).

Very best wishes,
Klaas