Chris,

Sorry for the delayed response. I'll also repost on the FSL/SPM lists.

As an FSL user you have 2 options: 
(1) Form a series of interactions, 1 for each condition by multiplying the condition by the seed region. Then form the model using N interaction terms, N task regressors, the seed region regressor, and any covariates you had in the original task model.

(2) Redo the first-level analysis in SPM.

The issue isn't so much doing the deconvolution as much as that the deconvolution portion is heavily integrated into the code.
As far as I know, no one has looked at the effect of deconvolution on event-related PPI.

Hope this helps.