extract variables from linear model

Hi,

I fitted a mixed linear model y = a + b + a*b + c + error, with c being 
the random factor:
lmefit <- lme(y ~ a * b - 1 , random = ~ 1 | c, na.action='na.omit')

Is there a way to omit some level combinations of the cross-term a:b? 
E.g. those that are not significant?

When I add the coefficients of a, b and a:b to get the combined effect 
without the effect of c, do then the standard errors of the coefficients 
also add?

best,

joerg