A self-loop on a state labeled with a will neither satisfy S1 nor S2 in the first example. The subformula !a& Xa holds on a path if in a state !a holds and on the successor state a holds. Hence, a must switch from false to true.
The formula a <-> X!a holds whenever there is a change on the truth value of a, i.e., if a switches from false to true as the above formula, but also if a switches from true to false. This is so, since a <-> X!a is equivalent to a&X!a | !a&Xa which contains the former formula. The past version of the X does not make a big difference here.
I guess the latter needs a bit more explanation: Looking at EXF((PWX!a)<->a), we can rewrite this to EFX((PWX!a)<->a) which is equivalent to EF(!a<->X a) and that is in turn equivalent to EF((!a&Xa | a&!Xa)), and also to EF(!a&Xa) | EF(a&!Xa) which makes S2 equivalent to S1 | EF(a&!Xa).
In the second case, we have EF((PWX!a)<->a) instead of EXF((PWX!a)<->a) and these two formulas are not equivalent as can be seen by the discussed structures. Note that PWX p holds initially for any p, and since a also holds on the single path in the second structure, (PWX!a)<->a holds initially on that single path, and thus also F((PWX!a)<->a) holds there. However, variable a never changes its value.
