Assigning to a field in a multi-dimensional struct array with indices in a cell array

I am curious about an inconsistency that I observe in the following assignment to a multi-dimensional struct array when the indices are generated from a comma separated list expansion of a cell array.

V(1:2,1:2,1:2) = struct('a',2);

U = V;
[U(:,2,:).a] = deal(3);
disp({U.a}); % (correct) output: [2]    [2]    [3]    [3]    [2]    [2]    [3]    [3]

I = {':',2,':'};
U = V;
[U(I{:}).a] = deal(3);
disp({U.a}); % (incorrect) output: [3]    [3]    [3]    [3]    [3]    [3]    [3]    [3]

J = {1:2,2,1:2};
U = V;
[U(J{:}).a] = deal(3);
disp({U.a}); % (incorrect) output: [3]    [3]    [2]    [2]    [2]    [2]    [2]    [2]

Any explanation as to what is going on? In my code, I need such an assignment where the index range to the struct array need to be dynamically generated. The only way that I have got things to work right now is to use a temporary struct to split the one step assignment to 3 steps as below.

U = V;
T = U(J{:}); % or T = U(I{:})
[T.a] = deal(3);
U(J{:}) = T;
disp({U.a}); % (correct) output: [2]    [2]    [3]    [3]    [2]    [2]    [3]    [3]