Why do the $U$ and $V^T$ have to be orthogonal matrix in SVD ? is it a definition or a theorem?

Question

$M=U\Sigma V^T$,and $U$ and $V$ are both orthogonal matrix ,that is ,$V^T=V^{-1}$, and $U^T=U^{-1}$

Why do the $U$ and $V^T$ have to be orthogonal matrix in SVD ? is it a definition or a theorem?if it is a theorem,can anyone prove it?