First question: How would one prove that by removing free (unbound) variables from lambda calculus, and allowing only bound variables, its power is not reduced (it is still Turing-complete) ?

Second question: Is the proposition given above really true? Is lambda calculus sans free variables really Turing-complete?