Your mistake is in assuming $-(a+b) = -a-b$, which isn't true in general.
However, if you add A5: $\lambda (x+y) =\lambda x+ \lambda y$ for $\lambda \in K$ ($K$ is the underlying field) and $x,y\in E$, and A6: $1_Kx = x$, and A7 $(\lambda +\mu)x= \lambda x+\mu x$ for $\lambda, \mu \in K$ and $x\in E$ then you can prove A4 from the rest, because then :
$0_Kx = 0_E$: indeed $0_Kx = (0_K+0_K)x= 0_Kx+0_Kx$ hence $0_Kx= 0_E$
$-x = (-1_K) x$: indeed $x+(-1)x = 1_Kx+(-1_K)x = (1_K-1_K)x= 0_Kx = 0_E$
$-(u+v) = -u-v$: follows from A5 and the previous line
$u+v= v+u$: $u+v -(v+u) = u+v -v-u = u-u = 0_E$.
As the other answers and comments mentioned, you can't prove A4 from A1, A2,A3 because that would mean any group is abelian, which is clearly not true (the group of permutations on more than $3$ elements is not abelian for instance, but also the group of affine invertible transformations of the euclidean plane, etc. there are many examples)
A5,A6 and A7 are axioms that are specific to vector spaces that make it so that indeed, you can prove A4 from the rest; but they add a specific structure to the group.