If $A$ is an abelian group, there are a number of characteristic subgroups of it that immediately come to mind: Any of the groups $nA$, $A[n]$ (the $n$-torsion), the full torsion subgroup $T(A)$, the maximal divisible subgroup $D(A)$ etc. Any of these examples, however, can be obtained by (possibly transfinite) iteration of forming verbal $(nA)$ and marginal ($A[n]$) subgroups and taking sums and intersections of those. Do you know any (specific or general) examples of characteristic subgroups that are not of this form?
To be precise: Let $\mathcal{E}(A)$ denote the smallest (nontrivial) family of subgroups of $A$ that is closed under sums, intersections and formation of verbal and marginal subgroups $-$ these characteristic subgroups I call expected. What are some unexpected characteristic subgroups?
Here is an example of what I am talking about: Consider the group $A=\mathbb{Z}/2^n\mathbb{Z}\times \mathbb{Z}/2\mathbb{Z}$ for $n\geq 3$. It is straightforward to see that $A$ contains exactly three distinct subgroups of order $2^r$ for $1\leq r\leq n$. Now, if both of these inequalities are strict we have the expectedly characteristic $2^{n-r}A$ and $A[2^{r-1}]$ among them. Hence, the third group (which is cyclic generated by $(2^{n-r},1)$) must be characteristic as well and is easily seen to lie outside of $\mathcal{E}(A)$.
If you happen to know that some class of groups cannot have such subgroups, you are more than welcome to share these as well. (For instance, underlying additive groups of vector spaces form such a class.)
Lastly, even though I posed the question only for abelian groups, you may feel free to interpret it more liberally.