The fusion calculi are a simplification of the pi-calculus in which input and output are symmetric and restriction is the only binder. We highlight a major difference between these calculi and the pi-calculus from the point of view of types, proving some impossibility results for subtyping in fusion calculi. We propose a modification of fusion calculi in which the name equivalences produced by fusions are replaced by name preorders, and with a distinction between positive and negative occurrences of names. The resulting calculus allows us to import subtype systems, and related results, from the pi-calculus. We examine the consequences of the modification on behavioural equivalence (e.g., context-free characterisations of barbed congruence) and expressiveness (e.g., full abstraction of the embedding of the asynchronous pi-calculus).