That holobionts are units of selection squares poorly with the observation that microbes are often recruited from the environment, not passed down vertically from parent to offspring, as required for collective reproduction. The taxonomic makeup of a holobiont’s microbial community may vary over its lifetime and differ from that of conspecifics. In contrast, biochemical functions of the microbiota and contributions to host biology are more conserved, with taxonomically variable but functionally similar microbes recurring across generations and hosts. To save what (...) is of interest in holobiont thinking, we propose casting metabolic and developmental interaction patterns, rather than the taxa responsible for them, as units of selection. Such units need not directly reproduce or form parent-offspring lineages: their prior existence has created the conditions under which taxa with the genes necessary to carry out their steps have evolved in large numbers. These taxa or genes will reconstruct the original interaction patterns when favorable conditions occur. Interaction patterns will vary in ways that affect the likelihood of and circumstances under which such reconstruction occurs. Thus, they vary in fitness, and evolution by natural selection will occur at this level. It is on the persistence, reconstruction, and spread of such interaction patterns that students of holobiosis should concentrate, we suggest. This model also addresses other multi-species collectively beneficial interactions, such as biofilms or biogeochemical cycles maintaining all life. (shrink)