Actually, what is a gain-of-function mutation?

For more than a century, scientists have worked to characterize, understand, and predict the consequences of mutations. For almost as long, scientists-always on the lookout for general principles-have categorized these mutations, hoping that putting them into labeled boxes might help reveal the molecular logic that governs mutational effects. Here, I will dive into one of these boxes, labeled "gain-of-function", a term that will ring familiar to undergraduates, (clinical) geneticists, and virologists alike. I will emerge from the box with a profound sense of bewilderment and the conclusion that its contents appear to have very little in common. What is a gain-of-function mutation? What do we know (or can reasonably assume) about a mutation once it has attracted this label? Do gain-of-function mutations share anything in common in terms of their molecular features or the consequences they cause? I will argue that the answers to these three questions are "I don't know," "not much," and "not really," and that the term gain-of-function tells us rather little. Worse, it often misleads our intuition regarding what a given mutation is or does. I will suggest that this is because the gain-of-function label has historically been applied, with liberal abandon, across different levels of biological complexity, from the behavior of individual proteins, to protein complexes, to cells, to whole-organism physiology. I will discuss the implications (all bad…) of this heterogeneous labeling history for recent efforts to train machine learning algorithms to discriminate different types of mutations. Above all, I hope to highlight that the myriad ways in which mutational effects can percolate through biological systems often defy easy categorization and that, while classifying things is often useful, it is best not to forget that molecular biology is a glorious mess.