In their Correspondence (A different perspective on alternative cleavage and polyadenylation. Nat. Rev. Genet. https://doi.org/10.1038/s41576-019-0198-z (2019))1, Xu and Zhang state that we have written our recent Review on alternative cleavage and polyadenylation (APA) in health and disease (Alternative cleavage and polyadenylation in health and disease. Nat. Rev. Genet. 20, 599–614 (2019))2 under the presumption that APA is generally beneficial or adaptive. This is not the case.

As we and others have found in early genomics studies of alternative splicing, many alternative isoforms can be explained by ‘noise’ in the processing of closely spaced sites, whose relative efficiency of processing could be predicted from the sequence3. More broadly, it is no longer disputed that stochasticity is pervasive at all steps of gene expression4 and that deep sequencing has brought to light a large number of rare variants with little functional relevance. The extensive discussion around the results of the ENCODE project5 illustrates this point (for example, ref.6). Thus, in contrast with what Xu and Zhang suggest, we are mindful that many of the unique APA variants observed in a genome-wide study are likely the result of imprecision in the selection of processing sites. For this reason, in our studies, we cluster closely spaced RNA 3′ ends that very likely result from imprecise processing, and we typically also require that analysed sites be identified in multiple experiments.

Beyond this, it is important to stress that in the absence of a specific model, such as the one we had for splicing, it is unclear what sort of patterns of isoform expression to expect. Furthermore, ‘noisy’ expression may be, in fact, beneficial in some situations7,8. Establishing whether particular molecular events are adaptive is notoriously difficult, and we thus avoided making claims to that effect in our Review. Instead, we focused on patterns that, in our opinion, warrant further investigation. Specifically, we highlighted (1) the coherent shift toward usage of proximal poly(A) sites in cancers and its potential mechanisms and (2) some examples of changes in poly(A) site usage for which the relationship to the disease has strong experimental support. How a systematic shift towards proximal poly(A) site usage arises in cancers is unclear. We and others hypothesized that specific regulators are involved (for example, refs9,10,11,12), but the question is by no means answered. Once generated, either owing to a general defect in polyadenylation or to the action of specific regulatory proteins, APA isoforms can be co-opted in pathogenic processes, as the IRF5 example illustrates13. Identifying such cases, and especially predicting at what level they affect cellular function, is not trivial. Our efforts to approach this issue are described in a previous study14.

To conclude, we do not hold the view that the majority of unique APA isoforms that are observed in a deep-sequencing study have specific functions. While our Review highlights some known examples of functional isoforms, more remain to be uncovered, as even those APA isoforms that result from noisy processing can play roles in pathogenic processes. Further progress in the field requires approaches for identifying functional variants and the mechanisms that generate them.