The composition and dynamics of plankton communities are critically affected by human-induced environmental changes. We analysed 33 years of phytoplankton monthly data collected in Lake Zurich (Switzerland), assigning organisms (genus level) to taxonomic groups (class, family), Reynolds associations and size categories. The aim was to understand how eutrophication and climate change have influenced taxa co-occurrence patterns within and between groups over the lake water column (14 depths, 0–135 m), using null-models to test for non-random spatial (depth) assembly. We found that the whole community showed high taxa co-occurrence levels, significantly deviating over time from random assembly concurrently with lake warming and reduced nutrient loading. This pattern was driven mostly by the depth structure of metalimnetic assemblages during summer and autumn. The prevalence of non-random spatial patterns changed for different taxonomic and functional groups, with only few significant deviations from null-model expectations. Within taxonomic and functional groups (particularly Classes and size categories), the frequency of spatial overdispersion of taxa decreased over time while the frequency of clustering increased. Our data suggest that the relative importance of mechanisms determining phytoplankton metacommunity dynamics have changed along with environmental gradients shaping water column structure.