Functional trait compositions of communities can adapt to altered environmental conditions ensuring community persistence. Theory predicts that the shape of trade-offs between traits crucially affects these trait dynamics, but its empirical verification from the field is missing. Here, we show how the shape of a defense-growth trade-off governs seasonal trait dynamics of a natural community, using high-frequency, long-term measurements of phytoplankton from Lake Constance. As expected from the lab-derived concave trade-off curve, we observed an alternating dominance of several fast-growing species with intermediate defense levels and gradual changes of the biomass-trait distribution due to seasonally changing grazing pressure. By combining data and modelling, we obtain mechanistic insights on the underlying fitness landscape, and show that low fitness differences can maintain trait variation along the trade-off curve. We provide firm evidence for a frequently assumed trade-off and conclude that quantifying its shape allows to understand environmentally driven trait changes within communities.