In the prolonged absence of catastrophic disturbance, ecosystem retrogression occurs, which is characterized by declining soil nitrogen (N) and phosphorus (P) availability, increasing plant and soil N to P ratios, and reduced plant biomass and productivity. It is, however, largely unknown as to how the effects of plant communities on soil nutrients change during retrogression or might contribute to declining nutrient availability as retrogression proceeds. We studied a well‐characterized system of 30 lake islands in northern Sweden that collectively represent a 5,000‐year post‐fire retrogressive chronosequence. For each island, we established an experiment that involved full factorial removal of three plant functional groups (tree roots, dwarf shrubs and mosses), and of three species of dwarf shrub (Vaccinium myrtillus, V. vitis‐idaea and Empetrum hermaphroditum). After 19 years, we took various measures of soil N and P availability, and measured foliar N and P for each dwarf shrub species, for each plot in the experiment. Although plant removal effects (and particularly removal of tree roots, shrubs and Vaccinium species) on below‐ground N and P measures sometimes changed during retrogression, this seldom happened in a way that explains the decline in nutrient availability and increase in N to P ratios that characterize ecosystem retrogression. The only exceptions were that the positive effects of tree roots on soil mineral N and P, and of V. myrtillus on soil mineral P, declined during retrogression. Plant removal effects on community‐level measures of shrub N and P varied greatly across the chronosequence, but these effects again did not align with the changes in soil nutrient availability or N to P ratios that characterize ecosystem retrogression. Synthesis. Our results suggest that retrogression, and associated changes in nutrient availability and soil N to P ratios, is driven mainly by longer‐term pedogenic processes as opposed to shorter‐term effects of plant communities on soil N and P availability. More generally, they illustrate the value of long‐term and large‐scale experimental manipulations of plant communities for showing how effects of biodiversity loss on ecosystem properties vary across contrasting ecosystems.