1. Using nurse shrubs to improve tree seedling establishment in stressed environments is a common practice in forestry. Recent refinements of the stress-gradient hypothesis suggest that positive nurse effects occur under intermediate stress and decline in the harshest conditions. Additionally, indirect facilitation is expected in low-stressed/productive systems according to the initial stress-gradient hypothesis. However, there have been fewer investigations into the use of nurse shrubs to decrease herbivore impacts in stressed systems compared with the role of nurses on stress alleviation. This raises the need to clarify conditions in which the use of nurse shrubs would be most appropriate. 2. We conducted a field experiment in the sand dune forest of southwest France characterised by severe summer water shortages and the presence of different types of wild herbivores. In 2015 and 2016, Pinus pinaster seedlings were planted under shrubs (Arbutus unedo) or in shrub-removed locations, in fenced/unfenced areas. Survival, cause of mortality, and water stress were monitored for one year after plantation. 3. Summer drought was the main cause of mortality. Water stress increased in both years as summer progressed. We found direct facilitation only in July 2016 at intermediate water stress, due to a decrease in vapour pressure deficit under shrubs. These positive effects declined in late summer when stress was at a maximum. Rodent damage occurred under shrubs in spring whereas ungulate browsing was observed in open areas in autumn/winter. Overall seedling survival was equivalent under or outside shrubs during the first year, but was better under shrubs in the second one. 4. Synthesis and applications: Our results partially validate the refinement of the stress-gradient hypothesis, with a decline of positive interactions in the most water-stressed conditions. This draws attention to the difficulty of using shrubs as nurses to improve tree regeneration in the most xeric systems. In addition, we show that indirect associational effects, both positive and negative, could be as important as abiotic stress alleviation in a stressful ecosystem, and using shrubs to facilitate seedling survival could be efficient depending on the type of herbivore involved. This highlights the fact that considering plant-plant interactions can help in the design of management options, providing that up-to-date ecological theories are considered and that the different biotic and abiotic constraints are accurately evaluated.