Deep seafloor communities, especially those from the ice-covered Arctic, are subject to severe food limitation as the amount of particulate organic matter (POM) from the surface is attenuated with increasing depth. Here, we use naturally occurring stable isotope tracers (δ15N) to broaden our rudimentary knowledge of food web structure and the response of benthic organisms to decreasing food supplies along the bathymetric transect of the deep-sea observatory HAUSGARTEN. Encompassing five trophic levels, the HAUSGARTEN food web is among the longest indicating continuous recycling of organic material typical of food-limited deep-sea ecosystems. The δ15N signatures ranged from 3.0 for Foraminifera to 21.4 (±0.4) for starfish (Poraniomorpha tumida). The majority of organisms occupied the second and third trophic level. Demersal fish fed at the third trophic level, consistent with results from stomach contents analysis. There were significant differences in the δ15N signatures of different functional groups with highest δ15N values in predators/scavengers (13.2 ± 0.2) followed by suspension feeders (11.2 ± 0.2) and deposit feeders (10.2 ± 0.3). Depth (=increasing food limitation) affected functional groups in different ways. While the isotopic signatures of predators/ scavengers did not change, those of suspension feeders increased with depth, and the reverse was found for deposit feeders. In contrast to the results of other studies, the δ15N signatures in POM samples obtained below 800 m did not vary significantly with depths indicating that changes in δ15N values could not be held responsible for the depth-related δ15N signature changes observed for benthic consumers. However, the δ15N signatures of sediments decreased with increasing depth, which also explains the decrease found for deposit feeders. Suspension feeders may rely increasingly on particles trickling down the HAUSGARTEN slope and carrying higher δ15N signatures than the decreasing POM supplies, which elevates the δ15N value of their tissues. Our results imply that a depth-stratified approach should be taken to avoid a misinterpretation of data obtained at different depths.