Experimental design:

The study was conducted in the Hengill valley, Iceland (N 64°03; W 21°18), which has been intensively studied over the past decade. The region consists of numerous spring-fed streams that occur within 1.5 km of each other and have similar physical and chemical properties, yet vary in mean annual temperature from 5-20 °C due to indirect heating of groundwater through the bedrock. The system thus acts as a space-for-time substitution to study eco-evolutionary responses of natural ecosystems at the endpoint of warming, but precludes transient responses that could occur during the warming process. Nevertheless, a whole stream warming experiment from the system showed that changes in community composition along the stream temperature gradient are similar to actual warming of a stream.

A field experiment was carried out in six geothermally heated streams in the Hengill valley. A split-plot experimental design was employed, with two levels of temperature (cold and warm) as the main plot crossed with two levels of a fish manipulation (presence and absence of brown trout, Salmo trutta) as the subplots within each of the main plots, for a total of four treatments with three replicates of each. There were three streams in each temperature category, yielding a mean temperature (± standard deviation, SD) over the course of the experiment of 6.8 ± 1.4 °C for the cold streams and 13.5 ± 2.4 °C for the warm streams.

Fish were manipulated by constructing three fences in each stream from metal rebar and extruded plastic netting (10 mm mesh), with each fence separated by a 15 m reach. The average width of the streams in the experiment was 1.5 m, equating to enclosure sizes of approximately 22.5 m^2. Whilst the mesh size controlled the presence or absence of fish in the experiment, it permitted drifting invertebrates to pass through the fences, which could have led to some upstream (i.e. non-treatment) biomass influencing each of the reaches. However, this is a natural process in flowing waters and must be considered a source of unavoidable “background noise” to the experimental treatments.

Electrofishing was carried out on 20th August 2012 to remove any pre-existing brown trout from the experiment, with at least three passes performed on every reach. Ten fish were removed from one of the warm streams (IS6) and six were removed from one of the cold streams (IS14). Fish have also been recorded in IS3 (warm) and IS13 (cold) in past sampling of the system, but never in IS9 (warm) or IS11 (cold). Fish for use in the experiment were electrofished from another stream in the system, IS12, which had an intermediate temperature to the experimental streams (10.5 °C). Whilst this precludes adaptive responses of fish to the temperature treatments over the longer term and thus equates more to shorter term, transient responses to temperature, it was a necessary compromise since fish were not already present in every experimental stream in the system. A total of 42 fish with a mean ± SD fork length of 17.5 ± 3.3 cm were captured on 22nd August and distributed evenly among the “Fish” reaches in the six streams (i.e. seven fish per stream). The experimental density of ~0.3 m^-2 was chosen to match typical densities of brown trout in the catchment. Note that the “Fish” reaches were always established downstream of the “No fish” reaches to minimise the chances of fish kairomones eliciting anti-predator behaviour amongst benthic invertebrates in those treatments.

An electrofishing survey was conducted on 7th September after heavy rainfall to ensure that the experimental treatments were still intact. No brown trout were found in any of the “No fish” reaches, although some fish were missing from the “Fish” reaches in each of the other streams. To restore the experimental densities, two new fish from IS12 were added to IS3, 6, 11, and 14, while three fish were added to IS9 and 13. The experiment was terminated after exactly five weeks on September 26th. Electrofishing was again performed, with no brown trout found in any of the “No fish” reaches and 5, 6, 3, 5, 4, and 6 fish recovered from IS3, 6, 9, 11, 13, and 14, respectively. These fish were returned to the stream from which they were originally captured (IS12).

Sampling:

Invertebrates and benthic algae were sampled on 21st August and 25th September, i.e. the day before the fish were added to the experiment at the beginning and the day before they were removed from the experiment at the end. Invertebrates were collected by taking five Surber samples (14 × 13.5 cm quadrat; 250 μm mesh) per experimental reach and preserving them in 70% ethanol. Benthic algae were sampled by taking two scrapes of a 2.3 × 3.5 cm micro-quadrat from each of five rocks per reach. We preserved one scrape in stream water with 2% Lugol’s solution for later identification of diatoms. We preserved the second scrape in 96% ethanol, immediately storing it in a black plastic bag, which was placed in a dark fridge at 4 °C upon returning to the lab. Chlorophyll pigments were allowed to extract for an 18-hour period before analysis on a DR5000 Hach-Lange spectrophotometer following established methodologies, including a correction for phaeophytin.

We quantified decomposition in the experiment using coarse mesh (5 mm) and fine mesh (250 μm mesh) litter bags. We placed 3.00 g of dried grass (Carex spp.) into each bag before sealing them. Note that there are no trees at our study site, so grass represents the major allochthonous input to the streams16. Three metal rebars were hammered 20 cm into the sediment in each experimental reach, with one coarse and one fine mesh litter bag attached near the base of each rebar with a cable tie. The litter bags were placed in the streams on 22nd August and collected on 26th September. The grass was removed from each litter bag, dried at 80 °C for 48 hours, and weighed. Litter breakdown rates [mg day-1] were calculated as the initial minus final weight of grass in the litter bags divided by the duration of the experiment (35 days). Microbial decomposition was taken as the breakdown rate in the fine mesh bags, while invertebrate decomposition was the difference between the breakdown rate in each pair of coarse and fine mesh bags.

Biomass estimation:

Invertebrates and diatoms were identified to species level in line with previous studies from the Hengill system. The biomass [mg m^-2] of each species was estimated as mean body mass multiplied by total abundance. The abundance of every invertebrate species was enumerated using a CETI Vari Zoom 10 stereo microscope and scaled by Surber quadrat area (m^-2). Photographs of every individual identified were taken with a CETI 1.3 megapixel digital USB camera at 80-800× magnification and one linear dimension was measured in Image J (n = 11,053 individuals from 31 species). Published length-weight relationships were used to estimate dry body mass [mg] from the linear measurements.
Diatom frustules were cleared of organic matter with 65% nitric acid, dried, and mounted on slides with naphrax. Abundances were estimated by counting the number of individuals of each species along a 15 × 0.1 mm transect of each slide, ensuring a transect contained at least 150 individuals. The sample dilution, transect area, and micro-quadrat area were used to calculate the abundance of each species (m^-2). Photographs of every individual diatom were taken with the above camera mounted on a CETI Magnum-B phase contrast microscope at 1,000× magnification. Two linear dimensions were measured in Image J (n = 25,248 individuals from 64 species). Every diatom species was assigned a shape, and cell biovolume (μm3) was calculated according to associated formulae. Cell carbon content was estimated from published cell volume to cell carbon relationships and converted to dry mass [mg] assuming an average carbon by dry weight content of 19% per cell.

Food web construction:

We aimed to construct five localised food webs for each experimental reach by pairing the list of invertebrate species from each Surber sample with the list of diatom species from the nearest rock scrape. Food webs were constructed from an established database of 49,324 gut content observations from the Hengill streams, supplemented with literature-based feeding links when yield-effort curves revealed the diet of consumers to be incomplete. 58% of feeding links were directly observed in each specific stream, 35% of links were inferred from direct observations in other streams in the Hengill system, and just 7% of links were inferred from the literature. A food web link was only included in the current study if both species were found in the paired Surber and rock scrape samples. This procedure ensured that temperature could alter the dietary preferences of the consumers, rather than solely inferring diet from the presence or absence of potential prey. Nevertheless, there was no comparison of invertebrate diets in the presence and absence of brown trout, precluding the possibility that behavioural factors such as fear of predation could have altered their diets. Localised food webs were analysed using the ‘cheddar’ package in R 3.5.0. We computed the following metrics: species richness (S), link richness (L), linkage density (L/S), directed connectance (L/S2), mean trophic level (using the ‘ShortWeightedTrophicLevel’ function), and the ratio of consumers to resources (i.e. the number of consumer species divided by the number of resource species).