Losses in freshwater fish diversity might produce a loss in important ecological services provided by fishes in particular habitats. An important gap in our understanding of ecosystem services by fishes is the influence of individuals from different size classes, which is predicted based on known ontogenetic shifts in habitat and diet. I used twenty experimental stream mesocosms located on Konza Prairie Biological Station (KPBS), KS, USA to assess the influence of fish size on ecosystem properties. Mesocosms included two macrohabitats: one riffle upstream from one pool filled with consistent pebble and gravel substrate. There were four experimental and one control treatment, each replicated four times (N = 20). I used two size classes of Central Stonerollers (Campostoma anomalum) and Southern Redbelly Dace (Chrosomus erythrogaster). Five ecosystem properties were assessed: algal filament length (cm), benthic chlorophyll a (µg/cm2), benthic organic matter (g/m2), macroinvertebrate biomass (g/m2), and stream metabolism (g O2/m2/day1). Size structure of fish populations affected some, but not all, ecosystem properties and these effects were dependent upon species identity. Size structure of both species had effects on algal filament lengths where stonerollers of both size classes reduced algal filaments, but only small redbelly dace kept filaments short. A better understanding of the relationship between these prairie stream minnows and their small stream habitats could be useful to both predict changes in stream properties if species are lost or size structure shifts, and to redbelly dace populations, a Species In Need of Conservation.
Mesocosm design and preparation: I used twenty experimental stream mesocosms located on Konza Prairie Biological Station (KPBS), KS, USA (also described in Matthews et al., 2006; Martin et al., 2016) to test how fish size influence their influence on stream ecosystems. The experiment ran for 30 days (1 to 31 October, 2012). Mesocosms included two macrohabitats: one riffle (area = 0.8 m2) upstream from one pool (area = 2.5 m2) filled with consistent pebble and gravel substrate (mean area 40 cm2). Flow was generated by a trolling motor pulling water through a 15.2 cm diameter solid black corrugated plastic drainage pipe connecting the pool to the riffle. Prior to the start of the experiment, streams were drained of any incidental water (e.g., rainwater) and left dry for four full weeks. Next, I scoured and washed the substrate and inside of each mesocosm with a pressure washer, after which streams were drained and left to dry for several days. Finally, mesocosms were filled with spring water using an underground PVC pipe system from the same springs that naturally fill nearby Kings Creek. Mesocosms take 2-3 days to fill completely, and once all mesocosms were filled, trolling motors were turned on and mesh baskets with open tops were buried in the substrate.
Mesocosm homogenization: Allowing mesocosms to be fully dry for long periods of time and then pressure washing the substrate and walls promoted homogeneity among units. Furthermore, prior to beginning the experiment, flow and depth were measured using a Marsh-McBirney flowmeter and meter stick along three transects in each riffle (n = 9) and at along four transects in each pool (n = 19) within each reach for roughly 10 measurements per m2 for each habitat, and a mean flow and depth were calculated for each macrohabitat. Based on an analysis of variance (ANOVA), there was no evidence for significant differences among habitat types across mesocosms for flow (i.e. all riffles were similar to each other and all pools were similar to each other; F6,42 = 0.97, P = 0.46) or depth (F6,42 = 0.18, P = 0.98).
Experimental design: There were four experimental treatments and one control treatment, each replicated four times (N = 20 mesocosms). Experimental treatments were additions of fish based the two most abundant herbivorous species (stonerollers and redbelly dace) in Kings Creek, the stream located on the KPBS, and one control treatment. All fish (Ntotal = 2,336) were collected from nearby Kings Creek and stocked on 28 September, 2012 and given two days to acclimate. Each species was represented by two size classes, creating four fish treatments: small stonerollers (mean total length (TL) = 56 mm ± 3.9 mm standard deviation (SD), mean weight (W) = 0.76 g ± 0.17 g), large stonerollers (TL = 81 mm ± 2.7 mm, W = 2.35 g ± 0.24 g), small redbelly dace (TL = 31 mm ± 1.6 mm, W = 0.13 g ± 0.01 g), and large redbelly dace (TL = 53 mm ± 1.1 mm, W = 0.65 g ± 0.09 g). Weights of fishes were calculated from standard length (SL) values using the standard length-weight equation W = aLb where a = 0.0037 and b = 3.08 (K. Gido, unpublished data). Cutoff lengths for juveniles versus adults (stonerollers = 65 mm, redbelly dace = 55 mm) were based on published accounts of length at maturity for both species (Hubbs and Cooper, 1936; Settles and Hoyt, 1978; Edwards, 1997; Martin et al., 2013). All fish were measured (total length) prior to stocking mesocosms. Fish biomass was 15 (± 3) g/m2, 16 (± 1.3), ) for juvenile and adult stonerollers, respectively and 29 (± 12.4) and 24 (± 1.7) for juvenile and adult redbelly dace, respectively. The total biomass of these fishes varies significantly, and our biomass reflects the highest end of this natural variation (Martin et al. 2016). Fish deaths and TL of dead were recorded and replaced with similar sized fish during the acclimation period. The fifth treatment served as a control with no fish added to the streams.
Response variables: Five variables were selected as they comprise the basic ecosystem functioning and food web dynamics in streams and are based on extensive work in natural and experimental streams (algal filament length [cm], benthic chlorophyll a [µg/cm2], benthic organic matter [g/m2], macroinvertebrate biomass [g/m2], and stream metabolism [g O2/m2/day1], Evans-White et al., 2001; Bertrand and Gido, 2007; Murdock et al., 2011; Martin et al., 2016). These ecosystem properties were measured in pools at the end of the experiment. To quantify structural properties of the periphyton, algal filament lengths (AFL) were measured in centimeters using a meter stick along four transects in each pool (n = 19) within each mesocosm for 10 measurements per m2 for each pool. Filament lengths were measured as the length of the longest filament attached to a pebble that occurred on the transect point. To assess benthic chlorophyll a, macroinvertebrates, and organic matter, the three 10 cm2 mesh baskets were filled with pebbles and buried in each pool prior to the start of the experiment were collected. One basket was removed at the end of the experiment for chlorophyll a analysis: three pebbles were collected from the basket, chlorophyll a was extracted where the concentration of chlorophyll a was corrected for cross-sectional area of pebbles (see Sartory and Grobbelaar, 1984; Bertrand and Gido, 2007 for detailed methods on chlorophyll a extraction). The second basket was removed to quantify coarse and fine benthic organic matter (CBOM and FBOM, respectively) analysis. The basket was placed into 8 L of water from the mesocosm and the substrate vigorously stirred by hand. Next, a 500 mL subsample of the slurry was collected. The subsample was preserved using 10% formalin, taken to the laboratory, filtered through two mesh sizes (coarse [1 mm mesh filter] and fine [GF/F 47 mm microfiber filter]), dried at 60 °C, weighed, ashed at 450 °C, and re-weighed to determine ash-free dry mass (AFDM) (Wallace et al., 2007). AFDM was extrapolated using the surface area of the basket. Macroinvertebrates were sampled by a similar procedure to organic matter, where the third basket was placed in 8 L of water from the mesocosm and vigorously stirred by hand to release attached macroinvertebrates. Here, the resulting slurry was elutriated to separate inorganic substrate from organic matter and poured through a sieve (250 µm mesh) to capture macroinvertebrates. Samples were preserved in 10% formalin and taken to the laboratory where invertebrates were counted and identified to order or family (Thorp and Covich, 2001; Merritt et al., 2008). Chironomids were initially classified as Tanypodinae or non-Tanypodinae; however, because Tanypodinae chironomids constituted < 1% of sample biomass, all chironomids were combined into a single group. Lengths of macroinvertebrates were estimated to calculate biomass for each macrohabitat using standard length-mass relationships (Benke, 1984). Macroinvertebrate biomass was calculated by dividing the total biomass of the sample by the surface area of the basket. Finally, whole stream metabolism (gross primary production [GPP]) was measured based on fluctuations in dissolved oxygen content (g O2 m-2 day-1) of the water measured every 60 min from 05:00 to 17:00 using a YSI550A handheld dissolved oxygen sensor. The sensor was deployed in the same location in each mesocosm pool near the observation window. These rates were corrected for variation in temperature, dissolved oxygen saturation, light, atmospheric pressure, and stream mesocosm morphology based on the single-station modeling technique outlined in Riley and Dodds (2013) and Dodds et al. (2013). In short, this method uses a standard equation (Marzolf et al., 1994) to predict dissolved oxygen concentration. Modeled dissolved oxygen (see Martin et al., 2016 for model example) is compared to observed dissolved oxygen using the Solver function in Microsoft Excel (version 2007; Microsoft Corporation, Redmond), which uses a Newton search method to minimize the sum of squares of error between modeled and observed values by changing the basic rates of GPP, ER, and gas transfer coefficient (k).
Data analysis: Differences among treatments for each of the five response variables was tested by performing single-factor Kruskal-Wallis (K-W) test to assess treatment effects (α = 0.05) comparing juvenile stonerollers, adult stonerollers, and the control. Separate single-factor K-W compared juvenile redbelly dace, adult redbelly dace, and control. If the K-W test was significant, a Dunn’s post-hoc test was performed. All statistical analyses were run using R 4.2.2 (R Core Team, 2022) with library FSA (Ogle et al., 2022).