Methods
Fish Collection
Fish were collected by pulsed-DC boat electrofishing (Midwest Lake Management, Inc. Missouri, USA) once every other month from April 2019 through February 2021 (a total of 12 bimonthly sampling events), with sampling at each site consisting of six, 600-sec transects. We sampled four sites (Figure 1)– three of these were located on the mainstem Tallapoosa River below Harris Dam (tailrace, Wadley, Horseshoe Bend), and one site was located upstream of Harris Reservoir to serve as an unregulated reference site (Lee’s Bridge). Output voltage was standardized between 700-900 volts with 100-120 pulses per second, and GPS coordinates were recorded at the start and end of each transect. A tow-barge electrofisher was used at the tailrace site given that it is inaccessible by boat; barge sampling consisted of one individual with the anode and 1-2 dip-netters wading alongside, with another individual pushing the barge itself. Barge electrofishing followed the same procedures, although a slightly lower voltage (500-700 volts) was used for safety.
For the first six sampling events, collected fish were returned to the lab for processing. For the remaining six sampling events, all initial processing was done in the field and fish were returned live to their sampling locations. In the lab, all collected fish were identified to species and up to 10 individuals of each species were weighed and measured. If more than 10 individuals of a given species were present in a transect, the remaining individuals were counted, and the group was bulk weighed. For processing in the field, collected fish were identified, measured (nearest mm TL), and weighed (nearest g) on site..
Data Analysis
Shannon’s diversity index (H) and total species richness were calculated for each site (with all samples combined) to allow comparison across sites and with previous studies (e.g. Travnichek & Maceina, 1994; Freeman et al., 2005). Given the potential for bias, estimates were generated both with non-native species included and omitted. Additionally, the proportion contribution by numbers for each fish family at each site was calculated. Catch per effort (fish/hour of electrofishing) of each species was calculated for each season where spring included March-May, summer included June-August, fall included September-November, and winter included December-February.
Overall assemblage structure was characterized using multivariate methods as in Kiraly et al. (2014), all of which were conducted using R statistical software using the Vegan and labdsv packages (R core team, 2020; Oksanen et al., 2020; Roberts, 2019). We fourth-root transformed CPE data to account for the extreme skew that was present in these data before using Bray-Curtis dissimilarity to conduct nonmetric multidimensional scaling on the transformed CPE data (NMDS; Goodsell & Connell, 2002; Kiraly et al., 2014). Bray-Curtis dissimilarity typically performs better than other measures of dissimilarity for ecological datasets (Kiraly et al., 2014; Orksanen et al., 2020). Function metaMDS in R was used to perform nonmetric multidimensional scaling. Several random starts were used with 50 iterations maximum and final dimensionality was determined by considering stress reduction and interpretability (Kiraly et al., 2014; Oksanen et al., 2020). Kendall’s tau correlation coefficient (T ) was calculated to determine the magnitude and direction of species correlations and MDS axes. Tau correlation coefficients were summed across species for each family to determine which family contributed the most to each MDS axis.
To quantify differences in fish assemblages across sites, we used a multiresponse permutation procedure (MRPP) based on the same Bray-Curtis dissimilarity (Kiraly et al., 2014). Only the three seasons (spring, summer, fall) during which sampling occurred at all sites were included. MRPP generates an A -statistic as well as a p-value, both of which must be considered to fully interpret results. The A statistic is a measure of effect size and describes within-group homogeneity compared to the random expectation; A= 1 if all units within groups are identical and A =0 if heterogeneity among groups equals the expected value by chance (McCune & Grace, 2002). If the null hypothesis is true, the p-value is the likelihood that the observed difference between groups is due to chance (McCune and Grace, 2002). Average dissimilarities both between and within groups were calculated to create a dendrogram describing the relationships between groups and to create group blocks. Additionally, MRPP generates a test statistic, δ, which is the overall weighted mean of group mean differences (Oksanen et al., 2020). MRPP serves as a hypothesis test of differences between groups of sampling units where p is the probability that δ is less than the observed value. A dendrogram was generated based on the Bray-Curtis dissimilarities where the vertical termination of each branch represented the within-group dissimilarity across seasons while the horizontal lines represented the dissimilarity between site blocks.
Indicator species values (IndVal) were calculated based on the formula given in Dufrene and Legendre (1997) and clarified by Roberts (2019) in the labdsv R package for each species given a significant overall MRPP result. This formula calculates the indicator values “d” of species as the product of the relative frequency and relative average abundance in clusters (Roberts, 2019) as follows:
\begin{equation} d_{\text{ic}}=\ f_{\text{ic}}*a_{\text{ic}}\nonumber \\ \end{equation}\begin{equation} f_{\text{ic}}=\ \frac{\sum_{j\in c}P_{\text{ij}}}{n_{c}}\nonumber \\ \end{equation}\begin{equation} a_{\text{ic}}=\ \frac{\sum_{j\in c}{x_{\text{ij}}/n_{c}}}{\sum_{k=1}^{K}{(\sum_{j\in c}{x_{\text{ij}}/n_{k})}}}\nonumber \\ \end{equation}
where: \(P_{\text{ij}}\) = presence/absence (1/0) of species i in sample j , \(x_{\text{ij}}\) = abundance of species i in sample j , and \(n_{c}\) = number of samples in cluster c , for cluster c ∈ K.
IndVal analysis accounts for species site specificity and fidelity and ranges from 0-1 (Dufrene & Legendre, 1997). The index equals one when a species is found in all sampling units (seasons) of a group (site). P-values were calculated for each species’ IndVal using a randomization procedure (Roberts, 2019). Species were randomly reassigned to sampling units and groups 1000 times and IndVals were recalculated to create a distribution of possible IndVal values from the given data. The p-value was the proportion of randomized IndVals that were greater than the observed value (i.e., the probability that the observed value was due to chance; Dufrene & Legendre, 1997).