3.3 | Colorado Pikeminnow diet composition
To estimate Colorado Pikeminnow contemporary and historical diet
composition, we used a Bayesian stable isotope mixing model in the R
package MixSIAR (Stock et al ., 2018). We used this model
to estimate the posterior probability distributions of the proportion
each fish and invertebrate prey contributed to Colorado Pikeminnow’s
isotopic signatures. The model accounts for uncertainty resulting from
multiple prey items, variation in signatures among prey and the
predator, and isotopic fractionation (trophic discrimination factors)
between predator and prey. However, for the mixing model to be
informative, isotope signatures among prey items need to be
statistically dissimilar (Stock et al ., 2018). Therefore, we
first estimated isotopic differences in prey items using two Bayesian
ANOVAs (one for δ13C and one for
δ15N) with vague priors and assessed significant
differences using posterior credible intervals with the R packagebrms (Bürkner, 2017). All prey items except Flannelmouth Sucker
and Speckled Dace were dissimilar by at least one isotope for both time
periods but we kept all species separate with the understanding that
Flannelmouth Sucker and Speckled Dace posterior probability
distributions would be similar and less uniquely informative. We
excluded Channel Catfish as potential prey because a laboratory feeding
study suggested it is unlikely to greatly contribute to Colorado
Pikeminnow’s diet (Gilbert et al ., 2018).
Because we lacked historical invertebrate samples, we used contemporary
invertebrate samples to estimate historical invertebrate isotopic
values. To do so, we assumed a mean isotopic shift in invertebrates from
the contemporary to historical period but retained the variability
observed in contemporary samples. Following this assumption, we
estimated historical invertebrate δ13C and
δ15N values by calculating the mean difference of
δ13C and δ15N between periods for
known invertivorous fishes (Flannelmouth Sucker, Speckled Dace, and
Fathead Minnow) and then subtracted this from each contemporary
invertebrate isotope value (mean value shift of δ13C =
+0.32 and δ15N = +3.14).
We used one mixing model for each time period to assess changes to the
resource use of Colorado Pikeminnow. To complete input data for each
mixing model, we used separate trophic discrimination factors (mean and
standard deviation) for potential fish prey (∆13C =
0.82 ± 0.10 and ∆15N = 2.76 ± 0.35) and invertebrates
(∆13C = 0.83 ± 0.29 and ∆15N = 2.93
± 0.23) as estimated by Franssen et al . (2017). Priors for the
proportion of prey in the diet for each model included 0.05 for
invertebrates with the remaining proportion distributed evenly among
fish prey species. We chose these prior probabilities because the most
comprehensive Colorado Pikeminnow diet study suggested the species
converts entirely to piscivory by 150 mm TL (Vanicek and Kramer, 1969),
although a more recent stable isotope analysis suggested invertebrates
contributed to the diet of fish as large as 300 mm TL (Franssen et
al ., 2019). We computed each mixing model with three chains of
3,000,000 iterations, discarded the first 200,000, and thinned chains by
retaining every 100th draw (Stock et al .,
2018). We again determined model convergence (all R-hat <1.01)
and report predicted posterior distributions of proportional
contributions of each prey item to the diet of Colorado Pikeminnow
between time periods. As with all prior analysis, we assessed changes
between periods by subtracting posterior probability distributions of
interests to determine the probability the difference was greater than
zero.