Following the experimental designs by Britton (2019) and Britton et al. (2018), we used 10 additive and substitutive treatments across a combination of allopatric and sympatric contexts and replicated each treatment three times (Table 1). Two control treatments used each species in allopatry (“Allopatry”; N=10; Table 1). One substitutive treatment combined these two species in sympatry (“Sympatry”; n =5 + 5, N=10; Table 1). Three additive treatments then used these two species in sympatry across three different abundances (“Interspecific competition”, n = 10 +10, 15+15, 20+20; N = 20, 30, 40; Table 1). Four intraspecific competitions used mud carp and mrigal carp in allopatry in higher abundances (“Intraspecific competition”; N =20, 30; Table 1). All the fish we used were juveniles and had been hatchery reared in May 2019 simultaneously for these two species. Their starting lengths were 3–3.5 cm and 1.5–2 cm, and they weighed 0.5–0.7 g (on average 0.63 g) and 0.1–0.2 g (on average 0.11 g) for mud carp and mrigal carp, respectively. The predicted stable isotope half-life for their dorsal muscle was 36 and 38 days for δ13C and δ15N, respectively (Thomas, Crowther 2015).
We completed the experiment using the treatments within enclosures in a large pond (50 m long × 50 m wide, 1–2 m deep) at the Pearl River Fisheries Research Institute, Chinese Academy of Fishery Sciences, Guangzhou, China. The enclosures were net cages (length: 2 m; width: 1 m; height: 2 m; mesh: 4 mm2, 10 mesh) fixed at the pond bottom using bamboo poles tied to the four top/bottom corners (Fig. S2). Our net cages prevent bird predation and fish ingress and egress but allow movements of small invertebrates and suspended materials. The 30 enclosures were numbered and placed neatly in the pond with a 2 m distance between them and 5 m distance from the pond edge. For each treatment, the enclosures were drawn randomly based on the enclosure number. During the experiment, we kept the pond without artificial disturbance; thus, sediments and organisms in the pond are the primary sources of food for mud carp and mrigal carp.
The experiment ran for one and half years (547 days) from 21 June 2019 to 19December 2020. This period enabled fish dorsal muscle to undergo more than 10 half-lives and would represent their diet in a pond (Thomas, Crowther 2015). During the experiment, the temperature in the larger pond ranged from 14.5 to 33.5 °C, as revealed by temperature loggers (Wdsen Electronic Technology Co., Ltd). All fish were weighed prior to release into the enclosures. In December 2020, all fish were recovered from the enclosures, euthanized and returned to the laboratory. For each individual, the standard length and total length were measured to 1 mm, and the weight was measured to 0.1 g. Then, a dorsal muscle sample was taken for stable analysis. In addition, the benthic surface grazing snail Cipangopaludina chinensis was sampled for stable isotope analysis.
The survival rate (SR) of each species in each cage net ranged from 0 to 1 but was not normally distributed. Thus, the difference in SR between species and treatments was examined using the Mann–Whitney test. The mean specific growth rate (SGR) in mass per replicate and species was calculated as ([lnWs+t]-[lnWs])/t (De Santis et al. 2020), where Ws = mean starting weight of the species in the replicate, Ws+t = mean end weight of the species in the replicate, and t = the duration of the experiment in days. By running the rlmer () function in the robustlmm package in R (Koller 2016), we performed robust linear mixed models to analyze SGR variation with species, treatments, and their interactions, in which the cage net number was set as a random factor, i.e., SGR ~ species * treatments + (1 | cage net number). The P value in robust linear mixed models was calculated for a two-tailed test following Cutler et al. (2021). Furthermore, multiple comparisons of the SGR between the allopatry treatment and each other treatment for each species were finished using the emmeans package (Lenth 2021). The P value in multiple comparison results was adjusted using the Bonferroni method.
Invaded wild fish communities
In four wild fish communities where they coexist, mud carp and mrigal carp were sampled. The sampling sites are located downstream of the Pearl River basin in China, i.e., the North River downstream (S1) and the central area of the Pearl River delta (S2-S4) (Fig. S3). At each site, we sampled fish using a combination of gillnets, fish traps, and cast nets (see details in Zhang et al. (2021)). Following their capture, they were euthanized and returned to the laboratory. In the laboratory, each of the individuals was identified to species and measured for the standard length (mm) and the total weight (g). For stable isotope detection, a small part (approximately 2 g) of dorsal muscle of each individual was cut and frozen at –20 °C.
Stable isotope analysis
Muscle samples of mud carp, mrigal carp and snails were freeze dried at –45 °C for 48 hours to constant weight in an FD-1-50 plus freeze-dryer (Beijing Biocool Experimental Instruments Co., Ltd.). The dried muscles were then ground to powder using a GT200 grinder (Beijing Grinder Instrument Co., Ltd.). Then, 0.5 mg of the ground muscle of each individual was encapsulated in a tin (Sn) capsule to ensure that they remained intact and did not leak or contaminate. The stable isotope ratios (δ13C and δ15N) and elemental concentrations of the tissues were detected using a continuous-flow isotope-ratio mass spectrometer (Delta V Advantage, Thermo Fisher Scientific, Waltham, MA, USA) coupled with an elemental analyzer (Flash EA 1112, Thermo Fisher Scientific, Milan, Italy).
The CN ratios of mud carp and mrigal carp ranged from 3.14 to 3.30, which was consistently below 3.5, indicating very low lipid content (Post et al. 2007); thus, the data were analyzed without lipid correction. Because both species in the mesocosm treatments were cultured in the same pond ecosystem, they had the same trophic baseline. δ15N was used to estimate fish trophic positions. Trophic position was calculated as follows: trophic position = (δ15N – δ15Nbase)/Δn+ λ, where δ15Nbase is the mean δ15N of the baseline organism of the benthic surface grazing snails C. chinensis, Δn is the enrichment in δ15N per trophic level and equals 3.4 ‰, and λ is the trophic position of the baseline organisms and equals 2 (Post 2002). Robust linear mixed models were constructed to analyze the effects of species, experimental treatment and their interactions on individual trophic position and δ13C through the robustlmm package (Koller 2016). Multiple comparisons of the dependent variables among each level of the treatments for each species were performed in the emmeans package (Lenth 2021).
To describe trophic niches, we selected the most commonly used stable isotope ratios, which are defined as quantitative indicators of trophic niches (Marshall et al. 2019). To determine the trophic niche of the studied fish across different experimental designs or different wild communities, we calculated the Layman et al. (2007) niche metrics of δ13C range (CR) and δ15N range (NR). The niche size was estimated using the corrected standard ellipse area (SEAc) and Bayesian standard ellipse area (SEAB) by running the SIBER package (Jackson et al. 2011). All data analyses were completed in R 4.1.1 (R Core Team 2020).