**2. Materials and Methods**

*2.1. Data Sources and Field Protocol*

The literature data included three populations of *P. castaneus* and two populations of *P. niger* studied in the Niger Delta, Nigeria [7,9]. The original data came from additional four populations of *P. castaneus*, two of *P. niger*, and one of *P. nanus*. The geographic positions of the various study areas are presented in Figure 2.

**Figure 2.** Map of the countries involved in the present study, showing the location of the sites where the diets of *Pelusios* spp. were studied. In Togo and Benin, only *Pelusios castaneus* were analyzed for this study; in Nigeria both *P. castaneus* and *P. niger*, in Zambia only *P. nanus* and in South Sudan only *P. adansonii*. Land use categories are also shown on the maps. Localities for both the literature and original data are pooled in this map.

Overall, original field studies were conducted between 1996 and 2020, in some savanna sites as well as in rainforest sites, in both perennial waterbodies (rivers, streams, lakes) and in temporary ponds (Figure 3). Concerning free-ranging turtles, the methodology used for obtaining the food items were carefully described in [9,15]. All captured turtles were sexed by examining their plastron and caudal shape, measured for curved carapace length, curved carapace width, plastron length and plastron width, and permanently individually marked by unique sequences of notches filed into the marginal scutes. Since the various morphometric measurements were significantly autocorrelated in all populations (*p* < 0.0001), we retained only the curved carapace length for our analyses involving turtle body size.

**Figure 3.** Typical habitats of *Pelusios* spp. in tropical Africa. (**a**,**c**) habitat types of *Pelusios adansonii* along the White Nile river course in South Sudan; (**b**) habitat type of *Pelusios niger* and *Pelusios castaneus* in Southern Nigeria; (**d**) habitat of *Pelusios nanus* in Zambia.

The dietary study is based on both the stomach analyses of a few dead specimens (generally offered in bush-meat markets or as roadkills), and stomach-flushing (as described in [16]) and fecal pellet analyses of living specimens (specimens were singly kept in plastic boxes until defecation occurred) [16]. Specimens captured into baited traps [15] were not included in the analyses, so we included in this study only those turtles that did not eat "artificially attractive" food during our studies. No specimen was killed or injured by the researchers. Each turtle was sexed by examining tail and plastron morphologies, measured for carapace length, and individually marked by scute notching and with a painted number on the carapace for identification and for excluding risks of data pseudoreplication. Food items or feces of each individual were deposited separately in test tubes (under alcohol) for laboratory dissection, and the reference number of each tube corresponded to the painted number on the turtle carapace. An example of the painted numbering on the turtle carapace is given in Figure 1c.

We included algae in our diet data analyses, although these may have been ingested secondarily by turtles, at least in some instances. Feces were examined under binocular microscope for the identification of any food items.

The diet composition of each population was described as the percentage of stomachs containing a given food item and not on the basis of the total number of items of each food category in the stomachs. This was necessary because it is often impossible to count the number of items from a feces analysis, and because the easiness of identification varied considerably by the various types of food item. Obviously, this methodology may have some shortcomings when comparing the data across studies. In fact, the stomach content corresponds to less processed material than in the feces, thus it could show different % when comparing the resources consumed. Therefore, we cannot exclude that the data obtained for the various dietary studies could be, in part, different also because of the different methodologies applied for gathering the food data, with the evidence on some consumed taxa that could have degraded and not be detected in the feces.

#### *2.2. Statistical Analyses*

We evaluated whether our sampling effort captured the true food items' richness and diversity within each study population by building a rarefaction curve for food type discoveries at each site, using the software PAST 4.0.

Generalized Linear Models (GLM, see [17]) were used to test the relationship between body size and vegetation cover in the diet of four species of turtles (*P. niger*, *P. castaneus*, *P. adansonii*, *P. nanus*). In the models, three different vegetation classes (savannah, derived forest and forest) and three body size classes as dependent variables and the frequencies of different prey species as predictors were used. For assigning the vegetation classes to each study site, we considered the dominant vegetation type along the banks of the concerned waterbody where turtles were captured and studied. In the models, computing by an all-effects procedure, the identity link function and a normal distribution of error were used [18].

Parametric tests were used only after having verified the normality and homoscedasticity of the used variables by the Kolmogorov–Smirnov test. Intersexual differences in the frequency of consumption of food items were assessed by an observed versus expected χ<sup>2</sup> test, with the *p*-value generated after 9999 Monte Carlo permutations. For analyses on the ontogenetic changes in diet composition, we divided the turtles into three body size categories: (i) <8 cm carapace length, (ii) 8.1–14 cm, and (iii) >14 cm. Since one of the study species (*P. nanus*) rarely exceeds 10 cm in carapace length and is one of the smallest turtle species in the world [1], for this latter species we had only size categories (i) and (ii) in our samples. In the text, the means are indicated as ± 1 Standard Deviation (S.D.), and alpha was set at 5%.
