Urbanization Shapes Dietary Composition and Snout Morphology of Salamanders Authors: A J D Marques, L Alarcón‐Ríos, V A Mata, G Velo‐Antón Published: 2026 ABSTRACT ABSTRACTUrbanization rapidly alters landscapes, generating habitat mosaics characterized by increased impervious surface, reduced soil moisture and diminished vegetation cover. These recurring urban environmental features are expected to reduce community diversity, favouring urban‐tolerant species, thereby promoting urban biotic homogenization. Although shifts in invertebrate community composition have been widely documented, less is known about how such changes propagate to higher trophic levels. In this study, we examined how urbanization relates to dietary and phenotypic variation in an amphibian species, the fire salamander ( Salamandra salamandra ), along an urban gradient in Porto, Portugal. Dietary diversity and composition were assessed through metabarcoding of faecal pellets, and geometric morphometric analyses of dorsal snout shape were conducted to determine whether dietary homogenization corresponds with morphological similarity and reduced phenotypic variation. Environmental indicators of urbanization were not correlated with a decline in overall dietary diversity; however, it was linked to an increased prevalence of isopods (Armadillidium spp.) in the diet. Morphological results showed a relationship between snout shape and dietary diversity, suggesting greater morphological similarity among urbanized sites. However, no significant relationship was detected between morphological differentiation and variation in dietary niche overlap or diversity. Overall, dietary and morphological traits were more similar among urban sites, suggesting emerging patterns consistent with urban biotic homogenization.Keywords: amphibians, biotic homogenization, diet, metabarcoding, phenotype, urbanization Introduction The process of urbanization is one of the most dramatic and recurring contemporaneous changes to natural landscapes (Gerten et al. 2019; McKinney 2006), leading to highly altered and fragmented habitats (Fenoglio et al. 2020, 2021) and introducing abiotic stressors, such as increased water scarcity (Shuster et al. 2005), higher average temperature (Frank and Backe 2023) and various sources of pollution (Karouna‐Renier and Sparling 2001). Under these conditions, species that can tolerate or utilize anthropogenic resources are deemed urban exploiters (McKinney 2006). As habitats undergo urbanization, it is expected that urban exploiters will proliferate, while endemic habitat specialists unsuited to the changes, deemed urban avoiders, will decline or risk extirpation (Marsh et al. 2023). This process, described as urban biotic homogenization (UBH) (Clergeau et al. 2006; Morelli et al. 2016; Olden and Rooney 2006), postulates that urbanization results in an overall reduction in species richness while simultaneously converging toward similar community assemblages, favouring the increased presence of generalist species and the decline or loss of endemic specialists (Lokatis and Jeschke 2022). The process of UBH has been reported in a variety of taxonomic groups. Among plants, it often corresponds to the increased prevalence of alien species with traits that confer tolerance to urban environments (Kühn and Klotz 2006; Padullés Cubino et al. 2019) along with the active maintenance of human‐favoured species (Groffman et al. 2017) to create recurring floral structures and assemblages. Similarly, among invertebrate communities, species that benefit from anthropogenic structures or plant assemblages can become urban exploiters (Chatelain et al. 2023). As an example, changes in soil compaction are expected to benefit hemiedaphic and epiedaphic taxa that are not impaired by the expansion of impervious surfaces (Perry et al. 2021; Szabó et al. 2023). The effect of UBH is also not limited to low dispersal taxa, as avian communities have also been reported to converge toward species traits that are advantageous in urban habitats, including increasingly grey and brown plumage colour (Leveau 2019), changes in evasive behaviour (Møller and Ibáñez‐Álamo 2012), and nesting preferences for built structures (Clergeau et al. 2006). Moreover, the niche variation hypothesis (Van Valen 1965) expects that individuals released from interspecific competition, either by the introduction of alien prey, loss of specialized competitors or supplementation by anthropogenic food sources, will expand their niches to exploit new community assemblages. This may result in predators assuming generalized foraging strategies with reduced niche segregation (De León et al. 2019; Palacio 2020) or through individual diet specialization, enabled by phenotypic variation within the community to mitigate intraspecific competition (Manlick and Newsome 2021). Amphibians are characterized as low dispersers (Smith and Green 2005) and vulnerable to many of the stressors associated with urbanization, including water scarcity (Pinto et al. 2023), fragmentation (Delaney et al. 2021; Lourenço et al. 2017), and any combination of noise, light or chemical pollution (Gallagher et al. 2014; Sievers et al. 2018; Sun and Narins 2005; Tennessen et al. 2014). Amphibians are thus expected to be particularly vulnerable to urbanization (Callaghan et al. 2021; Hamer and McDonnell 2008; Pillsbury and Miller 2008), and subsequently must occupy microrefugia within urban areas or be urban adapters (McKinney 2006), responding to changes in their habitat. Some accounts of possible adaptive processes among amphibians to environmental variations have been reported, including increasing the frequency and amplitude of mating calls to ambient noise (Halfwerk et al. 2019), changes in skin pigmentation (Alarcón‐Ríos, Álvarez, and Velo‐Antón 2024; Anderson et al. 2019; Polo‐Cavia et al. 2016; Troïanowski et al. 2015), or the increase of body size (Iglesias‐Carrasco et al. 2017) and phenotypic differentiation (Alarcón‐Ríos, Kaliontzopoulou, et al. 2024; Bettencourt‐Amarante et al. 2025). It remains to be explored how amphibians may respond to the process of urban biotic homogenization within their habitats, particularly in relation to concurrent shifts in associated community assemblages. Here we studied the effect of urbanization on a generalist amphibian predator by examining how diet and morphology change along an urban gradient. Focusing our study on the city of Porto, Portugal, we sampled local populations of the fire salamander ( Salamandra salamandra ), a generalist predator that primarily consumes terrestrial detritivores, including annelids, diplopoda, and a wide range of invertebrate species (Bas Lopez et al. 1979; Ferenti et al. 2010; Marques et al. 2022; Wang et al. 2021). We hypothesize that UBH leads to a decline in habitat‐specialist prey within salamanders' diets, replaced by more urban‐tolerant invertebrate species (Marques et al. 2022; Szabó et al. 2023). If so, we would expect reduced dietary diversity and increased similarity in prey composition among salamanders inhabiting urbanized sites. As salamanders' snout shape is closely linked to prey capture and handling performance (Sagonas et al. 2014), we similarly hypothesize that UBH‐driven changes in prey availability will promote functional homogenization of snout shape morphology in urban salamanders (Figueira et al. 2023; Iglesias‐Carrasco et al. 2017). Under this scenario, we would anticipate decreased morphological diversity and convergence toward a snout shape adapted to urban foraging conditions. We further predict a covariation between diet and morphology. Specifically, shifts in mean snout shape should correlate with changes in prey composition, while morphological variance should be positively associated with dietary diversity. Methods Study Area The city of Porto (northern Portugal) has undergone a period of rapid urbanization within the 20th century, coinciding with a greater than 90% decline in the amount of greenspace (Madureira et al. 2011) and a recently reported maximum urban heat island intensity of 3.5°C (Andrade et al. 2023). Ten sites were visited within the study area (the city of Porto and surrounding area; Table S1), however, adequate sample sizes were collected from only six, three sites from within the urban core and three sites from the urban periphery (Figure 1). The extent and impact of urbanization are difficult to capture using a discrete variable; therefore, we characterized habitats using environmental variables associated with urbanization. These include such variables as the mean proportion of impervious surface, mean proportion of tree cover, and proportion of small woody features (e.g., hedges, shrubs, etc.) around each site as determined by the Copernicus Land Monitoring Service (European Union 2018) (Figure 1). When considering the impact of each feature, we determined the proportion of features within a 100, 200 and 400 m buffer from the area searched at each site (Table S2). Sites varied in patch size, and so search area was also considered as a possible explanatory variable. Sites with low impervious surface but very little tree cover or woody features generally corresponded to agricultural areas which are considered unfavourable to salamanders (Lourenço et al. 2019). Among sites included in our analysis, the approximated search area applies to a 50 m buffer around the patch and the mean proportion of impervious surface, tree cover, and woody features estimated up to 400 m from the searched area. Site with low tree cover and impervious surfaces (white), generally correspond to agricultural areas considered unfavourable to salamanders. Photo courtesy Max Benito and Jose Vieira from ExSitu Project. Sampling For dietary analyses, adult male salamanders were collected between October and November 2022. Sites were visited on nights of high humidity, corresponding to a period with the highest activity among mating males in S. s. gallaica (Velo‐Antón and Buckley 2015), which were identified by their enlarged cloaca. Salamanders were most often located in green‐spaces, on the ground, with leaf litter or along stone walls that provide them with shelter. In most cases, sites were revisited 2–3 weeks later to increase the number of pellets collected per site. We therefore acknowledge the possibility of having re‐sampled individuals. To mitigate the potential impact of possibly pseudo‐sampling individuals, all analyses were repeated with a single sampling visitation to determine whether significant differences were observed. For morphological analysis, adult males were sampled in autumn 2021 from those same sites as those used for the diet analysis. When necessary, some sites were revisited multiple times. In such cases, we checked for resampled individuals by using pictures, thereby eliminating the possibility of pseudo‐replication in the morphology dataset. While comparisons between diet and morphology are made, these are based on summaries of sites, as the exact individuals used in either analysis cannot be guaranteed nor is it likely. Prey Detection To identify the contents of prey among sites, DNA metabarcoding of faecal pellets was selected as a cost‐effective, minimally invasive and expedient method for characterizing the diets of salamanders with genus‐level taxonomic resolution (Marques et al. 2022). To obtain faecal pellets, salamanders were placed inside large group containers (50 × 30 × 30 cm) with a bedding of moist paper towel and bisected PVC piping for shelter. Salamanders can take several days to produce a large faecal pellet and thus were monitored for up to 5 days before being returned to the place of capture. While we cannot exclude the possibility that multiple pellets derived from a single individual, we assumed, given the pellet size and infrequency of defecation, that an individual salamander only defecated once during holding. Since individuals were housed together, pellets should also be taken to represent samples from a site but cannot be linked to the individual that produced them. No more than 12 individuals were placed in a container at a time to avoid overcrowding. When more than 12 individuals were collected in a night, they were separated into different containers. After release, containers were sterilized with a 10% bleach solution and wiped down with a 70% ethanol solution before reuse. Individuals were hydrated twice daily with ddH20 and their paper towel bedding was replaced daily. Tweezers wiped down with 96% ethanol between each use were used to collect faecal pellets as they were produced. Faecal pellets were stored in 1 mL of 70% ethanol then stored at −20°C (Horng et al. 2018). Faecal pellets were gradually collected from sampled individuals and used to characterize diet through DNA metabarcoding. DNA was extracted with a modified version of Krinitsina et al. (2015) using conventional isopropanol extraction followed by an isolation of DNA using AMPure XP beads. To minimize sequencing bias, pellets from different sites and dates were randomly allocated to blocks of up to 23 pellets and a negative control for each extraction and amplification step (Efird 2011; O'Leary et al. 2018). Amplification of cytochrome oxidase I (COI) used the fwh1 fragment (Vamos et al. 2017) with a salamander specific blocking primer (Marques et al. 2022). While the choice of fragment may introduce a prey detection bias, this fragment results in similar estimates of prey diversity when compared to an alternative (Marques et al. 2022; Tournayre et al. 2020) and targets a smaller fragment, generally improving performance when applied to degraded DNA fragments. PCR mixes comprised 10 μL of Qiagen Multiplex PCR Master Mix, 0.3 μL or 0.4 μL of 10 mM target primer, 8.0 μL of 10 mM blocking primer, 2 μL of DNA template, and enough water for a final volume of 25 μL (Lopes et al. 2019). Thermocycling conditions included an initial denaturing step at 95°C for 15 min, followed by 40 cycles of denaturing at 95°C for 30 s, annealing at 45°C for 60 s, extension at 72°C for 30 s and a final extension at 72°C for 10 min. Initially, we attempted to amplify samples at a 1:5 dilution, but the presence of inhibitors meant successive dilutions (1:10 and 1:20) were occasionally required. Amplified samples were size selected with AMPure XP beads to reduce primer bias from non‐specific bands below that of our target fragment. External indexed primers were annealed to the cleaned PCR product, each containing a unique 7 bp long barcode that differed by at least 3 bp (da Silva et al. 2019). The indexing reaction comprised 7 μL of Kappa HiFi HotStart, 2.8 μL of PCR product and 1.4 μL of each external index. Thermocycling conditions included an initial denaturing step at 95°C for 3 min, followed by 8 cycles of denaturing at 95°C for 30 s, annealing at 50°C for 30 s, extension at 72°C for 30 s and a final extension at 72°C for 5 min. Once indexed, samples were cleaned with AMPure XP beads, quantified using an Epoch BioTek Microplate Spectrophotometer, and normalized to 20 nM of DNA per sample before being pooled into a single tube for sequencing submission on an Illumina MiSeq with a target of 30 k reads per sample. Paired reads were aligned with PEAR (Zhang et al. 2014) and passed to the ngsfilter function from OBITools (Boyer et al. 2016) to remove primer sequences and annotate sample information. Trimmed reads were then collapsed into unique sequence variants using obiuniq, and then denoised with–cluster‐unoise included in VSEARH (Rognes et al. 2016) using default parameters, with the exception of minimum sequence length, which was set to 150 bp. The resulting denoised sequences are zero‐radius operational taxonomic units (zOTU) to account for variable sequences within the same taxonomic group (Edgar 2016). Chimeras were removed from the resulting zOTUs with the –uchime3_denovo function and clustered into OTUs at 99% identity with –cluster_size; then reads were mapped to the remaining OTUs with 99% identity using –usearch_global. To further remove potential PCR and sequencing errors, LULU (Frøslev et al. 2017) was used with the default parameters. The maximum number of reads of any OTU identified in either extraction or PCR negatives was subtracted from the number of reads observed for that OTU in each sample to correct for contamination. OTUs were assigned to taxon at the genus level when possible using BOLDIGGER v.1.2.5 (Buchner and Leese 2020). Following taxonomic assignment, only OTUs corresponding to plausible prey, namely those in the phyla Arthropoda, Mollusca and Annelida were retained, excluding OTUs such as those assigned to bacteria, plants, fungi, nematodes or domestic animals. Sequenced pellets with fewer than 100 dietary reads were excluded, as were OTUs comprising fewer than 1% of the total dietary reads from a sample. Prey identities for OTUs were then subject to further inspection, and the final analysis was carried out with OTUs that could be assigned to a corresponding genus and with a minimum 97% similarity. Dietary Analysis Prey occurrences at the genus resolution were determined by the presence or absence of assigned OTUs in a faecal pellet, with multiple OTUs assigned to a common taxon counted as a single occurrence. Estimates of dietary diversity were based on empirical species richness, Shannon diversity index and Simpson diversity index using the iNext package (Chao et al. 2014; Hsieh et al. 2016) in the R language for statistical programming (R Core Team 2023). As the number of pellets retrieved from each site varied, standardized comparisons were made from interpolation of species diversity using rarefaction (Chao et al. 2014). Hill numbers incorporate relative abundance to estimate species diversity in a manner that allows for comparisons between multiple assemblages (Jost 2006). Each Hill number estimate was based on the minimum observed sample coverage among sites within a 95% confidence interval. Sample coverage in this context is likely the proportion of the community assembly that is captured by a given sampling effort, such that using minimum sample coverage entails that all estimates are based on either observation or rarefaction. Additionally, we compared the minimum sample coverage estimates to estimates based on the asymptote of the curve and twice the minimum sample coverage to determine whether variable thresholds impacted diversity estimates. Estimates were made for cumulative dietary diversity as well as each phylum of prey independently. Both a Pearson's correlation coefficient and Spearman's rank coefficient were used to test the relationship between Shannon diversity estimates at each site and environmental variables at a 100, 200 and 400 m buffer from the search area using the function rcorr included in the Hmisc package (Harrell Jr. 2025) with adjusted p‐values based on the false discovery rate. Environmental variables included scaled proportion of impervious surface and tree cover, estimated as the average pixel value (0%–100% impervious surface or tree cover) within the buffer area, while woody features are the proportion of the buffer area where woody features were present (European Union 2018). Search area (in hectares) was also scaled and included as a proxy for the effect of patch size. Additionally, correlations between each prey's frequency of occurrences among pellets from a site were tested against search area and environmental variables at each buffer using a Spearman's rank correlation while correcting for the false discovery rate. For subsequent analyses of variance in diet between sites, only prey taxa that were present in at least 5% of pellets were considered, representing the core diet and reducing the effect of rare species (Tournayre et al. 2021). Bray–Curtis dissimilarity between sites was computed based on this frequency of occurrence of prey taxa using the function vegdist. A multivariate homogeneity of group dispersion was performed with the betadisper function to generate a principal coordinates analysis (PCoA) for relative variances between sites. A permutational multivariate analysis of variance (PERMANOVA) was performed with adonis2 to test for significant variance between sites and each environmental variable included in our cumulative model. When applicable, 9999 permutations were used to test significance. All multivariate analyses were performed in R with functions included in the vegan package (Oksanen et al. 2024). To understand the contribution of each prey genus to variance between sites, a correspondence analysis (CA) was used. Additionally, a Pianka's niche overlap (Pianka 1974) was used to determine the similarity in prey occurrence between pairwise comparisons of sites, using the proportion of a given prey's occurrence in one site relative to the total prey occurrence among both (Farina et al. 2023). A standardized effect size (SES) was used to determine if the index is likely significant at either the upper (> 2) or lower tail possibilities (< −2). Pairwise comparisons of niche overlap between sites were performed with EcoSimR 1.00 (Gotelli and Entsminger 2003). Morphological Measures For morphological analysis we applied landmark‐based geometric morphometrics (GM). We followed the methodology for image acquisition and processing described in Alarcón‐Ríos et al. (2017). As we were solely interested in diet‐related structures in this study, the methodology described there was adapted to focus on landmarks defining the upper jaw, snout and eye shape, all of which are involved in the processes of food intake, manipulation and swallowing in urodeles (Herrel et al. 2019; Nishikawa and Schwenk 2001; Stinson and Deban 2017; Wake and Deban 2000). Nine fixed landmarks and 18 sliding semi‐landmarks were digitized on the anterior‐most part of the dorsal view of salamanders' snout (see also Alarcón‐Ríos et al. 2020; Figure S1) using tpsDig2 v 2.21 (Rohlf 2015) by one of us (L.A.‐R.). All individuals were digitized twice to take measurement error into account. To explore patterns of snout morphology across sites, the symmetric component of shape was extracted using the bilat.symmetry function from the R package geomorph (Baken et al. 2021). This procedure includes the alignment of snout shape coordinates of all individuals using a generalized least‐squares Procrustes superimposition to standardize in size, location, and orientation (Rohlf 1999; Rohlf and Slice 1990) and the calculation of the average coordinates of the reflected superimposed mean configuration for each individual to remove asymmetries. All GM procedures and statistical analyses were implemented using packages geomorph v. 4.0 (Baken et al. 2021) and RRPP v. 1.0 (Collyer and Adams 2018, 2019). Statistical significance was always assessed based on 10,000 random permutations, using Residual Randomization (Collyer et al. 2015). Diet‐Morphology Relationship We began by examining whether the relationship between snout size and shape differed across sites (i.e., whether they present different allometric slopes). A common slope could not be applied to adjust for the effect that size has on shape, however, the overall allometric differences effect (R 2 = 0.047, Z = 2.122) is considerably smaller than that of location (R 2 = 0.165, Z = 6.635; Table S3). Thus, all subsequent shape analyses included the effect of size on shape, excepting those that compare allometric slopes between sites and diet diversity. To evaluate patterns of variation in snout shape across the studied sites, an initial permutational ANOVA across all individuals was conducted with site as a factor. To further explore differentiation among sites, pairwise comparisons were performed among them and generated a principal components plot displaying least‐squares means for each site and 95% non‐parametric confidence ellipses for each site as a measure of the precision of mean estimation. Then, to explore the relationship between snout shape and dietary diversity within each site, we extracted PC1 values for each location and tested the correlation with dietary diversity measurements. Under the assumption that sites with similar diets may exhibit more similar morphologies, we tested for correlations between morphological (mean snout shape and allometry) and dietary similarity. Distance matrices of morphology (Euclidean distance between sites including allometric variation) and allometric slope attributes (pairwise differences in allometric distance, angle and length between sites) were compared with diet similarity matrices (Bray–Curtis dissimilarity and Pianka's niche overlap). As dietary and morphological distance matrices between sites derived from individuals sampled at different times, these analyses represent indirect site‐level associations. The correlation between matrices was assessed using the Mantel test as implemented in the R package vegan (Oksanen et al. 2024), with significance evaluated after Bonferroni correction. Finally, to evaluate whether greater dietary diversity is associated with increased morphological variation within each site, we tested for correlations among diet diversity (species richness, estimated diversity index for all prey, and for each of the three diet phylum) and phenotypic variation using the rcorr function (Harrell Jr. 2025). Phenotypic variation within each site was quantified using the symmetric component of snout shape variation (including size‐related variation), while accounting for site mean in the model (i.e., shape variance of individuals within each site relative to the site mean). This was implemented with the function morphol.disparity in R package geomorph. Results Diet Sampling and Prey Detection From each site a median of 10 individuals (range 6–57) and 6 pellets (range 3–29) were collected (Table S1). Initially 174 pellets were sequenced with approximately 3.793 M reads, of which 94.6% of reads were not associated with the 12 negative controls. Read lengths varied from 150 to 200 bp, with a median fragment length of 179 bp. Among the negative controls, a total of 96 OTUs were observed and their read counts were deducted from samples; however, only three of these corresponded to plausible prey taxa and had no more than two reads total among all negatives. Among the remaining OTUs, 41.2% had no taxonomic match, and only one OTU was associated with S. salamandra with a cumulative 23 reads across all samples, and absent among the negatives. Retaining only OTUs identified as plausible prey taxa, we retained 49.9% of the total starting reads. Finally, after removing pellets with fewer than 100 reads or less than 1% total dietary reads, we retained 163 pellets comprising 47% of the total starting reads. Sites where fewer than 10 pellets were collected were excluded from subsequent analysis, leaving us with six sites considered (Table 1). Of the 129 OTUs that remained post‐filtering, these corresponded to 87 distinct genera with between one and five OTUs per genus (68 Arthropoda, 11 Mollusca and 8 Annelida genera; Table S4). To reduce the influence of rare taxa in inflating estimates of niche overlap among sites, we retained only prey taxa present in at least 5% of all pellets. This threshold preserved 57% of all prey encounters, defined as instances in which a given taxon was detected among a pellet. This approach is often used to assess the core diet of insectivorous animals that tend to show a long tail of rare diet items (Tournayre et al. 2021). TABLE 1For each site, environmental information includes the estimated search area in hectares (ha), mean proportion of impervious surface (IS) within 400 m, tree cover (TC) within 200 m, and woody features (WF) within 100 m. Sampling results include the number of faecal pellets (n F) collected and individuals with snout morphologies (n M) measured.SitehaISTCWF n F n M P.Avioso4.78.330.624.22535Mindelo13.512.132.721.02447P.Cidade2.919.842.115.32519P.Oriental3.527.220.629.32520Ramalde4.750.19.311.21618Matosinhos1.663.49.35.52920 For each site, environmental information includes the estimated search area in hectares (ha), mean proportion of impervious surface (IS) within 400 m, tree cover (TC) within 200 m, and woody features (WF) within 100 m. Sampling results include the number of faecal pellets (n F) collected and individuals with snout morphologies (n M) measured. Dietary Diversity After comparing Hill numbers estimates we found that the Shannon's diversity index (q = 1) did not significantly differ from either the less conservative species richness (q = 0) or more conservative Simpson diversity index (q = 2) after applying a Bonferroni correction. Diversity estimates based on minimum observed sample coverage of 0.41 (Mindelo) did differ significantly after a Bonferroni correction (p < 0.05) from other thresholds (i.e., the asymptote and twice the minimum sample coverage), and we opted to use the former as a more conservative estimate of diversity for subsequent analyses (Table 2 and Figure S2). Dietary diversity increased with search area, but after correcting for the false discovery rate, neither search area nor any other environmental variables at any buffer distance were significantly correlated with dietary diversity (Figure S3). To determine what effect repeated visitations had on the results, we re‐ran all subsequent analyses with samples collected from a single visitation (Table S5). In nearly all cases, results were unchanged, but where a deviation was observed and it is noted below. TABLE 2Dietary results include observed species richness (Obs), the estimated sample coverage (SC), and the estimated Shannon Diversity Index for all prey (Est.) and each of the three identified phyla separately (Arth. = arthropoda, Anne. = annelida, Moll. = mollusca).SiteObsSCEst.Arth.Anne.Moll.P.Avioso2977.914.411.32.80.7Mindelo4057.435.930.13.49.6P.Cidade2484.19.37.51.41.9P.Oriental2871.913.611.11.11.7Ramalde2169.511.27.10.93.8Matosinhos2483.27.84.21.32.5 Dietary results include observed species richness (Obs), the estimated sample coverage (SC), and the estimated Shannon Diversity Index for all prey (Est.) and each of the three identified phyla separately (Arth. = arthropoda, Anne. = annelida, Moll. = mollusca). Diet Variation and Niche Overlap Frequency of prey occurrence varied significantly between sites in correlation with impervious surface (R 2 = 0.45; p < 0.05) and woody features (R 2 = 0.46; p < 0.05) based on the PERMANOVA analysis; however, the latter was not identified as significant when considering pellets from only a single visitation. Additionally, with a Constrained Correspondence Analysis (CCA), we identified isopods (Armadillidium spp.) as being overrepresented in the diets of salamanders from among sites with the highest impervious surface, Matosinhos and Ramalde. Other hard‐bodied prey, isopods (Eluma spp.) and beetles (Nalassus spp.), were overrepresented at sites P.Avioso and P.Oriental (Figure 2). Overlaying the contribution of environmental variables with that of prey occurrence, we observed that Armadillidium occurrence correlates with impervious surface (Figure S4), while other hard‐bodied detritivores correlate with wood cover. Meanwhile, sites like Mindelo and P.Cidade appear to share a higher prevalence of Lumbricus and proportional tree cover. Constrained correspondence analysis (CCA) illustrating the combined contribution of prey occurrence in the salamander diet. Loadings arrows indicate the contribution of environmental variables within a 100 m buffer to describing the variance in diet between sites. Focusing on the three most common prey, Armadillidium, Eluma and Lumbricus (Figure 3A), a Spearman's rank correlation identified a positive relationship between the proportion of pellets with Armadillidium and impervious surface (R 2 = 0.9, p < 0.01, Figure S4). Among pairwise comparisons of niche overlap, two pairs indicated a strong niche overlap (Ojk > 0.75 and SES > 2), namely P.Oriental and P.Avioso on the urban periphery, and the more urbanized sites of Ramalde and Matosinhos (Figure 3B). Mantel test results comparing distance matrices based on prey frequency of occurrence between sites for all prey and each prey phylum to those of environmental variables did not identify any significant correlations following a Bonferroni correction (Figure S5A). Differences in prey frequency of occurrence did not correlate with geographic distance between points. For each site, (A) ordered according to the proportion of impervious surface (IS), the frequency of occurrence among pellets is displayed for the most commonly observed prey taxa observed in at least 10% of all pellets. (B) Pairwise comparisons of niche overlap (Okj) between sites, where higher values indicate greater niche overlap. Standardized effect sizes > 2 are indicated by asterisks (*). Morphological Patterns and Diet‐Morphological Associations A total of 159 adult males (Table 1) were used for morphological analyses. Overall, observed morphological shape differences among sites (F (5,153) = 5.468, p < 0.001) were mostly driven by Mindelo, and to a lesser extent P.Avioso, site (Figure 4, Figure S6). Those sites present the highest estimates of Shannon index (Figure 4, Table 2), which significantly correlates with the first principal component axis of variation (t = −3.442, df = 4, p = 0.026). Main variation in snout shape along the first axis of variation is focused on snout length and width and eye size (Figure 4). Although the potential effect of size differences among populations on observed patterns should not be completely dismissed, snout size does not significantly correlate with Shannon index (t = −0.486, df = 4, p = 0.653), reinforcing that differences in size among sites (Figure S7) do not significantly influence the observed relationship between shape and dietary diversity. Means and 95% confidence ellipses of the first and second principal components of snout shape variation across individuals in each studied site, which are coloured by estimated Shannon index. Deformation grids depict shape change at the extremes of the first axis in comparison to the overall mean. Shape change has been magnified by a factor of seven to facilitate visualization. Shape includes size‐related variation. Mantel tests for correlation between morphological (snout shape and allometric attributes) and Bray‐Curtis dissimilarities of diet (Figure S5B) identified no significant differences. Nor was the dietary niche similarity matrix (Pianka's) correlated with morphological (snout shape and allometric attributes) difference matrices between sites (Table S6). No significant correlations were observed between phenotypic variation in snout shape (Table S7) and dietary diversity estimates (Figure S8). Unfortunately, because our analyses were limited to only six sites, our statistical power was very limited. Discussion By comparing diet and morphology along an urban gradient, our goal was to assess trophic interactions between fire salamanders and their terrestrial invertebrate prey. Rather than observing a clear decline in dietary diversity with increasing urbanization, we detected shifts in prey frequencies associated with variation in habitat structure. In particular, Armadillidium spp. were more prevalent in salamander diets from urbanized sites. Mean snout shape variation among salamanders correlates with dietary diversity values, which, along with an apparently more homogenous snout shape among sites with higher levels of urbanization, suggest a potential convergence in snout morphology, consistent with expectation under urban functional homogenization. Due to limited statistical power, however, we cannot confirm this trend conclusively. These tests therefore serve as an exploratory examination of correlation between diet, environment and morphology, and would require approximately eight sites per habitat in order to reach the statistical power to detect a large effect between any of these factors. Nonetheless, our results suggest that urban biotic homogenization may influence both diet and morphological variation in salamanders. Changes in diet can have important ecological and evolutionary implications for predators; however, metabarcoding alone cannot capture the functional and nutritional differences among prey types. Urban Biotic Homogenization of Trophic Niche Contrary to our initial hypothesis, dietary diversity was not correlated with any of the environmental variables, but was instead primarily associated with search area. Importantly, this refers to the area surveyed within each site and does not necessarily reflect total patch size, as the spatial limits of salamander occupancy remain unknown. For example, P.Cidade is a very large urban green space (83 ha), yet salamanders appear restricted to a relatively small portion of the park characterized by dense tree cover, standing water and abundant leaf litter, conditions favourable for salamander occupancy (Donati et al. 2022; Lourenço et al. 2017). Thus, surveyed area may better represent the extent of continuous suitable habitat or habitat complexity rather than total patch size (Jonsson et al. 2009). Regardless of the underlying mechanism, impervious surface alone cannot explain variation in dietary diversity unless patch integrity is considered, potentially as an indirect proxy for urbanization (Liu et al. 2016). While a decline in dietary diversity was not observed along the gradient, differences in prey frequency of occurrence in diets between sites were evident and were not simply explained by geographic distance. Environmental features were correlated with differences in prey frequency, resulting in three site groupings: (1) harsh environments (high impervious surface and low tree cover, Ramalde and Matoshinhos), (2) mild environments (low impervious surface and tree cover, P.Avioso and P.Oriental); and (3) semi‐natural environments (low impervious surface and high tree cover, P.Cidade and Mindelo). Harsh environments consist of small patches surrounded by highly impermeable surfaces, where salamander diets were dominated by Armadillidium spp. Like other isopod species, Armadillidium spp. are generalist detritivores that tolerate urban stressors and pollutants, thrive in disturbed soils, and are therefore well positioned as urban exploiters (Magura et al. 2008; Pouyat et al. 2010; Szlavecz et al. 2018). Mild environments, urban parks located near the urban core, exhibited a greater variety in hard‐bodied detritivores in diets, namely Nemobius spp., Eluma spp. and Nalassus spp. Finally, semi‐natural environments, which were structurally heterogeneous, as Mindelo and P.Cidade differ considerably. Mindelo is a large, continuous nature reserve with adjacent agricultural and forestry areas (Velo‐Antón 2020), whereas P.Cidade is a large urban park embedded within a densely urbanized matrix. Among the three aforementioned environments, this is the only one characterized by low dietary niche overlap. Instead, similarity within this category appears driven by increased occurrence of annelid prey, particularly Allolobophora spp. and especially Lumbricus spp. Among prey phyla, annelids were expected to be the most sensitive to urbanization, as increased impervious surfaces, soil compaction, and heavy metal contamination have been linked to declines in annelid richness (Perry et al. 2021; Szabó et al. 2023). However, no correlation was detected between annelid prey frequency and either impervious surface or tree cover. Other factors, such as leaf litter availability, soil moisture or soil compaction, may instead play a more important role in shaping annelid presence (Kotze et al. 2022). Dietary Diversity and Phenotypic Variation If predator diets were to homogenize across urbanized sites, a corresponding decline in functional variation might be expected (Manlick and Newsome 2021). Evidence from other taxa suggests that, alongside biotic homogenization in species richness, urbanization can also promote functional homogenization toward phenotypes that are advantageous in urban environments (Leveau 2019; Putman and Tippie 2020). In our study, environmental variables were not correlated with dietary diversity; however, we did detect an association between diet diversity and snout shape. Specifically, the site with the highest dietary diversity, Mindelo, and to a lesser extent P.Oriental and P.Avioso, were characterized by individuals with elongated snouts, slightly narrower jaws and larger eyes (Figure 4). Although dorsal snout shape appears to be associated with variation in diet diversity, it is important to note that Mindelo may represent an outlier. Its designation as a nature reserve may contribute to this pattern, potentially due to higher salamander densities and greater prey diversity relative to other sites. Thus, further comparisons to neighbouring salamander populations are needed to assess the robustness of this relationship. Furthermore, our analysis focused exclusively on dorsal snout shape which includes jaws and eyes. While this aspect of morphology is plausibly linked to prey capture and handling, it does not capture the full range of functional variation in salamander snout morphology. Although no significant relationship was detected between morphological differentiation and variation and dietary niche overlap and diversity, the limited morphological differentiation among sites from within the urban core relative to those in the rural periphery, suggests alignment with expectations under the urban biotic homogenization (UBH) hypothesis. This contrasts with previous findings in S. s. bernardezi from the city of Oviedo, where urban populations were more morphologically differentiated among themselves than non‐urban populations (Alarcón‐Ríos, Álvarez, and Velo‐Antón 2024). Differences in snout shape variability between cities may reflect differences in environmental heterogeneity among study sites. Urban habitats in Oviedo encompass a wide range of environments, from stony backyards to urban parks, whereas sites within Porto (S. s. gallaica) exhibit comparatively homogeneous environmental characteristics. Variation in the range of habitats occupied by urban salamanders may also stem from differences in life‐cycle strategies between subspecies (i.e., aquatic vs. terrestrial reproduction; see Velo‐Antón and Buckley 2015). Additionally, distinct ecological pressures, such as altered community assemblages (e.g., presence of competitors, novel predator), habitat modifications (e.g., soil modification, artificial lighting, shelter types), or the cumulative effects of multiple stressors (Alberti et al. 2017; Chatelain et al. 2023), may shape patterns of morphological differentiation differently across cities. In Oviedo, urban populations also showed a trend toward an increase within‐population phenotypic variation in snout shape (Alarcón‐Ríos, Kaliontzopoulou, et al. 2024), while the pattern here is not completely clear. Together, these findings might suggest that factors other than diet, such as competition, habitat structural variability or seasonal fluctuations in food resources (Chatelain et al. 2023; Herrel et al. 2019; Thompson et al. 2022), may contribute to snout shape diversification but were not explicitly captured in this study. Methodological Considerations Sample coverage estimates from Mindelo indicate that prey richness was likely undersampled. Low estimates of sample coverage suggest that observed prey richness may not fully represent true diversity, potentially limiting the model's ability to detect differences (Chao et al. 2014). This pattern may have been influenced by the higher diversity observed in Mindelo relative to other sites. However, excluding Mindelo did not alter the ordination or overall diversity patterns, suggesting that, despite low sample coverage, the majority of dietary diversity was captured at each site. Given that salamanders could not be maintained individually for extended periods, we were unable to directly link individual diet to morphology. Establishing such links would require alternative methodologies, such as collecting snout morphometrics alongside stomach flushing (Crovetto et al. 2012; Ferenti et al. 2010) or maintaining individuals in isolation until defecation (Wang et al. 2022). In general, variation in metabarcoding results must be anticipated, as prey detected in faecal samples may represent only a subset of consumed items. Running replicate PCR amplifications per sample could improve prey detection and reduce amplification bias prior to sequencing (Mata et al. 2019; Tournayre et al. 2020). Moreover, salamanders exhibit relatively long digestion times, often several days, compared with taxa such as bats, reptiles or frogs, which may defecate within hours of capture (Hernández et al. 2024; Mata et al. 2019; Pereira et al. 2021). Prolonged digestion may increase DNA degradation and reduce detection success. Metabarcoding is also agnostic to the developmental stages of prey (i.e., larval or adults) or size at consumption (Perry et al. 2021), limiting trait‐based dietary analyses. Although diet can vary seasonally (Barrett et al. 2012), all samples in this study were collected within a single season, and no effect of collection date on prey richness was detected. Furthermore, although we explored correlations between morphology and diet, it is important to note that datasets differ both in the individuals sampled and in the years of collection. Therefore, the observed patterns represent site‐level generalizations rather than direct individual‐level associations and should be interpreted with caution pending further study. Implications of Urban Biotic Homogenization As generalist predators, salamanders exert top‐down effects on macroinvertebrates, including isopods, millipedes and earthworms (Laking et al. 2021; Marques et al. 2022; Walton et al. 2006) which themselves contribute to leaf litter decomposition. The persistence of complex decomposer communities can accelerate carbon and nitrogen turnover in leaf litter (Handa et al. 2014). Leaf litter availability at sampled sites may also have influenced salamander detectability in the field. This was particularly evident in the urban parks of P.Cidade and P.Oriental, where salamanders were largely confined to forested patches that had not been intensively landscaped. From a management perspective, minimizing leaf litter removal in urban parks may benefit both macroinvertebrates and salamanders by providing shelter and nutrient resources (Korhonen et al. 2020; Le Roux et al. 2014). Preserving undisturbed greenspaces, either in peripheral areas or through intentional planning strategies (Li and Nassauer 2020), may help maintain stable salamander populations and diverse invertebrate communities. Overall, urbanization in this system appears to be associated with an increased prevalence of epiedaphic (i.e., Armadillidium and Eluma) detritivores in salamander diets within urban habitats, without a corresponding decline in species richness or overall diversity. To better understand the drivers of this variation, future studies should incorporate more detailed habitat characterization, including shelter availability, vegetation structure, land‐use or hydrology, all of which may influence both community assemblage and environmental pressures (Peng et al. 2020; Schrank et al. 2015; Zamora‐Camacho and Aragón 2023). Although challenging to assess in situ, further investigation into predator–prey interactions across habitat types may clarify how trophic dynamics reflect differences in community assemblage and ecological pressures (Stoler and Relyea 2020). While a direct mechanistic link between community and functional homogenization remains unresolved, the parallel patterns observed here support continued investigation into how trophic interactions respond to ecological and evolutionary change under urbanization. Author Contributions A.J.D.M., G.V.‐A. and V.A.M.: conceptualization. A.J.D.M., L.A.‐R. and V.A.M.: methodology. A.J.D.M. and L.A.‐R.: validation. A.J.D.M., L.A.‐R.: formal analysis. A.J.D.M., L.A.‐R., G.V.‐A. and V.A.M.: investigation. G.V.‐A. and V.A.M.: resources. G.V.‐A., A.J.D.M., L.A.‐R. and V.A.M.: data curation. A.J.D.M. and L.A.‐R.: writing – original draft preparation. G.V.‐A. and V.A.M.: writing – review and editing. A.J.D.M. and L.A.‐R.: visualization. G.V.‐A. and V.A.M.: supervision. G.V.‐A.: project administration. G.V.‐A.: funding acquisition. All authors have read and agreed to the published version of the manuscript. Funding This publication was supported by National Funds through the support of the Fundação para a Ciência e a Tecnologia (FCT, Portugal; ANTHROPOPHIBIAN project: Ref: PTDC/BIA‐CBI/2278/2020) and a Project of Excellency, Xunta de Galicia (Spain; Ref. ED431F 2022/10). A.J.D.M. was funded by the FCT (Ref. PTDC/BIA‐CBI/2278/2020). V.A.M. was funded by the FCT through the program ‘Stimulus of Scientific Employment, Individual Support—3rd Edition’ (2020.02547.CEECIND; https://doi.org/10.54499/2020.02547.CEECIND/CP1601/CP1649/CT0004). L.A.‐R. was supported by a ‘Margarita Salas’ contract funded by the European Union—NextGenerationEU, Recuperation, Transformation and Resilience Plan by Spanish Ministry of Universities, on the basis of the University of Oviedo (Spain) call (Ref: MU‐21‐UP2021‐030) and a post‐doctoral research contract in ANTHROPOPHIBIAN project (FCT—Foundation for Science and Technology to G.V.‐A.: grant number PTDC/BIA‐CBI/2278/2020). G. V.‐A. was supported by a Ramón y Cajal research grant (Ref. RYC‐2019‐026959‐I/AEI/10.13039/501100011033). Universidade de Vigo/CRUE‐CISUG covered the Article Processing Charge (APC). Ethics Statement Fieldwork for collecting individuals and faecal samples was carried out with the corresponding permits from the national administration (Portugal Ref. n° 544/2021/CAPT and 528‐530/2022/CAPT). Sampling procedures were carried out following the Guidelines for Use of Live Amphibians and Reptiles in Field and Laboratory Research, 2nd Edition, revised by the Herpetological Animal Care and Use Committee (HACC) of the American Society of Ichthyologists and Herpetologists, 2004. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: Landmarks (red‐filled points) and semi‐landmarks (white‐filled points) recorded on the dorsal view of salamanders' snout. Figure S2: Hill numbers 0 (species richness), 1 (Shannon diversity) and 2 (Simpson diversity) for each site were estimated based on rarefaction curves. Points correspond to the asymptote of the curve (qD.Asy), twice the minimum sample coverage threshold (qd.SC*2) or the minimum samples coverage threshold (qD.SCmin), with error bars corresponding to the upper and lower confidence intervals. Figure S3: Spearman's rank correlations between Shannon's diversity estimate using a minimum sample coverage threshold (qD.minSC), search area, and each environmental variable within 100, 200 and 400 m buffer distances from the search area. Significant correlations after correcting for the false discovery rate are indicated with an asterisk (*). Figure S4: Spearman's rank correlation between prey frequency of occurrence and environmental variables. Significant correlations after correcting for the false discovery rate are indicated with an asterisk (*). Figure S5: Results of Mantel tests testing the correlation of Bray‐Curtis pairwise distance matrices for (A) prey phylum and environmental variables (area = search area, IS = impervious surface, TC = tree cover, WF = woody features) at variable buffer distances matrices; (B) prey phylum and morphological traits (mean snout shape and allometric attributes). Significant correlations after correcting for the false discovery rate are indicated with an asterisk (*). Figure S6: Least‐Squares mean distances (LSmeans) between pairs of studied populations. Shape includes (a) size related variation. Pairwise differences in (b) distance, (c) angle and (d) length among allometric slopes of studied populations. Figure S7: Boxplots indicate the variance in centroid size (CS) of snouts among each sampled site. Figure S8: Bars indicate the R 2 coefficient for correlations between phenotypic variation and total species richness (Richness), estimated Shannon Diversity Index for all prey (qD.Total) and each of the three diet phylum (arthropoda = qD.ARTH, mollusca = qD.MOLL, annelida = qD.ANNE). All correlations were found to be not significant. Table S1: For each site visited the number of individuals collected, visitations, total pellets and the proportion of pellets collected per visit are listed. Sites Casa Tait, Vairão, Parque Real and Malta were subsequently excluded from analysis. Table S2: For each site considered in the analysis, the area searched in hectares as well as mean proportion of impervious surface (IS), tree cover (TC) and the proportion of area with wood features (WF) within the buffer area. Variable buffers include within 100, 200 and 400 m from the search area. Table S3: Results of the permutational ANOVA used to test the snout shape centroid size (CS) covariation, site differentiation and common allometric slopes among sites. Table S4: The taxonomic designations for all identified prey genera are listed up the genus resolution. For each, the number of corresponding zOTUs, the minimum % similarity to a reference sequence, and the proportion of samples were the taxa were observed are also listed. Table S5: Using pellets collected from only one visitation per site corresponds to the following changes in observed species richness (Obs), the estimated sample coverage (SC), and the estimated Shannon Diversity Index for all prey (Est.) and each of the three identified phylum separately (Arth. = arthropoda, Moll. = mollusca, Anne. = annelida). Table S6: Mantel tests results for correlation between Pianka niche similarity and distance matrices mean snout shape (Shape), allometric attributes (Allo), environmental variables (IS = impervious surface, SA = search area, TC = tree cover, WF = wood features). Table S7: Phenotypic variation (phenoV) values for each studied population. Acknowledgements We would like to thank Marco Dinis for assistance with GIS consulting and Maria João Magalhães for technical lab assistance. Contributor Information A. J. D. Marques, Email: adam.d.marques@gmail.com. G. Velo‐Antón, Email: guillermo.velo@uvigo.gal. Data Availability Statement The data presented in this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.17456884. Raw sequence data is also available through the Sequence Read Archive (PRJNA1428781). References Associated Data Supplementary Materials Figure S1: Landmarks (red‐filled points) and semi‐landmarks (white‐filled points) recorded on the dorsal view of salamanders' snout. Figure S2: Hill numbers 0 (species richness), 1 (Shannon diversity) and 2 (Simpson diversity) for each site were estimated based on rarefaction curves. Points correspond to the asymptote of the curve (qD.Asy), twice the minimum sample coverage threshold (qd.SC*2) or the minimum samples coverage threshold (qD.SCmin), with error bars corresponding to the upper and lower confidence intervals. Figure S3: Spearman's rank correlations between Shannon's diversity estimate using a minimum sample coverage threshold (qD.minSC), search area, and each environmental variable within 100, 200 and 400 m buffer distances from the search area. Significant correlations after correcting for the false discovery rate are indicated with an asterisk (*). Figure S4: Spearman's rank correlation between prey frequency of occurrence and environmental variables. Significant correlations after correcting for the false discovery rate are indicated with an asterisk (*). Figure S5: Results of Mantel tests testing the correlation of Bray‐Curtis pairwise distance matrices for (A) prey phylum and environmental variables (area = search area, IS = impervious surface, TC = tree cover, WF = woody features) at variable buffer distances matrices; (B) prey phylum and morphological traits (mean snout shape and allometric attributes). Significant correlations after correcting for the false discovery rate are indicated with an asterisk (*). Figure S6: Least‐Squares mean distances (LSmeans) between pairs of studied populations. Shape includes (a) size related variation. Pairwise differences in (b) distance, (c) angle and (d) length among allometric slopes of studied populations. Figure S7: Boxplots indicate the variance in centroid size (CS) of snouts among each sampled site. Figure S8: Bars indicate the R 2 coefficient for correlations between phenotypic variation and total species richness (Richness), estimated Shannon Diversity Index for all prey (qD.Total) and each of the three diet phylum (arthropoda = qD.ARTH, mollusca = qD.MOLL, annelida = qD.ANNE). All correlations were found to be not significant. Table S1: For each site visited the number of individuals collected, visitations, total pellets and the proportion of pellets collected per visit are listed. Sites Casa Tait, Vairão, Parque Real and Malta were subsequently excluded from analysis. Table S2: For each site considered in the analysis, the area searched in hectares as well as mean proportion of impervious surface (IS), tree cover (TC) and the proportion of area with wood features (WF) within the buffer area. Variable buffers include within 100, 200 and 400 m from the search area. Table S3: Results of the permutational ANOVA used to test the snout shape centroid size (CS) covariation, site differentiation and common allometric slopes among sites. Table S4: The taxonomic designations for all identified prey genera are listed up the genus resolution. For each, the number of corresponding zOTUs, the minimum % similarity to a reference sequence, and the proportion of samples were the taxa were observed are also listed. Table S5: Using pellets collected from only one visitation per site corresponds to the following changes in observed species richness (Obs), the estimated sample coverage (SC), and the estimated Shannon Diversity Index for all prey (Est.) and each of the three identified phylum separately (Arth. = arthropoda, Moll. = mollusca, Anne. = annelida). Table S6: Mantel tests results for correlation between Pianka niche similarity and distance matrices mean snout shape (Shape), allometric attributes (Allo), environmental variables (IS = impervious surface, SA = search area, TC = tree cover, WF = wood features). Table S7: Phenotypic variation (phenoV) values for each studied population. Data Availability Statement The data presented in this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.17456884. Raw sequence data is also available through the Sequence Read Archive (PRJNA1428781).