Introduction

Morphological integration and modularity have become prominent topics in evolutionary biology, especially since the beginning of this century. Integration refers to the cohesion degree between traits resulting from the interaction of biological processes (adaptive, developmental, among others) that produce them; while modularity refers to the varying degree of connectivity among traits, leading to the formation of modules1. Integration can be measured by the extent to which traits covary2, whereas modularity is assessed by the degree to which certain functional traits covary more than others3. As Hallgrímsson et al.4 highlight, the integration of functionally related body parts and the modular organization of development are fundamental determinants of their evolvability, i.e., the ability of organisms to respond to selective pressure.

Full-body integration and modularity in the skeleton of modern birds have been studied using geometric morphometrics2,5. For example, Orkney et al.2 examined patterns of integration in the shape and size of 13 skeletal elements across all birds, distinguishing between Telluraves and non-Telluraves, and comparing these patterns with diet, flying style, and foot use. They found that size integration is high within elements and that there is strong modularity between elements, while shape integration shows a more diffuse pattern, with occasionally strong covariances between skeletal parts2.

Networks have been used as graphical tools for representing and studying skeletal integration and modularity in animals3. For example, Bell et al.6 studied the skeletal correlations within and between the different bones of the hind limbs and forelimbs of birds, bats, and pterosaurs using linear measurements and partial correlation analysis, representing the correlations as links in networks.

Functional traits are defined as characteristics whose variation influences organismal performance or fitness7,8. Several morphological functional traits are recognized in birds7, but how coordinated their variation is remains unknown. In birds, these traits are continuous measurements of the body (the body mass) or of functionally relevant anatomical parts (such as the beak, wings, legs and tail). Functional trait network analysis9 uses the correlation between multiple functional traits in a group of organisms to build weighted and undirected networks, aiming to capture the functional integration, modularity and relative importance of these traits within a given anatomical part or across the whole body. Trait networks analysis is an approach toward resolving the relationships among multiple traits and their significance. In this abstraction, the nodes of the network are the functional quantitative traits and the links are the correlations between these traits, being the strength of the correlation represented as the weight of the link. The analysis of network parameters is then used to assess organismal integration9 (networks with high density, high average degree and shortest path lengths) and modularity10 (exploring algorithmic community detection). Additionally, node parameters like the degree and strength can identify hub nodes, which are traits that, when modified (for example, by natural selection), can trigger major phenotypic changes10.

Large datasets of animal traits with comprehensive species-level sampling enable the use of network-based tools as novel approaches to address longstanding issues in evolutionary morphology, such as integration and modularity. In this contribution, we aim to assess: (1) which functional morphological traits within the networks are more correlated with all others, and thus potentially triggering comprehensive phenotypic changes; (2) whether functional morphological traits in birds from different primary lifestyles and trophic niches are more integrated or not; and (3) if these ecological variables result in different functional trait modules.

Results and discussion

The networks represent how different functional traits are correlated according to various primary lifestyles (Fig. 1) and trophic niches (Fig. 2). Each node represents a functional trait; the size of the nodes varies based on the number of connections to other nodes, and the thickness of the links reflects the strength of the correlation. Node colors indicate which module they belong to.

Fig. 1
Fig. 1
Full size image

Network models for the different primary lifestyles. (a) Generalist, with an illustration of the meaning of the different measurements based on Tobias et al.7. (b) Aerial. (c) Aquatic. (d) Insessorial. (e) Terrestrial. Color of nodes according to their belonging to different modules. Link thickness according to the strength of the correlation between the nodes.

Fig. 2
Fig. 2
Full size image

Network models for the different trophic niches. (a) Aquatic predator. (b) Frugivore. (c) Granivore. (d) Herbivore. (e) Invertivore. (f) Nectarivore. (g) Omnivore. (h) Vertivore. Color of nodes according to their belonging to different modules. Link thickness according to the strength of the correlation between the nodes.

Most networks have all their nine nodes (i.e., functional morphological traits) interconnected, with the exception of four particular networks. Three of them have a large component with seven connected nodes and a separated small component with two nodes connected to each other (granivores, nectarivores, and aerials). There are no clear differences between the networks based on primary lifestyles and those based on trophic niches. Considering the different primary lifestyles (Fig. 1), the networks appear very similar, except for that of aerial birds, which has a smaller independent component containing the traits beak length from the tip to the culmen and beak length from the tip to the nares. Nevertheless, all the connections in aerial birds are strongly correlated, as indicated by the thickness of the links. Considering the different trophic niches (Fig. 2), the network of granivores has a smaller independent component containing the traits beak depth and beak width. A similar pattern can be observed in the network of nectarivores; however, in this case, the smaller component containing the traits beak length from the point to the nares and beak length to the culmen. The fourth exceptional network, the one of the vertivores, have all the nine nodes completely connected, meaning that all traits are correlated with each other.

Node centrality and functional traits triggering phenotypic changes

We obtained two node parameters that measure the centrality of different functional traits based on the number and strength of their high correlations with other traits. These parameters are the degree k_i, which measures the number of connections for each node, and the strength s_i, which accounts for both the number and the strength of these connections, corresponding to the strength of the correlations.

Body mass and wing traits are the ones with the highest degree and strength (Supplementary Tables S1 and S2). Traits with a high degree are considered hubs. A hub plays a central regulatory role in the network and influences the overall phenotype9,11. These are traits whose evolutionary modification triggers significant changes in other traits. Conversely, traits with a low degree considered as satellites are peripheral in the network, and their evolution occurs with a degree of independence from the rest of the phenotype, although all links within these networks are based on strong correlations.

While body mass has long been a fundamental trait in ecological and evolutionary research, other avian morphological traits have gradually been incorporated to enhance our understanding of ecological niches and trophic interactions7. Differences in beak morphology are related to the capture and processing of food; while differences in wings, tails, and legs are associated with locomotion12.

In most primary lifestyles (Supplementary Table S1), body mass is the most central trait, except in terrestrial and aquatic birds. In terrestrial birds, the most central are the wing traits (wing length from carpal joint to wingtip and wing length from carpal joint to tip of the outermost secondary), which may reflect the major role that flightlessness plays in their morphological evolution. In aquatic birds, the most central trait is the beak length from the tip to the nares, suggesting that the variety of aquatic morphotypes (such as non-divers, foot- and wing-propelled divers) may be primarily influenced by this trait.

In most trophic niches (Supplementary Table S2), the wing traits (and particularly, the wing length from carpal joint to tip of the outermost secondary) are the most central traits, except in nectarivores and frugivores. In nectarivore birds, like hummingbirds, lories and small passerines, the most central trait is body mass, which may reflect how this trait is closely tied to the physiological demands of this trophic niche. In frugivore birds, the most central trait is the length of the beak from the tip to the culmen, suggesting that the variety of beak morphologies and their biomechanical advantages, may be primarily influenced by this trait.

Integration of functional morphological traits

The basis from which the concepts of integration and modularity arise is that the whole body of organisms is integrated to some degree and that there is evolutionary covariation between all traits, nevertheless some body parts covary more than others and, thus, are more integrated than others. In network analysis, density is a parameter directly related to integration, as a group with more covariant traits will have a network with more links, resulting in higher density values9,10. Average strength is a similar metric, but unlike network density, it does not account only for the presence or absence of links, but also the weight of the links that connect each node. Therefore, a network with more but weaker links could potentially have a higher density and lower average strength. Average shortest path length is another measure of connectedness. Larger shortest paths can be interpreted as indicating more independent traits, thus, less integrated9. In our case, we considered the weight of the links in this measurement, where nodes with heavier shortest paths are considered “closer” to each other. More integrated phenotypes are expected to exhibit higher values of density and average strength, as well as smaller values for average shortest path length9. However, average shortest path length becomes an unreliable measurement to compare integration if there are unconnected nodes, as in that case the algorithm tends to include only the shortest paths within each cluster.

Considering the networks of the different primary lifestyles (Supplementary Table S3), the low density value for aerials was expected, as the network has a free element. Nevertheless, aquatics, besides having a low density value, also present the lowest average strength and the highest average shortest path length, making them the group of birds with the lowest integration between their functional traits. A loose network does not necessarily mean that there is independent evolution for every trait, but it may mask the existence of morphotypes or highly integrated yet distinct phenotypes in aquatic birds. This could potentially separate non-divers from foot- and/or wing-propelled divers, as significant differences between these groups —alongside strong convergences within each group— are well known13,14,15,16,17. This differentiation should be considered in future research.

The remaining primary lifestyles show very similar parameter values (Supplementary Table S3). Terrestrials have one additional link compared to the other two groups, reflected in a slight increase in density and a decrease in average shortest path length. The fact that terrestrials follow the same pattern as the rest suggests that, unlike aquatics and at least when considering functional traits, there are no clear morphotypes separating flying taxa from flightless taxa—even though, as mentioned above, wing traits are the most central. Generalists show a slight increase in average strength compared to the other groups, indicating slightly stronger correlations between traits. This highlights that avian functional traits are highly interconnected, forming an integrated phenotype. Specialization can develop through the freeing of a functional trait and its co-option for a new or specialized function.

Considering the networks for the different trophic niches (Supplementary Table S3), there is greater variation in the values of parameters related to integration compared to those for the different primary lifestyles. This may indicate greater independence among the different traits or the presence of several morphotypes tied to each trophic niche.

Three networks stand out for having the lowest density values: those of granivore, herbivore, and nectarivore birds. These networks also have the lowest average strength values. Nevertheless, granivore and nectarivore networks also exhibit relatively low values for average shortest path length, suggesting that the elements composing these networks may be integrated. At the opposite end of the integration spectrum is the network of vertivores. In this network, every node is connected to every other node, resulting in a density value of one. The values of average strength and average shortest path length further indicate high integration. The reason why the vertivore bird phenotype shows such high integration, at least in its functional traits, remains an open question. One possible explanation is the ecological constraint imposed by the habit of capturing and eating other land vertebrates, which may drive the strong covariation between skull and postcranial functional traits. This factor may also explain the high covariation observed between accipitrids and falconids18, although this category also includes many other birds, such as owls and butcherbirds.

All pairwise differences between the parameters of the different networks were significant (p < 0.05).

Functional modules within ecological variables

Despite its widespread use in evolutionary and developmental biology, the concept of modularity remains theoretically and operationally ambiguous. While modularity is often invoked to explain morphological diversification, evolvability, and complexity19, its empirical detection and definition vary widely across disciplines and methodological frameworks. In functional and morphological studies, modules are frequently defined post hoc, without clear a priori criteria or consensus regarding their biological reality20. This raises a fundamental question: is modularity a real and biologically meaningful property of organisms, or is it an artifact of the analytical tools and assumptions we impose? We draw on the concept of modularity to interpret patterns of morphological variation, while acknowledging ongoing debates about its epistemological and methodological foundations. Rather than treating modularity as a self-evident property of biological systems, we consider whether the notion of modules reflects an inherent aspect of organismal organization or a construct shaped by analytical frameworks and assumptions.

In our network analysis, when modules are present, they tend to separate beak traits from postcranial traits, following the same pattern observed when smaller traits are disconnected from the network (Supplementary Figs. S1 and S2). Moreover, networks that exhibit modules usually have only two beak characteristics in the smaller module: either beak width and beak depth or beak length from tip to the culmen and beak length from the tip to the nares. Regarding beak traits, although one beak measurement is included in the other (beak length from the tip to the nares within beak length from the tip to the culmen), these are the most correlated traits only in aerials (Supplementary Fig. S1) and in nectarivores (Supplementary Fig. S2). Within the trophic niches (Supplementary Fig. S2), there is greater variation in the exhibited modules. In two networks —omnivores and frugivores— beak length from the tip to the nares is grouped with the remaining beak traits, while beak length from tip to the culmen is part of the module containing postcranial traits. Additionally, in omnivores, tail length is grouped with all the beak traits except with beak length from tip to the culmen. Herbivores are the only example where all the beak traits are grouped in one module and all the postcranial traits in another, indicating a seemingly independent variation of the skull and the rest of the body. Finally, only terrestrial carnivorous birds (vertivores and invertivores) show no modularity whatsoever (Supplementary Fig. S2).

Conclusions

Trait network analysis is a valuable tool to understand the behavior of the avian phenotype when considering birds grouped by different ecological factors. In this case, the different primary lifestyles and trophic niches where functional traits can vary their influence over the rest of the phenotype, and the phenotype itself can be more or less integrated and exhibit different patterns of modularity.

Most primary lifestyles and trophic niches networks are connected, suggesting that functional traits tend to exhibit high covariation. In cases where the network includes free elements, these smaller components typically consist of two nodes representing two of the four beak traits. A similar pattern is reflected in the network modules: when present, these modules tend to separate certain beak traits from the rest.

Network configuration shows greater variation in trophic niches than in primary lifestyles, with two networks featuring free elements, and with one network being noticeably loose (herbivores) and another network especially tight (vertivores).

In all cases, body mass and wing traits are among the most central traits, making them key factors in driving major phenotypic variation and exerting significant influence over other traits. In some networks, additional traits also become more central: within primary lifestyles, aquatic birds show increased centrality in beak length from the tip to the nares; while in trophic niches, frugivores exhibit increased centrality in beak length from tip to the culmen. However, beak traits are not consistently central across different trophic niches. This could be because functional traits do not necessarily have a strong impact on overall phenotype; instead, they may be co-opted for specialization and thus become “freed” from the rest of the body.

Modularity and integration are not opposing processes but rather exist in tension. While having modules limits overall integration, the modules themselves can still be highly integrated. Birds with different primary lifestyles, and particularly those with different trophic niches, exhibit variation in both the number and nature of their modules. The ability of traits and body parts to evolve independently in response to a lineage’s adaptation to a new lifestyle or trophic niche appears to be a necessary condition for specialization, although some specialized lifestyles (such as the vertivores analyzed herein) demand a high degree of integration among traits, resulting in more constrained morphological variation.

Methods

Data acquisition

We extracted body mass, functional morphological traits and ecological variables data about all the avian species within AVONET7, a large dataset containing individual-level measurements of continuous morphological traits, ecological variables, and information on range size and location for 11,009 extant bird species across 36 orders worldwide.

The functional morphological traits are defined as those key external morphological characters that influence organismal performance or fitness7. These include: (1) beak length from tip to the culmen, (2) beak length from the tip to the nares, (3) beak depth, (4) beak width, (5) tarsus length, (6) wing length from carpal joint to wingtip, (7) wing length from carpal joint to tip of the outermost secondary; (8) tail length, and (9) body mass (Fig. 1). For details on measurements and protocols see Tobias et al.7.

Two categorical ecological variables (and their categories) were used7,12, including primary lifestyle PLS (i.e., predominant locomotory niche while foraging), and trophic niche TN. The categories within primary lifestyle include: aerial, insessorial (birds perching on trees and other substrates), terrestrial, aquatic, and generalist birds. The categories within the trophic niches include: aquatic predator, frugivore, granivore, herbivore, invertivore, nectarivore, omnivore, and vertivore birds. Details about the number of species and diversity of families within each category are in supplementary information. Further details on each ecological variable and their categories are in Tobias et al.7 and its supplementary data.

Network construction and analysis

We build adjacency matrices and undirected networks for each category within each of the two ecological variables (including only those categories with 100 species or more). Trait data were log-transformed before the confection of the networks. The functional morphological traits were formalized as nodes, and the correlation between the traits of all the species included in each category were formalized as the links. We only modeled the moderate to highly correlated traits (i.e., those with Pearson’s correlation coefficient ≥ 0.6) and the value of the correlations were considered as the weight of the links. The correlation of each trait with itself was discarded.

Node centrality was measured by two parameters: (1) the connectivity degree (k_i), which is the number of links connecting one node with all other; and (2) the strength (s_i), which is the analog of k_i in weighted networks, but considering the sum of weight of the links21.

For the networks we obtained three parameters related to integration: (1) density (D), which is the ratio between the number of links in a network and the number of total possible links; (2) average shortest path length (APL), which is the average of all the shortest paths in the network, that is, the minimal number of links between two nodes; and (3) average strength (AS), which is the average of all the strengths in the network.

The number and composition of modules within the networks was presented as dendrograms, and calculated using the cluster walktrap algorithm22. This algorithm performs short random walks between nodes, starting from different nodes and traversing a few links in each iteration. The short walks will tend to favor closely connected nodes, thereby detecting densely connected subgraphs22. In our analysis, we accounted for the weight of the links, treating heavier links as shorter walks between nodes.

The adjacency matrices, networks, node centrality parameters, network parameters, and modularity were obtained using the igraph package23,24 of R25.

Statistical analysis

We tested the significance of the differences between the several categories within each ecological variable. All trait data were log-transformed before the analyses. The categories were bootstrapped with 5000 iterations of random resamplings, each resampling recovering a number equal to the ninety percent of the species of the smallest category for each variable (i.e., 254 for primary lifestyle, and 161 for trophic niche). For each resampling, a trait network was constructed so that each trait network had the same number of species. This was done because the parameters revealing some of the relationships between the traits in the networks were found to be affected by the number of species in the network10. We made the hypothesis testing by pairwise comparisons for each network parameter using a Wilcoxon rank-sum test, using a level of significance of p < 0.05. All statistical analyses were performed using R25, and data manipulation within R was performed using the Tidyverse collection of packages26.