Cranial integration in the ring-necked parakeet, Psittacula krameri (Psittaciformes: Psittaculidae)

The study of integration and modularity aims to describe the organization of components that make up organisms, and the evolutionary, developmental and functional relationships among them. Both have been studied at the interspecific (evolutionary) and intraspecific (phenotypic and ontogenetic) levels to different degrees across various clades. Although evolutionary modularity and integration are well-characterized across birds, knowledge of intraspecific patterns is lacking. Here, we use a high-density, three-dimensional geometric morphometric approach to investigate patterns of integration and modularity in Psittacula krameri , a highly successful invasive parrot species that exhibits the derived vertical palate and cranio-facial hinge of the Psittaciformes. Showing a pattern of nine distinct cranial modules, our results support findings from recent research that uses similar methods to investigate interspecific integration in birds. Allometry is not a significant influence on cranial shape variation within this species; however, within-module integration is significantly negatively correlated with disparity, with high variation concentrated in the weakly integrated rostrum, palate and vault modules. As previous studies have demonstrated differences in beak shape between invasive and native populations, variation in the weakly integrated palate and rostrum may have facilitated evolutionary change in these parts of the skull, contributing to the ring-necked parakeet’s success as an invasive species.


INTRODUCTION
Although an organism must function as a complete integrated whole in order to survive, the component parts of that organism can in fact grow, function and evolve semi-independently from one another.This phenomenon is described by modularity: the degree to which collections of organismal traits form independently varying and evolving units (Olson & Miller, 1958;Klingenberg, 2009).The covariation between modules or among traits within a module, termed 'integration', arise from their genetic, developmental and functional associations (Olson & Miller, 1958).By quantifying aspects of integration and modularity such as strength, pattern and change over time, it is possible to gain insight into the functional, genetic and developmental systems that give rise to this variation (Felice et al., 2019a).
The relationship between the magnitude of phenotypic integration and the evolution of phenotypic variation is complex and seems to vary across clades and systems.In the archosaur skull some aspects of evolutionary integration are consistent across a wide range of taxa (such as high integration in the occipital region), and some parts of cranial integration vary greatly between groups.These differences are exhibited in the patterns of integration between the quadrate, pterygoid and jugal in birds and nonavian dinosaurs (Felice et al., 2019a).Previous work has suggested that variation in integration and modularity may impact phenotypic evolution and explain differences in evolutionary patterns observed across taxa (Cheverud, 1996).In carnivorans and primates, high levels of integration have been found to constrain disparity, restricting evolution by limiting the extent and direction in which selection can act on variation (Goswami & Polly, 2010).In some cases, however, the opposite effect is observed.In the felid axial skeleton, higher integration appears to facilitate greater response to selection (Randau & Goswami, 2017).Some studies have found no evidence of a relationship between strength of integration and disparity at all (Bardua et al., 2019a).
Recent research has begun to reveal the complex relationships among cranial traits in birds (Bright et al., 2016(Bright et al., , 2019;;Felice & Goswami, 2018;Navalón et al., 2020).Some studies have found the avian cranium to be highly modular, with modules of differing degrees of integration evolving at different rates.Some believe that this leads to a range of highly specialized species and that high-level modularity is associated with high evolutionary rate and low disparity (Felice & Goswami, 2018;Felice et al., 2019a).Others attribute the highly specialized forms seen in some birds to a more highly integrated pattern of modularity (Bright et al., 2019).Research that features honeycreepers and Darwin's finches, for example, suggests that high levels of integration between the rostrum and the rest of the skull are associated with a high rate of evolution (Navalón et al., 2020).Finally, some studies have found that allometry (the relationship between shape and size) is a significant predictor of variation in the cranium (Klingenberg, 2016;Bright et al., 2019).The results of these investigations of evolutionary integration are often interpreted as a reflection of the correlations imparted by the shared underlying developmental systems that generate variation in these traits.To truly understand the link between evolutionary integration and developmental/ population scale phenomena requires quantifying phenotypic integration at intraspecific scales.However, although studies into intraspecific integration and modularity are becoming more common (Marshall et al., 2019;Bon et al., 2020 andsee Parr et al., 2016), investigations in birds are still lacking.Here, we quantify cranial integration and modularity in a single species of parrot, the ring-necked parakeet, to test whether evolutionary integration is indeed a reflection of population-scale integration patterns.
The Psittaciformes (parrots) is an order containing approximately 390 species (del Hoyo et al., 2020), split into three 'superfamilies' of the Psittacoidea ('true' parrots), the Cacatuoidea (cockatoos) and the Strigopoidea (New Zealand parrots).Parrots exhibit a set of cranial novelties that include a vertical palate, ossified arcus suborbitalisas and a pseudoprokinetic cranio-facial hinge that is thought to provide increased agility and a greater range of movement (Tokita, 2003).It has been argued that this adaption is key to their survival as it allows them to access the mechanically restrictive food that they require, although this is debated (Tokita, 2003;Bright et al., 2016).
Studies investigating evolutionary patterns of integration and modularity in parrots provide support for several competing hypotheses of modularity in the skull.Felice & Goswami (2018) found a high level of modularity across bird species (including the Psittaciformes), with different modules evolving at different rates.Furthermore, this study proposed that the unique palate in parrots is a self-contained module that has evolved independently at a higher rate than other parts of the skull (Felice & Goswami, 2018).Other research, however, suggests that rather than high modularity, high levels of allometry and integration throughout the parrot cranium explain the majority of variation observed (Bright et al., 2019).Methodologies can differ greatly in their approaches to measuring shape data: Felice & Goswami (2018) did not control for allometry and used a high-density landmark approach, which is typically less influenced by allometric effects (Goswami et al., 2019).Allometric effects can be considerable regardless of the degree of modularity, especially within less inclusive clades where size-related variation may dominate.Given the size of the Psittaciformes order and variation within it, it is therefore likely that allometry has a stronger effect within the Psittaciformes than across all birds (Bright et al., 2016(Bright et al., , 2019;;Klingenberg et al., 2016;Marshall et al., 2019).
Evolutionary (interspecific) patterns of modularity and integration often replicate static (intraspecific) patterns (Klingenberg, 2014), although in some taxa this is not the case (Urošević et al., 2012).In the fire salamander, intraspecific modularity patterns closely match results of studies investigating intraspecific variation in caecilians, as well as interspecific patterns across the wider caecilians clade, despite osteological differences (Bardua et al, 2019a;Marshall et al., 2019;Bon et al., 2020).In contrast, cranial analysis in squamates appears to vary across the taxa; research focussing on evolutionary integration has revealed high levels of modularity across the clade (Watanabe et al., 2019), whereas research investigating static integration has found the cranium to be more integrated (Urošević et al., 2012).It has been suggested that a strong degree of static integration is associated with function and the more modular evolutionary patterns are more influenced by development; this implies that functional modularity can be adaptive (Urošević et al., 2019).
Here, we investigate static patterns of modularity and morphological integration in the cranium of the ring-necked parakeet, P. krameri, using a highdensity geometric morphometric approach.Our aim is to investigate the roles of integration and Downloaded from https://academic.oup.com/biolinnean/article/133/1/47/6188526 by Institute of Child Health/University College London user on 30 May 2022 modularity in the evolution of the specialized skull of the Psittaciformes order and determine whether patterns of phenotypic integration and modularity in this species reflect those of studies of evolutionary integration in birds.We test 13 different hypotheses (Supporting Information, Table S1) concerning the organization of nine cranial regions, based on results found in studies of avian evolutionary modularity.Furthermore, we investigate the influence of allometric size on cranial shape and integration and quantify the relationship between morphological diversity and within-module integration.We predict that the skull of the ring-necked parakeet will show high levels of modularity, with individual skull regions showing distinct patterns of disparity and within-module integration.Our results will enable us to compare and contrast intraspecific variation within this particular species with interspecific variation between other parrots, among birds and with tetrapods in general.

SpecimenS and data collection
The sample consisted of 37 alcohol-preserved adult specimens of P. krameri of unknown sex and age, retrieved from the collection of the Natural History Museum (NHM) at Tring, Hertfordshire, UK.A full list of specimens can be found in the Supporting Information (Table S2).Specimens were scanned using a Nikon Metrology X-Tek HMX ST 225 micro-CT scanner at NHM, London, UK.Tomographs were processed into surface files using Avizo Lite v.9.3 (FEI, Hillsboro, OR, USA) before being imported into Geomagic Wrap 2017 (3D Systems, Inc. Rock Hill, South Carolina, USA), where the lower mandible and postcranial skeleton were removed and any artefactual holes filled.Based on suture development, all specimens appeared to be sexually mature.Before being exported from Geomagic Wrap, the 'decimation' tool was used to reduce the specimen to under one million faces to reduce the file size without compromising quality of detail.Some specimens that had been damaged on the right-hand side were mirrored using the 'flip' function in Geomagic, so the same side of each specimen could be used for landmarking (Supporting Information, Table S2).

morphometric analySiS
Landmarks (Type I and II) and semilandmark curves were placed by a single researcher (M.J.M.) using Stratovan Checkpoint (Stratovan, Davis, CA, USA).Landmarks were placed bilaterally and semilandmark curves were placed on the right side only (Fig. 1; Supporting Information, Tables S3, S4).The skull was split into nine different regions, with landmarks and semilandmark curves used to define each area (Fig. 1; Supporting Information Tables S3, S4).Shape data were imported into RStudio v.1.1.419running R v.3.6.1.(R Core Development Team, 2020) and landmarks and semilandmarks were checked thoroughly before conducting any analysis.Any landmarks that could not be placed due to damage to the skull were marked as 'missing data' in Checkpoint and coordinates were estimated   S2). Curves were subsampled to a consistent number of points across specimens (Supporting Information, Table S4).
The 'createAtlas' function of morpho v.2.8 (Schlager, 2017) was used to create a template of the surface semilandmarks on a hemisphere model.The 'placePatch' function of morpho v.2.8 (Schlager, 2017) was used to project surface semilandmarks from the model onto the surface of the cranium of each specimen using 'inflate' and 'tol' values that were optimized for each region (Bardua et al., 2019b; Supporting Information, Table S5).The patch data from each part of the cranium were then combined so that the right side of each specimen was fully covered in patch points (Bardua et al., 2019b).The 'slider3D' function of morpho v.2.8 (Schlager, 2017) was used to slide the landmarks along the curves and surfaces of the skull to minimize bending energy (Gunz et al., 2005).We then used the 'mirrorfill' function of paleomorph v.0.1.4(Lucas & Goswami, 2017) to mirror landmarks from the right side of the skull over to the left, by linking any landmarks on the left side to their corresponding right-hand side landmarks.Mirroring landmarks aimed to remove any potential artefacts caused by Procrustes analysis when applied to only one side of the skull (Cardini, 2016).Some surface semilandmarks were consistently projected onto the wrong part of the skull; these were removed, leaving a total of 38 anatomical landmarks, 24 curves and 384 patch points placed onto each specimen (Supporting Information, Tables S3-S5).We then carried out a generalized Procrustes alignment on our morphometric data using the 'gpagen' function of geomorph v.3.2.1 (Adams et al., 2020) to remove the non-biological variation caused by differences in translation and rotation, as well as isometric scaling.Mirrored landmarks were then removed before further analysis.

allometry
To estimate the influence of allometry on shape variation in our data set, we fit permutational linear regressions using the 'procD.lm'function of geomorph v.3.2.1 (Adams et al., 2020), using skull centroid size as the size value and 1000 permutations.

principal componentS analySiS
A principal components analysis (PCA) was conducted on the shape data using the 'PlotTangentSpace' function of geomorph v.3.2.1 (Adams et al., 2020) to determine the main axes of shape variation within the data set and explore the distribution of specimens in morphospace.The 'plotRefToTarget' function of geomorph v.3.2.1 (Adams et al., 2020) was then used to warp a specimen to represent the extremes of PC1 and PC2 to help visualize the data.

modularity and integration
We tested 13 different hypotheses for modularity, from maximum modularity of nine modules (rostrum, naris, jugal, vault, quadrate, basisphenoid, pterygoid, occipital and palate) to a layout of just two: the face and neurocranium (Supporting Information, Table S1).Modularity was tested using both the 'Evaluating Modularity with Maximum Likelihood' (EMMLi) and Covariance Ratio (CR) methods.Firstly, the 'dotcorr' function of paleomorph v.0.1.4(Lucas & Goswami, 2017) was used to determine the correlation between landmarks.The 'EMMLi' function of the EMMLi v.0.0.3 package (Goswami et al., 2017) used these correlations to compare the different hypotheses and calculate the between-module and within-module integration for each module of each hypothesis.Because EMMLi may prefer overparameterized models, we then conducted a CR analysis using the 'modularity.test'function of geomorph v.3.2.1 (Adams et al., 2020) to quantify support for the two bestsupported hypotheses by testing which showed the higher modular signal.This approach quantifies the strength of modularity by comparing between-module covariance to within-module covariance (Adams, 2016).Each modularity test was run with 1000 iterations.The 'compare.CR' function of geomorph v.3.2.1 (Adams et al., 2020) was used to compare the results of these tests.

morphological variance
We used the 'morphol.disparity'function of geomorph v.3.2.1 (Adams et al., 2020) to quantify the variance within each module.The result was used to investigate the relationship between the Procrustes variance and within-module integration (calculated from EMMLi) for each module.We also used the 'per_lm_variance' function of the hot.dots package v.0.0.0.9 (Felice et al., 2018) to calculate the per-landmark variance of each individual landmark on the cranium, allowing us to visualize the distribution of variation across the entire skull.

RESULTS allometry
The effect of centroid size on shape of the P. krameri skull was not significant (R 2 = 0.025, P = 0.586).As such, no correction for the effects of allometric size were made for subsequent analyses.

principal component analySiS
The PCA of the cranium identified 36 principal components (PCs) that together explained 100% of the variation of the cranium; 95% of the variance was explained by PC1-PC28.The main axis of variation was represented by PC1, which exhibited 19.7% of the total variation, mostly describing variation in the depth of the curve of the cranial vault, the circumference of the orbit and the length and curvature of the rostrum.PC2 represented 9.5% of the total variation and was driven by variation in the width and curvature of the underside of the rostrum as well as the shape of the quadrate (Fig. 2).

EMMLi
The most supported hypothesis from the EMMLi analysis was the maximum modularity hypothesis (Supporting Information, Tables S1, S6).This suggests that all nine defined regions of the skull (rostrum, naris, jugal, vault, quadrate, basisphenoid, pterygoid, occipital and palate) are independent modules.Highest between-module integration was found between the rostrum and naris, and the pterygoid and occipital (Fig. 3).The highest within-module integration was found in the pterygoid (Fig. 3; Table 1).Withinmodule integration was greater than between-module integration for all nine modules.

Covariance ratio
The modularity.test for the 'all modules' hypothesis (nine distinct cranial modules) produced a CR score of 0.624 (P = 0.001).The modularity test for the sevenmodule hypothesis [e.g. the hypothesis supported in Felice & Goswami (2018)] showed a slightly higher CR score of 0.642 (P = 0.001), signalling less support.When 'compare.CR' was used to compare support for the hypotheses, the nine-module hypothesis was found to have the stronger signal (z-score of -28.4 compared to -28.0, P = 0.037).These results support those of the EMMLi analysis, that the nine-module hypothesis best reflects the pattern of integration within P. krameri.

morphological variance
The rostrum, cranial vault and palate showed the highest morphological disparity (Procrustes variance), with the lowest disparity found in the naris (Table 1).A linear regression between Procrustes variance and within-module integration found a significant negative relationship, suggesting that as withinmodule integration increases, morphological variance decreases (adjusted R 2 = 0.705, P = 0.0284; Fig. 4).Most per-landmark variance was found in landmarks around the edge of the orbital, the tip of the rostrum, the palate, and also on the ventral posterior of the cranium between the orbital and the occipital region (Fig. 5).

DISCUSSION
P. krameri exhibits a pattern of high modularity in the skull.From 13 hypotheses representing different patterns of modularity between the nine major cranial regions, both EMMLi and CR analyses found most support for our hypothesis of maximal modularity, suggesting that the cranium of P. krameri is split up into nine distinct modules.This provides evidence that the evolution of the unique cranial features possessed by parrots may be the result of each region of their cranium being a highly integrated subunit that can respond to selection and evolve somewhat independently from the others.
This pattern of phenotypic integration, found from measuring an intraspecific data set, can be compared to the results of evolutionary (interspecific) studies of integration and modularity in birds to determine whether evolutionary patterns replicate those we observed here at the intraspecific level.Felice & Goswami (2018) studied evolutionary integration and modularity in a wide range of birds and found the avian cranium to be highly modular, albeit with a slightly different pattern.Their results show the pattern of modularity in the avian cranium to be made up of seven modules, using the same modularity pattern used here, but with the rostrum and jugal combined as one module and the pterygoid and quadrate combined as one module.Importantly, this seven-module hypothesis was the hypothesis of maximal modularity in their study: no nine-module hypothesis was defined (Felice & Goswami, 2018).Here, we directly compared our hypothesis of maximal modularity (nine modules) with their most supported hypothesis of seven modules using compare CR, and found that our nine-module hypothesis had significantly more support.
Other studies have found higher levels of cranial integration in birds.Navalón et al. (2020) aimed to determine the relationship between beak and skull Table 1.Procrustes variance (× 10 -4 ) and within-module integration for each of the nine cranial regions, as well as within-module integration scores calculated using EMMLi  Unexpectedly, their results showed that integration was high between the beak and the rest of the skull, and that the whole cranium played a part in the fast   evolution associated with these species (Navalón et al., 2020).They found these results unique among land birds and concluded that both a high degree of modularity and a high degree of integration can facilitate evolution (Navalón et al., 2020).Bright et al. (2019) used different methods to find that variation in the skull was explained by significant allometry and identified high levels of integration between beak shape and skull shape.
A key reason for these differences in findings across integration and modularity studies could be the types of data and methods used.Bright et al. (2019) used an approach consisting of 20 landmarks across the midline and the right-hand side of the skull as well as semilandmark curves along the dorsal midlines of the beak and skull.Here, we used a similar high-density landmark approach to Felice & Goswami (2018) with considerably more anatomical landmarks and curves as well as surface landmarks that span the whole of the right side of the cranium.Previous research has found that results from only using anatomical landmark data (or limited curves) can vary from results that use surface patching approaches; results from a landmark-only data set may emphasize betweenregion correlations over within-region integration because landmarks are concentrated at the borders of regions, whereas surface semilandmarks sample between boundaries (Goswami et al., 2019;Bon et al., 2020).The consequence of this is that results from landmark-only analyses often show considerably weaker support for more modular organizations (boundary bias), simply because they primarily capture shape information at the most integrated regions of elements: their boundaries (Goswami et al., 2019).Moreover, Bright et al. (2016Bright et al. ( , 2019) ) did not specifically test alternate modular structures, but rather measured the integration between the two regions of interest, the beak and braincase.
In further contrast to Bright et al. (2016Bright et al. ( , 2019)), we also found no significant effect of allometry in our data set.This is likely because we were studying a single species rather than two different groups of parrots.As the superfamilies Psittacoidea and Cacatuoidea vary so greatly [and were both included in the study by Bright et al. (2019)], a more comparable study to ours would investigate whether allometry has any effect on an intraspecific data set consisting of cockatoos only or a different species of Psittacoidea.
We also recover a negative relationship between Procrustes variance and within-module integration, implying that high integration within modules restricts disparity.These results mirror research into evolutionary integration in bird taxa that have found high levels of cranial-wide integration restrict variance (Felice & Goswami, 2018).We found particularly high variance within the palate and rostrum of our specimens (as well as low within-module integration), implying that these modules have a high degree of variation and therefore a potentially high evolutionary rate (Felice & Goswami, 2018).These results reflect those found studying birds and certain other mammalian taxa such as carnivorans and primates (Goswami & Polly, 2010;Felice & Goswami, 2018 ), although this complex relationship appears to vary across clades.For example, no consistent relationship between disparity and within-module integration has been identified in squamates (Watanabe et al., 2019) or various clades of amphibians (Bardua et al., 2019b(Bardua et al., , 2020;;Fabre et al., 2020).
High levels of variation in the palate and rostrum may be linked to the ring-necked parakeet's success as an invasive species.A previous study investigating the differences between morphology of populations of ring-necked parakeets in their native range and the invasive populations that inhabit Europe has found that the beaks of those in non-native ranges tend to be bigger and stronger (Le Gros et al., 2016).The palate and rostrum are directly involved in feeding, so high variation in two specific modules may have helped the species adapt to a new environment and food source more easily.Thus, selection on specific beak morphologies may have been facilitated by high levels of rostrum and palate variation within the species, though the relationship between diet and skull shape is complex (Bright et al., 2019;Felice et al., 2019b).These results are in line with the hypothesis that high modularity can facilitate broad ecological tolerances and thus rapid invasion of new habitats (Adams et al., 2007).However, the integration and modularity patterns observed here in P. krameri are similar to those seen in all birds (Felice & Goswami, 2018).Direct comparison between this cosmopolitan taxon and related parrot species that are vulnerable to climate change or habitat loss are needed to test how cranial modularity might influence ecological flexibility and niche conservatism in this clade.
Using a three-dimensional approach, we analysed cranial shape and patterns of modularity and integration in P. krameri, the ring-necked parakeet, finding a highly modular pattern consisting of nine internally integrated modules.We found a significant negative relationship between withinmodule integration and disparity that mirrors results from analyses of evolutionary integration in the avian skull using high-density geometric morphometrics.Our findings support the hypothesis that high levels of modularity can facilitate the development of specialized cranial features that enable organisms to adapt to new ecological niches.Amongst parrots, our findings suggest that this high degree of modularity has facilitated the evolution of the unique palate and cranio-facial hinge that are Downloaded from https://academic.oup.com/biolinnean/article/133/1/47/6188526 by Institute of Child Health/University College London user on 30 May 2022 distinctive to their cranium.The palate and rostrum modules both show a low degree of within-module integration (Fig. 4), implying relaxed constraint.This low integration in turn leads to greater variation in these regions and provides greater opportunity for natural selection to act on them, allowing them to evolve independently into their specialized forms.It may also facilitate higher evolutionary rates than in a more integrated system (Felice & Goswami, 2018).To determine whether this effect is found more widely, further investigation could compare this intraspecific data set with intraspecific data sets of other parrots (particularly cockatoos) as well as other birds.Building on the results of this analysis and macro-scale analyses across birds, further studies into interspecific integration patterns that compare species across the Psittaciformes would clarify the role integration and modularity play in the specialization of the parrot cranium and in the evolutionary diversity and success of both native and invasive populations.

Figure 2 .
Figure 2. Morphological space showing the distribution of specimens according to variation described by PC1 and PC2.PC1 represented 19.7% of variation and PC2 represented 9.5%.

Figure 3 .
Figure 3. Network graph showing the results of the EMMLi analysis; degrees of the within-module and between-module integration in the P. krameri cranium.Circle size represents within-module integration and the thickness of the connective bars represents the degree of between-module integration.B: Basisphenoid region, J: Jugal, N: Naris, O: Occipital region, P: Palate, Pt: Pterygoid, Q: Quadrate, R: Rostrum, V: Vault.

Figure 5 .
Figure 5. Hot dots analysis projected onto specimen 25. 'Hotter' colours represent a greater degree of variation in a specific landmark compared to 'colder' colours, which show less.Most variation can be seen around and behind the orbital, around the palate and on the quadrate.a, lateral view; (b) inferior view; (c) posterior view.