Similarity of stream insect trait profiles across biogeographic regions

Habitat templet theory predicts that the functional niches of species evolved in response to selection pressures imposed by each species' spatial–temporal environment. Consequently, similar environmental conditions should lead to convergence in the biological trait composition of biogeographically independent assemblages. Given their high diversity and ubiquitous occurrence, stream insects represent an ideal group to test convergence. Such an analysis should provide insight into both how spatially variable stream insect traits are and how transferable trait–environment relationships are across large spatial scales. We tested two hypotheses: (1) functional niches of stream insects are similar across Australia, Europe, North America, New Zealand and Southern Africa, and (2) the variability in trait profiles of stream insects is positively related to climatic variability within regions.


| INTRODUC TI ON
The habitat templet theory (Southwood, 1977) posits that environmental conditions shape the functional niches of species by selecting for traits that best maximize individual fitness (e.g.McGill et al., 2006).Consequently, similar environmental conditions should result in convergent evolution of functional traits within assemblages, independent of phylogenetic history (Schluter, 1986;Vellend, 2010).If trait-environment relationships are consistent and strong, they can be used to understand how assemblages will respond to global change (Fukami et al., 2005).Indeed, trait-based approaches have been used to study how different anthropogenic stressors influence assemblages, for example among many others, for freshwater invertebrates (Statzner & Bêche, 2010) or fish (Colin et al., 2018).
In general, the specific trait composition occurring in habitats should depend on the cumulative effects of phylogenetic relatedness, environmental selection, biotic interactions, and stochastic establishment of species in different habitats (Vellend, 2010;Weiher & Keddy, 1995).Phylogeny can influence trait occurrence because species inherit traits from common ancestors, which should persist if there is no strong selection against those traits (e.g.Prinzing et al., 2001).Environmental selection should lead to convergence of traits that are required to withstand environmental conditions, whereas biotic interactions should lead to divergence in habitat and resource use traits via competition and may result in adaptive radiation in habitat and resource use, which could translate into high trait variability within assemblages (e.g.Gounand et al., 2016).
Stochastic processes should result in non-directional effects on the trait composition.
Stream invertebrates are a suitable taxonomic group for testing hypotheses about trait convergence at large scales because they occur ubiquitously and have adapted to live in the highly spatially and temporally heterogeneous habitats created by running waters.
River hydrology and hydraulics interact with geology and topography to form strong chemical and physical gradients and patchiness in streams and a diverse set of habitats-for example different hydrologic and thermal regimes; different channel units such as pools, riffles and runs; and different substrate types such as sand, gravel or cobble (Hynes, 1970;Palmer & Poff, 1997).Moreover, temporal variability can include marked extremes in the amount of water present in channels (floods and droughts, e.g.Aspin et al., 2019) and the temperature of the water.Stream biota have evolved a range of morphological, life history, and behavioural adaptations to cope with, and flourish in, these dynamic environmental conditions (Dijkstra et al., 2014;Lytle & Poff, 2004).These adaptations are expressed as observable traits and are thought to have evolved independently in different regions.The traits possessed by different stream invertebrate species have recently been compiled for different regions (for more details, see Section 2).
Environmental selection acts on entire organisms, and thus, combinations of traits should determine the performance, and ultimately, the distribution of organisms.We refer to the trait combinations that co-occur in organisms as trait profiles.They can be used to delineate groups of organisms that express a similar set of traits, which we refer to as trait profile groups (TPGs).Various synonymous terms for trait profiles to distinguish groups of organisms exist in the literature, for example functional groups or trait syndromes (Poff et al., 2006;Usseglio-Polatera et al., 2000a).Recent studies have started to account for interrelationships between traits when establishing trait-environment relationships (e.g.Pilière et al., 2016).Multiple traits may be linked and either act simultaneously or interact in response to environmental conditions (Poff et al., 2006;Verberk et al., 2013).However, only a limited set of all possible trait combinations seem to occur and appear to be adaptive (Verberk et al., 2013), and several researchers have tried to group stream insects based on these trait profiles (Pilière et al., 2016;Poff et al., 2006;Usseglio-Polatera et al., 2000a).
Previous studies that examined stream invertebrate trait profiles were largely restricted to taxa occurring within a single continent (Europe or North America), but see Brown et al. (2018) andCrabot et al. (2021).Statzner and Bêche (2010) compared invertebrate trait composition between Europe and North America and concluded that invertebrate trait patterns were broadly similar in the two continents, but their comparison was based on a limited number of genera (on average 25% of the most abundant genera per continent).They found the strongest differences in traits describing feeding mode, which might have been partly attributable to differences in how feeding groups were defined in the two continents.Comparing trait profiles of stream invertebrates across multiple continents should (1) help reveal if the patterns of assemblage organization are consistent and thus predictable (Lamouroux et al., 2002), (2) expand our knowledge of the spatial variability of trait profiles, and (3) provide insight into how general trait-environment relationships are across regions (Statzner et al., 2004).
We tested the main hypothesis that freshwater insect functional niches are similar across regions, an indication for their convergence.Therefore, we assessed the similarity between freshwater insect trait profiles in Australia, Europe, North America, New Zealand and Southern Africa.We expected that be climate, indicated by the higher trait profile variation in regions with more diverse climates.

K E Y W O R D S
large-scale comparison, stream insect traits, trait convergence the trait profiles of stream insects in Australia, New Zealand, and Southern Africa would be similar to each other because of their shared biogeographic history (Gondwana origin).Similarly, we expected that the trait profiles of the European and North American faunas would be more similar to one another than to other regions because of their shared biogeographic history (Laurasia origin).Additionally, we expected the trait profiles of the Australia and New Zealand faunas would be most similar to one another because of recent (in terms of evolutionary time scales) dispersalassociated, faunal exchanges (Wallis & Trewick, 2009).We also tested a second hypothesis, that the trait profiles of stream insects are more variable (i.e.broader functional niches) in regions with high climatic variability than in regions with low climatic variability.We expected the New Zealand fauna to exhibit less diverse trait patterns because it is the smallest region we examined with the shortest climatic gradient.

| General approach
To test for the indications of convergence, we used trait datasets from Australia, Europe, North America, New Zealand and Southern Africa (hereafter referred to as regions) covering the corresponding taxa pool.This approach was necessitated by the lack of comprehensive stream insect community datasets at these large spatial scales that are comparable (e.g. are based on the same methods) between regions.We note that while we can show similarity, we cannot elucidate the contribution of underlying mechanisms such as environmental selection, biotic interactions and stochasticity with the data at hand and thereby demonstrate convergence.We (1) established and compared functional niche spaces for each region and (2) clustered the taxa in each dataset based on the traits they possessed to identify groups of taxa with similar sets of traits, that is TPGs.For each region, we identified (1) the number of TPGs, (2) the trait combinations that defined each TPG and (3) the most important traits that distinguished TPGs.Our primary hypothesis would be supported if we observed similar numbers of TPGs, similar trait combinations characterizing TPGs and similar traits that were important in separating TPGs across regions.We used null models to test if observed differences in functional niche spaces and TPGs resulted from purely random processes.To test our second hypothesis that more climatically variable environments lead to more diverse trait profiles, we assessed if, within regions, the variability of trait profiles between TPGs was positively related to the number of climate zones (a coarse measure of climatic heterogeneity).
If data gaps occurred (e.g. for body form traits), we asked taxonomic experts to assign traits based on their best professional judgement.Different trait terminologies are used to this day in the ecological literature.We followed Schmera et al. (2015) in use of the terms grouping feature (i.e.general type of trait) and trait (i.e.trait state or category within a type of trait).The grouping features used in this study have been commonly applied in trait-based ecological studies (e.g.Crabot et al., 2021;Pilière et al. 2016) and describe different aspects of the biology of a species: life history (voltinism), morphology (body form, size), physiology (feeding type, respiration) and mobility (locomotion).Grouping features, traits, taxonomic resolution, coding of trait values and data completeness differed between the original trait databases and were harmonized prior to the analysis as described in Kunz et al. (2021).We harmonized the traits so that every trait dataset included the same 22 traits (Table 1) to ensure comparability across the trait datasets.We used either fuzzy coding or, when unavailable, binary coding (0 or 1) to represent the likelihood (affinities) that each taxon possessed each trait.Affinities from fuzzy-coded traits were normalized and converted to proportions (range between 0 and 1).Though the traits we used can be seen as different categories, the analysis was TA B L E 1 Grouping features and their traits used to compare the trait profiles across regions.and species within a family.This method has been shown to match expert assignments at the family level (Kunz et al., 2021).For a few families, aggregation with the median yielded affinities of zero for all traits within a grouping feature (one family in Australia, three families in North America).In these cases, we used mean values instead of medians.We restricted our analyses to the stream insect orders Coleoptera, Diptera, Ephemeroptera, Hemiptera, Odonata, Plecoptera and Trichoptera because too little trait data existed for the other orders (e.g.Planipennia, Megaloptera, Hymenoptera).

Voltinism
These orders also infrequently occur in stream ecosystems.The resulting trait datasets contained aggregated trait data for 107 families from Australia, 87 families from Europe, 90 families from North America, 67 families from New Zealand and 66 families from Southern Africa (Table S1).

| Functional niche space
We quantified functional niche spaces by creating an ordination of families in trait space.We first calculated a distance matrix (traits by family) derived from the combined trait datasets.We used Manly and Navarro's Overlap Index (Manly & Navarro Alberto, 2017), which is a distance metric that is suitable for proportional data that add up to one.When using this metric, a distance is first calculated for each grouping feature, which is then weighted across all grouping features.This means that each grouping feature was weighed equally, independent from the number of traits.We then applied a principal coordinate analysis (PCoA) to the distance matrix derived from the combined datasets.We computed separate convex hulls based on the first two PCoA axes for each region to visualize its trait structure and used the kernel density method to estimate the occurrence probabilities of each family given their trait profiles.To test for differences in trait structure among regions, we used permutational multivariate analysis of variance (PERMANOVA) (Anderson, 2001).Additionally, we compared the trait structure between major climatic regions (cold and temperate) in Europe.We found only minor differences between these two climatic regions, so we present methods and results for this analysis only in the Supporting Information (Text S1; Table S2).

| Null model functional space
To test the null hypothesis that the observed differences in niche space are purely random, that is would arise from any distribution of taxa across regions, we used a null model approach to compare the observed functional niche space overlap against we consider this as evidence against the null hypothesis that differences in functional niche space are purely random.

| Trait profile groups
To delineate TPGs, we calculated distance matrices based on the aforementioned overlap Index for each trait dataset separately.
We then applied hierarchical cluster analysis on the distance matrices (Ward's agglomeration method).To estimate the number of optimal groups, we used the 'gap' statistic (Tibshirani et al., 2001), which compares the within-cluster dissimilarities to expected within-cluster dissimilarities based on random reference datasets.
The reference datasets are drawn from a uniform distribution in the range of the original data.The optimal number of clusters is where the differences in within-cluster dissimilarities between original and reference data are maximal.We deemed a trait as characteristic for a TPG when more than 50% of the families within a TPG expressed the trait with an affinity >0.5 (we call these defining traits).Selecting at least 50% of families with an affinity >0.5 ensured that the majority of families expressed a particular defining trait.Setting the threshold for the affinity to a lower value, for example 0.3, would result in low discriminatory power (median number of defining traits seven to eight depending on the region), while selecting a higher value, for example 0.7, would result in TPGs characterized by only two traits.We used a null model to test if the observed TPGs are always similar in their defining traits irrespective of particular adaptations (see Text S2).To quantify the information lost through trait aggregation to family level (e.g. new defining trait combinations), we applied the previously described cluster analysis to the non-aggregated New Zealand trait dataset (n = 440), which contained complete information for our grouping features.In the non aggregated New Zealand trait dataset 86.8% of the taxa described were on the species level, 8.9% on the genus level and 4.3% on the family level.

| Most important traits for the separation of TPGs
We estimated how important each trait was to the separation of the stream insect families into TPGs with the permutation importance values derived after fitting random forest models in classification mode (Breiman, 2001).For these models, we used the derived TPGs as the response variable and the traits as predictor variables.The permutation importance measures the difference in the predictive performance of the random forest model for a particular predictor variable and its randomly permuted form.
The size of the difference indicates the association strength of this predictor variable with the response variable and thereby the permutation importance.However, correlated predictor variables can cause spurious importance rankings (Grömping, 2009).
We expected that many of the traits we used would be correlated because they often are jointly important ecologically (Poff et al., 2006) and because of the fuzzy coding employed, whereby affinities of traits belonging to a particular grouping feature and family sum up to one.Thus, to test the consistency of variable importance results, we used Boruta, a feature selection algorithm recommended for situations with correlated variables (Degenhardt et al., 2019;Kursa & Rudnicki, 2010).

TPGs in relation to climatic heterogeneity
We related the variation of trait profiles between and within TPGs with the number of Köppen-Geiger climate zones (Beck et al., 2018) within each region (Table S3).We used the number of climate zones as a proxy for the climatic heterogeneity of each region.First, we calculated distances between the mean trait profiles of the TPGs for each region (between TPG variation).We then calculated distances between the trait profiles of each family and the mean trait profile of their TPG (within TPG variation).All distances were calculated based on Manly and Navarro's Overlap Index.When trait profiles are completely different the distance score is 1, when trait profiles are identical the distance is 0. Finally, we regressed the between and within TPG variation with the number of climate zones per region.

| Functional niche space
The functional niche spaces of the different regions largely overlapped in the ordination space defined by the first two PCoA axes (Figure 1), although at least one region differed from one or more regions in the functional niche space occupied by taxa (PERMANOVA, p-value = .001,6.7% explained variance by region).The Australian and North American taxa exhibited slightly higher trait diversity than the European, New Zealand and Southern African taxa, which was indicated by slightly higher average distances to their group centroids (Table S4).The Australian taxa were slightly more distinct in their functional niche space as evidenced by non-overlapping families (black dots Figure 1) and the lower mean overlap with the other regions in terms of mean convex hull area shared (dotted line in Figure S1).The 11 Australian families contributing to this nonoverlapping region in functional niche space belonged to the orders Coleoptera, Hemiptera, Odonata and Plecoptera and tended to be bi/multivoltine, medium to large swimming predators with different body forms that use plastrons and spiracles for breathing (Figures S2   and S3; Table S5).
The null models showed that the mean overlap between the observed datasets was smaller than the mean overlap between the simulated null trait datasets, except for Southern Africa (Figure S1).The result for Southern Africa could be a consequence of the relatively small functional niche space calculated for Southern Africa that almost entirely overlapped with the other functional niche spaces.
Kernel density estimation indicated that the smallest region in the two dimensional functional niche space that contained 50% of all families occurred in Europe, North America, New Zealand and Southern Africa (48%-66% of their respective families), whereas only 37% of all Australian families occupied that part of functional niche space.Families in this core part of functional niche space mainly included small, cylindrical, univoltine, crawling taxa that used gills for breathing (Table S6; Figure S4).

| Similarity of TPGs between regions
The number of TPGs identified by hierarchical cluster analysis differed only slightly between regions.But the null models showed that TPGs were more structured by region than expected by chance (Figure S15).Seven TPGs occurred in New Zealand, eight in Australia and North America, 10 in Southern Africa and 12 in Europe (Table 2; Figures S5-S14).The number of optimal groups in cluster analysis is influenced by the number of objects, that is families in our case.
However, when we standardized cluster outcomes to 100 families, we obtained 10 TPGs for New Zealand, seven TPGs for Australia, eight TPGs for North America, 15 TPGs for Southern Africa and 14 TPGs for Europe.
The number of defining traits (methods) ranged between three to seven and was similar across regions.No identical TPGs were found across all regions, but some TPGs shared certain defining traits (Table 2).Twelve TPGs characterized by crawling, univoltine and cylindrical taxa occurred across all regions.This was the most frequently occurring trait combination in the delineated TPGs, but these traits also occurred along with different combinations of other traits in all regions, for example crawling, cylindrical and small size.Furthermore, eight TPGs were defined by predators breathing with plastrons and spiracles that also occurred across all regions.
Other traits frequently associated with the combination of predator and respiration with plastron and spiracle traits included small size, crawling or swimming mobility, or cylindrical body form.Trait combinations that occurred only in some regions often contained combinations of the above mentioned traits and other additional traits (e.g.herbivory with small size and crawling mobility, see Text S3).
No TPGs were characterized by the traits spherical body form and burrowing.
The analysis based on the non-aggregated New Zealand trait dataset (mainly on species level) identified 22 distinct TPGs indicating that taxonomic resolution affects the number of TPGs detected (Table S7).Many of these TPGs were similar to the aggregated dataset, but some new defining trait combinations also emerged.

| Taxonomic signal in the TPGs
Most TPGs contained families that were phylogenetically distant from one another, that is, families in the TPGs were often from different orders (Table 3).Only two TPGs were represented by families from a single order (Diptera or Ephemeroptera), and these TPGs contained only four to five families (NA_TPG8, NZ_TPG3).Odonata families, although abundant in the trait datasets, did often cluster together and were only found in one to four TPGs on every region.
A similar trend was observed for Plecoptera and Hemiptera, though families from these orders were less abundant in the trait datasets.In contrast, Coleoptera, Diptera and Trichoptera were distributed over many different TPGs both within and across regions (e.g.Diptera at least in five TPGs in every region).

| Most important traits for separating TPGs
The random forest analyses showed that the five most important traits separating TPGs differed across the trait datasets (Figure 2; Table S8).However, feeding mode and respiration traits were among the five most important traits for separating TPGs in every dataset.
Gill respiration was the only trait that was among the five most important traits in four trait dataset (except Europe).Body size traits were among the five most important traits in four trait datasets (except North America; Table S9).Furthermore, locomotion traits were among the five most important traits in Australia, Europe and New Zealand and body form traits in North America.The least important traits for separating TPGs varied between trait datasets (Table S9).
Consistently least important traits for TPG selection were spherical body form (all except Europe) and semivoltinism (North America, New Zealand and Southern Africa).Results based on the Boruta algorithm were largely consistent with these patterns but ranked trait importance slightly differently for some regions (Figures S16-S20; Text S4).

TPGs in relation to climatic heterogeneity
Overall, we found a weak positive, but statistically significant, relationship between TPG variation and the number of climate zones present in a given region (Figure 3).Mean trait profiles of the TPGs in New Zealand, the region with the lowest number of climate zones, were the most similar compared to the other regions.Conversely, the mean trait profiles of families in North America, which has the highest number of climate zones, were most distant from one another.The distances between mean trait profiles of Southern Africa TPGs were higher (Figure 3) than in the other regions and were an exception to the general trend despite the fact that Southern Africa has fewer climate zones than Australia and Europe.We only found a weak and statistically non-significant relationship between the within TPG variation and the number of climate zones (Figure S21).

| DISCUSS ION
We assessed if the trait profiles of stream insects are similar across large spatial scales (a potential indication for trait convergence) and if the variation in these trait profiles was higher in regions with more climate zones.The null models indicated that functional niche spaces and delineated TPGs are not the result of purely random processes.
Therefore, we established and compared functional niche spaces and delineated groups of families with similar trait profiles.Our results, while not entirely conclusive, generally support our hypothesis of convergence in stream insect trait profiles across major and biogeographically distinct regions.We observed (1) large overlap in the functional niche spaces occupied by families in the different regions, (2) similar numbers of TPGs across all regions and (3) that feeding mode and respiration traits were important in separating stream insect families into different TPGs in all regions.The partial separation of the Australian functional niche space from that of other regions also suggests that Australian stream insect faunas have diverged in trait profiles in response to continental-scale differences either in response to unique environmental selection pressures or as a consequence of biogeographic history.We also found weak evidence for our second hypothesis that higher variation in trait profiles would occur in regions with more diverse climates-that is more climatically diverse settings would support a broader range of ecological niches.

| The similarity of stream insect trait profiles across large scales: an indication of trait convergence?
Functional niche spaces were largely similar across the studied regions, except for Australia, whose stream insect fauna seems to partially deviate in trait structure from the other regions.No greater similarity appeared to exist between Australian, New Zealand, and Southern African trait compositions than between the other regions.This is despite the fact that these three regions have a common geological history (Gondwana origin).Moreover, Australia and New Zealand are geographically close and share many species, including stream invertebrates with genetic affiliations (Wallis & Trewick, 2009).Kernel density estimation confirmed that the functional niche spaces of stream insects in Europe, North America, New Zealand and Southern Africa were similar to one another.The partial separation of the Australian functional niche space from that of the other regions may be a result of the exceptionally high inter-annual rainfall variability that occurs across much of Australia, compared to the other regions.Some streams in North America and Southern Africa also experience highly variable flows (e.g.Merritt et al. 2021), but do not represent typical conditions for the region related to the considered species pool.The highly variable precipitation regime in Australia results in many river systems with non-perennial flow and limited freshwater refugia during dry seasons (Kennard et al., 2010;Puckridge et al., 1998).The 11 families occurring in the non-overlapping part of the Australian functional niche space possessed traits to cope with flow intermittency, such as swimming, bi/ multivoltinism and respiration with plastron and spiracle.Nine of these 11 families also occurred in other regions, but the Australian families had mostly higher affinities for these traits (Table S5).
The overall similarity in functional niche space across regions could indicate convergence of trait profiles, but similarities could also reflect conserved trait evolution, also termed niche conservatism (Blonder, 2018).However, the phylogenetic signal of the trait combinations making up the functional niche space was weak.
Most TPGs included families from multiple orders, suggesting that niche conservatism had a minor influence on TPG formation.The fact that the most important traits driving the partitioning into TPGs (feeding mode, body size and locomotion traits) are labile, that is they are relatively independent from other traits and unconstrained by phylogeny (see Poff et al., 2006;Wilkes et al., 2020) is also an indication of trait profile convergence.This is further reinforced by our finding that often the same traits from a grouping feature were important across multiple regions (e.g.predator in three regions).Labile traits are expected to respond to changing environmental conditions and convergent evolution should primarily act on these labile traits (Cavender-Bares et al., 2004).Nonlabile, that is evolutionarily constrained (Poff et al., 2006), traits such as respiration and body form traits were also important for TA B L E 2 Summary of the defining traits for each delineated TPG based on the criterion that at least 50% of the families within a TPG expressed a trait with an affinity >0.50.Note: Traits in light grey are those that did not quite meet our criterion displayed for non unique TPGs, for example in the case of NZ_TPG4 45% of the families had an affinity of >0.5 for crawling.Traits are ordered according to their occurrence in TPGs.
Abbreviations: bi/multivol, bi/multivoltinism; plas_spira, plastron spiracle; streaml, streamlined; semivol, semivoltinism; TPG, trait profile group; univol, univoltinism.with the theoretical expectation that the diversity of traits increases with land area (Smith et al., 2013).For example, the smallest region (New Zealand, 270,000 km 2 ) had the most similar TPGs, whereas the largest region (North America, 24,710,000 km 2 ) had the most different TPGs.The notable variability between Southern African TPGs may be an artefact of the relatively low number of described stream insect taxa to date (Odume et al., 2023).

Swimming
Studies defining stream insect TPGs on an intercontinental or continental scale have produced similar results to our study.Crabot et al. (2021) derived 15 TPGs in a study on functional responses to flow intermittence based on trait databases for Europe, North America and New Zealand using 12 grouping features including five of the grouping features we used (except body form).

F I G U R E 2
Permutation importance values of the random forest models.The five most important traits for the selection of trait profile groups are colored in purple for each region.Feed.mode, feeding mode.
Their most important grouping features were body size, feeding mode, voltinism, respiration and locomotion.Their results are consistent with our analysis, where feeding mode, respiration, body size and locomotion traits were among the most important traits in most regions.These traits might be important for the TPG classification because they are central for stream insects in coping with their environment, as individual traits but also in their mutual dependencies.Body size and feeding mode are both crucial traits in that they strongly determine the trophic niche and resource use of stream insects, whereas obtaining sufficient oxygen supply is crucial for aquatic organisms in general and may be the reason why this trait governed TPG classifications.Furthermore, body size is correlated with other important traits that could not be considered such as larval growth and fecundity.Climate seems to be a driver of body size variations, that is stream insects tend to be smaller in warmer environments (Peralta-Maraver & Rezende, 2021;Sweeney et al., 2018).TPGs for North America and Europe have been delineated in previous studies (Pilière et al., 2016;Poff et al., 2006;Usseglio-Polatera et al., 2000a).These studies identified some TPGs similar to the TPGs we found.For example, Poff

| Data limitations
We used stream insect traits aggregated to the family-level to deal with variability in (1) the taxonomic resolution across individual trait databases and (2) the number of taxa described in the multiple trait databases used (e.g.4067 taxa in Europe and 440 taxa in New Zealand).Through trait aggregation, we were able to study large-scale trait patterns and obtained a similar number of families across regions.Our use of the non-aggregated New Zealand trait dataset to evaluate how taxonomic resolution influences the results of cluster analyses showed that higher-resolution data produced a higher number of TPGs, but the TPGs had many similar defining trait combinations.This analysis also revealed a few new TPGs and defining traits, including traits that were rarely defining traits at family level (e.g.semivoltinism or sessile).Thus, the necessity of using family-level aggregated traits in our analysis may have obscured our ability to detect existing similarities and differences in TPGs between the regions.Using family-level traits in the study of trait convergence has another limitation.
Many stream insect families originate from ancient clades, that is they were already present in the late Jurassic era when the supercontinent Pangaea disintegrated (Wootton, 1988), though there are exceptions, such as some families of Plecoptera diverged later (Ding et al., 2019).While the most important traits driving TPG separation suggest trait convergence, we can therefore not completely rule out that the observed similarity in functional niche spaces based on family-level traits is, possibly only partially, a result of trait conservatism.After the breakup of Pangaea there was still considerable radiation and using a taxonomic resolution younger than the separation of the landmasses, such as genus or species level, would be ideal to test for convergence.However, the family level traits we used are mostly aggregated traits from higher taxonomic levels and thus account for the diversity within individual families.
With the help of taxonomic experts we were able to conduct analyses based on 60%-100% of the stream insect families listed in the different trait databases (Table S1).Network may in the future contribute to enhancing coverage of trait data (Gallagher et al., 2020).

| Potential adaptations to environmental conditions
Trait combinations that define groups of organisms can be interpreted as strategies of species to cope with their environment.In this context, different trait combinations can not only have the same function (Verberk et al., 2013), but also the same trait possessed by members of different TPGs may respond differently to environmental stressors, highlighting the dependency to other traits (Pilière et al., 2016).The trait combinations characterizing the delineated TPGs may be interpreted as a general adaptation to a wide range of environmental contexts.Across all regions, two combinations of defining traits characterized TPGs: (1) various combinations of crawling mobility, small size, predator and cylindrical body form and (2) respiration with plastron and spiracle and predator, with many of these TPGs also characterized by swimming mobility and small size.Crawling might be an adaptation to low flows and resistance to flow disturbances (Stanley et al., 1994), small size may reflect the ability to adapt quickly to variable environments, and cylindrical body form may reduce friction in slow or still waters (Lancaster, 2013).Increased predation has been linked to droughts and organic pollution (Mondy & Usseglio-Polatera, 2014;Statzner & Bêche, 2010), while plastron and spiracles provide adaptation to low-oxygen situations (Botwe et al., 2018).Moreover, swimming provides resistance to flow disturbance (Poff et al., 2018).
Therefore, this combination of defining traits may represent an adaptation to high temperatures and related disturbances, such as droughts and flow intermittence.

| Implications
Our study is the first that shows indications for stream insect trait profile convergence across Australia, Europe, New Zealand, North Furthermore, the TPGs we have delineated, and the approach how to establish such groups, can be used to identify stream insects that may be vulnerable to certain global change drivers.Knowing which trait combinations consistently respond to specific environmental filters will enable researchers and applied ecologists to derive general relationships on how stream insect assemblages respond to alterations of their habitats caused by natural or anthropogenic factors across multiple sites (Hamilton et al., 2020;Lamouroux et al., 2002).These relationships may be used as a diagnostic tool in freshwater restoration and conservation, for example to understand why certain stream insects, based on the TPG to which they belong, are disappearing from their habitats, and ultimately help to mitigate anthropogenic impacts on freshwater ecosystems.

ACK N O WLE D G E M ENTS
We thank Luic Damien for supporting the analysis of the comparison of trait composition between climatic regions in Europe and Verena Schreiner for discussing and reviewing the trait data.Funding was provided by the German Research Society (DFG, project number 338785727).Open Access funding enabled and organized by Projekt DEAL.

CO N FLI C T O F I NTER E S T S TATEM ENT
The authors declare that there are no conflicts of interest.

PE E R R E V I E W
The peer review history for this article is available at https:// www.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available at https:// anony mous.4open.scien ce/r/ Conve rgenc e-trait -profi les-7303/ .
proportions which describe the affinity a family has to express a certain trait.Hence, statistical methods were selected to handle the proportional data.To obtain trait data with comparable taxonomic resolution and sufficient coverage of trait information across the five regions, we aggregated the trait data to family level by calculating the median of the trait affinities for all genera random expectations.Poff et al. (2006) demonstrated that traits are strongly linked through evolution and phylogeny and should be treated as dependent.Therefore, to maintain the links between traits, we sampled 1000 times from the global family pool (n = 417, all families from all regions) instead of the trait pool, without replacement and randomly assigned families across regions matching the number of families in the observed trait datasets.For each sampling iteration, we used the related trait composition and calculated the functional niche space, convex hulls, the overlap and mean overlap in functional niche space following the same statistical methods we used for the observed trait datasets.Mean overlaps of the null trait datasets were then compared to the mean overlaps of the observed trait datasets.Finally, we derived a permutational p-value based on the number of mean overlaps of the null trait datasets that were smaller or equal to the mean overlaps of the observed trait datasets divided by the total number of sampling iterations (1000).If the mean overlaps of the observed trait datasets strongly deviate from the mean overlaps of the null trait datasets

F
Principal coordinate analysis ordination plot showing the functional niche space.The convex hulls mark the functional niche spaces covered by each region.The circles represent family locations on the first factorial plane.Black open dots mark the families of the Australian functional niche space that do not overlap with the other functional niche spaces.AUS, Australia; EUR, Europe; NA, North America; NZ, New Zealand; SAf, Southern Africa.
Percentages of families within each order present in the various TPGs.Cells that are empty indicate that the corresponding order was not represented in the respective TPG.Two example families from two different orders that mainly represented a TPG are given in column 'Example families'.If only taxa from one order represented a TPG, then two families from this order are given.Abbreviation: TPG, trait profile group.TA B L E 3 (Continued)TPG separation across multiple regions.However, the respiration traits that were consistently most important across regions (gillsandrespiration with plastron and spiracle) are clearly related to environmental conditions.They represent specialized techniques to take up oxygen in situations with low oxygen levels (e.g.high temperatures, stagnant conditions).While our findings suggest that trait profiles of stream insects may have converged, similarity alone is insufficient to demonstrate convergence.To conclusively demonstrate convergence would require replicated long-term (many generations) monitoring studies designed to assess the effect of different selective pressures on local adaptation.Alternatively, comparing the effects of different selective pressures on trait structure across different sites could also demonstrate convergence (Lamouroux et al., 2002).Our study had limitations in this regard since we were only able to assess the effect of one selective pressure (climate) on a coarse resolution, that is with information from trait databases and the Köppen-Geiger climate classification.Future research should aim to incorporate field data and assess other potential relevant selective pressures that influence the trait distribution of stream insect assemblages, such as geomorphological characteristics of streams.Despite the relatively coarse resolution of our data, we were able to demonstrate that greater variation exists between TPGs in regions with more climate zones.This finding suggests that environmental filters, particularly climate and related variables (e.g.hydrologic variables), play a critical role in shaping stream insect assemblages.We expected to observe greater niche breadths in regions with more climate zones because organisms have more environmental conditions to adapt to, leading to a wider range of trait values and an increased ability to occupy different parts of the functional niche space.To our knowledge, our study is the first to demonstrate a relationship between the trait profile variation and the heterogeneity of climate on such a large scale in stream insects.This result is consistent with previous studies that have shown the importance of climate and related variables as drivers of stream invertebrate trait composition on broader scales [e.g.Jourdan et al. (2018) for northern Europe, Poff et al. (2010) for western U.S.].Additionally, it aligns et al. (2006) identified a TPG characterized by large-sized crawling or swimming predators similar to our North American TPG3 and TPG4 and Usseglio-Polatera et al. (2000a) identified a TPG (their group f) similar to our European TPG8, consisting of medium-sized, univoltine, crawling shredders.
America and Southern Africa based on the trait profiles that describe life history, morphology, physiology and mobility.The overlap we observed in trait space and trait patterns across the different regions indicates that trait profile-environment relationships may be transferable across large scales.To examine possible universal environmental filters that shape trait compositions within stream insect assemblages, future research should examine the consistency of stream insect trait profile-environment relationships across different regions [see e.g.Bower and Winemiller (2019) orLamouroux et al. (2002) for examples for fish].Our results suggest that climate is such a filter for stream insects.
Nevertheless, for North ditions, such as traits on dispersal, resistance forms, or synchronization.Open science based initiatives such as the Open Trait