Does lake habitat alteration and land-use pressure homogenize European littoral macroinvertebrate communities?

Authors


Correspondence author. E-mail: mcgoffe@tcd.ie

Summary

  1. Beta diversity is the compositional heterogeneity of biotic assemblages among sites, and biotic homogenization is the decrease in beta diversity, facilitated by an increase in similarity of biotic communities over time. Environmental harshness decreases the importance of stochastic processes in structuring assemblages, resulting in a homogenization of the biota.
  2. We investigated if increasing nutrient enrichment, land-use pressure, and within-lake habitat alteration would decrease the beta diversity of macroinvertebrates in 46 lakes across Europe. Beta diversity was calculated using global multivariate dispersion. We utilized a structural equation modelling approach to account for hierarchical interdependence between potential impacts, that is the direct effects and correlations among the different impacts.
  3. We found clear indications that European macroinvertebrate communities are being homogenized by ongoing lake shore development. Increasing land-use pressure in the form of residential and commercial development had a direct negative effect on beta diversity (standardized coefficient = −0·40), as did roadways, albeit indirectly through an increase in engineering structures (standardized coefficient = −0·31). Increasing within-lake silt levels also homogenized macroinvertebrate communities (standardized coefficient = −0·18), independent of near shore land use. Our results indicate the negative effect of both the near shore land-use pressure and the within-lake habitat alteration on macroinvertebrate beta diversity, with significant interactions between these pressures.
  4. Habitat protection should take a more holistic approach to assessing lake development pressure, over a range of scales, as a solely site specific approach is not always biologically meaningful. Thus, future management plans should carefully control and mitigate ongoing development pressure if lake ecosystem health and resilience is to be maintained.
  5. Synthesis and applications. This study is the first of its kind to demonstrate European-wide homogenization of littoral macroinvertebrate lake communities with increasing habitat alteration and land-use pressure. Significant interactions occur between different habitat scales, with no one scale entirely accounting for the homogenization effect. To avoid further biotic homogenization, development pressure must be carefully managed at multiple scales, and where possible, minimized. This presents a challenge, as globally there is an increasing expansion of the human population and a consequent increase in anthropogenic pressure across all habitats.

Introduction

Beta diversity is the compositional heterogeneity of biotic assemblages among sites and along habitat gradients (Whittaker 1972; Loreau 2000; Donohue et al. 2009), and is an important measure for assessing ecosystem health (Passy & Blanchet 2007). Habitat heterogeneity is a major determinant of species diversity at both local and landscape levels (Poggio, Chaneton & Ghersa 2010), and physical habitat alteration is one of the biggest anthropogenic threats to aquatic ecosystems (Paulsen et al. 1997). Biotic homogenization is the reduction in beta diversity, as biotic communities increase in taxonomic, functional or genetic similarity over time. It is caused by both local extinctions of endemic species and introductions of non-native species (McKinney & Lockwood 1999; Olden, Poff & Bestgen 2006). Homogenization of biota is considered one of the most widespread forms of biotic impoverishment world-wide (Olden et al. 2004), and it is generally facilitated by humans (McKinney & Lockwood 1999).

Nutrient enrichment alters the ecological functioning and macroinvertebrate community structure of lakes (Brauns et al. 2007a), and reduces beta diversity of littoral macroinvertebrates (Donohue et al. 2009). Lake nutrient enrichment, habitat modification and catchment land-use development are often closely coupled. Agriculture directly affects macroinvertebrate communities through anthropogenic eutrophication (Brodersen, Dall & Lindegaard 1998) and indirectly through lake shore alteration (Brauns et al. 2007b) or reductions in littoral structural complexity (Scheffer et al. 1993; Egertson, Kopaska & Downing 2004; Donohue et al. 2009). Although, many studies have focused on the impacts of eutrophication, there has been less work on hydromorphological alteration effects. Residential development is a major increasing pressure on lakeshores world-wide (Peterson et al. 2003; Gonzalez-Abraham et al. 2007), but the consequences of this land-use change on lake habitat structure have received little or no attention. Specific shoreline habitat features are vital for the survival of macroinvertebrates, periphyton, birds and fish (Harrison & Hildrew 2001; Sass et al. 2006; Brauns et al. 2007b; Carey et al. 2010), and lakes with poor surrounding vegetation are more likely to have poor biological conditions [U.S. Environmental Protection Agency (USEPA) 2009]. However, despite this, the responses of lentic biota to habitat structure changes are poorly understood (Jennings et al. 2003), and lake habitat responds to changes occurring within the entire watershed (Jennings et al. 2003).

That taxonomic communities undergo homogenization with increasing impact has been well documented (Rahel 2000; Rooney et al. 2004; Donohue et al. 2009). Several authors have demonstrated how the alteration of specific habitats reduces species beta diversity and species richness (Bryan & Scarnecchia 1992; Passy & Blanchet 2007), and others examined lake eutrophication effects (Donohue et al. 2009). However, none have examined the effect that near shore land-use pressure has on littoral macroinvertebrate beta diversity.

Thus, we hypothesized that with increasing nutrient enrichment, near shore land-use pressure and within-lake habitat alteration, there would be a corresponding decrease in the beta diversity of lake macroinvertebrates. This reduction is expected to occur both through the direct effects of the nutrient enrichment, land-use pressure and littoral habitat alteration on the beta diversity, and through the indirect effects mediated through within-lake habitat alteration. We used a structural equation modelling (SEM) approach to model this, and to disentangle nutrient enrichment from habitat alteration and land-use pressure. Specifically, our first aim was to investigate if nutrient effects on beta diversity were correlated with lake siltation. Our second aim was to differentiate the direct effect of near shore roads and to development land use on the beta diversity of macroinvertebrates. Thirdly, we determined if near shore land-use pressure is more important than within-lake habitat alteration for determining the beta diversity, and whether these two habitat zones are strongly coupled. This was achieved using a SEM approach on habitat and macroinvertebrate data from 46 European lakes.

Materials and methods

Study Sites

Macroinvertebrates were sampled from 46 lakes across seven European countries along a north–south and east–west gradient, in Sweden, Denmark, Germany, Italy, Ireland and the UK (Fig. 1) (see Table S1, Supporting information). Numbers of animals per order per lake are outlined in Table S2 (Supporting information).

Figure 1.

Map of study lake locations, where points represent a study lake.

Environmental Gradients, Invertebrate Sampling and Biotic Data Analysis

Within each lake, three different shoreline types were chosen a priori for macroinvertebrate sampling to capture the full range of macroinvertebrate variability. These biotic communities were expected to become more similar to each other with increasing lake wide pressure. The shoreline types were as follows: (i) hard embankment, termed hard sites, (ii) recreational beaches or improved grassland, termed soft sites and (iii) natural sites with no alteration, termed unmodified. Unmodified sites were characterized by woody debris, abundant emergent macrophytes and underwater tree roots in the littoral zone. The hard sites were generally modified by hard bank shore engineering, including concrete in the shore and the littoral zone, riprap and shore zone embankment. The riparian zone was usually made up of urban or suburban land use, with some roads or railways. Soft sites comprised sites with sand in the riparian and the littoral zone, with a gently sloped shore line. There was usually a beach and some recreational activities present.

Within each of the lakes the different shoreline types were replicated three times, so from each lake nine different sites were chosen: three hard sites, three soft sites and three unmodified sites. Macroinvertebrate samples were taken from each of the nine sites in all lakes, in late summer–autumn of 2009. A 1-min kick sample was taken at each site in depths of <0·75 m, with the sample effort divided proportionally among the habitats present at the site, according to their relative areal proportion. The sediment was disturbed using the feet while constantly moving backwards to create a current, and continuously sweeping the net from side to side through the water above the sediment to catch disturbed macroinvertebrates. A 0·5-mm mesh net was used (European standard EN-27 828), and each of the samples was immediately preserved in ethanol. All individuals sampled were identified to the lowest feasible level, generally species, except for Diptera and Coleoptera larvae which were identified to family level, and Trichoptera and Odonata larvae which were identified to genus level.

Global multivariate dispersion (Clarke & Warwick 2001) was used as a measure of multivariate compositional heterogeneity (Anderson 2006), termed beta diversity in this article. This is a multivariate measure of variability in groups of samples, and it is computed by comparing ranked similarity measures within and among groups in a similarity matrix. It is not confounded by measures of alpha and gamma diversity (Lande 1996). This method was chosen as traditional measures of tests of homogeneity of variance are based on Euclidean or Mahalanobis distance, which may not always be appropriate for biological data. Multivariate dispersion is not sensitive to departures from non-normality and provides a robust test for homogeneity of multivariate dispersion (Anderson 2006). A single multivariate dispersion value was calculated for each lake based on the Bray–Curtis similarity matrices from proportional abundances, with the dispersion calculated based on distance between the nine samples within each lake. These analyses were performed in primer (Clarke & Warwick 2001).

Habitat Surveying, Habitat Alteration and Land-Use Pressure

The lake habitat was surveyed using the Lake Habitat Survey (LHS) protocol. This method surveys the lake habitat at 10 locations (habplots) around the lake perimeter; and the macroscale riparian and littoral habitats are described for the lake as a whole (Rowan et al. 2006). The first habplot is chosen randomly and the remaining sites are spaced evenly around the perimeter from the first. The habplots are 15 m wide and extend 15 m landwards from the edge of the bank, to 10 m offshore. The land use in the lake perimeter is also recorded up to 50 m inland from the waterline, termed near shore land use for the remainder of this article. This land use is surveyed by boat, the perimeter is divided into sections, and within each section the percentage of land use given over to certain uses is estimated. Several different categories of land use are listed in the LHS method, e.g. residential areas, parks and gardens, quarrying and mining etc. For full details see Rowan et al. (2008).

Riparian and littoral habitats were assessed for each lake. The littoral habitat is referred to as ‘within-lake habitat alteration’ for this article. Two variables which represent this are as follows: (i) the average percentage of silt and/or clay cover in each habplot, and (ii) the average occurrence of hard and soft bank engineering across habplots. In this study, no distinction was made between silt and clay, so all particles <0·063 are included in the classification (Rowan et al. 2008), termed ‘silt’ for this article. Hard bank engineering incorporates consolidated construction materials such as concrete and steel sheet piling, and features associated with docks and marinas. Soft bank engineering includes shoreline stabilization materials such as basket-work, dumped natural debris and soft synthetic materials (Rowan et al. 2008). Our study grouped hard and soft engineering structures, although the frequency of soft engineering structures over all the lakes was very low. These were collectively termed ‘engineering’ for this article.

Two measures were chosen to best represent near shore land use: land impacted by development and land impacted by rail and roadways. Developed land constitutes commercial and residential land, defined as hotels, car parks, factories or farming construction, and houses and schools, termed ‘developed land’. Roadways simply accounted for the amount of shoreline affected by roads and railways, termed ‘roadways’.

SEM and Model Construction

We utilized a Structural Equation Model (SEM) approach to account for hierarchical interdependence between our different scales (Parsons, Thoms & Norris 2003) of near shore land use and within-lake habitat alteration, and to more explicitly link process and spatial pattern (Grace et al. 2010).

Structural Equation Model is closely related to regression analysis, but incorporates a priori scientific knowledge when the model is developed, serving as a multivariate hypothesis test (McCune & Grace 2002). The advantage of SEM is that it allows the overall model to be tested, along with significance tests for the individual parameters of the model (McCune & Grace 2002). In this study we used SEM for confirmatory analysis, where the model-implied covariance matrix is tested against the sampled covariance matrix from the actual data, to see if the data support it. However, although a model fit adds strength to a multivariate hypothesis, it must be noted that the strength of SEM lies in the a priori model, which is based purely on theoretical knowledge (Grace & Bollen 2005). SEM analyses were carried out in R program (version 2.14.1) (R Development Core Team 2011), using the lavaan package (Rosseel 2012), and AICcmodavg package (Mazerolle 2012).

Chi-square maximum likelihood tests evaluated model fit. When the fit is ≥0·05, the difference between the observed and model-implied covariance matrices is not significant, and the model fit can be deemed acceptable (Grace 2006). The individual relationships between variables were examined using P-values derived from attending standard errors (Grace 2006). The strength of these relationships was assessed by the estimated coefficient, and significance tested using z-statistics. Owing to small sample size, paths with a P-value of <0·1 were considered significant, following Lamb, Kembel & Cahill (2009). Even those pathways which are deemed non-significant remain in the model, as they are part of the model structure.

The overall model fit was also assessed using root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR) statistics, to ensure the model was not overfitted. RMSEA analysis is a sample-size-adjusted measure of fit. It has an associated P-value that is interpreted in the same way as the Chi-square. For RMSEA to indicate a non-significant difference between model and data the minimum RMSEA should include 0 within its range (Grace 2006). The SRMR is defined as the standardized difference between the observed and predicted correlation, and it is an absolute measure of fit. A value of <0·08 is generally considered a good fit (Hu & Bentler 1998). Within our model our ratio of samples to parameters is 9, which falls above the minimum number of 5, and closer to 10, the number generally recommended in the literature (Grace 2006), thus our sample size can be deemed adequate.

Both unstandardized and standardized coefficients are presented, but this study focused just on the standardized coefficients as they allow direct comparisons between the different pathways. The standardized coefficients indicate the strength and direction of the relationship, and these values are valid only for this data set. The unstandardized coefficients should be used for comparison across models (Grace 2006). Significance tests are based on the unstandardized coefficients.

Data Preparation, Transformations and Spatial Autocorrelation

Data were first assessed for distributional characteristics, as SEM, similar to regression, is sensitive to nonlinear relationships. The bivariate relationships between all variables were first visualized to test for linearity, and outliers. In SEM, there is no requirement for predictor variables to be parametric (Grace 2006). Two of the dependent variables were non-normal, thus the robust Satorra Bentler (Satorra & Bentler 2001) method was applied to adjust for deviation from normality.

As the sites appeared geographically grouped (Fig. 1), spatial autocorrelation was tested to ensure the assumption of independence of observations was not violated (Legendre 1993). The standard method for analysing spatial patterns in this type of data is the Mantel test (Mantel 1967; Koenig 1999). This tests for the overall relationship between distance and similarity between sites, with significance tests obtained by randomly permuting the data and recalculating the test statistic. For our data two matrices were generated, one was a distance matrix, consisting of the Euclidean distance between all pairs of sites, based on latitude and longitude, and the second was a correlation matrix, consisting of the similarity between dispersion values across all pairs of sites. The Mantel test was carried out in R program (version 2.14.1) (R Development Core Team 2011), using the vegan package (Oksanen et al. 2008), with 9999 permutations. No correlation was found between the two matrices (r = −0·01, = 0·57), indicating a non-significant autocorrelation of proportional dispersion values against distances.

Model Construct

Our theoretical model was based on the hypothesis that with increasing nutrient enrichment, land-use pressure and within-lake habitat alteration there would be a corresponding decrease in the beta diversity of lake macroinvertebrates. We hypothesize that this reduction is mediated both through the within-lake habitat features, and with direct effects from the near shore land use and nutrient enrichment variables.

The theoretical model was then converted into a testable SEM model, known as model specification (Grace et al. 2010) (Table 1) (Fig. 2). In our case, the only nutrient measure available for all lakes was total phosphorus (TP), collected by one off sample from each lake and analysed following the laboratory method outlined in Eisenreich, Bannerman & Armstrong (1975). This nutrient sampling varied in different countries, but at the minimum the TP value was based on a single value collected as close to the macroinvertebrate sampling date as possible. Nutrient enrichment has previously been shown to negatively affect lake macroinvertebrate beta diversity (Donohue et al. 2009) likely mediated through the coating of substrates with silt. Thus, the first within-lake habitat alteration measure, the average percentage abundance of silt in each habitat, was calculated across habplots for each lake. As TP increased, the levels of silt were also expected to increase (Richards et al. 1997), decreasing the beta diversity of macroinvertebrates (Bass 1992). A correlation (double headed arrow) was drawn between the silt and TP variables.

Table 1. Description of variables used in the structural equation modelling model
VariablesVariable nameCalculatedTermedRangeMean
Nutrient enrichmentTotal phosphorusμg L−1TP9–197·743·7
Lake wide land use(1) Roads and railwaysPercentage of the 50 m riparian band around the lake comprising roads and railwaysRoadways0–65·212·3
(2) Human and residential land usePercentage of the 50 m riparian band around the lake comprising human and residential land useDeveloped land0·15–46·612·9
Within-lake habitat alteration(1) SiltAverage % of silt over 10 habplots in a lakeSilt0·6–4·62·9
(2) EngineeringAverage occurrence of hard and soft engineering structures over 10 habplots in a lakeEngineering0–1·20·5
Independent factorsLake latitudeLatitude of each lakeLatitude41·7–60·852·1
Lake surface areaSurface area of each lake (km2)Lake area0·3–114·513·7
Response variableMacroinvertebrate beta diversityMultivariate dispersion among nine samples from a lakeBeta diversity21·5–50·935·9
Figure 2.

Specified a priori model showing path directions, with causal arrows in solid lines, and correlations with dotted lines. Thicker solid lines indicate a positive relationship is hypothesized, regular solid lines indicate a negative relationship is hypothesized. All variables are observed variables.

In habitats which have been anthropogenically altered, heterogeneity will often also decrease, impacting macroinvertebrate beta diversity. For this reason we hypothesized that as our second within-lake measure, the amount of hard and soft bank engineering, increased, the beta diversity of macroinvertebrates would decrease, and a directed arrow was drawn from the engineering variable to the beta diversity variable.

Near shore land-use variables were also assessed for their impact on beta diversity of macroinvertebrates. Maloney, Munguia & Mitchell (2011) found that macroinvertebrate alpha diversity decreased with the percentage of impervious land cover adjacent streams in the US, thus roadways were chosen as a measure of impervious land-use pressure for our study. We hypothesized that these would have a direct negative effect on beta diversity, and an indirect effect mediated through an increase in engineering structures related to the road building, and through an increase in silt, as the building of roadways beside water bodies increases silt levels (Taylor & Roff 1986). Both residential and commercial land use (developed land), were also assessed as our second measure of near shore land use, with a directed arrow from the developed land variable directly to beta diversity, along with an indirect arrow mediated through both silt and engineering variables.

A correlation arrow was also drawn between silt and engineering, as it was expected that as one increased the other may also increase. Two independent variables were also added to the model, which could be expected to influence the beta diversity of macroinvertebrates. Lake area was given a directed arrow to beta diversity, and a positive relationship was expected; lake latitude was also given a directed arrow, and it was expected to negatively affect richness (Vinson & Hawkins 2003; Soininen, Lennon & Hillebrand 2007). All exogenous variables were allowed to freely correlate, and a positive correlation was expected between TP, roadways and developed land use.

Results

SEM Confirmatory Results

The model fit the data well (= 0·473, X2 = 4·55, with 5 d.f.), with no significant difference between model-implied and observed covariance matrices. The RMSEA value also indicated a good fit (RMSEA = 0·0, with 90% confidence interval = 0·0–0·2), as did the SRMR analysis (SRMR = 0·04). The chosen model explained 37% of the variation in the beta diversity (adjusted R2 = 0·37) (Table 2, Fig. 3).

Table 2. Standardized and unstandardized coefficients for directed and correlated pathways
 Unstandardized path coefficientStandard errorP-valueStandardized coefficients
  1. Positive values indicate a positive effect; negative values indicate a negative effect. Significant results indicated with ‘*’ and bold P-values.

Directed paths
Macroinvertebrate beta diversity
Lake area*0·1160·026 0·001 0·440
Latitude−0·0580·1270·646−0·056
TP−0·3382·3950·888−0·020
Developed land*−0·2120·070 0·002 −0·402
Roadways0·0150·0570·7940·031
Engineering*−9·7663·961 0·014 −0·314
Silt*−0·1420·082 0·085 −0·180
Engineering
Developed land−0·0010·0020·426−0·083
Roadways*0·0060·002 0·003 0·412
Silt
Developed land−0·0780·1020·442−0·117
Roadways−0·0190·0930·837−0·032
Correlations
Silt
TP0·3460·3550·3300·118
Engineering
Silt0·0800·1850·6670·056
TP
Developed land0·8210·6530·2090·187
Roadways*−1·4020·719 0·051 −0·287
Developed land
Roadways*29·47715·410 0·056 0·191
Figure 3.

Structural model for hierarchical effects of anthropogenic pressures on macroinvertebrate taxonomic beta diversity. Numbers next to single headed arrows are standardized path coefficients, and beside double headed arrows are partial correlation coefficients. R2 values are given in bold above the three endogenous variables. Non-significant pathways are indicated with a grey line, significant in black, correlations are indicated with a dotted line. X2 = 4·55, P = 0·473 with 5 d.f.

Lake area had the strongest effect on beta diversity, with a standardized path coefficient of 0·44, indicating increasing beta diversity with lake size (Fig. 3). Latitude and TP effects were not statistically significant. Although hypothesized, TP and silt showed no correlation (Table 2).

The amount of roadways increased with the amount of engineering (path weight = 0·41), and engineering had a negative relationship with beta diversity (path weight = −0·31) thus roadways significantly decreased the beta diversity of macroinvertebrates via engineering (total and indirect effects are listed in Table S3, Supporting information). Developed land had a negative relationship (path weight = −0·40) with beta diversity, but had no indirect effects through our within-lake variables (i.e. silt or engineering). To further elucidate this, developed land was correlated with both the percentage of woody debris in the lake (averaged over 10 habplots in a lake) and with the macrophyte percentage volume inhabited (PVI) (averaged over 10 habplots), as both are affected by shoreline development, in turn affecting biota (Bryan & Scarnecchia 1992; Christensen et al. 1996). Spearman rank correlations found no significant relationship between developed land and either woody debris, or macrophyte PVI (r = −0·04, = 0·78, n = 46; r = 0·16, = 0·30, n = 46 respectively).

Discussion

The implications of homogenization are far reaching (Collins, Vazquez & Sanders 2002), with species contributing both individually and collectively to the stability and functioning of ecosystems (Olden et al. 2004). This study provides a clear indication that European macroinvertebrate communities are being homogenized by ongoing development on and around lakeshores. Habitat alteration is possibly the greatest anthropogenic threat to lake ecosystems (Paulsen et al. 1997), and the ecological consequences of biotic homogenization are largely unexplored (Olden et al. 2004). Macroinvertebrates play a key role in food webs and nutrient cycling of lakes (Schindler & Scheuerell 2002; Brauns et al. 2011; Silva & Moulton 2011), and habitat urbanization has been shown to shorten riverine food webs (Eitzmann & Paukert 2010). However, although habitat effects were elucidated, nutrient enrichment was not found to homogenize macroinvertebrate communities for our lakes, contradictory to findings for Irish lakes (Donohue et al. 2009). Donohue et al. (2009) used additional nutrient measures in their study (TP, total nitrogen and chlorophyll a), possibly providing a more robust measure of nutrient enrichment. It must, however, be noted that structural habitat alteration and water quality changes occur concomitantly, confounding attempts to measure the effect of just one factor (Jennings et al. 1999). SEM is a useful tool for disentangling this, and our results indicate that habitat alteration plays a greater role in biotic homogenization than nutrient enrichment at a European scale.

Defining the appropriate spatial scale to measure the effects of habitat modification is essential for habitat protection (Jennings et al. 1999). Our results indicate the importance of both the near shore land use and the within-lake habitat alteration for beta diversity, and the interactions between these. Roadways had a negative impact on beta diversity, via engineering. As the percentage of roadways around the lake increased, so did the frequency of engineering structures within the lake. Engineering structures have been shown to decrease macroinvertebrate species richness by reducing habitat heterogeneity and available surface area (Brauns et al. 2007b). Our study concurs with these findings, albeit expanding them to a larger lake wide scale. However, this study did not consider different engineering types, and the impact of site level alterations can be somewhat mitigated by providing more heterogeneous habitat options such as niche rich riprap (Jennings et al. 1999; Brauns et al. 2007b). There is a widespread view that small habitat alterations have a negligible effect, but as highlighted by Jennings et al. (1999) most lakes do not undergo one single dramatic habitat alteration. They undergo a gradual shift in land use, with assemblage structure responding to cumulative habitat alteration. In this study, macroinvertebrate beta diversity responded on both a local scale, decreasing with the cumulative effect of engineering structures and siltation within a lake, and at a landscape scale, decreasing with an increased developed land-use pressure.

No relationship was found between roadways or developed land and silt, although, roadways have been shown to increase the amount of silt (Extence 1978), and residential pressure to increase fine sediments in lakes (Jennings et al. 2003). However, the majority of sediment run-off occurs during the construction phase (Garrison & Wakeman 2000). Although unrelated to landscape level variables, silt did negatively affect beta diversity. Siltation decreases attributes of habitat heterogeneity, such as macrophyte species richness (Scheffer et al. 1993; Egertson, Kopaska & Downing 2004), and several authors have highlighted the direct impacts of silt on macroinvertebrates (Doeg & Koehn 1994; Wood & Armitage 1997). The mechanism is likely natural substrate homogenization resulting in less colonization area for macroinvertebrates specialized to specific substrate characteristics (Voelz & McArthur 2000; Brunke, Hoffmann & Pusch 2002).

Although, roadways were not directly affecting the beta diversity of macroinvertebrates, developed land use was. While not explicitly measured for this study, developed land could be providing a more accurate indication of impervious land cover than roadways. Impervious structures have been shown to decrease macroinvertebrate alpha diversity (Maloney, Munguia & Mitchell 2011), and to cause a shift towards more tolerant species (Stanfield & Kilgour 2006). In addition, developed land can decrease both the volume of woody debris in lakes (Christensen et al. 1996), and lake macrophyte abundance (Bryan & Scarnecchia 1992). However, no such relationship was found in our study lakes. Although, the exact mechanism for the negative relationship between developed land use and beta diversity remains unclear, and would require additional study, previous work has shown increased temperature spikes with increasing impervious land (Nelson & Palmer 2007), and increases in conductivity, sulphate, chloride, pesticides and polycyclic aromatic hydrocarbons (PAHs) in urbanized streams (Cuffney et al. 2010).

Lake area increased beta diversity, agreeing with previous lake studies (Dodson 1992; Heino 2000; Hoffmann & Dodson 2005; Hrabik et al. 2005; Sondergaard, Jeppesen & Jensen 2005), attributable to higher variability in available habitats (Hoffmann & Dodson 2005). However, Donohue et al. (2009) found the inverse relationship between macroinvertebrate beta diversity and lake size, postulating that greater within-lake connectivity homogenized the community. This discrepancy could be attributable to our SEM model including more landscape-level variables, thereby revealing additional effects of factors such as developed land on beta diversity.

The original concept of biodiversity is one based on taxonomy (Terlizzi et al. 2009). The impacts of differing taxonomic resolutions in our study were not accounted for, and may be hypothesized to alter the diversity patterns found. However, previous work examining beta diversity (multivariate dispersion) in a marine setting found that genus and family level heterogeneity followed species level heterogeneity (Terlizzi et al. 2009). Although not specific to lake macroinvertebrates, this indicates that coarser taxonomic resolution may not impact on the overall beta diversity patterns.

This study is the first of its kind to demonstrate European-wide homogenization of littoral macroinvertebrate lake communities with increasing anthropogenic habitat alteration and pressure. Our study indicates that neither the near shore land use, nor within-lake habitat alteration can solely explain the reduction in macroinvertebrate beta diversity, with significant interactions between the two. If further biotic homogenization is to be avoided, habitat protection should take a more holistic approach, over a range of scales, as a solely site specific approach is not always biologically meaningful (Jennings et al. 1999). Future management plans should carefully control and mitigate ongoing development pressure if lake health and resilience is to be prioritized. This type of research and the resulting adaptive management policies must be given priority, if global ecosystem health and diversity is to be maintained (Olden et al. 2004).

Acknowledgements

We thank those involved with the field sampling and processing of the samples collected in this survey. This study was financially supported by WISER Water bodies in Europe: Integrative Systems to assess Ecological status and Recovery, funded by the European Union under the 7th Framework Programme, Theme 6 (Environment including Climate Change) (contract No. 226273). Leonard Sandin was funded by the Marie Curie Actions of the European Commission (FP7-2010-PEOPLE-IEF) through the FRESHCLIM project (project 273215). We thank anonymous reviewers for constructive criticism and suggestions on the manuscript. Statistics advice from Dr. James Grace and Dr. Jarrett Byrnes proved invaluable for this work.

Ancillary