SEARCH

SEARCH BY CITATION

Keywords:

  • Body size;
  • EIE;
  • everything is everywhere;
  • geographic variation;
  • marine eukaryotes;
  • meta-analysis

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

Aim  To analyse how important body size is for geographic variation in microscopic organisms in the marine environment and thereby determine the validity of the ‘everything is everywhere’ hypothesis.

Location  Marine environments, globally.

Methods  Studies on geographic variation in all marine eukaryotes smaller than 1 mm and all marine copepods were compiled from the literature and incorporated in multiple binomial regressions to analyse the effect of body size, lifestyle, environmental isolation and taxonomic affinity on the probability of different regions being reciprocally monophyletic. Sample size was also analysed, because a negative relationship between sample size and probability of being reciprocally monophyletic would indicate biases due to undersampling in the original studies. Separate analyses were performed for three potential types of barriers to gene flow for marine organisms (the water barrier of the middle of the oceans, the temperature barrier of the equator, and the land barrier of the continents).

Results  Environmental isolation was the only variable in the best-fitting model for the probability of populations from each side of an ocean being reciprocally monophyletic. The estimated importance of body size was quite large and the lack of this variable in the best-fitting model may be a power issue due to the small sample size (n= 40). Both environmental isolation and body size were important for the probability of monophyletic populations in different hemispheres. The analysis of isolation between different oceans did not produce clear results. The relationship between sample intensity and probability of being reciprocally monophyletic was positive for the isolation between different oceans and non-existent for the two other analyses.

Main conclusions  The results showed that body size was an important factor governing the potential for geographic variation but not the only important factor by far. These results clearly weaken the ‘everything is everywhere’ hypothesis as they showed that microscopic organisms may have geographic variation, albeit to a lesser degree than larger organisms.


INTRODUCTION

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

The existence of geographic variation in microscopic eukaryotes has been heavily debated in the recent literature. There are two main competing schools. Followers of the ‘everything is everywhere’ model (EIE) state that all free-living organisms smaller than about 1–2 mm are cosmopolitan (Finlay, 2002; Fenchel & Finlay, 2004; Fenchel, 2005), whereas opponents assert that endemism occurs in microscopic organisms although probably to a lesser degree than in larger organisms (Foissner, 1999, 2006, 2008).

Fenchel & Finlay (2004) report that a much higher proportion of all known microorganisms than macroorganisms can be found in a single locality. They argue that the difference can be explained by a lack of geographic isolation in microorganisms, caused by their large population sizes and presumably large dispersal capacities. However, their analysis requires the same species concept for micro- and macroscopic species, a requirement that in our view is not fulfilled. Fenchel & Finlay (2006) advocate an ecological morphospecies species concept that requires morphological differences of likely ecological importance between different species, whereas most taxonomists of macroscopic animals use a biological species concept with many allopatric species only identifiable on small, probably neutral, differences, e.g. minor differences in genital morphology.

Opponents of EIE argue that this apparent cosmopolitan distribution is an artefact largely caused by taxonomic lumping and misidentifications. They contend that conclusions can only be drawn for a few characteristic ‘flagship species’ which show geographic variation (Foissner, 2006). On the other hand, followers of EIE consider patterns recovered from these ‘flagship species’ to be artefacts caused by geographic undersampling and taxonomic over-splitting (Esteban et al., 2001; Finlay et al., 2004; Foissner, 2006; Smith & Wilkinson, 2007).

We suggest that the degree of geographical variation in microscopic organisms can be more satisfactorily analysed using molecular approaches. The ecological characteristics of most microscopic organisms are insufficiently known, making it difficult to determine whether the distributional limits of a species are caused by barriers to dispersal or lack of suitable habitat. Fenchel & Finlay (2006) argue that microorganisms with identical morphology are ecologically identical. If so, genetic studies of the deep lineages often found within morphospecies of microscopic organisms would be ideal for understanding species distributions, since geographic variation could only result from dispersal barriers. We disagree, however, with the opinion that morphologically identical populations will also be ecologically identical, since not all adaptations will necessarily induce changes in morphology and comparisons should therefore be restricted to similar habitats.

The main theoretical argument for the EIE model is the frequent recovery of an inverse correlation between body size and census population size (Fenchel & Finlay, 2006). This is assumed to correspond to a similar relationship between body size and effective population size (Ne), defined as the size of the corresponding ideal population (a population with Poisson distributed variation in reproductive output and without fluctuations in population size), resulting in very large effective population sizes for microscopic organisms. Such a large Ne would probably remove all geographic variation, but simulations indicate that under some conditions the correlations between census and effective population size may only be weakly positive or even negative (Pertoldi et al., 2007). Nucleotide diversity seems to be comparable in micro- and macroscopic organisms, which could also indicate that effective population sizes may be similar if similar mutation rates are assumed. Hitchhiking selection may, however, greatly reduce nucleotide diversity in large populations, especially if they have a large frequency of asexual reproduction (Gillespie, 2000; Bazin et al., 2005). Comparable nucleotide diversity could therefore be found in micro- and macroscopic organisms even if their population sizes are greatly different.

The applicability of the EIE model is, however, fairly easy to test, as frequent occurrences of reciprocal monophyly (i.e. all individuals from a larger region constituting a monophyletic clade) in different regions of microscopic organisms would be strong evidence against the model. Although persistent founder effects can lead to reciprocal monophyly on a small scale, such as within a pond, they will not lead to reciprocal monophyly of larger areas under normal circumstances (de Meester et al., 2002; Gómez et al., 2007); persistent founder effects can therefore be ignored in analyses of variation between regions.

Here we carried out a meta-analysis of all studies analysing the degree of geographic variation in microscopic organisms and investigated the importance of body size relative to other factors for the probability of recovering different regions as reciprocally monophyletic. We restricted our analysis to marine organisms for several reasons: (1) the potential for allopatric speciation is often assumed to be lower in marine environments because of greater connectivity among areas (Dawson & Hamner, 2008; but see Knowlton, 1993, 2000); (2) a finding of substantial geographic variation in marine environments would therefore provide stronger evidence against EIE than a similar finding in other habitat types; and (3) identification of specific isolating barriers, and estimation of the strength of these barriers, is easier in marine environments.

There are four main factors that could potentially contribute to geographic variation; three are dispersal barriers and the fourth is a consequence of undersampling. Advocates of EIE generally assume that insufficient sampling of localities or individuals per locality could explain the absence of a species from a suitable area. Undersampling may be common but is easily tested for, since the risk for undersampling is decreased by sampling more individuals. If undersampling is indeed a large problem, geographic variation would consistently be less common with larger sampling sizes, as increased sampling would increase the likelihood of sampling all haplotypes present in a population, removing false ‘endemic lineages’.

Water temperature is one of the main factors determining species distribution (Fenchel & Finlay, 2006; Foissner, 2006), although other environmental parameters are also important (see for instance Orbulina universa in de Vargas et al., 2001, and Darling & Wade, 2008). Genetic studies indicate that water temperature may be even more important than currently assumed, since cryptic species often occupy different temperature niches (e.g. de Vargas et al., 2001; Darling et al., 2007). The equator should therefore act as a strong barrier separating populations in the Northern and Southern Hemispheres; the strength of isolation caused by the equator is likely to be larger in more cold-adapted species. This barrier is hereafter called hemisphere isolation (HI).

Another potentially important barrier to dispersal is land. Nearly all aquatic organisms are confined to either marine or freshwater systems, with a few exceptions such as the ciliate Cyclidium glaucoma where the same genetic lineages seem to occur in both habitats (Finlay et al., 2006). The Americas and Eurasia–Africa are therefore potential barriers confining species to either the Atlantic or the Pacific/Indian Oceans. These barriers are likely to be stronger for tropical organisms as migration south of the land barriers of Africa or South America or north of North America or Asia is potentially possible for cold water species. This type of barrier is hereafter called land isolation (LI).

The final potential barrier is the ocean itself, which may isolate populations of coastal, tidal or estuarine taxa along different coasts but should be unimportant for taxa living in the open ocean. This type of barrier is hereafter called ocean isolation (OI).

Two species attributes are also likely to be important in shaping geographic variation. Planktonic organisms should show less geographic subdivision than benthic organisms as they have increased dispersal via water currents. Smaller organisms may be less likely to show geographic subdivision, either as a consequence of their proposed higher Nes or because passive long-distance dispersal may be more common for smaller organisms. The influence of size on geographic variation is the main focus of this article. Under the EIE hypothesis, size will be the main factor governing geographic variation with dispersal barriers and species ecology playing only a minor role. In contrast, recovery of a relative weaker effect of organism size would contradict the predictions of the EIE.

MATERIALS AND METHODS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

Selection of datasets

Studies of geographic variation in marine microscopic eukaryotes were compiled from the literature to test the EIE hypothesis. Studies on bacteria were excluded because of their known frequent exchange of DNA between distantly related taxa, which makes phylogenetic analyses difficult to interpret. Studies of geographic variation in symbionts were also excluded due to the confounding role of the geographic distribution of their host species. As the EIE implies that biogeography should be ignored for organisms below a certain size, we set an upper size limit for taxa to be included. Theoretical arguments for the actual size limit imposed by the EIE seem to be missing, although based on morphological data, Fenchel & Finlay (2004) argue that it is close to 1 mm. We follow the suggestion of Fenchel & Finlay (2004) and have generally applied this threshold in our analysis. Copepods and annelids contain species above and below this size threshold; we wanted to analyse complete taxonomic groups when possible while minimizing violations of the 1 mm threshold. We therefore chose to include all copepods even though the largest copepod, Neocalanus cristatus, is 2.8 mm but to exclude all annelids larger than 1 mm.

Geographic distribution was simplified by dividing the oceans of the world into eight regions corresponding to the eastern and western side of the Atlantic and Pacific/Indian Ocean on both hemispheres (Fig. 1).

image

Figure 1. Map showing the eight geographic areas used in this study. NEA (north-east Atlantic) is the coastal area surrounding Europe and the Atlantic side of Africa north of the equator including the Mediterranean. SEA (south-east Atlantic) is the Atlantic side of Africa south of equator. NWA (north-west Atlantic) is the area surrounding the eastern side of North America and South America north of the equator. SWA (south-west Atlantic) is the Atlantic side of South America south of the equator. NWP (north-west Pacific) is the Red Sea, Indian Ocean and western Pacific Ocean north of the equator. NEP (north-east Pacific) is the eastern part of the Pacific Ocean north of the equator. SWP (south-west Pacific) is the western part of the Pacific south of the equator and the Indian Ocean. SEP (south-east Pacific) is the eastern Pacific south of the equator.

Download figure to PowerPoint

Montresor et al. (2003) showed very similar ribosomal sequences of the dinoflagellate Polarella glacialis in the Arctic and Antarctic. Studies like this reporting similar or identical DNA sequences from organisms living in distant areas are often considered evidence for EIE, but without any knowledge of the substitution rate or degree of local variation all that can be concluded is that the two populations are fairly closely related. It is impossible to determine if their last common ancestor existed thousands or millions years ago. To allow a comparison of local and global variation and determine whether levels of genetic variation were greater within nearby populations or on a continental scale, we restricted our analysis to datasets that met the following criteria: (1) at least three populations sampled, (2) these populations represented at least two adjoining regions, (3) a minimum of five individuals in total were sampled from these populations, and (4) genetic variation was recovered in the analysed markers.

In all cases we scored whether populations on each side of one of our proposed barriers were reciprocally monophyletic in the phylogenies presented in the original papers. Using reciprocal monophyly as a proxy for geographic variation will underestimate the number of lineages identified as geographically variable, as the initiation of reproductive isolation and cessation of gene flow will substantially pre-date recovery of reciprocal monophyly (Moore, 1995). The expected time until two populations become reciprocally monophyletic once gene flow ceases is proportional to Ne (an average of Ne generations for a mitochondrial marker and 4Ne for a nuclear marker). Despite this time lag, we chose to use reciprocal monophyly because it is readily observable from the original analyses and its conservative nature strengthens inferences of geographic variation.

It should be noted that the goal of this study is not strictly a test of EIE but of the uniformly high levels of dispersal, which is the theoretical argument behind EIE. Generally, previous examinations of EIE have been focused at the species level; we included all studies with morphologically similar monophyletic genetic lineages. We chose to include currently recognized species as well as lineages with taxonomic status above or below the species level since all of these lineages represent independent evolutionary histories which may or may not show isolation across a barrier.

The chosen datasets generally corresponded to the smallest named lineages in the original studies. In some of the included datasets sister lineages with similar temperature preferences occurred in adjoining regions and were combined in a single dataset. A study of the intertidal foraminiferan Ammonia exemplifies how we selected datasets. Ammonia was split into 13 genetic types assumed to be cryptic species (Hayward et al., 2004). Three of these, T1, T2 and T6, were sampled in several regions; we included these datasets in our analysis. One of these and an additional type (the sister lineages T6 and T7) together form a monophyletic lineage occurring on each side of the Atlantic Ocean barrier; we combined these types as T6-7 in our analysis of this barrier.

A single dataset on the ciliate Cyclidium glaucoma consisted of both marine and non-marine isolates (Finlay et al., 2006); only the marine samples were included in our analysis. We excluded published datasets that recovered reciprocal monophyly which could not unambiguously be assigned to a temperature, land or ocean barrier. In some cases geographically based reciprocal monophyly could only be assessed for some of the sampled regions, and therefore only some of the data were included. For example, in the study by von Soosten et al. (1998) on the polychaete Petitia amphophthalma we excluded the population in Florida. Although this population appeared monophyletic in their study, too few populations were sampled to determine whether the level of genetic distinctiveness is high relative to other populations of this species. As a result, for this polychaete we could only analyse the strength of isolation between the north-west Pacific and the north-east Atlantic.

There are two potential problems with a meta-analysis like this, but neither is likely to influence our conclusions. The subjectivity inherent in the selection of datasets could potentially bias the number of lineages showing reciprocal monophyly; choice of datasets should have no effect, however, on the correlation between a lineage recovered as reciprocally monophyletic and the evaluation of potential causal factors. Secondly, meta-analyses can potentially generate biased results if negative and positive findings are not equally likely to be published. As the same authors have published studies showing both presence and absence of geographic variation (Schmidt & Westheide, 1998, 1999) and the degree of geographic variation in microscopic organisms is still a topic of intense discussion it is unlikely that a publication bias exists to the extent that it would influence our conclusions.

Three different subsets of the compiled data were analysed to evaluate the effect of HI, LI and OI. The ocean subset consisted of taxa with populations on both sides of oceans in the same hemisphere (NWA/NEA, NWP/NEP, SWA/SEA, NWP/NEP; see Fig. 1). The hemisphere subset consisted of taxa with populations from the same side of the oceans on both hemispheres (NWA/SWA, NEA/SEA, NWP/SWP, NEP/SEP). The land subset consisted of taxa with populations from both sides of a potential land barrier (NWA/NEP, NEA/NWP, SWA/SEP, SEA/SWP). The precise separation between adjoining regions was difficult to determine when not defined by continents or equator, but there were no cases where minor shifts in these borders changed the results.

Variable estimation

All taxa were scored for the strength of the isolating barrier, body size, lifestyle, phylogenetic affinity, the data type (sequence or non-sequence data) and the probability of sampling errors.

Sampling error (the probability that apparent reciprocal monophyly in different regions is caused by undersampling) was scored in two different ways. Sampling incompleteness is the logarithm of the probability that both region a and b comprise single distinct monophyletic unit given that a total of A individuals from region a and B individuals from region b have been sampled if there is no geographic barrier to gene flow:

  • image

Population incompleteness is similar, but instead of the total number of individuals sampled in each region, the number of populations is used instead to account for the non-independence of individuals sampled from the same population. Both measures were log-transformed and the number of individuals or populations sampled on each side of the relevant barrier was used in the calculation of sampling error. The measures are based on tree geometry and arise from the fact that the number of rooted trees with N taxa is (2N− 3)!/(2N−2× (N − 2)!) for N≥ 2 (Nei & Kumar, 2000). The numerator of the formula therefore represents the number of trees where both area a and area b are monophyletic and the denominator represents the total number of possible trees.

Sizes of organisms with different body plans are difficult to compare. We estimated the average volume of an organism assuming a cylindrical or ball shape for all organisms. Other studies have used body length as a measure of size (Wilkinson, 2001; Fenchel & Finlay, 2004). For comparison, the cubic root of the volume was calculated corresponding to the body length of a cube-shaped organism. When possible, size was estimated from the same dataset as the genetic data or from other recent sources. All taxa were also scored as holoplanktonic or as benthic or sessile during part of their life cycle.

A relationship between these factors and the probability of geographic isolation could potentially result from phylogenetic inertia if some taxonomic groupings are more likely to show geographic subdivision. To test for this, taxonomic structure was included in our analysis and coded two different ways: as (1) subdomain of eukaryotes – Plantae, Unikonta, Chromalveolata; or (2) major groups of eukaryotes as defined by the Tree of Life project (Keeling et al., 2009). All sampled Unikonta belonged to the Metazoa and will be referred to as such. Single representatives were merged with their closest relatives.

For the analysis of hemisphere isolation, the strength of the temperature barrier (TB) between individuals in different hemispheres was calculated as the difference between the average temperature in the warmest part of the known distribution of that species and the average temperature in the warmest part of the ocean in which it was found. Surface temperatures were generally used, except for three datasets on deepwater foraminiferans (Pawlowski et al., 2007) for which the temperature at 1000 m was used to reflect the lower variation in temperature of deeper water.

Strength of the land barrier (LB) reflects the difficulty of migration around the continents; although cross-continent dispersal is possible, it is probably rare and was ignored in this simple model. For all populations the difference between the average annual temperature of the locality and the temperature immediately north or south of the landmass separating the populations from the neighbouring area (the northern tip of North America for NEP/NWA, the northern tip of Eurasia for NEA/NWP, the southern tip of South America for SEP/SWA or the southern tip of Africa for SEA/SWP) were calculated and minimum value used as the estimate of LB. Ocean surface temperatures were used in all cases for LB. Temperatures used for both LB and TB were based on yearly averages from Locarnini et al. (2006).

The temperature range of each taxon was estimated from the populations that were sampled for DNA. Although this will underestimate the temperature range wherein a taxon is found, it cannot reasonably be assumed that the morphospecies reported in the literature correspond to the same taxa used in this study since cryptic species are common. This method resulted in an overestimation of the LB or TB of each taxon and therefore reduced the relative isolation strength of a given value of LB or TB (e.g. the isolation strength of an LB or TB of 5 °C).

For the analysis of oceanic isolation, an estimate of the water barrier (WB) was required. Organisms only sampled less than 80 km from the coast (corresponding to the average width of the continental shelf) were scored as coastal, while organisms sampled at least once more than 80 km from the coast were scored as oceanic. This barrier was scored as a dummy variable.

Statistical analysis

The impacts of temperature, land and ocean barriers were analysed separately using generalized linear models with the presence or absence of reciprocal monophyly as the dependent variable. This type of analysis is similar to standard ANCOVA analysis with two noteworthy exceptions: (1) the dependent variable is not assumed to be normally distributed (our data were binomially distributed); and (2) a linear function of the chosen variables (β01x1+ . . . βnxn) is not equal to the dependent variable but to a function of the dependent variable called the link function. In this analysis three different link functions were used: (1) logit [ln(p/1 −p) =β01x1+ . . . βnxn]; (2) probit [Φ(p) =β01x1+ . . . βnxn where Φ is the cumulative normal distribution function]; and (3) complementary log–log (ln[−ln(1 −p)]=β01x1+ . . . βnxn).

Four biological variables and all their first-order interactions were considered as potential explanatory variables. The variables were body length, lifestyle (benthic/planktonic), relevant barrier estimator (TB, LB, WB) and taxonomic structure. Size, land and temperature barrier were analysed as having either a linear (β1× LB; β1× TB; β1× size) or a logarithmic effect [β1× ln(LB + 1); β1× ln(TB + 1); β1× ln(size)]. Three sampling variables (sampling incompleteness, population incompleteness and data type) were also included as potential explanatory variables. Lifestyle, water barrier, taxonomic structure and data type were analysed as dummy variables. Sampling and population incompleteness were highly correlated with temperature and land isolation (r= 0.68–0.91) and therefore only one of the variables was kept in the model for the analyses. All other explanatory variables had correlations lower than 0.5, and although some of them were significant, it was assumed that this low correlation would not interfere with the analyses.

Models with limited sample sizes relative to a large number of explanatory variables risk overfitting the data by incorporating variables which may appear significant but do not represent a true relationship. For data that are binomially distributed, a recent simulation study indicates that such problems are not a major concern when there are at least five events per explanatory variable (Vittinghoff & McCulloch, 2007). In this study we had an expected sign of most potential variables (larger probability for geographic variation with increased land barrier, temperature barrier or size, lower probability with increased sampling intensity and higher probability for coastal or benthic organisms) and could effectively halve the probability of a type 1 error by looking at the sign of the effect. We therefore chose to set a maximum of 2.5 events per variable. This meant that the models considered contained a maximum of two variables for temperature isolation (six events of geographic variation), three variables for water isolation (eight events of geographic variation), and six variables for land isolation (14 events of geographic variation). We noted that this still left a risk of overfitting the data, but believed our solution to be the best compromise between fairly limited data and a large number of potential variables.

The best model was estimated as the combination of link functions and variables with the lowest Bayesian information criterion (BIC) [−2 × log-likelihood + ln(t) ×npar, where t is the number of taxa in the analysis and npar is number of explanatory parameters] (Schwarz, 1978; Hardin & Hilbe, 2007), with the restriction that one of the measures of body size was kept in the model even if the BIC value was lower for a model excluding body size. BIC is similar to the more widely applied Akaike information criterion (AIC) (−2 × log-likelihood + 2 ×npar) but is more conservative and was chosen to further reduce the risk of overfitting.

This procedure enabled the selection of the best model from our subset but did not guarantee that any of the models was sufficient. The absolute fit of the best model was therefore checked with several analyses of the Pearson residuals. Normality of the residuals was checked with D'Agostino tests for skewness (D'Agostino, 1970) and Anscombe–Glynn tests of kurtosis (Anscombe & Glynn, 1983) using the moments package in R 2.7.2 (R Development Core Team, 2009). Bias of the residuals was checked with standard t-tests. All P-values for D'Agostino, Anscombe–Glynn and t-tests were Bonferroni corrected (n= 3). Finally, we graphically checked whether the relationship between the residuals and predicted values fitted the expectations for a binomial distribution (largest variance for intermediate probabilities).

The importance of suspicious outliers was analysed by calculating the Pearson residuals of the best model and investigating all taxa with residuals greater than two standard deviations from the mean. If information was available that questioned our scoring of geographic variation of any variable the dataset was removed and the data reanalysed.

All analyses were performed both with the raw data and with all continuous explanatory variables standardized to have a mean of zero and a standard deviation of one. The analysis of standardized data was performed to help the comparison between the importance of different variables measured on different scales. The analyses of raw data were used to construct graphs that plot the probability of geographic variation as a function of body size and other variables. The best-fitting model for the standardized data was used to construct the graphs. All analyses were performed using generalized linear models in R 2.7.2 using the function glm (R Development Core Team, 2009) and the search for the best models was performed by manually specifying all alternatives.

RESULTS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

Ocean isolation

A total of 40 taxa were analysed. Of these, eight were reciprocally monophyletic on each side of an ocean. The average body size of non-reciprocally monophyletic taxa was 536 µm (range 2.8–2879 µm). Of non-reciprocally monophyletic taxa, 47% were coastal and 75% were planktonic. Moreover, 17 belonged to Chromalveolata, 14 to Metazoa and 1 to Plantae. Two datasets were based on non-sequence data. The mean sampling incompleteness was −13.5 (range −75 to −0.48) and the mean population incompleteness was −2.1 (range −7.3 to 0).

The average body size of the reciprocally monophyletic taxa was 250 µm (range 4.8–872 µm). All were coastal and 38% were planktonic. Four belonged to Chromalveolata and 4 to Metazoa. One dataset was based on non-sequence data. The mean sampling incompleteness was −14.5 (range −21 to −3.1) and the mean population incompleteness was −2.1 (range −7.4 to −0.7). See Appendix S1 in Supporting Information for detailed information on the taxa.

The best model for the data was a logistic regression with ocean isolation and the logarithm of body size as independent variables (Table 1). Body size was retained in the model even though the BIC score improved when body size was removed (BIC = 37.098 vs. 40.247). No apparent outliers were detected and there was no indication of model violations. The probability of geographic variation as a function of body size and ocean isolation is shown in Fig. 2.

Table 1.  Optimal models for probability of geographic variation.
BarrierFormulaDummy variablesTests of residuals
Residuals skewnessKurtosisMean
  • **

    0.05 > P > 0.01;

  • ***

    P < 0.01.

  • The water barrier (WB) of the middle of the oceans, the temperature barrier (TB) of the equator and the land barrier (LB) of the continents are further explained in Materials and Methods. Land isolation (LI) was analysed both with and without body size (BS) as a potential variable.

  • To test the reliability of the models, the skewness, kurtosis and mean of the residuals are shown as are the significance of these values.

  • OI, ocean isolation; SI, sample incompleteness; TI, temperature isolation.

OIln(p/1 −p) =−19.77 + WB + 0.413 × ln(BS)WB: 19.240.902.805.2 × 10−4
TIΦ(p) =−1.78 + 1.24 × ln(BS) + 1.03 ln(TB + 1) 0.484.03−0.04
LI with BSln(p/1 −p) =−3.06 + group −3.50 × SI −1.42 × ln(BS)Plantae: 2.01; Metazoa: 4.242.40**12.2***0.03
LI without BSln(p/1 −p) =−0.55 − 3.67 × SI + 0.94 × ln(LB + 1)    
image

Figure 2. Probability of reciprocal monophyly between populations from different sides of the same ocean. The lines are drawn based on an analysis of the raw data with the model that provided the best fit for the transformed data (logistic regression with the logarithm of body size and water barrier as independent variables). The different lines represent the probability of reciprocal monophyly as a function of body size for coastal and oceanic organisms.

Download figure to PowerPoint

Hemisphere isolation

A total of 36 taxa were analysed. Six of these were reciprocally monophyletic in each hemisphere. The average body size of non-reciprocally monophyletic taxa was 341 µm (range 4–1861 µm). The average temperature barrier was 3.7 °C (range 0–12°C) and 90% were planktonic. Moreover, 24 belonged to Chromalveolata and 6 to Metazoa. The mean sampling incompleteness was −10.2 (range −106 to 0) and the mean population incompleteness was −2.0 (range −6.4 to 0).

The average body size of the reciprocally monophyletic taxa was 600 µm (range 192–1297 µm). The average temperature barrier was 6.7 °C (range 2–12 °C) and 57% were planktonic. Two belonged to Chromalveolata and four to Metazoa. The mean sampling incompleteness was −33.6 (range −157 to −5.7) and the mean population incompleteness was −5.8 (range −25.4 to −0.7). See Appendix S1 for detailed information on the taxa.

The best model based on our criteria was a probit regression with the logarithms of temperature isolation and body size as independent variables (BIC = 30.653 including body size; BIC = 35.983 excluding body size; Table 1). No apparent outliers were observed and there was no indication of model violations. The probability of geographic variation as a function of body size and ocean isolation is shown in Fig. 3.

image

Figure 3. Probability of reciprocal monophyly between populations from different hemispheres in the same ocean. The lines are drawn based on an analysis of the raw data with the model that provided the best fit for the transformed data (probit regression with the logarithm of body size and the logarithm of water temperature barrier as independent variables). The different lines represent the probability of reciprocal monophyly as a function of body size for three different temperature barriers.

Download figure to PowerPoint

Land isolation

A total of 41 taxa were analysed. Fourteen of these were reciprocally monophyletic on each side of Eurasia or the Americas. The average body size of non-reciprocally monophyletic taxa was 298 µm (range 4–1861 µm). The average land barrier was 7.3°C (range 0–26°C) and 74% were planktonic. Moreover, 18 belonged to Chromalveolata, one to Plantae and eight to Metazoa. The mean sampling incompleteness was −4.24 (range −16.3 to −1) and the mean population incompleteness was −1.4 (range −3.8 to 0). One dataset was based on non-sequence data.

The average body size of the reciprocally monophyletic taxa was 511 µm (range 1.6–1861 µm). The average and barrier was 11.4°C (range 0–27°C) and 50% were planktonic. Two belonged to Chromalveolata, one to Plantae and eleven to Metazoa. The mean sampling incompleteness was −16.6 (range −93 to −0.7) and the mean population incompleteness was −2.0 (range −7.1 to 0). One dataset was based on non-sequence data. See Appendix S1 for detailed information on the taxa.

The best-fitting model including body size was a logistic regression with a eukaryotic subdomain and the logarithm of body size and sampling incompleteness as independent variables (BIC = 56.818; Table 1), while the best-fitting model overall had the logarithms of land isolation and sampling incompleteness as independent variables (BIC = 53.838; Table 1). In the model including body size, one taxon (the diatom Pseudo-nitzschia delicatissima) had residual values outside the threshold of two standard deviations. In the model not including body size two taxa (the red alga Rhodosorus marinus and the bryozoan Membranipora membranacea) had residual values outside the threshold of two standard deviations. For all three taxa there was nothing odd about the scoring.

For both the model including body size and the model without body size the likelihood of reciprocal monophyly was systematically higher with more intense sampling (Table 1; a more negative value for sampling incompleteness indicates a larger number of samples) indicating that the relationship cannot be explained by undersampling. In both cases D'Agostino (1970) tests for skewness, Anscombe & Glynn (1983) tests for kurtosis, and graphical analysis of the relationship between the residuals and predicted values indicated either model violations or a poor model fit. The results from analysis of land isolation will not be discussed further.

DISCUSSION

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

The results from ocean isolation only showed evidence for water barrier as a factor controlling the probability of geographic isolation. The estimated effect of body size was fairly large, and although the variable was not included in the best-fitting model it is possible that this is a power issue and that an effect of body size would be recovered with larger sample sizes. Both organism size and temperature barrier variables appeared to be important for generating geographic structure across the equator, but for isolation to be likely both a fairly large temperature barrier and body size were necessary and neither the largest body size nor temperature barrier in itself was sufficient to make isolation likely.

Surprisingly, the effect of lifestyle (benthic/holoplanktonic) was excluded in the best-fitting model in any subset, although dispersal ought to be easier for planktonic organisms. This result could be an artefact of our classification of lifestyle; holobenthic organisms may differ when compared with organisms with both planktonic and benthic life-cycle stages. It is possible that a more refined analysis of lifestyle might yield different results.

Some of the metazoan datasets used in this meta-analysis consisted of organisms larger than the often used upper limit of 1 mm for cosmopolitan species. This should not influence the conclusions, however, as the inclusion of larger organisms that display geographic variation can only increase the relative importance of body size compared with the other variables.

Although we applied stringent criteria to score the datasets used in this analysis in an unbiased manner, other researchers may interpret some of the results differently. Slapeta et al. (2006) found five deep lineages in the tiny Micromonas pusilla and, based on a molecular clock, suggested a Cretaceous origin. The occurrence of each of these lineages in several oceans has been used to argue that M. pusilla is a cosmopolitan taxon (Finlay et al., 2006). The distributions of lineages separated since the Cretaceous tell us little about present dispersal patterns. In fact, one of the lineages recovered by Slapeta et al. (2006) was included in our analysis and showed reciprocal monophyly between oceans.

One of the main problems in studies of geographic variation is the choice of the appropriate taxonomic level; the higher the taxonomic level, the lower the degree of geographic variation. Advocates of EIE consider morphospecies to be the appropriate level (Fenchel & Finlay, 2006) and at that level EIE may be well supported. Of the 28 cases of geographic variation examined in this study we only found five (three metazoan and two protozoan) examples of the co-occurrence of morphological variation. For the two protozoan cases, although the geographic varieties in the dataset are morphologically distinguishable from each other they are not distinct from less closely related taxa (Hayward et al., 2004; Katz et al., 2005). In the three metazoan examples, however, the diagnostic characters appear more consistent, albeit still small (Schmidt & Westheide, 1998; Dick et al., 2003). There are well-known cases of apparent ecological speciation without phenotypic differentiation in microscopic organisms (de Vargas et al., 2002). Additionally, recent evidence suggests morphological and genetic differentiation can be decoupled in microorganisms (Logares et al., 2007). Given these and other difficulties with the application of a morphospecies concept, we argue that the use of lineages represents a more appropriate approach.

Some of our results could potentially be caused by undersampling. Should undersampling be a general phenomenon, however, we would expect to consistently recover a negative correlation between sample size and likelihood of geographic variation. As we recovered a positive correlation between these variables, it is unlikely that undersampling has influenced these analyses. The smallest organisms for which we recovered isolation between the two sides of the Atlantic were the diatoms Skeletonema menzelii and Skeletonema pseudocostatum (Kooistra et al., 2008), which had small sample sizes (S. menzelii n= 4 vs. 5; S. pseudocostatum n= 1 vs. 20), yet the probability of recovering reciprocal monophyly by chance is low: 0.077% (1575 out of 2.0 × 106 possible trees with 9 taxa) and 2.6% (8.2 × 1021 out of 3.2 × 1023 possible trees with 21 taxa), respectively. Similarly, the smallest organisms showing reciprocal monophyly across a land barrier were M. pusilla clade A (n= 2 vs. 3) and P. delicatissima clade A (n= 1 vs. 5) (Lundholm et al., 2006; Slapeta et al., 2006) sampled from each side of Eurasia/Africa. In this case, the probability of recovery of random reciprocal monophyly is 2.9% (3 out of 105 possible trees with five taxa) and 11.1% (105 out of 945 possible trees with six taxa), respectively. Although it is possible that some of these lineages may turn out not to be reciprocally monophyletic, it is very unlikely that this will be the case for all of them.

Human-induced transport between areas that were formerly isolated could introduce endemic lineages into new regions. A future recovery of a lineage in a new region would therefore not necessarily support EIE. Several likely candidates for such transport, for instance some of the different lineages in the Alexandrium tamarense complex, have been incorporated in this analysis as datasets showing no geographic variation. Unnoticed occurrence of this organism is unlikely, since it causes paralytic shellfish poisoning which has only been noticed recently in several areas (Lilly et al., 2007). The natural frequency of cosmopolitan organisms may therefore be lower than suggested by this meta-analysis.

Our analysis does not provide support for the EIE hypothesis, and we propose that future work should focus on the specific groups where geographic variation is present. Geographic variation is indeed a real phenomenon in microscopic marine organisms but there are still a number of questions awaiting further analyses. The most important of these is the time-scale at which different lineages have been isolated. Few groups have reliable molecular clocks to estimate the time since the most recent common ancestor, with some separations between Arctic and Antarctic planktonic foraminiferans as noteworthy exceptions (Darling et al., 2007).

Furthermore, the variables included in this analysis only explained part of the variation. It could be illuminating to explore the likelihood of geographic variation between different taxonomic groupings or to investigate the relative importance of the different variables in different size groups. Such analyses are, however, not possible at the moment because the number of published studies is too small to make reliable analyses on smaller subsets possible.

ACKNOWLEDGEMENTS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

This work was supported by The Danish Natural Science Research Council (no. 272-06-0534) and National Geographic's Committee for Research and Exploration (no. 8488-08). We thank Sean Connolly and four anonymous referees for invaluable comments and suggestions on the manuscript. Thanks are due to Dave Parker, Vanessa Kellermann and Alison Hamilton for improving the English of the manuscript.

REFERENCES

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

BIOSKETCHES

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

S. Faurby is currently a post-doc in the Department of Ecology and Evolutionary Biology at the University of California, Los Angeles. He is interested in the geographic variation of microscopic animals at both regional and global scales. This work represents part of his dissertation research conducted in the Department of Biological Sciences at Aarhus University.

P. Funch studies meiofauna and evolution of the animal kingdom. His current research is mainly on marine phylogeography. He is an associate professor in the Department of Biological Sciences at Aarhus University.

S.F. designed the experiment and performed the analyses. S.F. and P.F. wrote the paper.

Editor: Sean Connolly

Supporting Information

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. REFERENCES
  9. BIOSKETCHES
  10. Supporting Information

Appendix S1 Raw data for the analyses and references to original studies.

FilenameFormatSizeDescription
GEB_609_sm_AppendixS1.doc175KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.