Introduction

‘Everything is everywhere, but, the environment selects’ (Baas-Becking, 1934; de Wit and Bouvier, 2006) suggests that dispersal is not limiting for microbes, but that the environment is the primary factor determining whether a particular microbe is actively participating in a given community. If true, microbial biogeographic distributions may be reflective only of technical limitations in detection, whereas their active or meaningful participation in a community will depend on the suitability of a given habitat to foster positive population growth (Gibbons et al., 2013; Hambright et al., 2014). The marine haptophyte Prymnesium parvum Carter is considered to be a globally invasive species in many freshwater systems in which it is now a community member and often dominant. However, recent study has suggested that P. parvum is not dispersal limited, as extremely low-density populations have been detected in habitats that do not experience P. parvum blooms, many of which are directly downstream of P. parvum bloom sites (Zamor et al., 2012). This phenomenon reflects the general notion of the rare or dormant microbial biosphere in microbes (Sogin et al., 2006; Caron and Countway, 2009; Jones and Lennon, 2010; Gibbons et al., 2013), and leads to the hypothesis that environmental conditions that foster high growth in P. parvum relative to other microbial constituents are the principle determinants of P. parvum blooms in inland freshwater systems.

P. parvum blooms and fish kills have been observed in many inland aquatic systems worldwide (Granéli et al., 2012), including waterbodies, mostly reservoirs, in at least 20 US states (Roelke et al., 2011; Hambright, 2012). This apparent incredible range expansion since the first North American report (Pecos River of southern Texas) in the 1980s (James and De La Cruz, 1989) represents an interesting enigma—while P. parvum has an ability to thrive across a broad range of environmental conditions (Edvardsen and Paasche, 1998), the conditions found in most North American bloom sites tend to be far removed from optimal conditions described from the laboratory study (Baker et al., 2009). Most inland blooms of P. parvum in the southwestern United States have occurred at relatively low salinities (1–3 partial salinity units, psu) during winters when temperatures range from 10 °C to 20 °C, yet laboratory studies suggest that inland strains of P. parvum are well suited to high salinities (8–30 psu) and temperatures (20–30 °C) (Baker et al., 2007, 2009; Hambright et al., 2010, 2014; Roelke et al., 2011; Patiño et al., 2014). This apparent paradox has led to speculation that its toxigenic capabilities provide a competitive edge to P. parvum over other algae, allowing blooms to develop during periods of stress, such as created by low nutrient availabilities (for a review, see Granéli et al., 2012). However, while toxicity may have important roles in predator avoidance and heterotrophy in this unicellular mixotroph, toxin production is unlikely to provide a competitive advantage to P. parvum to the degree necessary to lead to bloom formation under suboptimal environmental conditions (Jonsson et al., 2009; Remmel and Hambright, 2012).

Because it is a microbial eukaryote that can occur in immense numbers, encyst and be passively transported, its dispersal capabilities are potentially unlimited (sensu Finlay, 2002; but see Martiny et al., 2006; Hanson et al., 2012). Indeed, examination of the P. parvum distribution in Lake Texoma (Oklahoma–Texoma) reveals that P. parvum is dispersed throughout the lake, yet blooms and fish kills are common only in areas in which environmental conditions are conducive for growth (Hambright et al., 2010; Zamor et al., 2012). Thus, the recent expansion of P. parvum into new habitats, which is typically noted only after a bloom and fish kill, suggests that habitats with suitable environmental conditions (e.g., elevated nutrients and salinities, see below) may be becoming more abundant, particularly in the southwestern United States, where quality of surface water resources is subjected to the pressures of climate change and increasing freshwater demands that accompany growing human populations and development (Roelke et al., 2011).

Here we report results of a long-term study of P. parvum in subtropical Lake Texoma (Oklahoma–Texas, USA), in which P. parvum blooms are now commonplace. We use these data to test the hypothesis that environmental conditions that foster high growth in P. parvum are the principle determinants of P. parvum blooms in the lake. Our analysis provides further support for the Baas-Becking hypothesis ‘Everything is everywhere, but, the environment selects’, as we identify a primary role for environmental conditions, but not dispersal, in the lake-wide distributions and bloom patterns in P. parvum.

Materials and methods

Study site

Lake Texoma (Figure 1), an impoundment of the Red and Washita Rivers, was constructed in 1944 for flood control, hydropower generation and recreation. The lake is the 12th largest reservoir in the United States (at normal pool elevation) with a surface area of 360 km2, and mean and maximum depths of 8.7 and 26 m. The lake watershed occupies 87 500 km2 of the high plains of Texas and the rolling plains of Texas and Oklahoma. Owing to this extremely large watershed (watershed area: lake surface area=243), nutrient loading to the lake is high and the lake is eutrophic to hypertrophic, depending on season and location within this complex, dendritic reservoir (Oklahoma Water Resources Board, 2010). The watershed also contains abundant deposits of calcium carbonate, halite, gypsum, anhydrite and other Permian-Salado evaporites (Ground and Groeger, 1994), which lead to salinities that often exceed 1 psu (defined here as having a specific conductivity equivalent to 1 g l−1 NaCl), the general limit for fresh water. The phytoplankton are often dominated by filamentous and colonial cyanobacteria and the lake is home to a diverse array of zooplankton, including multiple daphniids, numerous copepod and rotifer species (Franks et al., 2001; Hambright et al., 2010), and more than 50 fish species, many of which are recreationally important (Matthews et al., 2004).

Figure 1
figure 1

Schematic of Lake Texoma location and sampling sites. L1–L8 are littoral sites and P1–P5 are pelagic sites.

Full size image

Lake sampling

We sampled eight littoral sites along the northern Oklahoma side of Lake Texoma in marinas and coves, and five pelagic stations in the Red and Washita River arms of the lake, as well as at the dam (Figure 1). Littoral samples were taken in shallow (usually <1 m) waters, usually from boat ramps or docks, if present. Temperature and salinity (as specific conductance), and dissolved oxygen, chlorophyll and phycocyanin concentrations were measured in situ with YSI and Hydrolab sondes, using a single mid-water point for littoral sites, and surface-to-bottom water column profiles for pelagic sites. Water samples (1 l; mid-water samples for littoral sites or 6 to 10 m, depending on depth, integrated samples for pelagic sites) were collected in acid-washed, deionized- and sample-rinsed Nalgene bottles, stored on ice in the field and refrigerated in the laboratory for subsequent analyses of P. parvum abundances by quantitative PCR (Zamor et al., 2012) (microscopy (hemocytometer, 6–12 fields per sample) was used in 2006–2007), and for concentrations of chlorophyll (acetone extraction), total nitrogen (TN) and total phosphorus (TP) (flow-injection autoanalysis) (American Public Health Association, 1998) and pH. Lake surface elevation data were taken from the 0800 hours measurements recorded at the Denison Dam (Army Corps of Engineers–Tulsa District, 2012). Further details of sampling and sample analyses and additional parameters monitored can be found in Hambright et al. (2010).

Statistical analyses

All statistical analyses were carried out in R (version 3.0.2; R Development Core Team, 2013). We used logistic regression analysis to examine relationships between Prymnesium distributions within the lake (categorized as either presence–absence or bloom–no bloom, where bloom was quantified as ⩾10 000 cells per ml) and environmental conditions, in which the predictor variables (identified from earlier studies; (Hambright et al., 2010; Zamor et al., 2012) consisted of temperature (°C), salinity (psu), TN (mg N l−1), TP (mg P l−1), molar TN:TP and site type (littoral or pelagic); N=1181 (rms package; version 4.0-0; Harrell, 2013). This approach of dichotomizing the response variable and using logistic as opposed to a continuous generalized linear modeling approach was taken because these response variables best fit our question (i.e., comparing variables predicting presence and blooms) and because our data are zero inflated owing to a preponderance of absence and non-bloom situations. Preliminary analysis of environmental predictors indicated correlations between TN and TP and between TP and TN:TP (see Supplementary Information and Supplementary Figure S1), but variance inflation factors were sufficiently low (all <5) to rule out potential effects of multicollinearity on parameter estimates or goodness-of-fit metrics (Davis et al., 1986). We quantified goodness of fit of logistic regression models using the model likelihood ratio χ2; Nagelkerke’s R2N as a measure of explained variance; the C index, a measure of concordance of model prediction, which ranges from 0.5 (random) to 1 (perfect), >0.8 indicates a useful model; and Somer’s Dxy, which measures the rank correlation difference between concordance and discordance of predictions to model, and ranges from 0 (random) to 1 (perfect). Logistic regression models were further validated using bootstrapping (n=100) to measure the degree of overfitting; slope ranges from 0 (extreme overfitting) to 1 (no overfitting). Predictors of P. parvum presence and blooms were further elucidated using classification tree analyses (rpart package; version 4.1-1; Therneau et al., 2013). Classification trees complement logistic regression well because they can capture complex interactions and non-monotonic relationships between predictors and response variables, and display these complex relationships in easily interpreted plots (De'ath and Fabricius, 2000). Classification trees predicting either presence or blooms of P. parvum were assembled using the default settings of rpart. To minimize overfitting, trees were pruned to minimize cross-validated error (for details, see De’ath and Fabricius, 2000).

Although chlorophyll, dissolved oxygen and pH are routinely measured in water quality monitoring programs owing to ease of measurement, any relationship between P. parvum and these predictors may be confounded by the direct influence of algal growth and abundances on these three variables. These variables were strongly correlated with other variables used in the models (Supplementary Figure S1), although, as before, the variance inflation factors were sufficiently low (all <5) to suggest multicollinearity was not likely a substantial problem. In addition to the previously described predictors, our previous analysis (Hambright et al., 2010) suggested that water level elevation might be a good predictor of P. parvum abundances in Lake Texoma, and Grover et al. (2012) have suggested a potential negative relationship between cyanobacterial and golden algal abundances. Therefore, we also ran logistic regression and classification tree analyses in which chlorophyll, DO, pH and elevation (N=1181), and chlorophyll, DO, pH, elevation and cyanobacterial abundances, as phycocyanin concentrations (N=560) were included as predictor variables in an effort to ascertain any increase in predictive capabilities for P. parvum presence and blooms by the inclusion of these additional variables. Full and reduced models were compared using the corrected Akaike Information Criterion (for details, see Burnham and Anderson, 2002).

Results

P. parvum bloomed in at least one cove of Lake Texoma in seven of the nine winters during 2004–2012 (Figure 2). Highest densities were observed at littoral sites, such as Lebanon Pool, on the western Red River arm of the lake, and lower densities further downstream and on the Washita River arm of the lake (see also Hambright et al., 2010; Zamor et al., 2012). As demonstrated previously, prevalent environmental conditions during the bloom years included relatively high salinities and nutrients. Net maximum population growth rates of P. parvum during the initial bloom phase in Lebanon Pool (calculated as the slope of the log of P. parvum densities over time during the initial growth period of each bloom) scaled positively with salinity, but were not significantly related (all P>0.05) to TN:TP, pH or temperature (Figure 3).

Figure 2
figure 2

Monthly mean abundances of P. parvum and temperature, salinity (as psu), total nitrogen and total phosphorus concentrations, and molar N:P in Lake Texoma littoral (black line) and pelagic (dashed line) stations during 2004–2012.

Full size image
Figure 3
figure 3

Net growth rates of P. parvum during initial bloom development in Lebanon Pool (L2) during the winters of 2004–2005 through 2011–2012 in relation to mean salinity, TN:TP, temperature and pH during the period of population increase. Note that the relationship between growth rate and pH is confounded by an expected increase in CO2 consumption (and concomitant increase in pH) with increased growth rates.

Full size image

Logistic regression analysis revealed that the presence of P. parvum was best predicted by salinity, site type, temperature, TP and TN:TP; TN was not a significant predictor (Table 1). Addition of chlorophyll, DO, pH and water elevation to the model resulted in a slight improvement of the overall model fit, with chlorophyll, pH and water elevation being added as significant predictors of P. parvum presence (assessed as Nagelkerke's R2; Supplementary Table S1). Model comparison using corrected Akaike Information Criterion suggested that the second, 10-predictor model (including chlorophyll, DO, pH and water elevation) was strongly preferred over the 6-predictor model excluding these factors (Supplementary Table S3). Addition of phycocyanin to the analysis resulted in a smaller data set (n=560) and did not improve either 10- or 6-predictor model results (Supplementary Table S4).

Table 1 Results from logistic regression analysis showing the independent variables, partial logistic regression coefficients (B), standard errors of the partial slope coefficients (s.e.), Wald test and the significance level (P)
Full size table

Blooms of P. parvum (defined as ⩾10 000 cells per ml) were best predicted by salinity, site, temperature, TN, TP and TN:TP (Table 2). Addition of chlorophyll, DO, pH and water elevation to the model also resulted in a slight improvement of the overall bloom model, with chlorophyll and water elevation being added as significant predictors of P. parvum blooms (assessed as Nagelkerke’s R2; Supplementary Table S2). Corrected Akaike Information Criterion model comparison suggested that the second, 10-predictor model (including chlorophyll, DO, pH and elevation) was preferred over the reduced, 6-predictor model for prediction of P. parvum blooms (Supplementary Table S5). Addition of cyanobacteria to the analysis reduced the size of the data set (n=560) and slightly weakened both 10- and 6-predictor model results (Supplementary Table S6).

Table 2 Results from logistic regression analysis showing the independent variables, partial logistic regression coefficients (B), standard errors of the partial slope coefficients (s.e.), Wald test and the significance level (P)
Full size table

The classification tree analysis of P. parvum presence generated a pruned tree with three splits and four terminal nodes with a misclassification rate of 0.243 (Figure 4a). Only one terminal node was classified by environmental conditions (salinity ⩾1.3 psu, temperature <21.7 °C and TP <0.29 mg l−1) in which P. parvum was likely to be present. The classification tree analyses of P. parvum blooms generated a pruned tree with only one split and two terminal nodes with a misclassification rate of 0.059 (Figure 4b). In this case, P. parvum blooms were predicted to be most common when salinity ⩾1.7 psu.

Figure 4
figure 4

Pruned classification trees for P. parvum presence or absence (a) and bloom or no bloom (b) using the predictors salinity (psu), site type (littoral, pelagic), temperature (°C), TN (mg l−1), TP (mg l−1) and molar TN:TP. Presence was defined as detectable by microscope (⩾166 cells per ml; 2006–2007) and quantitative PCR (⩾26 cells per ml; 2008–2012) analyses; misclassification rate=0.243 (presence) and 0.059 (bloom); N=1181.

Full size image

Discussion

Previously, based on a 3-year analysis of Lake Texoma littoral sites during winter (January–April) only, we concluded that salinity and TN:TP were the primary predictors of P. parvum densities (Hambright et al., 2010). Although all parameters examined in that study were correlated with winter P. parvum densities, a general lack of any significant relationships between P. parvum and environmental factors across the complete data set led to the conclusion that there was a great deal of stochasticity behind P. parvum dynamics in Lake Texoma. Now with an additional 6 years of data, much of the apparent stochasticity has become predictable. As before, but now with the complete data set, including both pelagic and littoral sites, P. parvum presence and blooms are related positively to salinity and negatively to TN:TP (Tables 1 and 2 and Supplementary Tables S1 and S2). Both presence and blooms are also negatively related to site type and temperature, indicating that P. parvum is more likely to be present or to bloom in littoral sites during the winter. While winter P. parvum abundances had scaled positively with both TN and TP (Hambright et al., 2010), in this study, P. parvum presence scaled negatively with both TN and TP, and P. parvum blooms scaled positively with TN, but negatively with TP. These negative relationships with nutrients seem contradictory to the nutrient-bloom paradigm embodied by the general concept of eutrophication. However, further analysis of the data reveals that the negative and positive relationships are due to unimodal relationships between the probabilities of P. parvum presence and P. parvum blooms with TN and TP (Supplementary Figure S2). Probabilities for presence and blooms would be predicted to occur at relatively high TN (1.2–1.7 mg l−1) and TP (100–250 μg l−1) concentrations, with both probabilities declining markedly at lower and higher values.

Additions of chlorophyll, DO, pH and water elevation resulted in improved model fits for both P. parvum presence and P. parvum blooms. Hambright et al. (2010) also noted that the inclusion of chlorophyll improved their multiple regression model of winter P. parvum densities. However, the potential correlation between P. parvum and its primary photosynthetic pigment is uninformative with respect to understanding environmental regulation of P. parvum dynamics. The predictive capability of pH, which has been reported to affect toxicity of P. parvum toxins (Valenti et al., 2010, but see, Cichewicz and Hambright, 2010), and of DO are similarly compromised, as elevated pH and O2 concentrations are indicative of high algal growth rates and blooms. However, pH was not a significant predictor of P. parvum blooms generally (Supplementary Table S2). Previously, we (Hambright et al., 2010) hypothesized that water elevation could be used to predict P. parvum blooms because of a strong negative relationship between salinity in the littoral regions of the lake and water levels, which reflected the balance between inflows and evaporation. Roelke et al. (2011) report a similar role for inflows in three Brazos River system reservoirs (Lakes Possum Kingdom, Granbury and Whitney), with blooms occurring only at low inflows and high salinities. Our present study also reveals relationships between water level elevation and both P. parvum presence and blooms (Supplementary Tables S2 and S3), as well as with salinity (Supplementary Figure S3), but unlike the Brazos system in which inflows and salinities are more tightly coupled, these relationships in Lake Texoma are nonlinear and very noisy, making useful interpretation difficult. For example, the counterintuitive positive relationships between water level and P. parvum presence or blooms detected by logistic regression analyses are driven by unimodal relationships (Supplementary Figure S2), the left-hand side of which resulted from a period of relative drought and lower than normal water levels during seasons when other conditions conducive to P. parvum (e.g., cooler temperatures, higher salinities) were absent. Finally, there has been considerable discussion pertaining to potential negative (via allelopathy; see references in Granéli et al., 2012) and positive (via indirect proliferation of bacterial prey for P. parvum; see references in Granéli et al., 2012) impacts of cyanobacteria on P. parvum. Although we only monitored cyanobacterial abundances during 2009–2012, addition of phycocyanin concentrations (a proxy for cyanobacterial abundances) to our logistic regression models did not provide any improvement in predictive capabilities, even though cyanobacteria are very common in Lake Texoma during much of the year.

As in previous inland studies, we found that P. parvum presence and blooms in Lake Texoma seemed to correlate with non-optimal growth conditions, at least with respect to temperature and salinity (Hambright et al., 2014; Patiño et al., 2014). During 2004–2012, net population growth rates in Lebanon Pool were highest at suboptimal cold temperatures, between 5 °C and 10 °C, as well as suboptimal salinities, ranging between 0.5 and 2.5 psu, during the bloom development phase, compared with laboratory optima of 27 °C and 22 psu measured for a P. parvum strain (UTEX-LB2792; originally cited using the temporary identification LBZZ181, D Nobles, UTEX Culture Collection, University of Texas, personal communication) isolated from the Colorado River, Texas (Baker et al., 2007). Other experiments using UTEX-LB2797 revealed a logarithmic growth–salinity relationship in which maximum potential growth rates of P. parvum declined at salinities of 6 psu or lower compared with 15 and 30 psu (Hambright et al., 2014). Similar findings have been documented for Norwegian and Danish strains of P. parvum (Larsen and Bryant, 1998). Not only were P. parvum growth rates during blooms in Lebanon Pool, Lake Texoma (Oklahoma–Texas) negatively related to temperatures between 7 °C and 16 °C, these blooms were large (up to 200 000 cells per ml) and quite toxic, as indicated by bioassays and large fish kills (Hambright et al., 2014; Zamor et al., 2014). In Hambright et al. (2014), both growth rates and general toxicity in cultures were found to increase with increasing salinity.

We hypothesized that environmental conditions that foster high growth in P. parvum relative to other microbial constituents are the principle determinants of P. parvum blooms in inland freshwater systems. The alternative hypothesis suggests dispersal limitation. Two aspects of our data are supportive of the environment side of the hypothesis: (1) downstream flow in Lake Texoma is repeatedly introducing P. parvum downlake, yet blooms remain restricted to uplake, Red River stretches of the reservoir (Hambright et al., 2014; Zamor et al., 2014, 2) both our logistic regression models (compare R2N, C and Somer’s Dxy) and classification tree analyses (compare misclassification rates) indicate that environmental data were better at predicting P. parvum blooms than P. parvum presence. We interpret this as meaning that presence is more dependent on factors such as detection limit or propagule pressure or both, whereas blooms are explicitly associated with appropriate environmental conditions. Because it is a microbial eukaryote that can occur in immense numbers, encyst and be passively transported, P. parvum’s dispersal capabilities are potentially unlimited (sensu Finlay, 2002; but see Martiny et al., 2006; Hanson et al., 2012) and subjected to probabilities related to various potential vectors, such as wind, migratory-animal-assisted propagation or even human intervention (Johnson et al., 2008). For example, Hambright et al. (2014) estimated that during a P. parvum bloom of 105 cells per ml in Lebanon Pool (volume ∼3 × 106 m3), the total P. parvum population would exceed 3 × 1011 individuals and represent an immense pool for downstream transport of invasive propagules via hydraulic flushing and advective downstream flow. Yet, P. parvum, which is often detected downstream, has yet to bloom in any area of the lake outside the environmental conditions identified here (Zamor et al., 2012). A similar argument can be made for areas downstream of Lake Texoma, as well as many downstream water bodies in the Canadian River watershed, which also experiences P. parvum blooms in upstream systems (Zamor, 2013).

Recently, there has been growing interest focused on biogeographic patterns in the distributions of microorganisms and whether general ecological principles based predominantly on the study of macrobial species also hold for microbes. For example, can microbes be invasive? Much of the discussion has centered on the longstanding credo in microbiology, ‘Everything is everywhere, but, the environment selects’ (Baas-Becking, 1934; de Wit and Bouvier, 2006), which suggests that dispersal is not limiting for microbes and that the environment is the primary factor determining whether a particular microbe is abundant enough to be detected in a given habitat. Indeed, early morphology-based taxonomic studies of protists tended to conclude that cosmopolitan distributions for microbes were the norm (Finlay, 2002; Fenchel and Finlay, 2004). By contrast, recent taxonomic studies using modern molecular genetics (i.e., pyrosequencing) have provided evidence that indeed some microbes seem to exhibit biogeographic patterns unrelated to environmental conditions (Hanson et al., 2012), although the underlying mechanisms which explain these patterns are not yet understood. Interestingly, a more recent study by Gibbons et al. (2013) has provided support for the Baas-Becking hypothesis, as they propose that extremely deep sequencing (i.e., on the order of 1011 sequences) of an individual marine bacterial community would reveal global phylogenetic diversity of the oceans. Within the P. parvum system, a recent phylogenetic analysis (Lutz-Carrillo et al., 2010) also provides evidence that long-distance dispersal in P. parvum is common. There, they found that isolates of P. parvum from populations in the US states of Texas, South Carolina and Wyoming share high levels of similarity in the first internal transcribed spacer in the nuclear ribosomal operon (ITS1) with P. parvum isolates from Scotland; isolates from a single lake in north Texas were similar to isolates from Denmark and Norway; and isolates from Maine were similar to isolates from England. Isolates collected from the US state of Washington were most similar to a clade comprised of isolates from Scotland, Denmark, Norway, Texas, South Carolina and Wyoming, whereas an isolate from Australia was most similar to the England and Maine clade. Thus, if everything is everywhere (or has the potential to be), perceived microbial invasions and range expansions may indeed be reflective only of technical limitations in detection or changing environmental conditions. Alternatively, spatial barriers to dispersal could vary in strength, from being negligible at local and regional scales, to stronger at continental, intercontinental and global scales (Lutz-Carrillo et al., 2010). These contrasting hypotheses should be testable with population genetic surveys.

Classical ecologic theory stipulates that the differences in taxonomic diversity and composition among sites will depend on whether species are excluded from meaningful participation in a particular habitat by local biotic and abiotic conditions or by dispersal limitation, in which a species does not arrive at that particular habitat (Shurin, 2000). Unfortunately, in microbial systems, the probability of type 2 error associated with rejecting dispersal as a driving mechanism is heavily dependent on technological detection limits of the commonly used molecular tools. This problem has led some researchers to suggest that dispersal be redefined as being both arrival to, and successful establishment in, a given habitat, as evidenced by metabolic activity and some level of reproduction (Hanson et al., 2012). Thus, mere presence would not suffice as evidence that dispersal had occurred and presumably the problem associated with detection limits would no longer be an issue. Unfortunately, this definition of dispersal would exclude many microbes that are present in a community but that are not active, biologically relevant participants in the community because they are either extremely rare (Sogin et al., 2006; Caron and Countway, 2009) or dormant (Jones and Lennon, 2010; Gibbons et al., 2013) under the current environmental conditions. However, these microbes could become important community participants if the environment changed to more favorable conditions (Caron and Countway, 2009). Furthermore, the use of active community participation to define dispersal renders ‘Everything is everywhere, but, the environment selects’ untestable. By contrast, for microbial taxa for which technical limitations of detection have been greatly reduced through development of new methodologies, we would argue that differentiation between the roles of dispersal and habitat suitability in microbial biogeographic patterns can and should be addressed experimentally (e.g., Ehrlén and Eriksson, 2000; Shurin, 2000).