Towards defining global ecotypes of the toxic cyanobacterium Raphidiopsis raciborskii

ABSTRACT Raphidiopsis raciborskii is a harmful bloom-forming cyanobacterium with strains that vary in ecophysiology within and between populations around the world. Understanding the extent of intraspecific diversity of strains is needed to design laboratory experiments that capture the breadth of responses the species has to abiotic and biotic interactions; therefore, choice of strains is a critical consideration for experimental design. In this paper, we identified major ecotypes of 12 R. raciborskii strains from three continents, characterizing their morphology via microscopic cell measurements; physiology via growth rates under nitrogen-replete and -free conditions; and genetic variation, via multiplex randomly amplified polymorphic DNA PCR (i.e., DNA fingerprinting). Euclidean distance plots based on morphological and physiological measurements showed three groupings of strains defined as the major ecotypes. Best groupings were obtained using a selection of both morphological and physiological traits. Ecotype groupings did not correlate with geographic location, implying that understanding the ecology of the species does not require in-depth local knowledge of the strains. This study indicates that general average physiology of the global species could be characterized, indicating the existence of major ecotypes across populations.

Intraspecific variation is crucial to the success of a species by providing a variable population-level response to ecological pressures, especially in bloom-forming scenarios (Burford et al., 2014). Plasticity in response to nutrient availability in species widens their realized niche (Berg & Ellers, 2010). For example, R. raciborskii adopts a high level of flexibility to the nutrients nitrogen and phosphorus (Burford, Willis, Chuang, Man, & Orr, 2018). This variable response allows rapid adaptation, provides resilience to changing environmental conditions and also influences interspecies competition. Xiao, Adams, Willis, Burford, and O'Brien (2017b) found the competition outcome to be vastly variable between several Microcystis aeruginosa and R. raciborskii strains. This variable competition and consequent effect on population dynamics of the cyanobacteria were underpinned by specific traits such as growth response to light and temperature. An understanding of these traits and their interactions is fundamental to ecosystem modelling and phylogenetic grouping (Litchman, Klausmeier, Schofield, & Falkowski, 2007).
Species traits are a functional measure of diversity (Litchman et al., 2007). Trait characterization may also inform management control strategies in waterbodies and lakes by describing the niche of dominant species (Mantzouki, Visser, Bormans, & Ibelings, 2016). Functional groupings such as ecotypes allow us to better understand the strategies of a species to perform along environmental gradients and thus which of these ecotypes will be selected (Litchman, de Tezanos Pinot, Klausmeier, Thomas, & Yoshiyama, 2010). Traits of interest include nutrient uptake and storage; for example, a four-fold difference in abundance of heterocysts was found among R. raciborskii strains in nitrogen (N)-free conditions (Willis et al., 2015). In bloom-forming species, differences in toxin cell quota and response to abiotic drivers in co-occurring strains can change the overall bloom toxin yield and thus risk level of the bloom (Burford et al., 2014). To more effectively predict toxin yields and bloom formation, an understanding of population strain diversity and major ecotypes is needed.
To understand a species' niche and the diversity of its responses to abiotic and biotic interactions, it is necessary to use cultured isolates in laboratory experiments; therefore, choice of strains is a critical consideration for experimental design (Xiao et al., 2017b). In this paper, we characterized 12 R. raciborskii strains isolated from Africa, Australia and Europe to determine the major ecotypes based on morphological, physiological and genetic differences. We address the query of how many traits are required to successfully differentiate ecotypes and how combinations of traits affect that outcome. Euclidean distance plots were used to characterize the ecotypes and to determine the minimum characterization required to differentiate strains with relevant physiological traits.

Cultures
Twelve strains of R. raciborskii, one strain of R. mediterranea and one strain of Chrysosporum ovalisporum (Table 1) were acquired from the Australian National Algae Culture Collection (ANACC). Cultures were maintained in MLA media (Bolch & Blackburn, 1996) at 20°C, under a 12:12 h light:dark cycle of 15 µmol photons m −2 s −1 .

Growth experiment
Each strain was grown in triplicate 4 ml cultures in 12well plates (Corning® Costar®, Merck KGaA, Darmstadt, Germany) in both nitrogen-replete (NR) MLA (170 mg l −1 NaNO 3 ) and nitrogen-free (NF) MLA (0 mg l −1 NaNO 3 ) treatments, under the conditions described above. Optical density absorbance at 750 nm (OD 750 ) measured with a plate-reader (Biotek Cytation 1) was used as a proxy for cell concentration. The initial optical density was OD 750 = 0.01, and three weekly OD 750 measurements were made for 12 days. Specific growth rates (GR) were measured during exponential phase using the standard growth equation GR = ln(c2/c1)/t2-t1, where c is the optical density and t is time (Andersen, 2005).

Morphological measurements
On day 12, a 2 ml subsample was preserved with~1% final volume Lugol's solution. Under 100× magnification using a Leitz microscope, the trichome length, cell length and cell width were measured using the Leica Application Suite (v.4.10.0) software, and biovolume was calculated based on the cylinder shape (Hillebrand, Durselen, Kirschtel, Pollingher, & Zohary, 1999). The number of heterocysts per trichome was counted for 30 trichomes for each sample. All strains had straight morphology except for CS-506 which was coiled.

DNA isolation and Hip1 PCR
Total genomic DNA was isolated from a 10 ml subsample of each culture using a DNeasy Blood and Tissue Kit (Qiagen) with a modified lysis step. Cells were collected with centrifugation, 5 min at 4000 rpm, and resuspended in 500 µl lysis buffer (10 mmol l −1 Tris-HCl, 0.1 mmol l −1 EDTA, 1% SDS) with glass beads. Cell mixtures were vortexed for 30 s, Lysozyme (16 µg µl −1 ) added and incubated at 37°C for 20 min. Proteinase K (250 µg) and SDS (0.5% final concentration) were then added and the mixture incubated at 50°C for 2 h. The mixture was then transferred to the DNeasy Blood and Tissue Kit and DNA extracted following the manufacturer's directions. Quality and quantity of DNA was determined with nano-drop and Qubit, following the manufacturer's directions. DNA fingerprinting was performed using the Hip1 TA/TC primers and cycle as previously described (Smith, Parry, Day, & Smith, 1998) using 2.5 ng DNA, 20 pmol primers, with GoTaq Green PCR Master Mix (Promega, Australia). 15 µl of amplification product was run on a 1% agar gel and visualized on an iBright FL1000 Imager (Invitrogen, California, USA). Band density was analysed on the iBright FL1000 Imager using the manufacturer's software.

Statistical analysis and Euclidean cluster analysis
One-way ANOVA, hierarchical cluster analysis and Euclidean distance plots were completed in R (version 3.5.3, R Core Team, 2013). Complete linkage hierarchical clusters were formed from normalized physiological and morphological measurements. The number of clusters for each dendrogram plot was determined via the elbow method using the total within group sum of squares (Thorndike, 1953). All strains were incorporated into four hierarchical clustering analyses with morphological, physiological or mixed traits (Table 2): (1) morphological traitsbiovolume, cells trichome −1 , coiled/straight, trichome length, NF biovolume, NF trichome length, NF cells trichome −1 = 7 traits; (2) physiological traitsheterocyst count (average number of heterocysts on 30 trichomes), NF heterocyst count, NF growth rate, growth rate, toxin presence = 5 traits; (3) morphological traits with NR physiologybiovolume, cells trichome −1 , coiled/straight, trichome length, NF biovolume, NF trichome length, NF cells trichome −1 , growth rate, toxin presence, heterocyst count = 10 traits; (4) morphological traits with NR and NF physiologybiovolume, cells trichome −1 , coiled/straight, trichome length, NF biovolume, NF trichome length, NF cells trichome −1 , growth rate, toxin presence, heterocyst count, NF heterocyst count, NF growth rate = 12 traits. Ecotypes were characterized by a group of strains clustered into dendrograms by the most influential characteristics, and C. ovalisporum CS-1034 was included as a control to ensure interspecific diversity was greater than intraspecific diversity. The most influential traits for each dendrogram were used to determine the minimum number of traits required to differentiate the ecotypes.
For the DNA fingerprinting, a Euclidean distance plot was completed with the same method as described above, using the PCR amplification band presence and density data.

Trait characterization
The 12 traits measured across the 14 strains/species were all found to be significantly different between strains (Table 3). More than half of all strains displayed higher values for biovolume, cell width and cell length traits in NF conditions compared with NR. Traits in NR conditions overall had more variability than the same traits in NF conditions. The largest variation in traits were in NR conditions, with trichome length, biovolume and growth rate varying 4-fold, 4-fold and 6-fold, respectively, between strains. CS-1034 had the smallest cell length, but cell width was above average under both NR and NF conditions. CS-1034 also displayed a cells trichome −1 value that was 12-fold larger than the lowest value. CS-1104 had the second highest heterocyst per trichome count, and CS-1108 had negative growth rates under NF conditions.

Ecotype characterization
Euclidean distance plots displayed a variety of cluster sizes and characterization depending on the traits included; the dominant trait that characterized each ecotype is shown by the dendrogram cluster labels (Figs 1, 2). Using only morphological traits yielded clusters characterized by trichome length and biovolume ( Fig 1A). Grouping using morphological traits also clustered 80% of non-Australian strains into one cluster that was characterized by "High trichome length + Low biovolume". The two remaining ecotypes contained three and four strains; they were clustered by trichome length and biovolume, respectively. The fourth cluster contained C. ovalisporum CS-1034 as a stand-alone strain. Dendrogram 1B showed clustering based on five physiological traits that was most commonly characterized by NF GR (Fig 1B); the cluster sizes were small with the exception of one large group that contained 10 strains from all three geographic regions. C. ovalisporum CS-1034 was clustered into the large ecotype characterized by "High NF GR + Low GR". Clustering using morphological measurements with NR physiology yielded mainly morphological characterizers, such as cells trichome −1 and biovolume (Fig 2A). Clusters were small, with one large grouping of 10 strains of various geographic origins. After NF physiological traits were added (Fig 2B), all three cluster characterizers contained at least one physiological trait or more. Out of six characterizing traits, only one was morphological; the most common characterizer was NF GR. Cells trichome −1 defined two clusters in dendrogram 2A and none in dendrogram 2B. Both dendrograms 2A and 2B differentiated C. ovalisporum CS-1034 from Raphidiopsis sp. and characterized three Raphidiopsis sp. ecotypes in a variety of cluster sizes from 1 to 10 strains. These three ecotypes were defined as the major ecotypes characterized by "High GR + High heterocyst count"; "Non-toxic + Low NF GR"; and "High NF GR + High biovolume".
Of all four dendrograms, morphology and morphology + NR physiology demonstrated the largest clusters containing 10 strains. No single cluster was found to be characterized strongly by toxicity; two clusters were characterized by non-toxicity across two dendrograms.
DNA fingerprinting DNA fingerprinting of the 12 R. raciborskii strains, one R. mediterranea strain CS-1039 and one C. ovalisporum strain CS-1034 was used to differentiate the strains genetically (Fig 3). All strains were differentiated by the number and density of the amplified bands, except for CS-1039 where no bands were amplified. Clustering defined three major groups. Table 3. Trait characterization of Raphidiopsis raciborskii, R. mediterranea and Chrysosporum ovalisporum strains under nitrogen-replete (NR) and nitrogen-free (NF) conditions (mean ± SD, n = 3 (triplicate cultures)). Superscript letters denote significant differences (different letters denote p < 0.05) of within treatment ANOVA with post-hoc Tukey's test. ǂ Toxicity data from ANACC database, toxin = cylindrospermopsin + deoxy-cylindrospermopsin.

Discussion
Three major ecotypes were defined from the 12 R. raciborskii strains characterized, with growth rate, trichome length and biovolume as the most influential traits. The three major ecotypes were characterized by (1) high growth rate and high heterocyst count; (2) nontoxic and low NF growth rate; and (3) high NF growth rate and high biovolume. The strains clustered within each ecotype were not correlated to the geographic origin of the strains, illustrating that the main ecotypes occur across populations. Understanding the major ecotypes will contribute to collaborative efforts across geographic regions to further manage the invasive potential of R. raciborskii with a changing climate. The global distribution and invasiveness of R. raciborskii is facilitated by the combination of physiological plasticity and the co-occurrence of multiple ecotypes with a wide tolerance to different environments (Antunes, Leão, & Vasconcelos, 2015;Padisák, 1997;Sinha et al., 2012). Population heterogeneity plays a crucial role in ecosystem resilience by allowing dynamic responses to environmental change and could underpin the success of this species . Management of this species will require an ecotype-based approach for understanding the environmental niche of the species. Ecotypes in this study were commonly grouped by biovolume. Biovolume or cell size has been dubbed a master trait, because a large number of other physiological traits and ecophysiological interactions depend on it in cyanobacteria (Reynolds, 2006). Biovolume and hence cell surface ratios and diffusion boundary layer impact upon nutrient uptake and growth rates (Litchman et al., 2007). Populations of cyanobacteria with ecotypes defined by small biovolume, for example in low nutrient environments, may perform better than one with a larger biovolume (Reynolds, 2006). Understanding of when these ecotypes are dominant can then inform ecological models and management practices in drinking-water reservoirs.
The inclusion of NR physiology traits was not crucial in characterization of ecotypes but the ecophysiological implications of measuring this trait are clear. Trait variability was highest amongst NR traits, with some exceeding 6-fold differences; however, Bolius, Wiedner, Figure 2. Hierarchical cluster dendrograms of Raphidiopsis raciborskii (12 strains), R. mediterranea CS-1039 and Chrysosporum ovalisporum CS-1034 based on (a) mixed morphology + NR physiology traits (10 traits), (b) mixed morphology + both NR/NF physiology traits (12 traits). Geographic strain origin 1 -Australia, 2 -Africa 3 -Europe. and Weithoff (2017) assessed 13 strains of R. raciborksii and found trait variability was highest in NF conditions. This discrepancy could be accounted for by the close geographic origin of these strains, with all isolated from lakes in north-west Germany. In the same vein,  found single population intraspecific variation using 24 strains of R. raciborskii; significant physiological and morphological differences were found without genetic differences, but the variation was less than that found in the present study. It is likely that these 24 cooccurring strains would be defined as a single ecotype if compared to the 12 strains in this study. This illustrates that within-population diversity will probably be less than the diversity across global populations.
Clustering of strains into ecotypes was significantly impacted by the traits used. Physiological traits were significant influencer traits, particularly growth rate, but morphological traits were required to differentiate C. ovalisporum CS-1034 from Raphidiopsis sp. C. ovalisporum CS-1034 was included to ensure that the traits used were sufficient to differentiate interspecific from intraspecific diversity. The differentiation of C. ovalisporum CS-1034 only occurred when morphological and physiological traits were combined, with morphology or physiology alone insufficient to differentiate interspecific diversity. This demonstrates the importance of combining both physiological and morphological traits for distinction of interspecific strains using Euclidean distance plots.
When values are attributed to traits in a hierarchical clustering algorithm, a binary trait such as "coiled/ straight" trichome morphology is more influential on cluster formation than other continuous traits. To reduce the dominance of binary factors, scaling and normalization were pre-applied to all data, but this method is less effective for binary traits. Weithoff and Beisner (2019) reviewed approaches in trait-based phytoplankton ecology and stated that measurement of continuous explicitly defined traits allows investigation of trade-offs and characterization of ecosystem functioning. To further mitigate the dominant role of binary traits, if available, the input of continuous values would be preferable, e.g., toxin cell quota in place of toxic (Y/N).
The number and combination of traits played an impactful role on cluster formation. The minimum number of traits required to differentiate functional ecotypes was 7; we found that the further addition of more traits served only to change the cluster characterizers according to the most influential traits present. Incorporating all 12 traits led to a more even spread of strains across clusters. At lower trait counts, single large clusters accounted for over 70% of strains. In these larger clusters, non-Australian strains were seen to be clustered but no pattern was observed. Physiological traits had the most influential role on cluster formation; thus, preliminary strain selection needs to include a relevant physiological measurement. This study investigated traits under constant laboratory conditions of light and temperature with the addition of nitrogen-free treatment, which had a major role in defining the ecotype groupings. It has previously been shown that R. raciborskii strains differ in the magnitude of their response to different abiotic factors, such as nitrogen and phosphorus (Willis et al., 2015;Willis, Chuang, Dyhrman, & Burford, 2019), light and temperature (Xiao et al., 2017a) and CO 2 (Willis et al., 2019). These differences in response are what gives R. raciborskii high species plasticity . However, the differences between strains do not necessarily correlate with each other, which makes choosing strains difficult, and means that ecotype groupings are likely to be specific to the conditions that they are tested under. We propose that future experimental designs include a preliminary strain screening under the conditions of interest to create specific ecotype groupings. An example of this would be nitrogen fixation for nitrogen experiments, or phosphorus uptake for phosphorus experiments. Transition from 7 to 12 traits, e.g., the inclusion of additional physiology, was influential but not crucial. By including NF traits, morphological traits such as biovolume became less dominant. The implications for trait selection in choosing strains for intraspecific variation experiments can be seen here; the next step to understanding how two or more traits interact is through ecophysiological modelling, a topic beyond the scope of this study.
DNA fingerprinting was used for genetic differentiation of the strains, by generating unique and identifying DNA profiles of cyanobacteria strains (Saker & Neilan, 2001;Smith et al., 1998). Comparison of PCR amplification band presence and density differentiated all the strains, showing that each is genetically distinct. The clustering of the strains based on the DNA fingerprinting correlated more strongly with geographic distance than the morphology/physiology clustering. There was only very faint amplification of a single band from R. mediterranea CS-1039 DNA (data not shown), indicating that despite the close phylogenomic relationship of this species with R. raciborskii (Stucken et al., 2010) the Hip1 TA/TC primers were not a good match. Comparison of the DNA fingerprinting Euclidean distance plot with the "best" morphological/physiological Euclidean distance plot showed that the clustering is broadly similar across both plots. The DNA fingerprinting also showed four major clusters: with Australian strains in one cluster; European, African, R. mediterranea CS-1039 and Australian CS-1101; and two outliers: R. raciborskii CS-1104 and C. ovalisporum CS-1034. However, R. raciborskii CS-1101 was a stand-alone ecotype in the morphology/physiology Euclidean distance plot, while CS-1104 was stand-alone in the DNA fingerprinting. It is not surprising that the Euclidean distance plots are not an exact match, as the randomly amplified PCR targets will be different to the physiological controls. However, it is useful that the clustering is broadly similar so that DNA-fingerprinting could also be used to differentiate major ecotypes.
Robust assessment of intraspecific variation requires strains isolated across a spatial and temporal spectrum to best understand the species. Furthermore, this raises the question of how many strains are enough? The number of strains to include in a trait-based investigational study has been a limiting issue for research in the past by only including one or two strains (Piccini et al., 2011). More recent studies have included a multitude of strains from multiple locations. Xiao et al. (2017aXiao et al. ( , 2017b) isolated a total of 14 strains of M. aeruginosa and R. raciborskii from two locations. The intraspecific variation was found to be greater than interspecific variation, with significant difference in growth rate ranges. By setting out clear and translational ecotypes, the number of strains used could have been reduced by avoiding similar strains and made more representative by selecting a broader set of ecotypes. In this study, we used 12 strains, with two additional species as outliers. The 12 strains used covered three continents and multiple geographical locations within each, but many more ecotypes probably exist. The spatial scale depends on the experimental design and the questions being asked. For this paper, we used a global perspective on intraspecific variation; three continents and multiple sites within those countries were included. Only three major ecotypes were found and a minimum of seven traits were required to differentiate them. The clustering could be considered too broad, and more sensitive clustering would define four to five ecotypes; this would become important in experimental design and the sensitivity of the ecological question being asked. This study aimed to define major ecotypes, a "broad brush" approach; more ecotypes doubtlessly occur within R. raciborskii populations and these will be defined by the inclusion of more strains in similar studies.
Defining ecotypes helps culture collections and researchers make informed choices about the type of strains to curate and which to include in different experiments. An understanding of intraspecific variation and ecotypes should be included in the background context of experimental design for cyanobacteria, HABs and climate change investigations in the future (Burford et al., 2020). Due to restrictions in time and space, culture collections holding multiple strains may use cluster ecotypes or genotypes to select for specific strains that span the breadth of the species intraspecific diversity and maintaining one or two strains of each ecotype may suffice.