Understanding and predicting harmful algal blooms in a changing climate: A trait‐based framework

The worldwide proliferation of harmful algal blooms (HABs) both in freshwater and marine ecosystems make understanding and predicting their occurrence urgent. Trait‐based approaches, where the focus is on functional traits, have been successful in explaining community structure and dynamics in diverse ecosystems but have not been applied extensively to HABs. The existing trait compilations suggest that HAB taxa differ from non HAB taxa in key traits that determine their responses to major environmental drivers. Multi‐trait comparisons between HAB‐forming and other phytoplankton taxa, as well as within the HAB groups to characterize interspecific and intraspecific differences will help better define ecological niches of different HAB taxa, develop trait‐based mechanistic models, and better identify environmental conditions that would likely lead to HABs. Building databases of HAB traits and using them in diverse statistical and mechanistic models will increase our ability to predict the HAB occurrence, composition, and severity under changing conditions, including the anthropogenic global change.

warming, changes in land use and nutrient inputs (Huisman et al. 2018;Ho et al. 2019). Whether HABs in marine and freshwater ecosystems are increasing globally or only in some regions is debated (Hallegraeff et al. 2021;Wilkinson et al. 2022), but understanding the mechanisms producing HABs is critical. HAB-forming taxa belong to several taxonomic groups, and, according to the global IOC-UNESCO Taxonomic Reference List of Harmful Micro Algae (Moestrup et al. 2009), include at least 89 species of dinoflagellates, 35 cyanobacteria, 29 diatoms, 8 haptophytes, 6 raphidophytes, 3 dictyochophytes, and 2 pelagophytes. Cyanobacteria are the major bloom producers in freshwaters, and dinoflagellates dominate marine HABs, some diatoms and haptophytes also form blooms in coastal regions (Moestrup et al. 2009).
HABs negatively affect water quality and human wellbeing, with tremendous undesirable economic and health consequences (Anderson et al. 2000). Although we know a lot about the ecology of several major HAB taxa in marine and freshwater environments, our ability to predict blooms remains inadequate, especially on longer time scales or under rapidly changing conditions. Even relatively short-term, seasonal predictions of HABs using model ensembles may not work; for example, smaller blooms of shorter duration were predicted for Lake Erie for 2021 than were observed that year (Stumpf 2021). Similarly, correlational models of red tides do not perform well in highly dynamic coastal systems (McGowan et al. 2017). In general, statistical models of HABs mostly work on short-time scales, and process-based models are more suitable for longer-term predictions, especially under changing conditions (Ralston and Moore 2020).
Existing process-based models of HABs are often parameterized inadequately, with parameter values not based on the systematic compilations of many studies but on single studies of individual species (Liu et al. 2019), and thus may not predict the dynamics correctly, if different species or even different strains dominate. Moreover, it is hard to predict if the same toxic cyanobacteria would dominate in the future or they would be replaced by other taxa (Kokocinski et al. 2010; . The existing models focusing on the taxa that cause HABs at present may not be capable of forecasting the potential proliferation of different taxa in the future. There are models that predict blooms of the cyanobacterium Microcystis in Lake Erie (Rowe et al. 2016) but the blooms of other cyanobacteria cannot be predicted by those models, as they are species-focused. Whether the same or different HAB taxa would dominate in the changing climate is an important question because different species differ in their toxicity and the effects on food webs and biogeochemical cycling (Herrero and Flores 2008;Chorus and Welker 2021). Thus, trait differences across different HAB and non HAB taxa should be at the core of our efforts to understand and predict HABs at present and in the future.
We critically need novel frameworks that would increase our predictive capabilities for HABs (Wells et al. 2020). A trait-based perspective where the focus is on functional traits of species may become such a predictive mechanistic framework because it can identify and disentangle the myriad forces, such as abiotic drivers, trophic interactions, etc., structuring phytoplankton communities that include HAB taxa (Edwards et al. 2013a,b). By being species-agnostic, such a framework should be much more capable at predicting the dominance of novel taxa. Such a framework would help transform the ecology of HABs into a more predictive discipline, augmenting existing predictive tools, and should help address numerous fundamental and applied problems associated with HABs.
Surprisingly, despite the increased use of trait-based approaches to explain diverse patterns in ecological communities and species abundance, both in terrestrial and aquatic ecosystems (McGill et al. 2006;Litchman et al. 2007;Funk et al. 2017;Li et al. 2021), such approaches are only now starting to make inroads into the ecology of HABs (Kruk et al. 2017; Van de Waal et al. 2022). Here, I will introduce a trait-based framework and how it can be used to answer major questions about HABs and thus help understand and predict HAB dynamics. I will also discuss the challenges of using traitbased approaches. A trait-based perspective should improve our understanding of the mechanisms that drive HABs and thus benefit the science and management of aquatic ecosystems. The present work is not a comprehensive review of the trait information or the existing models of the HAB taxa, but rather a set of ideas for developing the mechanistic trait-based framework. These ideas should be applicable to HABs in both freshwater and marine environments, enabling cross-system comparisons that may enrich HAB ecology in general.

Overview of trait-based approaches
Using traits to explain community composition has become a mainstay of community ecology, especially for terrestrial plants (Wright et al. 2004;McGill et al. 2006;Lamanna et al. 2014). Major traits that explain variation in community composition have been identified in plants and other organisms (Díaz et al. 2016;Falster et al. 2017;Edwards and Steward 2018). The combinations of different traits and the trade-offs between them can define contrasting ecological strategies, such as fast growers vs good resource competitors, or the gleaner-opportunist continuum (Tilman 1990;Grover 1997).
Traits can be defined as "any morphological, physiological, or phenological heritable feature measurable at the level of the individual, from the cell to the whole organism, without reference to the environment or any other level of organization" (Garnier et al. 2016). Traits that affect fitness are termed functional traits (Violle et al. 2007). Traits can be morphological, physiological, life history, behavioral, as well as genomic and metabolic traits (Litchman and Klausmeier 2008;Litchman et al. 2021;Walker et al. 2022). In microbes, including phytoplankton, genomic and metabolic traits and the trade-offs may be important for inferring ecological strategies

Litchman
Trait-based prediction of HABs (Litchman et al. 2015a). Traits can be grouped according to different organismal functions, such as growth and reproduction, resource acquisition and avoidance of consumers and parasites/pathogens (Litchman and Klausmeier 2008) ( Table 1). Traits can be classified as "response" and "effect" traits (Lavorel and Garnier 2002;Litchman et al. 2015b).
Response traits characterize the responses of an organism to the environment and vice versa, effect traits describe the effects of an organism on the environment. A better characterization of response traits allows using them to understand and predict how organisms respond to changing environments. Some traits can be response and effect traits at the same time.
For example, nitrogen (N) fixation is both a response trait (Nfixation increases under N-limitation) and an effect trait (Nfixation can increase N-availability in an ecosystem).

Relevant traits
The most important traits for predicting the dynamics of HAB taxa are the traits that characterize their responses to abiotic and biotic drivers and align with the major dimensions of their ecological niche. The main abiotic drivers for HAB taxa, as well as other phytoplankton, are nutrients (both macronutrients and micronutrients), light, temperature, and stability of the water column (Anderson et al. 1998;Litchman and Klausmeier 2008;Facey et al. 2019). pH may also be an important factor (Shapiro 1984;Raven et al. 2020). Biotic drivers include competition with other phytoplankton, zooplankton grazing, parasitism by viruses, fungi, and bacteria, and mutualistic interactions with bacteria and other phytoplankton (Work and Havens 2003;Hoke et al. 2021;Papoulis et al. 2021).
Consequently, key traits are those that describe how abiotic and biotic factors affect fitness (Table 1). They include cell size, nutrient-and light-utilization traits, temperatureresponse traits (Fig. 1), susceptibility to grazing, with cell/ colony size often being the proxy for that, buoyancy/motility, the ability to fix atmospheric nitrogen, toxicity, and others (Table 1) (Litchman and Klausmeier 2008;Litchman et al. 2021).
The focus on trait differences among phytoplankton has a rich history. The seminal work of Colin Reynolds in freshwaters (Reynolds 1984(Reynolds , 1997Reynolds et al. 2002), Ramon Margalef in marine environments (Margalef 1978), and others (Smayda 1997;Smayda and Reynolds 2001) described many morphological, physiological and life history traits and their associations in major taxonomic groups of phytoplankton. Although the modern trait-based framework continues this tradition, it became much more quantitative, comparing multiple quantitative traits and their distributions along environmental gradients (Edwards et al. 2012;Thomas et al. 2012;Mousing et al. 2018), similar to such approaches in terrestrial plants (Wright et al. 2004;Butler et al. 2017). Moreover, there is more focus on the eco-physiological traits directly connected to fitness (functional traits) that can be used to parameterize mechanistic models. Morphological traits, which are easier to characterize can sometimes be used to infer the eco-physiological traits (Kruk et al. 2010).
Recently, significant progress has been made in developing trait-based ecology of both marine and freshwater phytoplankton (Litchman and Klausmeier 2008;Schwaderer et al. 2011;Edwards et al. 2013a, Weithoff and Beisner 2019Kruk et al. 2021). One of the major advances is the demonstration that traits can explain diverse occurrence patterns of different phytoplankton. For example, several ecophysiological traits measured in the lab, such as nutrient affinity, light utilization traits, and maximum growth rates could explain more than 80% of temporal variation in abundances of phytoplankton species in a marine ecosystem (Edwards et al. 2013a). Similarly, for freshwater phytoplankton we found that light utilization traits and maximum growth rates predicted how species abundances vary along light and nutrient gradients on a continental scale (Edwards et al. 2013b). The maximum growth rates and nutrient competitive abilities inferred from morphological traits explained seasonal succession in a shallow hypertrophic lake (Segura et al. 2013).
Despite these advances, HAB research rarely applies traitbased approaches, where the information on many traits and their variation across taxa and along environmental gradients is synthesized and used to explain species distributions and community structure. Moreover, trait-based models that ask what trait combinations would be selected for in given environments as providing the highest fitness (Follows et al. 2007;Falster et al. 2017;Klausmeier et al. 2020a) have not been applied to HABs. I propose that such empirical and modeling approaches can be used systematically to better understand the mechanisms that lead to HABs in different environments. I formulate several major questions and discuss how they can be answered using traits. Finally, I propose that a trait-based framework can be used to predict HABs in the future, including under rapidly changing conditions and on longer time scales.

Comparing traits between HAB and non-HAB taxa
One of the major questions is whether and how HAB taxa differ in their eco-physiology from other phytoplankton. Answering this question will help determine the niche differences between HAB and non-HAB taxa and, consequently, the conditions leading to HABs, and requires a quantitative comparison of the relevant eco-physiological traits, with a comprehensive coverage of those groups (Fig. 3). Existing trait compilations allow us to make some inferences, but as far as we know, no extensive comparison that focuses specifically on HAB taxa vs. other groups exists. Here I provide examples of relevant trait comparisons and highlight the gaps.

Freshwater comparisons
Among the freshwater phytoplankton, cyanobacteria, major HAB-producing group, appear to have intermediate growth rates, high phosphorus uptake affinity but low nitrogen uptake affinity, likely influenced by the nitrogen fixers among the cyanobacterial species (Edwards et al. 2012). Cyanobacteria also have high initial slopes of the growthirradiance curves α and low optimum irradiances for growth I opt compared to other freshwater taxa, making them adapted to low light environments (Schwaderer et al. 2011). Interestingly, diatoms also tend to have high α, but also high I opt , which makes them adapted to fluctuating light (Schwaderer et al. 2011).
Freshwater cyanobacteria can regulate their position in the water column through collapsible gas vesicles (Klemer 1983). This provides an advantage in accessing the optimal light and  Litchman and Klausmeier (2008) and Litchman et al. (2021). Some representative references for select traits are also provided. Functions/traits in bold are often included in models of HABs.

Trait type
Function/trait(s) Example traits Genomic Genome size and structure Large (relative or absolute) genomes in some HAB taxa (Kudela et al. 2010;Larsson et al. 2011;Hong et al. 2016 (Reynolds 2006). Interestingly, in marine environments many HAB-forming taxa can also adjust their position in the water, but mostly through active swimming using flagella. However, there are many taxa that possess the ability to regulate their position in the water column, either through regulating buoyancy or active swimming, that do not form HABs.
Freshwater cyanobacteria appear to have high optimum temperatures for growth compared to other groups but this difference may only be significant for temperate strains ). In the tropics, the temperature-related traits are similar across major groups (e.g., cyanobacteria, diatoms and green algae in freshwater have similar T opt ), possibly because of the fewer distinct thermal niches due to small temperature variation ). This dependence of trait differences on the strain isolation latitude may lead to different predictions of how warming would influence freshwater cyanobacterial blooms in temperate vs. tropical environments: rising temperatures may stimulate cyanobacterial blooms in temperate latitudes but not in the tropics, because of the similar temperature traits of different phytoplankton groups in low latitudes. This hypothesis has not been investigated but may be important for assessing the role of climate change in HAB proliferation in different geographic regions. A recent metaanalysis showed that the population dynamics of the bloom-forming cyanobacterium Raphidiopsis (Cylindrospermopsis) raciborskii strongly depended on temperature in temperate regions but only weakly in a tropical lake (Recknagel et al. 2019), thus supporting our trait-based hypothesis. In the tropics, cyanobacterial HABs can occur for prolonged periods of time, as they are less constrained by temperature changes (Zohary and Robarts 1990;Roegner et al. 2020). Tropical strains from major groups can still differ in other traits, hence, changes in other environmental factors may alter bloom severity and duration in the future.

Marine comparisons
In marine environments, dinoflagellates are one of the main taxonomic groups that produce HABs, known as "red tides." Their key functional traits differ significantly from other marine groups. The maximum growth rates of dinoflagellates are the lowest of most other marine phytoplankton groups (Edwards et al. 2012). Their scaled nitrogen and phosphorus uptake affinities are also lower than in other groups, making them inefficient nutrient competitors (Edwards et al. 2012) (Fig. 4C). It is likely that mixotrophy and motility help offset this inefficiency through enabling improved access to nutrients. Dinoflagellates also tend to have low initial slopes α of the growth-irradiance curves, which makes them poor light competitors (Edwards et al. 2015) (Fig. 4C). Interestingly, this is different from the major HAB-forming group in freshwaters, the freshwater cyanobacteria, which are good light competitors. The ability of dinoflagellates to migrate vertically to gain access to optimal light conditions may similarly offset their poor light-competitive abilities.
Similar to the freshwater groups, the differences in temperature optima for growth across major marine groups appear greater in higher latitudes, compared to the tropics ). This may mean that rising temperatures are more likely to preferentially stimulate dinoflagellate blooms in temperate climate, and less so in the tropics.

Multitrait comparisons
Considering the ranges of the observed values for multiple traits simultaneously shows remarkable differences across major taxa in a multitrait space for both marine and freshwater groups (Fig. 4). Note that the cuboids in Fig. 4 result from combining the ranges of individual traits for each taxonomic group and, therefore, illustrate the possible regions of trait space, which may not all be filled by existing species. Due to the trade-offs among traits, some trait combinations may not light) level and the associated traits, the maximum growth rate μ max , the growth affinity (initial slope of the growth-resource curve), and the halfsaturation constant for growth k s . (b) Light-dependent growth curve with photoinhibition, the traits are the maximum growth rate, the light affinity α (slope), the optimum irradiance I opt , above which photoinhibition occurs. (c) Typical thermal performance curve (TPC) (growth dependence on temperature) and the associated traits, the optimum temperature for growth T opt , the minimum and maximum temperatures where growth is non negative, T min and T max , respectively, and the thermal niche width.
be possible, resulting in some parts of the cuboids being empty (Litchman et al. 2007;Edwards et al. 2011). Individual species align along the trade-off surfaces within this volume (Edwards et al. 2011). We found that a three-way trade-off leads to a plane that includes both the marine and freshwater species' traits, so one can envision points representing trait combinations of individual species or strains aligning on such a plane (Edwards et al. 2011). With more data, individual species or strains can be represented in a multitrait space, instead of the individual trait ranges delineated by the cuboids.

Trade-offs and trait inference
Trade-offs among traits (pairwise and higher-dimension) are important in defining diverse ecological strategies. For example, in the compilation of light traits, we can identify taxa that are good light competitors but sensitive to photoinhibition, with high alpha and low I opt = low light-adapted strategy vs. the high light-adapted strategy (Schwaderer et al. 2011). Toxicity also trades off with other traits, such as light competitive abilities (Kardinaal et al. 2007) or the ability to neutralize hydrogen peroxide (H 2 O 2 ), one of the reactive oxygen species (ROS) produced by phytoplankton (Diaz and Plummer 2018;Hellweger et al. 2022). Characterizing tradeoffs is also useful for inferring the missing traits, where the relationships between traits (often representing trade-offs) can be used to derive the unknown trait values when other traits are known.
Currently, many ecologically important traits, such as the resource-dependent or temperature-related traits, are not known even for key HAB species, so some traits need to be inferred. There are several methods to infer missing traits but all of them have limitations (Bruggeman 2011;Edwards et al. 2011;Penone et al. 2014). A better approach is to determine traits directly, through measurements (high amount of  effort), and possibly, infer traits from the environmental data using the artificial intelligence approaches (see "Future directions" section). Different taxa can then be compared using multivariate statistical techniques, and their trait differences visualized in the multidimensional trait space (e.g., nonmetric multidimensional scaling, Chuckran et al. 2021). (log 10 L cell À1 d À1 ), the affinity to use nitrogen and phosphorus, respectively, ranges (trait mean AE 95% CI) taken from Edwards et al. (2012), α is the initial slope of the growth-irradiance curve (μmol quanta À1 m 2 s d À1 ), values show ranges taken from Schwaderer et al. (2011) and the optimum temperature for growth, T opt ( C) (mean AE 95% CI) (from ). (c) Marine groups. The traits are scaled N and P affinity (log 10 L μmol À1 d À1 ), the affinity to use nitrogen and phosphorus, respectively, normalized by cell volume (trait mean AE 95% CI) taken from Edwards et al. (2012) and α (μmol quanta À1 m 2 s d À1 ) is the initial slope of the growth-irradiance curve normalized by temperature and daylength, values (ranges) taken from Edwards et al. (2015). Note the slightly different units for marine and freshwater groups. The circle and points in (a) represent the possible trajectories of trait combinations and specific combinations that may be selected for in different lakes or different seasons. Point 1: Low T opt , α, and P aff may be selected for in the beginning of the growth season, when water temperatures are low, and light and nutrients are high; over the course of the season, the optimal trait combinations change (shown as dark trajectories) and can reach point 2: this combination of traits may be advantageous later in the season, when temperatures are higher, but both P and light availability decline, favoring groups, species, or strains with high T opt , α, and P aff .
This very broad comparison of traits of major taxonomic groups that include HAB taxa in freshwater and marine environments suggests that there are key differences in several important functional traits, resulting in the ecological niche differences along major dimensions, such as nutrients, light, and temperature (Fig. 4A,B). Тhe regions of the trait space typical for major taxonomic groups differ, both in the freshwater and marine taxa, especially when you consider multiple traits simultaneously (Fig. 4).
In addition to traits that determine the responses of different taxa to abiotic environment, traits describing the relationships with other organisms, such as grazers, parasites, or mutualists, affect fitness and need to be compared across taxa Lürling 2021). A general trend is that HAB-forming taxa often have traits that make them resistant to grazing, through large cell size, coloniality, resting stage formation, and toxicity (Panči c and Kiørboe 2018; Lürling 2021). Viruses seem to readily infect marine and freshwater HAB species (John et al. 2002;Baudoux and Brussaard 2005;Morimoto et al. 2019;Pound and Wilhelm 2021), but tend to be more taxon-specific than grazers. It is possible that in a multidimensional (>3) trait space, that include traits for grazing or viral resistance (e.g., cell size, coloniality, toxicity), the differences between HAB and non-HAB taxa may be even more pronounced.
A better characterization of trait differences between HAB and non-HAB taxa in a multitrait space should help determine environmental conditions that favor HAB taxa (Fig. 3). The multitrait information should also help identify the trade-offs among traits, which define different strategies of HAB taxa and could be used to reduce the dimensionality of trait-based models (Litchman et al. 2015a;Ehrlich et al. 2020).
Trait differences between HAB-and non-HAB taxa are also relevant for detecting HABs remotely and understanding the dynamics of HABs. For example, not only chlorophyll fluorescence, but phycocyanin (pigment present in many cyanobacteria) and other pigment fluorescence is being actively used to estimate cyanobacterial abundance in water bodies (Simis et al. 2005;Wang et al. 2016). Monitoring HAB toxins or genes associated with toxin production (e.g., domoic acid synthesis associated genes dab, Brunson et al. 2018) is also informative.
I discussed how traits differ across the main taxonomic groups, some of which contain HAB-forming species. A related but largely unexplored question is whether HAB-producing species differ significantly in their traits from other species within the same taxonomic group that are not known to produce blooms. In cyanobacteria, just a handful of species and genera (e.g., Microcystis aeruginosa, Dolichospermum flos-aquae, Nodularia spumigena, etc.) are responsible for most of the HAB events globally. Many other cyanobacteria (e.g., Chroococcus, Synechococcus, Coelosphaerium, etc.) are not known to form toxic/harmful blooms. In marine environments, among dinoflagellates and diatoms, only some species are known to produce HABs (genera Alexandrium, Karenia, Pseudo-nitzschia, etc.). A comparison of the traits between HAB-and non-HAB taxa within the taxonomic groups may help explain why only some of the species produce blooms. HAB-forming taxa may have larger ranges of key traits which may make them better adapted to different conditions (Brandenburg et al. 2018). Many studies focus more on HAB taxa, so there is insufficient knowledge of trait values of the non HAB-producing taxa.

Trait differences across HAB taxa
Although HAB-forming taxonomic groups appear to have trait values that differ from the non-HAB taxa, they are not monolithic groups (Fig. 3). There is a considerable variation in key traits among HAB species within the taxonomic groups. Characterizing these trait differences is important for our understanding of why and when different species bloom. For example, several species/genera form most freshwater cyanobacterial HABs, with probably the most common of them being Dolichospermum (Anabaena) spp., Aphanizomenon flos-aquae and M. aeruginosa, sometimes called "Annie, Fannie and Mike". In Lake Mendota, Wisconsin, these three species bloom in different times of year. Aphanizomenon and Dolichospermum bloom earlier than Microcystis (Brock 1985). This may be due to differences in temperature optima among species (Konopka and Brock 1978). There is, however, significant interannual variability in bloom patterns where either a single HAB species dominates for most of the summer or the bloom does not develop at all (Brock 1985). This variability in the dominance patterns may be due to the interactions of various environmental drivers with the corresponding traits of different taxa.
The Laurentian Great Lakes also differ in the identity of their dominant HAB taxa: Lake Erie often experiences massive blooms of M. aeruginosa (Steffen et al. 2017), while in Lake Superior, Dolichospermum is the dominant HAB genus so far (Sterner et al. 2020). These differences may be associated with the differences in temperature, nutrients (L. Superior is colder and oligotrophic) or some other factors that currently are unidentified. More information on traits of these species and lakespecific strains should help elucidate the reasons. Even within a lake, HABs may consist of different species across space. In Sandusky Bay of Lake Erie, the blooms often consist of Planktothrix agardhii and not Microcystis (Hampel et al. 2019).
Other common bloom-forming freshwater cyanobacteria include Planktothrix spp., Raphidiopsis sp., and sometimes Gloeotrichia echinulata. Some of these genera/species are nitrogen-fixers, and the ability to fix nitrogen (N) is an important trait that can lead to different effects of environmental factors, nitrogen concentration in particular, on species abundance.
The ability to fix nitrogen is a key trait that can help understand what species can form blooms under different conditions. The seminal work by Val Smith highlighted the importance of N fixation in freshwater cyanobacterial HABs and used this trait to predict blooms, suggesting that low N : P ratios should lead to N-fixer blooms (Smith 1983). However, the blooms of N-fixers were also documented in lakes with high N : P ratios (Noges et al. 2008). Other factors, besides the N : P ratios, may play a role in determining N-fixer dominance. For example, light levels interact with N : P ratios to determine the N-fixer abundance: low N : P ratios stimulated N-fixers only at high light (De Tezanos Pinto and Litchman 2010). High light also stimulated blooms of the N-fixing N. spumigena in the Baltic Sea, and decreased light availability led to bloom termination (Wasmund 1997). Therefore, including other traits, such as light-related traits, susceptibility to grazing or temperature-related traits would allow a more nuanced view of the N-fixer blooms and could improve our predictions.
It is likely that differences in eco-physiological traits among different HAB-forming genera and species lead to their dominance under different conditions, both at present and in the future. For example, one possible explanation for the dominance of Planktothrix in Sandusky Bay of Lake Erie is its lower half-saturation constant for ammonium uptake and better competitive abilities for ammonium (Hampel et al. 2019). A study of cyanobacterial abundance in Swedish lakes showed that different environmental drivers, such as dissolved organic matter (DOM), pH, and N : P ratios stimulated the growth of different cyanobacterial genera, for example, increasing DOM stimulated Dolichospermum but suppressed Chroococcus (Freeman et al. 2020), likely due to their differences in traits.
Significant interspecific trait differences have also been reported across the marine HAB taxa. Within the genus Alexandrium, there is a significant variation in toxicity, with broad latitudinal trends (decreasing toxicity from north to south in the North Atlantic, Anderson et al. 1994). Three genera of HAB-producing dinoflagellates from the Gulf of Maine responded differently to increasing temperature and acidification: the growth rate of Alexandrium catenella decreased, while Scrippsiella sp. and Amphidinium carerae's growth rates increased under the projected warming and lower pH, suggesting that the taxonomic composition of HABs may shift in the future (Seto et al. 2019).
Therefore, a better knowledge of the relevant traits and their comparison across taxa in a multitrait space can help explain or even predict what HAB taxa would produce blooms. Comparing the traits of different bloom-forming taxa, as well as contrasting them with the nonbloom-forming species may help answer a general ecological question of why some species have the ability to reach extremely high densities and dominate communities.

Intraspecific trait differences in different HAB species
Although it is clear that common species and genera of HAB phytoplankton differ in their traits, there are also significant differences in traits within the HAB species (Schaeffer et al. 2007; Willis et al. 2016;Dick et al. 2021), which is similar to other groups of phytoplankton and other organisms (Rynearson and Armbrust 2005;Violle et al. 2012). Characterizing the intraspecific variation in traits is important for several reasons (Fig. 3). First, the ecophysiological trait diversity within a species is often due to the presence of multiple genotypes in a population, and this genotypic and phenotypic diversity allows species to potentially adapt to changing environmental conditions, including the anthropogenic global change. The genotypes with higher fitness under novel conditions would be selected for and the population would experience evolutionary rescue (Bell and Gonzalez 2009;Klausmeier et al. 2020b). The resulting trait values in a population would change compared to the initial values, thus leading to evolutionary adaptation. Determining the degree of trait variation within a population or a species can thus help assess the evolutionary potential of HAB taxa. For example, a comparison of thermal curves in Microcystis showed that although many strains isolated from Michigan lakes had temperature optima for growth below 30 C, there was also a strain that had a higher T opt , suggesting that there is relevant genotypic and phenotypic diversity that may allow this species to maintain high growth rates under warming conditions .
Second, assessing the intraspecific variation in traits will also inform us about the width of the environmental niches of HAB taxa, as environmental tolerances may widen if there are genotypes with different trait values; for example, considerable intraspecific variation in T opt could allow species to persist under a wide range of temperatures (Fig. 2, . A survey of photosynthetic traits in multiple strains of the toxic dinoflagellate Karenia brevis revealed low light-and high light-adapted strategies, suggesting that the species can survive in different light conditions (Schaeffer et al. 2007).
At the same time, characterizing and comparing intraspecific trait variation in different environments allows us to determine the degree of local adaptation (Kawecki and Ebert 2004). There are few studies exploring local trait adaptation in HAB taxa, but the existing evidence suggests that it may be common in both marine and freshwater environments. The toxic dinoflagellate Alexandrium ostenfeldii showed a significant genetic differentiation among different geographic locations, suggesting local adaptation (Tahvanainen et al. 2012;Brandenburg et al. 2021). Similarly, the freshwater cyanobacterium Microcystis showed genetic divergence across lakes in Michigan (Wilson et al. 2005). An invasive cyanobacterium R. (C.) raciborskii may be adapted to local temperatures, as the strain from the Midwest had lower T opt than the Florida strains ( Fig. 2)  . Studies that investigate the geographic patterns in both genetic and trait diversity would be especially illuminating.
A better characterization of the traits of different genotypes and trade-offs between traits within a species or a population can help predict possible changes in toxicity. Strains of Microcystis exhibited a trade-off between competitive ability for light and toxicity (Kardinaal et al. 2007). Consequently, with the progression of the Microcystis bloom and decreasing light availability, less toxic strains with greater light competitive ability replaced more toxic strains that were poor light competitors, thus decreasing the bloom toxicity (Kardinaal et al. 2007). The knowledge of the trait variation for multiple strains/genotypes within a species may help assess the outcomes of competitive interactions between different HAB genera, as the outcome of competition between Microcystis and Cylindrospermopsis depended on what strains were involved (Xiao et al. 2017).
Intraspecific trait variation arises either due to trait differences across genotypes or due to physiological plasticity (Violle et al. 2012). Many studies have documented plasticity of many traits in marine and freshwater HAB taxa, such as CO 2 uptake, degree of coloniality, inducible defenses, toxicity, nitrogen fixation, and so on (De Tezanos Pinto and Litchman 2010; Wohlrab et al. 2017;Ji et al. 2020;Lürling 2021). Different genotypes may have different degrees of phenotypic plasticity, as in a toxic dinoflagellate A. ostenfeldii (Brandenburg et al. 2021). Comparing the degree of phenotypic plasticity within (and across) genotypes to the trait variation across genotypes would help determine if the fitness under future conditions would be influenced more by evolutionary adaptation or by acclimation.
Information on the degree of intraspecific trait variation in HAB taxa is crucial for explaining the ecological success of major toxic bloom-forming species in marine and freshwater environments. Studies on key HAB taxa, such as the marine dinoflagellate Alexandrium spp., freshwater cyanobacterium Microcystis spp., and others show considerable genotypic and phenotypic diversity, which may be one of the reasons for their global distribution and frequent blooms (Anderson et al. 2012;Brandenburg et al. 2018;Dick et al. 2021). Interestingly, the degree of intraspecific trait variation can differ across populations from different geographic locations, such as in the toxic marine dinoflagellate A. ostenfeldii isolated from the Baltic Sea vs. the Netherlands (Brandenburg et al. 2018). It is unknown whether the trait/phenotypic diversity is considerably greater in HAB-producing vs. non-HAB taxa.

Predicting HABs using traits
Surprisingly, using trait-based approaches to predict the dynamics of species and communities in the future is much less developed compared to trait-based explanations of the observed past and present community composition. Incorporating trait-based approaches into predictions should help increase the reliability of predictions and extend their time horizon.
Predictive models for HABs, as for many other ecological phenomena, can be divided into statistical (data-driven) and . However, including the temperature trait information may lead to a change in the relationship (red dashed line) and predict a decrease in biomass because of the decline in growth rate of that cyanobacterium with increasing temperature. (b) The common focus on the individual HAB species (e.g., Sp. 1) may erroneously predict a decline in biomass when temperature increases above the T opt for Sp. 1. However, depending on what species are present, the cyanobacterial biomass may still increase with increasing temperature or decline. The temperature where the decline may happen will depend on the temperature traits of the HAB species in the community.
process-based (mechanistic) models (Ralston and Moore 2020). Statistical models may work reasonably well, especially for the shorter time scale predictions, but may fail on longer time scales, or when the conditions change rapidly and create novel combinations or ranges of environmental drivers, which is likely to happen given the anthropogenic global change. Process-based models that include the physiological dependence of growth and mortality of HAB taxa on environmental factors may predict the dynamics of HABs better (Ralston and Moore 2020;Wang et al. 2022). Adequately parameterizing such models is, however, difficult, as is including all the relevant processes without overly complicating the models.
There are at least three ways in which a trait-based framework can be used to improve predictions. First, the relevant traits can be included as predictors in statistical models, in addition to environmental variables (Fig. 5). Some models that predict species abundances and distributions started including traits as predictors, which improved model performance (Regos et al. 2019;Vesk et al. 2021). Edwards et al. (2013a) showed that several eco-physiological traits of phytoplankton, namely the affinity for light and a nutrient (nitrate), and the maximum growth rate were significant predictors of the species' abundance changes in response to seasonally changing environmental conditions in a marine ecosystem, explaining up to 80% of the variance in abundance. Statistical models that include both the environmental factors and functional traits as predictors may work better than the models that only include environmental variables. The multilevel (mixed, hierarchical) statistical models such as the one used by Edwards et al. (2013a) and others (Thorson and Minto 2014) can include variation between species due to environmental factors, variation between species not explained by environmental factors and variation not explained by species identity or environmental factors.
Second, empirical trait information can be used to systematically parameterize process-based, mechanistic models of HAB dynamics. Choosing the parameter values and their ranges based on multiple studies helps increase the reliability of parameterizations and guide sensitivity analyses where different parameter values are explored. We have previously used an extensive trait compilation to parameterize a dynamic model of the phytoplankton community composition in the present and future ocean and attempted to identify traits that were the most influential for the predictions (Litchman et al. 2006). Similar approaches can be used to parameterize the phytoplankton community models that include HAB taxa and to determine what traits are the most influential for predicting HAB dynamics under different conditions and thus require more thorough measurements.
Third, we can use trait-based models to augment the existing dynamical (e.g., differential equations-based models of population density dynamics) or statistical models, which should improve our predictions of the blooms on longer time scales and in novel conditions, and would allow the possibility for different HAB taxa to become dominant. Strictly speaking, trait-based models are models that predict what traits and trait combinations would be selected for under different conditions and are, thus, species/taxon agnostic (Falster et al. 2017;Klausmeier et al. 2020a). Therefore, contrary to the popular misconception, the models that include traits of species, as many dynamical models that have species or functional groups do, are taxon-based, not trait-based. Such commonly used species or functional group models are useful as well, especially when parameterized based on reliable empirical trait data and when they include relevant mechanisms, such as resource-dependent growth, mortality from grazing and other sources, and the effects of environmental drivers (temperature, pH, physical mixing, etc.). However, they may not adequately predict the rise of novel HAB taxa not included in the models.
Combining different types of models may be advantageous and provide better predictions of HABs/water quality under changing conditions. Such hybrid approaches may include agent-based modeling of the dynamics of toxin genes and more standard hydrodynamic models (Hellweger et al. 2022) or statistical models of oxygen concentration with varying coefficients (not the HAB dynamics per se) and a hydrodynamic model (Deyle et al. 2022). Linking trait-based models or species-based models parameterized using extensive empirical trait data with hydrodynamic models forced by the present or future climate scenarios may provide a robust way to predict when and what HAB taxa would proliferate.
Trait-based models can inform us what trait values would lead to high fitness (often approximated by growth rate) in different environments (Litchman and Klausmeier 2008). For the models that focus on HABs, relevant traits include traits that determine the growth dependence on environmental factors, both abiotic and biotic (e.g., grazing or parasitism) ( Table 1). The models could predict trait value combinations in a multidimensional trait space that lead to the highest fitness under given conditions. Consequently, the taxa (groups, species, or strains) that have the combinations of traits closest to the predicted combinations would likely be selected for under given environmental conditions. Then matching the predicted trait combinations with existing trait data for different species could help identify what species would likely dominate (Fig. 4).
Mechanistic trait-based models may predict a particular combination of traits that results in the highest fitness under given environmental conditions. It is possible that some range of trait combinations may have similar fitness, leading to the possibility of several species co-occurring, especially if species with lower fitness are supported by immigration or changing selection regimes (Koffel et al. 2022). Theoretical models may also predict more than one optimal trait value or trait combination (Litchman et al. 2009;Kremer and Klausmeier 2017), which may make coexistence of different ecological strategies, possibly represented by different taxonomic groups, likely.
One relevant trait-based modeling approach that was used successfully to explain different community composition of phytoplankton in the world ocean was to "seed" a community with species/phenotypes that have different trait combinations (e.g., from low-to high-light adapted, from low to high maximum growth rates, T opt , etc.) and simulate community dynamics under diverse environmental conditions to determine what trait combinations win under given conditions (Follows et al. 2007). Those predicted trait values can then be matched with the empirical trait distributions for different phytoplankton, including HAB taxa to determine the probability of a bloom.
The advantage of such models that explore all the possible trait space and determine trait combinations associated with high fitness under different environmental conditions is that they are not constrained by what species/taxa they model and, thus, allow new species to become dominant. For example, a model that only includes Microcystis sp. would not be able to predict the development of a Planktothrix or Dolichospermum bloom if the conditions change, while traitbased models could. Trait-based models should be especially useful for predicting HABs under rapidly changing conditions or on longer time scales, when an expansion of different species may be more likely.

Future directions
Combining the data-driven and modeling trait-based approaches to understand and predict HABs in freshwater and marine environments is a promising area of research. There are some key directions that should speed up progress. First, we need more information on trait values of different HAB taxa to compare them systematically with the traits of non-HAB taxa and to parameterize mechanistic models. The traits that describe the growth dependence of HAB taxa on environmental (abiotic and biotic) drivers (Table 1) are especially relevant. In addition to traditional ways of measuring traits (laboratory experiments), we need to develop new approaches, such as inferring traits that can be used in predictive models from the 'omic (genomic, transcriptomic, proteomic, and metabolomic) data (Hennon and Dyhrman 2020;Strzepek et al. 2022) or the environmental distribution data using machine learning or similar techniques (Newman and Furbank 2021). Using genomic information to predict the toxic potential of the blooms is widespread (Janse et al. 2004;Brunson et al. 2018;Panksep et al. 2020), but inferring other traits such as growth rates or nutrient uptake is much more complicated. The 'omic approaches are being increasingly used to characterize the responses of different HAB species and strains to environmental conditions, which may aid HAB predictions (Hennon and Dyhrman 2020). Combining 'omics data with trait information for many species may allow identifying the 'omic signatures for some traits and help derive the missing traits. In parallel, species responses to the environment and the associated traits can also be derived from the environmental time series using machine learning or similar computational approaches, but such potential remains underexplored (Thomas et al. unpublished).
Compared to the traits related to abiotic drivers, traits that characterize biotic interactions, such as predation, parasitism or mutualism, are less frequently characterized , and should be one of the research foci in HAB ecology. For example, little is known about what traits may protect from viral attacks or other parasites, and if there are trade-offs between the defense and other traits, such as nutrient uptake capabilities (Menge and Weitz 2009;Litchman et al. 2021).
Characterizing the multitrait surfaces for HAB and non-HAB taxa would allow us to understand whether the given environmental conditions (combinations of nutrients, temperature, light, water mixing, etc.) would lead to blooms of harmful algae. At the same time, the knowledge of traits within HAB taxa, across different species, would allow us to determine what species would be most likely to bloom. Moreover, better characterizing the intraspecific variation in traits is essential for predicting the ability of different HAB taxa to adapt to changing environmental conditions, including the anthropogenic global change. All this trait information should be compiled into a HAB trait database, similar to the trait databases for other organisms, such as terrestrial plants and fish (Angermeier and Frimpong 2011;Kattge et al. 2020). Such databases enabled numerous fundamental advances in community ecology and would be essential for HAB research.
Although at least some trait data for key planktonic HAB taxa exist, trait information on the benthic HABs is lacking. Benthic toxic cyanobacteria proliferate in many water bodies, especially running waters, but the measurements of their traits are scant (Wood et al. 2020). Future trait work should increase focus on benthic forms. In addition, many planktonic HAB taxa can be found in benthic habitats, for example, as resting stages (Ståhl-Delbanco et al. 2003;Ellegaard and Ribeiro 2018). Characterizing HAB traits during the benthic stage is necessary to understand different parts of the life cycle of HAB taxa.
Another direction is developing mechanistic models that include key traits and can predict trait selection under different future scenarios (mixing, nutrient levels, temperature, etc.). Such models should include not only the abiotic but biotic controls as well, such as grazing, parasitism, and mutualism. Parameterizing these models using extensive trait databases will increase model realism. The models should then be tested with different types of empirical data, with different temporal (e.g., from weekly to multiyear) and spatial (individual lakes vs. regional or global) resolution.