Deciphering the effects of nitrogen, phosphorus, and temperature on cyanobacterial bloom intensification, diversity, and toxicity in western Lake Erie

Although cyanobacterial harmful algal blooms (CHABs) are promoted by nutrient loading and elevated temperatures, the effects of these processes on bloom diversity are unclear. This study used traditional and next‐generation sequencing approaches to assess shifts in phytoplankton, cyanobacterial (16S rRNA), and microcystin‐producing (mcyE) communities during CHABs in western Lake Erie (Maumee and Sandusky Bays) in response to natural and experimental gradients of nitrogen (N), phosphorus (P), and temperature. CHABs were most intense near the Maumee and Sandusky Rivers and were dominated by Microcystis and Planktothrix, respectively. Sequencing of 16S amplicons revealed cryptic cyanobacterial diversity (47 genera) including high abundances of two distinct clades of Synechococcus in both bays and significant differences in community structure between nutrient‐rich nearshore sites and less eutrophic offshore sites. Sequencing of mcyE genes revealed low taxonomic (n = 3) but high genetic diversity (n = 807), with toxigenic strains of Planktothrix being more abundant than Microcystis and more closely paralleling microcystin concentrations. Cyanobacterial abundance significantly increased in response to elevated N, with the greatest increases in combined high N, P, and temperature treatments that concurrently suppressed green and brown algae. N significantly increased microcystin concentrations and the relative abundance of nondiazotrophic genera such as Planktothrix, while diazotrophic genera such as Dolichospermum and Aphanizomenon were, in some cases, enhanced by high P and temperature. While nutrients and elevated temperatures promote CHABs, differing combinations selectively promote individual cyanobacterial genera and strains, indicating management of both N and P will be required to control all cyanobacteria in Lake Erie, particularly as lake temperatures rise.

Since the mid-1990s, Lake Erie has become increasingly prone to the occurrence of cyanobacterial harmful algae blooms (CHABs; Conroy and Culver 2005;Bridgeman et al. 2013;Steffen et al. 2014;Watson et al. 2016). This intensification has been linked to accelerated anthropogenic nutrient loading (Carpenter et al. 1998;Paerl et al. 2001;O'Neil et al. 2012) and climatic warming (Heisler et al. 2008;Paerl and Huisman 2008;Carey et al. 2012) and can result in ecosystem disruption (Chorus and Bartram 1999;Paerl et al. 2001). Further, CHABs are a growing public health concern due to their ability to produce an array of toxins, the most well-studied being the microcystins, which are produced by a variety of genera and have been implicated in animal deaths and human poisoning events (Chorus and Bartram 1999;Carmichael 2001;Carmichael and Boyer 2016). Hence, CHABs can jeopardize the sustainability of socioeconomically important freshwater resources (Fuller et al. 2002).
Lake Erie, and in particular its western basin, is the most susceptible region in the Great Lakes to CHABs (Makarewicz and Bertram 1991) as it is the smallest, shallowest, and most southern, and therefore warmest and most highly influenced by nutrient enrichment (De Pinto et al. 1986;Mortimer 1987;Davis et al. 2015;Carmichael and Boyer 2016). The western basin receives the majority of Lake Erie's nutrient load (bioavailable P and N) via the Maumee and Sandusky Rivers (Dolan and Chapra 2012;Han et al. 2012;Allinger and Reavie 2013), which deliver nonpoint runoff from highly agricultural watersheds (Ohio EPA 2010). Nutrient concentrations near these river mouths can be four-times greater than in the open waters of the western basin (Chaffin et al. 2011;Bridgeman et al. 2012) and are believed to fuel the annual late summer blooms in Maumee and Sandusky Bays (Bullerjahn et al. 2016).
Traditionally, primary production in freshwater systems has been viewed as phosphorus (P) limited (Vollenweider 1968;Edmondson 1970;Schindler 1977), with increases in P loads resulting in shifts toward cyanobacteria-dominated communities as many bloom-forming cyanobacteria (i.e., Dolichospermum and Aphanizomenon) are diazotrophic and can fix atmospheric nitrogen (N) to meet their N demands (Smith 1983;Paerl et al. 2001;Schindler et al. 2008). In Lake Erie, external P loads have been successfully used to predict the interannual intensity of CHABs (Stumpf et al. 2012;Michalak et al. 2013) and P reduction strategies implemented in the 1970s significantly reduced cyanobacteria biomass for a 20-yr period (Makarewicz and Bertram 1991;Makarewicz 1993;Stumpf et al. 2012). In the past decade, the importance of N for CHABs has been noted due to the increasing dominance of nondiazotrophic genera (Paerl and Fulton 2006;Harke et al. 2016), such as Microcystis and Planktothrix, concurrent with global increases in N loading (Vitousek et al. 1997;Galloway et al. 2004;Glibert et al. 2014). Furthermore, N availability can influence the production and congener composition of the N-rich toxin, microcystin, in Microcystis and Planktothrix, thereby impacting bloom toxicity (Davis et al. 2010;Gobler et al. 2016;Chaffin et al. 2018). This has spurred a growing emphasis on the occurrence of N limitation (Conley et al. 2009;Gobler et al. 2016), as well as N and P colimitation (Elser et al. 2007;Sterner 2008;Davis et al. 2015;Müller and Mitrovic 2015) in CHAB development, although there still remains a debate regarding the most effective and costefficient mitigation strategies . Additionaly, given that global temperatues are expected to rise an additional 1.5-5 C this century (Houghton et al. 2001;Paerl and Huisman 2009;O'Neil et al. 2012) and the high thermal optima of many freshwater cyanobacteria (≥ 25 C; Paerl et al. 2011;Thomas and Litchman 2016), CHABs are expected to intensify with rising surface water temperatures. Recent longterm monitoring and modeling studies have alluded to the synergistic effects of temperature elevation and eutrophication on cyanobacterial growth (Anneville et al. 2005;Elliott et al. 2006;Kosten et al. 2012), suggesting that eutrophic systems are likely to become more prone to blooms as temperatures continue to rise (Rigosi et al. 2014), supporting the need for a deeper understanding of these complex interactions.
Although numerous studies have examined the effects of nutrient availability and temperature on the growth and toxicity of common bloom-forming genera and total cyanobacteria biomass (Rinta-Kanto et al. 2009;Davis et al. 2010;Chaffin et al. 2011), there remains a limited understanding of their impacts on cyanobacteria assemblage diversity and the associated microcystin-producing communities during blooms (Rinta-Kanto and Wilhelm 2006; Steffen et al. 2012). To gain further insight into the relationships among phytoplankton community dynamics, microcystin production, and environmental drivers, we used next-generation amplicon sequencing (NGS) to evaluate changes in western Lake Erie cyanobacterial assemblages in response to natural and experimental gradients in nutrients (N and P) and temperature. Algal communities were profiled using a multitiered approach to identify variations in the composition and abundance of the total phytoplankton (cyanobacteria and green and brown algae), cyanobacterial (16S rRNA), and potential microcystin-producing (mcyE) communities during field surveys and experiments. The goal of the study was to identify environmental conditions associated with the dominance of specific algal and cyanobacterial communities in western Lake Erie.

Study sites and field surveys
During August 2015, large CHABs developed in Sandusky and Maumee Bays within western Lake Erie, and persisted into September (National Oceanic and Atmospheric Administration 2015b). Sampling transects of the blooms were conducted across each bay during mid-September 2015 aboard the RV Erie Monitor (The Ohio State University), spanning from the respective river mouths into the open waters of the western basin (Fig. 1a). The Sandusky transect consisted of six sites, with two sites each in the inner (western) and outer (eastern) sections of the bay divided by its hourglass shape, as well as two sites beyond Cedar Point, which separates the bay from the western basin of Lake Erie ( Fig. 1a; Supporting Information Table S1). The Maumee transect consisted of five sites extending from the Maumee River to the Western Sister Islands located in the central portion of the western basin ( Fig. 1a; Supporting Information Table S1). Sites were selected to represent a gradient of nutrient and temperature conditions across a range of cyanobacteria densities approximated from the 15 September 2015 MODIS satellite image ( Fig. 1b; National Oceanic and Atmospheric Administration 2015a). Levels of cyanobacteria were confirmed on-site with a BBE Moldenke Fluoroprobe which estimates abundances of cyanobacteria, green algae, and brown algae (e.g., diatoms, dinoflagellates, raphidophytes, and haptophytes) based on differential fluorescence of photosynthetic accessory pigments (Beutler et al. 2002;Chaffin et al. 2013;. Pigment-based class identification was affirmed with 45 cultures of diatoms, cyanobacteria, dinoflagellates, raphidophytes, green algae, and haptophytes that revealed no cross over among the three fluorescence channels for these cultures. At each site, surface water temperature was measured using a handheld YSI sonde (model 556) and phytoplankton communities in the mixed surface layer (0.1-1 m) were assessed with a Fluoroprobe. One liter of subsurface water (~0.25 m) was collected from which triplicate samples were obtained for DNA-based community analysis of the cyanobacteria (16S) and microcystin-producing (mcyE) communities by filtering 30 mL of whole water on 0.22 μm polycarbonate filters. Filters were immediately frozen in liquid nitrogen and stored at −80 C until DNA extraction (see DNA Isolation, Sequencing, and Analysis section below). Additionally, whole water samples were preserved with Lugol's iodine solution (5% v/v) for cyanobacteria identification and enumeration by light microscopy (see Sample Analysis section below), and subsamples were collected are 5 and 10-m depth, dots indicate sampling sites, and hexagons indicate major cities. The insert shows the location of Lake Erie within the Great Lakes region. (b) Composite of MODIS cyanobacterial index image for 14 September 2015 from the NOAA experimental Lake Erie harmful algal bloom bulletin (https://www.glerl.noaa.gov/res/HABs_and_Hypoxia/bulletin.html), where black represents no cyanobacteria detected and the colored pixels indicate cyanobacteria density. and stored at −20 C until further processing for analysis of total and dissolved (combusted EMD Millipore APFB glass fiber filter) nutrients and total microcystins (whole water). Photosynthetic efficiency (F V /F M ) of the algal assemblage was measured using the maximum quantum yield technique (see Sample Analysis section below for details).

Nutrient-temperature amendment experiments
During the Maumee Bay transect, an additional 60 liter of subsurface water (~0.25 m) was collected in acid-washed 20 liter carboys at sites M2 and M4 for experiments. Site M2, which was in Maumee Bay, hosted a cyanobacteriadominated bloom community (Fluoroprobe-derived cyanobacteria concentrations above the 25 μg L −1 bloom threshold used by the New York State Department of Environmental Conservation), whereas site M4, which was within the open waters of the western basin, hosted a mixed-algal community that was more representative of a nonbloom population (< 25 μg L −1 ; Fig. 1a). Experiments were designed to assess the effects of elevated N, P, and temperature (T), as well as their interactions, on algal community composition, abundance, and toxicity. For each experiment, two sets of 1-liter polycarbonate bottles (n = 12 per set) were filled with the surface water via laminar flow and amended in triplicate with additions of nitrate (20 μmol L -1 NO 3 − ), phosphate (1.5 μmol L -1 K 2 PHO 4 ), nitrate and phosphate (20 μmol L -1 NO 3 − /1.5 μmol L -1 K 2 PHO 4 ), or left unamended as a control.
The nutrient additions were representative of pulses of nitrate and phosphate that have been previously described in the region (Wilhelm et al. 2003;Davis et al. 2015; and were administered at the start of the experiment and again after 48 h to prevent depletion before the conclusion of the experiment. The bottles were incubated for 96 h in clear, submersible containers at 0.25 m depth in Fishery Bay, Lake Erie, near the Franz Stone Laboratory (The Ohio State University), at two temperatures. The first set of bottles (n = 12) was incubated at the ambient lake temperature, achieved with openings in the container that allowed lake water to exchange, whereas the second set of bottles (n = 12) was incubated at an elevated lake temperature (+4 C) achieved with heating wands (ViaAqua Inc.) within the container. This 4 C increase is consistent with temperature projections for the 21st century (Houghton et al. 2001) and is representative of summer heatwaves within temperate latitudes (Joehnk et al. 2008). The bottles were covered with one layer of neutral density screening to mimic in situ light conditions at~0.25 m, which were measured continuously along with experimental temperatures (average 22.3 and 26.6 C) using in situ HOBO loggers (Onset Computer Corporation). Bottles were mixed daily to promote equal phytoplankton-nutrient distribution and were inspected for floating cyanobacterial colonies prior to mixing to affirm cyanobacterial cells were healthy within the experiment vessels.
At the 96 h time point, the bottles were processed in the same manner as the field samples (see above).

Sample analysis
Microscopic enumeration was used to support the pigment-and DNA-based analyses of cyanobacteria abundances. For each preserved sample, 1 mL aliquots was settled for 5 min on a gridded Sedgewick-Rafter chamber before being visually scanned on an inverted microscope at 20X and 40X. Cyanobacteria colonies and filaments were quantified and identified to the genus level, with a minimum of 200 units counted per sample. The biovolume (μm 3 L −1 ) of each taxa was then estimated from the area per unit measured with a Nikon DS-Vi1 camera and NIS-Elements imaging software (version 3.22.11). Specifically, Microcystis biovolume was calculated from individual colony radii with the assumption that each colony was a sphere and that colony biovolume resembled cellular biovolume as the colonies were too dense and three dimensional to count individual cells. Dolichospermum biovolume was calculated by converting filament area to cells per filament using the average cell area calculated from 10 filaments per sample and then multiplying by the average cell biovolume (assuming spherical cells). Planktothrix and Aphanizomenon filament areas were converted to biovolume using the average filament width calculated from 10 filaments per sample and the assumption that each filament was a cylinder (Hillebrand et al. 1999;. Green and brown algal relative abundances determined via Fluoroprobe analysis were also confirmed by microscopic examination (data not shown).
Nutrient samples were analyzed for nitrate, ammonium, orthophosphate, total nitrogen (TN), and total phosphorus (TP) on a Lachat Instruments autosampler (ASX-520 series) using standard wet chemistry (Valderrama 1981;Jones 1984;Parsons 2013). Recovery of standard reference material of at least 90% AE 10% was achieved for all nutrient analyses. Photosynthetic efficiency, a diagnostic measure of nutrient stress (Parkhill et al. 2001;Simis et al. 2012), was determined by the increase in fluorescence yield from dark-adapted in vivo fluorescence (F V ) to maximal DCMU-enhanced (3,4-dichlorophenyl-1,1-dimethylurea, photosynthetic inhibitor) fluorescence (F M ), measured on a Turner Designs TD-700 fluorometer (EM filter of > 665 nm and EX filter of 340-500 nm) that was blank corrected with filtered lake water. Total microcystins were extracted from whole water samples by chemical lysis using the Abraxis Quicklyse Cell Lysis Kit, then quantified using the congener-independent (Fischer et al. 2001) Abraxis Microcystin/Nodularian (ADDA) enzyme-linked immunosorbent assay kit per the manufacturer's instructions. Concentrations were measured on a SpectraMax plus 384 plate reading spectrophotometer, calculated using a logarithmic curve, and reported in microcystin LR-equivalents. This assay provided a 98.6% AE 5% average recovery of samples spiked with known concentrations of microcystin-LR standard.

Jankowiak et al.
Effects of N, P, and temperature on cyanobacterial bloom DNA isolation, sequencing, and analysis For molecular analyses, total nucleic acids were extracted from the field and experimental samples (biological triplicates) using the cetyltrimethyl ammonium bromide (CTAB) method outlined in Dempster et al. (1999). Briefly, frozen filtered samples were placed in 1 mL of CTAB lysis buffer with 4 μL betamercaptoethanol, heated to 50 C for 30 min and then frozen at −80 C overnight, followed by a single chloroform extraction and an isopropanol/sodium chloride precipitation. The nucleic acids were then resuspended in 25 μL of sterile nuclease free water, and double-stranded DNA was assessed for quality using a Nanodrop spectrophotometer and quantity using a Qubit fluorometer with a dsDNA BR Assay kit per the manufacturer's instructions. Samples were normalized and stored at −80 C until polymerase chain reaction (PCR) amplification and amplicon sequencing, performed at Molecular Research Labs (Shallowater). The Sandusky transect S6 sample was excluded from sequencing due to low extraction efficiency and therefore was not considered for 16S and mcyE analysis.
To examine the cyanobacteria community composition, the 16S rRNA gene (~349 bp) was amplified using the cyanobacteria-specific primer set CYA106F: 5 0 -CGG ACG GGT GAG TAA CGC GTG A-3 0 (Nübel et al. 1997) and 530R: 5 0 -CCG CNG CNG CTG GCA C-3 0 (Usher et al. 2014), developed to discriminate between cyanobacteria genera (Nübel et al. 1997;Castiglioni et al. 2004). To determine the composition of the microcystin-producing community, the mcyE gene (~472 bp) was amplified using the HepF 5 0 -TTT GGG GTT AAC TTT TTT GGG CAT AGT C-3 0 and HepR 5 0 -AAT TCT TGA GGC TGT AAA TCG GGT TT-3 0 ) primer set developed by Jungblut and Neilan (2006) to discriminate between toxic and nontoxic strains of microcystin-producing genera. The mcyE gene was chosen as it is one of several microcystin synthetase genes and encodes an aminotransferase that plays a key role in the production of microcystin (Tillett et al. 2000).
Prior to PCR amplification, an identifying barcode was placed on the forward primer for each sample and then amplification was conducted using the HotStarTaq Plus Master Mix Kit and cycling conditions as follows: 94 C for 3 min, followed by 28 cycles of 94 C for 30 s, 53 C for 40 s, and 72 C for 1 min and a final elongation step at 72 C for 5 min. The PCR products were visualized on a 2% agarose gel to determine the success of the amplification and the relative intensity of the product bands. The samples were then pooled together for each respective primer region (16S rRNA or mcyE) in equal proportions based on their molecular weight and DNA concentrations, and purified using calibrated Ampure XP beads. The purified products were used to prepare a DNA library following the Illumina TruSeq DNA library preparation protocol and sequenced on an Illumina MiSeq platform for paired end reads (2 × 300) following the manufacturer's guidelines.
The 16S and mcyE sequence data were processed using the Quantitative Insights into Microbial Ecology (QIIME) v1.9.1 pipeline (Caporaso et al. 2010). Briefly, the raw paired-ends reads were trimmed of their sample identification barcodes prior to joining using seqprep, and then the joined reads were depleted of both primers and demultiplexed into their respective samples based on the associated barcodes. Quality filtering was performed throughout this process using the default parameters in QIIME, and the resulting split library output was filtered for chimeric sequences using usearch61 (Edgar 2010;Edgar et al. 2011). The filtered sequences were then clustered into operational taxonomic units (OTUs) based on 97% (16S) or 99% (mcyE) sequence identity using UCLUST (Edgar 2010) and open reference clustering with their respective reference databases (see below). The resulting representative sequences were aligned using PyNAST (Caporaso et al. 2010) and taxonomically classified using BLAST (Altschul et al. 1990). To confirm taxonomic assignment, representative sequences for the most abundant OTUs were extracted and identified using the National Center for Biotechnology Information (NCBI) basic local alignment search tool (BLAST). Postprocessing in QIIME, the 16S dataset was refined to exclude all noncyanobacteria, chloroplast, and mitochondria OTUs, which were not considered for further analysis as the focus of this study was on the cyanobacteria assemblage. The 16S OTUs were reported at the genus level, whereas the mcyE OTUs were reported at the lowest taxonomic level identified.
The SILVA rRNA (16S SSU) release v119 database (Quast et al. 2012) was used as the reference set for OTU picking and taxonomic identification of the 16S dataset, with two taxonomic modifications conducted postprocessing. First, the informatically identified Prochlorococcus genus was reassigned as Synechococcus for several reasons. First, NCBI BLAST identifications of the "Prochlorococcus" OTU consensus sequences were most similar to Synechococcus strains (Synechococcus sp. MV0409 and Synechococcus sp. 1tu14s11) in the Cluster F/Clade VII (Loar 2009). Next, phylogenetic analysis of the QIIME identified Prochlorococcus and Synechococcus OTU representative sequences resulted in an intermixed group, further suggesting misidentification. Finally, Prochlorococcus has yet to be reported within freshwater systems (Ouellette et al. 2006;Takasu et al. 2015). As another distinct Synechococcus OTU group was already identified from our analyses, the Prochlorococcus OTUs were reassigned as "Synechococcus clade II." Second, 16S sequences that were not matched to a taxonomic assignment at the genus level were grouped into an uncultured group and consisted of sequences most similar to Pseudanabaena, Microcystis, and Nodosilinea in NCBI BLAST searches (Supporting Information Table S2). As the "uncultured" OTU was a mixed taxonomic group, caution was used when interpreting the responses of this group to environmental variables.
For the mcyE dataset, a custom reference database was created from available sequences (n = 507) in the NCBI nucleotide database. Briefly, BLAST search results for the mcyE gene were filtered to remove whole genomes (> 2000 bp, to decrease processing time) or any noncyanobacteria taxa, and converted into taxonomy and FASTA files using the gb2qiime. py script (McCann 2014). The FASTA sequences were aligned at a 100% identity in GeneiousPro v5.6.7 (Kearse et al. 2012) using clustalW (Larkin et al. 2007) and the default settings from which duplicate sequences were removed and the remaining sequences were trimmed to the amplicon region. In QIIME the trimmed sequences were then clustered at an 80% identity using UCLUST (v1.2.22q) (Edgar 2010) and a representative sequence set was generated using the default settings. The condensed representative set was aligned using MUSCLE (Edgar 2004) to create a core alignment file that was used along with the representative sequence file during open reference OTU picking. The OTU taxonomic assignments were modified postprocessing in the case of Oscillatoria, which was reassigned to Planktothrix, based on two criteria. First, in previous studies , Planktothrix sequences have been shown to be misidentified as Oscillatoria by molecular pipelines due to reference database errors, and second, in samples where Oscillatoria was detected in high abundances by the mcyE sequencing, it was not detected by 16S sequencing or microscopic identification, while Planktothrix was detected.
For the 16S rRNA marker, the 58 in situ and experimental samples generated a total of 9,874,402 sequences with an average length of 368 bp. After joining and quality filtering a remaining 7,922,628 sequences clustered into 21,508 OTUs at a 97% sequence identity, of which 2567 were classified as cyanobacteria (Supporting Information Table S3). For the mcyE marker, the 58 samples generated a total of 7,050,837 sequences of~413 bp length. A total of 4,172,528 sequences remained after quality filtering and joining and clustered into 868 OTUs at a 99% sequence identity (Supporting Information Table S3). Sequence data from this study has been deposited to GenBank (Accession numbers SPR8472474-SPR8472589).

Statistical analyses
All statistical analyses were performed using R software v3.2.3 (R Core Team 2013) unless otherwise noted. For the transect samples, Spearman correlations were used to compare the community abundances (Fluoroprobe biomass, biovolumes, and 16S and mcyE relative abundances) and biochemical metrics (F V /F M and microcystin concentrations) to the environmental parameters (T, nitrate, ammonium, orthophosphate, TN, and TP) to assess drivers of the biological variation. Additionally, Spearman's correlations were used to compare cyanobacteria genera abundances determined by microscopy (biovolume) and 16S rRNA sequencing (relative abundance) methods.
For the experiments, three-way analyses of variances (ANOVAs) were used to test for significant (α = 0.05) effects of the experimental treatments (N, P, and T) on the community abundances (Fluoroprobe, biovolumes, and mcyE relative abundances) and biochemical metrics (F V /F M and microcystin concentrations). Briefly, prior to ANOVA analysis, data were assessed for normality (Shapiro-Wilk test) and equal variance (Bartlett's test) and then modeled using linear models (Gaussian error structure) which were assessed for goodness of fit through examination of qqplots and histograms of the residuals. Right skewed datasets were natural log transformed to increase normality, and the mcyE relative abundance datasets were arc-sin-square root transformed. Datasets that did not fit a Gaussian distribution (failed assumptions of normality and equal variance tests) were modeled using a generalized linear model assessed for goodness of fit using 1 − (residual deviance/null deviance). Specifically, fluorescence abundances were modeled with a Poisson error structure while the proportion of OTU to total sample mcyE reads were modeled with a binomial error structure and logit link function. The model types and transformations used are noted in the ANOVA result tables. Three-way ANOVAs (car package) were then performed on the models to determine significant main effects and interactions of the N, P, and/or temperature treatments, followed by a Tukey post hoc multiple comparison analysis.
Differential abundance analysis was used to identify significant changes in the abundances of the individual cyanobacteria genera identified by the 16S sequencing due to the experimental treatments or between the field stations, using the Phyloseq and DESeq packages (Anders and Huber 2010;McMurdie and Holmes 2013;McMurdie and Holmes 2014). Briefly, raw 16S read counts were modeled using a negative binomial distribution and parametric fitting of the dispersions, followed by Wald significance testing to calculate significant log 2 fold changes in abundance (α = 0.05). Changes in abundances were examined between bays and between the nearshore and offshore sites in the field and compared to the control in the experiments. p values were adjusted using the Benjamini-Hochberg procedure to correct for multiple testing.
Multivariate statistical approaches were employed to analyze differences among the 16S and mcyE community structures (composition/abundance) in relation to environmental and experimental variables. Specifically, hierarchical clustering analysis (average linkage algorithm) was conducted using Euclidean and Bray-Curtis dissimilarity metrics of the physio-chemical and relative abundance (16S, mcyE) datasets, respectively, to determine groups of field sites with similar physical conditions and community compositions. The field and experimental 16S and mcyE relative abundances were then analyzed with nonmetric multidimensional scaling (nMDS; vegan package) analysis using the Bray-Curtis dissimilarity metric to identify variables whose communities behaved in a similar manner. To test for significant dissimilarities in the community composition between field groups identified via clustering or experimental treatments analysis of similarity (ANOSIM; rank based) and permutational analysis of variance (PERMANOVA; distance based) tests were performed on the 16S and mcyE Bray-Curtis dissimilarity metrics in PAST v2.17c (Hammer et al. 2001) with 999 permutations. Similarity percentage analysis (SIMPER) was then conducted to identify taxa driving the community dissimilarity between significantly different groups. Significant differences in diversity (genus richness) between the field clusters or experimental treatments were assessed using one-way ANOVAs followed by a Tukey post hoc analyses. Additionally, canonical correspondence analyses (CCA) were performed to identify associations between the environmental variables and the community compositions of the field samples complemented with PERMA-NOVA analysis (Adonis; vegan packages) to test for significant correlations between the physiochemical parameters and community variations.

Field surveys Physiochemical lake properties
In both the Maumee and Sandusky transects, most physiochemical parameters varied with proximity to the respective rivers, with distinct nutrient-temperature conditions observed between the nearshore and offshore sites (Fig. 2a). Surface water temperatures (19.0-21.6 C) significantly increased with distance offshore (t-test, p < 0.05; Table 1), while nutrient concentrations were typically higher nearshore. TN and TP concentrations were significantly higher at nearshore sites (M1-M2 and S1-S4; TN > 75 μM; TP > 3 μM) in both transects (t-test, p < 0.01; Table 1). Peak nitrate concentrations were three-fold greater near the mouth of the Maumee River (25.5 μM) compared to the Sandusky transect, whereas ammonium (3.47 μM) and orthophosphate (0.99 μM) concentrations were two-to three-fold higher along the Sandusky transect (Table 1). There was no significant difference between the nutrient and temperature conditions between the bays (t-test, p > 0.05).

Phytoplankton class dynamics
Cyanobacterial blooms were present in both the Maumee and Sandusky Bays, with similar spatial distributions of the phytoplankton classes across each transect. Cyanobacteria dominated the fluorometrically identified phytoplankton assemblages at all transect sites, comprising 40-100% of the community, with the highest abundances occurring nearshore (Fig. 3a). Along the Maumee transect, cyanobacterial biomass peaked at site M2 (66.3 AE 1.19 μg L −1 ) and decreased to lower levels in the open waters of the western basin (M3-M4, 10.9 AE 0.70 μg L −1 ; Fig. 3a), being dominated by Microcystis at all sites (as a percentage of biovolume; Supporting Information Fig. S1B). Along the Sandusky transect, cyanobacteria dominated the phytoplankton assemblages throughout the inner and outer bay (S1-S4, 114 AE 66.9 μg L −1 ), peaking at site S4 (189 μg L −1 ) and decreasing into the western basin (Fig. 3a). Cyanobacterial levels were three-fold greater in the Sandusky transect compared to the Maumee transect and was composed primarily of Planktothrix agardhii (Supporting Information Fig. S1b). Green and brown algae were only present at sites with lower cyanobacterial levels, with the highest abundance of green algae occurring near both river mouths and brown algae only occurring offshore within the Maumee transect (Fig. 3a). ANOSIM detected a significant dissimilarity (p < 0.05) of the pigment-defined phytoplankton communities between transects, as the Sandusky transect phytoplankton assemblage was almost exclusively cyanobacteria (> 93%), with higher cyanobacterial biomass, compared to the Maumee transect which exhibited a more mixed phytoplankton community (Fig. 3a).

Cyanobacteria community dynamics
Both microscopy and 16S sequencing identified Microcystis, Planktothrix, Aphanizomenon, and Dolichospermum as the most abundant cyanobacteria in Maumee and Sandusky Bays with significant correlations of the relative abundances quantified via microscopy and sequencing for Microcystis, Planktothrix, and Dolichospermum (p < 0.05; Supporting Information Table S4). Amplicon sequencing of the 16S rRNA gene identified diverse cyanobacterial assemblages associated with the Maumee and Sandusky blooms (Fig. 4) that varied between bays and spatially within each transect (Fig. 2b). Along the Maumee transect, 1435 OTUs corresponding to 30 genera were identified, with a core community of 128 OTUs shared among all sites (Fig. 5).
The potential microcystin-producing (mcyE gene containing) communities in both systems exhibited a low community diversity (Supporting Information Fig. S3), with only 486 genotypes corresponding to three cyanobacteria taxa-one Planktothrix strain and two Microcystis clades (Fig. 3d). The first Microcystis clade consisted of 81 OTUs most similar to Microcystis PCC 7005 (100% identity), Microcystis UTEX LB 2664 (100% identity), and Microcystis PCC 7806SL (99% identity) in NCBI BLAST search. The second Microcystis clade consisted of 500 OTUs most similar to Microcystis NIES-102 (99% identity) and Microcystis NIES-843 (99% identity). Planktothrix dominated the microcystinproducing community in both transects with a mean relative abundance of 64.9% AE 13.7% in Maumee Bay and 68.0% AE 6.72% in Sandusky Bay (Fig. 3d). Of the Microcystis clades identified, clade 2 was more abundant in the Maumee transect, with a relative abundance ranging from 15.6% to 44.6%, while clade 1 was more abundant along the Sandusky transect, with an average relative abundance ranging from 3.11% to 35.9% (Fig. 3d). Microcystin concentrations were most strongly correlated with the Planktothrix mcyE relative abundances and Planktothrix cell abundances (by biovolume and filament counts) in Sandusky Bay (p < 0.05 for all; Supporting Information Table S6).

Nutrient-temperature amendment experiments Maumee Bay experiment
After 96 h, the total phytoplankton community in the control differed from that of the initial bloom community, with a decrease in cyanobacterial biomass to 23 μg L −1 and an increase in green algae to 8.05 μg L −1 (Fig. 7a). Compared to the control, green algal biomass was significantly lower in all +N treatments (p < 0.001; Supporting Information Table S7A and Fig. S4; Fig. 7a), whereas cyanobacterial biomass significantly increased by three-to four-fold in all +N treatments (p < 0.001; Supporting Information Table S7A and Fig. S4;  Fig. 7a). Within the +N treatments, there was a significant, synergistic +N + T interaction (p < 0.001) on cyanobacterial abundance resulting in the greatest cyanobacterial biomass, whereas there was a significant, antagonistic interaction of +N + P at ambient temperature with cyanobacteria levels lower than the +N only treatment (p < 0.001, Supporting Information Table S7A and Fig. S4). Higher temperatures also yielded a significant increase in cyanobacterial biomass (p < 0.001; Supporting Information Table S7A) among all nutrient treatments (Supporting Information Fig. S4), with a synergistic interaction of T and P yielding levels higher than the individual treatments (p < 0.05; Supporting Information Table S7A and Fig. S4). Concurrent with the increase in cyanobacterial biomass, photosynthetic efficiency (F V /F M ; Table 2) was significantly higher in the N and T treatments compared to the control after 96 h (p < 0.05; Supporting Information Table S7C).
The cyanobacteria community structure (16S) was only significantly altered by N (PERMANOVA, p < 0.01), with an average community dissimilarity of 29.9% (SIMPER) between treatments with and without N, driven by variations in the relative abundance of Planktothrix (42.0%), Microcystis (20.1%), Synechococcus clade II (13.6%), and Aphanizomenon (11.4%; Fig. 7c). The relative abundance of seven genera were significantly altered by N (p < 0.05; Supporting Information Table S5B), with an increase in Planktothrix and Nodularia and decrease in Microcystis, Aphanizomenon, Dolichospermum, Limnothrix, and Chroococcidiopsis among the +N treatments (Supporting Information Table S5B; Fig. 7c). Elevated P caused a significant increase in the relative abundance of Aphanizomenon and Dolichospermum (p < 0.05; Supporting Information Table S5B; Fig. 7c) and decrease in Chroococcidiopsis, while the relative abundance  of Dolichospermum and Chroococcidiopsis significantly increased due to elevated temperature (p < 0.05; Supporting Information Table S5B).
The microcystin-producing (mcyE-containing) community structure was significantly altered by the +N + P treatments (p < 0.05; PERMANOVA; Fig. 7d), with an average community dissimilarity of 19.3% compared to the non + N + P treatments. Within the +N + P treatments, Planktothrix, which remained the dominant strain across all treatments (75.4% AE 9.48%; Fig. 7d), significantly increased in relative abundance (p < 0.01; Supporting Information Table S7B and Fig. S4), whereas the Microcystis clade 2 significantly decreased in relative abundance (p < 0.05;  Table S7B and Fig. S4). Additionally, the relative abundance of Microcystis clade 1 significantly increased under elevated temperature (p < 0.05; Supporting Information Table S7B). Microcystin concentrations significantly increased in response to +N (p < 0.001; Supporting Information Table S7C; Fig. 7b) and were significantly correlated with the relative abundance (16S and mcyE) of Planktothrix across all experimental bottles (p < 0.05, Spearmans Correlation).

Western basin experiment
The non-bloom total phytoplankton community structure was more stable than the bloom community in Maumee Bay over the 96-h incubation, with a slight increase in brown algal biomass (23.3 AE 2.05 μg L −1 ) and decrease in cyanobacterial biomass (7.21 AE 0.72 μg L −1 ) in the control relative to initial conditions (Fig. 8a). Green algal abundance significantly increased in the +N + P treatment compared to the control (p < 0.01; Supporting Information Table S7A; Fig. 8a) with a synergistic interaction occurring in the +N + P + T treatment, resulting in significantly higher abundances than in all other treatments (p < 0.001; Supporting Information Fig. S4). Cyanobacterial abundance significantly increased in all +N treatments compared to the control (p < 0.001; Supporting Information Table S7A; Fig. 8a), with significant interactions within the +N + P and +N + T treatments (p < 0.001; Supporting Information Fig. S4), and the highest abundances occurring in the +N + P + T treatment (Fig. 8a). Conversely, brown algal abundance was significantly lower in the +N and +T treatments (p < 0.001; Supporting Information Table S7A, Fig. 8A), as well as the +N + P + T treatment compared to the control (p < 0.001; Supporting Information Fig. S4). On average, the photosynthetic efficiency of the phytoplankton community was significantly higher under +N conditions (p < 0.05, Supporting Information Table S7C).
After 96 h, within the cyanobacteria assemblage, N enrichment resulted in a significant decrease in the relative abundance (16S) of Synechococcus clade I (p < 0.05) and the elevated temperature conditions resulted in a significant decrease in relative abundance (16S) of Microcystis (p < 0.05; Supporting Information Table S5C; Fig. 8c). Among the microcystin-producing cyanobacteria, Microcystis clade 2 became the most abundant taxa in all treatments after 96 h ( Fig. 8d) but had significantly lower relative abundances in the +N treatments (p < 0.05; Supporting Information Table S7B and Fig. S4). Planktothrix became more abundant in the +N, +P, and +T treatments (Fig. 8c); however, abundances were not significantly different in any individual treatments compared to the control (Supporting Information Fig. S4). Overall, the microcystin-producing community structure was significantly altered under the +N + P treatment (p < 0.05; PERMANOVA). Microcystin concentrations significantly increased in the N enrichment treatments (p < 0.05; Supporting Information Table S7C; Fig. 8b), with the highest concentration (5.64 AE 6.69 μg L −1 ) in the +N + P + T treatment (Table 2; Fig. 8b).

Discussion
Nutrient and temperature controls on cyanobacteria abundance and phytoplankton community dynamics The 2015 CHAB bloom in Lake Erie studied here was one of the most severe to date (National Oceanic and Atmospheric Administration 2017) and formed following a period of high rains associated with the record discharges from the Maumee River during June and July (National Oceanic and Atmospheric Administration 2015b). Findings indicated that cyanobacterial abundance in September 2015 was strongly influenced by nutrient availability, with peak cyanobacterial abundances present under high nutrient regimes. Across the Maumee and Sandusky transects, cyanobacterial abundance was significantly correlated with TN and TP (p < 0.05; Supporting Information Table S6), with the highest concentrations at the nearshore sites and decreasing with distance offshore, associated with reduced influence of the river discharges. Although this correlation is in part due to cyanobacterial cells contributing to these pools as particulate N and P, the individual nutrient species followed similar trends. These similar cyanobacteria-nutrient spatial distributions in both bays support the hypothesis that nutrient inputs from the Table 2. Response of the biological processes (F V /F M , Chl a, and microcystin concentration) to the nitrogen (N), phosphorus (P), and temperature (T) treatments in (A) the Maumee Bay experiment and (B) the Western Basin experiment. Maumee and Sandusky tributaries are important stimuli for cyanobacteria growth in this system and help fuel the latesummer bloom in Lake Erie's western basin (Millie et al. 2009;Michalak et al. 2013). Experiments further indicated the influence of nutrient enrichment on cyanobacterial abundance, as N additions were the primary driving factor of cyanobacteria growth. During the Maumee Bay and western basin experiments, N enrichment significantly increased cyanobacterial abundance threeand two-fold higher compared to the controls, respectively, with concurrently increased photosynthetic efficiency of the phytoplankton communities. Together, these findings suggest that the phytoplankton community throughout western Lake Erie was N limited (Parkhill et al. 2001;Simis et al. 2012), which has been previously described during late-summer blooms in this region . In contrast, P enrichment alone did not alter cyanobacterial abundance in either experiment. This response may be partially attributed to the ability of Microcystis, which was a primary component of the Maumee Bay communities, to persist under low P conditions due to efficient P sequestering strategies (i.e., luxury uptake, high-affinity P transporters, and alkaline phosphatases; Wilhelm et al. 2003). Spatial differences were observed, however, between Maumee Bay and the open waters of the western basin that had lower total P levels and the greatest increase in cyanobacteria abundance with the dual addition of N and P, suggesting cyanobacteria in the open waters of the western basin may be under dual N and P limitation. These results affirm the traditional paradigm that excess P is an important factor promoting the intensity of cyanobacteria blooms in western Lake Erie (Stumpf et al. 2012;Michalak et al. 2013), but indicates that N availability is an important driver for nondiazotrophic genera (Microcystis, Planktothrix), which dominate western Lake Erie cyanobacteria communities in late summer (Conley et al. 2009;Paerl and Huisman 2009;Gobler et al. 2016).
Cyanobacteria abundance was also found to be significantly enhanced by elevated temperature, with the combined increase in temperature and N yielding the highest cyanobacteria concentrations in both experiments, suggesting that although temperature is not the primary driver of cyanobacteria proliferation, it may additively enhance the impact of eutrophication. This finding has important ecological implications as it suggests that eutrophied freshwater systems will become more impaired with the projected increases in global temperatures (Paerl and Huisman 2009). Given that temperatures are expected to rise through this century regardless of global carbon measures (Edenhofer et al. 2014), restrictions to nutrient loads will be needed to mitigate cyanobacteria proliferation (Conley et al. 2009;Paerl and Huisman 2009;Michalak et al. 2013).
In both the field surveys and experiments, phytoplankton diversity as defined by pigment-based classes was inversely correlated with cyanobacterial abundance with reduced levels under high nutrient-temperature. Specifically, in the transects, green and brown algae were only detected at sites with lower cyanobacteria abundances, with green algae present at the more nutrient-rich sites near the Maumee and Sandusky Rivers (M1, S1) and brown algae in the nutrient-limited sites offshore (M3-M5), consistent with phytoplankton distributions previously reported for Lake Erie (Ghadouani and Smith 2005;Millie et al. 2009). Similarly, in the experiments, green and brown algae were not present (Maumee Bay experiment) or present at low abundances (Western basin experiment) in the N-enrichment treatments and reduced (to a lesser degree) by elevated temperature, conditions which significantly increased cyanobacteria abundance. However, in the western basin experiment, where cyanobacteria were less abundant, green algae abundance significantly increased under +N + P conditions and there was a smaller reduction in brown algae over time in the +P treatments indicative of N and/or P limitation among these communities. Together, these finding suggest the green and brown algae may benefit from nutrient additions but are outcompeted by cyanobacteria (Dokulil and Teubner 2000;Wilhelm et al. 2003;Paerl and Paul 2012), particularly in combination with warmer temperatures (Rigosi et al. 2014). Beyond nutrient competition (Elliott et al. 2006), these patterns may also be associated with allelopathic inhibition by cyanobacteria (Sukenik et al. 2002;LeBlanc et al. 2005;Chia et al. 2018), which can become more intense under nutrient-limited conditions (Chia et al. 2018) such as those described here.
Spatial dynamics of the Lake Erie cyanobacteria community structure Although similar nutrient, temperature, and fluorometric community trends were observed in the Maumee and Sandusky transects, statistically distinct cyanobacteria communities were found within each system. Of the 2567 cyanobacteria genotypes identified, 25% of the OTUs were unique to Maumee Bay, whereas 44% were unique to Sandusky Bay (Fig. 5), with more diverse communities in Sandusky Bay as well. Community dissimilarity was driven by different bloom-forming genera within each system, as the Maumee Bay 16S communities were dominated by Microcystis (43%), a well-documented pattern in Lake Erie's western basin since the mid-1990s (Brittain et al. 2000;Wilhelm et al. 2003;Millie et al. 2009;Harke et al. 2016), whereas Sandusky Bay communities were dominated by Planktothrix, which is less studied but increasingly reported in recent years Conroy et al. 2007;Davis et al. 2015). This difference may be, in part, associated with the differing orthophosphate levels between the bays, which were lower in the Maumee transect, and Microcystis' tolerability of low P conditions (Harke et al. 2012). Under Plimited conditions, Microcystis has been found to up-regulate several P-scavenging genes allowing it to persist in low P conditions (Harke et al. 2012), whereas Planktothrix has been observed to respond more conservatively, down-regulating alkaline phosphatase, photosynthesis, and cell division genes  and up-regulating fatty acid degradation genes as an energy-sparing mechanism (Kaczmarzyk and Fulda 2010). Furthermore, Planktothrix has often been reported to dominate phytoplankton communities in temperate systems which are N limited but P rich (Donald et al. 2013;Steffen et al. 2014;Andersson et al. 2015), supporting their higher demand for P. However, these differences may also be related to physical differences between the bays. Microcystis is more likely to benefit in deeper, stable water conditions, such as Maumee Bay, due to its ability to regulate its bouyancy (Dokulil and Teubner 2000), allowing it to remain within surface waters to prevent against light limitation as well as migrate to deeper nutrient-rich waters. In contrast Planktothrix, which does not have this ability, is often favored in turbid, shallow waters Berger 2017), such as Sandusky Bay. Factors such as light or carbon limitation (Poste et al. 2013;Bullerjahn et al. 2016) may further explain the variation exhibited in the cyanobacteria community composition during this study, as the parameters examined only accounted for 77% of the variability (Fig. 6). Noteably, however, Planktothrix was also relatively abundant in the Maumee transect, accounting for around 10% of the community, and has been recorded in high abundances (up to 90% of the cyanobacteria community) in the western basin of Lake Erie in recent years (Millie et al. 2009), indicating it is an important component of CHABs across the western basin of Lake Erie.
Although Planktothrix and Microcystis are nondiazotrophic cyanobacteria, both dominated under the relatively low dissolved inorganic nitrogen (DIN) conditions observed in this study (~4 μM), despite N-stress among these populations displayed during experiments. In the Maumee Bay experiment, the relative abundance of Planktothrix was significantly increased by N enrichment, a finding consistent with prior studies of Planktothrix communities from Sandusky Bay . Microcystis, conversely, did not increase in relative abundance in the N-enrichment treatments but remained a large portion of the cyanobacteria community. However, given that total cyanobacteria abundance increased two-to three-fold during experiments, the absolute abundance of Microcystis likely increased during N additions as well. Freeliving or epiphytic N-fixing heterotrophic bacteria may play an important role in providing N to these populations (Worm and Søndergaard 1998;Davis et al. 2010), thereby supporting their persistence under N-Limited conditions, although there remains a limited understanding of their importance in Lake Erie (Mou et al. 2013). Such bacteria have been found to be imbedded within Microcystis colonies (Song et al. 2017), which may help explain the muted response of Microcystis to N additions compared to Planktothrix. Additionally, it has been proposed that the presence of co-occurring diazotrophic taxa may help relieve N stress of these populations, as N fixation can "leak" N-rich amino acids and ammonium (Ohlendieck et al. 2000;Wetzel 2001). Consistent with this theory, in both transects, sites with peak cyanobacteria abundance exhibited higher abundances of diazotrophic genera (Aphanizomenon and Dolichospermum), a pattern previously described for Sandusky Bay . However, due to their relatively low abundance, the quantitative importance of dizaotrophs in this system remains unclear.
Beyond the toxic, bloom-forming genera detected during this study (i.e., Microcystis and Planktothrix), Synechococcus was a major component of the Lake Erie cyanobacterial communities, accounting for 25-75% of 16S sequences in both bays. This is an important finding as Synechococcus has often been overlooked in prior microscopy-based Lake Erie studies (Hopkins and Lea 1982;Makarewicz 1993;Postius and Ernst 1999), in part due to its small size (< 2.0 μm) and nondistinct morphology. Molecular methods have, however, revealed that Synechococcus can be abundant in freshwater systems ) and even dominant among picoplankton (Weisse 1993;Padisák et al. 1997). Our findings support those of Ouellette et al. (2006) and Wilhelm et al. (2006), which found Synechococcus-dominated 16S rRNA clone libraries across Lake Erie and suggests that although Synechococcus is likely a minor contributor to cyanobacterial biomass due to its small size, it likely plays an important role in Lake Erie's ecosystem and food web functioning due to its high relative abundance (Fahnenstiel et al. 1998;Postius and Ernst 1999;Gobler et al. 2008). For example, Wilhelm et al. (2006) found Synechococcus may contribute toward the development of hypoxic zones in Lake Erie's central basin where it dominates the prokaryotic assemblage during seasonal rapid draw-down of dissolved oxygen. The decreased and unchanged relative abundance of Synechococcus (clade I and II) in response to N in the Maumee Bay and western basin experiments, respectively, despite the increase in total cyanobacteria abundance indicates that, unlike bloom-forming cyanobacteria, the niche of Synechococcus is defined by low available N levels.
Nitrogen availability, bloom toxicity, and the potential microcystin-producing community This study provided insight regarding the toxicity of Lake Erie CHABs during the late summer by characterizing potential microcystin-producing taxa present, their associations with microcystin concentrations, and the effect of nutrients and temperature on both. Although several known microcystinproducing genera were detected during this study via microscopy and 16S sequencing (e.g., Dolichospermum, Planktothrix, and Microcystis), all 868 mcyE genotypes mapped exclusively to Planktothrix and Microcystis, a finding consistent with prior investigations of the microcystin-producing communities in Lake Erie (Rinta-Kanto and Wilhelm 2006) and elsewhere (Hotto et al. 2008;Ye et al. 2009). There was a significant correlation between microcystin levels and Planktothrix filament abundance and biovolume in Sandusky Bay (p < 0.05 for both; Supporting Information Table S6). Additionally, microcystin concentrations closely paralleled toxigenic (mcyE-containing) Planktothrix abundances across western Lake Erie, with peak microcystin concentrations coinciding with peak Planktothrix mcyE relative abundances in both transects, suggesting Planktothrix was an important contributor to microcystin production in both Maumee and Sandusky Bays. This finding was supported by experimental findings as microcystin concentrations more closely paralleled changes in the relative abundance of Planktothrix than the Microcystis clades within the mcyE containing community. The relative abundance of the Planktothrix mcyE strain was generally higher in the +N treatments in which microcystin concentrations significantly increased (p < 0.05). Although there was not a significant increase in the relative abundance of the Planktothrix mcyE strain in the +N treatments, cyanobacteria biomass significantly increased in the +N treatment, indicating that the absolute abundance of Planktothrix likely increased in parallel with microcystin concentrations. Further, microcystin concentrations were significantly correlated with the relative abundance (16S and mcyE) of Planktothrix across all experimental bottles in the western basin experiment. While Planktothrix has been previously identified as the primary microcystin producer in Sandusky Bay Davis et al. 2015), Microcystis has typically been described as the primary microcystin producer in Maumee Bay and has been a primary focus of CHAB studies conducted in the western basin of Lake Erie (Rinta-Kanto et al. 2005;Davis et al. 2014;Harke et al. 2016). Supporting our findings, however, an examination of microcystin-producing communities in Maumee Bay by Rinta-Kanto and  found that although all toxic genotypes detected mapped to Microcystis, a subgroup contained two additional amino acids which were similar residues to known Planktothrix sequences, suggesting that toxigenic Planktothrix was present but not detected in Maumee Bay due to prior methodological constraints (i.e., use of clone libraries). Additionally, Kutovaya et al. (2012) found that Planktothrix was an important component of the microcystin-producing community in the Maumee River and hypothesized this population may seed Maumee Bay. Collectively, these findings may have important implications for bloom toxicity as Planktothrix and Microcystis produce different microcystin variants that vary in potency and Planktothrix has been shown to have higher microcystin production rates than Microcystis in the field (Christiansen et al. 2003). This suggests Planktothrix, along with Microcystis, should be a focus for management strategies focused on bloom toxicity in Lake Erie. It should be noted, however, that while the mcyE gene plays a major role in microcystin synthesis, it is one of several genes necessary for production and, therefore, its detection does not absolutely confirm the presence of potentially toxigenic Planktothrix strains. Furthermore, this study only examined the microcystin producing community at one point during the bloom and thus may not be representative of the mcyE containing community at other times.
Several lines of evidences emphasized the importance of N in controlling microcystin levels in Lake Erie, with little to no effect of P and temperature. In both experiments, N additions yielded significantly higher microcystin concentrations (Supporting Information Table S7), and microcystin concentrations were significantly correlated with TN in Sandusky Bay (Supporting Information Table S6). These responses are similar to those observed in field studies by Davis et al. (2015) and Donald et al. (2011) that found a greater increase in microcystin concentrations in response to N than P additions and supports the hypothesis that N availability controls microcystin cellular quotas as microcystins are N-rich molecules (Harke and Gobler 2013;Horst et al. 2014; Van de Waal et al. 2014;. This direct control of N availability on microcystin production is consistent with molecular analyses that have indicated the microcystin biosynthesis gene cluster is regulated by a N uptake transcription promoter (NtcA; Ginn et al. 2010, Neilan et al. 2013 and that the mcy genes are upregulated in response to N additions . Increases in microcystin concentrations in response to N enrichment may also be related to an increase in the abundance of toxic strains, which have been found to be selectivity promoted over non-toxic strains under N enriched conditions in Microcystis (Vézie et al. 2002;Davis et al. 2009). In the field, the relative abundance of mcyE-containing Planktothrix was significantly correlated with the ammonium levels (p < 0.05). Further, in the experiments, N increased the levels of cyanobacteria and microcystin but significantly decreased the relative abundance of Microcystis clade 1, indicating that the other dominant mcyE-strain, Planktothrix, had a higher absolute abundant in the presence of excess N and contributed to the higher microcystin levels in the treatment. All of these findings indicate that N loading plays an important role in CHAB toxicity and that N management is necessary to control bloom toxicity as well as cyanobacterial biomass.

NGS as a tool for cyanobacteria community studies
Overall, NGS proved to be a powerful tool for examination of the cyanobacterial assemblages during this study, allowing for a more detailed assessment of cyanobacteria community diversity and composition. While microscopy facilitated the identification of the six larger cyanobacteria genera, NGS detected over 2500 genetically unique OTUs at a 97% identity, corresponding to 46 genera, across Lake Erie's western basin. This allowed for identification of both rare (e.g., Chroococcidiopsis) and difficult to identify taxa, such as Synechococcus, including the differentiation of two sequentially distinct clades of Synechococcus, that in some cases responded differently to environmental drivers. There are also, of course, some caveats to consider regarding amplicon sequencing data. Read counts can provide an inaccurate measure of cell abundances, as multiple copies of a gene may be present in a genome and copy numbers can differ between species and within species (Acinas et al. 2004). This has been shown to occur with the 16S rRNA gene in several common cyanobacteria, including as Microcystis (Kaneko et al. 2007) and Synechococcus (Fogel et al. 1999). Additionally, PCR biases can arise due to the inability of primer sets to uniformly amplify all taxa (Polz and Cavanaugh 1998;Farris and Olson 2007), even with the use of "universal" primers and normalization of amplified DNA, as was the case in this study. Finally, taxonomic identification of sequence data can be limited by the robustness of reference databases (Shendure and Ji 2008), and therefore genes that are not highly sequenced, such as the mcyE gene used in this study, can result in limited taxonomic identification. This accounts for the discrepancy between the low taxonomic but high genetic diversity identified in the microcystin-producing communities and suggests that other genera may be present within this community but are not yet distinguishable, which should be improved with further sequencing studies. Despite these potential pitfalls, there were statistically significant correlations between the microscopic (biovolumes) and 16S relative abundances of the most abundant cyanobacterial genera during this study (Microcystis, Planktothrix, and Dolichospermum), indicating the robustness of the two approaches. Given that 40 of the 46 genera detected via NGS were not observed microscopically, correlations among the three other, less abundant genera observed microscopically were not necessarily expected.

Conclusion
This study demonstrated that N was the most influential environmental factor on cyanobacteria blooms in Lake Erie in late summer, promoting the abundance of cyanobacteria, microcystin concentrations, and causing the greatest changes in cyanobacteria community structure. Given that the combined N, P, and T treatment yielded the highest levels of cyanobacteria in both experiments and that P was an important factor promoting several diazotrophic genera cyanobacteria, dual management of N and P will be important for mitigation of CHABs in Lake Erie, especially in the future as summer lake temperatures rise. More studies using NGS to characterize bloom communities and their responses to environmental drivers are needed in freshwater systems to better understand the onset and persistence of cyanobacteria blooms.