Cyanotoxin-encoding genes as powerful predictors of cyanotoxin production during harmful cyanobacterial blooms in an inland freshwater lake: Evaluating a novel early-warning system

Abstract

Freshwater harmful cyanobacterial blooms (HCBs) potentially produce excessive cyanotoxins, mainly microcystins (MCs), significantly threatening aquatic ecosystems and public health. Accurately predicting HCBs is thus essential to developing effective HCB mitigation and prevention strategies. We previously developed a novel early-warning system that uses cyanotoxin-encoding genes to predict cyanotoxin production in Harsha Lake, Ohio, USA, in 2015. In this study, we evaluated the efficacy of the early-warning system in forecasting the 2016 HCB in the same lake. We also examined potential HCB drivers and cyanobacterial community composition. Our results revealed that the cyanobacterial community was stable at the phylum level but changed dynamically at the genus level over time. Microcystis and Planktothrix were the major MC-producing genera that thrived in June and July and produced high concentrations of MCs (peak level 10.22 μg·L⁻¹). The abundances of the MC-encoding gene cluster mcy and its transcript levels significantly correlated with total MC concentrations (before the MC concentrations peaked) and accurately predicted MC production as revealed by logistic equations. When the Microcystis-specific gene mcyG reached approximately 1.5 × 10³ copies·mL⁻¹ or when its transcript level reached approximately 2.4 copies·mL⁻¹, total MC level exceeded 0.3 μg L⁻¹ (a health advisory limit) approximately one week later (weekly sampling scheme). This study suggested that cyanotoxin-encoding genes are promising predictors of MC production in inland freshwater lakes, such as Harsha Lake. The evaluated early-warning system can be a useful tool to assist lake managers in predicting, mitigating, and/or preventing HCBs.

Graphical Abstract

1. Introduction

Harmful cyanobacterial blooms (HCBs; also known as cyanobacterial harmful algal blooms even though cyanobacteria are not eukaryotic algae) frequently occur in inland freshwater bodies during warm seasons. In the current study, we define an HCB event (or HCB in short) to be the significant growth of cyanobacteria with cyanotoxin production during warm seasons. HCBs potentially produce a large number of cyanotoxins with high concentrations such as microcystins (MCs, the dominant freshwater cyanotoxins), cylindrospermopsin, anatoxins, nodularins, and saxitoxins (Brooks et al., 2016; Munoz et al., 2019; Pearson et al., 2010; Schmale et al., 2019; Watson et al., 2014). Cyanotoxins significantly threaten aquatic ecosystems (e.g., causing fish poisoning and fish/waterfowl/mammalian mortality) and potentially contaminate municipal drinking water, posing serious public health risks (Backer and Miller, 2016; Brooks et al., 2017; Ferrão-Filho and Kozlowsky-Suzuki, 2011; Grattan et al., 2016; Landsberg, 2002). Ingestion of cyanotoxins can cause acute and chronic toxicity effects on the human liver, kidney, lung, nervous tissues, gastrointestinal tract, and immune systems (Buratti et al., 2017; Funari and Testai, 2008; Kubickova et al., 2019; Vu et al., 2020; Zanchett and Oliveira-Filho, 2013). Therefore, authorities shut down municipal water supplies when serious HCBs occur in drinking water sources (e.g., the 2014 drinking water crisis in Toledo, Ohio, USA, caused by a Microcystis bloom in Lake Erie) (Pelley, 2016; Steffen et al., 2017; Wynne and Stumpf, 2015). The United States Environmental Protection Agency (US EPA) has issued ten-day non-regulatory health advisory limits for MCs (i.e., MC concentrations at which adverse health effects are unlikely to occur over a ten-day exposure) in drinking water (0.3 μg·L⁻¹ for bottle-fed infants and pre-school children; 1.6 μg·L⁻¹ for school-age children through adults) (US EPA, 2015) and recreational water (8.0 μg·L⁻¹) (US EPA, 2019). An early-warning system predicting HCBs is critical for timely or early actions to protect drinking water sources and ecosystems and to limit recreational exposure. Once massive cyanobacterial growth has occurred, effectively alleviating HCBs and cyanotoxin production is difficult or even impossible.

Monitoring the changes in abundances and transcript levels of cyanotoxin-encoding genes is a promising approach to predict cyanotoxin production (Lu et al., 2020). MCs have more than 279 variants and are encoded solely by the gene cluster mcy (Abdallah et al., 2021; Bouaïcha et al., 2019; Cheng et al., 2021; Overlingė et al., 2021; Pearson et al., 2007; Pearson et al., 2019; Rouhiainen et al., 2004; Tanabe et al., 2009; Tillett et al., 2000; Tooming-Klunderud et al., 2008). The abundances and transcript levels of mcy have direct, positive correlations with MC concentrations (Chen et al., 2016a; Christiansen et al., 2008; Pacheco et al., 2016). For instance, mcy abundances positively correlated with MC levels in two freshwater ponds in India (Singh et al., 2015), one river in the US (Otten et al., 2015), and two inland lakes in China (Li et al., 2014). Therefore, mcy is a potential predictor of MC production. We developed a novel early-warning system to forecast MC production in William H. Harsha Lake (Ohio, USA) with a history of HCBs (Lu et al., 2020). The early-warning system monitored the abundances and expression (i.e., the transcript levels) of mcy using quantitative polymerase chain reactions (qPCR) and quantitative reverse transcription PCR (RT-qPCR), respectively. We established and used thresholds of qPCR and RT-qPCR signals of mcy to predict whether and when total MC concentrations would exceed health advisory limits during the 2015 HCB.

Our novel early-warning system is different from and has clear advantages over others (Brown et al., 2020; Davidson et al., 2016; Zohdi and Abbaspour, 2019). First, our approach relies on the direct, significant correlations between cyanotoxin-encoding genes and cyanotoxin production and is thus more straightforward. Previous early-warning systems rely on correlating cyanobacterial growth with environmental factors and cyanobacterial physiological parameters. Those factors and parameters affect cyanobacterial growth and can be potential indicators of cyanotoxin production. However, those factors and parameters control HCBs in indirect, complex, and often unpredictable ways. Therefore, early-warning systems based on those factors and parameters are less reliable. Second, our early-warning system is straightforward, rapid, and cost-effective. Previous early-warning systems, which generally target cyanobacterial blooms/growth rather than cyanotoxin production, are mainly based on complex mathematical (i.e., neural network) models (Henrichs et al., 2021; Lee and Lee, 2018; McGillicuddy, 2010; Ralston and Moore, 2020), remote sensing (Karki et al., 2018; Seltenrich, 2014; Tian and Huang, 2019; Wynne et al., 2013), optical and/or fluorescence detection (Ezenarro et al., 2021; Wang et al., 2018), and underwater camera imaging (Song et al., 2020). These complicated systems are difficult to use and can be cost-prohibitive.

The current study evaluating the early-warning system is novel and important. First, to implement an early-warning system in the field, we must fully evaluate the system against HCBs in different years and waterbodies. If the early-warning system is only accurate or useful for one single bloom, it would have limited scientific and practical implications. Therefore, we have conducted comprehensive, systematic, serial studies to evaluate the early-warning system. The first work in the serial studies proposed the concept of the early-warning system and was based on the 2015 HCB in Harsha Lake (Lu et al., 2020). In the current work, we focused on evaluating the efficacy of the early-warning system in forecasting cyanotoxin production in 2016 in the same lake. We have also tested the performance of the early-warning system in Harsha Lake in 2017 and 2018 and lakes in different geographical regions in the US (Smithville Lake, Missouri, in 2017; Detroit Lake, Oregon, in 2019; and Spanaway Lake, Washington, in 2019) (Table S1). Along with the fieldwork, we have evaluated the early-warning system in more detail in laboratory experiments using a toxic Microcystis aeruginosa strain and tested the efficacies of various approaches in preventing or mitigating cyanotoxin production (Table S1). Our serial studies form a complete story from the initial proposal of the early-warning concept to the application of engineering strategies to prevent cyanobacterial growth. Second, while the proposal of a novel early-warning system forecasting HCBs is significant, follow-up work evaluating the system is even more important than initial development and testing (Lu et al., 2020). Without appropriate evaluation, the early-warning system cannot be applied in the field. Third, a unique, clear contribution of the current work to the field is that it established thresholds of qPCR and RT-qPCR signals. When those thresholds are reached, massive cyanotoxin production will likely happen after a short period. The period is approximately one week in Harsha Lake but could be shorter or longer in other freshwater bodies. Water treatment plants and recreational water services need those thresholds to predict massive cyanotoxin production so that timely or early actions can be taken.

This study aimed to evaluate the novel early-warning system that we developed to forecast the 2015 HCB in Harsha Lake (Lu et al., 2020). We first examined whether an HCB event reoccurred in Harsha Lake in 2016 by characterizing cyanobacterial community dynamics. We then tested whether the early-warning system could accurately predict cyanotoxin production during the 2016 HCB event. In addition, to understand how environmental factors shaped the cyanobacterial community and affected cyanotoxin production, we measured numerous physicochemical parameters (including but not limited to nutrient-metabolism-related parameters). We hypothesized that the abundances and expression of cyanotoxin-encoding genes (mainly the mcy cluster) are promising predictors of cyanotoxin production during HCBs in freshwater bodies.

2. Materials and methods

2.1. Study sites and lake sample collection

Harsha Lake is a multi-use reservoir in the interior low plateau ecoregion in central Clermont County, Ohio, USA. The lake has a surface area of 8.7 km² and an average depth of 13.1 m with a maximum depth of 34.0 m (Zhu et al., 2019). The lake provides for flood prevention, drinking water, and recreational and wildlife habitats. As a eutrophic inland freshwater lake heavily impacted by nutrient loading from urban and agricultural land, Harsha Lake experiences HCBs with diverse cyanobacterial taxa (Lu et al., 2020; Lu et al., 2019; Zhu et al., 2019).

We sampled lake water from two sites, Harsha Buoy (BUOY; latitude 39.0325, longitude −84.1377) and East Fork Lake Surface (EFLS; near the intake of a municipal water utility; latitude 39.0367, longitude −84.1381) with a prolonged (March to September) and more frequent sampling scheme. Water samples were taken from the lake surface (0 to 0.5 m depth) three times a week in June; weekly in May, July, and August; and biweekly in April and September (32 sampling events for BUOY and 31 sampling events for EFLS) from March 25th to September 28th, 2016. Chen et al. (2017) describes the sampling sites in detail. We used this sampling scheme because low-frequency, short-term sampling is an important but neglected reason for the poor performance of mcy and qPCR as predictors of MC production during HCBs (Beversdorf et al., 2015; Pacheco et al., 2016; Sabart et al., 2015). We sampled from only the lake surface for two reasons. First, while Planktothrix could live in deeper water than Microcystis (Reynolds and Rogers, 1976; Su et al., 2015; Yu et al., 2018), our previous study (Lu et al., 2020) and the current study showed that compared to Microcystis, Planktothrix was not a major MC producer at the two sampling sites in Harsha Lake in 2015 and 2016 as determined by the abundances and expression of the mcy genes. Second, in our 2015 study (Lu et al., 2020), MC concentrations at a depth of 8 m were extremely low (data not shown).

We collected a 5 L water sample at each sampling event. An aliquot (approximately 9 mL) of each sample was transferred to a glass container and frozen at −20 °C for total MC analysis. A second aliquot (50 mL) was transferred into a 50 mL conical tube (Thermo Fisher Scientific; Waltham, Massachusetts, USA) for nutrient analyses. The second aliquot was split into two sub-samples. The first sub-sample was transferred to a screw-top culture tube for total nitrogen (TN) and total phosphorus (TP) concentration determination. The second sub-sample was transferred to a disposable culture tube for total reactive phosphorus (TRP), nitrite (NO₂⁻), nitrate (NO₃⁻), and ammonium (NH₄⁺) concentration analysis. A third aliquot (40 mL) of each sample was transferred to a scintillation vial containing one drop of concentrated sulfuric acid for total organic carbon (TOC) determination. Duplicates of a fourth aliquot (100 to 300 mL per duplicate) of each sample were filtered for DNA and RNA isolation.

2.2. Assessment of physicochemical water quality

We placed an EXO Sonde device (YSI Inc.; Yellow Springs, Ohio, USA) at BUOY to determine the following physicochemical water quality parameters: optical dissolved oxygen (ODO), water temperature, pH, specific conductivity, chlorophyll relative fluorescence, turbidity, blue-green algae phycocyanin (BGA-PC), oxidation-reduction potential (ORP), and water depth (the waves and wind constantly changed the relative locations of the Sonde device to the water surface). Some environmental parameters provided by the EXO Sonde device (i.e., salinity, conductivity, total dissolved/suspended solids, and water pressure) were derivative and thus were omitted from data analysis. Following the Standard Methods (APHA, 2017), we analyzed the concentrations of dissolved nutrients (NH₄⁺, NO₂⁻, NO₃⁻, TN, TRP, and TP) using a Lachat Quickchem 8000 Flow Injection Analysis System (Hach Company; Loveland, Colorado, USA) and TOC using a Phoenix 8000 TOC Analyzer (Teledyne Tekmar; Mason, Ohio, USA). In addition, we freeze-thawed the first aliquots for three cycles and centrifuged them for 15 min (3000 ×g, 4 °C) to release intracellular MCs. We then used an Enzyme-Linked Immunosorbent Assay (ELISA) (Gaget et al., 2017) to determine total MC concentrations in raw water using a Microcystins/Nodularins (ADDA) SAES ELISA Kit (Eurofins Abraxis, Inc.; Warminster, Pennsylvania, USA) on a Chromate 4300 microplate reader (Awareness Technologies; Palm City, Florida, USA). The limit of detection (LOD) for total MC concentrations in lake water was 0.15 μg·L⁻¹.

2.3. DNA and RNA extraction, DNA sequencing, and qPCR and RT-qPCR

We filtered the fourth aliquot (100 to 300 mL per duplicate) of each sample using an EMD Millipore Durapore® membrane filter (pore size 0.45 μm; MilliPore; Foster City, California, USA). Each filter with captured biomass was transferred to a 1.5 mL microcentrifuge tube containing 600 μL of a cell lysis Buffer RLT Plus with RNase inhibitor (QIAGEN; Valencia, California, USA) and stored at −80 °C before total genomic DNA and RNA extraction. We lysed the stored cells by shaking and beating the tubes with a Mini-Beadbeater-16 (BioSpec Products, Inc.; Bartlesville, Oklahoma, USA). We then centrifuged the tubes (10,000 ×g, 3 min, room temperature) and transferred the supernatant to new microcentrifuge tubes. We used an AllPrep® DNA/RNA Mini Kit (QIAGEN) to purify total genomic DNA and RNA from the supernatant. DNA contamination was removed from the RNA extracts (TURBO DNA-free™ Kit; Life Technologies Co.; Carlsbad, California, USA). We reverse-transcribed the cleaned RNA to cDNA using a High Capacity cDNA Reverse Transcription Kit (Life Technologies Co.). DNA and RNA concentrations were determined using a Nanodrop ND-1000 Spectrophotometer (NanoDrop Technologies, Inc.; Wilmington, Delaware, USA). The DNA and RNA extracts were stored at −80 °C before use.

We assessed the temporal variations of microbial communities using high-throughput 16S rRNA gene sequencing to determine whether the microbial community structures during the 2015 and 2016 HCBs were similar. The sequencing also allowed us to determine how the relative abundance (RA) of cyanobacteria against total bacteria changed over time and how the RA of dominant cyanobacterial genera against cyanobacteria changed over time. The sequencing was conducted following an established procedure (Lamendella et al., 2018; Zhang et al., 2021a). Briefly, we prepared the sequencing libraries by amplifying the V4 region using Illumina iTag PCR Mixtures (Illumina, Inc.; San Diego, California, USA) on an MJ Research PTC-200 Thermocycler (Bio-Rad; Hercules, California, USA). Each PCR (25 μL) contained (final concentration or mass) 5 to 10 ng of template DNA, 1× PCR buffer, 0.8 mM dNTPs (each), 0.625 U of Taq polymerase, 0.2 μM barcoded forward primer (515F), and 0.2 μM universal Illumina reverse primer (806R) (Caporaso et al., 2012; Walters et al., 2016). Pooled PCR amplicons were gel-purified using a QIAquick Gel Extraction Kit (QIAGEN) and quantified using a Qubit 2.0 fluorometer (Life Technologies Co.). We then checked the quality of the libraries using a 2100 Bioanalyzer DNA 1000 Chip (Agilent Technologies; Santa Clara, California, USA). Library pools were size-verified using a Fragment Analyzer™ Automated CE System (Advanced Analytical Technologies, Inc.; Ames, Iowa, USA) and quantified using a Qubit High Sensitivity Double-Stranded DNA Kit (Invitrogen™, Thermo Fisher Scientific). Laragen, Inc. (Culver City, California, USA) sequenced the purified DNA libraries using an Illumina MiSeq Reagent Kit v2 (500-cycles; Illumina, Inc.) with a 16S rRNA gene library sequencing primer set for 300 base pair (bp) paired-end reads. The raw sequencing data are available at the National Center for Biotechnology Information’s Sequence Read Archive (SRA) website (ncbi.nlm.nih.gov/sra) under the BioProject ID PRJNA752140 (Accession numbers SAMN20584775 to SAMN20584893).

The raw sequencing reads were merged against a minimum overlap length of 190 bp. We conducted quality filtering for the merged reads against a maximum error rate of 1.0% and truncated the filtered reads at a length of 370 bp using USEARCH (version 7) (Edgar, 2010). We subsequently analyzed the reads using QIIME (version 1.9.0) (Caporaso et al., 2010; Caporaso et al., 2011). We identified and removed chimeras using USEARCH61 (Edgar, 2013). We identified the open reference operational taxonomic units (OTUs) using the USEARCH61 algorithm and performed taxonomy assignment using the Greengenes 16S rRNA gene database (the May 2013 release of gg_13_5_99) with a minimum similarity of 97% (DeSantis et al., 2006). Any water sample with less than 1000 sequences was removed from the OTU table before cumulative sum scaling (CSS) normalization.

We determined the abundances of the cyanotoxin-encoding and nutrient metabolism (such as N₂-fixation) genes in the water samples via SYBR® Green qPCR and their expression (i.e., transcript levels) via SYBR® Green RT-qPCR (Table S2) (Chen et al., 2017; Lu et al., 2020; Lu et al., 2019; Zhang et al., 2021b). Specifically, we quantified the abundances and expression of mcyA and mcyE to target cyanobacterial MC producers, mcyG to target MC producers in Microcystis, anaC to target cyanobacterial anatoxin-a producers, and sxtA to target cyanobacterial saxitoxin producers. To monitor N₂-fixation, we quantified the abundances and expression of the N₂-fixation gene nif in genera Dolichospermum (née Anabaena) (Li et al., 2016) and Nostoc. We also quantified the abundances and expression of the inorganic phosphate scavenging gene pstS in Dolichospermum. To facilitate reporting the hosts of these functional genes, we named the qPCR/RT-qPCR targets mcyA-Cya (Cya: cyanobacteria), mcyE-Cya, mcyG-Mic (Mic: Microcystis), anaC-Cya, sxtA-Cya, nif-Ana (Ana: Anabaena, a genus with a current name Dolichospermum), nif-Nos (Nos: Nostoc), and pstS-Ana.

qPCR and RT-qPCR were performed on a QuantStudio™ 6 Flex System (Life Technologies Co.). Each reaction (20 μL) contained (final concentration or volume) 10 μL of 2× qPCR SYBR® Green Master Mix (Life Technologies Co.), 0.2 μM primers (each; Integrated DNA Technologies, Inc.; Coralville, Iowa, USA), and 2 μL of template DNA or cDNA. The thermal cycling conditions were 40 cycles of (95 °C for 15 s; the annealing temperature of 54, 56, or 60 °C for 30 s; and 72 °C for 30 s) and a final hold step at 72 °C for 5 min (Table S2).

The DNA or cDNA was quantified against a standard series constructed in-house. The standard series for mcyA-Cya, mcyE-Cya, and mcyG-Mic were constructed from the genomic DNA of M. aeruginosa. The standard series for anaC-Cya was constructed from the genomic DNA of Aphanizomenon flos-aquae FACHB-1039 isolated from Dianchi Lake, Kunming City, Yunnan Province, China. We also identified Aph. flos-aquae from Hasha Lake (Zhu et al., 2019). The standard series of nif-Nos, nif-Ana, pstS-Ana, and sxtA-Cya were conventional PCR products generated from DNA isolated from water samples from Harsha Lake. Each quantification was run in triplicate, and each qPCR plate contained a triplicate, six-point standard curve with target gene concentrations ranging from 10¹ to 10⁶ copies per microliter (tenfold serial dilution). We checked PCR inhibition by analyzing 10-fold diluted DNA extracts using qPCR and removed qPCR datapoints where significant PCR inhibition was detected following an established protocol (Zhang et al., 2021b).

2.4. Data processing and statistical analysis

qPCR and RT-qPCR data were transformed to log₁₀(GCN·mL⁻¹) (GCN: genome or gene copy number). We performed statistical analysis using R 3.6.1 (R Core Team, 2018). Specifically, we analyzed the diversity and structure of microbial communities using the phyloseq package (McMurdie and Holmes, 2013). We calculated the Pearson correlation coefficients and the associated p-values among various functional genes using the corrplot package (Wei and Simko, 2021). We conducted a TOBIT analysis with left-censored values using the AER package (Kleiber and Zeileis, 2008) to determine the relationship between total MC concentrations and mcy abundances and expression. By generating receiver operating characteristic (ROC) curves and calculating the areas under the ROC curves (AUC) using the pROC package, we made 7- and 14-d lag correlations (95% confidence interval) between total MC concentrations and the abundances and transcript levels of mcy. We then used the ROC curves to predict total MC concentrations and compared the predicted concentrations with health advisory limits of 0.3, 1.6, and 8.0 (and/or 4) μg·L⁻¹ (Lu et al., 2020; Robin et al., 2011). To correlate physicochemical water quality parameters with the abundances and expression of functional genes and total MC concentrations, we performed a canonical correspondence analysis (CCA) using the vegan package (Dixon, 2003; Ji et al., 2019). We analyzed the effects of nutrients and other physicochemical parameters on MC concentrations via the MASS package based on Akaike Information Criterion (AIC) with a backward elimination procedure (Zhang, 2016) to rule out unrelated factors and optimize the multivariance linear regression model. The significance level was 0.05.

3. Results and discussion

3.1. Dynamic cyanobacterial community structures and potential cyanotoxin producers

The microbial community structures were monitored from March to September 2016 at BUOY and EFLS in Harsha Lake. We visualized the similarity among the bacterial community structures in different water samples with an Unweighted Pair Group Method with Arithmetic Mean (UPGMA) dendrogram (Fig. S1). Samples clustered by sampling time for both sites, suggesting significant temporal variations in the cyanobacterial community structures. We speculated that the significant variations were due to environmental variables such as pH, water temperature, light intensity, and nutrient levels. Other reasons or factors could also contribute to the significant temporal changes in the cyanobacterial community structures.

At the phylum level, the bacterial community structures were stable throughout the sampling period with limited variations in RA (Fig. S2). Proteobacteria, Cyanobacteria, Bacteroidetes, Actinobacteria, Planctomycetes, Verrucomicrobia, and Chloroflexi (approximate descending order of RA) were the dominant phyla at both sampling sites. The RA of cyanobacteria was greater between late May and late September than before late May and after late September, indicating significant growth of cyanobacteria during the warm season, especially from early June to late July.

The cyanobacterial community structures at the genus level changed substantially from early June to late July (Fig. S3). Dolichospermum, Synechococcus, Pseudanabaena, Prochlorothrix, Planktothrix, and Microcystis were the dominant genera and potential cyanotoxin producers at both sampling sites. Similar to 2015 (Lu et al., 2020), Microcystis and Planktothrix were the major MC producers in 2016 and more abundant in June and/or July than in other months, indicating the reoccurrence of an HCB event in 2016. In addition, the RA of Microcystis and Planktothrix significantly and positively correlated with each other at both sampling sites (Pearson correlation coefficients ≥ 0.80, p-values < 0.05) (Fig. S4). Therefore, the two major MC producers co-occurred and had a close relationship with each other. However, from early July to mid-November, the RA of Planktothrix was high and much greater than Microcystis (Fig. S3). We speculated that the greater RA of Planktothrix was for two reasons. First, Planktothrix is a better competitor for light (a major driver for phytoplankton growth) than Microcystis (de Araujo Torres et al., 2016; Oberhaus et al., 2007). Second, Microcystis prefers a warmer environment, while Planktothrix is adapted to a broader temperature range (Jöhnk et al., 2008; Paerl and Huisman, 2008). Water temperature in Harsha Lake typically starts decreasing significantly in late August.

The RA of other cyanobacterial genera also changed dynamically over time. For instance, the RA of Dolichospermum increased from early April to early June (peak RA approximately 40%) but subsequently decreased and stabilized starting in early July (RA approximately 10%). The RA of Synechococcus was generally greater after June than in earlier months. From late June to early November, Synechococcus was more abundant than other identified cyanobacterial genera. Previous studies similarly found that the abundances of Dolichospermum and Synechococcus had significant temporal variations in lakes (Becker et al., 2007; Zhang et al., 2020).

The dominant cyanobacteria, such as Microcystis and Synechococcus, grow well in eutrophic freshwater bodies (Akins et al., 2018; Beaulieu et al., 2018; Liu et al., 2019). The relatively high abundances of the dominant cyanobacteria (Fig. S3) were thus a result of the eutrophication in Harsha Lake in 2016 as they were in 2015. Taken together, the dynamic cyanobacterial community structures preliminarily evidenced a reoccurred HCB event and the associated microbial communities in 2016 because of consistent eutrophication in Harsha Lake (Figs. S1 to S3).

3.2. Temporal variations in cyanotoxin-encoding genes and cyanotoxin production

We monitored the abundances and expression (i.e., transcript levels) of multiple cyanotoxin-encoding genes (Table S3) while focusing on three MC-encoding genes (mcyA-Cya, mcyE-Cya, and mcyG-Mic). The qPCR/RT-qPCR target mcyE-Cya covers all major MC-producing cyanobacterial genera (such as Dolichospermum, Microcystis, Planktothrix, and Nostoc) and certain nodularin-producing genera (such as Nodularia) (Jungblut and Neilan, 2006; Lu et al., 2020). The qPCR/RT-qPCR target mcyA-Cya covers fewer genera (mainly MC-producing Microcystis and Planktothrix and possibly other MC-producing genera) (Hisbergues et al., 2003). The qPCR/RT-qPCR target mcyG-Mic is specific to MC-producing Microcystis (Ngwa et al., 2014). In agreement with previous studies in Harsha Lake (Lu et al., 2020) and the Macau storage reservoir (Zhang et al., 2014a), the abundances of these three qPCR/RT-qPCR targets were comparable at each sampling time. In addition, temporal variations in the abundances of these targets evidenced that an HCB occurred from late May to early August 2016 in Harsha Lake (Figs. 1A and 1C), similar to the 2015 HCB in the same lake. Since mcyA-Cya, mcyE-Cya, and mcyG-Mic had comparable abundances and transcript levels during the 2016 HCB (Fig. 1), Microcystis and Planktothrix were the dominant MC producers, while other MC-producing cyanobacterial genera such as Dolichospermum were in low abundances or potentially absent during the 2016 HCB. In addition, we found that Microcystis produced significantly more MCs than Planktothrix because both the abundances and expression of mcyE-Cya and mcyG-Mic were comparable.

Fig. 1.

Abundances (A and C) and transcript levels (B and D) of mcy and nif genes compared with total microcystin concentrations (A and C) at BUOY (A and B) and EFLS (C and D) in Harsha Lake in 2016. GCN: Genome or gene copy number. The error bars represent the standard errors of the means for duplicate measurements. Date format: Month/Day/Year.

The abundances of the three MC-encoding genes (mcyA-Cya, mcyE-Cya, and mcyG-Mic) (Figs. 1A and 1C) varied significantly from late March to late September. Between late March and late May, the abundances of each MC-encoding gene were at or below approximately 1 log₁₀(GCN·mL⁻¹) at both sampling sites. Starting in late May, the abundances of each MC-encoding gene dramatically increased and peaked in mid- to late June at approximately 5.5 log₁₀(GCN·mL⁻¹). The gene abundances then gradually decreased until early August, whereafter the abundances for each MC-encoding gene were stable at approximately 2.5 log₁₀(GCN·mL⁻¹). These significant temporal changes in the abundances of the three MC-encoding genes implied that an HCB reoccurred in 2016 (Lu et al., 2020). To further understand the occurrence of this HCB, we measured the transcript levels of these three MC-encoding genes. The transcript levels followed a similar temporal trend to that of the gene abundances. For instance, the transcript levels also peaked in late June to early July at approximately 2.5 log₁₀(GCN·mL⁻¹) at both sampling sites (Figs. 1B and 1D). Subsequently, the transcript levels decreased and were minimal after early August.

The temporal variations in the abundances and transcript levels of these three MC-encoding genes (mcyA-Cya, mcyE-Cya, and mcyG-Mic) suggested that MC-producing cyanobacteria (mainly Microcystis and Planktothrix, particularly Microcystis) significantly grew between late May and late July. However, the abundances and expression of these MC-encoding genes indicated only the production of total MCs by cyanobacteria but might not indicate the concentrations of MC-producing cyanobacteria. For instance, from late May to early July, the RA of Microcystis and Planktothrix was comparable, while the RA of Planktothrix was high and much greater than that of Microcystis from early July to mid-November (Fig. S3). Even though the RA of Planktothrix was comparable to or even greater than that of Microcystis during the HCB, Microcystis was still the dominant MC producer at the two sampling sites in Harsha Lake in 2016 as determined by the abundances and expression of the mcy genes (Fig. 1). In other freshwater bodies, however, Planktothrix could be a major or dominant MC producer. For instance, in 1995 and 1996 in 55 German freshwater bodies, the concentrations of MCs per unit dry cyanobacterial biomass were highest when P. rubescens was dominant, followed by P. agardhii and Microcystis spp. (Fastner et al., 1999; Tonk et al., 2005).

The temporal trend in MC production was similar to those of the abundances and expression of the three MC-encoding genes (Fig. 1). The significant MC production confirmed the reoccurrence of an HCB event in 2016. The MC concentrations were below the LOD (0.15 μg·L⁻¹) before early June at both sampling sites. Starting in early June, total MC concentrations dramatically increased, peaking at approximately 10 μg·L⁻¹ in mid-June at both sites, which is greater than the health advisory limit for recreational water of 8 μg·L⁻¹ (US EPA, 2019). From mid-June to late July, total MC concentrations dramatically decreased. After early August, total MC levels were below the LOD even though the abundances of the MC-encoding genes were relatively high and stable at approximately 2.5 log₁₀(GCN·mL⁻¹). The undetected total MCs when the MC-encoding genes had relatively high abundances were presumably because these genes were not highly expressed. Indeed, the transcript levels of these MC-encoding genes were minimal after early August. Overall, before total MC concentrations peaked, temporal changes in the abundances of MC-encoding genes, mRNA production by these genes, and total MC concentrations displayed similar patterns. The close correlations among the MC-encoding genes, their expression, and MC concentrations before total MC concentrations peaked confirmed that MC-encoding genes are powerful molecular markers to assess and predict MC production during HCBs (Lu et al., 2020). In addition, total MC levels in three (4.8%) of the 63 water samples (32 for BUOY and 31 for EFLS) exceeded the health advisory limit of 8 μg·L⁻¹ for recreational water (US EPA, 2019).

This study focused on the dominant cyanotoxins (i.e., MCs) and the MC-encoding genes (i.e., the mcy genes) in Harsha Lake. In addition, we assessed the concentrations of anatoxin-a and saxitoxin and genes involved in their production (i.e., anaC and sxtA, respectively) (Chia et al., 2018; Sabart et al., 2015) for two reasons. First, low levels of anatoxin-a (but not saxitoxin) were detected in 2015 from Harsha Lake (data not shown). Second, we used qPCR and RT-qPCR as an early-warning tool to target not only MCs but also other common cyanotoxins in freshwater bodies such as anatoxin-a and saxitoxin.

anaC-Cya had high abundances and was highly expressed from late May to late September (Fig. 2), but it did not produce anatoxin-a to a level greater than the LOD (0.05 μg L⁻¹) at both sampling sites throughout the HCB event. Although sxtA-Cya was detected throughout the HCB, its expression was low or even minimal (Fig. 2), resulting in undetectable saxitoxin concentrations (i.e., < 0.05 μg L⁻¹). The undetected saxitoxin could be presumably because some non-saxitoxin-producing species contain the sxtA gene (Ballot et al., 2010; Le Tortorec et al., 2016; Murray et al., 2011; Stüken et al., 2011; Zhang et al., 2014b), where sxtA might not be transcribed into mRNA (Akbar et al., 2020) or the function of sxtA is not limited to saxitoxin production (Geffroy et al., 2021; Hackett et al., 2013). Additionally, Microcystis as the dominant MC producer in the current study might have suppressed the production of anatoxin-a and saxitoxin (Al-Tebrineh et al., 2010; Chia et al., 2018).

Fig. 2.

Abundances and transcript levels of anaC and sxtA at BUOY (A) and EFLS (B) in Harsha Lake in 2016. The error bars represent the standard errors of the means for duplicate measurements. GCN: Genome or gene copy number. Date format: Month/Day/Year.

3.3. Temporal variations in nutrient metabolism

We tracked the temporal changes in the abundances and expression of two nutrient metabolism genes (the N₂-fixation gene nif and the inorganic phosphate scavenging gene pstS) using three qPCR/RT-qPCR assays (nif-Ana and pstS-Ana specific to Dolichospermum; nif-Nos specific to Nostoc). The abundances of nif-Ana and nif-Nos increased dramatically starting in mid-April and peaked in early June (Figs. 1A and 1C). Afterward, the abundances of nif gradually decreased until mid-August. From mid-August to late August (the end of the monitoring), the abundances of nif significantly increased again. The transcript levels of nif followed a similar pattern. The transcript levels of nif-Ana and nif-Nos significantly increased starting in mid-May and peaked in early June at approximately 5 log₁₀(GCN·mL⁻¹) (Figs. 1B and 1D). The transcript levels of nif then dramatically decreased to approximately 2.5 log₁₀(GCN·mL⁻¹) in mid-June and subsequently rapidly increased again to approximately 4.5 log₁₀(GCN·mL⁻¹) in late June. The transcript levels then gradually decreased until early to mid-August and significantly increased again until the end of the monitoring (late August).

The temporal changes in nif abundances and dissolved nitrogen concentrations were closely associated. For instance, the abundances of nif significantly increased starting in mid-April, and the transcript levels of nif significantly increased starting in mid-May (Fig. 1). NH₄⁺-N concentrations formed significant peaks between late April and late May (Fig. S5), suggesting active N₂-fixation (Bothe et al., 2010; Zehr, 2011). The significant increase in NH₄⁺-N concentrations could have promoted the growth of nitrifiers (phylogenetically not closely associated with cyanobacteria) (Chen et al., 2016b; Hampel et al., 2018; Purkhold et al., 2000), which consume NH₄⁺-N and generate NO₂⁻-N and NO₃⁻-N. Consequently, peaks for the concentrations of NO₂⁻-N and NO₃⁻-N occurred approximately one week after the peaks for the concentrations of NH₄⁺-N. Starting in early June, the concentrations of dissolved NH₄⁺-N, NO₂⁻-N, NO₃⁻-N, and TN gradually decreased, indicating that cyanobacteria (and other bacteria) significantly consumed dissolved inorganic nitrogen (mainly NH₄⁺-N) for growth (Rittmann and McCarty, 2020), decreasing inorganic nitrogen concentrations in this lake.

The abundances of the phosphate transporter gene pstS significantly increased starting in mid-April, peaked in early June, and subsequently gradually decreased until mid-August (Fig. 3). From mid-August to the end of the monitoring (late August), the abundances of pstS significantly increased again. Similarly, the transcript levels of pstS significantly increased starting in mid-May, peaked in early June, subsequently decreased until mid-August, and then significantly increased again. Therefore, pstS was active during the 2016 HCB.

Fig. 3.

Dissolved total phosphorus (TP) concentrations, dissolved total reactive phosphorus (TRP) concentrations, pstS-Ana abundances, and pstS-Ana transcript levels at BUOY (A) and EFLS (B) in Harsha Lake in 2016. GCN: Genome or gene copy number. The error bars (for the abundances and transcript levels of pstS-Ana) represent the standard errors of the means for duplicate measurements. Date format: Month/Day/Year.

The concentrations of dissolved TRP and TP changed significantly over time. The concentrations of dissolved TRP and TP significantly increased starting in late April, peaked in mid-May, and subsequently gradually decreased. The abundances of pstS-Ana were low before mid-April, dramatically increased since mid-April, peaked in early June, and then gradually decreased (Fig. 3). Therefore, pstS-Ana was well regulated by extracellular phosphorus concentrations (Harke et al., 2012; Pereira et al., 2019). For instance, when the concentrations of dissolved TRP and TP peaked in mid-May, the expression of pstS-Ana was suppressed. Afterward, the transcript levels of pstS-Ana increased when the concentrations of dissolved TP and TRP decreased. Therefore, pstS-Ana was more active when external phosphorus concentrations were low but was downregulated when external phosphorus levels were high (Dyhrman and Haley, 2006).

3.4. The effect of nutrient metabolism on cyanotoxin production

Nutrient metabolism is an important cyanobacterial activity that catalyzes and sustains cyanotoxin production and HCBs (Glibert et al., 2018; Lu et al., 2019). Therefore, closely monitoring cyanobacterial nutrient metabolism can help explain the mechanisms, dynamics, and kinetics of HCBs.

This study suggested that active nutrient metabolism, especially N₂-fixation, potentially promoted total MC production for two reasons. First, cyanobacterial nutrient metabolism and cyanotoxin production were closely correlated. The abundances of the MC-encoding gene cluster mcy (i.e., mcyA-Cya, mcyE-Cya, and mcyG-Mic) and the N₂-fixation gene nif (nif-Ana and nif-Nos) were positively correlated (R_Pearson 0.75 to 0.82, p-values < 0.05) (Fig. 4). The transcript levels of mcy and nif also had significant positive correlations (R_Pearson 0.49 to 0.65, p-values < 0.05). Moreover, the abundances (R_Pearson 0.74 to 0.81, p-values < 0.05) and transcript levels (R_Pearson 0.59 to 0.65, p-values < 0.05) of mcy and pstS-Ana had significant positive correlations. In addition to mcy, the genes involved in the synthesis of anatoxin-a (anaC) and saxitoxin (sxtA) had positive correlations with nif (R_Pearson 0.59 to 0.78 for gene abundances, 0.16 to 0.61 for transcript levels, p-values generally less than 0.05) and pstS-Ana (R_Pearson 0.56 and 0.66 for gene abundances, 0.23 and 0.55 for transcript levels, p-values generally less than 0.05) (Fig. 4). A significant positive correlation between the concentrations of two potential cyanotoxin producers (Aphanizomenon and Dolichospermum) and nif abundances existed in Utah Lake (a freshwater lake in Salt Lake City, Utah, USA) (Li et al., 2020). We also identified multiple potential cyanotoxin-producing species in the genera Aphanizomenon and Dolichospermum such as Aph. flos-aquae, D. circinale, D. crassum, and D. ellipsoides from Harsha Lake in 2015 (Zhu et al., 2019). However, a cyanobacterial species could include both cyanotoxin-producing and non-producing strains. The identification and confirmation of cyanotoxin-producing strains should be included in future studies. The significant correlations between nutrient metabolism genes and cyanotoxin-encoding genes suggested that nutrient-related factors (such as nutrient circulation, loading, and concentrations) and cyanotoxin (mainly total MC) production could have an inherent connection in freshwater bodies (Bartoli et al., 2018; Huisman et al., 2018; Lu et al., 2019; Paerl et al., 2011b; Park et al., 2017; Xu et al., 2017).

Fig. 4.

Pearson correlation coefficients between pairwise variables (i.e., gene abundances and transcript levels) for the 2016 harmful cyanobacterial bloom in Harsha Lake. The superscript ^abd indicates gene abundances [log₁₀(GCN·mL⁻¹)]. The superscript ^trans indicates transcript levels [log₁₀(GCN·mL⁻¹)]. GCN: Genome or gene copy number. The *, **, and *** on top of the Pearson correlation coefficients indicate that the p-values are in the ranges from 0.01 to 0.05, 0.001 to 0.01, and 0 to 0.001, respectively.

Second, nutrient metabolism genes became active before increases in the abundances and transcript levels of cyanotoxin-producing genes and total MC concentrations. The initial significant increases in the transcript levels of nif (on May 11th) were approximately three weeks before those of mcy, anaC-Ana, and sxtA-Cya (all on June 01st) at both sampling sites (Figs. 1B, 1D, and 2). Significant growth of N₂-fixing cyanobacteria and increases in the transcript levels of N₂-fixing genes before significant growth of cyanotoxin producers such as Microcystis and production of cyanotoxins are common in eutrophic lakes (Tanvir et al., 2021). The initial increases in the transcript levels of pstS on May 18th (BUOY) and May 11th (EFLS) (Fig. 3) were also earlier than those of mcy, anaC-Cya, and sxtA-Cya. In addition, during the HCB, the abundances and transcript levels of nif-Ana, nif-Nos, and pstS-Ana were generally much greater than those of mcy, anaC-Cya, and sxtA-Cya (Fig. 1, Fig. 2, Fig. 3). Thus, heterocystous cyanobacteria outgrew cyanotoxin (mainly MC) producers, and nutrient metabolism was more active than cyanotoxin production. The more active nutrient metabolism genes compared with cyanotoxin-encoding genes further suggested that nutrient metabolism could trigger cyanotoxin production and stimulate HCBs in freshwater bodies (Lu et al., 2019; Paerl and Otten, 2013). Therefore, controlling nutrient loadings and suppressing cyanobacterial nutrient metabolism, specifically N₂-fixation, are useful strategies for mitigating and preventing HCBs in freshwater aquatic ecosystems (Hamilton et al., 2016; Nwankwegu et al., 2019; Paerl et al., 2016; Paerl et al., 2018).

Even though nutrient-metabolism-related genes may promote cyanotoxin production, these genes are not good predictors of cyanotoxin production for multiple reasons. First, nutrient metabolism (N₂-fixation and inorganic phosphate scavenging) and cyanotoxin production are separate pathways. Second, certain non-N₂-fixing cyanobacterial species can produce cyanotoxins, and certain N₂-fixing cyanobacterial species do not produce cyanotoxins. Third, in the current study, anatoxin-a and saxitoxins remained below the LOD or were even absent after significant increases in the transcript levels of the N₂-fixing genes and the inorganic phosphate scavenging gene. Similarly, even though the nutrient metabolism genes and the genes involved in the synthesis of anatoxin-a and saxitoxins had significant positive correlations (Fig. 4), anatoxin-a and saxitoxins were undetected. Fourth, significant increases in the transcript levels of nif genes might indicate that cyanotoxin (such as MC) production could occur subsequently but might not predict cyanotoxin concentrations (i.e., no clear quantitative correlation). Indeed, because total MC production could be affected by various factors and those factors could affect one another (i.e., intercorrelations), a simple correlation between total MC concentrations and the abundances/transcript levels of nif was not found (data not shown). These factors include but are not limited to light intensity, sulfur concentration, nutrient (i.e., nitrogen and phosphorus) levels, nitrogen-to-phosphorus ratio, iron concentration, water temperature, pH, salinity, turbidity, the presence of xenobiotics, and the interacts between MC producers and their predators and competitors (Jähnichen et al., 2011; Kotak et al., 2000; Pineda-Mendoza et al., 2016; Sevilla et al., 2008). Therefore, nutrient-metabolism-related genes are not useful predictors of MC production.

3.5. mcy as a promising predictor of MC production: evaluating the early-warning system

This study focused on evaluating our recently developed early-warning system (Lu et al., 2020), which uses MC-encoding genes as molecular markers to predict MC production in Harsha Lake. At both sampling sites, the abundances of the three mcy genes (mcyA-Cya, mcyE-Cya, and mcyG-Mic) increased before the initial significant increases in total MC concentrations (Figs. 1A and 1C). Furthermore, the temporal variations of MC concentrations and the abundances/expression of the MC-encoding genes followed similar patterns before total MC concentrations peaked (Fig. 1). Therefore, the MC-encoding genes are promising molecular markers to predict and assess MC production during HCBs in Harsha Lake.

To evaluate the early-warning system, we correlated total MC concentrations with the same-day abundances and transcript levels of the mcy genes using an association model. We also correlated total MC concentrations with the past (7 d earlier) abundances and transcript levels of the mcy genes (i.e., the early-warning system). Specifically, we used TOBIT regression with left-censored values to determine mcy abundances and transcript levels that would indicate the same-day or predict 7-d later total MC concentrations to reach the three health advisory limits. These limits are 0.3 μg·L⁻¹ in drinking water for bottle-fed infants and pre-school children, 1.6 μg·L⁻¹ in drinking water for school-age children through adults (US EPA, 2015), and 8.0 μg·L⁻¹ in recreational water (US EPA, 2019).

For the same-day association model, the correlation coefficients between predicted and observed total MC concentrations when the abundances/transcript levels of mcyA-Cya, mcyE-Cya, and mcyG-Mic were used as the predictors were 0.80/0.82, 0.77/0.81, and 0.87/0.87, respectively (Fig. S6). Therefore, the abundances (determined via qPCR) and transcript levels (determined via RT-qPCR) of mcy are comparably powerful in indicating the same-day total MC concentrations. The same-day association model (confidence level 95%) indicated that total MC concentrations would exceed the health advisory limits of 0.3, 1.6, and 8.0 μg·L⁻¹ if the mean abundances and mean transcript levels of the three mcy genes are greater than 3.70 and 0.94, 4.59 and 1.81, and 5.44 and 2.64 log₁₀(GCN·mL⁻¹), respectively (Fig. S7A). These thresholds for the MC-encoding genes are comparable to those for the 2015 HCB (Fig. S7B) (Lu et al., 2020). Therefore, the abundances and transcript levels of mcy in two consecutive years both significantly correlated with the same-day total MC concentrations, supporting the feasibility of using the mcy genes to assess MC production in Harsha Lake and probably other freshwater bodies. In addition, the World Health Organization (WHO) assumes that each MC-producing cyanobacterial cell contains 0.2 pg of total MCs (US EPA, 2019; World Health Organization, 2003). If all cells with the MC-encoding genes produce MCs, total MC levels would exceed 0.3, 1.6, and 8.0 μg·L⁻¹ when mcy abundances are greater than 3.18, 3.90, and 4.60 log₁₀(GCN·mL⁻¹), respectively. These gene abundances of 3.18, 3.90, and 4.60 log₁₀(GCN·mL⁻¹) are comparable to the ones determined by the same-day association model [3.70, 4.59, and 5.44 log₁₀(GCN·mL⁻¹), respectively]. Therefore, the same-day association model can accurately indicate total MC levels on the basis of the abundances and transcript levels of MC-encoding genes.

Weekly and biweekly samplings are both common strategies for monitoring HCBs and cyanotoxin production in waterbodies. However, it is unknown which sampling frequency is more appropriate in Harsha Lake. We compared the abilities of our novel early-warning system based on 7- and 14-d sampling frequencies in predicting total MC production in Harsha Lake. We developed ROC curves against health advisory limits of total MC of 0.3, 1.6, and 8.0 μg·L⁻¹ for the 7-d (Fig. S8) and 14-d frequencies. The AUC for the 7-d frequency were generally greater than those for the 14-d frequency (Fig. S9). Therefore, monitoring mcy weekly (i.e., a 7-d sampling frequency) through qPCR and RT-qPCR will more accurately indicate MC production during HCBs, especially when HCBs and the production of cyanotoxin are significant (i.e., during the peak stage). However, weekly sampling is more costly and time-consuming, while biweekly sampling could still well predict, assess, and/or indicate cyanotoxin production and the propagation of HCBs. As a result, when HCBs and the production of cyanotoxin are less pronounced (e.g., when water is cooler), biweekly sampling can reduce the cost of monitoring. In other lakes with different depths, volumes, nutrient levels, and water temperatures, the adequate sampling frequencies should be individually determined on the basis of the early-warning system.

We evaluated the early-warning system by predicting total MC concentrations during the 2016 HCB against a 7-d sampling frequency (Fig. 5). The correlation coefficients between simulated and observed total MC concentrations when the abundances/transcript levels of mcyA-Cya, mcyE-Cya, and mcyG-Mic were used as the predictors were 0.70/0.68, 0.65/0.58, and 0.80/0.70, respectively. These relatively large correlation coefficients (≥ 0.65) suggested that the abundances of the cyanotoxin-encoding genes (determined via qPCR) can predict MC production. In addition, RT-qPCR results not only predict cyanotoxin production (correlation coefficients ≥ 0.58) but also the dynamics of an HCB event from its beginning, to its peak, and to its end. Therefore, both gene abundances and transcript levels of mcy are good predictors of total MC production during HCBs in Harsha Lake. This early-warning system indicated that when the mean abundances and mean transcript levels for the three mcy genes reached 3.05 and 0.28, 4.45 and 1.62, and 5.80 and 2.92 log₁₀(GCN·mL⁻¹), total MC concentrations would have a 95% probability to reach health advisory limits of 0.3, 1.6, and 8.0 μg·L⁻¹, respectively, after approximately one week in Harsha Lake (Fig. 6). On the other hand, the MC-encoding genes are not the sole factors determining or affecting total MC concentrations. Future studies might need to consider the impacts of other factors or parameters on MC production to more accurately predict HCBs.

Fig. 5.

Prediction of 7-d lagged (d_i) total microcystin (MC) concentrations (left censored) using the current (d_i-7) abundances (A, C, and E) and transcript levels (B, D, and F) of mcyA-Cya (A and B), mcyE-Cya (C and D), and mcyG-Mic (E and F) in Harsa Lake in 2016. The R's indicate the correlation coefficients between predicted and observed total MC concentrations. For each sub-figure, the sample size is 42.

Fig. 6.

Abundances and transcript levels of mcyA-Cya, mcyE-Cya, and mcyG-Mic that predicted 7-d lagged total microcystin (MC) concentrations at health advisory limits of 0.3, 1.6, and 8.0 μg L⁻¹ in Harsa Lake in 2016. GCN: Genome or gene copy number. The black open and red closed circles indicate the arithmetic means for the abundances and transcript levels, respectively, of the three mcy genes (mcyA-Cya, mcyE-Cya, and mcyG-Mic).

We developed this early-warning system to reduce the frequency of (rather than to replace) direct MC measurements and thus to decrease MC analytical costs. qPCR and RT-qPCR are more cost-effective and easier to perform than direct MC determination assays such as ELISA and liquid chromatography/mass spectrometry/mass spectrometry (Lu et al., 2020). Before HCBs and significant MC production, measuring the abundances and expression of MC-encoding genes via qPCR and RT-qPCR, respectively, is more economically feasible and easier than directly determining total MC concentrations. Therefore, before significant MC production, the frequency of direct measurements of MC concentrations can be reduced, and qPCR/RT-qPCR targeting mcy can be used as the major HCB monitoring approach. Once the abundances and transcript levels of mcy reach the thresholds established by the early-warning system, a series of preventative measures could be implemented to predict and prevent HCBs and minimize the risks of cyanotoxins (mainly MCs). Similar to direct MC analysis methods, which are unable to provide on-site data, qPCR and RT-qPCR also normally require a laboratory. However, the qPCR/RT-qPCR caveat may no longer be an issue with the fast development of portable PCR devices for rapid on-site sample processing and gene abundance/expression determination (Ahrberg et al., 2016; Gou et al., 2018; Kuske et al., 1998; Marx, 2015; Zhu et al., 2020).

Once an HCB event is predicted by our novel early-warning system, actions minimizing the risks of cyanotoxins should be taken such as direct measurements of MC concentrations, cyanobacteria and cyanotoxin removal from waterbodies, advisory notices on recreational uses (e.g., beach bathing and fishing), and implementing enhanced treatment protocols at drinking water utilities. In addition, bank filtration, coagulation, and/or the floc-and-sink assays are promising approaches that can remove cyanobacteria and cyanotoxins from raw surface water (de Lucena-Silva et al., 2019; Romero et al., 2014; Sukenik et al., 2017).

3.6. Effects of physicochemical water quality parameters on MC production

Physicochemical water parameters, especially water temperature and nutrient levels or loading, significantly affect cyanobacterial growth and MC production (Jing et al., 2013; Kaebernick and Neilan, 2001; Paerl et al., 2016). This study supported that warm water could stimulate cyanotoxin production during HCBs. For instance, total MCs were detectable (LOD 0.15 μg·L⁻¹) only when water temperatures reached approximately 20 °C (Figs. 1A, 1C, and S10). In addition, warm water promoted cyanobacterial nutrient metabolism. For example, increases in nif and pstS abundances (Fig. 1, Fig. 3) followed an approximate 10 °C elevation of water temperatures starting in mid-April (10 °C), to late April (18 °C), and to early June (27 °C) (Fig. S10). Previous studies also indicated that warm water (indexing heat resource in a region) significantly promotes and sustains HCBs (Jing et al., 2013; Paerl and Otten, 2013; Paul, 2008). This is presumably because cyanobacteria are well adapted to warm water and their maximal growth occurs at a relatively high temperature (Paerl et al., 2011a; Paerl and Paul, 2012; Xu et al., 2021).

We used multiple linear regression to assess the relationship between total MC concentrations and water quality parameters (Table S4 and Fig. S10), such as water temperature, pH, NH₄⁺, NO₂⁻, TN, and TP. However, the model was insignificant (p-value > 0.05). The insignificant linear correlation can also be concluded from the divergent concentration profiles of total MCs (Figs. 1A and 1C), nitrogen compounds (Fig. S5), and phosphorus compounds (Fig. 3). Other environmental parameters not monitored in this study, possibly including light intensity, wind intensity, flow rate, precipitation, stormwater runoff, nutrient loading, metal ions, and biological agents (predators and competitors), might have more strongly affected total MC production.

We also used CCA to assess the linear correlation between various environmental parameters and MC- and nutrient-metabolism-related factors. These factors include total MC concentrations, the abundances and expression of cyanotoxin-encoding genes (mcy, anaC, and sxtA), the abundances and expression of the N₂-fixation gene nif (nif-Ana and nif-Nos), and the abundances and expression of the inorganic phosphate scavenging gene pstS (pstS-Ana). The constrained CCA explained less than 50% of the total inertia, confirming that some environmental parameters not monitored in this study more strongly affected total MC production and nutrient metabolism. Taken together, except for the mcy genes, commonly measured water quality parameters (such as nutrient-metabolism-related parameters) are not useful predictors of MC production.

The CCA map (Fig. 7) showed that the abundances of mcyA-Cya and mcyE-Cya were positively associated with ORP and water temperature but negatively correlated with the concentrations of dissolved TRP, NH₄⁺, NO₂⁻, and NO₃⁻. The negative correlation between mcyA-Cya/mcyE-Cya abundances and nitrogen compound concentrations could possibly be because these two genes were upregulated under nitrogen stress conditions (Pineda-Mendoza et al., 2016; Zhou et al., 2020). mcyG-Mic abundances and total MC concentrations positively correlated with turbidity, BGA-PC, TOC, and TN, confirming that Microcystis as a non-N₂ fixing genus produces more MCs at higher levels of dissolved nitrogen (Harke and Gobler, 2013; Kaebernick and Neilan, 2001).

Fig. 7.

A Canonical correspondence analysis (CCA) map for the effects of physicochemical water quality parameters on the abundances and transcript levels of functional genes (mcy, anaC, sxtA, nif, and pstS) and total microcystin (MC) concentrations at BUOY in Harsha Lake in 2016.

4. Conclusions

We previously proposed a novel early-warning system that uses cyanotoxin-encoding genes to predict cyanotoxin (mainly MC) production during HCBs in Harsha Lake. To evaluate the system, we examined the HCB event in Harsha Lake in 2016. During the warm season, Microcystis and Planktothrix were the dominant MC producers, the MC-encoding genes had high abundances and were highly expressed, and total MC concentrations were relatively high (e.g., exceeding the US EPA health advisory limit in recreational water of 8 μg·L⁻¹ in 4.5% of the 63 water samples), suggesting the recurrence of an HCB event. Before total MC concentrations peaked, temporal variations in the abundances of the three MC-encoding genes (mcyA-Cya, mcyE-Cya, and mcyG-Mic), transcript levels of these genes, and total MC concentrations displayed similar patterns. In addition, total MC concentrations (time-lag) and the abundances and expression of the MC-encoding genes had statistically significant correlations. Moreover, approximately one week after the qPCR/RT-qPCR signals reached the “actionable” thresholds established by the early-warning system, total MC concentrations would exceed health advisory limits. For instance, when the mean abundances and mean transcript levels of the three mcy genes reached 3.05 and 0.28, 4.45 and 1.62, and 5.80 and 2.92 log₁₀(GCN·mL⁻¹) in Harsha Lake in 2016, total MC concentrations would reach health advisory limits of 0.3, 1.6, and 8.0 μg·L⁻¹, respectively, after approximately one week. These findings indicated that cyanotoxin-encoding genes are powerful molecular markers to predict MC production during HCBs in Harsha Lake, further supporting our qPCR/RT-qPCR-based early-warning system. Other than the MC-encoding genes, commonly measured water quality parameters (such as nutrient-related parameters), are not useful predictors of cyanotoxin production. Direct and frequent (i.e., weekly or even daily) determinations of MC concentrations along with additional measures to control, mitigate, and/or prevent HCBs should be implemented once the early-warning system indicates that the qPCR/RT-qPCR signals reach these thresholds.