Use of qPCR and RT-qPCR for Monitoring Variations of Microcystin Producers and as an Early Warning System to Predict Toxin Production in an Ohio Inland Lake

Abstract

Public concern over cyanobacterial blooms has increased due to their higher frequency of occurrence and their potential ecological and health impacts. Detection of microcystin (MC) producers (MCPs) using qPCR and RT-qPCR allows for the rapid identification of blooms by combining specificity and sensitivity with a relatively high throughput capability. Investigation of MCP population composition (correlation, dominance), toxin gene expression, and relationship to MC concentration was conducted using a panel of qPCR assays targeting mcyA, E and G on weekly and daily water samples collected from an Ohio inland lake. Further, these data were used to develop early warning thresholds for prediction of MC concentrations exceeding the US EPA Health Advisory cutoff value (>0.3μg L⁻¹) using receiver operating characteristic curves and tobit regression. MCP Microcystis genomic copy number made up approximately 35.1% of the total Microcystis spp. and was the dominant toxic subpopulation of MCPs and remained a stable proportion of the total (30-40%) regardless of sampling dates and sites. The expressed MCPs were 0.2% of the extant genomic copy numbers, while toxic Microcystis had higher expressed proportion (0.5%) than that of toxic Planktothrix (0.04%). Microcystis toxin genes increased in June and July, but decreased in August and September along with similar trends of cell replication. Quantities of both RT-qPCR and qPCR followed the same trend and were highly correlated with MC-ADDA, while RT-qPCR not only reflected the active toxin genes or toxic species, but also indicated the beginning and ending of toxin production. A one-week early warning of MC with a cutoff value (>0.3μg L⁻¹) was also made based on signaling of qPCR and RT-qPCR using receiver operating characteristic curves. Together, this study reports the use of qPCR and RT-qPCR as an early warning system (up to 7 days prior to) of extant and MC producing potentials (false positive rates ~70% for LC-MS/MS or ELISA) during a toxic algal bloom, with a significant predictive power of 50%−60% and 30%−40%, respectively.

1. INTRODUCTION

The recent increase in cyanobacterial harmful algal blooms (CyanoHABs) has caused worldwide concerns for human and ecosystem health due to the production of toxins by bloom species. Cyanobacterial toxins are potent actors with biological activity including effects on hepatocytes and neurotransmitters, for which, action levels have been established for microcystin (MC). A notable event illustrative of MC impact occurred in Toledo, OH in August 2014. Residents were asked to stop using tap water due to MC concentrations >1 μg/L measured in the finished water (Wynne and Stumpf 2015). The US Environmental Protection Agency has since issued 10-Day Drinking Water Health Advisories of 0.3 μg/L MC for pre-school age and younger, and 1.6 μg/L MC for school-age and older populations (USEPA 2015). The draft primary contact recreational guidance value is 4 μg/L (USEPA 2016).

Toxic cyanobacterial blooms are difficult to predict because the mechanisms of bloom formation and toxin production are unknown. Monitoring of CyanoHABs has relied on algal species/counts, biomass, and remotely-sensed imagery, among other parameters, and are effective for observation but are limited in short-term predictive utility (Recknagel et al. 1997, Allen et al. 2008, Downing et al. 2001). Chlorophyll or chlorophyll-a have been commonly used as indicators of cyanobacteria or phytoplankton in predictive models (Recknagel et al. 1997, Allen et al. 2008, McQueen and Lean 1987). Based on chlorophyll-a data of five blue-green algal species, Recknagel et al. (1997) developed phytoplankton models (ANNA model) for predicting phytoplankton succession and occurrence of cyanobacteria (Recknagel et al. 1997). Recently, a phycopigment remote sensing method used to monitor CyanoHABs has been developed, in which spectral band analysis allows for the detection of light absorption indicative of phycocyanin, the primary light collection pigment in cyanobacteriawhen present in sufficiently high quantities (Randolph et al. 2008, Kutser et al. 2006). However, those methods lack sensitivity due to the need of minimum biomass to indicate a bloom and rely on relatively low frequency collection rates and low spatial resolution, limiting their utility as early warning tools. These methods also have no power to reflect species composition, specific MC producers (MCPs), nor toxin production. Toxin analytic methods are also of limited predictive power as those methods are retrospective measuring extant toxins at measurable levels. PCR assays targeting mcyA~G genes for potentially toxic Microcystis have been developed to monitor MC and MCPs directly (Baker et al. 2001, Baker et al. 2002, Nonneman and Zimba 2002, Pan et al. 2002, Jungblut and Neilan 2006). Using PCR methods, we examined the presence of toxin (microcystin, anatoxin-a, saxitoxin, cylindrospermopsin and nodularin) producers and conducted phylogenetic analysis for those detected in water samples from William H. Harsha Lake, OH (Harsha Lake), which showed that MCPs were often dominant (Chen et al. 2017). Furthermore, several quantitative applications for toxin producers have also been developed and used. For example: the mcyE assay covers all potential MCPs including Anabaena, Microcystis, Planktothrix and the nodularin synthetase gene cluster (Jungblut and Neilan 2006); the mcyA assay covers less genera (Hisbergues et al. 2003), and the mcyG (Ngwa et al. 2014) and mcyE (Tillett et al. 2001) (Furukawa et al. 2006) assays were found to be sensitive and specific to the Microcystis and Planktothrix genera, respectively. These genes were thus considered suitable targets for qPCR-based estimation of toxigenic MCPs including most of the MCP species worldwide, due to their roles in the synthesis and incorporation of vital amino acids into the microcystin structure (Tillett et al., 2000; Vaitomaa et al., 2003). Cyanobacterial specific qPCRs have also been broadly used in the studies of cyano-ecology and cyanotoxin assessment. For example: using qPCR, the spatial and temporal variation in proportions of MCP cyanobacterial subpopulations in natural conditions were determined (Hotto et al. 2007, Briand et al. 2009, Rinta-Kanto et al. 2005, Sabart et al. 2015); toxic Microcystis spp. and their microcystin genotypes were characterized in western Lake Erie, OH (Rinta-Kanto et al. 2005); toxic Microcystis sp. was quantified based on the mcyA gene (Furukawa et al. 2006, Kurmayer and Kutzenberger 2003); and a relative abundance of potential to non-potential toxic genotypes in water was also examined (Briand et al. 2009, Koskenniemi et al. 2007, Briand et al. 2008, Rinta-Kanto et al. 2009, Orr et al. 2010). The development and application of qPCR to detect cyanobacteria have been fundamental to better determine the occurrence of CyanoHABs and understand the dynamics of these toxic populations during bloom formation.

In addition to determining the occurrence and quantity of toxigenic cyanobacteria, it is essential to examine the active components including toxin gene expression variability and their relationship to measured toxin concentrations. Quantification using reverse transcription quantitative polymerase chain reaction (RT-qPCR) as a standard technique has been successfully used for cultured cyanobacteria (Pinto et al. 2012, Barón-Sola et al. 2016). Barón-Sola et al. (2016) examined relationships between cylindrospermopsin production and toxin gene expression in Aphanizomenon ovalisporum and found positive correlations between expressions of four toxin genes and toxin production by RT-qPCR. However, the reports on detection and quantification of specific environmental cyanobacterial gene expressions are limited (Sipari et al. 2010). Active mcyE gene expression detected by RT-qPCR was found to be associated with microcystin concentrations for Lake Tuusulanjärvi samples (Sipari et al. 2010). The relationships between qPCR (extant quantity) and RT-qPCR (expression of MC-producing gene), and toxin concentrations are the basis for determining toxigenic cyanobacteria and their dynamics, understanding environmental drivers, and further, develop an accurate early warning system for detecting CyanoHABs. The aim of this study was to determine the quantity and dominance of MCPs; examine their population ratios, expression of toxin related genes, and observed toxin concentration; and finally, evaluate their potential role as the foundation for developing an early warning system for microcystin production in Harsha Lake.

2. MATERIALS AND METHODS

2.1. Sample collection

Samples were collected from Harsha Lake for biological and MC analysis. Among four sites (Table S1), three sites were located in the western basin (EMB, BUOY, and EFLS) and one site in the eastern basin (CGB) near the point of inflow of the East Fork of the Little Miami River. Weekly (from May to October) and daily (during the bloom: entire month of June) samples were collected at each site. Water samples for each analysis presented were subsampled from a 5L volume collected at a depth of approximately 6-12 inches in a Polyethylene Terephthalate Glycol (PTEG) vessel. For qPCR/RT-qPCR analysis, a 1000 mL subsample was collected and 100-200 mL were filtered using EMD Millipore Durapore™ membrane filter (0.40 μm, MilliPore, Foster City, CA). The filter with concentrated particles from water was inserted in a 1.5 mL micro-tube containing a 600 μL RLT plus buffer (QIAGEN, Valencia, CA) and stored at −80 °C. The filter was disrupted and lysed using a Mini-Beadbeater-16 (BioSpec Products, Inc., Bartlesville, OK) for 2 x 30 s. The mixture was then centrifuged at 10,000 x g for 3 min, the supernatant was carefully transferred to a new sterile tube and DNA and RNA was extracted and purified using AllPrep DNA/RNA (QIAGEN, Valencia, CA) per manufacture’s instruction. DNA/RNA concentrations were estimated with a Nanodrop ND-1000 Spectrophotometer (NanoDrop Technologies, Inc., Wilmington, Delaware). The extracts were stored at −80 °C until qPCR, RT-qPCR, and PCR were performed at the same time with the same prepared reagents.

2.2. qPCR and RT-qPCR

To amplify the major cyanobacterial groups or genera revealed from microscopic and Sanger chemistry sequencing data (Chen et al. 2017), SYBR Green based qPCR assays were selected from previously published primers as listed in Table S2. The assays used passed the following criteria: 1) similar parameters of qPCR assays (i.e. short length of PCR products and approximate Tm 60 °C); 2) high specificity: checked in silico sequence search, tested against various cyanobacterial strains (Table S3) and confirmed by disassociation curves on all the samples; 3) high sensitivity: tested to be high enough to detect the cyanobacterial level observed prior to bloom (Table S2). The general assays were to amplify total cyanobacteria (CYA) (Nübel et al. 1997), Microcystis (MIC) (Neilan et al. 1997), Planktothrix (PLA) (Baxa et al. 2010), Cylindrospermopsin (Cy_rpo) (Wilson et al. 2009) and Nostoc (NTSf/1492R) (Neilan et al. 1997, Moffitt et al. 2001). The general MCPs were amplified using the assay mcyEcya and mcyAcya, of which the former was used for broad coverage including all potential MCPs: Anabaena, Microcystis, Planktothrix and the nodularin synthetase gene cluster (Jungblut and Neilan 2006), while the latter covered only toxic Microcystis, Planktothrix and Anabaena (Hisbergues et al. 2003), and were used for comparing and confirming data obtained by both assays. For Microcystis, the assay mcyAmic (Tillett et al. 2001, Furukawa et al. 2006) and mcyGmic (Ngwa et al. 2014) were used. The toxic Planktothrix was amplified using mcyEpla (Vaitomaa et al. 2003, Rantala et al. 2006). The nonMCP assays were used for examining the dominance of MCPs. Those assays were not only tested to be specific and sensitive, but were also validated in our studies using the procedures described in the methods under SYBR green qPCR conditions. The set of assays were considered suitable targets for qPCR-based estimation of toxigenic genera.

To run qPCR against DNA for occurrence and quantification and RT-qPCR for examining activities of targeting genes, reverse transcriptions were prepared while preparing qPCR reactions. Total RNA was transcribed to cDNA using a High Capacity cDNA reverse transcription kit (Life Technologies Co., Carlsbad, CA) following manufacturer’s instruction after removing contaminated DNA from total RNA extracts using TURBO DNA-free™ (Life technologies). The qPCR reaction mixtures (20 μL) contained 10 μL 2× qPCR SYBR Green Master Mix (Life Technologies), 0.2 μM primers (final concentration) and 2 μL of template DNA. Initial DNA treatment consisted of 50 °C for 2 min with UNG (Uracil-N-Glycosylase) to prevent carryover contamination, followed by 95 °C for 10 min for DNA denaturing. The quantification cycling protocol using a QuantStudio™ 6 Flex system (Life Technologies) was 40 cycles at 95 °C for 15 s and at 60 °C or at the melting temperature (°C) specified by the assay developers referred in Table S2 or 30 s with an extension at 72 °C for 30 s and a final hold step at 72 °C for 5 min. DNA and cDNA copies were quantified relative to a standard curve using genomic DNA and some using qPCR products (Church et al. 2005). Each sample was run in replicate and each qPCR plate contained a duplicate six-point standard curve with values ranging from 10¹–10⁶ copies. The standard curves of targets were constructed with the culture of M. aeroginosa, Planktothrix, Anabaena, C. raciborskii and Nostoc.

For QA/QC purposes, each DNA extract was assayed for potential qPCR inhibitors with 10-fold dilution vs. neat DNA and with the addition of the TaqMan Exogenous Internal Positive Control Reagents (a VIC-labeled probe) (Life Technologies). For cDNA, in addition to examining inhibition, DNA contamination was also examined by running qPCR against treated RNA using TURBO DNA-free™ (Life Technologies). Genome copy numbers (gns) for each sample were calculated using an equation generated from the average of the standard curves in each plate. The final gns were multiplied by a dilution factor and a water-filtering factor to present data as gn L⁻¹ of filtered lake water. As mentioned in the above for MCPs and the dominant Microcystis, qPCR using the assays targeting two different genes, respectively, were performed to compare and confirm the qPCR signals from the same samples.

2.3. Analysis of microcystin (MC)

A modified version of EPA method 544 was used to quantify MC congeners using liquid chromatography-tandem mass spectrometry (LC-MS/MS). A subsample volume of 100-mL was collected in PETG vessels and preserved with 2 g/L 2-chloroacetamide (microbial inhibitor), 0.35 g/L ethylenediaminetetraacetic acid trisodium salt (inhibits binding to metals), buffered with Trizma (pH 7.2 ± 0.2) and kept at 4 °C until extraction. Water samples (fortified with the surrogate, ethylated MC-LR, d₅) were passed through 45 mm inner diameter, 0.8 μm pore size Nuclepore filters and both the filtrate and the filter were collected for each sample. The filter was placed in a solution of methanol containing 20% reagent water and held for at least one hour at −20 °C to release the intracellular toxins from cyanobacteria cells captured on the filter. The liquid was drawn off the filter and added back to the 100-mL aqueous filtrate. The 100-mL sample (plus the intracellular toxin solution) was passed through a 6 mL, 150 mg Waters Oasis HLB solid phase extraction (SPE) cartridge to extract the cyanotoxins and surrogate. Cyanotoxins were eluted from the solid phase with 10 mL of methanol containing 10% deionized water. The extract was concentrated to dryness by nitrogen evaporation in a heated water bath (60 °C), and then adjusted to a 1-mL volume with methanol containing 10% deionized water with addition of the internal standard (Cyclosporin-A, ¹³C₂, d₄). The conditions used on a Waters Micromass Quattro Premier triple quadrupole mass spectrometer equipped with a Waters Acquity Ultra-Performance Liquid Chromatograph (LC-MS/MS) to achieve separation and quantitation of the cyanotoxins are shown in Table S4.

Enzyme Linked Immunosorbent Assay (ELISA), was used to determine total MC-ADDA (PN 520011, Abraxis, Warminster, PA). The assay quantifies the α-amino acid ADDA (all-S all-E)-3-Amino-9-methoxy-2,6,8-trimethyl-10-phenyldeca-4,6-dienoic acid). Subsamples were collected in glass and frozen at −20 °C until processing and analysis. Extracellular concentrations were measured by subjecting each sample to 3 freeze-thaw cycles. Assays were performed manually and analyzed using a Biolog Microstation plate reader (Biolog, Hayward, CA).

2.4. Data analysis

Statistical analyses were performed on log₁₀ transformed data. Pearson correlations, comparisons with multiple comparison procedures, linear analysis under PROC GLM procedures were performed using SAS 9.4 Package for Windows (SAS Institute, Cary, NC). Considering CyanHAB and MC production were developed from initial stage to stable growing stage, and daily samples were collected in June, the correlations, which were the basis for early warning of potential MC production vs. qPCR/RT-qPCR signals, were focused on this month only. ELISA analysis resulted in left and right censored values as samples with MC-ADDA concentrations > 5μg L⁻¹ were not diluted for reanalysis. LC-MSMS data were left-censored at the minimum reporting level. Maximum Likelihood Estimation was used to correlate MC-ADDA concentration with genome copy numbers. Specifically, tobit analysis was performed using the AER package in R (Kleiber and Zeileis 2008, Team 2013) to calculate spearman’s rank correlation and develop regression models of the dependent variable MC concentration vs. the independent variable genome copy numbers. Thresholds of genome copy numbers resulting in MC concentration meeting guidance criteria of 0.3 and 4 μg/L were calculated using the observed tobit model.

The ability of qPCR results to predict a microcystin event in exceedance of guidance criteria was evaluated using logistic regression on one-week lagged data with respect to two qPCR assays: mcyEcya and mcyAmic for both DNA and RNA. SAS Proc Logistic procedure: log[p/(1-p)] = a + bx, where p = probability of exceedance 7 days hence, x = current qPCR values. Exceedance events based on US EPA 10-day health advisory (HA) level for microcystin in drinking water (0.3 ug/L) (USEPA 2015) were modeled against each candidate indicator. Exceedance results from two measurement methods, LC-MS/MS and ELISA, were considered separately as outcomes. For each 2×4 combination of measurement method and indicator a receiver operating characteristic (ROC) curve (Savino et al. 2009) was generated from observed indicator values with sensitivity (Se = true positive rate) and 1-specificity (1-Sp = false negative rate) as coordinates. An optimal cutoff value, c_o, was estimated based on the Youden Index, a frequently used criterion for screening test accuracy (Youden 1950), J=max[Se(c)+Sp(c)-1 over observed indicator values, c. Predictive power was measured as the area under the curve (AUC) compared to 0.5 for a coin toss. Statistical significance implies p-value > 0.05.

3. RESULTS

3.1. QPCR and RT-qPCR analyses of cyanobacteria in Harsha Lake

3.1.1. Occurrence and dominance of MCPs

All assays used for SYBR Green qPCR in field samples proved to be highly specific after the analysis of disassociation curves and sequences of their PCR products and sensitive with detection limits ranging from 5 to 10 cells per reaction (Table S2). Cyanobacteria were detected in every sample (100%) during the sampling period (n=499). RT-qPCR occurrence of 16S rRNA genes targeting cyanobacteria and Microcystis, was equal to or higher than that of qPCR (Fig. 1), indicating greater sensitivity in the RT-qPCR detection of 16S rRNA genes than qPCR. The targets representative of MCPs were, in order of greatest occurrence, 98% (mcyEcya) and 97% (mcyAcya ). MCP Microcystis occurred most frequently (mcyGmic: 89% and mcyAmic: 89%), followed by Planktothrix (mcyEpla: 87%) (Fig. 1), indicating that all or most samples (≥97%) contained MCP genotypes. Overall, Microcystis (1.5±2.0×10⁵ gn L⁻¹) and Planktothrix, (1.4±1.5×10⁵ gn L⁻¹) were the dominant genera, accounting for 10% of the total examined cyanobacteria (3.0±2.4×10⁶ gn L⁻¹) in surface water (Table S5, Fig. S1). There was no significant difference in the qPCR data assayed among sites (BUOY, EMB, CGB and EFLS), indicating cyanobacterial distributions were horizontally even and they could be representative of general surface water (Table S5).

Fig. 1.

Occurrence of detectible cyanobacteria (CYA), Microcystis (MIC) and microcystin producers: total (mcyEcya and mcyAcya), toxic Microcystis (mcyGmic and mcyAmic) and toxic Planktothrix (mcyEpla), using qPCR and RT-qPCR in 2015 summer (pooling data from the four sites, n=360).

3.1.2. MCP’s relationships

The quantities of MCPs in the surface water samples were 5.6±5.8×10⁴ gn L⁻¹. Specifically, MCP Microcystis (5.1±6.8×10⁴ gn L⁻¹) were approximately 35.1% of total Microcystis and 64.2% of MCPs, while MCP Planktothrix (4.1±5.7×10² gn L⁻¹) was 0.5% of total Planktothrix and 1.9% of total MCPs, indicating Microcystis spp. were the major MCPs (Table S5). There were high correlations between MCPs and total corresponding categories. For example, the correlations between MCPs and cyanobacteria (R_mcyEcya-CYA=0.90), and Microcystis (R _mcyEcya-MIC =0.95), Microcystis MCP and total Microcystis (R_{mcyAmic or mcyGmic} =1.00), and Planktothrix MCP and total Planktothrix (R_mcyEpla-PLA =0.80) were all significantly correlated (Table 1). The ratio of MCP subpopulations (mcyAmic or mcyGmic) to Microcystis (MIC) (Table 2) were not correlated to the whole population (R _mcyAmic/MIC = −0.59, R _mcyGmic/MIC = −0.37), although the ratios trended lower with increasing Microcystis spp., indicating a relatively constant ratio of MCP strains to Microcystis population (Fig. S2).

Table 1.

Pearson correlations (R) between variables based on logarithmic transformation based on data (n=235, P<0.0001) in June, 2015

Table 2.

Overall ratios of genome numbers between RT-qPCR to qPCR indicated by “-r” targeting16S rRNA, mcyA, mcyE and mcyG genes of cyanobacteria (CYA) and Microcystis (MIC), and between two qPCR indicated by “to”

Variable	N	Mean	Std Dev	Minimum	Maximum
CYA-r	392	9.50	92.43	0.00	1827.54
MIC-r	348	1454.38	1196.07	0.05	9823.16
mcyEcya-r	392	0.00	0.02	0.00	0.41
mcyAcya-r	384	0.01	0.02	0.00	0.22
mcyAmic-r	348	0.01	0.01	0.00	0.12
mcyGmic-r	348	0.00	0.01	0.00	0.05
mcyGmic to MIC	348	0.33	0.14	0.13	1.49
mcyAmic to MIC	348	0.35	0.10	0.11	1.17
mcyGmic to mcyEcya	392	0.60	0.43	0.00	3.96
MIC to CYA	392	0.10	1.02	0.00	20.27

3.1.3. Expressed and non-expressed MCPs

To examine whether all extant genes were also expressed, the occurrence of detectible signals of qPCR and RT-qPCR were measured and are shown in Fig. 1. MC-producing genes were not only present, but also actively expressed, especially for MCP or nonMCP Microcystis, in most samples (Fig. 1). The ratios of expressed versus extant genome copy numbers of MCPs varied with the range of 0-0.41, and they trended to increase in mid-June, peaked from late June to mid-July and decreased in late July in all four sites (Fig. S3). Specifically, expressed genome numbers of MCP Microcystis and Planktothrix were 0.8% and 0.1% of their extant numbers, respectively, indicating much lower expression, compared to replication (16S rRNA gene) of cyanobacteria (9.5±92.4 fold) and Microcystis (1454.4±1196.1). It should be noted that expressed Planktothrix were much lower than expressed Microcystis, which was 8 times higher than the former (Table 2). The expressions of Microcystis toxin genes reflected by monthly fold changes increased in June and July (1.0 to 81.8) compared to May or June (Table 3) but decreased in August and September (−0.8 to −1.0). Similar trends of 16S rRNA indicated cell replications decreased as well (−0.9 to −1.0). The expression of Planktothrix mcyE increased across most samples, though at a low level. There were also significantly positive correlations between qPCR and RT-qPCR of MCPs (R_mcyEcya/rt=0.82, R_mcyAcya/rt=0.78), Microcystis (R_mcyAmic/rt =0.88, R_mcyGmic/rt =0.80). The expressed Microcystis MCPs were more highly correlated to extant genes than the expressed cyanobacteria based on mcyA gene, indicating higher activities of toxic Microcystis during the CyanoHAB forming (Table 1).

Table 3.

Monthly fold change (current/previous −1) of RT-qPCR (-rt) targeting16S rRNA, mcyA, mcyE and mcyG genes of cyanobacteria (CYA), Microcystis (MIC) and Planktothrix (pla)

Variable	June/May		July/June		July/August		August/September
	mean	Std	mean	Std	mean	Std	mean	Std
CYA-rt	1.0	0.1	0.8	0.1	−0.3	−0.2	−0.8	0.8
MIC-rt	24.6	10.1	1.4	0.2	−0.9	−0.8	−1.0	1.0
mcyEcya-rt	4.6	2.4	1.4	0.9	−0.5	−0.6	−1.0	1.0
mcyAcya-rt	6.6	3.5	2.1	0.7	−0.8	−0.8	−0.9	0.9
mcyAmic-rt	81.8	62.0	1.0	0.0	−0.8	−0.7	−1.0	1.0
mcyGmic-rt	15.6	7.7	2.1	0.6	−0.7	−0.7	−1.0	1.0
mcyEpla-rt	8.8	2.4	1.3	0.5	0.1	0.3	−0.5	0.6

3.1.4. Variation of MCPs

As stated above, MCPs were present during the entire summer sampling period with a prolonged CyanoHAB from early June until mid-August, showing a significant polynomial trend (Fig. 2). There were significant differences of MCPs and their components in genome copy numbers over time as shown in Table S6 for the monthly means and variations. Generally, those in June and July were higher than in August and September, and were significantly higher from June to September than May. This trend followed the bloom of Microcystis, which was highest in July and lowest in September. In May, there was no detection of Microcystis, but significant detection for other MCPs (showing mcyAcya < mcyEcya) (Table S6). The variations of MCPs trended closely to MCP Microcystis, showing a small peak in late May and several peaks in June and July, indicating Microcystis was the main MCP. The presence of toxic Microcystis based on mcyAmic rose to an average of 1.4×10⁵ gn L⁻¹ and maintained relatively stable values with variance 8.2×10⁴ gn L⁻¹ in July (Table S6). RT-qPCR values indicating the expression of the toxin gene mcyA, (assay mcyAmic) increased in June through July, followed by a subsequent decreased. Generally, RT-qPCR also showed a significant polynomial trend (Fig. 3). There were no significant expressions of MCPs in May and September.

Fig. 2.

Variations of log₁₀ (gn mL⁻¹) of microcystin producers: total (mcyEcya in orange and mcyAcya in blue), toxic Microcystis (mcyGmic in black and mcyAmic in green) and toxic Planktothrix (mcyEpla in purple) in BUOY, EFLS, CGB and EMB in 2015 summer. The thin dash lines indicate the polynomials.

Fig. 3.

Variations of log₁₀(qPCR: gn mL⁻¹) (blue line) and log₁₀(RT-qPCR: gn mL⁻¹) (red line) targeting Microcystis mcyA (mcyAmic), MC by LC-MS/MS (green line) and MC-ADDA by ELISA (black line) levels for total raw water microcystin in raw water of the four sites.

3.2. PCR as an early warning indicator of a MCP bloom in Harsha Lake

3.2.1. Correlations between qPCR/RT-qPCR and MC

Correlations between MC-ADDA and toxin gene qPCR and RT-qPCR for total MCPs and Microcystis, respectively, were highly positive (Fig. 4), except for Planktothrix. Variation in observed mcyEcya and mcyAmic qPCR data was less than that of RT-qPCR due to higher variation in gene expression than extant DNA. Measured toxin concentrations tracked RT-qPCR, during the period (June) of rapidly increasing MCP gene expression, demonstrating the linkage between MCP genes and toxin production. Toxin production peaked in mid-July, as the bloom entered a stationary growth stage. This was also observed in steady activities of toxin genes (Fig. 3). There were significantly positive correlations between MC-ADDA and qPCR (R ≥0.84) and RT-qPCR (R ≥0.74) of MCP cyanobacteria (mcyEcya and mcyAcya) and MCP Microcystis (mcyAmic and mcyGmic). It should be noted that the data ranges used to express the relationships were narrower in qPCR (4-5.5 logs) than RT-qPCR (1-2.8 logs) (Fig. 4). Censored regression modeling demonstrated the relationship between MC concentrations and qPCR copy numbers or RT-qPCR transcripts (Fig. 4, Table 4). The qPCR data generally had higher correlation and generalized R² values. Prediction of qPCR copy numbers at guidance criteria (threshold) is found in Table 4. The correspondence between MCP genome copy numbers / gene transcripts and the three thresholds are ≥ 10⁴ /10 gn L⁻¹ at 0.3 μg L⁻¹ threshold, ≥ 10⁵ / 100 gn L⁻¹ at 1.6 μg L⁻¹ threshold and > 10⁵ / 1000 gn L⁻¹ at 4 μg L⁻¹ threshold, respectively. Considering the importance of starting and ending points of a CyanoHAB, it seems more direct and sensitive to use RT-qPCR to indicate the production and a progress of developing MC or MCPs. Compared to general MCPs and Microcystis, the weak correlations between MC-ADDA and mcyEpla qPCR or RT-qPCR (R_{mcyEpla qPCR} =0.44 or _RT-qPCR=0.32) indicated the less importance of Planktothrix, especially in potential production of MC (Fig. S7).

Fig. 4.

Regressions and correlations of log₁₀(MC-ADDA) with qPCR (upper panel) and RT-qPCR (lower panel) of microcystin producing genes (mcyE and mcyA) for cyanobacteria (mcyEcya) and Microcystis (mcyAmic) of pooling data from the four site (The dots in different colors indicate levels of detection for MC-ADDA: MDL-method detection level, MRL-minimum reporting limit, UCL-upper confidence limit and report limit.).

Table 4.

Estimated genomic copy numbers based on qPCR corresponding to criteria thresholds using ELISA

Marker	Type	0.3 μg L−1	1.6 μg L−1	4 μg L−1
CYA	qPCR	5.73	6.43	6.81
CYA	RT-qPCR	5.99	6.70	7.09
MIC	qPCR	3.56	4.86	5.58
MIC	RT-qPCR	6.69	7.93	8.61
mcyEcya	qPCR	3.74	4.59	5.06
mcyEcya	RT-qPCR	0.39	1.71	2.43
mcyAcya	qPCR	3.51	4.41	4.90
mcyAcya	RT-qPCR	0.70	1.91	2.57
mcyAmic	qPCR	3.13	4.39	5.08
mcyAmic	RT-qPCR	0.99	2.12	2.74
mcyGmic	qPCR	3.08	4.34	5.04
mcyGmic	RT-qPCR	0.42	1.68	2.36

3.2.2. Receiver operating characteristics (ROC) curves

To assess the utility of qPCR data as an early warning predictor for MC production, logistic analysis of instances of microcystin concentration exceeding HA levels (0.3 μg L⁻¹) as measured by either LC/MS/MS or ELISA was performed. Results indicate significant relationships between MC concentrations and mcyEcya and mcyAmic qPCR/RT-qPCR assay results from 7 days prior (Table 4). mcyEcya and mcyAmic are equivalent in terms of their predictive accuracy based on ROC curve (Fig. 5A–B) with a predictive power of about 60% (qPCR) and 40% (RT-qPCR) with respect to LC/MS/MS exceedance and 50% (qPCR) and 30% (RT-qPCR) with respect to ELISA.

Fig. 5.

Receiver operating characteristics curve for screening of future microcystin concentration in exceedance of MRL (0.3 μg L⁻¹) as determined by qPCR and RT-qPCR targeting mcyA and mcyE genes of cyanobacteria (cya) and Microcystis (mic) seven days prior (pooling data from the four sites).

Over the period of the study, 7-day advance mcyEcya qPCR prediction resulted in negative results among 118 of 241 samples. This represents a 49% reduction in number of analyses required and if used as a screening method, it could reduce monitoring expenses. The false negative rate was 4%, 5 of the 48 observed exceedances were missed. Similar results hold for toxic Microcystis specific assay mcyAmic qPCR/RT-qPCR. Correspondingly, with respect to ELISA-based monitoring, there were 128 negatives (a 47% reduction in analyses) and 11 of 78 (14%) ELISA-based exceedances missed. Here we only used the Youden criterion for cutoff, which may or may not be optimal depending on factors such as relative cost, expected benefits, and estimated likelihood of exceedance based on historical experience.

Models of the logistic modeling results indicate that when indicators exceed their respective presumed optimal cutoffs (> 10⁵ qPCR gn L⁻¹ or ≥ 100 RT-qPCR gn L⁻¹) the probability that criteria for LC/MS/MS will be exceeded one week later more than 21-26% of the time for the qPCR or RT-qPCR indicators. For ELISA criteria, the corresponding exceedance limits were all in the range of 31-38%. This resulted in higher false positive rates for mass spectrometry standards. However, indications were that in the case of the two reverse transcript indicators false negatives were reduced, meaning that fewer exceedances would be missed. Unlike sensitivity and specificity, which define the optimal cutoffs, false positive and false negative rates will be highly dependent on actual frequency of exceedances in the location where these criteria are applied. Behavior of the ROC curve over other lag periods is illustrated in Fig. 5. Only data from daily sampling from June 2 to July 1, which covered the different periods of MC presences (low, increased and high detection from bloom-beginning to forming) were used to avoid over representing seven-day lag periods between genomic and LC/MS/MS analyses. The mcyEcya RT-qPCR indicated the bloom-developing from early to late June in the aforementioned results. Based on the continuous daily data collected in June, behavior of the ROC curve over other lag periods is illustrated in Fig. 6. Strength of the mcyEcya qPCR/RT-qPCR predictive power appeared to wane initially, but rebounded at 6 days. This may be a periodicity effect related to variation in conditions by day-of-the week.

Fig. 6.

Observed predictive power based on area under the curve for lag times from continuous 0 to 14 days using qPCR and RT-qPCR targeting mcyA and mcyE genes of cyanobacteria (cya) and Microcystis (mic) with respect to LC/MS/MS measured microcystin concentrations (pooling data from the four sites).

4. DISCUSSION

The goal of this study was to examine the utility of qPCR data as an indicator of cyanoHAB risk and the potential use as an early warning predictive modeling of MCP cyanoHABs. Observations of the major cyanobacteria populations (Microcystis, Planktothrix, Cylindrospermopsis, Anabaena and Nostoc) quantified by qPCR were comparable in quantity, variation, and association with values obtained through microscopic identification (Chen et al. 2017). For example, the compositions of genera and dominant genera were in agreement, and the correlations between genomic numbers by qPCR and biomass by microscopy were highly positive (R_{Cyanobacteria}=0.86, R_Microcystis=0.82) across all samples. qPCR provides an efficient, accurate, and consistent means to quantify cyanobacterial populations as compared to the relatively laborious microscopic approach. Importantly, molecular or qPCR method is the only way for quantifying MCP populations, which were the focus of this study.

This study also demonstrated the potential predictive role of qPCR in the identification of elevated cyanotoxin risk as illustrated with ROC analysis. Harsha Lake MC congener concentrations, dominated by MC-LR (mean: 0.48, StDev: 0.42) μg L⁻¹, were similar to Lake Ontario (Hotto et al. 2007), where MC-LA, -RR and -LY were detected, but the most common congener was MC-LR, corresponding to both mcyA genotypes. Hotto et al. also observed instances of higher MC-LR concentrations > 1.0 μg L⁻¹ in mid-July, exceeding the World Health Organization (Organization 1999) provisional guideline 1.0 μg L⁻¹. MC production is controlled by the mcy genes, a bi-directionally transcribed complex of 10 open reading frames responsible for synthesis of polyketide synthases and peptide synthetases involved in MC synthesis (Dittmann et al. 1997, Kaebernick et al. 2002). MCPs in Lake Harsha included species across three major orders: Chrococales, Nostocales and Oscillatoriales and Microcystis. Microcystis is the major MCP, but not all strains produce MC. Nontoxic strains do not appear to carry mcy genes. In this study, the assays mcyEcya and mcyAmic were sensitive to most, if not all, MCPs. The significant correlations of qPCR of mcyEcya and mcyAmic with MC-ADDA (R: 0.84 and 0.91, respectively) indicated their importance in MC production and a role in predicting elevated MC risk. The principle for using qPCR to monitor MC is that toxic strains continuously produce MCs (Kaebernick et al. 2000) and their ratios in total populations vary with cell growth (Orr and Jones 1998), and genetic differences within mcy result in production variation. Approaches using qPCR to explore the correlations between cyanobacteria and toxin, especially MC, have been carried out in several experiments. A majority of previous studies on lake CyanoHABs provided the evidence to support the application of qPCR to monitor toxic genotypes and further predict MC production. For example, positive correlations of the sum of Microcystis and Anabaena mcyE copy numbers with MC (R_sum = 0.93, R_separate = 0.74-0.82) were found from both Lake Tuusulanjärvi and Lake Hiidenvesi (Vaitomaa et al. 2003) in agreement with correlations presented here. Furthermore, the proportion of mcyA containing P. agardhii cells was shown to significantly correlate with MC values (Briand et al. 2009), while the different MC variant concentrations could be predicted from the corresponding genotype concentrations (Ostermaier and Kurmayer 2010). As described above, such result regarding P. agardhii was not found in this study. In the Great Lakes, the densities of potentially toxic Microcystis cells significantly correlated with MC concentrations (Davis et al. 2014) and suggested it could be a better predictor of MC concentrations in aquatic ecosystems than parameters like total cyanobacteria cell counts, total Microcystis, or chlorophyll. Nevertheless, there are reports which suggest non-correlated results between MC and mcy copy numbers (Beversdorf et al. 2015). Specific growth phase or log phase has been observed to play a role in generating higher toxin concentrations than the other phases (Kurmayer and Kutzenberger 2003). Some field studies are limited by the insufficient sampling frequency and sites (e.g. monthly sampling with a few sites) which made it impossible to evaluate the relationships between qPCR and MC (Briand 2009). It is apparent that the presence of mcyEcya or mcyAmic qPCR signals indicating potential MC production is an inadequate indicator of MC production, while the use of MC gene expression signaling by RT-qPCR can reveal approaching MC production.

Regressing qPCR and RT-qPCR signal against MC-ADDA concentration provided numerical values of toxic genes, above which, MC would be expected to be found in concentrations greater than guidance criteria. ROC analysis of qPCR and RT-qPCR demonstrated relatively high power in predicting MC concentrations (UAC ≥ 62%). Mcy-targeted RT-qPCR reflected not only the potential of MC level, but also signaled a cycle of bloom-associated MC production. Under the extensive sampling scheme, which consisted of the routine weekly sampling and the daily sampling during an imminent bloom, the variation of toxin gene RT-qPCR signal and total MC along the time course (Fig. 3) reflected the various stages (initiation, growth, stagnation, and decrease) of potential MC production at the community level. The process can potentially be used to predict the occurrence of cyanotoxins. Taken together, understanding the timing of these periods of elevated cyanotoxin risk is critical for managing exposure.

The ratios of toxin gene expressions over the extant toxic genes indicated the activities of toxic populations. Genus or species-specific RT-qPCR signals indicated potential MC production. This study suggested a rapid and cost-effective monitoring process in which, proactive toxin measurement would be initiated when RT-qPCR targeting MCPs reach a threshold level. QPCR for the same genotypes reflect potential toxin levels. RT-qPCR would also signal the roles of each MCP. For example, lower quantities of toxic subpopulation ratio and expression ratio as observed in Planktothrix indicate a lower risk associated with this genus. This notion is supported by observations that under culture conditions, significant correlation of extant/expressed cylindrospermopsin (CYN) producing gene signals expression to intracellular, as opposed to intercellular total CYN accumulation in Aphanizomenon ovalisporum (Barón-Sola et al. 2016). Thus, MCP qPCR is not only used to reflect the occurrence and quantitative variations of MCP populations, but importantly also used to predict potential MC level. Taken together, mcyEcya and mcyAmic RT-qPCR should be considered as indicators to estimate the level of total MC produced by Microcystis.

MCPs presence, dominance, ratio, and relationships to the whole population or community also revealed utility of qPCR. The dominance of MCPs was evidenced by multiple lines of data including qPCR and toxin analytical data by this study and clone-sequencing by a previous study (Chen et al. 2017). MCPs have been considered the most common cyanobacterial species in freshwater, possibly due to their roles in the synthesis and incorporation of vital amino acids into the microcystin structure (Tillett et al. 2001, Vaitomaa et al. 2003). Various MCP species and strains including Microcystis, Planktothrix, and Anabaena have been reported. Of the MCPs in Harsha Lake, Microcystis spp., mainly M. aeruginosus according to the PCR-clone sequencing results (Chen et al. 2017), were the dominant genus based on mcyE and mcyG targeted qPCR. Another MCP, Planktothrix, was present throughout the sampling season, peaked along with Microcystis and became more dominant than Microcystis and other major groups from August to September. Planktothrix was also observed to be dominant repeatedly in autumn in other lakes (Hisbergues et al. 2003, Fastner et al. 1999). Other frequently observed genera like Cylindrospermopsis, Anabaena, and Nostoc were not dominant and not shown to be MC producing or detectible throughout the summer in Harsha Lake.

The ratios of toxic subpopulation to the total Microcystis were found to be stable regardless of sampling dates and sites in this study. It was a novel finding, which indicated that MCPs were a normal part of total population, and consistent with co-growth of nontoxic subpopulations in the lake. In previous studies, lower ratios of MCP subpopulations (mcyA, D, E or other genes) in Microcystis population (16S or counts) were reported in various water bodies. For example, the ratio ranges are 0.4 - 11.5% in Lake Erie samples (Rinta-Kanto et al. 2005), 0.37-20.2% in the San Francisco Estuary (Baxa et al. 2010), 1.7-22.6% in Missisquoi bay, Quebec, Canada (Ngwa et al. 2014), 0.5-35% in Lake Mikata, Japan) (Yoshida et al. 2007) and 1-38% in Lake Wannsee, Germany (Kurmayer and Kutzenberger 2003). However, the higher average ratio of MCP subpopulations in the total Microcystis population was reported as approximately 80% (Ha et al. 2008) in a fish pond. Thus, the ratio of toxic Microcystis population reported in this study was close to those in other eutrophic lakes dominated by nontoxic subpopulations. When compared to the total cyanobacterial biomass, toxigenic Microcystis seem to constitute a relatively small fraction of the cyanobacterial population in this lake (~2%). The ratio of reported toxic to nontoxic genotypes varies. A significant negative correlation between total MC and the ratio during a Microcystis bloom was reported (Briand et al. 2009). In this study, persistent rainy weather was not a favorable condition to form a higher Microcystis bloom (Chen et al. 2007), therefore the stable toxic and nontoxic subpopulations co-occurred in the system.

The significantly positive relationships of MCP qPCR (mcyEcya or mcyAcya) with the qPCR and counts of cyanobacteria (qPCR: CYA; count: cyano) and Microcystis (qPCR: MIC; count: micro) (Table S6), indicated that MCPs could be directly estimated through cyanobacterial biomass or total genomic copy numbers, or possibly other methods like chlorophyll or remote sensing during bloom season. Previous studies have also shown that there were significant correlations between qPCR signals and total counts of cyanobacteria and Microcystis. Ngwa et al. (2014) reported highly significant correlation (R² >0.90) of microscopically determined total Microcystis counts with qPCR-based Microcystis mcyA, mcyE, or mcyG cell number equivalents. Furthermore, cell concentrations were estimated using mcyA assay in Missisquoi Bay, Quebec, Canada (Ngwa et al. 2014). Using the same mcyE primers, Ngwa et al. quantified MCP Microcystis in two Finnish lakes wherein significant positive correlations were demonstrated between Microcystis mcyE copy numbers and microcystin concentrations (Vaitomaa et al. 2003).

This study provided the following information regarding Microcystis toxin gene expression in field samples: 1) the expressed genes were only small portion (0.2-0.4%) of extant genes, while their cell replication could reach 4911 fold; 2) fold change of toxin gene expression and cell replication were similar in trend and levels, especially during bloom; 3) Planktothrix was determined as a lower toxin producer in both extant and expressed toxin gene mcyE, which confirmed previous assumptions that Planktothrix is not a major contributor to MC production in other Ohio lakes (Francy et al. 2015); and 4) High extant toxin gene levels indicated greater toxin production potential during bloom. This study was in agreement with previous finding that the synthesis of MC is related to cell division and growth, although the amount of produced MC is regulated in response to environmental factors such as light (Kaebernick et al. 2000, Tonk et al. 2005) and trace elements (iron) (Sevilla et al. 2008).

5. CONCLUSION

In summary, MCPs were quantitatively estimated using a panel of SYBR-qPCR assays targeting toxin genes (mcyA, E, and G). Comparisons between toxic subpopulations and total cyanobacteria, non-toxic subpopulations, and non-MCP species were conducted for Harsha Lake for weekly (routine) and daily (during bloom) samples. Microcystis aeruginosa was the major species responsible for MC production, while others like Planktothrix and Anabaena contributed little or negligible toxin production. RT-qPCR can be a good predictive indicator for MC-production. RT-qPCR can be used to indicate the initiation/ending and peak period of toxin in raw water, while qPCR can reflect the levels of biomass, variation, bloom development, and toxin. The application of qPCR and RT-qPCR with high specificity, sensitivity, and throughput may be valuable in forecasting toxic blooms by determining the blooms of specific toxic species with genotypes harmful to human and ecosystem health.

Table 5.

Logistic results of qPCR and RT-qPCR (-rt) targeting mcyA and mcyE of cyanobacteria (CYA) and Microcystis (MIC) for endpoints of drinking water guideline exceedance (0.3 μg L⁻¹) for determination by (1) LC-MS/MS and (2) ELISA

MC method	Indicator (log10)	Odds Ratio	P-value	AUC1	Cutoff2	P(Exceed) @ Cutoff3 Estimate 95% CI		False Positives	False Negatives
LC-MS/MS	mcyEcya	5.43	<0.001	0.60	4.52	0.22	0.12 - 0.36	66%	6%
	mcyEcya-rt	1.50	<0.001	0.34	1.68	0.25	0.21 - 0.30	69%	13%
	mcyAmic	3.84	<0.001	0.60	4.29	0.21	0.13 - 0.32	65%	5%
	mcyAmic-rt	1.94	<0.001	0.44	2.20	0.26	0.20 - 0.33	63%	12%
ELISA	mcyEcya	3.34	<0.001	0.48	4.45	0.37	0.27 - 0.48	48%	11%
	mcyEcya-rt	1.45	<0.001	0.24	1.09	0.35	0.31 - 0.39	55%	16%
	mcyAmic	2.60	<0.001	0.50	4.27	0.38	0.30 - 0.47	46%	11%
	mcyAmic-rt	1.77	<0.001	0.34	1.47	0.31	0.26 - 0.37	52%	9%

¹Area under the ROC curve (bounded by line of equality).

²Cutoff value of the qPCR/RT-qPCR values at the Youden Index based optimum.

³Model estimated probability that 7-day future criterion (LC-MS/MS or ELISA) will exceed its guideline value when present indicator is at cutoff value.

Highlights.

• Microcystin producers were dominated by Microcystis and kept a stable ratio to nontoxic subpopulation.

• High correlations were found between qPCR and RT-qPCR, and between microcystin and qPCR/RT-qPCR.

• RT-qPCR signaled activities of toxin genes and toxic species indicating the initiation of bloom-formation and toxin production and the subsequent subsidence of toxin production.

• A prediction interval of one-week prior to MC production using qPCR/RT-qPCR signals was identified.