A qPCR-Based Tool to Diagnose the Presence of Harmful Cyanobacteria and Cyanotoxins in Drinking Water Sources

Harmful cyanobacteria have been an important concern for drinking water quality for quite some time, as they may produce cyanotoxins and odorants. Microcystis and Cylindrospermopsis are two common harmful cyanobacterial genera detected in freshwater lakes and reservoirs, with microcystins (MCs) and cylindrospermopsin (CYN) as their important metabolites, respectively. In this study, two sets of duplex qPCR systems were developed, one for quantifying potentially-toxigenic Microcystis and Microcystis, and the other one for cylindrospermopsin-producing cyanobacteria and Cylindrospermopsis. The duplex qPCR systems were developed and validated in the laboratory by using 338 samples collected from 29 reservoirs in Taiwan and her offshore islands. Results show that cell numbers of Microcystis and Cylindorspermopsis enumerated with microscopy, and MCs and CYN concentrations measured with the enzyme-linked immuno-sorbent assay method, correlated well with their corresponding gene copies determined with the qPCR systems (range of coefficients of determination R2 = 0.392−0.740). The developed qPCR approach may serve as a useful tool for the water industry to diagnose the presence of harmful cyanobacteria and the potential presence of cyanotoxins in source waters.


Introduction
Lakes and reservoirs are important drinking water sources in Taiwan and many other countries. Due to improper management of nutrients in their watersheds, many reservoirs are facing the risk of eutrophication, increasing the possibility of cyanobacterial blooms in the source waters [1]. Cyanobacteria have been reported to produce a wide variety of chemically unique secondary metabolites, such as hormones, antibiotics, allelochemicals, toxins, and taste and odor (T&O) compounds [2]. Among these metabolites, cyanotoxins may cause illness or death of domestic animals and humans when they are exposed to water contaminated with high cyanobacterial concentrations [3]. Microcystins (MCs) and cylindrospermopsin (CYN) are two well-known cyanotoxin groups found in the reservoirs of Taiwan [4][5][6] and the world [7][8][9][10][11]. MCs have been linked to the increased incidence of primary liver cancer [3,12,13] and acute cases of poisoning in humans and other animals [14]. CYN has been reported to be not only a general cytotoxin but also hepatotoxic and neurotoxic [15][16][17], and has also been linked to the death of domestic animals [18]. Therefore, detection of these two cyanotoxins is an important issue to safeguard drinking water quality.
Water samples were collected from 29 different drinking water reservoirs (DWRs) located throughout the main island of Taiwan and its two off-shore islands, Kinmen Island and Matsu Island. The sampling locations for all the studied reservoirs were all near the water intakes of their associated water treatment plants. The locations of the studied DWRs are shown in Supplementary Materials Figure S1.

Cell Enumeration
Cyanobacteria cells were enumerated following the procedures prescribed in the Standard Method 10200F [63]. In brief, the water samples were fixed with Lugol's solution (1% by volume) immediately after sampling and stored in the dark until enumeration. For cell counting, water samples were placed into a 1 mL Sedgwick-Rafter chamber (Pyser-SGI Ltd., Kent, UK) and allowed to settle for 30 min. The spherical Microcystis cells were then counted under a microscope (BX51, Olympus, Tokyo, Japan) atˆ400 magnification. For the filamentous Cylindrospermopsis, the cross walls between cells were often indistinct, making this measurement imprecise [43]. Thus, the cells were enumerated by their average cell number per unit length of trichome following the protocol reported in Marbun et al. [5].

Extraction of DNA from Cyanobacterial Cells
To extract DNA from cyanobacterial cells, cyanobacteria-laden water samples (10 mL) were first concentrated using a centrifuge (Hettich Mikro 20 Microfuge, Biotech Equipment, San Francisco, CA, USA) rotating at 16,060 rcf for 5 min. After centrifugation, the supernatant was removed and the cyanobacterial pellets remaining were re-suspended with 5 mg/mL of Lysozyme (Invitrogen™, Life Technologies, Carlsbad, CA, USA), and followed by a freezing-thawing process that was performed three times using liquid nitrogen and a 37˝C water bath for the lysis of cyanobacterial cells [64]. Then, DNA was extracted from the lysed cyanobacterial samples using Illustra™ triplePrep kit (GE Healthcare, Hatfield, UK), following the protocol provided by the manufacturer.

Duplex qPCR System
Each of the developed duplex qPCR systems was performed individually and separately with a qPCR device (Smart Cycler ® II; Cepheid, Sunnyvale, CA, USA). The first system helps differentiate potentially-toxigenic Microcystis cells from that of total Microcystis cells, while another system helps distinguish cylindrospermopsin-producing cyanobacteria from that of total C. raciborskii cells.
For the Microcystis system, primer and probe set Micr184F/Micr431R/Micr228 was used to target the 16S rRNA region unique to Microcystis cells [65], while mcyB#04F/mcyB#04R/mcyB#04 were used to specifically target the microcystin synthetase (mcyB) region unique to potentially-toxigenic Microcystis cells [4]. Detailed information of the primer and probe sets were summarized in Table S1. During analysis, 2 µL of extracted DNA was mixed with 12.5 µL of 2ˆRotor-Gene Multiplex PCR Master Mix kit (Qiagen, Hilden, Germany) and 0.1 µM of each primer and probe, with the volume being made up to 25 µL using sterile deionized water. The qPCR protocol includes pre-incubation at 95˝C (300 s), and a 40-cycle of 2-step amplification of denaturation (95˝C, 15 s) and annealing/extension (60˝C, 30 s). At the end of the extension, the fluorescence intensity was recorded at 519 nm for FAM and at 561 nm for Cy3. The threshold cycle (Ct) values were determined using Smart Cycler ® software version 2.0 (Cepheid, Sunnyvale, CA, USA).
For identification of Cylindrospermopsis, primer and probe set m4/k18/pks is efficient for targeting the polyketide synthase (pks) region of cylindrospermopsin-producing cyanobacteria [26,45,66], while cyl2/cyl4/rpoC1 helps to specifically target the rpoC1 region unique to C. raciborskii [45,67] (Table S1 (SM)). Also, 0.1 µM of each primer and probe was used during analysis. The qPCR protocol includes a pre-incubation at 95˝C (300 s), and a 40-cycle of 2-step amplification of denaturation (95˝C, 15 s) and annealing/extension (60˝C, 45 s). At the end of the extension, the fluorescence intensity was recorded at 665 nm for Cy5 and at 615 nm for Texas Red. The Ct values were calculated by Smart Cycler ® software.
The final reaction mixtures of both uniplex and duplex qPCR systems were optimized by testing the primer and probe concentrations. The primer and probe concentrations corresponding to the lowest Ct and highest fluorescent intensity, respectively, were chosen. Similar Ct's (difference < 1) were obtained for all the four targeted genes in both uniplex and duplex qPCR systems in this study.

PCR Amplification, Cloning, and Standard Curves
The PCR product of the 16S rRNA region unique to Microcystis cells and the mcyB region were amplified from the extracted DNA of M. aeruginosa PCC7820 strain using the primer sets of Micr184F/Micr431R and mcyB#04F/mcyB#04R, respectively. In a similar fashion, the PCR product for the rpoC1 region and the pks region was amplified from the extracted DNA of C. raciborskii CYP026J strain using the primer sets of cyl2/cyl4 and m4/k18, respectively.
Sterile deionized water was then added to a PCR mixture, which consisted of 5 µL of the extracted DNA solution, 2 µL of 10ˆEx Taq™ buffer, 200 µM of dNTP mix, 0.2 µM of the forward and reverse primers, and 0.1 µL of TaKaRa Ex Taq™ DNA polymerase (Code No. RR01A, Takara Biotechnology, Kusatsu, Japan), such that the volume totalling 50 µL was then placed in a 96-well thermal cycler (C1000™, Bio-Rad Corp., Hercules, CA, USA). The PCR experiment was executed with a pre-incubation at 95˝C (5 min), and a 30-cycle of 3-step amplification of denaturation (98˝C, 10 s), annealing (55˝C, 30 s), and extension (72˝C, 1 min), then a final extension at 72˝C for 10 min.
Then, four PCR products were cloned by using a commercial kit (TOPO ® TA Cloning ® , Invitrogen™, Life Technologies, Carlsbad, CA, USA) following the suggested protocol. The plasmid DNAs from cloning were used to develop the standard curve between the Ct values and the copy numbers. Longer sequences of the target region from plasmid DNAs were amplified by using the universal M13 primers to avoid the possibility of mispaired nucleotides at the ends of the target genes. The M13 PCR product was quantified by measuring the fluorescence intensity with PicoGreen ® dsDNA Reagent and Kits (Quant-iT™, Invitrogen™, Life Technologies, Carlsbad, CA, USA). The DNA copy number (copies/µL) was then calculated by dividing the DNA concentration (g/µL) by the molecular weight (g/mole) of the targeted gene and then multiplied by 6.02ˆ10 23 copies/mole [9].
Series of 10-fold dilutions of M13 PCR product of the target genes were prepared (10 2 to 10 8 copies/reaction), and the Ct values were plotted against the logarithm of the copy number of the target genes with an amplification efficiency (E) of 100%˘10%, where E = 10´1 /S´1 (S is the slope of the logarithm regression).

Test of Inhibition on Gene Detection in Duplex Systems
To determine the concentration ratios causing the inhibition, a set of experiments were conducted with different targeted gene concentration ratios for the duplex systems. In these experiments, standard genes of the four studied targets, 16S rRNA, mcyB, pks, and rpoC1, were prepared with varying amounts between 10 3 and 10 7 copies/reaction. The prepared standard genes were then measured with the two duplex qPCR systems for Microcystis and Cylindrospermospis.

Detection of Cyanotoxins
Cyanotoxins in the samples were detected using two commercial ELISA Kits, PN 520012 and PN 522011, both from Abraxis, US for MCs and CYN, respectively [49]. Samples were collected from DWRs then prepared to detect both total cyanotoxins (cell bound + dissolved phase) and dissolved cyanotoxins. For the unfiltered samples, the cyanobacterial cells were ruptured using liquid nitrogen treatments, similar to DNA extraction, to release cell-bound cyanotoxins and then filtered with the 0.22 µm nylon syringe filter to remove cell debris. Dissolved cyanotoxins were separated from cell-bound cyanotoxins using a 0.22 µm nylon syringe filter (Advangene, Lake Bluff, IL, USA). The unfiltered and filtered samples were analyzed for total MCs/CYN and dissolved MCs/CYN, respectively. Following the protocol provided by the manufacturer, 50 µL of the pre-treated MCs/CYN-laden water sample were analyzed. An ELISA reader (Multiskan FC, Thermo Scientific, Vantaa, Finland) was employed to measure the absorbance. The calibration curves were constructed using standard MCs/CYN provided with the ELISA kit.

Statistical Analysis
The field data obtained in this study were analyzed for the correlation between cell enumeration, toxin concentrations, gene copies and cell equivalents of the samples. Although the gene copies in Microcystis and Cylindrospermopsis may vary with growth status, species, and environmental conditions [68,69], in this study the cell equivalents for the two cyanobacteria were obtained by linking the gene copies of the two standard strains with their cell numbers under the logarithmic growth phase. The calculation of 95% prediction intervals for the linear regressions followed that of Chiu et al. [64]. The regression analysis was conducted with SPSS Statistics 17.0 software (IBM, New York, NY, USA). A statistical software from Minitab (Minitab ® , State College, PA, USA), was further used to compare the slope coefficients of different regression lines. The slopes are considered to be statistically significantly different when a P value is less than or equal to 0.025 with two-tailed 95% confidence intervals. Another statistical method, the Chow test [70], was also used to test the similarity of the regression coefficients obtained for different data.

Standard Curves
In a comparison of qPCR standard curves for Microcystis and Cylindrospermopsis systems, results clearly demonstrate that the calibration curves obtained from the duplex quantification system are almost identical to those for the uniplex system, suggesting the equal performance of both systems ( Figure 1). The calibration curves of duplex systems for the 16S rRNA gene and mcyB gene, as well as the rpoC1 gene and pks gene, were linear from 10 2 to 10 8 copies/reaction with high correlation coefficients, with R 2 values at 0.998, 0.996, 0.999, and 0.999, respectively. In addition to this, amplification efficiencies of the 16S rRNA gene, mcyB gene, rpoC1 gene and pks gene were 94%, 91%, 92% and 91%, respectively. Such relatively high correlation coefficients and amplification efficiencies suggest that the analytical protocol is reasonable.

Influence of Gene Abundance in the Duplex Systems
To test the applicability of the duplex qPCR systems for field monitoring, the systems were tested in Tai-Hu Reservoir (THR), Kinmen, Taiwan. In the meantime, the samples were also analyzed with the uniplex qPCR systems. The duplex system results were compared with their respective uniplex system for the detection of the targeted cyanobacteria. The data, which are shown in Tables  S2 and S3 for Microcystis and Cylindrospermosis, respectively, were collected and analyzed from February 2013 to August 2016. Among the 13 samples analyzed for both Microcystis and Cylindrospermopsis, most results from the duplex systems closely matched with those for a uniplex system. However, a discrepancy was found for three samples collected in February 2013, November 2013 and March 2014, which tested specifically for Cylindrospermopsis (Tables S3 and 1). Although high concentrations of CYN were measured, only the uniplex system was able to capture both rpoC1 and pks genes, while the duplex system failed to detect the pks gene ( Table 1). The failure of the duplex system may be caused by differences in primer concentration, probe concentration, polymerase/dNTP/buffer concentration, annealing/extension time and amount of template DNA [71][72][73][74]. Similarly, previous studies have also shown that the amount of template DNA affects its detection in multiplexed systems [51,[58][59][60], where the detection of low concentration genes was inhibited by the high concentration ones. In this study, the Ct values for the rpoC1 gene in these three uniplex-based samples were 9-10 cycles smaller than those for the pks gene, suggesting that abundance of the pks gene in the sample was much lower (Table 1). This is in accordance with those reported in Wang and Mustapha [58], Hyeon, et al. [59], Dai, et al. [60] and Te and Gin [51] that detection of a lower gene concentration was inhibited in multiplexed systems.

Influence of Gene Abundance in the Duplex Systems
To test the applicability of the duplex qPCR systems for field monitoring, the systems were tested in Tai-Hu Reservoir (THR), Kinmen, Taiwan. In the meantime, the samples were also analyzed with the uniplex qPCR systems. The duplex system results were compared with their respective uniplex system for the detection of the targeted cyanobacteria. The data, which are shown in Tables S2 and S3 for Microcystis and Cylindrospermosis, respectively, were collected and analyzed from February 2013 to August 2016. Among the 13 samples analyzed for both Microcystis and Cylindrospermopsis, most results from the duplex systems closely matched with those for a uniplex system. However, a discrepancy was found for three samples collected in February 2013, November 2013 and March 2014, which tested specifically for Cylindrospermopsis (Table S3 and Table 1). Although high concentrations of CYN were measured, only the uniplex system was able to capture both rpoC1 and pks genes, while the duplex system failed to detect the pks gene ( Table 1). The failure of the duplex system may be caused by differences in primer concentration, probe concentration, polymerase/dNTP/buffer concentration, annealing/extension time and amount of template DNA [71][72][73][74]. Similarly, previous studies have also shown that the amount of template DNA affects its detection in multiplexed systems [51,[58][59][60], where the detection of low concentration genes was inhibited by the high concentration ones. In this study, the Ct values for the rpoC1 gene in these three uniplex-based samples were 9-10 cycles smaller than those for the pks gene, suggesting that abundance of the pks gene in the sample was much lower (Table 1). This is in accordance with those reported in Wang and Mustapha [58], Hyeon et al. [59], Dai et al. [60] and Te and Gin [51] that detection of a lower gene concentration was inhibited in multiplexed systems.  1 Represents standard deviation of 2 replicates; 2 -Represents the result < detection limit (Table S1).
To determine the concentration ratios causing the inhibition, the results clearly show that, for the duplex Microcystis system when mcyB gene copies were 10 2 times higher than that of 16S rRNA gene, the system fails to detect the less abundant gene ( Table 2). In contrast, detection of the mcyB gene was not affected even when 16S rRNA gene abundance was 10 4 times higher. Fortunately, since the abundance of the mcyB gene, which is a gene that identifies a subset of Microcystis cells that are potentially toxigenic [4], should always be less than or equal to the total number of Microcystis cells as identified via the 16S rRNA gene [65], the issue of false negative results for 16S rRNA would not be the problem in the duplex Microcystis system (Table S2). 1 Represents standard deviation of 2 replicates; 2 -Represents the result < detection limit (Table S1).
For the duplex Cylindrospermopsis system, when pks gene copies were 10 3 times higher than that of rpoC1 gene, no rpoC1 gene will be detected. On the other hand, pks gene was not detected if the abundance of rpoC1 gene copies were 10 2 times higher. Since the pks gene primer was designed for cylindrospermopsin-producing cyanobacteria [26,45,66], and the rpoC1 gene primer was targeted specifically on C. raciborskii [45,67], those detected by the pks gene primer are not necessary all detected by the rpoC1 gene primer, as some cyanobacteria such as Anabaena bergii [26], Anabaena lapponica [27], Aphanizomenon gracile [28], Aphanizomenon flos-aquae [29], Aphanizomenon ovalisporum [30][31][32], Raphidiopsis curvata [35], and Umezakia natans [36] may also produce cylindrospermopsin. Therefore, in applying this duplex Cylindrospermopsis system, if either gene is detected at >10 5 copy/reaction and the other one is not detected, the uniplex system should be used to confirm the presence of the second gene.
Attempts have been made to improve the detection of the lower abundant targeted genes, including primer concentration, probe concentration, polymerase/dNTP/buffer concentration and annealing/extension time ( Figure S2 and Table S4). However, none of them can reduce the influence on the inhibition of the detection. Further studies are needed to improve the inhibition of detection for the lower abundant targeted genes in multiplexed systems.

Applications of the Duplex Systems in Reservoir Water Samples
The two duplex systems were further applied to quantify the targeted genes in 338 samples collected between February 2013 and August 2016 from 29 reservoirs throughout Taiwan, consisting of 10 drinking water reservoirs (DWRs) on the main island of Taiwan, 10 DWRs in the Kinmen islands, and 9 DWRs in the Matsu islands ( Figure S1). In addition, the samples were also analyzed for the two cyanotoxins, MCs and CYN, and enumerated for cell numbers of the two targeted cyanobacteria. Figures 2 and 3 show the monitoring results for one of the studied reservoirs, THR in Kinmen islands, for Microcystis and MCs and Cylindrospermopsis and CYN, respectively. It is noted that for the reservoir samples, the gene abundance (copies-mL´1) was transferred to the cell equivalents concentration (cells-mL´1) through the calibration curves constructed for the laboratory cultures [65], as shown in Figure S3. Based on this, the total number of Microcystis cell equivalents, potentially-toxigenic Microcystis cell equivalents, and the total number of Cylindrospermopsis cell equivalents in a sample can be estimated, from the abundance of the 16S rRNA gene, mcyB gene, and rpoC1 gene, respectively. It is noted that the primer and probe set for the pks gene is efficient for targeting the polyketide synthase (pks) region of cylindrospermopsin-producing cyanobacteria, including Cylindrospermopsis raciborskii, Aphanizomenon ovalisporum, Raphidiopsis curvata, Anabaena bergii and Oscillatoria sp. [26,45,66]. As the genome copies of cyanobacteria may vary in different genera [68,69], the gene copies were not transferred into cell equivalent of cylindrospermopsin-producing cyanobacteria in the samples.
Results showed that the cell-equivalent concentrations of field samples determined with duplex qPCR methods, 16S rRNA gene for Microcystis and rpoC1 gene for Cylindrospermopsis, were mostly similar to cell concentrations enumerated with microscopy, with only 5 of the 13 Kinmen samples having a difference of more than one order of magnitude (Figures 2 and 3). In natural water, Microcystis cells often grow in colonies [75,76], while Cylindrospermopsis trichomes are 50-250 µm long, and slightly constricted or not constricted at the cross walls between cells [43]. Colonies and trichomes may prohibit the uniform distribution of Microcystis and Cylindrospermopsis cells in the water samples, making the cell enumeration and gene detection difficult. The discrepancy observed for the two analytical methods, the qPCR and microscope methods may be caused by the non-uniform distribution of the cells in natural water samples.
The increase of the cell-equivalent concentration of toxigenic microcystis (based on the mcyb gene) and copy number of cylindrospermopsin producers (based on the pks gene) follow an increase in the concentration of MCs and CYN, respectively. Note that for each particular toxin, the data obtained for two of the sampling dates did not follow the trends. Figure 2 showed that, for the duplex microcystis system, only mcyb gene and no MCs were detected in the samples collected in February 2013 and May 2014; while in Figure 3, for the duplex cylindrospermopsis system, only CYN and no pks genes were detected in the 2 samples collected in December 2014 and August 2015. The discrepancy between the two methods, gene detection, and ELISA, may be attributed to a few reasons. First, detection of mcyb and pks genes represents the presence of potential producers, but not necessarily the presence of products (toxins) [77][78][79][80]. It is possible that some of the producers were not producing the cyanotoxins during the sampling. Although naked DNA from lysed cells might be present in the samples, in the current experimental protocol, the naked DNA was expected not to be extracted using centrifugation [64]. Thus, it is possible that in the Cylindrospermopsis system, the lysed cells or extremely low cell concentrations may lead to no detection of the pks gene in the samples. However, the released CYN was still measurable.
In Figure 2, the results also demonstrate that abundance of Microcystis (detected with 16S rRNA gene) was always higher than that of potentially toxigenic species (mcyB gene). Previous research demonstrated that toxic and non-toxic strains of the same cyanobacterial species often coexist in the environment [39][40][41][42]. The results of this study show that 1.53-17.14% of Microcystis cells detected were capable of producing MCs, confirming the observations of previous field studies that not all strains of Microcystis found in natural samples are capable of producing toxins [65,78]. In Figure 2, the results also demonstrate that abundance of Microcystis (detected with 16S rRNA gene) was always higher than that of potentially toxigenic species (mcyB gene). Previous research demonstrated that toxic and non-toxic strains of the same cyanobacterial species often coexist in the environment [39][40][41][42]. The results of this study show that 1.53-17.14% of Microcystis cells detected were capable of producing MCs, confirming the observations of previous field studies that not all strains of Microcystis found in natural samples are capable of producing toxins [65,78].

Correlations among Cell Numbers, Toxin Concentrations, Gene Copies and Cell Equivalents of the Samples
Microcystis and Cylindrospermopsis data collected from all the reservoirs were separated into three main groups based on differences in geographic location: Taiwan main island (Figure 4a In Figure 2, the results also demonstrate that abundance of Microcystis (detected with 16S rRNA gene) was always higher than that of potentially toxigenic species (mcyB gene). Previous research demonstrated that toxic and non-toxic strains of the same cyanobacterial species often coexist in the environment [39][40][41][42]. The results of this study show that 1.53-17.14% of Microcystis cells detected were capable of producing MCs, confirming the observations of previous field studies that not all strains of Microcystis found in natural samples are capable of producing toxins [65,78].

Correlations among Cell Numbers, Toxin Concentrations, Gene Copies and Cell Equivalents of the Samples
Microcystis and Cylindrospermopsis data collected from all the reservoirs were separated into three main groups based on differences in geographic location: Taiwan main island (Figure 4a

Correlations among Cell Numbers, Toxin Concentrations, Gene Copies and Cell Equivalents of the Samples
Microcystis and Cylindrospermopsis data collected from all the reservoirs were separated into three main groups based on differences in geographic location: Taiwan main island (Figure 4a In order to better understand the effectiveness of this study's new duplex system, two comparisons were made for Microcystis at each geographic location: cell equivalents concentrations (based on the 16S rRNA gene) was compared to the cell abundance (microscopy); potentially toxigenic cell equivalents (mcyB gene) was compared to MC concentration (ELISA). The comparisons for Cylindrospermopsis: cell equivalents (rpoC1 gene) were compared to cell abundance (microscopy); cylindrospermopsin producing the gene copy number (pks gene) was compared to CYN concentration (ELISA). Health 2017, 14, 547 10 of 17 was compared to the cell abundance (microscopy); potentially toxigenic cell equivalents (mcyB gene) was compared to MC concentration (ELISA). The comparisons for Cylindrospermopsis: cell equivalents (rpoC1 gene) were compared to cell abundance (microscopy); cylindrospermopsin producing the gene copy number (pks gene) was compared to CYN concentration (ELISA). Regardless of geographic location, correlation values between the cell equivalents of total Microcystis concentrations obtained via duplex and microscopy were relatively moderate to strong, ranging from R 2 = 0.566−0.769 (p < 0.01; Figure 4a,c,e). Similarly, the correlation value between the cell-equivalent concentration of potentially toxigenic Microcystis obtained via duplex and MCs concentration obtained via ELISA was also moderate to strong, ranging from R 2 = 0.620−0.731 (p < 0.01; Figure 4b,d,f). Furthermore, regardless of geographic location, the coefficient of cell abundance compared to conventional methodology remained consistent. Coefficients of cell equivalents of total Microcystis concentrations obtained via duplex compared to microscopy ranged from 0.605 to 0.741 (Figure 4a Regardless of geographic location, correlation values between the cell equivalents of total Microcystis concentrations obtained via duplex and microscopy were relatively moderate to strong, ranging from R 2 = 0.566´0.769 (p < 0.01; Figure 4a,c,e). Similarly, the correlation value between the cell-equivalent concentration of potentially toxigenic Microcystis obtained via duplex and MCs concentration obtained via ELISA was also moderate to strong, ranging from R 2 = 0.620´0.731 (p < 0.01; Figure 4b,d,f). Furthermore, regardless of geographic location, the coefficient of cell abundance compared to conventional methodology remained consistent. Coefficients of cell equivalents of total Microcystis concentrations obtained via duplex compared to microscopy ranged from 0.605 to 0.741 (Figure 4a,c,e); while coefficients of cell-equivalent concentrations of potentially toxigenic Microcystis obtained via duplex compared to MCs concentrations varied, within an even smaller range of 0.354 to 0.444 (Figure 4b,d,f). This suggests that regardless of differences in environmental factors due to changes in geographic location, slope value remains consistent. strong, ranging from R 2 = 0.528−0.792 (p < 0.01; Figure 5a,c,e). Similarly, the correlation value between the copy number of the cylindrospermopsin producing gene obtained via duplex and the CYN concentration obtained via ELISA was also moderate to strong, ranging from R 2 = 0.224−0.880 (p < 0.01; Figure 5b,d,f). Furthermore, the coefficients of the cell equivalents of total Cylindrospermopsis concentrations obtained via duplex compared to microscopy ranged from 0.638 to 0.861 ( Figure  4a,c,e); while coefficients of the copy number of cylindrospermopsin producing gene obtained via duplex compared to CYN concentration varied within an even smaller range of 0.109 to 0.183 ( Figure  4b,d,f). During analyses, the methodology developed in this study revealed consistent results, regardless of geographic grouping. In general, results showed that coefficients for linear regression For Cylindrospermopsis, the correlation values between the cell equivalents of total Cylindrospermopsis concentrations obtained via duplex and microscopy were relatively moderate to strong, ranging from R 2 = 0.528´0.792 (p < 0.01; Figure 5a,c,e). Similarly, the correlation value between the copy number of the cylindrospermopsin producing gene obtained via duplex and the CYN concentration obtained via ELISA was also moderate to strong, ranging from R 2 = 0.224´0.880 (p < 0.01; Figure 5b,d,f). Furthermore, the coefficients of the cell equivalents of total Cylindrospermopsis concentrations obtained via duplex compared to microscopy ranged from 0.638 to 0.861 (Figure 4a,c,e); while coefficients of the copy number of cylindrospermopsin producing gene obtained via duplex compared to CYN concentration varied within an even smaller range of 0.109 to 0.183 (Figure 4b,d,f).

Int. J. Environ. Res. Public
During analyses, the methodology developed in this study revealed consistent results, regardless of geographic grouping. In general, results showed that coefficients for linear regression amongst all data were similar, showing that the method in this study is consistent regardless of location. For cell abundance of the total Microcystis, comparison of the duplex qPCR method to microscopy (using the Minitab statistical software) showed that, regardless of geographic grouping, there were no significant differences for the three geological locations, with p = 0.704, 0.2620, and 0.206 (different if p ď 0.025 (two-tailed) for 95% confidence interval) for Taiwan main island and Kinmen islands, Taiwan main island and Matsu islands, and Kinmen islands and Matsu islands, respectively. Similar results were also observed for comparison of all other three cell/cyanotoxin to gene correlations for the three geological locations, with p ě 0.025 for the cases of mcyB and MCs concentration, rpoC1 and Cylindrospermopsis abundance, and pks and CYN concentration.
Several parameters may affect the correlation slope of the target compound to the genes, including toxin production rate and cell quota (toxin concentration per cell equivalents), toxin-producing species, and gene copy of the producing species [64]. Therefore, the correlation equations between cells/cyanotoxins and genes obtained for the three geological locations for both Microcystis and Cylindrospermopsis systems were further tested for regression stability, using Chow test [68]. The result suggests that the coefficients in all three linear regression models for three geological locations are equal (p < 0.01 in Figure 4a,c,e; p < 0.05 in Figure 4b,d,f; p < 0.001 in Figure 5a,c,e; and p < 0.05 in Figure 5b,d,f).
Although correlation values and slope values between the duplex qPCR method and conventional methods were moderate to strong, statistical analyses reveal that geological location does not necessarily determine the correlation values and slope values in this study. According to this result, the methodology developed in this study should not be limited by location, but rather applicable to other study areas.
In this study, the slopes of correlations obtained for Taiwan main island, Kinmen islands and Matsu islands fall in a narrow range. Therefore, all the data for the three geographical locations were further correlated. Figure 6 demonstrates the good correlations for the data collected from all geographic locations and may provide a simple way to link the data between cell numbers and gene copies, and toxin concentrations and gene copies or cell equivalents, for the two studied cyanobacteria in reservoirs. amongst all data were similar, showing that the method in this study is consistent regardless of location. For cell abundance of the total Microcystis, comparison of the duplex qPCR method to microscopy (using the Minitab statistical software) showed that, regardless of geographic grouping, there were no significant differences for the three geological locations, with p = 0.704, 0.2620, and 0.206 (different if p ≤ 0.025 (two-tailed) for 95% confidence interval) for Taiwan main island and Kinmen islands, Taiwan main island and Matsu islands, and Kinmen islands and Matsu islands, respectively. Similar results were also observed for comparison of all other three cell/cyanotoxin to gene correlations for the three geological locations, with p ≥ 0.025 for the cases of mcyB and MCs concentration, rpoC1 and Cylindrospermopsis abundance, and pks and CYN concentration. Several parameters may affect the correlation slope of the target compound to the genes, including toxin production rate and cell quota (toxin concentration per cell equivalents), toxinproducing species, and gene copy of the producing species [64]. Therefore, the correlation equations between cells/cyanotoxins and genes obtained for the three geological locations for both Microcystis and Cylindrospermopsis systems were further tested for regression stability, using Chow test [68]. The result suggests that the coefficients in all three linear regression models for three geological locations are equal (p < 0.01 in Figure 4a,c,e; p < 0.05 in Figure 4b,d,f; p < 0.001 in Figure 5a,c,e; and p < 0.05 in Figure 5b,d,f). Although correlation values and slope values between the duplex qPCR method and conventional methods were moderate to strong, statistical analyses reveal that geological location does not necessarily determine the correlation values and slope values in this study. According to this result, the methodology developed in this study should not be limited by location, but rather applicable to other study areas.
In this study, the slopes of correlations obtained for Taiwan main island, Kinmen islands and Matsu islands fall in a narrow range. Therefore, all the data for the three geographical locations were further correlated. Figure 6 demonstrates the good correlations for the data collected from all geographic locations and may provide a simple way to link the data between cell numbers and gene copies, and toxin concentrations and gene copies or cell equivalents, for the two studied cyanobacteria in reservoirs. In Figures 4-6, 95% prediction intervals were also calculated and displayed. As seen, the majority of data (94.7-100% in Figure 4, 95.6-100% in Figure 5, and 93.1-98.1% in Figure 6) fall within the 95% prediction intervals. This may suggest that the correlations developed are able to capture the variations of the tested cyanobacteria and metabolites in the reservoirs studied. As gene results for Microcystis and Cylindrospermopsis obtained from this study showed somewhat strong correlation values with the cell number information obtained from microscopy and the cyanotoxin concentration from ELISA, respectively, the qPCR-based method may be a better alternative for the monitoring of cyanobacteria and metabolites in natural water. In addition, since the qPCR system is able to distinguish toxic and nontoxic species, the approach has the potential to quickly provide useful information for water utilities to estimate the potential risks of toxins and toxigenic cells in the source waters.

Conclusions
Two sets of duplex qPCR systems were successfully developed for quantifying two commonly observed harmful cyanobacteria in this study; the first consisted of Microcystis and microcystin-producing Microcystis using 16S rRNA and mcyB, respectively; the second set consisted of Cylindrospermopsis and cylindrospermopsin-producing cyanobacteria using rpoC1 and pks gene, respectively. Field and laboratory tests showed that, in the duplex qPCR systems, presence of a highly abundant targeted gene may inhibit the detection of the other less abundant one. Although mcyB gene was detectable even when the concentration of 16S rRNA gene was four orders of magnitude higher, however, for the Cylindrospermopsis system, when either pks or rpoC1 gene copies were 10 2 -10 3 times higher than the other one, the system would fail to detect the lower abundant gene. Analysis of the samples collected from 29 reservoirs in Taiwan and her offshore islands show good correlation values between cell (equivalent) numbers and gene copies, and between toxin concentrations and gene copies for both Microcystis and Cylindrospermopsis systems, with R 2 mostly >0.60 for the reservoirs in three geographical locations. Relatively high coefficients of determination were found for the correlations between Microcystis cells and 16S rRNA gene (R 2 = 0.740), microcystin and mcyB (R 2 = 0.683), Cylindrospermopsis cells and rpoC1 gene (R 2 = 0.659), and cylindrospermopsin and pks gene (R 2 = 0.392), for all the samples collected in the reservoirs. The results demonstrate that the developed duplex qPCR systems were able to quantify the targeted cyanobacteria and toxin producing genes. In particular, the obtained gene information correlated well with cell numbers enumerated with microscopy and toxin concentrations measured with ELISA. The qPCR approach may be used as a tool for water utilities to diagnose the potential risk associated with toxigenic cyanobacteria and cyanotoxins in drinking water lakes and reservoirs.
Supplementary Materials: The following are available online at www.mdpi.com/1660-4601/14/5/547/s1, Figure S1: Locations of the studied reservoirs, Figure S2: Tests of inhibition on gene detection caused by different amounts of standard DNA using gel electrophoresis, Figure S3: The relationship between cell enumeration measured with microscopy and gene copy number with qPCR, Table S1: Detailed information of oligonucleotides, Table S2: Monitoring results of Microcystis and microcystins for the samples collected from Tai-Hu Reservoir (THR), Table S3: Monitoring results of Cylindrospermopsis and cylindrospermopsin for the samples collected from Tai-Hu Reservoir (THR), Table S4: The influence of primer concentration on the inhibition of gene detection, Table S5: Correlation between MCs/CYN concentrations and cell equivalents.