Extracellular microcystin prediction based on toxigenic Microcystis detection in a eutrophic lake

Existing models for predicting microcystin concentration in water body generally use chlorophyll or cyanobacteria concentration as input variables, although microcystins only originate from toxigenic strains of a few species. Moreover, the nonconcurrency between harmful algal growth and toxin release has yet to be quantified. Therefore, this study explored a new prediction method that considers these toxin production mechanisms for the eutrophic Yangcheng Lake, a large-scale drinking water source in China. The Lake was monitored weekly at six sampling sites from July to October in 2012, including the detection of toxigenic Microcystis (expressed as mcyA copy number) by qPCR. Compared with chlorophyll a, cyanobacteria, and total Microcystis abundance, toxigenic Microcystis concentration was more significant in predicting extracellular microcystin. Site-specific nonlinear regression models that link mcyA to microcystins were established. Parameters for toxin release delay (i.e., one or two weeks) were embedded in these models. Further analysis ascribed the different release timescale to NH3-N:TN and TN:TP ratios of approximately 0.015 and 9.2, respectively, which may decrease the delay in microcystin release. Model applications in determining mcyA monitoring frequency and its warning thresholds were discussed.

to represent all of the algae related to photosynthesis. Several cyanobacterial species can produce MCs during cyanobacterial bloom [8][9][10] . Toxins originate from toxigenic strains rather than all strains of these species 17 . Hence, Chla is apparently not the most relevant attributor to MC production. As a result, even if the correlation coefficient between Chla and the toxins is significant, the uncertainty level is relatively high, and the application of Chla is limited to lakes with absolute dominant toxigenic species of cyanobacteria 18,19 . In view of MCs originating from toxigenic cyanobacteria 20 , the amount of toxigenic strains is theoretically a suitable parameter for forecasting the MC level. Real-time quantitative PCR (qPCR) based on mcy gene sequences of Microcystis is a highly sensitive method for quantifying toxigenic strains of Microcystis 21,22 . This approach provides a solid foundation to elucidate the relationship between toxigenic Microcystis abundance and MC concentration.
Furthermore, a temporal discrepancy from strain growth to toxin release possibly exists because most MCs are only released into the water body after the lysis of toxigenic Microcystis cells. The inputs and outputs in existing models for extracellular MC prediction are generally synchronized variables. In other words, toxin level cannot be predicted without input assumed by model users. Therefore, a new model that describes MC release delay may allow the prediction of future toxin concentration on the basis of the current quantity of toxigenic Microcystis. Such a model contributes to warning the risk posed by MCs.
Toxigenic Microcystis, MCs, and other water quality items in the Yangcheng Lake were monitored in 2012 for this study. A new model that considers toxin release delay was proposed for predicting MC concentration on the basis of the amount of toxigenic Microcystis measured through qPCR. Quantification of the release delay enabled the estimation of MC concentration one or two weeks earlier than previous methods. Furthermore, the potential reason for the variation in delay period was discussed.

Results and Discussion
Temporal profile of toxigenic Microcystis and MC release delay. During the sampling period, total Microcystis (expressed as 16S rDNA copy number) and toxigenic strains (expressed as mcyA copy number) in the Yangcheng Lake exhibited similar temporal trends; that is, they both showed peaks on week 9 to 11 in September ( Fig. 1) with an average abundance of 1.5 × 10 7 and 7.5 × 10 6 copies/L, respectively. Total Microcystis grew gradually and steadily from week 4 to 11, while the toxigenic strains increased sharply after week 8. The proportion of toxigenic Microcystis to total strains reached approximately 69% at the peak compared with the 8% in other months, indicating that toxic genotypes achieved a higher growth in September. Because cell quotas for MCs vary with the growth rate of natural Microcystis 23,24 and the toxicity effect increases by 10-to 1000-fold when the Microcystis community is changed to toxic strain predominance 25,26 , the monitoring results indicated that toxigenic strains were more likely turning into the exponential growth phase by the end of August and at the beginning of September, while intracellular toxins began to accumulate in this phase.
Previous researchers found that the maximum extracellular MC concentration does not concurrently occur with peak value of mcyA copy number as well as of cyanobacterial abundance. Toxin concentration poorly correlates with these two variables (P > 0.05) 27,28 . In this work, the successive and long-term field monitoring study revealed the details of the inconsistency. Firstly, correlation analysis supported the previous findings. P-values of the hypothesis tests verified that neither the linear nor natural log-linear relationships between variables for abundance (including Chla, cyanobacteria, 16S rDNA, and mcyA) and MCs were significant (Table 1, see the column of τ = 0 for reference). Moreover, the monitoring data indicated a distinct time difference between the occurrence of the maximum values for cyanobacteria, 16S rDNA, mcyA, and MCs. Taking the mcyA as an example (Fig. 2), the mcyA and MC concentrations at the six sites gradually increased from week 7, and then the mcyA of all sites peaked around week 11. Afterward, MC concentrations steadily increased and reached the maximum values one or two weeks later, i.e. around week 12 or week 13, whereas mcyA copy numbers decreased.
In general, intracellular toxins increase with the increase in toxigenic Microcystis biomass and reach the maximal value at the end of the exponential growth phase 29 . By contrast, intracellular toxins are not immediately released. Most MCs are completely released into the water body when the mature cells of toxigenic Microcystis are injured to lyse or grow to death. For the nonconcurrency between toxigenic Microcystis growth and MC release   was verified in the Yangcheng Lake, the extracellular MC concentration could only be well estimated through toxigenic Microcystis abundance by considering delay period.

MC prediction models.
In the current management framework, chlorophyll concentration or cyanobacteria abundance is generally used to predict MC concentration in the water. Although chlorophyll concentration and cyanobacteria are easily monitored, model performance requires careful validation, and high uncertainty impedes further application. To identify the best predictor for MC concentration estimation for the Yangcheng Lake, regression models were tested with different independent variables (Chla, cyanobacteria, 16S rDNA, and mcyA concentration) as input and MC concentration as output. Release delay was added as another parameter into each model to quantify the nonconcurrent phenomenon discussed in the previous section. The general model structure was described using the following equation: where t is the sampling time (week), τ is the delay period of MC release (week), i is the site number, Y t, i is the concentration of MC (μ g/L) at week t of site i, and X t−τ, i is Chla (mg/L)/cyanobacteria (cells/L)/16S rDNA (copies/L)/ mcyA (copies/L) at week (t− τ) of site i. Monitoring data were recorded weekly; thus, different delay periods (1 to 4 weeks, i.e., τ = 1, 2, 3, 4 weeks) for each sampling site were tested for the best data fitting. Results demonstrated that mcyA concentration was the most significant input (with the least P-value) for predicting MC concentration (Table 1). An optimum regression relationship was found between mcyA and MC concentrations in each site when considering site-specific delay period (Table 1, Fig. 3). The MC concentrations at sites 1, 2, 5, and 6 strongly correlated with mcyA concentration obtained two weeks beforehand (P < 0.05). This result indicated that MCs were released into the surrounding water two weeks after the growth of toxigenic Microcystis. At sites 3 and 4, MC concentration strongly correlated with mcyA concentration obtained a week ahead of time (P < 0.05). This result demonstrated that the delay period of MC release was only one week at these two sites.
Since the underlying quantitative toxin release characteristics of MCs were identified, the models enabled the prediction of MCs one or two weeks in advance, when real-time qPCR tests were used on the basis of continuous weekly sampling. This prediction method supports a new framework for the risk management of water bodies with cyanobacterial bloom and MC pollution. Toxigenic Microcystis abundance can be used as a significant indicator for the early warning of potential MC hazards, with the corresponding qPCR test widely applied in the lake monitoring.

Different MC release delays and nutrient conditions. No relation was found between toxigenic
Microcystis abundance and different toxin release delay periods in the Yangcheng Lake. The delay of MC release possibly depends on other factors. The difference in time delays in the models is discussed below, with the focus on nutrient conditions.
In the Yangcheng Lake, Wilcoxon signed ranks tests were used to cross-verify differences in environmental factors that include TN, TP, NH 3 -N, TN:TP ratio, and NH 3 -N:TN ratio between sites with different toxin release delays. Sites with different τ values significantly differed (P < 0.1) in NH 3 -N:TN and TN:TP ratios ( Table 2). This result validates the previous finding that nutrients affect MC release. In addition, for the eutrophic Yangcheng Lake, nutrient ratios rather than individual nutrient concentration level influence the delay period.
In the Yangcheng Lake, statistical test results suggested that both NH 3 -N:TN and TN:TP ratios may cause the MC release delay to vary between one and two weeks. On the basis of the differences in delay period and the characteristics of nutrient ratios, the six sites were classified into three groups: sites 1, 2, and 5 in group one; sites 3 and 4 in group two; and site 6 in group three. Group 1 showed significantly higher NH 3 -N:TN level and lower TN:TP level than Group 2, whereas Group 3 had significantly lower NH 3 -N:TN and higher TN:TP ratios than Group 2. Descriptive statistics of nutrient ratios for three groups are shown in Table 3. This result suggests that the delay period is shortened to one week when the NH 3 -N:TN and TN:TP ratios are within a certain range. Taking the medians of nutrient ratios (0.015 for NH 3 -N:TN and 9.2 for TN:TP) in group 2 as references, we hypothesized that NH 3 -N:TN and TN:TP ratios of approximately 0.015 and 9.2 possibly shortened the delay in MC release. Further research should explore the quantitative mechanisms underlying the time scale of MC release delay and potential influential factors.

Potential model applications in MC risk management.
The data obtained from weekly monitoring of MC concentration in the Yangcheng Lake in 2012 did not exceed the WHO recommended value of 1.0 μ g/L for dissolved toxins in drinking water. However, MCs considerably accumulated in aquatic products, such as fish and crabs, and fish consumption was considered as a main channel for human exposure [30][31][32] . The highest estimated daily intake for bioaccumulation is 38 times higher than the WHO recommendable maximum daily intake 33,34 . The associated toxic effect posed by MCs on water and food from the lake, as a large-scale source for potable water and freshwater products, is bound to be a concern. Potential applications of the prediction model in MC risk management are discussed below.
The model is helpful to formulate site-specific monitoring plans on toxigenic Microcystis detection. The quantitative relationship between MCs and toxigenic Microcystis showed a variation in the slopes of the regression curves, with the largest slope at site 2 (0.094) and the smallest slope at site 1 (0.034). The slopes indicate the sensitivity of MC concentration to the change in the amount of toxigenic Microcystis. The MC concentration at sites with steep slopes (e.g., site 2) would sharply change even with a minimal variation in mcyA concentration. As a result, a high sampling frequency is needed for mcyA measurements until mcyA reaches its peak value to predict an unexpected augmentation of MC risk. Furthermore, the monitoring frequency of mcyA value must be increased at sites with short delay periods. For example, the MC concentration at site 3 changes with mcyA after a week of delay. Thus, monitoring should be performed more frequently than every week from late August to early October. By contrast, at site 1, the delay period for MC release is two weeks and the MC concentration would not change significantly when mcyA concentration varies. Therefore, mcyA content could be measured once every two weeks to predict MC risk. Consequently, a rough guidance on mcyA test frequency for the Yangcheng Lake aiming at MC risk warning was proposed ( Table 4).
The model can also be used to derive different warning levels of mcyA concentration in accordance with MC control requirements. To make the process convenient and time efficient, regression curves of various delay periods for the six sites were integrated into Y t = 0.056ln[X (t−τ) ] − 0.407 (τ = 1 or 2; P < 0.001) for the entire lake. Once mcyA is measured at a specific location, for the western and eastern areas of the Yangcheng Lake, the concentration of MC two weeks later can be estimated using the above equation. The concentration of MC one week later in the inner area of the Yangcheng Lake can also be estimated using the equation. The confidence interval for MC concentration was narrowed down compared with the separate equations for individual sites. The P-value also confirmed an apparent uncertainty decrease (P < 0.001). This generalized equation can be used to infer the threshold of toxigenic Microcystis abundance once the MC concentration limit is pre-set. For instance, a two-level warning threshold was obtained by the inverse calculation when the extracellular MC control limit was set to 1.0 μ g/L (standard compliance value for drinking water source) and 0.5 μ g/L (assuming daily MC exposure from drinking water is approximately equal to the exposure from fish consumption 31 ). The calculation results suggested that the lower and higher warning level of mcyA concentration should be set as 1.0 × 10 7 and 8.0 × 10 10 copies/L, whose corresponding 95% confidence intervals of predicted MC concentration are [0.45, 0.55] and [0.79, 1.20] μ g/L, respectively. Taking the model uncertainty into account, the threshold value can be lowered or raised for the sake of more conservative or more adventurous risk management. For instance, the threshold 1.0 × 10 7 copies/L could be lowered down to 7.2 × 10 6 copies/L or raised up to 2.3 × 10 7 copies/L, while the probability that MC concentration exceeds 0.5 μ g/L would be reduced from 50% to 10% or increased from 50% to 90%.

Methods
Sample collection and processing. The Yangcheng Lake, located between the Taihu Lake and the Yangtze River in China, covers an area of 120 km 2 . This lake is a drinking water source for Kunshan City and a backup water source for Suzhou City, serving a population of 11 million. Moreover, the lake is commercially important because of fishery production. The Yangcheng Lake has been repeatedly subjected to toxic algal blooms and microcystin pollution since the 1990 s because of wastewater inflow and aquaculture development. Monitoring and controlling the MCs in the Yangcheng Lake is a government priority in consideration of human health protection and economic growth in the region.
A total of 81 water samples from six sites of the Yangcheng Lake were collected weekly from July to October in 2012. The sampling sites covered the western, middle, and eastern sections of the lake (sites 1 to 6). The corresponding weeks were numbered as 1 to 3 (July), 4 to 7 (August), 8 to 12 (September), and 13 to 15 (October).
Routine water quality parameters, including water temperature, pH, dissolved oxygen concentration, Chla concentration, and cyanobacterial biomass, were measured on site with a multi-probe sensor (YSI 6600, USA). Approximately 1 L of water was collected with a vertical water sampler at 0.5 m depth under the water surface, instantly stored on ice, and then delivered to the laboratory for further analysis. Nutrients including NH 3 -N, TN, and TP were analyzed in accordance with standard protocols 35 . Real-time qPCR for total and toxigenic Microcystis quantification. Water samples (100 mL) were centrifuged at 12000 rpm for 10 min at 4 °C as previously described, and DNA was extracted using Universal Site 6* higher than site 6 (P: 0.007) higher than site 6 (P: 0.003) no significant difference lower than site 6 (P: 0.007)   Genomic DNA Extraction Kit Ver.3.0 in accordance with the manufacturer's instruction (TaKaRa Code: DV811A). For each sample, 50 μ L of DNA was obtained and stored at − 80 °C for preparation. The total and toxigenic Microcystis were quantified on the basis of the copy numbers of the 16S rDNA and mcyA genes, respectively [36][37][38] . Amplification was performed using 16S rDNA primers (16S-F: 5-TATTGGGCGTAAAGCGTCCT-3; 16S-R: 5-AACCACATACTCCACCGCTT-3) and mcyA primers (mcyA-F: 5-TTTCATCTCCATCGCCGCA-3; mcyA-R: 5-GGACTCCCACTTGTTTACCGA-3). The total and toxigenic Microcystis sequences were 396 and 376 bp long, respectively. Real-time qPCR was conducted using an Applied Biosystems 7500 Fast Real-Time PCR System (USA). Subsequently, 1 μ L of DNA template, 0.1 μ L of (20 μ M) primers, 5 μ L of 2 × SYBR Premix Ex Taq TM (TaKaRa, Dalian, China), and 3.8 μ L of dH 2 O were mixed. PCR parameters include 95 °C for 10 s; 40 cycles of 95 °C for 5 s, 59 °C for 30 s, and 72 °C for 30 s; melting curve analysis at 95 °C for 15 s; and annealing at 60 °C for 1 min. Standard curves were prepared on the basis of diluted plasmid DNA, and double distilled water was used as the negative control. Purified plasmid cDNA was quantified using UV spectrophotometry (260 nm). The copy numbers of 16S rDNA and mcyA were determined through qPCR to represent the relative abundance of total and toxigenic Microcystis, respectively 39 .
In our previous study, the relationships between the cell numbers of M. aeruginosa counted by microscopy and the copy numbers of the 16S rDNA and mcyA genes detected by qPCR indicated highly significant correlation coefficients. Linear regression relationships of 16S rDNA (y = 0.989x, R 2 = 0.8087) and mcyA (y = 1.102x, R 2 = 0.8863) to cell numbers were then used to estimate the cell number of Microcystis and potentially toxigenic Microcystis genotypes in each sample. These linear relationships suggest that M. aeruginosa (FACHB-905) contains one copy number each of 16S rDNA and mcyA genes. This suggestion is consistent with previous studies that showed equivalent cells of total and toxigenic Microcystis calculated from the copy number of the 16S rDNA 40 and mcyD genes 41 . However, the number of gene copies in each Microcystis cell may vary. Kaneko et al. 42 showed that two copies of the mcy gene are carried by each Microcystis cell and that 4 to 250 copies of the 16S rDNA gene are carried per Microcystis cell 40,43 . In the present study, the copy numbers of the 16S rDNA and mcyA genes were used to indicate the abundances of total and toxigenic Microcystis in water samples.

Detection of MCs.
Enzyme-linked immunosorbent assay kits (J&Q Environmental Technologies Co., Ltd.) were used to measure MC concentration in triplicate 44 . MC-monoclonal antibody was used, which can react with a majority of the microcystins-producing variants, thereby representing the total MCs 45 . The concentration of the complete antigen was 5 μ g/mL. The dilution of the monoclonal antibody and enzyme tracer was 1:3000. The detection limit was 0.05 μ g/L.
Statistical analysis. All data were processed and analyzed using SPSS Statistics 18.0 (USA). The quantitative relationship between toxigenic Microcystis and extracellular MCs was deduced from a nonlinear regression process (P < 0.05). Wilcoxon signed-ranks tests on the NH 3 -N:TN and TN:TP ratios between the sites were employed to identify possible effects of these nutrient levels on the differences in the delay period for MC release. Data obtained at the same time from two sites were paired and compared.