Analysis of multiclass cyanotoxins (microcystins, anabaenopeptins, cylindrospermopsin and anatoxins) in lake waters using on-line SPE liquid chromatography high-resolution Orbitrap mass spectrometry

An on-line SPE-UHPLC-HRMS method was optimized for filtration, on-line SPE, and HRMS conditions for the rapid screening of 17 cyanotoxins. 8 cyanotoxins were detected with 75% of lakes containing MC-LR and 38% containing anabaenopeptins (A or B).


Introduction
Cyanobacteria are on the radar worldwide due to the increase of harmful algal bloom (HAB) episodes caused by global climate change and eutrophication due to anthropogenic activities such as agriculture. [1][2][3][4][5] Under favorable conditions, these prokaryotic organisms produce harmful secondary metabolites known as cyanotoxins. These are noxious to aquatic ecosystems and may impact human activities through contamination of drinking water resources and recreational waters. 6 Some cyanotoxins from the two main groups, peptides and alkaloids, have been linked to acute, chronic and lethal poisoning of wild and domestic animals, and many cases were also reported in humans. 7 Considering the future global warming model, one can assume that these harmful episodes are on the rise. 8,9 A broad and effective monitoring of cyanotoxins is therefore needed to better understand and predict the emergence of HABs.
Among the secondary metabolites produced by cyanobacteria, the rst group includes non-ribosomal oligopeptides also called cyanopeptides, counting hundreds of identied bioactive compounds to date. 10 We mainly associate the presence of toxic cyanobacterial blooms with the appearance of hepatotoxic microcystins (MCs) (Fig. 1). These are the most documented cyanotoxins and are associated with numerous cases of human intoxication due to their high occurrence during HAB episodes. 11 Their hepatotoxicity ultimately results from the binding to protein phosphatases 1 and 2A, followed by disruption of cellular homeostasis. The most acute cases lead to liver necrosis and colorectal and liver cancer. 12 Other bioactive cyanopeptides, such as anabaenopeptins (APs), cyanopeptolins and microginins, are, so far, not considered in most monitoring plans. However, it was demonstrated that MCs rarely appear alone during HAB episodes, and specic congeners from other cyanopeptides can even be dominant. 13 Some studies suggest that some APs inhibit protease and protein phosphatases. 14 AP-B and F have also been shown to induce lysis of cyanobacteria, which could inuence the bioavailability of cell-bound cyanotoxins. 15 Over 500 cyanopeptides have been structurally identi-ed, including more than 240 MCs. 16,17 The factors that lead to their production and their impact on human and animal health have only been studied for a few compounds. Cyanopeptides other than MCs have received little attention in recent years, yet their impact on ecosystems and human and animal health is of concern. 10 For this reason, a more in-depth study of the presence of these compounds and their toxicological impact would be essential for a better implementation of risk management.
Besides bioactive and toxic peptides, some cyanobacteria can also produce toxic alkaloids. Among them, the most common are anatoxin-a (ANA-a), homoanatoxin-a (HANA-a) and cylindrospermopsin (CYN). ANA-a and HANA-a are neurotoxins with a vast distribution in water sources. 18,19 To date, many documented animal mortalities have been attributed to ANA-a. 20 They mimic the neurotransmitter acetylcholine and if ingested at high doses, continuous stimulation of the respiratory muscles may lead to immediate asphyxiation. 21 CYN is originally present in tropical and subtropical climates. Nevertheless, due to increasing temperature and cyanobacteria adaptation, the presence of CYN may be increasingly reported in originally temperate climates such as in Europe and North America. 22,23 CYN exhibits cytotoxic, neurotoxic and hepatotoxic effects, acts as a tumor promoter and carcinogen, and affects organs by inhibiting protein synthesis. 24 In recent years, new guidelines have been proposed for several cyanotoxins and different age groups. The World Health Organization (WHO) proposed a 1 mg L À1 threshold in drinking water for both free and cell-bound MC-LR. 25,26 Several countries have adopted this regulatory value for the sum of all MC congeners that can be quantied, but some countries have also suggested maximum levels for ANA-a and CYN. 27,28 The US EPA has implemented guidelines for MCs and CYN in drinking water for adults and children. Four US states (Minnesota, Ohio, Oregon and Vermont) have also included ANA-a and the "paralytic shellsh poison" saxitoxin in their guidelines. 29,30 Ohio has introduced "do not drink" and "do not use" notices when cyanotoxin thresholds for drinking and recreational waters are reached, respectively. 31 Quebec, on the other hand, has adopted a guideline for the quality of drinking water which must not contain more than 1.5 mg L À1 of MCs, expressed as MC-LR toxic equivalents. 32 While guideline harmonization is not always achieved between different states, upstream work is also needed for understanding and predicting HABs. Improving toxin tracking using effective analytical methods appears as a critical research need, which is still hindered by the limited availability of certied standards and materials.
Environmental monitoring authorities face major challenges in the analysis of cyanotoxins. The wide variety of cyanotoxins and increasing number of water sources affected by cyanobacteria require high-throughput analytical approaches able to quantify multi-class toxins in environmental matrices. 10,33 The method of choice to obtain such a degree of selectivity is liquid chromatography (LC) coupled to mass spectrometry (MS). Few methods have been able to quantify multi-class toxins, mainly due to extraction limitations and the range of properties of the targeted cyanotoxins. Solid-phase extraction (SPE) is usually preconized, as it provides better sensitivity and reproducibility. 13,[34][35][36] However, SPE sample preparation is time consuming and can be limiting when one has to promptly process hundreds of samples during a bloom season. One solution was to develop an automated extraction method directly coupled to LC-MS analysis. The use of on-line SPE has proven effective in past studies aiming at including diverse cyanotoxins in a single high-throughput method, without signicantly affecting sensitivity. [37][38][39][40][41][42][43] The methods proposed earlier can still be improved by expanding the scope of targeted cyanotoxins for a wider screening. Some studies have also explored the possibility of using high-resolution MS to quantify known cyanotoxins and identify unknown cyanopeptide variants using LC-QTOF or LC-Orbitrap. 35,40,41,44 HRMS is increasingly used as a powerful tool to perform targeted analysis, with the further advantage of allowing suspect-target and non-target screening when certied standards are not available.
This study presents the application of an optimized rapid chromatographic method using on-line SPE-UHPLC coupled to quadrupole-Orbitrap HRMS for the determination of multiclass cyanotoxins in environmental surface water samples. For method development and validation purposes, 17 quantitatively targeted cyanotoxins were rst listed, which included CYN, ANA-a, HANA-a, AP-A, AP-B and 12 MCs ([Asp 3 ]RR, RR, YR, HtyR, [Asp 3 ]LR, LR, HilR, WR, LA, LY, LW and LF). As part of the method optimization, different lters were tested for adequate sample pre-treatment. On-line SPE parameters were also investigated for a faster analysis without sacricing recovery and accuracy performance. In this respect, the addition of a washing step aer sample loading was critical for efficient matrix effect removal. Detection through HRMS allowed the mass distinction of ANA-a from its isobaric interference, the naturally encountered amino acid phenylalanine (Phe). 45 Full scan (FS) and parallel reaction monitoring (PRM) modes were investigated to keep a maximum of information in the acquired data and improve the identication of difficult-to-measure compounds. Samples were collected through Adopt a Lake, a citizens' campaign throughout Quebec and Canada where people volunteered to collect samples from a nearby lake or waterbody (Fig. 2). 46 The samples from the various lakes (N ¼ 16) were analyzed for the 17 quantitatively targeted cyanotoxins. In addition, a retrospective screening of extracted full scan HRMS chromatograms was conducted to investigate the presence of less frequently monitored compounds. To the best of the authors' knowledge, this method includes the widest range of cyanobacterial toxins in a single automated workow. MC-LR and MC-RR were kept at À20 C for a maximum of six months. All other individual stock solutions were prepared in methanol (MeOH) at a concentration of 25 mg L À1 and were kept at À20 C for a maximum of one year. Primary working solutions were prepared at a concentration of 100 mg L À1 for targeted cyanotoxins and 9 mg L À1 for internal standards (ISs: MC-LR( 15 N 10 ) and NOD-R) by dilution in MeOH of individual stock solution aliquots. Subsequent working solutions were prepared daily in water to give solutions of desired concentration. All organic solvents and water used for dilutions were of HPLC grade purity from Fisher Scientic (Whitby, ON, Canada).

Sample collection and preparation
Samples were collected through Adopt a Lake, a citizens' campaign throughout Quebec and Ontario, Canada. The participants of this campaign volunteer by collecting a sample from a lake. Samples were collected in 2018 between May and the end of September, mostly from recreational lakes (Fig. 2). At each sampling location, a duplicate set of samples was collected in 125 mL amber polyethylene terephthalate glycol-modied (PETG) bottles (Thermo Scientic Nalgene, Waltham, MA, USA). 47 The PETG bottles were previously rinsed three times with the surface water from the site. The bottles were then lled to the brim, sealed, stored at À20 C until shipment and sent to the laboratory within 3 days. Both free and cell-bound cyanotoxins were assessed by quantifying both fractions together. Upon arrival at the laboratory, the whole samples underwent three freeze-thawing cycles (to lyse cyanobacterial cells) and were subsequently ltered through 25 mm diameter, 0.2 mm pore size Acrodisc GH Polypro (GHP) lters (Waters, Milford, MA, USA) using a vacuum ltration unit with a vacuum pump. A volume of 1450 mL of each ltered sample was transferred into 2 mL amber glass vials and kept at À20 C until analysis. Prior to analysis, the internal standards were added for a nal concentration of 300 ng L À1 .  Table 1 and were adapted from a previous study with some modications. 38 A washing step was added to the initial method for matrix removal. The wash volume was investigated at three levels (0, 0.5 and 1.0 mL of wash volumes), using surface water (lake) samples spiked at 250 ng L À1 with native compounds. Under the optimized conditions, 1 mL of the sample was injected and the sample transfer time (loading speed) from the injection loop to the SPE column was 1 mL min À1 . An additional wash volume of 0.5 mL was allowed to ow through the column following the sample loading step. The on-line pre-concentration column was then eluted using the analytical pump gradient; the eluting analytes were transferred directly to the analytical column for separation. Chromatographic separation proceeded using a gradient program with solvents ACN (B) and water (A) (both with 0.1% formic acid) at a ow rate of 550 mL min À1 . The method run time was 8 min per sample.

Instrumental conditions
2.3.2 HRMS conditions. A heated electrospray ionization interface (HESI-II) operated in positive mode was used for analyte ionization. The source parameters were set as follows: the ionization spray voltage was set at +3500 V; capillary temperature was set at 350 C; the vaporizer temperature was set at 250 C; sheath gas and auxiliary gas ow were set at 60 and 15 arbitrary units, respectively. HRMS parameters such as resolving powers, automatic gain control (AGC) and maximum injection time (IT) were selected for optimal selectivity and sensitivity. CYN, ANA-a, HANA-a, MC-LA and MC-LW were monitored using the parallel reaction monitoring (PRM) scan mode to ensure a better selectivity. High noise was observed for complex matrices using the full scan (FS) mode at different tested resolving powers. Therefore, a PRM scan was set between 1.5 and 4.7 min for those difficult-to-measure compounds. For the PRM mode, the resolving power was set at 17 500 full width at half maximum (FWHM) at m/z 200 and the AGC and IT parameters were set respectively at 1 Â 10 5 ions capacity and 50 ms lling. Other targeted cyanotoxins were acquired using the FS mode. For the FS mode, a mass range of m/z 150-1200 was applied, and the resolving power was set at 70 000 FWHM at m/z 200 (for singly charged microcystins, the corresponding resolution at m/z 1000 was $31 000 FWHM). The scan rate was 3 Hz, yielding between 9 and 30 points per chromatographic peak depending on the particular compound (on average, 12 points per chromatographic peak). Table 2 presents the detailed HRMS optimized parameters including retention time (RT), selected

Analyte identication and quantication
The positive identication of target analytes was conrmed by matching chromatographic retention times within AE2% of those from standards spiked in matrix-matched surface water. For a suitable selectivity in data analysis, the exact mass tolerance was set to AE5 ppm for the extracted m/z values from acquisition. 48 The quantication of targeted cyanotoxins relied on relative responses of the area of chromatographic peaks, considering the area ratio of native analyte to the selected ISs, NOD-R or MC-LR( 15 N 10 ). The concentration in eld samples was derived from the area ratio of chromatographic peaks in the particular sample and the matrix-matched surface water calibration curve. If results were found to be outside the dynamic range, a new sample preparation was performed using a dilution factor so that the concentration can t in the calibration curve. The full scan HRMS extracted chromatograms were also subjected to retrospective screening of additional cyanotoxins. These compounds were previously reported in Meriluoto et al.
(2017) and were screened within a AE5 ppm window of their theoretical exact mass. 17

Method validation
For all the validation steps and quantication quality controls (QCs), analytes were spiked in composite surface waters consisting of analyte-free lake water sampled before HAB seasons (referred to as matrix-matched surface water) and the absence of analyte was conrmed by LC-MS analysis. A ltration step was used prior analysis to remove suspended cell debris from the surface water samples. To lessen the impact of the lter material on cyanotoxin concentrations, we set out to compare two lter types: 0.2 mm GH polypro (GHP) lters vs. 0.2 mm nylon lters (Sterlitech Corporation, Kent, WA), both reported to be optimal. 38,41 Five replicates were spiked at a midlevel concentration (200 ng L À1 ) and recovery values (%) were obtained with the ratio between samples spiked before ltration and aer ltration.
Method detection limits (MDLs) and method quantication limits (MQLs) were derived as 3 and 10 times, respectively, the standard deviation of replicate analyses of spiked surface water samples (n ¼ 10). The spike concentration was selected as approximately 5 times the estimated MDL (between 10 and 50 ng L À1 , depending on the analyte). Linearity was evaluated based on an 8-point internal calibration curve prepared with matrixmatched surface water at concentration levels ranging from the MQL to 1000 ng L À1 (IS concentration set at 300 ng L À1 ). Relative matrix effects were evaluated by using standard addition calibration curves. The slopes in matrix-matched surface water samples were compared to those in three selected samples from this study. Extraction recoveries were determined in a previous study by comparing the mean peak areas of the selected cyanotoxins from a direct chromatographic injection with those from on-line SPE injection. 38 Accuracy, dened as the quotient of the obtained vs. expected concentration (%), and intra-day precision, expressed as relative standard deviation (RSD, %), were determined in matrix-matched surface water at 4 spiking levels selected within the linearity range (75 and 120 ng L À1 , n ¼ 7; 200 and 800 ng L À1 , n ¼ 5). Inter-day precision was also evaluated at 4 spiking levels and was derived from the relative standard deviation (%RSD) obtained over 3 different days.

Data handling and statistical analyses
Data treatment was performed using Xcalibur 3.0 Soware (Thermo Fisher Scientic, Waltham, MA, USA). Chromatographic peaks of targeted cyanotoxins were integrated using Xcalibur Qual Browser according to their experimental mass (m/ z) and retention time ( Table 1). Calibration curves and sample quantication were based on area ratios of native compounds to their corresponding IS. Statistical tests were performed with the Statistical Package for Social Science (SPSS 21.0, Chicago, IL, USA). ANOVA and Tukey's post hoc tests were used with statistical signicance dened as P < 0.05.

Filtration conditions
The choice of an appropriate ltration material to remove suspended cell debris is critical to ensure sample integrity. Undesirable interactions between the lter material and the analytes of interest may negatively affect the total recovery of the method. This was extensively discussed in previous cyanotoxin studies and it was demonstrated that the mostly used membrane for MCs, glass ber lters (GF/F), was not showing optimal results when analyzing a higher number of multi-class cyanotoxins due to their diverse range of physicochemical properties. 38 Therefore, two types of materials have been reported as optimal for environmental waters, depending on the targeted cyanotoxins: nylon and polypropylene (GH polypro-GHP). 38,41 Nylon and GHP membranes are both hydrophilic, implying that lower interactions would occur with less hydrophilic compounds such as MCs. GHP lters are made of hydrophilic polypropylene, which is supposed to reduce unwanted interactions making them potentially universal lters compared with polypropylene whose surface is known to adsorb MCs. 49 Considering the broad range of cyanotoxins in the present study, analyte recoveries were evaluated using these two materials (Fig. 3). Membranes with a pore size of 0.2 mm can remove the majority of cyanobacterial cells while reducing the amount of particles entering the chromatographic columns to ensure a longer lifetime. Although nylon offers good recoveries for a number of compounds, there are some exceptions and mean recoveries are from 8 to 100%. GHP yielded better recoveries for all targeted compounds with mean recoveries varying between 82 and 108%, without exceptions. These results are in accordance with previous studies showing virtually no analyte losses using this membrane type. 41,50

Optimization of the on-line SPE procedure
On-line SPE conditions were optimized according to previous work that detected six MCs, ANA-a and CYN using an SPE-UHPLC coupled to a triple quadrupole mass spectrometer. 38 The injected volume was set at 1 mL for high-throughput convenience while limiting analyte breakthrough. Additionally to the optimized parameters, a wash step was added consecutively to sample loading. The on-line aqueous mobile phase is maintained aer sample enrichment, but before the elution step. The wash volume passing through the column should be optimized so that satisfactory matrix removal is attained while limiting analyte breakthrough. The inuence of wash volume (no wash applied vs. 0.5 or 1 mL) for selected cyanotoxins is illustrated in Fig. 4. Between 0 mL and 0.5 mL, the analyte signal improved signicantly (e.g., 2.5Â for MC-YR, 2.5Â for [Asp 3 ]MC-LR, 3Â for MC-LF, 1.5Â for CYN). The improvement between 0 mL and 0.5 mL is modest for more hydrophilic compounds such as ANA-a and HANA-a, presumably due to some breakthrough losses. Between 0.5 mL and 1 mL, the analyte signal either improved by a small margin (e.g., MC-YR), or decreased signicantly, especially for more hydrophilic compounds (e.g., CYN, ANA-a). In view of the above, we selected a 0.5 mL wash volume. Extraction recoveries depend on organic modiers and pH (which was maintained the same) and injection volume, which was lowered compared to our previous method. 38 According to this study, on-line SPE recoveries evaluated with spiked matrixmatched water were acceptable and ranged from 72 to 102% for the range of targeted cyanotoxins.
Historically, the most challenging chromatography step was the adjustment of the elution gradient conditions for analysis of ANA-a when the detection is conducted with standard resolution mass spectrometry. 24,[51][52][53] Phe is an isobaric compound to ANA-a, detected within algal blooms, and with the two most abundant fragments of ANA-a in common. If not well-separated from ANAa, Phe could thus result in false positives during MS detection. In our previous study using liquid chromatography coupled to tandem mass spectrometry, two different organic eluents were included in the chromatographic gradient to achieve resolution of ANA-a from its isobaric interference. 38 However, HRMS has adequate resolving power to distinguish both compounds. 45 In view of this, the choice of organic solvents and the elution gradient program was simplied without compromising the integrity of the analysis (Section 2.3).

HRMS conditions
The evaluation of MS scan modes used for quantication purposes is essential to ensure reliable data, depending on the compounds and matrices.  demonstrated that the PRM mode can provide better sensitivity and selectivity for a majority of targeted veterinary antibiotics in swine manure compared to the FS mode using a Q-Orbitrap mass spectrometer. 48 Moreover, the PRM mode is commonly used with triple quadrupole mass spectrometers for the determination of MCs using the common product ion m/z 135 for quantication. 17 However, using such a mode does not allow acquisition of expanded data on the samples in order to obtain a retrospective analysis of suspect MCs. Nonetheless, the FS mode can be efficient for quantitative purposes and this has been conrmed in several studies. [54][55][56] As mentioned above, the use of highresolution FS presents other advantages, for instance the retrospective analysis of suspect-target analytes in environmental samples. To determine which of the two scan modes is favored for better identication, the extracted chromatograms from unspiked surface water were compared to the same water spiked at 200 ng L À1 in both scan modes. ANA-a, HANA-a and CYN are early-eluting, low mass compounds and, unsurprisingly, undesirable peaks can be observed in their mass range. For this reason, these compounds were detected using the PRM mode to improve the condence in their identication. MC-LW and MC-LA were two other difficult-to-measure compounds for which the PRM mode proved useful. The illustrative chromatograms in Fig. S1 † demonstrate the improvement of iden-tication using the PRM mode. The other targeted compounds were satisfactorily extracted from the FS analysis. Both scan modes were operated in parallel between 3 and 4.7 min, but the data acquired for each chromatographic peak were sufficient to be quantitative despite the increased duty cycle. Using the combined PRM-FS approach, suitable chromatographic performances were obtained in the surface water matrix for all the targeted compounds (Fig. 5). The full scan MS mode was also useful for conducting a fast retrospective screening of additional cyanotoxins not included in the list of quantitatively targeted analytes.

Analytical method validation
Matrix-matched calibration curves were used to limit the impact of matrix effects on the reported results. The matrix-matched water consisted of composite surface water sampled in the lakes under study in spring, before the HAB season to ensure that it is analyte-free. Whether this composite matrix would correctly compensate matrix effects for individual lake samples should, however, be validated. These lakes are oen more loaded with organic matter and nutrients during the summer season in addition to the potential increase in biomass caused by the appearance of cyanobacterial cells. Therefore, the matrix effects caused by these potentially more complex summer water samples were assessed by comparison with the matrix-matched water sample. For this purpose, relative matrix effects were evaluated by comparing the slope of the matrix-matched calibration curve (used for quantication in the present study) vs. the slopes of standard additions to selected lake samples from this study: Lac Mimi, Lacà la Tortue, and Lac Juneau. Positive matrix effect values (%) indicate signal enhancement, while negative values indicate signal suppression. Matrix effects remained in most instances within AE20% (Table 3), an acceptable range for complex environmental samples, given the limited number of internal standard types available. Signicant matrix effects were observed for CYN when using NOD-R as the internal standard (unpublished data), but the matrix effect was lowered within AE30% when using isotopically labelled MC-LR (Table 3). Finding appropriate ISs for the analysis of emerging contaminants is oen a challenge, and cyanotoxins are no exception since only a few isotopically labelled compounds are available in the market. Matrix-matched calibration curves showed suitable linearity, with R 2 $ 0.998 (Table 3). Residuals for the different calibration levels were within AE20% of the linear t, except for the rst calibration level close to the LOQ. The precision (%RSD) and accuracy (%) were also suitable as shown in the ESI (ESI Table  S1 †). Intra-day precision ranged from 1.2 to 17%, with the exception of MC-LY with a RSD of 48% at the low spike level (120 ng L À1 ). Inter-day precision varied from 3.1 to 19% with the exception of MC-LY with a RSD of 31% at the low spike level (120 ng L À1 ). The MDL and MQL values ranged from 8 to 53 ng L À1 and from 27 to 176 ng L À1 , respectively (Table 3). These values are substantially lower than the established and proposed water quality guidelines from different jurisdictions. Water quality guidelines in drinking water can vary from 0.1 to 1.6 mg L À1 for MC-LR or total MCs (WHO, USA states depending of age groups), 0.7 to 3 mg L À1 for CYN (USA states, depending on age groups), and 0.7 to 6 mg L À1 for ANA-a (USA states depending on age groups, Canada and New Zealand). 25,26 The method appears as a valid and sensitive approach for the screening of targeted MCs, CYN and ANA-a for risk management purposes and likely for other non-targeted cyanotoxins. For internal standardization purposes, the isotopically labelled analog of the targeted compound is generally the best choice for an efficient correction. In our previous work, native NOD-R was, however, used as the internal standard for microcystins, the compound having a structure analogous to that of cyanopeptides. 38 The reasons for this choice were the low probability of nding this cyanotoxin in targeted freshwater lake samples and the fact that commercially available isotopically labelled compounds for cyanotoxins were not readily available or costly. In this work, MC-LR( 15 N 10 ) was evaluated and compared to NOD-R. For most of the targeted cyanotoxins, MC-LR( 15 N 10 ) offered better quantitative correction, but ANA-a, HANA-a, MC-LA MC-HtyR and MC-LY showed better results with NOD-R. In the present work focusing on freshwaters, we did not detect any samples containing NOD-R that could have contributed to the measured spiked NOD-R concentration used for internal standardization.

Method application to the fast screening of surface water
The SPE-LC-HESI-HRMS method was used to analyze multiclass cyanotoxins in surface water samples from lakes monitored across Québec and Ontario (Canada). The samples were collected through the Adopt a Lake crowd-sourcing campaign, an initiative conducted to raise awareness about the protection of natural resources affected by cyanobacteria. 46 Citizens interested in knowing the state of health of nearby lakes participated by monetary donation or by helping to sample a lake. The volunteers received a sampling box including protocols and necessary equipment.
From the lakes that were sampled between May and the end of September 2018, all were reported to contain cyanotoxins with concentrations varying between 9 and 10 000 ng L À1 (Table 4). Higher variability, shown with higher RSD on replicate eld samples, was observed in some instances, more particularly for the lower-range concentrations close to the limit of detection. This variability may also be attributed to the sample nature and the sampling procedures, independently of the analytical method. Citizens were individually in charge of collecting the samples, which could be locally inhomogeneous, thus increasing the probability of error during the sampling of eld replicates. Concentrations could also be underestimated if degradation occurred between sampling and analysis in the laboratory. The presence of concentration levels well above target criteria for drinking water, conrmed by the value of citizen-initiated sampling, allowed us to identify contaminated lakes that are not systematically tracked for toxic blue-green algae. The results of three samples analysed for taxonomy at the genus level by the 16S rRNA methodology also revealed that the Microcystis genus was present in these samples.
Nine lakes were positive to ANA-a, with concentrations ranging between 23 and 79 ng L À1 . These concentrations are relatively low and would not pose an immediate threat to human health. Nevertheless, the detection of ANA-a suggests that ANA-a cyanobacteria producers are present, and higher concentrations could be observed when conditions are favorable or if a larger number of sites were sampled. Out of the 12 quantitatively targeted microcystins, 5 were recurrently detected in these lake samples (Table 4). MC-LR was the most frequently detected microcystin (12/16 samples) and it was also found at the highest concentrations (range in positive samples: 29-3476 ng L À1 ). MC-HilR was the second most oen detected microcystin (6/16 samples), while other detected MCs were less recurrent. When 2 or more MCs were detected, MC-LR was oen predominant, except for Lac McKay where MC-LA dominated the MC prole (Table 4). Two particular samples, Réservoir Choinière and Lac Memphrémagog, had summed MC levels higher than the proposed recommendations for MC-LR equivalents by the World Health Organization (WHO) for drinking water (1 ug L À1 ). For recreational water, guidance and action levels vary between 0.8 mg L À1 and 20 mg L À1 in different American states and are set at 20 mg L À1 in Canada. The targeted MCs in Lac Memphrémagog and Réservoir Choinière had totalized concentrations of 4.1 and 1.4 mg L À1 , respectively, which are lower than the Canadian guidance level for recreational water. These values are still a potential concern for more vulnerable targets such as children and domestic animals that are exposed to such lakes.
In addition to providing quantitative data for the targeted cyanotoxins, the on-line SPE-UHPLC-HRMS method can be useful in the retrospective analysis of extracted full scan MS chromatograms (Section 2.4). The acquired data can be revisited to screen previously reported cyanotoxins not currently included in monitoring programs. 17 A summary of the tentative candidates qualitatively identied in samples from the present survey is provided in Table 5. A semi-quantication was performed using MC-LR as the reference compound, except for [Gly 1 , Asp 3 , ADMAdda 5 , Dhb 7 ]MC-RR for which MC-RR was used (see also ESI Table S2 † 6 ]MC-LR was detected at a relatively high level ($7500 ng L À1 ) in the particular sample where MC-LR also displayed the highest concentration may provide additional condence, although unambiguous conrmation was not attained here and would require high-resolution MS/MS experiments on sufficiently concentrated extracts. 17,57 Two less studied cyanopeptides were also monitored in the present study. Anabaenopeptins AP-A and AP-B were detected at relatively high concentrations in Lac Roxton and Lac Memphrémagog (at 1160 and 8280-10 000 ng L À1 , respectively). AP-A was detected in three lake samples and AP-B in six lake samples; overall, 38% of the surveyed samples contained one or both congeners. AP-A is reported to show a toxicity similar to that of MCs and AP-B has the potential to induce cell lysis, thus increasing the bioavailability of cell-bound cyanotoxins. APs are not yet included in cyanotoxin monitoring programs, but some studies showed their systematic presence in toxic cyanobacterial blooms when MCs are observed. 10,13,58 About 96 AP congeners are known to date and little information is available on their occurrence and impact on public health and ecosystems. In light of these results, it is also likely that other AP congeners are regularly present in algal blooms. More in-depth characterization studies could help underline the most abundant congeners present in contaminated waters following HABs, which would be useful for improved management and health safety of drinking water sources.

Conclusions
A simple and high-throughput methodology based on on-line SPE-UHPLC-HRMS has been developed for the analysis of 17 multi-class cyanotoxins in lake water. Two high-resolution scan modes, FS and PRM, were combined into a single acquisition method to improve the quantication of difficult-to-measure compounds. HRMS detection presented the further advantage of resolving ANA-a from its isobaric interference Phe. Suitable method performance was obtained, with low relative matrix effects and suitable accuracy and precision. This is a fast (8 min), high-throughput, and robust method that includes, to our knowledge, the widest scope of cyanotoxins in a single automated workow. Method detection limits in the range of 8-53 ng L À1 were obtained, which is relevant for environmental analysis of eld samples. The applicability of the method was veried on a range of lake samples collected in the summer 2018 through a citizen-science collaborative project (Adopt a Lake campaign). Cyanotoxins were found in sixteen samples with individual concentrations varying between 9 and 10 000 ng L À1 . ANA-a was found in more than half of the samples albeit at low concentrations (compliant with guidelines), which suggests a high occurrence of this toxin and its producers. Two particular samples, Réservoir Choinière and Lac Memphrémagog, contained summed microcystin levels higher than proposed recommendations for MC-LR equivalents by the WHO for drinking water (1 mg L À1 ) but lower than the Canadian guidance level for recreational waters (20 mg L À1 ), or the threshold proposed by the USEPA for recreational use (8 mg L À1 ). A retrospective analysis of extracted full scan MS chromatograms suggested the presence of some additional MC variants, at semi quantied concentrations in the mg L À1 range in the sample from Lac Memphrémagog. In addition, about 38% of the samples contained either anabaenopeptins AP-A, AP-B, or both, with concentrations of up to 10 000 ng L À1 , higher than the reported MCs. Little information is available about the environmental fate and effects on public health of these infrequently monitored cyanopeptides. In view of the systematic cooccurrence of APs and MC congeners in some of these samples, complementary research is needed to characterize their toxicological properties (individually and from mixtures with known compounds) and their occurrence at a broader spatial scale and properly evaluate their associated risks.

Conflicts of interest
There are no conicts of interest to declare. Table 5 Qualitatively identified microcystins in samples from the present survey, including retention time (RT, min), theoretical and observed exact m/z and corresponding mass accuracy (d (ppm)), occurrence frequency in lake samples (n out of 16), and semi-quantified a concentration levels (ng L À1 )