Worldwide, approximately 884 million people lack access to clean drinking water, the majority of which reside in semi-arid poor geographic areas of sub-Saharan Africa and Asia (e.g., Afghanistan and Pakistan)¹. In Latin America, some regions, such as in Northeastern Brazil, still rely on natural or man-made shallow standing water ponds. Extremely harsh climatic conditions and drought have resulted in dry-out of these water bodies. The current drought period began 6 years ago and is considered the worst of at least 10 large droughts in the last century (Martins et al., 2015). The Boqueirão pond, which has a capacity of approximately 411 million m³, experienced approximately 97% volume decrease in March 2017 (Brazil, 2017). Boqueirão experiences high temperatures and high turbidity throughout the year, which is associated with a constant state of eutrophication through nutrient inputs. This pond supplies water to 1 million people throughout 19 cities. Despite the need to remove harmful cyanobacteria and toxins and perform clarification (coagulation and flocculation), disinfection, and pH correction as preconized by Resolution 357/05 from the Brazilian National Environment Council (CONAMA) (Brazil National Environment Council/Conselho Nacional de Meio Ambiente - CONAMA, 2005), pond water is transported by means of water trucks to nearby cities households as drinking water without any treatment. The extremely poor water governance is the consequence of a complex context.

Water governance is a conceptual framework that comprises different aspects, including (i) how (federal, state, and municipality) institutions operate, (ii) how (federal, state, and municipality) regulations affect political actions and societal concerns through formal and informal instruments, and (iii) how the above mentioned (i) and (ii) enable practical management tools to be applied (Tortajada, 2010; UNDP, 2010). A recent study modeled political, economic, social, and environmental variables that impact water sector performance in Brazil (Kayser et al., 2015). It became clear from this study that simple actions such as coordination and data sharing between ministries that deal with drinking water services, monitoring and enforcement of water quality laws, and sufficient technical capacity to improve administrative and technical management of water services at the local level could improve dramatically water governance in Brazil. Relevant technical aspects to improve water governance include the monitoring of water quality by means of (in)organic chemistry analysis, toxicological tests, and metagenomics.

The combination of high nutrient loads and high temperature promotes the formation of toxic cyanobacterial blooms in freshwater bodies (Paerl and Huisman, 2008). The occurrence of cyanobacterial blooms in aquatic ecosystems has increased in extension and frequency and is becoming a potential threat to both human and ecosystem health worldwide (Sangolkar et al., 2009; Paerl and Otten, 2013). Even developed countries may face severe water shortages (Qin et al., 2010; Tanber, 2014). In 2007, a massive Microcystis bloom in the Lake Taihu, China, affected approximately 10 million people, of which more than 2 million have the water supply cut off for at least a week (Qin et al., 2010). High nitrogen and phosphorus levels control the development of blooms (Graham et al., 2004; Rinta-Kanto et al., 2009; Orihel et al., 2012; Berry et al., 2017), and the balance between these nutrients promotes a shift from non-toxic to toxic cyanobacterial species (Gobler et al., 2016). The most frequently detected and widespread cyanotoxins in freshwater are hepatotoxins (i.e., microcystins, nodularins, and cylindrospermopsins), which could bioaccumulate (Christen et al., 2013; Li et al., 2013; Cai et al., 2015a,b; Zhao et al., 2015). Severe human poisoning by cylindrospermopsins was first recorded in Australia (Byth, 1980) and England (Turner et al., 1990). In Brazil, the first documented combined microcystin and cylindrospermopsin poisoning episode occurred in Caruaru city in 1996, after a stronger drought, causing 76 deaths at a hemodialysis clinic (Pouria et al., 1998; Carmichael et al., 2001; Azevedo et al., 2002). Previous studies have shown that Microcystis may be found in ponds of Northeastern Brazil (Bouvy et al., 2000; Huszar et al., 2000; Vasconcelos et al., 2011). Microcystin is known to be produced by Microcystis, Anabaena, Planktothrix, and Nostoc toxic species (Sivonen and Jones, 1999; Hotto et al., 2007; Dittmann et al., 2013), which harbor microcystin synthetase (mcyABCD) gene clusters, directly involved in the microcystin production, and bi-directionally transcribed central promoters (mcyA/D) (Kaebernick et al., 2002). Microcystin and other cyanotoxins are known to produce malformations and killing zebrafish (Danio rerio) (Dao et al., 2013; Pavagadhi et al., 2013) and medaka embryos (Oryzias latipes) (Adámek et al., 2011) in controlled laboratory conditions.

The aim of the present study was to analyze the water quality and the toxicity potential of three major ponds Araçagi, Boqueirão, and Saulo Maia, and two minor ponds Galante and Mazagão, located in the semi-arid region of Campina Grande (Paraíba, Brazil). We hypothesized that potentially harmful cyanobacteria and other microbes might be a significant component of the water making it improper for consumption as drinking water without treatment.

Chemical and biological parameters segregated the three different ponds (Figure 1B). Levels of total nitrogen, Mn, and Ba were higher in Araçagi, whereas Boqueirão was mainly characterized by the higher abundance of total autotrophs and cyanobacteria (Supplementary Figure S1 and Supplementary Table S2). The ponds were hypereutrophic (total phosphate > 0.1 mg l^-1 and total nitrogen > 0.45 mg l^-1 for all ponds). The highest phosphorus (0.55 mg l^-1) and total nitrogen (2.43 mg l^-1) values were found in Araçagi (Supplementary Table S2). The eutrophic condition in semi-arid regions is established by values above 0.05–0.06 mg l^-1 of total phosphorus (Thornton and Rast, 1993), and the limit of 0.03 mg l^-1 of total phosphorus is established by the CONAMA (Resolution 357/05) for Class II lentic environments (reservoirs) – (Brazilian classification of water bodies for human supply established by CONAMA) (Brazil National Environment Council/Conselho Nacional de Meio Ambiente - CONAMA, 2005). High levels of phosphorus and nitrogen promote the formation of blooms of non-diazotrophic cyanobacteria, such as Microcystis, and the concomitant production of microcystin, and other secondary metabolites, such as aeruginosin, cyanopeptolin, and protease inhibitors (Gobler et al., 2016; Harke et al., 2016). Protease inhibitors discourage zooplankton grazing, facilitating bloom proliferation (Agrawal et al., 2005; Gobler et al., 2007). The measured nutrient loads clearly demonstrate that ponds are under a severe eutrophication process, possibly conditioned by both climatic factors (e.g., drought) and local pollution. The resulting high loads of nutrients (e.g., phosphorus) may promote the formation of potentially toxic cyanobacterial blooms. However, the mechanisms underlying cyanobacterial bloom formation and the massive toxins production remain to be further investigated (Paerl et al., 2011; Harke et al., 2016; Li et al., 2016).

A total of 8.34 million reads (raw sequences) were obtained for all ponds (Supplementary Table S3). Approximately 1.9 × 10⁶ sequences were annotated. Bacteria domain contributed an average of 81% of the sequences annotated, ranging from 80.6% (Araçagi, total N = 397,361), 80.1% (Saulo Maia, total N = 223,287), to 77.1% (Boqueirão, total N = 382,168) for the major ponds; and 82.8% (N = 238,497) to 77.4% (N = 207,391) for Galante and Mazagão ponds, respectively. A total of 29 distinct bacterial phyla and 3 candidate bacterial phyla were identified within all samples analyzed. Proteobacteria was the most abundant and largest phylum in all ponds, followed by the unclassified bacteria, Actinobacteria, Bacteroidetes, and Cyanobacteria (Supplementary Figure S2). The phylum Proteobacteria accounted for an average of 34.9% for the three major ponds, ranging from 27.8% (Boqueirão) to 43.8% (Saulo Maia) (Supplementary Figure S2). Bacteroidetes accounted for an average of 12.1%, ranging from 9.3% for Araçagi to 15.9% for Boqueirão, and 10.9% for Saulo Maia. Considering the total of Betaproteobacteria class sequences, the orders Burkholderiales (an average of 20.7%) and Methylophilales (an average of 5.8%) were the most abundant, and they are recently suggested to be more important in microcystin degradation than Sphingomonadales, an Alphaproteobacteria (Mou et al., 2013).

The taxonomic assignments of the metagenomic sequences revealed that cyanobacteria contributed an average of 10.6%, ranging from 3.7% in Saulo Maia to 16.2% in Araçagi. Cyanobacteria counts ranged from 2.10 × 10⁴ to 4.46 × 10⁵ cell ml^-1. Accordingly, Boqueirão pond had the highest cyanobacteria counts and can be classified as Class III, possibly requiring advanced water treatment (CONAMA; Brazil National Environment Council/Conselho Nacional de Meio Ambiente - CONAMA, 2005). Microcystis metagenomic sequence counts were approximately 100- and 6-fold more abundant (p < 0.01) in Araçagi than in Boqueirão and Saulo Maia, respectively. While the most abundant cyanobacteria belonged to the genus Microcystis in Araçagi (an average of 57.6 ± 4.08% of the total cyanobacterial sequences; N = 37,204), in Boqueirão, the community was more diverse: Synechococcus (an average of 13.8 ± 8.8%, N = 9,456), Anabaena (an average of 3 ± 0.4%, N = 1,219), Cyanobium (an average of 2.3 ± 0.45%, N = 1,210), and Cylindrospermopsis (an average of 0.6%, N = 127). Unclassified cyanobacteria are also an abundant group in the ponds, accounting for an average of 34.1 ± 12.4% (N = 15,483) in Boqueirão, 29 ± 1.8% (N = 2,429) in Saulo Maia, and 20.1 ± 3.64% (N = 12,978) in Araçagi, considering the total of Cyanobacteria. COG functional annotation of metagenomic sequences revealed gene sequences related to toxin production in all ponds, including cyanopeptolin synthetase (mcn), microcystin synthetase (mcy), and non-ribosomal peptide synthase genes (N = 688, ranging from 280 in Araçagi to 11 in Saulo Maia; Figures 2B,C; Supplementary Table S4). Not all mcn and mcy produce toxins, although their presence in the metagenomes hints to the toxicity potential of pond waters. To evaluate the potential toxicity of pond water, we went further and investigated the presence of toxins by ELISA and tested water toxicity in a standard zebrafish embryo model under controlled laboratory conditions.

Microcystins, nodularins, and cylindrospermopsins were detected by ELISA in small concentrations in all samples (Supplementary Table S5). Araçagi had the highest toxin concentration (0.5 ± 0.2 μg l^-1 microcystins/nodularins and 0.08 μg l^-1 for cylindrospermopsin). Toxin levels were similar to those obtained in previous studies conducted in Rio Grande do Norte (Fonseca et al., 2015), and in Pernambuco (Piccin-Santos and Bittencourt-Oliveira, 2012). These levels were below the threshold for human consumption (<1 μg l^-1), proposed by the World Health Organization (Chorus and Bartram, 1999), and followed by the Brazilian Ministry of Health (Ordinance 2914/2011) (Brazil Ministry of Health/Ministério da Saúde, 2011). Given that ELISA kits used cross-react only against eight microcystin congeners/isoforms (Microcystin-LA, Microcystin-LF, Microcystin-LR, Microcystin-LW, Microcystin-LY, Microcystin-RR, and Microcystin-YR), and one for nodularin (Nodularin-R), the values obtained here could be underestimates of the real toxin concentrations in the pond waters. There is an estimative of more than 80 other microcystin congeners being largely disregarded in ready-to-use ELISA kits (Dietrich and Hoeger, 2005). In addition, the sample treatment, preparation, and storage prior to cyanotoxins analysis could also affect results, leading, for example, to low recoveries of microcystin concentrations in the order of 40–70% (Kamp et al., 2016). Nevertheless, the detection of toxins in the Campina Grande ponds hints to the risk of water consumption prior treatment and the need for reliable water quality monitoring programs in this region. The need for a reliable governance program is even more evident by the zebrafish toxicological results obtained here.

The zebrafish embryo lethality rate was significantly higher (p < 0.01) in Araçagi (62.9 ± 0.8%) and Boqueirão (62.5 ± 0.8%) than in Saulo Maia waters (8.2 ± 1.0%) (Figure 3). Zebrafish malformations were found in Boqueirão (37.5 ± 0.8%) and Araçagi (37.1 ± 0.8%). Water from Saulo Maia did not result in zebrafish malformations (Figure 3G). Whereas 100% of the embryos presented heart edema and spine deformation when reared with Araçagi water; 100% of the embryos presented mouth deformations when reared with Boqueirão water (Figure 3). Water quality monitoring programs that evaluate toxin presence and toxicity are lacking in the semi-arid region studied here. Araçagi and Boqueirão waters were the most lethal and toxic to zebrafish (Figures 3A,B and C,D respectively), whereas the controls did not show any evidence of lethality and toxicity (Figures 3E,F). The observed diverse malformations and killing in the present study are in agreement with previous studies (Adámek et al., 2011; Dao et al., 2013; Pavagadhi et al., 2013). The observed effects include a variety of toxic cellular actions typical of microcystins, e.g., DNA damage, mitochondria dysfunction, endoplasmic reticulum disturbance, and cell cycle deregulation, all contributing to apoptosis/programmed cell death of hepatocytes as well as many other cell types (Chen and Xie, 2016) (Supplementary Figure S3). Microcystins have been implicated in neurotoxicity, hepatotoxicity, and damage to reproductive organs (Lone et al., 2015; Máthé, 2016; Valério et al., 2017). These cyanotoxins are potent inhibitors of protein phosphatases (phosphatase 1 and phosphatase 2A), which are key regulators of embryonic development, leading to changes in mRNA levels of genes that induce oxidative stress (endoplasmic reticulum stress) involving reactive oxygen species (ROS) generation in zebrafish (Faltermann et al., 2016).

Safe water supply for human consumption remains a challenge task in the northeast Brazil. Our comprehensive approach integrating metagenomics, biogeochemical analysis, and toxicity tests clearly demonstrates that the untreated pond water from semi-arid regions is not a safe source of drinking water. The toxicity and teratogenicity of pond water observed in the present study hints to possible harmful effects in human health. Finally, we highlight that reliable water quality monitoring may be an important tool to improve water management and governance.

JW, FL, ML-F, LV, LL, FM, GA, RO, AM, ASOM, CER, AT, and FT designed and planned the study. JW, FL, LV, and RO carried out the field work. JW and FL performed the bioinformatics analyses. JW, FL, ML-F, and LV compiled the data. JW, FL, ML-F, AM, ASOM, CER, AT, and FT analyzed the results. JW, FL, ML-F, and FT wrote the manuscript and all authors commented on the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. FT is a member of Frontiers Editorial Board.