3.1. Different spatial responses of the chemical lake composition after rain events
The analysis of the recorded rain rate at Aydat during the lake sampling period allows us to identify two wet periods: from August 05th, 2021, until August 12th, 2021, and from September 3rd, 2021, until September 27th, 2021, with a drought period of 22 days between them (rain rate < 0.1 mm.h− 1) (Fig. 2a). The first wet period is characterized by a rainfall amount of 45 mm, while the second wet period is characterized by rain event with higher intensities, reaching 70 mm.h− 1 during a 15-min interval on 25th September at 21:30 UTC and a rainfall amount of 560 mm. On the other hand, high mean wind speed (> 6 m s− 1) occurs only during the first period, especially on August 6th, 2021, reaching 6.4 m.s− 1 in a 1-hour interval (Fig. 2a).
During the sampling season, both sites exhibit similar temporal variations in water temperature which vary following those of atmospheric temperature (Suppl. Figure 2). Thus, the water temperature increases from August 8th until the 16th, reaching 20.9°C and 21°C at MP and WP sites, respectively. It gradually decreases at the end of the drought period, on August 30th (17 & 18°C at MP and WP, respectively) and finally drops at 16.3°C and 17.1°C at MP and WP sites at the end of the lake sampling, on September 27th, 2021. We also observe a mean temperature difference of 0.8°C between the two sampling sites, with higher temperatures recorded at WP compared to MP (Fig. 2b). This temperature difference is associated with a stronger thermal stratification at WP, where the mean depth of the thermocline is 5.7 m, compared to a mean thermocline depth of 7.3 m at MP. The higher temperature at WP is likely attributed to its shallower water column, which has a depth of 9 m, compared to 15 m at MP.
Before the drought period, similar variations of the concentrations of ions are observed at both sites, with low concentrations of \(\:{\text{N}\text{H}}_{4}^{+}\) and \(\:{\text{P}\text{O}}_{4}^{3-}\) consistently remain below the detection limit (LOD<16 µg.L−1 and 0.7 µg.L− 1, respectively) at both sites, while very high concentration of \(\:{\text{C}\text{a}}^{2+}\), reaching 9.1 µg.L− 1 and 9.2 µg.L− 1 at MP and WP, respectively. During the drought, and especially after 18 days without rain, a general decrease in the concentrations of all ions is observed (on August 30th, 2021) (Fig. 2c). Interestingly, strong disparities in the chemical water composition are reported across the sites following the rain events occurring after the drought, from September 3rd, 2021, until the end of the lake sampling. Indeed, we observed significant higher concentrations of \(\:{\text{N}\text{O}}_{3}^{-}\), \(\:\:{\text{C}\text{a}}^{2+}\), \(\:{\text{N}\text{a}}^{+}\), \(\:{\text{C}\text{l}}^{-}\), and \(\:{\text{S}\text{O}}_{4}^{2-}\) (p-value=4,5.10−2; 3.10− 2;1,2.10− 2; 2,9.10− 2; and 3.10−3, respectively) at WP compared to MP from August 30th 2021, until September 27th 2021. At WP, most of the ions show a recovery to pre-drought levels, especially at 1.5 m deep. However, the concentrations of \(\:{\text{N}\text{O}}_{3}^{-}\) not recover to pre-drought level; instead, they reached 88% and 55% of the pre-drought levels at the surface and at a depth of 1.5 meters on September 6th, 2021, respectively, at WP (Fig. 2c). In contrast, at MP, we do not observe a recovery of the ions after rain events following the drought.
Contrasting results have been reported in the literature regarding variations in ion concentrations following rain events, depending on factors such as rain intensities, the trophic state of the lake, and the geomorphology (Chorus et al., 2021). Huisman et al. (2018) suggested that intense rainfall enhances nutrient runoff, which can lead to profound nutrient enrichment of downstream waters, while Morabito et al. (2018) reported that pronounced rain events actually dilute nutrients rather than enrich them. Given the different mechanisms at play in various types of water bodies, the results are sensitive to the choice of water bodies and may not support generalizations (Chorus et al., 2021). Aydat Lake, seems to experience a significant influx of ions primarily from the inflowing water of the Veyre River near the wetland area. This inflow contributes mainly to the increase in ions concentrations following rain events at WP. In contrast, we observed a lower increase in ions concentrations in the middle of the lake (MP), indicating that some of the ions from the Veyre River may have been consumed before reaching that area, as also suggested by Ishikawa et al. (2022).
3.2. Temporal disparities in phytoplankton’ response following meteorological events
The presence of Euglenophyta remained consistently low during the lake sampling period, while the biovolume of Charophyta, Chlorophyta (green algae), Bacillariophyta (diatoms), Cyanobacteria, Cryptophyta, and Ochrophyta (brown algae) exhibited variations during the lake campaign (Fig. 3). Specifically, from August 5th until August 23rd, we observed higher biovolume levels of diatoms (Stephanodiscus, Cyclotella and Fragilaria), green algae (Sphaerocystis and Closterium) and brown algae (Uroglena), while that of cyanobacteria was much lower (Fig. 3). Although this phytoplanktonic composition is not typically observed in August in this area, the presence of diatoms is not surprising as this group are known to thrive in habitats that are frequently stirred up (Padisák et al., 2006; Blottière et al., 2017; Pannard et al., 2007). These atypical conditions are likely attributed to the low water temperatures (below 19°C) and the frequent occurrence of rain events accompanied by wind. These factors may explain the limited presence of cyanobacteria, as these species typically have a low tolerance to mixing and low water temperature (Reynolds, 2006; Elliott, 2010; Padisák et al., 2009). In contrast, chlorophyceae like Sphaerocystis is known for being less sensitive to temperature variations compared to cyanobacteria (Reynolds et al., 2002), potentially explaning their prevalence in August (Fig. 3a-c). Significant shifts in phytoplanktonic composition were reported at the end of the drought period, especially on August 30th, marked by the increase of water temperature and the decrease in \(\:{\text{N}\text{O}}_{3}^{-}\) concentration (Fig. 2). Specifically, we noticed the increase of the biovolume of diazotroph cyanobacteria, such as Aphanizomenon (Fig. 3c). These observations aligns with previous studies indicating that diazotroph cyanobacteria thrive in eutrophic lakes with low nitrogen content, as they can fix atmospheric nitrogen (Wei et al., 2023; Reynolds et al., 2002). Conversely, at the end of the drought period, the biovolume of diatoms, ochrophytes and cryptophytes were low at both sampling sites (Fig. 3b, d and e).
Interestingly, the two rain events occurring following this drought period, from 03rd until 04th September and from 10th to 12th September, have shown contrasted impacts on cyanobacterial biovolume despite having similar characteristics in terms of rain intensity (less than 10 mm.h− 1): i) the first rain events following the drought, occurring from 3rd until 4th September, led to an important decrease of cyanobacterial biovolume on 6th September, especially of Aphanizomenon across all sampling points; ii) the second rain events after the drought, occurring from 10th to 12th September, had the opposite effect, resulting in an increase in the biovolume of the dominant cyanobacteria Aphanizomenon on the 13th September, reaching its maximum value in this study (Fig. 3c). We observed a contrasting pattern with the cryptomonas genus (Fig. 3e), exhibiting an initial increase following the first rain events from 03rd until 04th September, followed by a decrease in its abondance on the 13th of September after the second rain events following the drought, from 10th to 12th September. So, rain event with low intensity (< 10 mm.h− 1) is probably not the main driver explaining the dynamics of phytoplankton species in our case. Surprisingly, the significant increase in nutrients brought by the first rain event did not lead to the expected enhancement in the total biovolume of cyanobacteria. However, there was an exception with the biovolume of Dolichospermum, which increased at the lake surface (Fig. 3c). In the same way, although the strength of thermal stratification is commonly recognized as a significant factor affecting cyanobacteria response to rain events (Chorus et al., 2021), our study found it to be less influential in our specific case as the first meters of the water column were already mixed before the rain events due to the unexpected decrease of water temperature from 21–22°C to 17–18°C at the end of August. This relatively low temperature for a month of August, could have induced a cold-stress response that appears to operate with a timescale of few days corresponding to the decrease of the cyanobacterial biovolume at the beginning of September, especially for Aphanizomenon which showed a higher decrease of its biovolume (Fig. 3c). Such cold-stress have already been shown concerning microcystin biosynthesis when temperature reduce from 26°C to 19°C and where authors indicated that the processes involved operated on a different timescale (Martin et al., 2020). The timescale due to the cold stress might not exceed a few days as the cyanobacterial biovolume grown again on the 13th of September when water temperature rises again (19.5°C).
Hence, it appears that the biovolume of Aphanizomenon is more influenced by water temperature rather than the volume of rainfall. Nevertheless, it's crucial to take into account the frequency and intensity of rain events: the rise in frequency towards the end of September led to a decrease in cyanobacteria at both sampling sites, even though the water temperature remained stable at the wetland point and slightly decreased at the middle point. These rain events were characterized by a high intensity up to 10 mm.h− 1, which could impact the global dynamics of phytoplankton by limiting the biovolume of cyanobacteria and increasing those of diatoms (Fragilaria and Navicula) (Fig. 3b-c), corresponding to the start of the increase in turbulence and autumnal conditions (Reynolds et al., 2002). Finally, on September 23rd, when the water temperature dropped below 19°C, the biovolume of all phyla was lower compared to the rest of the lake sampling period.
3.3. Intracellular metabolite profiles
In our study, we detected 4,446 untargeted metabolites from phytoplankton biovolume and were able to annotate 260 of them thank to GNPS molecular network approach. The number of annotated metabolites remains very challenging as it depends on the design of the experiment, the biomass collected, the analytical techniques used, and the presence of reference molecules within public chemical databases. Therefore, high differences in the quantification and annotation of metabolites can be found across the different studies. For example, from previous studies using LC-MS-based untargeted metabolomics approaches carried out under lake systems, Sadler et al. (2014) identified approximately 100 secondary metabolites, some of them being known as chemotrypsine inhibitor; while McNabney et al. (2023) focused on primary metabolites and were able to annotate 33 metabolites. The chemical libraries we used in the present case are public and comprise both generalist and cyanobacterial-specific databases, such as HMDB, GNPS or CyanoMetDB (Jones et al., 2021), respectively. Therefore, we attempt to find here a high number of primary metabolites together with secondary metabolites derived from cyanobacteria and other phytoplankton taxa. A serious challenge for eco-metabolomic studies is to determine and quantify the maximum number of metabolites as possible to satisfy the need to disentangle the biologically relevant components and response shifts under environmental changes (Sardans et al., 2011). However, generally only 2–5% of the features detected in untargeted mass spectrometry analysis were matched with known metabolites in public libraries so far (da Silva et al., 2015) and molecular network approach offer great opportunity to deeper explore the chemical diversity of natural microbial ecosystems. Indeed, the exact number of metabolites in natural samples remains challenging, even in the case of microorganisms with relatively simple and well-understood metabolism (Sardans et al., 2011).
We carry out multivariate analyses with either 4,446 untargeted metabolites and 260 annotated metabolites using PCA (Suppl. Figure 4) and confirm the similar patterns according to the dates observed with both datasets. Since regression models become infeasible when the number of metabolites exceeds the sample size (Antonelli et al., 2019), we carry out the subsequent statistical analyses using only the 260 annotated compounds to better explore the biological functions linked with environmental stressors. However, we keep in mind that our analysis focuses on annotated metabolites and thus does not reflect the entire metabolism pathways of the phytoplanktonic community. It is important to note that further work is required to annotate other unidentified metabolites, as various molecular clusters remain fully unannotated thank to automatic search with public database combining the already-described natural product part.
Our statistical analysis shows that our sampling and extraction methods allow to characterize the eco-metabolomes during the lake campaign. Indeed, the intracellular untargeted metabolomics data are strongly correlated with the biovolume of phytoplankton (R = 0.82) and with abiotic variables such as ion concentrations (R = 0.71). Additionally, the "environmental variables" group, which includes atmospheric parameters (wind speed and rain rate) as well as lake variables (water temperature, pH, and euphotic zone depth), also exhibits strong correlations with the intracellular metabolomic data (R = 0.7) (Figs. 4 & 5).
Interestingly, two distinct metabolome fingerprints are identified through the RDA (Fig. 4). The first metabolome fingerprint is explained by the biovolumes of species belonging to the phyla Bacillariophyta (Fig. 3b) and Ochrophyta (Fig. 3d) during dates characterized by high wind speed and pH, from August 2nd to August 23rd, 2021. Instead, the second metabolome fingerprint is explained by the biovolume of Cyanobacteria, from August 30th to September 27th, 2021 (Fig. 3c), characterized by high rainfall amount and deeper depth of the euphotic zone (Fig. 4). These phyla are indeed associated with annotated metabolites, but it is important to note that other phyla may also produce additional unidentified metabolites during these dates. Hence, the correlations derived from subsequent statistical models, between phytoplankton biovolumes and the relative abundance of untargeted metabolites, may do not capture the entire chemical diversity related to phytoplanktonic composition. These correlations particularly highlight the dynamics of specific species that are positively or negatively correlated with the relative abundance of annotated metabolites. Consequently, we will concentrate on the period explained by these phyla and refer to the first period as the "diatoms' co-occurrence period" and the second period as the "cyanobacteria co-occurrence period". A previously study by McNabney et al, (2023) has already identified a distinct metabolome fingerprint based on the phytoplankton community and sampling sites, which is characterized by specific abiotic variables such as nitrate and conductivity. In our current study, we provide clear evidence of differences in the metabolome based on changes in the phytoplankton community following summer meteorological events. These changes are influenced by abiotic variations, such as fluctuations in ions concentrations, which exhibit a decay over several days following drought. Notably, the concentrations of ions begin to decrease after 11 days without rain, probably indicating the time taken for consumption of nutrients by phytoplankton. Conversely, an increase in ions concentrations occurs immediately after rain events. It is worth noting that our monitoring was conducted 3 days after the rain events, so it is possible that this increase occurred even earlier.
To enhance our understanding of metabolome changes and unravel the biological functions and response shifts under environmental changes with a high temporal resolution, we employed multiblock sPLS models during both diatoms and cyanobacterial dominance periods (Fig. 5). Interestingly, the identification of metabolites during these two periods exhibits striking differences, as evidenced by the heatmap resulting from the models (Fig. 5e-f), with a high lipid content observed during the diatoms co-occurrence period (Fig. 5e) and a high cyanopeptides content observed during the cyanobacterial co-occurrence period (Fig. 5f). Our findings reveal strong positive correlations among the abundance of certain untargeted metabolites, abiotic variables, and phytoplankton biovolumes, as demonstrated by correlation circle plots (Fig. 5a-b & e-f). These plots depict the rapid changes in the metabolome during the sampling periods characterized by the co-occurrence of diatoms (Fig. 5a) and cyanobacteria (Fig. 5b). Consequently, we address these dynamics in two subsequent subsections, 3.3.1 and 3.3.2. in relation to meteorological events.
3.3.1. Physiological and molecular strategies of phytoplankton following the drought: insight on the lipid accumulation to cope with the decreased concentrations of ions.
The analysis from the multiblock sPLS-model carried out during the “diatoms’ co-occurrence period” highlight different metabolomes according to the phytoplankton composition. Indeed, the biovolume of Chroococcus is strongly correlated at the onset of the lake campaign with amino acids, such as guanosine, tyrosine, and inosine (Fig. 5e). Instead, the biovolumes of Closterium, Cyclotella and Scenedesmus are negatively correlated with amino acids but positively correlated during the drought with other kind of metabolites, such as sterols and glycerolipids, such as betaine lipids (lyso-diacylglyceryltrimethylhomoserine (LDGTS)) (Fig. 5e). The strong correlation between these lipids and phytoplankton may potentially suggest a physiological response due to the low levels of \(\:{\text{P}\text{O}}_{4}^{3-}\) and \(\:{\text{N}\text{O}}_{3}^{-}\). Indeed, it has been suggested that under nutrient limitations, such as phosphorus and nitrogen limitation, lipid can be accumulated by phytoplankton cells, which disrupt anabolic processes (Morales et al., 2021; Popko et al., 2016; Murakami et al., 2018). Furthermore, betaine lipids and sterol intracellular content increase from August 5th until 23rd, while the abundance of some phospholipids, such as glycerol-phosphatidylethanolamine (PE), and phosphatidylcholine (PC), decrease (Suppl. Figure 6a-c). Similar patterns have been previously reported by Giroud et al., (1988) suggested that betaine lipids are produced to complement the reduction in phospholipids occurring during phosphorus limitation. These authors suggested that accumulation of betaine lipids may aim to re-allocating phosphate use from membrane lipid synthesis to other metabolic pathways (Giroud et al., 1988).
Interestingly, we also report that abundance of amino acids is high during the onset the period, from August 5th until 9th, but then decrease during the drought. The negative correlation between the amino acid contents and the biovolume of Cyanothece, Closterium, and Cyclotella, may suggested a downregulation of the production of amino acids, such as observed from culture of Phaeodactylum tricornutum (diatom) during nitrogen-starvation (Popko et al., 2016). In addition, the study of Feng et al. (2015) had also reported a downregulation of amino acid production under nutrient limitation, with higher production of glycolipids compared to phospholipids, as well as an upregulation of protein degradation, lipid accumulation, and photorespiration. The authors also reported a downregulation of energy metabolism, photosynthesis, amino acid, and nucleic acid metabolism (Feng et al., 2015). In addition, the study of Gargallo-Garriga et al., (2020) also observed an upregulation of the pathway for lipid metabolism in the dry season with regard with the wet season by comparing leaf metabolomic profiles of 54 species in two rainforests of French Guiana. Therefore, our results suggest that during the drought, when the concentrations of available ions decrease, Cyanothece, Closterium, and Cyclotella may develop acclimatized strategies to cope with the decrease in ions, as shown by the increase of betaine lipids and decrease of glycerophospholipids.
Additionally, we observed the synthesis of antioxidant compounds, like polyphenols, which are recognized for their ability to scavenge reactive oxygen species (ROS) as a protective mechanism (Morales et al., 2021). Therefore, these molecules might be produced in reaction to oxidative stress induced by nitrogen limitation during drought, as observed in previous studies on microalgae cultures experiencing nitrogen starvation (Chokshi et al., 2017).
Since our samples are obtained from a natural phytoplankton community, we cannot definitively conclude whether these lipids are specifically produced by all species or only by the dominant ones. Betaine lipids have been widely reported in organisms such as diatoms, microalgae, and cyanobacteria (Sato, 1992; Popko et al., 2016; Künzler and Eichenberger, 1997). In addition, previous study from Heal et al., (2021) demonstrated that microalgae produce diverse sets of metabolites, with 17% of the untargeted metabolites commonly found in marine phytoplankton and identified 44 metabolites that were observed in over 80% of the phytoplankton species, including amino acids, primary metabolites and nucleic acids. Based on all these findings, we can suppose that the production of lipids might be a common response of the phytoplankton community to environmental changes, indicating strategies of acclimatation employed by phytoplankton to cope with decreasing ion concentrations and potentially counteract the increasing levels of reactive oxygen species due to nitrogen limitation.
3.3.2. Temporal shifts in metabolites production according to the sites
The sPLS model carry out during the “cyanobacterial co-occurrence” period reveal two distinct metabolome fingerprints according to the sites (Fig. 5d). Most of the metabolites annotated correspond to cyanopeptides and are positively correlated with the biovolumes of Dolichospermum, thriving at both sites following rain events. In contrast, Aphanizomenon, the dominant genus among cyanobacteria, exhibits a low correlation between its biovolume and most of the metabolites present during the rainy period. One possible explanation is that metabolites associated with Aphanizomenon are not identified in the databases we used as only a small part of metabolites are currently known and annotated. Moreover, Aphanizomenon exhibited contrasting dynamics in response to rain events (section 3.2), unlike Dolichospermum (Fig. 3c & Suppl. Figure 5). Therefore, correlations between the relative abundance of untargeted metabolites and the abundance of Dolichospermum are more robust compared to those observed with Aphanizomenon. These correlations need to surpass the correlation cutoff threshold (set at 0.4 for correlation circle plots and at 0.7 for the heatmaps) to be reported in the present statistical results (Fig. 5 & Suppl. Figure 5). On the contrary, these annotated metabolites are negatively correlated with the biovolumes of Closterium and Sphaerocystis (Fig. 5f). We do not find positive correlations between their biovolumes and any annotated metabolites, which is not surprising considering that we primarily utilized available generalist and specialist databases such as, respectively, HMDB and CyanoMetDB, which predominantly contain metabolites produced by cyanobacteria (Jones et al., 2021). However, this does not imply the absence of metabolites produced from Closterium and Sphaerocystis as, during the drought, the biovolume of Closterium have been observed to be also positively correlated with certain lipids (Fig. 5e). In addition, previous studies have reported that cyanopeptides can be rather produced during the growth phase of cyanobacteria. Therefore, this could explain the high proportion of identified cyanopeptides during this period (Chorus et al., 2021), especially on the 13th of September, when the cyanobacteria biovolume rich their maximum value (Fig. 3c).
Interestingly, we observed significant variations in the intra-cellular content in metabolites between the sites during the wet period (p-value = 0.014) while phytoplankton communities were similar (Fig. 6 & Suppl. Figure 5). To our knowledge, the only study that has demonstrated shifts in metabolome fingerprints in aquatic ecosystems is the recent work of McNabney et al. (2023). Their study showed contrasting metabolomes associated with the phytoplankton community, as well as the concentration of nitrate and conductivity across two sites separated by 200 km. In our study, the sites are only 200 m apart (Fig. 1b), highlighting a divergence in their respective metabolomic niches. This concept have been recently proposed by the study of Gargallo-Garriga et al. (2020) when demonstrating that trees at different location exhibit different functional niches.
Therefore, to better understand the differences across the sampling sites, we carry out a supervised multiblock sPLS-DA model (Fig. 6 & Suppl. Figure 4), since the metabolites appear significantly different across the sites (p-value = 0.04) (Suppl. Figure 4). This analysis shows, thank to hierarchical clustering, that endo-metabolome differs strongly across the sites, but only after August 30th (Fig. 6). Interestingly, most cyanopeptides have significant higher relative abundance at WP rather than at MP, such as anabaenopeptin, microginins, shinorine, and micropeptin B, as well as cyanotoxins, such as microcystin LR (Fig. 6). Most of these secondary metabolites are referred as bioactive compounds (Demay et al., 2019).
In our study, since the relative abundances of cyanopeptides are strongly correlated with the concentrations of inorganic ions (Fig. 6), we suggest that the synthesis of these peptides is primarily favored by nutrients availability and may be related to biotic interaction function (e.g. allelopathy), very likely providing competitive advantage to the producing cyanobacteria. In addition, these components may also act as potential as nitrogen reservoir, as previously suggested by several studies for microcystin-LR, aeruginosins, cyanopeptolins, and shinorine (Agrawal et al., 2005; Davis et al., 2010; Sadler and von Elert, 2014; Harke and Gobler, 2013; Peinado et al., 2004; Horst et al., 2014). Indeed, during nutrient limited conditions, it was hypothesized that the biosynthesis of arginine should be used by dinoflagellates and cyanobacteria as a nitrogen reservoir molecule.
After rain events, according to the geomorphology of the lake, spatial changes of abiotic conditions can be observed. Indeed, we recorded a higher water temperature and a stronger lake thermal stratification at the wetland area (WP) compared to the middle of the lake (MP). In this study, we report that during the wet period a higher abundance of polyphenol compounds and glycerophospholipids are detected at MP, while most cyanopeptides have been identified at WP (Fig. 6 & Suppl. Figure 6). This result seems also to indicate a distinctive molecular response of this cyanobacteria between the two sites, distant of only 200 meters. Therefore, the untargeted metabolomic approach is well-suited for highlighting ecological acclimatation by providing a measurable method to identify the corresponding responses of the phytoplankton community to abiotic variations triggered by meteorological events. Consequently, eco-metabolomic approaches can be utilized to study the metabolomic niches, as previously proposed by Gargallo-Garriga et al. (2020) in the context of trees located at different locations. To the best of our knowledge, this is the first time that we demonstrate the acclimatized response of phytoplankton in lake ecosystems following meteorological events, revealing a strong heterogeneity based on sampling dates and sites.