The Lake Bagno dell'Acqua is located within the caldera depression of the Caldera Cinque Denti in the northeastern part of the island of Pantelleria. It has a subcircular shape, approximately 450 m long and 350 m wide, with a maximum depth of 12.5 m. The inflow of the lake is determined by direct recharge from precipitation, small tributaries on the western portion of the watershed, as well as inflow from hydrothermal fluid vents rich in CO₂ primarily located in the southwestern part of the basin (D'Alessandro et al., 1994; Parello et al., 2000; Aiuppa et al., 2007). These hydrothermal fluids contribute to the high alkalinity, salinity, and unique geochemical composition of the lake, fostering microbial processes such as carbonate precipitation and biomineralization (Duchi et al., 1994; Mazzoni et al., 2024). Since the lake has no outlets, its hydrological balance is primarily controlled by meteoric precipitation and evaporation (Aiuppa et al., 2007).

The lake water chemistry is characterized by the following general pattern: Na⁺ > K⁺ > Mg²⁺ > Ca²⁺ for cations and Cl⁻ > HCO3- > SO42- > CO32- for anions (Cangemi et al., 2010; Ingrassia et al., 2024). Repeated geochemical surveys have documented the alkalinity of the lake over several decades, with the most recent measurements reporting values of up to 57 meq/L (Jácome Paz et al., 2016).

The geological context of Pantelleria Island further influences the lake's chemistry. The island is a Pleistocene stratovolcano, located within a continental rift system between North Africa and Sicily, shaped by tectonic and volcanic processes (Boccaletti et al., 1987). The alkaline nature of the lake results from interactions between hydrothermal fluids and sodium-rich peralkaline rhyolite (pantellerite) (Pecoraino et al., 2015). These processes contribute to its elevated chloride and sodium concentrations, high electrical conductivity (up to 40 mS/cm), and pH values reaching 9 (Pecoraino et al., 2015). The limited water volume makes its chemical composition highly sensitive to hydrological variations, including seasonal precipitation and evaporation cycles. The lake is primarily fed by subsurface springs, which release CO₂-rich hydrothermal fluids at temperatures between 34 and 58 °C (Cangemi et al., 2018). Thermal springs and bubbling gases are in the southwestern area of the lake (Figures 1A, B).

Water temperature (T), pH, and electrical conductivity (EC) were measured using the Hach HQ Series multiparameter probe (Hach Company, Loveland, CO, USA) at regular intervals of 2 m in depth along a vertical profile in the deepest central part of the lake. Moreover, measurements were carried out in correspondence with the inflow of a hot spring along the shore in the southern part of the lake (at 0, 3, 6 and 12 m from the shoreline). All physicochemical measurements were performed in situ simultaneously with water sampling during each sampling.

Water sampling at the center of the lake (36°8′58“N, 11°9′13”E) was conducted in spring (25/05/2022) and in September, hereafter named “late summer” (27/09/2022) using a Ruttner bottle along the vertical profile at 0 (CL0m), −2, −3, −5, −7, −9, and −11 m, where negative values indicate depth below the water surface.

Aliquots for nutrient analysis were collected in HDPE bottles and stored at 4 °C until further analysis. For dissolved organic carbon (DOC) analysis, samples were filtered on-site with pre-combusted GF/F Whatman fiberglass filters (Frisenette ApS, Denmark) and acidified with an HCl solution to pH 2 to remove inorganic dissolved carbon and inhibit microbiological carbon degradation. For microbial diversity analysis, lake water (500 mL) was filtered through polycarbonate filters with 0.2 μm pores (GTTP type; diameter 47 mm; Millipore, Eschborn, Germany) and the filters were stored at −20 °C until processing. Unfiltered and filtered aliquots (GFF Whatman) (2 mL) were fixed with a formaldehyde solution (Sigma Aldrich; final concentration 1%) and stored at 4 °C for cytometric analysis.

During the spring sampling, sediment samples and microbial mat (hereafter named ‘'SED1” and “green mat”, respectively) at the bottom were collected by an end-cut syringe operated by divers. Sediment and mat were either directly stored at −20 °C (for the DNA extraction) or fixed with ethanol (for CARD-FISH analysis) (Sigma Aldrich; final concentration 50%) and stored at −20 °C until further processing.

During the late-summer sampling period in addition to the sediment sample collected at the central lake site (SED2), 3 surface water samples were collected at the distances of 0 (HS0m), 3 (HS3m), and 12 m (HS12m) from the shore in correspondence with the inflow of the hot spring of the lake, and respective sediment samples (samples S-HS) were collected and treated as described before.

Dissolved organic carbon (DOC) and Total Nitrogen (TN) were analyzed with a N/C 3,100 analyzer (Analytik Jena, Germany) (detection limit is 0.1 mg/L). The content of chromophoric and fluoromorphoric organic moieties was estimated with a UV-visible spectrophotometer (UV1,700 Pharma Spec, Shimadzu) and spectrofluorometer (RF-5,301, Shimadzu, Japan) equipped with a xenon lamp and a light-source compensation system S/R mode), respectively.

Water samples were prefiltered with 0.22 μm pore-size PTFE membrane filters (Frisenette ApS, Denmark) to remove impurities, and all optical measurements were performed at room temperature.

Absorbance spectra were recorded in the 200–800 nm range with a 1 cm quartz cuvette. Ultrapure MQ water was used as blank. Two chromophoric descriptors are estimated: the specific aromaticity at 254 nm (SUVA₂₅₄; Weishaar et al., 2003); and the spectral slope at the interval 274–295 nm (S275–295; Helms et al., 2008). SUVA₂₅₄ describes the aromatic content of dissolved organic matter (DOM); S275–295 is typically inversely correlated with the molecular weight of the DOM.

Fluorescence spectra were obtained with a 1 cm quartz cuvette. A three-dimensional excitation-emission matrix (EEM) was obtained for each sample. An EEM consists of 21 synchronous scans with 1 nm increments both in emission and in excitation. The excitation and emission wavelengths ranged from 250 to 410 nm and from 310 to 530, respectively. A 5 nm bandwidth was used for both excitation and emission.

Goletz et al. (2011) protocol was executed to perform the correction and normalization of each EEM. A Raman curve of ultrapure MQ water {λex350 nm, λem371–428nm} (Lawaetz and Stedmon, 2009) was used to normalize. The inner filter effect was removed with absorbance measurements (Larsson et al., 2007).

Three DOM qualitative proxies were estimated with EEMs: the fluorescence index (FI), the freshness index, and the humification index (HIX), following previously published approaches (McKnight et al., 2001; Ohno, 2002; Huguet et al., 2009). Index calculations were performed according to the procedures described by Guarch-Ribot and Butturini (2016). Data are reported as mean ± standard deviation.

Nutrients, namely ammonia, nitrites, nitrates, and phosphate, were spectrophotometrically determined using a UV/VIS spectrophotometer (cell 5 cm, Lambda 25, PerkinElmer, Norwalk, CT, USA). Calibration in fresh and salty water was performed using NO2-, NO3- and PO4-3 stock solution 1 g/L (Carlo Erba, Italy) and preparing 1 g/L stock solution for ammonia by weighing ammonium chloride (NH₄Cl) salt (Sigma Aldrich).

In detail, ammonia nitrogen (N-NH₃) was determined using the Nessler reaction. To 10 mL of sample, 2 drops of Seignette solution were added, followed by 0.4 mL of Nessler reagent. In the case of seawater, a pretreatment step was included to remove interferences from Ca, Na and Mg. Hence, 20 mL of sample were mixed with 200 μL zinc sulfate solution (0.1 g/L ZnSO₄ · 7H₂O, Sigma Aldrich) and 150 uL NaOH 6M, prepared by NaOH pellets (Sigma, Aldrich). At higher pH interferents will precipitate and will be removed by centrifugation. 10 mL of surnantant solution will be then treated as above indicated for not saline water. Finally, in both cases (saline and not saline) the colored complex will form after 20 min, and absorption read at 420 nm (APHA, 1998).

Nitrates (N-NO₃) were determined as nitrites by previously reducing them by passing the solution through a metallic cadmium (Cd) reducing glass column (35 g metallic Cd washed with acetone, 2M HCl and 2% CuSO₄ solution). The column was finally conditioned with 0.625% NH₄Cl solution. Then, 100 ml sample mixed with 2 mL 0.625% NH4Cl were passed through the column, first 40 mL discarded and the following 50 mL collected (APHA, 1998). To 10 mL of reduced solution were added of 2 mL of 1% sulfanilamide solution (SA) and 2 mL of 0.1% naphthylethylenediamine solution (NEDA). After a 15 min nitrate read at 543 nm (APHA, 1998). Orthophosphate (PO4-3) was measured by taking 10 mL sample and adding 300 μL of mixed reagent (45 mL ammonium molybdate solution + 5 mL potassium antimonyl tartrate and 200 mL sulphuric acid 4.5M) then 300 μL ascorbic acid solution. After 20 min absorbance was read at 882 nm (APHA, 1998).

The abundance of total prokaryotic communities was determined by the Flow Cytometer A50-micro (Apogee Flow System, Hertfordshire, UK), equipped with a 20-mW solid-state blue laser (488 nm). The light scattering signals, including forward scatter (FSC) and side scatter (SSC), together with red fluorescence (> 610 nm), orange fluorescence (590/35 nm), and blue fluorescence (430–470 nm) were acquired and considered for the direct identification and quantification of distinct microbial groups by following harmonized protocols (Gasol and Morán, 2015). Total prokaryotic communities were quantified by following the staining procedure with SYBR Green I (1:10,000 dilution; Invitrogen™, Thermo Fisher Scientific; Carlsbad, CA, USA; code S7563) Based on the intensity of green fluorescence, prokaryotic cells were further discriminated in high nucleic acid (HNA) and low nucleic acid (LNA). Thresholding was set on the green channel, and the gating strategy was manually adjusted to exclude most of the unspecific signals according to negative unstained controls, consisting of unfiltered samples processed identically but without the addition of SYBR Green I. Thresholding was set on the red channel to exclude most of the unspecific signals according to 0.22-μm filtered control water samples. The gating strategy was manually adjusted on the density plots of SSC vs. Red and of Orange vs. Red channels. The volumetric absolute counting was carried out in density plots of SSC vs. blue channel. Data handling and visualization were performed by the Apogee Histogram Software (v89.0) (Amalfitano et al., 2018).

For molecular analyses, water and sediment samples were processed separately. For sediments, 1 g of material was processed per extraction to increase DNA yield. For water samples, 500 mL were filtered at each depth along the central-lake vertical profile (0, −3, −7, −11 m) in both sampling campaigns, and 1,500, 750, and 500 mL were filtered along the hot-spring horizontal transect (0, 3, and 12 m from the source, respectively). Filters were then used directly for DNA extraction. DNA was extracted using the PowerSoil Isolation Kit (MoBio, Carlsbad, CA) according to the manufacturer's instructions. DNA concentration and purity were assessed using a NanoDrop spectrophotometer (Thermo Fisher Scientific, USA), and DNA integrity was verified prior to amplification. A total of 17 samples were sequenced for molecular analyses. Amplicon libraries targeting the archaeal and bacterial 16S rRNA gene (variable regions V4–V8) were prepared using the primer set abeV48A, following a custom amplification protocol. Up to 25 ng of extracted DNA was used as a template for PCR amplification, and each PCR reaction (50 μL) contained 0.5 mM dNTP mix, 0.01 units of Platinum SuperFi DNA Polymerase (Thermo Fisher Scientific, USA), and 500 nM of each forward and reverse primer in the supplied SuperFI Buffer. PCR was done with the following program: Initial denaturation at 98 °C for 3 min, 25 cycles of amplification (98 °C for 30 s, 62 °C for 20 s, 72 °C for 2 min), and a final elongation at 72 °C for 5 min. The forward and reverse primers used included custom 24 nt barcode sequences followed by the sequences targeting abeV48A: [515FB] GTGYCAGCMGCCGCGGTAA and [1391R] GACGGGCGGTGWGTRCA (Apprill et al., 2015; Parada et al., 2016). The resulting amplicon libraries were purified using the standard protocol for CleanNGS SPRI beads (CleanNA, NL) with a bead-to-sample ratio of 3:5. DNA was eluted in 25 μL of nuclease free water (Qiagen, Germany). Sequencing libraries were prepared from the purified amplicon libraries using the SQKLSK114 kit (Oxford Nanopore Technologies, UK) according to manufacturer protocol with the following modifications: 500 ng total DNA was used as input, and CleanNGS SPRI (CleanNA, Waddinxveen, The Netherlands) beads for library cleanup steps. DNA concentration was measured using Qubit dsDNA HS Assay kit (Thermo Fisher Scientific, USA). Fragment size and purity of a subset of amplicon libraries were assessed using an Agilent TapeStation 2,200 automated electrophoresis system with D 1,000 and High Sensitivity D 1,000 ScreenTape (Agilent Technologies, USA). The resulting sequencing library was loaded onto a MinION R10.4.1 flowcell and sequenced using the MinKNOW 22.12.7 software (Oxford Nanopore Technologies, UK).

Reads were basecalled and demultiplexed with MinKNOW guppy g6.4.2 (Oxford Nanopore Technologies, Oxford, UK) using the super accurate basecalling algorithm (config r10.4.1_400bps_sup.cfg) and custom barcodes.

The sequencing reads in the demultiplexed and basecalled fastq files were filtered for length (320–2,000 bp) and quality (phred score > 15) using a local implementation of filtlong v0.2.1 with the settings –min_length 320 –max_length 2,000 –min_mean_q 97. The SILVA 16S/18S rRNA database SSURef NR 99 version 138.1, comprising full-length sequences and formatted using the RESCRIPt pipeline, was downloaded via QIIME 2 on 29 September 2022 (Yilmaz et al., 2014; Robeson et al., 2021). Potential generic place holders and dead-end taxonomic entries were cleared from the taxonomy flat file, i.e., entries containing uncultured, metagenome or unassigned, were replaced with a blank entry. The filtered reads were mapped to the SILVA database SSURef NR99 138.1, which corresponds to a 99% sequence similarity reference, with minimap2 v2.24r1122 (Li, 2018) using the –ax map-ont command and downstream processing using samtools v1.14 (Danecek et al., 2021). Mapping results were filtered such that query sequence length relative to alignment length deviated < 5 %. Noteworthy, low abundant Operational Taxonomic Units (OTUs) making up < 0.01 % of the total mapped reads within each sample were disregarded as a data denoising step. Rarefaction curves were generated to assess sequencing depth and sample coverage, and are reported in supplementary materials (Supplementary Figure 1). Bioinformatic and statistical analyses were conducted in R (v4.2.3) using the ampvis2 (2.7.27) (Albertsen et al., 2015) and iNEXT (2.0.20) (Hsieh et al., 2016) packages for amplicon data analysis and diversity calculations, respectively. The ShortRead (1.54.0) package (Morgan et al., 2009) was employed for quality control and trimming. Lastly, the Seqinr package (4.2.16) was used for the visualization of biological sequences (Charif and Lobry, 2007; Chao et al., 2014). The full OTU abundance dataset generated in this study is available as Supplementary Data 1. Putative ecological functions were inferred using FAPROTAX v.1.2.6 (Louca et al., 2016). The OTU table was reformatted in R v.4.3.2 to obtain a taxonomy-based input, and functional assignments were generated with the FAPROTAX database. The resulting functional matrix was imported in R, where missing values were set to zero, rows with no variance were removed, and data were row-normalized (z-score scaling) prior to visualization as clustered heatmaps with the package pheatmap. The sequencing dataset is available through the Sequence Read Archive (SRA) under accession PRJNA1039605.

In order to visualize bacteria within the green mat sample, Catalyzed Reported Deposition-Fluorescence in situ Hybridization (CARD-FISH) was applied as described previously (Fazi et al., 2007; Lupini et al., 2011), using a specific rRNA-target Horseradish peroxidase labeled oligonucleotidic probe (PLA46 targeting Planctomycetota, Biomers, Ulm, Germany) (Pizzetti et al., 2011). Cells were then stained with DAPI solution. The stained green mat sample was then observed under a confocal laser scanning microscope (CSLM; ZEISS LSM 900, Carl Zeiss Microscopy GmbH, Jena, Germany) at a magnification of 63 x. Both DAPI-stained cells were excited by 405 nm light and emitted at 430 to 470 nm (blue color). The hybridized bacterial cells were excited with the 488 nm line of an Ar laser (excitation) and observed in the green channel from 500 to 530 nm (emission).

Physicochemical parameters were processed and summarized using Microsoft Excel 2016 (Microsoft Corporation, Redmond, WA, USA). The OTU's table was processed in R Studio, and the taxonomic composition was visualized as relative abundances of the most prevalent bacterial and archaeal phyla. Ordination analysis was performed using Principal Coordinates Analysis (PCoA) based on Bray–Curtis dissimilarities (cmdscale, vegan) at the family level. PCoA plots were generated with ggplot2, where the distance between points reflects community dissimilarity, and convex hulls were drawn to highlight sample groupings by habitat and campaign.

Differences in microbial community composition between hot spring water and central lake water were tested using PERMANOVA (adonis2) on Bray–Curtis dissimilarities computed from median-scaled relative OTU abundances.

OTU-level community overlap among sites was visualized using a Venn diagram generated with the R package VennDiagram (Chen, 2022). Unique and shared OTUs were calculated from the non-rarefied OTU table, as sequencing depth was comparable across samples (approximately 80,000 reads per sample).

Observed richness was calculated using the specnumber () function directly from the OTU count table, while Shannon and Simpson diversity indices were computed using the diversity () function implemented in the vegan package (Oksanen et al., 2022). Pielou's evenness was calculated as the ratio between Shannon diversity and the natural logarithm of observed richness. Samples comparisons were visualized with ggplot2 as scatter plots.

Lake waters showed an average temperature of 24.2 °C during the spring and of 25.0 °C at the end of the summer with no change along the water depth (Table 1). The direct comparison during the late summer sampling period showed differences between the water column and the hydrothermal spring. In particular, pH and EC remained consistent along the water column, with pH values around 9 and EC around 40 mS/cm. At the hot spring site, the temperature of 52.9 °C decreased with the distance from the source. The pH was more acidic, with values of 6 at the spring, reaching a value of 7.6 at 12 m of distance from the shore. EC values were lower compared to the water column, averaging 15 mS/cm.

DOC in the water column averaged 8.1 ± 0.43 mg/L without any clear vertical pattern (Table 1). Absorbance spectroscopy revealed little availability of chromophores (SUVA254 averaged 1.56 ± 0.13 L/mg·cm) with the presence of more aromatic and larger DOM molecules (described with the S (275–295) index) at the upper 3 meters. Simultaneously, the fluorescence spectroscopy showed that the fluorescent DOM pool had a low level of humification (HIX < 0.63) with a remarkable signature of tryptophan-like fluorophore at all depths with a maximum at 3–5 m depth, together with the presence of fresh (1.45 < Freshness Index < 2.37) but of moderate microbiological origin (1.3 < FI < 1.36) fluorescent DOM.

DOC at hot spring was much lower (0.87 mg/L) than in the lake and almost free of chromophore substances (SUVA254 = 0.13 L/mg·cm). Moreover, fluorescence spectroscopy indicated that the active hot spring DOM was more humified (HIX = 0.9), older (Freshness Index = 0.92) and with higher contribution of microbiological origin (FI = 1.65) than that reported in the lake water column. As hot water flowed toward the lake, DOC increased up to 4.4 mg/L together with a notable leaching of chromophoric DOM moieties (SUVA254 increases up to 7.2 L/mg·cm). Concurrently, the FI and freshness Index descriptors tended to decrease, and HIX showed a little increase.

Inorganic solutes containing nitrogen and phosphorus showed elevated concentrations in the upper water column (0–5 m), peaking at 3 m depth (Table 1). Concerning nitrogen, ammonia was the dominant species (average 190 μg/L), while nitrites and nitrates were below detection limits in most of samples. In hydrothermal surface waters, nutrient concentrations showed a clear spatial gradient along the horizontal transect (Table 1). At the hot spring, nitrate and ammonium concentrations were highest (356 μg/L and 340 μg/L, respectively) and progressively decreased with distance from the source, reaching 87.9 μg/L of NO3- and 222 μg/L of NH₃ at 12 m. Phosphate concentrations in hydrothermal waters averaged 354 μg/L and showed no significant spatial trend.

The total bacterial cell concentration (TCC) in the central lake water varied with depth, reaching 6.8 × 10⁶ cells/ml at−5 m depth (Table 1). In contrast, hot spring water exhibits significantly lower cell concentrations (0.4 – 6.8 x10⁵ cells/ml). Low nucleic acid (LNA) cells dominate the lake samples, particularly at the surface (95.5%), with a progressive decrease at greater depths, while the proportion of high nucleic acid (HNA) cells increased. In the hot spring water, the microbial community showed a higher percentage of HNA cells along the gradient (up to 54.8% at 12 meters).

Microbial community composition varied with season and habitat, with bacterial phyla dominating all samples, while archaeal phyla were mainly detected in sediments and hydrothermal sites (Figure 2). In the central water column Bacteria dominated with highly similar phylum-level proportions in both seasons— Actinomycetota 50%, Cyanobacteriota 20%, Pseudomonadota 13%, and Bacteroidota 10%—while the spring profile additionally showed minor contributions from Bacillota and Planctomycetota. Actinomycetota were ubiquitous and abundant in both campaigns, with a strong imprint of Micrococcales (31% in spring and 40% in late summer sampling period) and a secondary contribution of Microtrichales detectable only in late summer (3%) (Figures 2A, B). Among Cyanobacteriota, Synechococcales dominated (24% in spring vs. 16% in late summer) but decreased at −11 m (10%), where Pirellulales, absent along the column, rose to 6%, consistent with the planctomycetal mat recorded at the water-sediment interface during the late summer sampling campaign. Burkholderiales increased in late summer waters (9% vs. 3%), while Archaea were not detected in the water column in both seasons.

In the sediments of the central lake (SED1 spring; SED2 late summer), phylum-level composition was broadly conserved across seasons, dominated by heterotrophic and anaerobic lineages: Pseudomonadota 18–22% (mainly Gammaproteobacteria, families Defluviicoccaceae and Ectothiorhodospiraceae, along with a fraction of Alphaproteobacteria), Bacillota 10–20% (classes Bacilli, Dethiobacteria; families Acholeplasmataceae, Dethiobacteraceae), Chloroflexota 10–12% (Anaerolineae, Dehalococcoidia), and Bacteroidota 6–9% (Bacteroidia, Rhodothermia). In sharp contrast to the water column, sediments hosted a substantial and seasonally modulated archaeal fraction. During the spring season, the archaeal component was higher than in late summer and dominated by Thermoplasmatota. In particular, Thermoplasmata accounted for 72.7% of archaeal reads during the spring season and 68.4% in late summer. Nanoarchaeota were the second most abundant group, represented by the class Nanoarchaeia (14% in spring, and 10% in late summer) followed by Micrarchaeota (6 % spring, and 2 % in late summer). Crenarchaeota, represented by the class Bathyarchaeia, were detected exclusively in spring samples (4%).

The green mat (Figure 3), present only during the spring season, physically separated the oxygenated deep water from the underlying anoxic sediment. Its community was dominated by Planctomycetota, as shown in Figures 3A, 3D (85%; families Pirellulaceae, Rubinisphaeraceae), while Cyanobacteriota (5.5%, mainly Cyanobiaceae) and Pseudomonadota (5%, families Rhodobacteraceae, Rhizobiaceae), occurred at lower relative abundances. The planctomycetal signature recurred both in the deep water at −11 m and in the overlying sediment (Figure 3D). No Archaea were detected in this sample.

In the hot spring samples (Figure 4), the horizontal transect revealed a strong water–sediment disjunction and pronounced spatial structuring. In the hot spring waters (Figure 4A), Cyanobacteriota (Cyanobacteria class, including Cyanobium and Synechococcus genus) and Pseudomonadota (mainly Gammaproteobacteria, including Pseudomonadales, Burkholderiales) dominated, with gradual changes along the transect. Cyanobacteria class was abundant in water near the vent (27% at 0 m) but declined sharply in sediments (3%). In the hot spring sediments (Figure 4B), taxa typical of reduced niches prevailed, including Chloroflexota (Anaerolineae), Desulfobacterota, and Acidobacteriota. At 3 m from the vent, sediments showed a marked peak of Acetothermia (22.4%), absent in the co-localized water. At 12 m, the water–sediment divergence was even stronger, with Chloroflexota reaching 29.6% in sediments vs. 7.6% in water, while Pseudomonadota accounted for 55% of the water but only 17.7% of the sediment.

The archaeal component in hot spring samples also followed distinct gradients. At 3 m in the water, communities were dominated by Nanoarchaeota (Nanoarchaeia, 41.3%), together with Iainarchaeota (25.8%) and Halobacterota (13.7%; classes Methanosarcinia, Methanomicrobia). In contrast, Crenarchaeota (38.7% Bathyarchaeia) and Thermoplasmatota (33% Thermoplasmata) dominated 3 m sediments. At 12 m from the spring, water sample showed the highest contribution of Halobacterota (56%), co-occurring with Nanoarchaeota (40%) and a smaller fraction of Altiarchaeota (4%, exclusive to water).

Community overlapped at the OTU level (Figure 5A) showed pronounced spatial structuring across the three surface sites. Only 2 OTUs were shared by the three habitats (Roiseinatrobacter sp. and Rhodobacteraceae_bacterium), whereas most OTUs were unique to HS12m (858; 45%), followed by the hydrothermal vent (607; 32%) and the central lake surface water (189; 10%). Alpha diversity metrics varied across samples along the surface transect spanning the hydrothermally influenced area and the central lake water (Table 2). Observed richness was lowest in the central lake surface water (CL0m), increased markedly at the hot spring (HS0m), and reached the highest value at 12 m from the hydrothermal spring (HS12m). Shannon diversity showed a different pattern, with the highest value observed at HS0m (Shannon = 5.00), followed by HS12m (Shannon = 4.91), and lower diversity in the central lake surface water (CL0m; Shannon = 3.83). Similarly, Simpson diversity and Pielou's evenness were highest at HS0m, with a more even distribution of taxa at the hot spring compared to the other sites.

Beta-diversity analysis (PCoA based on family level) further highlighted differences in microbial community composition among habitats (Figure 6). Axis 1 (42.6% variance) clearly separated hot spring waters from all other samples, while Axis 2 (16.8% variance) distinguished hot spring sediments, central lake sediments, the green mat, and water column. Pairwise PERMANOVA analyses revealed significant differences in microbial community composition at the family level between water column and central lake sediment (R² = 0.65, p_adj = 0.014), and between water column and hot spring water (R² = 0.59, p_adj = 0.014), with no significant differences in multivariate dispersion between groups (p = 0.07).

FAPROTAX predictions indicated an enrichment of sulfur cycling pathways, methanogenesis, and hydrocarbon degradation in hot spring sediments (Figure 7). In contrast, the central lake water column, both in spring and late summer, was enriched in functions associated with phototrophy, chemoheterotrophy, and nitrogen cycling (nitrification and nitrite respiration). Central basin sediments showed higher contributions of fermentation, anaerobic respiration, and aromatic compound degradation, with the spring samples displaying a stronger representation of functions linked to anoxic processes compared to the late summer samples.