Phylotypic Diversity of Bacteria Associated with Speleothems of a Silicate Cave in a Guiana Shield Tepui

The diversity of microorganisms associated with speleological sources has mainly been studied in limestone caves, while studies in silicate caves are still under development. Here, we profiled the microbial diversity of opal speleothems from a silicate cave in Guiana Highlands. Bulk DNAs were extracted from three speleothems of two types, i.e., one soft whitish mushroom-like speleothem and two hard blackish coral-like speleothems. The extracted DNAs were amplified for sequencing the V3–V4 region of the bacterial 16S rRNA gene by MiSeq. A total of 210,309 valid reads were obtained and clustered into 3184 phylotypes or operational taxonomic units (OTUs). The OTUs from the soft whitish speleothem were mostly affiliated with Acidobacteriota, Pseudomonadota (formerly, Proteobacteria), and Chloroflexota, with the OTUs ascribed to Nitrospirota being found specifically in this speleothem. The OTUs from the hard blackish speleothems were similar to each other and were mostly affiliated with Pseudomonadota, Acidobacteriota, and Actinomycetota (formerly, Actinobacteria). These OTU compositions were generally consistent with those reported for limestone and silicate caves. The OTUs were further used to infer metabolic features by using the PICRUSt bioinformatic tool, and membrane transport and amino acid metabolism were noticeably featured. These and other featured metabolisms may influence the pH microenvironment and, consequently, the formation, weathering, and re-deposition of silicate speleothems.


Introduction
Bacterial communities exist in every ecosystem on Earth, and the species compositions of the communities can be adjusted to adapt to various environmental conditions, including caves, as shown by culture-independent pyrosequencing [1]. Bacterial diversity and metabolic strategies in cave ecosystems can improve the comprehension of the biodiversity of different ecosystems around the world, as demonstrated by a metagenomic approach [2]. Although high microbial diversity in caves has been revealed by, for example, clone-byclone sequencing [3,4], cave microbial communities are still among the least studied [5]. Cave microbial communities are affected by rock types and surface-soil richness/poorness, resulting in various geochemical and hydrochemical features of cave streams, such as pH, organic/inorganic nutrient availability, and buffering action [6,7]. On the other hand, cave microbial communities are likely involved in the formation of various geological forms, such as stalactites in limestone caves [8] and "champignons", i.e., mushroom-like white speleothems, uniquely found in silicate caves [9]. The roles of microorganisms have been well studied but perhaps are still poorly understood [10].
Silicate caves have been discovered in 1972 in the table mountains, or tepuis, in Guiana Highlands in South America, especially in southeastern Venezuela and western Guiana [11][12][13][14][15][16][17]. Tepuis are table mountains or mesetas composed of Paleoproterozoic quartzites and sandstones, surrounded by steep cliffs [18]. In the process of exploring these isolated environments, karst structures with numerous cave systems and unique silica deposits have been found [15,19]. Due to long-term weathering and water erosion, a variety of silicate speleothems have been formed inside tepui caves. At present, three sets of the world's largest cave systems in quartzites have been found: one of them in the most popular Roraima-tepui [11,20]; another in the Auyán-tepui, which is known for the world's tallest waterfall, i.e., 979 m high Angel Fall or Kerepakupai Merú [21]; and the third in the Churi-tepui in the Chimantá massif, Venezuela [14]. Other complex networks of underground passages have been found but mainly with deep vertical development; in this case, it is usual to note the extreme scarcity of big speleothems.
In the Roraima-tepui cave system, the microbial involvement in the dissolution of quartz has been implied [15,22]. In another cave in the Roraima-tepui, the silicate cave microflora has been suggested to be affected by limited nitrogen and the poor buffering action of sandstones compared with surface-soil-rich carbonate caves [23]. A microbiological study performed in a cave in the Auyán-tepui showed that both quartz weathering and silica mobility were affected by chemotrophic bacterial communities [24].
This study provides the profiles of MiSeq-generated bacterial phylotypes, or operational taxonomic units (OTUs), based on partial 16S rRNA gene sequences (V3-V4 region) associated with three speleothem samples, two of which were closely similar (i.e., two speleothem types), from the silicate cave system in the Churi-tepui. The compositions and diversity of the retrieved OTUs were generally consistent with those reported for other silicate and limestone caves. The candidate metabolisms of the speleothem-associated bacteria were predicted by the PICRUSt2 bioinformatic tool [25], resulting in the metabolic implication for possible microbial involvement in silicate speleogenesis.

Speleothem Sample Collection
The sampling site was located in a silicate cave of table mountain "Churi-tepui" (ca. 05 • 15 N, 62 • 00 W) in Chimantá Massif, Gran Sabana, Bolívar State, Venezuela. Chimantá Massif is a dissected plateau with 11 table mountains or tepuis, according to the Pemón Indians' name, 8 and 3 of which are located in the northern and southern areas, respectively. Churi-tepui belongs to the southern group, having maximum elevation of ca. 2500 m [26] or 2420 m [27] with a summit area of ca. 47.5 km 2 [26]. Churi-tepui is known for a cave system that accumulates more than 20 km of passages genetically related but currently disconnected in separate caves by breakdowns [14,27]. Charles Brewer Cave is the southernmost of the system and has been mostly studied for geology, geochemistry, and hydrochemistry [15,27], and the pH values of 2.47-2.54 of the acidic cave water was reported for dripping water and "underground river" [27] (p. 38, Table 2). We sampled speleothems formed on the wall of the cave during the Japan-Venezuela Joint Expedition in October 2016.
The speleothems samples were collected at sites on the right bank of the stream between two waterfalls, which were located ca. 540 m from piedra del helicóptero at the only cave entrance by straight-line distance ( Figure 1). Three speleothem samples were collected from an area of about 5 m 2 in range and coded as GM1, GM2, and GM3, and their appearances can be roughly seen in Figure 2. GM1 was characterized as whitish and soft-textured (fragile), while GM2 and GM3 were blackish and hard-textured. These speleothem samples were collected by using a flame-sterilized field knife and stored in situ pre-sterilized Whirl-Pak ® plastic bags. The collected samples were then stored below 4 °C for subsequent laboratory experiments.

Figure 2.
Morphological appearances of the speleothem samples. GM1: speleothem shows a morphology similar to "champignons". GM2 and GM3: speleothems show coral-like morphology. The lengths of the knife blade and handle in the left and middle photos were 9 cm and 12 cm, respectively.

Crystallographic and Geochemical Analyses
Two speleothem samples, GM1 and GM2, were ground into powder using pre-autoclaved mortar and pestle for crystallographic and geochemical analyses, as well as for the DNA extraction mentioned below; GM3 was wholly expensed for DNA extraction to ensure the recovery of maximally extractable DNA. Although GM3 was not used for the analyses, it was located nearby horizontally to GM2 and resembled GM2 in its blackish appearance and hard texture. The ground powders were analyzed by: energy-dispersive X-ray spectroscopy (EDS) at 10 kV and 15 kV for soft whitish and hard blackish speleothems, respectively; and powder X-ray diffraction (PXRD) by Cu Kα radiation (1.54059Å) for crystallographic characterization. EDS and PXRD were conducted using JED-2300T (JEOL Ltd., Tokyo, Japan) and RINT2500 (Rigaku Corp., Tokyo, Japan), respectively, at Natural Science Center for Basic Research and Development (N-BARD) of Hiroshima University. In addition, an elemental analysis on the cut surfaces of the speleothems was performed with an electron probe micro-analyzer (EPMA; JXA-iSP100; JEOL Ltd., Tokyo, Japan) at 15 kV at N-BARD. For the EPMA observation, small intact (not powdered) speleothems were embedded in epoxy resin, cut, and polished mechanically at Thin Section Workshop, Craft Plaza of Hiroshima University.

DNA Extraction, PCR Amplification, and MiSeq Sequencing
Bulk DNAs were extracted from approximately 10 g of the ground-powdered GM1, GM2, and GM3 samples with the ISOSPIN Soil DNA extraction kit (Nippon Gene Co. Ltd., Figure 1. Site of sample collection in Cueva Charles Brewer (Charles Brewer Cave), Churi tepui. The cave map was modified from [9,12].
Microorganisms 2022, 10, x FOR PEER REVIEW 3 pre-sterilized Whirl-Pak ® plastic bags. The collected samples were then stored below for subsequent laboratory experiments.

Crystallographic and Geochemical Analyses
Two speleothem samples, GM1 and GM2, were ground into powder using pre-a claved mortar and pestle for crystallographic and geochemical analyses, as well as fo DNA extraction mentioned below; GM3 was wholly expensed for DNA extraction to sure the recovery of maximally extractable DNA. Although GM3 was not used fo analyses, it was located nearby horizontally to GM2 and resembled GM2 in its blac appearance and hard texture. The ground powders were analyzed by: energy-dispe X-ray spectroscopy (EDS) at 10 kV and 15 kV for soft whitish and hard blackish spe thems, respectively; and powder X-ray diffraction (PXRD) by Cu Kα radiation (1.540 for crystallographic characterization. EDS and PXRD were conducted using JED-2 (JEOL Ltd., Tokyo, Japan) and RINT2500 (Rigaku Corp., Tokyo, Japan), respectivel Natural Science Center for Basic Research and Development (N-BARD) of Hiroshima versity. In addition, an elemental analysis on the cut surfaces of the speleothems was formed with an electron probe micro-analyzer (EPMA; JXA-iSP100; JEOL Ltd., Tokyo pan) at 15 kV at N-BARD. For the EPMA observation, small intact (not powdered) sp othems were embedded in epoxy resin, cut, and polished mechanically at Thin Sec Workshop, Craft Plaza of Hiroshima University.

DNA Extraction, PCR Amplification, and MiSeq Sequencing
Bulk DNAs were extracted from approximately 10 g of the ground-powdered G GM2, and GM3 samples with the ISOSPIN Soil DNA extraction kit (Nippon Gene Co.

Crystallographic and Geochemical Analyses
Two speleothem samples, GM1 and GM2, were ground into powder using preautoclaved mortar and pestle for crystallographic and geochemical analyses, as well as for the DNA extraction mentioned below; GM3 was wholly expensed for DNA extraction to ensure the recovery of maximally extractable DNA. Although GM3 was not used for the analyses, it was located nearby horizontally to GM2 and resembled GM2 in its blackish appearance and hard texture. The ground powders were analyzed by: energy-dispersive X-ray spectroscopy (EDS) at 10 kV and 15 kV for soft whitish and hard blackish speleothems, respectively; and powder X-ray diffraction (PXRD) by Cu Kα radiation (1.54059 Å) for crystallographic characterization. EDS and PXRD were conducted using JED-2300T (JEOL Ltd., Tokyo, Japan) and RINT2500 (Rigaku Corp., Tokyo, Japan), respectively, at Natural Science Center for Basic Research and Development (N-BARD) of Hiroshima University. In addition, an elemental analysis on the cut surfaces of the speleothems was performed with an electron probe micro-analyzer (EPMA; JXA-iSP100; JEOL Ltd., Tokyo, Japan) at 15 kV at N-BARD. For the EPMA observation, small intact (not powdered) speleothems were embedded in epoxy resin, cut, and polished mechanically at Thin Section Workshop, Craft Plaza of Hiroshima University.

DNA Extraction, PCR Amplification, and MiSeq Sequencing
Bulk DNAs were extracted from approximately 10 g of the ground-powdered GM1, GM2, and GM3 samples with the ISOSPIN Soil DNA extraction kit (Nippon Gene Co. Ltd., Tokyo, Japan) and precipitated in 70% ethanol with precipitation facilitator Ethachinmate (Nippon Gene, Tokyo, Japan). The DNA precipitate was resuspended in sterilized ultrapure water. The concentration and purity of the extracted DNAs were checked with NanoDrop (Thermo Fisher Scientific, Waltham, MA, USA) for subsequent procedures and stored at −20 • C. PCR amplicons were generated using the Kapa HiFi HotStart ReadyMix PCR kit (Kapa Biosystems, Wilmington, MA, USA) and the bacterial V3-V4 region-specific primer pair (S-D-Bact-0341-b-S-17, 5 -CCTACGGGNGGCWGCAG-3 /S-D-Bact-0785-a-A-21, 5 -GACTACHVGGGTATCTAATCC-3 ) [28]. The PCR conditions were as follows: 95 • C for 3 min with the lid being heated to 110 • C; 25 cycles of 95 • C for 30 s, 55 • C for 30 s, and 72 • C for 30 s; and a final elongation at 72 • C for 5 min. The sequence library was constructed following our previous method [29]. Pair-end 300 bp sequencing by MiSeq (Illumina, San Diego, CA, USA) was performed using a Nextera XT Index Kit (Illumina) at Department of Biomedical Science, N-BARD, Hiroshima University.

Statistical and Bioinformatic Analyses of the MiSeq-Generated V3-V4 Sequences
Raw sequence data, or raw reads, generated by MiSeq were processed with the Microbiome Taxonomic Profiling (MTP) pipeline by EzBioCloud (https://www.ezbiocloud. net/contents/16smtp) [28]. Briefly, the merging of the pair-end reads as well as PCR primer trimming were conducted using the EzBioCloud in-house pipeline; in this step, unmerged reads as well as short (<100 bp) or low-quality (averaged Q value < 25) reads were omitted. For quality-checked reads, the identical sequences were de-replicated; then, the non-redundant reads obtained were compared to EzBioCloud 16S rRNA gene sequence database PKSSU4.0, with the option of the target taxon of "bacteria". Note that, in this curated database, the uncultured taxonomic group is tentatively given by the hierarchical name assigned to the DDBJ/ENA/GenBank sequence accession number with the following suffixes: "_s" (for species), "_g" (genus), "_f" (family), "_o" (order), "_c" (class), and "_p" (phylum). The taxonomic assignment of the reads was performed based on the following sequence similarity cut-offs: species (≥97%), genus (>97% > x ≥ 94.5%), family (> 94.5% > x ≥ 86.5%), order (>86.5% > x ≥ 82%), class (>82% > x ≥ 78.5%), and phylum (>78.5% > x ≥ 75%), where x corresponds to the sequence identity with sequences in the database. Note that these cut-offs were taken from previous studies [30,31] and are default parameters of the MTP pipeline. The reads below those cut-offs at the species or higher levels were appended with the suffix "_uc" (for unclassified). Next, all reads that could not be identified at the species level (<97% similarity) were subjected to chimera sequence detection through comparison with the EzBioCloud chimera-free reference database (https://help.ezbiocloud.net/mtp-pipeline/), and the chimera reads identified were discarded; unmatched and eukaryotic plastid reads were also excluded. The resulting final dataset was used to pick phylotypes, or operational taxonomic units (OTUs), based on the 97% similarity cut-off value. Using the EzBioCloud MTP pipeline, the rarefaction curve was computed and visualized; the alpha-diversity indices, i.e., Shannon, Simpson, and Chao1 indices, were calculated to estimate the evenness/richness of the OTUs of each speleothem.
The beta diversity showing principal component analysis (PCA) and hierarchical clustering based on the UniFrac distance matrix was also calculated to compare OTU compositions among the speleothem samples. Venn diagrams at the levels from phylum to species were also drawn. Biomarker OTUs that discriminated the speleothem microbiomes were specified by the linear discriminant analysis (LDA) [32] and LDA-Effect Size algorithm (LEfSe; http://huttenhower.sph.harvard.edu/galaxy/) [33]. The threshold on the logarithmic LDA score for discriminative features is generally set to 2 and was set to 2.5 in a rock varnish study [34]; this study set it to 4 to only focus on biomarkers with large statistical differences between samples.

Crystallographic and Geochemical Characteristics of Speleothems
The EDS spectra ( Figure 3) showed that the major elements of the two speleothem samples were silicon, oxygen, and carbon, clearly indicating that the speleothems were silicate rather than carbonate. Carbon was also detected but as a minor element. The relative abundances of Si and O were different between GM1 and GM2, but the reason was unclear. The sharp peaks in the PXRD patterns ( Figure 4) corresponded to those of quartz according to Powder Diffraction File 00-046-1045 (quartz) of International Center for Diffraction Data (https://www.icdd.com). The difference in the quartz peak intensity between  The sharp peaks in the PXRD patterns ( Figure 4) corresponded to those of quartz according to Powder Diffraction File 00-046-1045 (quartz) of International Center for Diffraction Data (https://www.icdd.com). The difference in the quartz peak intensity between the two samples was due to a greater or lesser inclusion of quartz grains. The lump with a maximum at 22 degrees was the most important feature; it is typical of opal-A [37] and was observed in Venezuelan tepui cave speleothems [27] (p. 84, Figure 79A). The EPMA images showed interesting contrasts between GM1 and GM2 (Fig The elemental co-occurrence of oxygen and aluminum rather than silicon was s GM1, while the co-occurrence of sulfur and calcium was only seen in GM2, but the very minor and did not affect the overall SiO2-dominant features of the speleothem The EPMA images showed interesting contrasts between GM1 and GM2 ( Figure 5). The elemental co-occurrence of oxygen and aluminum rather than silicon was seen in GM1, while the co-occurrence of sulfur and calcium was only seen in GM2, but they were very minor and did not affect the overall SiO 2 -dominant features of the speleothems.
Combining the results of the FE-SEM/EDS, PXRD, and EPMA allowed us to identify samples GM1 and GM2 as opal-A, with traces of quartz as impurities from quartz bedrock. The Al element present in sample GM1 probably belonged to pyrophyllite, a metamorphic mineral that has been extensively found in tepui bedrock. Sample GM2 also showed S and Ca, which fits with gypsum also reported for other tepui caves [27,[38][39][40]. The EPMA images showed interesting contrasts between GM1 and GM2 ( Figure 5). The elemental co-occurrence of oxygen and aluminum rather than silicon was seen in GM1, while the co-occurrence of sulfur and calcium was only seen in GM2, but they were very minor and did not affect the overall SiO2-dominant features of the speleothems. Figure 5. EPMA images of cross-sections of two speleothems, GM1 (left) and GM2 (right). Warm and cool colors indicate high and low relative abundances, respectively; black areas reveal the absence or less-than-detectable presence of elements.
Combining the results of the FE-SEM/EDS, PXRD, and EPMA allowed us to identify samples GM1 and GM2 as opal-A, with traces of quartz as impurities from quartz bedrock. The Al element present in sample GM1 probably belonged to pyrophyllite, a metamorphic mineral that has been extensively found in tepui bedrock. Sample GM2 also showed S and Ca, which fits with gypsum also reported for other tepui caves [27,[38][39][40].

Evaluation of MiSeq-Generated V3-V4 Sequences (Reads) and OTUs
An overall total of 225,321 sequences, or reads, were generated from the three speleothem samples, 210,309 (93.3%) of which were validated for quality with an average length of 454.0 bp; 15,012 non-validated reads (6.7%) were filtered out due to low quality. For each speleothem, the number and average length of valid reads for GM1 were 87,168 (92.5% of total) and 454.5 bp; for GM2, 49,279 (99.0%) and 458.4 bp; and for GM3, 73,862 (90.8%) and 452.0 bp, respectively (Table 1). Table 1. Numbers of MiSeq-generated reads, derived OTUs, and annotated taxa in each speleothem sample and the corresponding overall total numbers. Note that the overall total taxa numbers are smaller than simple sums of those for GM1, GM2, and GM3 due to overlaps among samples. Based on the cutoff at 97% similarity, the valid reads of GM1, GM2, and GM3 were grouped into 474, 1576, and 1134 phylotypes, or operational taxonomic units (OTUs), respectively, for a total of 3184 OTUs. The OTUs were annotated to a total of 1122 bacterial species, 516 genera, 262 families, 142 orders, 76 classes, and 30 phyla, and the hierarchical composition of each speleothem is summarized in Table 1.

Sample
The rarefaction curves for reads-OTUs relationships were drawn to calculate the coverage of retrieved OTUs over the predicted total OTUs (=the Chao1 values mentioned below). The coverages for GM1, GM2, and GM3 were 99.33%, 96.66%, and 99.10%, respectively, which showed that the sequencing depth reached in this study was sufficient to describe and characterize microbiomes in the silicate cave speleothems ( Figure 6).
The rarefaction curves for reads-OTUs relationships were drawn to calculate t erage of retrieved OTUs over the predicted total OTUs (=the Chao1 values mentio low). The coverages for GM1, GM2, and GM3 were 99.33%, 96.66%, and 99.10%, tively, which showed that the sequencing depth reached in this study was suffi describe and characterize microbiomes in the silicate cave speleothems (Figure 6). The distribution of OTUs among the samples was visualized using a Venn d (Figure 7) and the associated table ( Table 2) that summarizes the numbers and na phyla in each intersection and relative compliment (Venn diagrams at the levels fro cies to class are shown in Figures S1-S5). The intersection of GM1, GM2, and GM ∩ GM2 ∩ GM3) had the largest numbers of phyla and reads, which showed the similarity of the phylum compositions among the speleothems. Eight phyla were to GM1 (GM1 / GM2 / GM3), while GM2 and GM3 had only one specific phylum. The distribution of OTUs among the samples was visualized using a Venn diagram ( Figure 7) and the associated table ( Table 2) that summarizes the numbers and names of phyla in each intersection and relative compliment (Venn diagrams at the levels from species to class are shown in Figures S1-S5). The intersection of GM1, GM2, and GM3 (GM1 ∩ GM2 ∩ GM3) had the largest numbers of phyla and reads, which showed the relative similarity of the phylum compositions among the speleothems. Eight phyla were specific to GM1 (GM1/GM2/GM3), while GM2 and GM3 had only one specific phylum.  Table 2. Venn diagrams for species, genera, families, orders, and classes are sho supplementary Figures S1-S5.   Table 2. Venn diagrams for species, genera, families, orders, and classes are shown in Supplementary Figures S1-S5.

Alpha and Beta Diversity Analyses
The Chao1, Shannon, and Simpson indices were calculated with EzBioCloud to estimate the alpha diversity (within each sample) of GM1, GM2, and GM3 ( Table 3). The Chao1 index, an estimator of species richness, corresponding to the predicted OTU numbers in the rarefaction curve analysis (Figure 6), ranged from 490.40 in GM2 to 1586.63 in GM1. The Chao1 index values were almost equivalent to the number of retrieved valid OTUs, which agrees with the "high coverages" of 96.66-99.33% in the rarefaction curve analysis, as "coverage" is defined as the ratio of the number of valid reads to the Chao1 index value of the corresponding sample. The Shannon index is an estimator of species evenness or diversity of equally abundant species. The highest Shannon index value of 5.19 was calculated for GM1, which was converted to an effective number of species (ENS) of 179.47 by e 5.19 (=2.718 5.19 ) [41], that is, the equivalent diversity with 179 equally common OTUs, corresponding to 11% of retrieved OTUs of GM1. Applying this approach to the lower Shannon index values of 3.06 for GM2 and 4.05 for GM3, ENS values as low as 21.33 and 57.40 were calculated, respectively.
The Simpson index is also an estimator of both species richness and evenness, with lower values for higher diversity. As the reciprocal of the Simpson index is regarded as a different expression of the ENS, the highest Simpson index (lowest diversity) of 0.16 for GM2 was converted to an ENS of 6.25, and the lowest Simpson index (highest diversity) of 0.02 in GM1 yielded an ENS of 50 [37].
Beta diversity, i.e., similarity/dissimilarity among samples, was represented by PCA and hierarchical cluster analysis. The PCA showed that the OTU populations of the GM1, GM2, and GM3 speleothems were distinguishable (Figure 8, left). The hierarchical cluster analysis showed a closer similarity between the GM2 and GM3 OTUs, rather than similarities with GM1 (Figure 8, right). Table 3. Alpha diversity indices, i.e., Chao1, Shannon, and Simpson indices, for the OTUs for the GM1, GM2, and GM3 speleothems. Effective number of species (ENS) values were calculated from the Shannon and Simpson indices, showing the same tendency of the highest value in GM1 and the lowest value in GM2.

Sample
Valid Beta diversity, i.e., similarity/dissimilarity among samples, was represented by PCA and hierarchical cluster analysis. The PCA showed that the OTU populations of the GM1, GM2, and GM3 speleothems were distinguishable (Figure 8, left). The hierarchical cluster analysis showed a closer similarity between the GM2 and GM3 OTUs, rather than similarities with GM1 (Figure 8, right). The LEfSe analysis identified the biomarker OTUs that affected PCA-based grouping and hierarchical clustering and discriminated the speleothem microbiomes. The cladogram generated by LEfSe showed that the discriminative biomarker OTUs were affiliated with taxa at various ranks ( Figure 9). For example, OTUs affiliated with candidate phylum The LEfSe analysis identified the biomarker OTUs that affected PCA-based grouping and hierarchical clustering and discriminated the speleothem microbiomes. The cladogram generated by LEfSe showed that the discriminative biomarker OTUs were affiliated with taxa at various ranks ( Figure 9). For example, OTUs affiliated with candidate phylum AD3 or "Candidatus Dormibacteraeota" (code "f" in the cladogram) were biomarkers of the GM1 microbiome.
LEfSe identified a total of 100 biomarkers with LDS scores >4 (p = 0.01183), of which 14 biomarkers had LDA scores >5, i.e., log 10 5 or 10 5 ( Table 4). The GM1 microbiome was characterized by the OTUs affiliated with acidobacterial class Solibacteres, while the GM2 and GM3 microbiomes were characterized by the OTUs affiliated with gammaproteobacterial genus Dyella. GM3 was also partly influenced by the OTUs affiliated with Solibacteres.
Microorganisms 2022, 10, x FOR PEER REVIEW 11 of 20 AD3 or "Candidatus Dormibacteraeota" (code "f" in the cladogram) were biomarkers of the GM1 microbiome. Figure 9. LEfSe cladogram showing taxonomic biomarkers for speleothems GM1, GM2, and GM3. The innermost node corresponds to the Bacteria domain, followed by the concentrically arranged nodes of class, order, family, genus, and species. Red, green, and blue nodes/shades indicate taxa that are significantly higher in relative abundance. The diameter of each node is proportional to the abundance of the taxon. Codes in the cladogram with corresponding taxonomic ranks/names are listed in supplementary Figure S11.
LEfSe identified a total of 100 biomarkers with LDS scores >4 (p = 0.01183), of which 14 biomarkers had LDA scores >5, i.e., log105 or 10 5 ( Table 4). The GM1 microbiome was characterized by the OTUs affiliated with acidobacterial class Solibacteres, while the GM2 and GM3 microbiomes were characterized by the OTUs affiliated with gammaproteobacterial genus Dyella. GM3 was also partly influenced by the OTUs affiliated with Solibacteres.  The innermost node corresponds to the Bacteria domain, followed by the concentrically arranged nodes of class, order, family, genus, and species. Red, green, and blue nodes/shades indicate taxa that are significantly higher in relative abundance. The diameter of each node is proportional to the abundance of the taxon. Codes in the cladogram with corresponding taxonomic ranks/names are listed in Supplementary Figure S11. Table 4. Taxonomic biomarkers having LDA scores >5 and their corresponding codes in Figure 9 and Figure S12. All biomarkers with LDA scores >4 are shown in Supplementary Table S1. Figure 9 and Figure S12 Rank
Pseudomonadota, Acidobacteriota, and Actinomycetota represented >70% of the studied microbiomes and as high as 95.99% of the GM1 microbiome. However, the relative abundance among samples was different ( Figure 10). The other phylum present in all samples was Chloroflexota, which accounted for more than 5% in the GM1 and GM3 samples. For other low-abundance phyla, the proportion of the Nitrospirota phylum detected in the GM2 sample was greater than 5%, and the proportion of Planctomycetota phylum detected in the GM3 sample was greater than 2%. Figure 10. Taxonomic compositions of OTUs from the GM1, GM2, and GM3 speleothems. Nine bacterial phyla and two phylum-level lineages (AD3 and OMAN) were observed with >1% sequence abundance in at least one sample in the MiSeq read data. Pseudomonadota, Acidobacteriota, and Actinomycetota represented >70% of the studied microbiomes and as high as 95.99% of the GM1 microbiome. However, the relative abundance among samples was different ( Figure 10). The other phylum present in all samples was Chloroflexota, which accounted for more than 5% in the GM1 and GM3 samples. For other low-abundance phyla, the proportion of the Nitrospirota phylum detected in the GM2 sample was greater than 5%, and the proportion of Planctomycetota phylum detected in the GM3 sample was greater than 2%.

Microbial Metabolic Functions Predicted by PICRUSt2
All OTUs of each speleothem were used for the PICRUSt2 analysis to characterize the functional metabolic profile of each microbiome. At Level 2, which is a high hierarchical level of the KEGG metabolism category classification, each speleothem microbiome had the same 39 pathways (Figure 11). The "membrane transport" pathway accounted for the highest relative abundance in GM1, while "amino acid metabolism" was the highest in GM2 and GM3 and the second highest in GM1. In both pathways, GM1 showed higher relative abundances, while that of "carbohydrate metabolism" was higher in GM2 and GM3.
At KEGG Level 1 (Supplementary Figure S12), each microbiome had seven large pathways, which were "metabolism", "genetic information processing", "unclassified", "environmental information processing", "cellular processes", "human diseases", and "organismal systems" in the order of relative abundance. The relative abundance of the "metabolism" pathway was close to 50% in all the microbiomes.
At Level 3 (Supplementary Figure S13), each microbiome had 251 individual pathways. The "transporters" pathway had the highest proportion (4.36%) in GM1 and the second highest proportion in GM2 and GM3. The pathway categorized as "general function prediction only" had the second highest proportion (ca. 3%) in GM1 and the highest proportion (3.85%) in GM2 and GM3. Another transport pathway, "ABC transporters", was the third most abundant in GM1 (2.83%) but was relatively low in GM2 and GM3. In contrast, GM2 and GM3 had relatively high abundances (ca. 3%) of "bacterial motility proteins", "secretion systems", and "two-component system", which were higher than in GM1. the functional metabolic profile of each microbiome. At Level 2, which is a high hierarchical level of the KEGG metabolism category classification, each speleothem microbiome had the same 39 pathways (Figure 11). The "membrane transport" pathway accounted for the highest relative abundance in GM1, while "amino acid metabolism" was the highest in GM2 and GM3 and the second highest in GM1. In both pathways, GM1 showed higher relative abundances, while that of "carbohydrate metabolism" was higher in GM2 and GM3. Figure 11. KEGG Level 2 metabolic pathways of GM1 (red), GM2 (green), and GM3 (blue) speleothem microbiomes. Pathway categories are shown in the order of relative abundances in GM1. Figure S12), each microbiome had seven large pathways, which were "metabolism", "genetic information processing", "unclassified", "environmental information processing", "cellular processes", "human diseases", and

Discussion
Guiana Shield tepuis present a unique geographical location, special geomorphological features, and some of the longest and deepest silicate caves in the world [20,42]. The mineralogical processes had a very long timespan to occur. The first date record from tepui silica speleothems revealed ages from 53 to 390 thousand years [43,44], so the scarcity of elements and the harsh conditions are counterweighted by the availability of longer periods for biological and mineralogical interactions. The carbon only insignificantly detected in EDS spectra of the speleothems was probably due to organic materials that are abundant in the soil of Chimanta Massif tepuis [27] (p. 42). The relative abundances of Si and O were different between the two speleothem types (Figure 3), which could be related to the hardness or softness (fragility) of the speleothems in relation to the water of crystallization and the age of speleothems. The quartz grains found in the speleothems (Figure 4) were probably impurities derived from sandstone bedrock [27] (p. 80, Figures 78-80, Table 5).
Only the results of culture-independent, MiSeq-generated V3-V4 sequences were reported in this study, although cultivation using the R2A agar plates was tried and resulted in the Sanger sequencing of 16S rRNA genes of four isolates, which were related to Bacillus and Paenibacillus spp., as shown in previous studies on limestone cave microflorae [45,46]. However, their V3-V4 sequences were not found in our dataset from Charles Brewer Cave. They might have been cultured with certain bias or as contaminants and thus were not included in this study.
This study focused on MiSeq-generated V3-V4 sequences and found a total of 30 bacterial phyla in the speleothems of a Guiana tepui silicate cave. In general, the representative phyla were mostly similar to those reported in previous studies (Table 5). Phylum Pseudomonadota accounted for the largest proportion of OTUs in this cave, followed by Acidobacteriota and Actinomycetota. Actinomycetota is generally the dominant phylum, accounting for 60% of the bacterial community of limestone caves [6]. In other calcareous cave studies, Pseudomonadota was identified as the major phylum [47][48][49][50], which is consistent with the result of this study. A study in a Venezuelan orthoquartzite cave showed the dominance of classes Actinomycetales and Alphaproteobacteria in endolithic bacterial communities close to the cave entrance [23].
Phylum Chloroflexota is a ubiquitous phylum and is often found in caves [50,51]. Bacterial communities in a Venezuelan orthoquartzite cave were dominated (82-84%) by class Ktedonobacterales of phylum Chloroflexota [23], which was the first identification of class Ktedonobacterales; we also identified the OTUs of this class, with the read number accounting for >10% of total reads.
The OTUs affiliated with phylum Nitrospirota were mainly found in GM1 ( Figure 10). Nitrospirota occur in different cave systems, such as the extremely acidic Frasassi Cave in Italy [52], as well as the limestone Pajsarjeva jama Cave and Tito Busillo Cave in Spain [3]. In Oylat Cave in a Turkish marble formation, Nitrospirota was the fourth abundant phylum [53]. Species of Nitrospirota contribute to nitrogen cycling by nitrite oxidation and provide nitrogen sources to oligotrophic lava cave habitats [54].
The OTUs affiliated with photoautotrophic phylum Cyanobacteria were found in the speleothem samples. Possible routes of cyanobacterial intrusion are air flow, infiltration with the water flow, and macrobiological vectors such as spiders carrying certain microorganisms and nutrients to underground [39] (p. 34, Photo 3), while entomological resources and the influence of trogloxen fauna such as bat colonies are generally scarce in tepui caves. Silica precipitation and speleothem formation are attributed to the activities of filamentous bacteria, including cyanobacteria [24]. In this study, Cyanobacteria comprised as high as 5% of the total OTUs in three samples.
The OTU-based metabolic prediction of the speleothem microbiomes suggested the presence of chemoautotrophic bacteria, which would support the sustenance of complex microbial communities in the studied silicate cave. Natural nutrient inputs and biogeochemical microenvironments together may have a mutual influence between silica amorphization and microbiomic composition [21]. Local changes in pH and the production of metabolites that influence silica solubility can result from bacterial metabolic processes related to chemoautotrophic activities, e.g., CO 2 fixation and inorganic nitrogen transformation [55,56]. In this respect, GM2 and GM3 showed high abundances of the OTUs affiliated with genus Rhizobiales (family Methylocystaceae) capable of N 2 fixation. The OTUs affiliated with nitrite-oxidizing Nitrospirota, whose presence might be correlated to CO 2 -fixation-coupled ammonia oxidation [57], were detected in all the samples. Previous studies also indicated members of Rhizobiales and Actinomycetota to be involved in biomineralization processes and rock weathering in cave environments [58]. In this study, the amount of Rhizobiales in the three speleothem samples was as high as 19%. In addition, ammonia-oxidizing archaea, i.e., members of Thaumarchaeota, are generally dominant in the microflora of silicate caves [23] and biodegraded limestone walls (not limestone caves) [59]; therefore, archaeal OTUs should be studied with our DNA samples in future studies.
Samples GM2 and GM3 belonged to the same type of speleothems, and the microbiomic functions as predicted by PICRUSt2 showed high similarity. At KEGG Level 1, "metabolism" accounted for the highest proportion of all pathways in all the speleothem samples. Bacterial metabolisms produce corresponding metabolites, and metabolites secreted from cells can affect the near-bacteria microenvironment. At KEGG Level 2, the high proportion of the "membrane transport pathway" was consistent with the secretion of metabolites. At Level 2, "amino acid metabolism" accounted for the highest proportion, and amino acid metabolisms can change the pH of the microenvironment, which may promote the dissolution and re-precipitation of speleothems; the pH values of the studied cave waters as macroenvironments were reported as 2.52 for dripping water and 2.47-2.54 for "underground river" [27] (p. 38, Table 2). Pathways related to "carbon metabolisms" such as glycolysis/gluconeogenesis, citric acid cycle (TCA cycle), pentose phosphate pathway, and pyruvate metabolism were found at KEGG Level 3. The nitrogen metabolism, sulfur metabolism, and carbon fixation pathways were also found at Level 3. These metabolisms can produce a variety of organic and inorganic acids, thereby changing the near-bacterial microenvironment, where discrete condensation may allow elements to stay longer than in flowing or dripping water.
At present, cave microbiology belongs to a new field in biology and geology. A large part of various and numerous cave ecosystems has not been studied, and the diversity and functional potentials of cave microorganisms remain to be explored by both culturedependent and -independent methods [60]. In our study, while understanding the diversity of bacteria in the silicate cave, we also suggested bacterial functions in rock transformation, particularly the formation/dissolution of speleothems; in addition, we aim to elucidate archaeal functions in future studies.

Conclusions
MiSeq V3-V4 microbiomics was performed for the first time on opal speleothems from a silicate cave. The microbiomes of the soft whitish and hard blackish speleothems were separated by differential analysis; however, dominant phyla such as Acidobacteriota, Actinomycetota, Chloroflexota, and Pseudomonadota were mostly similar to those reported for limestone caves as well as silicate caves. The occurrence of Nitrospirota was specific to the soft whitish speleothem, which may be related to the biogeochemical cycling of nitrogen. The metabolic features of the speleothem-associated microbiomes were inferred based on the V3-V4 sequences, and the inferred membrane transport and amino acid metabolism may have influences on speleological processes.
Author Contributions: Q.L. performed the bioinformatic analyses and prepared the manuscript; Z.H. performed the bioinformatic analyses and edited the manuscript; T.N. planned the study, performed sample collection, and reviewed/edited the manuscript; R.N. supported the bioinformatic analyses and edited the manuscript; L.M.R., R.C. and F.U. supported the sample collection and reviewed/edited the manuscript. All authors have read and agreed to the published version of the manuscript.
Funding: Analytical parts of the study were self-funded by T.N.