Samples for microscopy and cultivation were taken from moonmilk deposits originating from three sampling points (collection points COL1, COL3, COL4, Supplementary Figure 1 from Maciejewska et al., 2016) in the upper Viséan limestone cave “Grotte des Collemboles” (Springtails' Cave), Comblain-au-Pont, Belgium (more detailed cave description in Supplementary Figure 1).

Moonmilk samples for strains isolation were brought to the laboratory on ice and stored at 4°C prior to lyophilization. Extensive attempts at the cultivation of Actinobacteria led to the isolation of 31 phylogenetically distinct Streptomyces strains representing phylogenetically distinct phylotypes (as previously described in Maciejewska et al., 2016 and Supplementary Figure 2). Moonmilk samples for scanning electron microscopy (SEM) were preserved in two separate fixative solutions, 2.5% glutaraldehyde in 0.1 M Na-acetate buffer (pH 7.4), and 100% ethanol, and stored at 4°C until analysis. To exclude the effect of fixatives on the crystalline structure of moonmilk, the lyophilized samples were also observed under the microscope as controls (data not shown).

Glutaraldehyde-fixed moonmilk samples were post-fixed in 1% osmium tetroxide (OsO₄) in distilled water, rinsed, and dehydrated through a graded ethanol series (30–100%). Glutaraldehyde-fixed samples and ethanol-preserved samples were then processed by critical point drying, prior to mounting on the glass slides covered by double-side carbon tape together with lyophilized samples. Parts of each sample were mounted to expose the outer surface, a vertical section or fracture made with a scalpel blade. Dry samples were subsequently sputter coated with platinum on Balzers sputtering Unit SCD 030 (Balzers, Lichtenstein).

The production of mineral deposits by isolates MM24 and MM99 following growth in two different culture conditions—calcite precipitation agar (CPA) and modified B-4 medium (more details in CaCO₃ precipitation section) was detected from the living bacterial colonies after being air-dried. Morphological observations were performed by light microscopy (reflected, transmitted, and polarized light) using an Olympus Provis AX-70 microscope fitted with a Visicam 5.0 videocamera for image capture. SEM observations were performed in an environmental scanning electron microscope FEI XL30 ESEM-FEG (Eindhoven, The Netherlands). Platinum-coated samples were observed under high vacuum (HV) conditions with the ET-secondary electron (SE) detector at 10 mm working distance and 15 kV accelerating voltage. Air-dried cultures, were observed under low vacuum (LV) conditions (0.4 Torr) with the large field gaseous secondary electron (GSE) detector and the backscattered electron (BSE) detector at 10 mm working distance and 10 and 20 kV accelerating voltage, respectively. As a result of applying two types of detection—GSE and BSE, the contrast of the images due to the surface morphology or due to the atomic number of the elements and the density of minerals could be obtained. Elemental X-ray microanalysis and mapping was carried out using a Bruker silicon drift energy dispersive detector (SDD Quantax 129 eV, Billerica, MA, USA) at 10–20 kV accelerating voltage with the Esprit 1.9 software. Semi-quantitative analyses of the elemental composition were done using the standard-less ZAF method with automatic background subtraction.

The genes of interest (see Figure 1) including those coding for (i) the ureolytic system (ure), (ii) the Ca²⁺/2H⁺ antiporter system (chaA), (iii) the nitrate/nitrite reductases (nar/nas/nir) with the corresponding transporters (narK), and (iv) the carbonic anhydrase (CA) together with sulfate transporter in cluster with CA (sulP; Felce and Saier, 2005), were retrieved from the genomes of moonmilk Streptomyces sequenced at the Luxembourg Institute of Science and Technology, as previously described (Maciejewska et al., 2016). These genes were first identified within the chromosome of the model Streptomyces species—Streptomyces coelicolor (Bentley et al., 2002). The designations of the selected genes are listed in Supplementary Table 1. Subsequently, genes sequences encoding the corresponding proteins were identified within the genomes of additional 54 reference Streptomyces strains (Supplementary Table 2) for which completely assembled genomes are available in NCBI FTP server (data retrieved on January 8th, 2016).

A total of 407,461 protein sequences were organized in clusters of orthologous groups (COGs) using Proteinortho v 5.12 (Lechner et al., 2010) with the PoFF extension to further discriminate similar sequences based on synteny. Created COGs were used as models to screen moonmilk Streptomyces genomes. For every S. coelicolor gene, the collection of protein sequences in the corresponding COG was used to build a hidden Markov model (HMM) profile (Eddy, 1998). As an example, the gamma sub-unit of the urease metallo-protein of S. coelicolor (UreA, SCO1236) clusters in a COG with 52 sequences from other Streptomyces (Supplementary Table 1). This COG is used to construct a HMM profile representing the UreA protein that is used to search a database of translated predicted coding sequences of the moonmilk Streptomyces. Partial coding sequences resulting from the fragmented nature of the moonmilk Streptomyces genomes were also considered in the screening. The moonmilk Streptomyces coding sequences were predicted with Prodigal v2.6.2 (Hyatt et al., 2010). HMM profile building and HMM search were carried out using the HMMER3 software package (v3.1b2, http://hmmer.org/). The accession numbers of genes recovered from moonmilk Streptomyces are compiled in the Supplementary Table 3.

The assessment of nitrate and nitrite reduction was performed as described by Li et al. (2016), with small modifications. Briefly, the phylotypes were inoculated in nitrate or nitrite agar slants (potassium nitrate/sodium nitrite, 1 g/l; peptone, 5 g/l; beef extract, 3 g/l; agar, 12 g/l; pH 7.0) and nitrate/nitrite broth tubes (potassium nitrate/sodium nitrite, 1 g/l; peptone, 5 g/l; beef extract, 3 g/l; pH 7.0) equipped with the Durham tubes. Escherichia coli ATCC 25922 was used as a control strain for exclusive reduction of nitrate to nitrite, while Pseudomonas aeruginosa ATCC 27853 as a complete reducer of nitrate into the nitrogen gas, collected in the Durham tube. Uninoculated media were used as an additional control. Inoculated test tubes were incubated at 28°C for 9 days in static conditions, to reduce amount of dissolved oxygen. After incubation time few drops of sulfanilic acid and alpha-naphthylamine were added to each test tube, which together react with nitrite generating red/pink color. Reduction of nitrate (NO3-) is indicated by appearance of red color in nitrate broth, while red color disappearance in nitrite broth indicates nitrite (NO2-) reduction.

The glucose oxidative test was carried out according to Hugh and Leifson (1953). Briefly, each representative of the phylotypes was spot-inoculated on Hugh and Leifson's Oxidative-Fermentative (OF) basal medium, prepared as follows: peptone, 2.0 g/l; NaCl, 5.0 g/l; bromothymol blue, 0.03 g/l; K₂HPO₄, 0.3 g/l; agar, 3.0 g/l; pH 7.1. Filter-sterilized glucose was added after autoclaving to a final concentration of 1%. The OF medium not supplemented with glucose was used as a negative control, and oxidative Pseudomonas aeruginosa ATCC 27853 strain as a positive control. The inoculated plates were incubated at 28°C during 7 days and monitored continually to observe development of yellow color indicating on the acid production due to glucose metabolism.

The screening for isolates able to precipitate calcium carbonate through ureolysis was performed using calcite precipitation agar (CPA) as previously described (Stocks-Fischer et al., 1999). The CPA medium was prepared as follows: (g/l); nutrient broth, 3.0 g/l; urea, 20.0 g/l; CaCl₂*2H₂O, 28.5 g/l; NaHCO₃, 2.12 g/l; NH₄Cl, 10.0 g/l; agar, 15.0 g/l. All components were autoclaved apart from urea, which was added filter-sterilized. The same medium, but without urea, was used as an additional control. Alternatively, precipitation of CaCO₃ was tested on modified B-4 medium (pH 7.0) composed of: yeast extract, 4 g/l; calcium acetate, 2.5 g/l; agar, 15 g/l. The plates inoculated with MM24 and MM99, following incubation at 28°C, were examined under ESEM after 2 months of incubation for CPA media, and 1 month of incubation for B-4.

Isolates were tested for their ability to solubilize calcium carbonate on two different solid media including (i) minimal medium (MM; Kieser et al., 2000) containing CaCO₃ (2 g/l), supplemented or not with glucose (5 g/l), and (ii) 1:100 diluted nutrient agar (Portillo et al., 2009), containing CaCO₃ (2.5 g/l), supplemented or not with glucose (2 g/l). The spot-inoculated plates were incubated for 4 weeks at 28°C and the capability of calcium carbonate solubilization was confirmed by observation of the clear halo around a colony.

Filamentous bacteria, particularly Actinobacteria, have been proposed to participate in the genesis of moonmilk deposits by serving as nucleation sites for carbonate deposition (Canaveras et al., 2006). Regardless the fixation procedure (glutaraldehyde, ethanol, and freeze-drying), classical SEM observations revealed the same crystal morphologies that have also been described in the literature (data not shown; Canaveras et al., 1999, 2006; Bindschedler et al., 2010, 2014). The surface of moonmilk samples revealed the presence of dense, unstructured meshes of micrometer-size filaments known as needle-fiber calcite (Figure 2a), which, based on EDS analysis, were shown to be mainly composed of calcium, carbon, and oxygen (Supplementary Figure 3A). The structure of moonmilk deposits was characterized by the presence of abundant, randomly oriented, monocrystalline rods, and polycrystalline fibers composed of stacked rhombohedra (Figure 2a). Those crystals showed variable dimensions, ranging from 0.5 to 1 μm width and 30–100 μm length for monocrystals, and 2–20 μm width, 10–100 μm length for polycrystals, as previously reported (Canaveras et al., 1999). Interestingly, we observed within the moonmilk microscopic composition, abundant networks of filaments with nano-sized width (50–150 nm), which either formed compacted pellets (Figure 2b) or mats (Figure 2c). The organized networks of nano-fibers, similar to the one observed in Figure 2c, were previously reported from moonmilk and associated with fungal wall polymers (Bindschedler et al. 2010; Bindschedler et al., 2014). However, randomly-oriented, compacted pellets of nano-fibers as presented on Figure 2b, were comparable to filamentous Actinobacteria, but characterized by a much smaller cell size which could be a result of the oligotrophic nature of the cave environment. It has been demonstrated that the cell size is largely dependent on the nutritional status of the environment, with resource-poor ecosystems stimulating dwarfism (Young, 2006; Portillo et al., 2013). In this work, the observed size of the putative bacterial nano-filaments, ranging from 0.05 to 0.15 μm, could also be a consequence of the oligotrophic nature of the moonmilk niche as cell elongation/filamentation has been shown to be the result of nutritional stress in some bacteria (Steinberger et al., 2002), including Actinobacteria (Pine and Boone, 1967; Wills and Chan, 1978; Deutch and Perera, 1992). However, even in nutrient-rich soils the majority of bacteria can display a diameter <0.2 μm (Hahn, 2004). In addition, the imprints of nano-sized microorganisms were already reported from other geological formations, such as sedimentary rocks (Folk, 1993; Folk and Chafetz, 2000). Finally, recent findings of ultra-small marine Actinobacteria with an average diameter of about 0.3 μm, provides an additional evidence for the existence of nano-bacteria in oligotrophic ecosystems (Ghai et al., 2013). Unlike crystal fibers, the tiny filaments observed in glutaraldehyde- and ethanol-fixed samples, were not completely straight and regular in their shape but rather displayed plasticity and were often curved, either without preferential orientation (Figure 2b) or in the same direction (Figures 2d,e). The tiny, curved filaments were not observed—probably not preserved—in freeze-dried samples that suffered of ice crystal growth (data not shown). One-directional growth is not a typical behavior of growing actinobacterial filaments that are commonly branching in diverse directions in order to form a complex mycelial network. This rigid and unidirectional growth might suggest that they represent calcified filaments potentially still actively growing at their tip, which remains curved (Figure 2d). Elemental analysis performed with EDS revealed that the observed nano-size filaments contained higher content of carbon and oxygen in comparison to the surrounding crystals, suggesting a possible biological origin (Supplementary Figure 3B). However, these EDS analyses should be viewed with great prudence as the difference in the elemental compositions between filaments and crystals could simply be a consequence of the structure and not the nature (organic vs. mineral) of the analyzed areas. Additionally, along (Figure 2f) or on the tip (Figure 2g) of some filaments, a possible initiation of calcium carbonate deposition was observed. Besides nano-sized filaments, reticulated filaments were sporadically observed within moonmilk crystals (Figure 2h). Those particular filamentous forms, with the size of about 0.5 μm width and up to 75 μm length, are often found in subsurface environment, including limestone or lava caves, and possess higher carbon content than the one typically observed for calcite minerals, suggesting their biogenic origin. However, the associated microorganisms are not yet identified (Melim et al., 2008, 2015; Northup et al., 2011; Miller et al., 2012). Overall, these observations tend to confirm the hypothesis of filamentous microorganisms (bacteria and fungi) serving as a nucleation sites for moonmilk mineral deposition (Canaveras et al., 2006).

A nucleation site itself is not sufficient to promote CaCO₃ precipitation, as dead bacterial cells lose the ability to precipitate minerals (Banks et al., 2010). The dominant role of bacteria in calcification is attributed to metabolic activities which increase the pH of the environment (above pH 8) and therefore favor a shift of the CO₂ – HCO3- – CO32- equilibrium toward carbonate ions which precipitate with Ca²⁺ ions. We screened 31 representative Streptomyces strains isolated from moonmilk (see Supplementary Figure 2 retrieved from Maciejewska et al., 2016) for their metabolic activities that could lead to a raise in pH, including ureolysis, peptides/amino acids ammonification, and dissimilatory nitrate/nitrite reduction to ammonia (Figure 1). Although these assays were performed under laboratory conditions, a qualitative assessment of these processes allowed ranking the tested phylotype representatives according to their metabolic performance, and therefore their potential to drive biomineralization through an increase in the extracellular pH. The results are shown in Figure 3, along with a compilation of activities and phylogenetic relationships (Figure 4). The results of each metabolic activity assay for all the tested strains are presented in Supplementary Figure 4.

Among the 31 isolates tested, 15 showed an ability to increase the pH by hydrolysis of urea (Figure 4). MM99 and MM122 were the strongest ureolytic strains, with comparable metabolic performance to the positive control strain—Klebsiella pneumoniae ATCC 13883 (Figure 3A and Supplementary Figure 4A). The majority of urease-positive moonmilk isolates displayed weak and moderate activities, which were observed either within 3 days of incubation (five strains) or after extended time (more than 1 week; six strains; Figure 3A and Supplementary Figure 4A).

In order to know if a relation could be established between the assessed in vitro activity and the genetic predispositions for ureolysis, we examined the genomes of the phylotype representatives for the presence of urease genes. We used HMM profiles constructed from Clusters of Orthologous Groups (COGs) of proteins of the urease structural subunits [UreA/UreB/Ure(AB)/UreC], as well as the accessory proteins (UreF/UreG/UreD). The corresponding genes originate from functional clusters of three types i.e., ureABCFGD. ure(AB)C, and ure(AB)CFGD, which are all present in the urease-positive species. The large majority of moonmilk strains (90%) was found to encode all the urease genes, in some cases present in several copies (Table 1), suggesting that even the strains displaying urease-negative phenotypes are capable of urease activity. This suggests that in urease-negative strains urease activity is not expressed under the conditions tested, or that they harbor mutations that prevent expression or activity. Nonetheless, urea transport appears to be functional in all urease-negative isolates as urea exerted a toxic effect in the growth media of these strains (an effect that could be reversed by growth in the same medium lacking urea; data not shown).

Urease activity has been recently linked to an activity of the zinc-containing enzyme—carbonic anhydrase (CA; Achal and Pan, 2011). While urease maintains an alkaline environment by generating ammonia, carbonic anhydrase would provide carbon dioxide for biomineralization through dehydration of carbonic acid (H₂CO₃) also produced during ureolysis (Figure 1). It was recently demonstrated that the activities of CA and urease were correlated along the bacterial growth and corresponded to maximum calcite production (Banks et al., 2010; Achal and Pan, 2011), while inhibition of carbonic anhydrase activity decreased the rate of calcification, with calcite precipitation occurring more efficiently with the synergistic action of both enzymes—CA and urease (Dhami et al., 2014).

The presence of this highly efficient enzyme in cave settings was previously confirmed through metagenomic study of a speleothem in Tjuv-Ante's Cave (Sweden; Mendoza et al., 2016), and highly acidic cave biofilms, known as “snottites” (Jones et al., 2011). However, it has never been directly associated to a specific taxonomic group through genome-based approach. The only link was suggested by Cuezva et al. (2012), who proposed CA to be responsible for CO₂ sequestration by gray spot colonization found on the walls of Altamira cave, which were dominated by Actinobacteria.

Evaluation of the genomes of moonmilk phylotypes for the presence of β-CA (Smith and Ferry, 2000), together with the sulfate transporter family protein in cluster with CA (SulP-type permease) revealed that all the investigated strains encode at least one copy of carbonic anhydrase [100% strains encoded β-CA (1) and 87% β-CA (2)] and 90% of them possessing sulfate transporter (Table 1). High copy number of this intracellular zinc metalloenzyme, reaching up to seven copies in MM17, revealed their ubiquitous distribution among karstic bacteria, and suggests their applications in multiple and essential cellular processes beyond their presumed role in carbonate precipitation. Interestingly, a high representation of sulfate permeases of SulP family was also characteristic for the studied population. These broad specificity inorganic anion transporters were suggested to assist bicarbonate (HCO3-) transport (Felce and Saier, 2005), which was experimentally confirmed in marine cyanobacteria (Price et al., 2004). While in cyanobacteria SulP transporter mediates HCO3-/Na²⁺ symport, the substrate specificity in the case of Streptomyces is highly speculative. Nevertheless, fusion of this protein with carbonic anhydrases indeed strongly suggests its participation in HCO3- transport. However, whether it is an importer or exporter remains an open question. It could be possible that HCO3- uptake through SulP increases the intracellular pool of this inorganic carbon species, which potentially might be efficiently transformed into carbon dioxide by carbonic anhydrases, and be exported outside the cell, unless required for cellular metabolism. Alternatively, SulP-dependent export of intracellularly formed HCO3- through CA-mediated hydration of carbon dioxide, would be assisted with import of other ions (Figure 1). Altogether, these findings reveal abundant distribution of genes involved in the inorganic ionic transport and metabolism, which might be related to the biomineralization phenomenon.

Moonmilk Streptomyces were also evaluated for their ability to raise the pH through peptide/amino acid mineralization. Ninety-four percent of tested strains efficiently decomposed nitrogenous compounds into ammonia (Figures 1, 3B, 4 and Supplementary Figure 4B). The majority displayed a strong metabolic phenotype (Figure 5), which is unsurprising given that Streptomyces are well-known ammonifying bacteria in soils, where they actively participate in the decomposition of organic matter (Prakash et al., 2012). In the isolated cave environment, with limited organic matter input from the surface, the source of such macromolecules is unclear, although peptides and amino acids might be entering the cave through water that has percolated through the soil, making such molecules more readily available than urea (Northup and Lavoie, 2001). Amino acid/peptide ammonification was found to be more widespread among the moonmilk isolates than ureolysis, supporting this hypothesis.

Dissimilatory nitrate reduction to ammonium (DNRA) is another nitrogen cycle-related process considered to be involved in calcification (Castanier et al., 2000; Figure 1). This pathway operates in oxygen-limited environments, which could be encountered in the inner layers of the moonmilk deposits. Moreover, moonmilk develops alongside dripping water and its often pasty structure can become fluid, based on its water content, which can drastically reduce oxygen availability. Although Streptomyces are obligate aerobes, they are genetically capable of survival under oxygen-limited conditions, encoding genes related to anaerobic respiration, the so-called “anaerobic paradox” (Borodina et al., 2005). Indeed, the model species Streptomyces coelicolor was reported to anaerobically respire nitrate (Fischer et al., 2010, 2014). Though nitrite, the product of this process, was not reduced to ammonia, but detoxified through extrusion via NO3-/NO2- antiporter system (Fischer et al., 2010, 2012). In S. coelicolor the NirBD reductase was demonstrated to participate in nitrogen assimilation; however a nirBD null-mutant grown in the presence of nitrite and excess ammonium was still able to reduce nitrite suggesting the activity of an alternative and yet unknown enzyme (Fischer et al., 2012). We therefore questioned whether moonmilk isolates would be able to mediate DNRA and reduce nitrate and nitrite under oxygen-limited conditions. For this purpose, we incubated each strain in static (without agitation) liquid culture conditions to limit oxygen availability. Overall, 68 and 77% of strains revealed a capacity to reduce either nitrate (NO3-) or nitrite (NO2-), respectively, with 52% able to reduce both nitrate and nitrite (Figures 3C,D, 4, 5). No N₂ gas production was observed, excluding denitrification, which has been only rarely reported for Streptomyces (Albrecht et al., 1997; Shoun et al., 1998; Kumon et al., 2002). The lack of N₂ generation would suggest a complete reduction of nitrate/nitrite to ammonia via dissimilatory nitrate reduction. Ammonia produced by this pathway, unless not incorporated by other bacteria or oxidized to other nitrogenous compounds, might alkalinize the extracellular environment and thus stimulate CaCO₃ precipitation (Figure 1). In order to evaluate the presence of a DNRA pathway in moonmilk isolates, we screened their genomes for the presence of genes coding for respiratory nitrate reductases (narGHJI) and their associated NarK-type nitrate/nitrite transporter—NarK2 (Figure 1). While 40% of moonmilk strains encoded NarK2 transporter, 30% of them possessed the respiratory nitrate reductases genes (narGHJI), with isolates MM7, MM10, MM48, MM106, MM109, MM111, encoding a complete nar cluster (Table 1). Most of those strains were found to be among the strongest nitrate reducers under oxygen-limited conditions (Figure 4). The fact that only a minority of the moonmilk Streptomyces possessed the genetic material to perform the first step of DNRA, while a majority (68%) was able to reduce nitrate, suggests that another nitrate reduction pathway was operating under the condition tested. This alternative pathway is most likely the NO3- and NO2- assimilatory process that uses nitrate and nitrite as nutrient sources through assimilatory nasA (nitrate reductase) and nirBD (nitrite reductase) genes, with the assistance of additional Nark-type NO3- transporter (Tiffert et al., 2008; Amin et al., 2012; Fischer et al., 2012). Assimilatory reduction of nitrate and nitrite to ammonium is highly plausible as 93% of moonmilk strains possess nasA, nirBD, and narK NO3- importer genes (Table 1). When monitored on solid medium and thus without oxygen limitation, 68 and 77% of the tested strains were able to reduce nitrate and nitrite, respectively (Supplementary Figures 4C,D), which confirmed the high potential of moonmilk Streptomyces to use nitrate and nitrite as a nitrogen source, as previously reported for terrestrial Streptomyces (Pullan et al., 2011; Fischer et al., 2012). Altogether, although moonmilk Streptomyces possess metabolic ability to reduce nitrate and nitrite, without production of gas, pointing on ammonia as a final product, we cannot at this point conclude whether this process represents assimilatory or dissimilatory pathway and whether DNRA is fully functional.

In addition to processes passively influencing carbonate precipitation, bacteria can also actively impact this phenomenon, through an active transport of calcium ions. Banks et al. (2010) suggested that the calcium-toxicity driven removal of this ion outside the bacterial cell is a factor driving calcification phenotype. Therefore, we have also retrieved through an in silico search ChaA, the Ca²⁺/2H⁺ antiporter system suggested to be involved in CaCO₃ deposition (Hammes and Verstraete, 2002; Banks et al., 2010; Figure 1). The presence of the chaA gene was confirmed for 50% of moonmilk cultivable phylotypes (Table 1), extending in those strains the calcium-detoxification system to their spectrum of biomineralization-related processes.

In order to confirm whether the moonmilk cultivable Streptomyces could indeed produce mineral deposits, we selected two phylotype representatives to be first investigated by polarized light microscopy then by ESEM in low vacuum mode for the presence of calcium carbonate precipitates. The selection of strains was based on their predispositions for CaCO₃ precipitation as judged by the sum of metabolic performance observed for ureolysis and peptide/amino acid ammonification—the two most significant activities observed for moonmilk Streptomyces (Figure 4). Strains MM24 and MM99, amongst one of the best isolates based on the metabolic ranking (Figure 4), were simultaneously cultivated on urea-containing CPA medium for 2 months, as well as for 1 month on the modified B-4 medium commonly used for CaCO₃ precipitation assays. Combined microscopic observations of bacterial colonies surfaces with BSE and GSE detectors revealed highly abundant calcite deposits produced by both strains in both media tested (Figures 6, 7). Ureolysis-mediated mineral precipitation was confirmed by the lack of any calcite in urea-deficient CPA medium for both isolates (data not shown). All the produced mineral deposits appeared bright under polarized light (data not shown). The morphology of the calcite polymorphs was comparable between the isolates, however differed between the two culture conditions. On BSE-images, bacterial colonies grown in CPA medium showed a rocky surface with discoidal- or oval-shaped structures of variable diameter that were almost completely encrusting microbial colonies (Figures 6a,c). The mineral nature of the deposits was confirmed through their high (white) contrast on BSE-images compared to the surrounding dark organic matter of the colonies (Figures 6a,c). In addition, their CaCO₃ mineral composition was confirmed by the elemental X-ray analyses (Supplementary Figure 5) and elemental mapping (Figures 6b,d). Calcium, carbon and oxygen were present roughly in stoichiometric proportion of CaCO₃ in the spectra (Supplementary Figure 5), and the distribution of those elements was clearly visualized on the mapping (Figures 6b,d). The presence of calcium was associated with CaCO₃ minerals, while higher proportion of carbon was associated with organic colony biomass (Figures 6b,d). Interestingly, just on the mineral surface, dense webs of nano-sized filaments were observed (Figures 7a–e). On high resolution BSE-images, they appeared either as dark curved filaments at the mineral surface or in the middle of mineralized wrinkles suggesting that bacteria produced minerals in which they got progressively entombed (Figures 7a–d). The filaments free of mineral were also clearly seen interconnecting together and connecting adjacent mineral deposits (Figure 7e). While on the CPA medium both strains prolifically produced relatively small-sized calcite polymorphs (up to 100 μm; Figures 6a,c), on the B-4 medium the observed precipitates although being more scarce, were much larger (up to 400 μm; Figures 6e,g). The inorganic nature of the larger CaCO₃ deposits (Figures 6e,g) were confirmed by X-ray elemental spectra (Supplementary Figure 5) and elemental mapping (Figures 6f,h), which clearly distinguished CaCO₃ minerals from the surrounding bacterial biomass rich in carbon and oxygen. On BSE-image (Figure 7f) and under polarized light (data not shown), tiny CaCO₃ deposits were also detected along randomly distributed bacterial filaments. Additionally, unlike in CPA medium where the mineral surface was rather irregular and unstructured, mineral produced by both strains in B-4 showed morphologically distinct areas, either with a smooth, radiating texture (Figure 7h), with visible filamentous imprints (Figure 7j) or wrinkles (Figure 7g), that presumably corresponded to bacterial nano-sized filaments.

In addition to constructive processes, bacteria are also believed to induce cave bedrock dissolution. As oppose to precipitation, a dissolution phenomenon is related to the acidification of the bacterial microenvironment, most likely as a result of organic acid production, which are the by-products of microbial carbon metabolism. The presence of detectable levels of organic acids in cave environment was previously demonstrated by in situ analysis via ATR-FTIR spectroscopy (Bullen et al., 2008). Released organic acids are able to bind cations such as Ca²⁺ and liberate carbonates, which can be subsequently re-precipitated to form cave secondary deposits or be used by bacteria. Therefore, we have tested the 31 phylotypes for their ability to decrease the pH of the medium through oxidative degradation of glucose based on the standardized oxidative/fermentative test (Hugh and Leifson, 1953). Over 68% of isolates were found to induce media acidification by this pathway, which was observed as yellow color to transparent halo development around the inoculum (Figures 8A, 9). Among them, 52% of the Streptomyces strains exhibited either good or strong oxidative glucose respiration abilities, strongly reducing the extracellular pH (Figures 5, 9 and Supplementary Figure 6).

We have further tested solubilization abilities related to carbon metabolism by cultivating the phylotypes in calcium carbonate containing media—either in minimal medium or in diluted nutrient agar—supplemented or not with glucose. The clear effect of glucose breakdown was observed in diluted nutrient agar in which the supplementation with the carbohydrate induced CaCO₃ dissolution in 26% of the isolates, while none of the phylotypes promoted dissolution in the non-supplemented medium (Figures 8B, 9). In the non-supplemented nutrient agar, rich in organic nitrogen source, the cellular energy comes from amino acid utilization, releasing ammonia as the by-product (Figure 1), which increases the pH of the medium and thus promotes precipitation rather than dissolution. The addition of glucose clearly induces the opposite effect but only in a minority of the isolates which suggests the preference toward amino acids as carbon source over glucose in the large majority of the studied strains (Figures 8B, 9). On the contrary, when assays were performed in the minimal medium, CaCO₃ dissolution was instead inhibited by the exogenous supply of glucose, as only 29% of isolates showed ability to solubilize CaCO₃ under this condition, while in minimal media without glucose supply the dissolution phenotype was characteristic for 81% of isolates (Figures 8C, 9). This might be related to the fact that in minimal medium not supplemented with glucose the only carbon source constitutes the carbonate/bicarbonate from CaCO₃, which is probably efficiently scavenged by cave-dwelling bacteria for autotrophic growth resulting in the high dissolution rate observed under this condition. Although Streptomyces are mainly heterotrophic microorganisms, autotrophic growth using CO or CO₂ as a sole carbon source within members of this genus has already been reported (Gadkari et al., 1990; Kim et al., 1998). The identification of a high number of carbonic anhydrases together with SulP transporters among moonmilk Streptomyces (Table 1) could propose a mechanism via which extracellular bicarbonate would be incorporated into the cell and subsequently converted to CO₂ (Figure 1). In the moonmilk niche that is deprived of organic carbon, the uptake of inorganic carbon could thus be a possible scenario, which would primarily promote CaCO₃ precipitation, and in a second step would lead to CaCO₃ dissolution, as a result of organic acids excretion. The availability of glucose also induces a dissolution phenotype via release of organic acids from glucose breakdown, however only in 3 out of the 31 strains tested. This indicates that, although carbohydrate metabolism might somehow play a role in rock weathering, it is probably not the only operating system leading to this phenomenon, particularly in carbon-limited cave environment, where the source of carbon might be the rock itself.

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