The Samail Ophiolite provides a complete cross-section through 7 km of oceanic crust and 15 km of underlying upper mantle rocks (Boudier and Coleman, 1981; Coleman and Hopson, 1981; Lippard et al., 1986; Glennie, 1996; Nicolas et al., 2000). Partially serpentinized peridotite in the ophiolite actively undergoes hydration and carbonation reactions at low temperatures (≤60?), producing hyperalkaline fluids enriched in dissolved methane and hydrogen gas in the subsurface (Barnes et al., 1978; Neal and Stanger, 1983, 1985; Clark and Fontes, 1990; Kelemen and Matter, 2008; Kelemen et al., 2011, 2013; Paukert et al., 2012; Streit et al., 2012; Miller et al., 2016). The ophiolite’s size, exposure, and accessibility create unique opportunities for studying a broad range of geologic and hydrologic conditions supporting subsurface microbial life (Rempfert et al., 2017). Previous geochemical studies of the Samail Ophiolite groundwater observed a strong partitioning between two subgroups of peridotite-hosted fluids and classified them as “Type I/alkaline” (pH 8–10) or “Type II/hyperalkaline” (pH > 10), while wells positioned at the “contact” between peridotite and gabbro displayed more geochemical variability (Rempfert et al., 2017; Supplementary Figure 1). Alkaline waters have greater availability of oxidants and DIC and more meteoric influences than hyperalkaline, while contact waters appear to be a mixture of the two (Nothaft et al., 2021). We continue to use these designations to refer to well types throughout this study.

In February 2017, a submersible pump was used to collect water samples from five previously drilled wells in the Samail Ophiolite (Table 1). Approximately 100 L of flush water were pumped from at least 20 M below the air-water interface in each well prior to sample collection. Sample volume varied according to the amount of water that could be acquired from each well while continuously pumping: 20 L were collected from NSHQ14, 7 L from WAB71, 18 L from WAB55, 4 L from WAB104, and 20 L from WAB105. Each sample was filtered through 0.3 μM glass fiber filters in a muffled stainless steel inline filter holder using PTFE tubing. Approximately 30 ml filtrate was quenched and stored in liquid nitrogen for transport back to the laboratory, where total organic carbon analysis was performed on a Shimadzu total organic carbon analyzer. The rest of the filtrate was acidified to pH 3 using concentrated HCl for solid phase extraction in the field. Biomass for DNA extraction was collected using in-line 0.2 μM polycarbonate filters as described previously (Fones et al., 2019). A total of 10 ml samples of unfiltered waters from each well were preserved in the field for enumeration of planktonic cells by adding formaldehyde to a final concentration of 10% vol/vol, and transported back to the laboratory for direct cell counting via 4′,6-diamidino-2-phenylindole (DAPI) staining on black polycarbonate 0.2 μM filters. Water temperature, conductivity, pH, oxidation-reduction potential (ORP), and dissolved oxygen concentration were measured during sampling using a Hach HQ40D Portable Multi Meter (Fones et al., 2019).

Dissolved organics were captured from the acidified filtrate on a solid phase extraction Bond Elut-PPL cartridge (Agilent, Santa Clara, CA, USA) (Dittmar et al., 2008; Fiore et al., 2015; Seyler L. M. et al., 2020). 1/8″ × 1/4″ PTFE tubing was used to pull the supernatant into the cartridge to minimize the possibility of contamination from plastic leaching, and discarded flow-through was removed via Viton tubing with a peristaltic pump (flow rate not exceeding 40 ml/min). The cartridges were then rinsed with at least two cartridge volumes of 0.01 M HCl to remove salt, and the sorbent dried with air for 5 min. Dissolved organic matter (DOM) was eluted with 2 ml 100% methanol via gravity flow into muffled amber glass vials. These extracts were then shipped back to the laboratory at Michigan State University, where they were dried down via vacuum centrifugation, and resuspended in 495 μL 95:5 water:acetonitrile and 5 μL 5 μg/ml biotin-(ring-6,6-d₂) standard (Rabinowitz and Kimball, 2007; Fiore et al., 2015). This protocol was repeated for 4 L milliQ water as an extraction blank. In addition, an extraction solvent blank was run consisting of 495 μL of 95:5 water:acetonitrile plus 5 μL 5 μg/ml biotin-(ring-6,6-d₂).

Samples were analyzed via quadrupole time-of-flight liquid chromatography tandem mass spectrometry (QToF-LC/MS/MS) on a Waters Xevo G2-XS UPLC/MS/MS instrument at the MSU Metabolomics Core facility. Each sample was separated chromatographically in an Acquity UPLC BEH C18 column (1.7 μM, 2.1 mm × 50 mm) using a polar/non-polar gradient made up of 10 mm TBA and 15 mm acetic acid in 97:3 water:methanol (solvent A), and 100% methanol (solvent B). The gradient was run at 99.9/0.1 (% A/B) to 1.0/99.0 (% A/B) over 9 min, held an additional 3 min at 1.0/99.0 (% A/B), then reversed to 99.9/0.1 (% A/B) and held another 3 min. At a rate of 400μL/min, the sample was fed into a quadrupole time-of-flight mass spectrometer using a Waters Xevo G2-XS MS/MS in negative ion mode using a data independent collection method (m/z acquisition range 50–1,500 Da). Mass spec data were then imported into Progenesis QI software for statistical analysis and correlation. Quality control was performed by checking ion intensity maps and producing alignments of all sample runs. An aggregate data set was then created from the aligned runs to produce a map of compound ions that was applied to each sample for peak matching. Compound ions were deconvoluted and ion abundance measurements normalized so that runs could be compared. This approach yielded a table of peak areas that are identified by a unique mass to charge (m/z) and retention time (rt) pair referred to as features. Multivariate statistics was then performed on the data including principal components analysis and correlation analysis. A table of relative abundances of features was generated for further analysis. Features found in the blanks were subtracted from the sample data (Broadhurst et al., 2018). In order to be considered present in a sample, a feature had to have at least a two-fold greater abundance in the sample than in both blanks. Features were putatively annotated in Progenesis QI against the ChemSpider database using mass, charge, retention time, and fragmentation spectra. Further annotation of features was achieved using mummichog (Li et al., 2013), which predicts functional activity by compiling metabolic networks using mass spec data. Whenever possible, we used autonomous METLIN-Guided In-source Fragment Annotation (MISA) (Domingo-Almenara et al., 2019), an algorithm that compares detected features against low-energy fragments from MS/MS spectra in the METLIN database (Smith et al., 2005), to increase our confidence in the putative annotation of representative metabolites; however, most of the putative metabolites did not have MS/MS spectral data available in the database. Raw spectrometry data are available in the MetaboLights database (Haug et al., 2020) under study identifier MTBLS6081.

DNA extraction, generation of sequencing libraries, and analysis of metagenomic sequence data, including binning, was described previously (Fones et al., 2019; Kraus et al., 2021).

Bins were taxonomically classified using GTDB-Tk (Chaumeil et al., 2020). Bin quality was assessed using CheckM version 1.0.11 (Parks et al., 2015); bins with 50% or greater completion were selected for further downstream analyses. The relative abundance of each bin in the metagenomes was calculated using a competitive recruitment approach. A bowtie2 index was built from a concatenated file of all bin contigs using bowtie2-build (Langmead and Salzberg, 2012). Quality-filtered metagenome reads were mapped against the bin bowtie2 index using bowtie2 with the -no-unal flag. The resulting SAM files (containing only the reads which mapped) were converted to BAM files, then BamM¹ was used to remove all alignments with <95% identity and <75% alignment coverage. The number of reads in each remaining BAM file was counted using Binsanity-profile (v0.3.3) (Graham et al., 2017). The read counts were then used to calculate the normalized relative fraction of the metagenomes that mapped to the bins using the following equation:

Z-scores of the normalized relative fractions were calculated using the average and sample standard deviation of the relative fraction of each bin across all metagenomes. The bin relative abundance data was run through the Non-negative Matrix Factorization (NMF) package (Gaujoux and Seoighe, 2010) in R to run a standard non-negative matrix factorization algorithm based on the Kullback–Leibler divergence (Brunet et al., 2004). This algorithm generated a heatmap of the bins across samples and a dendrogram using a Pearson correlation distance function. Bins that were representative of particular wells or geochemical conditions were selected using the method defined by Carmona-Saez et al. (2006). MetaSanity (Neely et al., 2020) was used to annotate functional orthologies against the KEGG database (Kanehisa and Goto, 2000); these annotations were run through KEGGDecoder (Graham et al., 2018) to determine the completeness of metabolic pathways of interest. Genes and pathways not detected by KEGGDecoder were searched for in the bins using protein sequences available in UniProt (UniProt Consortium, 2021) and the Conserved Domain Database (Lu et al., 2020), the MUSCLE alignment tool (Edgar, 2004), and the hmmsearch function in HMMER v 3.3.2 (Eddy, 2008).

Metabolomic feature abundance and bin abundance were standardized for ordination according to Legendre and Gallagher (2001) using the decostand function in the R vegan package (v2.4-2).² Regularized Discriminant Analysis (RDA) was performed on the standardized data using the klaR package (v0.6-14).³ These RDAs were then subjected to Procrustes rotation using function Procrustes with symmetric scaling.

Feature annotations acquired using mummichog were cross-referenced with intermediate compounds from pathways identified in representative bins. As for the bins, NMF analysis was performed, and a heatmap of features across samples and a dendrogram using a Pearson correlation distance function were generated. Features that were representative of particular wells or geochemical conditions were extracted using the procedure described by Kim and Park (2007).

Putative annotations of mass spectrometry features included glycan sugars (L-fuculose, L-rhamnose, L-rhamnulose) and their degradation products (L-fucono-1,5-lactone and 2-dehydro-3-deoxy-L-rhamnonate), sugar alcohols, polysaccharides, butanoate, and isocitrate (Figure 5). We further confirmed the annotation of isocitrate using MISA.

Isocitrate was more abundant in the hyperalkaline wells than in the contact or alkaline wells. Isocitrate is an intermediate of multiple metabolic pathways, most notably the citric acid (TCA) cycle, the reverse TCA (rTCA) carbon fixation pathway, the glyoxylate cycle, and coenzyme B biosynthesis in methanogens. A partial TCA cycle was annotated in nearly every bin generated from the metagenomic data, but the rTCA pathway was not identified in any of the bins (Figure 6). The glyoxylate cycle, which uses five of the eight enzymes of the TCA cycle to synthesize macromolecules from two-carbon compounds (e.g., acetate), was annotated in 25 bins, three of which were representative bins of the alkaline wells, three which were representative of the contact well, and one which was representative of the hyperalkaline wells (Figure 6). A complete pathway for coenzyme B biosynthesis was detected in only one bin, a Methanobacterium that was representative of NSHQ14 (Figure 6).

Butanoate was also most abundant in the hyperalkaline wells. Butanoate may be produced by the fermentation of pyruvate, but while fermentation pathways were found in 31 (23 representative) bins (Figure 6), the complete set of enzymes required for pyruvate fermentation to butanoate were not detected in any of the bins (Supplementary Table 3).

A complete pathway for L-fuculose or L-rhamnose degradation was not annotated in any of the bins (Supplementary Table 3). The gene for D-threo-aldose 1-dehydrogenase, which produces L-fucono-1,5-lactone during L-fucose degradation, was only annotated in three bins, but the gene for L-fuconolactonase, which further degrades L-fucono-1,5-lactone to L-fuconate, was annotated in fifteen bins (Supplementary Table 3). Similarly, the gene for the L-rhamnonate dehydratase enzyme that produces 2-dehydro-3-deoxy-L-rhamnonate was only annotated in two bins, but the enzyme that acts on this product, 2-dehydro-3-deoxy-L-rhamnonate dehydrogenase, was annotated in seven bins (Supplementary Table 3).

Chitobiose, a disaccharide produced by the hydrolysis of chitin, was also annotated in the mass spectrometry data (Figure 5). MISA further confirmed the annotation of chitobiose; however, we question the validity of the annotation as the peaks eluted very far apart, suggesting that they did not arise from the same parent molecule (Supplementary Figure 1). Genes for chitinase and diacetylchitobiose deacetylase, enzymes that hydrolyze chitin and chitobiose, were identified in 11 and 16 bins, respectively, though only two bins contained both (Figure 6). Beta-N-acetylhexosaminidase, an exoglycosidase that catalyzes the hydrolysis of chitobiose, was detected in 54 out of 128 total bins, 14 of which were representative bins (six of the seven alkaline representative bins, three hyperalkaline representative bins, and five contact representative bins) (Figure 6). Genes for enzymes involved in the catabolism of other polysaccharides, including alpha-amylase, glucoamylase, beta-glucosidase, D-galacturonate epimerase, D-galacturonate isomerase, oligogalacturonide lyase, and pullulanase were also detected in the metagenomic data, and tended to be associated with bins that displayed the greatest abundance in the hyperalkaline wells (Supplementary Table 3).

Five features were putatively annotated as either cis-homoaconitate, an intermediate in coenzyme B biosynthesis, or cis-2-methylaconitate/2-methyl-trans-aconitate, intermediates in propanoate degradation to pyruvate and succinate via the 2-methylcitrate cycle (Figure 5). These features were more abundant (by up to eight orders of magnitude) in the hyperalkaline wells compared to the contact and alkaline wells. An intermediate of coenzyme M production, (2R)-phosphosulfolactate, was also putatively annotated in the mass spectrometry data and was detected in all of the wells, though it was most abundant in the alkaline wells and was designated a representative feature of the alkaline/contact wells by NMF analysis (Figure 5). A feature putatively annotated as 2-hexaprenyl-6-methoxyphenol, an intermediate in methanogenesis from methoxylated aromatic compounds, was almost exclusively found in NSHQ14 (Figure 5). The enzyme that produces this intermediate has not been characterized.

Complete pathways for methanogenesis via CO₂ and coenzyme M reduction to methane were found in two out of the 128 bins analyzed (Figure 6; Fones et al., 2019). Both of these bins were taxonomically annotated as Euryarchaeota belonging to the Methanobacteriales; one bin was designated a representative bin for NSHQ14, and was one of the most abundant bins in that well (Figure 6; Fones et al., 2019). Both bins also contained the acetoclastic methanogenesis pathway, as well as a nearly complete pathway for methanogenesis from methoxylated aromatic compounds, except for the gene for the first enzyme in the pathway, mtoB (Figure 6 and Supplementary Table 3). The mtoB gene was detected in 51 other bins; 80% of these bins displayed the greatest abundance in the hyperalkaline or contact wells (Supplementary Table 3). A complete 2-methylcitrate cycle for propanoate degradation was detected in four bins, including a representative bin of WAB71 (a member of the Microtrichales) and a representative bin of WAB105 (Acinetobacter junii) (Figure 6).

Fourteen features in the mass spec data were annotated as intermediates of pigment, porphyrin, and quinone biosynthesis pathways (Figure 5). A feature putatively annotated as protoporphyrin IX, a central metabolite in the production of heme and chlorophyll, was detected in all wells and was particularly abundant (3–5 orders of magnitude greater) in NSHQ14 (Figure 5). Another feature, putatively annotated as coproporphyrinogen III (a precursor to protoporphyrin IX) was detected in every well but displayed the lowest abundance in NSHQ14, by more than two orders of magnitude (Figure 5). Complete pathways for the biosynthesis of heme from glutamate and/or glutamyl-tRNA were annotated in bins detected in every well; however, bins containing complete or nearly complete (≥90%) heme pathways were generally more abundant in the hyperalkaline and contact wells (Figure 6).

Three features annotated as intermediates in ubiquinol (4-hydroxybenzoate) and menaquinone synthesis (2-succinyl-5-enolpyruvyl-6-hydroxy-3-cyclohexene-1-carboxylate and (1R,6R)-6-hydroxy-2-succinylcyclohexa-2,4-diene-1-carboxylate, intermediates in the synthesis of menaquinol and phylloquinol from isochorismate) were detected in all of the wells but were most abundant in NSHQ14 (Figure 5). Genes for the enzymes that produce these intermediates were detected in 14, 11, and 6 bins, respectively; these bins tended to be most abundant in the hyperalkaline or contact wells (Supplementary Table 3). The complete pathway for ubiquinone biosynthesis was detected in thirteen bins, including two representative bins of NSHQ14 (Methanobacterium and Pseudomonas) and two representative bins of WAB104 (a member of the Acidobacteria, and Pseudomonas alcaliphila) (Figure 6). The complete pathway for menaquinone biosynthesis was detected in eleven bins, including a representative bin of every well except NSHQ14 (although one representative bin of that well, a member of the Moorellia, had 90% of the pathway) (Figure 6). The most commonly found cytochrome genes in the bins encoded for the cytochrome bd complex and NADH-quinone oxidoreductase. A total of 41 of the 128 bins contained ≥90% of the genes encoding one or both of these cytochromes (Figure 6).

Features putatively annotated as (2′S)-deoxymyxol 2′-alpha-L-fucoside and glycosyl-4,4′-diaponeurosporenoate, metabolites produced in the biosynthesis of the carotenoids myxol 2′-dimethyl-fucoside and staphyloxanthin, respectively, were also detected in all the wells (Figure 5). The biosynthesis of (2′S)-deoxymyxol 2′-alpha-L-fucoside is a multi-step reaction that has not been fully characterized, but we were able to annotate the enzyme that catalyzes the conversion of (2′S)-deoxymyxol 2′-alpha-L-fucoside to myxol 2′-dimethyl-fucoside, beta-carotene hydroxylase, in 18 bins (Figure 6). Genes in the pathway for staphyloxanthin biosynthesis were annotated in nearly every bin, but the complete pathway was only detected in one bin, a member of the Anaerolineales that was most abundant in WAB71 (Figure 6).

A feature found exclusively in NSHQ14 and designated as a representative feature of that well was putatively annotated as tetraprenyl-beta-curcumene, an intermediate in the biosynthesis of a sesquiterpene found in Bacillus spp. (Figure 5). The gene for tetraprenyl-beta-curcumene synthase was detected in seven bins, two of which were most abundant in WAB71, and the rest of which were most abundant in WAB105 (Figure 6).

Two features that were among the most abundant features in the mass spec data were annotated as intermediates of chlorophyll metabolism: magnesium protophorin and 13(1)-hydroxy-magnesium-protoporphyrin IX 13-monomethyl ester (Figure 5). Other putatively annotated features associated with photosynthesis and chlorophyll metabolism include phylloquinone, phytyl diphosphate, and phylloquinol (Figure 5). A feature annotated as either 3,4-dihydroxy-L-phenylalanine or N,N-dihydroxy-L-phenylalanine, precursors to pigments/cytochromes in fungi and plants, was also detected in all of the wells and was a representative feature of the alkaline/contact wells (Figure 5). Most of these features were relatively more abundant in the alkaline and contact wells. However, the feature putatively annotated as phylloquinone was exclusively detected in the hyperalkaline wells and was particularly abundant in NSHQ14 (Figure 5). Because the pathway for phylloquinol/phylloquinone synthesis and menaquinol synthesis involve the same/highly homologous enzymes, we could not distinguish between these pathways in the bins.

The complete pathway for chlorophyll biosynthesis was not detected in any bin (Supplementary Table 3). However, genes for two enzymes in this pathway, magnesium chelatase subunit I and anaerobic magnesium-protoporphyrin IX monomethyl ester cyclase, were annotated in 27 bins and 29 bins, respectively (Figure 6). Magnesium chelatase is a three-subunit enzyme responsible for inserting a Mg⁺² into protoporphyrin IX in the first step of chlorophyll biosynthesis; genes for the other two subunits (H and D) were found in two and five bins, respectively, and none of the bins contained the genes for all three subunits of the enzyme (Supplementary Table 3). Anaerobic magnesium-protoporphyrin IX monomethyl ester cyclase produces 13(1)-hydroxy-magnesium-protoporphyrin IX 13-monomethyl ester in the chlorophyll biosynthesis pathway found in anaerobic photosynthetic bacteria. None of the bins contained the complete suite of genes required for photosynthesis, but two bins (one diagnostic) in the alkaline wells did contain the genes for an anoxygenic type-II reaction center (Supplementary Table 3). Thirteen bins contained the genes for the RuBisCo enzyme, nine of which also contained the complete Calvin–Benson–Bassham pathway for carbon fixation, but none of which included the anoxygenic type-II reaction center (Supplementary Table 3).

A feature putatively annotated as phytyl diphosphate, an intermediate of geranylgeranyl diphosphate biosynthesis, was also detected in all the wells. Geranylgeranyl diphosphate is a precursor to phylloquinol, chlorophyll, archaeol, and multiple pigments across all domains of life.

Eighteen features were annotated as fatty acids and intermediates in fatty acid pathways, five of which were designated characteristic features of the alkaline and contact wells by NMF analysis. These included four peaks annotated as dolichyl diphosphate, lipids involved in N-glycan production in archaea, which were detected in all the wells but were most abundant in NHSQ14 and WAB71. We also detected a feature putatively annotated as 12-oxo-9(Z)-dodecenoic acid (an intermediate of alpha-linolenic acid metabolism), a feature putatively annotated as 30-oxolanosterol (an intermediate of sterol biosynthesis in methylotrophs), and three features annotated as dodecanoic acid (Figure 5). There were also three features found in all the wells but particularly abundant in WAB71, which were annotated as prostaglandins by mummichog. MISA further confirmed the annotation of the latter two features as prostaglandin H2 or prostaglandin D2 (Figure 5 and Supplementary Figure 1). However, we suspect that this annotation is incorrect and that these features are more likely intermediates or secondary metabolites of arachidonic acid and/or eicosapentaenoic acid (see section “Discussion”).

Genes for polyunsaturated fatty acid (PUFA) synthases (arachidonate and/or eicosapentaenoate synthase) were annotated in 18 of the bins analyzed in this study; however, a complete PUFA synthase operon (pfaABCD) was only found in two bins, one taxonomically annotated as belonging to the genus Bellilinea and the other as genus Levilinea, which are in the family Anaerolineaceae (phylum Chloroflexota, class Anaerolineae, order Anaerolineales) (Figure 6 and Supplementary Table 3). The pfaE gene that encodes the phosphopantetheinyl transferase required for catalytic activation of acyl carrier protein domains was not present in these bins, but was found in two other bins belonging to phylum Chloroflexota, class Dehalococcoidia (Supplementary Table 3). Genes for fatty acid synthesis initiation and elongation (fabB, fabD, fabF, fabG, fabH, fabI, fabZ) and long chain acyl-CoA biosynthesis (ACSL, fadD), which produce the precursors to arachidonate and eicosapentaenoate, were annotated in all but two of the representative bins (Figure 6).

Thirteen features were putatively annotated as intermediates of amino acid/polyamine biosynthesis or degradation pathways (Figure 5). Most of these features were detected in all of the wells, with a few exceptions. A feature annotated as agmatine, an intermediate of arginine degradation and putrescine/spermidine biosynthesis, was present in every well except NSHQ14, and was designated a representative feature of the alkaline/contact wells by NMF analysis (Figure 5). Complete putrescine biosynthesis pathways were annotated in 12 bins; interestingly, nearly half of these bins displayed the greatest abundance in NSHQ14 (Figure 6). Conversely, a feature annotated as 5′-S-methyl-5′-thioinosine, a metabolite produced by the deamination of 5′-methylthioadenosine (an intermediate of the methionine salvage cycle and spermidine biosynthesis), was detected in all the wells but was particularly abundant in NHSQ14 (Figure 5). The gene for enzyme that performs the deamination reaction, 5-methylthioadenosine/S-adenosylhomocysteine deaminase, was detected in 50 of the 128 bins analyzed in this study, including four representative bins of NSHQ14, one representative bin of WAB71, three representative bins of WAB55, two representative bins of WAB104, and one representative bin of WAB105 (Figure 6). This enzyme has also been implicated in the recycling of 5′-deoxyadenosine and the biosynthesis of aromatic amino acids in methanogens (Miller et al., 2014). The methionine salvage cycle is also linked to the production of putrescine and spermidine. Only four bins contained ≥90% of the methionine salvage pathway, but 60 bins contained the portion of the pathway that results in 5′-methylthioadenosine/spermidine biosynthesis (Figure 6).

Two representative features of NSHQ14 were putatively annotated as isethionic acid, a product of sulfoacetaldehyde degradation. These features were found only in NSHQ14 and WAB105, but at eight orders of magnitude greater abundance in the former versus the latter well (Figure 5). Sulfoacetaldehyde is both a product of taurine degradation and an intermediate of coenzyme M biosynthesis. The gene for NADH-dependent sulfoacetaldehyde reductase, the enzyme that catalyzes the conversion of sulfoacetaldehyde to isethionic acid, was detected in 77 bins, including four representative bins from NSHQ14, one from WAB71, five from WAB55, three from WAB104, and two from WAB105 (Figure 6).

Two features annotated as either shikimate, 2-isopropylmaleate, or (2S)-2-isopropyl-3-oxosuccinate were detected in all of the wells but were most abundant in the alkaline and contact wells (Figure 5). Shikimate is an intermediate of the shikimate pathway for the production of chorismate, a precursor to aromatic amino acid and menaquinone/ubiquinone biosynthesis. 2-Isopropylmaleate and (2S)-2-Isopropyl-3-oxosuccinate are intermediates of leucine biosynthesis. Two other features, annotated as alpha-isopropylmalate/2-isopropylmaleate/3-isopropylmalate, all of which are intermediates in leucine biosynthesis, were also detected in all the wells but most abundant in the alkaline wells (Figure 5). Complete shikimate and leucine biosynthesis pathways were detected in bins in every well (Figure 6). Nine of the 15 representative bins of the contact and alkaline wells contained one or both of these pathways, but only one of the eight representative hyperalkaline bins contained these pathways (Figure 6).

Other putatively annotated features related to amino acid metabolism include intermediates of lysine biosynthesis (2,3,4,5-tetrahydrodipicolinate), tryptophan degradation (3-hydroxyanthranilate and 2-aminomuconate), and phenylalanine metabolism [phenylacetic acid and 3-(3-hydroxyphenyl)propanoic acid] (Figure 5). Lysine biosynthesis pathways were detected in multiple bins (Figure 6). The complete pathway for tryptophan degradation via 3-hydroxyanthranilate to 2-aminomuconate was annotated in only one bin, which was taxonomically annotated as a member of the Bipolaricaulia (Supplementary Table 3). This bin was more abundant in WAB105 than any other well, but had a low relative abundance compared to other bins in that well. The gene amiE, which encodes the amidase enzyme that converts 2-phenylacetamide to phenylacetic acid, was detected in almost every bin; the gene katG, which encodes the catalase-peroxidase enzyme that converts phenylalanine to 2-phenylacetamide was detected in 43 bins (Figure 6). The enzyme in phenylalanine metabolism that produces 3-(3-hydroxyphenyl)propanoic acid has not been characterized, but the gene for 3-(3-hydroxy-phenyl)propionate hydroxylase, which converts 3-(3-hydroxy-phenyl)propionate acid to 3-(2,3-dihydroxyphenyl)propanoate in the same pathway, was detected in seven bins, only one of which was designated a representative bin of a well (WAB105) (Figure 6).

A feature putatively annotated as pimelate, a precursor of biotin, was detected in every well, with a slightly greater relative abundance in the alkaline wells (Figure 5). The complete pathway for biotin synthesis from pimelate was annotated in only one bin, a member of the Thermodesulfovibrionales that displayed the greatest relative abundance in WAB71 (Figure 6). Twenty-one bins possessed four out of the five genes necessary for the pathway but were missing the gene for the enzyme 6-carboxyhexanoate-CoA ligase (bioW), which initiates the pathway by converting pimelate to pimeloyl-CoA (Supplementary Table 3). Two other bins contained this gene but otherwise lacked the complete pathway (Supplementary Table 3). The mechanism for the biosynthesis of pimelate is currently unknown, though it is speculated to begin with malonyl-CoA (Manandhar and Cronan, 2017). This same study found that deletion of the bioW gene results in a biotin auxotrophic phenotype in Bacillus subtilis.

Another feature, putatively annotated as 7-carboxy-7-carbaguanine (the precursor to preQ0 in preQ0 biosynthesis), was also detected in every well but was most abundant in WAB55 and WAB105 (Figure 5). A total of 36 bins contained the complete pathway for preQ0 biosynthesis, including one representative bin for NSHQ14, four representative bins for WAB55, two representative bins for WAB104, and three representative bins for WAB105 (Figure 6).

Bins we identified as representative of alkaline wells belonged to the Bacillales (family Bacillaceae), Rhizobiales (family Rhizobiaceae), Pseudomonadales (Acinetobacter junii and Pseudomonas alcaliphila), Acidobacteria, Bacteroidota (UBA10030), and Chloroflexota A (Figure 6). The most abundant bins in the alkaline wells also included members of the Thaumarchaeota, Actinobacteriota, Methylomirabilota, and Saccharimonadales (Figure 6). Representative hyperalkaline bins included members of the Thermodesulfovibrionales, Bipolaricaulia (formerly candidate phylum OP1), Chloroflexota (Dehalococcoidia), Moorellia, Methanobacteriales, Microtrichales, Gemmatimonadetes, and Pseudomonadales (Pseudomonas aeruginosa) (Figure 6). Members of the Deinocococales, Burkholderiales, and Nitrospirales were also among the most abundant bins in the hyperalkaline wells (Figure 6). Bins representative of WAB55 (the only contact well sampled in this study) included Nitrospirales, Brocadiales, Chloroflexota (Anaerolineales), Gammaproteobacteria (UBA4486), Firmicutes (CSP1-3), and Poribacteria (Figure 6). A complete list of the bins, including bin completeness, contamination, strain heterogeneity, and phylogenetic classification, is presented in Supplementary Table 4.

Characterization of the DOM pool in the Samail Ophiolite groundwater reveals the presence of numerous metabolic products and intermediates that may provide clues to the survival strategies of microbial communities in environments associated with serpentinization. Among the features we were able to annotate and confirm with metagenomic data, the pigments, porphyrins, fatty acids, sterols, and intermediates/byproducts of pathways associated with methanogenesis (such as coenzyme B and coenzyme M biosynthesis) were the compounds whose abundance was most strongly correlated to fluid type. In particular, pigments are a promising and intriguing biosignature of life in this system, due to both their specificity to particular processes and their longevity in the geological record (Schwieterman et al., 2015).

We chose methods for characterization that we hoped would allow us to capture as many DOM features as possible. These methods naturally have limitations. Many of the metabolic pathways of interest in these wells, in particular those involved with C1 metabolisms, consist of intermediate compounds that are involved in multiple overlapping pathways (Peregrín-Alvarez et al., 2009), many of which are too small to be detected using the liquid chromatography methods used in this study. These compounds are also cycled rapidly and retained within the cell (Bar-Even et al., 2011), precluding them from an exometabolome analysis such as this one. Additionally, many of the features we detected could not be annotated using available metabolite databases- a caveat which reflects both the under-characterization of microbial secondary metabolites on the whole, but also the uniqueness of the environment and the microbial communities that inhabit it. The annotation of features based on mass and fragmentation is also a potential source of error, and structural assignments based on these criteria may not be valid. Metagenomic data can be used to reinforce these annotations but does not provide absolute certainty.

Linking DOM composition to environmental characteristics has exciting implications for the search for potential biomarkers for life fueled by serpentinization, in particular because the greatest numbers of unique features were found in wells that are the most heavily influenced by serpentinization. Metabolomics studies of environments of astrobiological significance, such as serpentinizing rock, hold great promise in the search for biomarkers that may be used as evidence of life on other worlds (Seyler L. et al., 2020). Further efforts to characterize the unique features detected in hyperalkaline fluids in the Samail Ophiolite may thus provide researchers with the means to detect life in serpentinizing systems elsewhere in our Solar System, such as the subsurface of Mars or the icy moons of Jupiter and Saturn. Such techniques hold the possibility of uncovering biomarkers of processes and relationships that are representative of habitats influenced by serpentinization, aiding the search for life in similar environments on other worlds in our Solar System.