Two worlds beneath: Distinct microbial strategies of the rock-attached and planktonic subsurface biosphere

TL;DR

Rock-attached microbes in groundwater ecosystems are more abundant and functionally distinct from planktonic microbes, playing a key role in subsurface biogeochemical processes.

Contribution

This study reveals the distinct taxonomic and functional profiles of attached versus planktonic microbial communities in carbonate aquifers.

Findings

Attached microbial communities are dominated by Proteobacteria and enriched in genes for biofilm formation and chemolithoautotrophy.

Planktonic communities are dominated by Cand. Patescibacteria and Nitrospirota with lower functional versatility.

Attached microbes likely contribute significantly to inorganic carbon sequestration in carbonate aquifers.

Abstract

Microorganisms in groundwater ecosystems exist either as planktonic cells or as attached communities on aquifer rock surfaces. Attached cells outnumber planktonic ones by at least three orders of magnitude, suggesting a critical role in aquifer ecosystem function. However, particularly in consolidated carbonate aquifers, where research has predominantly focused on planktonic microbes, the metabolic potential and ecological roles of attached communities remain poorly understood.

To investigate the differences between attached and planktonic communities, we sampled the attached microbiome from passive samplers filled with crushed carbonate rock exposed to oxic and anoxic groundwater in the Hainich Critical Zone Exploratory and compared it to a previously published, extensive dataset of planktonic communities from the same aquifer ecosystem. Microbial lifestyle (attached vs. planktonic) explained more variance in community composition than redox conditions, prompting us to further investigate its role in shaping functional and activity profiles. Metagenomic analysis revealed a striking taxonomic and functional segregation: the 605 metagenome-assembled genomes (MAGs) from attached communities were dominated by Proteobacteria (358 MAGs) and were enriched in genes for biofilm formation, chemolithoautotrophy, and redox cycling (e.g., iron and sulfur metabolism). In contrast, the 891 MAGs from planktonic communities were dominated by Cand. Patescibacteria (464 MAGs) and Nitrospirota (60 MAGs) and showed lower functional versatility. Only a few genera were shared, and even closely related MAGs (> 90% average nucleotide identity) differed in assembly size and metabolic traits, demonstrating lifestyle-specific functional adaptation. Analysis of active replication indicated that the active fraction of the attached community was primarily represented by the most abundant MAGs. Planktonic communities featured a higher fraction of active MAGs compared to attached communities, but overall with lower relative abundances.

The high abundance, metabolic specialization, and carbon fixation potential of attached microbes suggest that they are key drivers of subsurface biogeochemical processes. Carbonate aquifers may act as much larger inorganic carbon sinks than previously estimated based on CO2 fixation rates of the planktonic communities alone. Our findings underscore the need to incorporate attached microbial communities into models of subsurface ecosystem function.

Video Abstract

Video Abstract

The online version contains supplementary material available at 10.1186/s40168-025-02325-1.

Figures

Sampling of attached groundwater communities. Passive sampler casing (left) deployed in two groundwater wells (well H41, oxic groundwater, and well H52, anoxic groundwater) filled with polyester mesh bags with autoclaved crushed rock material (right, 2–4 mm in diameter)

Distinction of microbial community composition of attached and planktonic aquifer samples. Principal Component Analysis (PCA) depicts groundwater (blue: 2022, cyan: 2019) and passive sampler (orange) communities from groundwater wells under oxic (light shades) and anoxic (darker shades) conditions. Data is derived from 16S rRNA gene amplicon sequencing

Microbial community composition of attached and planktonic communities based on metagenomic data. A Distribution of MAGs across major phyla in attached (orange) and planktonic (blue) metagenomes. Mean relative abundance of (B) overall phyla and (C) CPR classes in the attached and planktonic communities based on metagenomic coverage of MAGs. For details see Supplementary Table S3 & S4

Biofilm-related functions in attached and planktonic MAGs. Comparative analysis of genetic potential for motility, attachment, quorum sensing, biofilm formation, and biofilm degradation enzymes between attached (orange, 605 MAGs) and planktonic (blue, 891 MAGs) microbial communities. Functions were considered present if at least 50% of the key genes were identified. For details, see Supplementary Table S5 & S6

Functions associated with chemolithoautotrophic growth in attached and planktonic MAGs. A Relative abundance of putative autotrophs with different CO_2_ fixation pathways in attached and planktonic communities, based on metagenomic coverage. MAGs were included when pathways were at least 50% complete and the following key enzymes were present: RuBisCO for Calvin cycle, CODH for Wood Ljungdahl pathway, ATP citrate lyase for rTCA cycle, Pyruvate synthase and PEP carboxylase for DC-HB cycle. The most abundant taxa for each pathway for attached and planktonic communities are indicated. B Presence

Taxonomic and functional overview of actively replicating attached and planktonic MAGs based on iRep analysis. A Taxonomic comparison of active MAGs. Each square represents one MAG, with its size corresponding to its mean relative abundance based on metagenomic coverage. Colors indicate taxonomy on phylum level, and darker shades indicate actively replicating MAGs. Lower panels show a functional overview of active MAGs from attached (B) and planktonic (C) communities. Orange and blue circles show the percentage of MAGs associated with each functional category. Asterisks indicate functions that

Functional comparison of closely related MAGs. The heatmap displays functional contrasts between closely related attached (orange) and planktonic (blue) bacterial taxa, represented by high-quality MAGs with ANI values > 90%. The columns correspond to the MAGs of respective taxa, while each row indicates the presence or absence of specific functions, with categories derived from DRAM and Metabolic-G analyses. For details, see Supplementary Table S5 & S6

Attached versus planktonic microbial lifestyles in carbonate aquifers. In attached communities, microbes drive CO_2_ fixation and EPS formation, harnessing energy from rock-derived minerals. Planktonic microbes rely on groundwater-dissolved nutrients as electron donors for CO_2_ fixation, and heterotrophs might scavenge nutrients from EPS

Full text

Introduction

Earth’s ecosystems host diverse microbial communities that exist in either planktonic or attached lifestyles [1, 2]. Planktonic microbes are free-floating in the water phase, while attached microbes adhere to solid surfaces like rocks, mineral grains, or organic particles [3–6]. Research often focuses on planktonic communities due to ease of access and handling. However, biofilms represent the major lifestyle of prokaryotes on Earth [2]. Despite this universality, the mechanisms behind biofilm formation can vary depending on the environmental conditions present. Different energy sources like organic matter, light, or inorganic compounds can drive the costly biofilm formation process [3, 4, 7]. Especially under the oligotrophic conditions in pristine groundwater, the mechanisms of biofilm formation and the functional traits of the attached community are not well understood.

Microbial attachment to the aquifer rock surfaces may offer several benefits. It provides access to rock-derived minerals, which can serve as electron donors for energy generation [6, 8]. This process is not only sustaining microbial growth but also drives essential biogeochemical processes like rock dissolution and mineral precipitation [9, 10]. In addition, biofilms formed on the rock surfaces can provide protection from environmental stressors and predators [7, 11]. Following an initial attachment and coordinated by intercellular communication via quorum sensing, microbes produce extracellular polymeric substances (EPS) that form the biofilm matrix. After establishment of the biofilm, dispersal can occur to colonize new surfaces, perpetuating the biofilm cycle [2, 12]. In the nutrient-limited groundwater, EPS furthermore represents a source of organic carbon for growth of heterotrophic microbes [1, 2], providing them an advantage compared to a planktonic lifestyle. These benefits might explain why attached communities contain a significantly larger proportion of the aquifer microbiome, outnumbering planktonic cells by at least three orders of magnitude, independent of lithology [1, 13–15]. Given this dominance, the attached communities likely are crucial for providing groundwater ecosystem functions like contaminant degradation and nutrient recycling [9, 10, 16].

To gain insights into the processes driving the attached microbial community in a groundwater ecosystem, we focused on the carbonate aquifers within the Hainich Critical Zone Exploratory (CZE) in Thuringia, Germany [17]. Previous bioreactor experiments with groundwater and rock material demonstrated rapid colonization of rock surfaces [15]. However, only a small subset of groundwater microbes attached in this artificial bioreactor setting, raising the question of whether such differentiation between attached and planktonic communities also occurs in situ, and if so, whether it translates into distinct functional traits with consequences for subsurface ecosystem processes.

Using metagenomics, we provide an in-depth characterization of the taxonomy and functional potential of attached microbes in comparison to their planktonic counterparts. We hypothesize that attached and planktonic communities harbor distinct taxonomic groups and functional traits that reflect their contrasting lifestyles. To test this, we sampled microbial communities using passive samplers filled with crushed limestone rock, deployed in oxic and anoxic groundwater wells for up to one year. The resulting metagenome-assembled genomes (MAGs) were compared with a comprehensive dataset of previously recovered planktonic MAGs [18, 19]. By resolving the functional and ecological differentiation between attached and planktonic communities, our study addresses a major gap in subsurface microbial ecology and advances our understanding of how microbial lifestyles shape biogeochemical processes in carbonate aquifers.

Materials and methods

Passive sampler setup and DNA extraction

Crushed rock materials with 2 to 4 mm diameter were prepared from limestone, specifically calcium mudstones of the Trochitenkalk formation, sourced from an outcrop located in the low-mountain ridge of the monitoring well transect within the Hainich CZE in northwestern Thuringia, Germany. The rock material was filled into small polyester mesh bags of approximately 1 mm mesh size and subsequently autoclaved for sterilization of the rock surfaces. Each bag contained approximately 100 g of crushed rock material. Four bags each of crushed rock material, enclosed in passive sampler casings as previously described (Fig. 1, [20]), were deployed in wells H41 (51.1150842 N, 10.4479713 E, 48 m sampling depth) and H52 (51.1193392 N, 10.4691776 E, 65 m sampling depth) in January 2022 for up to one year. These wells access aquifer layers with distinct redox states: groundwater from H52 in the upper Meissner Formation features anoxic conditions, while groundwater from H41 in the lower Trochitenkalk Formation features oxic conditions (5.1 mg/L O₂) (see Supplementary Table S1 for details on oxygen concentrations). Groundwater from H41 features a higher redox potential (373 ± 65 mV vs. 230 ± 32 mV), lower ammonium concentration (0.08 ± 0.08 mg/L vs. 0.37 ± 0.15 mg/L), and higher nitrate concentration (8.5 ± 4.9 mg/L vs. 1.0 ± 4.0 mg/L) compared to H52, as well as different levels of metal cations [21]. These differences reflect variations in surface connectivity, permeability, and flow dynamics of the sloping strata [21, 22]. Groundwater from both wells is highly oligotrophic, with total organic carbon (TOC) levels remaining below 2 mg/L throughout [21]. Sampling occurred after 6, 18, 30, and 42 (H41) or 54 (H52) weeks of exposure, with subsequent immediate freezing in dry ice and storage at −80 °C. As a control, autoclaved rock material was also sampled without incubation and treated in parallel to the other samples. DNA was extracted using a phenol/chloroform-based protocol as previously described [15]. In brief, 10 g of passive sampler material was treated with 2.6 mL 2 × SET buffer, 30 μL PMSF, and 350 μL 10% SDS, incubated at 55 °C for 2 h with shaking, then centrifuged (4000 × g, 10 min). The supernatant was extracted with phenol:chloroform:isoamyl alcohol (25:24:1), followed by chloroform:isoamyl alcohol (24:1) extraction. DNA was precipitated overnight with glycogen and ethanol, pelleted (4000 × g, 30 min), washed with 80% ethanol, air-dried, and resuspended in 50 μL TE buffer. DNA concentrations were quantified with a Qubit 3.0 fluorimeter (Thermo Fisher Scientific, Waltham, USA) and the Qubit dsDNA HS assay kit (Thermo Fisher Scientific), where DNA concentrations from unincubated rock material were found to be below the limit of detection.Fig. 1. Sampling of attached groundwater communities. Passive sampler casing (left) deployed in two groundwater wells (well H41, oxic groundwater, and well H52, anoxic groundwater) filled with polyester mesh bags with autoclaved crushed rock material (right, 2–4 mm in diameter)

Microbial community profiling

To select samples for metagenomic sequencing, first an amplicon-sequencing-based screening targeting bacterial 16S rRNA genes was conducted. Amplicon libraries were prepared using a two-step barcoding strategy as previously described [23]. The first PCR employed primer pair S-D-Bact-0341-b-S-17/S-D-Bact-0785-a-A-21, which amplify the V3–V4 region of the 16S rRNA gene [24, 25], each modified with Illumina adaptor overhangs to enable subsequent indexing. Reactions (15 μL) were run in triplicate using 1–5 ng of template DNA, 0.4 μM of each primer, 1 μg/μL BSA, and 2 × HotStartTaq Master Mix (Qiagen, Germany). Thermal cycling included initial denaturation at 95 °C for 15 min, followed by 25–30 cycles of 94 °C for 45 s, 55 °C for 45 s, 72 °C for 45 s, and a final extension at 72 °C for 10 min. Samples from unincubated rock material yielded no product at this stage and were hence not considered further. Obtained amplicons were barcoded in a second PCR as previously described [23], using 1 μL of first-round product, 0.5 μM of each barcoding primer, and 2 × Ruby Taq Master Mix (Jena Bioscience, Germany). This step involved 6 cycles of 95 °C for 45 s, 55 °C for 45 s, and 72 °C for 45 s. PCR products were checked for size and integrity via agarose gel electrophoresis. Sequencing was performed on an Illumina MiSeq platform (Illumina, Eindhoven, The Netherlands) with v3 chemistry in paired end mode (2 × 300 bp). Sequence reads were combined with previously published amplicon datasets from the planktonic groundwater community of the Hainich CZE (0.2 µm filter fraction), covering a time series from January 2019 to December 2019 and January 2022 to January 2023 [26]. Data analysis was performed with the DADA2 (Divisive Amplicon Denoising Algorithm 2) pipeline (v1.22.0) using default settings following the core DADA2 algorithm using R software (v4.2.1) [27, 28]. The Vegan software package (v2.6–2) was used to analyze bacterial community patterns [29]. Permutational multivariate analysis of variance (PERMANOVA), implemented via the adonis function [30], was used to assess differences in community composition between attached and planktonic communities. The ggplot2 software package (v3.3.6) was used for graphical representation of the data [31].

Metagenomic sequencing

Based on results from amplicon sequencing, DNA from passive sampler material at 6, 18, and 30 weeks from both wells (10 samples) was selected for metagenomic sequencing. Additionally, to improve the recovery of microbial genomes, metagenomic sequencing was performed on samples from a previous bioreactor experiment ( [15], 10 samples, incubation times 4 to 44 days). The integrity of extracted DNA was determined on a 4200 TapeStation System (Agilent, Santa Clara, CA, USA) using the Genomic DNA ScreenTape (Agilent). Paired-end libraries were generated with an NEBNext Ultra II FS DNA preparation kit (New England Biolabs, Ipswich, MA, USA), and sequenced (2 × 150 bp) on an Illumina NovaSeq 6000 system (Illumina, San Diego, CA, USA) at the Leibniz Institute on Aging– Fritz Lipmann Institute, Jena, Germany. Yields of metagenomic sequencing per sample ranged from 20.5 to 35.4 Gbp (mean = 25.7 Gbp) (Supplementary Table S2).

Metagenomic assembly, binning, and refinement for attached communities

The reads obtained from metagenomic sequencing were trimmed using BBDuk (v39.01) [32] with the following parameters: statscolumns = 5, ktrim = r, qtrim = rl, trimq = 20, minlen = 50, k = 23, mink = 11, and hdist = 1. The quality-controlled reads of each individual sample were assembled using metaSPAdes (v3.15.2) [33] with kmer sizes 21, 33, and 55. Scaffolds larger than 1 kb were used for downstream analyses. Genome binning was carried out using five binning algorithms: Abawaca (v.1.0.7) [34], metabat2 (v2.12.1) [35], CONCOCT (v.1.0.0) [36], BinSanity (v.0.2.7) [37], and MaxBin2 (v.2.2.6) [38] with default parameters. Both the 40 and 107 marker gene sets were utilized in MaxBin2. BinSanity and Abawaca were used to generate bins using contigs of 3000 bp and above. For Abawaca binning, tetranucleotide frequencies were calculated from contigs with a minimum size of 5,000 bp and 10,000 bp. The generated bins were subsequently refined using metaWRAP (v1.3.2) [39], with ≥ 50% completeness and ≤ 10% contamination, representing described thresholds for medium quality bins [40]. Bins were de-replicated using dRep (v3.4.0) [41] at 99% average nucleotide identity (ANI) at secondary clustering to remove strain-level redundancy across samples, resulting in 631 representative MAGs of the rock-attached communities. From these MAGs, we excluded 26 MAGs that were exclusively present in bioreactor samples to avoid the introduction of strains from the laboratory setting, leading to a final set of 605 MAGs from the attached community.

Selection of planktonic groundwater MAGs for comparative analysis

To compare the obtained MAGs from the rock-attached microbiome to planktonic MAGs from the same ecosystem, a dataset of previously recovered Hainich groundwater MAGs based on metagenomic sequencing of 12 samples from January 2019, including wells H14, H32, H41, H43, H51, and H52, was employed [18, 19]. From this dataset, a total of 891 MAGs, dereplicated with the same ≥ 50% completeness and ≤ 10% contamination threshold as the attached MAGs, were selected, as well as quality-filtered metagenomic reads from six samples of the 0.2 μm filter fraction of the groundwater, which contained the majority of the groundwater microbiome. These planktonic MAGs were analyzed in parallel to the attached MAGs as described in the following sections.

Determination of MAG taxonomy, abundance, and functional potential

The 605 attached MAGs and the 891 planktonic MAGs were taxonomically classified using GTDB-Tk (v2.3.0; [42]) with the Genome Taxonomy Database (release 214) as reference. Furthermore, Kaiju (v.1.9.0) [43] with the database nr_euk was used for taxonomic classification of metagenomic datasets on read level. To estimate the relative abundances of each attached MAG in the passive sampler metagenomic datasets, first genome coverages were calculated based on quality-filtered metagenomic data with coverM (v.0.6.1) [44] in genome mode (parameters: –coupled –min-covered-fraction 0 –methods mean). Relative abundances were then calculated by dividing each MAG’s coverage by the total coverage of all genomes in the respective dataset (attached or planktonic), as previously described [45]. This approach was chosen to ensure comparability with values published for the planktonic dataset [18, 19]. We note that this approach reports abundance relative to the binned fraction of the community rather than the entire metagenome (i.e., it does not account for unmapped reads). Relative abundances throughout the paper are hence expressed as per all MAGs combined, and intended to enable a comparison between the attached and planktonic dataset. The fraction of reads mapping to binned genomes per sample is reported in Supplementary Table S2. In addition, SingleM (v0.13.2) [46] was used to determine abundances of taxa not covered on MAG level directly from metagenomic quality-filtered reads. SingleM identifies and classifies microbial taxa by detecting reads of single-copy marker genes.

Functional annotation focusing on biofilm-related functions, biogeochemical cycling, and nutrient transporters, using pre-defined functional categories and individual genes derived from KEGG orthology, was done using DRAM (v.1.4.6) [47] and Metabolic-G (v.4.0) [48] with default settings. For flagella, three key genes responsible for flagellar motor switch protein (FlgG, FliM, and FliN/Y) were considered. For functions linked to iron oxidation and reduction, gene clusters defined by the FeGenie database (v1.2) [49] were used in conjunction with Metabolic-G. A pathway was considered present if more than 50% of the genes associated with it were identified. For CO_2_ fixation pathways, in addition to this cutoff, the presence or absence of key enzymes in the pathway, e.g., ribulose bisphosphate carboxylase/oxygenase (RuBisCO) for Calvin cycle, carbon monoxide dehydrogenase (CODH) for Wood Ljungdahl pathway, ATP citrate lyase for reverse tricarboxylic acid (rTCA) cycle, was also required.

Assessment of MAG quality metrics and genome statistics

The quality (completeness and contamination/redundancy) of the MAGs was calculated based on domain-level bacterial/archaeal specific single copy marker gene sets using the checkM workflow (v.1.2.2) [50], with parameters –min-covered-fraction 0 and –methods mean. For MAGs affiliated with the Candidate Phyla Radiation (CPR), a CPR-specific single copy marker gene set was used. Corrected assembly sizes of the MAGs were calculated by dividing by the number of marker gene sets present and multiplying with the expected number of marker gene sets based on the checkM analysis. The Wilcoxon signed-rank test was used to determine differences between sizes of attached and planktonic MAGs on phylum level, as for lower taxonomic levels, the number of MAGs would have been too low for meaningful comparisons. To evaluate whether observed differences in MAG assembly sizes could be biased by abundance, strain heterogeneity, or contamination, we performed three complementary analyses: (i) we compared relative abundances of attached vs. planktonic MAGs using Welch’s t-test, (ii) tested for correlations between MAG assembly size and relative abundance, and (iii) tested for correlations between MAG assembly size and estimated contamination, both using Spearman’s rank correlation.

Determining active MAGs based on replication indices

To identify actively replicating MAGs, first quality-filtered reads were mapped to MAGs using bowtie2 (v2.4.4) [51]. Mapping files were sorted and indexed via samtools (v1.13) [52]. Subsequently, indices of replication were determined based on the sequencing coverage trend resulting from bi-directional genome replication from a single origin of replication using iRep (v1.10) [53]. Indices were only determined for MAGs with > 75% completeness and < 2% contamination, as previously suggested [53]. A MAG was considered to be active when at least in one metagenomic sample of the respective dataset, R^2^ values calculated between the coverage trend and the linear regression were ≥ 0.9, and reads were covering ≥ 98% of the respective MAG with at least 5 × coverage [53]. To identify functions enriched in active MAGs, ordinary least squares linear models fitted with the lm() function in R (v4.2.1) [28] were used to assess the correlation between functional categories and MAG activity. Functional categories with positive correlations and p-values below 0.05 were considered as significant predictors of microbial activity.

Comparison of shared taxa in attached and planktonic communities

MAGs that were present in both attached and planktonic communities were selected to investigate the adaptation mechanisms that facilitate their survival across lifestyles. For this, first microbial taxa present in both attached and planktonic communities were counted, revealing a total of 27 genera shared between 100 attached and 134 planktonic MAGs. Of these shared taxa, MAGs were selected to calculate pairwise average nucleotide identities using fastANI (v1.33) [54]. Matches between planktonic and attached MAGs with > 90% ANI, a threshold chosen to enable genus-level or low-divergence comparisons [55], were selected for comparative analysis of their functional potential and assembly sizes. Integrity of these MAGs was verified using anvi’o (v7.1) [56]. A fold change in assembly size was calculated between matched pairs of attached and planktonic MAGs. A Wilcoxon signed-rank test was then applied to evaluate whether these fold changes across all pairs were significantly different from zero.

Results

Attachment is the primary determinant of groundwater microbial community composition

When comparing microbial communities from groundwater and passive samplers of the same aquifers based on 16S rRNA gene amplicon data, the attachment preference (attached vs. planktonic) explained the largest proportion of variance (18.5%) in community composition (R^2^ = 0.185, F value = 15.4) (Fig. 2). The presence of oxygen explained 16.7% of variance (R^2^ = 0.167, F value = 13.6) and the year of sampling explained only 2.3% of variance (R^2^ = 0.023, F value = 2.4). These patterns were also evident in the 16S rRNA gene taxonomic profiles of the communities (Supplementary Figure S1). Hence, even with the strong differences in groundwater hydrochemistry across the Hainich CZE, including but not limited to oxic as well as anoxic conditions, attachment preference emerged as a stronger factor shaping community composition than redox conditions. The lack of distinct changes in the planktonic community between 2019 and 2022 also agreed with long-term metabarcoding data at the site showing consistent functional and taxonomic patterns over time [57]. This supports the comparison of MAG datasets acquired at different sampling times (2019 for planktonic communities, 2022 for attached communities) to elucidate the differences between these community types.Fig. 2. Distinction of microbial community composition of attached and planktonic aquifer samples. Principal Component Analysis (PCA) depicts groundwater (blue: 2022, cyan: 2019) and passive sampler (orange) communities from groundwater wells under oxic (light shades) and anoxic (darker shades) conditions. Data is derived from 16S rRNA gene amplicon sequencing

Attached and planktonic communities show drastic taxonomic differences

Distinct community composition was also observed in metagenomic datasets of attached versus planktonic samples of the Hainich CZE groundwater. In the attached communities, Proteobacteria contributed 358 of the total of 605 MAGs (Fig. 3A). In terms of relative abundance, these accounted for 60.3 ± 8.1% (mean ± st. dev.) of the total coverage across all MAGs in the attached community (Fig. 3B). In contrast, Proteobacteria represented only 9.62 ± 5.64% relative abundance in the planktonic community, with 52 MAGs. Instead, 465 of the 891 MAGs in the planktonic community were affiliated with Cand. Patescibacteria, with 39.1 ± 11.8% relative abundance. In the attached community, Cand. Patescibacteria accounted for only 10.0 ± 4.6% relative abundance with 30 MAGs. The planktonic community also featured 102 archaeal MAGs, and 5.5% of the sequencing reads were classified as archaeal (Supplementary Figure S2). From the attached community, only two archaeal MAGs were recovered, with 0.48% of reads being archaeal.Fig. 3. Microbial community composition of attached and planktonic communities based on metagenomic data. A Distribution of MAGs across major phyla in attached (orange) and planktonic (blue) metagenomes. Mean relative abundance of (B) overall phyla and (C) CPR classes in the attached and planktonic communities based on metagenomic coverage of MAGs. For details see Supplementary Table S3 & S4

The dominant Proteobacteria in the attached communities featured highest abundances of genera like Rhodoferax (8.6 ± 6.7%), Aquabacterium (6.4 ± 2.1%), Hydrogenophaga (5.2 ± 3.6%), and Undibacterium (4.6 ± 4.0%). In planktonic communities, only two proteobacterial genera occurred at noteworthy proportions: Phenylobacterium (2.7 ± 5.4%, 17 MAGs) and Nitrosomonas (0.9 ± 1.4%). As the most abundant group in the planktonic community, Cand. Patescibacteria were dominated by Cand. Paceibacteria (19.5 ± 9.3%, 295 MAGs), ABY1 (4.5 ± 2.2%, 65 MAGs), and Cand. Microgenomatia (3.0 ± 1.1%, 61 MAGs) (Fig. 3C). In attached communities, four classes affiliated with Cand. Patescibacteria were present, Cand. Saccharimonadia (4.2 ± 3.5%, 3 MAGs), JAEDAM01 (3.1 ± 3.5%, 6 MAGs), Cand. Paceibacteria (2.4 ± 1.4%, 17 MAGs), and ABY1 (0.2 ± 0.3%, 4 MAGs). In particular, in planktonic communities no MAG of class JAEDAM01 was observed, and this group had less than 0.01% read coverage, making it specific to the attached communities.

Attached microbes feature larger MAGs and widespread biofilm-related functions

We found an attached lifestyle to be reflected by larger genomes. Significantly higher sizes of attached compared to planktonic MAGs were observed for the phyla Nitrospirota (p-value = 1.0 × 10^–4^), Proteobacteria (p-value = 2.1 × 10^–15^), and Bacteroidota (p-value = 1.6 × 10^–2^) (Supplementary Figure S3). The mean MAG size of Proteobacteria in the attached fraction was 4.48 Mb, compared to 1.80 Mb in the planktonic fraction, corresponding to a 2.49-fold difference. Bacteroidota MAGs were 1.87-fold larger (3.89 Mb vs. 2.08 Mb), and Nitrospirota showed a 1.15-fold difference (3.31 Mb vs. 2.88 Mb). We did not find evidence that abundance and strain heterogeneity biased MAG sizes (Figure S4), suggesting that the observed differences are likely to reflect underlying biological variation. These higher MAG sizes also coincided with an increase in functional potential, in particular considering biofilm-related functions (Fig. 4). Attachment functions were encoded in up to 80% of attached MAGs: Pili were mostly present in Gammaproteobacteria, while Nitrospirota featured type 1 and type 3 secretion systems and Alphaproteobacteria showed a greater variety of these attachment genes. Only 8.2% of planktonic MAGs featured attachment genes, primarily for type 3 secretion systems. For the formation of the biofilm matrix, various pathways were present in 98.7% of the attached MAGs: The Raetz pathway was primarily found in Gammaproteobacteria, while Nitrospirota featured Vibrio-type polysaccharide (VPS) biosynthesis. In planktonic communities, only 25.4% of the MAGs, with representatives of Nitrospirota, Proteobacteria, and Planctomycetota, featured such biofilm formation genes. Among Cand. Patescibacteria, only class JAEDAM01 featured biofilm-related functions, including pili, quorum sensing, VPS, and the Raetz pathway. Furthermore, diverse genes related to the degradation of biofilms were present in nearly 65% of attached MAGs (Fig. 4). Some of these genes, primarily for glucosidases, were also present in 32% of planktonic MAGs, representing the only biofilm-related function prevalent in the planktonic microbes.Fig. 4. Biofilm-related functions in attached and planktonic MAGs. Comparative analysis of genetic potential for motility, attachment, quorum sensing, biofilm formation, and biofilm degradation enzymes between attached (orange, 605 MAGs) and planktonic (blue, 891 MAGs) microbial communities. Functions were considered present if at least 50% of the key genes were identified. For details, see Supplementary Table S5 & S6

Distinct CO2 fixation pathways were prevalent in attached vs. planktonic communities

Strikingly, we found a substantially higher fraction of putative chemolithoautotrophs in the attached than in the planktonic community. A MAG was considered to be putatively autotrophic if a CO_2_ fixation pathway was more than 50% complete and the respective key enzyme was present. These criteria were fulfilled by 15.7% of attached MAGs but only 6.6% of planktonic MAGs. Putative autotrophs accounted for a significantly higher relative abundance of 20.7 ± 0.93% (mean ± st.dev) of the attached community compared to 12.7 ± 0.55% in the planktonic community (Mann–Whitney U test, p = 0.0448) (Fig. 5A). The distribution of pathways was distinctly different: The Calvin cycle was by far the most prevalent CO_2_ fixation pathway in the attached community, with 18.4 ± 1.02% of relative abundance. It was primarily found in the abundant proteobacterial genera like Rhodoferax, Aquabacterium, and Undibacterium. In contrast, only 2.4 ± 0.12% of the planktonic community featured this pathway (Fig. 5A). Instead, the Wood–Ljungdahl (WL) pathway, with 4.9 ± 0.62%, and the Arnon–Buchanan cycle, with 4.7 ± 0.91%, had the highest relative abundances in the planktonic community. These pathways primarily occurred in MAGs affiliated with Nitrospirota and Omnitrophota.Fig. 5. Functions associated with chemolithoautotrophic growth in attached and planktonic MAGs. A Relative abundance of putative autotrophs with different CO_2_ fixation pathways in attached and planktonic communities, based on metagenomic coverage. MAGs were included when pathways were at least 50% complete and the following key enzymes were present: RuBisCO for Calvin cycle, CODH for Wood Ljungdahl pathway, ATP citrate lyase for rTCA cycle, Pyruvate synthase and PEP carboxylase for DC-HB cycle. The most abundant taxa for each pathway for attached and planktonic communities are indicated. B Presence of CO_2_ fixation pathways and genes associated with biogeochemical cycling (nitrogen, sulfur, and iron), based on DRAM and Metabolic–G analysis, across the attached (orange) and planktonic (blue) MAGs. Functions were considered present if at least 50% of the required genes were found. For CO_2_ fixation pathways, shades indicate completeness. C Percentage of MAGs from attached versus planktonic communities featuring transport functions for inorganic electron donors and acceptors as well as organics and other compounds. For details, see Supplementary Table S5 & S6

Higher abundance of genes associated with other biogeochemical cycles in attached communities

The metabolic potential for reduction and oxidation of inorganic compounds (Fe, S, N), essential for microbial energy acquisition in groundwater, was generally more widespread in attached than planktonic MAGs (Fig. 5B). The cyc1 gene for iron oxidation was found in 42% of attached MAGs (mainly Gammaproteobacteria), compared to only 6% in planktonic MAGs. Iron reduction genes (mtrBC, dmkAB, and cytochromes DFE_0448-0451 and DFE_0461-0465) were more widespread, found in 91% of attached MAGs and 61% of planktonic MAGs. The sox cluster for sulfur oxidation was likewise present in 38% of attached MAGs, mainly Gammaproteobacteria and Myxococcota, but only 5% of planktonic MAGs, including Nitrospirota and Chloroflexota. The sat-apr-dsr system, linked to both dissimilatory sulfate reduction and sulfur oxidation, was found in comparable percentages of MAGs, in 10.5% of attached MAGs and 8.6% of planktonic MAGs.

For nitrogen metabolism, primarily reductive pathways were found, mostly in Proteobacteria and Nitrospirota of both communities. Genes specific for denitrification (napAB, nirK/nirS, norBC, nosZ) were found in up to 39% of attached MAGs and 11% of planktonic MAGs. The nirBD or nrfAH genes specific for dissimilatory nitrate reduction were present in 11.4% of attached MAGs and 2.6% of planktonic MAGs. In contrast, nitrification-specific genes (amoCAB, hao) were found in similarly low numbers in both attached (5.6%) and planktonic MAGs (5.1%).

The most abundant attached genera, like Aquabacterium, Rhodoferax, and Undibacterium, possessed genes for both iron and/or sulfur oxidation as well as denitrification. While a comparable number of attached and planktonic autotrophs harbored genes for nitrite oxidation (47% and 42%, respectively) and ammonia (14% and 12%) oxidation, a significantly higher proportion of attached autotrophs possessed genes for sulfur oxidation (73% vs. 28%) and iron oxidation (64% vs. 10%) than planktonic autotrophs. This provided evidence that sulfur and iron oxidation played a more important role in fueling autotrophy in the attached community.

Nutrient transporters were more abundant in the attached microbial community

Transporters for uptake of inorganic electron donors and acceptors were likewise more widespread in the attached community: Iron transporters were present in 90% of attached but only 24% of planktonic MAGs (Fig. 5C). Transporters for sulfate and reduced sulfur compounds were present in 55% of attached but only 11% of planktonic MAGs. The distribution of transporters for nitrogen compounds was less skewed, with 27% attached and 13% of planktonic MAGs encoding them. Likewise, transporters responsible for the uptake of simple and complex sugars, such as arabinosaccharide, glucose, and mannose, were present in both attached (17%) and planktonic (10%) MAGs.

Active replication in a higher portion of planktonic vs. attached MAGs

To identify actively growing taxa, we calculated indices of replication (iRep) as previously described (Brown et al. 2016). Replication indices were generally below the thresholds typically associated with strain heterogeneity (Supplementary Table 2 and 3). Overall, the active attached community consists of fewer but more abundant MAGs compared to the planktonic community and possessed a broad functional repertoire to thrive in biofilms. A higher proportion of planktonic MAGs (42%, 373 MAGs) compared to attached MAGs (25%, 151 MAGs) featured iRep values indicating growth. In the attached community, the most abundant MAGs were active (Fig. 6A). In comparison, in the planktonic community, a large proportion of the active MAGs were of low abundance. As a result, in terms of relative abundance, a higher proportion of the attached community (57%) compared to the planktonic community (38%) was active. The active attached taxa included the proteobacterial key players such as Aquabacterium, Rhodoferax, and Undibacterium, as well as the most abundant Actinobacteria, Myxococcota, Nitrospirota, and Verrucomicrobiota MAGs. Contrastingly, in the planktonic community, the diverse Cand. Patescibacteria made up 61% of the active MAGs, while in the attached community, only Cand. Patescibacteria MAGs of group JAEDAM01 were active.Fig. 6. Taxonomic and functional overview of actively replicating attached and planktonic MAGs based on iRep analysis. A Taxonomic comparison of active MAGs. Each square represents one MAG, with its size corresponding to its mean relative abundance based on metagenomic coverage. Colors indicate taxonomy on phylum level, and darker shades indicate actively replicating MAGs. Lower panels show a functional overview of active MAGs from attached (B) and planktonic (C) communities. Orange and blue circles show the percentage of MAGs associated with each functional category. Asterisks indicate functions that are significantly enriched in the respective community (linear model, p-value ≤ 0.05). For details, see Supplementary Table S7

To identify which metabolic traits were linked to activity, we specifically compared functional gene presence and active replication across MAGs. More than 50% of the active attached MAGs exhibited functions related to biofilm formation (Fig. 6B). Attached MAGs with genes for the Raetz pathway for EPS biosynthesis, the Calvin–Benson–Bassham (CBB) cycle, and denitrification were active significantly more often than MAGs without these functions (Fig. 6B). In contrast, less than 15% of active planktonic MAGs carried biofilm formation-related functions (Fig. 6C), while their activity was more often linked to EPS degradation (especially via glucosidases), nitrate, and iron reduction. Across both communities, CO_2_ fixation pathways (WL, rTCA) were positively correlated with activity, indicating that autotrophy was a common predictor of activity regardless of lifestyle.

Overlapping taxa exhibited functional differences

To determine whether taxa present in both attached and planktonic communities represent microbes transitioning through the planktonic state to colonize new surfaces, we compared the genomic functions of these overlapping taxa. We found that only 27 genera (~ 7% of the total genera) were shared. These taxa were mostly affiliated with Proteobacteria, Nitrospirota, and Bacteroidota. Of these, only seven pairs of MAGs featured ANI values above 90%, indicating close relatedness. All attached MAGs exhibited a broad range of functions for flagellar biosynthesis and chemotaxis, as well as biofilm formation and quorum sensing (Fig. 7). In contrast, these functions were mostly absent in planktonic MAGs. Genes for biofilm degradation, sugar uptake, and other transport systems were likewise more abundant in the attached compared to the planktonic MAGs.Fig. 7. Functional comparison of closely related MAGs. The heatmap displays functional contrasts between closely related attached (orange) and planktonic (blue) bacterial taxa, represented by high-quality MAGs with ANI values > 90%. The columns correspond to the MAGs of respective taxa, while each row indicates the presence or absence of specific functions, with categories derived from DRAM and Metabolic-G analyses. For details, see Supplementary Table S5 & S6

Functions for oxidation and reduction of iron, nitrogen, and sulfur compounds showed strong differences in their distribution in the closely related attached and planktonic MAGs. For example, the planktonic Bacteroidota MAG contained nitrification genes that were missing from its attached counterpart, and the attached Nitrospirota 9FT-COMBO.42.15 contained nitrogen cycling genes absent in its planktonic relative. Attached MAGs had significantly larger assembly sizes (1.40 ± 0.31 times, mean ± st. dev.) than planktonic MAGs (p-value = 0.01073, Wilcoxon signed-rank test) despite comparable completeness above 90% and contamination below 10% (Supplementary Table S8). Thus, although being closely related, the overlapping taxa showed lifestyle-specific differences.

Discussion

Understanding the ecological principles driving the life of groundwater microbial communities is crucial for unraveling biogeochemical processes in the subsurface. Our results show that across taxonomic, functional, and activity-based analyses, microbial lifestyle, i.e., the preference for attached or planktonic growth, consistently emerged as the strongest determinant of community structure and function in carbonate rock aquifers. Even severe differences in the hydrochemical conditions, such as redox gradients from oxic to anoxic settings, exert weaker effects compared to lifestyle. These findings are surprising in light of prior work emphasizing hydrochemistry and redox gradients as the primary drivers of groundwater community structure [9, 10], and suggest that a paradigm shift towards a focus on microbial lifestyles (attached vs. planktonic) is necessary for a better understanding of groundwater microbial ecology.

Many groundwater studies have interpreted microbial community patterns through the lens of the seed bank hypothesis, which posits that the planktonic community acts as a reservoir of dormant or low-abundance taxa capable of colonizing surfaces when conditions permit. [58, 59]. This concept is supported in porous, unconsolidated aquifers, where dynamic exchange between water and sediment allows frequent microbial dispersal [59–61]. However, our genome-resolved analysis of a consolidated carbonate aquifer reveals a fundamentally different picture. Attached and planktonic communities were not only taxonomically distinct, but functionally segregated, with minimal overlap at the MAG level. Even these closely related taxa (with > 90% ANI) exhibited pronounced differences in MAG size and metabolic potential depending on lifestyle, including biofilm formation, redox metabolism, and environmental sensing. We note that MAG recovery is biased toward abundant organisms, and low-abundance taxa relevant for rock surface colonization therefore might be overlooked in the planktonic community. Nevertheless, within these methodological limits, our findings suggest that in fractured rock aquifers, biofilms are not seeded from the planktonic microbiome, but rather form functionally distinct and relatively isolated communities due to selective and stable ecological filtering. Our results thus challenge the broad applicability of the seed bank model in groundwater microbiology, and call for revised conceptual frameworks that recognize the limited connectivity and strong functional divergence between lifestyles in consolidated aquifers.

The functional adaptations for biofilm formation observed in the attached community are widespread in the bacterial domain. Traits including adhesion functions and EPS production are distributed throughout taxa and environments [2, 4, 62–64]. This prevalence aligns with the general tendency of microorganisms to colonize surfaces and live in biofilms, from ecosystems in the natural environment to host- and disease-associated communities [2, 65, 66]. The presence of biofilm-associated genes, together with larger MAG sizes and metabolic versatility, positions attached microbes as functionally rich and ecologically stable anchors of the subsurface ecosystem. These biofilms are not passive but can influence mineral weathering and nutrient fluxes, reflecting the capacity of these communities to interact with and modify their local geochemical environment, acting as specialized ecosystem engineers [10, 67, 68].

Biofilm formation by the attached organisms seems mainly driven by chemolithoautotrophic growth, based on CO_2_ fixation via the CBB cycle and oxidation of sulfur and iron compounds (Fig. 8). Reduced sulfur and iron are available to attached microbes from minerals on the rock surfaces [69, 70], and their release might be promoted by microbially mediated dissolution processes [71, 72]. Attached key species like Rhodoferax and Undibacterium are known for chemolithoautotrophic growth on these electron donors [73, 74]. The sticky EPS matrix of the biofilms allows cells to remain in close proximity, promoting efficient nutrient exchange and uptake while reducing loss through groundwater flow [7, 11], thus offering various advantages in the oligotrophic conditions of the aquifer. The high abundance of genes for biofilm degradation furthermore implies that for subsequent heterotrophic colonizers, EPS can act as a source of organic carbon. Such genes for degradative enzymes were significant predictors of activity in planktonic MAGs as well, indicating that scavenging of dissolved biofilm material might play an important role also for the planktonic community (Fig. 8).Fig. 8. Attached versus planktonic microbial lifestyles in carbonate aquifers. In attached communities, microbes drive CO_2_ fixation and EPS formation, harnessing energy from rock-derived minerals. Planktonic microbes rely on groundwater-dissolved nutrients as electron donors for CO_2_ fixation, and heterotrophs might scavenge nutrients from EPS

The autotrophically-driven nature of the aquifer biofilms is likely linked to the oligotrophic conditions present in the groundwater, and is in contrast to observations made in environments with a higher availability of organic carbon: In marine ecosystems, organic particles are first colonized by heterotrophic degraders, and autotrophs might join at a later stage [75]. Similarly, in stream sediments, organic matter recycling and respiration often precede primary production, but the combined activity of autotrophs and heterotrophs can make the system self-sufficient in terms of carbon, if energy is present [4, 76]. As such, autotrophy-driven biofilms might also play a different role in global carbon cycles. The high abundance of 20% attached autotrophs in the groundwater suggests that these biofilms form a sink for carbon in the subsurface. Efforts to assess subsurface CO_2_ fixation so far primarily focused on planktonic communities [19, 77–79], leaving a gap in the global balances. As current methods for assessment of CO_2_ fixation rates are reliant on the extraction of sufficient biomass, they might not be easily adaptable to attached communities. However, given the observed twofold higher relative abundance of autotrophs in the attached compared to the planktonic fraction, as well as the often drastically higher number of attached cells [14, 15], the carbon sequestration by groundwater ecosystems might be substantially higher than previously assumed. The fate of this fixed carbon within attached biofilms remains an open question. Given the slow turnover and physical protection of biofilm matrices in fractured rock aquifers [7, 9], a fraction of the fixed carbon may persist within extracellular polymeric substances or as mineral-associated biomass. While our results indicate distinct attached and planktonic communities, the presence of biofilm degradation functions in planktonic fractions suggests active recycling at the biofilm-water interface. Nevertheless, physical stabilization and slow microbial turnover could enable partial retention of carbon within attached reservoirs over hydrologically relevant timescales. Thus, autotrophic biofilms in the subsurface may contribute to effective carbon sequestration, particularly in low-flow, energy-limited environments where mineral interactions and anoxic conditions favor long-term carbon preservation. Future work combining isotopic tracing and in situ turnover measurements could help constrain the residence times and contribution of these attached reservoirs to the subsurface carbon budget.

Beyond chemolithoautotrophy, the attached community also exhibited broader general capacities for redox transformations of sulfur, iron, and nitrogen. Genes related to dissimilatory sulfite and nitrate reduction, iron reduction, and ammonia oxidation were consistently more abundant in attached MAGs, indicating a capacity for utilizing mineral-derived electron donors and acceptors [10, 80, 81]. These capabilities suggest that attached microbes not only serve as primary producers but also contribute actively to long-term nutrient turnover and geochemical transformations at the rock–water interface. In contrast, in both the total and active planktonic community, the functions investigated were rare and scattered, being present in only 6.8% of MAGs on average. Reduced nitrogen compounds seemed to play the primary role for sustaining the lower fraction of planktonic autotrophs (Fig. 8). Originating from surface inputs and biomass recycling [82, 83], reduced nitrogen might be more available in the groundwater than rock-derived compounds. Overall, the scattered functions in the planktonic community imply that the central functions sustaining it are carried out by only a small fraction of the organisms present, or that it might rely on the activity of the attached community.

When comparing activity and abundance, we found distinct patterns in the two communities. In the attached community, replication activity was present primarily in the most abundant MAGs, whereas low-abundance MAGs were less often active. In contrast, in the planktonic community, active and inactive MAGs showed a more random distribution, with some phyla explicitly showing activity in less abundant MAGs. These patterns indicate a higher dynamic of activity/inactivity in planktonic communities, supporting the argument that attachment supports more stable conditions for microbial growth [84–86]. Key active organisms in the attached community, like Rhodoferax, Aquabacterium, Hydrogenophaga, and Undibacterium, have previously been reported as abundant not only in rock-attached [15, 87], but also in endolithic communities in these carbonate aquifers [88]. They might hence be key contributors to the stability and resilience of the attached community over time, contrasting the more dynamic conditions in the planktonic communities.

A striking feature was also the activity of MAGs from the parasitic Cand. Patescibacterial group JAEDAM01. This group appeared exclusively in the attached community, where it was the only group of Cand. Patescibacteria that showed active replication, indicating that the ecological conditions facilitate their specialized, host-associated lifestyle. Previous studies have shown that JAEDAM01 establish close physical associations with their hosts, often forming multicellular stacks on the host cell surface [89, 90]. Such a strategy may be particularly effective in biofilms, where the spatial proximity of cells enhances opportunities for host contact and transmission between neighboring cells. Moreover, reported hosts of JAEDAM01 belong to autotrophic lineages, which renders the sulfur- and iron-dependent chemolithoautotrophs abundant in the attached community plausible candidates for interaction. The combination of higher host availability and dense spatial structuring may explain why JAEDAM01 was proliferating in attached communities.

In our experiments, we ensured a de novo colonization of the exposed rock material, excluding pre-existing biofilms by using rock material sterilized by autoclaving. The attachment of compositionally and functionally distinct organisms from the co-occurring planktonic community hence raises a key ecological question: Where do these colonizing microbes originate, if they are not abundant in the groundwater? One possibility is that they derive from the rare biosphere, existing at low abundance in the planktonic phase and below detection thresholds [58]. Alternatively, microbes may disperse through episodic detachment from other rock surfaces within the aquifer [60], representing a low-frequency surface–to–surface exchange rather than continuous seeding from the water phase. It is also conceivable that colonizers arise from unsampled microhabitats such as rock pores or interfacial mineral zones that are not captured by standard planktonic sampling methods [68, 80]. Our findings suggest colonization by long-resident, ecologically specialized taxa that persist on mineral surfaces throughout the aquifer. These microbes appear to exhibit low dispersal potential, rarely entering the planktonic phase, and are adapted for surface-associated life, as evidenced by their enriched functional repertoire (e.g., biofilm formation, chemolithoautotrophy), larger MAG sizes, and active replication. Together, these patterns point to a structured, lithic microbiome that operates largely independent of the free-living groundwater community, challenging assumptions of high connectivity and functional redundancy between microbial lifestyles in consolidated aquifers.

This observed lifestyle-driven divergence may carry important evolutionary implications. In spatially structured environments such as consolidated aquifers, physical separation between microenvironments (rock pores, mineral surfaces, the water phase) can act as a barrier to gene flow, facilitating ecological and evolutionary differentiation [91]. Our genome-resolved comparisons of closely related MAGs revealed substantial differences in assembly size and metabolic functions, suggesting that consolidated aquifers host parallel microbial lineages undergoing independent evolutionary trajectories, rather driven by ecological selection than by dispersal. These patterns align with broader evidence that microbial populations in heterogeneous systems evolve along parallel, niche-specific trajectories shaped more by selection than dispersal [92].

The effect of attached vs. planktonic lifestyle as a key determining factor for microbial communities, however, varies greatly across environments. A comparable dissimilarity to our results has been observed in low-porosity granite aquifers, where Proteobacteria dominate attached communities while Cand. Patescibacteria and Desulfobacterota were more abundant in planktonic fractions [93]. In contrast, in highly porous aquifers like sandstone and gravel, attached and planktonic communities tend to be similar: In sandstone aquifers, Proteobacteria and Bacteroidota were identified as the two most abundant groups, constituting up to 70% of both the attached and planktonic communities [94]. Similarly, Proteobacteria, Geothrix, Burkholderiales, and Desulfuromonadaceae were reported in both communities in sand-gravel aquifers [10].

Differences in porosity hence may contribute to microbial differentiation in groundwater: In high porosity aquifers, higher permeability and fluid flow lead to a more uniform nutrient distribution, preventing the establishment of a free-living community distinct from their attached counterparts. Hydrodynamic effects were previously suggested to also drive interactions between attached and planktonic communities in groundwater and stream ecosystems [60, 95, 96]. A recent study using in situ bioreactors in groundwater, conversely, found higher differences between biofilms and planktonic communities in shallow alluvial aquifers with low porosity, but more similarity in anoxic bedrock groundwater [84], attributing this distinction to be related to redox conditions. It can therefore be assumed that more complex interactions between lithology, porosity, and hydrodynamics, as well as redox conditions, drive the differentiation of attached and planktonic communities.

In summary, our findings call for a reevaluation of the seed bank model across aquifer types. While it may explain microbial dispersal in porous, hydrologically connected sediments [59, 97], it does not capture the dynamics of consolidated aquifers, where dispersal is limited and biofilm-based communities are stable and functionally distinct [10, 80]. Given the distinction between attached and planktonic communities, a targeted evaluation of functions in attached aquifer microbiomes is crucial to assess their role in carbon sequestration and biogeochemical cycling on a global level. Understanding such rock-hosted systems will also require different conceptual frameworks that account for lithological structure, hydrodynamics, and physical separation of microenvironments [9, 92], as well as novel methodological approaches to resolve the mechanisms of microbial activity, adaptation, and persistence in spatially structured subsurface environments.

Supplementary Information

Additional file 1. Supplementary_Figures.docx, containing supplementary Figs. 1—4.Additional file 2. Supplementary_Tables.xlsx, containing Supplementary Tables S1—S8.