2.1 Sampling Sites

Biofilm samples were collected from various river sites across England as part of the Environment Agency's routine surveillance monitoring. Samples were gathered using standardized methodologies for sampling diatoms from rivers (Comité Européen de Normalisation 2003; Kelly et al. 2018; Kelly et al. 2020; Taylor et al. 2023). Briefly, biofilm-covered stones were collected in a tray and scrubbed with deionized or tap water using a clean toothbrush. A pipette was used to transfer 5 mL of the biofilm suspension to a 15 mL tube containing 5 mL of RNAlater-based preservative (3.5 M ammonium sulfate, 17 mM sodium citrate, and 13 mM Ethylenediaminetetraacetic acid). The samples were then transported to the laboratory via an overnight courier at a temperature of 5°C ± 3°C and stored frozen at −20°C prior to DNA extraction within 12 months (Warren et al. 2024).

Sample selection was based on two different environmental variables, pH, and conductivity. Samples were grouped in high and low pH and high and low conductivity. These two variables were selected to represent extremes of environmental factors that shape microbial communities and formed either the top or bottom 5% of sample values from the Environment Agency's routine surveillance monitoring in 2021 (Kim et al. 2016; Weigel et al. 2023). The low pH group had a range of pH measured at the site of sampling of pH 4.03–6.30, and the high pH group had a range of pH 8.33–8.50. The low conductivity group had a range of 18.66–44.33 μs/cm, and the high conductivity group had a range of 1945.33–4817.68 μs/cm (see Appendix A: Table A1 for full details). Geographical mapping of sample site distribution was performed in R package ggplot2 v3.5.1 using Map data of the UK from package rnaturalearthdata v 1.0.0. (South et al. 2025). Jittering was applied with a width of 0.1 and a height of 0.1 to reduce overplotting while maintaining the general structure of the data distribution (Appendix B: Figure B1).

2.2 DNA Extraction

The effectiveness of the Eland et al. (2012), Qiagen Blood and Tissue (BT) based method (Kelly et al. 2018) was compared to three popular commercial DNA extraction kits. These kits were chosen based upon a literature review and practical experience and included the Qiagen DNeasy PowerSoil Pro Kit (PS), Thermo MagMAX Microbiome Ultra Nucleic Acid Isolation Kit (MM) and the Zymo Quick-DNA Fecal/Soil Microbe Miniprep Kit (FS). The Eland et al. (2012) method employs both chemical and enzymatic lysis with overnight digestion with proteinase K, whilst the PS, MM, and FS all employ chemical and mechanical lysis, and proteinase K digestion is included with MM and FS kits. To account for sample type, 100 μL of homogenized and resuspended biofilm was used in all extraction procedures. All kits followed the manufacturer's protocol, with the exception of the FS kit, where the protocol was amended to include DNA/RNA shield in place of standard lysis buffer, as experience has shown it to be optimal for this sample type and kit. Further details are provided in Data S1.

2.3 DNA Yield and Quality Evaluation

The presence of DNA was first verified by gel electrophoresis stained with GelRed nucleic acid stain on a 1% agarose gel, run at 85 V for 45 min. DNA was accurately quantified using a Qubit 3 fluorometer (Thermo Fisher, UK) and its companion Qubit 1X dsDNA BR Assay Kit. DNA quality was assessed by UV–VIS spectra determined by Thermo Scientific Nanodrop 8000.

2.4 Quantification of Microbial Abundance via qPCR

See Table 1.

To compare the amount of amplifiable DNA from bacterial, eukaryote, fungal, and phototrophic communities, quantitative polymerase chain reaction (qPCR) was used to amplify target regions of the genes coding for the 16S rRNA (16S), 18S rRNA (18S), the Internal Transcribed Spacer region (ITS), and the ribulose 1,5-bisphosphate carboxylase (rbcL), respectively (see Table 1 and Methods for reaction conditions and methods (Data S1)). Reactions were performed in duplicate, and mean values were used for analysis. For samples where one of the duplicates was unsuccessful (cq≥water controls), or a duplicate was > 2 cq values apart, the successful or highest concentration sample was used in subsequent analysis. While it is acknowledged that this is outside of the normal qPCR acceptance levels, we included these settings as a measure of amplification success without any further purification. If assays were to be used for monitoring, more rigorous cutoffs would need to be applied. Full details of Log copies calculations are given in Data S1.

2.5 Metabarcoding of Biofilm Communities

The full metabarcoding protocol is given in Data S1. To summarize, bacterial, fungal, eukaryotic, and phototrophic community structure was assessed using rarefied sequence abundance of 16S, ITS, 18S, and rbcL, respectively. Library preparation followed a two-step amplification approach, with the first step using Illumina Nextera tagged primers based upon the universal metabarcoding primers outlined in Table 1 and fully described in Data S1. In concurrence with other studies (Kelly et al. 2024), we found that the primers, although targeting diatoms, also amplified additional taxa, and have therefore chosen to refer to the amplified community as phototrophs. The second-step PCR integrated unique custom barcode combinations corresponding to each sample (Kozich et al. 2013). PCR products were normalized using the NGS Normalization 96-Well Kit (Norgen Biotek Corp). Pooled amplicon libraries were vacuum concentrated and gel purified. Resultant libraries were quantified using a Qubit dsDNA HS Assay kit (Invitrogen) and sequenced at a concentration of 7.0 pM with the addition of 10% Illumina PhiX control library. Sequencing was performed on an Illumina MiSeq platform using V2 chemistry (Illumina Inc., CA, USA).

2.6 DNA Sequence Processing

Sequences were processed using DADA2 (Callahan et al. 2016) pipeline in R V.3.0.17 (R Core Team 2012) to quality filter, merge, denoise, and assign taxonomies. 16S amplicon reads were trimmed to 250 and 220 bases, forward and reverse, respectively. ITS2 and 18S amplicon reads were trimmed to 220 and 220 bases, forward and reverse, respectively. rbcL amplicon reads were trimmed to 230 and 210 bases, forward and reverse. Filtering settings were maximum number of Ns (maxN) = 0, maximum number of expected errors (maxEE) = 2,2 (16S, 18S and rbcL) 1,1 (ITS). The primer sequences were removed using trimLeft = c (20). Sequences were dereplicated and the DADA2 core sequence variant inference algorithm was applied. mergePairs was used to merge sequences, and ASV tables were constructed. Chimeric sequences were removed using removeBimeraDenovo default settings. ASVs were subject to taxonomic assignment using assignTaxonomy at default settings; training databases were Silva v138.1 (Quast et al. 2013), Unite v7.2 (Koljalg et al. 2005), PR2 V 4.14.1 (Guillou et al. 2013) and diat.barcode (Rimet et al. 2019) for 16S, ITS, 18S, and rbcL, respectively.

After quality filtering, a total of 4,979,362 bacterial (16S), 3,525,207 fungal (ITS2), 9,045,927 eukaryotic (18S), and 9,399,909 phototrophic community (rbcL) sequences were used in the analysis. To account for the effect of sequencing depth bias, the resultant ASV tables were rarefied to an even depth of 19,821 (16S), 12,229 (ITS2), 11,598 (18S), and 8839 (rbcL), based on the sample with the lowest number of reads in each experiment after the rarefaction curve asymptote was reached. After rarefaction, samples were phylotyped to the genus level using the tax_glom command and with the NArm = FALSE setting enabled, resulting in 1524 (16S), 1331 (18S), 1032 (ITS2), and 329 (rbcL) genera-level phylotypes.

2.7 Statistical Analysis

All post sequencing sample quality filtering were performed in R v2.2, R Studio v 2022.02.2 (R Core Team 2012; RStudio Team 2020), with and package Phyloseq v1.48.0 (McMurdie and Holmes 2013). Data were visualized through packages; Phyloseq v1.48.0 (McMurdie and Holmes 2013), DESeq2 v 1.44.0 (Love et al. 2014) GGbreak V0.1.2 (Xu et al. 2021), microeco v 1.6.0 (Liu et al. 2021), and GGplot2 v3.5.1 (Wickham 2009). The following statistical analyses were performed in R v2.2, R Studio v 2022.02.2, as the following: Kruskal–Wallis and Dunn's tests (package rstatix v 0.7.2, Kassambara 2023), Kolmogorov–Smirnov tests (package dgof v1.4, Arnold and Emerson 2011), R² values for Regression of relative abundance comparisons by kit were determined through linear modeling using lm function (package stats version 3.6.2, R Core Team 2012), PERMANOVA was performed on NMDS of Bray–Curtis distance using the Adonis 2 command (package vegan 2.6–4, Oksanen et al. 2024), To determine which community members are most impacted by different extraction conditions, differential expression analysis was performed. Differential expression analysis based on the Negative Binomial (a.k.a.Gamma-Poisson) distribution, was performed upon phylotyped genera, with significance determined through hypothesis testing performed using Wald test was performed in DESeq2 v 1.44.0 (Love et al. 2014).

3.1 DNA Quantity and Quality

DNA quantity and quality were determined as a primary measure of kit success. Values for total nucleic acid (NA) concentration assessed through UV–Vis spectrophotometry were higher than that of fluorometry (Qubit), suggesting that the BT method had the highest total NA concentration, mirroring gel electrophoresis. However, these methodologies are known to be far less accurate than fluorometry and skewed by the presence of free nucleotides, RNA, residual protein, and environmental contaminants such as humic acids (Li et al. 2014; Paul et al. 2021). Therefore, fluorometry was performed to determine dsDNA extraction success. This was highest in the Fecal Soil (FS) and Blood and Tissue (BT) kits with concentrations of dsDNA within kit detection range (21 and 20 samples levels, respectively), whereas the MagMAX (MM) and Power Soil (PS) kits performed worse in comparison (18 and 15 samples with high concentrations of dsDNA). When input biofilm and lysate volume was accounted for (Figure 1A), there was a significant difference in concentration recovered (Kruskal–Wallis p = 0.006). The highest median concentration of dsDNA across all kits was from the FS kit. However, between kits, only the differences between FS and MM were significant (Dunn's test p = 0.003) and this is likely skewed by high concentration samples.

When the 260/280 ratio determined with UV–Vis spectrophotometry was used to assess sample quality (Figure 1B, and Appendix C: Table C1), there was a highly significant difference across kits (Kruskal–Wallis p = 1.2e⁻¹⁰). When comparing between kit difference (Dunn's test) significantly lower quality was found in the MM and PS kits, when compared to the BT (MM p = 3.3e⁻⁷, PS p = 8.6e⁻⁵) and FS (MM p = 2.8e⁻⁷, PS p = 7.4e⁻⁵) kits in the median values. However, very few of the samples reached the accepted quality standard of > 1.8. This low 260/280 ratio is common in environmental samples and is likely due to environmental contaminants or extraction kit carryover. This was particularly pronounced in the MM kit where visible magnetic bead carryover was present in the final elution steps despite careful laboratory handling.

3.2 Microbial Community Quantification (qPCR)

Key community members were quantified through representative genes using qPCR (Figure 2). Significant differences in log copies amplified across kits were detected using the Kruskal–Wallis test irrespective of target gene (16S p = 1.57e⁻⁸, ITS p = 2.71e⁻⁵, 18S p = 0.004, and rbcL p = 0.02). Pairwise kit comparisons (Dunns test, Appendix C) suggested that significantly higher gene copies were recovered from kits that employed mechanical lysis when compared to the BT kit, although significance did vary by target amplicon. However, amplification success was more variable for 16S, ITS, and rbcL using the BT kit without mechanical lysis compared to mechanical lysis kits optimized for environmental microbes (MM, PS, and FS); as such, when accounting for spread, none of these differences were significant when using the Kolmogorov–Smirnov test (p ≤ 0.05). In terms of community prevalence, there were higher mean Log Copies of 16S compared to all other genes irrespective of kit. This would indicate prokaryotes are more abundant than eukaryotes in the biofilm communities. The lowest Log Copies were found in the rbcL assay.

3.3 Richness and Diversity

To investigate community richness across sample extraction kits, phylotype distribution was compared for each amplicon using Venn diagrams, whereby regions of overlap represent the count and percentage of phylotypes shared (Figure 3). Greater than 96% of all ASVs from each community were detected in all extraction kits, indicating the communities were broadly similar irrespective of extraction methodology. The next highest percentage of shared ASVs was unique to the mechanical lysis kits (16S = 0.7%, 18S = 1.9%, ITS = 2.2% and rbcL = 0.4%), suggesting that up to 2% of richness would not be detected without using mechanical lysis.

When abundance and distribution of phylotypes were accounted for, minimal observable differences in community α diversity (as measured by Shannon–Weiner index) are present (Appendix D: Figure D1). Similar to the qPCR samples, the chemical lysis only method (BT) had lower levels in bacterial (16S) and fungal (ITS) communities. However, none of the differences were significant when using the Kolmogorov–Smirnov test (p ≤ 0.05). The highest mean diversity was seen within the bacterial (16S) communities, irrespective of the DNA extraction kit used.

3.4 Microbial Community Taxonomic Composition

As biofilms comprise multikingdom microbial communities, it was important to ensure that no taxonomic bias was imposed through kit choice. We therefore examined the impact of the lysis method upon the key constituent communities (Prokaryotes, Eukaryotes, Fungi, and Phototrophs) separately. When using sequence references, levels of taxonomic classification and nomenclature vary between taxonomic grouping; therefore, we used the terminology given by the related sequence identification database (Silva v138.1, Pr2 v4.14.1, UNITE v7.2, and diat.barcode v11.1) (Koljalg et al. 2005; Quast et al. 2013; Guillou et al. 2013; Rimet et al. 2019).

The communities detected through metabarcoding are consistent with the complex community structures described previously (Fechner et al. 2010; Besemer 2015; Romero et al. 2020; Li et al. 2023). Bacterial/Archaeal communities are dominated by known phototrophic sequence signatures (Eukaryotic chloroplasts and Cyanobacteria) and contained chemoheterotrophic taxa such as Burkholderiales, Rhizobiales, Cytophagales, and Rhodobacteriales (Figure 4A).

The ITS assay detected multiple fungal groups (the usual target for this assay) and numerous algal groups, particularly Chlorophyceae and Ulvophyceae (Figure 4B). The most dominant fungal group detected belonged to the Dothideomycetes. This fungal class is one of the largest and most diverse groups and is a common saprobe (feeding on dead or decaying organic matter) of freshwater systems (Krauss et al. 2011).

Eukaryote 18S community composition detected through metabarcoding (Figure 4C) was again (Besemer 2015) dominated by sequence signatures from phototrophic orders such as Zygnemophyceae, Chlamydomonas, Bacillariophyta, and Sphaeropleales. Heterotrophic fungal orders such as Pezizomycotina and Chytridiomycotina and grazers such as members of Annelida, Gastropoda, and Crustacea were also present. Overall, class-level eukaryotic composition was largely consistent within samples from the same source, irrespective of extraction kit, but varied much more between samples than that of Prokaryotes.

Of all the sequencing assays, the poorest sample amplification was seen in the rbcL assay (Figure 4D). Here, the diatom selective primers did favor members of the Bacillariophyceae, as they showed a higher proportionate contribution to the phototrophic community than with the 18S primer set. Yet, it should be noted that this primer set was the only one of the four where none of the extraction methods were able to successfully sequence all 23 samples after quality thresholds were applied.

3.5 Variation Between Methods

Nonmetric multidimensional scaling (NMDS) using Bray–Curtis distance was used to visualize the relationships between communities (Figure 5). For 16S, samples cluster significantly (p = 0.001, df = 22, PERMANOVA) by sample type rather than by kit (p = 0.999, 3 df, PERMANOVA), indicating that the kit type played a relatively minor role in determining community composition (Figure 5A). This pattern continues to hold true for the 18S (Figure 5B), ITS (Figure 5C) and rbcL assays (Figure 5D) whereby samples from the same biofilm sample cluster significantly (all p = 0.001, df = 22) rather than NMDS loadings displaying any impact of extraction kit type (p = 1, p = 1 and p = 0.998).

3.6 Differences in Relative Abundance of Community Members

The impact of extraction methodology on genera relative abundance was visualized through linear regression (Appendix E: Figures E1-E4). The resultant R² values are shown in Table 2.

When examining the Bacterial/Archaea communities, no apparent differences were observed between kits, as demonstrated by the similar R² values (0.90–0.98). However, when examining the total eukaryotic community, there was a higher similarity between kits that employed a mechanical lysis step R² (0.73–0.83) than the BT kit, which employed enzymatic and chemical lysis alone (R² 0.56–0.61). This pattern was repeated in both the fungal ITS (R² 0.85–0.95 compared to R² 0.56–0.61) and phototrophic rbcL (R² 0.90–0.94 compared to R² 0.53–0.63). It should be noted, however, that in general, the goodness of fit (R²) values in this analysis are high, showing a strong association between individual samples irrespective of the extraction method employed.

3.7 Differential Abundance of Community Members

Differential abundance analysis was performed (Appendix F: Figures F1-F4). Table 3 summarizes the results from these analyses. For brevity, only taxa that are significantly impacted in at least three of the six (BT/MM, BT/PS, BT/FS, MM/PS, MM/FS, and PS/FS) comparisons are shown, and the mean values of these comparisons are reported. Of the 4216 genera-level phylotypes identified, the abundance of only 19 genera was significantly different.

In three or more kit comparisons, there was a clear distinction between kits that employed mechanical lysis and those that did not. In fact, no other comparison groupings met our cutoff of greater than three, again suggesting that kits employing additional mechanical lysis produced highly similar community profiles when compared to enzymatic lysis alone. More specifically, within the bacterial populations, the majority of significantly affected taxa were more abundant in kits employing mechanical lysis, most of which belong to known spore-forming groups. Only the taxa identified as the cyanobacterial genus Halospirulina showed greater abundance in the traditional biofilm extraction technique. This pattern is mirrored in the fungal ITS and rbcL amplicons, where all affected taxa were of significantly higher abundance in mechanical lysis kits. Interestingly, this pattern reversed when studying the 18S amplicon library, with impacted communities being largely more abundant in samples extracted using the traditional biofilm extraction methodology. However, it should be noted that, except for Melosira and Monostroma, the overall community abundance of the identified taxa was low and unlikely to be core community members.