Six contrasting European sites (two Mediterranean marine sites – Spain, Croatia; four European river sites – France, Spain, Germany, Finland) were selected for sampling, based on differences in water quality and typology (Table 1). Sampling at all sites was done on the same day (18 September 2017).

Freshwater phytobenthos was sampled from biofilms following the European standard (CEN 2014a) at each of the four river sites (OF, MS, EG, HF). The resulting biofilm suspension was transferred to a sterilised 1 litre bottle that was stored in a cool box with ice packs for immediate express shipment within 1 day to INRAE lab (Thonon, France) (Fig. 1). After the four river samples had been received, they were left to settle for three hours at 5 °C and concentrated by removing water supernatant to a final volume of around 900 ml of suspended biofilm per sample. Samples were then homogenised and subsampled into three sterilised bottles, one bottle of 300 ml per preservation method.

Marine phytoplankton was sampled by one vertical net haul at both marine sites (LC and ES) with a phytoplankton net (50 μm mesh size) from 15 m deep to the surface. Each net sample was suspended and evenly filtered until complete filter saturation (30 ml per filter for station LC and 60 ml per filter for station ES), on 1.2 µm cellulose (Millipore) (LC site) or GF/F glass microfibre filters (Whatman) (ES site) (Fig. 1). Net samples were well-homogenised during filtration to ensure an even distribution of the sample on each filter. For each site, 18 filters were obtained in total and were placed in marked tubes and stored in a cool box with ice packs for immediate express shipment within one day to the Center for Marine Research (CMR) lab (Rovinj, Croatia). All filters were cut in half, resulting in 12 half-filter subsamples per site and per preservation method (see details below).

Three preservation methods were applied to phytoplankton and phytobenthos samples (Table 2, Fig. 1): 1) deep-frozen (hereafter “FR”) (Visco et al. 2015), 2) preservation with a nucleic home-made acid preservative solution, based on RNAlater storage solution (Merck, Kenilworth, USA) (hereafter “RL”, standing for “RNAlater” style) (Kelly et al. 2018) and 3) preservation with ethanol (hereafter “ET”) (Vasselon et al. 2017a).

FR preservation method : For freshwater samples, twelve 2 ml subsamples of the biofilm suspension were obtained from one 300 ml bottle under agitation for each site. Subsamples were then centrifuged, supernatant was discarded and pellets were frozen and stored at -20 °C (Suppl. material 1: Data 1C and D). For marine samples, all 12 half-filters were frozen and stored in tubes at -80 °C for each site.

RL preservation method : A nucleic acid preservation solution was home-made with 3.5 M ammonium sulphate, 17 mM sodium citrate and 13 mM ethylene-diamine-tetra-acetic acid (EDTA). pH was adjusted to 5.2 using 1 M H₂SO₄ and the solution was sterilised by filtration with 0.2 µm filter. For freshwater samples, one volume of the nucleic acid preservative solution was added to one volume of sampled biofilm, for one 300 ml bottle under agitation. 24 × 2 ml subsamples of the preserved biofilm suspension were then stored for each site (Suppl. material 1: Data 1C and D). For marine samples, 2 ml of the preservative solution was added in the 12 tubes with half-filters for each site. All samples were then stored at -20 °C.

ET preservation method : For freshwater samples, three volumes of 96% ethanol were added to one volume of biofilm, in order to obtain a final ethanol concentration of 70%. This was applied to one 300 ml bottle under agitation. Six 17 ml subsamples of the preserved biofilm suspension were then stored for each site (Suppl. material 1: Data 1C and D). For marine samples, 2 ml of 96% ethanol were added in 12 tubes with half-filters. All samples were then stored in the dark at +4 °C.

In all subsampling phases for freshwater biofilm samples (Suppl. material 1: Data 1C), special attention was paid to homogenisation by: 1) permanent agitation of the solution to subsample and 2) sequential subsampling, adding solution to all subsamples in succession, 1 ml per 1 ml.

The samples, preserved with the three methods, were further processed at six different storage durations (1 day, 1 week, 1 month, 3 months, 6 months and 1 year) during one year (Fig. 1). For each duration, two replicates were retrieved per preservation method and per site (i.e. 36 samples) and independently processed for DNA extraction.

For freshwater samples, DNA extraction was performed on biofilm pellets, either those directly preserved (FR samples, Table 2, Suppl. material 1: Data 1) or those obtained by centrifugation just before DNA extraction (RL and ET samples, Table 2, Fig. 1). In order to minimise the dilution effect of RL and ET methods (respectively ½ and ¼), DNA extraction was processed on the same total amount of 2 ml biofilm pellets for each preservation method (1 pellet for FR, 2 pooled pellets for RL and 4 pooled pellets for ET). For marine samples, wet half-filters were used directly for DNA extraction.

DNA extractions were performed using a commercial kit (Macherey–Nagel NucleoSpin Soil kit, Düren Germany) with purification columns following Vautier et al. (2020). In short, biofilm pellets and phytoplankton filters were re-suspended in lysis buffer and mechanically disrupted using ceramic beads. After proteins and undissolved sample material precipitation, supernatant with dissolved DNA was first passed through inhibitor removal columns and next through NucleoSpin Soil columns for DNA binding where PCR inhibitors were removed by efficient washing. Finally, DNA was eluted in the NucleoSpin Soil elution buffer (Tris/HCl buffer) and stored at -20 °C prior to PCR amplification and sequencing (Fig. 1). DNA extractions were processed at INRAE lab (Thonon, France) for freshwater samples and at CMR lab (Rovinj, Croatia) for marine samples. All DNA extracts from marine samples were then sent with express shipment to INRAE lab for downstream analysis at the end of the 1-year period. In total, for this 1-year experiment, 216 DNA extracts were obtained (2 replicates × 6 storage durations × 3 preservation methods × 6 sites).

At the end of the 1-year storage period, DNA quality and quantity were assessed on all 216 DNA extracts (Fig. 1). To evaluate the DNA quality, the 260/280 nm ratio was measured by spectrophotometry with a Nanodrop ND-1000 (Nanodrop Technologies, Wilmington, Delaware). To evaluate the DNA quantity, DNA concentration (ng/µl) was measured for each of the two DNA extract replicates with the Quant-iTTM PicoGreen dsDNA assay kit (Life Technologies, Carlsbad, California) using a microplate reader (Fluoroskan AscentTM FL; Thermo Scientific, Waltham, Massachusetts) following the manufacturer’s instructions. The mean concentration value of the two replicates was used in subsequent analyses.

A 312 bp fragment of the rbcL chloroplastic gene was amplified from DNA extracts using Takara LA Taq polymerase and an equimolar mix of the forward primers Diat_rbcL_708F_1, 708F_2, 708F_3 and the reverse primers R3_1, R3_2 (Vasselon et al. 2017a), following the protocol of Chardon et al. (2020). For each DNA sample, a single step PCR amplification was performed in triplicate in a final volume of 25 μl. After validating PCR amplification, 19 µl of each PCR product per sample were pooled together and 50 µl of this pool were transferred into an individual well of a 96-well microplate. The resulting three plates with 216 samples were sent to “GenoToul Genomics and Transcriptomics” (GeT-PlaGe, Auzeville, France) where library preparation, final library pool and the sequencing with Illumina MiSeq System with paired-end sequencing kit (V2, 250 bp × 2) were performed (Fig. 1).

Demultiplexing and a quality check (FastQC, Andrews 2010) was performed by the sequencing service and two fastq files (forward and reverse) were provided for each sample (2 × 216 fastq files in total). Fastq files were processed using Mothur version 1.43.0 software (Schloss et al. 2009). Contigs were made by merging forward and reverse reads (make.contigs) trimmed to only the overlapping section (trimoverlap = T). Reads in every fastq file were then quality filtered (screen.seqs) by excluding sequences with an overlap shorter than 180 bp (minoverlap = 180) and all sequences with ambiguities (maxambig = 0) and mismatches (mismatches = 0). From the resulting good quality reads, exact duplicates were removed (dereplication) (unique.seqs). Subsequent bioinformatics steps on good quality reads included alignment (align.seqs) of these sequences to the reference alignment (Diat.barcode v.7 reference sequence library) and preclustering (pre.cluster) that enabled de-noising of sequences (one difference for every 100 bp of sequence was allowed). Chimera removal (chimera.vsearch) was done using the VSEARCH algorithm with default parameters (de novo chimera detection). Sequence classification (classify.seqs) was made using the naïve Bayesian Method (Wang et al. 2007) with bootstrap confidence score set to 85% and Diat.barcode v.7 as the reference sequence library (Rimet et al. 2016, 2019). Sequences classified to taxa other than diatoms (Bacillariophyta) were removed from further processing (remove.lineage). Operational Taxonomic Units (OTUs) were defined by calculating distances between sequences (dist.seqs) and clustering (cluster) these distances, based on 0.05 difference cut-off (95% similarity) implementing the furthest neighbour algorithm. OTUs containing one single sequence (singletons) were removed. Significant correlation between replicates was detected for the number of reads (Spearman R = 0.71, p < 0.001) and the number of OTUs (R = 0.93, p < 0.001). Sample replicates were merged (merge.groups) and replicate reads were summed to retain read abundances. Using the classify.otu command, consensus taxonomy for each OTU was assigned with an 85% confidence threshold using previously classified reads taxonomy. A list of classified OTUs and their read abundances in each sample was produced. OTUs with identical highest taxonomy level were merged (merge.otus) to obtain the list of different taxa detected in the dataset. Random subsampling was performed to normalise the data with size (number of reads) set to the size of the smallest sample (14,190 reads).

Statistical analyses, as well as graphical presentations of the results, were performed using the R software version 3.6.0 (R Core Team 2017). Sequencing data import and manipulations in R were conducted with the phyloseq package. Graphical presentations were produced using ggplot2. T-tests and ANOVA were used to test for differences in DNA concentration and diatom assemblage diversity among different storage durations and preservation methods. Spearman’s rho statistic was used to estimate a rank-based measure of association in correlation analyses. Patterns of sample dissimilarity were visualised using unconstrained ordinations of non-metric multidimensional scaling (NMDS). Ordinations were based on taxa presence/absence and abundance using Jaccard and Bray-Curtis indices. Further tests of the differences in the diatom assemblages used permutational multivariate ANOVA (PERMANOVA) on the normalised dataset with the adonis function in vegan version 2.4.2 (Oksanen et al. 2018) with 9999 permutations. PERMDIST with the betadisper function, also in vegan, was used to test multivariate homogeneity of group dispersions (variances). Pairwise differences within groups were determined by pairwise Adonis function (Martinez Arbizu 2020) in vegan. Differences in community composition were visualised using Venn diagrams (vegan) and taxa lists for each sample were produced with merge.otus. Finally, the presence/absence of dominant and rare (< 100 reads) taxa for different storage durations and preservation methods were analysed.

For freshwater river sites, we assessed their ecological quality using the Specific Pollution-sensitivity Index (SPI) (Cemagref 1982), based on species inventories (species composition and relative abundances, based on read numbers; or genus, if species level was not reached) obtained by metabarcoding. SPI values were calculated using the OMNIDIA 5 software (Lecointe et al. 1993).

Fastq files are available at https://doi.org/10.57745/M3PESU.

Spectrophotometry measurements confirmed good DNA quality with 260/280 nm ratios between 1.8 and 2 for all samples. Measured DNA concentrations differed among samples and ranged from 1 to 160 ng/µl (Fig. 2) with marine samples having significantly lower concentrations (mean value 7.1 ng/µl, Welch’s two-sample t-test, p < 2.2e-16) than the freshwater ones (mean value 65.8 ng/µl). The preservation method had an effect on DNA concentration of freshwater samples, but not on that of marine samples (F_2,33 = 0.44, p > 0.05). Freshwater samples stored in ethanol (ET) had significantly lower DNA concentrations (F_2,69 = 28.86, p < 0.001) compared to the other two types of preserved samples (FR, RL). Storage duration had no effect on DNA concentration for either the marine samples (F_5,30 = 0.78, p > 0.05) or the freshwater samples stored in FR and RL (Fig. 2). However, for freshwater samples stored in ET, a significant decrease in DNA concentration was observed after 3 months (F_5,18 = 7.55, p < 0.001).

Figure 2.

DNA concentrations (mean values of 2 replicates) over time (x axis: 1D – 1 day; 1W – 1 week; 1M – 1 month) for the three preservation methods (blue: ET, yellow: FR, grey: RL) for marine sites (top row) and freshwater sites (middle and last rows).

High-throughput sequencing of the 216 samples resulted in a total of 7.9 million reads. Only one sample (Ebro bay-RL-1 month) could not be sequenced successfully.

After bioinformatics processing, a total of 3.9 million (49.4%) reads were conserved with an average of 36,621 reads per sample. Read clustering (95% sequence similarity threshold) resulted in an average of 357 OTUs per sample and classification (85% bootstrap confidence score threshold) identified an average of 97 taxa per sample. Rarefaction curves indicated sufficient sequencing depth for most of the samples (Suppl. material 2: Data 2).

When all preservation methods and storage durations were considered, freshwater sites were, on average, characterised with higher number of reads (41,222 reads/sample), higher OTU and taxa richness (433 OTUs/sample and 111 taxa/sample) and higher diversity index (Shannon) values compared to marine sites (27,155 reads/sample, 199 OTUs/sample and 67 taxa/sample) (Fig. 3). The average number of OTUs and taxa across all sites is very similar for all three different preservation methods. Preservation methods had no significant impact on read numbers (F_2,104 = 0.533, p > 0.05), OTU (F_2,104 = 0.413, p > 0.05) and taxa (F_2,104 = 0.436, p > 0.05) richness and Shannon index values (F_2,104 = 0.439, p > 0.05) (Fig. 3). These diversity parameters did not change significantly over time (read numbers (F_5,101 = 0.38, p > 0.05), OTU (F_5,101 = 0.069, p > 0.05) and taxa (F_5,101 = 0.129, p > 0.05) richness and Shannon index values (F_5,101 = 0.025, p > 0.05)).

Figure 3.

Diversity parameters of diatom assemblages: box plots for (a) read numbers, (b) Shannon index, (c) number of OTUs and (d) number of taxa for the three preservation methods (blue: ET, yellow: FR, grey: RL) for marine (LC, ES) and freshwater sites (OF, MS, EG, HF). Boxes represent the interquartile range, with the median indicated with a line and whiskers extending to the highest and lowest values.

Using the Diat.barcode reference library, 289 OTUs were assigned at species level, 77 at genus level, 21 at family level, nine at order level and two at class level. Overall, 102 different diatom genera were detected in the dataset. The diatom assemblage composition differed among sites (Suppl. material 3: Data 3), reflecting their environmental characteristics (freshwater vs. marine). The genus Nitzschia was the most abundant in terms of read numbers and included the highest number of species (28). About 50% of the 102 genera were represented by a single species in the dataset.

Do preservation methods and/or storage duration affect assemblage structure ?

Community ordination analyses taking read abundance into account (Bray-Curtis distance) showed that the samples differed primarily according to sampling sites (PERMANOVA, pseudoF_5,101 = 573.08, R2 = 0.96, p = 0.001) (Fig. 4). Sampling sites significantly influenced the community structure when presence/absence was considered (Jaccard distance: PERMANOVA, pseudoF_5,101 = 67.45, R2 = 0.9076, p < 0.001). Since sampling sites were very different (Bray-Curtis and Jaccard distance R2 > 0.90), the effects of the preservation method and storage time on community structure were tested with site-by-site analyses. Permutation tests for homogeneity of multivariate dispersions confirmed significant differences in dispersion (p < 0.001) between sampling sites, but not between methods (p = 0.786) and time (p = 0.975).

Figure 4.

Non-metric multidimensional scaling (NMDS) ordinations for all sites (above) and for each site (next page), based on Bray-Curtis distances, taking read abundance into account. Samples are marked according to the preservation method (colour) and storage duration (shape). The three preservation methods (ET, FR and RL) are visualised by ellipses for the six sites (next page).

Differences in assemblage composition between different methods can be seen in NMDS plots for each site (Fig. 4). Site-by-site analysis of read abundance (Bray-Curtis distance) and taxa presence/absence (Jaccard distance) showed that the preservation method had a significant effect (PERMANOVA, p = 0.001 for both distance matrix) on diatom assemblage structure at all sampling sites, while storage duration did not (PERMANOVA, p > 0.3 for both distance matrices). The preservation method explained on average 64% of the total variance in distance between samples, while the storage duration explained on average 11% (Table 3). Pairwise comparisons of Bray–Curtis dissimilarity between different methods (Table 4) for each site indicate significant differences in assemblages for all comparisons (p < 0.05), except FR-RL for sites MS and HF and ET-RL for site LC.

Are some taxa differentially detected ?

Assemblage changes are mainly due to changes in relative abundances for abundant taxa (Suppl. material 3: Data 3) and to changes in presence-absence for low abundant taxa (not shown). There is no significant difference in the number of taxa detected between the preservation methods (ANOVA, F_2,104 = 0.436, p > 0.05) and the overall number of taxa detected for each method is around 300 per sample. In the dataset, 81% of detected taxa were shared by all the three methods, while each method is characterised with less than 5% of taxa detected only by one method (ET: 11 taxa, RL: 13 taxa, FR: 5 taxa, Fig. 5). Taxa that are unique to one method are phylogenetically diverse and do not share evident ecological characteristics (cell size and shape, colony formation, habitat preference etc.). Taxa with < 100 reads were also mostly common to all three methods (36%, Fig. 5). A total of 75% of the diatom assemblage was shared in the dataset irrespective of storage durations, with those taxa unique to each duration mostly of low abundance (less than 100 reads). Only one species with > 100 reads was unique to the shortest duration (1 day) and the RL method: Actinoptychus splendens, which is a marine species detected only at the ES site (Ebro Bay, Spain). Rare taxa were often method-specific and usually appeared and disappeared over time without any obvious pattern.

Figure 5.

Comparison of the number of diatom taxa shared by the three preservation methods (ET: blue, RL: yellow, and FR: green). Rare taxa (less than 100 reads in the dataset) are presented in red.

SPI scores were calculated for freshwater sites, based on OTUs assigned at species (73%) or genus (19%) levels and their read abundances. They ranged from 14.2–18.9 (Fig. 6). Eutrophic (EG) and humic (HF) freshwater sites had lower mean SPI values, 14.52 and 14.90, respectively. Mesotrophic (MS) and oligotrophic (OF) sites had higher mean SPI values, 17.51 and 18.23, respectively. The influence of sampling site on SPI values was significant (ANOVA, F_3,68 = 548.8, p < 0.001). Most importantly, at each site, SPI values were very stable regardless of the preservation method (ANOVA, F_2,69 = 0.039, p > 0.05) and the storage duration (ANOVA, F_5,66 = 0.006, p > 0.05) and small variations did not translate into quality class changes.

Figure 6.

SPI (Specific pollution-sensitivity index) index values over the three preservation methods (ET: blue, FR: yellow, RL: grey) and the six storage durations (x axis).

When metabarcoding is used to assess biodiversity or ecological quality indices, based on diatom assemblages, our results show an overall robustness of the approach that is only slightly affected by the method used to preserve the samples or by the storage duration. Overall, diatom assemblage composition differed among sampling sites, rather than due to preservation methods or storage duration. Detecting an important impact of sampling site on assemblage composition is not surprising since sites were chosen to represent very diverse environments with various trophic status. Diatoms are known to have specific ecological preferences; thus, their assemblages are shaped by local environmental properties. This is the reason why these assemblages are used as proxies of phytobenthos when monitoring the ecological status of waterbodies for the WFD (Rimet 2011). In our study, little effect of preservation methods and storage duration was observed on the final end-points delivered by metabarcoding: assemblage composition, diversity indices and SPI values. Although the water chemistry was different from one site to another, this did not have an impact on preservation methods. In all cases, the site effect appears to predominate, based on their diatom assemblage structure. The longest storage duration tested here was one year after sampling, which is a valid cut-off point in the context of a monitoring program where results are awaited each year. As no significant changes have been observed throughout this period for any end-points, it suggests that longer storage may also not affect results. This was observed by the authors in previous studies where samples were stored for longer periods of time and still provided results consistent with the morphotaxonomy results for RL (Kelly et al. 2020) and for ET (Bailet et al. 2019; Kahlert et al. 2021). When evaluating future needs, managers would be mainly concerned with storage contingencies (e.g. space in freezers or cold-rooms) and need to know whether raw samples or DNA extracts have to be stored, both of which would require further experiments.

In most cases, the preservation methods we explored did not affect the quantity and quality of the DNA extracted from preserved samples. The exception is the ET method applied to freshwater samples. Preservation with ethanol seems to lead to lower DNA yield than other methods. Ethanol acts both as a killing and a preservative agent, replacing water molecules in biological tissues (Carter 2003) and has been successfully used for macroinvertebrate specimen preservation for biomonitoring (Stein et al. 2013). A minimum of 70% ethanol is necessary to ensure the fixation of the samples, preventing their degradation through time due to biotic processes. To attain that, one volume of suspended biofilms needs the addition of three volumes of 96% ethanol. For the two other methods, dilution was reduced (1/2 for RL method) or absent (FR method). However, this higher dilution in ethanol (1/4), compared to RL and FR methods, was compensated by more material (4 biofilm pellets) for DNA extraction, instead of two or one, for RL and FR, respectively. Therefore, different dilutions should not impact on the final performance of the DNA extraction. In any case, these DNA-based methods should be compared with the usual practices of morphotaxonomy which require only a small subsample of suspended biofilm, to prepare one slide on which 400 valves are identified to establish species relative abundances in the sample. Thus, a considerably smaller proportion of the sample is used for current monitoring.

Recent studies on macroinvertebrate have shown that organismal DNA is released from cells into the ethanol used for preservation during sample storage (Martins et al. 2019; Zizka et al. 2019), confirming the replacement of water molecules by ethanol in biological tissues (Carter 2003). The DNA released in ethanol has been suggested and tested as an efficient method to characterise macroinvertebrate communities (Hajibabaei et al. 2012). In our case, some DNA may have also been released from diatom cells and consequently not included in the centrifuged sample that is used for DNA extraction, thus limiting the initial DNA yield. This may have also been facilitated by the absence of a freezing step in the ET method. However, samples conserved at 4 °C will be more prone to abiotic degradations than frozen ones. Finally, biofilms are complex samples that include many organisms (mainly bacteria, other microalgae, fungi) embedded in an exopolysaccharide matrix (Flemming and Wingender 2010) and even hosting eDNA from external organisms (Rivera et al. 2021a, 2021b). The extracted DNA is, thus, a mix of DNA from different organisms. Differential DNA release may occur during the preservation phase in ethanol, potentially adding one more dilution for diatoms.

Sample storage duration, from one day to one year, does not affect the quantity and quality of the DNA extracted from preserved samples, except when preserving freshwater samples with ethanol. In that case, a decrease of DNA concentration was observed that was marked during the three first months of preservation, but this trend did not continue over the subsequent nine months. We can hypothesise that this decrease is linked to the release of DNA from the cells to the ethanol solution. This could be further evaluated by extracting DNA from the pellets and DNA from the ethanol in parallel.

For all methods and dates, even in the “worst case” of ethanol preservation for the samples that have been stored the longest, the final end-points were not affected. Indeed, the assemblage composition is largely homogeneous at each site, whatever the method and the storage duration. The small percentage of taxa that differ from one method to the other or from one date to the other are among those that are rare (< 100 reads). When diatom metabarcoding is dedicated to the evaluation of ecological status to compare changes in assemblage structure through time and space, which is currently its main application, it is definitely an approach that is not affected by the sample conservation. If diatom metabarcoding is dedicated to the detection of rare species (e.g. invasive, endangered or toxic species), then the choice of the sample preservation method may be more critical. However, in our study, we could not identify a specific trend and derive best practice which is in line with previous observations that rare OTUs may be random and are poorly reproducible (e.g. Leray and Knowlton 2017). For that purpose, a higher sequencing depth is probably required to avoid overlooking rare species and needs to be calibrated appropriately. Moreover, in such cases, a specific study design (e.g. biological replicates, positive and negative controls) and biomolecular methods focusing on the target species may be more adapted (e.g. dPCR, qPCR).

Following results from numerous pilot studies, we can be confident that diatom metabarcoding is robust and can replace or complement the current approach, based on morphotaxonomy. To do so, stakeholders call for guidelines and/or standards to accompany the deployment of the method for biomonitoring purposes (Blancher et al. 2022). General guidelines have been recently published, providing best-practices for many applications of DNA metabarcoding (Bruce et al. 2021; Pawlowski et al. 2021), representing an important milestone. However, guidelines need to be even more precise to be easily and reliably handled by the operators in a variety of contexts. Diatom metabarcoding includes several steps which require different skills (e.g. field sampling, molecular analysis, bioinformatics, ecology). For that reason, a step-by-step standardisation would be necessary to: 1) enable various actors to be involved and 2) leave the door open to future technological changes and evolutions that are numerous in the biomolecular sector.

However, methods have to remain operational and, as far as possible, user-friendly. Concerning sample conservation, depending on the context, one method or the other maybe more adapted. Methods requiring freezing or deep-freezing conservation (FR and RL) imply immediate storage and have to avoid multiple freeze-thaw cycles. To be usable, they require fast access to -80 °C or -20 °C frozen facilities. For organisations that have to organise extensive field campaigns or to access adverse environments, without access to laboratory facilities for several days, the more practicable process is the addition of a preservative solution (nucleic acid preservative or ethanol) directly in the field. In this study, we did not test the impact of the conservation time of biofilms in nucleic acid preservative, prior to centrifugation and -20 °C storage, which may be an interesting alternative. Field collections become compromised when sample processing cannot be completed within short critical time periods when essential capabilities are unavailable. However, we assume that the conservation of samples in preservative is little affected by storage temperature (frozen or room temperature) in the first week following sampling (Ladell et al. 2019), thus providing flexibility in the first sampling and storage step of the metabarcoding process. Preservation with ethanol appears to be the most practicable strategy for large or remote field campaigns, whilst storage in the dark at +4 °C facilitates the easy storage of a large number of samples. For studies orientated towards the characterisation of diatom assemblages, our results show that ethanol is well-suited to long-term preservation.

A first attempt for standardisation was done in 2018, with the publication of a technical report (CEN 2018b) in line with an existing standard developed in CEN (CEN 2014a). This technical report concerned adapting the initial sampling phase to ensure it evolved in a manner consistent with DNA metabarcoding. Indeed, the sample conservation proposed for analysis by light microscopy (e.g. Lugol’s, formaldehyde) was preventing the extraction of DNA from conserved samples. Based on ongoing studies at this time, a large choice of already-tested DNA-friendly conservation methods was proposed in a technical report (CEN 2018b). From that starting point, the present study was conducted to ascertain and optimise the use of these conservation methods. Based on our results, the initial sampling step of the diatom metabarcoding process appears to be robust whatever the preservation method used and whatever the storage duration, up to one year and probably longer. Consequently, preservation method choice in future diatom metabarcoding studies dedicated to ecological quality assessment could be primarily based on experimental design, field and lab facilities, shipping constraints and available funding and less on necessary choice of one optimal sample preservation method.

Such robustness has been already observed for other steps of diatom metabarcoding: DNA extraction methods (Vasselon et al. 2017b), PCR amplification (Vasselon et al. 2021), bioinformatic processing (Bailet et al. 2020; Rivera et al. 2020). In all these studies, final end-points, especially ecological quality indices, were seldom affected by changes in methods. An open-access reference library, Diat.barcode (Rimet et al. 2019) and related standards (CEN; Rimet et al. 2021) are also completing the tool-box.