Research Article |
Corresponding author: Cristina Di Muri ( cristina.dimuri@iret.cnr.it ) Academic editor: Florian Leese
© 2020 Cristina Di Muri, Lori Lawson Handley, Colin W. Bean, Jianlong Li, Graeme Peirson, Graham S. Sellers, Kerry Walsh, Hayley V. Watson, Ian J. Winfield, Bernd Hänfling.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Di Muri C, Lawson Handley L, Bean CW, Li J, Peirson G, Sellers GS, Walsh K, Watson HV, Winfield IJ, Hänfling B (2020) Read counts from environmental DNA (eDNA) metabarcoding reflect fish abundance and biomass in drained ponds. Metabarcoding and Metagenomics 4: e56959. https://doi.org/10.3897/mbmg.4.56959
|
The sampling of environmental DNA (eDNA) coupled with cost-efficient and ever-advancing sequencing technology is propelling changes in biodiversity monitoring within aquatic ecosystems. Despite the increasing number of eDNA metabarcoding approaches, the ability to quantify species biomass and abundance in natural systems is still not fully understood. Previous studies have shown positive but sometimes weak correlations between abundance estimates from eDNA metabarcoding data and from conventional capture methods. As both methods have independent biases a lack of concordance is difficult to interpret. Here we tested whether read counts from eDNA metabarcoding provide accurate quantitative estimates of the absolute abundance of fish in holding ponds with known fish biomass and number of individuals. Environmental DNA samples were collected from two fishery ponds with high fish density and broad species diversity. In one pond, two different DNA capture strategies (on-site filtration with enclosed filters and three different preservation buffers versus lab filtration using open filters) were used to evaluate their performance in relation to fish community composition and biomass/abundance estimates. Fish species read counts were significantly correlated with both biomass and abundance, and this result, together with information on fish diversity, was repeatable when open or enclosed filters with different preservation buffers were used. This research demonstrates that eDNA metabarcoding provides accurate qualitative and quantitative information on fish communities in small ponds, and results are consistent between different methods of DNA capture. This method flexibility will be beneficial for future eDNA-based fish monitoring and their integration into fisheries management.
biodiversity, eDNA monitoring, freshwater, lake, lentic
Environmental DNA (eDNA) metabarcoding is revolutionising biomonitoring in aquatic environments (
Environmental DNA metabarcoding has been recently suggested as a complementary biomonitoring strategy for the European Union Water Framework Directive (WFD, 2000/60/EC) which requires member states to assess the ecological status of freshwater bodies. Currently established WFD methodologies include the morphological identification and counting of phytoplankton, phytobenthos and benthic invertebrates or gillnetting and electrofishing for fish (
The ability of eDNA metabarcoding to provide information on abundance and biomass is more controversial, and uncertainties regarding the quantitative power of eDNA metabarcoding are still present among the scientific community and monitoring agencies (
A second key question is how replicable eDNA metabarcoding is with different field and laboratory protocols. Standardisation of protocols may overcome this issue, but a “one‐size fits all” protocol would be unrealistic (
To evaluate the efficiency and suitability of different eDNA capture, a number of published studies have compared different approaches (precipitation versus filtration; on-site versus in laboratory), and a variety of filtration equipment, filters material and filters pore size (e.g.
In the present study we tested whether eDNA metabarcoding can provide accurate information on the community composition and fish biomass and abundance in ponds that were drained as part of an invasive species eradication programme. During the drain down, all fish were counted, measured and weighed, providing absolute measures of species abundance and biomass, and so avoiding the biases of established techniques used in previous studies. Secondly, we tested whether estimation of fish abundance and biomass with eDNA metabarcoding is consistent between different methods of DNA capture, by comparing Sterivex (hereafter also STX) enclosed filters preserved with three different buffers (ethanol, Longmire’s solution and RNAlater) and open filtration (using Mixed Cellulose Ester; MCE filters and a vacuum pump) followed by freezing preservation at -20 °C.
The study was carried out at a UK fishing venue which consisted originally of three hydrologically-isolated stocked ponds (Upper, Middle and Lower Lake; Fig.
Map and fish diversity of the site surveyed. (A) Map of eDNA collection sites (in red) at the fishery venue. Map was downloaded and edited from Digimap (https://digimap.edina.ac.uk). (B) Fish species composition of the New Lake and Middle Lake after re-stocking (species with asterisk only). Ring pie charts (outer circles) show proportion of species composition by fish abundance (number of individuals); pie charts (inside circles) indicate proportion of species composition by fish biomass (kg).
Water samples were taken on three separate occasions applying different strategies based on the goal of each occasion (see Fig.
Experimental design. Panels show eDNA collection at different ponds (New Lake and Middle Lake) and processing strategies (Sterivex filters [STX] vs. Mixed Cellulose Ester open filters [MCE]). Numbers within the panels indicate the workflow from water sampling (1) to filtration (2) and DNA extraction (3).
All precautions to avoid contamination were taken while sampling. Fieldwork equipment was sterilised using 10% v/v chlorine-based commercial bleach (Elliott Hygiene Ltd, UK) and sterile gloves (STARLAB, UK) were changed at each sampling location. Blanks, consisting of 2 L sampling bottles filled with ultra-purified water (Milli-Q), were included for each sampling occasion. Blanks were opened once in the field and then kept and processed alongside other water samples.
On each sampling occasion, eight 2 L water samples were collected equidistantly (~30 m apart) around the perimeter of each pond (Fig.
Samples for open filtration were placed inside cool boxes with ice packs, transported back to the laboratory and processed within 12 hours of collection. Environmental DNA was captured on 0.45 μm MCE membranes (47 mm diameter, Whatman, GE Healthcare) using a vacuum-pump and NalgeneTM filtration units. Filtration equipment was sterilised in 10% v/v chlorine-based commercial bleach (Elliott Hygiene Ltd, UK) for 10 min, then rinsed with 5% v/v MicroSol detergent (Anachem, UK) and with purified water. Filtration was stopped after 45 min and approximately 500 mL of water was filtered through each of two MCE open filter membranes per sample (i.e. 1 L of the 2 L total sample was filtered). Filter membranes were then stored in sterile 50 mm Petri dishes (Fisher Scientific UK Ltd, UK) sealed with parafilm (Bemis, Fisher Scientific UK Ltd, UK) and kept at -20 °C until DNA extraction.
Sterivex filtration was carried out in the field. Environmental DNA was captured using 0.45 μm Sterivex filter units (PVDF membrane, Merck Millipore) connected to a peristaltic pump (Easy Load II Peristaltic Pump, In-situ Europe Ltd, UK). On-site filtration was also carried out until an individual filter became clogged, otherwise it was stopped after 45 min. Approximately 350 mL were filtered through each Sterivex filter and three Sterivex units were used per sample. Each filter was then preserved using 2 mL of one of three different buffers: ethanol (≥ 99.5% v/v), Longmire’s solution, and RNAlater.
All DNA extractions were carried out using the Mu-DNA protocol for water samples following adaptation for Sterivex as recommended in
Contamination during laboratory procedures was minimised by using separated laboratories, located on different floors, for pre-PCR and post-PCR work. Pre-PCR procedures (DNA extraction and PCR preparation) were performed in a dedicated laboratory where only eDNA samples are handled. This laboratory has separated work stations for DNA extraction and PCR preparation. All equipment, instruments and benches are sterilised with 10% commercial bleach solution and 70% ethanol solution prior and after any work. PCR preparation occurred under UV-sterilised hoods with dedicated PCR pipettes.
Library preparation included a two-step PCR with a nested-tagging approach as described in
After visualisation, PCR triplicates were combined and samples belonging to the same collection site were pooled and normalised using different volumes as deduced from strength of PCR products on gels (no/very faint band = 10 µL, faint band = 7.5 µL, bright band = 5 µL) using 1 μL of the positive samples and 5 μL of blanks/negatives for each pool (
Amplicon pools were cleaned using a double-size selection magnetic beads protocol (
Raw sequencing data were demultiplexed using a custom Python script and subsequently analysed with metaBEAT (metaBarcoding and Environmental Analysis Tool) v0.97.11 (https://github.com/HullUni-bioinformatics/metaBEAT), an in-house developed pipeline. Quality trimming, merging, chimera detection, clustering and taxonomic assignment against a custom-curated 12S reference database (
Total read count per sample was calculated as the sum of assigned and unassigned reads. The proportion of reads assigned to each fish species over the total read counts was then calculated on a sample by sample basis. A low-frequency noise threshold of 0.001 (0.1%) was applied across the dataset to reduce the probability of false positives arising from cross-contamination or tag-jumping (
Morphological identification of fish species revealed that a substantial amount of F1 hybrids (Fig.
As read counts and site occupancy data were not normally distributed, Spearman’s rank correlation coefficient was used to calculate correlations between biomass/abundance data and species average read counts and site occupancy for filter types and treatments. Graphs were plotted using ggplot2 (
VEGAN package v2.5-4 (Oksanen et al. 2019) was then used to test differences of fish communities between filter types (Sterivex and MCE membranes) and treatments (preservation buffers and freezing). Betadisper was used to investigate compositional variance of each group, and homogeneity of group dispersions was tested using ANOVA. Distances from the centroids of each treatment and the variance within treatment were visualised with a Principal Coordinates Analysis (PCoA). To test groups for compositional differences, a permutational multivariate analysis of variance (PERMANOVA), with replicates nested into each filter type, was carried out using the adonis function. Tests were performed on a square-root transformed abundance-weighed dissimilarity matrix (Bray-Curtis) of species composition.
Kernel density plots of fish species richness distribution across eDNA samples for each pond (New Lake and Middle Lake) and eDNA filtration/preservation strategy (Sterivex with buffers and MCE open filters replicates) were used to evaluate the number of fish species detected in the mixed samples compared to the mean species richness of eight individual samples. Density plots were built using the function geom_density implemented in ggplot2 (
Lastly, sample-based species accumulation curves (SACs) were built using the function specaccum for each filter type and replicate.
The total number of forward and reverse sequences across 98 samples (81 eDNA samples and 17 controls) was 10,751,170. Of these, 6,398,530 paired-end sequences passed the trimming quality filter and 92% were subsequently merged. 3,389,668 sequences remained after chimera detection and clustering with an average read count per sample of 40,042 (excluding control samples). Excluding the cichlid species used as positive controls, 16 Operational Taxonomic Units (OTUs), and 1,314,623 sequences were identified as fish taxa, with 100% match to the custom-curated fish reference database with thirteen OTUs remaining after applying the thresholds. All fish OTUs were identified to species level with the exceptions of records matching the family Percidae. Percidae records were manually assigned to P. fluviatilis as this was the only species of the family identified in the study area during fish relocation.
P. parva reads found in two Middle Lake-STX samples (279 and 148 reads) were also excluded from further analyses as after eradication this species was not physically present at the site surveyed.
OTUs from eleven of the twelve fish species translocated to New Lake were detected in eDNA samples, but two records were removed after applying thresholds. Sequences from the following taxa were detected at all eight sites within New Lake: A. brama, C. carassius, C. carpio, P. fluviatilis, R. rutilus, Silurus glanis and Tinca tinca (Fig.
All nine possible OTUs corresponding to the species reintroduced were detected beyond threshold limits in Middle Lake in both sampling occasions (16th and 17th of February). Eight OTUs (A. brama, R. rutilus, C. carassius, C. carpio, T. tinca, B. barbus, P. fluviatilis, S. cephalus) were detected in both Middle Lake-STX and Middle Lake-MCE, and with all filter replicates (Figs
We evaluated the relationship between fish eDNA read counts/site occupancy of different filter replicates and fish biomass and abundance in New Lake and Middle Lake.
We observed a strong positive association between fish read counts and fish biomass (r = 0.75; p = 0.052; Fig.
Spearman’s correlations were calculated separately for each filter type (Sterivex/filter membranes) and filter replicate for samples collected from Middle Lake (Middle Lake-STX, Middle Lake-MCE). Fish read counts for all replicates and filters were positively correlated to both fish biomass and abundance. The highest associations were observed when read counts of Sterivex filter replicates were compared with biomass (Ethanol: r = 0.89, p = 0.019; Longmire: r = 1, p < 0.001; RNAlater: r = 0.93, p = 0.0025; Fig.
For MCE open filter membranes (Middle Lake-MCE), there was a significant correlation between read counts and biomass for both filter replicates (r = 0.79, p = 0.036; r = 0.94, p = 0.0048; Fig.
Positive but weaker correlations of New Lake eDNA samples were observed when species site occupancy was associated with fish biomass (r = 0.58, p = 0.17; Fig.
Fish site occupancy of Middle Lake filter replicates (Middle Lake-STX, Middle Lake-MCE) was also positively correlated to both fish biomass and abundance with, however, weaker associations. Correlation coefficients and significance of the Spearman’s correlations varied between filter replicates of both filter types. The strongest associations were observed when site occupancy of Sterivex filters preserved with ethanol were correlated with abundance and biomass (Ethanol: r = 0.94, p = 0.0051; Fig.
Correlations between eDNA metabarcoding read counts and fish abundance/biomass. Scatterplots showing lines of best fit and Spearman’s correlations of fish species average read counts with abundance (number of individuals, on the left) and biomass (kg; on the right) at different sampling occasions. Panel (A) and (B) Spearman’s correlations for New Lake; (C) and (D) Spearman’s correlations for Middle Lake with Sterivex filters (STX); (E) and (F) Spearman’s correlations for Middle Lake with open filter membranes (MCE). Plot axes were log transformed for better visualization. Significance codes: ***0.001; **0.01; *0.05.
Correlations between eDNA metabarcoding site occupancy and fish abundance/biomass. Scatterplots showing lines of best fit and Spearman’s correlations of fish species site occupancy with abundance (number of individuals, on the left) and biomass (kg; on the right) at different sampling occasions. Panel (A) and (B) Spearman’s correlations for New Lake; (C) and (D) Spearman’s correlations for Middle Lake with Sterivex filters (STX); (E) and (F) Spearman’s correlations for Middle Lake with open filter membranes (MCE). Significance codes: ***0.001; **0.01; *0.05. Note: mixed samples were not included in the analyses.
To evaluate the effect of different sampling strategies the mean species richness of individual samples was compared to the species richness of the mixed sample at each sampling occasion and treatment (Fig.
There were no differences between fish community composition of different filter types (ANOVA F = 0.8521, p = 0.3611; Fig.
There was no significant difference between centroids of Middle Lake fish communities described by eDNA metabarcoding when using different filter types (PERMANOVA; R2 = 0.23278; p = 0.7231) or different preservation methods (buffers and freezing; R2 = 0.03795; p = 0.7231). However, more variation (23%) was explained by the use of different DNA capture methods (MCE versus Sterivex), compared to within filter treatment (3.8%).
Species accumulation curves of both Sterivex and MCE filters showed that approximately six samples are required to detect all fish species when filter replicates are combined (Fig.
Environmental DNA metabarcoding fish community plots for different filter types and treatments. (A) Kernel density plots showing distribution of species richness across eDNA samples collected from different ponds and with different filtration and filter preservation strategies. The dashed blue lines indicate the mean species richness of individual eDNA samples (n = 8), the dotted black lines indicate the species richness of mixed samples (pooled aliquots of individual samples) at each sampling occasion and filtration/preservation strategy. The x axes represent the fish species richness in each pond surveyed (New Lake = 12 species; Middle Lake = 9 species) (B) PCoA plot showing distances from centroids of filter types (MCE and Sterivex [STX]; ANOVA F = 0.8521, p = 0.3611) and treatments (buffers and freezing; ANOVA F = 0.6495, p = 0.6305). Distances from centroids were calculated upon a dissimilarity matrix (Bray-Curtis) of fish species read counts. (C) Species accumulation curves for filter replicates of Sterivex filters preserved with buffers (top) and MCE filters with freezing preservation (bottom). In both figures, golden curves are calculated based on the sum of species when filter replicates/treatments of the same filter type are combined. 95% confidence intervals refer to the golden curves and boxplots of these curves show distribution of species diversity as inferred from the method “random”, which add sites in random order and was used for the SACs. Asterisks represent outliers.
With the advent of the next-generation eDNA-based monitoring surveys there is a growing interest in whether eDNA metabarcoding can generate accurate semi-quantitative data. Previous studies in natural environments have focussed on indirect estimates of fish abundance from established surveys which have their own inherent biases. Here, we used absolute data on fish abundance and biomass from drained ponds and found that read counts from eDNA metabarcoding consistently correlate with both fish abundance and biomass. Moreover, the present study suggests that the use of different eDNA capture (Sterivex vs. MCE open filters) and storage methods (buffers and freezing) produce repeatable results of fish diversity, composition and biomass/abundance estimates. We additionally show that the collection of spatial and filter replicates enhances species detection probability for rare species, thus sample coverage and replication are an important consideration in experimental design.
All 12 fish taxa were successfully detected in both fishery ponds surveyed with the only exception of S. cephalus in New Lake (single specimen of 0.7 kg; Fig.
Of particular interest is the detection of P. parva DNA in Middle Lake samples as this invasive species was the target of the eradication programme and present in extremely high abundance before the ponds were drained and treated with a piscicide. The persistence of P. parva as living organisms within the pond appears extremely unlikely due to the effective eradication methods used in combination with the relatively small size of the water body (
To our knowledge, this is the first published study to date where the correlation between eDNA metabarcoding data and actual measures of species biomass and abundance in semi-natural lentic systems has been investigated.
Positive associations were observed between species site occupancy and fish biomass/abundance, however, less significant than correlations with read counts (Fig.
Current uncertainties regarding the quantitative power of eDNA metabarcoding ultimately originate from our lack of knowledge on the origin and fate of eDNA in aquatic systems (
A lack of robust sampling and metabarcoding protocols may also contribute to a distortion of the observed diversity patterns. Insufficient sampling effort, inhibition, primer biases, sequencing artefacts, database inaccuracy and contamination are the main methodological sources of bias (
Suitable eDNA metabarcoding data for quantitative fish monitoring require comparable measures of biomass and abundance across studies and over time, for example, to detect trends in abundance of fish populations. In light of this, the use of site occupancy appears a more practical approach as abundance/biomass estimates from site occupancy are easily comparable even across studies with uneven sampling efforts (
In our study the correlations between sequence read counts and species abundance/biomass were consistently high for all filtration treatments with average correlation coefficients of 0.93 for Sterivex filters and 0.84 for MCE filters (Fig.
The higher species richness found in Sterivex filters preserved with RNAlater and open filter membrane replicate 1 resulted from the detection of only one low-abundant taxon within the pond (S. erythropthalmus; Fig.
Overall, we found that both filter types showed a good representation of fish diversity and community composition and, consequently, we suggest that they can be used interchangeably depending on time, resources and location of the study. Sterivex filters, for instance, are effective for field processing of water samples, facilitating collection in remote locations. After sample collection, Sterivex are immediately filtered on-site (using peristaltic pumps or sterile syringes) and the risk of contamination is reduced because of the lack of filter handling (
This study underpins valuable considerations for the quantitative estimates of eDNA metabarcoding data. We demonstrated that eDNA metabarcoding data correlate with actual abundance and biomass of fish communities within small freshwater systems with high fish density.
Established methods (i.e. hydroacoustic, electrofishing, gillnetting) for obtaining quantitative estimates of fish abundance are resource intensive and may not be suitable for all water bodies and species (
Environmental DNA metabarcoding is arguably a more flexible tool, adaptable to all aquatic environments and fish species, is non-lethal, and the sources of errors can be minimised through a careful optimisation of field and laboratory protocols.
Monitoring trends in population size and community structure is paramount to the assessment of species health and viability, and the outputs are required to undertake management actions and to guide conservation decisions (
Details of protocols, bioinformatics, R script and supplementary material used for the analyses can be found on Open Science Framework at: DOI https://doi.org/10.17605/OSF.IO/ZWPSQ. Sequencing data have been submitted to NCBI (Bioproject: PRJNA638011; SRA accession numbers: SRR11949830–SRR11949928).
We would like to express our thanks to the EA national and local staff working at the site during the eradication programme and the fishery farm owner for providing us the opportunity of using the site for this research study. We are also very grateful to Peter Shum, Jairo Arroyave and Stephanie McLean for the constructive feedback on the manuscript before its initial submission.
This work was funded by the UK Environment Agency (collaborative agreement 171024).
Table S1 and Figure S1
Data type: Protocols, figures and table
Explanation note: Additional laboratory protocol information and fish diversity and biomass/abundance data of the ponds surveyed; Figure S1 – Read counts and site occupancy barplots; Table S1 – Fish biomass/abundance data; Protocol of DNA extraction from Sterivex filters; Magnetic beads purification protocol.
Table S2. Fish taxonomic assignment metaBEAT
Data type: Table
Explanation note: metaBEAT results of taxonomic assignment (100% identity) against the custom-curated 12S fish database.
Table S3. Unassigned blast 1.0
Data type: Table
Explanation note: Results of unassigned reads blast (100% identity) against GenBank.