Optimizing target-to-total DNA ratio in eDNA studies: effects of sampling, preservation, and extraction methods on single-species detection

Field sampling, filtration, preservation, extraction

Optimizing the target-to-total DNA ratio

Our other metric of interest in addition to absolute target DNA concentrations is the ratio of target DNA (as measured by qPCR) to total DNA (as measured by Qubit). The units of target DNA are copies per volume whereas total DNA is mass per volume. We converted the total DNA from mass to copy number by using the length of the fragment, the average mass of each base pair, and Avogadro’s number (Eq. 1). This gives us the concentration in copy/µL if all the total DNA in the extract were 80 base pair fragments (i.e., the length of the target DNA amplified by the T. tursiops assay). Rather than using the exact molecular weight of the target DNA, we use an average molecular mass of a DNA base pair (618 g/mole) given that we know that all genomic DNA is not target DNA. This allows us to make a ratio by having both quantities in the same units, using: (1)${\displaystyle C_{totalDNA}\left[copies/\mu L\right]=\frac{C_{totalDNA}\left[ng/\mu L\right]\mathrm{\ast }6.022\mathrm{\ast }10^{23}\left[copy/mol\right]}{MW\left[g/mol\right]\mathrm{\ast }10^9\left[ng/g\right]}.}$

Then the ratio of target to total DNA is calculated simply by: (2)${\displaystyle Ratio=\frac{C_{targetDNA}\left[copy/\mu L\right]}{C_{totalDNA}\left[copy/\mu L\right]}.}$

Linear models to compare different methodological choices

Here, we present two linear models to quantify how different methodological choices (volume filtered, filter pore size, preservation method, extraction method) impact target DNA quantification. Given that for each field campaign, the water sampled all comes from a common pool, we can attribute the differences in target quantification to (1) sampling variability, and (2) the differences in methodological choices in collection and processing. Because we did not have a full factorial sampling design over our two field campaigns, we developed two closely related models: one predicting concentration as a function of volume filtered and pore size, and the other predicting concentration as a function of preservation and extraction methods.

In both models, (log) observations of concentrations of target DNA are treated hierarchically, with observations from technical replicate i, bottle j, treatment k, and campaign m drawn from a nested series of normal distributions: ${\displaystyle y_{ijkm}\sim N\left({\mu }_{jkm},\sigma \right)}$ ${\displaystyle {\mu }_{jkm}\sim N\left({\theta }_{km},\tau \right)}$ ${\displaystyle {\theta }_{km}\sim N\left({\varphi }_m,\nu \right)}$

where μ_jkm is the mean of each biological replicate (i.e., bottle-level mean) and σ is the standard deviation among technical replicates within a biological replicate. Bottle-level means are in turn treated as samples from a treatment-level distribution of mean θ_km and standard deviation τ. Finally, all treatment-level means are treated as draws from a campaign-level (overall) distribution of mean ϕ_m and standard deviation ν. The model reflects the fact that the samples are nested, with samples from each campaign collected from a common pool of water, but each treatment (here, pore size and volume) has a set of biological replicates, and those biological replicates are subsamples of the common pool and technical replicates are subsamples of each biological replicate.

We first model the effects of volume filtered and pore size filter across two field campaigns as: (3)${\displaystyle {\theta }_{km}=\beta X}$

where β is a vector of regression coefficients having the same length as θ, and X is the design matrix mapping different combinations of volume filtered and pore size (i.e., treatments) to the data.

The second model is identical, but for treatments of different preservation and extraction methods, and having only a single campaign, such that it requires only the first two hierarchical levels (i.e., there is no campaign-level mean to separate the two pools of water).

The priors for both models were the same and were as follows: ${\displaystyle {\mu }_{jkm}\sim N\left(0,10\right)}$ ${\displaystyle \sigma \sim gamma\left(1,1\right)}$ ${\displaystyle \tau \sim gamma\left(1,1\right)}$ ${\displaystyle \beta \sim N\left(0,5\right).}$

The models were both implemented in RStan.

Total DNA, target DNA, ratio of target to total DNA

For the samples collected in Campaign 1, we compared volume filtered and filter pore size (Table 2, Fig. 1). Given the same volume of water filtered, the 1 µm filters had higher total and target DNA as compared to the 5 µm filters. Given the same pore size, the 1 L filters had less total and target DNA than the 3 L filters. However, the 1 µm filters with 3x volume filtered had more than 3x total DNA (340% of mean value) and more than 3x target DNA (434% of mean value). The 5 µm filters showed a similar pattern with the 3x volume samples having more than 3x total DNA (325% of mean value) but showed much more than 3x target DNA (514% of mean value). When converting these to ratios of target to total DNA, 1 µm filters were lower than 5 µm filters and the 1 L samples were lower than the 3 L samples, resulting in the 5 µm and 3 L filter having the highest ratio of target to total DNA.

Here, the 1 µm filters captured more total DNA than the 5 µm filters did, presumably across a range of particle sizes; we assume that the material captured on a 5 µm filter is a subset of the material on a 1 µm filter, with the larger pore size selecting for larger particles and not capturing particles <5 µm (at least initially; at some point as the filter clogs it will have an effective pore size smaller than 5 µm and capture smaller particles). To the extent that our target eDNA fragment—from vertebrate mtDNA—is more likely to occur in larger particles (due to the size of mammalian cells, etc.), we expect and observe a larger target:total ratio with 5 µm filters (Power et al., 2023). In terms of inhibition, all 1 L samples required 1:20 dilutions except two technical replicates requiring 1:100 dilutions, whereas the 3 L samples required dilution from 1:10 to 1:40 (Figs. S1 and S2). However, there was no significant effect of either volume filtered or pore size on the dilution factor required to alleviate inhibition.

For the second campaign comparing preservation and extraction methods, the total DNA varied across methods, with PCI extractions having consistently higher yields than either kit across preservation methods (Fig. 2). However, target DNA was similar across preservation and extraction methods. The samples preserved via −80 °C had the highest target DNA recovery, but the resulting ratio of target to total DNA exhibits the opposite trend of the total DNA yield, where PCI has the lowest ratio, then the Qiagen kit, and finally the Zymo kit with the highest ratio (i.e., the most desirable for a particular targeted assay). For all samples extracted with the Zymo kit, the Zymo DNA/RNA Shield had the highest ratio of the three preservatives tested. We found that the samples extracted via PCI had the highest total DNA concentrations however had the most samples that were not inhibited at 1:1 compared to Zymo and Qiagen (Figs. S3 and S4).

Biological variability with homogenized and non-homogenized samples

In the first campaign, a large volume of water was collected and homogenized before splitting into the different treatments of pore size and volume filtered. In the second campaign, individual water samples were grabbed from the source (well mixed, relatively small pool) but not homogenized before filtering. In both experiments, the coefficient of variation between technical replicates (Fig. 3, X markers) was less than the coefficient of variation between biological replicates (Fig. 3, colored circles), which was less than the coefficient of variation between treatments (Fig. 3, dashed lines; except in two sets of biological replicates). This trend demonstrates that it is important to have replication to be able to separate technical and biological variability from the treatment effect (here, volume/pore size and preservation/extraction).

For the volume and pore size experiment, we also calculated the coefficient of variation assuming a linear increase in concentration with volume filtered (Fig. 3, Panel A, dotted line). The coefficient of variation across treatments in the homogenized water in Campaign 1 (pore size and volume) and adjusting for the volume filtered was 0.24, compared to a coefficient of variation of 0.35 across treatments in Campaign 2. For the homogenized samples, the mean coefficient of variation between biological replicates was 0.1, whereas the mean coefficient of variation between biological replicates in the non-homogenized samples was 0.2 (Fig. 3). The observed biological variability did not seem to be related to the methodological choices (Fig. 3). Given how close the biological variability was to the variability across treatments especially in the preservation and extraction experiment, the linear models are particularly helpful to distinguish sources of variability.

Models to combine samples processed with different methods

Our two sets of experiments provided an opportunity to use two models to combine samples with different methodological choices from the same common reality (e.g., different pore size and volume filtered in the first experiment and different preservation and extraction methods in the second experiment). Because in a single campaign, the samples came from the same source water (and in the case of the first experiment, the source water was homogenized), we can assess how the different methodological choices affect differences in target DNA recovery through the use of linear models.

Preservation and extraction methods

Here, the source water was the same for all samples so there is a single intercept, which represents preservation by −80 °C and extraction by PCI (i.e., the base case for treatment type); again, the coefficients for treatment effects indicate deviation from those methodological choices (Table 4, Fig. 5, Fig. S6). We find no meaningful effects of preservation or extraction method—including interactions among a subset of these—on our target eDNA concentration, with 1 exception: desiccation as a preservation method yielded systematically less target eDNA than other methods (mean posterior coefficient estimate = −0.894, 95% posterior CI [−1.37 to −0.429]). Desiccation retains approximately 40% of target eDNA relative to what would be retained by preserving filters at −80 °C immediately after filtering. Across a broad range of methodological choices, our results suggest rougly even performance in recovery of target DNA with only a few departures.

Table 3:

Model parameters for volume and pore size experiment.

Coefficients in bold are meaningfully different than zero.

Parameter	Mean estimate	Standard deviation	2.5% CI	97.5% CI
Campaign 1	8.81	0.134	8.55	9.09
Campaign 2	11.2	0.271	10.6	11.7
Pore size (5 μ m)	−0.0937	0.157	−0.409	0.214
Log volume	1.43	0.140	1.14	1.71

DOI: 10.7717/peerj.20127/table-3

Table 4:

Model parameters for preservation and extraction experiment.

Coefficients in bold are meaningfully different than zero.

Parameter	Mean estimate	Standard deviation	2.5% CI	97.5% CI
(Intercept)	12.8	0.166	12.5	13.1
Desiccation	−0.894	0.239	−1.37	−0.429
Shield	−.0330	0.236	−0.807	0.133
Longmires	−0.130	0.234	−0.593	0.320
Qiagen	0.195	0.238	−0.281	0.660
Zymo	0.105	0.236	−0.365	0.555
Desiccation/Qiagen	0.167	0.339	−0.492	0.840
Shield/Qiagen	−0.00194	0.337	−0.673	0.672
Desiccation/Zymo	0.400	0.333	−0.261	1.05
Shield/Zymo	0.124	0.337	−0.523	0.821

DOI: 10.7717/peerj.20127/table-4

More (total) DNA is not always better

The natural inclination is to maximize capture of all DNA in order to capture more target DNA, either by using a smaller pore size filter or using a different extraction protocol. We found here that, as expected, both target and total DNA increased with both increased sample volume and with smaller pore size filters (Capo et al., 2020). However, the increase in target DNA recovery was small compared to the increase in total DNA recovery. In other words, the target was a smaller percentage of total DNA (in the case of 1 µm filters, total DNA increased by 340% from 1 to 3 L while target DNA increased by 434% and in the case of 5 µm filters, total DNA increased by 318% from 1 to 3 L while target DNA increased by 514%). By having a rare target in a larger pool of off-target DNA, other issues can arise associated with the concentration of various inhibitors. In both experiments, we demonstrate that the optimal methods for maximizing total DNA capture do not maximize the target DNA nor the ratio of target to total DNA. Here, we did not find that samples with more total DNA had more inhibition, but in cases where the target DNA is very rare, the maximization of target/total DNA should be carefully considered. For example, if the absolute target DNA concentrations is 9 copies/L and a 1 L sample (9 copies total) is not inhibited but a 3 L sample (27 copies total) requires a 1:100 dilution, the target DNA would be diluted (perhaps beyond the limit of detection) in the inhibition treatment.

Though inhibition was not at a threshold where the rarity of the target DNA in a larger pool of total DNA was an issue, there are remarkable patterns in particularly the extraction method (Fig. 2C) demonstrating the increasing ratio of target to total DNA recovery. As noted above, we encourage researchers to consider not only the absolute recovery of either target or total DNA, but to consider when target is particularly rare and inhibition is likely to choose a methodology focusing on maximizing the ratio rather than absolute concentration.

The same concepts apply with the preservation and extraction experiment with different methodological choices resulting in different total DNA capture, target DNA capture, and ratio of target to total DNA capture (Spens et al., 2017; Deiner et al., 2015; Deiner et al., 2018; Djurhuus et al., 2017). Longmire’s buffer with PCI extraction yielded the highest total DNA concentration while −80 °C preservation with Qiagen extract yielded the highest target DNA concentration, however Shield preservation with Zymo extraction yielded the highest ratio of target to total DNA. As with any ecological sampling, different methods can and do result in different reflections of the environment being sampled. However, a quantitative framework accounting for such methodological variability—such as the one we present here—offers a simple means of combining information across protocols.

Some hypotheses of the differences in total, target, and ratio of total to target DNA across different pore sizes, preservation, and extraction methods include consideration of the mechanisms of filtration and the combinations and compatibility of preservatives and extraction kits (Capo et al., 2020). For the filter pore size, smaller pore sizes (here, 1 µm) will capture more bacteria which will contribute to the total DNA yield, but as in this study targeting dolphin DNA, are non-target (Power et al., 2023). A larger pore size filter (here, 5 µm) will result in less capture of smaller organisms such as bacteria and will leave room for capturing more of the desired larger animal cells before the filter clogs or the designated volume has been filtered. It should be noted that this study was conducted in relatively shallow, near-shore, nutrient rich water. Deep sea or oligotrophic waters might concentrate less off-target DNA and the ratios found here are not necessarily portable to different environments.

We note that it would have been informative to have a wider range of absolute concentrations of target DNA and a wider range of ratios of target to total DNA to explore these methodological choices more. In particular, it would be informative to have lower target DNA concentrations where the trade-offs between inhibition treatments (namely, dilution) and target recovery are empirically demonstrated. We also acknowledge that a wider range of pore sizes and volumes filtered for the first campaign would have been particularly informative to confirm the trends observed. Future studies could expand to even smaller pore size filters (e.g., 0.22 or 0.45 µm) and even larger pore size filters (e.g., 10 µm). Finally, we acknowledge the limitation of using a single species target and quantification via qPCR, which limits the applicability to other single-species targets with quantification via qPCR or to metabarcoding studies. However, we expect that the mechanisms we discuss here are broadly applicable and generalizable to macro-organisms. For single-species targets and quantification via qPCR for micro-organisms, in particular, the pore size of the filter should be smaller for optimal capture. For metabarcoding, we expect the findings here to be broadly generalizable to primers targeting macroorganisms as well, though we hypothesize that there could be a larger effect in lower concentration samples given the compositional nature of metabarcoding data.

Even when collecting water while looking at a dolphin, we are still looking for a needle in a haystack

We discussed maximizing the target to total DNA ratio to minimize the rarity of the target relative to non-target DNA, however it is worth noting the absolute values of that ratio, especially given how sampling was conducted. In both experiments, dolphins were present while the water was being collected. The absolute concentration of dolphin DNA was high, as expected, and similar order of magnitude concentrations found in other studies where sampling is conducted in close proximity to the target organism (Brasseale et al., 2025) and lower than others with many more individuals present (e.g., Capo et al., 2020; Eichmiller, Miller & Sorensen, 2016). To generate the ratio of target to total DNA, the total DNA concentration had to be converted from units of mass per volume as measured by the Qubit fluorescence reader to copies per volume to match the units of the target DNA concentration as measured by qPCR. This conversion requires a fragment length, which here we use the length of the fragment targeted by the qPCR assay. Therefore, the denominator used in the ratio can be thought of as the total number of fragments that could possibly be target (here, dolphin).

These percentages are very low. Even when sampling water while looking at the species of interest and using methods intended to maximize the target to total ratio (here, larger pore size filter and larger volume of water), the target is just 0.00001% of the total DNA. This corresponds relatively well with a study that used shotgun sequencing and found fish to be 0.00004% of the total reads from a 1 L water sample taken from a reef in Australia and filtered on a 0.2 µm nylon filter, frozen at −20 °C, and extracted using the Qiagen Blood and Tissue Kit (Stat et al., 2017). It is worth noting that water samples contain genetic material from many, many species that we are not interested in, and this is important to bear in mind within the context of very rare targets and the possibility of false negatives.

Here, our molecular assay reliably detected 10 copies/µL of extract. Most extraction protocols (including the one we followed) elute extracts in 100 µL. If 1 L of water was filtered, this becomes 1,000 copies/L filtered, whereas if 3 L of water was filtered, this becomes 333 copies/L, so by filtering larger volumes, more dilute DNA can be detected. Furthermore, a single cell can have 100s to 10,000s of mitochondria, and in humans, each mitochondrion has 2–10 copies of mtDNA, resulting in a range from 100s to 100,000s of copies of mtDNA per cell (Zhang et al., 2015; Castellani et al., 2020; Rath et al., 2024). A single cell is therefore easily detectable given the observed limit of detections.

There are many choices and most are just fine

Though the total DNA yield varied with preservative and extraction choice, the amount of target DNA recovered was similar across methodological choices. Some preservative and extraction protocol combinations are less than ideal (i.e., desiccation with PCI), but most methodological combinations perform similarly. In contract, the −80 °C preservation had the highest target DNA recovery, but can provide to be logistically challenging or impossible in the field. Therefore, practicality of methodological choices must also be considered. Additionally, when selecting a combination of preservation and extraction methods, it is important to keep in mind the mechanisms of the different preservatives. For example, Longmire’s buffer and DNA/RNA Shield are both lysis buffers, meaning that the DNA is actually preserved in the buffer and removed from the filter while the filter is submerged in the buffer. On the other hand, the self-preserving filters and storing filters in the −80 °C both work by desiccation and therefore the DNA is still on the filter when starting the extraction.

Accordingly, as long as the extraction method is compatible with whether the DNA is still on the filter or in the buffer, there should not be large differences in target DNA yield. Again, the total DNA yield will differ based on extraction protocol (e.g., PCI will recover more total DNA), but the target DNA seems relatively robust to different preservation and extraction methods assuming compatibility between the two. Note, there were differences in target DNA recovery with preservation method with −80 °C having higher target DNA concentrations, but logistics also must be considered. Access to reagents and infrastructure like freezers may vary and logistical constraints may impact the decisions for preservation and extraction methods.

Responsibly combining data from different methods

Particularly relevant for time series data or combining data from different projects, it is important to keep methods consistent. However, there are many reasons why one might want to combine data generated from different protocols. Here, we demonstrate the use of simple linear models to “correct” for different protocols and make data comparable. It is important to have a calibration experiment where all possible combinations of protocols are sampled from a common reality in order to make the corrections. However, once that has been done, this allows for extrapolating any scenario from the linear model for unknown samples. Importantly, any set of samples can be translated to the equivalent concentration of a different methodological choice. This is essential to responsibly combine quantitative data collected via different methods, thereby facilitating the generation of larger sample sizes and larger spatial and temporal coverage for exploring broad-scale hypotheses. Future work could look at doing something similar with different species-specific assays for the same target species and determining how portable these parameter estimates are in relation to different assays and other water samples, especially considering other environmental factors like turbidity, salinity, or other parameters that might affect portability.