4.1. Number of fish species detected
The amount of fish species detected was a function of the number of
samples, species present and environmental conditions. The number of
environmental samples varies from 16 (Wells) to 152 (Great Bay), and the
number of fish species expected to be present ranges from 34 (Wells) to
105 (Apalachicola). Accumulation curves developed for each site indicate
an increasing number of species with more samples (as expected), with
the number of new species detected decreasing with more samples (Figure
5). We expected that the region with the highest number of known
(observed) resident fish species (Apalachicola) would also have the
highest density of species detected per eDNA sample; however, this is
not the case, likely due to interference in the samples. In 2019 the
northeast Pacific experienced a ‘marine heatwave’ (Scannell et al.,
2020) resulting in higher than normal temperatures at South Slouth, the
NERR site in Oregon. The warm, turbid waters at this site and at
Apalachicola in Florida yielded high concentrations of DNA, but fewer
fish species than might be expected. Several factors likely contributed
to this: 1) high organic content can inhibit PCR reactions, and reduce
or block the replication of target sequences; 2) co-amplification of
non-target sequences may overwhelm the relatively weak fish signal,
resulting in samples with a high read count for off-target sequences.
Difficulties detecting specific target species in warm, turbid waters
have been reported by others (Kumar et al., 2021; Sanches & Schreier,
2020; Williams et al., 2017) and there are potential solutions,
including collecting smaller samples, and utilizing more sophisticated
methods of removing inhibition or DNA size selection. Primers can be
designed to reduce or block amplification of non-target species, for
example the MiFish 12S primer used in this study has recently been
modified to reduce bacterial amplification (Stoeckle et al., 2022). Some
of these methods increase the per sample cost of analysis, and may not
be appropriate for broad application across a standardized water
monitoring network. Managers interested in applying eDNA across a range
of sites may want to consider a decision tree protocol, where a baseline
methodology is developed for all sites, enhanced by additional
laboratory or analysis steps when specific conditions are present.
We also noted that because the eDNA method detects ‘allochthonous’ DNA,
or DNA that has been transported from outside the ecosystem (e.g., DNA
from freshwater species in an estuarine environment), it is possible to
detect more species than actually reside in the immediate sampling area.
We found that even in Great Bay, where we collected samples from 15
locations, the curve does not become level, and it is likely that
transient marine species, very rare organisms, species shifting ranges,
and arriving invasives, may continuously add to total detections. This
also may be a function of seasonal and annual variation of estuaries,
particularly in the Gulf of Maine where climate is changing faster than
most regions.
4.2. Comparison to traditional fish surveys
We compared species detected with eDNA to species expected to be present
using data from traditional, net-based surveys, and managers’ knowledge
of the system. Information on fish species has been collected over time
at each of the participating NERRs, using a variety of traditional
sampling methods. For example, Apalachicola conducts trawl surveys;
Great Bay and South Slough conduct seining; Heʻeia employs cast net and
visual surveys; and Wells conducts zooplankton tows. For each method,
there are some species which are generally undercounted; for example,
larger fish can escape nets, while very small fish may pass through the
mesh. Local managers augment this data with non-quantitative information
on fishes that are likely to be present but are undercounted in surveys.
Figure 6 compares species detections from seine surveys and eDNA in
Great Bay. Both methods detect relatively few fish at a single site, and
although all of the fish in these surveys are common, the methods do not
fully overlap (6a). The agreement becomes stronger as more samples are
added, and when all of the samples collected in Great Bay in 2019 are
compared to the fish species expected to be present in the system
(including larger fish that are rarely caught in seine nets), there is
strong agreement (6b). When we compare the specific species detected by
the two methods, we find that some species are well represented in each
data set (Fundulus heteroclitus , Menidia menidia ), and
several species (Brevoortia tyrannus , Anguilla rostrata ,Urophycis regia ) are detected only in eDNA. Others, such asAlosa sapidissima seem to be more ready detected by seining. The
limited number of sampling events precludes making definitive
conclusions about the strength of each method with regards to specific
species, but it is clear that the combined methods result in more
diverse detections.