Needle in a haystack? A comparison of eDNA metabarcoding and targeted qPCR for detection of great crested newt (Triturus cristatus)

Environmental DNA (eDNA) analysis is a rapid, cost-effective, non-invasive biodiversity monitoring tool which utilises DNA left behind in the environment by organisms for species detection. The method is used as a species-specific survey tool for rare or invasive species in a broad range of ecosystems. More recently, eDNA and ‘metabarcoding’ have been combined to describe whole communities as opposed to focusing on single target species. However, whether metabarcoding is as sensitive as targeted approaches for rare species detection remains to be evaluated. The great crested newt Triturus cristatus is a flagship pond species of international conservation concern and the first in the UK to be routinely monitored using eDNA. We evaluate whether eDNA metabarcoding has comparable sensitivity to targeted real-time quantitative PCR (qPCR) for T. cristatus detection, and examine the implications for the future of community eDNA monitoring. Extracted eDNA samples (N = 532) were screened for T. cristatus by qPCR and analysed for all vertebrate species using High-Throughput Sequencing technology. Of all ponds, 50% tested positive for T. cristatus with qPCR when the detection threshold was set to 1/12 positive qPCR replicates. Detection rate decreased to 32% when the threshold was increased to 4/12 positive qPCR replicates. With metabarcoding, newts were detected in 34% of ponds when no detection threshold was applied, and in 28% of ponds when a threshold (0.028%) was applied. Therefore, qPCR generates a higher detection rate than metabarcoding, but the metabarcoding detection rate with no threshold is broadly equivalent to that of qPCR with the stringent detection threshold. The proportion of T. cristatus sequences in each sample was positively associated with the number of positive qPCR replicates (qPCR score) suggesting eDNA metabarcoding may be indicative of species abundance. eDNA metabarcoding holds enormous potential for holistic biodiversity assessment and routine freshwater monitoring. We advocate this community approach to freshwater monitoring to guide management and conservation, whereby entire communities can be initially surveyed to best inform use of funding and time for species-specific surveys.

However, eDNA metabarcoding is confounded by potential amplification bias during PCR, 78 preventing capture of all species present in a given area (Kelly et al. 2014). Species' DNA in 79 community samples is also in competition to bind to metabarcoding primers during PCR, where 80 more common templates are more likely to be amplified. High abundance species may thus prevent  Although qPCR provided higher detection probability for two species, metabarcoding achieved 92 comparable results for a third species, and allowed simultaneous detection of IMS and other taxa 93 in a single sequencing run without development of multiple species-specific markers. and can be implemented in large-scale citizen science monitoring programmes without loss of 109 species detection (Biggs et al. 2015). T. cristatus is the first species to be routinely monitored using 110 eDNA in the UK (Natural England 2015) and targeted eDNA assays are now offered as a 111 commercial service by ecological consultancies. The targeted eDNA assay is highly effective for 112 T. cristatus detection; however, should metabarcoding have comparable sensitivity, this approach 113 would allow detection of T. cristatus alongside pond communities and potentially enable more 114 cost-effective monitoring of entire ecosystems and ecological hypothesis testing.

115
Here, we perform a large-scale comparison (N = 532 ponds) of eDNA metabarcoding and 116 targeted qPCR for T. cristatus detection to compare method sensitivity. A single primer pair that 117 is vertebrate specific for mitochondrial DNA (mtDNA) and requires no a priori knowledge of 118 species composition was employed for eDNA metabarcoding. The metabarcoding results were 119 then compared to results obtained using the standard T. cristatus qPCR assay (Biggs et al. 2015).

120
Our hypotheses are: (1) eDNA metabarcoding will give equivalent results to qPCR for T. cristatus 121 6 detection, (2) eDNA metabarcoding sequence read count for T. cristatus will increase as qPCR 122 score (the number of positive qPCR replicates) increases, therefore eDNA metabarcoding may be 123 indicative of species abundance, and (3) metabarcoding primers will amplify DNA from all taxa 124 equally well and no bias towards amplification of T. cristatus will occur (bias would be indicated 125 by a positive association between the proportion of T. cristatus sequence reads and PCR product 126 concentration). We also examined cost and investigator effort required by each approach to 127 determine whether a trade-off between cost, time and amount of data generated exists.

217
To facilitate comparison with current qPCR scoring (our NT interpretation) and eDNA 218 metabarcoding, we applied a stringent qPCR threshold of >= 4/12 positive qPCR replicates to infer 219 species presence, and termed the new analysis qPCR TA (Threshold Applied).

220
The raw eDNA metabarcoding dataset with no detection thresholds applied was termed  Table S2). For example, the species-specific false positive sequence 227 threshold for T. cristatus was 0.028% to omit all false detections in the PCR positive controls. The 228 resultant dataset was termed metabarcoding TA (Threshold Applied). 229 We tested whether mineral oil reduced contamination by analysing the distribution of  We next tested the null hypothesis of no significant difference in sensitivity of qPCR and 247 metabarcoding. Overall agreement between eDNA metabarcoding and qPCR for T. cristatus 248 detection was measured using Cohen's kappa coefficient (Cohen 1960 The relationship between read count and qPCR score has not been examined previously, 252 and whether read production is indicative of DNA concentration remains unknown. qPCR score 253 is positively correlated with T. cristatus abundance and can be proxy for amount of DNA present 254 (Biggs et al. 2015). We hypothesised samples with higher qPCR score would have increased T. 255 cristatus read count. First, the average number of T. cristatus reads produced by eDNA 256 metabarcoding per qPCR score (1-12 out of 12) was calculated. A Spearman Rank Correlation was 257 then used to test for a relationship between average read count and qPCR score.     Table S7); seven occurred in more than one NTC and 308 eight had high maximum read counts (> 100 reads).

342
Agreement between eDNA approaches is summarised in   (Table 2). A significant positive relationship was observed between qPCR score and 358 the proportion of T. cristatus reads within total reads per sample (P < 0.001) (Fig. 3a). Conversely,

394
In our study, 12 independent qPCR replicates were performed for each sample but due to limited 395 resources, metabarcoding was based on three pooled PCR replicates which were sequenced once 396 only. Therefore, to enable a fair comparison between methods in terms of PCR effort, a threshold 397 of >= 4/12 positive replicates (qPCR TA) was applied to the qPCR data. Detection sensitivity was 398 most similar between methods with the qPCR threshold and without the metabarcoding threshold. that has reached its price ceiling, HTS is relatively new technology and prices will continue to 531 drop, meaning higher sample throughput and more technical replication will be possible. We 532 therefore argue that metabarcoding will become more cost-efficient in the long-term, providing 533 more data at lower cost and comparable sensitivity to qPCR. However, where samples cannot be information, and provide broad-scale distribution data for multiple species simultaneously.

544
Nonetheless, the method at present appears to be less sensitive than qPCR for single-species 545 monitoring, and species detection by molecular and conventional methods was incongruent.

546
Comprehensive study of the influence of water volume, eDNA capture and extraction method, and 547 sample storage on single-species and community detection in lentic and lotic systems is required. processed in large batches, metabarcoding may suffer from false negatives due to reduced 558 sequencing depth and replication. However, in many cases, the biodiversity information generated 559 by this approach, and its implications for community ecology and conservation, will eclipse lower 560 sensitivity. This passive screening approach would be most effective for initial survey of water 561 bodies to generate broad-scale multi-species distribution data. This holistic data can then inform 562 best use of funding and time for targeted species-specific survey.