Advances in Environmental DNA Sampling for Observing Ocean Twilight Zone Animal Diversity

The ocean’s vast twilight, or mesopelagic, zone (200–1,000 m depth) harbors immense biomass consisting of myriad poorly known and unique animal species whose quantity and diversity are likely considerably underestimated. As they facilitate the movement of carbon from surface waters to the deep sea through feeding and migratory behaviors, ocean twilight zone (OTZ) animals are vital to regulating Earth’s climate (Ducklow et al., 2001). However, anthropogenic threats, such as climate change, ocean acidification, pollution, and overfishing pose an imminent threat to OTZ animals. Long-term spatially and temporally intensive observations are essential to our understanding of biodiversity in the OTZ, to resolving global carbon cycles, and to monitoring ocean health. Environmental DNA (eDNA) analysis, which involves studying the trace genetic signatures of organisms (Figure 1), is a promising approach to filling this urgent need. eDNA can be sampled and diagnostic genetic markers (“barcodes”) can be sequenced in order to detect the animals inhabiting a given water parcel. Other laboratory protocols (e.g., quantitative PCR, or “qPCR” and “digital droplet PCR”) can be applied to facilitate quantitative assessments of specific target species (Eble et al., 2020). In seagoing oceanographic research, eDNA assessment is transitioning from being considered an experimental approach to becoming an established routine that can be scaled up to match ocean observing needs.


TECHNOLOGY Environmental DNA THE OCEAN TWILIGHT ZONE
The ocean's vast twilight, or mesopelagic, zone (200-1,000 m depth) harbors immense biomass consisting of myriad poorly known and unique animal species whose quantity and diversity are likely considerably underestimated. As they facilitate the movement of carbon from surface waters to the deep sea through feeding and migratory behaviors, ocean twilight zone (OTZ) animals are vital to regulating Earth's climate (Ducklow et al., 2001). However, anthropogenic threats, such as climate change, ocean acidification, pollution, and overfishing pose an imminent threat to OTZ animals. Long-term spatially and temporally intensive observations are essential to our understanding of biodiversity in the OTZ, to resolving global carbon cycles, and to monitoring ocean health. Environmental DNA (eDNA) analysis, which involves studying the trace genetic signatures of organisms (Figure 1), is a promising approach to filling this urgent need. eDNA can be sampled and diagnostic genetic markers ("barcodes") can be sequenced in order to detect the animals inhabiting a given water parcel. Other laboratory protocols (e.g., quantitative PCR, or "qPCR" and "digital droplet PCR") can be applied to facilitate quantitative assessments of specific target species (Eble et al., 2020). In seagoing oceanographic research, eDNA assessment is transitioning from being considered an experimental approach to becoming an established routine that can be scaled up to match ocean observing needs.

TECHNOLOGICAL NEEDS FOR OTZ eDNA ANALYSES
As eDNA analyses are incorporated into mid-and deepwater oceanographic research and observing platforms, new technologies and an improved understanding of eDNA distributions and their relationships to animal distributions and sampling methods are required. Each step of the eDNA analysis process, from experimental design and sample collection to data analysis, impacts the final result ( Figure 2). Of the different steps that could introduce bias (Eble et al., 2020;Shelton et al., 2022), sampling approaches and strategies are particularly underexplored (Govindarajan et al., 2022). FIGURE 1. The ocean twilight zone (OTZ) is home to a diverse faunal assemblage, including fish, crustaceans, and gelatinous animals, all of which leave behind genetic traces (eDNA, conceptualized as particle clouds in the background) as they move through the water. eDNA can originate from sloughed cells and scales, fecal pellets, gametes, tissue fragments from sloppy feeding, and other processes. Animal images are not to scale.

AUTONOMOUS eDNA SAMPLERS AND SAMPLING PLATFORMS
Autonomous sampling with in situ filtration is an alternative approach for sample collection. In situ filtration offers numerous advantages, including reduced labor, fewer opportunities for contamination, larger sample volumes, and fewer experimental design constraints (Yamahara et al., 2019;Govindarajan et al., 2022;Truelove et al., 2022). Autonomous samplers must be mounted on platforms that access the target environment, such as BOX 1. LARGE-VOLUME eDNA SAMPLING ON MESOBOT We coupled large-volume eDNA sampling with the robot Mesobot, a recently developed autonomous underwater vehicle designed for studying midwater regions (Yoerger et al., 2021). Mesobot carries a wide variety of sensors, can track and image small animals and particles, and can carry eDNA samplers in its payload section. Figure B1 shows three 16-sample modules integrated into Mesobot's payload section. The unit on the far right is loaded with cartridges (dark blue) containing collection filters. The sampler has a high filtration rate (approximately 2 liters/minute) and uses large-area filters to accommodate large sample volumes. By sampling larger volumes, more animal taxa can be detected, making this approach well suited for environments where the eDNA signal is dilute and when it is important to detect rare species (Govindarajan et al., 2022). We took advantage of Mesobot's Lagrangian motion capabilities and our large-volume samplers to collect several time series of eDNA samples at a constant depth from the same water parcel to determine the timing of ocean twilight zone animal migrations. Our time-series experiments consisted of continuous, consecutive collection of ~3 0-liter samples before, during, and after the evening migration period. We conducted these experiments during cruises on E/V Nautilus in the Santa Monica Basin in September 2021 and on R/V Endeavor in the Slope Sea in the Northwest Atlantic Ocean in August 2022. Analysis of both data sets is in progress. autonomous underwater vehicles (AUVs), remotely operated vehicles (ROVs), or towed instruments. For example, the Environmental Sample Processor (ESP) has been incorporated into a long-range AUV (Yamahara et al., 2019;Truelove et al., 2022) and can take up to 50 eDNA samples, filtering seawater at a rate of about 1 liter per hour.
In a different approach, a large-volume eDNA sampler that could take 12 samples was mounted on the midwater robot Mesobot to filter seawater at a rate of about 2 liters per minute (Govindarajan et al., 2022).

SAMPLING STRATEGIES
Standard CTD-mounted Niskin bottle sampling strategies typically involve a vertical profile with samples collected at targeted depths. Water is collected instantaneously at a point when the designated bottle is triggered. Animals are patchily distributed and often found in layers, and real-time acoustic backscatter data from an echosounder can be used to guide the selection of sampling depths by identifying concentrations of biomass . Niskin bottles may be mounted on other platforms, such as ROVs or AUVs, permitting greater sampling flexibility (Everett and Park, 2018).
In contrast to instantaneous Niskin bottle sampling, samplers with in situ filtration filter continuously over a FIGURE 3. Typical oceanographic eDNA collection strategy. Seawater in Niskin bottles on a CTD rosette is filtered for eDNA collection. In this example, tubing connects the Niskin bottle spigots directly to filters with submicron pore sizes. Seawater is pumped with multi-head peristaltic pumps so that several samples can be filtered simultaneously. The filtering outflow is collected in carboys to determine the exact volume of water filtered. Photo credit: Erin Frates (WHOI)  (Yoerger et al., 2021), can act as a Lagrangian sampling platform (Box 1). In contrast, non-Lagrangian sampling is integrative. For example, "Eulerian" sampling is a type of non-Lagrangian, integrative sampling where the sampler is at a fixed location (e.g., from a mooring or a stationary vessel), filtering as the water flows past. Sampling is also integrative if the sampler is mounted on non-Lagrangian mobile platforms, such as some AUVs and towed instruments like Deep-See (Box 2). These platforms may traverse water parcels over the duration of sampling, resulting in "crosscutting" spatially and temporally integrated samples. This is the case with powered autonomous vehicles that must maintain forward motion to stay in control while surveying. These often move in a "sawtooth" pattern (e.g., Govindarajan et al., 2015), although the integration can be minimized by containing the vehicle movements within the moving flow. For example, the vehicle can travel forward while turning continuously in tight circles (Truelove et al., 2022). As a result, the vehicle remains with a water mass of known size, rendering it effectively Lagrangian. Integrative sampling also occurs with Deep-See, which, as a towed instrument, moves with the ship, potentially against the prevailing flow and traversing water parcels (Box 2). It is important to recognize that as these approaches (instantaneous, Lagrangian, and non-Lagrangian) may be sampling different entities, they may not be directly comparable.

FILTRATION TIME AND SAMPLE VOLUME
The OTZ environment is immense, and the animals that live there (and their eDNA traces) are patchily distributed. Current eDNA sampling strategies may not yield representative results as they may not match the appropriate temporal and spatial scales of eDNA variation. Field observations have demonstrated that eDNA concentrations decrease with depth and that there can be high variability in the types and proportions of taxa found in sampling replicates (Easson et al., 2020;Govindarajan et al., 2022).
These observations indicate that typical sample volumes (1 to 5 liters) are not aligned with eDNA distributions and that sampling strategies used in shallow and coastal environments may need to be increased for mid-and deepocean waters. Autonomous sampling can facilitate the filtration of larger sample volumes. To obtain these larger volumes while minimizing unwanted spatial and temporal sample integration, higher sampler filtration rates are needed (Govindarajan et al., 2022). Ideally, a sample should be taken within the confines of the target water parcel. However, the size of the target parcel is usually undefined or unknown, given the lack of knowledge about eDNA distributions. Determining optimal sample volume, the number of replicates, and the spatial and temporal sampling frequencies will require further advances in sampling technology, including developing low-cost samplers that can be intensively deployed over large areas. These studies typically find that both methods recover many of the same animal taxa, while also recovering taxa unique to each method. It is important to note that these are "apples to oranges" comparisons that fundamentally measure different entities (organisms vs eDNA) and the volume of water parcels that they are sampling. Net tows are integrative over time and space (i.e., integrative sampling from a mobile, non-Lagrangian platform)-but at a much larger scale than integrative eDNA sampling.

CO-COLLECTION OF eDNA WITH NET TOW AND ACOUSTIC SENSOR DATA
Net tows sample very large volumes, covering hundreds or thousands of meters over the course of an hour or more, and sampling orders of magnitude more water . As such, eDNA and net tow comparisons should be viewed as complementary, not as "calibrations" of each other.
Data collected with echosounders, which measure acoustic backscattering from OTZ animals, can be used to estimate distribution and biomass and are also complementary to eDNA sampling. Both data types are especially valuable for the OTZ, where migrating biomass can be identified from the movements of sound-scattering layers.
Typically, shipboard echosounders for OTZ studies operate at frequencies of 18 kHz and 38 kHz. Real-time observations of acoustic backscatter can be used to guide eDNA sampling Govindarajan et al., 2021; Box 2). Conversely, eDNA data can potentially provide taxonomic resolution for interpretation of acoustic signals, which at these frequencies are typically dominated by animals with gas-bearing structures such as fish with swim bladders and siphonophores (Lavery et al., 2007), and may identify the broader animal communities associated with these signals. Traditionally, depth-stratified net tows have been used for this purpose. However, net tows may integrate across acoustic layers, and animals such as active swimmers and delicate gelatinous species are undersampled. eDNA can potentially provide more spatially precise information about species occurrence relative to the observed acoustic backscatter.

INTERPRETING eDNA SIGNALS
An essential issue for eDNA observing is the ability to interpret a signal appropriately. eDNA is collected from a dynamic fluid medium, and it is important to understand whether the resulting genetic signatures reflect the biodiversity from the sampling location or elsewhere. This is critical for stand-alone assessments and when the eDNA data are meant to inform co-collected data (e.g., acoustics). Laboratory studies on eDNA persistence are especially valuable for understanding the fate of eDNA and, consequently, the origin of eDNA signals. The rate at which eDNA decays (which affects how long the signal can be detected) is primarily controlled by temperature (Allan et al., 2020;McCartin et al., 2022). Thus, the cold temperatures of the OTZ and deeper waters likely result in greater BOX 2. ADAPTIVE eDNA SAMPLING AND ACOUSTICS WITH DEEP-SEE We paired autonomous eDNA sampling with the towed instrument Deep-See ( Figure B2a) to enable the co-collection of eDNA, acoustics, and imaging data (Lavery et al., 2019). Deep-See combines wide-band, split-beam acoustics (1-500 kHz) with optical and environmental sensors and enables acoustic and image data collection from mesopelagic depths. We mounted one 16-filter large-volume sampler unit on the tail section of Deep-See ( Figure B2b). During an August 2022 cruise on NOAA Ship Henry B. Bigelow to study the ocean twilight zone in the Slope Sea in the Northwest Atlantic, we conducted adaptive eDNA sampling by triggering sampling from a shipboard computer in response to observed acoustic backscatter from both Deep-See ( Figure B2c) and the shipboard echosounder ( Figure B2d). Unlike the narrowband signals used with shipboard echosounders, the wide-band capabilities of Deep-See allow the spectral responses of individual animals to be measured. To identify eDNA signatures associated with these unique spectral signatures, we are analyzing samples collected inside and outside observed biomass patches. The ability to take eDNA samples adaptively in response to real-time acoustic backscatter will allow us to better target patchily distributed biomass and transient phenomena such as vertical migration behavior. Sampling inside and outside of biomass patches could also potentially facilitate understanding of the relationship between eDNA metabarcoding results and biomass. High-resolution sampling with a suite of samplers and diverse platforms is superimposed on 38 kHz acoustic backscatter data obtained over a 24-hour period (data credit: Andone Lavery, WHOI). The acoustic backscatter data demonstrate diel vertical migration of a significant portion of the OTZ biomass. At dusk, a portion of the biomass is seen transiting from mesopelagic to shallow depths. The biomass moves back to mesopelagic depths at dawn. Diverse approaches to eDNA sampling will help to determine the depth distributions and migratory behaviors of different species. eDNA persistence and signal transport than do the warmer waters of coastal systems. However, further investigation is needed into other factors that may influence eDNA persistence, such as the state of the eDNA (i.e., intracellular vs. disassociated), the rate of microbial degradation, interactive effects among the microbial community, and abiotic conditions (e.g., Jo and Minamoto, 2021).
A particular challenge for interpreting eDNA results from the OTZ is the vertical migration behavior of many mesopelagic species. The regular, daily transit of these animals between surface waters and depth may result in their genetic traces being left throughout the water column, making it difficult to ascertain their sources. Allan et al.
(2021) conducted a modeling study using realistic parameters for a temperate system to address this issue. They found that eDNA signatures remain close to their depth of origin despite the potential for movement. This finding supports the use of eDNA to study vertical phenomena in the mesopelagic, including identifying which taxa migrate (Canals et al., 2021)

86
As sampling technologies develop, it is important to simultaneously pursue a better understanding of how eDNA is distributed in the ocean. This understanding is necessary both to guide the development of sampling technologies and to shape sample collection strategies.
Computational modeling studies that consider the influence of oceanographic environments (e.g., temperature and flows) and incorporate eDNA shedding, persistence, transport, and dispersal are critical for linking measured eDNA signatures with their sources and for inferring species distributions and biomasses. They should be coupled with lab and field experiments for calibration, data analysis, and designing representative and effective field sampling strategies. For inferring biomass of multiple species (e.g., quantitative metabarcoding), incorporating an understanding of PCR dynamics into overall data interpretation will be especially important (Shelton et al., 2022). More research is needed to determine appropriate spatial and temporal sampling scales, which may depend on regional ocean currents; the biology of the target organisms; and specific research questions. In most, if not all, cases, we will need to vastly scale up our sampling efforts to enable spatiotemporally resolved data and obtain the replication required for statistical analyses. These steps are crucial for drawing scientific conclusions and translating eDNA observations into actionable conservation and policy insights.