Effects of consumer surface sterilization on diet DNA metabarcoding data of terrestrial invertebrates in natural environments and feeding trials

Abstract DNA metabarcoding is an emerging tool used to quantify diet in environments and consumer groups where traditional approaches are unviable, including small‐bodied invertebrate taxa. However, metabarcoding of small taxa often requires DNA extraction from full body parts (without dissection), and it is unclear whether surface contamination from body parts alters presumed diet presence or diversity. We examined four different measures of diet (presence, rarefied read abundance, richness, and species composition) for a terrestrial invertebrate consumer (the spider Heteropoda venatoria) both collected in its natural environment and fed an offered diet item in contained feeding trials using DNA metabarcoding of full body parts (opisthosomas). We compared diet from consumer individuals surface sterilized to remove contaminants in 10% commercial bleach solution followed by deionized water with a set of unsterilized individuals. We found that surface sterilization did not significantly alter any measure of diet for consumers in either a natural environment or feeding trials. The best‐fitting model predicting diet detection in feeding trial consumers included surface sterilization, but this term was not statistically significant (β = −2.3, p‐value = .07). Our results suggest that surface contamination does not seem to be a significant concern in this DNA diet metabarcoding study for consumers in either a natural terrestrial environment or feeding trials. As the field of diet DNA metabarcoding continues to progress into new environmental contexts with various molecular approaches, we suggest ongoing context‐specific consideration of the possibility of surface contamination.


| INTRODUC TI ON
Biological communities and ecosystem function are shaped by interactions between organisms (Hooper et al., 2005). Among the many interaction types, consumptive interactions (including herbivory, predation, and parasitism) can shape the stability of biologically diverse communities (Delmas et al., 2019). Until recently, consumptive interactions were most often measured by visual observations of feeding or by gut dissection or inspection of fecal contents (Baker et al., 2014;Nielsen et al., 2018), which made it challenging or impossible to conduct diet analyses for many consumer groups. Specifically, these diet analyses are not possible for consumers that (a) are too small for dissection and food identification and (b) have feeding habits or food items which make diet visually unidentifiable (Sheppard & Harwood, 2005). This group of consumers, including terrestrial insects, spiders, and other arthropods, form the base of most terrestrial food webs and are integral to maintaining biodiversity and ecosystem functioning in ecosystems worldwide (Wilson, 1987). For these consumer groups, the use of high-throughput sequencing is one of the most promising emerging approaches for determining gut contents. High-throughput sequencing (hereafter referred to as "diet DNA metabarcoding") can identify a suite of diet species at once and provides a comprehensive and efficient method for determining intrapopulation, intraspecific, and interspecific diets (Lucas et al., 2018;Pompanon et al., 2012;Quéméré et al., 2013;Soininen et al., 2015). These methods have already illuminated new interactions and ecological trends in a variety of environments (e.g., host-parasitoid: (Wirta et al., 2014); plant-herbivore: (Kartzinel et al., 2015); host-parasite: (Schnell et al., 2012); and predator-prey: (Toju & Baba, 2018).
As diet DNA metabarcoding methods continue to advance, however, they need to be validated so that the ecological inference made from them is robust. Focusing on the challenges of small organisms where small body size has limited other diet analysis methods, DNA diet analyses are often performed on full organisms or body parts without gut dissection (e.g., Jacobsen et al., 2018;Toju & Baba, 2018). The necessity to use full organisms or body parts increases the possibility of surface contamination, altering detection and species composition of presumed diet items. Surface sterilization, the use of chemical treatments or physical action to remove surface contaminants, is systematically used in other fields to reduce the risk of contamination in DNA metabarcoding datasets (Burgdorf et al., 2014;Zimmerman & Vitousek, 2012). However, surface sterilization has not been systematically used in diet metabarcoding studies. While some fields have developed informed protocols based on decades of research into best practices and study-specific considerations (Brown et al., 2018), the field of diet DNA metabarcoding has not developed a similarly systematic approach (e.g., ethanol : Doña et al., 2019, bleach: Anslan et al., 2016, and no sterilization: Jacobsen et al., 2018Wirta et al., 2014). The lack of systematic surface sterilization in diet DNA metabarcoding when using full individuals or body parts may be due to the desire to avoid DNA destruction in relatively permeable animal cells (Greenstone et al., 2012). However, without considering surface sterilization as a treatment for surface contamination, we have limited ability to confidently assign DNA sequences to ingested diet items (Greenstone et al., 2011(Greenstone et al., , 2012Linville & Wells, 2002).
In this study, we look at the effects of surface sterilization to remove surface contaminants on our understanding of consumer diets where the DNA of full body parts (no internal dissection) is used for diet DNA metabarcoding. Targeting the CO1 gene region, we produced high-throughput sequencing results from the full body parts (opisthosomas without gut dissection) of an invertebrate consumer species (the spider, Heteropoda venatoria). We surface sterilized half of the consumers prior to DNA extraction using a series of washes in a 1:10 dilution of bleach (10% commercial bleach) and deionized water; we left the other half of consumers unsterilized. We first determined how surface sterilization to remove contaminants impacts presumed diet from consumers collected in their natural environment, comparing surface sterilized individuals to those which were not surface sterilized, to ask whether surface sterilization influences (a) detection, (b) rarefied abundance, (c) richness, and (d) composition of potential diet items. We then performed a laboratory feeding trial, comparing surface sterilized individuals to those which were not surface sterilized to ask whether surface sterilization influenced (a) detection or (b) rarefied abundance of offered diet items. Exploring these questions in natural and contained settings addresses whether surface contamination alters interpretations of feeding interactions and thus whether it should be incorporated into standard protocols in diet metabarcoding.

| Field site and collections
We conducted fieldwork on Palmyra Atoll National Wildlife Refuge, Northern Line Islands, USA (5°53′N, 162°05′W). Palmyra Atoll has a well-characterized species list and is relatively species poor, allowing for relatively complete characterization of consumer and diet items (Handler et al., 2007). We targeted a generalist, active hunting spider species (Heteropoda venatoria) because (a) it occurs in high abundance on the atoll and is easy to collect, (b) it is a generalist species that feeds on a wide suite of organisms (including spiders, other invertebrates, and two geckos in the genus Lepidodactylus), and (c) it is the only species in its family on the atoll, meaning consumer DNA can be differentiated from potential diet DNA. All individuals were stored individually in sterilized containers (Greenstone et al., 2011).

| Natural environment consumer collection
In 2015, we collected consumers (n = 47) from natural environments, which had fed on available diet items and come into contact with environmental surfaces, to test whether DNA metabarcoding detects diet DNA effectively. Consumers were collected at night via eye shine while they were actively hunting. We collected the first individuals we observed in each survey period and so they represent the distribution of body size and population demographics of this species that actively hunt in that environment. We froze all individuals at −80℃ immediately following collection until surface sterilization and DNA extraction in 2019.

| Feeding trial consumer setup and feeding
In 2017, we conducted laboratory trials (n = 26) to test whether DNA metabarcoding detects DNA from diet items offered in a contained environment. We created feeding environments from one-liter plastic yogurt containers with holes for air transfer and placed one H. venatoria in each container. After 12 hr, we placed one large grasshopper (Oxya japonica, a likely diet item (Handler et al., 2007)) in each container and left all containers for 24 hr. We then froze (−20℃) each H. venatoria that had killed the grasshopper (n = 25, consumption was not easily detectable and thus not considered in analyses).
We cleaned all containers between trials with 10% bleach solution.
To test surface sterilization's efficacy at removing possible contaminants, we used a surface sterilization treatment (Burgdorf et al., 2014;Schulz et al., 1993) on ~half the consumers for each set: those collected from the natural environment and those subjected to controlled feeding trials. We submerged and stirred each (whole) consumer in 10% commercial bleach by volume for 2 min and washed each in deionized water for 2 min. Similar bleach submersion leads to undetectable DNA degradation in similar soft-exoskeleton consumers (Greenstone et al., 2012;Linville & Wells, 2002). Natural environment consumers (2015) had been frozen at −80°C since collection; we surface sterilized these consumers in a sterilized laminar flow hood in 2019 just before DNA extraction (n = 22 surface sterilized, n = 25 not surface sterilized; Table 1). We surface sterilized feeding trial consumers (2017) in the laboratory on the atoll in 2017 following freezing at −20℃ and then stored each in individual vials of 95% ethanol in a −20°C freezer until DNA extraction (no −80°C freezer was available at the field station that year) (n = 10 surface sterilized; n = 14 not surface sterilized). Prior to DNA extraction, we dried all samples for 1-3 hr in a sterilized laminar flow hood and then removed the full opisthosoma (containing the hind gut region) using a sterilized scalpel. Between all steps, tools were sterilized with either ethanol and flame (scalpels and forceps) or 10% bleach (surfaces) between handling each individual.

| DNA extraction and removal of consumer DNA with AMPure XP beads
We extracted DNA from each consumer following a modified CTAB extraction protocol (Fulton et al., 1995). We quantified DNA using a Qubit (Invitrogen) fluorometer with the high sensitivity doublestranded DNA quantification kit. We followed Krehenwinkel et al. (2017) to isolate a proportion of lower molecular weight DNA with AMPure XP beads prior to PCR (Appendix S5, Figure S1). We diluted each DNA sample to 20ng/μl (creating a total sample volume of 40μl), mixed each sample using AMPure XP beads (0.75x beadto-DNA ratio), and kept the supernatant. With the supernatant, we precipitated the DNA pellets with isopropanol and 5 M potassium acetate and washed DNA pellets with ethanol (Appendix S6). We quantified this cleaned DNA again using a Qubit fluorometer and diluted all samples to 10 ng/μl prior to PCR steps. All DNA pellets were stored in and diluted with TE buffer.
We performed all PCR preparation steps in a UV-sterilized biosafety cabinet. We used PCR volumes of 25μl (9μl nuclease free water, 12.5μl GoTaq Green Master Mix (Promega Corp.), 1.25 μl of each of the primers (at 10 mM), and 1 μl of DNA template (at 10 ng/μl)). We ran each sample in duplicate along with duplicated negative samples each PCR run. PCRs are as follows: initial denaturation step at 95℃ for 3 min and then 35 cycles of (a) 95℃ for 30 s, (b) 46℃ for 30 s, and (c) 72℃ for 1 min, followed by a final 5 min at 72℃. We cleaned PCR products with AMPure XP beads at a 0.8x bead-to-DNA ratio and resuspended from beads using a 10 mM TRIS buffer.
We attached Illumina index primers with an additional PCR step following standard protocols (Nextera XT Index Kit v2, Illumina, 2019). We combined duplicate samples for which both duplicates successfully amplified and diluted to a concentration of 5 nM. We multiplexed all samples with one negative control and two fungal clone positive controls (GenBank accession numbers: MG840195 and MG840196; Apigo & Oono, 2018;Clark et al., 2016;Toju et al., 2012

| Sequence merging, filtering, and clustering with UNOISE3
We merged, filtered (max ee = 1.0), and denoised (clustered) our sequences around amplicon sequence variants (ASVs) using the UNOISE3 algorithm (unoise3 command in the open-source USEARCH 32-bit version 11.0.667; Edgar, 2016, Appendix S5, Figure S3). Prior to denoising with UNOISE3, we used cutadapt (version 1.18, Martin, 2011) to remove primers from each sequence. We also repeated analyses with the DADA2 algorithm run through R (dada2 package version 1.1.14.0; Callahan et al., 2016) and with a data cleaning step run through BBSplit (Bushnell, 2019) to remove consumer DNA prior to ASV assignment (because ASV assignment is abundance-sensitive). We considered analyses from the UNOISE3 algorithm only because UNOISE3 assigned more sequence reads to positive controls than DADA2 (on average, 3× as many reads per positive control) and the cleaning step paired with either DADA2 or UNOISE3 did not increase potential diet DNA detection (summary and comparisons in Appendices S1 and S2).
We created a list of unique ASVs and a matrix of ASV abundances across samples. We matched ASVs to taxonomies in the GenBank and BOLD databases. For GenBank, we used BLAST

| Detection of potential diet items
For consumers from both natural environment and feeding trials, we asked whether surface sterilization altered detection of potential diet items for each consumer. For natural environment consumers, we examined all potential diet items (which could represent either diet or surface contaminants). For feeding trial consumers, we fo- As all ASVs received family-level taxonomic assignment, we pooled ASVs that matched at the family level into one taxonomic unit using cumulative read abundance (i.e., all ASVs matched to diet family A were pooled into diet family A taxonomic unit), a practice common in diet metabarcoding (Kartzinel et al., 2015) and predator-prey interaction (Brose et al., 2019) studies.

| Statistical analyses
For potential diet detection and rarefied abundance in both sets of consumers (natural environment and feeding trial), we used Fol-degen-rev GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGTANACYTCNGGRTGNCCRAARAAYCA Leray et al. (2013) generalized linear models to assess the effect of surface sterilization treatment. For prey detection, we used all potential (natural environment) or offered (feeding trial) diet item detection (presenceabsence per sample) as the response variable in the full model with surface sterilization as a fixed effect and a binomial distribution.
For rarefied diet abundance, we only assessed consumers for which we had detected diet and not those with no diet detection (n = 33 of 37 for natural environment; n = 14 of 19 for feeding trials). For for heteroscedasticity, and for count models (Poisson or negative binomial), zero inflation and overdispersion (Bolker et al., 2009;Zuur et al., 2009). We performed the CCA using the vegan package in R, comparing a model with surface sterilization as a fixed effect to a null model using an ANOVA. All raw data, data cleaning, and data analyses are available online (Miller-ter Kuile, 2020a, 2020b, and model outputs for primary and supplemental models can be found in Appendices S3 and S4.

| PCR success, sequence merging, filtering, and clustering with UNOISE3 and DADA2
We successfully extracted DNA from 100% of samples (n = 72).  (Saitoh et al., 2016). There were no conflicting taxonomic assignments at the family level or higher between the BOLD and BLAST assignments.

| Detection of potential diet items
We detected potential diet in 89% ( We detected offered prey in 50% of consumers that had been surface sterilized compared to 91% of those consumers that were not surface sterilized.

| Proportion of potential diet DNA
For natural environment consumers, potential diet rarefied DNA sequence reads represented 2.0% (±1.0%) of total per-sample DNA sequence abundance (Figure 2). In feeding trial consumers, offered diet DNA sequence reads represented 0.8% (±0.7% SE) of total persample DNA sequence abundance. For both natural environment and feeding trial consumers, the null models that did not include surface sterilization treatment as a fixed effect were the best models of diet DNA read abundance. (b) Detection of offered diet (Oxya japonica) DNA in feeding trial consumers that were and were not surface sterilized. While the best-fitting model based on AICc values indicated an effect of surface sterilization treatment (a decrease from 91% without surface sterilization to 50% with surface sterilization), the effect of this term in the model was statistically unclear (p-value = .07)

F I G U R E 2
Neither the (a) proportion of total potential diet DNA in natural environment consumers or the (b) proportion of offered diet item DNA in feeding trial consumers significantly changed with surface sterilization treatment F I G U R E 3 In natural environment consumers, surface sterilization did not alter per-sample diet richness of either familylevel or ASV-level taxonomic units models which did not include surface sterilization treatment as a predictor (Figure 4, Figure S1). Diet families came from insect, arachnid, and centipede orders (insects: Diptera (5) Geophilomorpha (1), Figure 4). with food web studies in this field, e.g., Brose et al., 2019) and when considering richness of molecular taxonomic units (ASVs). We detected diet across 84% of the total consumers in our study (n = 47 of 56), including 20 diet families. Diet DNA metabarcoding has high potential to contribute diet information for small consumers with cryptic feeding habits. Furthermore, it appears that current protocols that do not include surface sterilization steps are sufficient to determine potential diet for these consumers.

| D ISCUSS I ON
The field of diet DNA metabarcoding has not universally adopted surface sterilization practices into common protocols, in particular for studies including DNA extraction of full organisms or body parts without dissection (e.g., Jacobsen et al., 2018;Wirta et al., 2014).
We demonstrate that surface sterilization does not seem necessary to avoid contamination effects. The evident lack of the effects of surface contaminants in our study contrasts with obvious surface contaminants that alter ecological interpretations in other fields using high-throughput sequencing to determine community diversity, particularly fungal endophyte studies (Burgdorf et al., 2014).
One reason for this difference may be that fungal spores are widespread on and in the surfaces of most environments and organisms (Després et al., 2012) and likely to contaminate studies targeting specific subgroups of these communities. Indeed, even in our dataset, some sequences matched to fungal taxonomies. The fact that these nontarget sequences did not alter our DNA metabarcoding data by hiding target diet DNA, even with the relative rarity of diet DNA compared to consumer DNA (0.006%-26% of each sample), is likely due to differences in biomass of these sources of DNA in our samples and the specificity of our DNA size-selection protocol and PCR primers (Elbrecht et al., 2017;Krehenwinkel et al., 2017).
Therefore, our results are promising both in validating the robustness of findings from past diet DNA studies that have not implemented surface sterilization treatments, but also highlight that diet DNA metabarcoding using broad, universal primer sets (e.g., those in this study) is an effective tool even when DNA sequence data contain potential environmental contaminants (Appendix S5, Figure S5).
While we saw no widespread support of the necessity for surface sterilization in our study, a model from the feeding trial that includes surface sterilization performed slightly better than one without this treatment (ΔAICc = 1.59). Thus, it is possible that contained environments may be more prone to contamination than open terrestrial environments. We see this result as an ideal starting point for next steps in validating diet DNA metabarcoding in similar contexts.
Specifically, because this study had a relatively limited sample size (n = 8 and 11 in each sterilization treatment group) and because we did not confirm ingestion, a similar trial including crossed treatments of sterilization with different forms of diet item contact (e.g., Greenstone et al., 2012) would provide additional evidence of the effects of surface sterilization or surface contamination. Further exploration of these results might reveal that the decision to surface sterilize prior to diet DNA metabarcoding may matter more in some F I G U R E 4 For natural environment consumers, surface sterilization did not alter the composition (either with a presenceabsence of abundance model) of potential diet items of either family-level taxonomic units or ASV-level taxonomic units. In this figure of family-level taxonomic units by surface sterilization treatment, presence is indicated by a colored box and abundance is indicated by color depth (divided by quartiles due to wide variation in DNA sequence abundance) environments and experiments than others (e.g., where diet items are in high density or consumers have long handling times (Abrams & Ginzburg, 2000;Samu & Biro, 1993). Furthermore, as earlier studies targeting particular consumer diet pairs explored (e.g., Greenstone et al., 2012), the field of diet DNA metabarcoding is ripe for a comparison of surface sterilization techniques.
Diet DNA metabarcoding can empirically provide diet descriptions for a suite of consumers important to food web ecology and the maintenance of biodiversity on the planet (Stork, 2018).
Characterizing consumptive interactions for small, cryptic species for the first time will build a better picture of nature's complexity and allow ecologists to confidently query how species interactions will change with continued anthropogenic disturbance (Tylianakis et al., 2008). Like any method for determining consumptive interactions in nature, DNA metabarcoding continues to be refined, especially as tools and data emerge (Krehenwinkel et al., 2019;Kvist, 2013). This study builds on past efforts to refine diet DNA metabarcoding by using surface sterilization to pinpoint potential sources of error in diet DNA data. Here, we found that, on the whole, surface sterilization seems unnecessary in two contexts (terrestrial environments and contained feeding trials) when extracting DNA from body parts of invertebrate taxa. Continued context-specific refinement of surface sterilization protocols, and of other steps in diet DNA metabarcoding, will improve the widespread utility of diet DNA metabarcoding across consumer groups and environments. iting this manuscript. We thank four anonymous reviewers for their help revising this manuscript. This is publication number PARC-160 from the Palmyra Atoll Research Consortium.

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw sequence data are available on GenBank (BioProject: PRJNA639981). Cleaned sequence data and analyses are available on Dryad (DOI: https://doi.org/10.5061/dryad.gqnk9 8snc).