Scoop That Poop: Optimising Faecal Sample Pre‐Processing for Parasite Metabarcoding

ABSTRACT

Molecular techniques such as DNA metabarcoding are increasingly used to characterise parasite communities. However, relatively few studies have examined how sample processing methods influence detection rates. We used faecal samples collected from free‐roaming horses to evaluate how parasite density and pre‐processing methods influenced quantification of parasite richness, community composition, and detection of different taxa. Methods that concentrated parasites substantially increased parasite detection, especially at the low infection levels often seen in wildlife. Specifically, we found that DNA extracts from larval coproculture and a newly developed egg concentration approach detected approximately twice as many species and genera as extracts made directly from faecal matter. Although parasite richness was consistently lower in these faecal subsamples, overall parasite communities were still similar between pre‐processing methods. Ultimately, the optimal method depends on research constraints and goals. Working with parasite larvae is more time intensive, but lower cost as larval parasites can be extracted using a lysis buffer approach which performs similarly to commercial extraction kits. DNA extraction from faecal subsamples misses rare and common taxa but minimises field processing. Our novel egg concentration method offers a compromise between relatively rapid processing and high sensitivity.

1. Introduction

All animal species host parasites which potentially impact animal health (Coulson et al. 2018; Hasik and Siepielski 2022), population stability (Dobson and Hudson 1992; Pedersen and Greives 2008), species interactions (Graham 2008), and ecosystem processes (Fischhoff et al. 2020; Koltz et al. 2022). Reliable infection detection is critical for understanding disease in natural systems, managing wildlife populations (Belant and Deese 2010), and predicting human disease risk (Farrell et al. 2013). However, detection can be challenging in wild animals, especially when individuals cannot be captured (Ryser‐Degiorgis 2013). As a result, a variety of non‐invasive approaches have been developed to detect infection (Bergner et al. 2019; Miller et al. 2024). With their expanding use in ecological research, it is increasingly necessary to validate and optimise these methods, especially for helminth parasites where both infection burden and egg shedding can be highly variable.

Historically, most non‐invasive assays have relied on visual identification of eggs or larvae in host faeces (Zajac and Conboy 2006). For example, in faecal egg counting methods, floatation solutions are used to separate eggs from heavier debris prior to identification under a microscope (Bosco et al. 2018; Ghafar et al. 2021). For parasites that develop into free‐living larvae, eggs can be cultured from host faeces allowing resulting larvae to be identified based on morphology. While more species can be identified using this coproculture approach, it is time intensive, requires highly trained observers, and still provides limited taxonomic resolution, especially in systems without identification keys (Roeber and Kahn 2014). Although microscopy‐based assays remain widely used (Ezenwa et al. 2021; Hayward et al. 2022), sequencing‐based approaches offer a powerful alternative for parasite identification and quantification (Titcomb et al. 2019).

Among available sequencing approaches, metabarcoding is increasingly used for characterising parasite communities (Miller et al. 2024). For example, recent studies using this technique have shown that more closely related hosts harbour related parasites (Gogarten et al. 2020; Titcomb et al. 2022), wild ruminants are parasite reservoirs for domestic livestock (Barone et al. 2020), and parasitism is associated with altered microbiomes in wild rodents (Kim et al. 2022). Metabarcoding is more sensitive than microscopy‐based approaches and, for many parasite groups, has better taxonomic resolution (Davey et al. 2021, 2023). However, methodology varies substantially between studies. Although differences in sample storage, DNA extraction, and primer selection have been explored for metabarcoding of parasites (Ayana et al. 2019; Bourret et al. 2021; Davey et al. 2023) and other targets (e.g., Pollock et al. 2018; Ando et al. 2020; Kumar et al. 2020), the effects of sample pre‐processing (i.e., steps taken to prepare samples before DNA extraction) on parasite detection have received relatively little attention (but see Pafčo et al. 2018; Davey et al. 2021).

In parasite metabarcoding, several approaches are commonly used for sample pre‐processing. Studies targeting parasites with motile larvae often use methods such as coproculture or Baermann's method to collect larvae, which are then concentrated and rinsed prior to DNA extraction (e.g., Aivelo et al. 2015; Barone et al. 2020; Poissant et al. 2021). Others concentrate eggs using methods that typically include an initial centrifugation step to suspend eggs in a floatation solution (Davey et al. 2021; Abbas et al. 2023). Finally, many studies simply extract DNA from faeces (e.g., Mitchell et al. 2019; Gogarten et al. 2020; Titcomb et al. 2022; Davey et al. 2023).

Methods that concentrate eggs or larvae may be more sensitive than those that do not, particularly when parasite burdens are low. However, concentration approaches typically require samples to be pre‐processed shortly after collection and each method has limitations. For example, coproculture misses taxa that do not develop or migrate out of faeces (Roeber and Kahn 2014; Pafčo et al. 2018). Pre‐processing methods also influence DNA extraction protocols as isolated larval parasites can be extracted using a low‐cost lysis buffer approach (Avramenko et al. 2015) while samples containing eggs and faeces typically require commercial kits to lyse eggs and remove PCR inhibitors (Miller et al. 2024). Although parasites have been detected using a variety of approaches, it is unclear whether different methods generate comparable information.

To test how pre‐processing approaches affect parasite detection, we used parasite communities from free‐roaming horses. Horse parasite communities are dominated by a diverse assemblage of strongylid nematodes (Chapman et al. 2003; Lichtenfels et al. 2008), with worm burden and community composition varying between individuals and populations (Wood et al. 2013; Sargison et al. 2022; Abbas et al. 2023; Ahn et al. 2024). Although some strongylids (e.g., Strongylus vulgaris) can cause high pathology (McCraw and Slocombe 1976; Love et al. 1999), most free‐roaming equids appear to be relatively tolerant of infection and often have high parasite loads (Harvey et al. 2019; Jenkins et al. 2020). These parasites are well represented in reference databases (Workentine et al. 2020) and metabarcoding targeting the internal transcribed spacer (ITS2) region of the rDNA is increasingly used to characterise nematode communities in horses (e.g., Poissant et al. 2021; Ahn et al. 2024; Ochigbo et al. 2025). Combined, the high parasite diversity, individual variation, and a well‐established metabarcoding region make free‐roaming horses an ideal system for testing how pre‐processing methods and egg density influence parasite detection.

Responding to calls for continued optimisation of parasite metabarcoding (Aivelo and Medlar 2018; Poissant et al. 2021; Miller et al. 2024), we tested three approaches for preparing faecal samples for parasite metabarcoding. Specifically, we tested how larval coproculture and a new, simplified egg concentration method compared to DNA extraction directly from faeces. For larval parasites, we also tested whether lysis buffer extractions performed similarly to commercial DNA extraction kits. We predicted that concentrating parasites would increase detection sensitivity, particularly for rare taxa or samples with low egg densities. Our results provide insight into the best methods for pre‐processing samples for parasite metabarcoding and highlight the limitations of using faecal subsamples.

2. Materials and Methods

2.1. Sample Collection

To test how sample pre‐processing influenced parasite detection, we collected faecal samples from free‐ranging horses near Vernal, UT (40.1636° N, 109.2177° W). We collected samples into clean plastic bags and vacuum‐sealed each sample the night of collection to prevent parasite development. Samples were stored in a cooler in the field and initial pre‐processing began within 48 h of collection. To test pre‐processing methods, we collected samples from 13 horses over 2 days in January 2024 and, to test DNA extraction methods, we collected samples from eight horses on 1 day in January 2025.

2.2. Parasite Egg Counts

To determine egg density in each sample, we measured eggs per gram (EPG) of faeces using both triple‐chamber McMaster slides (8 EPG sensitivity) and double centrifugation faecal floats (1 EPG sensitivity). For McMaster counts, we mixed 4 g of faeces with 26 mL of Sheather's sugar solution (specific gravity: 1.27), removed large particulates using a tea strainer, pipetted a well‐mixed subsample into the McMaster slide and then counted eggs under 100× magnification. For double centrifugation floats, we mixed 3 g of faeces with ~8 mL of water, strained the mixture as above and then centrifuged the solution at 1500 rpm for 5 min. We removed the supernatant, resuspended the washed faecal material in Sheather's sugar solution, placed a coverslip on top of the tube before mixing, and then centrifuged again at 1500 rpm for 5 min to force eggs to the top of the solution. We mounted coverslips onto microscope slides and counted eggs under 100× magnification. As faecal water content can influence EPG estimates, we calculated faecal dry weights by drying approximately 7 g of each sample at 55°C. We weighed faeces at 4 and 10 days and then calculated the percent water in each sample.

2.3. Sample Pre‐Processing

To compare the three pre‐processing methods (Figure 1), we homogenised each sample from the 2024 collection by thoroughly kneading the sealed bags. From each bag we then froze a 150 mg subsample at −25°C, prepared a second sample using an egg concentration approach, and prepared a third sample using coproculture to concentrate parasite larvae, as described below. Hereafter, we refer to these three subtypes of samples as (1) faecal subsample (no pre‐processing), (2) eggs (egg concentration), and (3) larvae (coproculture). To determine whether lysis buffer extractions produced the same results as commercial DNA extraction kits, we homogenised each sample from the 2025 collection and used this material to prepare two larval coprocultures from each host.

FIGURE 1.

To compare the three pre‐processing methods, we (A) collected fresh faeces from free‐roaming horses and then homogenised each sample. (B) From each bag, we pre‐processed one sample using larval coproculture (‘larvae’), one sample using an egg concentration (‘egg’) approach, and preserved one subsample without additional pre‐processing (‘faecal subsample’). (C) We extracted DNA using a commercial kit to compare larvae, egg, and faecal subsamples. For larval DNA extractions, we also compared the commercial kit to the Single Worm Lysis (SWL) buffer approach. (D) All samples were then sequenced using Next Generation Sequencing.

2.3.1. Egg Concentration via Double Centrifugation

To collect a concentrated egg sample from faeces, we modified the double centrifugation approach commonly used to quantify gastrointestinal parasite loads (Zajac and Conboy 2006). We mixed 4 g of faeces and ~8 mL of water and then strained the mixture to remove large particulates. We centrifuged the strained solution for 5 min at 1500 rpm, removed the supernatant, and then resuspended the washed faecal pellet in 13 mL of Sheather's sugar solution. This faeces‐sugar solution was mixed and then centrifuged again for 5 min at 1500 rpm to force eggs to the top of the solution. We collected material from this top layer (0–10 mm depth) using a sterile cotton swab (Puritan HydraFlock Sterile Flocked Collection Devices, 25‐3206‐H) and stored swabs at −25°C until DNA extraction.

2.3.2. Larval Coproculture

To collect larval parasites, we cultured faecal samples following standard methods (Avramenko et al. 2015) with minor modifications. For each sample, we mixed ~40 g of faeces with an equal volume of vermiculite and enough water to obtain a slightly tacky consistency. We covered samples with a loose‐fitting lid and incubated at 25°C, adding water as needed to prevent desiccation. After 3 weeks, we harvested larval nematodes (L3s) by filling cups with warm water, placing a glass petri dish on top, and then inverting the cup into the dish. We added additional water to the petri dish and left samples overnight for L3s to migrate into the dish. We collected this liquid and estimated the number of L3s using a stereo microscope. To concentrate L3s, we transferred solution to a 50 mL falcon tube and chilled it at 4°C for at least 1 h before centrifuging at 5820 rcf for 5 min. We reduced the volume to 10 mL and then added ~40 mL of near‐boiling water to kill L3s. After chilling again at 4°C, we centrifuged the sample at 5820 rcf for 5 min and then reduced the volume to 5 mL. At this point, we quantified the number of L3s in three 30 μL subsamples using a compound microscope. We concentrated the remaining sample to 0.3 mL using multiple rounds of centrifugation at 13000 G for 3 min. We then transferred L3s into Single Worm Lysis buffer (SWL solution; 121 mg tris base, 380 mg KCl, 51 mg MgCl2, 460 μL NP‐40 alternative, 460 μL tween‐20 and water to a final volume of 100 mL). We added 1 mL of buffer to each sample, vortexed to mix, centrifuged at 13000 G for 4 min and then reduced the volume to 0.3 mL. We repeated this rinse twice to remove residual water before reducing samples to a final volume of 0.1 mL and storing at −80°C.

2.4. DNA Extraction, Library Preparation, and Sequencing

To compare the three pre‐processing methods, we extracted frozen faecal subsamples, larvae, and egg swabs using the Zymo Quick‐DNA Faecal/Soil Microbe DNA Miniprep Kit, including a swab control, SWL buffer control, and three extraction blanks. To ensure extraction of all material from larval and egg samples, we rinsed each storage tube with 750 μL of BashingBead Buffer and then transferred this buffer to the BashingBead Lysis Tube for extraction with its respective egg swab or larval sample.

To compare DNA extraction methods, we extracted the paired L3 samples using either the above kit or the lysis buffer approach (Avramenko et al. 2015). For lysis buffer extractions, L3s in 100–150 μL of SWL buffer were incubated on a thermomixer for 15 min at 95°C and 1000 rpm, frozen at −80°C for at least 1 h and then defrosted on ice. We added 6 μL of Proteinase K, incubated samples for 2 h on a thermomixer at 55°C and 800 rpm, and then denatured Proteinase K by incubating at 95°C for 20 min. We placed samples on ice and then made a 1:10 dilution with molecular grade water. We quantified DNA yields from kit‐based extracts and 1:10 lysates using the Qubit High Sensitivity DNA assay.

We submitted extracted DNA to the USU CIB Genomics Core Facility for library preparation and sequencing. Samples were prepared using a protocol adapted from Avramenko et al. (2015). First stage PCR reactions were run in triplicate using the Kapa HiFi Hotstart PCR mix and the NC1‐NC2 primer set (Gasser et al. 1993; Avramenko et al. 2015). First stage PCR thermocycling parameters were 95°C for 2 min, followed by 25 cycles of 98°C for 20 s, 62°C for 15 s, and 72°C for 15 s followed by a final extension of 72°C for 2 min. Triplicate reactions were pooled and a 1:50 dilution was used for second stage PCR. This reaction used the Platinum II Taq Hot Start master mix and attached a unique set of TruSeq indexing primers to each sample. The thermocycling parameters were 94°C for 60 s, followed by 12 cycles of 94°C for 15 s, 64°C for 15 s, and 72°C for 60 s, followed by a final 72°C incubation for 3 min. Products were cleaned with AMPure beads using a reaction‐to‐bead ratio of 0.8:1. Final amplicons were quantified using PicoGreen (Invitrogen, Carlsbad, CA) before being pooled at equimolar concentrations. Pools were sequenced on an Illumina MiSeq using the MiSeq v2 500 cycle Nano kit, with samples from the pre‐processing and extraction method comparisons prepared as separate pools and sequenced on two runs.

2.5. Bioinformatics

We processed sequencing data using Cutadapt v4.0 to remove primers, DADA2 v1.32.0 to infer sequence variants, and phyloseq v1 for additional filtering (Martin 2011; McMurdie and Holmes 2013; Callahan et al. 2016). Primers were trimmed from forward and reverse reads using a 10% error rate and 3 base pair minimum overlap. We discarded untrimmed reads but applied no additional length filtering. In DADA2, we used the filterAndTrim function with default settings and maxEE = (2,5). Paired reads were merged using the mergePairs function with a minimum overlap of 12 basepairs and a maximum of one mismatch allowed in the overlap region. To assign taxonomy, we used the Nematode ITS2 1.0.0 database (Workentine et al. 2020) and the IdTaxa function from DECIPHER (Wright 2016) using 100 bootstraps and both strands. As our primary goal was to compare between pre‐processing methods, we used a 50% minimum bootstrap confidence for taxonomic assignments to increase rates of taxonomy assignment. However, we acknowledge that this threshold increases the risk of misidentifying assigned sequence variants (ASVs). The first sequencing run generated 790,996 read counts assigned to 479 ASVs. In total, 76.17% of reads were retained following the filtering steps in the DADA2 pipeline, resulting in 15,447 ± 7247 read counts per sample, excluding controls which averaged 5.80 ± 7.95 read counts per sample (range 0–15 reads/sample). Of these read counts, 99.99%, 99.98%, 97.57%, and 96.22% were assigned to phylum, family, genus and species, respectively. Additionally, 96.65%, 96.37%, 84.36%, and 77.37% of ASVs were assigned to phylum, family, genus, and species. We removed reads not assigned to family and, for each sample, any ASV with ≤ 15 reads. We categorised samples (n = 3) with ≤ 1000 sequencing reads as failed reactions. After removing failed reactions and controls, we rarefied remaining samples to the lowest observed sequencing depth (3413 read counts).

Due to the smaller sample size, sequencing depth was higher on the second run with an average of 51,182 ± 10,433 read counts per sample, excluding kit and SWL buffer controls which retained 6 and 70 reads, respectively, after initial processing. The percentage of reads and ASVs assigned to each taxonomic level was similar to values from the first sequencing run. As for the first run, we removed any ASV with ≤ 15 read counts from each sample before subsequent analyses.

2.6. Statistical Analyses

We used a combination of generalised linear models and multivariate approaches to compare results from the three pre‐processing approaches. We first tested whether eggs per gram (EPG) predicted the number of recovered larvae per gram of faeces using Pearson's correlation. We then used a negative binomial regression to test whether different pre‐processing methods generated different read counts and a Poisson regression to test whether methods differed in detected parasite richness at the ASV, species, and genus levels. We tested whether more prevalent genera and species were found at higher intensity (i.e., at higher abundance in infected individuals), using Pearson's correlation and relative read abundance in infected animals as a proxy for intensity. We used k‐means clustering (k = 2) of these data to differentiate rare and common taxa (Figure S1) and then used Poisson regression to test whether pre‐processing methods differed in their ability to detect rare and common genera and species. We also tested whether evenness, measured using Pielou's evenness (J′ = (Shannon index)/log(number of taxa)), differed by pre‐processing method using beta regression and repeating the analysis at the ASV, species, and genus level.

We next tested whether the three pre‐processing methods detected different communities of ASVs using the adonis2 function, calculating community dissimilarity with Bray‐Curtis and Jaccard distances to test for differences in the relative abundance and presence of taxa, respectively. Finally, we tested whether detection of individual species and genera differed between methods. We ran a separate binomial regression model for each species (or genus) found in at least five hosts and controlled for multiple comparisons using the Benjamini and Hochberg method, with a significance threshold set at an FDR of < 0.05 (Benjamini and Hochberg 1995). We included egg density (double centrifugation eggs per gram of dry weight) as an interaction effect in all initial models, identified best models by backwards model selection and performed post hoc pairwise comparisons using the Tukey adjustment in the emmeans package (Lenth 2020).

We used a paired t‐test to test whether lysis buffer and kit extractions differed in number of observed ASVs, species, and genera, repeating the analysis using richness calculated from unrarefied read counts and after rarefying all samples to the minimum sequencing depth of 16,108 read counts. We next tested whether the two extraction methods detected different communities of ASVs using the adonis2 function, calculating community dissimilarity using both Bray‐Curtis and Jaccard distances. Finally, to test whether detection of individual species or genera differed between extraction methods, we used the DESeq package (Love et al. 2014), shrinking log fold changes using the ashr package (Stephens et al. 2024).

We performed all analyses in R v4.4.0 (R Core Team 2024), set significance at α = 0.05, and report average values as mean ± SD. Model assumptions and performance were evaluated by visual inspection and residual diagnostics.

3. Results

3.1. EPG Quantification & Nematode Infection

For animals sampled in 2024, we detected strongylid eggs in all hosts using double centrifugation faecal floats (29.90 ± 33.64 EPG, range: 0.65–107.67), but in only 12 of 13 individuals using the McMaster slide approach (57.85 ± 62.75 EPG, range: 0–200). Samples with higher egg density generated more L3s (Pearson's correlation: r = 0.58, 95% CI [0.045, 0.858], p = 0.037). Strongylid eggs were detected in all eight animals sampled in 2025, with McMaster EPG values ranging from 16 to 328 (86 ± 101.71 EPG). No non‐strongylid parasites were detected during either survey.

3.2. DNA Yields

DNA yield significantly differed between the three processing methods (linear regression, F(2,39) = 136.3, p < 0.001). Yields from extracted egg swabs (2.24 ± 1.11 ng/μL) and larvae (1.95 ± 2.48 ng/μL) were similar and, not surprisingly, both significantly lower than yields from extracted faecal subsamples (44.9 ± 13.7 ng/μL; post hoc pairwise comparisons, faecal—egg: p < 0.0001, and faecal—larvae: p < 0.0001, larvae—egg: p = 0.99). DNA yields did not significantly differ for the eight replicated L3 samples extracted using the commercial kit or SWL buffer approach (paired t‐test, t(7) = −1.08, p = 0.32; kit: 0.71 ± 1.30 ng/μL; SWL: 0.28 ± 0.27 ng/μL).

3.3. Comparing Pre‐Processing Methods

We identified 298 ASVs assigned to 10 genera and 22 species of strongylid nematode from the 13 horses sampled in 2024. Combining replicates from each individual, we detected an average of 58.31 ± 22.01 ASVs, 11.38 ± 2.56 species, and 6.77 ± 1.17 genera per horse. The most frequently detected genera were Cylicocyclus, Cylicostephanus, Cyathostomum, Triodontophorus, and Coronocyclus. More prevalent genera and species also occurred at higher mean intensity (genus: Pearson's correlation: r = 0.54, t(8) = 1.80, p = 0.11, 95% CI [−0.14, 0.87]; species: Pearson's correlation: r = 0.45, t(20) = 2.24, p = 0.037, 95% CI [0.03, 0.73], Figure S1). K‐means clustering using this data identified six common and four rare genera and 10 common and 12 rare species (Figure 2).

FIGURE 2.

Sampled horses hosted a mixture of common and rare taxa, with differential detection between pre‐processing methods. Detection prevalence of each parasite genus (A) and species (B) by method (larval coproculture: L, egg concentration: E, faecal subsample: F) in the 13 horses sampled in 2024. Common taxa, based on k‐means clustering, are bolded. Prevalence of each taxon, given in parentheses, was calculated by merging larval coproculture, egg, and faecal subsample data for each individual.

The larval coproculture method produced the highest number of successfully sequenced samples, with all 13 samples generating > 1000 read counts and averaging 19,000 ± 3700 reads per sample. The egg concentration method generated 201–27,000 reads per sample (17,000 ± 8000) and successfully amplified 11 samples. Faecal subsamples ranged from 0 to 21,000 read counts per sample (11,000 ± 7000) and successfully amplified 12 samples. Although larval extracts generated the highest number of reads per sample, read counts did not differ significantly by method (see Table 1).

TABLE 1.

Summary of model outputs listing the best supported model (selected by backwards model selection) and the estimate, standard error (SE), z‐value, and p‐value for each factor retained in this model. Significance codes: *p < 0.05; **p < 0.01; ***p < 0.001.

Best Model	Estimate	SE	z	p
Read counts ~1
Intercept	9.67	0.11	87.50	< 0.0001***
ASV richness ~ Method + EPG
Intercept	2.47	0.08	30.40	< 0.0001***
Larvae	0.94	0.09	11.00	< 0.0001***
Eggs	0.76	0.09	8.31	< 0.0001***
EPG	0.03	0.00	8.56	< 0.0001***
Species richness ~ Method + EPG
Intercept	1.14	0.17	6.90	< 0.0001***
Larvae	1.03	0.17	5.91	< 0.0001***
Eggs	0.76	0.19	4.08	< 0.0001***
EPG	0.02	0.01	2.48	0.013*
Genus richness ~ Method
Intercept	1.13	0.16	6.85	< 0.0001***
Larvae	0.69	0.20	3.48	0.0005***
Eggs	0.57	0.21	2.73	0.006*
Common Species richness ~ Method
Intercept	0.81	0.19	4.21	< 0.0001***
Larvae	1.08	0.22	4.89	< 0.0001***
Eggs	0.72	0.24	3.04	0.0024**
Rare Species richness ~ Method + EPG
Intercept	−1.42	0.48	−2.97	0.003**
Larvae	1.90	0.48	3.97	< 0.0001***
Eggs	1.51	0.50	3.00	0.0027**
EPG	0.05	0.01	3.93	< 0.0001***
Common Genera richness ~ Method
Intercept	1.10	0.17	6.59	< 0.0001***
Larvae	0.43	0.21	2.04	0.041*
Eggs	0.38	0.22	1.70	0.089
Rare Genera richness ~ Method
Intercept	−2.49	1.00	−2.49	0.13
Larvae	2.92	1.03	2.85	0.004**
Eggs	2.57	1.04	2.47	0.014*

ASV richness was significantly influenced by pre‐processing method and positively correlated with EPG (Table 1, Figure 3A). Larvae, egg swab and faecal subsample approaches detected an average of 39.15 ± 15.41, 31.64 ± 14.06, and 15.58 ± 10.53 ASVs per sample, respectively. Larval coproculture yielded significantly higher ASV richness than egg swab and faecal subsamples (post hoc pairwise comparisons, larvae—egg: p = 0.02, larvae—faecal: p < 0.0001). However, the egg swab method still generated significantly higher ASV richness than extraction of faecal subsamples (post hoc pairwise comparison, p < 0.0001).

FIGURE 3.

Faecal subsampling detected fewer parasite taxa. Regression plots, on the left, show detected parasite richness by EPG at the ASV (A), species (B), and genus level (C) by method (larval coproculture: L, egg concentration: E, faecal subsample: F). Euler plots, on the right, show the number of taxa detected by each method at the ASV, species, and genus level.

Species richness significantly differed between pre‐processing methods and increased with EPG for all methods (Table 1, Figure 3B). Larvae, egg swab, and faecal subsampling approaches detected an average of 10.15 ± 2.76, 7.64 ± 1.96, and 3.67 ± 1.92 species per sample, respectively. We found no significant difference in species richness between larval coproculture and egg swab methods (post hoc pairwise comparison, p = 0.13). However, both larval coproculture and egg swab approaches detected significantly more species than faecal subsamples (post hoc pairwise comparisons, larvae—faecal: p < 0.0001, egg—faecal: p = 0.0001). Reduced richness in faecal subsamples was driven by significantly lower detection of both common and rare species (Table 1, post hoc pairwise comparisons, common species: larvae—faecal: p < 0.0001, egg– faecal: p = 0.007, larvae—egg: p = 0.1, rare species: larvae—faecal: p = 0.0002, egg—faecal: p = 0.008, larvae—egg: p = 0.4), with detection of rare (but not common) species increasing at higher EPG values.

Genus‐level richness also differed between pre‐processing methods but was not significantly influenced by EPG (Table 1, Figure 3C). Larvae, egg swab and faecal subsample approaches detected an average of 6.15 ± 1.28, 5.45 ± 1.21, and 3.08 ± 1.31 genera, respectively. Genus‐level richness did not significantly differ between larval coproculture and egg swab approaches (post hoc pairwise comparison, p = 0.76), and both larvae and egg swab methods contained significantly more genera than extracted faecal subsamples (post hoc pairwise comparisons, larvae—faecal: p = 0.002, egg—faecal: p = 0.02). Lower genus‐level richness in faecal samples was primarily due to reduced detection of rare genera, with significantly fewer rare genera seen in faecal samples compared to egg swab and larvae extracts (post hoc pairwise comparisons, rare species: larvae—faecal: p = 0.01, egg—faecal: p = 0.04, larvae—egg: p = 0.6). Although pre‐processing method significantly influenced the richness of common genera (Table 1), post hoc comparisons found no significant pair‐wise differences between methods (all p > 0.05).

Although parasite richness was consistently lower in faecal subsamples, overall parasite communities and community evenness were similar across pre‐processing methods (Figure 4). We found no significant effect of pre‐processing method on evenness at the parasite ASV, species, or genus level (all model p‐values > 0.05). Pre‐processing method also had no significant effect on community composition, regardless of whether community similarity was measured in terms of ASV presence (PERMANOVA, Jaccard distance, pseudo‐F(2,33) = 1.14, p = 0.265, n = 999 permutations) or relative abundance (PERMANOVA, Bray‐Curtis distance, pseudo‐F(2,33) =1.07, p = 0.362, n = 999 permutations). None of the 14 species and eight genera found in at least five animals (i.e., > 38% prevalence), were significantly more likely to be detected in one method compared to the others (adjusted for multiple comparisons, all model p‐values > 0.05).

FIGURE 4.

Parasite communities did not significantly segregate by method. Samples, coloured by horse and shaped by method, are shown on a PCoA plot. Parasite community similarity was calculated using Bray‐Curtis distances and larval coproculture (L), egg concentration (E), and faecal subsamples (F) from each animal are connected by lines.

3.4. Comparing Extraction Methods

Coprocultured faeces from the eight horses sampled in 2025 contained a total of 267 ASVs, 21 species, and nine genera. We detected no significant difference in the number of ASVs, species, or genera in samples extracted using the SWL buffer approach compared to the commercial kit (paired t‐tests, all p‐values > 0.05), with similar results seen for rarefied and unrarefied read counts. We also detected no significant difference in parasite communities, regardless of whether similarity was measured in terms of ASV presence (PERMANOVA, Jaccard distance, pseudo‐F(1,14) = 0.351, p = 0.99, n = 999 permutations) or relative abundance (PERMANOVA, Bray‐Curtis distance, pseudo‐F(1,14) = 0.257, p = 0.97, n = 999 permutations). Although we found no difference in overall communities detected by the two DNA extraction approaches and no evidence of differential abundance of any taxa at the genus level (all p > 0.05), differential abundance analysis did show significantly more reads assigned to Triodontophorus brevicauda and Cylicocyclus ultrajectinus in samples extracted using the SWL approach (both adjusted p‐values < 0.0001).

4. Discussion

Metabarcoding is widely used in ecological research, particularly for monitoring biodiversity, characterising microbial communities, and estimating animal diets. However, its integration into parasitology has been slow, likely due to a lack of reference sequences and commercially available parasite metabarcoding services, high establishment and training costs for setting up new protocols, and challenges associated with capturing diverse and minute infectious agents that often occur at low density within hosts. To address this latter challenge, here, we show that pre‐processing samples to concentrate parasite larvae or eggs substantially improves parasite detection, especially for rare species.

Of the tested methods, larval coproculture consistently provided the most comprehensive representation of the parasite community. It yielded the highest sequencing success, recovered the greatest richness across taxonomic levels, and was least affected by egg density in faecal samples. This differs from earlier studies in primates that detected higher parasite diversity in extracted faeces (Pafčo et al. 2018). Primate parasites may be more difficult to culture than horse strongylids, which all exhibit relatively high hatch rates under similar coproculture conditions (Mfitilodze and Hutchinson 1987). Alternatively, and perhaps more likely, is that the improved performance of larval culturing in our study comes from greater amounts of starting material used in larval culture (40 g) compared to egg swabs (4 g) or faecal subsample extraction (0.15 g). These starting amounts reflect what is commonly used for each pre‐processing protocol and thus provide a practical comparison of sensitivity for typical use‐cases.

Larval culturing offers several advantages compared to other pre‐processing methods. It produces a relatively clean and highly concentrated sample of parasite larvae which can be easily extracted with inexpensive lysis buffers. Moreover, methods which rely on parasite movement (e.g., coproculture and Baermannisation) selectively recover live parasites, potentially reducing contamination from environmental DNA or parasites in target species' prey (Aivelo et al. 2018). However, because parasites must migrate out of faeces, these methods are limited to easily cultured parasites with free‐living, motile larval stages. Although we did not detect differences in community composition or evenness between pre‐processing methods in this study, larvae can also exhibit substantial variation in size and optimal culture conditions (Roeber and Kahn 2014), potentially biasing relative abundance estimates. Finally, coproculturing also requires large amounts of fresh faeces and may be less feasible in field settings due to the lengthy processing time, physical space required to incubate samples, and the need for high‐speed centrifuges and ultracold storage.

Egg swab samples performed nearly as well as larval coproculture, with similar diversity detected at the species and genus level. Like larval coproculture, this method concentrated infective stages and was less sensitive to differences in egg density. Concentrating infective stages likely also decreases host and environmental DNA contamination (Davey et al. 2021), potentially reducing the need for host‐blocking primers or other enrichment techniques. In contrast to larval coproculture, egg concentration can be done relatively rapidly, potentially even using a hand‐driven centrifuge, and can detect parasites that do not produce motile stages. Although egg concentrating protocols are simpler than coproculture, processing still requires fresh samples, several hours of lab work and some specialised lab equipment. Furthermore, like faecal subsamples, DNA extraction from concentrated eggs requires more expensive and time‐intensive protocols that lyse eggs and remove PCR inhibitors (Ayana et al. 2019). Potential modifications to egg concentration approaches may make it even more accessible for field collections. Specifically, preservatives like DNA/RNA shield, RNAlater, or ethanol could be used to inactivate infectious agents and facilitate storage at ambient temperature. It may also be possible to prepare egg swabs from frozen faecal samples; however, further testing is needed as freezing can reduce egg recovery (Schurer et al. 2014).

DNA extractions from faecal subsamples proved to be the least sensitive approach. Surprisingly, this method failed to detect both rare and common taxa. Our results differ from previous work in reindeer showing no difference in parasite detection between extractions of faeces and infectious stages concentrated via flotation or sedimentation (Davey et al. 2021). This discrepancy likely reflects the lower diversity of parasites infecting reindeer and the amount of faeces used for DNA extractions. Although Davey et al. extracted DNA from 3 g of faeces, we used just 0.15 g to match protocols from the commercial kits most commonly used for faecal DNA extractions. For the diverse parasite community in free‐roaming horses, extracting DNA from 0.15 g of faeces consistently underestimated parasite diversity, especially when egg counts were low. As most wildlife hosts low‐intensity infections (Shaw et al. 1998), for many systems, data derived from small faecal subsamples risks substantially underestimating both infection prevalence and the interconnectedness of parasite sharing networks.

Despite reduced sensitivity, there are advantages to extracting DNA directly from faecal samples. Preserving faeces requires little expertise, no pre‐processing prior to DNA extraction, and reduces cross‐contamination risk. Faecal extraction protocols that use faecal subsamples require only a small amount of material and may also be able to detect DNA from parasites that are not actively shedding infectious stages (Doherty et al. 2023). Finally, faecal DNA extracts can be used for multiple assays (e.g., microbiome, diet, parasites) and can provide additional information on host sex, species, and individual identity. Although less sensitive than other approaches, extracted faeces still recovered an average of 59% of the species and 70% of the genera present in coprocultures and showed no significant difference in overall community composition nor biases against specific taxa. This suggests that, when limitations are acknowledged, small faecal extracts can provide insight into parasite prevalence and community composition, particularly when other approaches are not feasible.

Ultimately, all non‐invasive parasite detection methods have limitations. Regardless of pre‐processing method, faeces can typically only be used to detect adult gastrointestinal parasites that produce infectious stages. Hosts infected with only immature, male, or intermittently shedding female parasites can generate false negatives (Gibson 1965). Use of faecal samples can also generate false positives by detecting parasites present in prey items or transient parasites stages from the environment (Tercel et al. 2021; Davis et al. 2021). Furthermore, as parasitism represents a trophic strategy rather than a taxonomic group, no primers exist to amplify all parasite taxa. Even the most universal primers struggle to amplify some parasite taxa (Miller et al. 2024), and sequences can be difficult to identify as parasites are substantially underrepresented in reference libraries (Li and Poulin 2025). Finally, because DNA inputs are normalised prior to sequencing, NGS approaches provide no information on total parasite burden and must be coupled with methods like faecal egg counts of qPCR to quantify parasite loads. These parasite‐specific limitations, combined with the well‐documented caveats inherent to all DNA metabarcoding studies (e.g., Deagle et al. 2019) underscore the need for continued traditional parasitological surveys. Ultimately, parasite metabarcoding is part of a suite of non‐invasive tools for assessing infection and, despite its limitations, still provides a powerful stand‐alone approach for detecting infection in systems where no other method is feasible.

5. Conclusion

Collectively, these results have important implications for the use of metabarcoding in disease research. Sample pre‐processing methods substantially influence detection of both rare and common parasites. Larval coproculture and our novel egg swab method most accurately capture parasite diversity, but both require at least some processing at the time of sample collection. DNA extraction directly from faecal subsamples allows for rapid sample collection and preservation but fails to detect a substantial proportion of the parasite community. Ultimately, the optimal pre‐processing approach for parasite metabarcoding depends on the specific research question and logistical constraints.

Author Contributions

M.E.P. contributed to designing the research, conducting field and lab work, analysing data and writing the manuscript. G.B.G. contributed to designing the research, conducting field and lab work, and editing the manuscript. S.B.W. contributed to designing the research, conducting field work, analysing data, and writing the manuscript. All authors gave their approval for the final version of the manuscript.

Funding

This work was supported by a Utah State University Extension Grant (Ext00172) and a USU Undergraduate Research and Creative Opportunity Grant.

Disclosure

Benefit‐sharing statement: Benefits from this research accrue from the sharing of our data and results on public databases as described above.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: More prevalent genera and species occurred at higher mean intensity (Genus: Pearson's correlation: r = 0.54, t(8) = 1.80, p = 0.11, 95% CI [−0.14, 0.87]; Species: Pearson's correlation: r = 0.45, t(20) = 2.24, p = 0.037, 95% CI [0.03, 0.73]). K‐means clustering was used to identify the six common and four rare genera and 10 common and 12 rare species.

Patch, M. E. , Goodman G. B., and Weinstein S. B.. 2026. “Scoop That Poop: Optimising Faecal Sample Pre‐Processing for Parasite Metabarcoding.” Molecular Ecology Resources 26, no. 2: e70112. 10.1111/1755-0998.70112.

Data Availability Statement

Raw data can be found on the Sequence Read Archive (Accession number PRJNA1309108). Further data and scripts are openly available on Github (https://github.com/patchmadison/URCO‐Horse‐Project).