Assessing detection of New Zealand mudsnails at low densities in Arizona streams

461 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 Assessing detection of New Zealand mudsnails at low densities in Arizona streams Crosby K. Hedden1, Caroline E. Mallinson1, Crystal Castillo1, Alexander D. Loubere1, Ryan D. Mann1 1 Arizona Game and Fish Department, 5000 W. Carefree Highway, Phoenix, AZ, 85086, USA Corresponding author: Crosby K. Hedden ([email protected]) Aquatic Invasions 2025 Volume 20, Issue 4: 461–476 Copyright: © Crosby K. Hedden et al. This is an open access article distributed under terms of the Creative Commons Attribution License (Attribution 4.0 International – CC BY 4.0). Research Article Academic editor: Ian Duggan Received: 16 April 2025 Accepted: 24 June 2025 Published: 12 November 2025 Citation: Hedden CK, Mallinson CE, Castillo C, Loubere AD, Mann RD (2025) Assessing detection of New Zealand mudsnails at low densities in Arizona streams. Aquatic Invasions 20(4): 461–476. https://doi. org/10.3391/ai.2025.20.4.164778 Abstract The New Zealand mudsnail (NZMS) is a small-bodied gastropod that has successfully invaded waters across multiple continents. This species has the ability to reach extremely high densities in streams and exclude other aquatic macroinvertebrates which higher trophic levels rely on as a food source. While the effects of NZMS are well studied, early detection methods for this species are limited almost entirely to environmental DNA (eDNA) testing. While eDNA is a valuable tool for the early detection of this species, low density sampling protocols are also essential to verify positive eDNA detections and to determine precise distributions so that management may be implemented in these areas during an invasion. The goal of our study is to evaluate and compare the efficacy of various quadrat sampling protocols to detect NZMS at low densities, and to determine the densities below which detection may become uncertain using these protocols. We tested 10-, 20-, and 30-quadrat grids within 100 m stream reaches, using both random and strategic selection of quadrat sites, to assess each design’s performance in overall probability of detection. We found that a non-random strategic sampling design was significantly more effective at detection of NZMS than a random design. Additionally, we found that, across study streams with different snail densities, taking 14 quadrat Surber samples using non-random strategic site selection consistently led to capture probabilities over 99%, with one exception in the stream with the lowest densities. To account for heterogeneity in habitat and snail density, we recommend using 30 quadrats with non-random strategic site selection to maximize detection in systems with unknown presence. This study outlines a sampling protocol to verify the physical presence of NZMS that can be adapted into monitoring programs or to confirm presence of this species following a suspected introduction. Key words: Bootstrapping, Invasive Species, Quadrats, Physical Detection, Surber Sampler Introduction Ecosystem health and proper functioning is imperative to the continued persistence of aquatic organisms and is largely influenced by having essential trophic levels intact and interacting in a stable manner (Scott et al. 2012). Ecosystem processes can be affected by multiple stressors, including disturbance and the introduction Low density detection of New Zealand mudsnails 462 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 of invasive species (Cardinale and Palmer 2002; Gallardo et al. 2016). Aquatic invasive species have become increasingly common throughout North America, including within the American Southwest, where they have caused ecosystem health to decline in numerous watersheds (Dukes and Mooney 2004; Thomaz et al. 2015). The New Zealand mudsnail (Potamopyrgus antipodarum; hereafter NZMS), a gastropod native to freshwater streams and lakes throughout New Zealand, is one such aquatic invasive species that has been introduced throughout the region and has the ability to cause negative effects to proper ecosystem functioning of many waterbodies (Geist et al. 2022; Alonso et al. 2023). Within its native range, NZMS is a host to multiple trematode parasites which render their hosts infertile. This heavily limits the reproductive rate of NZMS, keeping their populations in check (Hechinger 2012). With the absence of these parasites outside of their native range, NZMS have the ability to reproduce unchecked and reach extremely high densities in streams (Hechinger 2012; Geist et al. 2022). The NZMS has been documented to alter the structure of the native invertebrate communities in both laboratory and field studies (Larson and Black 2016; Rakauskas et al. 2017), which has led to bottom-up effects on higher trophic levels that rely upon these invertebrate communities as a prey base. The expansion of this species, which now encompasses numerous waters in five continents, can be attributed to life history and physical characteristics of the species. This includes an optimized reproductive strategy, small size and ability to attach themselves to surfaces that are moved between waters, as well as their operculum, which provides the snail a strong resistance to desiccation when removed from aquatic systems (Richards et al. 2004; Alonso and Castro-Díez 2008; Geist et al. 2022). Additionally, this species is of little nutritional value in fish diets, with individuals of NZMS closing their opercula upon consumption and commonly passing through the fish intestinal tract undigested and still alive (Vinson and Baker 2008; Bruce et al. 2009). Because they outcompete aquatic invertebrates and are largely inaccessible as a food source, NZMS operate as a trophic sink to higher trophic levels, drastically interrupting the food chain. The difficulty in suppressing this species following an introduction, coupled with its high potential for invasion between waterbodies makes them extremely difficult to manage (Richards et al. 2004; Geist et al. 2022). Given the largescale impact this species can have on the food chain and long-term viability of native aquatic communities, it is imperative to be able to detect NZMS effectively and efficiently in systems outside of their native range to mitigate the effects as quickly as possible. Environmental DNA (eDNA), which is DNA that has been shed into the environment by organisms, has been commonly used to detect various aquatic invasive species (AIS), including NZMS, in the critical early stages of invasion (Woodell et al. 2021). Compared to traditional methods of detection, eDNA offers advantages, such as low cost and the ability to detect presence of a target species at low densities without physically locating a species within a system where density is unknown (Woodell et al. 2021). Advancements in eDNA sampling methods, equipment, and assays have decreased sample processing times and have made it possible to obtain in situ results (Thomas et al. 2020). Despite the benefits of eDNA technology, the potential for false positives and false negatives remain prominent challenges in eDNA analyses (Bochove et al. 2020; Wood et al. 2021). Erroneous survey results can result in inappropriate management decisions for a particular stream. For example, false positives can be costly and pull finite resources away from where they are needed, while false negatives can delay detection and make it more difficult to respond to and control AIS species once they are established. Such errors can also be an issue in other methods of AIS detection. Presence of AIS is also sometimes reported by members of the public, and these reports are at times the first evidence Low density detection of New Zealand mudsnails 463 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 of AIS in a particular waterbody. While these public reports can greatly assist in documenting the spread of AIS, particularly in waters that may not be monitored regularly, physical verification is still needed prior to implementing management actions (i.e., mobilizing staff, identifying treatment options, public postings, etc.) due to the potential for a false identification. Physical confirmation of NZMS presence addresses concerns related to false positives, and also provides instantaneous confirmation of presence, allowing for immediate action. Accurate records of occurrence following a suspected invasion are essential for managers to quickly define the severity of the problem and refocus management activities to suppress and/or control the species and minimize their spread within the ecosystem. Low species densities can complicate and impact both the spatial determination of their distribution in a stream system and the accuracy of their initial detection, particularly using eDNA. While eDNA remains a valuable tool in regard to detecting species introductions throughout aquatic ecosystems, physical verification following a positive eDNA detection or any other unsubstantiated report is necessary to verify presence before further management actions are implemented. While the initial detection of an invasive species, such as via eDNA sampling, is crucial for directing rapid response efforts to control its spread, long-term management of NZMS requires the ability to confirm their presence via physical sampling. A sampling protocol that provides guidance on the sampling intensity and effort required to physically confirm presence will allow managers to better allocate survey resources and also have confidence in survey results. Thus, the goal of this study was to establish a standardized sampling method to physically and reliably detect NZMS in streams at low densities following a suspected introduction. To accomplish this, we tested multiple quadrat sampling designs for the purpose of physically detecting low-density populations of NZMS. Our specific objectives were: 1) to evaluate the number of quadrats needed within sampling designs to ensure detection at low densities in a stream setting, 2) to compare random versus non-random strategic site selection on detection, and 3) using the outcomes of Objective 1 and 2, to determine the lowest NZMS density that can be detected by these quadrat methods. Results from this study will provide managers with guidance on standardized methods for assessing the distribution of NZMS after positive eDNA detections or when physical surveys are necessary to verify the presence of the species. Methods Study area We sampled three streams in Arizona for NZMS: Oak Creek, Canyon Creek, and Tonto Creek (Fig. 1). Each of these streams contained high recreational value due to their Brown Trout (Salmo trutta) and Rainbow Trout (Oncorhynchus mykiss) populations and angling popularity. Oak Creek is a tributary to the Verde River while both Tonto and Canyon Creek are tributaries to the Salt River. Oak Creek has the lowest elevation (1066 meters above sea level [MASL]) and a higher percentage of fine substrates (sand and silt) relative to both Tonto and Canyon Creeks. Canyon and Tonto Creeks are both high elevation streams (1785 and 1653 MASL, respectively) situated near the Mogollon Rim in Arizona. All three streams have had positive eDNA results for NZMS, which were confirmed by physical observations prior to the initiation of this study, although snail densities remain relatively low (<50 NZMS/m2) in some areas of each stream and at sites distant from presumed introduction sites. We qualitatively targeted and prioritized low density and/or heterogeneous distributions of NZMS in selection of study reaches because Low density detection of New Zealand mudsnails 464 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 the presence of snails is easily verified in high density areas. Scouting trips were taken to each stream before the onset of sampling to identify stream reaches that met the desired low density criteria (i.e., NZMS are not yet abundant via visual observation but still can be found throughout the reach) by taking preliminary benthic samples using Surber samplers (Wildlife Supply Company; Suffield, CT) and visual surveys to identify the extent of NZMS in the reach. Sampling design We explored two primary sampling design approaches to determine whether random or non-random strategic site selection is more effective at detecting NZMS, and whether the number of quadrats taken (i.e. 10-, 20-, or 30-quadrats) using each sampling technique provided sufficient detection probability, while minimizing effort needed. To do this, we investigated a section of stream prior to the initiation of sampling to identify a low-density area of NZMS. Once a low-density section was identified, we systematically selected a single 100-meter reach within this section Figure 1. Map of NZMS sampling areas and major streams in the Gila River Basin of Arizona. Black dots indicate the location of sampling in Oak, Tonto, and Canyon Creek. Major cities throughout the state are indicated by red stars. Low density detection of New Zealand mudsnails 465 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 that had adequate access to serve as the reach for all Surber sampling (Fig. 2). The only location where this was slightly modified was Tonto Creek, where two 100-meter reaches that were directly adjacent to one another were sampled due to the narrow-wetted width of this stream to avoid overlapping quadrat samples. Following the identification of the sampling reach, Surber samplers (30.48 cm by 30.48 cm) were used to sample NZMS in the stream. Sampling was conducted such that random sites were sampled first and non-random strategic sites were sampled second. For both site selection methods, three replicates of each 10-, 20-, and 30-quadrat group samples (i.e., 10-, 20-, and 30-sample group) were collected to increase the overall samples size within the 100-meter sampling reach (180 random samples and 180 non-random strategic samples per stream). For random sampling, a random number generator between 0–100 (0 being the most downstream extent of the sampling reach, 100 being the most upstream extent) was used to determine the longitudinal distance in the stream to sample. Another random number generator was used between zero and the maximum width of the stream in the sampling reach to determine the horizontal distance (from river right) to place the Surber sampler. If the horizontal distance of the random number generator exceeded the maximum width of the stream at the selected sampling point, the Surber sample was taken as close to the margin of the stream as possible. For non-random strategic site selection, each biologist was briefed on preferable habitats of NZMS based upon existing literature outlining habitat preference (Geist et al. 2022). The biologist then identified a location within the sampling reach to place the Surber sampler based upon visual identification of the most ideal NZMS habitat available (i.e., areas with low water velocity, intermediate depth, and small to intermediate substrate sizes). Non-random strategic sites were still distributed throughout the 100-meter reach (i.e., sites were not clustered within the same area of the sampling reach). Once a sampling site had been selected, either randomly or strategically, the Surber sampler was placed as close to parallel with water flow as possible, with the opening facing upstream and the base frame embedded in the substrate. The area within the sampler base was disturbed into the collection net, and all rocks, vegetation, and woody debris that could be moved were picked up and rubbed to dislodge any attached organisms into the net. When possible, the substrate was excavated down to a depth of two inches inside the sampler in order to capture any snails that had burrowed below the surface. After all surfaces within the base frame had been disturbed, the sampler was removed and the contents of the collection net were processed through a sieve (4000 μm) to filter out coarse substratum, which were visibly inspected for snails. Each sample was then placed in a labelled Whirl-Pak™ (Whirl-Pak Filtration Group; Pleasant Prairie, Wisconsin) and preserved in 90% ethanol for later processing in the laboratory. For each Surber sampler collected at a sampling point, distance from the bottom of the reach, horizontal distance into the stream relative to river right, sample depth, and substrate size were recorded. The substrate size categories used for this study were based upon Wentworth (1922): clay (<0.0039 mm), silt (0.0039–0.063 mm), sand (0.063–2 mm), gravel (2–4 mm), Figure 2. Schematic defining key terms for the NZMS detection study. Low density detection of New Zealand mudsnails 466 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 pebble (4–64 mm), cobble (64–256 mm), and boulder (>256 mm). After all samples were taken and packaged, the sieves and Surber samplers were thoroughly cleaned and rinsed before proceeding to the next sampling point. Equipment was decontaminated between waterbodies to prevent further spread and establishment of NZMS following Hazard Analysis and Critical Control Point (HACCP) protocols and using Virkon S Disinfectant and Virucide (Antec International; Sudbury, England). After returning to the laboratory, samples were run through a series of stacked sieves of progressively smaller mesh (4000 μm, 2000 μm, 500 μm, and 250 μm) to filter the sample and isolate any snails. All samples were then individually examined visually and snail presence-absence was recorded for each sample. All NZMS captured within a sample were counted and placed in a vial containing ethanol solution for sample preservation. Analysis All analyses were completed using Program R Version 4.4.1 (R Core Team 2024). We used a mixed effects logistic regression model to test for differences in NZMS presence-absence across 10-, 20-, or 30-quadrat groups, as well as between random and non-random strategic site selection. Models were fit using the glmer function in the lme4 package implemented in R, version 4.4.1 (Bates 2014; R Core Team 2024). We fit binomial mixed effects models with presence-absence of NZMS in each Surber sample as the response variable. Independent variables included random or non-random strategic site selection, sampling location, and quadrat group size (10, 20, or 30). Quadrat group replicate (replicate 1, 2, or 3) was used as a random effect to account for varying catch rates in successive passes. We then utilized post-hoc comparisons to assess the differences among groups of our independent variables for detecting NZMS, pairwise comparisons were performed using the emmeans function in the package emmeans (Lenth 2021). All model assumptions were evaluated using the DHARMa package in R (Hartig and Lohse 2020). Additionally, we fit separate models to determine the effectiveness of detecting NZMS using each method (i.e. random or non-random strategic site selection, as well as 10, 20, or 30 samples) in each stream. To do this, data were divided by stream (to account for mean NZMS densities in each stream), random or non-random strategic site selection, as well as 10-, 20-, or 30-quadrat groups for a total of 18 individual models (six models per stream [10-, 20-, or 30-quadrat group observing both random and non-random strategic sites]). We then fit a model using the glm function with presence-absence of NZMS in each Surber sample as the response variable and a random sample point number (i.e. 1–10, 1–20, or 1–30 randomly assigned to samples depending on quadrat group number) as an ordered factor for the independent variable. Model diagnostics showed that the addition of quadrat group as a random effect did not contribute to the variance in the model, so it was not included for this portion of the analysis. For each group, we utilized bootstrap resampling (with 1,000 iterations) to account for variability in detection estimates. Following this, we used the predict function to determine an estimated detection probability and 95% confidence intervals for each individual sampling point at all locations across all sampling types. The equation: was used to determine sequential probability of detection for each successive sample where pi is the predicted probability of detection in sample i. Using this equation, we were able to estimate the detection probability of the quadrat group as a whole using each sampling method within each stream of varying densities. Low density detection of New Zealand mudsnails 467 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 Lastly, we compared the sequential detection probability (i.e., the cumulative probability of detecting NZMS with sequential samples) across all three different streams and sampling strategies. These comparisons were made by plotting the sequential probability of detection derived from the equation above based upon the predicted detection probability per net (i.e. the value of pi in sample 1, sample 2, etc.), and the sequential detection probabilities obtained from the glm models of each sampling strategy which utilized the equation. We analyzed the results based upon a minimum threshold of 99% detection probability to minimize the risk for false negatives while using these sampling protocols. Results Sampling for NZMS occurred from June to July, 2024. The distribution of substrate sizes across the three streams sampled varied slightly, with Canyon Creek having a higher percentage of large substratum (i.e., cobble and larger) while Tonto and Oak Creek had a higher proportion of fine sediments (Fig. 3). Mean sampled depth also differed between streams with Oak Creek being the deepest site and Tonto Creek being the shallowest (Fig. 3). Simulated scale residuals indicated no evidence of deviation from uniformity for our glmer model (p-value > 0.05) and no significant over dispersion was detected (alp-values > 0.05). Mean densities of NZMS observed across the three study streams were as follows: Oak Creek = 47.0 snails/m2, Canyon Creek = 28.0 snails/ m2, and Tonto Creek = 8.2 snails/m2 (Table 1). The proportion of NZMS captured also varied between sampling strategies (Table 1). Probability of detecting NZMS significantly differed between random site selection and non-random strategic site selection, with the latter resulting in significantly higher probability of detection (Z value = -6.38, p-value < 0.001; Table 2; Fig. 4). There was no statistical difference in detection probability by net across the 10-, 20-, or 30-quadrat groups in Figure 3. Distribution of substrate size (top) and mean sampled depth (bottom) of the three streams sampled for New Zealand mudsnails. Whiskers represent standard deviation. Low density detection of New Zealand mudsnails 468 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 Table 1. Summary of NZMS detected from each stream, quadrat group, and replicate over the course of this study in each stream. Numbers below stream name indicate overall density of NZMS in each stream (total snails captured in stream/total area sampled in stream). Columns indicate quadrat number, observed density within each quadrat group (total snails captured in quadrat group/total area sampled in quadrat group), replicate, and the number of Surber samplers where NZMS were present and absent. Waterbody Sample Type Quadrat Number Observed Density of Quadrat Group (Snails/m2)Replicate Number Percent of Samples with NZMS Present Canyon Creek (28.02 snails/m2) Random 10 9.69 1 40.0% 10 15.07 2 50.0% 10 18.30 3 60.0% 20 50.05 1 60.0% 20 28.52 2 35.0% 20 13.99 3 45.0% 30 25.83 1 40.0% 30 28.34 2 50.0% 30 30.86 3 33.3% Strategic Non-Random 10 26.91 1 80.0% 10 17.22 2 50.0% 10 101.18 3 60.0% 20 22.60 1 75.0% 20 37.67 2 55.0% 20 30.14 3 55.0% 30 20.09 1 53.3% 30 19.38 2 56.7% 30 26.91 3 36.7% Oak Creek (47.0 snails/m2) Random 10 2.15 1 10.0% 10 4.31 2 30.0% 10 2.15 3 10.0% 20 0.54 1 30.0% 20 12.92 2 5.00% 20 41.44 3 45.0% 30 34.44 1 40.0% 30 9.69 2 36.7% 30 26.19 3 23.3% Strategic NonRandom 10 88.26 1 80.0% 10 27.99 2 40.0% 10 82.88 3 90.0% 20 59.20 1 70.0% 20 75.35 2 60.0% 20 39.83 3 50.0% 30 29.78 1 23.3% 30 138.14 2 73.3% 30 103.69 3 66.7% Low density detection of New Zealand mudsnails 469 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 either the random or non-random strategic design (Z value = 0.22, p-value = 0.34; Table 2; Fig. 4), indicating that heterogeneity (i.e. density and distribution) was likely similar across sets of quadrats. Further, non-random strategic sampling was also significantly better at detecting NZMS in all streams sampled for this study when making pairwise comparisons using the function emmeans function (Oak Creek: Z value = -5.74, p-value <0.001; Canyon Creek: Z value = -2.12, p-value = 0.034; Tonto Creek: Z value = -0.96, p-value <0.001; Fig. 4). We found no Waterbody Sample Type Quadrat Number Observed Density of Quadrat Group (Snails/m2)Replicate Number Percent of Samples with NZMS Present Tonto Creek (8.19 snails/m2) Random 10 0.00 1 0.0% 10 4.31 2 20.0% 10 0.00 3 0.0% 20 1.61 1 10.0% 20 1.08 2 10.0% 20 2.15 3 15.0% 30 2.87 1 13.3% 30 6.82 2 30.0% 30 1.79 3 16.7% Strategic NonRandom 10 19.38 1 20.0% 10 1.08 2 10.0% 10 0.00 3 0.0% 20 5.38 1 35.0% 20 7.00 2 25.0% 20 5.38 3 25.0% 30 15.43 1 33.3% 30 29.78 2 60.0% 30 18.30 3 23.3% Figure 4. Estimated probability of detection by individual sampling point in each stream, with whiskers representing 95% confidence intervals, of New Zealand mudsnails across all streams for random (panel A) and non-random strategic (panel B) sampling designs. Probabilities of detection represent the likelihood of capturing a NZMS in a single Surber sample derived from the bootstrapped glm models. Low density detection of New Zealand mudsnails 476 Crosby K. Hedden et al. (2025), Aquatic Invasions 20(4): 461–476, 10.3391/ai.2025.20.4.164778 Rakauskas V, Šidagytė E, Butkus R, Garbaras A (2017) Effect of the invasive New Zealand mud snail (Potamopyrgus antipodarum) on the littoral macroinvertebrate community in a temperate mesotrophic lake. Marine and Freshwater Research 69: 155–166. https://doi.org/10.1071/MF17059 Richards DC, O’Connell P, Shinn DC (2004) Simple control method to limit the spread of the New Zealand mudsnail Potamopyrgus antipodarum. North American Journal of Fisheries Management 24: 114–117. https://doi.org/10.1577/M02-133 Scott SE, Pray CL, Nowlin WH, Zhang Y (2012) Effects of native and invasive species on stream ecosystem functioning. Aquatic Sciences 74: 793–808. https://doi.org/10.1007/s00027-012-0263-6 Thomas AC, Tank S, Nguyen PL, Ponce J, Sinnesael M, Goldberg CS (2020) A system for rapid eDNA detection of aquatic invasive species. Environmental DNA 2: 261–270. https://doi. org/10.1002/edn3.25 Thomaz SM, Kovalenko KE, Havel JE, Kats LB (2015) Aquatic invasive species: general trends in the literature and introduction to the special issue. Invasive Species 746: 1–12. https://doi. org/10.1007/s10750-014-2150-8 Victorian Government (2010) Invasive plants and animals policy framework. Department of Primary Industries, Melbourne, Australia. Vinson MR, Barker MA (2008) Poor growth of rainbow trout fed New Zealand mud snails Potamopyrgus antipodarum. North American Journal of Fisheries Management 28: 701–709. https://doi.org/10.1577/M06-039.1 Wentworth CK (1922) A scale of grade and class terms for clastic sediments. The Journal of Geology 5: 377–392. https://doi.org/10.1086/622910 Wood ZT, Lacoursière-Roussel A, LeBlanc F, Trudel M, Kinnison MT, McBrine CG, Pavey SA, Gagné N (2021) Spatial heterogeneity of eDNA transport improves stream assessment of threatened salmon presence, abundance, and location. Frontiers in Ecology and Evolution 9: 650717. https://doi.org/10.3389/fevo.2021.650717 Woodell JD, Neiman M, Levri EP (2021) Matching a snail’s pace: successful use of environmental DNA techniques to detect early stages of invasion by the destructive New Zealand mud snail. Biological Invasions 23: 3263–3274. https://doi.org/10.1007/s10530-021-02576-7