Robust data to refute or support claims of global insect decline are currently lacking, particularly for the soil fauna in the tropics. DNA metabarcoding represents a powerful approach for rigorous spatial and temporal monitoring of the taxonomically challenging soil fauna. Here, we provide a detailed field protocol, which was successfully applied in Barro Colorado Island (BCI) in Panama, to collect soil samples and arthropods in a tropical rainforest, to be later processed with metabarcoding. We also estimate the proportion of soil/litter ant, springtail and termite species from the local fauna that can be detected by metabarcoding samples obtained either from Berlese-Tullgren (soil samples), Malaise or light traps. Each collecting method detected a rather distinct fauna. Soil and Malaise trap samples detected 213 species (73%) of all target species. Malaise trap samples detected many ant species, whereas soil samples were more efficient at detecting springtail and termite species. With respect to long-term monitoring of soil-dwelling and common species (more amenable to statistical trends), the best combination of two methods were soil and light trap samples, detecting 94% of the total of common species. A protocol including 100 soil, 40 Malaise and 80 light trap samples annually processed by metabarcoding would allow the long-term monitoring of at least 11%, 18% and 16% of species of soil/litter ants, springtails and termites, respectively, present on BCI, and a high proportion of the total abundance (up to 80% of all individuals) represented by these taxa.

Barro Colorado Island, Berlese-Tullgren, Collembola, Formicidae, Isoptera, light trap, Malaise trap

Arthropods represent the majority of macroscopic terrestrial animal life, both in terms of species richness (Stork 2018) and biomass (Pimentel and Andow 1984), particularly in the tropics (Basset et al. 2012). They also provide critical services, including ecosystem functioning and human food security. Despite their ecological and economic importance, arthropods are still comparatively neglected as a primary focus of scientific research, limiting our understanding of the functional ecology of ecosystems (Basset et al. 2019). Alarmingly, recent reports suggest catastrophic declines in current insect abundance, with potentially serious implications for ecosystem functioning. This includes declines in the richness of insect species in temperate countries (Sánchez-Bayo and Wyckhuys 2019), insect biomass in Germany and Puerto Rico (Hallmann et al. 2017; Lister and Garcia 2018), butterflies and moths in the United Kingdom (McDermott Long et al. 2017), or insect pollinators worldwide (Goulson et al. 2015). Although some of these articles have attracted considerable attention in the media (often with dramatic epithets such as “Insect Armageddon”), they have also been criticized for lack of scientific rigor (Leather 2017; Didham et al. 2020). What is clear is that we currently lack robust data to refute or support claims of global insect decline, particularly in the tropics (Basset and Lamarre 2019; Crossley et al. 2020; van Klink et al. 2020; Wagner 2020), not to mention scenarios of how insect populations will be affected by anthropogenic changes in the future (Lamarre et al. 2020).

In the tropics, especially in tropical rainforests, arthropods may face significant threats due to habitat loss (Wagner 2020), as well as climate change (Deutsch et al. 2008). High quality data on the population dynamics of tropical insects are urgently needed to understand the implications of insect decline (Basset and Lamarre 2019; Crossley et al. 2020; Wagner 2020). This appears even more important for the soil fauna in the tropics, which remains poorly surveyed, known and understood (André et al. 2002). For example, out of the 17 insect monitoring studies performed in the tropics with time-series > 5 years reviewed by Basset and Lamarre (2019), only one study targeted the soil fauna (litter ants: Donoso 2017). To the best of our knowledge, no long-term monitoring program is currently focused on the soil fauna in the tropics.

DNA metabarcoding (Shendure and Ji 2008; hereafter “metabarcoding”) represents a powerful approach for the screening of numerous environmental samples rich in species and for rigorous spatial and temporal monitoring (Leray and Knowlton 2015; Tang et al. 2015; Beng et al. 2016). In recent years, metabarcoding has proven useful for comparing local soil communities (Arribas et al. 2016, Oliverio et al. 2018), including in tropical rainforests (Zinger et al. 2019).

A rare effort to monitor arthropods in tropical rainforests in the long term is epitomized by the ForestGEO Arthropod Initiative in Panama (Lamarre et al. 2020). Arthropod surveys are performed within permanent forest dynamics plots monitored by the Forest Global Earth Observatories (ForestGEO; http://www.forestgeo.si.edu/; Anderson-Teixeira et al. 2015). In Panama, the ForestGEO Arthropod Initiative has been monitoring several focal taxa on Barro Colorado Island (BCI) since 2009, including important soil taxa such as ants and termites, as well as some species whose larvae are developing in the soil (Lamarre et al. 2020). A recent study on BCI compared 100 pairwise soil samples either sorted with traditional taxonomy or processed with metabarcoding, considering ants, termites and springtails extracted with Berlese-Tullgren (Bano and Roy 2016) as target taxa (Basset et al. 2020, in prep.). These three taxa represent an important proportion of animal biomass in the soil of tropical rainforests. They also play critical roles in the maintenance and regeneration of the forest, including soil turnover, nutrient cycling and decomposition, plant protection and seed dispersal (Hopkin 1997; Abe et al. 2000; Lach et al. 2010).

The BCI study indicated that (a) a positive correlation existed between the abundance of all species in taxonomic samples and their occurrence in metabarcoding samples. (b) Seasonal shifts in species occurrence and changes in faunal composition between the dry and wet seasons were correlated between taxonomic and metabarcoding samples. (c) False positive and negative species (i.e., species identified positively in the samples but unlikely to be present or species not occurring in the samples but likely to be present) represented a low proportion of species surveyed overall, owing in part to the availability of good reference libraries. These results indicated that metabarcoding could be used for the long-term monitoring of soil arthropods in tropical rainforests (Basset et al. in prep.). However, some challenges emerged. In particular, ant species were not well detected by metabarcoding as compared to samples sorted by traditional taxonomy, as well as compared to the local ant fauna known to inhabit BCI (37% of ant species not detected: Basset et al. in prep.).

It is well-known to entomologists that a range of different methods and traps are needed to provide sound estimates of arthropod diversity in tropical rainforests (Basset et al. 2012). Of interest in this context are the methods, such as Malaise or light traps, able to collect alates of social insects nesting in soil (ants, termites) during dispersal events. Since DNA barcoding can match up different castes within species of social insects (workers, soldiers, alates; Smith et al. 2015), metabarcoding data obtained from Malaise or light traps may also conveniently complement metabarcoding data obtained from soil samples.

The aims of this contribution are twofold. First, we provide a detailed field protocol to collect soil samples and arthropods in a tropical rainforest, to be later processed with metabarcoding. Field protocols to collect soil arthropods and assess soil quality exist (e.g., Römbke et al. 2006), but they are not adapted to processing samples with metabarcoding. Studies targeting the soil fauna and using metabarcoding are typically not adapted to sampling in tropical rainforests, do not provide recommendations for spatial replication, or provide insufficient details about actual sampling and sample handling in the field (e.g., Arribas et al. 2016; Saitoh et al. 2016; Oliverio et al. 2018; Zinger et al. 2019). In particular, Arribas et al. (2016) advocated using a flotation method, followed by extraction with Berlese-Tullgren apparatus. It is well known that flotation yields low capture rates, is biased against macrofaunal specimens that are less likely to survive the flotation procedure (Macfayden 1953; Southwood and Henderson 2000) and is rather tedious as it involves several filtration procedures (Arribas et al. 2016). In addition, the volume of soil surveyed by flotation is rather small (5 liters in Arribas et al. 2016 against 200 liters in this study, see methods). We therefore hope that details about our protocol will stimulate and guide further studies initiating long-term monitoring of the soil fauna in the tropics. In this article we do not discuss the laboratory protocols related to metabarcoding.

Second, we explore the detectability of soil/litter ants (Formicidae), springtails (Collembola) and termites (Isoptera) with metabarcoding and high-throughput techniques. Specifically, we take advantage of two surveys, performed on BCI and processed with metabarcoding or high-throughput barcoding, to estimate the proportion of soil/litter ant, springtail and termite species from the local fauna that can be detected by metabarcoding or barcoding samples obtained either from Berlese-Tullgren, Malaise or light trap. We then estimate whether each method provides a complementary survey of the local BCI fauna, with respect to the species richness and faunal composition of samples processed by metabarcoding. Eventually, we discuss whether these data can help us to recommend a field protocol with metabarcoding that can detect an adequate number of soil/litter ant, springtail or termite species in this tropical rainforest. In this context, we answer the question “can the soil protocol with Berlese-Tullgren extraction be significantly improved regarding the detectability of focal species by appending one or two additional field protocols, such as Malaise or light traps?”

All species occurrence in soil, Malaise and light trap samples are detailed in Suppl. material 2: Appendix S2. A nearly equal number of species was detected in Malaise trap and soil samples, whereas only half of that number was detected in light trap samples (Table 1). However, the sampling effort with light traps was clearly lower than for other methods (Table 1). Each collecting method detected a rather distinct fauna (Fig. 1). Samples from the Malaise traps detected about twice the number of ant species than the soil and light trap samples (Fig. 1a). A high number of springtail species were detected in soil samples, followed by Malaise samples, but very few species were detected in light trap samples (Fig. 1b). Termites were best detected in soil samples, and alates of termites were better detected in light trap than in Malaise samples (Fig. 1c). Overall, for the three target taxa, 87 species were uniquely detected by Malaise samples, 83 by soil samples and 35 by light trap samples (Fig. 1).

Ant species nesting in the soil were similarly well detected in all three types of samples, along the series Malaise trap > soil samples > light trap, whereas arboreal ants were mostly detected in Malaise trap samples (Suppl. material 1: Fig. S1). The pattern was rather different for termites nesting in the soil, which were mostly detected in soil samples. Arboreal termites were mostly detected in light trap samples (Suppl. material 1: Fig. S1). The pattern was also different for common species nesting or dwelling in the soil, which represented 22–36% of all species detected, depending on the taxa (Suppl. material 1: Fig. S2). The most common ant species were detected in light trap and soil samples. Common species of springtails were mostly detected in soil samples, whereas common species of soil termites were well detected in soil samples and light trap samples.

Species accumulated faster in Malaise trap samples for ants (total estimated species richness with iNEXT = 168 ± 32 [s.e.]) than for soil (74 ± 15) and light trap samples (54 ± 7; Fig. 2). This was also apparent when considering the minimum number of samples to be performed within one year for each collecting method. By contrast, springtails showed faster species accumulation in soil samples (estimated richness 43 ± 6), followed by Malaise trap samples (30 ± 2) and light trap samples (6 ± 2). The pattern was also different for termites, which showed faster species accumulation in soil samples (estimated richness 48 ± 9), followed by light trap samples (20 ± 4) and Malaise trap samples (14 ± 11; Fig. 2). For common species nesting or dwelling in the soil equivalent species accumulation curves were rather different. For Formicidae, common species accumulated faster in samples in the series light trap > soil samples > Malaise trap (Suppl. material 1: Fig. S3). For springtails the equivalent series was soil samples > Malaise trap > light trap and for termites it was soil samples > light trap (sampling effort was too low to compute an accumulation curve for termites collected by Malaise trap; Suppl. material 1: Fig. S3).

Figure 2.

Accumulation of species richness vs. the number of samples for (a) ants, (b) springtails and (c) termites, detailed for each collecting method: blue = soil samples; green = Malaise trap samples; red = light trap samples. Solid lines are interpolated curves, dotted lines are extrapolated curves to 150 samples. The vertical lines indicate, for each collecting method, the targeted minimum number of samples within one year (see text).

The NMDS plots confirmed that faunal composition was rather different in samples obtained by the three collecting methods. Differences in faunal composition were most marked when all species were considered together (F_2,135 = 29.17, p = 0.001; Fig. 3), then for ants and springtails (F_2,123 = 29.74, p = 0.001 and F_2,109 = 29.76, p = 0.001, respectively), and for termites (F_2,115 = 26.56, p = 0.001).

Figure 3.

NMDS plots for (a) all species, (b) Formicidae, (c) Collembola and (d) Isoptera. Plots of samples (blue = soil samples, green = Malaise trap samples, red = light trap samples) in the first two axes of the ordinations. The ellipses represent 95% confidence limits around the centroids of each method.

Entomologists are well aware that an array of collecting methods are necessary to survey most arthropod species thriving in even relatively small areas (Basset et al. 2012), and this is also true for soil arthropods (André et al. 2002). These challenges are compounded when processing samples by metabarcoding because detection rates may differ among species (Saitoh et al. 2016; Elbrecht et al. 2019). Hence, the faunal composition of soil samples sorted manually may differ from that in soil samples processed with metabarcoding. This is not necessarily detrimental to long-term monitoring, but the possible bias and limitations of metabarcoding must be acknowledged, and perhaps corrected either in the field, laboratory or via bioinformatics. Here, we are concerned with field protocol and provided a detailed protocol to collect soil samples and arthropods in a tropical rainforest to be processed with metabarcoding. We then evaluated whether samples processed with metabarcoding and obtained with other methods commonly used by entomologists, Malaise and light traps, may be able to detect a substantial number of soil-dwelling species not detected in soil samples, for the target taxa of Formicidae, Collembola and Isoptera in this tropical rainforest. We show that a substantial number of additional species are detected in metabarcoding samples obtained with Malaise and light traps, including species that are common and amenable to statistical analyses of long-term trends. This validates the approach of using different field protocols to obtain samples processed by metabarcoding with the purpose of monitoring in the long term a reasonable share of the focal species inhabiting BCI.

We restricted our study to three dominant taxa of soil/litter arthropods. We do not have sufficient information to discuss cogently how well metabarcoding may detect other invertebrates in the soils of BCI, for lack of sound DNA libraries for these other taxa. Among others they may include earthworms, mites, spiders, myriapods, nematodes and annelids.

We showed that each collecting method detected a rather distinct fauna, in terms of species richness or faunal composition. Overall, when considering the three target taxa, soil and Malaise trap samples detected 213 species (73%) of all species detected by the three collecting methods. However, patterns were different between taxa, with Malaise trap samples detecting many ant species, whereas soil samples were more efficient at detecting springtail and termite species. Both Malaise and light traps detected a high proportion of arboreal ants and termites, representing 35% of the total of species of social insects detected. Further, with respect to long-term monitoring of soil-nesting or soil-dwelling common species (more amenable to statistical trends), the best combination of two methods, in terms of maximizing the detection of common species, were soil and light trap samples, detecting 94% of the total of the species common in the soil. This emphasizes that different methods may be suitable for different research goals. For a snapshot survey of the local area (including the soil and arboreal fauna), metabarcoding samples obtained with all three collecting methods would obviously be best, followed by a combination of soil plus Malaise samples, if funding is limited. For generating time-series in the long-term, soil samples detected 72% of common species and would be the choice method, followed by a combination of soil plus light trap samples. These results were not too different when considering species accumulated and the ideal minimum number of samples that should be performed within a year to monitor common species adequately (Basset et al. 2013).

Several factors may hinder a direct comparison of the data available, even when considering the number of samples collected in species accumulation curves. First, we cannot consider reliable estimates of species abundance in samples processed with metabarcoding with the present technology (Lamb et al. 2018; Creedy et al. 2019). Second, it is difficult to compare samples from the different methods with respect to spatial (soil samples > light trap > Malaise trap and seasonal replication (Malaise trap > soil samples > light trap). Hence, we can only conjecture what would happen in our BCI system if we were to obtain with an ideal protocol each year 100 soil samples (10 spatial replicates, two sampling seasons), 40 one-week Malaise trap samples (10 replicates, 4 annual surveys) and 80 light trap-night samples (10 two-nights replicates, 4 annual surveys). In this case, we can assume that (1) more ant species nesting in the soil would be detected in our samples (the soil ant species detected in this study represent about 50% of the total species richness of soil-nesting ants on BCI; D. Donoso, unpubl. data); (2) we may also detect more springtail species with more replicates of Malaise trap (and complete analysis of springtails, see methods), although not all these species may be dwelling in the soil/litter (see below); and (3) a higher number of species may be amenable to statistical analysis (common species). Third, de Kerdrel et al. (2020) noted, in their comparison of arthropod samples obtained with different methods and processed with high-throughput barcoding, that fewer sequences were recovered from soil trapping methods (Berlese and pitfall traps) as compared to methods targeting flying insects or arthropods perched on vegetation (Malaise traps, beating). These authors suggested that samples with a considerable amount of soil and debris may include compounds that may inhibit PCR. Water in the soil may also increase DNA degradation (de Kerdrel et al. 2020).

We used high-throughput barcoding for Malaise trap samples instead of the metabarcoding used for soil and light trap samples. Because longer sequences were obtained with the former, BIN assignment was probably more accurate for Malaise trap samples, resulting in better quality data (deWaard et al. 2019a). Further, in this study, we compared species with BINs among methods for a convenient and straightforward approach, as opposed to also considering operational taxonomic units (OTUs) of unknown species without BINs. It is legitimate to also ask whether our results would be different when considering these possible extra OTUs. We believe that at least for ants and termites, differences would have been negligible because of the near complete DNA reference libraries available for these taxa inhabiting BCI.

Irrespective of the collecting method used, the number of ant species recovered by metabarcoding remained rather low. This may be related to lower extraction rate and sequence recovery. For instance, out of 14 arthropod orders screened by de Kerdrel et al. (2020), Hymenoptera (including ants) showed the lowest sequence recovery rate. Implementing our three collecting methods with an ideal sample size of 100 soil, 40 Malaise and 80 light trap samples annually, would generate reliable time-series for the 24 ant species that we labeled as being common and amenable to statistical analysis of long-term trends. These 24 species represent 11% of the total number of soil/litter ant species inhabiting BCI. This proportion is similar or higher to that for the soil/litter ant species amenable to similar statistical analysis (8%) when acquiring each year 50 Winkler samples sorted manually. Replacing Winkler samples by soil samples and adding Malaise trap samples would increase field work slightly (i.e., human labor to survey traps and handle samples) as compared to our current annual sampling scheme (Lamarre et al. 2020). However, this would decrease the time related to specimen processing and identification considerably, not to mention the value added of obtaining metabarcoding data of non-target taxa for all three collecting methods.

In this contribution, springtails were all assumed to be soil-dwelling. This is certainly an oversimplification, given that many species also thrive in arboreal habitats, even though their biology remains poorly known (Hopkin 1997). Although springtails are often neglected in Malaise trap samples, they are by no means infrequent in these samples (Greenslade and Florentine 2013) and can locally represent a substantial proportion of catches albeit with low species richness (Palacios-Vargas and Gómez-Anaya 1993; Palacios-Vargas et al. 1998). This concerns mainly the family Entomobryidae and in the Neotropics genera such as Seira, Entomobrya or Lepidocyrtus, as well as Paronellidae such as Trogolaphysa and often Salina, which may climb on the mesh of the trap (J.G. Palacios-Vargas, pers. obs.). By contrast, this is hardly possible in light traps as they were suspended at 1.3 m height (Lucas et al. 2016). While both springtails and termites may fall victim to predatory ants (Basset et al. 2020), and thus be detected in metabarcoding samples, this is not necessarily detrimental to monitoring, as it ascertains the presence of the prey in samples. There are at least 104 species of springtails on BCI (Basset et al. 2020), but currently only 33% of these species have BINs. Hence, we can conservatively estimate that a sampling scheme based on 100 soil samples annually would allow for the monitoring of at least the 19 species that we labeled as being common and amenable to statistical analysis of long-term trends (18% of the total number of species). This proportion can probably be increased as springtail BINs are added to the reference libraries.

There are at least 62 species of termites on BCI (Basset et al. 2020; Y, Basset et al., unpubl. data), including about 31% of specialized humus-feeding species (mostly Apicotermini) species, 53% of wood-eating species in the understorey and 16% of wood-eating arboreal species. Soil metabarcoding samples detected many of the humus-feeding species. A protocol based on 100 soil and 80 light trap samples annually would probably allow for the monitoring of about 16% of termite species (the 10 soil species labeled as common), although wood-eating species may be underrepresented. Light trap data would allow for monitoring alates of a few more arboreal species.

When time and funding do not allow the combined use of the three collecting methods tested in this study, Malaise traps have the advantage over light traps of not requiring electrical power in the field, as well as requiring less time to set up and run in the field. Yang et al. (2014) proposed that soil and Malaise trap samples processed with metabarcoding be tested as a candidate method for rapid environmental monitoring in terrestrial ecosystems. Relevant to this, it is also legitimate to ask whether it is sound to monitor workers (mostly in soil samples) or alates (mostly Malaise and light trap samples) of social insects, since metabarcoding does not allow for distinguishing between the two casts (but barcoding does after verification of vouchers). For example, out of the +16,000 specimens of ants sorted from the 10 BCI Malaise traps surveyed during the period of 2002–2017, 95.54% were male alates, 4.40% were female alates and only 0.06% were workers (D. Donoso et al. unpubl. data). There are undeniable advantages to monitor alates as, in some cases, workers are far more difficult to survey than alates. Because workers’ swarms are constantly on the move, the spatial distribution of army ants is highly patchy (Kaspari et al. 2011) and they are difficult to survey in soil or Winkler samples, whereas alates are readily caught in numbers in light traps. Likewise, all termites collected in light traps samples are alates and include a good proportion (60%) of species that are arboreal and hence rarely collected in monitoring samples obtained from the forest understorey.

Based on our results, a multiple protocol including 100 soil, 40 Malaise and 80 light trap samples processed annually by metabarcoding would allow monitoring of at least 11%, 18% and 16% of species of soil/litter ants, springtails and termites, respectively, present on BCI. All these species have been formally vouchered, and many are identified (Suppl. material 2: Appendix S2), so that in the future it may be possible to discuss cogently potential reasons for population decline or increase over time. Although these common species may not represent a large share of species richness, they represent a high proportion of the BCI fauna in terms of abundance. For example, common soil/litter ant species as defined in this study represent about 80% of all ant individuals collected annually by Winkler samples for our long-term monitoring scheme. The sampling effort suggested may be suitable for the area of BCI but may vary for other rainforest locations. Figure 4 summarizes how we envision the long-term monitoring of soil arthropods at BCI. This provides a starting point for pilot studies concerned with the long-term monitoring of arthropods in tropical rainforests.

Figure 4.

Long-term monitoring scheme for the soil arthropods in the tropical rainforest of Barro Colorado Island, Panama. Horizontal arrows colored in blue refer specifically to this study, pointing to results).

We thank ForestGEO and the Smithsonian Tropical Research Institute for logistical support. Richard Ctvrtecka, Petr Blazek and Pamela Polanco helped with light trap samples. Scott Miller commented on an early draft of the manuscript.

This study was supported by SENACYT (FID16-070 to YB and HB), the Czech Science Foundation (GAČR 20-31295S to YB) and ForestGEO. Grants from the Smithsonian Institution Barcoding Opportunity FY012, FY015 and FY018 (to YB) and in-kind help from the Canadian Centre for DNA Barcoding as part of the iBOL program allowed sequencing specimens. YB and HB were supported by the Sistema Nacional de Investigación, SENACYT, Panama. G.P.A.L. was supported by a GAČR Junior Grant (19-15645Y).

The field protocol was deposited at protocols.io (dx.doi.org/10.17504/protocols.io.bj9gkr3w). Sequences and BINs used as reference libraries are publicly accessible in BOLD projects as indicated in the text, the dataset for Collembola as dx.doi.org/10.5883/DS-COLLMAL. Sequences are also deposited in GenBank databases under accession numbers KP845288–KP849461, KU745531–KU745532, KX072335–KX072563, MF922335–MF970719, MG030727, MK758129–MK770080, MN345316–MN621065 and MT357731. Suppl. material 2: Appendix S2 provides all the information about the identity of species and their occurrence in samples. The raw sequences were deposited at the Sequence Read Archive (SRA; https://www.ncbi.nlm.nih.gov/bioproject/PRJNA668155).