A high‐throughput amplicon sequencing approach for population‐wide species diversity and composition survey

Abstract

Management of agricultural pests requires an understanding of pest species diversity, their interactions with beneficial insects and spatial‐temporal patterns of pest abundance. Invasive and agriculturally important insect pests can build up very high populations, especially in cropping landscapes. Traditionally, sampling effort for species identification involves small sample sizes and is labour intensive. Here, we describe a multiprimer high throughput sequencing (HTS) metabarcoding method and associated analytical workflow for a rapid, intensive, high‐volume survey of pest species compositions. We demonstrate our method using the taxonomically challenging Bemisia pest cryptic species complex as examples. The whiteflies Bemisia including the”tabaci” species are agriculturally important capable of vectoring diverse plant viruses that cause diseases and crop losses. Our multiprimer metabarcoding HTS amplicon approach simultaneously process high volumes of whitefly individuals, with efficiency to detect rare (i.e., 1%) test‐species, while our improved whitefly primers for metabarcoding also detected beneficial hymenopteran parasitoid species from whitefly nymphs. Field‐testing our redesigned Bemisia metabarcoding primer sets across the Tanzania, Uganda and Malawi cassava cultivation landscapes, we identified the sub‐Saharan Africa 1 Bemisia putative species as the dominant pest species, with other cryptic Bemisia species being detected at various abundances. We also provide evidence that Bemisia species compositions can be affected by host crops and sampling techniques that target either nymphs or adults. Our multiprimer HTS metabarcoding method incorporated two overlapping amplicons of 472 bp and 518 bp that spanned the entire 657 bp 3' barcoding region for Bemisia, and is particularly suitable to molecular diagnostic surveys of this highly cryptic insect pest species complex that also typically exhibited high population densities in heavy crop infestation episodes. Our approach can be adopted to understand species biodiversity across landscapes, with broad implications for improving transboundary biosecurity preparedness, thus contributing to molecular ecological knowledge and the development of control strategies for high‐density, cryptic, pest‐species complexes.

1. INTRODUCTION

The ecology of a species is linked by definition to its species status. It underpins our understanding of species diversity in conservation genetics, biosecurity preparedness and developing effective pest management strategies. The correct identification of cryptic species can be challenging and often involves the use of molecular approaches. Timely identification and characterisation of pest species will greatly assist the development of effective management strategies to minimise impacts of crop losses due to pests on food security. Furthermore, techniques that enable the rapid screening and accurate identification of large numbers of individuals can accelerate research on a target pest species and lead to important insights into species ecology and biology.

The use of the maternally‐inherited mitochondrial cytochrome c oxidase subunit I (mtCOI) partial gene 5' (N) terminal region (i.e., typically one of 648, 650, or 658 bp) in “DNA barcoding” via the Sanger sequencing method has contributed to molecular diagnostics of species (e.gBerry et al., ²⁰⁰⁴; Boykin et al., ²⁰⁰⁷; Dinsdale et al., ²⁰¹⁰; Hebert et al., ²⁰⁰³; Ward et al., ²⁰⁰⁵), and provided insights into the potential species diversity where cryptic species may coexist (e.gHebert et al., ²⁰⁰³; Rao et al., ²⁰¹⁸; Tay et al., ²⁰¹⁶). Cryptic species such as the hemipteran whitefly Bemisia tabaci and related “non‐tabaci” species are excellent examples where putative species identification can be achieved via molecular characterisation of the partial mtCOI gene (e.gLee et al., ²⁰¹³; Mugerwa et al., ²⁰¹⁸), provided that PCR artefacts, such as NUMTs and/or poor quality sequence trace files and/or gene regions, were first removed (e.g., Elfekih et al., ²⁰²¹; Kunz, Tay, Court, et al., ²⁰¹⁹; Kunz et al., ²⁰¹⁹; Tay, Elfekih, Court, et al., ²⁰¹⁷; Vyskočilová et al., ²⁰¹⁸). Within the B. tabaci cryptic species complex, over 34 putative B. tabaci cryptic species have been reported (Kunz, Tay, Elfekih, et al., 2019; Lee et al., ²⁰¹³; Mugerwa et al., ²⁰¹⁸; Vyskočilová et al., ²⁰¹⁸). The cryptic species status of the B. tabaci complex reflected the significant challenges associated with the correct identification of species status (reviewed by Boykin et al., ²⁰¹⁸; De Barro et al., ²⁰¹¹; Tay & Gordon, ²⁰¹⁹). Species of the B. tabaci complex are notorious around the world as effective vectors of viruses that cause significant diseases of crop plants (Polston et al., 2014). In east Africa, B. tabaci species transmit diverse viruses that cause diseases in cassava. Cassava mosaic disease (CMD) and cassava brown streak disease (CBSD) have led to significant crop losses and loss of production by smallholder farmers (Legg et al., 2011; Legg, Shirima, et al., ²⁰¹⁴). Understanding the Bemisia species present in east Africa, and their patterns of abundance and distribution in different farming landscapes, is a necessary first step to managing this pest problem (Macfadyen et al., 2018).

In African nations such as Tanzania, Malawi and Uganda, cassava is typically grown by smallholder farming families, surrounded by noncassava crops such as tomatoes, sweet potatoes, groundnuts, soybean, and maize. Native ecosystems that include unmanaged grazing lands, weeds, and shrubs, are also present and contain plants which are capable of supporting a diversity of whitefly species. Some may be novel species, which are yet to be characterised by morphological, behavioural and molecular approaches (e.g., Mugerwa et al., ²⁰¹⁸; Vyskocilova et al., ²⁰¹⁹, Vyskocilova et al., 2018). Globally, cassava is an important carbohydrate source for over 500 million people (ARC 2014, accessed 21 April 2019), and is planted in Africa, South America, and Asia/South East Asia. Thirteen of the top 20 global cassava producers are African countries (i.e., Nigeria, Angola, Ghana, Democratic Republic of Congo, Malawi, United Republic of Tanzania, Cameroon, Mozambique, Benin, Sierra Leone, Madagascar, Uganda and Rwanda (<www.worldatlas.com>up‐dated 25 April 2017; accessed 21 April 2019). In other countries such as Thailand, Vietnam and Cambodia, it is grown as a source of high‐quality starch and is used in diverse food products (Delaquis et al., 2018; Gotz & Winter, ²⁰¹⁶; Graziosi et al., ²⁰¹⁶; Kumar et al., ²⁰¹⁷). Therefore, having techniques that enable rapid and cost‐effective surveillance to investigate pest species compositions especially in cassava (and other food crops, that is, tomatoes; other cash crops, i.e., cotton, ornamental plants) production landscapes are incredibly valuable. While identifying cryptic species diversity can best be achieved through molecular characterisation of appropriate genomic regions, adapting the concept of high‐throughput sequencing (HTS) platforms to answer questions around species diversity and evolutionary interactions can be challenging.

The HTS platform (e.g., Illumina's iSeq100, MiniSeq, MiSeq, NextSeq 550, NextSeq 1000 & 2000, NovaSeq 6000; Oxford Nanopore Technologies [ONT] range that is, Flongle, MinION, GridION, and PromethION; PACBIO's Single Molecule, Real‐Time [SMRT] Sequencing and Sequel System/Sequel II System/Sequel IIe System) offers a range of scalable solutions (i.e., to best suit the research institute's and target country's infrastructure) for surveillance of insect species (e.g., see Hebert et al., ²⁰¹⁸; Srivanthsan et al., ²⁰²¹), including cryptic species that reach high population densities, provided that methods are sufficiently flexible to accommodate study designs tailored to address targeted research questions. For example, understanding host‐parasitoid relationships, cryptic species diversity, host‐plant relationships, and identifying keystone species are fundamental ecological questions common to diverse ecosystems. Addressing these ecological and evolutionary questions can be challenging and requires scaling up of the data collection processes. HTS platforms represent a solution when combined with appropriate species diversity databases, flexible bioinformatics and associated analytical pipelines, and optimised molecular methods.

The metabarcoding approach that incorporates the power of HTS platform and the utility of the DNA barcoding gene region aims to provide species identification at both population and community levels (i.e., not aimed at characterising all known mtCOI haplotypes at the individual level). In the cryptic B. tabaci species complex however, molecular diagnostics is particularly susceptible to suboptimal primer efficacies (Elfekih et al., 2018; Mugerwa et al., ²⁰¹⁸; Shatters et al., ²⁰⁰⁹; Tay, Elfekih, Court, et al., ²⁰¹⁷). Partial or inconsistent Polymerase Chain Reaction (PCR) amplification, especially when targeting coextracted DNA of individual cryptic species that may be found together, can lead to inaccurate conclusion of species composition within the surveyed agroecosystems. For example, a minor species may become overrepresented within a population if coamplified within a pool of major species but enjoys superior primer efficacies compared with the major species. Also, given the time, effort and cost necessarily associated with the preparation of HTS DNA libraries, especially where accurate input genomic DNA concentration is needed, metabarcoding should incorporate a “back‐up” step. Such a step may involve a second set of primers, ensuring the amplification of the same population, should the first set of primers fail and thereby avoiding a total PCR failure. Such a back‐up design also enables capturing of species that are missed by the first primer pairs, and vice versa, or where longer DNA region is needed to improve molecular diagnostics of cryptic species complex. Provided that the primer pairs developed were sufficiently robust, within a pest insect species complex such as in the whitefly cryptic species, beneficial insects such as parasitoid species can also be investigated with appropriate sampling of pest‐insect life stages (e.g., whitefly nymphs).

Here, we describe the HTS metabarcoding method and primers developed for surveying the Bemisia cassava whitefly, related B. tabaci cryptic species and associated parasitoid species complex, and the workflow for processing the large volume of HTS amplicon sequence reads generated. Our goal was to develop a multiprimer metabarcoding technique that enabled us to determine the dominant Bemisia species in any cassava field quantitatively, using a robust sample size in terms of numbers of individuals. We provide a verification of primer efficacies and both successful and failed species delimitation rates of the multiprimer method, and confirmed the method and primer efficacies via Sanger sequencing of individual field‐sampled novel whiteflies species. Our samples were collected from geographically diverse cassava cropping landscapes in Uganda, Malawi and Tanzania, so test the limitations and usefulness of this approach especially for countries with significant infrastructure and economic challenges for researching agricultural pests impacting food security.

2. MATERIALS AND METHODS

2.1. Samples used in PCR primer efficacy tests

The DNA of whitefly cryptic species from Africa, Asia, South America, and Australia were individually extracted using the Qiagen Blood and Tissue DNA extraction kit (Cat no. 69506). Extracted DNA of all samples to be used for testing of PCR primer efficacies were individually eluted in 27.5 µl of EB. Whitefly species status was confirmed by PCR and Sanger sequencing using the primer set wfly‐COI‐F1/R1 and/or wfly‐COI‐F2/R2 (Supporting Information for Online Publication I). Primer development and Sanger sequencing followed the methods of Elfekih, Tay, et al. (2018). Sanger sequencing was carried out at the BRF at JCSMR ANU, Canberra, ACT. Whitefly species used to verify primer efficacies and their respective GenBank accession numbers are listed in Table 1.

TABLE 1.

Bemisia species used to confirm PCR primer efficacies

Species	GenBank	Field sample identifiers	Country/source	Reference
MED‐ASL	MH357342	60,132	Malawi (sweet potato) Lat −12.90243413, long 34.28411598	Vyskocilova et al. (2019)
MEAM1	KY951449		Peru	Tay, Elfekih, Polaszek, et al. (2017)
B. afer	AF418673	60,266	Malawi, on ornamental tree cassava in township.	Maruthi et al. (2004)
Bemisia sp01	MN646951	30,350	Tanzania (Lantana weed) Lat −2.742276, Long 33.021902	New Bemisia sp. (this study)
SSA1	JQ286457	40,011	Tanzania (cassava) Lat −6.5221814, Long 38.924329	Mugerwa et al. (2012)
SSA2	AY057141		NRI laboratory culture	Legg et al. (2002)
SSA3	KM377923		NRI laboratory culture	Ghosh, S., Gowda, M., & Bouvaine, S. (September 27, 2014)
NW1	MK386668		Barbados	Kunz, Tay, Court, et al. (2019)
Asia II_1	MF497064		Bangladesh	Khatum, M. F. & Lee, K. Y. (May 15, 2018)
Asia1	MF497045		Bangladesh	Khatum, M. F. & Lee, K. Y. (May 15, 2018)
Lipaleyrodes (Bemisia)	MN056066		Australia	Kunz, Tay, Elfekih, et al. (2019)

With the exception of MN646951, all GenBank accession numbers provided represent partial mtCOI sequences of Bemisia species tested where 100% sequence identity against publicly available sequences have been detected. Note: KY951449 and MK386668: nucleotide position: 782–1,438.

To ascertain PCR efficacies of the redesigned metabarcoding primers, various hypothetical field scenarios were simulated. These scenarios involved pooling of different genomic DNA (gDNA) that belonged to 11 separately extracted individuals of Bemisia species (Table 2) with predetermined species status based on the standard partial mtCOI data set (Kunz, Tay, Elfekih, et al., 2019). These hypothetical scenarios included: (I) a two‐species co‐occurrence scenario, in which there is a dominant Bemisia whitefly species (e.g., SSA1) and a minor species present in different predetermined proportions (i.e., Table 2: Mixed‐Ia [5%], Mixed‐Ib [2.5%], and Mixed‐Ic [1%]), and (II–IV) scenarios assuming co‐occurrence of multiple mixed Bemisia cryptic pest species at different proportions based on estimates of gDNA concentration. With the exception of the test scenario “Mixed‐III” and “Mix‐IV” which were each replicated twice, all remaining test scenarios were replicated three times to estimate error rates and ensure protocol robustness.

TABLE 2.

Test scenarios to ascertain primer efficacies and species detection sensitivity by high‐throughput sequencing platform

Detection sensitivity	Selected known Bemisia species											Others a
	SSA1	SSA2	SSA3	Bemisia. Sp01 (MN646951)	MEAM1	MED	NW1	Asia1	AsiaII−1	Bemisia sp. (MN056066)	B. afer
Mixed‐Ia (1:20)	95 (88.9 ± 0.3)	0	0	0	0	5 (6.1 ± 0.7)	0	0	0	0	0	0 (5 ± 0.01)
Mixed‐Ib (1:40)	97.5 (94.8 ± 0.3)	0	0	0	0	2.5 (1.6 ± 0.1)	0	0	0	0	0	0 (3.6 ± 0.2)
Mixed‐Ic (1:100)	99 (92 ± 0.7)	0	0	0	1 (1.2 ± 0.4)	0	0	0	0	0	0	0 (7.5 ± 1.3)
Mixed‐II	0	0	0	0	0	0	25 (11.4 ± 0.7)	25 (12.7 ± 0.9)	25 (28 ± 1)	25 (20.2 ± 1.2)	0	0 (27.9 ± 2.1)
Mixed‐III	0	0	0	0	0	0	30 (24.1 ± 0.6)	50 (30.3 ± 0.4)	10 (8.8 ± 0.5)	10 (14 ± 0.4)	0	0 (24.1 ± 1.4)
Mixed‐IV	58 (42.5 ± 4.9)	8.8 (12.5 ± 1.8)	8.8 (10.5 ± 0.7)	6 (3 ± 1.8)	6 (7 ± 0.6)	6 (3.1 ± 0.7)	0	0	0	0	6 (2.9 ± 0.6)	0 (18.2 ± 3.1)

Average ± standard deviation of detection sensitivity (i.e., detection of observed proportion of taxa) for each category is provided based on amplicon reads as generated. “Others” include amplicons associated with NUMTs (e.g., MEAM2; Tay, Elfekih, Court, et al., ²⁰¹⁷), where the MEAM1 species was used in the mixed‐Ic and mixed‐IV scenarios), unknown Bemisia genomic regions, parasitoids (Encarsia, Eretmocerus) partial mtCOI gene region, bacteria including secondary symbionts, fungal, and of unknown origins. “0”, species not considered in test scenarios.

^a Only the first 10–15 amplicon contigs were searched for sequence identity by nucleotide blast (blastn) (i.e., optimised for somewhat similar sequences) search against the nonredundant (nr) NCBI DNA database. Verification of sequence identity was accomplished using the sanitized nucleotide database of Kunz, Tay, Elfekih, et al. (2019) as subject sequences, and using the NCBI's Basic Local Alignment Search Tool (BLAST) “blastn suite‐2sequences” option and using only top hits (i.e., those with 97%–100% sequence identity). PCR primer amplification efficacies were ascertained on species representing significantly divergent evolutionary lineages (Kunz, Tay, Elfekih, et al., 2019) that included a novel Bemisia species (MN646951; From Tanzania field code: 30350, see Macfadyen et al., ²⁰²¹), Lipaleyrodes (Bemisia) from Queensland, Australia (MN056066), B. afer, the African cassava whitefly species (SSA1, SSA2, SSA3), and species within the B. tabaci cryptic complex from diverse geographic origins (e.g., New World 1 (NW1); Mediterranean (MED), Middle East Asia‐Minor 1 (MEAM1); Asia II‐1, and Asia 1).

2.2. Field samples survey methods

Cassava fields in Uganda, Tanzania, and Malawi were surveyed for Bemisia cryptic species between 2015–2016 (Figure 1). In total, we conducted three data collection trips across a 2‐year period (trip 1, 1 August 2015–26 August 2015, trip 2, 5 April 2016–25 April 2016, trip 3, 29 October 2016–9 November 2016). In each region, we selected up to 10 cassava fields to survey and sample for nymph and adult Bemisia whiteflies (for sampling details see also Macfadyen et al., ²⁰²¹). Briefly, up to c. 50 adults were collected using an aspirator from the top leaves of cassava plants of known variety in four Malawi field sites (Table 2). Nymphs were collected from all sites by selecting leaves that were lower down on the plant (and so had visible nymphs in the second to fourth instar age range on the underside of the leaf). The top lip of 2 cm diameter circular plastic vial (5 ml volume) was used to cut small leaf discs that contained nymphs. These discs were placed in the vial filled with 99% ethanol. As a comparison, nymphs from other crops surrounding the focal cassava field were also collected. All ethanol‐preserved samples were stored in –20°C prior to being processed in the laboratory, where nymphs were dislodged from leaves and placed in fresh Eppendorf (1.5 ml) test tubes in batches of between 20 and 40 individuals per field (note that in some cases, fields sampled for Bemisia contained <20 individuals). For our research question we selected nymphs that had no obvious signs of parasitism by hymenopteran parasitoids in order to optimize the B. tabaci DNA in the sample. We randomly selected 26 field‐collected Bemisia populations from Malawi, Uganda and Tanzania for analysis to demonstrate the efficacies of the HTS mtCOI amplicon approach under field conditions (Table 3).

FIGURE 1.

(a) Bemisia cryptic species sampling sites from Uganda, Tanzania and Malawi, with field site codes from Tanzania (i.e., T1–T6) provided in a, Uganda (i.e., U1–U9) in panel (b), and Malawi (i.e., M1–M11) in panel (c). GPS coordinates of each sampling site are provided in Table 3. The maps of Africa were produced using Mapchart <https://mapchart.net>

TABLE 3.

Twenty‐six field‐collected whitefly samples from Malawi, Uganda and Tanzania analysed by high‐throughput amplicon sequencing

Country	Site ID (site code)	Life stage	Host plants (variety)	Lat	Long	Species composition	Total reads	Reads Identified (%)
Malawi	60,109/60,121 (M1)	N, n = 26	Cassava (Gomani)	–12.639611255	34.17544665	SSA1: 12.7, Baf: 12.8, Enc: Y, NUMTs: 0.4	43,652	43,383 (99.35%)
	60,109/60,121 (M2)	A, n = 7	Cassava (Gomani)	–12.639611255	34.17544665	SSA1: 6.4, Baf: 0.4, NUMTs: 0.01	54,392	54,135 (99.53%)
	60,013/60,126 (M3)	N, n = 22	Cassava (Gomani)	–13.03129506	34.29746802	SSA1: 21.7, NUMTs: 0.17	68,562	68,224 (99.51%)
	60,013/60,126 (M4)	A, n = 8	Cassava (Gomani)	–13.03129506	34.29746802	Baf: 7.84, NUMTs: 0.09	44,424	44,166 (99.42%)
	60,127/60,250 (M5)	N, n = 20	Cassava (Gomani)	–13.11325369	34.32143017	SSA1: 5.1, Baf: 14.5, NUMTs: 0.07	34,660	32,169 (92.81%)
	60,127/60,250 (M6)	A, n = 20	Cassava (Gomani)	–13.11325369	34.32143017	SSA1: 15.9, Baf: 3.8, NUMTs: 0.13, Bac: 0.01	62,600	62,193 (99.35%)
	60,032/60,263 (M7)	N, n = 18	Cassava (Manyokola)	–14.11054382	33.81826233	SSA1: 0.63, Baf: 14.7, U: 0.08, NUMTs: 1.45	43,156	40,341 (93.48%)
	60,032/60,263 (M8)	A, n = 15	Cassava (Manyokola)	–14.11054382	33.81826233	SSA1: 5.3, Baf: 5.9, U: 0.56, NUMTs: 2.6	72,172	66,784 (92.53%)
	60,010/60,242 (M9)	N, n = 30	Cassava (Beatrice)	–12.804086285	34.21788623	SSAS1: 18.7, Baf: 10.1, NUMTs: 0.91, Enc: Y, Eret: Y	63,506	62,888 (99.03%)
	60,037/60,265 (M10)	N, n = 40	Cassava (Manyokola)	–14.1593299559021	33.8348985175844	SSA1: 0.6, Baf: 38.7, Enc: Y, NUMTs: 0.16	44,808	44,228 (98.71%)
	60,039/60,040 (M11)	N, n = 40	Cassava (Manyokola)	–13.9888087453993	33.6391004272323	SSA1: 38.4, Baf: 0.7, Eret: Y, T: 0.2, NUMTs: 0.52	36,530	36,207 (99.12%)
Uganda	10,109/10,230 (U1)	N, n = 40	Cassava (TME_14)	0.93830798	33.16688609	SSA1: 32.8, SSA2: 3.4, Eret+Enc: Y (0.5), NUMTs: 2	49, 601	47,906 (96.58%)
	20,018/20,019 (U2)	N, n = 40	Cassava (Omongole)	–0.98750407	31.42085652	SSA1: 32.8, Baf: 2.5, Eret+Enc: Y (4.1), NUMTs: 0.3	38,082	37,752 (99.13%)
	10,085/10,336 (U3)	N, n = 40	Cassava (TME_14)	0.87832027	33.19462426	SSA1: 35.6, Baf: 0.13, Eret+Enc: Y (2.8), NUMTs: 0.96	68,176	67,185 (98.55%)
	431,30 (U4)	N, n = 20	Cassava (TME_14)	2.35231436	32.93393845	SSA1: 9.9, SSA2: 3.1, Baf: 2.4, Eret: Y, NUMTs: 1.8	7,486	6,531 (87.24%)
	43,131 (U5)	N, n = 40	Cow pea	2.35246646	32.93482051	SSA1: 8.1, MED: 21.8, U: 1.6, Eret: Y, NUMTs: 2.1	6,674	5,883 (88.15%)
	53,371 (U6)	N, n = 40	Sweet potato	–0.52519054	31.6441164	SSA1: 1, MED: 26.2, IO: 0.4, Ug1: 11, NUMTs: 0.14, Eret: Y (0.1)	58,646	56,426 (96.21%)
	10,362 (U7)	N, n = 8	Cassava (Nase_14)	0.92017081	33.21595328	SSA1: 5.1, SSA2: 1.6, NUMTs: 0.14	7,305	6,202 (84.90%)
	33,026 (U8)	N, n = 20	Other crop, Malakwang Rosella‐like	0.97703406	33.04814937	SSA1: 0.6, MED: 19.2, Eret: Y, NUMTs: 0.2	54,877	54,726 (99.72%)
	33,042 (U9)	N, n = 20	Cassava (Nase_14)	1.00622674	32.98948722	SSA1: 18.4, MED: 0.03, Eret+Enc: Y (1.3), NUMTs: 0.18	47,014	46,812 (99.57%)
Tanzania	30,145/30,157 (T1)	N, n = 40	Cassava (Busanagulwa)	–2.388214545	33.05174421	SSA1: 33, Eret+Enc: Y (6.5), NUMTs: 0.27	58,826	58,407 (99.29%)
	30,052/30,061 (T2)	N, n = 40	Sweet potato in cassava field	–2.41603623925181	33.0826151071912	SSA1: 39.3, Baf: 0.4, Enc: Y, NUMTs: 0.3	62,988	62,589 (99.37%)
	40,013/40,014 (T3)	N, n = 40	Cassava (Kiroba)	–7.15118632229587	39.1807717492072	SSA1: 31.5, Baf: 6, NUMTs: 1.1	71,134	68,629 (96.48%)
	40,279/40,288 (T4)	N, n = 40	Cassava (Kiroba)	–6.58507491183717	38.8436025425142	SSA1: 32.4, Baf: 1.1, Eret+Enc: Y (5.8), NUMTs: 0.34	49,610	49,154 (99.08%)
	30,359 (T5)	N, n = 12	Cassava (unknown)	–2.802858945	33.05936783	SSA1: 9.5, U: 1.2, NUMTs: 0.1	6,931	6,219 (89.73%)
	70,120 (T6)	N, n = 20	Cassava (Marfaransa)	–6.59705482	38.81119587	SSA1: 14.2, MED: 0.1, Eret+Enc: Y, NUMTs: 4.3	6,029	5,912 (98.06%)

To calculate the estimated detected individuals for each species per sampling site, the proportion of amplicons assigned to each putative species was first determined by mapping to the reference species sequence (outlined in Figure 2) and then divided by the total Illumina raw output fastq (i.e., Total reads) reads from the pooled population.

Life stages were nymphs (N); and adults (A). Sample size (n). Detected individuals for each species were estimated from proportions of amplicons. Bemisia cryptic species detected were: B. tabaci Indian Ocean (IO), Mediterranean (MED), B. sp. unknown (U), B. afer (Baf), B. sp. “Uganda” (Ug1), African cassava whitefly sub‐Saharan Africa 1 (SSA1), sub‐Saharan Africa 2 (SSA2), Trialeurodes sp. (T), Encarsia spp. (Enc), Eretmocerus spp. (Eret), Bacterial (Bac). Nuclear mitochondrial sequences/PCR artefacts (NUMTs) were detected in all cases. GenBank accession numbers of all novel Erectmocerus and Encarsia species were MN646915–MN646927 and MN646928–MN646950 respectively, GenBank accession numbers of all novel Bemisia and Trialeurodes whitefly species detected were MN660053–MN660056 and MN660057–MN660058, respectively. Raw HTS amplicon reads can be downloaded from CSIRO's public data access portal <https://data.csiro.au/collection/csiro:53348>. Site codes are as shown in Figure 1.

2.3. DNA extraction and HTS library preparation

DNA extraction of pooled individuals was carried out using the Qiagen Blood and Tissue DNA extraction kit (cat. No. 69056) following the manufacturer's instructions. Extraction of pooled individuals was life‐stage specific (i.e., extraction of pooled adults or pooled nymphs were carried out separately). Extracted gDNA from each pooled sample was quantified for DNA concentration and standardised to 0.5 ng/µl prior to PCR amplification of target mtCOI region. Traditionally, molecular diagnostics for B. tabaci cryptic species have relied on the F‐C1‐J‐2195 and the TL2‐N‐3014 primers (Crozier et al., 1989; Frohlich et al., ¹⁹⁹⁹; Roehrdanz, ¹⁹⁹³; Simon et al., ¹⁹⁹⁴) that amplified approximately 750 bp of the 3' COI terminal region. This primer set was developed based on Diptera, Lepidoptera, Coleoptera, and Hymenoptera but was not originally reported for Hemiptera (Roehrdanz, 1993; Simon et al., ¹⁹⁹⁴). Recent evaluations of these primers have found poor efficacies when applied in B. tabaci cryptic species (e.gElfekih, Tay, et al., ²⁰¹⁸; Mugerwa et al., ²⁰¹⁸) that contributed to the misidentification of pseudo species (Kunz, Tay, Elfekih, et al., 2019; Tay, Elfekih, Court, et al., ²⁰¹⁷). The B. tabaci cryptic species includes >34 well‐defined species with interspecific nucleotide distances that ranged from ca. 3% to >18% (Dinsdale et al., 2010; Kunz, Tay, Elfekih, et al., ²⁰¹⁹; Lee et al., ²⁰¹³). To ensure maximal PCR efficacies of our primer sets, we aligned complete COI genes as obtained via HTS platform, from a range of species that included SSA1, SSA2 (Kunz, Tay, Elfekih, et al., 2019); MED cryptic species (Rossitto De Marchi et al., 2018; Tay, Elfekih, Court, et al., ²⁰¹⁷; Vyskocilova et al., ²⁰¹⁹); Asian species (Tay et al., 2016); Bemisia “JpL” (Tay, Elfekih, Polaszek, et al., 2017), Australian (AUS), Indian Ocean, and Middle East Asian Minor 1 (MEAM1) species (Tay, Elfekih, Court, et al., 2017), to ascertain most conserved regions for designing and developing our two sets of complementary primers (wfly‐PCR‐F1/R1 and wfly‐PCR‐F2/R2; Supporting Information for Online Publication I). Primers were developed using Oligo 7 (Molecular Biology Insights) with minimal primer duplex and hairpin structure formation, with 55–65°C Tm, for expected amplicon size of 550–600 bp, and with minimal false primer annealing sites.

Two PCR reactions were carried out for each pooled sample using primers wfly‐PCR‐F1/R1 and wfly‐PCR‐F2/R2 for the first and second reaction, respectively. For the PCR amplification of each pooled sample, we used 3.5 ng gDNA template in a 35 µl PCR reaction volume and 30 PCR cycles (PCR conditions described in Supporting Information for Online Publication I). Amplicons from each of the pooled samples were ascertained for PCR amplification success on 1.25% 1× TAE agarose gel stained with GelRed and visualised on a UV‐transilluminator. Amplicons were then purified using AMPure XP beads as instructed in the Illumina 16S Metagenomic Sequencing Library Preparation (Part # 15044223 Rev. B) protocol and resuspended in 25 µl Qiagen EB (instead of the recommended 52.5 µl). Purified pooled amplicons were individually quantified for DNA concentration using a Qubit v2.0 fluorometer (Life Technologies) and Qubit HS dsDNA assay kit and indexed by amplifying 5.0 ng of the purified amplicon in 50 µl PCR volume over 11 PCR cycles (see Supporting Information for Online Publication I). Indexed amplicons were purified using AMPure XP beads as described above and concentration ascertained using Qubit v2.0. Equal proportions of purified indexed amplicon (i.e., 100 ng) from each pooled sample were mixed with 6× Loading Dye and run on 1.25% low melt agarose at 75 V for 2 h. The F1/R1 combined amplicons and the F2/R2 combined amplicons were excised and purified using the Zymoclean Gel DNA Recovery Kit (cat. no. D4001, D4002, D4007, D4008) following the manufacturer's protocol. Purified indexed amplicons were quantified for double stranded DNA concentration using Qubit and fragment size was estimated using TapeStation (Agilent Technologies). Amplicons were calculated for their nano‐molarity and diluted to 4 nM (see Illumina 470–2016–007‐B) for HTS run on MiSeq sequencer using the Illumina MiSeq Reagent Kit Version 3 (600 bp pair‐ended). A detailed protocol for the amplicon library preparation and related primer sequence information is provided as Supporting Information for Online Publication I.

2.4. HTS amplicon data processing

Fastq sequences representing individual populations post MiSeq runs were trimmed using Trimmomatic version 0.36 (Bolger et al., 2014) by quality (Leading:3; Trailing: 3; Slidingwindow:4:15, minlength: 125 bp) and imported into Geneious v11.1.5 as pair‐ended reads of the two metabarcoding primer sets for simultaneous mapping to reference sequences. Amplicon sequence reads from each population were mapped to curated barcode reference sequences of representative Bemisia species (see below) using the Geneious Mapper option, and selecting to trim amplicon reads with options set to “Annotate new trimmed region”, error probability limit set to 0.05, and trimmed off ≥23 bp at both 5' and 3' ends. Customed mapping options within Geneious Mapper function were: no gaps allowed, word length set to 16; randomly map multiple best matches, index word length set at 15, maximum ambiguity option set to 8, and with two separate runs of 3% and 5% maximum mismatches per read for all B. tabaci species; 7% for B. afer, and 12% for parasitoids (Figure 2). Unmapped reads after each round were used in mapping against other Bemisia and parasitoid reference sequences. A Geneious de novo assembly step with default setting was carried out separately for each primer‐pair, and only so when all reference whitefly and parasitoid sequences had been mapped and when unmapped reads were typically low (i.e., ≤4,000 pair‐ended reads). For the de novo assembly, the top 10 most common contigs were determined for their sequence identity using nucleotide blast (blastn) within NCBI database. A schematic representation of the described workflow is provided in Figure 2.

FIGURE 2.

Workflow for analysis of MiSeq generated high‐throughput sequencing of mtCOI amplicon reads for estimating proportions of Bemisia species in the African cassava cultivation landscape, including assessment of hymenopteran parasitoid species. Optional steps are indicated in dashed box depending on whether amplicon libraries were gel‐purified (refer to Supporting Information for Online Publication I) to remove small fragments. Quality of amplicons can also be improved if desired, by using an Illumina sequence trimming program such as Trimmomatic (Bolger et al., 2014). The de novo assembly step enables novel Trialeurodes‐, parasitoid‐ and bacterial‐related sequences, as well as NUMTs, to be identified. NUMTs were identified through detection of INDELs, premature stop codons, and presence of unexpected amino acid residues at highly conserved COI regions following the methods of Kunz, Tay, Elfekih, et al. (2019)

2.5. Reference sequences for identification of Bemisia mtCOI

Within the African cassava cultivation landscape context, the proportion of specific Bemisia and related cryptic species detected was estimated as represented by the proportion of total amplicon fragments. Our reference mtCOI sequences were from the following Bemisia species: SSA1 (GenBank JQ286457, nt 59–715), SSA2 (GenBank AY057141, nt 29–685), SSA3 (GenBank KM377923, nt 20–676), MEAM1 (GenBank KY951449, nt 782–1,438), IO (GenBank KY951448, nt 782–1,438), B. afer (GenBank AF418673_, nt 59–715), Uganda1 (GenBank KX570857, nt 34–690),”African” MEAM2 (GenBank KX570778, nt 34–690), MED (GenBank KY951447, nt 782–1,438), SSA6 (GenBank KX570852, nt 34–690), SSA9 (GenBank KX570856, nt 34–690), SSA10 (GenBank KX570804, nt 34–690), SSA13 (GenBank KX570833, nt 34–690) and novel sub‐Saharan Bemisia species (i.e., MN646951; MN646952) detected by Sanger sequencing from field‐surveyed populations (Macfadyen et al., 2021; this study). We confirmed the authenticity of these reference sequences as free from NUMTs following the methods of Kunz, Tay, Elfekih, et al. (2019).

2.6. Parasitoid reference sequences

Eretmocerus and Encarsia parasitoid mtCOI reference sequences from the sub‐Saharan region surrounding our field survey sites were built up (MN646919, MN646949) during the course of the analysis workflow (see Figure 2). Due to the anticipated significant whitefly hymenopteran parasitoid species diversity in the sub‐Saharan region, we arbitrarily determine a cutoff value of 6% within the Geneious “Map to Reference” function.

2.7. Assessment of species delimitation using a stagged over‐lapping amplicon approach

To assess efficacies of our two‐primer set approach for metabarcoding of cryptic Bemisia and related whitefly species, we downloaded the clean B. tabaci mtCOI database of Kunz, Tay, Elfekih, et al. (2019) and performed phylogenetic analyses based on 315 bp aligned nucleotides at 50 bp sliding window sizes for primer pairs “wfly‐PCR‐F1/R1” (i.e., 3' mtCOI barcode nucleotide position (nt) 1–315; nt 51–365; nt 101–415; nt 151–465), and “wfly‐PCR‐F2/R2” (i.e., nt 140–454; nt 190–504; nt 240–554; nt 290–604; nt 340–654), as well as for the full length of wfly‐PCR‐F1/R1 amplicon (473 bp) and the wfly‐PCR‐F2/R2 amplicon (518 bp), and compared that with the species delimitating powers based on uncorrected pair‐wise nucleotide similarity (i.e., p‐dist) distance method of the full 657 bp 3' mtCOI B. tabaci barcoding gene. Sequence alignment using MAFFT Alignment (selecting “Auto” option for the Algorithm setting, 200PAM/K = 2 scoring matrix, and gap open penalty and offset value set at 1.53 and 0.123, respectively) (Katoh & Standley, 2013) and trimming were carried out in Geneious v11.1.5. Confidence of clustering of identified species was estimated using Ultrafast bootstrap as implemented in IQ‐Tree, and followed the methods as described in Kunz, Tay, Elfekih, et al. (2019). Briefly, aligned and trimmed sequences were imported as fasta formatted file and imported to the IQ‐Tree website <http://iqtree.cibiv.univie.ac.at> (Trifinopoulos et al., 2016). The “Automatic” substitution model selection option was selected and we specified 1,000 ultrafast bootstrap approximation (Minh et al., 2013) with 1,000 maximum iterations and 0.99 as the minimum correlation coefficient. Visualisation and assessment of correct clustering of species were carried out using FigTree v1.4.4 (2006–2018 Andrew Rambaut, Institute of Evolutionary Biology, University of Edinburgh).

3. RESULTS

3.1. Primer efficacies

The primers wfly‐COI‐F1/R1 and/or wfly‐COI‐F2/R2 were shown to be highly efficient in successful PCR amplification of diverse representative Bemisia species that included Australian species (Lipaleyrodes (Bemisia)), Asian, New World (NW), Mediterranean (MED), Middle East Asian Minor (MEAM1), sub‐Saharan African (SSA) species, and “non‐tabaci” species (i.e., B. afer). Sanger sequencing of PCR amplicon from test Bemisia cryptic species (Table 1) partial mtCOI confirmed species identity when compared against both the updated standard B. tabaci mtCOI reference data set and the uncorrected pairwise nucleotide distances (p‐dist; see Kunz, Tay, Elfekih, et al., ²⁰¹⁹). While Sanger sequencing confirmed sequence integrity such as through conservation of amino acid residues against published B. tabaci cryptic species HTS‐derived full mtDNA genomes (e.gKunz, Tay, Court, et al., ²⁰¹⁹; Tay, Elfekih, et al., ²⁰¹⁶; Tay, Elfekih, Court, et al., ²⁰¹⁷; Tay, Elfekih, Polaszek, et al., ²⁰¹⁷; Vyskočilová et al., ²⁰¹⁸), results from HTS amplicon analyses readily detected NUMTs such as those similar to “MEAM2” (Delatte et al., 2005; Tay, Elfekih, Court, et al., ²⁰¹⁷), as well as partial mtCOI genes with INDELs. Detection of coamplified pseudogenes/NUMTs demonstrated the complexity of the Bemisia genome (Chen et al., 2016; Xie et al., ^{2017, 2018}). HTS‐amplicon analyses also shed light on the unexpected coamplification of nontarget whitefly genomic regions, as well as fungal, bacterial symbionts, and parasitoids genomic regions, and highlighted the complexity of trophic interactions in Bemisia whiteflies (e.gRao et al., ²⁰¹⁸; Shamimuzzaman et al., ²⁰¹⁹; Tay, Elfekih, Polaszek, et al., ²⁰¹⁷).

3.2. HTS‐amplicon protocol and analysis workflow

Test scenarios for detection and identification of Bemisia cryptic species in predetermined proportions (i.e., 20:1; 40:1; 100:1) showed that the HTS large scale sampling protocol and the analysis workflow (Figure 2) confidently identified the dominant as well as minor species in approximate expected proportions (Table 2). Similarly, in the hypothetical mixed‐species composition scenario that included different proportions of mixed species (i.e., scenarios “Mixed‐II”, “Mixed‐III”, “Mixed‐IV”; Table 2), the protocol also successfully detected all test species while the expected proportions were affected by low amount of coamplified NUMTs/pseudogenes, bacterial, fungal, parasitoid, and nonspecific whitefly genomic regions (Table 2). The test scenarios therefore provided significant insights into characteristics associated with our primers and the robustness of our multiprimer metabarcoding protocol. For example, across various scenarios, detected amplicon coverage regularly fell below the expected values due to nonspecific PCR coamplification of nuclear genome regions, endosymbionts, and environmental DNA contaminants.

3.3. Field application of HTS‐amplicon method

Applying the HTS multiprimer metabarcoding protocol to field‐collected samples from Malawi, Tanzania, and Uganda showed that in the African cassava fields, differences in Bemisia species compositions existed (Kalyebi et al., 2018; Macfadyen et al., ^{2018, 2021}). A total of eight whitefly species (i.e., SSA1, SSA2, IO, MED, B. afer, Uganda 1, two novel Bemisia species), and a novel Trialeurodes species, were detected in different field sites (Table 3, Figure 3). Multiple novel mtCOI haplotypes for Eretmocerus species (MN64915–MN64927) and Encarsia species (MN646929–MN646950) were also detected (Table 3). Parasitoid sequences were only ever detected in nymphal samples, and this was despite targeted exclusion of parasitised nymphal samples that could be visually identified prior to genomic DNA extraction. Detection of parasitoids is therefore probably the result of early stages of parasitism where visual identification would be most challenging. While our findings suggested high parasitism rates in the agricultural landscapes in Uganda, Tanzania, and Malawi, we nevertheless advise caution in interpreting the Bemisia parasitism rates, parasitoid species, and genetic diversity in these natural agricultural settings, as our primary experimental design was to develop a metabarcoding approach for surveys of the hemipteran Bemisia cryptic pest species only.

FIGURE 3.

Summary results of B. tabaci cryptic species from Uganda (a), Tanzania (b), and Malawi (c). Site codes are as provided in Table 3. Bemisia species are sub‐Saharan Africa 1 (SSA1), sub‐Saharan Africa 2 (SSA2), Mediterranean (MED), Uganda 1 (UG), Indian Ocean (IO), B. afer (Baf), and unknown (“Uknw”) species from Bemisia and Trialeurodes genera. Low detection rates of COI‐related pseudogene, parasitoid (Eretmocerus and Encarsia genera), and bacterial/fungal sequences are grouped in the “others” category. Whitefly species compositions are also compared between cassava and noncassava host plants (d and e, respectively), and between nymphal (M1, M3, M5, M7) and adult (M2, M4, M6, M8) life stages from four cassava sites in Malawi (f)

The sub‐Saharan African 1 (SSA1) Bemisia species was identified as the dominant species in cassava fields from Uganda and Tanzania (Figure 3a,b), while in Malawi both SSA1 and B. afer were present in approximately similar proportions (Figure 3c). Overall, the SSA1 species was most prevalent in cassava landscape (Figure 3d). We also detected the SSA1 species on noncassava crop planted within a cassava field (“T2” in Figure 3e; Table 3). The B. tabaci MED cryptic species complex was present as the dominant species in Uganda from three noncassava host crop sampling sites (Figure 3e, Table 3). Sweet potato crops also hosted other Bemisia species including Uganda 1 (“UG”) and Indian Ocean (“IO”), although both UG and IO species were not detected on sweet potato within cassava fields in Tanzania (“T2”), and could suggest potential behavioural differences between SSA1 and UG/MED species (Figure 3e) impacting on species‐specific abundances. In Malawi, B. afer and SSA1 species dominated different field sites (Table 3, Figure 3c,f). Their detection showed significant variability in species composition within the same field sites depending on the life‐stage sampled (i.e., nymphal vs. adults; Figure 3f). Field sample analyses also detected NUMTs and bacterial/fungal origins in the African populations, similar to our hypothetical scenarios (Table 2), thereby demonstrating consistencies between our HTS workflow and primer‐efficacy assessments.

3.4. Assessment of species delimitation efficacies between amplicon sizes

Short amplicons of between 130–319 bp (e.gBrandon‐Mong et al., ²⁰¹⁵; Leray et al., ²⁰¹³) but also longer amplicon lengths (e.g., 421 bp, Hajibabaei et al., ²⁰¹⁹) have been proposed for COI metabarcoding. Our assessment of species delimitation powers by short (i.e., 315 bp) versus full amplicon lengths based on distance methods (i.e., p‐dist) of our wfly‐PCR‐F1/R1 (473 bp) and wfly‐PCR‐F2/R2 (518 bp) primer sets, as well as their combined 657 bp barcode gene length for the B. tabaci cryptic species complex showed that metabarcoding involving the full 657 bp was needed to fully delimit between Bemisia species (Figure 4). Species detection and delimitation involving the various shorter 315 bp nucleotide lengths along the 657 bp barcoding gene region often failed to differentiate certain cryptic species complexes such as those within the Mediterranean (MED), New World (NW), sub‐Saharan Africa (SSA), and Australia (AUS) clades (Figure 4). Interestingly, other invasive species such as the SSA2 (e.g., Hajibabaei et al., ²⁰¹⁹) and the MEAM1 (De Barro et al., 2011; Elfekih, Etter, et al., ²⁰¹⁸), as well as the Asia II‐1 species that has recently emerged as important vector species in the transmission of the Sri Lankan Cassava Mosaic Virus (SLCMV) in South East Asia (Chi et al., 2020) could be readily identified by any of the 315 bp COI region.

FIGURE 4.

Summaries of efficacies of species delimitation powers based on the p‐dist method across the 657 bp 3' mtCOI barcoding gene region in the cryptic B. tabaci and non‐tabaci species complex. The metabarcoding primer pair “wfly‐PCR‐F1/R1” (472 bp) was divided into four overlapping regions of 315 bp across a 50 bp sliding window size, and the 518 bp amplicon generated by the “wfly‐PCR‐F2/R2” primer pair was divided into five overlapping regions of 315 bp. We also assessed full amplicon lengths from both primer pairs (i.e., F1/R1: 472 bp; F2/R2: 518 bp, blue coloured cells). Cryptic species that could not be defined were represented by grey coloured cells, and sequences that were successfully defined into their respective species clades were shown by green cells. Dark grey cells indicated failed species delimitation by both primer pairs and full species delimitation will require the complete 657 bp barcoding gene. The number of clean mtCOI sequences in the database (db) of Kunz, Tay, Elfekih, et al. (2019) for each species analysed are indicated (total: 224 sequences)

4. DISCUSSION

This study successfully demonstrated the applicability, transferability and powers of the multiprimer metabarcoding approach to investigate species compositions within an agroecological context, using the highly challenging Bemisia whitefly cryptic pest species as a study model system. These complementary sets of primers developed to amplify across the large genetic distances among the B. tabaci species complex also detected non‐Bemisia hemipteran whitefly‐nymphs, while showing potential to also simultaneously survey for associated parasitoid species diversity and parasitism rates, thereby further demonstrating the wide applicability and contribution to our understanding of trophic interactions between pests, beneficial insects and host plant usage across highly heterogeneous agricultural landscapes. Our multiprimer metabarcoding approach that involved high‐volume sampled biological material demonstrated host plant preference and species composition differences in African Bemisia species (Table 3, Figure 3), and that ascertaining a Bemisia species' taxonomic status will require the standard length (i.e., ≥657 bp) mtCOI barcoding 3' gene region (Supporting Information for Online Publication II; Figure 4), in addition to considering also developmental/life stages (Figure 3f), and the species' plant virus vectoring capabilities (e.g., Chi et al., ²⁰²⁰). We also outlined the development of an easy‐to‐use tailored bioinformatic analysis workflow, and provided the Bemisia research communities with two sets of highly efficient PCR primers that have the capability to amplify the alternative but widely used 3' mtCOI Bemisia barcoding region. Our primer sets when used in metabarcoding of Bemisia nymphs, also detect various Aphelinid hymenopteran parasitoid species within the Eretmocerus and Encarsia genera. Unexpected coamplification of various fungal and bacterial homologous gene regions, albeit at low frequencies, was also detected.

DNA metabarcoding approaches have been reported for integrative taxonomy (e.gCruaud et al., ²⁰¹⁷; Sigut et al., ²⁰¹⁷), their potential for invasive species surveys reviewed (Piper et al., 2019), for understanding relationships between species abundances and associated ecological factors (e.g., Macfadyen et al., ²⁰²¹), and impact from aquacultural activities demonstrated (He et al., 2021). Adoption of such metabarcoding approaches for landscape ecological investigations under field conditions between insect pest hosts and beneficial insects may also be possible (e.g., Macfadyen et al., ²⁰²¹), although appropriate experimental designs, primer improvement and modifications to analysis workflow including building of relevant barcoding database will be needed to increase species survey and detection sensitivity. For example, DNA barcoding primers for arthropods including the Hymenoptera (e.g., Folmer et al., ¹⁹⁹⁴; Hebert et al., ²⁰⁰⁴; see also Woodcock et al., ²⁰¹³) could be included as internal checks to compare frequencies of parasitism and species diversity, with concurrent analysis following the HTS pipeline as outlined (Figure 2). Studies that aimed to understand host‐parasitoid interactions involving Aphalinid parasitoids faces other challenge such as high species diversity in the Aphelinidae (e.gGebiola et al., ²⁰¹⁷; Hernández‐Suárez et al., ²⁰⁰³), while different reproductive modes (sexual vs. asexual vs. Rickettsiales‐induced sex ratio biases) in these haplodiploid parasitoids further complicate our understanding and interpretation of such biological interactions. Efficacies of these two complementary primer sets described in this study on hymenopteran parasitoids requires further assessment, as primer choice are known to affect species diversity recovery rates (e.g., Hajibabaei et al., ²⁰¹⁹) and is currently beyond our research aims.

Primer efficacies is a significant issue for Bemisia molecular diagnostics (e.gElfekih, Tay, et al., ²⁰¹⁸; Mugerwa et al., ²⁰¹⁸; Shatters et al., ²⁰⁰⁹), and have contributed to misidentification of pseudospecies (e.g., MEAM‐K, Roopa et al., ²⁰¹⁵; MEAM2, Delatte et al., ²⁰⁰⁵; Karut et al., ²⁰¹⁵; Ueda et al., ²⁰⁰⁹; SSA4, Berry et al., ²⁰⁰⁴) as demonstrated through genomic approaches (Elfekih et al., 2021; Kunz, Tay, Elfekih, et al., ²⁰¹⁹; Tay, Elfekih, Court, et al., ²⁰¹⁷; Vyskočilová et al., ²⁰¹⁸). Our two primer sets while successful in test scenarios in amplifying diverse Bemisia cryptic species (Table 2), by themselves, have nevertheless failed to resolve species status within various cryptic species complexes (Figure 4) due to insufficient DNA length. In a previous study by Wang et al. (2018), it was proposed that cost‐effective NGS barcoding solutions were available albeit with shorter (e.g., 313 bp) partial mtCOI sequences, however this would result in poor study outcome due to the biological and taxonomical complexity associated with the B. tabaci species complex (see Figure 4). When used together however, was able to differentiate all but the closely related MED‐ASL/MED‐Burkina Faso species complex, SSA11, and also the two closely related NW2/NW2‐2 species complex. Differentiating between certain cryptic Bemisia species may benefit from using alternative DNA markers and gene regions. For example, Vyskočilová et al. (2018) showed that species delimitation of the MED‐ASL from other MED cryptic species was best achieved using the standard (i.e., 5') barcoding gene region or via full mitogenomes. Furthermore, delimiting the true identity of the SSA1 species complex in our study sites will probably require genome‐wide single nucleotide polymorphic markers (Elfekih et al., 2019). For the putative NW2/NW2‐2 species complex identified by Kunz, Tay, Elfekih, et al. (2019), it remained to be seen if the 5' COI gene region or if genome‐wide single nucleotide polymorphic makers would prove to be the more effective diagnostic marker system.

Test scenarios involving evolutionarily divergent whitefly species (Tables 1 and 2) and field data on diversity of whitefly species detected (e.g., included novel Bemisia and Trialeurodes species; Table 3) suggested that overall nondetection rates of species could be potentially low. Our primers further demonstrated unexpected detection of novel Encarsia and Eretmocerus parasitoid species in the African cassava and noncassava fields, however the true genetic and species diversity of these minute parasitoids will require independent assessment for confirmation, such as to include the standard barcode COI gene region using standard primers (e.g., C_LepFoIF/C_LepFoIR; Woodcock et al., ²⁰¹³) as internal control. Nevertheless, understanding of trophic interactions between crop hosts, pest/beneficial insects and their potential biological control agents represents a molecular ecological network research area that could yield significant insights via metabarcoding approach (e.g Bansch et al., ²⁰²⁰; Evans et al., ²⁰¹⁶; Sow et al., ²⁰¹⁹).

Determining which insect represents the major species within diverse agricultural landscape settings requires adequate population‐wide sampling (Kalyebi et al., 2018, 2021; Macfadyen et al., ²⁰²¹). Conclusions based on few individuals sampled at a large number of sites are inherently problematic; there is a need, therefore, for the development of techniques to scale up data collection and processing of samples, and to develop associated bioinformatic workflows and analytical pipelines to provide robust findings at economic scales. For the cassava cultivation landscape in sub‐Saharan Africa, determining the dominant pest whitefly cryptic species represents a significant challenge due to difficulties with species identification without the help of molecular DNA markers (De Barro et al., 2011; Kunz, Tay, Elfekih, et al., ²⁰¹⁹). Furthermore, the often‐high population sizes associated with individual fields, and poor PCR primer efficacies (Elfekih, Tay, et al., 2018; Mugerwa et al., ²⁰¹⁸; Shatters et al., ²⁰⁰⁹) regularly resulted in suboptimal PCR amplification, and the difficulty encountered in sequence quality control (Kunz, Tay, Elfekih, et al., 2019; Tay, Elfekih, Court, et al., ²⁰¹⁷; Vyskočilová et al., ²⁰¹⁸) have been persistent problems in Bemisia molecular systematics and diagnostics.

In this study, we showed that by applying the DNA barcoding and the species “genetic gaps” concepts first demonstrated by Dinsdale et al. (2010) (see also Kunz, Tay, Elfekih, et al., ²⁰¹⁹), and with careful redesigning of DNA markers for the HTS metabarcoding platform, it was possible to develop a highly efficient method for large‐scale, effective and economic sampling and identification of the cryptic whitefly species to contribute to their management and control in challenging agricultural landscape settings (Kalyebi et al., 2018; Macfadyen et al., ^{2018, 2021}). Our species survey findings showed that SSA2 and B. afer were also present in the East Africa's cassava landscape in addition to the dominant SSA1 species, and contrasted the reported absence of SSA2 based on sampling of low numbers of individuals in the Ugandan cassava landscape (Ally et al., 2019). Co‐existence of both SSA1 and SSA2 may have important implications to genome evolution through introgression (Elfekih et al., 2021) and emergence of cassava disease epidemic (Legg, Sseruwagi, et al., 2014; Patil & Fauquet, ²⁰⁰⁹). Significant Bemisia whitefly species diversity such as the Asia species complex (e.g., see Kunz, Tay, Elfekih, et al., ²⁰¹⁹) as well as invasive MED and MEAM1 species existed across the Asian continent. Adopting our metabarcoding approach to delimit species status, composition, prevalence, and their association with asymptomatic/diseased cassava plants could provide knowledge necessary to assist with the development of management strategies for the recent SLCMV disease outbreaks in the South East Asian region (Chi et al., 2020; Minato et al., ²⁰¹⁹).

The importance of using standardised approach to sample sufficiently large quantities of B. tabaci cryptic species (Macfadyen et al., 2021; Sseruwagi et al., ²⁰⁰⁴) especially across diverse agricultural landscape and between different life stages (e.g., nymph vs. adult) is also demonstrated (e.g., Table 3, Figure 3f), where we detected significant differences in species compositions (e.g., between B. afer and SSA1). The approach developed here is therefore useful for processing nymphal data that may have strong links to host plants. Standardized sampling techniques for nymphs are less well‐developed than for adult B. tabaci on cassava (which are considered easier to count and collect). In reality, the focus of the research question should drive the sampling approach selected (be that nymphs, parasitized nymphs, or adults); however, in all cases scaling up the numbers of individuals processed is critical. Another important distinction of specific sampling of whiteflies from different host crops was also demonstrated in cassava versus noncassava host crops (Table 3, Figure 3d,e), where differences in species compositions, especially between B. afer and SSA1 cf. MED, Ug, and IO, respectively, were evident. Ecological and biological factors therefore underpinning Bemisia species compositions in Africa agricultural landscapes. Similarly, ecological factors and agricultural practices will probably impact on pest species compositions in other agroecological landscape systems and can be assessed by adopting the multiprimer metabarcoding method outlined here.

While our method enables efficient ascertainment of species composition at the landscape level, this method should not be used to determine individual maternal lineages of mtDNA haplotypes due to the high sensitivity of detecting random PCR‐introduced SNPs in individual amplicon molecules. At the landscape survey level, ascertaining specific maternal lineages does not represent a criterion needed to enable species identity to be determined, provided that sequence quality was adequately managed to avoid misidentification of species status due to mtDNA pseudogenes. Elfekih et al. (2021) have shown that NUMTs and pseudogenes could lead to significant misinterpretation of Bemisia population structure in the African cassava landscape regardless of whether partial mtCOI or genome‐wide SNPs were used (see e.g., Wosula et al., ²⁰¹⁷). A significant limitation on our approach is the quality of the library used to interrogate the sequences generated. Sanger sequencing‐generated partial mtCOI gene sequences in the public database (e.g., NCBI; see Vyskočilová et al. (2018) regarding MED‐related pseudogenes/NUMTs; see also Kunz, Tay, Elfekih, et al. (2019) for reanalysis of B. tabaci DNA database [Boykin et al., ²⁰¹⁷]) are gradually being recognised for containing high numbers of problematic sequences (see e.gNacer & do Amaral, ²⁰¹⁷; Song et al., ²⁰⁰⁸; Triant & Dewoody, ²⁰⁰⁷). Such readily available public DNA global databases will need to be critically reviewed and assessed, given that a large scientific community relies on these sequences to assist with species identification.

A further complicating factor in molecular diagnostics of Bemisia species is the significant challenge associated with their nomenclature, cryptic taxonomy (Boykin et al., 2018; De Barro et al., ²⁰¹¹), and the preferred partial COI gene region widely adopted by the B. tabaci research community (i.e., 3' COI region) against the standard barcode 5' COI region. For example, against the 34 B. tabaci cryptic species and five non‐tabaci species presented by Kunz, Tay, Elfekih, et al. (2019) based on sanitized data set, Barcode of Life data system v4 (BOLDSYSTEMS) did not identify 30 of these, while eight cryptic species were all identified as B. tabaci, and only one was correctly identified as B. afer (Supporting Information for Online Publication II). Furthermore, within the BOLDSYSTEMS database, the invasive MEAM1 was identified as B. tabaci, while the Mediterranean (MED) species, shown as the real B. tabaci species as originally described by Gennadius in 1889 (Tay et al., 2012), returned as “no species level match”. While BOLDSYSTEMS database and the NCBI database are linked, inherent problems associated with reporting cryptic species remained (e.g., unable to provide cryptic species identity in NCBI sequence submission process) and will continue to impact on research outcomes especially as metabarcoding becomes more widely and rapidly adopted. Reporting of full COI gene such as through assemblies of draft mitogenomes in the Bemisia whitefly species complex (e.gKunz, Tay, Court, et al., ²⁰¹⁹; Tay, Elfekih, Court, et al., ²⁰¹⁷; Tay, Elfekih, Polaszek, et al., ²⁰¹⁷; Thao & Baumann, ²⁰⁰⁴; Vyskočilová et al., ²⁰¹⁸; Wang et al., ²⁰¹⁶) can consolidate the different databases for B. tabaci cryptic species. Linking of other nonstandard gene regions (e.g., mitochondrial DNA 12S rRNA and ND2 gene regions) to the standard 5' COI gene region will also further resolve our understanding of species diversity from diverse ecosystems (e.gMiya et al., ²⁰²⁰; Stefanni et al., ²⁰¹⁸; Weitemier et al., ²⁰²¹).

Our metabarcoding platform can be adapted to understand other pest species complexes in agricultural systems and may be especially effective for invasive insect pests with small body size and where high density of population sizes can build up rapidly, such as in invasive whiteflies, thrips, mites, and aphids. We propose, therefore, to apply this approach to: (1) greatly increase number of individuals that can be sampled, processed and characterised simultaneously for their widely applied proposed 5' COI (Hebert et al., 2003) or alternative “DNA barcoding” genes and gene regions (e.g., the mtCOI 3' region), and the earlier (Woese et al., 1990) proposed RNA genes and related regions (i.e., ITS rRNA region for fungi, e.g., Xu, ²⁰¹⁶; Microsporidia, e.g., Ghosh & Weiss, ²⁰⁰⁹; Klee et al., ²⁰⁰⁶; O'Mahony et al., ²⁰⁰⁷; Tay et al., ²⁰⁰⁵; Velasquez et al., ¹⁹⁹⁶; 16S rRNA, e.g., Shelomi & Chen, ²⁰²⁰; 18S rRNA, e.g., Popovic et al., ²⁰¹⁸; 28S rRNA, e.g., Pawlowski et al., ²⁰¹²); (2) estimate individual species proportion across large agricultural landscapes, and (3) in nymphal/larval samples, identify parasitoid species that target the agricultural pests of interest. Contrasting the majority of metabarcoding studies published to‐date, our approach utilises significantly longer partial COI gene sequence through two sets of complementary primer sets (i.e., c. 650 bp cf. 130–421 bp; Brandon‐Mong et al., ²⁰¹⁵; Hajibabaei et al., ²⁰¹⁹; Miya et al., ²⁰²⁰; Stefanni et al., ²⁰¹⁸) which was important for defining cryptic Bemisia species status. The use of this multiprimer metabarcoding approach to generate longer sequence length and targeting the 5' COI barcoding gene region could enable greater species identification with the support of the BOLDSYSTEMS.

While the high‐throughput amplicon sequencing method in a species composition study in agroecological landscapes represents a relatively new field of research, the use of this method to identify genes of biosecurity importance has been demonstrated for understanding the spread of pyrethroid resistance genes in the red‐legged earth mites Halotydeus destructor (Edwards et al., 2018). The method described by Edwards et al. (2018) can be incorporated into our high‐throughput amplicon sequencing platform for rapid species composition identification and insecticide resistance allele frequency estimates. This would enable a genomic approach that would simultaneously allow the identification of species composition at landscape scales, the survey of insecticide resistance profiles (e.g., Hopkinson et al., ²⁰²⁰), and assessment of the implication of chemical control strategies on diversity and density of beneficial insects such as parasitoids. Such studies would be especially informative when temporal scales are included within the study design (e.gKalyebi et al., ²⁰²¹; Macfadyen et al., ²⁰²¹).

Developmental stages sampled impact Bemisia species compositions and the conclusions drawn from a survey, as demonstrated in the Malawi cassava sampling sites for adult and nymphal Bemisia populations. The extent of such variability in target species composition between immature and adult life stages for many other biological systems can also benefit from the HTS metabarcoding method described here. In this study, Bemisia nymphs that visibly showed signs of being parasitised by hymenopteran parasitoids were first excluded prior to genomic processing. Despite this, Encarsia and Eretmocerus parasitoid mtCOI sequences were detected, thereby highlighting the importance of biological control of Bemisia pest species by parasitoids in the sub‐Saharan African agricultural system. While we have not optimised this workflow to address host‐parasitoid interaction research questions, we have demonstrated the power of HTS genomic approaches to document and facilitate an understanding of these biological interactions. Improved understanding of the ecological impact of pest species systems from genomic perspectives will further transform our understanding of sustainable agricultural production. Appropriate experimental designs including careful development of sampling procedures will be needed for meaningful interpretation that takes advantage of the large volume of genomic data generated by the HTS platform. Our study demonstrates the successful use of multiprimer HTS metabarcoding methods in classical field‐ecological studies and for novel cryptic species discovery. The adoption of this approach more widely has the potential to help integrate global biodiversity with genomic (Arribas et al., 2021) and transform global biosecurity preparedness to impact understanding of ecological network, conservation biology, human health, and food security.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

OPEN RESEARCH BADGES

This article has earned an Open Data, for making publicly available the digitally‐shareable data necessary to reproduce the reported results. The data is available at <https://data.csiro.au/collection/csiro:53348>, and from GenBank accession numbers: MN646915‐MN646927; MN646928‐MN646950; MN660053‐MN660056 and MN660057‐MN660058.

Supporting information

App S1

App S2

Tay, W. T. , Court, L. N. , Macfadyen, S. , Jacomb, F. , Vyskočilová, S. , Colvin, J. , & De Barro, P. J. (2022). A high‐throughput amplicon sequencing approach for population‐wide species diversity and composition survey. Molecular Ecology Resources, 22, 1706–1724. 10.1111/1755-0998.13576

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from GenBank accession numbers MN646915‐MN646927; MN646928‐MN646950; MN660053‐MN660056, MN660057‐MN660058, and from published GenBank accession numbers cited throughout. Raw HTS sequence data can be downloaded from CSIRO's public data access portal <https://data.csiro.au/collection/csiro:53348>.