The Cassava Mealybug (Phenacoccus manihoti) in Asia: First Records, Potential Distribution, and an Identification Key

Phenacoccus manihoti Matile-Ferrero (Hemiptera: Pseudococcidae), one of the most serious pests of cassava worldwide, has recently reached Asia, raising significant concern over its potential spread throughout the region. To support management decisions, this article reports recent distribution records, and estimates the climatic suitability for its regional spread using a CLIMEX distribution model. The article also presents a taxonomic key that separates P. manihoti from all other mealybug species associated with the genus Manihot. Model predictions suggest P. manihoti imposes an important, yet differential, threat to cassava production in Asia. Predicted risk is most acute in the southern end of Karnataka in India, the eastern end of the Ninh Thuan province in Vietnam, and in most of West Timor in Indonesia. The model also suggests P. manihoti is likely to be limited by cold stress across Vietnam's northern regions and in the entire Guangxi province in China, and by high rainfall across the wet tropics in Indonesia and the Philippines. Predictions should be particularly important to guide management decisions for high risk areas where P. manihoti is absent (e.g., India), or where it has established but populations remain small and localized (e.g., South Vietnam). Results from this article should help decision-makers assess site-specific risk of invasion, and develop proportional prevention and surveillance programs for early detection and rapid response.


Introduction
The cassava mealybug, Phenacoccus manihoti Matile-Ferrero (Hemiptera: Pseudococcidae), is one of the most severe pests of cassava (Manihot esculenta) in the world [1]. It is native to South America [2], but it has become naturalized throughout sub-Saharan Africa since its inadvertent introduction into the continent in the early 1970s ( Fig. 1) [3]. P. manihoti was not known to occur in Asia until 2008, when it was first detected in Thailand. Since that year, it has spread aggressively throughout Thailand's cassava-growing region [4], also invading its neighboring countries and Indonesia [5] (Fig. 1), and raising significant concern over its potential arrival to more countries [6]. Responding to this concern, we present the first records of P. manihoti invading Asia and use them to estimate the climatic suitability for its establishment throughout the region. To further support detection and response efforts, we also provide a taxonomic key that differentiates all mealybug species recorded from the genus Manihot.
Several non-preferred host species can support P. manihoti reproduction, but only cassava is known to experience significant damage by this insect [7]. When it feeds on cassava, P. manihoti causes severe distortion of terminal shoots, yellowing and curling of leaves, reduced internodes, stunting, and weakening of stems used for crop propagation (Fig. 2). In the absence of its natural enemies and other control measures, this damage can reduce yields by more than 80% [8]. No cassava cultivars are known to be fully resistant to P. manihoti [9]. Explorations for P. manihoti natural enemies within its native range identified four hymenopterous parasitoids, twelve predators and one entomopathogenic fungus [2,10] out of which the parasitoid Anagyrus lopezi appeared to be one of the most promising [2]. The introduction of this parasitoid into Africa in the 1980s reduced high infestations by 90%, becoming a highly-successful case of classical biological control [11][12][13]. A similar outcome is expected from its recent introduction to Thailand, in November 2009 [4].
P. manihoti is parthenogenic, producing only female offspring. Hence, a single immature or adult may be sufficient to start an outbreak. Under optimal conditions, adults can deposit between 200-600 eggs [14,15] within ovisacs on the undersides of leaves and around apical and lateral buds. Ovisacs are sticky and can adhere to clothing, facilitating long-distance mealybug dispersal. Eggs hatch into mobile crawlers that can spread over the plant or be passively dispersed to neighboring plants by wind. Crawlers commence feeding from phloem fluids in young leaves and stems, and pass through three nymphal instars before reaching maturity.
Under laboratory conditions at 25uC, egg to adult development takes an average of 31-33 days [16,17]. Development is optimal around 27uC [15], and significant mortality occurs below 15uC [10] and above 33uC [16,18]. Rainfall is a key determinant of P. manihoti abundance and population dynamics: dry regions, years and seasons favor outbreaks [19,20]. Rainfall is thought to suppress P. manihoti mainly by causing mechanical mortality [21], but also by favoring insect pathogens and reducing cassava's suitability as a host [9,22].
Pest risk maps, based on models predicting climatic suitability for a species, are important decision-support tools for the management of invasive pests like P. manihoti [23]. Two modeling approaches are often used to develop them. The correlative or inductive approach estimates a species' climatic preferences based  [2]. Regional distribution in Africa was adapted from figure 1 in Herren and Neuenschwander [30]. Point locations in Africa correspond to Anagyrus lopezi releases at locations with high P. manihoti infestations, and were georeferenced from Neuenschwander [3]. Point locations in Asia correspond to reports listed on Table 1. doi:10.1371/journal.pone.0047675.g001 on analyses of geographic occurrence data [24]. By contrast, the mechanistic or deductive approach estimates its climatic preferences based on laboratory experiments [25]. Outputs from correlative models often align more closely with known distributions without demanding any biological data, but mechanistic models are thought to be superior in predicting distributions in novel environments [26]. Hence, integrative approaches that draw upon the complementary strengths of both can provide a very good approximation to the potential distribution of an invasive pest [27][28][29]. In this article, we use an integrative modeling approach to predict P. manihoti's potential distribution in Asia, in order to support decisionmaking in the management of this invasive pest.

Materials and Methods
Known distribution map P. manihoti's distribution records in South America and Africa were obtained from the published literature. Native distribution records were obtained from Löhr et al. [2]. Naturalized distribution records in Africa correspond to A. lopezi release sites [3], presumably at locations with high P. manihoti infestations, and to a regional distribution map adapted from Herren and Neuenschwander [30]. Geographic coordinates were approximated either by georeferencing published maps or by searching locations on Google Maps or on the MarkSim shape file for African towns (afrtowns.shp) [31]. Invasive distribution records in Asia correspond to specimens either collected by or submitted to the authors for identification. The first specimens were submitted for authoritative taxonomic identification to experts Dr. Gillian Watson (Department of Food and Agriculture, California) and Dr. Douglas J. Williams (Natural History Museum, London). Subsequent identifications were verified by TK and/or Maria del Pilar Hernandez (see reports in bold on Table 1). These latter specimens are deposited at the CIAT Entomology Collection, Palmira, Colombia. Invasive distribution records were georefer-

Potential distribution model
We modeled P. manihoti's potential distribution using CLIMEX Version 3 [32], a software widely used with positive results in the fields of biological control, pest risk assessment and climate change [33,34]. The CLIMEX Compare Locations module uses an integrative inductive-deductive approach to estimate climatic suitability for a species based on both (1) geographic occurrence data and on (2) the species' growth response under experimentally-manipulated conditions [32,35]. Climatic suitability is estimated by the ecoclimatic index (EI  [35]). Theoretically, EI is scaled between 0 and 100, and the larger the EI the more suitable the location for the species. In practice, EI values below 10 indicate marginal suitability, EI values above 20 indicate high suitability, and EI values above 50 are rare and usually confined to the tropics [33]. Formulas governing the CLIMEX Compare Locations model have been published by Sutherst and Maywald [35].

Climate data
CLIMEX models demand weekly, temporally-interpolated data from averages of five variables: maximum and minimum temperatures, 9 a.m. and 3 p.m. relative humidity, and rainfall (i.e., 260 data points per location). We used two metereological databases to provide this data. To streamline model development iterations, we first used the less computationally-demanding station database built into CLIMEX. This is a point location database with records from about 2,400 metereological stations worldwide. We then used CliMond 109 interpolated climate database for CLIMEX [36] to project model results globally.

Model fitting
Population growth parameters. We used eight parameters to define conditions suitable for P. manihoti population growth. Four parameters (DV0-DV3) captured the temperature optima and bounds for growth. An initial range of values for these parameters were obtained from reviewing published experimental studies on P. manihoti development [14][15][16][17][18]37]. Four additional parameters (SM0-SM3) captured the moisture optima and bounds for growth, in proportional units of soil water holding capacity. Values for these parameters were assigned under the assumption that P. manihoti growth is not directly limited by moisture, but it is optimal when its host is under drought stress [22]. We used a final parameter (PPD) to denote the degree days above the lower threshold for development (DV0) needed by P. manihoti to complete one generation. This parameter was used to estimate the potential number of generations P. manihoti can complete in one year at a given location. Parameter values for population growth were assigned so as to allow stress indices to explain a greater proportion of EI. For example, we set the upper threshold of soil moisture (SM3) at 2.5, fully aware that mortality by rainfall probably begins at much lower soil moisture levels.
Mortality or ''stress'' parameters. After parametrizing population growth, we used seven mortality parameters in a stepwise inductive process to confine the predicted distribution of P. manihoti, reconciling predictions with known distribution patterns in Figure 1. Three parameters captured mortality due to extreme cold, limiting sub-tropical distributions in South America and Africa without affecting distributions in northern Thailand. Parameter values were adjusted such that cold stress (CS) accumulates at a rate (DHCS) of 20.0015 week 21 when the total weekly number of degree days above a threshold (DVCS) of 16uC is below the cold stress threshold (DTCS) of 21uC days. These values are conservative, rendering a location unsuitable for P. manihoti only after eight consecutive weeks at an average weekly minimum of 15uC, a temperature leading to very high P. manihoti mortality in the laboratory [10]. Two parameters captured mortality due to extreme heat, mainly limiting distributions in the African Sahel. Parameter values were adjusted such that heat stress (HS) accumulates at a rate (THHS) of 0.001 week 21 when the average weekly maximum temperature is above the heat stress threshold (TTHS) 35uC. These values are also conservative relative to laboratory experiments, which suggest P. manihoti cannot survive prolonged periods at or above 33uC [16]. Finally, two parameters captured mortality due to rainfall; limiting distributions in the Congo Basin but not in the west, south and southwest of the Democratic Republic of the Congo (previously Zaire); thereby approximating mealybug distribution maps for that country [38]. Parameter values were adjusted such that wet stress (WS) accumulates at a rate (HWS) of 0.00125 week 21 when soil moisture is above the threshold (SMWS) of 80% water holding capacity.
Model validation. We validated our model qualitatively, by evaluating the ability of its weekly output indices for population growth (GI) and rainfall mortality (WS) to match P. manihoti seasonal population dynamics observed at specific locations in Paraguay [2] and in the Democratic Republic of the Congo [39]. For this evaluation, we selected the locations within the CLIMEX station database that were closest to the study sites.

Description and identification key
According to the scale insect database ScaleNet [40] there are currently 26 mealybug species (Hemiptera: Pseudococcidae) recorded on the genus Manihot, of which 23 have been recorded on cassava, Manihot esculenta (Euphorbiaceae). An additional species, Phenacoccus solenopsis Tinsley, not listed in ScaleNet, has been reported on cassava [41], increasing the number of species recorded on cassava to 24 and 27 on the genus Manihot.
In order to facilitate the identification of mealybugs that may be found on cassava, TK prepared a taxonomic key that differentiates all mealybug species hitherto recorded from the genus Manihot worldwide. Morphological features of mealybugs needed to prepare the key were taken from descriptions by Williams [42] and Williams & Granara de Willink [43], and the key was constructed mainly by adapting the keys to mealybugs by Williams & Granara de Willink [43]. The key should be used by a trained person or by a specialist since basic knowledge on the morphology of Pseudococcidae is needed in order to interpret the different morphological features used in the key. There is always a possibility that a species not included in the key may be found feeding on cassava, thus the following key should be used with caution.

Results
Known distribution map P. manihoti's native distribution in South America, naturalized distribution in Africa and invasive distribution in Asia is presented in Figure 1. Distribution points in Asia correspond to reports listed in Table 1. The first authoritatively-verified specimens of P. manihoti from Asia were collected between October and November of 2008. The distribution map was not intended to be comprehensive, but rather to capture sufficient environmental heterogeneity to guide model parametrization.

Potential distribution model
Parameter values for the P. manihoti distribution model are presented in Table 2. Spatio-temporal predictions for locations within CLIMEX's station database are shown in Figure 3. Spatial predictions adequately match the known distribution map for P. manihoti in South America, Africa and Asia. Weekly growth (GI) and wet stress (WS) indices for Asuncion (Paraguay) and Noqui (Angola) match P. manihoti seasonal population dynamics at nearby locations [2,39]. Specifically, the model adequately predicts population peaks from August-November around Asuncion [2], and from June to October around Noqui [39]. Weekly indices at Barumbu (Zaire) explain the unsuitability of the Congo Basin for P. manihoti.
Predictions for Asia based on CliMond 109 interpolated climate database are shown in Figure 4. All distribution records in Asia fall within predicted suitable regions, mostly within regions predicted to be at high risk of outbreaks (EI.20). The highest predicted suitability within cassava-growing regions in Asia is found within the southern end of Karnataka in India, the eastern end of the Ninh Thuan province in Vietnam, and in most of West Timor in Indonesia. Cold stress (CS) explains predicted unsuitability across Vietnam's northern regions and in the entire Guangxi province in China (Fig. 4B). Also, wet stress (WS), or rainfall mortality, explains predicted unsuitability across much of the wet tropics in Indonesia and the Philippines (Fig. 4D). The model also suggests that in Asia P. manihoti is not limited by heat stress (Fig. 4C) and can potentially complete up to 17 generations in one year (Fig. 4E).  [46].

Phenacoccus manihoti Matile-Ferrero
Materials examined. Specimens used to verify the key are reported in bold in Table 1.
Diagnosis. In life, species pinkish, covered in a white mealy secretion, with tufts of flocculent waxy secretion at posterior end and around margins (Fig. 2A). The species always reproduces parthenogenetically. The species is most similar in life to Phenacoccus herreni Cox & Williams which is yellowish and reproduces bi-parentally.
Slide preparations remarkably similar to those of P. herreni Cox & Williams. Phenacoccus manihoti possess 18 pairs of cerarii (Fig. 5), each with two enlarged lanceolate setae; the dorsal setae are minute and lanceolate; without aggregations of trilocular pores around the setal collars. Quinquelocular pores are numerous on the venter as in P. herreni, but there are always 32-68 on the head in the area immediately anterior to the clypeolabral shield, whereas P. herreni has 0-20 in this area. Normally the dorsal multilocular pores around the margins are more numerous in P. manihoti than in P. herreni. They are sometimes present on the thorax in P. manihoti but have not been observed in P. herreni. Groups of tubular ducts are present around the dorsal margins as in P. herreni, although in most specimens they are more numerous in P. manihoti. Other important features are the 9-segmented antennae, denticles on the claws and a circulus that is distinctly 'ox-yoked' shaped (diagnosis adapted from reference [43]).

Discussion
The potential spread of P. manihoti into more Asian countries remains a prime concern for cassava production in the region [6].
In an effort to support decisions to manage this invasive pest, this article reports P. manihoti's known invasive distribution, predicts its potential distribution in Asia, and presents a taxonomic key that distinguishes it from all other mealybug species associated with the genus Manihot.
To our knowledge, our article is the first to report P. manihoti's occurrence in Cambodia and Vietnam, suggesting the pest is rapidly spreading in the region. We know of only one additional study predicting P. manihoti's potential distribution, but based on correlative models [47]. This article complements the previous effort by parametrizing a mechanistic model (CLIMEX), using an integrative inductive-deductive model fitting approach. Prediction patterns from both models are very similar, with a tendency of the CLIMEX model to be more conservative (e.g., predicting no suitability where the correlative model predicts low suitability). One important advantage of the CLIMEX model is that it allowed us to formulate specific hypotheses on the climatic factors potentially limiting P. manihoti's spread in Asia. The model is also temporally explicit, and could therefore be instrumental in the design and planning of early detection programs.
Results suggest P. manihoti is (1) broadly adapted to the Southeast Asian climates, but is likely to be limited by (2) cold in northern latitudes (.20uN) and (3) high rainfall around the Equator. In ecological terms, our pest risk map represents a hypothesis of what environments in Asia fall within P. manihoti's fundamental niche. The fundamental niche is a concept representing the full range of environmental conditions where a species can survive and reproduce in the absence of negative interactions with other species [48]. Accordingly, the risk map does not take into account the effects of natural enemies and human intervention, among other limiting factors that should further restrict P. manihoti's distribution. In that respect, our study, combined with previous mechanistic modeling work for P. manihoti [20], could be used as the basis of a more comprehensive model that also accounts for the potential suppressive role of A. lopezi [49].
Our model predictions should be particularly important to guide management decisions for high risk areas where P. manihoti is absent (e.g., India), or where it has established but populations remain small and localized (e.g., South Vietnam). For those locations, management options include prevention, eradication and containment (Fig. 6) [50][51][52]. The development of plant quarantine measures to prevent introductions at likely entry pathways is the first and most cost-effective option where a pest is absent [50][51][52][53]. It can be achieved by intercepting, treating or prohibiting the entry of contaminated or potentially-contaminated material (e.g., cassava planting stakes [54]) [52]. When prevention fails, eradication is the preferred course of action [50][51][52][53]55]. Insect eradication can be achieved with insecticide or biopestide treatments designed to eliminate the pest from a delimited area [52,56]. Finally, containment involves managing the spread of invasion either by reducing dispersal, reducing population growth or a combination of both [51,56]. Viable containment tactics include domestic quarantines, insecticide treatments and classical biological control at the expanding population front [56]. Successful eradication and containment rest on the ability to detect low-density populations, demanding the development of species-specific surveying methods that are practical and cost effective [52,53,56]. In that respect, our model could be used as a tool to design a risk-based surveying program, specific in space and in time, that improves the probability of detecting nascent P. manihoti populations.
The window of opportunity for P. manihoti early detection and rapid response may be small once the invasion reaches its spread phase. In Africa, the cassava mealybug spread at a rate of 150 km/ year [4], contrasting the less than 30 km/year for other invasive Hemiptera [56]. Similarly in Thailand, P. manihoti spread widely and began causing yield losses as high as 50%, estimated at roughly US$ 30 million, within two years of first detection [4,47]. This aggressive spread is poorly explained by the insect's dispersal biology. Instead, anthropogenic mechanisms such as the movement of contaminated planting stakes, where mealybugs can survive feeding on buds [54], are more likely drivers. Based on this hypothesis, we believe promoting the soaking of cassava cuttings on an aqueous solution of thiamethoxam (0.2 g/L), imidacloprid (0.8 g/L) or dinotefuran (8 g/L) may be an effective tactic to slow mealybug spread. Ultimately, however, successful management of P. manihoti spread will require a better understanding of the mechanisms contributing to its long-distance dispersal.
In summary, the arrival of P. manihoti in Asia imposes an important, yet differential, threat to cassava production in the region. The identification key presented in this article should help qualified experts accurately distinguish it from similar species associated with cassava. Our mechanistic model accurately matched P. manihoti's known distribution and a previous correlative distribution model, suggesting it is good working hypothesis on the mealybug's potential distribution in Asia. This new model, in addition to the recent sightings reported in this article, should help decision-makers assess site-specific risk of invasion, and develop proportional prevention and surveillance programs for early detection and rapid response.