Introduction

Development of a classical weed biological control program is a multifaceted process that includes estimating inherent risks associated with the introduction of exotic species. Risk assessments involve, but are not limited to, quantifying a prospective biological control agent’s host range to estimate the potential for direct feeding and damage to vulnerable non-target species. This process is an exercise in maximizing predictive precision within the bounds of practicality (Forno and Heard 1997). Intuitively, increasing the number of test plant species presented to a herbivore is expected to increase the precision of host range predictions. In practice, however, most truly relevant information regarding a herbivore’s host range is derived from test species that are phylogenetically close relatives of the target weed (Wapshere 1974; Pemberton 2000; van Klinken and Heard 2000; Barratt et al. 2010). As described by Briese and Walker (2008), plant phylogenies can be used to refine test plant lists that may include many unnecessary species to a shorter and more informative list that accurately and efficiently characterizes a herbivore’s host range (Briese 2005; Sheppard et al. 2005).

A second factor used to refine test plant lists is the geographic area of the biological control program. All classical weed biological control programs have a strong spatial component, a geographic footprint that delineates the region of study and implementation. This geographic footprint plays a critical role in development of risk assessments and interpretation of inferences drawn from pre-release studies. The geographic region of interest is often defined by the distribution of the weed species within certain spatial constraints. These constraints may include natural geographic barriers like oceans, mountain ranges, deserts or other geological features that inhibit landscape level dispersal of released biological control agents (Briese and Walker 2002). Alternatively, physiological constraints (lower or upper lethal temperatures, diapause, etc.) that limit survival of a candidate biological control agent at certain latitudes, elevations, or climates further delineate a program’s geographic footprint (Boughton et al. 2009). It may be argued, therefore, that suitable host plants beyond these geographical and physiological barriers are not at risk from biological control introductions due to a lack of spatial overlap or a perceived inability of the agent to gain access to these areas even though they may be suitable for establishment (Briese and Walker 2002).

An underlying assumption of classical biological control implies that intentionally introduced natural enemies will remain within the identified boundaries that delineate the program’s footprint. Examples from the literature, however, demonstrate that this assumption is not uniformly valid. The South American cactus moth’s circuitous range expansion to North America, where it now threatens native cacti, illustrates this point. Prickly pear cacti in the genus Opuntia are native to the new world but have become invasive weeds elsewhere (Goeden and Andres 1999). Cactoblastis cactorum (Berg) is a specialist of cacti in the genus Opuntia and was imported from Argentina to Australia in 1925 for control of exotic Opuntia spp., ultimately resulting in complete suppression of nearly 24 million hectares of infested land (Dodd 1940; Goeden and Andres 1999). This success led to transfers of C. cactorum to other Opuntia infested regions, including the Caribbean island of Nevis in 1957 (Pemberton 1995; Stiling and Simberloff 2000). The moth eventually spread, either naturally or inadvertently through the ornamental trade, from Nevis to southern Florida, USA, where it developed new associations with native Opuntia species including the endangered O. corallicola (Small) Werdermann (Stiling 2002). Thus, risk to American Opuntia species from the introduction of C. cactorum into Australia was minimal due to multiple significant geographic barriers that limited movement between intended and “at risk” regions. The moth’s introduction to Nevis, in contrast, markedly increased risk to endemic American Opuntia populations due to regional proximity (or overlap) coupled with the herbivore’s inherent or human-mediated dispersal capabilities.

The role of plant phylogenies in host range testing has benefited from repeated scientific scrutiny (Pemberton 2000; van Klinken and Edwards 2002; Louda et al. 2003; Sheppard et al. 2005; Barratt et al. 2010). In contrast, less attention has been given to the influence of biogeographical barriers in host range predictions and the resulting inferences for widely distributed pest species (Louda and Stiling 2004; Petit et al. 2009). The assumption that weed biological control agents will remain within regional constraints has largely gone untested for most biological control programs. Herein, this assumption is evaluated for the introduced natural enemies of the internationally distributed tree Melaleuca quinquenervia (Cav.) S.T. Blake.

The myrtaceous tree M. quinquenervia occurs naturally along Australia’s eastern coast from Sydney in New South Wales to the northern tip of Queensland, in New Guinea, and in New Caledonia (Boland et al. 1987). Australian habitats that support M. quinquenervia populations typically include low-lying coastal wetlands behind heath-dominated headlands, riparian zones and brackish estuaries behind mangrove swamps (Rayamajhi et al. 2002).

Melaleuca quinquenervia has been widely disseminated over the course of the last century for ornamental, revegetation, and agroforestry purposes (Turner et al. 1998; Serbesoff-King 2003; Dray et al. 2006). This exotic tree was introduced into various locations in the United States and Caribbean but was planted and propagated extensively in southern Florida (Dray et al. 2006). After its introduction, M. quinquenervia spread at an estimated rate of 2850 ha year−1 (Center et al. 2000) and has proven to be a superior competitor to most, if not all, native vegetation occurring in forested and sawgrass dominated wetlands of the Florida Everglades (Turner et al. 1998). These M. quinquenervia wetland forests typically form dense stands characterized by continuous upper canopies with depauperate understories (Rayamajhi et al. 2009).

A classical weed biological control program targeting M. quinquenervia in Florida was initiated in the mid 1980s, with the expectation that introduced herbivores would limit invasion and complement conventional control tactics (Balciunas et al. 1994). The biological control program’s area of implementation was identified as the geographic range of M. quinquenervia in Florida, which encompassed much of the state’s peninsula. The adventive range of the exotic tree also includes various nearby Caribbean islands (Table 1) so the flora of these neighboring countries was also considered during development of test plant lists. Based on the flora of these regions, it was determined that biological control agents would require genus level specificity to be suitable for introduction into Florida. The curculionid weevil Oxyops vitiosa Pascoe (Coleoptera: Curculionidae) was the first candidate selected for quarantine-based host range assessments (Purcell and Balciunas 1994) and, once deemed sufficiently host specific, was released in south Florida in 1997 (Center et al. 2000; Pratt et al. 2003). Feeding by the weevil markedly reduces the tree’s reproductive potential and growth (Pratt et al. 2005; Tipping et al. 2008), but O. vitiosa pupates in the soil so persistent populations are rare in permanently flooded habitats where some M. quinquenervia stands persist. To enhance landscape-level suppression of M. quinquenervia, a second biological control agent, the psyllid Boreioglycaspis melaleucae Moore (Hemiptera: Psyllidae), was released in south Florida during the spring of 2002 (Center et al. 2006). By completing its life cycle entirely on the plant, B. melaleucae is less vulnerable to hydrological conditions and exploits a wider range of leaf ages than the weevil (Wineriter et al. 2003). Feeding by psyllids induces leaf senescence, eventually resulting in mortality of coppicing stumps and seedlings (Morath et al. 2006; Franks et al. 2006). Host specificity testing revealed that O. vitiosa and B. melaleucae are specialists of a species complex within the genus Melaleuca, which is restricted to Australasia. Following establishment, common garden experiments confirmed that feeding and development by O. vitiosa and B. melaleucae was restricted to Melaleuca species, as predicted in quarantine-based host range testing, and posed no direct threat to native or economically important species in the New World (Center et al. 2007; Pratt et al. 2009).

Table 1 Adventive populations of the Australian tree Melaleuca quinquenervia and the years monitored for colonization by the biological control agents Oxyops vitiosa and Boreioglycaspis melaleucae
Full size table

An areawide release effort from 2001 through 2008 resulted in 3.3 million M. quinquenervia biological control agents (combined total) redistributed to 407 locations and among 15 Florida counties (Balentine et al. 2009). Post release evaluations indicate that the geographic distribution of O. vitiosa encompasses 71% of the M. quinquenervia infestation in Florida. The distribution of B. melaleucae is slightly greater than its predecessor, despite being released five years later, with a range including 78% of the tree’s range in Florida. Although widely distributed, highest population densities of both herbivores occur in southern portions of the state (Pratt et al. 2003; Balentine et al. 2009).

Materials and methods

Surveys for spreading herbivores

An international survey was implemented to monitor the long range dispersal of O. vitiosa and B. melaleucae beyond the herbivores’ intended geographic range of Florida. First, a combination of herbaria searches, literature reviews, and various unpublished reports were used to develop a database of adventive M. quinquenervia populations (Table 1). Herbaria responding with geographic data included: ARCH, BISH, BRIT, EAP, F, FLAS, FSU, FTG, G, GH, HNMN, JBSD, LSU, MO, MOL, MU, NY, PIHG, SWF, USF, US (see Thiers 2011 for interpretation of herbaria codes). Surveys of adventive populations consisted of two phases: (1) confirming the existence of M. quinquenervia at reported locations and (2) searching the trees for the presence of the herbivores or signs of their feeding (see Balentine et al. 2009). Global positioning system (GPS) data were gathered for each confirmed M. quinquenervia population as potential areas of colonization. Their distances from the Florida coast line were measured using ArcMap (ver. 9.3, ESRI, Redlands, CA, USA). For each confirmed location, M. quinquenervia trees were randomly selected and herbivore presence or absence was determined by searching canopy vegetation during a 30-min evaluation period. Surveys were conducted once per year, during the dry season when herbivore densities and signs of damage are greatest (Balentine et al. 2009). The frequency of annual surveys varied among sites based on local cooperator availability (Table 1). It should be noted that a few sites were surveyed only once or twice during the monitoring process (i.e. Cuba, Costa Rica, Jamaica, etc.) and therefore caution should be taken when interpreting these results. Infrequently monitored sites, however, were surveyed later in the sampling effort (2009–2011), affording more time for colonization, population buildup, and thus increased likelihood for detection. Biological control agents discovered during surveys were collected, identified by the authors based on morphological features, and voucher specimens were deposited with the California State Collection of Arthropods or Florida State Collection of Arthropods (see Pratt et al. 2006; Pratt et al. 2008; Pratt and Arakelian 2011).

Dispersal pathways radiating from south Florida to surrounding regions include natural as well as human-mediated mechanisms. Confirming pathway use requires intercepting the dispersing herbivore in transit and was beyond the scope of this survey (Work et al. 2005). However, patterns of dispersal and detection from among the possible recipient locations can provide insight to the likelihood of pathway use (Petit et al. 2009). Long range dispersal from Florida to unintended areas may include active flight or other more passive forms of dispersion including “rafting” on debris, propulsion from wind currents, or “hitchhiking” on vessels of transportation (Browne and Peck 1996; Drake and Farron 1998; Kiritani and Yamamura 2003). Long range dispersal is strongly influenced by distance between the propagule source and recipient areas suitable for colonization, as well as propagule pressure (density) and life stage (Simberloff 2009). Based on principles of island biogeography, we hypothesize that the probability of colonization is negatively correlated with distance and infer that detection of herbivores at “near” versus “far” recipient M. quinquenervia stands is evidence of natural dispersal. For the purposes of this study, “near” M. quinquenervia populations were those sites <1,000 km from the Florida coastline (the Bahamas, Cuba, Jamaica) while “far” populations included those beyond 1,000 km.

Airline transportation systems also serve as important invasion pathways for the long range dispersal of insects and are indicators of trade connectivity between two regions (Kiritani and Yamamura 2003; Work et al. 2005). Successful colonization of new habitats is often attributed to the frequency of invasion attempts or opportunities (Crawley 1989; Wilson et al. 2009). By assuming that flight frequency was a metric for colonization opportunities, we hypothesize that the probability of M. quinquenervia herbivore colonization is positively correlated with the number of flights arriving in foreign locations from southern Florida. To investigate the frequency of airline connections, transportation statistics for direct (non-stop) flights between international airports in southern Florida and destinations that harbored adventive populations of M. quinquenervia were tabulated. This was done by searching the air carrier database of the USA Department of Transportation (http://www.bts.gov/). These statistics include monthly data reported by certified USA and foreign air carriers on the combined number of passengers, freight, and mail transported flights departing the USA. Data were sorted by the flight destination, origin, year, and number of departures performed. Origin was limited to the four international airports that are sympatric with the M. quinquenervia biological control agents’ ranges in southern Florida (Fort Lauderdale Hollywood International Airport (FLL), Miami International Airport (MIA), Palm Beach International Airport (PBI), Southwest Florida International Airport (RSW)). All Florida airports are <1 km from M. quinquenervia stands that support high densities of the introduced natural enemies. Destination was represented by the nearest international airport to the monitored M. quinquenervia population. The mean distance between monitored M. quinquenervia populations and the nearest international airports was 55.2 (SE: 17.5, range: 1–232) km. The resulting data represented the number of all flights (passenger, cargo, and mail combined) originating from southern Florida and arriving in an area of interest per year. The mean number of flights per year was calculated by averaging data from years 2005 through 2007. The non-parametric Kruskal–Wallis test was used to investigate the influence of distance and flight frequency on the probability of herbivore recruitment (PROC NPAR1WAY, SAS ver. 9.1, SAS 1999).

Results

Thirteen M. quinquenervia populations were confirmed and monitored from 2005 to 2011 (Table 1). Surveys conducted in January 2006 resulted in the discovery of B. melaleucae on leaves of M. quinquenervia trees growing near San Juan, Puerto Rico. An island-wide assessment by Pratt et al. (2006) documented that psyllid densities and associated damage were greatest near the San Juan (Luis Muñoz Marín International) Airport but decreased with increasing distance from the greater San Juan area. In 2007, O. vitiosa was observed on the island of New Providence in the Bahamas but neither herbivore was detected on nearby Grand Bahama or Andros islands (Pratt et al. 2008). Upon first discovery, O. vitiosa was observed within 4 km of the Nassau International Airport. In 2009, B. melaleucae was also observed attacking M. quinquenervia trees within a neighborhood community approximately 5 km west of the Los Angeles, California (USA) International Airport (Arakelian 2009). To date, O. vitiosa and B. melaleucae have not been detected in Cuba, Jamaica, Texas (USA), Costa Rica, Brazil, Hawaii (USA) or South Africa.

Winged dispersal represents one pathway by which O. vitiosa and B. melaleucae may have spread beyond their intended geographic range. Distances from Florida’s coastline to possible recipient M. quinquenervia populations are listed in Table 1. At least six populations of the exotic tree are less than 1,000 km from peninsular Florida, with approximately 200 km of open water separating Grand Bahama from Florida. The recruitment of Florida’s biological control agents, however, was not influenced by linear distance between Florida and the recipient M. quinquenervia stand (χ1 2 = 0.01, P = 0.93), suggesting that flight by biological control agents across the open ocean was not likely.

Accidental anthropogenic transportation is another common mechanism of long range dispersal of insects worldwide. The mean number of flights departing from the four international south Florida airports was nearly 350,000 annually when averaged across 2005–2007. Of the total number of flights, 15% landed at airports within 200 km from known adventive M. quinquenervia stands or trees. The most common destination of direct flights that linked south Florida with other exotic populations of the tree included the Bahamas, with nearly 20,000 flights annually (Table 1). Other common destinations included Puerto Rico, Jamaica, Colombia, and Los Angeles (USA). Various adventive populations of the tree are not or only weakly linked by direct air transit with south Florida, including the Hawaiian islands, South Africa and various Caribbean islands of the Lesser Antilles (Table 1). Evidence suggests that the destination of dispersing M. quinquenervia biological control agents was influenced by the mean number of annual flights linking south Florida with other adventive tree populations (χ1 2 = 4.88, P = 0.03).

Discussion

The M. quinquenervia biological control agents that were intentionally introduced into Florida have dispersed far beyond their intended adventive range. While unexpected, the movement of O. vitiosa and B. melaleucae to Caribbean islands highlights the biotic connectivity between the Antilles and peninsular Florida. Intuitively, one may assume that factors related to island biogeography (island size and proximity) may be important drivers influencing biotic exchange and serve as a framework to explain the movement of the dispersing herbivores. One may hypothesize that among the Caribbean islands, for instance, dispersing biological control agents are more likely to spread to the island of Cuba due to its close proximity to the mainland (330 km) and much larger surface area in relation to the other islands within the Greater and Lesser Antilles. A similar hypothesis may be generated for the Bahamian Archipelago with the largest (Andros) or the nearest (Grand Bahama) islands more likely to recruit dispersing herbivores as compared to the smaller and relatively more distant island of New Providence. In contrast to these predictions, B. melaleucae bypassed nearer islands harboring M. quinqueneriva and was detected over 1665 km south of Florida on the island of Puerto Rico in 2005. Similarly, O. vitiosa was observed on the island of New Providence in 2006, which is the smallest, more distant, and least infested of the Bahamian islands surveyed. Most recently (2009) B. melaleucae was recovered from M. quinqueneriva trees in Los Angeles California (USA) and, assuming the psyllid originated from Florida populations, this dispersal event spanned the North American Continent with no observed satellite colonies distributed between donor and recipient locations. Alternatively, the B. melaleucae population in Los Angeles may have originated in the native range of Australia rather than Florida. Franks et al. (2011) examined mtDNA sequence data for Australian and Floridian populations of B. melaleuca and determined that the adventive range possessed two (A and B) of the eight haplotypes sampled in eastern Australia (A–H). Therefore, evaluation of the mtDNA for the Los Angeles psyllid population can provide insight to the herbivore’s origin if the haplotypes differ from the two types that occur in both Australia and Florida (A and B). Regardless of origin, these data underscore the unpredictability of long range dispersal events based on spatial and geographical constraints. Emigration of the M. quinquenervia biological control agents was not restricted by large water bodies, mountain ranges, or extended expanses of host-free lands.

Although spatial proximity or island size may not account for the observed pattern of the herbivore’s recruitment, frequency of trade and tourism may. Human activities play an important role in accidental insect invasions, with the most common introduction pathways including international transportation of airplane luggage and cargo (Kiritani and Yamamura 2003). Dobbs and Brodel (2004) reported, for instance, that over 10% of airplanes on the Miami tarmac contained insects in their cockpits, cabins or cargo areas. Nearly 32,000 airline flights depart Florida for the Bahamas annually and approximately 20,000 of these arrived on the three islands studied herein. Within the Bahamas, the mean number of annual flights to New Providence was markedly greater (13,982) than those destined for Grand Bahama (5,452) or Andros (559). A similar pattern of long range dispersal also exists among the other adventive M. quinquenervia stands monitored herein (Table 1), with herbivore recruitment associated with locations that experience a high level of connectivity with southern Florida, as indicated by the large numbers of direct flights originating from the region. Considering the frequent transport of tourists and cargo between southern Florida and the newly colonized locations, the premise that the biological control agents were inadvertently carried or “hitchhiked” to their new ranges is a plausible or even likely explanation.

These long range dispersal data may also provide insight to the likelihood that the remaining natural enemy-free M. quinquenervia populations will recruit these herbivores without intentional human assistance. The exotic tree populations in Colombia and to a lesser degree Jamaica, Dominican Republic, and Costa Rica experience relatively high levels of transportation connectivity with southern Florida and are therefore predicted to have a greater probability of colonization by the biological control agents than other populations (Table 1). The island of Grand Bahama, however, may have the highest probability of colonization due to the second highest number of flights originating from southern Florida but also due to the new colonization of O. vitiosa within the archipelago, which further increases its connectivity with adventive weevil populations in the Caribbean. Therefore, these new adventive populations of the M. quinquenervia herbivores serve as foci that may facilitate expansion to areas that were previously less likely to be invaded due to lack of transport connectivity and intra-island connectivity within the Caribbean may play an important role of the continued spread of these biological control agents. The islands of Hawaii, for instance, were previously considered less vulnerable to unintentional spread by the M. quinquenervia biological control agents due to limited pathways of invasion yet adventive satellite populations in Los Angeles (USA) markedly increase linkages between the mainland and herbivore-free tree populations in Hawaii.

The presence of these newly arrived herbivores is interpreted differently among the recipient land managers. In the Bahamas and Puerto Rico, for instance, this unintended spread is considered by some to be a fortuitous benefit to their ongoing and underfunded effort to control the spread of M. quinquenervia (Pratt et al. 2006; Pratt et al. 2008). In California, where the tree is not considered invasive, land managers perceive the psyllid’s arrival as an added complication to the aesthetic maintenance of the communities’ street trees (Pratt and Arakelian 2011).

Unintended spread of introduced biological control agents is not limited to the M. quinquenervia system. Geographic range expansions of biological control agents can be categorized into at least two general groups: those that spread beyond political boundaries and those that overrun geographical barriers that were assumed to curtail their spread. Recently, the houndstongue root weevil Mogulones cruciger Herbst, which was released in Canada in 1997, has dispersed south across the US border where it may feed on native Boraginaceae (Andreas et al. 2008). Similarly, the seed head fly Urophora quadrifasciata (Meigen) was introduced into Canada in the early 1970s but spread across the USA border and can now be found in much of North America, including the states of Arizona and Arkansas (Story 1985; Duguma et al. 2009). In contrast, the eriophyid mite Eriophyes chondrillae (Canestrini) and the rust fungus Puccinia chondrillina Bubak and Sydow were released for control of Chondrilla juncea L. in the western USA but spread north to British Colombia, Canada (Julien and Griffiths 1998). The tephritid fly Procecidochares utilis Stone was introduced into India in 1963 for control of Ageratina adenophora (Sprengel) but has spread to neighboring Nepal and China (Wang 1989).

There are also numerous examples of biological control agents that have dispersed beyond geographic barriers that historically were considered impermeable. In addition to the South American cactus moth cited earlier, the scale parasitoid Aphytis lepidosaphes Compere was intentionally introduced into California in 1948–1949 but has since been recovered at various locations where it was not intentionally released including: Florida, Hawaii, Puerto Rico, El Salvador, Argentina, Turkey, Israel and Australia (DeBach 1974). Various biological control agents of Lantana camara L., including the agromyzid flies Calycomyza lantanae (Frick) and Ophiomyia lantanae (Froggatt), have spread from areas of intentional release (i.e. Australia) to Malaysia and Micronesia (Julien and Griffiths 1998; Muniappan and Reddy 2003). The gracillariid Dialectica scalariella (Zeller) was introduced to Australia for the control of Echium candicans L. but has since dispersed to New Zealand (Julien and Griffiths 1998).The South American bruchid beetle Acanthoscelides macrophthalmus (Schaeffer) was introduced to South Africa for the control of Leucaena leucocephala (Lam.) but is now found in Australia as well as Cyprus (Vassiliou and Papadoulis 2008).

Over time, regulatory and advisory organizations that oversee weed biological control agent introductions have demonstrated greater awareness for the potential of herbivore dispersal beyond its intended range. The Technical Advisory Group (TAG) for biological control agents of weeds in the USA, and its predecessors, conducted informal and reciprocal reviews of proposed introductions with Canadian officials beginning in 1962 (APHIS-PPQ 1998). The reviewing body began requesting formal comments from both Canadian and Mexican officials in 1971 based on the knowledge “that an introduced organism recognizes no political boundaries and its introduction need(s) to be considered on a continental basis” (APHIS-PPQ 1998). Considering the data presented herein, we propose that this process can be more inclusive through formal consultation and comment from Caribbean or other nearby countries to address the potential of unintended spread. In addition, petitions for the introduction of weed biological control agents can be improved by delineating the attainable geographic range of the introduced herbivore based on the target weed’s and alternative host plant’s distributions rather than environmental barriers that restrict dispersal. A recent example of this process involves a Longitarsus sp. that was proposed for biological control of Heliotropium amplexicaule Vahl but was not permitted for release in Australia. Its rejection was attributable to the questionable permeability of geographic barriers expected to limit the herbivore’s dispersal and its use of allopatric native non-target species even though they did not overlap with the weed’s current geographic range (Briese and Walker 2008).

The assumption that introduced biological control agents will remain within the identified boundaries that delineate the program’s geographic footprint is not supported by the M. quinquenervia system as well as other biological control projects. Considering the ever-increasing levels of globalization, are there are any geographical barriers that can meaningfully restrict the spread of introduced biological control agents (Vermeij 2005)? We propose that the dramatic increase of international trade and tourism has resulted in the development of complex pathways that render historic barriers irrelevant to curtailing the spread of biological control agents. The reality of long-range dispersal and unintended spread in biological control underscores the need to conduct risk assessments that focus less on “at risk” species within strict geographical barriers and more on broadly defining the agent’s host range (Briese and Walker 2008). This will be accomplished as greater attention is placed on exploring patterns of a herbivore’s host plant use in comparison to the degree of phylogenetic relatedness to the target weed over larger geographic ranges. Inferences drawn from host use patterns in relation to plant phylogenies provide greater insights to the risk of direct non-target damage across geographic and political barriers.