Invasive lionfish dispersal between shallow- and deep-water habitats within coastal Floridian waters

Invasive lionfish threaten native fishes and ecosystem health in the Atlantic, Gulf of Mexico, and Caribbean. Controlling their spread and population growth can be difficult given their early maturity, high fecundity, a long larval dispersal period, and preference for structure. Mitigation efforts are further complicated by the existence of large and numerous deep water refugia. Despite the potential importance of these refugia, their connection with shallow water populations and their role in recruiting pelagic larvae is poorly understood. We examined the post-settlement dispersal patterns of invasive lionfish using otolith δ13C and δ18O stable isotope analysis and Bayesian stock mixture analysis. We find that there is settlement within both deep- and shallow-water habitats. It is estimated that 34.5% of the shallow-water population is composed of individuals that moved from deep-water habitat and into these shallow regions post-settlement. Only 4.1% of the deep-water population is composed of individuals that settled in shallow-water habitat before dispersing. These results demonstrate a link between deep and shallow habitats within the coastal waters of Florida, with the flow of individuals from deep to shallow water being the predominate dispersal direction of post-settlement individuals. We suggest that post-settlement dispersal can contribute to the invasion of new habitat and potentially hinder shallow-water removal efforts in currently colonized areas.


Introduction
Invasive lionfish (Pterois volitans and P. miles) are widely distributed across a diverse set of habitats throughout the Caribbean and western Atlantic (Schofield 2010;Luiz et al. 2021) where temperatures are above ~ 10 °C (Kimball et al. 2004;Dabruzzi et al. 2017): including mangroves (Barbour et al. 2010), coral reefs (Schofield 2010; Barbour et al. 2010;Biggs and Olden 2011), seagrass beds (Barbour et al. 2010;Biggs and Olden 2011), and deep coastal waters (Andradi-Brown et al. 2017;Luiz et al. 2021). Invasive lionfish cause large reductions in the biomass of native fishes by exploiting more than half of gape-appropriate prey species in their invaded range (Green et al. 2012). So, understanding their distribution and dispersal is critical in managing native communities. In their invaded range, lionfish first colonized shallow habitats (Schofield 2010;Biggs and Olden 2011) and tend to exhibit high site fidelity (Jud and Layman 2012;Akins et al. 2014;Tamburello and Côté 2015). Shallow regions are commonly used by juvenile reef fish as nurseries (Nagelkerken et al. 2000;Mumby et al. 2004;Dorenbosch et al. 2005) as they provide abundant food resource and protection from predators. As the invasion progresses, lionfish populations appear in deeper coastal regions in dense populations with larger individuals (Claydon et al. 2012;Nuttall 2014;Switzer et al. 2015;Andradi-Brown et al. 2017;Gress et al. 2017).
Persisting source populations may decrease the efficacy of removal efforts (Johnston and Purkis 2015). Thus, spillover from deep-water refugia into shallow-waters may influence near-shore removal efforts. In the western Atlantic and Caribbean, lionfish populations are 35 m deep on average but exist as deep as 150 m (Johnston and Purkis 2011). On the Western Floridian Shelf (WFS), models predict that the invasion began further offshore, as larvae shed off the Florida Loop Current (FLC) into deep waters (Switzer et al. 2015;Johnston and Purkis 2015). Unfortunately, deep-water removal efforts to manage populations are logistically more difficult than shallow-water spear fishing efforts. However, developments in lionfish-specific trapping technologies are allowing for deep-water lionfish removal with minimal harm to native species (Harris et al. 2020).
Lionfish are highly fecund (Fogg et al. 2017) and produce larvae that travel long distances (Wilson Freshwater et al. 2009;Ahrenholz and Morris 2010). While larval dispersal is a significant component of overall lionfish dispersal, post-settlement dispersal is common in coastal marine fish (Cocheret de la Morinière et al. 2002;Franco et al. 2012). In addition, invasive species like rainbow trout (Oncorhynchus mykiss) colonize new territories along rivers through post-settlement dispersal (Thibault et al. 2010). Unlike salmonids, lionfish are not typically a highly mobile species. But they may use post-settlement dispersal to colonize new territories or resupply existing populations. Early mark-recapture studies on lionfish show high site fidelity on time scales from days to months (Jud and Layman 2012;Akins et al. 2014;Tamburello and Côté 2015). Recently, fine-scale acoustic tracking methods demonstrated that lionfish individuals have larger home-ranges (McCallister et al. 2018;Dahl and Patterson 2020;Green et al. 2021) than expected and are capable of moving up to ~ 800 m a day and 2 km over the course of 89 day (Dahl and Patterson 2020). However, both broad patterns of post-settlement dispersal, including any post-settlement connections between deep and shallow waters, are not well understood.
Stable isotope analysis on otoliths is used as a proxy for the environmental conditions a fish experiences across time (Thorrold et al. 1997;Weidman and Millner 2000). Changes in microchemistry along an otolith's growth axis reflect shifts in a fish's chemical environment and can indicate dispersal events (Campana and Thorrold 2001;Reis-Santos et al. 2015;Kitchens et al. 2018;Bouchoucha et al. 2018;Pracheil et al. 2019). Environments along the inshore-offshore gradient should be more different from each other than those that fall within chemical isoclines across similar depths (Kalish 1991;Michener and Lajtha 2008;Paillon et al. 2014). To investigate the post-settlement dispersal patterns of invasive lionfish, we use otolith stable isotope compositions from lionfish captured throughout Florida's coastal waters: namely the Florida Keys and a subset of the Western Floridian Shelf. Specifically, this study tests: (1) if individuals tend to settle in deeper or shallower habitat; and (2) if post-settlement dispersal connects deep and shallow water habitats. Given the earlier work indicating high site fidelity, we expect to see most individuals inhabiting a similar environment throughout time. As densities of juveniles are often higher in shallow habitat and there are indications of potential ontogenetic shifts towards deeper habitat, we expect that a small fraction of otoliths from individuals captured in deep water habitats will reflect the early use of shallow, warmer habitats and subsequent offshore dispersals. Invasive lionfish dispersal between shallow-and deep-water habitats within coastal Floridian…

Methods
We sampled sagittal otoliths from 32 lionfish from the Florida Keys (N = 9) and the West Florida Shelf (WFS) (N = 23) (Fig. 1). Otoliths from the Florida Keys were sampled from carcasses as byproducts of recreational fishers and dive shops. As the majority of these carcasses were cleaned and filleted by the anglers, data on fish total length and weight was not always available, but depth and location of capture was available for all individuals. Otoliths from the WFS were obtained as a subset from a collection of samples previously acquired for the comparison of lionfish age and growth across the Gulf of Mexico (Fogg et al. 2019). These individuals were collected opportunistically by recreational and commercial spear fishers and trawlers, and the depth and location of capture was recorded for each sample (Fogg et al. 2017).
Fish were sampled between 0 m and 93 m and were binned into two "stocks" for the classification of shallow and deep-water "stocks" in our Bayesian stock mixture analysis. Individuals collected above 35 m of depth were categorized as part of the shallow-water stock. Deep-water stocks were collected between 35 and 93 m. Shallow-water stocks are closer to shore, subject to warmer year-round temperatures, and are within safe recreational diving limits. Individuals from deep-water stock (N = 15) were sampled exclusively along the WFS and individuals from shallow-water stocks (N = 17) were sampled from both within the WFS (N = 8) and the Florida Keys (N = 9). The shallow and deep stocks were assessed using a Wilcoxon Signed-Ranks Test and found to be significantly distinct from eachother for both δ 13 C (W = 49, p < 0.001) and δ 18 O (W = 260, p < 0.001). See below for additional information on stock delineation and isotopic discrimination across groups.
A small subset of the WFS otoliths were aged as part of a previous study (Fogg et al. 2019). The median age of all samples included in this study was 1.25 years and the maximum age was 3.5 years. The median age of lionfish obtained from the shallow-water sites was 1.875 years and these fish had a median total length of 296 mm. The median age of lionfish obtained from the deep-water sites was 1.25 years and these fish had a median total length of 260 mm.
The otolith material is a proxy for the environmental chemistry each fish was exposed to during its growth. We sampled two points along the otolith growth axis, the center of otolith growth (core) and the otolith edge (rim), to assess chemical differences between two periods of time for each individual. Samples were drilled with a micro-mill within 1 mm of the core and then once along the rim within 1 mm of the otolith edge. The center of otolith growth represents early otolith growth, and the rim represents the most recent growth. Otoliths were cleaned using a Sonic Vibra-Cell. Each otolith was sonicated for 6-7 s at 60% amplitude to remove contaminants. Whole otoliths were dried and mounted to glass slides with superglue for the micro-mill process. The drill and working area were cleaned between samples to avoid contamination. Between 50 and 80 µg of otolith material were analyzed for δ 13 C and δ 18 O in a Thermo-Fisher Delta V + mass spectrometer with dual-inlet and Kiel IV carbonate reaction device at the Lamont-Doherty Earth Observatory. Weighed sample powders were dissolved in ~ 100% H3PO4 at ~ 70 °C. Measurements were calibrated through repeated measurements to the National Bureau of Standards (NBS)-19 (TS-Limestone) standard (Schloesser et al. 2009(Schloesser et al. , 2010Brand et al. 2014) and reported relative to the Vienna Peedee Belemnite (VPDB) standard. To assess internal precision, NBS-19 standards were run every 10th sample where internal precision for δ 13 C was ± 0.03 ‰ and ± 0.06 ‰ for δ 18 O. To assess external precision and sample homogeneity, three replicate core and three replicate rim samples were run for an individual from the deep-water stock of the WFS. External precision for rim δ 13 C and core δ 13 C was ± 0.61 ‰ and ± 0.45 ‰ respectively and was ± 0.03 ‰ and ± 0.28 ‰ for rim δ 18 O and core δ 18 O respectively.
Over the last two decades, otolith microchemical profiling has proven to be a powerful tool for assessing the composition of unknown stocks (Millar 1990;Schloesser et al. 2010;Smith and Campana 2010;Correia et al. 2014;Brophy et al. 2016;Tanner et al. 2016). Stock mixture analysis is a means of assessing the extent to which known populations, or stocks, contribute to a mixed stock where individuals originate from unknown origins (Pella and Masuda 2001;Smith and Campana 2010;Brophy et al. 2016). By assuming that the individuals are residents in the locations of capture (Pella and Masuda 2001;Smith and Campana 2010;Brophy et al. 2016), otolith rim compositions function as known stocks. Differences between core (i.e., the unknown stock) and rim otolith samples reflect differing habitat use across time and a mixing model is used to estimate what proportion of the unknown stock originated from known stocks. Using the growth axis of the otolith, this method is useful for understanding the adolescent habitat use of fishes and their dispersal across landscapes (Pella and Masuda 2001;Rooker et al. 2008;Smith and Campana 2010;Brophy et al. 2016).
To remove individuals from the dataset that likely inhabited an unsampled habitat in the past, we plotted isotopic biplots of the data with 99% confidence ellipses around the rim compositions of the shallow and deep stocks. Any individuals (N = 6) with core isotopic compositions that fell outside of these 99% confidence ellipses for either of the known stocks were removed. This process only removed individuals from the shallow-water stocks and left 11 shallow-water individuals and 15 deep-water for analysis. We used a Bayesian mixing model as described by Smith and Campana (2010) and in their R package 'mix.Fish'. This model was run with WinBUGS software (Bayesian inference Using Gibbs Sampling, Lunn et al. 2000) through the R package R2WinBUGS (v2.1.21, Sturtz et al. 2005). Analyses were conducted in R (v3.5.1; R Core Team 2019). The Bayesian stock mixture model was run for 50 K iterations across 5 chains and the first 25 K iterations were discarded as burn-in. 90% credible intervals were calculated using posterior distributions for our unknown, mixed stocks.
Additionally, there must be differentiation between the known base stocks. The isotopic composition of δ 18 O increased with increasing depth (R 2 = 0.42, F(1,33) = 23.95, p < 0.001). δ 13 C decreased with increasing depth (R 2 = 0.172, F(1,33) = 6.857, p = 0.013) (Fig. 2). For these regressions, the assumption of homoscedasticity is not met and does not inform the p-values. Isotopic differentiation between regions was confirmed through the comparison of otolith rim isotopic compositions using Wilcoxon Signed-Ranks Test as mentioned above. There was a significant (α < 0.05) difference between WFS deepwater and WFS shallow-water otolith δ 13 C (W = 15, p < 0.001) and δ 18 O (W = 113, p < 0.05). Similarly, there was a significant difference between WFS deepwater and Florida Keys shallow-water otolith δ 13 C (W = 34, p < 0.05) and δ 18 O (W = 147, p < 0.001). The WFS shallow-water otolith δ 18 O was not significantly different the Florida Keys shallow-water otolith δ 18 O (W = 46, p = 0.37). However, the WFS shallow-water δ 13 C and the Florida Keys shallow-water otolith δ 13 C were slightly different between groups (W = 57, p = 0.046). Additionally, we assessed niche overlap between all groups using SIBER (v2.6.1, Jackson et al. 2011) in R. We report the median overlap between known stocks (iterations = 20,000, burn-in = 1000) as the percent of overlapping area between two ellipses. 63.40%, 90% credible interval [8.57, 100], of the shallow Keys' standard ellipse overlapped with the shallow WFS standard ellipse (Fig. 2) (Fig. 2). Given the similarity in δ 18 O between both shallow-water sampling regions and the high overlap in niche area, the shallow WFS individuals and the shallow Keys individuals were grouped together as the "shallow" stock for the stock-mixture analysis. For an illustration of the offset between core and rim values for individual lionfish, please see Figure S1.

Discussion
Overall, our stable isotope analysis of lionfish otoliths demonstrates the early use of deep-water habitat and indicates that post-settlement dispersal away from deep-water habitat is a life-history strategy utilized by lionfish to move across their invaded range. Previous studies have suggested that most new invasions of lionfish begin in shallow waters (Schofield 2010;Biggs and Olden 2011) and that densities of adults are higher in deep-water refugia when compared to nearby shallower habitats (Biggs and Olden 2011;Claydon et al. 2012;Nuttall 2014;Switzer et al. 2015;Andradi-Brown et al. 2017;Gress et al. 2017). Here we assess if (1) individuals tend to settle into deeper or shallower habitat and (2) post-settlement dispersal connects deep and shallow water habitats. Our study highlights the potential importance of both shallow and deep habitats and the post-settlement dispersal across those depths.
Juvenile lionfish often use nursery habitats like seagrass beds (Biggs and Olden 2011;Fig. 2 Isotopic differentiation across depths and stocks. A Isotopic biplot depicts standard ellipses and centroids of the three sampling regions. There is high overlap between the shallow-Keys and shallow-WFS stocks, demonstrating the systematic differences between the otolith rim isotopic compositions between deep water (WFS only) and and shallow water (shallow WFS and Keys) sourced individuals. B Isotopic biplot depicts standard ellipses and centroids of the two stocks used in the stock-mixture analysis (shallow WFS and Keys combined). C d 18 O of otolith rim increases with increasing depth depth (R 2 = .42, F(1,33) = 23.95, p < .001). D) d13C of otolith rim decreases with increasing depth (R 2 = 0.172, F(1,33) = 6.857, p = 0.013). For both regressions, the assumption of homoscedasticity is not accomplished and does not inform the p-values. Vertical line at 35 m represent the boundary of the depth steps Andradi-Brown et al. 2017). However, in areas along the WFS, juveniles and adults utilize the same habitats in the absence of established nursery grounds in close proximity (Stevens et al. 2006;Dahl et al. 2018). We demonstrate that lionfish use both shallow and deep habitats throughout their lives in the WFS and Florida Keys. While deeper waters have been cited as important refugia for adult lionfish (Biggs and Olden 2011;Claydon et al. 2012;Nuttall 2014;Switzer et al. 2015), these findings indicate that deep water habitats may also be important for lionfish recruitment and dispersal. Settlers from deep-water habitat largely contributed to the sampled deep-water population. These deep-water settlers also contributed to the sampled shallow-water populations, indicating post-settlement dispersal. Shallow-water settlers composed a majority of shallow-water populations and were predicted to make very limited contributions to deep-water populations. Given that these shallow-to deep-water dispersals made up such a small overall percent of the modelled individuals, it is likely that post-settlement dispersal away from shallow regions may not be a dominant life history strategy for lionfish in this region.
Our results demonstrate that while many lionfish are lifelong residents of a certain depth a non-trivial portion disperses away from habitat used in their early life stages. Lionfish tend to exhibit high site fidelity (Jud and Layman 2012;Akins et al. 2014;Tamburello and Côté 2015), but our results suggest that around a fifth of all modelled individuals demonstrate differences in habitat use through time. The remaining proportion of the modelled population may be exhibiting site fidelity, may be making dispersals that are too small to be detected, or may be dispersing across isotopically similar regions. Given that very few individuals leave shallow-water habitat to disperse towards deeper-water habitats, the shallowwater settlers appear to exhibit more habitat fidelity and recruit retention.
Our isotopic analysis connects deep and shallow lionfish populations. Fish may disperse as individuals outgrow predation pressures (Laegdsgaard and Johnson 2001), develop new dietary preferences (Laegdsgaard and Johnson 2001), or as a product of density dependent spillover (Travis et al. 1999;Abesamis and Russ 2005). In areas of high density, invasive lionfish grow at slower rates (Benkwitt 2013), have higher incidence of cannibalism (Dahl et al. 2018), and demonstrate decreased site fidelity (Tamburello and Côté 2015;Dahl et al. 2016). Although, other reports demonstrate no evidence of density-dependent trends in Fig. 3 The connections between the stock sources and the mixed stocks represent the mean estimated proportion of individuals from each stock that disperse across the studied region and compose mixed stocks.  (Benkwitt 2013). High population density within our sampled deepwater region may explain the high contribution of this depth zone to those adjacent to it.
The Florida Loop Current (FLC) is a fast-moving current that transports water through the Gulf of Mexico and around Florida. It has been implicated, in theoretical work, as a potential source of lionfish larvae to the Gulf of Mexico and WFS, allowing for early colonization of the area through deep water colonies that move shoreward (Switzer et al. 2015;Johnston and Purkis 2015). High density populations in the offshore WFS are predicted sources while lower density sinks are predicted to persist inshore (Johnston et al. 2017). Along the continental shelf, currents are slower and more variable in direction, allowing lionfish to radiate throughout the region (Switzer et al. 2015;Johnston and Purkis 2015). Slower moving currents would allow a flow from offshore to inshore environments, even without impressive swimming capabilities. The FLC runs close to our sampled sites, and given the previous theoretical work supporting this mechanism, we conjecture this is the source of many larvae in the region. Lionfish individuals may then use these slower currents, outside of the FLC, to disperse away from their early habitats.
Our results suggest that lionfish are settling across a range of depths. Shallow recruitment, implicated to be a major driver of the lionfish invasion throughout other Caribbean systems, plays a surprisingly low role in contributing individuals throughout our sampled WFS region. Focusing removal or conservation effort on invasive populations is a well-known method of controlling the growth of a target species (Lepak et al. 2006;Weidel et al. 2007;Conner and Morris 2015). As removal of lionfish from deeper water is more difficult than shallow water removal, the existence of important deep-water lionfish settlement and the transfer of deep-water settlers into shallow-water populations may hinder lionfish population control.
While our study provides evidence of long-term habitat specificity and settler retention, our localized data suggest that post-settlement dispersal may be an overlooked component of lionfish invasion biology. Our study area reflects only a subset of usable habitat within Floridian coastal waters and only presents data from a small number of lionfish individuals, so we are unable to speak to broader spatial and temporal patterns within this system. Further work should be directed towards understanding if larval settlement is passive or deliberate in search of a certain habitat: potentially a preference for a certain terrain or rugosity. Increased sampling of areas throughout the entire study region could aid in identifying other settlement locations. While our results are specific to smaller regions within a broader landscape, this study may inform broader trends concerning the role of deep-water refugia throughout the invaded western Atlantic and the role of currents as enablers of postsettlement dispersal.