The contribution of egg-bearing female American lobster (Homarus americanus) populations to lobster larvae collected in Long Island Sound by comparison of microsatellite allele frequencies.

ABSTRACT In 1999, the massive die-off of the American lobster (Homarus americanus) populations in western Long Island Sound (LIS) caused complete collapse of the commercial lobster fishery. Reestablishment of a commercial fishery in western LIS requires, in part, knowledge of lobster populations that provide larvae to western LIS. To address this issue, larvae collected from sites within LIS were assigned to egg-bearing female lobsters by comparison of microsatellite allele frequencies. Egg-bearing female lobsters were collected from three sites within LIS and from a site in the Hudson Canyon (Crivello et al. 2005). Tissue samples were collected from female lobsters in spring, summer, and fall of 2001, and lobster larvae were collected in late summer of 2001 and 2002 from five sites within LIS. Differences in microsatellite allele frequencies were used to assign larvae to female lobster populations by a Bayesian and neural network approach. Larval assignments indicated that a high percentage (35% to 45%) of the larvae collected in the eastern or central LIS originated in female lobsters collected in the Hudson Canyon. In contrast, very few larvae collected in western LIS originated from females in western LIS (<30%) and the overall contribution of western female lobsters to the collected larvae was very low. The relative contribution of all sampled female lobsters to collected larvae was greatest from the Hudson Canyon area (35%), followed by central LIS (25%), eastern LIS (20%), western LIS (13%) and the Stratford Shoals area (7%). These results are discussed in context of reestablishment of a lobster commercial fishery in western LIS.

KEY WORDS: lobster, Homarus americanus, parentage, larvae

INTRODUCTION

The American lobster (Homarus americanus H. Milne Edwards, 1837) represents one of the most important commercial fisheries in the Gulf of Maine and along the eastern United States. The American lobster is found at intertidal depths to 720 m, but most frequently at 4-50 m, along the continental shelf throughout much of the western North Atlantic from southern Labrador to offshore North Carolina (Herrick 1909). Major coastal concentrations of lobster are in the Gulf of Maine and the coastal waters of New Brunswick and Nova Scotia, Canada (Cooper & Uzmann 1980). Major offshore concentrations are along the outer edge of the Continental Shelf and upper slope between the eastern part of Georges Bank and the Delaware Bay (Schroeder 1959). Small numbers inhabit the outer edge of the Nova Scotia shelf (Cooper & Uzmann 1980).

Informed management of the lobster fishery requires better understanding of the physical processes responsible for larval dispersal by shelf and coastal currents, and the exchange of larvae between offshore waters and coastal bays, where settlement and eventual harvest can occur (Brooks 2003).

Previous work has examined the effects of circulation models, winds, tidal mixing and hydrographic structures on lobster larval dispersion. The hope is to better understand the pathways of transport and delivery of larval lobsters from nearshore and coastal egg production. Rogers et al. (1968) were the first to suggest that coastal lobster populations may originate from larvae released offshore and carried shoreward where onshore surface currents exist. This hypothesis was based on the observation that stage 1 larvae were abundant offshore, but the final planktonic 4th stage larvae were far more common in coastal waters. Stage 1 lobster larvae migrate from 15-30 m depth during daylight to the upper 10 m at night (Incze 2000). Supporting this was evidence that there are unusually high levels of stage 4 lobster larvae in Long Island Sound (LIS).

The issue of larval movement has become pressing in areas of western LIS. In 1999, there was a massive die-off of lobsters from this area that corresponded with several years of above-normal water temperatures, application of pesticides for West Nile virus-carrying mosquito control and Paramoeba sp. infections (CT DEP 2003). The commercial lobster fishery in western LIS has not recovered in recent years. The lack of a recovery may be due, in part, to the required reseeding of the area with lobster larvae from other areas. Previous work has shown that lobsters collected in western LIS are more genetically differentiated from lobsters collected in other areas of LIS or the Hudson Canyon than can be explained because of geographical distance (Crivello et al. 2005).

The presence of genetic differences between egg-bearing female lobsters can be used to examine the parentage of lobster larvae. Fisheries that exploit mixed stocks are common and the contribution of specific stocks can be determined by analysis of genetic differences (Pella & Milner 1987). An early approach to characterize the contribution of specific stocks by differences in allele frequencies was through the use of the conditional maximum likelihood method. This approach maximizes a likelihood function of the stock-mixture genotype if their relative frequency in baseline stocks is known without error. This approach, and variants such as unconditional maximum likelihood and unconditional least squares, produce estimates of stock proportions that are increasingly biased as true stock-mixture proportions become more uneven (Pella & Milner 1987, Xu et al. 1994).

A Bayesian approach has been developed (Pella & Masuda 2000) that provides a probability distribution for stock composition estimates. Bayesian classifiers assume that the distribution of inputs within the target group is known exactly and that the prior probabilities of the classes are known, so posterior probabilities can be computed by a simple application of the Bayes' theorem (Ripley 1996). This approach has proven effective in analyzing difficult stock mixtures and provides results that are in line with other estimations of stock mixtures.

Recently, supervised artificial neural networks (ANN) composed of many simple processing units, connected by communication channels that carry numeric information (Bishop 1995, Masters 1993), have been used to identify stock-mixture genotypes (Brosse et al. 1999, Lek & Guegan 1999, Manel et al. 1999, Wu-Catherine & McLarty 2000, Brosse et al. 2001, Obach et al. 2001, Crivello et al. 2004). Neural networks are based on supervised procedures, the construction of a model based on examples of data with known outputs. These networks construct models solely from provided examples, which are assumed to implicitly contain the information necessary to establish the correct relationship. The sample (or training sets) is used to fit the parameters (weights) to minimize the generalization error. The performance of the ANN is compared by evaluating the generalization error using an independent validation set, and the ANN with the smallest error is selected for analysis of unknown samples. The structure of these models is a layered feed-forward network in which nonlinear elements (neurons) are arranged in successive layers and information flows unidirectionally, from input to output layer, through hidden layers. In a feed-forward ANN, the connections between processing units do not form cycles and rapidly produce a response to an input. Networks with hidden neurons have been shown to be universal approximators (Cybenko 1989, Hornick et al. 1989) for continuous maps and can be used to implement any defined function. In these types of networks, during the learning process, errors are corrected by back-propagation algorithms (Shuurmann & Muller 1994).

A key component of all of these approaches (Bayesian and ANN) is that they have the ability to correctly assign individuals, but a common drawback is that if the origin of the individual is not represented in the reference populations, most methods will designate a wrong population of origin (Cornuet et al. 1999). The Bayesian method is based on the assumptions that all loci are at Hardy-Weinberg equilibrium and at linkage equilibrium. Other constraints are the levels of differentiation between tested populations (the genetic differences), the ability to sample all, or virtually all of the potentially contributing stocks, the temporal stability of the microsatellite markers and a large sample size that contains representation for all donor populations (Letcher & King 1999). The neural network approach is not dependent on all loci at Hardy-Weinberg equilibrium.

Previous work (Crivello et al. 2005) has demonstrated that there is sufficient genetic difference between egg-bearing female lobsters collected from within LIS and the Hudson Canyon to use these statistical approaches to classify lobster larvae to the most likely source population. These statistical differences were used in a Bayesian approach (GeneClass), and a neural network approach (NeuroShell Classifier) to assign collected lobster larvae to the most likely source of egg-bearing female lobsters.

METHODS

Collections

Lobster larvae (stages 1-4) were collected with surface plankton nets in 2001 and 2002 from 5 sites within LIS: the Race, eastern, central, western LIS and Stratford Shoals area (Table 1 and Fig. 1). Larvae were preserved whole in propranol and then transported to the laboratory for genetic analysis. Adult lobsters are long-lived and it is reasonable to compare adults collected in 2001 with larvae collected in 2001 and 2002.

[FIGURE 1 OMITTED]

Genomic DNA Isolation

A small section (approximately 2.5 mm piece) of adult pereiopod soft tissue and the entire lobster larva were processed for genomic DNA. Genomic DNA was quantified with PicoGreen (Molecular Probes Inc.) and each sample was adjusted to 2 ng/[micro]L genomic DNA.

Microsatellite Loci

Lobster-specific microsatellite loci and flanking PCR primer sequences described in Jones et al. (2003) were used to characterize microsatellite alleles at nine loci (Crivello et al. 2005). To analyze each microsatellite locus, 10 ng of genomic DNA was mixed with a stock solution containing 0.5 [micro]M forward & reverse primer (with the forward primer tagged with a [D.sub.2], [D.sub.3] or [D.sub.4] fluorescent tag, Beckman Coulter, Palo Alto, California), 0.2 mM dNTPs, 10 [micro]L Tris, 50 mM KCl, 2.5 mM Mg[Cl.sub.2] and 0.5 units of a thermostable DNA polymerase in a final 10 [micro]L volume. Each sample was heated to 94[degrees]C for 30 sec, to the annealing temperature for 30 sec and then 72[degrees]C for 45 sec for 35 cycles.

The reaction products were diluted with 30 [micro]L of water and then 10 [micro]L of a [D.sub.2], [D.sub.3] and [D.sub.4] reaction were combined and precipitated. The precipitated products were washed with 70% ethanol and redissolved in formamide that contained 60-400 base pair size markers labeled with [D.sub.1] (Beckman Coulter, Palo Alto, California). The samples were then analyzed in a Beckman Seq2000 capillary electrophoresis system. Microsatellite alleles were identified by size with a resolution of 0.25 base pairs by comparison with size standards.

Statistical Approaches

The GeneClass program, freely available at (http://www.montpellier.inra.fr/URLB/geneclass/geneclass.html), uses multilocus genotypes to select or exclude populations as origins of individuals (Rannala & Mountain 1997). The GeneClass program uses the Bayes theorem to address assignment of individuals to populations.

An artificial neural network (ANN) supplied by NeuroShell Classifier (Ward Systems Group) was used. The data were entered as comma-separated values (csv) of the size (in base pairs) of each microsatellite allele produced by each individual. The generalization error was estimated by a cross-validation method (Masters 1993, Tibshirani 1996). The program calculates a fitness coefficient matrix to minimize the total number of incorrect classifications (a minimization of the generalization error) (Tibshirani 1996). A neural network of this type has been used to classify juveniles to their most likely spawning group (Lek & Guegan 1999, Wu-Catherine & McLarty 2000, Brosse et al. 2001). In the ANN, the generalization error is minimized by a cross-validation procedure and the ANN with the least generalization area is used to classify unknown samples. After validation, the trained neural network is then used to classify all unknown samples.

RESULTS

Egg-bearing females and lobster larvae were examined for allele frequencies at nine microsatellite loci and differences in allele frequencies were used to determine the parentage of lobster larvae with a neural network and Bayesian statistical approach.

The estimated parentage of collected larvae (all stages) from different areas by both methods is given in Table 2 and Figure 2. Though there were some differences between the parentages determined by the two approaches, the differences were not great. The majority of lobster larvae collected in central LIS in 2002 were determined to have arisen from egg-bearing females found in the Hudson Canyon (between 39% and 45%), followed by eastern and central LIS (between 21% and 35%). There were very few larvae collected in central LIS that had arisen from egg-bearing females in western LIS or in the Stratford Shoals area (between 3% and 8%). Approximately 90% of the larvae collected in central LIS arose from egg-bearing females from eastern and central LIS and the Hudson Canyon area.

[FIGURE 2 OMITTED]

In contrast, an equal number of lobster larvae collected in eastern LIS in 2001 and 2002 arose from egg-bearing females in the Hudson Canyon and central LIS sites (approximately 25% to 40%, by both approaches). The parentage was very similar from year-to-year. Approximately 15% of the lobster larvae arose from egg-bearing females in western LIS and the Stratford Shoals area. Between 70% and 90% of the larvae arose from egg-hearing females from eastern and central LIS and the Hudson Canyon area, slightly less than in the case of larvae collected from central LIS.

Most larvae collected in the Race area arose from egg-bearing females from western LIS (approximately 30%). Only 58% to 67% of the larvae arose from egg-bearing females from eastern and central LIS and the Hudson Canyon area.

Most larvae collected in the Stratford Shoals area arose from egg-bearing female lobsters collected in the Hudson Canyon area (2002) and central LIS (2001). The parentage of larvae collected in the Stratford Shoals area differed substantially year-to-year. The larval input from western LIS and Stratford Shoals egg-bearing females was low in both years.

Most larvae collected from sites in western LIS in 2001 and 2002 arose from the Hudson Canyon and central LIS female lobsters. The relative input of larvae from the Hudson Canyon was greater in 2002 than 2001. Once again, the input from western LIS and Stratford Shoals egg-bearing females was low in both years.

An examination of the parentage of larval stages is given in Table 3 and Figure 3. An examination of the parentage of larvae collected in central LIS reveals that the parentage of stage 2, 3 and 4 larvae are very similar with most of them originating from females in the Hudson Canyon area. Most stage 1 larvae arose from females in central LIS. One other notable difference was that there was little input from central LIS females into collected stage 2 larvae. Few stage 1 larvae originated from either western L1S or Stratford Shoals females (0% to 6%), but more stage 2 larvae were from these areas (10% to 17%).

[FIGURE 3 OMITTED]

An examination of the parentage of larvae collected in eastern LIS reveals that the parentage of stage 3 and 4 larvae did not change significantly from year-to-year, with one exception. Most of the stage 3 and 4 larvae originated in females collected from the Hudson Canyon, and eastern and central LIS. A greater percentage of stage 3 larvae originated from western LIS females in 2001 than 2002 (37% vs. 20% in 2001; 31% vs. 21% in 2002).

There were significant differences between the parentage of stage 3 and 4 larvae collected in the Race. The greatest source of stage 3 larvae was from western LIS and the Stratford Shoals area. The greatest source of stage 4 larvae was from central LIS female lobsters.

In the Stratford Shoals area, the greatest source of stage 1 larvae in 2001 and 2002 was from females in the Hudson Canyon area. This may not necessarily represent the true major source of stage 1 larvae because few stage 1 larvae were collected in 2001. The greatest source of stage 3 larvae in 2001 was from central and eastern LIS (60% to 70%). The greatest source of stage 4 larvae in 2001 and 2002 was from central LIS, but there was an increased contribution from western LIS in 2002.

In western LIS, the greatest source of stage 1 larvae in 2001 and 2002 was from the Hudson Canyon area with very little input from western LIS or the Stratford Shoals area. Very few stage 2 and stage 3 larvae were collected in 2001, with the greatest source of stage 2 larvae from western LIS and the greatest source of stage 3 larvae from central LIS.

When larval parentage was examined as a function of stage, regardless of the collection site (Table 4) there were significant differences. Most of stage 1 larvae (51%) arose from the Hudson Canyon area, followed by central and eastern LIS. There were very few stage 1 larvae arising from western LIS or from the Stratford Shoals area.

There were a limited number of stage 2 larvae analyzed and most arose from the Hudson Canyon, followed by western LIS. Stage 3 and stage 4 larvae arose equally from central and eastern LIS and from the Hudson Canyon area.

DISCUSSION

Previous work has suggested that there was little genetic population structure among lobster populations in Long Island Sound and the Gulf of Maine (Harding et al. 1997). Recent work has demonstrated that lobster populations in western LIS are more genetically different from other Long Island Sound populations or lobster populations in the Hudson Canyon than can be explained by geographical distance (Crivello et al. 2005). These genetic differences may be caused by selective pressures, factors reducing gene flow or because of a massive die-off in 1999 (the founder effect). The complete collapse of the commercial lobster fishery in Long Island Sound in 1999 has led to efforts to develop a management plan to return the fishery to commercial viability (CT DEP 2003). A critical component of this process is to determine the source or sources of lobster larvae within Long Island Sound and their impact on commercial fisheries.

H. americanus females begin to reproduce at 5-7 y and carry their eggs for %1 I mo. Lobster larvae are present in the water column for 4-8 wk during June,through August. Larval movement is influenced by wind-driven transport, passive drift via coastal currents and the ability of stage 3 and 4 lobster larvae to swim effectively. All of these processes are believed to be responsible for the movement of lobster larvae from offshore canyons into coastal waters (Clancy & Cobb 1996). Examination of offshore and coastal larval distribution reveals a gradient of larval stages with a greater proportion of early larval stages in the vicinity of offshore canyons and more mature stage 3 and 4 larvae closer to coastal habitats (Katz et al. 1994).

To examine the parentage of lobster larvae collected within LIS, and the adjacent Race area, differences in microsatellite allele frequencies were compared between lobster larvae and previously characterized egg-bearing female lobsters collected from the same areas and the offshore Hudson Canyon area. Parentage was determined by a neural network approach (Crivello et al. 2004) and a Bayesian approach (Rannala & Mountain 1997). The neural network approach has proven to be successful in identifying parentage when microsatellite loci do not conform to Hardy-Weinberg equilibrium (Crivello et al. 2004). The Bayesian approach has been used successfully to address parentage issues within salmonids (Marcot et al. 2001).

Examination of parentage by both methods gave very similar results even though they are based on different statistical assumptions. These experiments support earlier work that suggested that Hudson Canyon lobster larvae are carried into LIS (Schroeder 1959, Tracey et al. 1975). Hudson Canyon lobster larvae are transported to the western reaches of LIS and contribute to populations throughout LIS. This east-to-west flow of larvae is supported by currents and the flood tides that carry large volumes of water toward its western end of LIS (Brown 1984, Kenefik 1985, Signell et al. 2000). Ebb tides are likely responsible for the transport of western LIS larvae to the race area.

The poor contribution of western LIS and Stratford Shoal area to larvae collected in other areas may be caused by reduced numbers of egg-bearing females after the massive die off in 1999, or an artifact introduced by the examination of only eleven egg-bearing females collected in western LIS. Future experiments will have to examine many more egg-bearing fei9males from western LIS to resolve this point. The demonstration of a large genetic difference between egg-bearing females from western LIS and the Stratford Shoals area suggests another possible reason. If the genetic differences were not caused by the massive die-off but instead were an indication of anthropogenic pressures and geographical limits to gene flow, this would suggest that western LIS larvae were selected for growth in western LIS and were less well adapted for growth in other areas of LIS. Western LIS receives greater anthropogenic pressures that eastern LIS, and is the site of greater population numbers, sewage treatment plants and industrial operations (CT DEP 2003). These types of anthropogenic pressures have been shown previously to induce resistance to pesticides in a wide range of insect species (Baker & Argobast 1995, ffrench-Constant et al. 2000). The geographical contours of western LIS, near the Stratford Shoals area, have been suggested to be geographical barriers to movement of lobsters from central to eastern LIS (Howell et al. 2003). Western LIS is also at the geographical limit of the American lobster's natural range and has seen several years of increased water temperatures.

The demonstration of larval movement from other areas of LIS and the Hudson Canyon clearly indicates high levels of gene flow between areas within LIS. For such genetic differences to exist in the presence of high levels of gene flow, strong selective pressures must be present. Unfortunately, these experiments and previous work (Crivello et al. 2005) cannot discriminate between these two possibilities.

Examination of lobster larval parentage as a function of larval stage did indicate stage-specific differences. There were a limited number of stage 2 larvae collected from any site within Long Island Sound or the Race, making it difficult to draw many conclusions about parentage. In support of previous work, there were many more stage 3 and 4 larvae collected in LIS, than stage 1 and 2 larvae (Incze 2000, Brooks 2003). The parentage of stage 3 and 4 larvae were essentially the same as the parentage of all larval stages, which is not surprising because they represented 80% of all collected larvae. Interestingly, most stage 1 larvae were from the Hudson Canyon area with very little input from western parts of LIS. Because stage 1 larvae do not have the swimming ability of possibly stage 3 and stage 4 larvae, this distribution is presumably because of the effects of tides and currents. Even though it seems that fewer stage 2 larvae are from central and eastern areas of LIS, this may be because of the few numbers of tested stage 2 larvae.

The patterns of larval parentage were temporally consistent in 2001 and 2002 in eastern LIS suggesting that the factor or factors affecting larval input are not an entirely random process. There are some minor stage-specific parentage differences, predominantly an increase in the input from western LIS and the Stratford Shoals in stage 3 larvae.

The recent evidence that the lobster populations in western LIS have not rebounded since 1999 may be due in part to the requirement of larval reseeding from other areas or because conditions in western LIS may have selected for populations resistant to anthropogenic influences. This work clearly establishes that lobster larvae from the Hudson Canyon and other areas of LIS travel to the western reaches of LIS. It is unclear if these larvae will he successful in developing commercially viable populations.

ACKNOWLEDGMENTS

The authors thank Penny Howell and the scientists at the Connecticut DEP for the collection of lobster larvae. This work was funded by the Connecticut DEP and the National Marine Fisheries Service.

LITERATURE CITED

Baker, J. E. & R. T. Arbogast. 1995. Malathion resistance in field strains of the warehouse pirate bug (Heteroptera: Anthocoridae) and a prey species Tribolium castaneum (Coleoptera: Tenebrionidae). J. Ecol. Entomol. 88:241-245.

Bishop, C. M. 1995. In: Neural networks for pattern recognition. England: Oxford University Press.

Brooks, D. 2003. Dispersal of lobster larvae within and between coastal bays in the eastern Gulf of Maine: preliminary model studies. Geophys. Res. Abs. 5:01758.

Brosse, S., J. F. Guegan, J. N. Tourenq & S. Lek 1999. The use of neural networks to assess fish abundance and spatial occupancy in the littoral zone of a mesotrophic lake. Ecol. Model 120:299-311.

Brosse, S., J. L. Giraudel & S. Lek. 2001. Utilization of non-supervised neural networks and principal component analysis to study fish assemblages. Ecol. Model 146:159-166.

Brown, W. S. 1984. A comparison of Georges Bank, Gulf of Maine, and New England shelf tidal dynamics. J. Phys. Ocean 14:145-167.

Connecticut Department of Environmental Protection (CT DEP). 2003. The status of the Long Island Sound lobster resource. In: N. C. Balcom, editor. Program and abstracts of the 3rd Long Island Sound lobster health symposium. March 7, Bridgeport, CT. Connecticut Sea Grant Publication No. CT-SG-03-03. pp. 2-4.

Clancy, M. & J. S. Cobb. 1996. Recruitment strategies in marine decapods: A comparative approach. J. Shellfish Res. 15:493-497.

Cooper, R.A. & J.F. Uzmann. 1980. Ecology of juvenile and adult Homarus americanus. In: J. S. Cobb & B.F. Phillips, editors. The biology and management of lobsters. New York: Academic Press. pp. 215-276

Cornuet, J. M., S. Piry, G. Luikart, A. Estoup & M. Solignac. 1999. New methods employing multilocus genotypes to select or exclude populations as origins of individuals. Genetics 153:1989-2000.

Crivello, J. F., D. F. Landers & M. Keser. 2005. The genetic stock structure of the American lobster (Homarus americanus) in the Long Island Sound and the Hudson Canyon. J. Shellfish Res. 24(3):841-848.

Crivello, J. F., D. F. Danila, E. Lorda, M. Keser & E. F. Roseman. 2004. The genetic stock structure of larval and juvenile winter flounder (Pseudopleuronectes americanus) larvae in Connecticut waters of eastern Long Island Sound and estimations of larval entrainment. J. Fish Res. 64:1-15.

Cybenko, G. 1989. Approximation by superpositions of a sigmoidal function. Math. Cont. Sign. Sys. 2:303-314.

ffrench-Constant, R. H., N. Anthony, K. Aronstein, T. Rocheleau & G. Stilwell. 2000. Cyclodiene insecticide resistance: from molecular to population genetics. An. Rev. Entom. 48:447464.

Harding, G. C., E. L. Kenchington, C. J. Bird, D. S. Pezzack & D. C. Landry 1997. Genetic relationships among subpopulations of the American lobster (Homarus americanus) as revealed by random amplified polymorphic DNA. Can. J. Fish. Aquat. Sci. 54:1762-1771.

Herrick, F.H. 1909. Natural history of the American lobster. Bull. U.S. Bureau Fish 29:149-408.

Hornick, K., M. Stinchcombe & H. White. 1989. Multilayer feedforward networks are universal approximators. Neur. Net. 2:359-366.

Howell, P., C. Giannini, K. Gottschall, D. Pacileo & J. Benway 2003. Assessment and monitoring of the American lobster resource and fishery in Long Island Sound. Semi-annual Performance Report, NOAA/ MNFS, NA16FW1238.

Incze, L. S. 2000. Post-larval lobster (Homarus americanus) distributions in Penobscot Bay in relation to hydrography, circulation and remote sensing information. Annual Report for 2000, NOAA/NESDIS.

Jones, M. W., P. T. O'Reilly, A. A. McPherson, T. L. McParland, D.E. Armstrong, A. J. Cox, K. R. Spence, E. L. Kenchington, C. T. Taggart & P. Benzten. 2003. Development, characterization, inheritance, and cross-species utility of American lobster (Homarus americanus) microsatellite and mtDNA PCR-RFLP markers. Genome 46:59-69.

Katz, C. H., J. S. Cobb & M. Spaulding. 1994. Larval behavior, hydrodynamic transport, and potential offshore-to-inshore recruitment in the American lobster (Homarus americanus). Mar. Ecol. Prog. Ser. 103: 265-273.

Kenefik, A. M. 1985. Barotropic [M.sub.2] tides and tidal currents in Long Island Sound: a numerical model. J. Coast. Res. 1:117-128.

Letcher, B.H. & T.L. King. 1999. Targeted stock identification using multilocus genotype "familyprinting." Fish. Res. 43:99-111.

Lek, S. & J.F. Guegan. 1999. Artificial neural networks as a tool in ecological modeling, an introduction. Ecol. Model. 120:65-73.

Manel, S., J.M. Dias & S.J. Ormerod. 1999. Comparing discriminant analysis, neural networks and logistic regression for predicting species distributions, a case with a Himalayan river bird. Ecol. Model. 120: 337-347.

Marcot, B. G., R. S. Holthausen, M. G. Raphael, M. M. Rowland & M. J. Wisdom. 2001. Using Bayesian belief networks to evaluate fish and wildlife population viability under land management alternatives from an environmental impact statement. For. Ecol. Man 153:29-42.

Masters, T. 1993. In: T. Masters, editor. Practical neural network recipes in [C.sub.++]. San Diego, CA: Academic Press.

Obach, M., R. Wagner, H. Werner & H. Schmidt. 2001. Modelling population dynamics of aquatic insects with artificial neural networks. Ecol Model. 146:207-217.

Pella, J. J. & G. B. Milner. 1987. Use of genetic marks in stock composition analysis. In: N. Ryman & F. Utter, editors. Population genetics and fisheries management. Seattle: University of Washington Press. pp. 247-276.

Pella, J. J. & M. Masuda. 2000. Bayesian methods for analysis of stock mixtures from genetic characters. Fish. Bull. 99:151-167.

Rannala, B. & J. L. Mountain. 1997. Detecting immigration by using multilocus genotypes. Proc. Natl. Acad. Sci. USA 94:9197-9201.

Ripley, B.D. 1996. Pattern recognition and neural networks. In: B.D. Ripley, editor. Cambridge, England: Cambridge University Press. pp. 1-399.

Rogers, B. A., J. S. Cobb & N. Marshall. 1968. Size cmnparisons of inshore and offshore larvae of the lobster, Homarus americanus, off southern New England. Proc. Natl. Shell. Coun. 58:78-81.

Schroeder, W.C. 1959. The lobster, Homarus americanus, and the red crab, Gervon quinquedens, in the offshore waters of western North Atlantic. Deep Sea Res. 5:266-282.

Schuurman, G. & E. Muller. 1994. Back-propagation neural networks, recognition vs. prediction capability. Environ. Toxicol. Chem. 13:743-747.

Signell, R. P., J. H. List & A. S. Farris. 2000. Bottom currents and sediment transport in Long Island Sound: A modeling study. J. Coast. Res. 16:551-566.

Tibshirani, R. 1996. A comparison of some error estimates for neural network models. Neur. Comp 8:152-163.

Tracey, M. L., K. Nelson, D. Hedgecock, R. A. Shleser & M. Pressick. 1975. Biochemical genetics of lobsters: genetic variation and the structure of the American lubster (Homarus americanus) populations. J. Fish. Res. Board Can. 32:2091-2101.

Wu-Catherine, H. & J. W. McLarty. 2000. Neural networks and genome informatics. In: Catherine H. Wu & J. W. McLarty, editors. Methods in computational biology and biochemistry, New York: Elsevier. pp. 1-205.

Xu. S., C.J. Kobak & P.E. Smouse. 1994. Constrained least squares estimate of mixed population stock composition from mtDNA haplotype frequency data. Can. J. Fish. Aquat. Sci. 51:417-425.

JOSEPH F. CRIVELLO, (1) * DONALD F. LANDERS JR. (2) AND MILAN KESER (2)

(1) Department of Physiology and Neurobiology University of Connecticut, Storrs, Connecticut 06269; (2) Millstone Power Station Resource Services Inc. Environmental Laboratory, Waterford, Connecticut 06385

* Corresponding author: E-mail: [e-mail]

TABLE 1.
Lobster larval collections from eve sites within Long Island Sound.

               Collection                Larval
Site           Date                      Number

Race           6/25/02 through 7/11/02   State 1-0     Total = 145
                                         Stage 2-8
                                         Stage 3-59
                                         Stage 4-78
Eastern LIS    7/5/01 through 8/9/01     Stage 1-1     Total = 106
                                         Stage 2-1
                                         Stage 3-23
                                         Stage 4-81
               6/20/02 through 7/18/02   Stage 1-1     Total = 250
                                         Stage 2-3
                                         Stage 3-57
                                         Stage 4-189
Central LIS    6/20/02 through 6/24/02   Stage I-14    Total = 112
                                         Stage 2-18
                                         Stage 3-33
                                         Stage 4-47
Stratford      6/20/01 through 6/25/01   Stage 1-15    Total = 133
  Shoals LIS                             Stage 2-0
                                         Stage 3-67
                                         Stage 4-51
               6/17/02 through 7/22/02   Stage 1-24    Total = 43
                                         Stage 2-0
                                         Stage 3-0
                                         Stage 4-19
Western LIS    6/12/01 through 6/25/01   Stage 1-24    Total = 210
                                         Stage 2-3
                                         Stage 3-50
                                         Stage 4-0
               5/28/02 through 7/30/02   Stage 1-72    Total = 120
                                         Stage 2-0
                                         Stage 3-31
                                         Stage 4-17

TABLE 2.
Analysis of the parentage of lobster larvae to egg-bearing female
lobsters based on differences of microsatellite allele frequencies.

                              Fractional Assignment
                              (NN = Neural Network;
                                  B = Bayesian)

                   Central          Eastern
                     LIS              LIS
Parental            Larvae          Larvae
Sources             (2001)          (2001)

                      NN       B      NN       B
Hudson Canyon        0.45    0.39     0.32    0.32
Eastern LIS          0.19    0.17     0.19    0.19
Central LIS          0.21    0.35     0.25    0.40
Stratford Shoals     0.06    0.03     0.10    0.05
Western LIS          0.08    0.06     0.14    0.05
N                  112              104

                              Fractional Assignment
                              (NN = Neural Network;
                                  B = Bayesian)

                   Central          Eastern
                     LIS              LIS
Parental            Larvae          Larvae
Sources             (2001)          (2001)

                      N        B
Hudson Canyon        0.29    0.39     0.24    0.21
Eastern LIS          0.20    0.14     0.15    0.22
Central LIS          0.24    0.36     0.19    0.24
Stratford Shoals     0.13    0.06     0.09    0.06
Western LIS          0.14    0.05     0.33    0.27
N                  250              148

                   Stratford
                    Shoals
                    Larvae
                    (2001)            (2002)

                     NN         B       NN      B
Hudson Canyon         0.20     0.14    0.49    0.46
Eastern LIS           0.31     0.28    0.16    0.11
Central LIS           0.36     0.27    0.21    0.20
Stratford Shoals      0.04     0.16    0.05    0.11
Western LIS           0.08     0.14    0.09    0.12
N                   133               43

                        Fractional Assignment
                        (NN = Neural Network;
                            B = Bayesian)

                  Western
                    LIS
Parental           Larvae
Sources            (2001)          (2002)

                     NN      B       NN      B
Hudson Canyon        0.39   0.28     0.57   0.44
Eastern LIS          0.09   0.14     0.08   0.16
Central LIS          0.33   0.31     0.24   0.28
Stratford Shoals     0.06   0.10     0.03   0.04
Western LIS          0.13   0.18     0.08   0.06
N                   54              60

TABLE 3.
Analysis of the parentage of lobster larvae to previously characterized
egg-bearing female lobsters as a function of larval class.

                   Fractional Assignment (NN = Neural
                         Network; B = Bayesian)

                     NN       B       NN       B

Parental Sources        Central LIS Larvae (2001)

Lobster Larvae     Stage 1          Stage 2
Hudson Canyon       0.29     0.32    0.5      0.47
Eastern LIS         0.29     0.32    0.14     0.14
Central LIS         0.35     0.30    0.07     0.12
Stratford Shoals    0.00     0.06    0.14     0.17
Western LIS         0.06     0.00    0.14     0.10
N                  18               14

                         Eastern LIS Larvae (2001)

Hudson Canyon       *        *      *         *
Eastern LIS         *        *      *         *
Central LIS         *        *      *         *
Stratford Shoals    *        *      *         *
Western LIS         *        *      *         *
N                   1               1

                         Eastern LIS Larvae (2002)

Central LIS         *        *      *         *
Eastern LIS         *        *      *         *
Hudson Canyon       *        *      *         *
Stratford Shoals    *        *      *         *
Western LIS         *        *      *         *
N                   1               3

                            Race Larvae (2002)

Hudson Canyon       *        *      *         *
Eastern LIS         *        *      *         *
Central LIS         *        *      *         *
Stratford Shoals    *        *      *         *
Western LIS         *        *      *         *
N                   0        0      8

                      Stratford Shoals Larvae (2001)

Hudson Canyon       0.67     0.53   *         *
Eastern LIS         0.13     0.20   *         *
Central LIS         0.13     0.20   *         *
Stratford Shoals    0.00     0.00   *         *
Western LIS         0.07     0.07   *         *
N                  15               0

                      Stratford Shoals Larvae (2002)

Hudson Canyon       0.80     0.61   *         *
Eastern LIS         0.08     0.08   *         *
Central LIS         0.12     0.06   *         *
Stratford Shoals    0.00     0.00   *         *
Western LIS         0.00     0.19   *         *
N                  24               0

                         Western LIS Larvae (2001)

Hudson Canyon       0.49     0.35    0.25     0.27
Eastern LIS         0.09     0.15    0.08     0.17
Central LIS         0.33     0.39    0.25     0.29
Stratford Shoals    0.06     0.06    0.00     0.00
Western LIS         0.03     0.05    0.42     0.28
N                  33               12

                         Western LIS Larvae (2002)

Hudson Canyon       0.57     0.44   *         *
Eastern LIS         0.08     0.16   *         *
Central LIS         0.24     0.30   *         *
Stratford Shoals    0.03     0.06   *         *
Western LIS         0.08     0.04   *         *
N                  60               0

                   Fractional Assignment (NN = Neural
                        Network; B = Bayesian)

                     NN       B       NN       B

Parental Sources        Central LIS Larvae (2001)

Lobster Larvae     Stage 3          Stage 4
Hudson Canyon       0.47     0.43    0.49     0.39
Eastern LIS         0.23     0.20    0.17     0.17
Central LIS         0.20     0.26    0.21     0.27
Stratford Shoals    0.00     0.00    0.10     0.08
Western LIS         0.10     0.11    0.06     0.08
N                  33               47

                         Eastern LIS Larvae (2001)

Hudson Canyon       0.25     0.24    0.35     0.24
Eastern LIS         0.17     0.21    0.21     0.17
Central LIS         0.21     0.25    0.27     0.30
Stratford Shoals    0.04     0.09    0.12     0.13
Western LIS         0.33     0.21    0.08     0.06
N                  23               81

                         Eastern LIS Larvae (2002)

Central LIS         0.14     0.28     0.27     0.35
Eastern LIS         0.24     0.14     0.18     0.14
Hudson Canyon       0.31     0.29     0.29     0.29
Stratford Shoals    0.19     0.12     0.11     0.07
Western LIS         0.12     0.17     0.15     0.14
N                  57               189

                            Race Larvae (2002)

Hudson Canyon       0.24     0.24     0.16    0.24
Eastern LIS         0.25     0.09     0.20    0.23
Central LIS         0.16     0.21     0.37
Stratford Shoals    0.14     0.05     0.08    0.05
Western LIS         0.21     0.42     0.19    0.16
N                  59                78

                      Stratford Shoals Larvae (2001)

Hudson Canyon       0.19     0.13     0.15    0.17
Eastern LIS         0.31     0.21     0.33    0.23
Central LIS         0.37     0.26     0.39    0.30
Stratford Shoals    0.06     0.19     0.03    0.10
Western LIS         0.08     0.21     0.15    0.20
N                  67                51

                      Stratford Shoals Larvae (2002)

Hudson Canyon      *         *        0.11    0.26
Eastern LIS        *         *        0.26    0.11
Central LIS        *         *        0.31    0.31
Stratford Shoals   *         *        0.05    0.05
Western LIS        *         *        0.26    0.26
N                  0                 19       0

                         Western LIS Larvae (2001)

Hudson Canyon      0.22      0.20     *       *
Eastern LIS        0.11      0.11     *       *
Central LIS        0.44      0.46     *       *
Stratford Shoals   0.11      0.11     *       *
Western LIS        0.11      0.11     *       *
N                  9                  0

                         Western LIS Larvae (2002)

Hudson Canyon      *         *       *        *
Eastern LIS        *         *       *        *
Central LIS        *         *       *        *
Stratford Shoals   *         *       *        *
Western LIS        *         *       *        *
N                  0                 0

TABLE 4.
Analyxis of the parentage of lobster larvae by stage.

Parental Sources    All Larvae Regardless of Collection Site

                   Stage 1   Stage 2   Stage 3   Stage 4

Hudson Canyon        0.51      0.38      0.27      0.29
Eastern LIS          0.14      0.15      0.25      0.22
Central LIS          0.26      0.15      0.23      0.31
Stratford Shoals     0.03      0.08      0.10      0.10
Western LIS          0.05      0.27      0.14      0.15
N                  147        26       251       443