Zebra mussel, Dreissena polymorpha, in the Great Lakes is being monitored as a bio-indicator organism for environmental health effects by the National Oceanic and Atmospheric Administration’s Mussel Watch program. In order to monitor the environmental effects of industrial pollution on the ecosystem, invasive zebra mussels were collected from four stations—three inner harbor sites (LMMB4, LMMB1, and LMMB) in Milwaukee Estuary, and one reference site (LMMB5) in Lake Michigan, Wisconsin. Nuclear magnetic resonance (NMR)-based metabolomics was used to evaluate the metabolic profiles of the mussels from these four sites. The objective was to observe whether there were differences in metabolite profiles between impacted sites and the reference site; and if there were metabolic profile differences among the impacted sites. Principal component analyses indicated there was no significant difference between two impacted sites: north Milwaukee harbor (LMMB and LMMB4) and the LMMB5 reference site. However, significant metabolic differences were observed between the impacted site on the south Milwaukee harbor (LMMB1) and the LMMB5 reference site, a finding that correlates with preliminary sediment toxicity results. A total of 26 altered metabolites (including two unidentified peaks) were successfully identified in a comparison of zebra mussels from the LMMB1 site and LMMB5 reference site. The application of both uni- and multivariate analysis not only confirmed the variability of altered metabolites but also ensured that these metabolites were identified via unbiased analysis. This study has demonstrated the feasibility of the NMR-based metabolomics approach to assess whole-body metabolomics of zebra mussels to study the physiological impact of toxicant exposure at field sites. Electronic supplementary materialThe online version of this article (doi:10.1007/s11306-015-0789-4) contains supplementary material, which is available to authorized users.

Using a combination of population- and individual-based analytical approaches, we provide a comprehensive examination of genetic connectivity of Dungeness crab (Cancer magister) along ~1,200 km of the California Current System (CCS). We sampled individuals at 33 sites in 2012 to establish a baseline of genetic diversity and hierarchal population genetic structure and then assessed interannual variability in our estimates by sampling again in 2014. Genetic diversity showed little variation among sites or across years. In 2012, we observed weak genetic differentiation among sites (F range = -0.005-0.014) following a pattern of isolation by distance (IBD) and significantly high relatedness among individuals within nine sampling sites. In 2014, pairwise F estimates were lower (F range = -0.014-0.007), there was no spatial autocorrelation, and fewer sites had significant evidence of relatedness. Based on these findings, we propose that interannual variation in the physical oceanographic conditions of the CCS influences larval recruitment and thus gene flow, contributing to interannual variation in population genetic structure. Estimates of effective population size (N ) were large in both 2012 and 2014. Together, our results suggest that Dungeness crab in the CCS may constitute a single evolutionary population, although geographically limited dispersal results in an ephemeral signal of IBD. Furthermore, our findings demonstrate that populations of marine organisms may be susceptible to temporal changes in population genetic structure over short time periods; thus, interannual variability in population genetic measures should be considered.

No abstract

No abstract

No abstract

No abstract

Marine populations are often typified by large annual variations in the number of larvae that return to the adult population. The Dungeness crab Cancer (Metacarcinus) magister is an important economic and ecological species along the western seaboard of the continental USA. Research suggests larval returns of Dungeness crabs vary annually by a factor of 1000, strongly influencing the population dynamics of the species. To understand how hydrographic conditions affect population dynamics, a light trap in Coos Bay, Oregon, was monitored daily during the recruitment season (April to September) from 1997 to 2001 and from 2006 to the present. Using an individual-based biophysical model, we tested the hypothesis that more Dungeness crab larvae recruit during negative-phase Pacific Decadal Oscillation (PDO). The model uses the Regional Oceanic Modeling System to simulate circulation in the California Current and an offline Lagrangian particle-tracking algorithm (Larval TRANSport Lagrangian Model, LTRANS) to model larval dispersal. We validated our model by comparing the model data to the light trap data. Our findings support the hypothesis that more megalopae (pelagic postlarvae) recruit during the negative phase of the PDO. In addition, megalopae appear to spend longer in the water column during positive-phase PDO as a result of faster development rates likely due to warmer seawater temperature. Lastly, our model suggests that the population experiences more self-recruitment than previously thought, albeit not to an extent to suggest there are multiple metapopulations.

Hierarchical Reinforcement Learning (HRL) allows interactive agents to decompose complex problems into a hierarchy of sub-tasks. Higher-level tasks can invoke the solutions of lower-level tasks as if they were primitive actions. In this work, we study the utility of hierarchical decompositions for learning an appropriate way to interact with a complex interface. Specifically, we train HRL agents that can interface with applications in a simulated Android device. We introduce a Hierarchical Distributed Deep Reinforcement Learning architecture that learns (1) subtasks corresponding to simple finger gestures, and (2) how to combine these gestures to solve several Android tasks. Our approach relies on goal conditioning and can be used more generally to convert any base RL agent into an HRL agent. We use the AndroidEnv environment to evaluate our approach. For the experiments, the HRL agent uses a distributed version of the popular DQN algorithm to train different components of the hierarchy. While the native action space is completely intractable for simple DQN agents, our architecture can be used to establish an effective way to interact with different tasks, significantly improving the performance of the same DQN agent over different levels of abstraction.