A general deep learning model for bird detection in high-resolution airborne imagery.

Advances in artificial intelligence for computer vision hold great promise for increasing the scales at which ecological systems can be studied. The distribution and behavior of individuals is central to ecology, and computer vision using deep neural networks can learn to detect individual objects in imagery. However, developing supervised models for ecological monitoring is challenging because it requires large amounts of human-labeled training data, requires advanced technical expertise and computational infrastructure, and is prone to overfitting. This limits application across space and time. One solution is developing generalized models that can be applied across species and ecosystems. Using over 250,000 annotations from 13 projects from around the world, we develop a general bird detection model that achieves over 65% recall and 50% precision on novel aerial data without any local training despite differences in species, habitat, and imaging methodology. Fine-tuning this model with only 1000 local annotations increases these values to an average of 84% recall and 69% precision by building on the general features learned from other data sources. Retraining from the general model improves local predictions even when moderately large annotation sets are available and makes model training faster and more stable. Our results demonstrate that general models for detecting broad classes of organisms using airborne imagery are achievable. These models can reduce the effort, expertise, and computational resources necessary for automating the detection of individual organisms across large scales, helping to transform the scale of data collection in ecology and the questions that can be addressed.

Accounting for food availability reveals contaminant-induced breeding impairment, food-modulated contaminant effects, and endpoint-specificity of exposure indicators in free ranging avian populations.

It remains unclear how sub-lethal effects of contaminants play out in relation to other stressors encountered by free-ranging populations. Effects may be masked or influenced by interactions with field stressors such as food availability. We predicted that (1) including food availability, and particularly its interaction with Hg, would reveal or enhance associations between Hg and breeding endpoints. We further predicted that (2) breeding impairment associated with Hg would be higher under food stress conditions. We monitored Hg and nest success of great egrets (Ardea alba) in eight breeding colonies in the Florida Everglades over 11 years. We characterized variation in local food availability among colonies and years using fish biomass and recession range -a proxy to fish vulnerability. We used two Hg exposure indicators (egg albumen Hg and nestling feather Hg) and six breeding endpoints (clutch-size, brood-size, fledged-size, hatching success, post-hatching success and fledglings per egg) to assess whether variation in food availability influenced associations between Hg and these endpoints. Accounting for interactions between Hg and food availability, we identified statistically significant associations in all 12 indicator-endpoint combinations, while only three were detectable without food. Further, 10 combinations showed interactions between Hg and components of food availability. Our results also indicated an endpoint-specific affinity, with albumen [Hg] explaining more variation in hatching success while nestling feather [Hg] explained more variation in post-hatching survival. Both Hg indicators accounted for relevant (6-10%) amounts of variation in fledglings produced per egg laid, an integrative endpoint. Increased Hg exposure resulted in overall reduced reproductive success when food availability was low, but our models predicted low or no effects of increasing Hg exposure when food availability was high. Our results indicate that Hg induced impairment is strongly driven by food availability, providing a framework that accommodates previously contradictory results in the literature.

AHR2-Mediated transcriptomic responses underlying the synergistic cardiac developmental toxicity of PAHs.

Polycyclic aromatic hydrocarbons (PAHs) induce developmental defects including cardiac deformities in fish. The aryl hydrocarbon receptor (AHR) mediates the toxicity of some PAHs. Exposure to a simple PAH mixture during embryo development consisting of an AHR agonist (benzo(a)pyrene-BaP) with fluoranthene (FL), an inhibitor of cytochrome p450 1(CYP1)--a gene induced by AHR activation--results in cardiac deformities. Exposure to BaP or FL alone at similar concentrations alters heart rates, but does not induce morphological deformities. Furthermore, AHR2 knockdown prevents the toxicity of BaP + FL mixture. Here, we used a zebrafish microarray analysis to identify heart-specific transcriptomic changes during early development that might underlie cardiotoxicity of BaP + FL. We used AHR2 morphant embryos to determine the role of this receptor in mediating toxicity. Control and knockdown embryos at 36 h post-fertilization were exposed to DMSO, 100 µg/l BaP, 500 µg/l FL, or 100 µg/l BaP + 500 µg/l FL, and heart tissues for RNA were extracted at 2, 6, 12, and 18 h-post-exposure (hpe), prior to the appearance of cardiac deformities. Data show AHR2-dependent BaP + FL effects on expression of genes involved in protein biosynthesis and neuronal development in addition to signaling molecules and their associated molecular pathways. Ca(2+)-cycling and muscle contraction genes were the most significantly differentially expressed category of transcripts when comparing BaP + FL-treated AHR2 morphant and control embryos. These differences were most prominent at 2 and 6 hpe. Therefore, we postulate that BaP + FL may affect cellular Ca(2+) levels and subsequently cardiac muscle function, potentially underlying BaP + FL cardiotoxicity.

Zebrafish cardiotoxicity: the effects of CYP1A inhibition and AHR2 knockdown following exposure to weak aryl hydrocarbon receptor agonists.

The aryl hydrocarbon receptor (AHR) is a ligand-activated transcription factor that mediates many of the toxic effects of dioxin-like compounds (DLCs) and some polycyclic aromatic hydrocarbons (PAHs). Strong AHR agonists, such as certain polychlorinated biphenyls and 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), cause severe cardiac teratogenesis in fish embryos. Moderately strong AHR agonists, such as benzo[a]pyrene and ß-naphthoflavone, have been shown to cause similar cardiotoxic effects when coupled with a cytochrome P450 1A (CYP1A) inhibitor, such as fluoranthene (FL). We sought to determine if weak AHR agonists, when combined with a CYP1A inhibitor (FL) or CYP1A morpholino gene knockdown, are capable of causing cardiac deformities similar to moderately strong AHR agonists (Wassenberg and Di Giulio Environ Health Perspect 112(17):1658-1664, 2004a; Wassenberg and Di Giulio Res 58(2-5):163-168, 2004b; Billiard et al. Toxicol Sci 92(2):526-536, 2006; Van Tiem and Di Giulio Toxicol Appl Pharmacol 254(3):280-287, 2011). The weak AHR agonists included the following: carbaryl, phenanthrene, 2-methylindole, 3-methylindole, indigo, and indirubin. Danio rerio (zebrafish) embryos were first exposed to weak AHR agonists at equimolar concentrations. The agonists were assessed for their relative potency as inducers of CYP1 enzyme activity, measured by the ethoxyresorufin-O-deethylase (EROD) assay, and cardiac deformities. Carbaryl, 2-methylindole, and 3-methylindole induced the highest CYP1A activity in zebrafish. Experiments were then conducted to determine the individual cardiotoxicity of each compound. Next, zebrafish were coexposed to each agonist (at concentrations below those determined to be cardiotoxic) and FL in combination to assess if CYP1A inhibition could induce cardiac deformities. Carbaryl, 2-methylindole, 3-methylindole, and phenanthrene significantly increased pericardial edema relative to controls when combined with FL. To further evaluate the interaction of the weak AHR agonists and CYP1A inhibition, a morpholino was used to knockdown CYP1A expression, and embryos were then exposed to each agonist individually. In embryos exposed to 2-methylindole, CYP1A knockdown caused a similar level of pericardial edema to that caused by exposure to 2-methylindole and FL. The results showed a complex pattern of cardiotoxic response to weak agonist inhibitor exposure and morpholino-knockdown. However, CYP1A knockdown in phenanthrene and 3-methylindole only moderately increased pericardial edema relative to coexposure to FL. AHR2 expression was also knocked down using a morpholino to determine its role in mediating the observed cardiac teratogenesis. Knockdown of AHR2 did not rescue the pericardial edema as previously observed with strong AHR agonists. While some of the cardiotoxicity observed may be attributed to the combination of weak AHR agonism and CYP1A inhibition, other weak AHR agonists appear to be causing cardiotoxicity through an AHR2-independent mechanism. The data show that CYP1A is protective of the cardiac toxicity associated with weak AHR agonists and that knockdown can generate pericardial edema, but these findings are also suggestive of differing mechanisms of cardiac toxicity among known AHR agonists.

Effect-directed analysis of Elizabeth River porewater: developmental toxicity in zebrafish (Danio rerio).

In the present study, effect-directed analysis was used to identify teratogenic compounds in porewater collected from a Superfund site along the Elizabeth River estuary (VA, USA). Zebrafish (Danio rerio) exposed to the porewater displayed acute developmental toxicity and cardiac teratogenesis, presumably because of elevated sediment levels of polycyclic aromatic hydrocarbons (PAHs) from historical creosote use. Pretreatment of porewater with several physical and chemical particle removal methods revealed that colloid-bound chemicals constituted the bulk of the observed toxicity. Size-exclusive chromatography and normal-phase high-performance liquid chromatography were used to fractionate Elizabeth River porewater. Acute toxicity of porewater extracts and extract fractions was assessed as the pericardial area in embryonic zebrafish. The most toxic fraction contained several known aryl hydrocarbon receptor (AhR) agonists (e.g., 1,2-benzofluorene and 1,2-benzanthracene) and cytochrome P450 A1 (CPY1A) inhibitors (e.g., dibenzothiophene and fluoranthene). The second most toxic fraction contained known AhR agonists (e.g., benzo[a]pyrene and indeno[1,2,3-cd]pyrene). Addition of a CYP1A inhibitor, fluoranthene, increased toxicity in all active porewater fractions, suggesting synergism between several contaminants present in porewaters. The results indicate that the observed acute toxicity associated with Elizabeth River porewater results from high concentrations of AhR agonistic PAHs and mixture effects related to interactions between compounds co-occurring at the Elizabeth River site. However, even after extensive fractionation and chemical characterization, it remains plausible that some active compounds in Elizabeth River porewater remain unidentified.

Knockdown of AHR1A but not AHR1B exacerbates PAH and PCB-126 toxicity in zebrafish (Danio rerio) embryos.

Various environmental contaminants are known agonists for the aryl hydrocarbon receptor (AHR), which is highly conserved across vertebrate species. Due to gene duplication events before and after the divergence of ray- and lobe-finned fishes, many teleosts have multiple AHR isoforms. The zebrafish (Danio rerio) has three identified AHRs: AHR1A and AHR1B, the roles of which are not yet well elucidated, and AHR2, which has been shown to mediate the toxicity of various anthropogenic compounds including dioxins, polychlorinated biphenyls (PCBs), and polycyclic aromatic hydrocarbons (PAHs). In this study, we sought to explore the role of the two AHR1 isoforms in PAH- and PCB-induced toxicity in zebrafish embryos utilizing morpholino gene knockdown of the AHR isoforms. Knockdown of AHR1B did not affect the toxicity of PAH mixtures or PCB-126, whereas knockdown of AHR1A exacerbated the cardiac toxicity caused by PAH mixtures and PCB-126. Knockdown of AHR1A did not impact the mRNA expression of CYP1A, CYP1B1, and CYP1C1 in exposed embryos, but it did result in increased CYP1 activity in exposed embryos. As has been shown previously, knockdown of AHR2 resulted in protection from PAH- and PCB-induced cardiac deformities and prevented CYP1 enzyme activity in exposed embryos. Co-knockdown of AHR1A and AHR2 resulted in an intermediate response compared to knockdown of AHR1A and AHR2 individually; co-knockdown did not exacerbate nor protect from PAH-induced deformities and embryos exhibited an intermediate CYP1 enzyme activity response. In contrast, co-knockdown of AHR1A and AHR2 did protect from PCB-126-induced deformities. These results suggest that AHR1A is not a nonfunctional receptor as previously thought and may play a role in the normal physiology of zebrafish during development and/or the toxicity of environmental contaminants in early life stages.

Glutathione transferase pi class 2 (GSTp2) protects against the cardiac deformities caused by exposure to PAHs but not PCB-126 in zebrafish embryos.

Glutathione transferases (GSTs) are phase II enzymes that detoxify a wide range of toxicants and reactive intermediates. One such class of toxicants is the ubiquitous polycyclic aromatic hydrocarbons (PAHs). Certain PAHs are known to cause developmental cardiac toxicity in fish. Herein, we explored the role of GST pi class 2 (GSTp2) in PAH- and PCB-induced cardiac toxicity in zebrafish (Danio rerio) embryos. We measured expression of GSTp2 in embryos exposed to individual and co-exposures of the PAHs benzo[k]fluoranthene (BkF), benzo[a]pyrene (BaP), and fluoranthene (FL) as well as 3,3',4,4',5-pentachlorobiphenyl (PCB-126). GSTp2 mRNA expression was induced by exposure to BkF, BaP, PCB-126, and BaP+FL and BkF+FL co-exposure. A splice junction morpholino was then used to knockdown GSTp2 in developing zebrafish. GSTp2 knockdown exacerbated the toxicity caused by co-exposures to BkF+FL and BaP+FL. However, GSTp2 knockdown did not affect PCB-126 toxicity. These results further suggest that pi class GSTs serve a protective function against the synergistic toxicity caused by PAHs in developing zebrafish.