skip to main content

Attention:

The NSF Public Access Repository (NSF-PAR) system and access will be unavailable from 5:00 PM ET until 11:00 PM ET on Friday, June 21 due to maintenance. We apologize for the inconvenience.


Title: Ecoimmunology at spatial scales
Abstract In Focus

Becker, D. J., Albery, G. F., Kessler, M. K., Lunn, T. J., Falvo, C. A., Czirják, G. Á., Martin, L. B., & Plowright, R. K. (2020). Macroimmunology: The drivers and consequences of spatial patterns in wildlife immune defence.Journal of Animal Ecology,89, 972–995. Ecoimmunology seeks to identify and explain natural variation in immune function. Most research so far has focused on differences among individuals within populations, which are often driven by trade‐offs in resource allocation between energetically costly immunity and competing processes such as reproduction. In their review article, Becker et al. (2020) have proposed a framework to explicitly address habitat‐ and population‐level differences in wildlife immune phenotypes. Termed macroimmunology, this concept integrates principles from ecoimmunology and macroecology. Becker et al. (2020) have highlighted three non‐mutually exclusive habitat features that are likely to vary at spatial scales and influence immune function: (a) parasite pressure, (b) abiotic and biotic factors and (c) anthropogenic changes. However, a large and robust body of literature suitable for synthesis to detect macroimmunology patterns and effect sizes is not yet available. Through their systematic review and critical assessment, Becker et al. (2020) identified common problems in existing research that hinders spatial inferences, such as a need for spatial replication in study design and statistical analyses that account for spatial dependence. Overall, macroimmunology has the potential to identify and even predict spatial patterns in immune phenotypes that form the mechanistic underpinnings of important wildlife disease processes, and this review represents an important step to realizing these goals.

 
more » « less
Award ID(s):
1911925
NSF-PAR ID:
10456137
Author(s) / Creator(s):
 ;
Publisher / Repository:
Wiley-Blackwell
Date Published:
Journal Name:
Journal of Animal Ecology
Volume:
89
Issue:
10
ISSN:
0021-8790
Page Range / eLocation ID:
p. 2210-2213
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    The prevalence and intensity of parasites in wild hosts varies across space and is a key determinant of infection risk in humans, domestic animals and threatened wildlife. Because the immune system serves as the primary barrier to infection, replication and transmission following exposure, we here consider the environmental drivers of immunity. Spatial variation in parasite pressure, abiotic and biotic conditions, and anthropogenic factors can all shape immunity across spatial scales. Identifying the most important spatial drivers of immunity could help pre‐empt infectious disease risks, especially in the context of how large‐scale factors such as urbanization affect defence by changing environmental conditions.

    We provide a synthesis of how to apply macroecological approaches to the study of ecoimmunology (i.e. macroimmunology). We first review spatial factors that could generate spatial variation in defence, highlighting the need for large‐scale studies that can differentiate competing environmental predictors of immunity and detailing contexts where this approach might be favoured over small‐scale experimental studies. We next conduct a systematic review of the literature to assess the frequency of spatial studies and to classify them according to taxa, immune measures, spatial replication and extent, and statistical methods.

    We review 210 ecoimmunology studies sampling multiple host populations. We show that whereas spatial approaches are relatively common, spatial replication is generally low and unlikely to provide sufficient environmental variation or power to differentiate competing spatial hypotheses. We also highlight statistical biases in macroimmunology, in that few studies characterize and account for spatial dependence statistically, potentially affecting inferences for the relationships between environmental conditions and immune defence.

    We use these findings to describe tools from geostatistics and spatial modelling that can improve inference about the associations between environmental and immunological variation. In particular, we emphasize exploratory tools that can guide spatial sampling and highlight the need for greater use of mixed‐effects models that account for spatial variability while also allowing researchers to account for both individual‐ and habitat‐level covariates.

    We finally discuss future research priorities for macroimmunology, including focusing on latitudinal gradients, range expansions and urbanization as being especially amenable to large‐scale spatial approaches. Methodologically, we highlight critical opportunities posed by assessing spatial variation in host tolerance, using metagenomics to quantify spatial variation in parasite pressure, coupling large‐scale field studies with small‐scale field experiments and longitudinal approaches, and applying statistical tools from macroecology and meta‐analysis to identify generalizable spatial patterns. Such work will facilitate scaling ecoimmunology from individual‐ to habitat‐level insights about the drivers of immune defence and help predict where environmental change may most alter infectious disease risk.

     
    more » « less
  2. Obeid, Iyad Selesnick (Ed.)
    Electroencephalography (EEG) is a popular clinical monitoring tool used for diagnosing brain-related disorders such as epilepsy [1]. As monitoring EEGs in a critical-care setting is an expensive and tedious task, there is a great interest in developing real-time EEG monitoring tools to improve patient care quality and efficiency [2]. However, clinicians require automatic seizure detection tools that provide decisions with at least 75% sensitivity and less than 1 false alarm (FA) per 24 hours [3]. Some commercial tools recently claim to reach such performance levels, including the Olympic Brainz Monitor [4] and Persyst 14 [5]. In this abstract, we describe our efforts to transform a high-performance offline seizure detection system [3] into a low latency real-time or online seizure detection system. An overview of the system is shown in Figure 1. The main difference between an online versus offline system is that an online system should always be causal and has minimum latency which is often defined by domain experts. The offline system, shown in Figure 2, uses two phases of deep learning models with postprocessing [3]. The channel-based long short term memory (LSTM) model (Phase 1 or P1) processes linear frequency cepstral coefficients (LFCC) [6] features from each EEG channel separately. We use the hypotheses generated by the P1 model and create additional features that carry information about the detected events and their confidence. The P2 model uses these additional features and the LFCC features to learn the temporal and spatial aspects of the EEG signals using a hybrid convolutional neural network (CNN) and LSTM model. Finally, Phase 3 aggregates the results from both P1 and P2 before applying a final postprocessing step. The online system implements Phase 1 by taking advantage of the Linux piping mechanism, multithreading techniques, and multi-core processors. To convert Phase 1 into an online system, we divide the system into five major modules: signal preprocessor, feature extractor, event decoder, postprocessor, and visualizer. The system reads 0.1-second frames from each EEG channel and sends them to the feature extractor and the visualizer. The feature extractor generates LFCC features in real time from the streaming EEG signal. Next, the system computes seizure and background probabilities using a channel-based LSTM model and applies a postprocessor to aggregate the detected events across channels. The system then displays the EEG signal and the decisions simultaneously using a visualization module. The online system uses C++, Python, TensorFlow, and PyQtGraph in its implementation. The online system accepts streamed EEG data sampled at 250 Hz as input. The system begins processing the EEG signal by applying a TCP montage [8]. Depending on the type of the montage, the EEG signal can have either 22 or 20 channels. To enable the online operation, we send 0.1-second (25 samples) length frames from each channel of the streamed EEG signal to the feature extractor and the visualizer. Feature extraction is performed sequentially on each channel. The signal preprocessor writes the sample frames into two streams to facilitate these modules. In the first stream, the feature extractor receives the signals using stdin. In parallel, as a second stream, the visualizer shares a user-defined file with the signal preprocessor. This user-defined file holds raw signal information as a buffer for the visualizer. The signal preprocessor writes into the file while the visualizer reads from it. Reading and writing into the same file poses a challenge. The visualizer can start reading while the signal preprocessor is writing into it. To resolve this issue, we utilize a file locking mechanism in the signal preprocessor and visualizer. Each of the processes temporarily locks the file, performs its operation, releases the lock, and tries to obtain the lock after a waiting period. The file locking mechanism ensures that only one process can access the file by prohibiting other processes from reading or writing while one process is modifying the file [9]. The feature extractor uses circular buffers to save 0.3 seconds or 75 samples from each channel for extracting 0.2-second or 50-sample long center-aligned windows. The module generates 8 absolute LFCC features where the zeroth cepstral coefficient is replaced by a temporal domain energy term. For extracting the rest of the features, three pipelines are used. The differential energy feature is calculated in a 0.9-second absolute feature window with a frame size of 0.1 seconds. The difference between the maximum and minimum temporal energy terms is calculated in this range. Then, the first derivative or the delta features are calculated using another 0.9-second window. Finally, the second derivative or delta-delta features are calculated using a 0.3-second window [6]. The differential energy for the delta-delta features is not included. In total, we extract 26 features from the raw sample windows which add 1.1 seconds of delay to the system. We used the Temple University Hospital Seizure Database (TUSZ) v1.2.1 for developing the online system [10]. The statistics for this dataset are shown in Table 1. A channel-based LSTM model was trained using the features derived from the train set using the online feature extractor module. A window-based normalization technique was applied to those features. In the offline model, we scale features by normalizing using the maximum absolute value of a channel [11] before applying a sliding window approach. Since the online system has access to a limited amount of data, we normalize based on the observed window. The model uses the feature vectors with a frame size of 1 second and a window size of 7 seconds. We evaluated the model using the offline P1 postprocessor to determine the efficacy of the delayed features and the window-based normalization technique. As shown by the results of experiments 1 and 4 in Table 2, these changes give us a comparable performance to the offline model. The online event decoder module utilizes this trained model for computing probabilities for the seizure and background classes. These posteriors are then postprocessed to remove spurious detections. The online postprocessor receives and saves 8 seconds of class posteriors in a buffer for further processing. It applies multiple heuristic filters (e.g., probability threshold) to make an overall decision by combining events across the channels. These filters evaluate the average confidence, the duration of a seizure, and the channels where the seizures were observed. The postprocessor delivers the label and confidence to the visualizer. The visualizer starts to display the signal as soon as it gets access to the signal file, as shown in Figure 1 using the “Signal File” and “Visualizer” blocks. Once the visualizer receives the label and confidence for the latest epoch from the postprocessor, it overlays the decision and color codes that epoch. The visualizer uses red for seizure with the label SEIZ and green for the background class with the label BCKG. Once the streaming finishes, the system saves three files: a signal file in which the sample frames are saved in the order they were streamed, a time segmented event (TSE) file with the overall decisions and confidences, and a hypotheses (HYP) file that saves the label and confidence for each epoch. The user can plot the signal and decisions using the signal and HYP files with only the visualizer by enabling appropriate options. For comparing the performance of different stages of development, we used the test set of TUSZ v1.2.1 database. It contains 1015 EEG records of varying duration. The any-overlap performance [12] of the overall system shown in Figure 2 is 40.29% sensitivity with 5.77 FAs per 24 hours. For comparison, the previous state-of-the-art model developed on this database performed at 30.71% sensitivity with 6.77 FAs per 24 hours [3]. The individual performances of the deep learning phases are as follows: Phase 1’s (P1) performance is 39.46% sensitivity and 11.62 FAs per 24 hours, and Phase 2 detects seizures with 41.16% sensitivity and 11.69 FAs per 24 hours. We trained an LSTM model with the delayed features and the window-based normalization technique for developing the online system. Using the offline decoder and postprocessor, the model performed at 36.23% sensitivity with 9.52 FAs per 24 hours. The trained model was then evaluated with the online modules. The current performance of the overall online system is 45.80% sensitivity with 28.14 FAs per 24 hours. Table 2 summarizes the performances of these systems. The performance of the online system deviates from the offline P1 model because the online postprocessor fails to combine the events as the seizure probability fluctuates during an event. The modules in the online system add a total of 11.1 seconds of delay for processing each second of the data, as shown in Figure 3. In practice, we also count the time for loading the model and starting the visualizer block. When we consider these facts, the system consumes 15 seconds to display the first hypothesis. The system detects seizure onsets with an average latency of 15 seconds. Implementing an automatic seizure detection model in real time is not trivial. We used a variety of techniques such as the file locking mechanism, multithreading, circular buffers, real-time event decoding, and signal-decision plotting to realize the system. A video demonstrating the system is available at: https://www.isip.piconepress.com/projects/nsf_pfi_tt/resources/videos/realtime_eeg_analysis/v2.5.1/video_2.5.1.mp4. The final conference submission will include a more detailed analysis of the online performance of each module. ACKNOWLEDGMENTS Research reported in this publication was most recently supported by the National Science Foundation Partnership for Innovation award number IIP-1827565 and the Pennsylvania Commonwealth Universal Research Enhancement Program (PA CURE). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the official views of any of these organizations. REFERENCES [1] A. Craik, Y. He, and J. L. Contreras-Vidal, “Deep learning for electroencephalogram (EEG) classification tasks: a review,” J. Neural Eng., vol. 16, no. 3, p. 031001, 2019. https://doi.org/10.1088/1741-2552/ab0ab5. [2] A. C. Bridi, T. Q. Louro, and R. C. L. Da Silva, “Clinical Alarms in intensive care: implications of alarm fatigue for the safety of patients,” Rev. Lat. Am. Enfermagem, vol. 22, no. 6, p. 1034, 2014. https://doi.org/10.1590/0104-1169.3488.2513. [3] M. Golmohammadi, V. Shah, I. Obeid, and J. Picone, “Deep Learning Approaches for Automatic Seizure Detection from Scalp Electroencephalograms,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York, New York, USA: Springer, 2020, pp. 233–274. https://doi.org/10.1007/978-3-030-36844-9_8. [4] “CFM Olympic Brainz Monitor.” [Online]. Available: https://newborncare.natus.com/products-services/newborn-care-products/newborn-brain-injury/cfm-olympic-brainz-monitor. [Accessed: 17-Jul-2020]. [5] M. L. Scheuer, S. B. Wilson, A. Antony, G. Ghearing, A. Urban, and A. I. Bagic, “Seizure Detection: Interreader Agreement and Detection Algorithm Assessments Using a Large Dataset,” J. Clin. Neurophysiol., 2020. https://doi.org/10.1097/WNP.0000000000000709. [6] A. Harati, M. Golmohammadi, S. Lopez, I. Obeid, and J. Picone, “Improved EEG Event Classification Using Differential Energy,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium, 2015, pp. 1–4. https://doi.org/10.1109/SPMB.2015.7405421. [7] V. Shah, C. Campbell, I. Obeid, and J. Picone, “Improved Spatio-Temporal Modeling in Automated Seizure Detection using Channel-Dependent Posteriors,” Neurocomputing, 2021. [8] W. Tatum, A. Husain, S. Benbadis, and P. Kaplan, Handbook of EEG Interpretation. New York City, New York, USA: Demos Medical Publishing, 2007. [9] D. P. Bovet and C. Marco, Understanding the Linux Kernel, 3rd ed. O’Reilly Media, Inc., 2005. https://www.oreilly.com/library/view/understanding-the-linux/0596005652/. [10] V. Shah et al., “The Temple University Hospital Seizure Detection Corpus,” Front. Neuroinform., vol. 12, pp. 1–6, 2018. https://doi.org/10.3389/fninf.2018.00083. [11] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. https://dl.acm.org/doi/10.5555/1953048.2078195. [12] J. Gotman, D. Flanagan, J. Zhang, and B. Rosenblatt, “Automatic seizure detection in the newborn: Methods and initial evaluation,” Electroencephalogr. Clin. Neurophysiol., vol. 103, no. 3, pp. 356–362, 1997. https://doi.org/10.1016/S0013-4694(97)00003-9. 
    more » « less
  3. null (Ed.)
    Coastal salt marshes are distributed widely across the globe and are considered essential habitat for many fish and crustacean species. Yet, the literature on fishery support by salt marshes has largely been based on a few geographically distinct model systems, and as a result, inadequately captures the hierarchical nature of salt marsh pattern, process, and variation across space and time. A better understanding of geographic variation and drivers of commonalities and differences across salt marsh systems is essential to informing future management practices. Here, we address the key drivers of geographic variation in salt marshes: hydroperiod, seascape configuration, geomorphology, climatic region, sediment supply and riverine input, salinity, vegetation composition, and human activities. Future efforts to manage, conserve, and restore these habitats will require consideration of how environmental drivers within marshes affect the overall structure and subsequent function for fisheries species. We propose a future research agenda that provides both the consistent collection and reporting of sources of variation in small-scale studies and collaborative networks running parallel studies across large scales and geographically distinct locations to provide analogous information for data poor locations. These comparisons are needed to identify and prioritize restoration or conservation efforts, identify sources of variation among regions, and best manage fisheries and food resources across the globe. Introduction Understanding the drivers of geographic variation in the condition and composition of habitats is crucial to our capacity to generalize management plans across space and time and to clarify and perhaps challenge assumptions of functional equivalence among sites. Broadly defined wetland types such as salt marshes are often assumed to provide similar functions throughout their global range, such as providing nursery habitat for fishery species. However, a growing body of evidence suggests substantial geographic variation in the functioning of salt marsh and other coastal ecosystems (Bradley et al. 2020; Whalen et al. 2020). Variation in ecological patterns and processes within habitat types can alter community structure and dynamics. Local-scale patterns and processes (e.g., patch [10s of meters], local [100s of meters]) can be influenced by processes that occur at larger spatial scales (e.g., regional [kms], global), thereby causing geographic differences in the function and ecosystem service delivery of a given habitat type. Salt marshes (which include vegetated platform, interconnected tidal creeks, fringing mudflats, ponds, and pools) are widely distributed (Fig. 1) and function as valuable nursery habitats by providing key resources for many estuarine species that transition to marine or aquatic habitats as adults (Beck et al. 2001; Minello et al. 2003; Sheaves et al. 2015). However, factors that underlie variability in the delivery of ecological functions are still inadequately understood. Previous studies have explored geographic variation in the function of salt marshes for fish and mobile crustaceans (“nekton”; e.g., Minello et al. 2012, Baker et al. 2013). However, field studies that compare multiple sites across a geographical gradient are typically limited in duration and scale. In addition, the explanatory variables (e.g., elevation, flooding duration, plant structure) collected by smaller scale studies are often inconsistent and therefore limit generalizations across sites. 
    more » « less
  4. ABSTRACT Mercury (Hg) is a widely distributed, toxic heavy metal with no known cellular role. Mercury toxicity has been linked to the production of reactive oxygen species (ROS), but Hg does not directly perform redox chemistry with oxygen. How exposure to the ionic form, Hg(II), generates ROS is unknown. Exposure of Thermus thermophilus to Hg(II) triggered ROS accumulation and increased transcription and activity of superoxide dismutase (Sod) and pseudocatalase (Pcat); however, Hg(II) inactivated Sod and Pcat. Strains lacking Sod or Pcat had increased oxidized bacillithiol (BSH) levels and were more sensitive to Hg(II) than the wild type. The Δ bshA Δ sod and Δ bshA Δ pcat double mutant strains were as sensitive to Hg(II) as the Δ bshA strain that lacks bacillithiol, suggesting that the increased sensitivity to Hg(II) in the Δ sod and Δ pcat mutant strains is due to a decrease of reduced BSH. Treatment of T. thermophilus with Hg(II) decreased aconitase activity and increased the intracellular concentration of free Fe, and these phenotypes were exacerbated in Δ sod and Δ pcat mutant strains. Treatment with Hg(II) also increased DNA damage. We conclude that sequestration of the redox buffering thiol BSH by Hg(II), in conjunction with direct inactivation of ROS-scavenging enzymes, impairs the ability of T. thermophilus to effectively metabolize ROS generated as a normal consequence of growth in aerobic environments. IMPORTANCE Thermus thermophilus is a deep-branching thermophilic aerobe. It is a member of the Deinococcus - Thermus phylum that, together with the Aquificae , constitute the earliest branching aerobic bacterial lineages; therefore, this organism serves as a model for early diverged bacteria (R. K. Hartmann, J. Wolters, B. Kröger, S. Schultze, et al., Syst Appl Microbiol 11:243–249, 1989, https://doi.org/10.1016/S0723-2020(89)80020-7 ) whose natural heated habitat may contain mercury of geological origins (G. G. Geesey, T. Barkay, and S. King, Sci Total Environ 569-570:321–331, 2016, https://doi.org/10.1016/j.scitotenv.2016.06.080 ). T. thermophilus likely arose shortly after the oxidation of the biosphere 2.4 billion years ago. Studying T. thermophilus physiology provides clues about the origin and evolution of mechanisms for mercury and oxidative stress responses, the latter being critical for the survival and function of all extant aerobes. 
    more » « less
  5. Abstract

    An understudied aspect of vertebrate ecoimmunology has been the relative contributions of environmental factors (E), genetic background (G) and their interaction (G × E) in shaping immune development and function. Environmental temperature is known to affect many aspects of immune function and alterations in temperature regimes have been implicated in emergent disease outbreaks, making it a critical environmental factor to study in the context of immune phenotype determinants of wild animals.

    We assessed the relative influences of environmental temperature, genetic background and their interaction on first‐year development of innate and adaptive immune defences of captive‐born garter snakesThamnophis elegansusing a reciprocal transplant laboratory experiment. We used a full‐factorial design with snakes from two divergent life‐history ecotypes, which are known to differ in immune function in their native habitats, raised under conditions mimicking the natural thermal regime—that is, warmer and cooler—of each habitat.

    Genetic background (ecotype) and thermal regime influenced innate and adaptive immune parameters of snakes, but in an immune‐component specific manner. We found some evidence of G × E interactions but no indication of adaptive plasticity with respect to thermal environment. At the individual level, the effects of thermal environment on resource allocation decisions varied between the fast‐ and the slow‐paced life‐history ecotypes. Under warmer conditions, which increased food consumption of individuals in both ecotypes, the former invested mostly in growth, whereas the latter invested more evenly between growth and immune development.

    Overall, immune parameters were highly flexible, but results suggest that other environmental factors are likely more important than temperature per se in driving the ecotype differences in immunity previously documented in the snakes under field conditions. Our results also add to the understanding of investment in immune development and growth during early postnatal life under different thermal environments. Our finding of immune‐component specific patterns strongly cautions against oversimplification of the highly complex immune system in ecoimmunological studies. In conjunction, these results deepen our understanding of the degree of immunological flexibility wild animals present, information that is ever more vital in the context of rapid global environmental change.

     
    more » « less