Distinguishing between nectar and non-nectar odors is challenging for animals due to shared compounds and varying ratios in complex mixtures. Changes in nectar production throughout the day and over the animal's lifetime add to the complexity. The honeybee olfactory system, containing fewer than 1000 principal neurons in the early olfactory relay, the antennal lobe (AL), must learn to associate diverse volatile blends with rewards. Previous studies identified plasticity in the AL circuits, but its role in odor learning remains poorly understood. Using a biophysical computational model, tuned by in vivo electrophysiological data, and live imaging of the honeybee's AL, we explored the neural mechanisms of plasticity in the AL. Our findings revealed that when trained with a set of rewarded and unrewarded odors, the AL inhibitory network suppresses responses to shared chemical compounds while enhancing responses to distinct compounds. This results in improved pattern separation and a more concise neural code. Our calcium imaging data support these predictions. Analysis of a graph convolutional neural network performing an odor categorization task revealed a similar mechanism for contrast enhancement. Our study provides insights into how inhibitory plasticity in the early olfactory network reshapes the coding for efficient learning of complex odors.

Mammalian brains handle complex reasoning by integrating information across brain regions specialized for particular sensory modalities. This enables improved robustness and generalization versus deep neural networks, which typically process one modality and are vulnerable to perturbations. While defense methods exist, they do not generalize well across perturbations. We developed a fusion model combining background and foreground features from CNNs trained on Imagenet and Places365. We tested its robustness to human-perceivable perturbations on MS COCO. The fusion model improved robustness, especially for classes with greater context variability. Our proposed solution for integrating multiple modalities provides a new approach to enhance robustness and may be complementary to existing methods.

Odorants binding to olfactory receptor neurons (ORNs) trigger bursts of action potentials, providing the brain with its only experience of the olfactory environment. Our recordings made in vivo from locust ORNs showed that odor-elicited firing patterns comprise four distinct response motifs, each defined by a reliable temporal profile. Different odorants could elicit different response motifs from a given ORN, a property we term motif switching. Further, each motif undergoes its own form of sensory adaptation when activated by repeated plume-like odor pulses. A computational model constrained by our recordings revealed that organizing responses into multiple motifs provides substantial benefits for classifying odors and processing complex odor plumes: each motif contributes uniquely to encode the plume’s composition and structure. Multiple motifs and motif switching further improve odor classification by expanding coding dimensionality. Our model demonstrated that these response features could provide benefits for olfactory navigation, including determining the distance to an odor source.

Animals are constantly bombarded with stimuli, which presents a fundamental problem of sorting among pervasive uninformative stimuli and novel, possibly meaningful stimuli. We evaluated novelty detection behaviorally in honey bees as they position their antennae differentially in an air stream carrying familiar or novel odors. We then characterized neuronal responses to familiar and novel odors in the first synaptic integration center in the brain-the antennal lobes. We found that the neurons that exhibited stronger initial responses to the odor that was to be familiarized are the same units that later distinguish familiar and novel odors, independently of chemical identities. These units, including both tentative projection neurons and local neurons, showed a decreased response to the familiar odor but an increased response to the novel odor. Our results suggest that the antennal lobe may represent familiarity or novelty to an odor stimulus in addition to its chemical identity code. Therefore, the mechanisms for novelty detection may be present in early sensory processing, either as a result of local synaptic interaction or via feedback from higher brain centers.

Mammalian brains handle complex reasoning tasks in a gestalt manner by integrating information from regions of the brain that are specialised to individual sensory modalities. This allows for improved robustness and better generalisation ability. In contrast, deep neural networks are usually designed to process one particular information stream and susceptible to various types of adversarial perturbations. While many methods exist for detecting and defending against adversarial attacks, they do not generalise across a range of attacks and negatively affect performance on clean, unperturbed data. We developed a fusion model using a combination of background and foreground features extracted in parallel from Places-CNN and Imagenet-CNN. We tested the benefits of the fusion approach on preserving adversarial robustness for human perceivable (e.g., Gaussian blur) and network perceivable (e.g., gradient-based) attacks for CIFAR-10 and MS COCO data sets. For gradient based attacks, our results show that fusion allows for significant improvements in classification without decreasing performance on unperturbed data and without need to perform adversarial retraining. Our fused model revealed improvements for Gaussian blur type perturbations as well. The increase in performance from fusion approach depended on the variability of the image contexts; larger increases were seen for classes of images with larger differences in their contexts. We also demonstrate the effect of regularization to bias the classifier decision in the presence of a known adversary. We propose that this biologically inspired approach to integrate information across multiple modalities provides a new way to improve adversarial robustness that can be complementary to current state of the art approaches.

Global climate change has made what were seemingly extraordinary environmental conditions, such as prolonged droughts, commonplace. One consequence of extreme environmental change is concomitant changes in resource abundance. How will such extreme resource changes impact biodiversity? We developed a trait-based consumer-resource model to examine how resource abundance affects the potential for adaptive evolution and coexistence among competitors. We found that moderate changes in resource abundance have little effect on trait evolution. However, when resource scarcities were sufficiently extreme, a critical transition—a tipping point—occurred, which caused consumer traits to diverge and restructured the community in a way that outlasted the scarcity. Therefore, even though traits can evolve in response to minor resource fluctuations, large environmental shifts may be necessary for producing long-lasting impacts on community structure. These results may also help to illuminate patterns of stasis frequently observed in nature, despite the considerable evidence demonstrating rapid evolutionary change.

Adaptation of neural responses is ubiquitous in sensory systems and can potentially facilitate many important computational functions. Here we examined this issue with a well-constrained computational model of the early olfactory circuits. In the insect olfactory system, the responses of olfactory receptor neurons (ORNs) on the antennae adapt over time. We found that strong adaptation of sensory input is important for rapidly detecting a fresher stimulus encountered in the presence of other background cues and for faithfully representing its identity. However, when the overlapping odorants were chemically similar, we found that adaptation could alter the representation of these odorants to emphasize only distinguishing features. This work demonstrates novel roles for peripheral neurons during olfactory processing in complex environments. NEW & NOTEWORTHY Olfactory systems face the problem of distinguishing salient information from a complex olfactory environment. The neural representations of specific odor sources should be consistent regardless of the background. How are olfactory representations robust to varying environmental interference? We show that in locusts the extraction of salient information begins in the periphery. Olfactory receptor neurons adapt in response to odorants. Adaptation can provide a computational mechanism allowing novel odorant components to be highlighted during complex stimuli.

Many tumors are characterized by genetic instability, producing an assortment of genetic variants of tumor cells called subclones. These tumors and their surrounding environments form complex multi-cellular ecosystems, where subclones compete for resources and cooperate to perform multiple tasks, including cancer invasion. Our recent empirical studies revealed existence of such distinct phenotypes of cancer cells, leaders and followers, in lung cancer. These two cellular subclones exchange a complex array of extracellular signals demonstrating a symbiotic relationship at the cellular level. Here, we develop a computational model of the microenvironment of the lung cancer ecosystem to explore how the interactions between subclones can advance or inhibit invasion. We found that, due to the complexity of the ecosystem, invasion may have very different dynamics characterized by the different levels of aggressiveness. By altering the signaling environment, we could alter the ecological relationship between the cell types and the overall ecosystem development. Competition between leader and follower cell populations (defined by the limited amount of resources), positive feedback within the leader cell population (controlled by the focal adhesion kinase and fibronectin signaling), and impact of the follower cells to the leaders (represented by yet undetermined proliferation signal) all had major effects on the outcome of the collective dynamics. Specifically, our analysis revealed a class of tumors (defined by the strengths of fibronectin signaling and competition) that are particularly sensitive to manipulations of the signaling environment. These tumors can undergo irreversible changes to the tumor ecosystem that outlast the manipulations of feedbacks and have a profound impact on invasive potential. Our study predicts a complex division of labor between cancer cell subclones and suggests new treatment strategies targeting signaling within the tumor ecosystem.

In many cancers, such as Chronic Myelogenous Leukemia (CML), pancreatic, and colorectal cancer, long delays exist between the initiation of the disease and the onset of debilitating symptoms. The early stages of these diseases present manageable symptoms and, in the case of CML, highly effective treatment options. Progression to the more aggressive stages of the diseases limits effective treatment and significantly exacerbates patient prognosis. The mechanisms causing delay and disease progression are largely unknown. The later stages of these diseases are characterized by excessive build up of primitive cell types, indicating a disruption in the normal cell differentiation process that is commonly regulated through feedback from differentiated types. In this study, we propose a mechanism where mutated primitive cells produce a feedback interference signal that desensitizes them to a normal homeostatic feedback. Using a mathematical model, we show that this mechanism can account for the long delay period between occurrence of genetic changes and symptomatic onset characterized by fast growth of cancerous cell population. Finally, we explore novel concepts for potential treatment of chronic cancers.

Debate remains on the contributions of niche and neutral processes in structuring biological communities. Temporal variation in the extent to which these two processes may jointly operate makes the problem of resolving their roles even more daunting. Here, we gain insight into this problem by using deterministic and stochastic models of competitors to investigate how the occurrence of rare niches, in what is usually a neutrally structured community, affects species diversity. Rare niches are modeled by allowing each species access to unique resources, which occur with temporal variability. While results from the deterministic model are clear (rare niches provide stable coexistence to otherwise neutral competitors), demographic stochasticity complicates this picture. Stochastic rare niche models show parameter regimes where increases in rare niches actually increase extinction risk by amplifying the variance in population counts. We also use our stochastic model to evaluate the effectiveness of current empirical methods in resolving the difference between rare niche and neutral systems. We find that in many cases, stochastic variation makes niche and neutral systems indistinguishable, allowing for the possibility of niche systems to masquerade as neutral ones. These results highlight the need to better understand how demographic stochasticity and environmental variation can affect the maintenance of species diversity.

BACKGROUND:Interconnected cell signaling pathways are able to efficiently and accurately transmit a multitude of different signals, despite an inherent potential for undesirable levels of cross-talk. To ensure that an appropriate response is produced, biological systems have evolved network-level mechanisms that insulate pathways from crosstalk and prevent 'leaking' or 'spillover' between pathways. Many signaling pathways have been shown to respond in an ultrasensitive (switch-like) fashion to graded input, and this behavior may influence specificity. The relationship of ultrasensitivity to signaling specificity has not been extensively explored. RESULTS:We studied the behavior of simple mathematical models of signaling networks composed of two interconnected pathways that share an intermediate component, asking if the two pathways in the network could exhibit both output specificity (preferentially activate their own output) and input fidelity (preferentially respond to their own input). Previous results with weakly-activated pathways indicated that neither mutual specificity nor mutual fidelity were obtainable in the absence of an insulating mechanism, such as cross-pathway inhibition, combinatorial signaling or scaffolding/compartmentalization. Here we found that mutual specificity is obtainable for hyperbolic or ultrasensitive pathways, even in the absence of an insulating mechanism. However, mutual fidelity is impossible at steady-state, even if pathways are hyperbolic or ultrasensitive. Nevertheless, ultrasensitivity does provide advantages in attaining specificity and fidelity to networks that contain an insulating mechanism. For networks featuring cross-pathway inhibition or combinatorial signaling, ultrasensitive activation can increase specificity in a limited way, and can only be utilized by one of the two pathways. In contrast, for networks featuring scaffolding/compartmentalization, ultrasensitive activation of both pathways can dramatically improve network specificity. CONCLUSIONS:There are constraints to obtaining performance objectives associated with signaling specificity; such constraints may have influenced the evolution of signal transduction networks. Notably, input fidelity (preferential response to an authentic input) is a more difficult objective to achieve than output specificity (preferential targeting to an authentic output). Indeed, mutual fidelity is impossible in the absence of an insulating mechanism, even if pathways are ultrasensitive. Ultrasensitivity does, however, significantly enhance the performance of several insulating mechanisms. In particular, the ultrasensitive activation of both pathways can provide substantial improvement to networks containing scaffolding/compartmentalization.

Neural circuits exploit numerous strategies for encoding information. Although the functional significance of individual coding mechanisms has been investigated, ways in which multiple mechanisms interact and integrate are not well understood. The locust olfactory system, in which dense, transiently synchronized spike trains across ensembles of antenna lobe (AL) neurons are transformed into a sparse representation in the mushroom body (MB; a region associated with memory), provides a well-studied preparation for investigating the interaction of multiple coding mechanisms. Recordings made in vivo from the insect MB demonstrated highly specific responses to odors in Kenyon cells (KCs). Typically, only a few KCs from the recorded population of neurons responded reliably when a specific odor was presented. Different odors induced responses in different KCs. Here, we explored with a biologically plausible model the possibility that a form of plasticity may control and tune synaptic weights of inputs to the mushroom body to ensure the specificity of KCs' responses to familiar or meaningful odors. We found that plasticity at the synapses between the AL and the MB efficiently regulated the delicate tuning necessary to selectively filter the intense AL oscillatory output and condense it to a sparse representation in the MB. Activity-dependent plasticity drove the observed specificity, reliability, and expected persistence of odor representations, suggesting a role for plasticity in information processing and making a testable prediction about synaptic plasticity at AL-MB synapses.

Recordings from the mushroom body demonstrate high specificity in Kenyon cell (KC) responses during odor processing. We explored the possibility that a form of plasticity may control and tune synaptic weights to the mushroom body to provide specificity of KCs responses to familiar or meaningful odors. To test our hypothesis, we compared two potential mechanisms: (1) spike-rate dependent plasticity (SRDP) learning mechanism; and (2) spike-timing dependent plasticity (STDP), both in a network model that included a one-dimensional layer of simplified KCs each receiving N partially overlapping inputs with realistic spike trains. The input to the KCs was characterized by slow temporal patterns and transient synchronization between spike trains in different afferents, as observed experimentally. In the model, mechanisms 1 and 2 yielded similar patterns of spiking in KCs, which usually included a small proportion of odor-specific neurons, independent of the initial (pre-training) synaptic weights to the KCs. However, at high initial weights greater specificity was attained in the network with the STDP-based learning rule. When the networks were trained by a set of multiple odors, the STDP network responded to any single odor from the set with firing patterns that were highly similar to those in the network trained by this odor alone. The SRDP network could reproduce this behavior only partially. The finding indicates that STDP preserves the network s ability to respond with different patterns for different odors from a training set. The study suggests a possible role for synaptic plasticity in enabling efficient decoding of afferent temporal patterns and in favoring sparsening of odor representations, as observed in the locust mushroom body. Support Contributed By: the Swiss NSF, the NIH and the HHMI.