Analyzing Protein-Protein Spatial-Temporal Dependencies from Image Sequences Using Fuzzy Temporal Random Sets

SummaryFluidity is essential for many biological membrane functions. The basis for understanding membrane structure remains the classic Singer-Nicolson model, in which proteins are embedded within a fluid lipid bilayer and able to diffuse laterally within a sea of lipid. Here we report lipid and protein diffusion in the plasma membrane of live cells of the bacterium Escherichia coli, using Fluorescence Recovery after Photobleaching (FRAP) and Total Internal Reflection Fluorescence (TIRF) microscopy to measure lateral diffusion coefficients. Lipid and protein mobility within the membrane were probed by visualizing an artificial fluorescent lipid and a simple model membrane protein consisting of a single membrane-spanning alpha-helix with a Green Fluorescent Protein (GFP) tag on the cytoplasmic side. The effective viscosity of the lipid bilayer is strongly temperature-dependent, as indicated by changes in the lipid diffusion coefficient. Surprisingly, the mobility of the model protein was unaffected by changes in the effective viscosity of the bulk lipid, and TIRF microscopy indicates that it clusters in segregated, mobile domains. We suggest that this segregation profoundly influences the physical behaviour of the protein in the membrane, with strong implications for bacterial membrane function and bacterial physiology.

The last stage of the constitutive secretory pathway is the delivery and fusion of post-Golgi transport vesicles (carriers) to the plasma membrane. However, it is still unclear where at the plasma membrane these carriers fuse, in a living cell. In this work we used total internal reflection fluorescence microscopy (TIR-FM) to study the delivery and fusion of post-Golgi carriers, as well as of recycling vesicles from the endosomal compartment, to the cell surface in live cells. TIR-FM is suited to image fluorescent molecules near the cell-substrate interphase, since it selectively illuminates the contact surface of cells. One aim of this work was to establish a quantitative method for the microscopic detection of single exocytic fusion events. Following the localization of fusion we answered basic, previously unresolved, questions in the field of membrane traffic. 1\. Imaging fusion of single post-Golgi carriers. A TIR-FM system was optimized to image exocytosis of single post- Golgi carriers. By labeling the carriers with a GFP-tagged membrane protein (vesicular stomatitis virus glycoprotein, VSVG), quantitative criteria for the microscopic detection of single carrier fusion events were established for the first time. Quantitative analysis of time-lapse images could clearly distinguish fusion of the carriers from both movement of carriers relative to the plasma membrane as well as lysis of carriers. The flattening of the carriers into the plasma membrane as well as the subsequent diffusion of the membrane cargo into the plasma membrane was resolved. The duration of the flattening process was found to depend on the size of the carrier, distinguishing small spherical from large tubular carriers. 2\. Role of microtubules in post-Golgi traffic of fibroblasts. The simul-taneous imaging of post-Golgi carriers and microtubules using a novel dual-color TIR-FM system showed that post-Golgi carriers are transported along microtubules to the fusion sites at the plasma membrane. The data strongly suggested that the carriers are capable of undergoing fusion while still attached to the microtubules and that the carriers do not have to reach the end of the microtubules in order to fuse. In contrast to stationary fibroblasts, migrating fibroblasts were shown to have a microtubule-mediated mechanism for polarized insertion of post-Golgi carriers (here LDL-receptor-GFP) close to the leading edge. Disrupting the microtubules restricted this directed delivery of the carriers to regions of the plasma membrane close to the Golgi complex, making the distribution of fusion sites in stationary and migrating cells indistinguishable. Disrupting the microtubules also decreased the overall fusion frequency, increased the frequency of "partial" fusions, and increased the amount of cargo delivered per fusion. We conclude that the microtubule cytoskeleton is necessary for the domain-specific delivery of post-Golgi membrane cargo in fibroblasts. 3\. Role of microtubules in post- Golgi traffic of polarized epithelial cells. Time-lapse fluorescence microscopy was used to analyze the delivery of apical and basolateral membrane proteins to the cell surface in both non-polarized and polarized epithelial cells. We demonstrated that post-Golgi carriers containing either apical or basolateral membrane proteins fuse to the basal membrane in non-polarized cells. Upon polarization, exocytosis of all carriers to the basal membrane was abrogated. Basolateral carriers were seen to fuse to sites at the lateral membrane, while apical carriers presumably fused to the apical membrane. This selective targeting is concomitant with redistribution of the t-SNAREs, syntaxin 3 and 4, upon polarization. Furthermore, we showed that both the targeted exocytosis of apical proteins and the exclusive localization of syntaxin 3 at the apical plasma membrane are dependent on intact microtubules in polarized epithelial cells. In contrast, targeted exocytosis of basolateral proteins and the basolateral distribution of syntaxin 4 and sec6 are maintained independently of microtubules in polarized cells. 4\. Insulin- regulated recycling of glucose transporter. We studied the insulin-regulated release of the glucose transporter (GLUT4) from the endosomal recycling compartment (ERC) in live cells. We show that GLUT4 is retained within the transferrin receptor-containing general ERC in fibroblasts. Using dual-color TIR-FM, we demonstrate that the transferrin receptor and GLUT4 are transported from the ERC in separate vesicles. This provides the first functional evidence for the formation of distinct classes of vesicles from the ERC. We propose that GLUT4 is dynamically retained within the ERC in fibroblasts because it is concentrated in vesicles that form more slowly than those that transport transferrin receptor.

IML-UNet: A brain-inspired spatiotemporal collaborative encoding method for medical image sequence registration.

Force generation by actin assembly shapes cellular membranes. An experimentally constrained multiscale model shows that a minimal branched actin network is sufficient to internalize endocytic pits against membrane tension. Around 200 activated Arp2/3 complexes are required for robust internalization. A newly developed molecule-counting method determined that ~200 Arp2/3 complexes assemble at sites of clathrin-mediated endocytosis in human cells. Simulations predict that actin self-organizes into a radial branched array with growing ends oriented toward the base of the pit. Long actin filaments bend between attachment sites in the coat and the base of the pit. Elastic energy stored in bent filaments, whose presence was confirmed by cryo-electron tomography, contributes to endocytic internalization. Elevated membrane tension directs more growing filaments toward the base of the pit, increasing actin nucleation and bending for increased force production. Thus, spatially constrained actin filament assembly utilizes an adaptive mechanism enabling endocytosis under varying physical constraints.

Background and PurposeNeurosteroids influence neuronal function and have multiple promising clinical applications. Direct modulation of postsynaptic neurotransmitter receptors by neurosteroids is well characterized, but presynaptic effects remain poorly understood. Here, we report presynaptic glutamate release potentiation by neurosteroids pregnanolone and pregnanolone sulfate and compare their mechanisms of action to phorbol 12,13‐dibutyrate (PDBu), a mimic of the second messenger DAG.Experimental ApproachWe use whole‐cell patch‐clamp electrophysiology and pharmacology in rat hippocampal microisland cultures and total internal reflection fluorescence (TIRF) microscopy in HEK293 cells expressing GFP‐tagged vesicle priming protein Munc13‐1, to explore the mechanisms of neurosteroid presynaptic modulation.Key ResultsPregnanolone sulfate and pregnanolone potentiate glutamate release downstream of presynaptic Ca2+ influx, resembling the action of a phorbol ester PDBu. PDBu partially occludes the effect of pregnanolone, but not of pregnanolone sulfate. Calphostin C, an inhibitor that disrupts DAG binding to its targets, reduces the effect PDBu and pregnanolone, but not of pregnanolone sulfate, suggesting that pregnanolone might interact with a well‐known DAG/phorbol ester target Munc13‐1. However, TIRF microscopy experiments found no evidence of pregnanolone‐induced membrane translocation of GFP‐tagged Munc13‐1, suggesting that pregnanolone may regulate Munc13‐1 indirectly or interact with other DAG targets.Conclusion and ImplicationsWe describe a novel presynaptic effect of neurosteroids pregnanolone and pregnanolone sulfate to potentiate glutamate release downstream of presynaptic Ca2+ influx. The mechanism of action of pregnanolone, but not of pregnanolone sulfate, partly overlaps with that of PDBu. Presynaptic effects of neurosteroids may contribute to their therapeutic potential in the treatment of disorders of the glutamate system.

Total internal reflection fluorescence (TIRF) microscopy excites fluorophores within a few hundred nanometers of the sample–substrate interface, enabling high-contrast imaging near the cell membrane. When cultured cells differentiate, the membrane in contact with the coverslip generally acquires basal characteristics, while the opposite membrane develops apical features. Consequently, conventional TIRF microscopy is limited to imaging the basal surface. We developed an immersed-prism TIRF (IP-TIRF) microscope, in which a prism immersed in the culture medium generates TIR at the cell/medium–prism interface, illuminating the apical membrane and reducing cytosolic background. In proof-of-principle experiments, we imaged fluorescent beads and 3xmNeonGreen-tagged intraflagellar transport (IFT) particles in cilia, and compared the performance with confocal microscopy. In cellular regions where both methods can be applied (such as the IFT base pool), on average, IP-TIRF achieved approximately 1.8 times the contrast-to-noise ratio (CNR~31) compared to confocal microscopy. Furthermore, IFT-particle motion was detected in IP-TIRF image sequences and Kymographs of cilia, with adequate spatial resolution. Kymograph analysis revealed an average anterograde IFT velocity of 0.156 ± 0.071 µm/s and an average retrograde velocity of 0.020 ± 0.007 µm/s, approximately one-quarter and one-twentieth, respectively, of the values reported for mammalian primary cilia, which we attribute to acquisition at room temperature rather than physiological conditions. Therefore, these velocity measurements should be regarded as proof-of-principle demonstrations obtained at room temperature, not as validated physiological transport rates. Our IP-TIRF method provides a high-resolution, cost-effective, and broadly accessible approach for imaging the apical membrane in live cells.

The biogeoscience community has increasingly embraced the application of machine learning models across various domains from fire prediction to vegetation forecasting. Yet, as these models become more widely used, there is sometimes a gap in understanding between what we assume the model learns and what the model actually learns. For example, Long-short Term Memory (LSTM) models are applied to long time series, hoping they benefit from access to more information, despite their tendency to rapidly forget information. This can lead to erroneous conclusions, misinterpretation of results, and an overestimation of the models, ultimately eroding trust in their reliability.  To address this issue, we employ an explainable artificial intelligence (XAI) post hoc perturbation technique that is task-agnostic and model-agnostic. We aim to examine the extent to which the model leverages information for its predictions, both in terms of time and space. In other words, we want to observe the actual receptive field utilized by the model. We introduce a methodology designed to quantify both the spatial impact of neighboring pixels on predicting a specific pixel and the temporal periods contributing to predictions in time series models. The experiments take place after training the model, during inference. In the spatial domain, we define ground-truth pixels to predict, then examine the increase in prediction error, caused by shuffling their neighboring pixels at various distances from the selection. In the temporal domain, we investigate how shuffling a sequence of frames within the context period at different intervals relative to the target period affects the increase in prediction loss. This method can be applied across a broad spectrum of spatio-temporal tasks. Importantly, the method is easy-to-implement, as it only relies on the inference of predictions at test time and the shuffling of the perturbation area.  For our experiments, we focus on the vegetation forecasting task, i.e., forecasting the evolution of the Vegetation Index (VI) based on Sentinel-2 imagery using previous Sentinel-2 sequences and weather information to guide the prediction. This task involves both spatial non-linear dependencies arising from the spatial context (e.g., the surrounding area, such as a river or a slope, directly influencing the VI) and non-linear temporal dependencies such as the gradual onset of drought conditions and the rapid influence of precipitation events. We compare several models for spatio-temporal tasks, including ConvLSTM and transformer-based architectures on their usage of neighboring pixels in space, and context period in time. We demonstrate that the ConvLSTM relies on a  restricted spatial area in its predictions, indicating a limited utilization of the spatial context up to 50m (5 pixels). Furthermore, it utilizes the global order of the time series sequence to capture the seasonal cycle but loses sensitivity to the local order after 15 days (3 frames). The introduced XAI method allows us to quantify spatial and temporal behavior exhibited by machine learning methods.

Proper membrane localization of ion channels is essential for the function of neuronal cells. Particularly, the computational ability of dendrites depends on the localization of different ion channels in specific subcompartments. However, the molecular mechanisms that control ion channel localization in distinct dendritic subcompartments are largely unknown. Here, we developed a quantitative live cell imaging method to analyze protein sorting and post-Golgi vesicular trafficking. We focused on two dendritic voltage-gated potassium channels that exhibit distinct localizations: Kv2.1 in proximal dendrites and Kv4.2 in distal dendrites. Our results show that Kv2.1 and Kv4.2 channels are sorted into two distinct populations of vesicles at the Golgi apparatus. The targeting of Kv2.1 and Kv4.2 vesicles occurred by distinct mechanisms as evidenced by their requirement for specific peptide motifs, cytoskeletal elements, and motor proteins. By live cell and super-resolution imaging, we identified a novel trafficking machinery important for the localization of Kv2.1 channels. Particularly, we identified non-muscle myosin II as an important factor in Kv2.1 trafficking. These findings reveal that the sorting of ion channels at the Golgi apparatus and their subsequent trafficking by unique molecular mechanisms are crucial for their specific localizations within dendrites.

The Detection Transformer (DETR) has revolutionized the design of CNN-based object detection systems, showcasing impressive performance. However, its potential in the domain of multi-frame 3D object detection remains largely unexplored. In this paper, we present STEMD, a novel end-to-end framework that enhances the DETR-like paradigm for multi-frame 3D object detection by addressing three key aspects specifically tailored for this task. First, to model the inter-object spatial interaction and complex temporal dependencies, we introduce the spatial-temporal graph attention network, which represents queries as nodes in a graph and enables effective modeling of object interactions within a social context. To solve the problem of missing hard cases in the proposed output of the encoder in the current frame, we incorporate the output of the previous frame to initialize the query input of the decoder. Finally, it poses a challenge for the network to distinguish between the positive query and other highly similar queries that are not the best match. And similar queries are insufficiently suppressed and turn into redundant prediction boxes. To address this issue, our proposed IoU regularization term encourages similar queries to be distinct during the refinement. Through extensive experiments, we demonstrate the effectiveness of our approach in handling challenging scenarios, while incurring only a minor additional computational overhead.

It is well recognized that air quality inference is of great importance for environmental protection. However, due to the limited monitoring stations and various impact factors, e.g., meteorology, traffic volume and human mobility, inference of air quality index (AQI) could be a difficult task. Recently, with the development of new ways for collecting and integrating urban, mobile, and public service data, there is a potential to leverage spatial relatedness and temporal dependencies for better AQI estimation. To that end, in this paper, we exploit a novel spatio-temporal multi-task learning strategy and develop an enhanced framework for AQI inference. Specifically, both time dependence within a single monitoring station, and spatial relatedness across all the stations will be captured, and then well trained with effective optimization to support AQI inference tasks. As air-quality related features from cross-domain data have been extracted and quantified, comprehensive experiments based on real-world datasets validate the effectiveness of our proposed framework with significant margin compared with several state-of-the-art baselines, which support the hypothesis that our spatio-temporal multi-task learning framework could better predict and interpret AQI fluctuation.

Spatiotemporal modeling of the expansion of mechanized agriculture in the Bolivian lowland forests

This study proposes a hybrid sea level prediction model by coupling a dynamically optimized variational mode decomposition (VMD) with a convolutional bidirectional gated recurrent unit (CNN-BiGRU). The VMD decomposition is fine-tuned using the grey wolf optimizer and evaluated via entropy criteria to minimize mode mixing. The resulting components are processed by CNN-BiGRU to capture spatial features and temporal dependencies, and predictions are reconstructed from the integrated outputs. Validated on monthly sea level data from Kanmen and Zhapo stations, the model achieves high accuracy with an RMSE of 13.857 mm and 16.230 mm, MAE of 10.659 mm and 13.129 mm, and NSE of 0.986 and 0.980. With a 6-month time step, the proposed strategy effectively captures both periodic and trend signals, demonstrating strong dynamic tracking and error convergence. It significantly improves prediction accuracy and provides reliable support for storm surge warning and coastal management.

Human trajectory prediction is a crucial yet challenging problem, which is of fundamental importance to robotics and autonomous driving vehicles. The core challenge lies in effectively modeling the socially aware spatial interaction and complex temporal dependencies among crowds. However, previous methods usually model spatial and temporal information separately and only use individual features, while ignoring group-level motion characteristics. We propose a novel trajectory prediction framework termed GA-STT, a group aware spatial-temporal transformer network to address these issues. Specifically, we first get the individual representations supervised by group-based annotations. Then, we model the complex spatial-temporal interactions with spatial and temporal transformers separately and fuse spatial-temporal embedding through the cross-attention mechanism. Results on the publicly available ETH/UCY datasets show that our model outperforms the state-of-the-art method by 19.4% in average displacement error(ADE) and 16.9% in final displacement error(FDE) and successfully predicts complex spatial-temporal interactions.

Climate variability and shifting atmospheric patterns have emerged as critical determinants of vector-borne and waterborne disease incidence across tropical and subtropical regions. This paper presents a comprehensive machine learning- based early warning framework that integrates multi-source climate parameters—including temperature, relative humidity, rainfall intensity, wind velocity, and atmospheric pressure gradients—with epidemiological surveillance records to forecast disease outbreak risk at district and sub-district levels. The proposed system employs an ensemble architecture combining Random Forest, Gradient Boosting Machine (GBM), Long Short-Term Memory (LSTM) networks, and an Attention-Enhanced Convolutional Neural Network (ACNN) to capture both spatial heterogeneity and temporal dependencies inherent in climate- disease relationships. Training data spans seventeen years of merged meteorological observations sourced from the India Meteorological Department (IMD) and ISRO's MOSDAC satellite telemetry, cross-referenced with disease incidence records from the Integrated Disease Surveillance Programme (IDSP). Experimental results demonstrate an overall outbreak prediction accuracy of 93.7%, area under the receiver operating characteristic curve (AUC-ROC) of 0.96, and a mean absolute error of1.4 cases per 10,000 population for weekly incidence estimation. The framework incorporates an automated alert generation module that classifies health risk into four tiers—green, yellow, orange, and red—enabling district health officers to pre- position medical resources and initiate preventive interventions up to three weeks in advance of a projected epidemic threshold crossing. This work contributes a replicable, data-driven methodology applicable to resource-constrained public health systems in developing nations.

In the context of intelligent power system construction, power grid operational data exhibits highdimensional, non-linear, and strong stochastic fluctuation characteristics. To address the urgent demand for high-precision signal feature extraction, data privacy, and cross-domain collaborative analysis in complex environments, Federated MetaReinforcement Learning (FMRL) has emerged as an advanced paradigm for solving heterogeneous data analysis challenges. This paper constructs a signal feature extraction and fluctuation analysis framework based on FMRL for scenarios involving multiple grid operators. Distinct from traditional methods that rely on fixed optimization strategies, we formulate the crossdomain training process as a bilevel optimization problem. We define the optimization objective (i.e., the reward function) as maximizing the validation accuracy improvement of the signal reconstruction task. Accordingly, a Meta-RL Agent is designed to act as a hyper-parameter controller, which dynamically adjusts the local learning rate by perceiving the training status (State), thereby enabling rapid adaptation to Non-IID (Non-Identically and Independently Distributed) data distributions. The framework utilizes a hybrid ResNet-BiLSTM neural network as the base feature extractor, where ResNet captures spatial structural features and BiLSTM extracts long-term temporal dependencies. Targeting communication security, we propose a privacy-preserving algorithm based on meta-policy aggregation and Homomorphic Encryption (HE). Through a secure metapolicy update mechanism on the central server, efficient convergence and knowledge fusion are achieved without local raw signals leaving their respective domains. Simulation results demonstrate that, compared to state-of-the-art federated metalearning benchmarks, the proposed FMRL algorithm significantly improves signal fluctuation fitting ability and cross-domain adaptability, while effectively guaranteeing data privacy and model security.