High-throughput cryo-electron tomography enables multiscale visualization of the inner life of microbes

Abstract

Cryo-electron tomography (cryo-ET) is an advanced and rapidly evolving imaging technique that enables three-dimensional visualization of biological structures in their native state. Although cryo-ET has historically faced significant challenges, including limited applications, tedious data acquisition, labor-intensive image processing, and lower resolution when compared with single particle cryo-electron microscopy (cryo-EM), recent breakthroughs in hardware and software development have significantly improved the entire cryo-ET workflow to enable higher throughput and resolution. These advances have accelerated discoveries in structural and cellular biology, particularly in microbiology, where cryo-ET has unveiled unprecedented insights into the inner life of microbes. This review presents pivotal advances propelling high-throughput cryo-ET and the visualization of microbial architecture. As innovations in imaging technologies, workflow automation, and computational methods continue progressing rapidly, cryo-ET is expected to be increasingly utilized across various fields of life sciences, shaping the future of biological research and biomedical discoveries.

Introduction

Single particle cryo-electron microscopy (cryo-EM) has emerged as the method of choice for determining atomic structures of macromolecules[1]. Originally developed in the 1970s, cryo-EM enabled the cryopreservation and three-dimensional (3D) visualization of frozen-hydrated macromolecules in a thin layer of vitrious ice[2–4]. Nevertheless, single particle cryo-EM was long considered “blobology” due to its limited, nanometer to subnanometer resolution[1]. This limitation was ultimately overcome with recent breakthroughs in hardware and software, termed the “resolution revolution”[5], that enabled cryo-EM to transform the field of structural biology and provide mechanistic insights into macromolecules and their functions at the atomic level [1].

While single particle cryo-EM is primarily used to study purified macromolecules in isolation, cryo-electron tomography (cryo-ET) has become a leading method for revealing macromolecules and their spatial organization within their cellular context [6–8]. This is achieved by mechanically rotating the specimen to capture a series of images of the same object at different tilted angles, typically ranging from −60° to +60°. The tilted images are then computationally reconstructed into a tomogram, providing a detailed 3D view of macromolecules and their native context. Subtomogram averaging (STA) is then used to align and classify multiple copies of the same macromolecules extracted from tomograms to achieve in-situ structures of the macromolecules at resolutions ranging from nanometer to near atomic[9]. Importantly, recent advances in hardware and software have enabled higher throughput and high-resolution visualization of macromolecules within intact cells, particularly in microorganisms (Fig. 1). As microbes possess many complex proteinaceous nanomachines that facilitate survival and adaptation to various environments[10], the high-throughput cryo-ET pipeline has provided unprecedented insights into such complex nanomachines, notably in the characterization of bacterial type III secretion system[11].

Figure 1. Overall workflow of high-throughput cryo-ET.

Sample preparation for cryo-ET involves applying the sample to a glow-discharged EM grid with holey carbon film. After removing excess liquid with filter paper, the grid is plunged into cryogen to form vitrified, electron-transparent ice, preserving the sample for cryo-ET imaging. To achieve detailed structural analysis, the frozen sample is tilted along an axis perpendicular to the electron beam, typically spanning from −60° to +60°. At each tilt angle, a 2D image is captured, providing unique perspectives of the sample. This process enables the collection of around 200 tomograms in a single day, generating a comprehensive dataset for 3D reconstruction and structural analysis. Beam-induced sample motion and misalignment during tilt series acquisition are common challenges that can affect image quality and reconstruction accuracy. To overcome these issues, software tools are used for motion correction and alignment of the tilt series. With automation, these processes can be completed within a single day, significantly improving efficiency and throughput. Aligned tilt stacks can then undergo CTF correction, refining image contrast and resolution to further enhance the accuracy and quality of the final 3D reconstruction. Following the pre-processing of raw data, a 3D tomogram is reconstructed from the 2D projection images using computational algorithms. Widely used software tools such as IMOD [66,80] and EMAN2 [74] facilitate this process, producing tomograms that provide detailed 3D visualizations of the sample. These tomograms can then be used for particle picking, enabling the extraction of individual features of interest for further analysis. Following tomogram reconstruction and particle picking, STA is a pivotal step in structural analysis [74,75]. Extracted subtomograms are subjected to STA, where individual particles are iteratively aligned and averaged in 3D space to improve the signal-to-noise ratio. This approach corrects for variations in particle orientation and accounts for minor structural inconsistencies, resulting in a map that unveils detailed features of the target macromolecular complexes at higher resolution.

Key technological and methodological advances that have facilitated these discoveries can be broadly categorized into three areas: sample preparation, data acquisition, and image processing (Fig. 1). High-throughput data collection systems, driven by automated tilt-series acquisition, have made it feasible to analyze large numbers of samples in a reasonable time frame. In parallel, sophisticated algorithms now enable highly accurate 3D reconstruction, noise reduction, and tomographic data interpretation, revealing previously inaccessible features of microbial cells. This review highlights recent advances in the high-throughput cryo-ET pipeline as well as examines how these breakthroughs support multiscale visualization of microbial architecture, deepening our understanding of the inner life of microbes.

Advances in cryo-ET sample preparation

Preserving the ultrastructure of cells without introducing structural disruption is a fundamental prerequisite for studying protein complexes, a goal uniquely achieved through vitrification, which rapidly immobilizes samples in a near-native state. The most widely used vitrification technique is plunge-freezing[4], which began as a manual method and has since evolved with the advent of semi-automated devices such as the Vitrobot (Thermo Fisher Scientific) and GP2 (Leica), enabling more precise and reproducible cryo-EM sample preperation. In this method, samples are applied to an EM grid covered with a thin holey carbon film and blotted by filter papers to form a thin liquid layer. The EM grid is then plunged into liquid ethane cooled by liquid nitrogen, which ensures ultra-fast cooling rates necessary to achieve vitreous ice without forming crystalline structures[4]. While plunge-freezing is well-suited for thin specimens, such as single cell layers, it is not applicable to thicker biological samples like tissue or multicellular organisms. In such cases, high-pressure freezing (HPF) serves as an alternative vitrification method, allowing the rapid freezing of specimens up to 200 μm thick by applying ~2,000 bar pressure[12].

A longstanding bottleneck in cryo-ET sample preparation is managing the thickness of biological specimens to stay below 400–500 nm[8]. While vitrified samples that exceed this limit do not completely block the electron beam, increased inelastic scattering and degraded image contrast result in hazy images with poor resolution. Consequently, specimens must be substantially thinner for high-resolution imaging. Many thin bacteria are ideal for high resolution cryo-ET imaging. An intact flagellar motor was first visualized in the thin spirochete Treponema primitia by cryo-ET and STA[13]. Subsequently, many in-situ flagellar motor structures from different bacteria such as Helicobacter pylori (Fig. 2A–C) have been determined at higher resolution, revealing conserved core components as well as highly diverse species-specific adaptions[14–18]. Bacterial type IV secretion systems (T4SS) have also been visualized in relatively thin bacteria, including H. pylori[19,20], Legionella pneumophila[21–23], and Coxiella burnetii[24]. However, most bacteria, including Escherichia coli and Salmonella enterica, are too large for high-resoluton cryo-ET imaging. This limitation necessitates a post-vitrification sample thinning step, introducing additional complexity and reducing overall process efficiency. Notably, an innovative strategy to circumvent this thinning step employs bacterial minicells, whose characteristics and applications are well established[25]. Minicells have enabled in-situ structural determination of the receptor arrays in the chemotaxis system of S. enterica[26] and E. coli[27,28] as well as T3SSs in S. enterica[11] and Shigella flexneri[29](Fig. 2D–F). Importantly, cryo-ET imaging of minicells has provided detailed structural views of T3SS-mediated Salmonella-host interactions[30].

Figure 2. High-throughput cryo-ET enables multiscale visualization of bacterial nanomachines in situ.

(A) Schematic representation of Helicobacter pylori movement toward the gastric epithelium via flagella-driven motility, crucial for bacterial colonization, adhesion, and infection. (B) Tomographic slice of H. pylori showing polar flagella. (C) Surface rendering of the H. pylori flagellar motor structure obtained from STA showing key structural components, including the disk, cage, and C-ring (EMD-40405) [16]. (D) Illustration of a Shigella minicell approaching to the host cell via the Type III Secretion System (T3SS), a needle-like apparatus that facilitates the injection of bacterial effector proteins into the host cytoplasm. This interaction is crucial for Shigella pathogenesis, enabling host cell manipulation and intracellular replication. (E) A central slice of a tomographic reconstruction of representative minicell shows secretion machine embedded in the cell envelope. (F) The 3D surface rendering of the T3SS average structure reveals the intricate architecture of the secretion system, highlighting its molecular assembly and spatial organization (EMD-2667) [11].

While bacterial minicells offer a powerful solution for bypassing the thickness limitation in cryo-ET, their application is restricted to specific bacterial species and experimental contexts. Earlier cryo-ultramicrotomy approaches, such as cryo-electron microscopy of vitreous sections (CEMOVIS), enabled the physical sectioning of vitrified samples but were technically demanding and often introduced artifacts[31]. A significant advance in the field was the adaptation of focused ion beam (FIB) milling to cryogenic conditions[32,33]. Combining cryo-FIB milling with cryo-ET (cryo-FIB-ET) enables high-resolution in situ visualization of microbial samples and host-pathogen interactions. For instance, it enabled visualization of endospore-forming bacteria that were previously too thick to image. In Bacillus subtilis, this approach revealed a surprising asymmetry in the localization of cell division proteins (FtsZ-FtsA) between vegetative cells and sporulating cells, providing new, otherwise unattainable insights into the sporulation mechanism[34]. Moreover, cryo-FIB-ET can also be applied to bacteria in biofilms to expose their internal structures for high-resolution imaging[35]. Notably, in the context of host-pathogen interactions, cryo-FIB-ET has been applied to visualize bacterial secretion systems within intracellular contexts, including the T3SS of Yersinia enterocolitica in primary immune cell phagosomes[36], T4SS of L. pneumophila in amoebae [37](Fig. 3A), and T6SS of Amoebophilus asiaticus in amoebae[38]. Cryo-FIB milling has now become the thinning method of choice, particularly in the investigation of intricate host-pathogen interactions.

Figure 3. Cryo-FIB-ET visualization of bacterial secretion machineries during infection or interaction with other bacteria.

For both figures, the left panel represents a schematic illustration, while the right panel displays a tomographic slice. (A) Tomographic slices of cryo-FIB lamellae of intracellular L. pneumophila, demonstrating interaction sites between the bacterial outer membrane and the vacuolar membrane mediating by the Icm/Dot T4SS (white box) in amoeba host [37]. (B) A tomographic slice of Aureispira, a predatory marine bacterium, visualizing bacterial nanomachineries involved in prey capture and killing. The Type VI secretion system (T6SS) is observed in association with extracellular antennas, appearing in two distinct conformations (black arrows): open (extended) and closed (contracted). Grappling hook is shown with brown arrow. Scale bar: 100 nm [42].

Cryo-FIB milling is still challenging due to the slow milling rates associated with low-current Ga+ ion beams used at cryogenic temperatures to avoid sample damage or the introduction of artifacts. These limited ablation rates, while essential for preserving ultrastructure, render lamellae preparation a slow and labor-intensive process. To address the need for efficient lamellae preparation, efforts were initiated to develop an automated FIB milling approach capable of reproducibly generating lamellae on grids while minimizing manual intervention. This advance has been facilitated by software tools such as AutoTEM^™ (Thermo Fisher Scientific) and SmartFIB (Carl Zeiss), which enable semi- or fully automated workflows for lamellae preparation. A pivotal study demonstrated that lamellae prepared using automated cryo-FIB milling with AutoTEM^™ are suitable to resolve the ribosome structure of E. coli at 14 Å[39]. This study demonstrates that automated lamellae preparation can reliably produce samples of sufficient quality for high-resolution structural analysis. To further enhance the throughput of automated cryo-FIB milling, a fully automated sequential pipeline has been developed that involves the stepwise preparation of lamellae across multiple regions of grids, enabling parallelization of the workflow and reducing overall preparation time. This method involves a two-step process: coarse (rough) milling to rapidly remove bulk material, followed by fine polishing to achieve the desired lamellae thickness. The workflow utilizes the software SmartFIB to automate the milling process across multiple targets, incorporating features such as drift compensation to ensure precision. Notably, this approach significantly reduces operator time from approximately 10 hours in manual sessions to about 2.4 hours for an automated sequential milling session[40]. The resulting lamellae were further evaluated using cryo-ET and STA to visualize cyanobacterial septal junctions. The results were then compared to those from a previous study that characterized the same structures using manual cryo-FIB milling[41]. Remarkably, the averaged structure obtained from the automated dataset exhibited a resolution comparable to that of the structure derived from the manual milling study, underscoring the effectiveness and reliability of the automated cryo-FIB preparation workflow[40]. Additionally, this approach, fully automated sequential cryo-FIB, was employed in a separate study to prepare lamellae from plunge-frozen Aureispira-Vibrio mixtures, followed by cryo-ET and STA to visualize ultrastructures involved in ixotrophy, a predatory strategy in bacteria. The analysis revealed that the T9SS mediates prey capture by anchoring to flagellar filaments, functioning like a grappling hook, while the T6SS is responsible for the delivery of effector proteins required to kill prey cells[42](Fig. 3B). This study highlights the effectiveness of automated lamellae preparation for high-resolution structural analysis of complex interbacterial interactions.

Another challenge is the precise identification and localization of specific regions of interest within vitrified samples, particularly when these targets are rare. Fluorescence light microscopy (FLM), when integrated with diverse labeling or tagging strategies, provides a valuable complementary approach for overcoming this limitation. Correlative light and electron microscopy (CLEM)[43] effectively combines the contextual specificity of FLM with the high-resolution structural capabilities of cryo-EM/ET. By enabling the accurate detection of target areas within complex biological systems, CLEM becomes especially critical in studies where the regions of interest are infrequent. Studies have successfully implemented a combination of CLEM and cryo-FIB-ET to investigate bacterial cells[44,45]. Notably, a super-resolution cryo-CLEM workflow integrated with cryo-FIB milling and SEM volume imaging was developed to study the extremophilic bacterium Deinococcus radiodurans. This approach enabled in situ visualization of bacterial unique ultrastructure, including a distinctive diderm cell envelope[46].

In addition to throughput limitations, conventional cryo-FIB milling faces common challenges such as the small size of lamellae, which can constrain the ability to visualize the full architecture of entire cells, and the preferred orientation of specimens on grids, which limits the diversity of views in tomograms. The recently developed Waffle Method[47] addresses both issues. This approach employs HPF to vitrify concentrated, cell-dense specimens directly on EM grids, which populates grid squares with highly concentrated samples to increase the cross-sectional area suitable for milling. Furthermore, by enabling the preparation of lamellae containing numerous randomly oriented bacterial cells, the Waffle Method overcomes the problem of preferred orientation. As a result, it enhances the likelihood of capturing diverse ultrastructural features.

An increasingly pivotal technique in in situ structural analysis is cryo-lift-out, which significantly broadens the range of sample sizes and complexities accessible for cryo-ET, thereby expanding the range of cryo-FIB-ET applications[48–50]. The cryo-lift-out workflow involves first extracting a thick lamellae, typically tens of micrometers thick, from a high-pressure frozen bulk specimen containing the region of interest. This slab is then mounted perpendicularly onto a specialized receiver grid, half-moon TEM grid, for thinning. The resulting electron-transparent lamellae exposes previously inaccessible internal regions. Impressively, a “serial cryo-lift-out” strategy was recently introduced[51] whereby multiple thin sections are extracted sequentially from a single bulk sample to generate several lamellae from one volume. This serial approach increases the yield and preserves more contextual information across the sample. Despite its advantages, cryo-lift-out remains technically challenging and labor intensive, requiring specialized equipment and considerable operator expertise. Moreover, the multiple transfer steps inherent to the workflow increase the risk of sample contamination or devitrification, particularly if the integrity of the cryogenic chain is compromised at any stage.

Altogether, ongoing advances in sample preparation methodologies are directly addressing existing challenges and extending the frontiers of what can be visualized by cryo-ET.

Advances in cryo-ET data acquisition

Beyond advances in sample preparation, developments in cryo-ET data collection and processing have markedly enhanced both efficiency and resolution, further expanding the technique’s applicability and impact. Achieving detailed structural information using cryo-ET often necessitates the collection of a large number of tomograms, sometimes thousands. This requirement arises due to the inherently limited resolution of individual tomograms caused by factors such as electron dose limitations, low signal-to-background noise ratio (SNR), and the missing wedge artifact. Historically, cryo-ET faced significant limitations in data collection speed, with each tilt series taking 20 to 60 minutes[52]. Several cryo-ET software packages, such as SerialEM[53], Leginon[54], and Tomography(Thermo Fisher Scientific), have improved data collection by automating tracking and focusing for each tilt, ensuring the target remains centered. While these advances represent significant progress over manual methods, the process remains relatively slow and continues to pose a challenge for high-throughput studies.

A critical factor influencing both the quality and resolution of cryo-ET data is the tilt scheme: the specific angles at which images are acquired and the order in which those angles are collected. Among the various strategies developed, the dose-symmetric tilt-scheme[55] has gained widespread adoption for its ability to optimize the distribution of electron dose across the tilt series. This method begins near zero degrees and alternates incrementally between positive and negative tilts, thereby minimizing radiation damage at high tilt angles and preserving fine structural details. Importantly, this approach has been shown to bolster the resolution of STA reconstructions by reducing cumulative sample deformation and alignment errors, making it particularly advantageous for high-resolution in situ structural studies. However, accurate alignment and elucidation of structural heterogeneity, crucial for analyzing dynamic complexes or flexible regions, still rely on a substantial number of subtomograms. Consequently, large datasets are essential, driving the need for extensive data collection and contributing to the overall complexity of cryo-ET workflows. In recent years, improved tilt-series acquisition schemes have been introduced to boost throughput. The fast-incremental scheme[52] is an effective strategy to increase tilt-series acquisition speed (a few minutes per series) by omitting autofocus and tracking steps; movie frames are continuously recorded, stopping only to unblank the electron beam at each tilt angle. This approach successfully resolved the structure of the 70S ribosome in E. coli at subnanometer resolution using a dataset of only 12 tomograms[56]. While this approach has demonstrated significant advantages, it has yet to be integrated into widely used data acquisition platforms such as SerialEM[53].

Other advanced approaches have recently been developed to enhance high-throughput tilt series acquisition. These include beam image-shift electron cryo-tomography (BISECT)[57], parallel cryo-electron tomography (PACEtomo)[58], and multishot-tomography[59]. All these approaches employ beam-tilt induced image-shift (BIS), which leverages controlled tilts of the electron beam to shift the imaging field of view rather than mechanically moving the stage, collecting multiple tilt series simultaneously from neighboring regions of the grid. These approaches are designed to minimize or eliminate beam overlap, ensuring efficient and accurate imaging of multiple areas at high resolution[57,60].

Capturing high-resolution signals demands imaging at high magnification, but this naturally comes at the cost of a significantly reduced field of view, restricting it to only a small fraction of the cell. To overcome this challenge, the BIS technique was advanced to support “montage tomography”, whereby a large region is divided into partially overlapping tiles for tilt series acquisition. These tiles are then computationally stitched together to generate a large high-resolution tomogram[60]. Wide-area acquisition serves two key purposes in cryo-ET. First, it enables the visualization of extensive cellular landscapes, preserving the spatial organization and functional context in which molecular structures operate. Second, it increases the number of particles available for STA, which is critical for improving resolution of macromolecular complexes. Importantly, these aims are not mutually exclusive. In fact, the ability to extract high-resolution molecular structures while simultaneously mapping them within their native cellular environment is central to in situ structural biology, offering unique insights into the relationship between structure and cellular context. In other words, this approach reveals the high-resolution structure of a molecule as well as pinpoints its precise location and interactions within the cellular environment, offering a more comprehensive understanding of its biological function.

Advances in cryo-ET data collection have also progressed toward the goal of fully unattended data collection. The steps preceding data acquisition, such as sample screening and target selection, currently demand human effort and expertise. These challenges are being addressed by AI-driven workflows such as SPACEtomo[61], the latest version of PACEtomo, which incorporates deep learning algorithms to fully automate the entire cryo-FIB-ET data collection process. This approach facilitates automated grid navigation to identify and assess the quality of lamellae, identify features of interest, and acquire parallel tilt series, marking a transformative step toward faster, smarter, and more precise data collection. All these advances are contributing to cryo-ET data acquisition by paving the way for more efficient and insightful scientific discoveries.

Advances in cryo-ET data processing

Cryo-ET and cryo-EM are considered ‘big data’ technologies for their ability to produce massive, complex datasets[62]. As a result, the focus is shifting from data acquisition to the more demanding tasks of data management and processing. Cryo-ET data processing in particular is impeded by low SNR, artifacts from the missing wedge, significant storage and computational requirements, and the intrinsic heterogeneity of native biological samples. Traditionally, cryo-ET has faced significant challenges due to the complexity of the processing pipeline, including tilt-series alignment, contrast transfer function estimation, 3D reconstruction, denoising, particle picking, STA, and the management of large datasets. However, breakthroughs in AI, machine learning (ML), and software development are overcoming these obstacles, revolutionizing the efficiency of workflows. Efficient strategies for movie-frame alignment, contrast transfer function determination, and missing wedge restoration[63–66] have become standard approaches implemented in cryo-ET software packages. A significant advance in data processing is the development of AreTomo[67], a software package that automates marker-free tilt-series alignment and tomogram reconstruction. AreTomo is particularly beneficial for lamellae samples where introducing fiducial markers is challenging. Its rapid processing capabilities often allow for preliminary reconstructions to be completed before the acquisition of the next dataset concludes, thereby enhancing throughput in high-resolution cryo-ET workflow.

The inherent noise from low-dose imaging and missing wedge artifacts often obscures structural details. Deep learning-based denoising algorithms such as Topaz-Denoise[68], integrated into the Topaz suite, and IsoNet[69] facilitate clearer interpretation of biological features and improved reliability of downstream analysis.

In the realm of particle picking, ML methods have shown clear advantages over traditional template matching, particularly in single-particle analysis. However, in cryo-ET, where identifying individual particles and performing STA are significantly more challenging, these advantages are less pronounced and remain highly dependent on data quality and specimen characteristics. Tools like TomoTwin[70], Warp[64], and DeepFinder[71] leverage deep learning for particle localization, while emClarity[72] and PyTom[73] continue to utilize template matching approaches. Despite these advances, the performance of these tools varies considerably based on the quality of the dataset and characteristics of the particle of interest, and thus manual intervention is still required to ensure accuracy.

The STA approach, which involves iterative particle alignment, averaging, and classification to achieve structural details from tomograms[74,75], has been successfully implemented in software packages such as Relion[76], emClarity [72], and EMAN2[74]. Addressing structural heterogeneity remains a significant challenge in cryo-ET. A notable advance in this domain is the development of tomoDRGN[77], a deep learning framework specifically designed to tackle this issue. Derived from cryoDRGN[78], originally developed for analyzing structural variability in single-particle cryo-EM, tomoDRGN is capable of learning per-particle structural heterogeneity directly from cryo-ET datasets, while concurrently reconstructing structurally distinct states that are grounded in the underlying tomographic data. Application of this tool enabled the visualization of distinct ribosomal density in E. coli, characterized by varying levels of tRNA occupancy, which are indicative of active translational states[79]

These advances are accelerating structural analyses as well as expanding the boundaries of cryo-ET applications, from studying viral assembly and host-pathogen interactions to unraveling the complexity of cellular architecture.

Future directions

Cryo-ET has emerged as a powerful and multiscale imaging technique, significantly advancing our understanding of macromolecular organization in situ. Improvements in instrumentation and software are positioning cryo-ET as a method of choice for routine visualization of protein complexes within their native cellular environment. One of the most promising areas of development lies in the implementation of fully automated workflows that encompass sample preparation, data acquisition, and computational processing. These efforts aim to improve reproducibility, reduce user bias, and significantly enhance throughput. As automation becomes more robust, high-throughput cryo-ET is expected to become increasingly accessible to a broader range of laboratories, enabling researchers across diverse disciplines to explore the structural and functional intricacies of biological systems more efficiently. Crucially, the future of cryo-ET is closely intertwined with advances in AI and ML. These technologies are being applied to multiple steps in the workflow, improving both the speed and outcome of data processing, accelerating discoveries across the biological sciences. Equally important is the integration of cryo-ET with complementary cutting-edge technologies, which is poised to significantly broaden the biological scope and impact of structural discoveries. One particularly exciting direction is the convergence of cryo-ET with spatially resolved transcriptomics and proteomics, a rapidly evolving frontier in structural cell biology. This integration enables researchers to visualize macromolecular assemblies at near-atomic resolution in situ as well as map their associated gene and protein expression profiles within the same spatial context. Such synergistic approaches promise to transform our understanding of cell biology by bridging gaps among structure, function, and molecular context.