Visualizing Cells and Molecules

Introduction

Cells are far too small to see with the naked eye, and the macromolecules that carry out their functions are thousands of times smaller still. Yet understanding biology depends on seeing — seeing where molecules are, how they are organized, and what they look like at atomic resolution. A remarkable hierarchy of visualization techniques spans the scale from tissue architecture down to individual atoms, and computational tools now extract quantitative information from images at a scale and precision that were unimaginable a generation ago.

This lesson surveys the major techniques for visualizing cells and molecules: light microscopy (including super-resolution), electron microscopy (including cryo-EM), and the structural methods that determine protein structures at atomic resolution. It also covers the computational tools for bioimage analysis and structural bioinformatics.

The Light Microscope Can Resolve Details 0.2 µm Apart

The light microscope resolves details separated by more than about 200 nm (0.2 µm) — the diffraction limit, set by the wavelength of visible light (~400–700 nm). This is sufficient to see individual cells (typically 10–100 µm), nuclei (~5–10 µm), mitochondria (~1 µm), and large intracellular structures, but not individual macromolecules.

Photon noise creates additional limits to resolution, particularly when imaging dim or weakly labeled specimens. At low light levels, the statistical fluctuations in photon counts dominate the image, requiring longer exposures or brighter labels.

Phase Contrast and DIC Microscopy

Living cells are nearly transparent, making them difficult to see in a standard bright-field microscope. Two techniques convert invisible phase differences (caused by differences in cell thickness and refractive index) into visible contrast:

• Phase-contrast microscopy (Zernike, 1953 Nobel Prize) converts phase shifts into brightness differences, revealing cell boundaries, nuclei, and organelles in unstained living cells
• Differential interference contrast (DIC) (Nomarski optics) produces an apparent 3D relief image with no halo artifacts, ideal for thick specimens

Both methods allow live-cell imaging without staining or fixation, preserving normal cell behavior.

Digital Enhancement and Image Analysis

Images can be enhanced and analyzed by digital techniques. Deconvolution computationally removes out-of-focus blur by modeling the optical system’s point spread function, improving contrast and resolution in 3D image stacks. Computational clearing (e.g., background subtraction, noise filtering) further improves image quality.

Time-lapse imaging generates 4D datasets (3D + time) that capture dynamic cellular processes — cell division, migration, organelle transport — over hours to days.

Tissue Preparation: Fixation and Sectioning

Intact tissues are usually fixed (preserved with formaldehyde or glutaraldehyde) and sectioned (cut into thin slices, 5–10 µm for light microscopy, ~70 nm for TEM) before imaging. Sections are mounted on slides and stained with dyes that highlight specific structures:

• Hematoxylin and eosin (H&E) — the standard histological stain; hematoxylin stains nuclei blue, eosin stains cytoplasm pink
• Tissue clearing methods (CLARITY, iDISCO) make intact tissues transparent, enabling 3D fluorescence imaging of whole organs without sectioning

Fluorescence Microscopy

Fluorescence microscopy is the most widely used technique for localizing specific molecules in cells. A fluorophore absorbs light at one wavelength (excitation) and emits it at a longer wavelength (emission). By selecting appropriate filters, only the fluorescence signal is detected against a dark background.

Labeling strategies include:

• Immunofluorescence — fluorophore-tagged antibodies bind target proteins in fixed cells (indirect immunofluorescence uses a primary antibody + fluorescent secondary antibody for signal amplification)
• Fluorescent proteins (GFP and variants: YFP, CFP, mCherry) — genetically encoded fluorophores fused to the protein of interest, enabling live-cell imaging without fixation
• Small-molecule dyes — DAPI (binds DNA), phalloidin-rhodamine (F-actin), MitoTracker (mitochondria), LysoTracker (lysosomes)
• Fluorescent chemical tags (SNAP-tag, HaloTag) — genetically encoded protein tags that covalently bind cell-permeable fluorescent substrates, combining the specificity of genetic encoding with the brightness and photostability of organic dyes

Confocal and Multi-Photon Microscopy

Confocal microscopy uses a pinhole aperture to reject out-of-focus fluorescence, enabling optical sectioning — imaging one thin plane at a time. By collecting a stack of optical sections, a 3D reconstruction of the specimen can be built. Confocal microscopy dramatically improves contrast and resolution in thick specimens.

Two-photon (multi-photon) microscopy uses long-wavelength infrared light; two photons must be absorbed simultaneously to excite the fluorophore. This occurs only at the focal point (where photon density is highest), providing intrinsic optical sectioning. Two-photon microscopy can image deep within living tissue (up to ~1 mm), making it the method of choice for in vivo brain imaging in neuroscience.

Super-Resolution Fluorescence Microscopy

Several techniques break the classical diffraction limit, achieving nanoscale resolution with fluorescence microscopy:

• STED (Stimulated Emission Depletion) — a depletion laser beam with a doughnut-shaped intensity profile switches off fluorophores at the periphery of the excitation spot, effectively shrinking it to ~50 nm
• PALM/STORM (Photoactivated Localization Microscopy / Stochastic Optical Reconstruction Microscopy) — individual fluorophores are switched on stochastically, one at a time, and their positions are determined with ~20 nm precision from their diffraction patterns; the final image is reconstructed from millions of single-molecule localizations
• SIM (Structured Illumination Microscopy) — patterned illumination provides ~2× improvement in resolution (~100 nm) with relatively simple hardware

Super-resolution microscopy has revealed nanoscale details of the nuclear pore complex, synaptic vesicle clusters, chromatin organization, and cytoskeletal architecture that were invisible with conventional fluorescence microscopy.

The Electron Microscope

Transmission electron microscopy (TEM) uses electrons instead of photons. Because electrons have much shorter wavelengths than visible light, TEM resolves details at 1–2 nm in biological specimens. Samples must be fixed (glutaraldehyde, osmium tetroxide), embedded in resin, sectioned (ultrathin, ~70 nm), and stained with heavy metals (uranium, lead) that scatter electrons to produce contrast.

TEM reveals the ultrastructure of cells: membrane bilayers, ribosomes (~25 nm), nuclear pores, endoplasmic reticulum, Golgi cisternae, and cytoskeletal filaments.

Immunogold electron microscopy localizes specific proteins by labeling them with antibodies conjugated to gold nanoparticles (typically 5–15 nm), which appear as electron-dense dots in the TEM image.

Scanning electron microscopy (SEM) scans a focused electron beam across the surface of a specimen coated with a thin metal layer (gold or platinum), producing dramatic three-dimensional images of cell surfaces, tissue architecture, and organism morphology.

Cryo-Electron Microscopy

Cryo-EM has undergone a “resolution revolution” that has transformed structural biology. The method:

1. Protein particles are flash-frozen in vitreous (amorphous) ice on a thin grid — no staining or crystallization needed
2. The electron microscope images thousands of particles in their native hydrated state
3. Computational methods classify the images and reconstruct a 3D density map at near-atomic resolution (2–4 Å)

Cryo-EM excels at determining structures of:

• Large complexes (ribosomes, spliceosomes, proteasomes, ion channels) that are difficult or impossible to crystallize
• Membrane proteins in nanodiscs or detergent micelles, preserving their native lipid environment
• Heterogeneous or flexible assemblies — computational classification can sort different conformational states from a single dataset
• Complexes with therapeutics — cryo-EM enables structure-based drug design for targets that resist crystallization

Electron tomography extends cryo-EM to thicker specimens (cells, organelles): a tilt series of images is collected as the specimen is rotated, and computational reconstruction produces a 3D volume (tomogram) of the cellular interior at ~2–5 nm resolution.

X-Ray Crystallography and NMR

X-ray crystallography remains the most prolific method for atomic-resolution structures. A protein is crystallized, the crystal is bombarded with X-rays, and the resulting diffraction pattern is computationally inverted to produce an electron density map from which the atomic model is built. Resolution depends on crystal quality: the best protein crystals diffract to 1–1.5 Å.

NMR spectroscopy determines structures of proteins in solution (up to ~40 kDa), providing unique information about:

• Protein dynamics — which parts of the structure are rigid and which are flexible
• Conformational changes upon ligand binding
• Weak interactions (e.g., intrinsically disordered regions interacting with partners)

The Protein Data Bank (PDB) is the central repository for experimentally determined 3D structures, containing over 200,000 structures as of 2024, determined by X-ray crystallography (~85%), NMR (~8%), and cryo-EM (~7%, but rapidly growing).

Protein Sequence and Structure Provide Clues About Function

Genome and protein sequences contain an enormous amount of useful information. Sequence comparison identifies related proteins (homologs), suggesting shared function. Domain identification (Pfam, InterPro) recognizes modular functional units. Structure comparison (TM-align, DALI) identifies structural similarities even between proteins with no detectable sequence similarity.

let hemoglobin = "MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH"
let props = Struct.protein_props(hemoglobin)
print(props)

The physical and chemical properties of a protein — molecular weight, isoelectric point, hydrophobicity, charge — are fundamental to choosing the right experimental approach for visualizing it. Proteins that share significant sequence similarity almost always share the same 3D fold and often the same function.

Bioimage Analysis

Modern microscopy generates terabytes of image data. Bioimage analysis uses computational tools to extract quantitative information:

• Cell segmentation — identifying individual cells in images
  • CellProfiler — a versatile open-source pipeline for measuring cell features (size, shape, intensity, texture)
  • StarDist — deep learning-based nuclear segmentation that handles touching nuclei
  • Cellpose — a generalist deep learning model that segments cells across diverse image types without retraining
• Object tracking — following cells or molecules across frames in time-lapse experiments, measuring velocity, division rates, and lineage
• Fluorescence quantification — measuring intensity at specific locations to quantify protein abundance, colocalization, or FRET efficiency
• Machine learning for image classification — convolutional neural networks (CNNs) trained to classify cell phenotypes, identify structures, or detect rare events
• Image registration and stitching — aligning images from multiple fields of view or channels to create large-scale mosaics

Cryo-EM Data Processing

Cryo-EM data analysis involves specialized computational pipelines:

• RELION (REgularized LIkelihood OptimizatioN) — the most widely used software for single-particle cryo-EM, handling particle picking, 2D classification, 3D reconstruction, and iterative refinement using Bayesian statistical methods
• cryoSPARC — combines speed (GPU-accelerated algorithms) with ease of use; particularly efficient for ab initio 3D reconstruction
• Model building — atomic coordinates are fitted into the electron density map using tools like Coot, ISOLDE, and phenix
• Integration with molecular dynamics — MD simulations (using GROMACS or OpenMM) refine and validate cryo-EM structures, revealing conformational flexibility

Structural Bioinformatics

Computational tools extend structural analysis beyond experimental determination:

• AlphaFold Protein Structure Database provides predicted structures for nearly every known protein; ColabFold enables fast predictions on local hardware
• Structure-based drug design uses molecular docking (AutoDock, Glide) to predict how small molecules bind to protein targets
• Molecular dynamics simulations (GROMACS, OpenMM) model protein motion, conformational changes, and ligand binding at atomic detail
• Protein-ligand interaction analysis identifies binding site residues, hydrogen bonds, and hydrophobic contacts
• Protein design tools (RFdiffusion, ProteinMPNN) use deep learning to design novel proteins with desired structures and binding properties
• Binding affinity prediction methods estimate how strongly a protein binds a ligand, guiding drug optimization

Each structural technique operates at a characteristic resolution. Plotting representative structures by their resolution and molecular weight reveals the complementary strengths of each method:

let data = '[{"x": 1.2, "y": 14, "label": "Lysozyme (X-ray)"}, {"x": 1.8, "y": 65, "label": "Hemoglobin (X-ray)"}, {"x": 3.0, "y": 2500, "label": "Ribosome (cryo-EM)"}, {"x": 3.5, "y": 380, "label": "Spliceosome (cryo-EM)"}, {"x": 2.2, "y": 55, "label": "GFP (X-ray)"}, {"x": 4.0, "y": 800, "label": "Ion channel (cryo-EM)"}]'
let plot = Viz.scatter(data, '{"title": "Resolution vs Molecular Weight by Technique", "x_label": "Resolution (Å)", "y_label": "Molecular weight (kDa)", "color": "#8B5CF6"}')
print(plot)

Cryo-EM excels at large complexes that resist crystallization, while X-ray crystallography still achieves the highest resolution for smaller, crystallizable proteins. Structural bioinformatics leverages these complementary approaches to predict structures, identify functional sites, and design drugs.

Exercise: Protein Properties for Structural Characterization

Before choosing a visualization technique, structural biologists analyze a protein’s physical properties. Examine hemoglobin and a membrane protein — which is more hydrophobic, and why does that matter for structure determination?

let soluble = "MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH"
let membrane = "LLILALLAVFAGAILIDPKRRIAQWLAV"
let sol_props = Struct.protein_props(soluble)
let mem_props = Struct.protein_props(membrane)
print("Soluble protein properties:")
print(sol_props)
print("Membrane protein properties:")
print(mem_props)
// Which protein is more hydrophobic and harder to crystallize?
let answer = "membrane"
print(answer)

Exercise: Dimensionality Reduction for Image Feature Analysis

In bioimage analysis, machine learning reduces high-dimensional image features to interpretable components. Use PCA to reduce a dataset of cell measurements (area, intensity, roundness, texture) to two principal components that capture the most variance:

// 6 cells, 4 features each: [area, intensity, roundness, texture]
let data = '[120, 0.85, 0.92, 0.3, 115, 0.82, 0.89, 0.28, 350, 0.45, 0.55, 0.72, 340, 0.42, 0.58, 0.69, 125, 0.80, 0.90, 0.31, 360, 0.40, 0.52, 0.75]'
let result = ML.pca(data, 4, 2)
print("PCA of cell image features:")
print(result)
// How many principal components did we extract?
let answer = "2"
print(answer)

Exercise: Comparing Microscopy Techniques by Resolution

Different microscopy techniques cover different resolution ranges. Build a visualization comparing the resolving power of five techniques to understand which biological structures each can reveal:

let res_data = '[{"label": "Light microscopy", "value": 200}, {"label": "SIM", "value": 100}, {"label": "STED", "value": 50}, {"label": "PALM/STORM", "value": 20}, {"label": "Cryo-EM", "value": 3}]'
let chart = Viz.bar(res_data, '{"title": "Resolution Limits (nm)", "color": "#8B5CF6"}')
print(chart)
// Which fluorescence technique achieves the best resolution?
let answer = "PALM/STORM"
print(answer)

Knowledge Check

Summary

In this lesson you covered visualization techniques for cells and molecules:

• Light microscopy resolves details to ~200 nm; photon noise further limits resolution at low light
• Phase contrast and DIC microscopy visualize live cells without staining
• Digital enhancement (deconvolution) and time-lapse imaging capture dynamic 3D cellular processes
• Tissue preparation (fixation, sectioning, H&E staining) enables histological analysis; tissue clearing allows 3D imaging of whole organs
• Fluorescence microscopy localizes specific molecules using immunofluorescence, fluorescent proteins (GFP), and chemical dyes
• Confocal microscopy provides optical sectioning; two-photon microscopy images deep into living tissue
• Super-resolution methods break the diffraction limit: STED (~50 nm), PALM/STORM (~20 nm), SIM (~100 nm)
• TEM reveals ultrastructure at 1–2 nm; immunogold EM localizes specific proteins; SEM images 3D surfaces
• Cryo-EM determines near-atomic structures (2–4 Å) without crystallization; excels for large complexes, membrane proteins, and flexible assemblies
• Electron tomography produces 3D volumes of cellular interiors
• X-ray crystallography provides atomic-resolution structures; NMR reveals protein dynamics in solution
• The PDB houses >200,000 structures; AlphaFold predicts structures for nearly all known proteins
• Bioimage analysis tools (CellProfiler, StarDist, Cellpose) segment cells and extract quantitative data using deep learning
• Cryo-EM pipelines (RELION, cryoSPARC) process particle images into 3D reconstructions
• Structural bioinformatics: molecular docking (AutoDock, Glide), MD simulations (GROMACS), protein design (RFdiffusion, ProteinMPNN), and binding affinity prediction extend structural analysis computationally