# Microscopy and Quantitative Images

> Why can an ordinary light microscope not separate two vesicles 150 nm apart, however much you enlarge the picture? This lesson derives the diffraction limit, works the numbers for a real objective, and compares confocal, electron and super-resolution microscopes by what each one shows and what it costs. Afterwards you can compute a resolution and a pixel size, and read a fluorescence image the way a careful reviewer would: asking what was labelled, which controls were done, and what an overlap of two colours can and cannot mean.

Course: Cell Biology (Seeing and testing cells). Level: core university (third year). About 18 minutes.
Lesson page (link this one when citing): https://stemhop.com/lessons/microscopy-and-images/

## What you'll learn

- Define resolution as the smallest separation at which two point objects can be told apart, explain why diffraction sets it, and compute it with d = 0.61λ/NA (lateral) and dz ≈ 2λn/NA² (axial).
- Explain why magnification without resolution adds nothing, and apply the rule of at least two camera pixels per resolved distance.
- Compare widefield, confocal, TEM, SEM, cryo-EM, STED, SIM and PALM/STORM by what each shows, what it costs, and its typical artifacts.
- Judge a fluorescence image by its labelling controls, matched settings, saturation, background and noise, and state what colocalization can and cannot show.

The first film showed why two 150 nm sources become one recorded peak; now put numbers on that picture. The rest of the page tests camera sampling, methods that change the measurement, and what an image can support; each follows what enters the record.

## Put numbers on the blur

Here λ is the label’s fluorescence **emission wavelength**, and **numerical aperture** (NA) describes the cone of light the objective collects. **Resolution** is the smallest separation at which two point objects can be distinguished under a stated criterion. The Rayleigh criterion calls them resolved when one Airy pattern’s centre falls on the other’s first dark ring.

Use the ideal widefield setup behind the first film’s 227 nm figure: GFP-like emission at λ = 520 nm, an oil objective with NA 1.40, and 100 nm of specimen per camera pixel. These are exercise values, not a measured **point-spread function** (the instrument’s response to a point); aberrations and thick samples can do worse.

**Step 1, lateral resolution.**

> d = 0.61 × 520 / 1.40 = 317.2 / 1.40 = 226.6 nm, about 0.23 µm

The vesicles are 150 nm apart, below 226.6 nm, so they are unresolved by this criterion.

**Step 2, camera sampling.** Optics can preserve detail that a camera then samples too coarsely. A useful introductory rule is at least two camera pixels across one resolved distance.

**Predict.** Which camera samples d = 226.6 nm adequately by that rule: 100 nm specimen pixels, 150 nm specimen pixels, or both?

> **Check your prediction.** The 100 nm camera gives 226.6 / 100 = 2.27 pixels across d, so it just meets the rule. The 150 nm camera gives only 1.51 pixels across d, so the camera loses detail that the optics could pass.

**Step 3, change the objective.** With the same 520 nm emission but NA 1.00, d = 317.2 nm. Losing 0.4 in NA costs about 90 nm of lateral resolution in this model.

**Step 4, look along the optical axis (the depth direction).** The first film used the simple lateral case; here is the fuller three-dimensional one. With immersion index n = 1.515, the approximate axial estimate is

> dz ≈ 2λn/NA² = 2 × 520 × 1.515 / 1.40² = 804 nm

That is about 3.5 times the lateral value. A point’s three-dimensional blur is therefore elongated along the optical axis.

*Interactive on the lesson page:* Resolution and sampling: two points through a microscope

## Change the measurement, not the zoom

**Reminder:** in the scoped widefield model, λ and numerical aperture (NA) set the Rayleigh distance. Once the camera samples finely enough, further magnification and software zoom only enlarge the record: magnification without resolution is empty magnification. To do better, change the measurement.

### Remove out-of-focus haze

Widefield collects fluorescence throughout the illuminated depth, producing haze above and below focus in thick, densely labelled specimens. A **confocal microscope** scans a focused spot; a detector pinhole blocks most out-of-focus light. Each scan is an optical section, and a stack can reconstruct a three-dimensional cell, one way to measure organelle volumes. The [previous lesson’s volumes](https://stemhop.com/lessons/measured-eukaryotic-cell/) came from other methods, including areas in electron-microscope sections.

**Predict.** If a smaller confocal pinhole rejects more out-of-focus light, why not close it as far as possible?

> **Check your prediction.** Below one Airy unit—a pinhole as wide as the Airy pattern’s central disc—a smaller opening can buy modest resolution but costs signal. The main gain is optical sectioning and contrast; a conventional confocal remains diffraction limited, unable to separate arbitrarily close points.

### Use electrons for finer structural detail

**[Electron microscopes](https://stemhop.com/methods/electron-microscopy/)** focus an electron beam with magnetic lenses. Transmission EM (TEM) sends it through a thin specimen. Conventionally, cells are fixed, dehydrated, plastic-embedded, cut into 50 to 100 nm sections and stained with heavy metals. TEM can show membranes and organelles, but its image is a two-dimensional projection of dead, prepared tissue; fixation can create artifacts. Scanning EM (SEM) rasters a coated surface to show its shape.

*Figure (AI-generated illustration).* A thin-section electron micrograph is a two-dimensional projection of dead, prepared tissue, so a structure that fixation could have made must be checked with a second method. For conventional transmission EM, cells are chemically fixed, dehydrated, embedded in plastic, cut into sections about 50 to 100 nm thick and stained with heavy metals. Image: a four-step strip showing a cell being fixed, dehydrated and embedded in a plastic block, cut with a knife into a very thin section, and stained on a grid with an electron beam passing through it, with a dashed folded membrane pocket marked as a possible artifact.

In **[cryo-EM](https://stemhop.com/methods/cryo-em/)**, rapid freezing traps water as glass-like vitreous ice rather than crystals. Most near-atomic cryo-EM structures come from averaging many images of purified particles. Thinned frozen cells show molecules in place, usually at lower resolution. Averaging many copies of an abundant complex such as the ribosome has produced near-atomic structures inside human cells, but not an atom-by-atom whole cell.

### Encode close fluorescent emitters

[Super-resolution](https://stemhop.com/methods/super-resolution-microscopy/) changes what reaches the record without making diffraction disappear.

- **STED** uses a ring-shaped depletion beam to force excited dyes around the focus back to the dark state by stimulated emission, leaving a smaller centre able to fluoresce.
- **SIM** uses striped illumination to shift finer detail into the range the objective can pass, then computes an image.
- **PALM and STORM** switch on sparse fluorophores so nearby molecules do not glow together, estimate each isolated blur’s centre, and combine positions from thousands of cycles.

*Figure (AI-generated illustration).* STED beats the blur by controlling which molecules glow, not by shrinking the beams: diffraction still sets their size. A ring-shaped beam forces excited dye molecules back to the dark state by stimulated emission, leaving only a small centre fluorescent. Image: a top-down field of dye molecules with a large excitation spot, a ring-shaped depletion beam laid over its outer part, and a small bright gold centre where molecules are still free to glow; an inset compares the larger excitation-only spot with the smaller spot recorded with the ring beam.

**Predict.** One isolated molecule makes a broad blur, but its centre is estimated to 10 nm. Does that mean two molecules 50 nm apart are resolved when they glow together?

> **Check your prediction.** No. The 10 nm value is **localization precision**, the uncertainty in one isolated centre, not resolution between simultaneous emitters. PALM and STORM separate neighbours in time. Final resolution also depends on label density and size, drift and blinking; one fluorescent protein may be counted several times. With enough photons, an isolated centre can be found to a few nanometres, and shot-noise-limited uncertainty falls as 1/√N for N detected photons.

Each of these methods still ends in a record that someone must interpret.

The second film picks up here: what can two overlapping colours in that record claim?

**Film transcript: What an Overlap Can Claim**

### Film transcript: What an Overlap Can Claim

Red and green image channels overlap. Does the yellow patch mean the two proteins bind? Chapter one showed why close objects merge into one spot. Now follow the evidence from label to claim. Start with the blur from chapter one.

Across the image, it is about 230 nanometres wide. Along the depth direction, it stretches to about 800 nanometres. A yellow pixel alone cannot show that the two sources are closer than this scale. These dimensions use chapter one's green-light settings. A red channel usually has a longer emission wavelength and a larger blur. For a two-colour claim, use the measured resolution of each channel. Here is a question to hold. At typical cellular protein density, how many protein molecules would this volume contain? A handful, a few hundred, or tens of thousands?

Treat the blur as an ellipsoid. Its volume is roughly 0.02 cubic micrometres. At typical cellular protein density, it contains tens of thousands of protein molecules. So, after the controls, an in-focus overlap can support proximity within about this volume. It does not say the two proteins touch. In widefield, out-of-focus haze can create apparent overlap from farther away. Even that weaker claim needs evidence. Colour overlap is a claim about proximity. It is only as strong as each link that produced it. Check three links in turn.

First, the label. An antibody can bind the wrong protein. Signal loss after knockout supports specificity. With knockdown, the signal should fall with verified target reduction. A secondary-only sample checks nonspecific signal from the fluorescent secondary. Second, the channels. A dye's light can leak into another colour channel. That leak is called bleed-through. Single-label samples measure it, and multicolour beads test whether the channels line up. Misregistration can separate matching signals or superimpose different ones. Third, the record.

Image every condition with identical settings. Keep the relevant pixels below the detector maximum. Background subtraction removes the average background, but not its photon noise. Now return to yellow. If every link holds, an in-focus overlap supports proximity at roughly one resolution volume.

Binding is a separate claim, and it needs other experiments. Co-immunoprecipitation supports membership in the same complex, but not direct contact. A test with purified proteins addresses direct interaction. So read a colour overlap the way chapter one read the blur. Ask what was labelled, what leaked between channels, and what the detector recorded. The overlap reports what the measurement could separate, and nothing finer. Overlap supports proximity, not binding.

From the film

- A controlled, in-focus colour overlap supports proximity at roughly one resolution volume.
- In the green-light exercise, that volume is about 230 nm across and 800 nm deep, containing tens of thousands of proteins at typical cellular density.
- Overlap means something only if the label, the channels and the record each pass their controls.
- Colour overlap supports proximity, not binding; binding needs other experiments.

## Read the image as a claim

The second film named the controls for each link from label to claim; here are the cases it did not work through.

### 1. Is the signal the claimed target?

An antibody can bind the wrong protein (see [protein purification](https://stemhop.com/lessons/protein-purification-techniques/)). Signal loss after target knockout supports specificity; with knockdown, the signal should fall with independently verified target reduction. A secondary-only sample checks nonspecific signal from the fluorescent secondary. A genetic tag such as GFP avoids antibodies but can alter protein location ([hacking the cell](https://stemhop.com/lessons/hacking-the-cell/) and [detecting a signal](https://stemhop.com/lessons/detecting-a-signal/)).

**Predict.** An antibody still gives the same mitochondrial pattern after its target gene is deleted, although a blot detects no target. The secondary-only sample is dark. What failed?

> **Check your prediction.** The primary antibody is detecting something other than the target at mitochondria. The dark secondary-only control rules out the secondary antibody, but it cannot rescue the primary antibody’s specificity.

### 2. Are the channels separate?

Single-label samples measure **bleed-through** (one dye’s light leaking into another channel), and multicolour beads test whether the channels are aligned, or **registered**.

### 3. Are intensities comparable?

The second film gave the rules: identical acquisition, pixels below the detector maximum, and background subtraction that leaves photon noise. The hard case is saturation.

**Predict.** Treated cells contain many pixels at the detector maximum. Can their mean fluorescence be compared quantitatively with unsaturated control cells?

> **Check your prediction.** No. A saturated pixel has stopped increasing with the light that reaches it, so its value no longer reports how much label is present. Background subtraction cannot restore clipped information.

### 4. What does overlap support?

After the label, channel and acquisition controls pass, an in-focus two-colour overlap can support proximity at roughly one resolution volume; widefield haze can create apparent overlap from farther away. For the first film’s green-light exercise, that volume is about 230 nm laterally by 800 nm axially. A red channel usually has a longer emission wavelength and a larger blur, so a real two-colour analysis needs the measured resolution of each channel. Treating the green-light volume as an ellipsoid gives roughly 0.02 µm³, which contains tens of thousands of protein molecules at typical cellular protein density. Colocalization supports proximity, not binding. The design problem asks what it would take to claim more.

### Design problem

A paper shows an antibody to protein P staining mitochondria. Yellow overlap with a mitochondrial marker is offered as proof that P binds the marker. What would you require before accepting either claim?

> **One way to do it.** First test the stain. In P knockout or knockdown cells imaged identically, mitochondrial signal should fall with P on a blot. Add a secondary-only control and independent placement: tagged P checked for function or biochemical fractionation placing P in mitochondria.
>
> Then test proximity with single-label samples, registration beads and unsaturated raw images, so any overlap coefficient uses real intensities.
>
> Passing those tests places the proteins only within one resolution volume (about 230 nm laterally by 800 nm axially here). Co-immunoprecipitation supports the same complex, and a proximity method supports close approach; neither proves direct contact. Purified-protein binding tests direct interaction, and interaction-breaking mutations should remove binding without changing protein amount or location.

## Going deeper

The optical assumptions, sampling convention and experimental exceptions behind the main line.

### Where NA and the Airy pattern come from

Light diffracts through the objective’s circular opening, so one point forms a bright disc with faint rings, an **Airy pattern**. Two independent emitters produce patterns whose intensities add. Ernst [Abbe](https://stemhop.com/methods/light-microscopy/)’s 1873 theory described fine detail sending light to wider angles; an objective can rebuild only the detail carried by light it collects.

Numerical aperture is

> NA = n × sin θ

Here n is the refractive index between sample and lens, and θ is half the collected cone angle. In air n = 1.00, so NA cannot exceed 1. Immersion oil has n ≈ 1.515, allowing high-NA oil objectives near 1.4. Deep in thick, watery samples, a water objective that better matches the sample can give a sharper image. A fluorescent label absorbs one colour and emits a longer wavelength; that emission wavelength sets the detected blur.

That is why “the light limit is 200 nm” is incomplete: state the wavelength, NA, direction and criterion.

### Rayleigh and Nyquist are different questions

The two-pixel rule in the main line is introductory. Strict Nyquist sampling follows the finest spatial frequency the objective passes. Here λ/(2NA) = 186 nm, so pixels should be about 93 nm or smaller. Thus 100 nm pixels give more than two pixels across the 226.6 nm Rayleigh distance but still slightly under-sample by this stricter rule; careful work often uses smaller pixels.

### Evidence and history behind super-resolution

Hell and Wichmann proposed [STED](https://stemhop.com/methods/super-resolution-microscopy/) in 1994. In 2000 Klar, Jakobs, Dyba, Egner and Hell reported a fluorescent spot about 90 to 110 nm across, with smaller gains in live yeast and *E. coli*. Gustafsson reported a doubling of lateral resolution with SIM in 2000. PALM, STORM and FPALM were published in 2006 by Betzig and colleagues, Rust, Bates and Zhuang, and Hess, Girirajan and Mason, respectively. MINFLUX reported about 1 nm localization precision in 2017, but on DNA origami, not in a cell. The 2014 Nobel Prize in Chemistry went to Eric Betzig, Stefan Hell and William E. Moerner; no one person invented super-resolution.

### Electron wavelength is not specimen resolution

At 100 kV an electron’s wavelength is around 0.004 nm, but preparation, contrast and beam damage make practical biological resolution far coarser. The probe’s wavelength is only one link in the measurement chain.

### Labelling can move the target

Fixation is an experimental variable, and incomplete fixation can let labelling move proteins. In primary human lymphatic endothelial cells made to overexpress LYVE-1, antibodies clustered the receptor on unpermeabilised cells fixed with formaldehyde alone; adding glutaraldehyde largely prevented the clustering. The long argument over Golgi’s silver stain in nerve tissue, described in [the neuron doctrine](https://stemhop.com/lessons/neuron-doctrine/), was likewise an argument about what a stain reveals.

---

From STEMhop (https://stemhop.com), free lessons by Christopher Deppmann. Lesson text © the authors. Practice questions: https://stemhop.com/lessons/microscopy-and-images/
