Emission Filter

Filter datasheets lead with transmission — 93%, 95%, sometimes 98% — and that is the figure people compare. It is almost never the figure that determines whether the image is usable.

Consider the ratio you are working against. Of the excitation photons you pour into a sample, the fraction that comes back as detected fluorescence is somewhere between one in a hundred thousand and one in ten million: most excitation light passes straight through, and of the small part absorbed, some is lost to heat, and of the light emitted, your objective collects a fraction and your camera detects a fraction of that.

So if any meaningful portion of the excitation beam reaches the detector, it does not degrade the image; it replaces it. That is what out-of-band blocking, quoted as optical density, is for. OD6 means one part in a million gets through. Against a million-to-one background that puts stray excitation exactly level with your signal, which is not good enough. You need the dichroic's three or four orders and the filter's six, working together, to arrive somewhere near ten orders and a genuinely black background.

Read blocking as specified at the laser line, not averaged. Average blocking is a number that can be made to look good by being excellent where it does not matter.

There is an optimum bandwidth, it is never the widest available, and the reason is that the two things you are trading move at different rates.

Signal grows sublinearly as you open the filter. You start centered on the emission peak, where the spectrum is tallest, so the first ten nanometers of extra width collect a great deal and each subsequent ten collect less; you are working down the flanks of the band.

Contamination grows superlinearly. Your neighbor's emission is rising as you move toward it, so each extra ten nanometers of width admits more than the last did. One curve is flattening while the other is steepening, which guarantees they cross.

The asymmetry that decides where: emission bands are not symmetric. They fall away steeply on the blue side and drag a long tail to the red, because a molecule can drop to the ground state with a vibration still running and emit a photon short of the full energy, but it has nothing to borrow from to emit one above it. So widening a filter toward the red costs far more in contamination than widening it toward the blue. If you must open a filter, open the blue edge first.

The filter does not only decide how much of your dye and your neighbor's dye reach the detector. It decides how much of the tissue reaches the detector, and tissue is not optically silent.

Fixed tissue fluoresces on its own. Collagen and elastin contribute, flavins and NADH contribute, lipofuscin contributes heavily in aged or neural tissue, and formalin fixation adds its own. The bulk of it is excited in the blue and emits broadly across roughly 500 to 600 nanometers, which is to say, directly on top of the green channel almost everyone uses.

This changes the optimum. In cell culture, where autofluorescence is slight, you can afford a generous green filter. In FFPE tissue the same filter collects background that scales with tissue thickness rather than with your marker, so signal-to-background falls even as signal-to-noise improves. Those are different quantities and only one of them is on the histogram.

It is also the honest argument for pushing a panel red. A far-red channel is not intrinsically better at anything — quantum efficiency is usually worse there, but almost nothing in the tissue is emitting to compete with it.

Alexa 488 on FFPE tissue, emission peak 519 nm, with an Atto 565 marker in the next channel. Three candidate filters.

A 525/50 — 500 to 550 nm — is the default in most filter drawers. It collects a large share of the 488 emission and its red edge at 550 sits in the thick of the tissue background, and reaches toward where Atto 565 is climbing.

A 512/25 — 500 to 525 nm — gives up perhaps a third of the signal relative to the 525/50. In exchange it clears the autofluorescence hump and stops well short of the neighbor. If the marker is bright, this is usually the better image, because the third you gave up was signal you could recover with exposure while the background you excluded was not recoverable at all.

A 535/70 collects the most and is the right answer only if 488 is the sole dye in the sample and the tissue is clean.

The general shape of the decision: exposure time can buy back signal. Nothing buys back background, because it rises with the signal and carries its own shot noise. Given a choice between a filter that is dim and clean and one that is bright and contaminated, take the dim one and expose longer, right up until photobleaching or the frame rate says otherwise.