High-content imaging system selection: reproducibility first

Confocal or widefield: sample thickness and schedule decide

Neither has an advantage in itself. The thickness of your sample and what you need to see in the experiment decide.

A widefield system also collects light from outside the focal plane, so it images each field faster. For early screening of monolayer cells against tens of thousands of compounds, that is enough. A confocal system removes light from outside the focal plane and forms an image of a thin section only. To see inside spheroids and organoids, which are cell masses grown in three dimensions and can be hundreds of micrometers thick, you need confocal. The cost is longer imaging time.

Representative confocal systems are Revvity's Opera Phenix Plus and Yokogawa Electric's CellVoyager CV8000. Molecular Devices' ImageXpress Micro Confocal can switch between confocal and widefield. Culture of 3D models is covered on the organoid culture kits page. A switchable system lets you run widefield early and confocal only for hit confirmation, but values change with the method, so be careful when comparing across stages.

Decide by which failure costs more in your assay: if blurry images hurt more, lean toward confocal, and if missing the schedule hurts more, lean toward widefield.

Lenses, plates and automation belong to the same decision

Once the optical method is chosen, performance still depends on whether the lens and the plate fit it.

The number that catches the eye in a catalog is the NA (numerical aperture, a measure of light-collecting power) of the objective lens. Higher NA shows finer detail and a brighter image, but the field of view gets narrower. On the Opera Phenix Plus, choosing a water-immersion high-NA lens narrows the field of view a great deal and also shortens the working distance, the gap between lens and sample. To image the same well you must stitch together many more fields, so imaging time can grow instead of shrinking. NA is a number about image quality, not a guarantee of speed.

Catalog speeds also come with fixed conditions, such as plate type and number of colors. Yokogawa Electric publishes 1 minute for imaging a 96-well plate in four colors on the CV8000. How many minutes it takes with your cells and your plates is best measured on a demo unit.

The plate bottom is part of the system too. A water-immersion high-NA lens pairs with a plate whose bottom is a thin film about 0.2 mm thick (Revvity's PhenoPlate, for example). As the well count rises to 384 and 1536, you need automated dispensing equipment such as Tecan's Fluent or Hamilton's Microlab STAR. A model that can carry a built-in pipettor, such as the CV8000, can image changes a few seconds after a stimulus without moving the plate. A laboratory without an automation line may choose a benchtop type such as Yokogawa Electric's CQ1. You can compare dispensers on the liquid handling page, and compare systems for following changes over time on the live-cell imaging page.

Choose lens, plate and liquid handler together as one system.

Reproducibility beats feature count: what Cell Painting and JUMP show

If staining is not standardized, data become impossible to compare every time you change instruments. The best-known standardization of staining is Cell Painting.

In the procedure Bray and colleagues published in Nature Protocols in 2016, six fluorescent dyes stain eight structures, including the nucleus, the endoplasmic reticulum, mitochondria, the actin cytoskeleton and the Golgi apparatus, and the cells are imaged in five channels. The images are analyzed with the free CellProfiler software, which extracts about 1,500 morphological features per cell. That number is the ceiling of what you can capture. If the values shift each time the imaging day, the plate or the reagent lot changes, the usable information shrinks. This shift is the batch effect. Imaging data are multidimensional, so a single number such as the Z′ factor (in Japanese) has trouble revealing the shift.

A real example of how large the shift can be is the JUMP Cell Painting Consortium, led by the Broad Institute. Ten pharmaceutical companies, six technology companies and two nonprofit organizations took part, and 12 sites did the imaging. At the preprint stage, the set held about 116,750 compounds, 12,602 gene overexpressions and 7,975 CRISPR knockouts. The 136,000 perturbations in the title is the collective name for these. One aim of gathering this scale with the same procedure was to validate, on real data, correction methods that hold across sites and reagent lots.

So the question to put to a vendor is not "how many colors can you image?" It is "when I image the same plate again days later, how much do the features vary?" To check it yourself, put shared control wells on every plate and see whether the same values come out across days.

Instrument performance is set in one imaging run, but the value of the data is set by how far it holds when you image again.

Analysis software: bundled with the instrument, or free and vendor-neutral?

Analysis software is sold together with the instrument.

Revvity's instruments come with Harmony, Molecular Devices' with MetaXpress, and Yokogawa Electric's with CellPathfinder. Harmony can measure volume and surface area from 3D confocal data, and its machine learning plug-in, PhenoLOGIC, can classify phenotypes that are hard to tell apart by eye. CellPathfinder can also analyze unstained brightfield images. Free software such as CellProfiler, on the other hand, can analyze images from instruments of different makers in the same pipeline. That compatibility is why Cell Painting is used across sites.

The question to ask when you choose is whether you can export images in a standard format and run the same analysis in other software. The answer matters when you replace instruments or merge data with another site. The choice of software can be changed after imaging, so you do not have to fix it at the same time as the instrument.

When you choose, ask whether images can be exported in a standard format and re-analyzed elsewhere, because that protects you the day the instrument changes.

The order of decisions and what to check at a demo

Decide the sample thickness and the purpose of the experiment first. Lean toward confocal if you need to see inside a thick 3D model, toward widefield if you need to cover large numbers of monolayer wells, and choose a switchable system if you need both. Next decide the lens and plate that fit the optics, and the well count and the dispensing scale. Then choose the analysis software. Last, image the same plate again on a different day and pin down the variation in features as numbers. Teams that postpone this last check often find the problem only after installation.

At the demo or quotation stage, confirm the following.

1. Have you measured how long one plate takes with your own cells and plates, and are the catalog conditions the same as yours?
2. If you work with thick samples, does the inside of your own spheroids come out as an image?
3. Have you imaged the same plate again days apart and looked at the variation in features?
4. Can you export images in a standard format and analyze them in other software?
5. Do the dispensing automation and the plate bottom match the instrument you chose?

The main point of this order is not to look at the number of features first.