Are you regularly generating spectral cytometry data? If so, do you:

If you answered "yes" to any of the above, chances are you've had conversations similar to ones we've had inside Ozette. In fact, these are the kinds of conversations that led us to build Ozette Resolve Spectral Unmixing, our instrument-agnostic solution for evaluating unmixing controls and performing adaptive unmixing of spectral cytometry datasets.

Time and again we found that errors introduced in the unmixing process were the root cause of complex issues we observed in downstream data analysis. Sometimes a stain wasn't behaving as expected. At other times, undetected cross-contamination of single-color controls compromised the estimated spectral signatures. In still other cases, fluors with highly similar spectra interacted with the specific mathematical unmixing model to produce strange artifacts.

Whatever the error, we decided we needed a system to surface and correct these issues. Our solution is Ozette Resolve. We use it to reliably deliver high-quality unmixed outputs for all spectral datasets we generate or analyze, and our customers use it to unmix their own data to achieve similarly high-quality outputs.

We've realized there is significant value in using Resolve to monitor the quality of unmixing controls over time: changes in disease states, reagents, buffering systems, stimulation conditions, and biological matrices can all affect unmixing outputs. By monitoring control performance through the Resolve workflow, we identify issues that would produce unmixing errors, and remediate them before unmixing the raw data — preventing errors that would confound downstream analysis. To help you reach this goal, we'll discuss our own best-practices for optimizing unmixing outputs. Our first topic is a critical but often overlooked component of any experiment: unstained controls.

Unstained Controls

Unstained controls are used to estimate the autofluorescent properties of materials measured on a cytometer. These controls are particularly valuable in Resolve, since it calculates event-specific autofluorescence estimates from them.

Many conventional approaches account for autofluorescence by using an unstained control to estimate one or several autofluorescent profiles, either on the bulk population or subdivided by light scatter (e.g. lymphocytes, monocytes, granulocytes), then adjoining these profiles to a fixed unmixing matrix. In Resolve, unstained controls are instead used to estimate event-specific autofluorescence contributions. The system therefore benefits when you use paired unstained controls for each sample you unmix.

To see this, consider four unstained PBMC samples from four subjects. We can immediately see differences in the scatter properties of the four samples.

Four unstained PBMC samples showing differing scatter properties

Some differences can be explained by the number of events acquired (ranging from ~62,000 in sample A to ~295,000 in sample C). By gating down to lymphocytes,

Gating down to lymphocytes within the four samples

we can compare the median measured autofluorescence intensity within lymphocytes across the four subjects.

Median autofluorescence intensity within lymphocytes across four subjects

There are subject-specific differences in median autofluorescence. The magnitude varies by detector, but the difference between the brightest and dimmest median signal summed across detectors represents about 9,100 fluorescence units — caused by sample autofluorescence alone. Plotting the 5% and 95% bands on the dimmest and brightest samples shows a lot of overlap:

5% and 95% bands on the dimmest and brightest samples

This overlap implies many events can have their autofluorescence reasonably approximated by either sample. But the dimmest events in the dimmest sample would not be well-approximated using the brightest sample, and vice versa, since each sample lacks example events with those empirical autofluorescence characteristics.

Observing variable autofluorescence profiles between subjects and runs has led us to the following recommendations for using unstained controls in Resolve:

We have achieved high-quality unmixing through Resolve under all three scenarios, but paired unstained controls are particularly valuable for calculating the recently developed positivity scores included in Resolve outputs — a novel metric that can be used to inform objective gating, to be discussed in a future post.

In our next article, we discuss how to use Resolve to detect and remediate cross-contamination between single stain controls. In the meantime, if you'd like to try these recommendations with Resolve today, ask us about our free demo.

← All posts