Written and maintained by CASRAI Editorial Board
Last updated
Running a colocalization analysis in Fiji is a software workflow question, not a statistics question — the underlying choice between Pearson’s and Manders’ coefficients is covered separately in Pearson vs. Manders Coefficients: Choosing and Defending a Colocalization Statistic. This page covers the practical side: which plugin to use, how the Costes automatic thresholding method actually works, and the concrete steps from a raw two-channel image to a reportable coefficient.
The standard tool: Fiji’s Coloc2 plugin
Coloc2 is Fiji’s built-in colocalization plugin and the de facto standard for this analysis — it implements Pearson’s, Manders’, Li’s, and related pixel-intensity-correlation metrics in one tool, alongside automatic thresholding and a randomization-based significance test. It supersedes the older, more limited colocalization tools bundled with earlier ImageJ releases and is maintained as part of the core Fiji plugin set, so it’s already installed in a standard Fiji distribution — no separate download is required.
Find it at Analyze > Colocalization > Coloc 2, or press L to open Fiji’s command finder and type “coloc” to jump straight to it. A separate plugin, JACoP (Just Another Colocalization Plugin), computes an overlapping set of coefficients and is worth knowing about if you’re reproducing a paper that specifically cites it, but Coloc2 is the tool most current Fiji workflows and methods sections default to, and it’s the one this page walks through.
Coloc2 requires a two-channel image as input. If your acquisition produced a multi-channel composite, split it first via Image > Color > Split Channels before launching the plugin.
The Costes automatic threshold method
Manders’ coefficients (M1, M2) are calculated as a ratio of above-threshold pixels, which means the reported value depends directly on where the background/signal cutoff is drawn. Setting that threshold by eye is subjective and hard to reproduce — two people looking at the same image can reasonably pick different cutoffs, and the same person can pick differently on different days. The Costes automatic threshold method (Costes et al., 2004) exists to remove that subjectivity: it’s an algorithm that determines, from the statistical relationship between the two channels themselves, the intensity value above which pixels are treated as genuine signal for the Manders’ calculation, rather than leaving that cutoff to manual judgment.
In practice, this means the same image, run twice through Coloc2 with Costes thresholding selected, returns the same threshold and the same coefficients — which is the entire point. It’s not infallible: the documentation for the method notes it can behave poorly on images with a high zero-offset or very strong, non-uniform background, in which case it can settle on a threshold below the image’s real noise floor. If your Coloc2 output threshold looks implausibly low relative to what you can see in the image, that’s the failure mode to suspect, and a manual threshold with the choice explicitly documented is the more honest fallback in that case.
Coloc2 also runs a Costes randomization-based significance test alongside the threshold calculation. It works by scrambling the image into blocks (sized to roughly your point-spread function, which you specify before running the analysis) and repeating the colocalization calculation on the shuffled version many times over, building a distribution of the coefficient you’d expect from spatially randomized signal. Comparing your actual, unscrambled result against that distribution gives a measure of whether the colocalization you observed is greater than what chance overlap alone would produce. Fiji’s own documentation suggests running at minimum 10 iterations for this test, with 100 recommended for a more stable result — more iterations cost more computation time but give a steadier estimate.
Step-by-step workflow
1. Image preparation and ROI selection
Start from raw, unprocessed channel data where possible — avoid running colocalization on images that have already been contrast-stretched or gamma-adjusted for display, since that changes the intensity relationships the calculation depends on. Split multi-channel composites into separate single-channel images (Image > Color > Split Channels) if they aren’t already separate. If you want to restrict the analysis to a specific structure or cell rather than the whole field of view, draw a region of interest with Fiji’s selection tools, or prepare a binary mask image (matching the x/y/z dimensions of your data, white = include, black = exclude) if the region is irregular or you’re batch-processing multiple images with the same mask.
2. Running Coloc2
Open Analyze > Colocalization > Coloc 2, assign your two split channels as Channel 1 and Channel 2, and apply your ROI or mask if you prepared one. Enable Costes thresholding and set the PSF size (in pixels) and the number of randomization iterations for the significance test — 100 is a reasonable default unless computation time is a constraint. Run the analysis.
3. Interpreting the output
Coloc2 returns a results window containing the numerical coefficients (Pearson’s r, Manders’ M1 and M2, and related statistics), the Costes threshold values it selected for each channel, the randomization-based significance result, and a 2D intensity scatter plot with a fitted regression line. The scatter plot is worth reading visually, not just skimming the numbers: a tight, linear cloud of points supports a high Pearson’s r; a cloud with two separate populations (points along one axis with none along the diagonal, plus a separate diagonal cluster) suggests your image contains distinct colocalized and non-colocalized populations that a single whole-image coefficient will average together and potentially obscure. For what the resulting Pearson’s r and Manders’ M1/M2 values actually mean and when each is the more defensible statistic to report, see the comparison referenced above.
Document the threshold values and the number of randomization iterations used alongside your reported coefficients — per the comparison page linked above, inconsistent or unreported thresholding is one of the most common reasons colocalization results can’t be reproduced from a methods section alone.
Frequently asked questions
Do I need to install anything beyond a standard Fiji download to run Coloc2?
No. Coloc2 ships as part of the standard Fiji distribution’s plugin set, unlike some third-party colocalization tools that require adding an update site. If it doesn’t appear under Analyze > Colocalization, updating Fiji to the current release typically resolves it.
Can I skip Costes thresholding and just set the threshold manually?
Coloc2 allows a manual threshold as an alternative. It isn’t wrong to use one, but it reintroduces the subjectivity Costes thresholding exists to remove, so if you do set one manually, document the specific values you used and, ideally, that you tried Costes automatic thresholding as a cross-check.
What does it mean if the Costes significance test comes back non-significant?
It means the colocalization coefficient measured on your actual image wasn’t distinguishable from what the randomization procedure produced on spatially scrambled versions of the same image — in other words, the apparent overlap could plausibly be explained by chance spatial arrangement rather than genuine co-occurrence, given the amount and distribution of signal present.
Should I report Pearson’s r, Manders’ coefficients, or both?
Both, in most cases — they answer different questions, and Coloc2 computes both from the same run. See Pearson vs. Manders Coefficients for the full breakdown of when each is the more meaningful statistic.
For the broader practical bench-science context this workflow sits in, see the Lab Operations hub, and for image-formation background relevant to colocalization imaging, see Confocal Microscopy: Principle, Setup, and When to Use It.
