1  Image Analysis

Confocal microscopy images were analyzed using a custom Python pipeline. Maximum intensity Z-projections were generated for both DAPI and GFP channels. For granule detection, the GFP channel was preprocessed using Difference of Gaussians filtering (σ₁ = 3.0, σ₂ = 4.8) to enhance punctate structures while suppressing background noise. Granules were segmented using a global intensity threshold of 0.0009, followed by removal of objects smaller than 15 pixels and those touching image borders.

Cell segmentation was performed using Cellpose v3.1 (Stringer et al. 2021) with the cyto3 model (Stringer and Pachitariu 2024). Prior to segmentation, images were preprocessed with Gaussian smoothing (σ = 0.5 for GFP and σ = 1.0 for DAPI channels) and intensity rescaling to the [0,1] range. The model was configured with a target cell diameter of 100 pixels and minimum size threshold of 30 pixels. Cell masks were expanded by 5 pixels and border-touching cells were excluded from analysis.

Morphometric measurements including area, perimeter, and eccentricity were calculated for both granules and cells using scikit-image v0.24 (Walt et al. 2014). All measurements were converted to physical units using the microscope’s calibrated pixel size (0.18 µm/pixel). For each cell, the number of contained granules was counted and normalized to cell area. The analysis pipeline was implemented in Python using cellpose, scikit-image, numpy, pandas and plotting libraries matplotlib, seaborn and microfilm v0.2.1.