Writing in Nature communications, Seo and collaborators presented PICASSO as a method to achieve 15-color imaging of spatially overlapping proteins using no reference emission spectra in a single staining and imaging round. This accessible tool has the potential to be applied to diverse applications within the spatial biology field without neglecting accuracy.
Mixing it upMultiplexed biomolecular imaging of heterogeneous tissues is essential for a variety of biological and biomedical research. In particular, high-plex immunofluorescence imaging such as CODEX 1 , CyCIF 2 , 4i 3 , immunoSABER 4 , and NanoString CosMx 5 can detect tens of protein markers to identify heterogeneous cell types and cell-cell interactions at cellular or even subcellular resolution, enabling the study of architectural and spatial relationships of tissue in situ. Central to most of these approaches is the cyclic imaging which is time consuming and may result in signal loss due to photobleaching or possible loss of tissue morphology during the washing and/or quenching steps. Although it has been demonstrated to achieve much higher plex protein imaging using metal isotope labeling for mass spectrometry 6,7 or organic dyes for Raman microscopy 8 , these techniques require complex antibody tagging and different imaging modalities that are not readily accessible.When spectrally overlapping fluorophores are used to label biomolecules such as proteins for multicolor fluorescence imaging, it may become difficult to distinguish real signals from false positive signals due to bleed-through. Thus, techniques to accurately and efficiently tell apart the signal from each fluorophore become indispensable. Researchers had developed approaches to mitigate signal noise and overlap in multiplexed imaging studies. The signal in each channel is modeled as a linear combination of the contributing fluorophores 9 . By employing a mixing matrix, linear unmixing can produce an unmixed image. The major drawback with the linear unmixing approach is that a reference spectrum is needed. For example, in a multiplexed immunofluorescence experiment with two or three fluorophores that have overlapping emission spectra, to quantify the signal unique to each fluorophore, a reference, which could be a region of the tissue section of interest with 'pure' fluorophore can be used. This can be problematic in highly heterogeneous tissues such as the brain. Recently developed algorithms like LUMoS employ unsupervised machine learning approaches to identify the spectral signatures unique to each fluorophore in a method termed as blind unmixing precludes the need to have a reference spectrum 10 . However, challenges still remain with this approach making higher-level multiplexing difficult.