Quantitative modeling of autofluorescence and non-specific staining allows for improved cell phenotyping and marker assessment in highly multiplexed immunofluorescence studies

While fundamental in cancer diagnostics, immunohistochemistry (IHC) has several significant limitations, including the inability to simultaneously detect multiple markers per slide, which limits true co-expression analysis. In contrast, sequential immunofluorescence (seqIF), a multiplexed protein biomarker detection method, allows for simultaneous detection of multiple markers on a single slide by using antibodies with distinct fluorophores combined with the repeated automated cycles of staining, imaging, and elution. While seqIF has quantitative and scalability advantages over IHC, its use of fluorescence introduces various challenges for downstream computational analysis such as tissue's autofluorescence and non-specific fluorophore accumulation, which are compound by multiple cycles of staining, imaging and elution. The Lunaphore COMET, a seqIF system, allows for an initial background autofluorescence acquisition at the beginning of the protocol. However, we observed that this single baseline measurement did not account changes in the fluorescence of different regions in various tissue types with subsequent imaging and elution cycles, which significantly affected downstream analysis including cell typing.To further investigate these artifacts, we evaluated both autofluorescent quenching on a variety of tissue types during the imaging cycles as well as the non-specific accumulation of fluorophores in all channels (FITC, TRITC, Cy5, and Cy7). We found that the background autofluorescence decreased with sequential cycles in some, but not all cycles, which led to artificially decreased signal intensity for downstream marker analysis. Additionally, we found that non-specific fluorophore accumulation was channel specific, tissue specific, and subcellular localization specific, further confounding interpretability. Overall, these findings suggest that utilizing a single baseline autofluorescence cycle for background subtraction is insufficient and can lead to significant downstream analytical errors.While performing an additional imaging cycle after each elution cycle and subtracting background accordingly may remedy this problem, it is not always feasible or efficient due to increased time, reagent usage, and file size. We found that a total of 5 imaging cycles post-elution accurately predicted the background fluorescence for 20 cycles using a cell-level cubic spline model (R 2 = 0.94). These background signal modeling results are then utilized to calculate background-adjusted marker expression values at a cell level for individual cycles, which improved clustering resolution and interoperability. Overall, we demonstrate issues encountered with autofluorescence in seqIF and present a novel method for mitigating them to improve results.