DLBCL cells with ferroptosis morphology can be detected with a deep convolutional neural network

Pyry Kotkaranta; Mikko Chan; Tero Vuolio; Ilkka Miinalainen; Hanne Kuitunen; Taina Turpeenniemi-Hujanen; Hanna-Riikka Teppo; Outi Kuittinen; Milla E L Kuusisto

doi:10.1016/j.biopha.2024.117785

DLBCL cells with ferroptosis morphology can be detected with a deep convolutional neural network

Biomed Pharmacother. 2024 Dec 24:182:117785. doi: 10.1016/j.biopha.2024.117785. Online ahead of print.

Authors

Pyry Kotkaranta¹, Mikko Chan², Tero Vuolio², Ilkka Miinalainen³, Hanne Kuitunen⁴, Taina Turpeenniemi-Hujanen⁵, Hanna-Riikka Teppo⁶, Outi Kuittinen⁷, Milla E L Kuusisto⁸

Affiliations

¹ Translational Medicine Research Unit, University of Oulu, Oulu, Finland. Electronic address: [email protected].
² Translational Medicine Research Unit, University of Oulu, Oulu, Finland.
³ Biocenter Oulu, University of Oulu, Aapistie 5A, Oulu 90220, Finland.
⁴ Department of Oncology, Oulu University Hospital, Oulu, Finland.
⁵ Translational Medicine Research Unit, University of Oulu, Oulu, Finland; Department of Oncology and Radiotherapy and Medical Research Center, Oulu University Hospital, Oulu, Finland.
⁶ Department of Pathology, Medical Research Center Oulu, Oulu University Hospital, Finland.
⁷ Institute of Clinical Medicine, Faculty of Health Sciences, University of Eastern Finland, Kuopio, Finland; Department of Oncology, Kuopio University Hospital, Kuopio, Finland; Medical Research Center, Oulu University Hospital, Oulu, Finland.
⁸ Medical Research Center, Oulu University Hospital, Oulu, Finland; Department of Internal Medicine, Länsi-Pohja Central Hospital, Kemi, Finland; Biomedicine and Internal Medicine Research Unit, University of Oulu, Oulu, Finland.

PMID: 39721323
DOI: 10.1016/j.biopha.2024.117785

Abstract

It has been demonstrated that diffuse large B-cell lymphoma (DLBCL) is especially sensitive to ferroptosis. Currently, confirming the presence of ferroptosis requires flow cytometry, which is a time consuming and labor-intensive task. Blistering of the cell membrane has been shown to be a ferroptosis-specific morphological change. In this study we developed a deep convolutional neural network to detect the blistering of cell membrane. Buthionine sulfoximine treatment increased the percentage of blistering cells from 2 % to 38 % (p < 0.001) when glutathione was deprived from the culture media. Ferrostatin-1 treatment completely reversed the effect. Imidazole ketone erastin (IKE) and auranofin treatment increased blistering cells gradually in dose response manner from 5.4 % to 18.1 % (p < 0.05) and 6.1-50.1 % (p < 0.0001) respectively. We also tested malignant melanoma and breast cancer cell lines to confirm that the blistering phenomena can also be observed in adherent cell lines. We used fluorescence-activated cell sorting to measure the lipid peroxidation associated with ferroptosis and found a significant increase of bodiby-C11_oxidized mean compared to DMSO controls for IKE (345 vs 462, p < 0.01) and auranofin (345 vs 686.5, p < 0.05).

Keywords: Cancer therapy; DLBCL; Ferroptosis; Machine learning; Morphology.