Phenotyping senescent mesenchymal stromal cells using AI image translation

Abstract

Mesenchymal stromal cells (MSCs) offer promising potential in biomedical research, clinical therapeutics, and immunomodulatory therapies due to their ease of isolation and multipotent, immunoprivileged, and immunosuppersive properties. Extensive efforts have focused on optimizing the cell isolation and culture methods to generate scalable, therapeutically-relevant MSCs for clinical applications. However, MSC-based therapies are often hindered by cell heterogeneity and inconsistency of therapeutic function caused, in part, by MSC senescence. As such, noninvasive and molecular-based MSC characterizations play an essential role in assuring the consistency of MSC functions. Here, we demonstrated that AI image translation algorithms can effectively predict immunofluorescence images of MSC senescence markers from phase contrast images. We showed that the expression level of senescence markers including senescence-associated beta-galactosidase (SABG), p16, p21, and p38 are accurately predicted by deep-learning models for Doxorubicin-induced MSC senescence, irradiation-induced MSC senescence, and replicative MSC senescence. Our AI model distinguished the non-senescent and senescent MSC populations and simultaneously captured the cell-to-cell variability within a population. Our microscopy-based phenotyping platform can be integrated with cell culture routines making it an easily accessible tool for MSC engineering and manufacturing.

Keywords: AI image translation; Cell manufacturing; MSC phenotyping; Senescence.

Fig. 1.. Doxorubicin-induced senescence marker prediction.

(a) Machine learning model schematic. (b) Doxorubicin-treatment experimental timeline. Doxorubicin-treated (Dox) MSCs (orange) were cultured with Doxorubicin culture media for 48 hours followed by standard culture media for 24 hours prior to fixation. Control MSCs (blue) were cultured with standard culture media for 72 hours prior to fixation. (c) Fluorescent images and AI prediction for SABG. Left to Right: Phase-contrast images (Input), antibody-stained SABG immunofluorescence images (Target), and ML-produced SABG immunofluorescence images (Prediction). Top to Bottom: Doxorubicin-treated MSCs and Control MSCs. (d-g) Left to Right: Antibody-stained SABG immunofluorescence images (Target) and AI-produced SABG immunofluorescence images (Prediction). Top to Bottom: Doxorubicin-treated MSCs and Control MSCs. (d) p16. (e) p21. (f) p38. (g) CD44. Scale bar for c-g is 100 μm. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2.. Doxorubicin-induced senescence marker quantification.

(a-e) Doxorubicin-induced senescence scatter plots and bar charts. (a) SABG. (b) p16. (c) p21. (d) p38. (e) CD44. Scatter plots demonstrate a strong positive correlation between target and prediction images for control MSCs and Doxorubicin-treated MSCs stained for p16, p21, and CD44. Scatter plots demonstrate a moderate positive correlation between target and prediction images for control MSCs and Doxorubicin-treated MSCs stained for SABG and p38. Bar charts demonstrate a corresponding significant difference in SABG, p16, p21, and p38 between control-MSCs and Doxorubicin-treated MSCs for both prediction and target. Bar charts demonstrate a significant difference and no significance in CD44 between control-MSCs and Doxorubicin-treated MSCs for prediction and target, respectively. The gray diagonal lines denote a perfect prediction-target correlation. N.S. not significant; *p < 0.05; **p < 0.001; ***p < 0.0001. (f) Single cell Pearson correlation between prediction and target iamges for Doxorubicin-treated MSCs stained for SABG, p16, p21, p38, and CD44. Each Pearson correlation box plot contains 30 target-prediction image pairs. All markers demonstrate a moderate to strong Pearson correlation coefficient (r) greater than 0.5. Tgt: Target, Pred: Prediction (g, h) Multi-marker composite images demonstrating CD44, p21, p38, p16, SABG, and nucleus predictions from single phase-contrast images. (g) Doxorubicin-treated. (h) Control group. Scale bar is 100 μm for g and h. (i) Doxorubicin-treated MSC characteristic heatmap. Outlined cells were hierarchically clustered according to marker intensity and eight morphological features leading to clear clustering patterns. AR: Aspect-ratio; Nuc: Nucleus; Circ: Circularity; Round: Roundness.

Fig. 3.. Irradiation-induced senescence marker prediction and quantification.

(a) Irradiation-treatment experimental timeline. Irradiation-treated MSCs (purple) were exposed to irradiation followed by culture in standard culture media for 8 days prior to fixation. Control MSCs (blue) were cultured in standard culture media for 8 days prior to fixation. (b, c) Left to Right: Phase-contrast images (Input), antibody-stained immunofluorescence images (Target), and AI-produced immunofluorescence images (Prediction). Top to Bottom: Irradiation-treated MSCs and Control MSCs. (b) SABG. (c) CD44. Scale bar is 100 μm for b and c. (d, e) Irradiation-induced senescence scatter plots and bar charts. (d) SABG. (e) CD44. Scatter plots demonstrate a strong positive correlation between target and prediction images for SABG IR, CD44 Control, and CD44 IR. Scatter plots demonstrate a weak correlation between target and prediction images for SABG Control. Bar charts demonstrate a corresponding significant difference in SABG between control-MSCs and irradiation-treated MSCs for prediction and target. Bar charts demonstrate no significance and a significant difference in CD44 between control-MSCs and irradiation-treated MSCs for prediction and target, respectively. N.S. not significant; * p < 0.05; ** p < 0.001; *** p < 0.0001. Tgt: Target, Pred: Prediction.

Fig. 4.. Replicative senescence marker prediction.

(a) Replicative senescence experimental timeline. MSCs were cultured in standard 6-well culture plates (6 Well) using either FBS-supplemented DMEM or StemFit MSC culture media for 9 passages, passage 2 to passage 10. Every even passage (e.g. P2, P4, etc.), one 6 well plate of MSCs was fixed. (b, c) Left to Right: Phase-contrast images (Input), antibody-stained immunofluorescence images (Target), and AI-produced immunofluorescence images (Prediction). Top to Bottom: Passage 2 MSCs and passage 10 MSCs. (b) p16. (c) SABG. Scale bar 100 μm.

Fig. 5.. Replicative senescence marker quantification.

(a-d) Replicative senescence scatter plots and bar charts. (a) p16. (b) SABG. (c) p38. (d) CD105. Scatter plots demonstrate a strong positive correlation between target and prediction images for passage 2 MSCs to passage 10 MSCs stained for p16, SABG, p38, and CD105. Bar charts demonstrate a corresponding significant difference in p16, SABG, p38, and CD105 between passage 2 MSCs and passage 10 MSCs for prediction and target. N.S. not significant; *p < 0.05; **p < 0.001; *** p < 0.0001. Tgt: Target, Pred: Prediction. Scatter plots and bar charts were created using Python version 3.7. (e) Single cell Pearson correlation for replicative senescence MSCs stained for p16, SABG, p38, CD105, and CCND2. Here, CCND2 was also tested since it has been commonly found to be upregulated in senescent MSCs (Bertolo et al., 2019). Each Pearson correlation box plot contains 30 target-prediction image pairs. All markers demonstrate a moderate to strong Pearson correlation coefficient (r) with r values greater than 0.5. (f) Replicative senescence MSC UMAP. UMAP demonstrates separation of passage 2 MSCs from passage 10 MSCs and MSCs cultured in DMEM culture media from MSCs cultured in Stemfit culture media.