Quantifying signaling pathway activation to monitor the quality of induced pluripotent stem cells

Abstract

Many attempts have been made to evaluate the safety and potency of human induced pluripotent stem cells (iPSCs) for clinical applications using transcriptome data, but results so far have been ambiguous or even contradictory. Here, we characterized stem cells at the pathway level, rather than at the gene level as has been the focus of previous work. We meta-analyzed publically-available gene expression data sets and evaluated signaling and metabolic pathway activation profiles for 20 human embryonic stem cell (ESC) lines, 12 human iPSC lines, five embryonic body lines, and six fibroblast cell lines. We demonstrated the close resemblance of iPSCs with ESCs at the pathway level, and provided examples of how pathway activity can be applied to identify iPSC line abnormalities or to predict in vitro differentiation potential. Our results indicate that pathway activation profiling is a promising strategy for evaluating the safety and potency of iPSC lines in translational medicine applications.

Keywords: Gerotarget; algorithm; bioinformatics; embryonic stem cells; iPSC; pathway activation.

Figure 1. Distribution of statistically significant PAS in ESC and iPSC lines

PAS values were calculated with the Regeneration Intelligence software suite for ESC (orange) and iPSC (green) cell lines from GSE25970. Mean PAS values ± SD for 97 pathways with p-value < 0.001 are shown.

Figure 2. Pathways showing high and low variability within ESC or iPSC lines

A. The top 20 most variable pathways in ESC and iPSC lines. B. The top 20 least variable pathways in iPSC and ESC lines. The most and least variable pathways common to both ESC and iPSC lines are listed.

Figure 3. ESC and iPSC lines exhibit a common profile of pathway activation

A. Hierarchically clustered heat map of the top-50 most variable pathways for all iPSC and ESC lines (using fibroblast cell lines as reference). Blue indicates down-regulation, and red indicates up-regulation. B. Scatterplot comparing the deviation of each pathway in ESC vs. iPSC lines (measured relative to the ESC references; to calculate deviations for ESC, each ESC line was excluded one at a time from the averaged ES PAS reference to prevent comparing cell lines to themselves).

Figure 4. PAS-based quality scores for 12 iPSC lines

Shown for each iPSC line is the number of PAS that fall within the ES PAS quality score that are indistinguishable from ES PAS. The upper and lower limits for each ES PAS quality score were calculated as the average of the PAS values across all ESC lines for each pathway, plus or minus SD. The red line corresponds to 50% (95 pathways falling within corresponding ESC PAS quality range) of the maximum (190) iPSC quality score calculated using 190 pathways. Note that, the iPSC lines hiPS11b, hiPS17a, and hiPS27e lines received the lowest PAS-based quality scores; this is consistent with the observation in Bock et al. [17] that hiPS11b and hiPS27e show impaired, whereas hiPS17a showed enhanced differentiation capacity.

Figure 5. LASSO model using PAS scores can discriminate between iPSCs and fibroblasts

Lasso model coefficients for the nine pathways selected by LASSO as strong predictors of iPSC lines vs. fibroblasts. Pathways with negative coefficients have lower PAS scores in fibroblasts, and pathways with positive coefficients have higher PAS scores in fibroblasts. Classifier accuracy estimated with cross-validation is 100%.