Yet at the same time the extent of human variability

Yet at the same time the extent of human variability, of which we are only beginning to take the full measure, poses an important practical problem for the field of iPSC-based modeling. Given background genetic differences between individuals, the key question is which experimental designs and degrees of replication are necessary for robust and sensitive results, and what are the best measures to guide their selection. Here, we address such questions by harnessing the large datasets recently made available by the Human Induced Pluripotent Stem Cell Initiative (HipSci; Streeter et al., 2016; Kilpinen et al., 2016) and the National Heart, Lung, Blood Institute (NHLBI) NextGen consortium (Carcamo-Orive et al., 2016). We derive empirically grounded methodological recommendations for the design and analysis of iPSC-based studies. Given the accessibility of expression profiling and the critical importance of gene expression for the regulation of cellular identity and function, transcriptomic dysregulations have emerged for their rapid and informative insight into the molecular underpinning of diseases, representing for many a first high-content in vitro phenotype. This is especially true for many diseases such as neurodevelopmental disorders, for which the iPSC technology is of prime relevance due to the inaccessibility of the relevant tissues, and whose main disease-associated genes are strongly enriched for transcription factors and LY2835219 regulators (De Rubeis et al., 2014). Transcription is therefore of key importance in itself, as well as representing a powerful stand-in for other, more directly functional assays.
Results
Discussion The predominance of inter-individual differences in explaining transcriptional variability has already been reported (Rouhani et al., 2014; Burrows et al., 2016; Carcamo-Orive et al., 2016). These datasets are likely to underestimate human genetic variability due to the populations sampled (Figure S3); in particular, all individuals from which the HipSci lines used were derived are described by the consortium as “white,” and most of them are labeled as being of English origin. The predominance of inter-individual differences is in line with the surprising degree of genetic variation in human gene expression regulation (Kasowski et al., 2013; Melé et al., 2015; Barrera et al., 2016). Indeed, Barrera et al. (2016) showed that the median human genome harbors 60 heterozygous and 20 homozygous missense SNPs that change the amino acid sequence of transcription factor (TF) DNA binding domains, resulting in changes in affinity and/or specificity in at least 75% of the cases. Together with variations in TF target sites and other changes affecting DNA conformation, these can lead to wide differences in gene expression. While the present results were obtained from the transcriptome of pluripotent cells, the impact is unlikely to be smaller in differentiated cell types. Indeed, Banovich et al. (2016) recently reported that regulatory variation between individuals is lower in iPSCs than in two differentiated cell types. Furthermore, given the difficulty in obtaining samples from patients harboring rare mutations, it is not uncommon for probands\' and control lines to have different origins, and hence potentially confounding background genomic differences. It is therefore likely that the “spurious” differences observed here in the permutation DEAs are relatively conservative estimates. On the other hand, while the present study aimed at measuring the impact of spurious differential expression, it is possible that spurious differences passing multiple testing correction in this context, namely with little real differences between groups, might not necessarily pass it if the two groups show overriding transcriptional differences related to the condition of interest. This is due to the fact that popular correction methods (e.g., Benjamini-Hochberg) are rank based; hence the larger the transcriptional effect of the studied condition, the smaller the number of spurious DEGs will be. Finally, for specific cellular or functional assays with a proven more direct relationship to the ultimate traits under fitness selection, it is plausible that compared with transcription they are more robust and less influenced by genetic variation.