Global Identification and Characterization of REAP[+] Exons
REAP was applied to identify AS events in NPs compared to hESCs: Cyt-NP versus Cyt-ES; HUES6-NP versus HUES6-ES; hCNS-SCns versus Cyt-ES, and hCNS-SCns versus HUES6-ES. After removing potential FPs, 11,348 genes containing 158,657 probesets were scored by REAP.
As described above, for each pair of cell lines compared, each probset was represented by five points, where each point was defined a significant outlier if it had a high residual (p < 0.01), low influence, and high leverage. Points per probeset should be correlated; in other words, if one point was a significant outlier, the other points were expected to be outliers as well. To ensure that this was the case, we counted the number of probesets with N significant outliers, where N was varied from 0 to 5. Next, the identity of the probesets and points derived from them were exchanged with other probesets, keeping constant the total number of points that were considered significant outliers. At N = 0, we observed approximately equal numbers of probesets in the actual versus shuffled controls. In contrast, we observed that there were 1.5 times more probesets with N = 2 significant outliers relative to shuffled controls; 12–31 times more probesets with N = 3; and 17–612 times more probesets that had N = 4 significant outliers (Figure 5A; see Table S1). For example, in hCNS-SCns compared to Cyt-ES, approximately 0.39% (490 of 124,604) of probesets had three significant outliers and 0.25% (308 probesets) had four significant outliers, relative to 0.02% and 0% of shuffled controls, respectively.
Figure 5  Correlation between “Outliers”
(A) The number of probesets with N significant “outliers” was determined for hCNS-SCns versus Cyt-ES, hCNS-SCns versus HUES6-ES, Cyt-NPs versus Cyt-ES, and HUES6-NPs versus HUES6-ES (N = 0, 1, 2, 3, 4, 5). For comparison, points to probeset relationships were randomly permuted, retaining the same number of “outliers.” Vertical bars represent the ratio between the number of actual points and the randomly permutated sets.
(B) Similar to (A), except points were counted as “outliers” only if they were “outliers” in both hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES (combined hCNS-SCns versus hESC; blue bars); in both HUES6-NP versus HUES6-ES and Cyt-NP versus Cyt-ES (combined derived NP versus hESC; red bars); and in all four comparisons (combined NP versus hESC; yellow bar).   Next we asked whether the overlap between related comparisons was higher than expected. Comparing the significant probesets between hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES revealed 672 significant probesets (N ≥ 2), whereas if we shuffled the associations between probeset identity and significant outliers, only four significant probesets (N ≥ 2) were identified—a 168-fold enrichment (Figure 5B, Table S1). A total of 236 significant probesets overlapped when we compared the derived NPs to hESCs (Cyt-NP versus Cyt-ES and HUES6-NP versus HUES6-ES), relative to seven significant probesets (34-fold enrichment).
At a cutoff of two significant outliers, 1,737 probesets contained in internal exons were defined as positive REAP predictions (hereafter called REAP[+]) exons—candidate AS events that distinguished NP from hESC. Surprisingly, we observed that the majority of REAP[+] exons were specific to the pair of hESC and NP that was compared, likely reflecting differences in genetic origins and/or culturing and differentiation conditions of the cell lines: 614 REAP[+] events were unique to hCNS-SCns versus HUE6-ES; 220 were unique to hCNS-SCns versus Cyt-ES; 439 were unique to HUES6-NP versus HUES6-ES; and 250 were unique to Cyt-NP versus Cyt-ES. The shared events between pairs of comparisons made up a minority of the total number identified: 102 REAP[+] events were found to be in common between hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES; 48 between hCNS-SCns versus HUES6-ES and HUES6-NP versus HUES6-ES; and only 17 between hCNS-SCns versus Cyt-ES and Cyt-NP versus Cyt-ES (Table S2).