To screen for gene c interac ons that may be complementary in producing differen a on delay in N2B27, we carried out an inhibitor screen using X of our knockout lines in the presence of one of five inhibitors. Firstly, the two components of 2i medium (Ying et al. 2008) : CHIR99021 (CH), which mimics Wnt signalling ac va on by inhibi ng Gsk3 and preven ng the degrada on of β-catenin (Ying et al. 2008) , and PD184352 (PD), an inhibitor of Mek (Allen et al. 2003) . Secondly, LIF, which acts through ac va on of at least three pathways: Jak-Stat3 (Niwa et al. 1998) , MAPK, (Matsuda et al. 1999; Burdon et al. 1999) and PI3K-Akt-mTOR (Welham et al. 2007) . Thirdly, Cdk8i, which inhibits Cdk8 expression, an oncogene and posi ve regulator of β-catenin and Myc (Firestein et al. 2008; Adler et al. 2012) . Lastly, Gö, an inhibitor of atypical protein kinase C (PKC), specifically Prkci (Leeb et al. 2014) . Suppression of Prkci counteracts differen a on through modula on of the conserved Notch pathway (Du a et al. 2011; Mah et al. 2015) .
Our collaborators measured the percentage of undifferen ated cells (Zfp42-GFP posi ve) a er 48h (two replicates) and 72h (one replicate) a er placing cells from 2i/LIF into N2B27 medium. Addi onally, all cell lines were placed in unmodified N2B27 medium as control. As a gene c control, RC9 cell lines were included in all condi ons. As the frac ons of undifferen ated cells are by defini on bounded between 0 and 1, they are expected to follow a beta distribu on. To es mate the whether any interac on of knockout ( K ) and condi on ( C ) resulted in a greater or lower frac on of undifferen ated cells than expected from the behaviour of their controls, we therefore fi ed beta regression models to our data ( Eq 19 ) using the betareg R package (Cribari-Neto & Zeileis 2010) .
We then tested the significance of the interac on coefficients β 1 using a partial Wald test . A
nega ve coefficient indicates a lower propor on of undifferen ated cells than expected, whereas a posi ve coefficient indicates an increase over expecta on.
Fig 69 : Inhibitor interactions of pluripotency regulation.
Left: Interaction coefficients of the beta regression model interaction term, for all inhibitor-knockout combinations.
Asterisks indicate significance of the coefficient (.p ≤ 0.1, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001). A significant positive value indicates that a higher fraction of undifferentiated cells was observed than expected (dependent on both wild type levels and N2B27 levels), while a significant negative value indicates the opposite. Right: Heatmap showing the average fraction of undifferentiated cells in each inhibitor-knockout combination as well as the RC9 wild type and N2B27 (no inhibitor) control medium. Shaded bars represent the fractions in each combination (redundant with colour information). Dotted lines indicate wild type levels for easier comparison.
Under PD condi on, most significant interac on coefficients (Wald test p-value ≤ 0.05) were nega ve: those of Trp53, Myc, Ctbp2, Csnk1a1, Ptpn11, Fgfr1 and Zfp281 knockouts ( Fig 69 ). We inves gated the Zfp281 knockout more closely, as it has a strong control (N2B27) phenotype that is not exacerbated by PD ( Fig 69 ). Zfp281 is essen al for stem cell priming and known to be
cell differen a on through overlapping mechanisms. Surprisingly, Mapk1 (Erk2) knockout cells showed a drama c increase in the number of undifferen ated cells in the presence of only PD. As PD inhibits signal transduc on via the Mapk/Erk pathway (Allen et al. 2003) , we expected a redundant phenotype similar to Zfp281 knockouts. This suggests that in the absence of a signal from Mek , Mapk1 might be able to be ac vated through an alterna ve mechanism. Inhibi ng both Mek and knocking out Mapk1 might therefore more completely inhibit the downstream ERK pathway that posi vely regulates cell prolifera on.
In the CH supplemented medium, three knockouts showed a posi ve interac on: Leo1 , Zfp281 , and Jmjd1c . The synergis c effect with Zfp281 further supports the no on that Zfp281 may be par ally redundant with PD and therefore fulfill a similar func on in tandem with CH. On the other hand, five knockouts were less responsive than expected: Tcf7l1, Eed, Csnk1a1, Pten, and Ctbp2.
In the LIF condi on, we iden fied five knockouts that fell short of the expected increase in undifferen ated cells: Tsc2 , Rbpj , Smg5 , Eed, and Trp53 ( Fig 69 ) . These knockouts also show differen a on-delay phenotypes in the N2B27 control condi on, indica ng that their effects may be par ally redundant with LIF ac vity.
4.4 Discussion
Dissec ng the precise mechanisms by which pluripotent stem cells exit their naive pluripotent state and choose their differen a on trajectory is of crucial importance for applica ons in regenera ve medicine (Tabar & Studer 2014; Mizuno et al. 2012) . A mul tude of studies have focused both on the naive state itself (Young 2011) , or on the condi ons that trigger
differen a on (Ying et al. 2003; Wang et al. 2012; Leeb et al. 2014) . The genera on of haploid mouse embryonic stem cells (mESC’s) has greatly facilitated the search for the causal drivers of priming and differen a on (Elling et al. 2011; Leeb & Wutz 2011; Leeb et al. 2014) . However, these screens are s ll restricted to iden fying the responses to single knockouts. Complex molecular signalling networks are characterized by a high degree of redundancy (Gi er et al. 2009; Macneil & Walhout 2011) , sugges ng that mul ple parallel or interlocking mechanisms may govern cell fate decisions (Leeb et al. 2014) . Due to the exponen al search space of
possible experiments with two or more simultaneous knockouts, as well as the associated costs, a brute-force or random sampling strategy is infeasible. Therefore it is crucial to pursue a
systems level approach and characterize the interdependencies of regulatory network components to maximize informa on from single knockout screens.
In this project, we carried out the analysis of RNAseq data from one such large single knockout screen. Clonal cultures of 74 knockout mutants were kept in 2i medium, thereby maintaining pluripotency, or placed in N2B27 differen a on medium. The knockouts were selected by our collaborators based on their expression of a canonical marker of naive pluripotency, Zfp42 (Rex1), in N2B27 culturing condi ons, indica ng different degrees of differen a on delay. As controls, knockouts that accelerate differen a on, KO Myc and KO Nes , were included. We then integrated these data with informa on from two suppor ng experiments: the iden fica on of LIF-specific target genes, and the direct quan fica on of differen a on delay caused by a combina on of mutants and one of five inhibitors that affect differen a on. These data will be made available to the research community, both directly and through an interac ve web interface, providing a high quality and comprehensive resource.
At the me of wri ng, this project is ac vely being developed. However, through the analysis presented in this thesis, we have already achieved substan al progress towards several key goals.