LIF is an interleukine-6 cytokine that plays a crucial role in the maintenance of the naive pluripotency state. Supplementa on of LIF contributes to the ability of cells to self-renew by several avenues: Firstly, by ac va on of the Jak-Stat3 pathway, which is sufficient to maintain self-renewal of mESCs (Niwa et al. 1998) . Secondly, by ac va on of the MAPK pathway, which is involved in mESC differen a on; inhibi on of MAPK by MEKi promotes, but is not necessary for self-renewal (Matsuda et al. 1999; Burdon et al. 1999) . Thirdly, by ac va on of the
PI3K-Akt-mTOR pathways, which promote mESC prolifera on (Welham et al. 2007) . In our knockout screen, we used 2i medium without LIF, which robustly maintains pluripotency in cell lines from diverse gene c backgrounds (Ying et al. 2008) . However, supplementa on of LIF in 2i medium showed addi onal benefits for self-renewal (Ying et al. 2008) , indica ng that LIF and 2i downstream processes are par ally independent. Moreover, knockouts that affect the
pluripotency state might do so in a manner that recapitulates, at least in part, the effects of LIF. To quan fy the baseline LIF response, we first determined the addi onal effect of LIF
supplementa on on our RC9 control cell lines in 2i medium. For this, our collaborators carried out addi onal Quantseq sequencing (Moll et al. 2014) on biological duplicates of RC9 wild type cell lines in both 2i and 2i/LIF condi ons and provided us with the raw counts. Quantseq is restricted to sequencing a region close to the 3’ end of polyadenylated transcripts and produces only one fragment per gene (Moll et al. 2014) . As therefore no bias should arise from gene length we did not use RPKM for filtering lowly expressed genes as before. Instead, we excluded transcripts with a median count of ≤ 5, which resulted in 13432 remaining genes. We then processed the count values using voom and limma as described before. We used TREAT to test the difference between RC9 2i and RC9 2iL directly against an absolute log-fold-change threshold of
in RC9 cells upon addi on of LIF (at a significance level of α = 0.1).
Fig 66 : Wild type response to LIF in 2i medium.
Cell plot showing the 30 most enriched functional terms enriched in the RC9 LIF-response in 2i medium (RC9 2i/LIF vs.
RC9 2i ). All terms were found significant after multiplicity correction (Fisher test, adj. P. Value ≤ 0.05).
We carried out GO enrichment analysis on the set of thus iden fied LIF-specific genes using SETHRO, confirming that LIF affected the expression of genes involved in differen a on and development ( Fig 66 ). Socs3, a canonical downstream target induced by the Jak-Stat3 pathway (Starr et al. 1997) , was increased by more than 128-fold (log2-fold-change = 7.2). While
normally crucial for mESC self-renewal, Jak-Stat3 and Socs3 induc on is known to be
dispensable in 2i-medium (Ying et al. 2008) . Conversely, Lefty1 expression was reduced 26-fold.
Lefty1 is a transforming growth factor β (TGF-β) inhibitor that disrupts BMP signalling and
thereby promotes differen a on (Ulloa & Tabibzadeh 2001; Ying et al. 2003) . Lefty1 is present in both undifferen ated cells and differen a ng cells (Ulloa & Tabibzadeh 2001) and is known to increase upon removal of LIF (D.-K. Kim et al. 2014) . However, BMP is not necessary for
self-renewal in 2i medium (Ying et al. 2008) . These data suggest that our results are of high quality and in line with expecta ons from literature. Addi onal inves ga on is likely to uncover high confidence target genes not previously associated with LIF.
Fig 67 : LIFlikeness of knockouts is associated with delayed differentiation in N2B27.
(A) Scatter plot showing average change in naive marker expression induced by knockouts in 2i (y-axis) and their
LIF-likeness in 2i (x-axis, correlation of RC9 LIF response with knockout-induced response in 2i). (B) Scatter plot showing average change in naive marker expression induced by knockouts in N2B27 (y-axis) and their LIF-likeness in 2i (x-axis, correlation of RC9 LIF response with knockout induced response in 2i).
We next asked if we could iden fy knockout mutants that behaved in a LIF-like manner: does the respec ve knockout (in 2i medium) induce changes that are similar to the specific effect of LIF on wild type cells? For this, we selected the log-fold-changes from all 74 KO 2i vs. RC9 2i
comparisons for the 260 LIF-specific genes. We then correlated each knockout’s expression pa ern with the log-fold-changes induced by LIF in the wild type ( RC9 2i/LIF vs. RC9 2i ), determining
its “LIF-likeness”. To determine whether LIF-likeness of a knockout could predict the strength of the differen a on delay phenotype in N2B27, we further correlated this correla on against the mean of naive marker log-fold-changes ( KO N2 vs. RC9 N2 ). We found that mutants that induced
LIF-like effects in 2i were indeed more likely to show abnormally high naive marker expression in N2B27 (r = 0.69)( Fig 67 B). No such rela onship was found with marker expression in 2i (r = 0.14)( Fig 67 A).
One notable outlier was the Tcf7l1 knockout cell line, which showed a strong differen a on delay phenotype as measured by marker expression, but whose effects in 2i were not correlated to LIF effects. Tcf7l1 (aka Tcf3 ) is a repressor of several pluripotency marker genes including
Esrrb and Nanog (Martello et al. 2012) , and can be inac vated through Wnt/β-catenin
signalling. Tcf7l1 suppression, just like Wnt ac va on (Ohtsuka et al. 2015) , may therefore be independent of and complementary to LIF effects.
Fig 68 : Clusters of the knockout response of LIF-dependent genes in N2B27.
Heatmaps showing average log-fold-changes (KO N2 vs. RC9 N2 ) of LIF N2 gene clusters. Asterisks indicate that the
absolute row-wise z-value of the respective mean expression value was above 1.96 (p ≤ 0.05). For reference, naive marker log-fold-changes from the same comparison are shown in a separate heatmap (column ordering derived only from LIF N2 clusters).
To dissect the knockout response of LIF-dependent genes further, we first selected the 165 genes of the significant LIF response in 2i that were also significantly changed compared to their control in at least one knockout in N2B27. We then carried out cluster analysis using the Ward method as described in previous sec ons. We iden fied four clusters and calculated their mean log-fold-changes ( KO N2 vs. RC9 N2 ) as well as row-wise z-values ( Fig 68 ). Clusters one through
three were regulated in only a subset of knockouts, while cluster four regula on was posi vely correlated to marker expression in most knockouts ( Fig 68 ). For each cluster, we carried out GO enrichment analysis using SETHRO ( Table 8 ).
Cluster
(genes) KO specific regulation Enriched functions
LIF N2 /1
(29)
Up: Csnk1a1, Eed, Suz12 Down: Rbpj
-
LIF N2 /2
(40)
Up: Csnk1a1, Eed embryonic placenta development
LIF N2 /3
(59)
Down: Csnk1a1, Rbpj, Smg5,
Jarid2n, Zfp281, Tsc2
limb morphogenesis, glial cell differen a on
LIF N2 /4
(37)
Up: Csnk1a1, Eed, Suz12, Etv5 regula on of angiogenesis, regula on of vasculature development