4 Regulatory networks of mouse stem cell pluripotency
4.2 Project aims and contribu ons
4.3.9 Func onal clusters of the condi onal knockout response
In contrast, a gene shows a condi onal knockout response if it responds to the medium change differently from the RC9 control (i.e. significant interac on). In total, we found that 2274 genes responded in this manner in at least one KO condi on ( Fig 60 A). We dis nguished two types of interac on: firstly, genes that were also significantly changed only in the respec ve KO 2i vs. RC9 2i
to differences between KO and control in N2B27, and were therefore excluded from further considera on. Secondly, genes that were also significantly changed in at least one KO N2 vs. RC9 N2 comparison. We detected 1340 such genes, which we called the Interac on N2 subset ( Fig 60 A). We then checked how many of the Interac on N2 genes were also significantly regulated during the normal course of differen a on ( Fig 60 B). We found a highly significant overlap of 80% of all
Interaction N2 genes ( Fig 60 B). To stringently evaluate whether any genes were unchanged during
wild type differen a on, we performed an addi onal TOST equivalence test. We could only iden fy 12 genes of the Interaction N2 subset that were changed by less than 20% during normal
differen a on ( abs. log-fold-change ≤ log2(1.2), adj. P-Value ≤ 0.05). This indicates that differen a on delay in knockouts is most likely mediated by the perturba on of the normal re-wiring of differen a on networks.
We then inves gated whether it was possible to further subdivide the Interaction N2 gene subset into dis nct func onal clusters of genes. As with the subset of cons tu vely changed genes, we used Ward’s method to learn the gene cluster hierarchy based on minimiza on of within-cluster variance. Then, from visual inspec on of the cluster dendrogram and the differenced
within-cluster variances, we par oned the Interac on N2 subset into 12 clusters.
Fig 64 : Determining the optimal number of clusters of the Interaction N2 knockout response.
Line chart showing total within-cluster-variances (W, points) for different numbers of clusters k. The differenced values of W, i.e. the change of W from k-1 to k, are represented by triangles.
We expected three general categories of clusters: Firstly, clusters whose expression correlated or an -correlated with the differen a on phenotype (as characterized by marker expression or an -differen a on correla on). Genes classified this way may mirror results from our earlier regression analysis, which also screened for genes linearly dependent on pluripotency marker expression. Secondly, clusters that were en rely uncorrelated with the differen a on
phenotype. This category therefore had the biggest poten al to complement our previous analysis: clusters par ally linked to the differen a on delay phenotype may indicate dis nct mechanisms that affect differen a on in parallel and may thus be used to further characterize subgroups of knockouts.
To summarize cluster expression, we calculated the mean of the interac on log-fold-changes in each cluster; i.e. the differen a on response of each knockout normalized to the wild type response ( Fig 65 ). We also calculated the cluster-wise z-values of each knockout’s mean expression in that cluster ( Fig 65 ). This allowed as to classify knockouts that had an unusually high impact on the respec ve cluster’s expression, i.e. whose mean expression absolute z-value exceeded 1.96 (equivalent to 95% confidence)( Fig 65 ). The strongest knockout-specificity was seen in cluster 10, which was highly up-regulated specifically in Nmnat KO and Aff3 KO ( Fig 65 ). In
others, the distribu on was smoother, indica ng a gradual change in the degree of regula on. This was the case in cluster 3, which is up-regulated to a moderate to high degree in most differen a on-delayed knockouts ( Fig 65 ).
Fig 65 : Clusters of the constitutive knockout response in N2B27.
Heatmaps showing average log-fold-changes (KO N2 vs. RC9 N2 ) of constitutive response 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 constitutive response gene clusters).
SETHRO, recovering informa ve annota ons in every cluster except cluster 10 ( Table 7 ), which consisted of only 26 genes. Cluster 3, which smoothly correlated with naive pluripotency marker changes in N2B27 ( Fig 65 ), was enriched for response to growth factor s mulus and stem cell popula on maintenance and included the naive pluripotency markers themselves ( Table 7 ). Several other pathways strongly connected to pluripotency were enriched in at least one of the clusters (BMP signalling, Wnt signalling, SMAD protein signal transduc on, ERK signalling, MAPK ac vity)( Table 7 ). This confirms the ability of our approach to recover meaningful func onal clusters of pluripotency, both on the gene-level and the knockout-level.
Table 7 : Knockout specifically regulated clusters of the Interaction N2 knockout response.
Cluster
(genes) KO specific regulation Enriched functions
1 (280)
Down: Csnk1a1 TAP binding, glutamate receptor signaling pathway, calcium channel ac vity, nerve development, axon guidance, platelet-derived growth factor receptor signaling pathway, posi ve regula on of extrinsic apopto c signaling pathway, skeletal system morphogenesis, protein tyrosine/serine/threonine phosphatase ac vity, phospha dylinositol bisphosphate binding, …, nega ve regula on of MAP kinase ac vity, regula on of ERK1 and ERK2 cascade
2 (135)
Up: Csnk1a1, Eed SMAD protein signal transduc on, growth factor receptor binding, cellular response to tumor necrosis factor, posi ve regula on of transmembrane receptor protein serine/threonine kinase signaling pathway, regula on of cell-substrate adhesion, sphingolipid metabolic process, nega ve regula on of epithelial cell prolifera on, BMP signaling pathway, pallium development, organonitrogen compound catabolic process
3 (78)
Up: Csnk1a1, Etv5, Fgfr1, Rbpj response to temperature s mulus, nega ve regula on of cellular response to growth factor s mulus, stem cell popula on maintenance, amino acid transport, skin development, morphogenesis of a branching epithelium
4 (96)
Down: Csnk1a1, Jarid2n, Smg5,
Zfp281, Zfp423
glial cell differen a on, cytokine ac vity, anchored component of membrane, cell chemotaxis, B cell ac va on, synap c membrane, striated muscle cell differen a on, regula on of lipid metabolic process
5 (67)
Up: Raf1, Smg5, Trim71, Zfp281 Down: Jarid2n, Msi2
cell chemotaxis, epithelial cell migra on
6 (52)
Up: Aff3, Nmnat2 cytosolic ribosome, structural cons tuent of ribosome, cytosolic large ribosomal subunit, cytosolic small ribosomal subunit, ribosomal small subunit biogenesis, rRNA processing, rRNA binding
7 (99)
Up: Csnk1a1, Eed, Suz12, Zfp281 skeletal system morphogenesis, bone morphogenesis, car lage development, mesenchymal cell development, heart morphogenesis, posi ve regula on of smooth muscle cell prolifera on, embryonic skeletal system morphogenesis, biomineral ssue development, epithelial to mesenchymal transi on, G-protein coupled receptor ac vity, ... , nega ve regula on of Wnt signaling pathway
8 (299)
Up: Csnk1a1, Etv5, Rbpj, Zfp423 heparin binding, embryonic pa ern specifica on, regula on of Notch signaling pathway, stem cell popula on maintenance, secretory granule, response to steroid hormone, stem cell differen a on, neuroepithelial cell differen a on, soma c stem cell popula on
9 (58)
10 (26)
Up: Csnk1a1, Eed, Suz12 Down: Jarid2, Smg5
-
11 (88)
Down: Csnk1a1, Jarid2n, Zfp281,
Zfp423
mesenchymal cell development, posi ve regula on of cell growth, nega ve regula on of neuron differen a on, cellular calcium ion homeostasis
12 (62)
Up: Eed, Fbxw7, Suz12 Down: Smg5, Trim71
second-messenger-mediated signaling