A central question in biology is whether variability between genetically identical cells exposed to the same culture conditions is largely stochastic or deterministic. Using image-based transcriptomics in millions of single human cells, we find that while variability of cytoplasmic transcript abundance is large, it is for most genes minimally stochastic and can be predicted with multivariate models of the phenotypic state and population context of single cells. Computational multiplexing of these predictive signatures across hundreds of genes revealed a complex regulatory system that controls the observed variability of transcript abundance between individual cells. Mathematical modeling and experimental validation show that nuclear retention and transport of transcripts between the nucleus and the cytoplasm is central to buffering stochastic transcriptional fluctuations in mammalian gene expression. Our work indicates that cellular compartmentalization confines transcriptional noise to the nucleus, thereby preventing it from interfering with the control of single-cell transcript abundance in the cytoplasm.

Genome-wide RNA interference (RNAi) screens have identified near-complete sets of genes involved in cellular processes. However, this methodology has not yet been used to study complex developmental processes in a tissue-specific manner. Here we report the use of a library of Drosophila strains expressing inducible hairpin RNAi constructs to study the Notch signalling pathway during external sensory organ development. We assigned putative loss-of-function phenotypes to 21.2% of the protein-coding Drosophila genes. Using secondary assays, we identified 6 new genes involved in asymmetric cell division and 23 novel genes regulating the Notch signalling pathway. By integrating our phenotypic results with protein interaction data, we constructed a genome-wide, functionally validated interaction network governing Notch signalling and asymmetric cell division. We used clustering algorithms to identify nuclear import pathways and the COP9 signallosome as Notch regulators. Our results show that complex developmental processes can be analysed on a genome-wide level and provide a unique resource for functional annotation of the Drosophila genome.Genome-wide RNAi screens have been performed in cultured cells 1 -4 or by ubiquitous gene silencing in worms 5 , 6 or planarians 7 . To study complicated developmental processes, however, genes need to be inactivated in a tissue-specific manner in intact animals. This has become possible through the creation of a transgenic RNAi library targeting 88% of the Drosophila 8 protein-coding genes. To test the feasibility of this new approach, we focused on the Notch pathway, one of the most important regulators of development 9 , 10 . Notch is activated by binding to its ligands Delta or Serrate. After ligand binding, Notch is cleaved by Presenilin and the intracellular domain acts in the nucleus as a transcriptional co-activator. Europe PMC Funders Group Genome-wide RNAi screenHairpin constructs in the RNAi library are expressed under UAS/GAL4 control 18 . We tested scabrous-GAL4 18 , pannier (pnr)-GAL4 19 and fzIII-GAL4 (also known as MS248-GAL4 or P{GawB}MS248) 20 using a set of 40 RNAi lines targeting 21 genes involved in external sensory organ development (Supplementary Table 1). Consistently, phenotypes were stronger and lethality lower with pnr-GAL4, and this line was selected for large-scale analysis.A total of 20,262 transgenic RNAi lines were screened; these are predicted to target 11,619 of the 14,139 protein-coding genes (82.2%) in release 5.7 of the Drosophila genome 21 . Ten flies each were analysed and phenotypic abnormalities were recorded in a database (http:// bristlescreen.imba.oeaw.ac.at). Because pnr-GAL4 is only expressed in a central region of the notum (Fig. 1b), lateral areas were unaffected and served as internal controls. Phenotypes were described using controlled vocabulary (Fig. 1b, Supplementary Table 2). Phenotypic strength (P x ) was expressed on a scale of 0 (not affected) to 10 (completely affected) as the fraction of the pnr-GAL4 expression a...