Embryonic stem cell-based mapping of developmental transcriptional programs

Study of developmentally regulated transcription factors by chromatin immunoprecipitation and sequencing (ChIP-seq) faces two major obstacles: availability of ChIP grade antibodies and access to sufficient number of cells. We describe a versatile method for genome-wide analysis of transcription factor binding sites by combining directed differentiation of embryonic stem cells and inducible expression of tagged proteins. We demonstrate its utility by mapping transcription factors involved in motor neuron specification. The study of transcriptional networks provides an opportunity to gain fundamental insight into complex molecular processes that govern cell fate specification and embryonic development. While numerous transcription factors controlling cell differentiation have been functionally characterized, their cell type specific patterns of DNA binding remain largely unknown. The method of choice for genome-wide mapping of transcription factor binding sites is chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) 1. Although powerful, current ChIP-seq technology is limited by two critical factors when applied to developmental studies. First, ChIP-seq profiling demands a large number of cells (20–50 million) separated from other cell types expressing the transcription factor of interest, and second, it requires antibodies with high affinity and specificity that recognize transcription Competing financial interest The authors declare no competing financial interests. Author contribution EOM and GM generated the transcription factor inducible lines and EOM performed phenotypic analysis of the derived lines. EOM, WAW and CAM performed ChIP experiments. MI and MK developed the ICE cell lines and vectors. EOM performed expression analysis. EOM and MC performed the WB and protein binding to immobilized DNA. SM analyzed the ChIP-seq data. EOM, RAY, DKG and HW designed the experiments. EOM, SM and HW wrote the manuscript, DKG revised the manuscript. NIH Public Access Author Manuscript Nat Methods. Author manuscript; available in PMC 2012 June 01. Published in final edited form as: Nat Methods. ; 8(12): 1056–1058. doi:10.1038/nmeth.1775. N IH PA Athor M anscript N IH PA Athor M anscript N IH PA Athor M anscript factors in their native form bound to DNA. To overcome these two hurdles, we combined a versatile system for generating mouse embryonic stem cell (ESC) lines harboring inducible and epitope-tagged transcription factors with directed differentiation of ESCs along defined cellular lineages. This system presents several advantages: 1) the use of tagged transcription factors or DNA binding proteins obviates the need for validated factor-specific antibodies; 2) the use of pluripotent cells allows analysis of any developmental cell lineage; and 3) the inducible expression makes it possible to examine binding of developmentally regulated transcription factors in their correct developmental context, as well as to study tagged transcription factors by gain-of-function analysis. To overcome the inconsistency and inefficiency of classical transgenic ESC line production, we relied on a recently developed inducible cassette exchange (ICE) system 2. The resulting transgenic lines harbor a single copy of the transgene recombined into a defined expressioncompetent locus. To further streamline the generation of inducible cell lines, we introduced Gateway (Invitrogen) landing sites into the shuttle vector and a short epitope tag either at the amino(Flag-Bio) or carboxy-terminus (His-V5) of the protein (Fig. 1a). Because of the high efficiency of all steps, parallel production of multiple inducible tagged lines can be accomplished in as little as three weeks. Differentiation of mouse ESCs to spinal motor neurons yields scalable and largely homogeneous populations of cells mirroring developmentally relevant motor neuron differentiation states in mouse 3. To test our approach, we first investigated genome-wide binding of the bHLH transcription factor Olig2 in motor neuron progenitors (pMNs) 4, a rare population of cells (<1% of spinal cells on e9.5) found in the embryonic ventral spinal cord 5. We generated an inducible Olig2 ESC line in which Olig2 protein is carboxy-terminal tagged with the V5 epitope (iOlig2-V5). To mimic the normal Olig2 pattern of expression, doxycycline (Dox) was administered late on Day 3 and the expression of the transgene was analyzed on Day 4 (Fig. 1b) when cells reach pMN stage. The transgenic Olig2-V5 protein was expressed uniformly in pMNs, exhibited correct nuclear localization, and is expressed at levels ~4 fold higher than native Olig2 (Suppl. Fig. 1a–b). The V5 sequence did not perturb the function of the tagged Olig2-V5 protein. As expected, ectopic expression of Olig2-V5 resulted in the repression of Nkx2.2 in ventral interneuron progenitors (Fig. 1c) 4 and in the repression of Pax6 and Irx3 in dorsal interneuron progenitors (Fig. 1d) 6. Therefore, a tagged version of Olig2 recapitulates in differentiating ESCs the normal function of native Olig2 during spinal cord development 7. To profile Olig2 binding, we induced Olig2-V5 in pMNs and performed a ChIP-seq experiment with an anti-V5 antibody. We observed that Olig2-V5 binds in the proximity of the downregulated genes Irx3, Nkx2.2 and Pax6 (Fig 2a and Suppl. Fig. 1c), indicating that Olig2 specifies pMN identity by direct repression of interneuron transcriptional programs. The overexpression of the Olig2 transgene or the addition of the short tag sequence might affect the genomic binding pattern of the Olig2-V5 protein. For comparison, we therefore performed a ChIP-seq experiment in ESC-derived pMNs with antibodies against the native Olig2 protein in the absence of Dox. The endogenous Olig2 and inducible Olig2-V5 ChIPseq experiments revealed a remarkable level of agreement. The proteins bind to the same regulatory sequences of Irx3, Nkx2.2 and Pax6 (Fig 2a and Suppl. Fig. 1c). As expected for a bHLH transcription factor, motif discovery within the ChIP-enriched sites revealed an Ebox motif consensus (Fig 2b) that is present at 58.8% of Olig2-V5 and 60.4% of native Olig2 binding sites (10% false discovery rate (FDR) motif scoring threshold 8). To determine whether enriched sequences lacking E-box motif represent real binding events we Mazzoni et al. Page 2 Nat Methods. Author manuscript; available in PMC 2012 June 01. N IH PA Athor M anscript N IH PA Athor M anscript N IH PA Athor M anscript employed an in vitro ELISA based DNA-protein interaction assay. We demonstrate that Olig2 transcription factor can be recruited to all tested ChIP-seq identified sequences regardless whether they contain E-box motif or not (Suppl. Fig. 1d), supporting the notion that ChIP-seq data reflect Olig2 binding events. The distribution of binding sites found in both experiments is also highly coincident (Fig. 2c). Comparing the read counts at enriched peaks shows that only 0.2 % and 1.1% are differentially enriched in the native Olig2 and Olig2-V5 ChIP experiments, respectively (Fig. 2c). Globally, the levels of ChIP-seq enrichment are highly correlated between experiments with a Pearson’s correlation coefficient of 0.83, indicating that neither the overexpression of Olig2-V5 in Olig2+ pMNs, nor the addition of an epitope tag, affects Olig2 activity or DNA binding preference. Next we compared the binding site preference of tagged transcription factors in a postmitotic motor neuron stage. We have previously demonstrated that Hoxc9 represses cervical programs and promotes specification of thoracic motor neurons 9; the study of Hoxc9-V5 (iHoxc9-V5) (Fig. 1e) revealed a direct repression of cervical Hox genes 9. We compared binding sites of Cand N-terminally epitope tagged Hoxc9, reasoning that overlapping sites are most likely to reflect native Hoxc9 binding events. We modified the inducible system to accommodate a Flag-Bio (FlagB) amino-terminal tag (Fig. 1a) that can be used for ChIP pull-downs either with anti-Flag antibodies or streptavidin-based purification in combination with the biotinylation enzyme BirA 10. We determined that FlagB tagged Hoxc9 retained its ability to repress cervical Hoxc4 and Hoxa5 genes (Fig. 3e and data not shown). Importantly, the genome-wide binding of the iFlagB-Hoxc9 with anti-Flag antibodies shows a high degree of agreement with the Hoxc9-V5 binding profile. Both Hoxc9 proteins associate with rostral Hox genes regulatory elements, indicating their direct repression (Fig. 2d). At the genomic level, both proteins share an identical sequence preference, depicted by a typical Hox binding primary motif (Fig. 2e). Moreover, 47.1% of the peaks in the V5tagged and 39.1% in the Flag-tagged experiments contain the primary motif at a 10% FDR scoring threshold (Suppl. Fig. 2). Although we estimate that the proportion of ChIP-seq reads located in enriched regions is approximately three times higher in the FlagB-Hoxc9 experiment than in the Hoxc9-V5 experiment, the detected peaks are highly coincident across experiments (Fig. 2f). Out of 22,458, only 156 peaks (0.7%) are differentially enriched in V5 ChIP and 799 peaks (3.6%) are differentially enriched in the Flag experiment (Fig. 2f). We conclude that genomic regions shared between Cand Nterminally tagged ChIP-seq experiments are likely to represent native Hoxc9 binding events. The high degree of overlap between ChIP-seq experiments for one transcription factor contrasts with the binding profiles of two unrelated transcription factors. The comparison of the Olig2-V5 and Hoxc9-V5 ChIP experiments revealed a large fraction of non-overlapping peaks, which is in striking contrast to biological replicates of Hoxc9-V5 ChIP-seq experiments that are virtually indistinguishable (Suppl. Fig 3a–c). Detailed analysis of predicted binding positions by the GPS algorithm 8 in co-bound peaks reveals that Olig2 and Hoxc9 occupy proximal but distinct sites within the peaks (Suppl. Fig 2e and f). Because a typical ChIP-seq peak covers ~200 bp, these experiments might be revealing enhancers that are active in both motor neuron progenitors and postmitotic motor neurons. The system we present here is robust and allows the generation of multiple inducible cell lines in parallel. Of twenty-four generated lines, only three exhibited problems with inducible protein expression, likely due to the inherent toxicity of introduced transgenes (data not shown). While the system is versatile and can be employed to study both progenitors and differentiated cells, we observed that the efficiency and homogeneity of transgene induction declines in postmitotic neurons. Inducing the transgene at late Mazzoni et al. Page 3 Nat Methods. Author manuscript; available in PMC 2012 June 01. N IH PA Athor M anscript N IH PA Athor M anscript N IH PA Athor M anscript progenitor stage results in maintained and homogenous expression in postmitotic neurons, offering a reasonable workaround for this problem. Some transcription factors control their targets in a concentration dependent manner. In those instances, it will be important to first establish the Dox concentration and timing of the treatment that result in desired phenotypes, to ensure that the transcription factor binding studies produce biologically relevant information. In summary, we provide a set of tools for rapid generation of ESC lines and production of unlimited quantities of isogenic differentiated cells that enable identification of developmentally relevant transcription factor binding sites in a genome-wide manner. The cell lines can also be utilized for other biochemical studies, including the isolation and identification of transcription factor binding partners by coimmunoprecipitation followed by mass spectrometry 11. We believe that the combination of these powerful techniques will pave the way to a detailed mechanistic understanding of transcriptional networks that govern mammalian development.