Synergistic binding of transcription factors to cell-specific enhancers programs motor neuron identity

Efficient transcriptional programming promises to open new frontiers in regenerative medicine. However, mechanisms by which programming factors transform cell fate are unknown, preventing more rational selection of factors to generate desirable cell types. Three transcription factors, Ngn2, Isl1 and Lhx3, were sufficient to program rapidly and efficiently spinal motor neuron identity when expressed in differentiating mouse embryonic stem cells. Replacement of Lhx3 by Phox2a led to specification of cranial, rather than spinal, motor neurons. Chromatin immunoprecipitation–sequencing analysis of Isl1, Lhx3 and Phox2a binding sites revealed that the two cell fates were programmed by the recruitment of Isl1-Lhx3 and Isl1-Phox2a complexes to distinct genomic locations characterized by a unique grammar of homeodomain binding motifs. Our findings suggest that synergistic interactions among transcription factors determine the specificity of their recruitment to cell type–specific binding sites and illustrate how a single transcription factor can be repurposed to program different cell types.

1 2 2 0 VOLUME 16 | NUMBER 9 | SEPTEMBER 2013 nature neurOSCIenCe a r t I C l e S Hb9 gene, mutations in Isl1 and Lhx3 result in distinct phenotypes, indicating that the two transcription factors may also possess independent functions 16,17 . Thus, the question of whether the NIL factors act primarily synergistically or independently at the genomic level remains unanswered.
To overcome the low efficiency of cell programming that limits biochemical analysis of the process, we established inducible ESC lines that harbor the NIL programming module or a module in which Lhx3 is replaced by the cranial motor neuron determinant Phox2a (the NIP programming module) [18][19][20] . We found that NIL induction in differentiating ESCs resulted in rapid and highly efficient specification of spinal motor neurons and that NIP induction in an identical cellular context programed cranial motor neuron identity. Taking advantage of these robust and efficient programming systems, we mapped genome-wide binding sites of programming factors in both inducible lines. Computational analysis of occupied cis-regulatory elements revealed that Isl1 directly interacts and synergizes with Lhx3 or Phox2a in the relevant cellular contexts. The Isl1-Lhx3 and Isl1-Phox2a heterodimers exhibited different DNA-sequence preferences, forming the basis of cell-specific programming module activities and indicating that synergistic interactions between programming factors underlie specification of alternate motor neuron fates.

Ngn2, Isl1 and Lhx3 program spinal motor neuron fate
To study programming of spinal and cranial motor neuron identity, we generated two doxycycline (Dox) inducible ESC lines 21 : one line harboring a polycistronic expression construct in which the open reading frames of spinal motor neuron determinants Ngn2, Isl1 and Lhx3 (refs. 12-14) are separated by 2A peptides (iNIL line), and a second line in which we replaced Lhx3 with a cranial motor neuron determinant Phox2a (iNIP line) (Fig. 1). NIL factors have been shown to activate specification of motor neuron identity in retinoic acid-treated differentiating ESCs 13,14 . We found that NIL factors were sufficient to induce expression of spinal motor neuron markers even in the absence of retinoic acid. Treatment of differentiating ESCs with Dox resulted in robust induction of the tricistronic transgene 24 h later (Supplementary Fig. 1a,b). Notably, despite continuing Dox treatment, Ngn2 expression was extinguished in most cells by 48 h, consistent with its transient pattern of expression in cells transitioning from progenitors to postmitotic motor neurons 22,23 (Supplementary Fig. 1c).
Ngn2, Isl1 and Phox2a program cranial motor neuron fate Cranial motor neurons share many features with spinal motor neurons: they are cholinergic and express the transcription factor Isl1 (ref. 17). However, unlike ventral spinal motor neurons, they express Tbx20 instead of Hb9, and their specification depends on the pairedlike homeodomain transcription factors Phox2a and Phox2b instead of LIM homeodomain factor Lhx3 (refs. 19,24). Misexpression of the Phox2a and Phox2b transcription factors in the developing spinal cord is sufficient to induce ectopic cranial branchiomotor neurons 18 . However, the expression of Phox2a alone in differentiating ESCs resulted in only a small increase in the number of Isl1-positive cells, and most of the cells failed to acquire neuronal identity, as shown by the lack of Tubβ3 expression (Supplementary Fig. 1d).
We reasoned that joint expression of Ngn2 and Isl1 with Phox2a (NIP module; Fig. 1a) might lead to a more robust and uniform specification of cranial motor neurons. Indeed, iNIP cells treated for 48 h with Dox, dissociated and plated on laminin acquired uniform neuronal morphology and identity (Tubβ3 and NeuN expression) and expressed Phox2b in the absence of Hb9 (Fig. 1c,d and Supplementary Fig. 2a,b). The high efficiency of NIP programming was comparable to that of NIL programming: 99.77% ± 0.22 V5 + cells expressed NeuN, 99.03% ± 0.08 expressed Phox2b and 0.11% ± 0.11 expressed Hb9 were detected (Supplementary Fig. 2a,b). Together, these data indicate that replacement of Lhx3 in the programming module with Phox2a results in efficient specification of neurons that acquired molecular properties of cranial motor neurons.

Functional characterization of induced NIL and NIP neurons
To determine whether transcriptionally programmed cells acquired key properties of mature motor neurons, we cultured induced NIL and NIP cells alone or on monolayers of primary cortical mouse astrocytes for 7-10 d. Immunostaining of NIL and NIP cells cultured on monolayers of astrocytes revealed dense arrays of synapses marked by the synaptic vesicle marker SV2 (Fig. 1f). Notably, many of the synapses exhibited accumulation of vesicular acetylcholine transporter (Vacht), a marker of mature cholinergic synapse (Fig. 1f). Cholinergic identity of NIL and NIP induced cells was further documented by an approximately 70-fold increase in the levels of choline acetyl transferase (Chat) mRNA (Fig. 1e) and by Chat immunostaining (Fig. 1f).
Electrophysiologically mature motor neurons fire trains of action potentials following depolarization 25,26 . Whole-cell patch currentclamp recordings of NIL and NIP induced cells cultured on astrocytes for 7 d revealed that action potentials could be evoked by 20-150-pA, 1-s current injection in all cells tested (12 NIL cells,12 NIP cells). Furthermore, nearly all patched cells (11 of 12 NIL cells, 11 of 12 NIP cells) fired trains of action potentials, sustained for the duration of the depolarizing current step (Fig. 1g). Together, these observations suggest that inducible expression of NIL and NIP programming modules is sufficient to differentiate ESCs into electrically mature cholinergic neurons.
Motor neurons project axons outside of the CNS to innervate peripheral synaptic targets. To examine whether induced motor neurons acquired this defining characteristic, we implanted control, iNIL and iNIP cells treated with Dox from days 2 to 4 of differentiation into the developing cervical and brachial neural tube of developing chick embryos 27,28 . We detected robust outgrowth of axons (labeled by mouse specific NCAM antibody) exiting spinal cord via the ventral root and extending along all major spinal motor nerves 2 d after implantation of iNIL neurons (four of five successfully transplanted embryos; Fig. 1h). In contrast, axons of control transplants stayed in the spinal cord and failed to project to the periphery (Fig. 1h). To further test the specificity of iNIL cell axonal pathfinding, we examined projections of iNIP cells transplanted into the same region of the chick neural tube. Unlike spinal motor neurons, cranial motor neurons do not exit the CNS through the ventral horn, preferring a more dorsal exit point 19 . iNIP axons accumulated selectively at the lateral region of the developing spinal cord, coalescing with spinal accessory nerve npg a r t I C l e S populated by branchiomotor cranial motor axons originating from the spinal accessory nucleus in the lateral cervical spinal cord (four of four successfully transplanted embryos; Fig. 1h) 29 . The same axonal trajectory has been observed for ectopic cranial motor neurons formed in the developing spinal cord following misexpression of Phox2a or Phox2b 18 . These results indicate that induced expression of the NIL and NIP modules programs cell phenotypes that are, by all examined criteria, consistent with spinal and cranial motor neuron identities (induced cranial and spinal motor neurons).
Expression profiles of motor neuron programming Effective programming of ESCs into motor neurons should be accompanied by a repression of the stem cell expression program and induction of the spinal or cranial motor neuron specific transcriptome. Global expression profiling (Affymetrix GeneChIP ST arrays) revealed that 48 h of Dox treatment of iNIL and iNIP cells resulted in a marked change in gene expression profile (3,185 and 1,852 genes were more than twofold differentially expressed following NIL and NIP induction, respectively, P < 0.001; Fig. 2a). Induction of NIL and NIP programming modules extinguished the expression of pluripotency genes (Oct4, Nanog), upregulated generic motor neuron genes (endogenous Isl1, Ebf1/3, Onecut1/2), cholinergic genes (Vacht (also known as Slc18a3), Chrnb4) and genes encoding axon guidance molecules (Nrp1, Robo1/2, Dcc) ( Fig. 2b and Supplementary Fig. 3a). Comparison of iNIL and iNIP induced cells revealed significant differences between the two samples (2,731 differentially expressed more than twofold, P < 0.001; 1,878 genes were upregulated in iNIL cells compared with iNIP, 857 genes were upregulated in iNIP cells compared with iNIL; Fig. 2a). Although spinal motor neuron genes Hb9, Isl2, endogenous Lhx3 and Slit1/2 (ref. 30) were selectively expressed in iNIL cells, iNIP cells upregulated expression of cranial motor neurons markers Tbx20, endogenous Phox2a, Phox2b, Rgs4 and Gal 31-33 (Fig. 2b, Supplementary Fig. 3a). Unsupervised clustering of expression profiles revealed that Dox-treated iNIP cells segregated from Dox-treated iNIL cells (Fig. 2c), indicating that their identities are molecularly distinct. Spinal motor neurons can be generated from ESCs by directed differentiation controlled by patterning signals retinoic acid and sonic hedgehog (Shh) 27 . Retinoic acid-and Shh-driven differentiation recapitulates the normal process of motor neuron development. ESCs first acquire neural progenitor identity on day 3 (Sox1 + , Olig2 − ), followed by motor neuron progenitor stage on day 4 (Olig2 + , Hb9 − ), and differentiate to postmitotic motor neurons (Olig2− -, Hb9 + ) on days 5-6 (refs. 21,27). To follow spinal motor neuron induction, we introduced a GFP transgene driven by the Hb9 promoter into the iNIL cell line 27 . Although control cultures of iNIL cells contained few or no GFP-positive cells, ~40% of the cells treated with retinoic acid and Shh for 4 d, and the majority of cells treated with Dox for 48 h, became GFP positive (Supplementary Fig. 3b).
To examine how closely programmed neurons correspond to retinoic acid-and Shh-generated motor neurons, we compared expression profiles of FACS-purified Hb9-GFP + retinoic acid-and Shh-generated motor neurons on day 5 of differentiation with Hb9-GFP + cells purified from iNIL cultures treated with Dox for 48 h. The induced motor neurons were markedly similar to retinoic acid-and Shh-generated motor neurons (Fig. 2b-d). Most genes (97.4%) were expressed at levels that were not significantly different between the two samples (P < 0.001), and only 1.6% of all genes exhibited divergent expression (that is, are induced in one cell type and repressed in the other). The similarity of induced and control motor neurons is further supported by unsupervised clustering of gene expression profiles (Fig. 2c).
Although key motor neuron-specific genes were correctly regulated, a set of genes controlling rostro-caudal neural identity and motor neuron subtype identity was differentially expressed in retinoic acid-and Shh-generated cells and induced iNIL cells (Fig. 2e). Induced iNIL motor neurons expressed low levels of Hox transcription factors and high levels of rostral neural markers (Otx1, Otx2). To rectify this difference, we asked whether programmed iNIL motor neurons are responsive to the caudalizing signal retinoic acid 27,34 . Treatment of iNIL cells with retinoic acid during the Dox treatment resulted in correct specification of cervical spinal identity, marked by the expression of Hox genes from paralogous groups 4 and 5 and suppression of rostral markers Otx1/2 (Fig. 2e). Thus, programmed cells acquire generic motor neuron identity following induction of NIL factors, but specification of rostro-caudal subtype identity depends on the treatment of the cells with caudalizing patterning signals.
Motor neuron programming bypasses neural progenitor stages Rapid activation of postmitotic motor neuron markers following NIL and NIP induction raised the question of whether transcriptionally programmed cells transit through neural progenitor stages. To capture cells during the transition from ESCs to spinal motor neurons, we profiled induced cells 24 h after Dox treatment. At this time point, NIL factors had effectively repressed key stem cell genes (for example, Oct4, Nanog; Fig. 3a) and had already induced expression of markers associated with postmitotic spinal motor neuron identity, such as Hb9 (Mnx1), Isl2, Lhx4, VAChT (Slc18a3), Robo2, Slit2 and Nrp1 (Fig. 3a).
Neither NIL nor NIP induced expression of genes associated with progenitor stages (Sox1, Olig2 and Ngn2) that were highly expressed at 24 Fig. 3a,c and Supplementary Fig. 3c). These results indicate that programming modules initiate a state transition from ESCs to postmitotic motor neurons that bypasses key steps in the normal motor neuron developmental program.
Isl genome binding is dependent on programming partners Efficient and rapid programming of ESC differentiation into phenotypically distinct neurons by modules that differ only in one transcription factor provides an ideal system in which to study whether individual transcription factors act independently or engage in synergistic interactions. If the individual factors are recruited to DNA independently, replacing Lhx3 with Phox2a in the programming module should not affect the DNA binding preference of Isl1. To test this independent model, we performed chromatin immunoprecipitationsequencing (ChIP-seq) analyses of Isl1 in iNIL and iNIP cells 48 h after Dox induction. Inducible Isl1 was not epitope tagged and we optimized ChIP using a pool of monoclonal antibodies to Isl1. As these antibodies cross-react with both Isl1 and the closely related Isl2, we refer to the data as Isl ChIP-seq.
We observed extensive condition-specific Isl recruitment to genomic loci in the iNIL and iNIP induced cells (Fig. 4a). We identified 18,187 significant Isl binding events (Online Methods) in the two conditions, of which 38% were significantly differentially enriched (P < 0.001) between iNIL and iNIP lines (Fig. 4b). In contrast, only 9.6% of the Isl binding sites were differentially enriched between iNIL cells and retinoic acid-and Shh-derived motor neurons. To further test the synergistic model, we profiled the binding of Isl when ectopically expressed alone in differentiating ESCs. The genomic occupancy of Isl was substantially different from that of Isl expressed in the context of either iNIL or iNIP cells (Supplementary Fig. 4b): 67.5% of the Isl ChIP peaks were differentially enriched between iNIL and iIsl1 cells, and 48.8% were differentially enriched between iNIP and iIsl1 cells. These results indicating that recruitment of Isl1 to DNA binding sites depends on the composition of programming modules. Our findings are consistent with the synergistic model, implicating functional interactions between programming factors.
Next we examined whether identified Isl binding sites are distributed randomly across the genome or whether their positions correlate with tissue specific cis-regulatory elements. We took advantage of ENCODE project data that identified putative regulatory regions in mouse ESCs, whole brain, heart, kidney, liver and spleen, defined using combinations of DNaseI hypersensitivity and enrichment in H3K4me1 and H3K27ac histone modifications 35 . Of all of the tissues examined, Isl binding sites correlated best with whole brain putative regulatory regions (Supplementary Fig. 4a). Notably, the overlap with regulatory regions in ESCs was as low as in unrelated tissues (Supplementary Fig. 4a). These findings indicate that expressed NIL and NIP factors are not passively recruited to existing stem cell regulatory regions, but that these factors actively engage neuronal regulatory regions.

Cell-specific Isl binding correlates with gene expression
The identification of condition-specific Isl binding prompted us to examine whether differentially occupied sites in the iNIL and iNIP cells are associated with the establishment of cell type-specific gene expression profiles. We observed condition-specific Isl binding in the vicinity of developmental genes that were selectively induced by the NIL or NIP programming modules. For example, three sites that were bound by Isl in the iNIL cells, but not in the iNIP cells, were located downstream of the endogenous Lhx3 gene; conversely, two sites that were bound in the iNIP cells, but not in the iNIL cells, were located near the Phox2b transcription start site (TSS) (Fig. 4a). Meanwhile, we observed shared Isl binding sites near a subset of genes that are induced in both cell types, such as Chat (Fig. 4a).  We then asked what fraction of genes differentially expressed between the two conditions were proximal to sites that were differentially occupied by Isl. To extend this analysis, we asked what fraction of genes differentially expressed between the two conditions are proximal to sites that are differentially occupied by Isl. We first subdivided all induced genes that have nearby (overlapping the gene or <10 kbp upstream or downstream from the gene TSS) Isl binding sites into three categories: those that were induced in both cell lines, those that were induced selectively in iNIL cells (NIL induced) and those that were selectively induced in iNIP cells (NIP induced). Similarly, Isl binding sites were subdivided into condition-specific sites that were most differentially enriched (P < 0.001) in iNIL cells (5,285 sites, NIL>NIP), those most differentially enriched in iNIP cells (1, 657 sites, NIP>NIL; Fig. 4b) and condition-independent sites that were similarly enriched in the two cell lines (1,705 sites, NIL = NIP; Online Methods). Of all of the NIL induced genes, 57% had a nearby NIL>NIP site, 26% had a nearby NIL = NIP site and only 13% had a nearby NIP>NIL site (Fig. 4c). Conversely, 70% of genes induced selectively in iNIP cells had a nearby NIP>NIL Isl binding site, 22% had sites similarly occupied in both cell lines and only 14% had a nearby NIL>NIP (Fig. 4c). Based on the correlation between condition-specific Isl binding and condition-specific activation of gene expression, we propose that a subset of Isl binding sites function as context-dependent enhancers contributing to the establishment of the observed cell type-specific pattern of gene expression and to cell fate programming.
Sequence motifs explain differential Isl binding Given that Isl genomic binding depends on the programming module context, we reasoned that Isl1 might partner with different transcription factors during NIL-and NIP-mediated cell fate programming, resulting in a global change in its DNA binding preference. To elucidate the mechanisms underlying differential recruitment of Isl1 to genomic sites, we analyzed the DNA motifs enriched in the conditionspecific and condition-independent binding sites. Motif analysis identified a monomeric sequence with consensus TAAKKR under the condition-independent (NIL = NIP) sites, which is identical to the in vitro binding preference characterized for Isl2 ( Supplementary  Fig. 4c) 36 .
The analysis of differentially enriched sites revealed more complex dimeric motifs composed of a combination of two homeodomain binding sites (Fig. 5a,b). Notably, the motifs associated with iNIL and iNIP condition-specific sites exhibited different motif grammar. Although the homeodomain half-sites formed an inverted repeat in the motif enriched under NIL-specific sites, the motif enriched under NIP-specific sites contained an everted half-site configuration (Fig. 5b). The motifs were highly enriched under NIL-and NIPspecific Isl ChIP-seq peaks, with 60.1% of NIL and 33.5% of NIP peaks containing the NIL-and NIP-specific motif, respectively (2.5 × 10 −23 false discovery rate). The marked specificity of the ordering of homeodomain binding motifs in selectively occupied sites suggests that Isl1 partners with two different homeodomain transcription factors in iNIL and iNIP cell lines. The differences in the structure of these transcription factor complexes likely underlie their sequence specific recruitment to DNA, providing a physical mechanism by which one transcription factor can regulate different targets to establish alternate cellular identities.
Lhx3 and Phox2a co-occupy Isl-binding sites Previous analysis of the spinal motor neuron-specific Hb9 enhancer revealed that Isl1 forms a multimeric complex with Lhx3, Ldb1 and Ngn2 or Neurod4 12 . We asked whether Lhx3 co-occupies other sites selectively bound by Isl in iNIL cell line. Taking advantage of the V5 epitope tag on the Lhx3 transgene 21 , we performed ChIP-seq analysis of Lhx3 binding in the iNIL cells 48 h after Dox induction. We identified 47,908 Lhx3 binding sites in the genome and found that these sites were highly coincidental with the sites occupied by Isl in the iNIL cell line. We observed that only 1.7% of all sites were significantly differentially enriched (P < 0.001) in one experiment compared with the other (Fig. 5a,c). These findings suggest that Isl1 and Lhx3 bind to DNA as an obligatory heterodimer during spinal motor neuron differentiation.
Although there is no prior evidence that Phox2a heterodimerizes with Isl1, we examined whether V5 epitope tagged Phox2a might pair Context-specific Isl1 genome association in iNIL and iNIP cells correlates with differential gene expression. (a) Isl binding at developmentally regulated genes in iNIL and iNIP cells treated with Dox for 48 h. Isl ChIPseq signals over Lhx3, Chat and Phox2b are shown. Blue peaks represent significant (P < 0.01) read enrichment over control. Genomic loci coordinates are shown next to the x axis. (b) Isl genome association was NIL and NIP specific. Shown is a comparison of Isl read enrichment from iNIL and iNIP cells at all detected peaks. Blue represents peaks that were significantly differentially enriched in one experiment over the other (log 2 ; Online Methods). (c) Condition-specific Isl binding was associated with condition-specific gene expression. Differentially expressed genes were divided into ones selectively induced in iNIL and iNIP cells and ones induced in common in both cell types. The bar graph represents the percentage of genes containing a proximal Isl peak that was condition specific (Fig. 2b) for each group. npg a r t I C l e S with Isl1 in the iNIP cell line. Although ChIP-seq analysis revealed only 1,568 significant Phox2a binding events, Phox2a and Isl binding events were highly coincident and the magnitude of ChIP enrichment at the co-bound sites was also highly correlated, mirroring the cobinding of Isl and Lhx3 in iNIL cells. We observed that only 4.6% of all sites were significantly differentially enriched in one experiment compared with the other (Fig. 5a,c). The high degree of co-binding of Isl and Phox2a raised the possibility that the two factors might be parts of the same transcriptional complex. It has been shown that purified Isl1 and Lhx3 transcription factors interact in solution 126 . Co-immunoprecipitation experiments confirmed the Isl1-Lhx3 interaction in induced iNIL cells and revealed that Isl1 and Phox2a are members of the same transcriptional complex in iNIP cells (Fig. 5d).
Together, these results indicate that the alternate cellular fates  ChIP-seq npg a r t I C l e S produced by NIL and NIP programming modules are encoded by cooperative recruitment of Isl-Lhx3-and Isl-Phox2a-containing complexes to enhancers with distinct motif grammar.

DISCUSSION
We exploited the differentiation potential of pluripotent ESCs to study how transcription factor modules control specification of distinct neuronal cell types. Inducible expression of two programming modules differing in one transcription factor led to a rapid and efficient specification of cells expressing key molecular and functional properties of spinal and cranial motor neurons. Isl1 transcription factor changed its genome binding preference when expressed alone or in the context of either the NIL or NIP programming modules. Because the factors were expressed in identical cellular context, the different binding preference of Isl cannot be attributed to differential chromatin accessibility or initial presence of distinct cofactors. Our data support a model in which Isl forms transcriptional complexes with Lhx3 or Phox2a. The complexes are recruited to condition-specific enhancers with differential motif grammar leading to activation of cell type-specific expression programs and to specification of spinal or cranial motor neurons (Supplementary Fig. 5). These findings have broader consequences for the rational design of programming modules, as mapping an individual transcription factor's DNA binding preference is insufficient to predict its binding and its potential for cellular programming when it is coexpressed with other cooperating programming factors. Systematic computational and experimental analysis of interactions among programming factors, along with decoding the grammar of their cooperative binding motifs, will be a fundamental step toward rational design of programming modules for predictable production of diverse cell types of interest. The synergistic nature of the programming module's activity could explain why collections of factors are typically required to program terminal cell fate [2][3][4][5][6][7]37 . It is of interest that Oct4, Klf4 and Sox2 (core module) co-occupy regulatory elements in ESCs [38][39][40] , suggesting that combinatorial programming modules may be a general developmental strategy. A second set of transcription factors (Myc module) appears to operate in parallel to the core module in pluripotent stem cells 40 . We therefore anticipate that additional transcriptional modules besides NIL and NIP will contribute to the establishment of terminal motor neuron expression profiles. Notably, the NIL programming module does not activate expression of Hox transcription factors that control specification of motor neuron subtype identity 41 . This is consistent with our recent demonstration that rostro-caudal patterning signals specify motor neuron positional identity by remodeling Hox chromatin landscape during early neural progenitor stages that are bypassed during direct programming by NIL factors 42 . Thus, generic motor neuron identity can be experimentally uncoupled from the Hox-driven program controlling subtype-specific motor neuron properties. Evolution of a generic motor neuron program that operates in parallel with transcription factors controlling subtype-specific programs would provide a versatile and efficient system for diversification of generic motor neurons into distinct subtypes necessary for the assembly of a functioning motor system.
Currently, the identification of effective programming modules relies on empirical testing of combinations of transcription factors expressed in the target cell type. In contrast, the most effective programming module for specification of motor neuron identity is composed of transcription factors expressed only transiently during the transition from motor neuron progenitor to postmitotic state. We propose that selection of effective programming modules for other types of nerve cells should focus on transcription factors expressed during similar developmental windows. Without doubt, direct programming of cellular identity will have a substantial effect on human stem cell applications 13 . Differentiation of human pluripotent stem cells to neurons is currently relatively inefficient and slow, taking weeks to months of in vitro culture [43][44][45] . Understanding the logic and function of programming modules might not only inform ways to generate cell types refractory to efficient programming by extrinsic patterning signals, but might also substantially accelerate production of homogenous cell populations necessary for human disease modeling, cell-based drug screening and transplantation therapy.

METHODS
Methods and any associated references are available in the online version of the paper.