DIGIT is a conserved long noncoding RNA that regulates GSC expression to control definitive endoderm differentiation of embryonic stem cells

Abstract

Long noncoding RNAs (lncRNAs) exhibit diverse functions, including regulation of development. Here we combine genome-wide mapping of SMAD3 occupancy with expression analysis to identify lncRNAs induced by activin signaling during endoderm differentiation of human embryonic stem cells (hESCs). We find that DIGIT is divergent to Goosecoid (GSC) and expressed during endoderm differentiation. Deletion of the SMAD3-occupied enhancer proximal to DIGIT inhibits DIGIT and GSC expression and definitive endoderm differentiation. Disruption of the gene encoding DIGIT and depletion of the DIGIT transcript reveal that DIGIT is required for definitive endoderm differentiation. In addition, we identify the mouse ortholog of DIGIT and show that it is expressed during development and promotes definitive endoderm differentiation of mouse ESCs. DIGIT regulates GSC in trans, and activation of endogenous GSC expression is sufficient to rescue definitive endoderm differentiation in DIGIT-deficient hESCs. Our study defines DIGIT as a conserved noncoding developmental regulator of definitive endoderm.

Graphical abstract

Introduction

Definitive endoderm (DE) differentiation is one of the earliest steps in lineage commitment, leading to development of the gastrointestinal organs, lungs and thymus (Zorn and Wells, 2009). Activin or Nodal signaling, through the transforming growth factor beta (TGF-β) receptors, is the primary event initiating DE differentiation of embryonic stem cells (ESCs) (D’Amour et al., 2005; Kubo, 2004). Activation of the TGF-β receptors leads to phosphorylation of the transcription factors SMAD2 and SMAD3, which translocate from the cytoplasm to the nucleus to regulate gene expression (Massagué et al., 2005). SMAD2/3 tend to occupy unique enhancers in different cell types by associating with key transcription factors that determine cell identity (Mullen et al., 2011). As a result, SMAD2/3 switch locations to occupy many new enhancers during DE differentiation (Brown et al., 2011; Kim et al., 2011). Key mesendoderm and endoderm transcriptional regulators including EOMES, MIXL1, SOX17, and FOXA2 are each direct targets of the activin/SMAD2/3 pathway during differentiation (Brown et al., 2011; D’Amour et al., 2005; Kim et al., 2011).

Long noncoding RNAs (lncRNAs) are increasingly recognized as regulators of development and differentiation. lncRNAs are greater than 200 nucleotides (nt) in length, are often polyadenylated, have the same features as messenger (m) RNAs (Guttman et al., 2009), and can localize to the nucleus or cytoplasm without being translated into proteins (Cabili et al., 2015). Recent studies have identified lncRNAs as key contributors to the specification of lineages derived from all three germ layers including neural (Sauvageau et al., 2013), epidermal (Kretz et al., 2012), cardiovascular (Grote et al., 2013; Klattenhoff et al., 2013), dendritic (Wang et al., 2014), skeletal muscle (Gong et al., 2015), and lung (Herriges et al., 2014).

Although many lncRNAs have been identified and cataloged in human cells (Derrien et al., 2012; Xie et al., 2014), in most cases, their functions have yet to be defined. Two studies have identified lncRNAs induced in human endoderm differentiation (Jiang et al., 2015; Sigova et al., 2013). However, the lncRNAs directly targeted by activin signaling that contribute to regulation of DE differentiation remain unknown. We mapped the genome-wide occupancy of SMAD3 during endoderm differentiation and identified DIGIT (Divergent to GSC, Induced by TGF-β family signaling) as an lncRNA regulated by an enhancer bound by SMAD3 following activin signaling. We find that DIGIT is required for productive DE differentiation of both human and mouse ESCs. DIGIT is divergently transcribed from the mesendoderm regulator Goosecoid (GSC), and DIGIT and GSC transcripts are induced coordinately in the same cells during differentiation. GSC expression is regulated by DIGIT in trans, and the defect in DE differentiation present in DIGIT-deficient hESCs can be rescued by induction of GSC. These results identify DIGIT as an early regulator of DE differentiation for both human and mouse ESCs.

Results

DIGIT is induced by activin signaling

We first defined the lncRNAs that were induced by activin signaling during endoderm differentiation. Chromatin immunoprecipitation and sequencing (ChIP-seq) identified 252 SMAD3 enhancers activated with induction of endoderm (Figures 1A and 1B) (Table S1). Steady-state levels of 1387 lncRNAs were elevated at least two fold (Sigova et al., 2013) with endoderm differentiation, and 14 of these lncRNAs were in close proximity to SMAD3 enhancers, suggesting that they may be direct targets of activin signaling. Of these 14 candidates, four showed transcriptional activation during endoderm differentiation as measured by global run on sequencing (GRO-seq, p value <0.01) (Sigova et al., 2013). Only one lncRNA had an ortholog annotated in the mouse genome (GRCm38/mm10), suggesting possible functional conservation. Transcription of this lncRNA was activated during endoderm differentiation as measured by GRO-seq, together with the divergently transcribed developmental transcription factor GSC (Figure 1C). Formaldehyde-Assisted Isolation of Regulatory Elements (Faire) (Giresi et al., 2007; Simon et al., 2012) showed that DIGIT and GSC are transcribed from bidirectional promoters (Scruggs et al., 2015), characterized by nucleosome depletion between the two transcription start sites (TSSs) during endoderm differentiation (Figure S1A).

Figure 1. DIGIT is divergently transcribed from GSC and is activated by an enhancer bound by SMAD3 during endoderm differentiation.

A) Schematic showing the identification of DIGIT as a candidate lncRNA that regulates endoderm differentiation. ChIP-seq, RNA-seq and GRO-seq analysis were combined to identify four lncRNAs that were directly targeted by activin signaling, and only one lncRNA had a mouse ortholog.

B) ChIP-seq was performed to identify sites of SMAD3 occupancy in hESCs and after 48 hours of endoderm differentiation. The x-axis represents the linear sequence of genomic DNA, and the y-axis represents the relative number of mapped reads. The genomic scale in kilobases (kb) is indicated above the tracks. The site of SMAD3 occupancy is located 5 kb upstream of the DIGIT TSS. The SMAD3 site is enriched for H3K27ac (Tsankov et al. 2015, Figure S1F), which marks active enhancers. The locations of GSC and DIGIT are shown at the bottom of (C).

C) GRO-Seq was analyzed from Sigova el al., 2013, for hESCs (day 0, top) and hESCs differentiated toward endoderm for 48 hours (bottom). Transcription of the Watson (+) strand is indicated in red and transcription of the Crick (−) strand is indicated in green. Arrows show the direction of transcription. The structure of the GSC gene and the predicted structure of the DIGIT gene (labeled polyA RNA-seq) are shown below the tracks. DIGIT was cloned after RACE-PCR to define the 5’ and 3’ ends of the DIGIT transcript (Figure S1B), and the structure of the gene encoding this transcript is shown in black (labeled cloned). The cloned DIGIT transcript is shown for remainder of the manuscript.

D) DIGIT (red) and GSC expression (green) were analyzed by qRT-PCR in hESCs (Day 0) and for the first four days of endoderm differentiation. Fold enrichment is indicated on the y-axis, and error bars represent standard deviation.

E) Single-molecule RNA-FISH was performed for hESCs (Day 0, left) and on day 4 of endoderm differentiation (center). Red probes identify DIGIT and green probes identify GSC mRNA. Nuclei are stained with Hoechst (blue). Each dot represents a transcript, and white arrows indicate two foci of overlapping dots at sites of transcription (Levesque and Raj, 2013). The percentage of transcripts (y-axis) in the nucleus (black) and cytoplasm (white) is shown for DIGIT and GSC (far right).

F) The positions of two gRNAs flanking the enhancer occupied by SMAD3 (black box) are shown. The TSS of DIGIT (red) and GSC (green) are indicated on the left. Arrows connected by dotted lines indicate the location of PCR primers. Following deletion of the region occupied by SMAD3, the PCR product decreases from 580 bp to 130 bp (bottom). Genomic PCR was performed on two independent hESC lines with deletion of the SMAD3 enhancer (EnDel1, EnDel2) and is compared to wild-type hESCs (WT).

G) OCT4 and NANOG are markers of undifferentiated hESCs, while SOX17, FOXA2 and CXCR4 together are markers of DE.

H) Expression of OCT4 (gray) and NANOG (white) was quantified in two hESC controls (Ctrl, AAVS1) and two SMAD3 enhancer deletions (EnDel1, EnDel2). Wild-type hESCs are shown as the first control. hESCs that have undergone genome editing to insert a GFP construct into the AAVS1 locus were used as a second control. Tukey’s multiple comparison statistical test showed no significant difference between the means of any pair.

I) hESCs were differentiated towards endoderm for 4 days. RNA levels (y-axis) were quantified for the indicated genes (x-axis) in the AAVS1 control cells and in hESCs in which the SMAD3 enhancer is deleted (EnDel1). * indicates p <0.05. Analysis of EnDel2 is shown in Figure S1G.

J) Flow cytometry was performed to quantify protein expression of SOX17 (x-axis), FOXA2 and CXCR4 (y-axis) in AAVS1 control hESCs and hESCs containing the SMAD3 enhancer deletion (EnDel1) after 4 days of endoderm differentiation. The percentage of double positive cells is indicated in the gated areas. The percentage of cells expressing all three markers is quantified for control hESCs and hESCs with the SMAD3 enhancer deletion on the right. * indicates p <0.05. Analysis of EnDel2 is shown in Figure S1H.

To validate the genomic features of DIGIT, we generated complementary (c) DNA using polyadenylated (polyA) RNA isolated from differentiating hESCs and performed rapid amplification of cDNA ends (RACE) to define the location of the 5’ cap and the 3’ terminus of the DIGIT transcript (Figure S1B; the full sequence is contained in Supplemental Experimental Procedures). We then cloned the full-length transcript from polyA RNA to confirm expression during endoderm differentiation. This analysis showed that DIGIT is encoded by two exons, as predicted by RNA sequencing (RNA-seq) (Sigova et al., 2013), and defined the 3’ end of exon 2 to be downstream of the 3’ end predicted by RNA-seq (Figure 1C, cloned). We then analyzed DIGIT expression during endoderm differentiation using quantitative reverse transcription PCR (qRT-PCR). Both DIGIT and GSC are induced during the first four days of endoderm differentiation (Figure 1D), and both DIGIT and GSC are highly restricted to endoderm compared to other human tissues (Figures S1C-S1E).

We performed single molecule RNA fluorescent in-situ hybridization (FISH) and found that DIGIT transcripts are primarily retained in the nucleus while GSC mRNA transcripts are most abundant in the cytoplasm (Figure 1E). Quantification of this distribution across 50 cells revealed that about 90% of DIGIT transcripts are retained in the nucleus compared to 15% of GSC transcripts. Nuclear localization of DIGIT provides further support that DIGIT is not an mRNA, and this analysis also revealed that DIGIT and GSC are induced in the same cells during differentiation.

DIGIT and GSC are coordinately induced during endoderm differentiation, and we asked if the proximal enhancer occupied by SMAD3 (Figures 1B and S1F), which is located 5 kb downstream of the DIGIT TSS, regulates DIGIT and GSC activation. We used the CRISPR system (Cong et al., 2013) to delete both copies of the enhancer occupied by SMAD3 during hESC differentiation (Figure 1F). hESCs in which this enhancer was deleted maintain expression of the ESC markers OCT4 and NANOG (Figures 1G and 1H), but show a defect in activation of DIGIT and GSC upon endoderm differentiation (Figures 1I and S1G). Furthermore, deletion of the enhancer occupied by SMAD3 was also associated with a significant reduction in mRNA and protein expression of SOX17, FOXA2 and CXCR4 (Figures 1I, 1J, S1G and S1H), which together identify DE (D’Amour et al., 2005; Green et al., 2011; Loh et al., 2014; Ogawa et al., 2013). Passage-matched hESCs and hESCs containing a green fluorescent protein (GFP) expression system (Sim et al., 2015) were used as controls. We used the CRISPR system to insert the components of the GFP expression system into the AAVS1 loci to create a control that had undergone the same manipulations used to create the SMAD3 enhancer deletions. These findings show that the enhancer newly-occupied by SMAD3 during endoderm differentiation regulates DIGIT and GSC expression and suggest that depletion of DIGIT may inhibit DE differentiation.

DIGIT is a regulator of definitive endoderm differentiation

If DIGIT is required for DE differentiation, we would expect depletion of the DIGIT transcript to inhibit expression of genes that mark the DE fate. We used the CRISPR system to create hESC lines with constitutive expression of short hairpin (sh) RNAs against DIGIT by inserting the shRNA expression cassette along with a drug resistance cassette into the AAVS1 locus. Knockdown of DIGIT expression resulted in a defect in induction of SOX17, FOXA2 and CXCR4 after four days of differentiation compared to controls (Figure 2A), indicating that DE differentiation is compromised. In addition, depletion of DIGIT using locked nucleic acids (LNAs) demonstrated a similar defect in endoderm differentiation (Figure S2A). The LNA experiments were performed on day 2 of endoderm differentiation using two LNA constructs that targeted DIGIT. Transient transfection with LNAs was less effective in depleting DIGIT compared to stable integration of shRNAs, but still showed that reduction in the DIGIT transcript was associated with decreased expression of SOX17 and FOXA2. CXCR4 expression was not assessed because it is not induced on day 2 of endoderm differentiation.

Figure 2. Loss of DIGIT expression inhibits definitive endoderm differentiation.

A) The CRISPR system was used to insert a vector into the AAVS1 locus that expresses an shRNA targeting LACZ (control KD) or an shRNA targeting DIGIT (DIGIT KD). hESCs lines were established from single colonies. Control and DIGIT KD hESCs were differentiated towards endoderm for 4 days prior to RNA analysis. * indicates p <0.05.

B) Insertion of a GFP-polyA (pA) sequence into the gene encoding DIGIT allows transcription at the DIGIT locus while inhbiting production of the DIGIT transcript. Genomic PCR was performed (lower right) using wild-type hESCs (Ctrl) and cells with the GFP cassette knocked into both copies of DIGIT (gfp/gfp) and demonstrates loss of the 180 bp wild-type product and insertion of the GFP-pA cassette (1.3kb). The red line (bottom left) indicates that transcription is terminated after the GFP-pA sequence. Please see Figure S2B and S2C for additional details.

C) GFP mRNA levels (y-axis) were quantified in wild-type hESCs (Ctrl) and DIGIT^gfp/gfp (gfp/gfp) hESCs prior to endoderm differentiation (d0) and after 4 days of differentiation (d4). * indicates p <0.05.

D) RNA-FISH was performed after 4 days of endoderm differentiation for wild-type hESCs (Ctrl, top) and DIGIT^gfp/gfp hESCs (bottom). Cells were probed for RNA encoding GFP (green), DIGIT (red). Nuclei were stained with Hoechst (blue).

E) GFP expression was quantified by flow cytometry in DIGIT^gfp/gfp hESCs prior to endoderm differentiation (Day 0, top) and after 4 days of differentiation (Day 4, bottom). Green fluorescence is shown on the y-axis and forward scatter (FSC) is shown on the x-axis. The percentage of GFP⁺ cells in each condition is indicated.

F) Wild-type hESCs (Ctrl) and two independently derived DIGIT^gfp/gfp lines were differentiated towards endoderm for 4 days prior to RNA analysis. RNA levels (y-axis) are shown for the indicated genes (x-axis). * indicates p <0.05.

G) Wildtype (Control), DIGIT^gfp/gfp 1 and DIGIT^gfp/gfp 2 cells were differentiated towards endoderm for 4 days prior to analysis by flow cytometry. The percentage of double positive cells is indicated for each condition. The percentage of cells expressing all three endoderm markers is shown on the far right. * indicates p <0.05.

H) Wildtype hESCs, DIGIT^gfp/gfp 1 and DIGIT^gfp/gfp 2 hESCs were differentiated towards endoderm for 4 days prior to preparation of RNA-seq libraries. Two replicates for each condition were analyzed. Heat maps display genes that change in expression by at least two fold (FDR <0.05).

I) RNA-seq data were analyzed to quantify expression of the genes repressed (top) and activated (bottom) with loss of DIGIT expression (Figure 2H). RNA-seq datasets from hESCs, hESCs undergoing endoderm differentiation, hESCs differentiated towards cardiomyocytes, and hESCs differentiated towards neural progenitor cells (NPCs) are displayed (Jiang et al., 2015; Loh et al., 2014; Palpant et al., 2015; Schwartz et al., 2015).

J) hESCs were harvested after EB formation (Day 0) and 12 days later. RNA levels (y-axis) were quantified for OCT4 (marker of pluripotency) and SOX17 and Albumin (markers of endoderm). hESCs that underwent CRISPR targeting to insert a GFP cassette into the AAVS1 locus were used as controls. * indicates p <0.05.

To further investigate the role of DIGIT in DE differentiation, we asked how disruption of the gene encoding DIGIT affects differentiation. We used the CRISPR system to insert a sequence encoding GFP followed by a polyA signal 44 base pairs (bp) downstream of the DIGIT TSS (Figure 2B). This sequence was inserted without deleting endogenous DNA to avoid inadvertently removing regulatory elements that might affect nearby genes. Two drug selection markers were used to identify colonies with GFP-polyA insertions in both DIGIT alleles. hESC lines were expanded from single colonies and then transfected with a plasmid expressing Cre recombinase to remove the drug resistance cassettes (Figures S2B and S2C). Single cells were sorted and expanded to establish DIGIT-deficient (DIGIT^gfp/gfp) hESC lines. The insertion of GFP-polyA allows activation of transcription of GFP at the DIGIT locus during endoderm differentiation but leads to termination at the polyA signal, preventing expression of the full-length DIGIT transcript. Analysis of RNA expression by PCR and RNA-FISH demonstrates that GFP transcripts are induced during endoderm differentiation (Figures 2C and 2D) and are translated into protein (Figure 2E). The presence of GFP transcripts and the dramatic reduction of DIGIT (Figure 2D) indicate that transcription is activated at the DIGIT locus with little production of the DIGIT transcript.

DIGIT^gfp/gfp hESC lines maintain expression of hESC master regulators (Figure S2D) (Boyer et al., 2005). However, these cells show significantly reduced levels of DIGIT when differentiated towards endoderm and also demonstrate a defect in the induction of SOX17, FOXA2 and CXCR4 mRNA (Figure 2F). Reduced DIGIT expression also inhibited activation of DEANR1, an lncRNA recently described to regulate FOXA2 expression (Figure S2E) (Jiang et al., 2015). The primers detecting the DIGIT transcript are located downstream of the GFP-polyA insertion, and the low level of DIGIT still detected in DIGIT^gfp/gfp cells (Figure 2F) is likely the result of a low frequency of read through of the polyA sequence.

hESCs were also analyzed by flow cytometry after four days of endoderm differentiation (Figure 2G). Forty-eight percent of wild-type hESCs expressed SOX17, FOXA2 and CXCR4, which together identify DE. In contrast, only four and twenty-seven percent of two independent DIGIT^gfp/gfp cell lines co-express the DE markers (Figure 2G, right). DIGIT^gfp/gfp hESCs that do differentiate into DE maintain expression of SOX17 and FOXA2 at levels comparable to wildtype cells (Figure S2F). Thus, DIGIT appears to regulate the fraction of cells differentiating into DE but not the level of FOXA2 and SOX17 proteins in DE cells. We also quantified DE by another set of markers and found that reduced expression of DIGIT led to reduced expression of DE (58-64% reduced to 3-7%), as quantified by cells co-expression of c-KIT and CXCR4 (Figure S2G) (Green et al., 2011; Jiang et al., 2013; Nostro et al., 2011; Ogawa et al., 2013). In contrast to genes that together identify DE, reduced DIGIT levels did not have a significant effect on expression of SOX7, a marker of primitive and visceral endoderm (Kanai-Azuma et al., 2002), PLAT(T-PA), a marker of parietal endoderm (Cheng and Grabel, 1997), or MEOX1, a marker of mesoderm (Candia et al., 1992) (Figure S2H), suggesting that the activity of DIGIT is specific for DE.

We performed RNA-seq to determine the effect of depletion of DIGIT on the transcriptome during endoderm differentiation. Compared to wild-type controls, 225 genes were repressed and 45 genes were induced in DIGIT^gfp/gfp cells after four days of differentiation (Figure 2H and Table S1). Sequencing was performed in duplicate for DIGIT^gfp/gfp 1 and DIGIT^gfp/gfp 2 cells and compared to control hESCs. The genes that were repressed with depletion of DIGIT tend to be induced in endoderm differentiation compared to differentiation into ectoderm or mesoderm lineages (Figure 2I, top). The small number of genes induced with loss of DIGIT were enriched in neural progenitor cells (NPCs) (Figure 2I, bottom, p-value < 0.0018).

To assess the role of DIGIT in non-directed differentiation (Osafune et al., 2008), we created embryoid bodies (EBs) (Figure S2I) using DIGIT^gfp/gfp 1 and DIGIT^gfp/gfp 2 hESC lines and one control hESC line matched to the same passage number. EBs were allowed to spontaneously differentiate for 12 days. OCT4 expression was similar between all groups on day 0 and decreased dramatically by day 12. The endoderm markers SOX17 and Albumin are induced in wild-type cells on day 12, but this induction was inhibited in DIGIT^gfp/gfp hESCs (Figure 2J). Expression of PAX6 and MEOX1, which mark ectoderm and mesoderm, respectively were unchanged between all groups (Figure S2J). Overall, these results reveal that the DIGIT transcript is required for both directed and spontaneous differentiation of hESCs towards DE.

Definitive endoderm differentiation is regulated by an ortholog of DIGIT in mice

We next asked if DIGIT is conserved in mammalian development. We identified a 3002 nt mouse transcript divergent to Gsc that was isolated from day 9.5 embryos (Gm10000) and annotated to contain a 357 nt open reading frame (ORF). We performed RACE-PCR and then cloned the full-length lncRNA from polyA RNA harvested from mESCs that were differentiated towards endoderm for five days. This analysis confirmed a single exon transcript that contained a 5’ cap and 3’ polyA signal (Figure 3A; the full sequence is contained in Supplemental Experimental Procedures).

Figure 3. An ortholog of DIGIT is expressed in mouse and regulates definitive endoderm differentiation.

A) Digit (Gm10000) is divergent to Gsc in mouse. The 5’ and 3’ ends of the Digit transcript were defined by RACE-PCR, and the sequence of the 5’ and 3’ ends of the transcript are shown below. The full-length transcript cloned from differentiating mESCs using polyA RNA is shown in black.

B) Digit levels were quantified in mESCs (Day 0) and mESCs that were differentiated towards endoderm for 5 days.

C) BLAST alignment of the sequences from the human and mouse DIGIT orthologs was performed. Yellow areas mark the region of homology in which 294 of 335 nt in DIGIT are conserved with Digit.

D) RNA was extracted from mouse embryos at the indicated developmental stage. RT-PCR was performed to detect Gsc and Digit. ActB was used as a loading control. A water blank, serving as a negative control is on the far right.

E) RNA was extracted from nuclear (gray) and cytoplasmic fractions (white) on day 5 of differentiation. Digit expression in each fraction was quantified in comparison to Gapdh mRNA.

F) Digit is annotated with a 357 nt ORF. A sequence encoding hemagglutinin (HA) was inserted in frame in the annotated Digit and hemoglobin (Hb) ORFs. Plasmids containing the HA-tagged transgenes were transiently transfected into HEK 293T cells. HA expression was quantified by Western blot. The Digit product is predicted to be 17kD. ACTIN is shown as a loading control. RNA levels of Digit and Hb were comparable after transfection (Figure S3D).

G) Plasmids encoding Digit-HA and Hb-HA were transfected into mESCs 48 hrs prior to analysis by IF microscopy. HA is shown in red, and nuclear staining is shown in blue (Hoechst).

H) Wildtype (control mESC) and Digit^gfp/gfp mESCs were differentiated towards endoderm for 5 days prior to analysis by IF microscopy. Images show nuclear staining (HOECHST, top) and expression of Foxa2 (middle) and Sox17 (bottom). Digit^gfp/gfp mESCs were created by inserting a GFP-pA sequence to disrupt Digit transcription, and these experiments were performed before drug selection markers were removed (Figure S3E).

I) Wildtype mESCs (Ctrl) and two independently derived Digit^gfp/gfp lines were differentiated towards endoderm for 5 days prior to RNA analysis. RNA levels (y-axis) are shown for the indicated genes (x-axis). These experiments were performed after transient transfection with a plasmid expressing Cre recombinase to remove the drug resistance cassettes.

It was unclear if this transcript was induced during endoderm differentiation, if it was a coding or noncoding transcript, or if it shared conserved function with DIGIT. We found that mouse Digit was induced after 5 days of endoderm differentiation compared to mESCs (Day 0) (Figure 3B). Similar to human transcripts, mouse Digit and Gsc are also highly restricted in expression to endoderm compared to other tissues (Figures S3A and S3B). Analysis of nucleic acid sequence conservation between the human and mouse orthologs also revealed a 335 nucleotide sequence with 88% identity (Figure 3C). These results show that both DIGIT and GSC are induced during endoderm differentiation of hESCs and mESCs. We then isolated RNA from mouse embryos from embryonic day (E) 5.5 to 8.5 (Figure S3C) to quantify Digit and Gsc expression during in vivo development. RT-PCR results show that both Digit and Gsc are induced on E6.5, which corresponds to formation of the anterior primitive streak (Lawson et al., 1991). The highest levels of Digit expression are reached on E7.5, corresponding to formation of definitive endoderm (Tam and Beddington, 1987) and continue to be expressed on E8.5 (Figure 3D).

We then asked if the Digit transcript encoded a protein. First, we prepared nuclear and cytoplasmic extracts from mESCs differentiated towards endoderm for 5 days and found that Digit was retained in the nucleus (Figure 3E), which suggests that Digit is not an mRNA despite containing an annotated ORF. Next, we assessed the coding potency of the annotated ORF. CPAT analysis (Wang et al., 2013) provided further evidence that Digit is not a protein-coding transcript (p < 0.08). Finally, we tried to detect protein encoded by the annotated ORF. We generated a transgene in which sequence encoding a hemagglutinin (HA) tag was inserted in frame within the annotated ORF immediately upstream of the annotated stop codon. As a control, a transgene was created in which sequence encoding an HA-tag was inserted at 3'end of mouse Hemoglobin (Hb) cDNA immediately upstream of the stop codon. The Hb cDNA was chosen because it encodes a protein of similar molecular weight to that predicted for Digit.

Ectopic expression of these transgenes in HEK 293T cells resulted in production of RNA for both transgenes (Figure S3D), but Western blot analysis detected the HA tag only in cells expressing the Hb transgene and not in cells expressing the Digit transgene (Figure 3F). It is possible that HEK 293T cells may not express all the factors necessary to promote translation of a mouse transcript, so we transfected mESCs with the same constructs. We could detect the Hb-HA protein by Immunofluorescence (IF) microscopy, but not the digit-HA product (Figure 3G). These results establish Digit as an lncRNA that shares a conserved genomic location and pattern of expression with DIGIT.

We applied the GFP-polyA knock-in strategy previously used for premature termination of DIGIT transcription in hESCs to determine if Digit is required for DE differentiation in mESCs (Figure S3E). Mouse Digit was also required for DE differentiation, as Digit^gfp/gfp mESCs were deficient in induction of Sox17, Foxa2 and Cxcr4 compared to controls (Figures 3H and 3I). Thus, DIGIT is an lncRNA with conserved function in differentiation of mammalian ESCs.

DIGIT contributes to definitive endoderm differentiation through regulation of GSC expression

Divergent lncRNAs can positively and negatively regulate neighboring protein-coding genes (Guil and Esteller, 2012), and we asked if GSC expression is dependent on DIGIT. We quantified expression of GSC in hESCs during endoderm differentiation with depletion of DIGIT by shRNAs and LNAs and found that depletion of DIGIT was also associated with reduced GSC expression (Figures 4A left and S4A). Furthermore, DIGIT^gfp/gfp hESCs (Figure 4A, right) and Digit^gfp/gfp mESCs (Figure S4B) also showed reduced induction of GSC during endoderm differentiation.

Figure 4. DIGIT regulates definitive endoderm differentiation through control of GSC in hESCs.

A) GSC mRNA levels were quantified in hESCs expressing shRNA targeting GFP (Ctrl shRNA) and shRNA targeting DIGIT (DIGIT shRNA) after 4 days of endoderm differentiation (left). DIGIT expression is shown in Figure 2A. GSC mRNA levels were quantified in DIGIT^gfp/gfp 1 and DIGIT^gfp/gfp 2 hESC lines and compared to expression in wild-type hESCs after 4 days of differentiation (right). DIGIT expression is shown in Figure 2F. * indicates p< 0.05.

B) RNA levels were quantified for the indicated genes after 4 days of endoderm differentiation in hESCs containing an shRNA recognizing LACZ inserted into the AAVS1 locus (Ctrl shRNA) and hESCs containing an shRNA targeting GSC inserted into the AAVS1 locus (GSC shRNA). * indicates p< 0.05.

C) The location of the gRNA used to induce GSC expression with dCas9-VPR.

D) hESCs were transfected with a plasmid expressing the gRNA (in C) or a control gRNA recognizing LACZ along with a second plasmid expressing dCas9-VPR. GSC (left) and DIGIT expression (right) was quantified by qRT-PCR after 48 hours. * indicates p< 0.05.

E) The positions of two gRNAs flanking the second exon of DIGIT and used to create DIGIT^−/− hESCs are shown.

F) Control (gray) and DIGIT^−/− hESCs (white) were differentiated towards endoderm for 4 days and DE differentiation is quantified by co-expression of FOXA2, SOX17 and CXCR4 by flow cytometry (left). Control and DIGIT^−/− hESCs were differentiated towards endoderm for 2 days prior to analysis of RNA expression (right). * indicates p< 0.05.

G) DIGIT ^−/− hESCs were transiently transfected with dCas9-VPR and the gRNA targeting the GSC promoter (white) or a LACZ gRNA, which does not target the GSC promoter (gray). hESCs were maintained in mTESR1 for 2 days and then differentiated towards endoderm for 2 days before analysis of RNA expression. * indicates p< 0.05.

H) hESCs were transfected with a plasmid encoding DIGIT. RNA-FISH was performed after 2 days and shows nuclear localization (Hoechst, blue) of DIGIT (red) with transient transfection.

I) DIGIT^−/− hESCs were transfected with a plasmid expressing GFP and either DIGIT or a scrambled sequence of DIGIT (scrambled). hESCs were transfected and GFP⁺ cells were sorted after 2 days of endoderm differentiation. RNA expression was analyzed for the indicated genes. * indicates p< 0.05.

J) RNA-FISH was performed after 4 days of endoderm differentiation for wild-type hESCs (control) and DIGIT^+/gfp hESCs (DIGIT het). Cells were probed for RNA encoding GSC. The transcription sites of GSC were counted in nuclei with either single or double sites of transcription for both wild-type and DIGIT^+/gfp hESCs. Fisher exact test showed no significant difference between the ratio of single to double transcription sites between the two genotypes (P-value = 1).

K) RNA-FISH was performed after 4 days of endoderm differentiation for DIGIT^+/gfp hESCs. Cells were probed for RNA encoding GSC (green), DIGIT (red). Nuclei were stained with Hoechst (blue).

GSC is a homeobox gene induced by activin/Nodal signaling in early gastrulation in Xenopus and the primitive streak in mammals (Blum et al., 1992; Cho et al., 1991), but its role in hESC differentiation is not fully understood. We used an shRNA to deplete GSC (Figure 4B) and determine how depletion of GSC affected differentiation of DE. Depletion of GSC expression inhibited SOX17, as previously described (Kalisz et al., 2012) as well as FOXA2 and CXCR4, which together mark DE. The results suggest that GSC is required for formation of DE during hESC differentiation.

The DIGIT^gfp/gfp phenotype is rescued by induction of endogenous GSC

If DIGIT contributes to DE differentiation by regulating GSC, then induction of GSC in DIGIT-deficient hESCs should rescue the defect observed during endoderm differentiation. We first transfected hESCs with a plasmid encoding a dead Cas9 fused to the VP64-P65-Rta transcription activation domain (dCas9-VPR) (Chavez et al., 2015) and a plasmid encoding a gRNA that directs dCas9-VPR to the GSC promoter (Figure 4C) to determine if recruiting dCas9-VPR to the GSC promoter could activate GSC expression. hESCs transfected with both plasmids showed a nearly 200-fold induction of GSC after 48 hours (Figure 4D, left). dCas9-VPR was directed to bind 102 bp upstream of the GSC TSS, but this was also 376 bp upstream of the DIGIT TSS (Figure 4C), and targeting dCas9-VPR to this location also increased DIGIT expression, but to a lower extent (Figure 4D, right). These results showed that dCas9-VPR primarily activated GSC, but also increased expression of DIGIT. DIGIT^gfp/gfp hESCs rely on a polyA signal to terminate transcription and prevent production of DIGIT. We were concerned that increased transcription at the DIGIT locus in DIGIT^gfp/gfp cells would lead to a proportional increase in the small number of transcripts reading through the polyA and complicate interpretation of the experiment. To avoid this effect, we used the CRISPR system to generate hESC lines with homozygous deletion of the second exon of the DIGIT (Figure 4E and S4C). We then confirmed that these DIGIT ^−/− hESCs do not express DIGIT during endoderm differentiation (Figure S4D) and are also deficient in DE differentiation (Figure 4F). We next transiently transfected DIGIT ^−/− hESCs with plasmids expressing dCas9-VPR and GSC gRNA. We cultured cells in hESCs media for two days to allow expression of GSC prior to induction of differentiation. Because these were transient transfections, we analyzed gene expression after 2 days of endoderm differentiation and found that the DE markers SOX17 and FOXA2 were induced in DIGIT ^−/− cells with activation of GSC (Figure 4G). We repeated the experiments in DIGIT^gfp/gfp hESCs with the same results (Figures S4E and S4F). These findings demonstrate that activation of endogenous GSC can rescue the defect in DE differentiation created by loss of DIGIT expression and suggest that DIGIT controls DE differentiation of hESCs, at least in part, through regulation of GSC.

DIGIT regulates GSC in trans

Depletion of DIGIT by shRNAs, LNAs, and insertion of polyA termination sequences shows that it is the DIGIT transcript and not transcription at the DIGIT locus that is required to regulate GSC expression. To provide further evidence that the DIGIT transcript regulates GSC expression, we asked if ectopic expression of DIGIT could induce GSC expression in DIGIT-deficient cells. We first transfected hESCs with a plasmid expressing DIGIT behind a PGK promoter and performed RNA-FISH to confirm that DIGIT expressed from the plasmid is localized to the nucleus (Figure 4H). Next, we transiently transfected DIGIT ^−/− hESCs with a plasmid expressing DIGIT or a scrambled DIGIT sequence along with GFP. One day after transfection, hESCs were differentiated towards endoderm, and GFP+ cells were sorted after 2 days of differentiation. We found that DIGIT ^−/− hESCs expressing ectopic DIGIT (Figure 4I, left) showed increased expression of GSC as well as FOXA2 and SOX17 (Figure 4I, right). DIGIT ^−/− cells were used for this experiment instead of DIGIT^gfp/gfp hESCs because they did not already express GFP, and expression analysis was performed on day 2 of differentiation because these were transient transfections. CXCR4 expression was not assessed, as it is not induced on day 2 of endoderm differentiation. These results show that DIGIT does not need to be transcribed adjacent to GSC to promote GSC expression and provides further evidence that DIGIT regulates GSC expression.

The results from ectopic expression of DIGIT suggest that the DIGIT transcript may function in trans to regulate GSC. To further evaluate this possibility, we created DIGIT heterozygous (^+/gfp) hESCs and asked if GSC was expressed from both alleles or only the allele adjacent to the wildtype copy of DIGIT. We performed RNA-FISH to analyze GSC expression in wildtype and DIGIT^+/gfp hESCs undergoing endoderm differentiation (Figure 4J). Sites of GSC transcription are identified by bright foci of GSC RNA (Figure 1E) (Levesque and Raj, 2013) and can be observed at one or two sites in differentiating cells. The distribution of cells demonstrating one or two foci to mark active GSC transcription are not statistically different between wildtype and DIGIT^+/gfp cells, and DIGIT^+/gfp cells show a similar frequency of cells expressing GSC from two alleles compared to wildtype cells (Figure 4J). These results suggest that transcription can occur at both GSC alleles even when one allele is no longer producing DIGIT (Figure 4K). Together, the results from ectopic expression of DIGIT in DIGIT ^−/− cells and the expression of GSC from both alleles in DIGIT heterozygous cells suggest that the DIGIT transcript can function in trans to regulate GSC expression.

Discussion

In this study we performed genome-wide analysis of SMAD3 occupancy to identify enhancers targeted by activin signaling during endoderm differentiation. This approach combined with previous analysis of transcriptional responses (Sigova et al., 2013) allowed us to identify DIGIT out of over 1000 lncRNAs induced during endoderm differentiation. Deletion of the SMAD3-occupied enhancer proximal to DIGIT showed a profound reduction in DIGIT and GSC expression and inhibition of DE differentiation. These results show that the enhancer is required for normal activation of DIGIT and GSC, but targeted disruption of SMAD3 binding elements within this enhancer will be required to determine whether or not SMAD3 occupancy is required for activation. Our analysis defined 252 sites activated for SMAD3 binding within 48 hours of endoderm differentiation, and it is likely that investigation of these sites will identify additional enhancers that are also required for endoderm differentiation. Applying this concept more broadly suggests that focusing on lncRNAs that are direct targets of the signaling pathways that control differentiation will identify the lncRNAs that are likely to regulate development.

DIGIT is expressed divergently from GSC and is coordinately induced with GSC during endoderm differentiation when analyzed at the population level by RNA expression (Figure 1D) (Sigova et al., 2013), and expression of both genes are highly restricted to endoderm differentiation (Figures S1C-S1E and S3A-S3B). We also observe both DIGIT and GSC transcripts in the same cells during differentiation (Figure 1E). In other cell types, divergent genes are not always associated with co-expression (Cabili et al., 2015). This coordinated induction of divergent lncRNA and coding gene pairs may be more common with activation of loci during differentiation (Lepoivre et al., 2013; Ponjavic et al., 2009; Sigova et al., 2013) where sets of genes are turned on in a coordinated fashion to regulate changes in cell identity and may be less fixed in cells maintaining homeostasis.

DIGIT is not only divergently transcribed from the gene encoding GSC, it also regulates GSC expression. Depletion of the DIGIT transcript inhibited induction of GSC during endoderm differentiation of both human and mouse ESCs (Figures 4A, S4A-B), and ectopic expression of DIGIT in DIGIT ^−/− cells was also sufficient to induce GSC expression (Figure 4I). These findings, coupled with the observation that GSC can be expressed from both alleles in DIGIT heterozygous cells, show that DIGIT regulates GSC in trans rather than by being transcribed in close proximity to GSC.

GSC is a homeobox gene that is induced in early gastrulation with formation of the dorsal lip of the blastopore in Xenopus and the primitive streak in mammals, and is activated as a target of activin/Nodal signaling (Blum et al., 1992; Cho et al., 1991). Surprisingly, GSC^−/− mice do not show a defect in gastrulation and survive to gestation with craniofacial defects (Rivera-Pérez et al., 1995). Loss of GSC expression in hESCs was previously show to be associated with significant reduction of SOX17 (Kalisz et al., 2012), which is expressed in both definitive and visceral endoderm (Shimoda et al., 2007). By quantifying the effects of GSC on additional genes, our results provide more convincing data that GSC is required for DE differentiation of hESCs. Thus, while loss of GSC does not appear to be required for in vivo gastrulation in mice, we demonstrate that GSC is required for DE of ESCs. Furthermore, the defect in DE differentiation that occurs with loss of DIGIT expression can be rescued by induction of GSC (Figure 4G), showing that DIGIT controls DE differentiation by promoting expression of GSC. Further experiments will be required to determine the function of Digit during in vivo development.

There are increasing examples of lncRNAs that positively regulate proximal protein-coding genes, including those divergently transcribed from developmental genes (Arnes et al., 2016; Herriges et al., 2014; Jiang et al., 2015; Luo et al., 2016), making it necessary to have genomic tools to disrupt lncRNAs without genomic deletions that could affect regulatory elements of neighboring genes. We have established a method for efficient loss of lncRNA function via insertion of a GFP-polyA sequence into the lncRNA gene. Whether transcription initiation at divergently transcribed genes is mediated through bidirectional or adjacent unidirectional promoters (Core et al., 2014; Duttke et al., 2015; Grzechnik et al., 2014), the ability to allow transcription initiation and elongation at the endogenous lncRNA TSS while preventing production of the lncRNA transcript provides an essential tool to dissect lncRNA function. This system can be applied to many divergently transcribed lncRNAs or those in close proximity to protein-coding genes and allows preservation of transcriptional activation at the lncRNA locus while preventing production of the lncRNA product. It also avoids the need to delete any endogenous sequences that might act as a regulatory element of nearby genes. This approach has led to the identification of DIGIT as a key regulator of human and mouse DE differentiation and can be applied to investigate the function of lncRNAs in any developmental system.

Experimental Procedures

ESC culture and differentiation

hESCs were cultured and differentiated as previously described (Sigova et al., 2013). mESCs were cultured on irradiated MEFs and differentiated with 100 ng/ml Activin A. Please see Supplemental Experimental Procedures for additional details.

Sequencing Analysis

ChIP-seq libraries were prepared and analyzed as described (Mullen et al., 2011). GRO-seq (Sigova et al., 2013) analysis was performed using HTseq (Anders et al., 2014) and DESeq2 (Love et al., 2014). RNA-seq analysis of wild-type and DIGIT^gfp/gfp hESCs and comparison to other data sets of hESC differentiation was performed using HTseq and DESeq2. Please see Supplemental Experimental Procedures for additional details.

Genome editing

Genome editing was performed by transfecting cells with the px330 plasmid (Cong et al., 2013) containing the indicated gRNAs along with the indicated homology plasmids. The sequences of all gRNAs are listed in Table S2. Please see Supplemental Experimental Procedures for additional information including maps of the homology plasmids.

RNA in-situ hybridization

Custom Stellaris® FISH probes (Biosearch Technologies) were designed against DIGIT, GSC and GFP using the Stellaris® FISH Probe Designer (www.biosearchtech.com/stellarisdesigner) (Table S2). The samples were hybridized with Stellaris FISH Probe sets labeled with Quasar570 or Quasar670.

Cell sorting and flow cytometry analysis

hESCs and differentiated cells were separated into single cells by Accutase treatment prior to cell sorting using a FacsAria II (BD) or flow cytometry using a using an Accuri C6 (BD). Antibodies used for FACS: SOX17-PerCP (BD 562387) FOXA2-PE antibody (BD 561589) CXCR4-APC antibody (R&D Systems FAB173A). Please see Supplemental Experimental Procedures for additional details.

Analysis of RNA expression in developing embryos

Female CD1 mice were mated with male CD1 mice and checked daily for vaginal plugs indicating embryonic day 0. Pregnant mice were euthanized 5-8 days later, and embryos were dissected, carefully staged, and imaged (stages indicated in Figure S3C).

Total RNA was isolated using the Roche High Pure RNA Isolation Kit (Roche 11828665001) following the manufacturer’s protocol with the addition of a second DNAse treatment following the first wash. cDNA synthesis was performed using SuperScript II reverse transcriptase (Invitrogen 18064014) with both oligo(dT) primers (Promega C1181) and random primers (Promega C1101). RT-PCR was performed though 35 cycles at 94 degrees C for 30 seconds, 60 degrees C for 30 seconds, and 72 degrees C for 30 seconds. Products were electrophoresed on 2% agarose gel.

Immunofluorescence microscopy

IF was performed using the following antibodies: Oct4 (Abcam, ab19857), Cxcr4 (BD, 551852), FoxA2 (Abcam, ab40874), Sox17 (R&D, AF1924), HA (Abcam, ab9110). Please see Supplemental Experimental Procedures for additional details.

Induction of endogenous GSC expression.

A plasmid expressing dCas9-VPR (Chavez et al., 2015) was transiently transfected along with a plasmid expressing either a gRNA specific to LACZ or a gRNA specific to promoter of GSC. Cells were maintained in mTESR1 for 48 hours after transfection and were then differentiated for an additional 48 hours.

Previously Published RNA Datasets used in this Study

Human tissue datasets from GTEx (The GTEx Consortium, 2013) were obtained through dbGAP (Table S3). Additional datasets of differentiating hESCs were obtained from GSE63935, GSE52657, and GSE44875. GRO-seq data for hESCs and endoderm differentiation were obtained from GSE41009. Mouse tissue datasets were obtained from GSE36025 and datasets for differentiating mESCs were from GSE36114.

Highlights.

• - The lncRNA DIGIT is induced by activin signaling during endoderm differentiation

• - Depletion of DIGIT inhibits definitive endoderm differentiation in culture

• - DIGIT regulates differentiation of human and mouse embryonic stem cells

• - DIGIT regulates expression of Goosecoid to control endoderm differentiation