Minimal in vivo requirements for developmentally regulated cardiac long intergenic non-coding RNAs

A substantial proportion of the mammalian genome is transcribed throughout development, although only a small fraction of this yields functional protein (ENCODE Project Consortium, 2012; Hon et al., 2017; Wong et al., 2001). The remaining noncoding RNA is arbitrarily classified into long (lncRNA) and short transcripts based upon length greater or less than 200 nt. A few lncRNAs have been implicated to be important for cardiac development (Anderson et al., 2016; Grote et al., 2013; Han et al., 2014; Kurian et al., 2015). However, these RNA molecules are often products of the pervasive bidirectional transcription taking place at most genes (Katayama et al., 2005), which makes independently dissecting their function difficult. On the other hand, thousands of putative intergenic lncRNAs (lincRNAs) with little protein-coding potential exist as stand-alone units (Carninci et al., 2005). They can exhibit characteristics indicative of epigenetic control, such as histone H3 trimethylation at lysine 4 (H3K4me3) and acetylation of lysine 27 (H3K27Ac) at promoters, and trimethylation at lysine 36 (H3K36me3) throughout their gene body, splicing, 5′ m7G capping and polyadenylation (Derrien et al., 2012; Quinn and Chang, 2016; Sati et al., 2012). LincRNAs also can display considerable sequence conservation and are dynamically expressed in specific tissues at developmentally discrete times (Diederichs, 2014; Mattioli et al., 2019; Perry and Ulitsky, 2016). For example, the lincRNAs braveheart (Bvht), meteor (Themis^M4Btlr) and carmen (Carmn) seem to play significant roles in precardiac mesodermal differentiation, at least in cellular differentiation systems in vitro (Alexanian et al., 2017; Hou et al., 2017; Klattenhoff et al., 2013; Ounzain et al., 2015). However, meteor knockout in vivo resulted in milder phenotypes than observed in vitro (Guo et al., 2018), while braveheart and carmen have not been tested in embryos. Thus, few lincRNAs have been shown to be required for cardiac development in vivo. Given the great number of lincRNAs expressed within the cell, the energy investment into the processing and maintenance of these transcripts suggests their putative importance. Therefore, efforts must be taken to interrogate specific lincRNA requirements for proper embryogenesis.

We were most interested in lincRNAs that might act to influence the early commitment of nascent mesoderm into the cardiac lineage. We hypothesized that as yet unstudied transcripts were important for this most fundamental stage of cardiac development. Therefore, we screened for the expression of candidates during mouse embryonic stem cell (mESC) in vitro differentiation into cardiomyocytes (CM) through nascent mesoderm (MES), cardiac mesoderm (cMES) and cardiac progenitor (CP) intermediates. Of more than 114,000 long noncoding RNA annotations, we identified a small cohort of lincRNAs with epigenetic regulation, clear splice structure and cardiac progenitor specificity, which we then validated in vivo in the early embryo. Ablation of these noncoding genes revealed regulatory roles within their topologically associated domains (TADs) but very mild or undetectable phenotypes in heart development or postnatal function.

We hypothesized that, like many canonical genes, a subset of lincRNAs would be specifically expressed in the cardiac lineage. We also predicted that those most crucial for heart formation would function early in its development. To find candidate lincRNAs, we re-mapped stranded raw RNA-seq reads from differentiations of mouse ESCs into cardiomyocytes (Fig. 1A,B) (Devine et al., 2014; Wamstad et al., 2012) against Noncode version 4.0-annotated transcripts (Xie et al., 2014). Additionally, we integrated parallel histone modification ChIP-seq data (Wamstad et al., 2012) to discriminate loci under epigenetic regulation. To screen for cardiac developmental specificity, we chose to focus on elements that were lowly expressed in ESCs (FPKM<0.5), while strongly upregulated in cardiac mesoderm or cardiac progenitors (FPKM>1.0). The majority of protein coding genes display antisense transcription from their promoters (Carninci et al., 2005), including numerous studied lncRNAs (Anderson et al., 2016; Daneshvar et al., 2016; Gore-Panter et al., 2016; Grote et al., 2013; Han et al., 2014; Kino et al., 2010; Kurian et al., 2015; Ramos et al., 2015; Xu et al., 2016). We focused instead on genomic elements that could be altered independently from their nearby protein-coding genes. Therefore, we filtered for RNA annotations whose transcriptional start site (TSS) began more than 1 kilobase (kb) from the TSS of known protein-coding genes. To avoid spurious transcripts, we required candidates be spliced and then further refined the list to those displaying histone H3 lysine-4 trimethylation (H3K4me3) and H3 lysine-27 acetylation (H3K27Ac) at their promoters. After removing annotated transcripts that splice into nearby protein coding genes (i.e. A930006K02Rik into Ifnar1), we found that these criteria narrowed candidates to only nine total lincRNAs out of 114,104 considered transcripts (Fig. 1B).

The lincRNA Rubie (RNA upstream Bmp4 in the inner ear, Gm15219) is known to co-express with the TGFβ signaling protein Bmp4 (Wozney et al., 1988) after E15.0 in the mouse inner ear, and its perturbed splicing has previously been implicated in ear vestibule malformation and consequent circling behavior (Roberts et al., 2012). However, our candidate screen revealed it to be expressed much earlier in the developing cardiac mesoderm (Fig. 2A). As in the inner ear, its expression in vitro overlapped Bmp4, and these genes, which are separated by ∼176 kb, co-occupied a strongly interacting region within the same topologically associated domain (TAD; Fig. 2B,C) (Dixon et al., 2012; Nora et al., 2012).

Hand2, a transcription factor that is crucial for heart development (Srivastava et al., 1997), has previously been shown to be regulated by antisense transcription of the noncoding RNA upperhand (Uph; Hand2os1) locus 5′ from its promoter (Anderson et al., 2016). Our search identified 5033428l22Rik as a candidate lincRNA ∼8 kb downstream of Hand2, which we named Handlr (Hand2-associated lincRNA). During the course of our study, others also studied this locus, which they named handsdown (Hdn) (Ritter et al., 2019), in the cardiac lineage. This region displayed numerous transcribed splice forms, but 3′ rapid amplification of cDNA ends (RACE) of E9.5 cDNA revealed a single predominant 5-exon, polyadenylated isoform that varied from its Ensembl- and RefSeq-annotated structures (Fig. 1D, Fig. S1A,B). The expression of Handlr overlapped that of Hand2 in vitro (Fig. 2E), but these genes sat near a TAD border (Fig. 2F) and were separated by a CTCF insulation site (Martin et al., 2011), suggesting a potential topological division between the two.

Seven additional annotated lincRNAs met the criteria for subsequent analyses. We discovered that Atcayos (2310050B05Rik) transcription spanned the important cardiomyocyte metabolic regulator Nmrk2 (Diguet et al., 2018) and preceded its expression in differentiating cardiac progenitors and cardiomyocytes (Fig. S2A,B). Gm12829, named HrtLincR4 (heart lincRNA of chromosome 4) was correlatively expressed within a genomic domain in frequent proximity to Trabd2b, a Wnt protein-binding metalloprotease (Zhang et al., 2012b). In addition, its expression was only transiently detected within an 18 h window at the cardiac mesoderm (cMES) stage of differentiation (Fig. S2C-E). E130006D01Rik, named HrtLincR5 (heart lincRNA of chromosome 5), was expressed within a Mn1-interacting DNA domain ∼275 kb downstream of this transcriptional co-activator (van Wely et al., 2003). This transcript displayed highly stereotypic splicing and was only detected at the cardiac progenitor stage of differentiation (Fig. S3A-C). C430049B03Rik, named HrtLincRX (heart lincRNA of X chromosome) was highly expressed early in our differentiation model and contained a miRNA cluster in its 3′ tail that had previously been shown to drive cardiomyocyte specification (Shen et al., 2016). This lincRNA also lies ∼12.5 kb downstream of – and overlapped expression with – the essential placental gene Plac1 (Jackman et al., 2012), which is not normally expressed in most somatic tissues (Fant et al., 2010) (Fig. S2D-F). Finally, 5033406O09Rik, 9630002D21Rik and 2810410L24Rik also fulfilled the criteria of our screen (Fig. S4A-C).

All nine lincRNAs contained regions with highly homologous sequence to human and/or mammalian genomes (Fig. 2, Figs S2, S3 and S4). To assess the protein-coding potential of these candidates, we employed multiple tests. First, we evaluated PhyloCSF (Lin et al., 2011) codon scores in all three frames for each transcript. Whereas this algorithm readily detected stretches with coding potential in known genes such as Bmp4 and the micropeptide-containing Apela/Toddler (Pauli et al., 2014) and Dworf (Strit1) (Nelson et al., 2016), we found no evidence for protein-coding potential in our lincRNA cohort, with one exception. A 28 amino acid reading frame in the second exon of HrtLincR4 was predicted to represent a possible conserved coding region (Fig. S5A), although HrtLincR4, as well as each additional member of the cohort displayed negative coding-non-coding indices (CNCI, Fig. S5B) (Sun et al., 2013) similar to the known paraspeckle-associated lincRNA Neat1 (Hutchinson et al., 2007). However, CPAT (Wang et al., 2013) calculation of hexamer usage bias (Wang et al., 2013) and Ficket nucleotide composition and codon usage bias (Fickett, 1982) could not differentiate our lincRNA group from Apela or Dworf (Fig. S5C). We next tested Rubie, Handlr, Atcayos, HrtLincR4, HrtLincR5 and HrtLincRX localization in fractionated cardiac progenitor cells. These lincRNAs were biased to the nucleus, where Rubie and Handlr were enriched even more so than Neat1 (Fig. S5D). Additionally, these six lincRNAs could generate cDNA using oligo dT primers at least as efficiently as Actb and Neat1, which are known to be polyadenylated (Fig. S5E, Table S1) (Sasaki et al., 2009). Taken together, we classified these lincRNAs as nuclear enriched, polyadenylated, transcripts with little translational capacity. However, we could not rule out the coding potential of the minority HrtLincR4 fraction that reaches the cytoplasm.

We examined the spatiotemporal expression patterns in the developing embryo for each of the nine candidate lincRNAs by whole-mount in situ hybridization from E7.25 through E10.5 using transcript-specific probes (Table S2). The expression patterns observed in vitro were largely predictive of those observed in vivo. Rubie was first observed in the E7.75 embryo, where, similar to Bmp4 (Perea-Gomez et al., 1999), it strongly demarcated the extra-embryonic boundary and flanked the eventual heart field. From E8.0 to E8.5, its expression became less focused, spreading throughout the developing cardiac crescent and heart tube, respectively. Rubie transcription at E8.75 was largely non-cardiac, and by E9.5, it was strongly localized to posterior mesoderm and the otic vesicle (Fig. 3A). This patterning overlapped a somewhat refined subset of what had previously been established for Bmp4 (Danesh et al., 2009).

From E8.5 through E9.5, Handlr was transcribed in the developing heart tube, posterior cardiac progenitors, branchial arches and lateral plate mesoderm (Fig. 3B). These patterns overlapped what was also shown for Hand2 at this developmental stage (Charite et al., 2000), suggesting common regulation between Hand2 and Handlr. Atcayos, as predicted by in vitro expression patterns, was weakly expressed during early stages of heart tube formation, while it was dramatically upregulated after E9.5 in the developing ventricles, as well as cranial structures and somitic mesenchyme (Fig. 3C).

From E8.25 through E9.5, HrtLincR4 displayed strong expression in developing pharyngeal mesoderm, immediately dorsal to the developing cardiac crescent (Fig. 3D). Given its highly transient expression within differentiating cardiac mesoderm in vitro, these data suggested that HrtLincR4 was quickly specified to the secondary heart field and/or adjacent tissues during the onset of cardiac lineage commitment. HrtLincR5 was broadly expressed throughout the mesoderm, including the nascent cardiac crescent, at E8.25. As expected by its short-lived in vitro expression pattern, HrtLincR5 was predominantly lost in vivo by E9.5 (Fig. 3E).

HrtLincRX was strongly expressed by E7.5 during cardiac lineage formation in anterior mesoderm at the extra-embryonic boarder, as well as in extra-embryonic tissues. At E8.25, it was strongly expressed in the cardiac crescent, amniotic membranes and the developing allantois. Although expression of the adjacent miR322/503 cluster was previously shown to be cardiac specific (Shen et al., 2016), this lincRNA was widely expressed throughout the heart, forelimb and somitic mesoderm at E9.5 and E10.5 (Fig. 3F). This suggested divergent regulation and/or compounding roles for HrtLincRX versus its miRNA components. We could not effectively validate the expression of 5033406O09Rik, 9630002D21Rik or 2810410L24Rik beyond diffuse low levels in the developing mouse embryo (Fig. S6A-C). These experiments established the striking expression patterns of numerous tissue-specific lincRNAs identified from our screen of in vitro cardiac differentiation. Therefore, we aimed to test the developmental importance of Rubie, Handlr, Atcayos, HrtLincR4, HrtLincR5 and HrtLincRX expression during embryonic development.

To determine the requirement for the six lincRNAs that displayed compelling in vivo expression, we generated knockout mouse lines through pronuclear Cas9 mRNA and tru-sgRNA (Fu et al., 2014) injections. For each knockout, paired tru-sgRNAs were co-injected to induce 2-3 kb deletions flanking the respective lincRNA transcriptional start site (TSS) (Fig. 4A, Table S3), which successfully generated heritable alleles for all six target regions. After substantial outbreeding (more than three backcrosses into C57Bl/6j background), we crossed heterozygotes and harvested the anterior half of E8.25 embryos for RT-qPCR (Fig. 4B). We found that these deletions ablated downstream transcription of each lincRNA (Fig. 4C-H). As these lincRNAs were nuclear enriched, we hypothesized they might be involved in transcriptional regulation within their local genomic environments. To test this, we measured expression of neighboring protein-coding genes sharing the same respective TADs (Table S1). While Rubie was previously associated with BMP4 signaling in the inner ear (Roberts et al., 2012), its requirement for Bmp4 expression had not been established. We found that loss of Rubie resulted in significant reduction of Bmp4 expression during cardiac specification. Furthermore, the amount of transcribed Rubie was directly correlated with Bmp4 levels in this region at the same time point. This effect was maintained even within equivalent underlying genotypes, whereby Rubie and Bmp4 transcript levels were still significantly correlated among Rubie^+/− offspring only (Fig. 4C). These data strongly suggested that either the act of Rubie transcription and/or its physical RNA molecule were responsible for quantitative regulation of Bmp4 expression.

Despite proximity to and co-expression with Handlr, Hand2 activation was not dependent on Handlr lincRNA (or its underlying promoter DNA sequence, Fig. 4D). We speculated that the TAD architecture and CTCF boundary between these genes may introduce complex dynamics within the region. We also could not find a correlation between the expression of Mn1 and HrtLincR5 (Fig. 4E). In contrast, Nmrk2 and Trabd2b expression was dependent on Atcayos (Fig. 4F) and HrtLincR4 (Fig. 4G), respectively. Furthermore, Plac1 transcription was significantly and inversely correlated to HrtLincRX levels, whereby loss of HrtLincRX resulted in an approximately twofold increase in expression of Plac1 (Fig. 4H). However, using IntaRNA (Mann et al., 2017) analysis, we calculated stable RNA-RNA interactions between all three miRNAs constituents of its 3′ tail (miRNA-322, miRNA-351 and miRNA-503) and the 5′- and 3′-untranslated regions (UTRs) of Plac1 (Fig. 4I). Therefore, this relationship could likely be explained by the loss of inhibitory miRNA binding to Plac1 primary transcript. Nonetheless, these data indicate a potential role for HrtLincRX and/or its miRNAs in demarcating embryonic from extra-embryonic mesoderm during gastrulation and early cardiogenesis.

To determine the requirement of our lincRNA cohort for viable embryonic development in vivo, we bred heterozygotes for each gene and examined ratios of expected offspring that survived to weaning (Table S4). We could not establish any reduction in viability within null progeny for any tested lincRNA (Fig. 5A,C,D, Fig. S7A-C). Furthermore, homozygous offspring for these loci lived to adulthood and were fertile. However, the Rubie-null genotype did sporadically recapitulate the circling behavior described by Roberts et al., which they observed as a result of aberrant Rubie splicing in the SWR/J genetic background (Roberts et al., 2012). Although rarely observed, circling was only present in Rubie^−/− mice after more than 2 years of colony breeding (Fig. 5B). Nonetheless, despite their clear expression within the developing heart, we concluded that none of the lincRNAs were individually required for viable development or fertility in the FVBn; C57BL/6j mixed background.

Gene knockout models often require external stressors to materialize overt phenotypes. Handlr and Atcayos were the only cohort lincRNAs expressed in adult hearts, and the very high expression of Atcayos was reduced ∼50% after transverse aortic constriction (TAC)-induced cardiac hypertrophy (Fig. S7D,E) (Duan et al., 2017). Therefore, we performed TAC experiments on Handlr- and Atcayos-null mice, and compared their responses to wild-type littermates. At baseline, calculated left ventricular (LV) masses (from echocardiographic measurements) were modestly reduced in Handlr^−/− and Atcayos^−/− adults (−18.4%, P<0.05; −22.4%, P<0.01, respectively; Fig. 5E; Tables S5 and S6). Additionally, Handlr-null adults had slightly but significantly increased fractional area contractility (FAC) over Handlr^+/+, Atcayos^+/+ and Atcayos^−/− genotypes (46.5% versus 37.8%, P<0.01; 37.4%, P<0.005; and 40.1%, P<0.05, respectively; Fig. 5F, Tables S5 and S6). However, neither loss of Handlr nor Atcayos could invoke a significant alteration to the LV mass increase or to the LV fractional shortening decrease after TAC-mediated stress. Of note, the severity of the TAC-response in Handlr-related experiments was greater than that for Atcayos due to genetic background and/or surgical differences for each cohort. This resulted in a sharper fractional shortening decrease and a longer duration under cardiac failure (%FAC<30%) for these mice. Consequently, Handlr^+/+ males displayed greater incidence of progression into heart failure-induced lung hypertrophy versus Handlr^−/− individuals (which began the experiment with greater contractile function), while less frequent lung hypertrophy arose in the Atcayos groups (Fig. 5G-J). This result was not interpreted to indicate an altered cardiac hypertrophic response in Handlr-null individuals but further supports their increased LV FAC measured at baseline. We also could not detect any noticeable changes versus wild-type in the expression of canonical hypertrophic response genes Nppa, Nppb or Acta1 due to Handlr or Atcayos knockout (data not shown). Therefore, despite strong Atcayos expression in the adult heart and the known hypertrophic involvement of Hand2 (the neighbor of Handlr) loss of these transcripts did not induce an altered response to LV pressure overload. Therefore, we concluded that neither of these lincRNAs played important roles in the physiological response to heart failure.

Despite the lack of overt lethality in lincRNA-deficient offspring, we carefully examined morphological heart development in Handlr- and Rubie-null embryos, the only conditions that produced noticeable physiological effects. We harvested E15.5 hearts and examined transverse histological sections to establish any change to chamber septation, myocardial trabeculation and/or compaction, or to ventricular outflow tract (OFT) development. Handlr^−/− adults exhibited increased left ventricular fractional shortening over wild-type controls, but we could not associate this functionality with overt changes in cardiac anatomy (Fig. S8A). Owing to overlapping expression patterns between Hand2 and Handlr in the developing heart, we next tested embryos from Hand2^+/−×Handlr^+/− crosses to eliminate one allele of either Hand2 or Handlr per chromosome. However, neither Hand2 heterozygosity nor Hand2^+/−; Handlr^+/− compound heterozygosity resulted in any clear effects on heart morphogenesis (Fig. S8B,C). In addition, we did not notice any elevated lethality in Hand2^+/−; Handlr^+/− offspring (data not shown, n=63).

Bmp4 expression in the embryo was correlated to the amount of Rubie transcript. Numerous studies have established the requirement of the correct BMP4 dose for normal septation of the atria, ventricles and outflow tract (OFT), as well as for viable embryo development (Dunn et al., 1997; Goldman et al., 2009; Jiao et al., 2003). Therefore, we tested the hypothesis that compound haploidy of Bmp4 and Rubie together would result in an exacerbated onset of resulting phenotypes. For this, we bred Bmp4^fl/fl×Rubie^+/−; Actb-Cre⁺ (single transgene integration) in the FVB/n; C57BL/6j mixed genetic background. When we examined E15.5 hearts, neither the Rubie^−/− background nor loss of a single Bmp4 allele could induce an abnormal cardiac phenotype. We also recovered these offspring in expected ratios at weaning (Fig. 6A-D). However, we did find a sustained ∼20% reduction in recovered pups carrying the Bmp4^+/−/Rubie^+/− compound genotype (Fig. 6D; n=186), although dramatically more breeding rounds would be required to deem this result statistically significant (n>585 needed for P<0.05 at observed lethality rate). In addition, these offspring exhibited incidences of OFT distortion out of the right ventricle beyond its typical boundary. In these cases, the origins of the pulmonary artery skewed toward the left ventricular OFT and aortic valve (Fig. 6E). Although we were unable to clearly establish communication between the pulmonary and aortic outflow systems in these instances, the data point toward a modest genetic interaction between Rubie and Bmp4 in cardiac morphogenesis.

With thousands of uncharacterized noncoding transcriptional elements expressed throughout the genome, efforts must be taken to better understand the functional relevance of unstudied lincRNAs. To achieve this, these experiments intended to identify and test the requirement for cardiac progenitor-specific lincRNAs in the developing embryo. Out of numerous considered annotated lncRNAs that we initially surveyed, our selection criteria led to a highly restricted set that contained epigenetically regulated promoter signatures and cardiac-specific expression in vivo. Ablation of these transcripts in the developing mouse revealed modulatory roles of Rubie, Atcayos, HrtLincR4 and HrtLincRX within their local genomic environments. In particular, we found a requirement for the Rubie locus for normal Bmp4 dose. Furthermore, loss of Handlr/Hdn and Atcayos led to minor reduction in LV mass in adult mice, along with modestly increased contractility among Handlr^−/− individuals. Loss of either of these two transcripts did not influence the hypertrophic or LV contractile response to pressure overload. Most importantly, despite clear transcription in the developing heart, none of the tested lincRNAs alone were required for embryo viability. However, when we generated compound heterozygotes for Rubie and Bmp4, we observed a slight yet consistent reduction in recovered offspring and a modest perturbation of right ventricular outflow tract orientation. Of course, this cohort of lincRNAs that we ablated does not include the entirety of all relevant cardiac transcripts. Nonetheless, we argue this set of transcripts represents a sufficient consideration of the early cardiac lincRNA landscape to predict modest individual roles for most noncoding transcripts of this type.

Similar to our observations, an independent study (Ritter et al., 2019) reported that the Handlr/Hdn locus produced multiple transcripts. Namely, Hdn was described as originating from the same TSS but splicing ∼15 kb 3′ beyond Handlr, along with Hdna, which shares TSS and splice patterning (albeit one less exon) with Handlr. Hdn/Hdna was described as localized to the cytoplasm near the nuclear envelope. Our results clearly supported Handlr enrichment in the nucleus, though we did not interrogate to which subnuclear component. Most importantly, they showed that complete deletion of the >20 kb Hdn locus produced embryonic-lethal cardiac and extra-embryonic phenotypes. This accompanied perturbation of a cardiac gene expression program, including hyper-expression of Hand2. They attributed the act of transcription from the Hdn/Hdna TSS as an important regulatory component of this phenotype, whereas Hdna-specific RNA molecules were dispensable. In our work, ablation of the Handlr/Hdn/Hdna TSS, which terminated downstream transcription, resulted in no lethality, dramatic cardiac phenotype or altered Hand2 expression. Therefore, both sets of experiments agree on the lack of requirement for Handlr/Hdna as a functional lincRNA during cardiogenesis, but disagree on a link between Handlr/Hdn/Hdna transcription and Hand2 expression. Importantly, a clear regulatory mechanism/effect for transcription per se from the Handlr/Hdn/Hdna TSS has yet to be established. We predict the large genomic deletion reported by Ritter et al. (2019) affects crucial cardiac enhancer elements within the Hdn locus. Therefore, future work must more definitively assign a mechanistic role for RNA synthesis, splicing and/or molecular function at this locus before attributing them to the mild or severe phenotypes observed by either study.

The subtle effects created by ablation of our collection of cardiac lincRNAs are consistent with the results most often obtained by the efforts of others to knockout developmentally-specific lincRNAs (Amândio et al., 2016; Goff et al., 2015; Goudarzi et al., 2019; Lai et al., 2015; Nakagawa et al., 2012, 2011; Sauvageau et al., 2013; Zhang et al., 2012a). In most instances where strong phenotypes have been observed, conflicting regulatory mechanisms, in vivo versus in vitro effects, lack of in vivo reproducibility, and/or conflated DNA-/RNA-/transcription-/splicing-based mechanisms prevent clean parsing of the underlying importance of lincRNA molecules. However, several lincRNAs in our cohort did seem to function within the nucleus to impact gene expression in their local environments, including the influence of Rubie on Bmp4. Consequently, future experiments are needed to dissect the physical mechanisms that underlie these effects. We speculate that the vast majority of singular lncRNAs fit as cogs into the multifaceted regulatory architecture of the nucleus, which individually can be compensated for during organogenesis in vivo. More so we hypothesize that overt phenotypic impacts in most lincRNA-centric experiments will be observed only after additional contextual molecular components are also manipulated. It may prove more beneficial to interpret most nuclear lincRNAs as collective epigenomic components analogous to histone/DNA modifications – with potential capabilities to alter the physical properties and orientation of the genome, and recruit regulatory machinery, more so than as individual gene-like elements.

Raw stranded, total RNA-seq reads from ESC (day 0), MES (day 4), CP (day 5.3) and CM (day 10) stages (Wamstad et al., 2012), and cMES (day 4.75) stage (Devine et al., 2014) of mESC in vitro differentiation into cardiomyocytes were mapped to the mouse genome (mm9) and aligned to Noncode v4 (Xie et al., 2014) annotated lncRNAs using the Cufflinks suite of software (Trapnell et al., 2010). ChIP-seq domains positive for trimethylation of histone 3 lysine 4 (H3K4me3), acetylation of histone 3 lysine 27 (H3K27Ac) and trimethylation of histone 3 lysine 27 (H3K27me3) for ESC, MES, CP and CM stages were obtained from Wamstad et al. (2012). The following criteria were used to generate a candidate list of lincRNAs: (1) fewer than 0.5 fragments per kilobase per million reads (FPKM) in mESCs; (2) greater than 1.0 FPKM at the CP or cMES stage of differentiation (expression at other time points was not factored into selection); (3) positive H3K4me3 ChIP-seq signal at TSS during CP stage; (4) positive H3K27Ac at TSS during CP stage; (5) at least one exon splice in transcript; (6) no splice events into neighboring protein coding genes; and (7) TSS at least 1 kb from nearest protein-coding gene TSS. Screened candidates tracks were then visually inspected via the UCSC Genome Browser (Kent, 2002) to filter for lincRNAs with expression patterns that matched the assigned lincRNA structure and not simply spurious reads at the general locus.

PhyloCSF (Lin et al., 2011) browser tracks were uploaded to the UCSC Genome Browser for interrogation. Ficket and hexamer scores were calculated with CPAT software (Wang et al., 2013) and visualized with the ‘ggplot2’ package (Wickham, 2016) in R version 3.4.0 (R Core Team, 2017). CNCI scores were obtained from NONCODEv4 annotations (Xie et al., 2014).

Directed cardiomyocyte differentiations were performed as previously described (Wamstad et al., 2012) using the Smarcd3-F6nlsEGFP mESC line (Devine et al., 2014) with minor modifications to improve differentiation efficiency. Briefly, 3 days before differentiation induction (day −3), mESCs were split into 2i+LIF media on gelatin. The following day (day −2), 2i+LIF was replaced with 15% FBS (HiClone) in DMEM, 1× non-essential amino acids, 1× sodium pyruvate 1× GlutaMAX, 1× β-mercaptoethanol, 1× penicillin/streptomycin+1000 U/ml LIF (ESGRO, EMD). The following day (day −1), cells were fed again with the same 15% FBS-LIF media to complete their conversion to epiblast-like stem cells. One day later (day 0), cardiac differentiation was initiated as per Wamstad et al. (Wamstad et al., 2012). On day 5.3, CP cells were dissociated with TrypLE (Gibco) and quenched in DMEM:F12 (Gibco)+10% FBS (HiClone). After washing in D-PBS, nuclei were isolated using the Nuclei EZ Prep kit (Sigma Aldrich), pelleted and supernatant was collected as a cytoplasmic fraction. Nuclei were washed again, and the pellet was harvested with Trizol (Invitrogen) as a nuclear fraction in subsequent qPCR quantification experiments. Cytoplasmic fractions were processed equivalently.

Primers were designed to amplify in situ probe templates between 440 bp and 1.5 kb for each candidate lincRNA off cDNA from the CP stage of in vitro differentiation (Table S2). Templates were electrophoresed in 1.0% agarose gel and purified using QIAquick gel extraction kit (Qiagen). These templates were then TOPO TA cloned into pCR4-TOPO using the TOPO TA cloning kit (Invitrogen) and Sanger sequenced to validate orientation in plasmid and proper composition. Linearized vector for each lincRNA template (2 µg) was then input into digoxigenin (DIG) RNA synthesis kit reactions (Roche) in a 40 µl total volume using either T7 or T3 primers, depending on template orientation. Transcription was carried out for 2 h at 37°C. Afterwards, 8 U DNase I (NEB) were added to each reaction and incubated for 15 min at 37°C to degrade DNA. DNase reactions were quenched with 1.5 µl EDTA, and DIG-RNA probes were cleaned and concentrated with RNeasy Mini Columns (Qiagen), ethanol precipitated and washed and resuspended in 20 µl H₂O. DIG probes were then diluted to 100 µg/ml in HYB buffer [50% formamide, 5× SSC (pH 4.5), 50 µg/ml yeast tRNA, 75 µg/ml heparin, 0.2% Tween-20, 0.5% CHAPS, 5 mM EDTA]. E7.5 through E12.5 mouse embryos were liberated from the uterus and dissected from extra-embryonic tissues and membranes. Embryos were washed with D-PBS and fixed overnight in 4% paraformaldehyde and then washed three times in in PBT (PBS+0.1% Tween-20) on ice. Embryos were dehydrated in a methanol series (25%, 50%, 75% and twice in 100% for 5 min each). Then, samples were rehydrated by reversing this series, including two extra PBT washes. Embryos were bleached in 6% H₂O₂ in PBT for 15 min at room temperature with rocking. Embryos were washed for 3×5 min in PBT and treated with 10 μg/ml proteinase K for 5 min (E7.5), 10 min (E8.5), 20 min (E9.5) or 30 min (E10.5+) with rocking at room temperature and then quenched twice with 2 mg/ml glycine in PBT followed by 3×5 min washes in PBTw. Embryos were re-fixed in 4% paraformaldehyde and 0.2% glutaraldehyde for 20 min with rocking, and washed for an additional 5×5 min with PBT. Embryos were then rinsed twice in 65°C HYB buffer and incubated in HYB buffer for 3 h at 65°C. LincRNA-specific probes (in HYB) were the added to final concentration of 1 µg/ml and hybridized overnight at 65°C. Embryos were rinsed for 3×5 min at 65°C in WASH1 buffer [50% formamide, 5×SSC (pH 4.5), 1% SDS] and then incubated again for 2×30 min at 65°C in WASH1 buffer. Next, embryos were washed for 2×30 min at 65°C in WASH2 buffer [50% formamide, 2×SSC (pH 4.5), 0.1% Tween-20], followed by 3×5 min washes at room temperature in TTBS [25 mM Tris HCl (pH 7.4), 135 mM NaCl, 2.5 mM KCl, 0.1% Tween-20]. Embryos were then blocked in TTBS containing 20% sheep serum for 3 h at room temperature and stained overnight with alkaline phosphatase (AP)-conjugated anti-DIG Fab fragments in TTBS+1% sheep serum (1:5000, Roche). Embryos were then rinsed for 3×5 min in RT TTBS, followed by 6×1 h TTBS washes at room temperature. A final TTBS wash was then performed overnight at 4°C. Embryos were then washed for 2×30 min in AP buffer [100 mM Tris (pH 9.5), 50 mM MgCl₂, 100 mM NaCl, 0.1% Tween-20] at room temperature. Boehringer Purple AP substrate was then added to embryos to initiate staining reactions. Reactions were allowed to progress in the dark until suitable contrast was observed. AP reactions were quenched with three PBT washes containing 1 mM EDTA, followed by multiple PBT (pH 5.5) washes. A final fixation was then performed overnight in 4% paraformaldehyde and 0.1% glutaraldehyde at 4°C. Finally, embryos were dehydrated again in methanol series and stored in 100% methanol at −20°C. Embryos were imaged on an upright microscope, and images were white balanced using Adobe Photoshop.

All mouse experiments were carried out in accordance with IACUC protocols and cared for by the UCSF LARC. For each lincRNA, two cut sites were targeted to induce a 2-3 kb deletion flanking the TSS/promoter. Two sequence-specific truncated single guide RNA (tru-gRNA, Table S3 (Fu et al., 2014)) regions were separately cloned into pX330 (Addgene). After generating T7 promoter-containing sgRNA templates by PCR using Phusion TAC polymerase (NEB), tru-sgRNAs were transcribed using the Hiscribe T7 High Yield RNA Synthesis Kit (NEB). Tru-sgRNA was extracted using Trizol reagent (Invitrogen) and dual chloroform purifications before immunoprecipitating with isopropanol. Each tru-gRNA pair was then resuspended in sterile 5 mM Tris-HCl before pronuclear injection by the Gladstone transgenic mouse core. Injections were carried out as previously described (Yang et al., 2014). To increase efficiency of obtaining deletions for each target site, all pairs were co-injected into each of 70 Mus musculus FVB/n pronuclei. All genotyping was performed on tail clips stored at −20°C. To extract gDNA, tail clips were suspended in 100 µl 50 mM NaOH in H₂O and incubated at 95°C for 40 min. Tubes were agitated to break up tissue, and remaining solids were allowed to settle before use. pH was normalized by addition of 7.0 µl of 1 M Tris HCl pH 7.4. 1.5 µl was then input into PCR reactions using Q5 2X master mix (NEB) and 3 gene-specific primers for simultaneous WT and KO allele amplification (Table S4). Thus, any KO allele products larger than WT reflect primer positioning and not inserted DNA sequence. Reactions were carried out according to manufacturer-specified recommendations. F0 founders for single lincRNA deletions were first identified and bred into Mus musculus C57BL/6j to establish germline transmission (F1). Separate F1 heterozygotes for each individual lincRNA deletion were then outbred into the C57BL/6j background for multiple generations to reduce off-target effects. Handlr- and Atcayos-null alleles were generated in collaboration with the Jackson Laboratory using the same targeting strategy but in a homogenous C57BL/6j background. Statistical significance of deviation from expected Mendelian ratios in offspring was analyzed using a χ² test.

At 8 weeks post-surgery, mice were sacrificed for analysis. First, left ventricle, lung and body weights were measured. Subsequently, a 10-20 mg concentric short axis slice of the left ventricle was collected and preserved in RNAlater reagent (ThermoFisher). Heart sections were disrupted in PureZOL (Bio-Rad) on a TissueLyser II (Qiagen). RNA was then purified with Aurum purification kit (BioRad). qRT-PCR was performed using TaqMan chemistry, including FastStart Universal Probe Master (Roche), labeled probes from the Universal Probe Library (Roche) and gene-specific oligonucleotide primers run on a 7900HT (ThermoFisher) cycler with absolute quantification. Gene expression levels were normalized to cycloB and Actb internal controls using the ΔCt method. Statistical significance was determined by Student's two-tailed t-test (P<0.05), except for week 8 organ weights due to bimodal hypertrophy progression (P values calculated using a Z-test).

At E8.25, embryos were removed from the uterus and dissected from extra-embryonic tissues and membranes. Only embryos displaying late cardiac crescent formation before heart tube expansion and cavitation were kept and deemed to be at E8.25. The anterior half of each embryo was washed twice in ice-cold PBS and transferred into Trizol (Invitrogen), while the posterior half was washed in PBS and stored at −20°C for genotyping. RNA from Trizol samples was precipitated using standard protocols and further purified/condensed using Qiagen RNeasy MinElute columns. RNA (250 ng) was reverse transcribed using the AffinityScript Reverse Transcription kit (Agilent) with 200 ng random hexamer and/or 100 ng dT₂₀ primers, where appropriate. RT-qPCR was subsequently performed with 5.0 ng cDNA and 500 nM gene-specific primers (Table S1) in PowerUP SYBR Green master mix (Thermo Fisher). Reactions were run on a 7900HT (Thermo Fisher) cycler with absolute quantification. Gene expression levels were normalized to Actb internal controls using the ΔCt method. Statistical significance was determined by Student's two-tailed t-test (P<0.05), except for correlation analysis, which was calculated using Pearson's r statistic and N-2 degrees of freedom.

At E15.5, embryos were liberated from the uterus and dissected from extra-embryonic tissues and membranes. Whole hearts were removed, rinsed twice in D-PBS and fixed overnight in 4% paraformaldehyde. Each heart was then embedded in paraffin wax and sectioned at an oblique transverse plane for four chamber visualization. Hematoxylin and Eosin staining and imaging were performed by the Gladstone Histology Core (UCSF).

We thank Junli Zhang (Gladstone Transgenic Core) for pronuclear injection, Hazel Salunga for help with echocardiography, and the Gladstone Histology and Microscopy Core for embryo heart sectioning and histology. We are also grateful to Judy Morgan, Leslie Goodwin and Laura Reinholdt (JAX) for mouse production.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.185314.reviewer-comments.pdf