A human ESC-based screen identifies a role for the translated lncRNA LINC00261 in pancreatic endocrine differentiation

Long noncoding RNAs (lncRNAs) are a heterogenous group of RNAs, which can encode small proteins. The extent to which developmentally regulated lncRNAs are translated and whether the produced microproteins are relevant for human development is unknown. Using a human embryonic stem cell (hESC)-based pancreatic differentiation system, we show that many lncRNAs in direct vicinity of lineage-determining transcription factors (TFs) are dynamically regulated, predominantly cytosolic, and highly translated. We genetically ablated ten such lncRNAs, most of them translated, and found that nine are dispensable for pancreatic endocrine cell development. However, deletion of LINC00261 diminishes insulin+ cells, in a manner independent of the nearby TF FOXA2. One-by-one disruption of each of LINC00261's open reading frames suggests that the RNA, rather than the produced microproteins, is required for endocrine development. Our work highlights extensive translation of lncRNAs during hESC pancreatic differentiation and provides a blueprint for dissection of their coding and noncoding roles.


Introduction
Defects in pancreatic endocrine cell development confer increased diabetes risk later in life (Bakhti et al., 2019). Therefore, a detailed understanding of the factors that orchestrate endocrine cell differentiation is highly relevant to human disease. Many of the molecular mechanisms that underlie the formation of pancreatic endocrine cells have been defined (Romer and Sussel, 2015;Schiesser and Wells, 2014). However, despite some evidence that long noncoding RNAs (lncRNAs) are important for proper development and function of pancreatic beta cells (Arnes et al., 2016;Morán et al., 2012;Wong et al., 2019), a systematic functional assessment of the noncoding transcriptome during pancreas development is lacking.
Most lncRNAs with to date demonstrated roles in the regulation of fundamental developmental processes are active in the cell's nucleus (Daneshvar et al., 2016;Jiang et al., 2015;Klattenhoff et al., 2013;Kurian et al., 2015;Lin et al., 2014;Luo et al., 2016;Ramos et al., 2015). However, a large proportion of lncRNAs is predominantly cytosolic (Cabili et al., 2015;van Heesch et al., 2014), and the functional relevance of these lncRNAs has remained unexplored in the context of human development. It is now widely accepted that many cytosolic lncRNAs possess short, 'non-canonical' open reading frames (sORFs) that are translated (Bazzini et al., 2014;Makarewich and Olson, 2017;Ruiz-Orera et al., 2014). What fraction of these non-canonical ORFs is functional, and whether sORF translation serves a pure regulatory purpose or results in the production of stable microproteins, remains an active topic of debate (Levy, 2019;Ruiz-Orera et al., 2018). Since high rates of conservation have historically been employed for the identification and annotation of canonical protein coding sequences (Lin et al., 2011;Mudge et al., 2019), a primary reason for doubting the protein-coding capacity of sORFs in presumed lncRNAs is their generally poor sequence conservation across species. To address these questions, several recent studies have systematically assessed the biological activity of newly discovered sORFs, revealing that many produce evolutionary young microproteins with roles across cellular organelles and processes, and a subset being essential for cell survival (Chen et al., 2020;Martinez et al., 2020;Prensner et al., 2020;van Heesch et al., 2019). This previously unrecognized coding capacity of supposedly noncoding RNAs illustrates their functional diversity and has called into question the noncoding classification of some lncRNAs. Thus, there is a need for careful investigation and dissection of any gene's coding and noncoding functions.
LncRNAs, translated or fully noncoding, are not randomly distributed in the genome but are frequently located close to, and coregulated with, canonical protein-coding genes in cis (Luo et al., 2016;Neumann et al., 2018;van Heesch et al., 2019). For example, the lncRNAs DIGIT (also known as GSC-DT) and Gata6as (also known as lncGata6 or GATA6-AS1) have been reported to enhance expression of divergently expressed endoderm regulators Goosecoid (GSC) and Gata6, respectively (Daneshvar et al., 2016;Luo et al., 2016;Neumann et al., 2018). Similarly, the Pax6associated lncRNA Paupar promotes pancreatic islet alpha cell formation through the alternative splicing of Pax6 transcripts in mice (Singer et al., 2019). Furthermore, LINC00261 (also known as DEANR1) and its neighboring TF FOXA2 are both induced in endoderm formation, during which LINC00261 has been proposed to positively regulate FOXA2 expression (Jiang et al., 2015). However, whether such cis-acting lncRNAs are translated and may exert cytosolic functions through trans-acting, microprotein-dependent mechanisms relevant for endoderm and pancreas development is not known.
In this study, we classified lncRNAs based on their dynamic regulation, subcellular localization, and translation in a hESC differentiation system that recapitulates in vivo pancreas development. Next, we used this classification to prioritize select dynamically regulated and highly translated lncRNAs for deletion in hESCs, followed by extensive phenotypic characterization across multiple intermediate states of pancreas development. Nine out of the ten selected lncRNAs were not essential for pancreatic development and, despite their vicinity to lineage-determining TFs, none of these lncRNAs regulated the expression of these TFs in cis.
The deletion of one lncRNA, LINC00261, impaired human endocrine cell development and led to a significant reduction in the number of insulin-producing cells. Contrary to previous studies of LINC00261 knockdown hESCs (Jiang et al., 2015), deletion of LINC00261 had no effect on the expression of nearby TF FOXA2 or other proximal genes, suggesting control of endocrine cell formation through a trans-rather than cis-regulatory mechanism. LINC00261 was among the most highly translated lncRNAs based on ribosome profiling (Ribo-seq) and produced multiple microproteins with distinct subcellular localizations upon overexpression in vitro. To systematically assess LINC00261's coding and noncoding functions, we separately introduced frameshift mutations into each of seven identified LINC00261 sORFs. However, rigorous phenotypic characterization revealed no apparent consequences of loss of each of the seven LINC00261-sORF-encoded microproteins on endocrine cell development. Our comprehensive assessment of functional lncRNA translation identified a likely trans-regulatory role for LINC00261 in endocrine cell differentiation that appears to be independent of the seven microproteins that were individually deleted. With this detailed investigation we provide a blueprint for the proper dissection of a gene's coding and noncoding roles in a human disease-relevant system.

LncRNAs and nearby lineage-determining transcription factors exhibit dynamic coregulation during pancreas development
To identify lncRNAs involved in the regulation of pancreas development, we profiled RNA expression at five defined stages of hESC differentiation toward the pancreatic lineage: hESCs (ES), definitive endoderm (DE), primitive gut tube (GT), early pancreatic progenitor (PP1), and late pancreatic progenitor (PP2) ( Figure 1A). While some lncRNAs were constitutively expressed (n = 592; 25.3%), the majority showed dynamic expression patterns (n = 1745; 74.7%), being either strongly enriched in (n = 874; 37.4%) or specific to (n = 871; 37.3%) a single developmental intermediate of pancreatic lineage progression ( Figure 1B and Figure 1-source data 1A). The expression of many of these  dynamically regulated lncRNAs correlated with that of proximal coding genes (Figure 1-figure supplement 1A-D and Figure 1-source data 1B,C), further exemplified by a subset of lncRNAs that was specifically coregulated with the key endodermal and pancreatic TFs GATA6, FOXA2, PDX1, and SOX9 ( Figure 1C,D). The expression coregulation of these lncRNA-TF pairs is likely explained by a shared chromatin environment (Figure 1-figure supplement 1E-H), which raises the possibility that like the TFs, the function of the lncRNAs is also required for endoderm and pancreas development.
Many pancreatic progenitor-expressed lncRNAs are cytoplasmically enriched and translated Although most functional roles described for lncRNAs to date have been predominantly nuclear (Marchese et al., 2017), multiple recent studies have shown that many lncRNAs are cytosolic and translated into sometimes biologically active microproteins (reviewed in Makarewich and Olson, 2017). To further characterize the above-identified dynamically regulated lncRNAs, we analyzed their subcellular localization and translation potential using fractionation RNA-seq and Ribo-seq across multiple hESC clones independently differentiated into PP2 stage pancreatic progenitors ( Figure 2A). Of all lncRNAs expressed in two replicate differentiations into PP2 cells, we classified 21% (n = 347) as localized to the nucleus, whereas a larger number (n = 563; 34%) primarily resided in the cytosol ( . This subcellular distribution of pancreatic lncRNAs is in agreement with previous lncRNA localization studies by us and others (Cabili et al., 2015;Clark et al., 2012;Sun et al., 2015;van Heesch et al., 2014). LncRNAs enriched in the cytosol were expressed at higher levels than nucleus-localized lncRNAs, with expression levels similar to canonical protein-coding mRNAs (Figure 2-figure supplement 1B). Intriguingly, almost half (49.4%) of all cytosol-enriched lncRNAs (278 out of 563) displayed dynamic expression regulation during the differentiation of hESCs to pancreatic progenitors, raising the possibility that many lncRNAs with putative developmental functions do not act in the nucleus, but instead in the cytosol where they may be translated.
To investigate the translation potential of these cytosolic lncRNAs, we used Ribo-seq, through which we obtained exceptionally deep and high quality translatome coverage across six replicate differentiations ( Requiring stringent reproducibility criteria (the exact ORF needed to be detected by RiboTaper (Calviello et al., 2016) in at least four out of six replicates), we identified a total of 625 new sORFs in lncRNAs with a median length of 47 amino acids (aa) (Figure 2-source data 1D). The majority of detected sORFs (76%; n = 477/625) is currently not present in the sORFs.org database (Olexiouk et al., 2016). The translated sORFs are located within 285 cytosolically localized lncRNAs (25.3% of all expressed lncRNAs) ( Using approaches similar to ours, non-canonical sORFs have previously been characterized in multiple immortalized human cell lines (Bazzini et al., 2014;Calviello et al., 2016;Chen et al., 2020;Ji et al., 2015;Martinez et al., 2020;Prensner et al., 2020;Raj et al., 2016) and human tissues (van Heesch et al., 2019). However, to our knowledge, our data constitute the first comprehensive set of non-canonical human ORFs generated from a non-transformed human cell model of development, providing a valuable resource for future functional studies.

Translated lncRNAs in pancreatic progenitors produce microproteins with distinct subcellular localizations
Having established that many stage-specific pancreatic lncRNAs are translated, we next sought to validate their translation potential through independent experimental approaches, additionally Replicates from six independent differentiations to PP2 stage each for total (polyA) RNA-seq and Ribo-seq experiments, and two biological replicates for the subcellular fractionation were analyzed. The histogram on the far right depicts the size distribution of the sORF-encoded small peptides as number of amino acids (aa). The pie Figure 2 continued on next page investigating the production of the predicted microproteins at the protein level. To this end, we first performed coupled in vitro transcription:translation assays on endogenous and complete transcript isoforms of four of the most highly translated lncRNAs (LINC00261, RP11-834C11.4, LHFPL3-AS2, and MIR7-3HG; Figure 2-figure supplement 1I; expression and ORF information in Figure 2B-E). Second, we generated a series of in vivo translation reporter constructs to assess the subcellular localization of microproteins translated from each of ten sORFs derived from the same four lncRNAs. Transient expression of individual constructs carrying in-frame GFP fusions in HEK293T cells produced GFP signal for all ten assayed microproteins, which was abolished upon introduction of a frameshift within the sORF or a stop codon following the sORF sequence ( Figure 2F and

Deletion phenotypes of translated lncRNAs during hESC pancreatic differentiation
To identify potential functional roles of translated lncRNAs during pancreas development, we selected ten candidates for CRISPR/Cas9-based genome editing in hESCs through excision of the lncRNA promoter or entire lncRNA locus ( Figure 3A,B). These ten lncRNAs were prioritized based on (i) high expression and endodermal tissue-specificity, (ii) dynamic regulation during pancreas development, (iii) abundant translation of sORFs, and (iv) proximity to TFs with known roles in endoderm and pancreas development. For seven of the selected lncRNAs, translation was highly abundant and reproducibly detected across Ribo-seq replicates: LINC00617 (also known as TUNAR; Lin et al., 2014), GATA6-AS1 (also known as GATA6-AS; Neumann et al., 2018), LINC00261, RP11-834C11.4, SOX9-AS1, MIR7-3HG, and LHFPL3-AS2. Although for two additional lncRNAs the translation potential could not be determined, they were nonetheless included because of a previously reported requirement for definitive endoderm formation (DIGIT, also known as GSC-DT) (Daneshvar et al., 2016) and genomic localization adjacent to the definitive endoderm TF LHX1 (RP11-445F12.1, also known as LHX1-DT). Lastly, LINC00479 was chosen as a non-translated control with expression dynamics and a subcellular localization similar to LINC00261. Of note, for each of the ten selected lncRNAs, we generated at least two independent hESC knockout (KO) clones and used different combinations of single guide RNAs where possible ( Figure 3-source data 1A).
We next differentiated each of the lncRNA KO hESC lines stepwise toward the pancreatic endocrine cell stage, conducting up to 16 replicate differentiations for each KO clone. Because . Data are shown as mean, with individual data points represented by dots (n = 2 biological replicates). Right: Subcellular fractionation RNA-seq, Ribo-seq, and P-site tracks (ribosomal P-sites inferred from ribosome footprints on ribosome-protected RNA) for loci of the depicted lncRNAs. Identified highest stringency sORFs (ORF in 6/6 replicates) are shown in red. For LINC00261, visually identified sORFs 1 and 2 are also shown. Heatmaps in the top right visualize the relative expression of the shown lncRNAs during pancreatic differentiation (means of two biological replicates per stage), on a minimum (white)/maximum (dark blue) scale. (F) In vivo translation reporter assays testing whether sORFs computationally defined in (A) give rise to translation products in HEK293T cells when fused in-frame to a GFP reporter. Left: Schematic of the constructs (gray: PGK promoter, black: lncRNA sequence 5' to sORF to be tested, red: sORF, green: GFP ORF). Right: Representative DIC and GFP images of HEK293T cells transiently transfected with the indicated reporter constructs. Scale bars = 50 mm. See also Source data 1. RNA-seq after subcellular fractionation and Ribo-seq in PP2 cells.  , arguing against cis-regulation by these lncRNAs. These findings are in contrast to prior reports that have shown a requirement for LINC00261 and DIGIT in definitive endoderm formation and the regulation of neighboring TFs FOXA2 and GSC, respectively (Amaral et al., 2018;Daneshvar et al., 2016;Jiang et al., 2015;Swarr et al., 2019).
Next, we further differentiated control and KO lines for eight out of ten lncRNAs toward the endocrine cell stage, excluding DIGIT and RP11-445F12.1 because they are not expressed after the definitive endoderm stage ( Figure 3A). In KO hESC lines of seven out of these eight lncRNAs, we observed no effect on pancreatic progenitor cell formation or gene expression, with the exception of a handful of dysregulated genes in LHFPL3-AS2 and RP11-834C11.4 KO cells (Figure 3-figure supplement 1C and Figure 3-source data 1E-K). Furthermore, deletion of seven out of the eight lncRNAs did not impair endocrine cell formation, as determined by quantification of insulin + cells and insulin mRNA levels ( Figure 3G-I). Similar to the RNA expression results obtained at the definitive endoderm stage, deletion of none of the lncRNAs close to pancreatic TFs (e.g. GATA6-AS1 and SOX9-AS1) altered the expression of these TFs, once more arguing against cis-regulation of these TFs by the neighboring lncRNA ( Figure 3-figure supplement 1C). Thus, nine out of ten endodermand pancreatic progenitor-enriched lncRNAs functionally investigated here appear to be nonessential for induction of the pancreatic fate and formation of insulin + cells. Furthermore, these lncRNAs do not appear to control the transcript levels of proximal TFs.

LINC00261 knockout impairs endocrine cell development
The exception was the endoderm-specific lncRNA LINC00261, which is highly expressed and translated in pancreatic progenitors (Figure 4-figure supplement 1A and Figure 2C). While deletion of LINC00261 caused no discernable phenotype in definitive endoderm ( Figure 3C-F and Figure 3figure supplement 1C), we observed a significant 30-50% reduction in the number of insulin + cells at the endocrine cell stage ( Figure 4A,B). This reduction in insulin + cell numbers was consistent Source data 1. Differentially expressed genes after lncRNA deletion. Source data 2. Source data used for the qRT-PCR quantification of gene expression presented in Figure 3A. Source data 3. Source data used for the qRT-PCR quantification of gene expression presented in Figure 3D. Source data 4. Source data used for the qRT-PCR quantification of INS expression presented in Figure 3H.  across four separately derived LINC00261 KO hESC lines, each independently differentiated to endocrine cell stage 5-8 times. In agreement with reduced insulin + cell numbers, insulin content and insulin mRNA levels were also reduced in LINC00261 KO endocrine stage cultures ( Figure 4C,D). Analysis of insulin median fluorescence intensities by flow cytometry further showed no reduction in insulin levels per cell in one LINC00261 KO clone and a mild reduction in the three other clones ( Figure 4E), indicating that LINC00261 predominately regulates endocrine cell differentiation rather than maintenance of insulin production in beta cells.
To determine the molecular effects of LINC00261 deletion, we performed RNA-seq in pancreatic progenitors derived from LINC00261 KO and control hESCs. Similar to the absence of cis-regulatory functions observed in the other lncRNA KOs, we found no evidence for cis-regulation of FOXA2 by LINC00261 ( Figure 4F and  (Artner et al., 2007;Sosa-Pineda et al., 1997). Of note, genes differentially expressed in LINC00261 KO cells mapped to all chromosomes and showed no enrichment for chromosome 20 where LINC00261 resides ( Figure 4G). These results suggest a trans-rather than cis-regulatory function for LINC00261, consistent with its predominantly cytosolic localization, translation, and diffuse distribution within the nucleus ( Figure 2C and Figure 4-figure supplement 1D). Trans-regulatory roles of LINC00261 have also been observed in previous studies (Aguet et al., 2019;Shi et al., 2019;Wang et al., 2019;Wang et al., 2017;Yan et al., 2019). This potential trans functionality prompted us to further investigate whether LINC00261's coding or noncoding features are essential for endocrine cell differentiation. Source data 1. Characterization of LINC00261 knockout and LINC00261-sORF3-frameshift PP2 cells. Source data 2. List of oligonucleotides and synthetic gene fragments used in this study. Source data 3. Source data used for the insulin measurements presented in Figure 4. One-by-one disruption of LINC00261's sORFs does not impact endocrine cell differentiation We established that LINC00261 harbors multiple distinct and highly translated sORFs, which raises the possibility that the translation of these sORFs is functionally important for endocrine cell differentiation. To systematically discriminate LINC00261's coding and noncoding roles, we individually mutated its seven most highly translated sORFs independently in hESCs, leaving the lncRNA sequence, and hence any noncoding function coupled to RNA sequence or structure, grossly intact. Each of these hESC lines either carried a homozygous frameshift mutation near the microprotein's N-terminus (for sORFs 1-6) or a full sORF deletion (sORF7; Figure 4-source data 1B). After verifying that CRISPR editing of the LINC00261 locus did not impact LINC00261 transcript levels (Figure 4-figure supplement 1E), we quantified (i) insulin mRNA levels, (ii) insulin + cells, and (iii) total insulin content in endocrine cell stage cultures. We observed no difference between sORF loss-offunction and control hESC lines for most of these endpoints ( Figure 4H,I and Figure 4-figure supplement 1E), although we noticed that the number of insulin + cells, but not the amount of insulin produced, was reduced in one of the two sORF4 and sORF7 KO clones. Transcriptome analysis of pancreatic progenitors with frameshifts in sORF3 (the most highly translated LINC00261-sORF; Figure 2C and Figure 2-source data 1D) revealed no differentially expressed genes between LINC00261-sORF3 frameshift and control cells ( Figure 4J and Figure 4-source data 1C), contrasting observations in LINC00261 RNA KO pancreatic progenitors ( Figure 4F and Figure 4-source data 1A). These results indicate that there is not one dominant LINC00261 sORF that is required for endocrine cell formation, suggesting a functional role of the LINC00261 transcript and not the individual sORFs mutated here. However, it is possible that the different sORFs, or the microproteins translated from these sORFs, are functionally redundant and capable of phenotypic rescue.
It has been suggested that ribosome association can control lncRNA transcript levels by inducing nonsense-mediated decay (NMD) (Carlevaro-Fita et al., 2016;Tani et al., 2013). Therefore, we determined whether the presence of multiple sORFs could regulate LINC00261 stability. To this end, we simultaneously mutated start codons of all seven sORFs (DATG sORF1-7 LINC00261) and expressed either wild type or DATG sORF1-7 LINC00261 ectopically in HEK293T cells, where LINC00261 is normally not expressed. LINC00261 half-life measurements upon transcriptional inhibition with actinomycin D revealed no difference in LINC00261 levels between wild type and DATG sORF1-7 LINC00261 ( Figure 4K), suggesting that the translation of the seven sORFs does not reduce LINC00261 transcript stability.
In sum, through the systematic, one-by-one removal of sORFs within a highly translated lncRNA with functional importance for pancreatic endocrine cell formation, we found no evidence to implicate the individual sORFs, or the microproteins they produce, in endocrine cell development. Although LINC00261's sORFs may share functional redundancy or have developmental roles that do not affect the production of insulin + cells, our findings strongly suggest that by themselves, these sORFs are not functionally required for endocrine cell formation.

Discussion
Limited cis-regulatory consequences of lncRNA deletion In this study we globally characterized molecular features of lncRNAs expressed during progression of hESCs toward the pancreatic lineage, including their subcellular localization and potential to be translated. We performed a phenotypic CRISPR loss-of-function screen, focusing on ten developmentally regulated, highly expressed, and highly translated lncRNAs proximal to TFs known to regulate pancreas development. The first important observation from this screen is that we found no evidence to implicate the lncRNAs LINC00261, DIGIT, GATA6-AS1, SOX9-AS1, and RP11-445F12.1 in the cis-regulation of their neighboring TFs FOXA2, GSC, GATA6, SOX9, and LHX, respectively, despite tight transcriptional coregulation of the lncRNA-TF pairs.
Contrasting our findings, a number of studies have reported cis-regulation of FOXA2 by LINC00261 (Amaral et al., 2018;Jiang et al., 2015;Swarr et al., 2019). However, several lines of evidence strongly support the conclusion that FOXA2 is not regulated by LINC00261 in our experimental system. First, we examined FOXA2 mRNA expression in LINC000261 -/cells at both the definitive endoderm and pancreatic progenitor cell stages. Second, we analyzed FOXA2 expression using two independent methods, namely qRT-PCR and RNA-seq. Third, immunofluorescence staining in definitive endoderm revealed no difference in FOXA2 protein expression between control and LINC00261 -/cells.
While different cellular contexts and species could explain the discrepancy between our findings and the ones by Amaral et al., 2018and Swarr et al., 2019, Jiang et al., 2015 reported FOXA2 regulation by LINC00261 in hESC-derived definitive endoderm. One important difference between our study and the study by Jiang et al. is that we employed CRISPR-Cas9-mediated deletion, whereas Jiang et al. used shRNA-mediated knockdown to inactivate LINC00261. It is possible that lncRNA deletion triggers compensatory mechanisms that are not activated after shRNA-mediated knockdown. For coding genes, mutant mRNA degradation has been shown to trigger genetic compensation (El-Brolosy et al., 2019). Another difference between our study and the one by Jiang et al. is that our differentiation protocol was more efficient in generating definitive endoderm. It is conceivable that the stability of the cell fate and identity of neighboring cells could influence how LINC00261 loss-of-function affects gene regulation.
Although lncRNAs are now appreciated as a novel and abundant source of sORF-encoded biologically active microproteins (Makarewich and Olson, 2017), it remains largely unknown which translation events lead to the production of microproteins, which solely have regulatory potential, or which have no functional roles, but are not negatively selected against. The cytosolic localization and translation of many RNAs classified as lncRNAs provides a strong rationale for considering both, coding and noncoding functions.
In this study, we identified the translated lncRNA LINC00261 as a novel regulator of pancreatic endocrine cell differentiation, as evidenced by a severe reduction in insulin + cell numbers upon LINC00261 deletion. We show that LINC00261 transcripts are highly abundant in pancreatic progenitors and, albeit present in the nucleus, are predominantly localized to the cytoplasm. Here, they frequently associate with ribosomes which leads to the translation of multiple independent sORFs. We show that the sORFs are capable of producing microproteins with distinct subcellular localizations upon expression in vitro. In contrast to LINC00261 deletion, individual frameshift mutations in each of LINC00261's sORFs did not impair endocrine cell development, suggesting that the requirement of LINC00261 for endocrine cell development can be uncoupled from the translation of its multiple sORFs. However, this does not exclude the possibility that these sORFs or the microproteins they produce could possess functions that become relevant under specific environmental, developmental, or disease conditions not examined in this study.
We found that mutating all translated LINC00261 sORFs simultaneously, thereby likely reducing LINC00261's ability to bind ribosomes, did not affect LINC00261 transcript levels in HEK293T cells. This indicates that, in contrast to reports suggesting that translated sORFs can regulate RNA stability by promoting nonsense-mediated RNA decay (Carlevaro-Fita et al., 2016;Tani et al., 2013), the high translation levels and multiple sORFs of LINC00261 are unlikely to be part of a LINC00261 decay pathway. It would have been interesting to determine how concurrent mutation of all sORFs in LINC00261 affects pancreatic cell differentiation. However, given the size of the LINC00261 locus and the many sORFs, such an approach comes with technical challenges and significant caveats.
LINC00261 -a potential trans regulator of endocrine cell differentiation?
Several lines of evidence suggest that LINC00261 regulates endocrine cell differentiation in trans: (i) LINC00261 transcripts show a diffuse distribution in multiple subcellular compartments, (ii) genes differentially expressed in LINC00261 KO cells are randomly distributed throughout the genome, (iii) expression of the nearby TF FOXA2 is not affected by LINC00261 deletion. Such a trans regulatory mechanism for LINC00261 is supported by a recent study from the GTEx Consortium, where LINC00261 is highlighted as one of a few lncRNAs that forms a potential trans regulatory hotspot through genetic interactions that influence the expression of multiple distant genes (Aguet et al., 2019). Consistent with its preferential cytosolic localization, and further supporting the notion of a trans regulatory mechanism, LINC00261 has been suggested to regulate gene expression through non-nuclear mechanisms, e.g. by preventing nuclear translocation of b-catenin (Wang et al., 2017) or by acting as a miRNA sponge Wang et al., 2019;Yan et al., 2019). Although our observations and current literature strongly hint to a function in trans independent of the produced microproteins, the exact mechanism by which LINC00261 regulates gene expression in pancreatic progenitors remains to be determined.

Limitations and future directions
In this study, we have characterized the role of translated lncRNAs, and in particular LINC00261, in a hESC differentiation system that mimics pancreas development. However, there are several potential limitations that need to be considered when interpreting the results. First, a small subset of analyses in this study was based on low numbers of replicate differentiations, in particular the cytosolic versus nuclear fractionation RNA-seq experiments, where only two replicate differentiations into pancreatic progenitor cells were analyzed. Second, although we provide evidence that LINC00261 can produce microproteins using Ribo-seq, which is further supported by in vitro translation assays and overexpression of LINC00261 constructs with different in-frame tags, we provide no protein-level evidence for the endogenous production and stability of LINC00261's microproteins in this differentiation system or in human pancreas development in vivo. Moreover, due to its highly specific expression pattern, LINC00261 has not been previously detected by sORF analyses in other cell types (Bazzini et al., 2014;Calviello et al., 2016;Chen et al., 2020;Ji et al., 2015;Martinez et al., 2020;Prensner et al., 2020;Raj et al., 2016;van Heesch et al., 2019). Even though we show microprotein production in vitro, it is possible that the act of translation has a key regulatory role rather than the protein products of LINC00261's sORFs. Lastly, LINC00261's microproteins and sORFs may have redundant functions, which could explain why deletion of individual sORFs produces no apparent phenotype. Thus, despite limited sequence similarity and stark differences in translation rates between the identified translated sORFs in LINC00261, we cannot rule out that different microproteins produced by LINC00261 compensate when one sORF is deleted. Future studies of LINC00261's precise mechanisms of action could be aimed at further dissecting the potential regulatory features of sORF translation and possibility of redundancy between sORFs.

Conclusions
In summary, we here present a rigorous, in-depth characterization of dynamically regulated and translated lncRNAs in a disease-relevant cell model of human developmental progression. Our combination of ultra-high-coverage RNA-and Ribo-seq, in vitro protein-level validation of microprotein production and localization, and the systematic, one-by-one deletion of all individual microproteins encoded by a single translated lncRNA, not only provides a detailed resource of translated 'noncanonical' sORFs and their microproteins in pancreatic development, but also serves as a blueprint for the systematic functional interrogation of translated lncRNAs. hESC culture and maintenance H1 hESCs (male) were obtained from WiCell (NIHhESC-10-0043, RRID:CVCL_9771) and tested for mycoplasma on a yearly basis. H1 hESCs were grown in feeder-independent conditions on Matrigelcoated dishes (Corning, Cat# 356231) with mTeSR1 media (STEMCELL Technologies, Cat# 85850). Propagation was carried out by passing the cells every 3 to 4 days using Accutase (eBioscience, Cat# 00-4555-56) for enzymatic cell dissociation. hESC research was approved by the University of California, San Diego, Institutional Review Board and Embryonic Stem Cell Research Oversight Committee.
To generate sORF frameshift mutations, sgRNA sequences targeting the N-terminal region of the predicted small peptides were inserted into pSpCas9(BB)À2A-GFP (Addgene plasmid #48138, RRID: Addgene_48138, gift from Feng Zhang) via its BpiI cloning sites. 3 mg of the resulting plasmids were then transfected into 500,000 H1 cells plated into Matrigel-coated six-wells the day prior, using Xtre-meGene 9 Transfection Reagent (Sigma-Aldrich) according to the manufacturer's instructions. 24 hr post-transfection, 10,000 GFP + cells were sorted on an Influx Cell Sorter (BD Biosciences) into Matrigel-coated six-wells containing 1 mL mTeSR1 media supplemented with 10 mM ROCK inhibitor and 1X penicillin/streptomycin. Seven days after sorting, emerging colonies were hand-picked and transferred into 96-well plates for genotyping. Frameshifts inside the targeted sORFs were confirmed by PCR-amplification of the sORF sequence with GoTaq Green Mastermix (Promega, Cat# M7123) and subsequent subcloning the PCR products into pCR2.1 (Thermo Fisher Scientific). For each hESC clone, at least six pCR2.1 clones were Sanger sequenced. Oligonucleotide sequences for sgRNA cloning are provided in Figure 4-source data 2A.

PCR genotyping of CRISPR clones
Four days after transfer of single cell-derived clones into 96-wells, cell culture supernatants containing dead cells were collected from each well prior to the daily media change. Cell debris was then pelleted and used for gDNA extraction with 10-20 ml QuickExtract DNA Extraction Solution (Lucigen, Cat# QE09050) according to the manufacturer's instructions. 1 ml DNA was then PCR-amplified with GoTaq Green Mastermix (Promega, Cat# M7123) and locus-specific primers that anneal either within or outside of the excised genomic DNA. PCR products generated with 'inside' primers were visualized on a 2% agarose gel, PCR bands generated with primers flanking the deletion were gelpurified and submitted for Sanger sequencing (see Figure 4-source data 2B for genotyping and sequencing primers).
For genotyping of sORF frameshift clones, PCR amplicons designed to encompass the Cas9 cut site were amplified and Sanger sequenced (Figure 4-source data 2B). If out-of-frame indels were apparent in the sequencing chromatogram, the sequenced PCR product was ligated into pCR2.1-TOPO via TOPO-TA cloning. A minimum of six clones were Sanger sequenced in order to determine the genotype at both alleles with high confidence.

Generation of sORF translation reporter plasmids
The four lncRNAs tested were PCR-amplified with KOD Xtreme DNA Hotstart Polymerase (Millipore) from their 5' end up until the last codon of the sORF to be tested, omitting its stop codon (primer sequences are listed in Figure 4-source data 2D). cDNA was used as PCR template for LINC00261 and LHFPL3-AS2; RP11-834C11.4, and MIR7-3HG were amplified from a gBlock synthetic gene fragment (Integrated DNA Technologies; see Figure 4-source data 2F). The GFP coding sequence (without start codon; amplified from pRRLSIN.cPPT.PGK-GFP.WPRE; RRID:Addgene_12252) was then fused in-frame to the sORF via overlap extension PCR. The resulting fusion product was cloned into pRRLSIN.cPPT.PGK-GFP.WPRE (Addgene plasmid #12252, gift from Didier Trono) via BshTI and SalI restriction sites included in the PCR primers. Due to the 3'-location of sORF7 within LINC00261, not the entire LINC00261 cDNA was amplified but only 65 bp preceding sORF7.

Flow cytometry analysis
For intracellular flow cytometry, single cells were washed three times in FACS buffer (0.1% (w/v) BSA (Thermo Fisher Scientific in PBS) and then fixed and permeabilized with Cytofix/Cytoperm Fixation/ Permeabilization Solution (BD Biosciences) for 20 min at 4˚C, followed by two washes in BD Perm/ Wash Buffer. Cells were next incubated with either PE-conjugated anti-SOX17 antibody (BD Biosciences; Cat# 561591, RRID:AB_10717121), or PE-conjugated anti-INS antibody (Cell Signaling Technology, Cat# 8508S, RRID:AB_11179076) in 50 ml BD Perm/Wash Buffer for 60 min at 4˚C. Following three washes in BD Perm/Wash Buffer, cells were analyzed on a FACSCanto II (BD Biosciences) cytometer.

Insulin content measurements
To measure total insulin content of endocrine cell stage control and lncRNA KO cells, adherent cultures were enzymatically detached from a 24-well at day 16 of differentiation. Upon quenching with FACS buffer (0.1% (w/v) BSA (Thermo Fisher Scientific in PBS), the cells were pelleted and extracted over night at 4˚C in 100 ml acid-ethanol (2% HCl in 80% ethanol). Insulin was measured by Insulin ELISA (Alpco, Cat# 80-INSHU-E10.1) and normalized to total protein, as quantified with a BCA protein assay (Thermo Fisher Scientific, Cat# 23227).

Quantitative reverse transcription PCR (qRT-PCR)
Total RNA was isolated from hESC-derived cells and HEK293T cells using either TRIzol (Thermo Fisher Scientific, Cat# 15596018) or the RNAeasy Mini Kit (Qiagen, Cat# 15596018), respectively. Upon removal of genomic DNA (TURBO DNA-free Kit, Thermo Fisher Scientific, Cat# AM1907 or RNase-free DNase Set, Qiagen, Cat# 79254) cDNA was synthesized using the iScript cDNA Synthesis Kit (Bio-Rad, Cat# 1708890). PCR reactions were run in triplicate with 6.25-12.5 ng cDNA per reaction using the CFX96 Real-Time PCR Detection System (BioRad). TATA-binding protein (TBP) was used as endogenous control to calculate relative gene expression using the DDCt method. Primer sequences are provided in Figure 4-source data 2C.

Transient transfection of HEK293T cells with polyethylenimine (PEI)
Two hours prior to transfection, fresh pre-warmed DMEM medium (Corning, Cat# 45000-312) was added to each well. Transfection mix was prepared by combining PEI (Polysciences Cat# 23966-1) and plasmid DNA (4:1 ratio; 4 mg PEI per 1 mg DNA) in Opti-MEM Reduced Serum Medium (Thermo Fisher Scientific, Cat# 31985062) followed by brief vortexing. After five minutes, the transfection complex was added dropwise to the cells.
To express LINC00261 (wild type) and LINC00261 (DATG sORF1-7 ) in HEK293T cells, the cells were plated in 6-well plates and transduced with lentivirus the following day. Two days post infection, the cells were passaged for RNA half-life measurements.

LINC00261 RNA half-life measurement
HEK293T cells transduced with either LINC00261 (wild type) or LINC00261 (DATG sORF1-7 ) lentivirus were seeded in six 24-wells. 48 hr after plating, cells from one well were collected for RNA isolation as the '0 hr' time point. To the remaining five wells, 100 ml growth media supplemented with 10 mg/ ml actinomycin D (Sigma-Aldrich Cat# A9415) were added to inhibit transcription. At 2, 4, 6, 8, and 9 hr following actinomycin D addition, samples were collected for RNA isolation. Total RNA was then reverse transcribed and analyzed by qPCR, where the abundance of each time point was calculated relative to the abundance at the 0 hr time point (DCt). The half-life was then determined by non-linear regression (One phase decay; GraphPad Prism).
Single molecule RNA fluorescence in situ hybridization (smRNA FISH) H1-derived PP2 stage cells (control and LINC00261 KO) were cultured on Matrigel-coated 12 mm diameter coverslips in a 24-well plate. Following 10 min fixation in 1 mL Fixation Buffer (3.7% (v/v) formaldehyde in PBS) at room temperature, the cells were washed twice in PBS and subsequently permeabilized in 70% (v/v) ethanol for one hour at 4˚C. Following a five minute wash in Stellaris RNA FISH Wash Buffer A (LGC Biosearch Technologies, Cat# SMF-WA1-60; 1:5 diluted concentrate, with 10% (v/v) formamide added), the coverslips were incubated in a humidified chamber at 37˚C for 14 hr with probes diluted in Stellaris RNA FISH Hybridisation Buffer (LGC Biosearch Technologies, Cat# SMF-HB1-10; with 10% (v/v) formamide added) to 125 nM. After a 30 min wash at 37˚C in Wash Buffer A, the cells were counter-stained with Hoechst 33342 (Thermo Fisher Scientific) for 15 min and washed in RNA FISH Wash Buffer B (LGC Biosearch Technologies, Cat# SMF-WB1-20) for 5 min at room temperature. The coverslips were mounted in Vectashield Mounting Medium (Vector Laboratories, Cat# H-1000) and imaged on a UltraView Vox Spinning Disk confocal microscope (Perki-nElmer) using a 100X oil objective.

In vivo translation assays
Reporter plasmids were transfected into HEK293T cells using PEI, and 36 hr post transfection live cells were imaged on an EVOS Cell Imaging System (Thermo Fisher Scientific) equipped with a 20X objective. Additional constructs were generated that served as negative controls (no GFP fluorescence): 1) a LINC00261-sORF3-GFP construct with a single 'T' insertion inside sORF3, causing a frame-shift, 2) a LINC00261-sORF2-GFP construct with a stop codon preceding the GFP coding sequence, and 3) a LINC00261-sORF1-GFP construct with a frame-shift mutation within the GFP coding sequence.
Stranded mRNA-seq library preparation for lncRNA KOs Total RNA from PP2 cells differentiated with the Rezania et al., 2012 protocol was isolated and DNase-treated using either TRIzol (Thermo Fisher Scientific), or the RNAeasy Mini kit (Qiagen) according to the manufacturer's instructions. RNA integrity (RIN > 8) was verified on the Agilent 2200 TapeStation (Agilent Technologies), and 400 ng RNA was used for multiplex library preparation with the KAPA mRNA HyperPrep Kit (Roche; Cat# KK8581). All libraries were evaluated on TapeStation High Sensitivity DNA ScreenTapes (Agilent Technologies; Cat# 5067-5584) and with the Qubit dsDNA High Sensitivity (Life Technologies; Cat# Q10212) assays for size distribution and concentration prior to pooling the multiplexed libraries for single-end 1 Â 51 nt or 1 Â 75 sequencing on the HiSeq 2500 or HiSeq 4000 System (Illumina). Libraries were sequenced to a depth of > 20M uniquely aligned reads.
Cell fractionation and ribo-minus RNA-seq H1 hESCs were differentiated to the PP2 stage with the Rezania et al., 2012 protocol, then nuclear and cytosolic RNA was isolated with the Paris Kit (Thermo Fisher Scientific). Unfractionated total RNA was set aside as a control. All samples were DNaseI-treated prior to further processing (TURBO DNA-free Kit; Thermo Fisher Scientific). rRNA-depleted total RNA-seq libraries were prepared with TruSeq Stranded Total RNA Library Prep Gold (Illumina; Cat# 2002059), and sequencing was performed on a HiSeq4000 instrument.
Alignment of lncRNA KO mRNA-seq samples and processing for gene expression analysis Using the Spliced Transcripts Alignment to a Reference (STAR) aligner (STAR 2.5.3b; Dobin et al., 2013), sequence reads were mapped to the human genome (hg38/GRCh38) with the Ensembl 87 annotations in 2-pass mapping mode, allowing for up to six mismatches. Cufflinks (part of the Cufflinks version 2.2.1 suite; Roberts et al., 2011;Trapnell et al., 2010), was then used to quantify the abundance of each transcript cataloged in the Ensembl 87 annotations in reads per kilobase per million mapped reads (RPKM). For plotting expression values, a pseudocount of 1 was added to all RPKM values prior to log 2 -transformation.
Genes with RPKM ! 1 across two replicates were deemed expressed. Differential gene expression was tested using the DESeq2 v1.10.1 Bioconductor package (Love et al., 2014) with default parameters. Input count files for DESeq2 were created with htseq-count from the HTSeq Python library (Anders et al., 2015). Genes with a > 2 fold change and an adjusted p-value of <0.01 were considered differentially expressed.
The chromosomal localization of genes differentially expressed upon LINC00261 KO was visualized with the RCircos package in R (https://cran.r-project.org/web/packages/RCircos/index.html).

Ribosome profiling and matching RNA-seq
Ribosome profiling was performed on PP2 cells obtained from six independent differentiations of H1 hESCs with the Rezania et al., 2014 protocol, yielding an average of 89% PDX1-positive cells. Ribosome footprinting and sequencing library preparation was done with the TruSeq Ribo Profile (Mammalian) Library Prep Kit (Illumina, Cat# RPYSC12116, currently out of production) according to the TruSeq Ribo Profile (Mammalian) Reference Guide (version August 2016). In short, 50 mg of PP2 aggregates were washed twice with cold PBS and lysed for 10 min on ice in 1 mL lysis buffer (1 Â TruSeq Ribo Profile mammalian polysome buffer, 1% Triton X-100, 0.1% NP-40, 1 mM dithiothreitol, 10 U ml À1 DNase I, cycloheximide (0.1 mg/ml) and nuclease-free H 2 O). Per sample, 400 mL of lysate was further processed according to manufacturer's instructions. Final library size distributions were checked on the Bioanalyzer 2100 using a High Sensitivity DNA assay (Agilent Technologies), multiplexed and sequenced on an Illumina HiSeq 4000 producing single end 1 Â 51 nt reads. Ribo-seq libraries were sequenced to an average depth of 85M reads.
Total RNA was isolated using TRIzol Reagent (Thermo Fisher Scientific) from the exact same cell cultures processed for ribosome profiling (10% of the total number of cells). Total RNA was DNase treated and purified using the RNA Clean and ConcentratorÀ25 kit (Zymo Research). RIN scores (RIN = 10 for all six samples) were measured on a BioAnalyzer 2100 using the RNA 6000 Nano assay (Agilent Technologies). Poly(A)-purified mRNA-seq library preparation was performed according to the TruSeq Stranded mRNA Reference Guide (Illumina), using 500 ng of total RNA as input. Libraries were multiplexed and sequenced on an Illumina HiSeq 4000 producing paired-end 2 Â 101 nt reads.

Alignment of Ribo-seq and matched mRNA-seq samples
Prior to mapping, ribosome-profiling reads were clipped for residual adapter sequences and filtered for mitochondrial, ribosomal RNA and tRNA sequences (Figure 2-source data 1). Next, all mRNA and ribosome profiling data were mapped to the Ensembl 87 transcriptome annotation of the human genome hg38 assembly using STAR 2.5.2b (Dobin et al., 2013) in 2-pass mapping mode. To avoid mRNA-seq mapping biases due to read length, the 2 Â 101 nt mRNA-seq reads were next trimmed to 29-mers, and those mRNA reads were processed and mapped with the exact same settings as the ribosome profiling data. For the mapping of 2 Â 101 nt RNA-seq reads six mismatches per read were allowed (default is 10), whereas two mismatches were permitted for the Ribo-seq and trimmed mRNA-seq reads. To account for variable ribosome footprint lengths, the search start point of the read was defined using the option -seedSearchStartLmaxOverLread, which was set to 0.5 (half the read, independent of ribosome footprint length). Furthermore, -outFilterMulti-mapNmax was set to 20 and -outSAMmultNmax to 1, which prevents the reporting of multimapping reads.

Detecting actively translated reading frames
Canonical ORF detection using ribosome profiling data was performed with RiboTaper v1.3 (Calviello et al., 2016) with standard settings. For each sample, we selected only the ribosome footprint lengths for which at least 70% of the reads matched the primary ORF in a meta-gene analysis. Following the standard configuration of RiboTaper, we required ORFs to have a minimum length of 8aa, evidence from uniquely mapping reads and at least 21 P-sites. The final list of translation events was stringently filtered requiring the translated gene to have an average RNA RPKM ! 1 and to be detected as translated in all six profiled samples. Furthermore, we required the exact ORF to be detected independently in at least 4 out of 6 samples.

Translational efficiency estimates
Translational efficiency (TE) estimations were calculated as the ratio of Ribo-seq over mRNA-seq DESeq2 normalized counts, yielding independent gene-specific TEs for each of the six individual replicate differentiations. For this, mRNA-seq and Ribo-seq based expression quantification was calculated for (annotated and newly detected) coding sequences (CDSs/ORFs) only, using RNA reads trimmed to footprint sizes as described above.