Long non-coding RNA GRASLND enhances chondrogenesis via suppression of interferon type II signaling pathway

Long non-coding RNAs (lncRNAs) play critical roles in regulating gene expression and cellular processes; however, their roles in musculoskeletal development, disease, and regeneration remain poorly understood. Here, we identified a novel lncRNA, Glycosaminoglycan Regulatory ASsociated Long Non-coDing RNA (GRASLND) as a regulator of mesenchymal stem cell (MSC) chondrogenesis, and we investigated its basic molecular mechanism and its potential application towards regenerative medicine. GRASLND, a primate-specific lncRNA, is upregulated during MSC chondrogenesis and appears to act directly downstream of SRY-Box 9 (SOX9), but not Transforming Growth Factor Beta 3 (TGF-β3). Utilizing the established model of pellet formation for MSC chondrogenesis, we showed that the silencing of GRASLND resulted in lower accumulation of cartilage-like extracellular matrix, while GRASLND overexpression, either via transgene ectopic expression or by endogenous activation via CRISPR, significantly enhanced cartilage matrix production. GRASLND acts to inhibit interferon gamma (IFN-γ) by binding to Eukaryotic Initiation Factor-2 Kinase EIF2AK2. We further demonstrated that GRASLND exhibits a protective effect in engineered cartilage against interferon type II across different sources of chondroprogenitor cells. Our results indicate an important role of GRASLND in regulating stem cell chondrogenesis, as well as its therapeutic potential in the treatment of cartilage-related diseases, such as osteoarthritis. Significance Long non-coding RNAs (lncRNAs) play critical roles in gene regulation and cellular physiology; however, the role of lncRNAs in controlling stem cell chondrogenesis remains to be determined. Here, we utilized next generation sequencing of adult stem cell chondrogenesis to identify a set of potential lncRNA candidates involved in this process. We identified lncRNA Glycosaminoglycan Regulatory ASsociated Long Non-coDing RNA (GRASLND) and characterized its molecular mechanism of action. We described a novel role of GRASLND in positive regulation of chondrogenesis via its inhibition of type II interferon. Importantly, we showed that overexpression of GRASLND augments stem cell chondrogenesis, providing a promising approach to enhancing stem cell chondrogenesis and cartilage regeneration.


/body
Type II, on the other hand, relies on activation of the JAK/STAT pathway following the binding of 23 IFN gamma (IFN-γ) to Interferon Gamma Receptors (IFNGRs). This subsequently results in the 24 phosphorylation and dimerization of STAT1 that translocates into the nucleus and induces 25 downstream targets via the gamma activated sequence (GAS) DNA binding element (32-34).
Although interferons (IFN) are widely known for their antiviral response, they can also act in 1 other aspects of cellular regulation (33). Interestingly, IFN-γ has been implicated in non-viral 2 processes, most notably its priming effect in auto-immune diseases such as lupus nephritis, 3 multiple sclerosis, or rheumatoid arthritis (35). An additional goal of this study was to elucidate 4 the link between IFN-γ and our lncRNA candidate, and how this interaction could potentially play 5 a role in MSC chondrogenesis and cartilage tissue engineering.

6
In a recent publication, we used high-depth RNA sequencing to map the transcriptomic 7 trajectory of MSC chondrogenesis (36). This dataset provides a unique opportunity to identify 8 candidate genes for subsequent functional characterization as regulators of chondrogenesis.

9
Here, we used bioinformatic approaches to integrate our RNA-seq data with other publicly 10 available datasets, applying a rational and systematic data mining method to define a 11 manageable list of final candidates for follow-up experiments. As a result, we identified 12 RNF144A-AS1 to be a crucial regulator of chondrogenesis, and proposed the name 13 Glycosaminoglycan Regulatory ASsociated Long Non-coDing RNA (GRASLND). We showed 14 that GRASLND enhances chondrogenesis by acting to suppress the IFN-γ signaling pathway, 15 and this effect was prevalent across different adult stem cell types and conditions. Together, 16 these results highlight novel roles of GRASLND and its modulation of IFN in stem cell 17 chondrogenesis, as well as its therapeutic potential to enhance cartilage regeneration.

11
To validate these gene expression findings, we performed RNA fluorescence in situ 12 hybridization (FISH) throughout the time course of MSC chondrogenesis. Pellets exhibited 13 RNF144A-AS1 FISH signals at later time points during chondrogenic differentiation, consistent 14 with RNA-seq data (Figure 2A). Next, to confirm RNF144A-AS1 subcellular location, we 15 performed qRT-PCR on isolated nuclear and cytoplasmic fractions of day 21 MSC pellets 16 (Figure 2B). We compared the subcellular expression patterns of RNF144A-AS1 to NEAT1 17 (Nuclear Paraspeckle Assembly Transcript 1) and GAPDH (Glyceraldehyde 3-Phosphate 18 Dehydrogenase). NEAT1 is a lncRNA previously characterized to localize to the nucleus (38, 19 39), and GAPDH is an mRNA and thus should be exported to the cytoplasm for protein 20 synthesis. Consistent with previous findings, NEAT1 displayed lower expression in the 21 cytoplasmic compared to the nuclear fraction, in contrast to GAPDH. By this measurement,

25
we speculate that this lncRNA may function in the form of an RNA-protein complex.

Characterization of RNF144A-AS1
1 We examined the characteristics of RNF144A-AS1 by first exploring its evolutionary 2 conservation. Except for exon 1, the genomic region of RNF144A-AS1 is highly conserved in 3 primates (Homo sapiens, Pan troglodytes, and Rhesus macaque) whose common ancestor 4 traced back to 25 million years ago (40), while sequences are less conserved in other mammals 5 (Figure 3A). This suggests that RNF144A-AS1 may belong to a group of previously identified 6 primate-specific lncRNAs (41, 42).

7
Per GENCODE categorization, the AS (antisense) suffix indicates a group of lncRNAs that 8 are positioned on the opposite strand, with overlapping sequences to their juxtaposed protein-9 coding genes. Often, these lncRNAs play a role in regulating the expression of their proteincoding counterparts (22). Therefore, we set out to examine whether this is also the case for

18
Next, we explored the signaling axis of GRASLND. Data mining and computational analysis 19 on earlier published data suggested that GRASLND was a downstream effector of SOX9 20 (GSE69110) (37). When SOX9 was overexpressed in fibroblasts, GRASLND expression was 21 increased (~ 2-fold). We further confirmed this by utilizing SOX9 transgene overexpression in our MSCs culture ( Figure 3E). Interestingly, while TGF-β3 has been demonstrated to act 23 upstream of SOX9, exogenous addition of this growth factor alone did not result in enhanced 24 GRASLND expression. It is notable that SOX9 levels in GFP controls were indistinguishable 25 between TGF-β3 conditions at the time of investigation (1 week in monolayer culture), consistent with our previous finding that SOX9 was not upregulated until later timepoints in MSC 1 chondrogenesis (36). Therefore, TGF-β3, despite being a potent growth factor, is not sufficient 2 to elevate GRASLND expression. Instead, GRASLND appeared to be a downstream target of 3 SOX9.

4
Enhanced chondrogenesis for cartilage tissue engineering with GRASLND 5 As knockdown of GRASLND inhibited GAG and collagen deposition, we sought to 6 investigate whether overexpression of GRASLND would enhance chondrogenesis. We

10
We designed our lentiviral transfer vector to carry a BGH-pA (Bovine Growth Hormone 11 Polyadenylation) termination signal downstream of GRASLND to allow for its correct processing 12 ( Figure S4A). Additionally, GRASLND was also driven under a doxycycline inducible promoter, 13 enabling the temporal control of its expression. We utilized this feature to induce GRASLND 14 only during chondrogenic culture (Figure 4A). This experimental design focused solely on the 15 role of GRASLND during chondrogenesis, while successfully eliminating its effect in MSC 16 maintenance and expansion from our analysis. As control, a vector encoding the Discosoma sp.

20
To determine whether GRASLND would improve chondrogenesis at lower doses of growth 21 factor or at earlier time points, we compared DNA and GAG levels from pellets cultured under different TGF-β3 concentrations on day 7, day 14, and day 21 . In agreement 23 with our knockdown data, DNA content was unaffected. On the other hand, increases in GAG 24 were observed at higher doses and at later time points, especially at 10 ng/mL of TGF-β3. It deposition, and GRASLND may act in concert with other downstream effectors, which were not 1 present at lower doses of TGF-β3 or at earlier time points in the process.

2
Elevated levels of GRASLND resulted in higher amounts of GAG deposition (p < 0.001) 3 (Figure 4B), consistent with our data on the gene expression level ( Figure 4D). We observed a 4 slight increase in chondrogenic markers (COL2A1, ACAN), and a slight decrease in the 5 apoptotic marker CASP3, while cellular senescence was not different between the two groups 6 (TP53) ( Figure 4D). Histologically, pellets derived from dsRed-transduced MSCs exhibited 7 normal GAG and collagen type II staining, indicating successful chondrogenesis. The control 8 pellets were indistinguishable from those derived from GRASLND-transduced MSCs ( Figure   9 4F), albeit macroscopically smaller at the time of harvest.

10
These findings were further confirmed using CRISPR-dCas9-VP64 mediated activation of 11 endogenous GRASLND. This system had been previously utilized to upregulate various 12 transcription factors that efficiently induce embryonic fibroblasts into neurons (43, 44). After 13 screening eleven synthetic gRNAs, we selected the one with highest activation level ( Figure   14 S5). When GRASLND was transcriptionally activated with CRISPR-dCas9, chondrogenesis was 15 enhanced as evidenced by elevated amount of GAG deposition (p < 0.01); DNA amount may 16 also be slightly increased, albeit not statistically significant ( Figure 4C). Similar trends were 17 detected by qRT-PCR ( Figure 4E) and histology (Figure 4F). It is worth noting that CRISPR-dCas9 mediated activation only resulted in a moderate up-regulation of GRASLND relative to 19 transgene ectopic expression (2-fold vs 100-fold). However, the functional outcome was more 20 pronounced with CRISPR-dCas9. We observed approximately 50% increase in the level of 21 GAG produced when normalized to DNA (9.4 ± 2.19 mg/mg vs 16.3 ± 2.08 mg/mg), compared 22 to 30% detected with ectopic expression (10.5 ± 0.84 mg/mg vs 13.9 ± 0.52 mg/mg).  Figure 5B). Surprisingly, pathways pertaining to interferon response were highly

20
To further confirm this relationship, we performed luciferase reporter assays for interferon 21 signaling upon GRASLND knockdown. Utilizing specific reporter constructs, we were able to 22 determine whether GRASLND acted on type I or type II IFN. Our results indicated that 23 decreased level of GRASLND led to heightened type II (IFN-γ) ( Figure 5E) response but not 24 type I (IFN-β) ( Figure 5D). Importantly, luminescence activities between scrambled control and This indicates that at basal level, the two groups responded similarly to lentiviral transduction, 1 and the observed difference in IFN signal was a consequence of GRASLND downregulation.
part of an RNA-protein complex. To test this, we performed an RNA pull down assay, followed 4 by mass spectrometry. Here, streptavidin beads were used as control, or conjugated to sense or 5 antisense strands of GRASLND. Naked or conjugated beads were then incubated with lysates 6 from day 21 pellets, from which bound proteins were eluted for further analyses. We found that 7 Interferon-Induced Double-Stranded RNA-Activated Protein Kinase (EIF2AK2) peptides were 8 detected at elevated levels in sense samples as compared to antisense controls (p < 0.05); 9 peptides were undetected in naked bead controls. Subsequent RNA pull-down followed by 10 western blot confirmed EIF2AK2 as a binding partner of GRASLND ( Figure S6). We detected an 11 increased level of EIF2AK2 bound to the sense strand of GRASLND relative to the antisense or 12 the pellet lysate control. We speculate that this association of GRASLND RNA to EIF2AK2 13 could potentially result in downregulation of IFN-γ signaling.
IFN-related genes were highly elevated in cartilage tissues of osteoarthritis patients: STAT1, 16 IFNGR2, NCAM1, MID1 ( Figure S7A). Since the microarray did not contain probes for 17 GRASLND, no information on its expression could be extracted. In addition, we identified 18 another independent study that reported changes in the transcriptomes of intact and damaged 19 cartilage tissues (E-MTAB-4304) (46). Similarly, a cohort of IFN-related genes were also 20 upregulated in damaged cartilage, especially STAT1 and IFNGR1 ( Figure S7B). Interestingly, 21 we identified a negative correlation between GRASLND and a few IFN related genes (IFNGR1, ICAM1) in damaged cartilage ( Figure S7C). Therefore, we proposed that GRASLND may 23 possess some therapeutic potential through suppressing IFN signaling in osteoarthritis. To 24 evaluate this possibility, we implemented the use of the GRASLND transgene in engineered 25 cartilage cultured under IFN addition (100 ng/mL of IFN-β or 5 ng/mL of IFN-γ). We determined doses of IFN-b and IFN-γ by selecting the lowest concentration at which day 21 pellets exhibited 1 GAG loss compared to no IFN control. Consistent with luciferase reporter assays, the protective 2 effect of GRASLND was observed upon IFN-γ challenge but not IFN-β ( Figure 5F, G). However, 3 we observed a reduced level of GAG production compared to normal conditions, suggesting 4 that GRASLND can protect the ECM from degradation, but not completely to control levels.

GRASLND enhanced the chondrogenesis of adipose-derived stem cells 6
To determine if the function of GRASLND is unique to MSCs or present in other adult stem 7 cells, we addressed whether modulating GRASLND expression could also improve 8 chondrogenesis of adipose stem cells (ASCs). We observed an increase in GAG production 9 when GRASLND was overexpressed in ASCs compared to control (p < 0.0001) ( Figure 6A),

10
although ACAN levels were not significantly increased. Importantly, COL2A1 expression was 11 significantly elevated (~ 5-fold) with overexpression of GRASLND ( Figure 6B). Histologic 12 examination of the engineered cartilage showed a similar level of collagen type II in pellets with

13
GRASLND overexpression compared to the dsRed control ( Figure 6C). Based on these data, it 14 appears that GRASLND utilized the same mechanism across these two cell types, asserting a 15 pan effect on potentiating their chondrogenic capabilities.

22
GRASLND and IFN-γ signaling pathway in this process, which was confirmed by the 23 identification of EIF2AK2 as its binding partner. Unfortunately, lack of a known murine homolog makes it difficult to study GRASLND in vivo, and thus future studies may require GRASLND 1 transgenic models in primate species.

2
In the context of the musculoskeletal system, IFN is mostly recognized for its role in bone 3 development and homeostasis (23-27, 30), myogenesis (29,47,48), as well as its crosstalk with 4 TGF-β in wound healing (49). Notably, IFN-γ has been suggested to inhibit collagen synthesis in 5 dermal fibroblasts, myofibroblasts, and articular chondrocytes (49-53). Furthermore, the 6 JAK/STAT pathway, which involves IFN downstream effectors, has also been shown to inhibit 7 chondrocyte proliferation and differentiation (28, 31). Here, we found that GRASLND acts to 8 suppress the IFN mechanism. In addition, we also present evidence indicating an interaction 9 between GRASLND and EIF2AK2 (also referred to as PKR). Canonically a crucial player in 10 protein synthesis, PKR has also been reported to control STAT signaling by directly binding to 11 and preventing its association with DNA for gene activation (54, 55). Additionally, several 12 studies have suggested that highly structured, single stranded RNA can also activate PKR via 13 its double stranded RNA binding domains (dsDRBs) (56-60). Our RNA-seq data suggested that 14 upon GRASLND knockdown, a cohort of downstream targets of STATs were upregulated.

15
Based on the presence of DNA binding motifs in investigated targets, we identified both STAT1 16 and STAT2 as potential regulators of genes disrupted by GRASLND knockdown. However, our 17 luciferase reporter assays pointed towards a mechanism in IFN type II (gene activation by 18 STAT1 homodimer) rather than type I (gene activation by STAT1/STAT2 heterodimer). Thus,

19
we hypothesized that GRASLND could form a secondary structure to bind and activate PKR, 20 which in turn inhibits STAT1-related transcriptional function. This mechanism supports the 21 hypothesis that modulation of IFN-γ via the JAK/STAT pathway, achieved by the GRASLND/PKR RNA-protein complex, is important for cellular proliferation and differentiation 23 during chondrogenesis. available databases provide evidence corroborating similar patterns of IFN in degenerated cartilage. As GRASLND inhibits IFN, utilization of this lncRNA offers potential in both MSC 1 cartilage tissue engineering and in OA treatment. As a proof of concept, we showed that 2 GRASLND could enhance matrix deposition across cell types of origin, with and without 3 interferon challenge in vitro. It would be interesting to next investigate whether GRASLND can 4 protect cartilage from degradation in a milieu of pro-inflammatory cytokines in vivo.
Since lentivirus was employed to manipulate the expression of GRASLND, it is possible our  indistinguishable. This suggests that altered levels of interferon signaling can be attributed to 10 experimentally varied levels of GRASLND and not to the presence of lentivirus. Our data 11 indicate that GRASLND acts through type II rather than type I IFN. We found that 5 ng/mL of IFN-γ was still more detrimental to chondrogenic constructs compared to 100 ng/mL of IFN-β.

13
One potential explanation for this phenomenon may be the skewed distribution of available 14 surface receptors between type I and type II (IFNAR vs IFNGR). Indeed, MSCs express a much 15 lower level of IFNAR2 compared to IFNAR1, IFNGR1, or IFNGR2 (both in GSE109503 (36) and 16 in GSE129985 (this manuscript)). As these receptors function as heterodimers (32, 34), 17 response to type I may be stunted due to IFNAR2 deficiency.

18
Furthermore, we showed that a modified CRISPR-dCas9 system could successfully be used 19 for endogenous transcriptional activation of lncRNA. This system had been previously used in 20 other cell types to regulate expression of both protein-coding and non-coding genes (43,44,65,66). We showed that CRISPR may be more effective than transgene expression, as indicated by a larger increase in GAG production, despite lower levels of overall gene activation. As 23 GRASLND does not regulate RNF144A, it is evident that GRASLND acts in trans. However, we 24 speculate the CRISPR-dCas9 system could also be useful for gain of function studies to investigate lncRNAs acting in cis, as well as lncRNAs that are difficult to obtain via molecular 1 cloning due to their secondary structures, high repeated sequence or GC-rich content.

2
In conclusion, we have identified GRASLND as an important regulator of chondrogenesis.     Table S2.
Guide RNA sequences were designed using the UCSC genome browser

24
Harvested pellets were stored at -20°C until further processing. Collected samples were 25 digested in 125 µg/mL papain at 60°C overnight. DMMB assay was performed as previously described to measure GAG production (76). PicoGreen assay (ThermoFisher) was performed to 1 measure DNA content following manufacture's protocol.
Harvested pellets were fixed in 4% paraformaldehyde for 48 hours, and processed for 4 paraffin embedding. Samples were sectioned at 10 µm thickness, and subjected to either   Table   14 S3 (RNF144A-AS1). GAPDH probe set was pre-designed by the manufacturer. Staining was 15 carried out according to manufacturer's protocol for frozen tissues. Slides were mounted with Microscope VS120 (Olympus) at lower magnification and with the confocal microscope (Zeiss) 18 at higher magnification. processing. For pellets, harvested samples were snap frozen in liquid nitrogen and stored at -using a bead beater (BioSpec Products) at 2,500 oscillations per minute for 20 seconds for a 1 total of three times. Subsequent steps were performed following manufacturer's protocol.

2
Nuclear and cytoplasmic fractions from day 21 MSC pellets were separated with the NE-3 PER Nuclear and Cytoplasmic Extraction Reagents (ThermoFisher) following manufacturer's 4 protocol. Resulting extracts were immediately subjected to RNA isolation using the Norgen Total 5 RNA Isolation Plus Micro Kits (Norgen Biotek) by adding 2.5 parts of buffer RL to 1 part of 6 extract. Subsequent steps were carried out following manufacturer's protocol.

7
Reverse transcription by Superscript VILO cDNA master mix (Invitrogen) was performed 8 immediately following RNA isolation. cDNA was stored at -20°C until further processing. qRT-9 PCR was carried out using Fast SyBR Green master mix (Applied Biosystems) following manufacturer's protocol. A complete list of primer pairs (synthesized by Integrated DNA 11 Technologies, Inc.) is reported in Table S4.

13
MSCs were plated at 8.5 x 10 4 cells per well in 24-well plates (Corning). Lentivirus carrying 14 the response elements for type I (ISRE -#CLS-008L-1) or type II (GAS -#CLS-009L-1) upstream of firefly luciferase was purchased from Qiagen. Twenty-four hours post plating, cells 16 were co-transduced with virus in the following groups: ISRE with scrambled shRNA, ISRE with GRASLND shRNA, GAS with scrambled shRNA, GAS with GRASLND shRNA. Twenty-four 18 hours post-transduction, cells were rinsed once in PBS and fresh medium was exchanged.

19
Three days later, medium was switched to expansion medium with 100 ng/mL IFN-β 20 (PeproTech) for wells with ISRE or with 5 ng/mL IFN-γ (PeproTech) for wells with GAS. MSCs 21 were cultured for another 22 hours, and then harvested for luminescence assay using Bright-Glo Luciferase Assay System (Promega). Luminescence signals were measured using the
For knockdown experiments, lentivirus was titered by determining the number of antibiotic resistant colonies after puromycin treatment. For overexpression experiments, lentivirus was