Introduction

Each organism and each cell have to maintain homeostasis to ensure their biological integrity. Homeostasis is constantly challenged by endogenous and exogenous stress conditions, such as hypoxia, DNA damage, oxidative stress, heat shock, nutrient deprivation or viral infections. To overcome these harmful conditions and to maintain homeostasis and ensure healthy aging, every cell must activate a range of conserved stress response pathways to prevent pathophysiological consequences (Kourtis and Tavernarakis 2011). While the response to cellular stress can be either specific for or shared between different stress and cell types, all stress response pathways have in common that they enable the cell to integrate different inputs of cellular stress, adapt to them and finally overcome them by either restoring homeostasis or inducing cell death (Fulda et al. 2010). Several key stress proteins and pathways have been identified that can integrate various stress signals and induce a multifaceted stress response. Among those regulators are for example the tumor suppressor protein p53, the mechanistic target of rapamycin (mTOR) or the eukaryotic initiation factor 2 (eIF2) protein kinases of the integrated stress response. p53 is activated upon various types of stress, such as DNA damage, hypoxia or oncogene activation, and its activation leads to the induction of the cyclin-dependent kinase inhibitor 1A (CDKN1A encoding p21) and other key response genes that regulate cell cycle arrest, apoptosis and senescence (Levine et al. 2006). mTOR is critical for cell growth and proliferation under nutrient- and growth factor-rich conditions and is turned off upon metabolic stress, nutrient starvation, hypoxia and DNA damage (Aramburu et al. 2014). The integrated stress response, on the other hand, is induced by a broad range of cellular stress conditions, including amino acid deprivation, viral infections and endoplasmic reticulum (ER) stress. It results in phosphorylation of eIF2α and consequently in global reduction of protein translation and preferential translation of stress-related proteins, which initiates an adaptive program to overcome the stress (Pakos-Zebrucka et al. 2016). In addition to these molecularly defined stress response pathways, the cell can also employ a less defined response that relies on the formation of dynamic cellular condensates that emerge from multivalent RNA and protein interactions. Such ribonucleoprotein condensates, also referred to as stress granules, have been shown to play an integral role in response to several different types of stress and in the control of normal physiology and disease (Jeon et al. 2022; Lee and Namkoong 2022). Independent of the underlying molecular mechanism of the different cellular stress response pathways, they all need to be tightly regulated with the ability to quickly adapt to stress, to integrate multiple stress signals and to fine-tune the response to certain cellular stress conditions in order to ensure an adequate output of the stress response.

The human genome is extensively transcribed into several types of non-protein-coding RNA species, some of which have been shown to act as potent regulators and mediators of the cellular stress response in animals as well as in plants (Jha et al. 2020; Mendell and Olson 2012). A relatively novel class among these noncoding RNA species represent the long noncoding RNAs (lncRNAs), which are defined by a sequence length of more than 500 nucleotides (Mattick et al. 2023). With regard to protein-coding genes, lncRNAs can be produced as intergenic, antisense, intronic or overlapping transcripts. They lack a detectable open reading frame, are mostly transcribed by RNA polymerase II, often spliced and polyadenylated, and may also occur in a circular form. Therefore, lncRNAs comprise a heterogeneous class of noncoding RNAs that have diverse physiological and molecular functions (Ulitsky and Bartel 2013). While tens of thousands of lncRNA genes have been annotated (Volders et al. 2015), the biological and molecular characterization of the vast majority of them is still missing. This is, at least in part, due to the fact that each lncRNA candidate gene has to be studied on a case-by-case basis to infer a meaningful biological and molecular function for the encoded transcript. Hence, there is only a very limited number of lncRNAs that have well-established biological roles with solid experimental validation of their molecular modes of action (Rinn and Chang 2020). In principle, lncRNA functions can be categorized into two different modes of action: (I) regulation of gene expression and transcription of neighboring genes in cis and (II) regulation of nuclear and cytoplasmic processes in trans. Such functions in trans include for example the modulation of gene expression at distant genomic loci, the formation of nuclear and cytoplasmic structures and compartments, or the regulation of interacting protein and/or RNA molecules. Of note, some trans-acting lncRNA genes have been also shown to encode small peptides that can exert important biological functions (Makarewich and Olson 2017; Patop et al. 2019; Wright et al. 2022). Depending on the proposed molecular mode of action, a lncRNA has to fulfill certain important criteria. For instance, if a lncRNA is proposed to function in trans via the interaction with other biomolecules, such as proteins or other RNAs like miRNAs, the respective lncRNA often needs to be expressed at sufficient abundance with ideally conserved sequences or secondary structures that confer specificity for its interaction with other biomolecules (Kopp and Mendell 2018). On the other hand, if lncRNAs are proposed to function in cis to regulate transcription and gene expression of neighboring genes, it needs to be demonstrated whether the accumulation of the actual encoded lncRNA sequence is necessary for the observed biological effect or whether the sole act of transcription or even DNA elements within the lncRNA locus are sufficient to control the neighboring gene locus independent of the encoded lncRNA (Anderson et al. 2016; Engreitz et al. 2016).

Increasing evidence suggests that lncRNAs can function as versatile regulators of the mammalian stress response via diverse molecular mechanisms that facilitate a time- and dose-sensitive response to various cellular stress conditions. In this review, we describe novel insights into lncRNA function in the mammalian stress response and homeostasis. We highlight and elaborate on the different molecular modes of action of lncRNAs that have been experimentally validated to regulate, mediate or titrate the response to a range of cellular stress conditions. We aim to generate a framework for the functional classification of stress-related lncRNAs based on their molecular modes of action, which we divide into three general classes: (I) lncRNAs regulating gene expression in cis or trans, (II) lncRNAs acting as scaffolds or tethers in ribonucleoprotein complexes, and (III) lncRNAs regulating and forming cellular condensates.

lncRNA-mediated regulation of gene expression upon cellular stress

Gene regulatory functions in the nucleus were among the first described roles for lncRNAs (Lee 2012). It has been suggested as a common mode of action that lncRNAs bind to certain RNA binding proteins or components of the polycomb repressor complex and guide these to either local (regulation in cis) or distant (regulation in trans) genomic sites, where they can modulate the chromatin state and repress or activate gene expression of the respective gene locus. Technological advances, including genome editing and rigorous biochemical testing of specific RNA-protein interactions, have led to the awareness that this initial model of how lncRNAs generally regulate gene expression in the nucleus may need to be revised (Kopp and Mendell 2018). The oftentimes low copy numbers of lncRNAs together with the usually promiscuous binding of RNA binding proteins to lncRNAs have challenged the proposed gene regulatory functions of some lncRNAs acting in trans, whereas careful studies of lncRNA loci using precise genetic models to distinguish between the role of transcription, splicing and the actual lncRNA transcript in lncRNA-mediated regulation of local gene expression have questioned the proposed modes of action for some lncRNAs acting in cis. Although these originally predicted gene regulatory functions of lncRNAs in cis and trans may not be as common as initially anticipated, they still represent an important mode of action that is also utilized and critical for lncRNA-mediated stress response.

Several noncoding RNAs, including lncRNAs, are regulated by the tumor suppressor protein p53 (Chaudhary and Lal 2017). The p53 pathway is typically activated upon DNA damage, which results in pleiotropic cellular effects, ranging from cell cycle arrest to pro-survival and pro-apoptotic signals. One such lncRNA represents the p53-induced noncoding RNA (PINCR), a ~ 2.2 kilobase (kb) long intergenic lncRNA that is predominantly localized to the nucleus, expressed from the X-chromosome and highly induced upon doxorubicin treatment in a p53-dependent manner (Chaudhary et al. 2017). PINCR has been shown to be critical for G1 cell cycle arrest and anti-apoptotic effects upon DNA damage in colorectal cancer cells, and PINCR loss-of-function results in increased sensitivity to doxorubicin and 5 fluorouracil (5-FU) treatment. On a molecular level, PINCR interacts with and recruits the nuclear protein matrin-3 through a yet unknown mechanism to a subset of p53 target genes. Matrin-3 then binds to p53 in a DNA- and RNA-independent fashion and forms regulatory chromatin loops with surrounding enhancer elements to regulate a subset of p53 target genes (Fig. 1a). PINCR therefore represents a stress-induced lncRNA that can modulate the p53-mediated response to DNA damage in trans via the regulation of selected p53 target genes that are important for G1 cell cycle arrest and cell survival. In a screen for RNAs transcribed from cell cycle-related gene promoters, several new putative lncRNAs have been identified, five of which are induced by DNA damage and arise from the p53-regulated CDKN1A (p21) promoter region (Hung et al. 2011). Among those lncRNAs is a ~ 1.5 kb long noncoding transcript termed p21 associated noncoding RNA DNA damage activated (PANDA) or promoter of CDKN1A antisense DNA damage activated RNA (PANDAR) that is located ~ 5 kb upstream of the CDKN1A transcription start site and highly induced upon DNA damage in a p53-dependent manner. Silencing of PANDAR does not affect the expression of its neighboring gene CDKN1A and results in increased sensitivity of human fibroblasts to doxorubicin treatment. PANDAR binds to and sequesters the transcription factor nuclear transcription factor Y subunit alpha (NF-YA), thereby limiting the expression of pro-apoptotic NF-YA target genes (Fig. 1b). Although the specificity of the interaction between PANDAR and NF-YA is not fully clear, it has been shown that siRNA-mediated knockdown of NF-YA is able to rescue and reduce apoptosis in PANDAR-depleted cells, demonstrating a critical role of NF-YA downstream of PANDAR. Taken together, these findings give an example of a DNA damage-induced lncRNA from the CDKN1A locus that acts in trans as a molecular decoy for the transcription factor NF-YA. PANDAR fine-tunes the DNA damage response of p53 via the regulation of pro-apoptotic NF-YA target genes, whereas other p53 effectors, such as p21, may rather regulate other aspects like cell cycle arrest. Hence, PINCR as well as PANDAR are two examples of lncRNAs that are integrated into the p53-mediated DNA damage response and that regulate the targeted expression of selected p53 response genes in trans through two distinct mechanisms.

Fig. 1
figure 1

lncRNA-mediated regulation of gene expression upon cellular stress. a PINCR-mediated recruitment of matrin-3 to selected p53 target genes upon DNA damage and subsequent regulation of their transcription in trans. b Induction of PANDAR expression from the CDKN1A locus upon DNA damage and regulation of NF-YA and its target genes in trans. c Local gene regulation of CDKN1A by either nascent transcription through conserved DNA/RNA elements or activating DNA enhancer elements in the lincRNA-p21 locus. d Promoter competition between the MYC and PVT1 locus (1) and RNA- or transcription-dependent regulation of MYC by PVT1b (2)

Full size image

Another lncRNA that is transcribed upstream of the Cdkn1a locus and induced by DNA damage in a p53-dependent manner is the noncoding RNA lincRNA-p21. Several different modes of action have been proposed for lincRNA-p21 since its discovery in 2010 (Huarte et al. 2010), ranging from gene regulatory functions in trans (Huarte et al. 2010) to post-transcriptional regulation of selected mRNA targets (Yoon et al. 2012) and gene regulation of its neighboring gene Cdkn1a (p21) in cis (Dimitrova et al. 2014; Groff et al. 2016; Winkler et al. 2022). Initially, lincRNA-p21 was reported to globally repress genes in the p53 pathway necessary for proper apoptosis induction upon DNA damage (Huarte et al. 2010). It was proposed that, upon p53-mediated induction, lincRNA-p21 binds to and recruits the RNA binding protein heterogeneous nuclear ribonucleoprotein K (hnRNPK) to selected p53 target genes, which are subsequently repressed. In a later study (Dimitrova et al. 2014), however, Dimitrova et al. showed that lincRNA-p21 rather activates the transcription of Cdkn1a (p21) in cis, presumably via interaction with hnRNPK, which then leads to a global repression of polycomb repressor complex 2 (PRC2) target genes, affecting the G1/S cell cycle checkpoint and cell proliferation. Importantly, this was the first study demonstrating a cis regulatory function of lincRNA-p21 using an allele-specific mouse model of lincRNA-p21 and Cdkn1a deficiency. Later, it was found that lincRNA-p21 deletion in mice results in global changes of local gene expression, including repression of the Cdkn1a gene, which is independent of the respective lincRNA-p21 expression in the tested tissues (Groff et al. 2016). These effects were attributed to activating enhancer elements within the lincRNA-p21 locus, which can form chromosomal loops with neighboring genes and induce their transcription, suggesting a role for lincRNA-p21 as an enhancer RNA (eRNA) whose expression is not required for the gene regulatory function of this locus. In a recent study (Winkler et al. 2022), the model of lincRNA-p21’s mode of action was further refined. Through a series of elegant genetic models that each independently address the role of transcription termination, transcript degradation, splicing inhibition, deletion of conserved regions, deletion of the lincRNA-p21-resident p53 response element, and transcriptional interference, it could be shown that lincRNA-p21 regulates Cdkn1a expression in cis through transcription of conserved DNA or RNA elements in the nascent lincRNA-p21 transcript, while transcription and splicing as well as the accumulation of full-length lincRNA-p21 are not required. Taken together, lincRNA-p21 can fine-tune the response to cellular stress in the p53 pathway via the regulation of the neighboring gene Cdkn1a. Whether the nascent transcript recruits factors, such as hnRNPK, to the Cdkn1a locus or whether rather the transcription of the encoded conserved DNA elements confer the cis regulatory effect remains an open question. In some cell and tissue types, the sole existence of regulatory DNA enhancer elements in the lincRNA-p21 locus may already suffice to regulate Cdkn1a in cis without the necessity of transcriptional activity (Fig. 1c).

Plasmacytoma variant 1 (Pvt1) represents a lncRNA transcribed approximately 50 kb downstream of the myeolocytomatosis (Myc) oncogene that has been initially identified as a hotspot for chromosomal translocations in murine and later also in human lymphomas (Cory et al. 1985; Graham and Adams 1986; Graham et al. 1985). Since the human genomic locus on chromosome 8 (8q24), including the PVT1 gene, has been shown to be amplified in human breast cancer (Curtis et al. 2012) and PVT1 itself has been reported to stabilize the MYC protein (Tseng et al. 2014), PVT1 has been described as a potential oncogene. However, this notion has become challenged by the finding that the PVT1 promoter competes with the MYC promoter for enhancer elements within the PVT1 locus, a function that appears to be independent of the PVT1 transcript (Cho et al. 2018). This negative regulation of MYC expression together with the fact that mutations in the PVT1 promoter frequently occur in human cancers suggest a rather tumor suppressive role for the PVT1 locus and promoter in human cancers. Interestingly, there is also a p53 response element within the PVT1 locus that results in p53-mediated repression of MYC upon genotoxic stress. MYC repression is thereby important for the general reduction in transcription and the proper induction of cell cycle arrest and apoptosis (Porter et al. 2017). Recently, it has been shown that oncogenic as well as genotoxic stress can specifically induce the expression of a certain p53-dependent Pvt1 isoform, called Pvt1b, in mouse embryonic fibroblasts (Olivero et al. 2020). Pvt1b negatively regulates the expression of the neighboring Myc oncogene and inhibits tumor growth, but not progression, in a murine lung tumor model. Interestingly, the Pvt1b RNA appears to be critical for stress-induced repression of Myc. Enforced production of the Pvt1b transcript, even in the absence of cellular stress and functional p53, is sufficient to inhibit Myc expression in cis. Based on the existing literature, and besides reported oncogenic functions, the PVT1 locus appears to act as a tumor suppressor through at least two distinct molecular mechanisms: one is p53- and RNA-independent via promoter competition between PVT1 and MYC (Cho et al. 2018), and another one is activated upon cellular stress, which depends on p53 as well as on the RNA (specifically, the Pvt1b isoform) (Olivero et al. 2020) (Fig. 1d). The complex regulation of MYC by the PVT1 locus illustrates that a lncRNA gene can modulate cellular programs in different cellular conditions, such as tumorigenesis and response to genotoxic stress, through various modes of action depending on the cellular context.

lncRNAs as stress-induced scaffolds or tethers

In addition to directly regulating transcription and gene expression in cis or trans, lncRNAs can form ribonucleoprotein complexes in order to control the composition of higher order cellular complexes as well as the activity or stability of the interacting protein and/or RNA molecules. This role of lncRNAs, which can be described as a scaffolding function, has been shown to be also utilized by noncoding RNAs that are involved in the mammalian stress response.

In a screen for transcripts that originate from promoters of human cell cycle genes (Hung et al. 2011), a novel ~ 1 kb long lncRNA has been identified as a divergent transcript originating from the CDKN1A promoter. This lncRNA has been shown to be highly induced by DNA damage in a p53-dependent manner and has been named damage induced noncoding or DINO (Schmitt et al. 2016). DINO induced by DNA damage or by ectopic overexpression binds to the C-terminal region of p53 via a distinct p53 binding site, resulting in the stabilization of the p53 protein. It is not yet fully understood how DINO binding stabilizes p53 protein and whether there are other proteins involved in the DINO-p53 ribonucleoprotein complex. However, it has been demonstrated that this positive feed forward loop and p53 stabilization are required for robust p53 target gene expression and proper induction of cell cycle arrest and apoptosis upon sustained DNA damage. Dino knockout mice display similarly reduced p53 levels as p53 heterozygous knockout mice and a comparable decreased survival upon lethal doses of irradiation, highlighting the importance of appropriate p53 protein levels under steady state and stress conditions. Interestingly, later studies (Marney et al. 2021, 2022) revealed that the Dino/Cdkn1a locus represents a tumor suppressor locus as loss of Dino can promote lymphoma in an Eµ-Myc mouse model as well as induce the formation of a subset of spontaneous tumors, including sarcomas and lymphomas. In line with Dino’s role in the p53 pathway, Dino requires intact p53 for its tumor suppressor function, while loss or inactivation of p53 abrogates its tumor suppressive role in lymphoma (Marney et al. 2021). Consistently, DINO promoter hypermethylation can be often found in human cancers in a mutually exclusive manner with p53 mutations, potentially explaining the escape from tumor suppression in human cancers with intact p53. These findings suggest that DINO forms a positive feed forward loop in the p53 pathway to titrate the response to DNA damage via the direct interaction with and stabilization of p53, leading to an amplification of the output of the DNA damage response once it exceeds a certain threshold (Fig. 2a). Another lncRNA that is induced upon DNA damage in a p53-dependent manner and that regulates p53 stability is the p53 upregulated regulator of p53 levels or PURPL (Li et al. 2017). In contrast to the afore-discussed lncRNA DINO that stabilizes p53, PURPL has been shown to suppress p53 protein stability and basal expression levels. PURPL associates with the p53 stabilizing protein MYBBP1A (MYB binding protein 1a) via the RNA binding protein HuR, preventing the formation of the MYBBP1A-p53 complex and hence the stabilization of p53. Accordingly, PURPL loss-of-function results in elevated basal p53 protein levels in colorectal cancer cells and consequently in reduced cell and tumor growth as well as in increased sensitivity to DNA damage. Of note, silencing of MYBBP1A in PURPL knockout cells partially restores basal p53 protein levels and normal cell proliferation, confirming an important role of MYBBP1A downstream of PURPL (Fig. 2b). These results suggest that PURPL is part of an autoregulatory, negative feedback loop that keeps basal p53 levels in check and facilitates proper cell proliferation in colorectal cancer cells. Overall, the modes of action of these two lncRNAs, DINO and PURPL, are intriguing examples of how lncRNAs can directly and timely influence the pathway they are integrated in, which can have important implications in human health and disease, including tumorigenesis.

Fig. 2
figure 2

lncRNAs as stress-induced scaffolds or tethers (I). a DINO-mediated stabilization of the p53 protein upon DNA damage. b Maintenance of basal p53 levels through p53 destabilization by the lncRNA PURPL. c SPARCLE-catalyzed cleavage of PARP1 by caspase 3 upon DNA damage. d Dual mode of action of the lncRNA GUARDIN upon DNA damage, (1) as a competing endogenous RNA for miR-23a, and (2) as a scaffold in a ribonucleoprotein complex stabilizing BRCA1

Full size image

Besides the direct regulation of p53, lncRNAs can also act as downstream effectors of p53 in the DNA damage response. The lncRNA SPARCLE is a ~ 770 nt long, nuclear RNA that is located upstream of the microRNA-34b/c (miR-34b/c) cluster on chromosome 11 (Meza-Sosa et al. 2022). Upon DNA damage, SPARCLE is induced in a p53-dependent manner together with miR-34b/c using the same p53 response element just upstream of the SPARCLE transcription start site. Interestingly, SPARCLE lacks polyadenylation and may be processed from the miR-34b/c precursor. Its expression level, even after induction of DNA damage, is low at around 10 copies per cell on average. Nevertheless, loss of SPARCLE enhances DNA repair and inhibits DNA damage-induced apoptosis, which is comparable to p53 loss-of-function and can be rescued by SPARCLE overexpression. While SPARCLE does not appear to regulate transcription, it binds with high affinity to poly(ADP-ribose) polymerase 1 (PARP1) and serves as a cofactor for caspase 3. In the presence of SPARCLE, caspase 3 efficiently cleaves PARP1, which results in impaired DNA repair and increased apoptosis. Of note, SPARCLE can potently enhance caspase 3-mediated cleavage of PARP1 at very low molar ratios compared to PARP1, corroborating the proposed molecular mechanism even at low copy numbers. In line with these findings, expression of cleaved N-terminal PARP1 in SPARCLE-deficient cells results in reduced DNA repair and enhanced apoptosis upon DNA damage, suggesting that SPARCLE mainly acts by promoting caspase 3-mediated cleavage of PARP1. As SPARCLE is not induced before day one after DNA damage, it may be responsible to ensure cell death of cells with extensive DNA damage at later stages of the DNA damage response. SPARCLE thereby acts as a scaffold that facilitates PARP1 cleavage by caspase 3, regulating DNA repair and apoptosis (Fig. 2c). Through its high affinity to PARP1 in the nanomolar range and its presumably reversible binding to full length as well as cleaved PARP1, SPARCLE may function as a catalytic lncRNA that can perform its molecular mode of action even at the observed low copy numbers.

A novel lncRNA termed GUARDIN has been recently characterized as another downstream effector of p53 and described to be induced upon genotoxic stress and oncogene activation in a p53-dependent manner (Hu et al. 2018). GUARDIN is expressed as a divergent transcript from the promoter of the miR-34 host gene (MIR34HG), sharing the same promoter and p53 binding region. Importantly, interfering with GUARDIN levels does neither affect the expression of the neighboring and partly overlapping MIR34HG nor the encoded miR-34, ruling out a gene regulatory function in cis. GUARDIN displays nuclear as well as cytoplasmic subcellular localization and shows higher expression levels in human colon tumors with wild-type than mutant p53. Inhibition of GUARDIN in different human cancer cells impairs cancer cell proliferation and survival as well as tumor growth in vivo. Furthermore, GUARDIN inhibition promotes chromosome end-to-end fusion and DNA damage, resulting in increased apoptosis and senescence as well as augmented sensitivity to genotoxic stress (Hu et al. 2018). On a molecular level, a dual mode of action was proposed for GUARDIN function (Fig. 2d): (I) GUARDIN can act as a competing endogenous RNA (ceRNA) regulating miR-23a and its downstream mRNA target telomeric repeat-binding factor 2 (TRF2), a critical component of the shelterin protein complex that is necessary for the protection of chromosome ends; (II) GUARDIN interacts with the proteins BRCA1 (breast cancer gene 1) and BARD1 (BRCA1 associated RING domain 1) via distinct binding regions, forming a ternary complex that stabilizes BRCA1, a key protein in the repair of DNA double strand breaks, by preventing its ubiquitination and proteasome-mediated degradation. Of note, the stoichiometry in terms of copy numbers and binding sites is plausible for the proposed modes of action as a ceRNA for miR-23a as well as a scaffold for the formation of the BRCA1-BARD1 complex. Additional evidence for a dual mode of action is provided by the fact that both TRF2 and BRCA1 overexpression is required to restore the cytoprotective effect of GUARDIN in GUARDIN-depleted cells. Taken together, this lncRNA is an example of a noncoding RNA that can exert pleiotropic functions in a ribonucleoprotein complex to maintain DNA integrity not only after genotoxic stress, but also at steady state under homeostatic conditions to preserve chromosomal stability.

Another example of a lncRNA that has been associated with cellular stress response and that has been ascribed more than one molecular mode of action is the lncRNA growth arrest specific 5 (GAS5). GAS5 was initially identified in a cDNA library of RNAs enriched in growth-arrested cells (Schneider et al. 1988). It is a highly abundant, spliced and polyadenylated lncRNA that hosts several small nucleolar RNAs (snoRNAs) in its introns (Coccia et al. 1992; Smith and Steitz 1998). Nutrient starvation often leads to growth arrest and eventually to cell death. Cell growth and survival depend on nutrients and can be modulated by glucocorticoid receptors upon starvation. Upon serum withdrawal, GAS5 is strongly induced and binds to the DNA binding domain of the glucocorticoid receptor via a conserved glucocorticoid response mimic, modulating glucocorticoid receptor-regulated genes as well as cell survival and metabolic activity (Hudson et al. 2014; Kino et al. 2010). A structural analysis of the GAS5 RNA using selective 2’ hydroxyl acylation analyzed by primer extension by mutational probing (SHAPE-MaP) revealed a modular structure of this noncoding RNA that consists of three modules, indicating that this lncRNA may serve pleiotropic functions (Frank et al. 2020): (I) a 5’ module that inhibits cell growth independent of steroid receptors, (II) a steroid receptor module that inhibits steroid-dependent cell growth, and (III) a structured core module that regulates mTOR-mediated inhibition of cell growth. In a recent study (Sang et al. 2021), an additional mode of action for GAS5 was identified. In a screen for subcellular localization and organelle-enrichment of lncRNAs, it was shown that GAS5 is localized to the mitochondrial fraction and that this mitochondrial localization is even more enriched upon glucose starvation. GAS5 interacts with MDH2 and negatively regulates the formation of the FH-MDH2-CS (fumarate hydratase-malate dehydrogenase 2-citrate synthase) complex, which catalyzes the reaction of fumarate to citrate in the tricarboxylic acid (TCA) cycle, via sirtuin 3 (SIRT3)-mediated deacetylation of MDH2. Nutrient deprivation leads to GAS5-mediated reduction of the TCA flux and hence to growth arrest. These findings demonstrate once more that a lncRNA can have more than one mode of action and that these different molecular functions can be defined by the modular architecture of the lncRNA. In case of GAS5, the lncRNA can function in different cellular compartments either as a molecular decoy for glucocorticoid receptors or as a scaffold that modulates the activity of an enzyme complex of the TCA cycle, depending on the respective cellular demands and stress condition (Fig. 3a).

Fig. 3
figure 3

lncRNAs as stress-induced scaffolds or tethers (II). a Induction of GAS5 upon glucose and serum starvation and subsequent regulation of mitochondrial and nuclear functions. (FH = fumarate hydratase, MDH2 = malate dehydrogenase 2, CS = citrate synthase, Ac = acetylation, TCA = tricarboxylic acid, GR = glucocorticoid receptor, GRE = glucocorticoid response elements). b Regulation of the NF-κB pathway by the lncRNA NKILA in different cellular stress conditions. (AICD = activation-induced cell death). c Activation of mTOR and cholesterol biosynthesis by the cholesterol-induced lncRNA SNHG6

Full size image

The immune-responsive NF-κB interacting lncRNA NKILA is another example of a lncRNA that can exert various physiological functions depending on the cellular and stress-related context. NKILA has been initially identified as a cytoplasmic lncRNA that is induced by inflammatory cytokines via the NF-κB pathway in human breast cancer cells (Liu et al. 2015). It binds to the transcription factor NF-κB and forms a stable ternary complex together with IκB, which prevents IκB phosphorylation and NF-κB activation. While decreased NKILA expression levels in breast tumors are associated with metastasis and poor prognosis of breast cancer patients (Liu et al. 2015), NKILA has been more recently shown to be also critical for immune evasion of breast and lung cancer cells (Huang et al. 2018). Tumor-specific cytotoxic T lymphocytes can detect and eliminate tumor cells at early stages. This tumor immunosurveillance, however, can be impaired by activation-induced cell death (AICD) of T lymphocytes, a process that removes activated T lymphocytes and contributes to tumor immune evasion. NKILA is induced in activated T cells by STAT1 (signal transducer and activator of transcription 1) in a calcium/calmodulin-dependent fashion and promotes AICD in tumor-specific cytotoxic T lymphocytes and T helper cells type 1 (TH1) via the inhibition of the NF-κB pathway. Interestingly, tumor-specific cytotoxic T cells and TH1 isolated from breast and lung cancer patients display a high expression of NKILA and a high sensitivity to AICD. Conversely, silencing NKILA in the context of an adoptive cell therapy in a patient-derived xenograft model of breast cancer can reduce AICD in T lymphocytes and improve therapy efficacy by preventing tumor immune evasion (Huang et al. 2018). Moreover, NKILA has been shown to be downregulated upon ischemia–reperfusion injury in cardiomyocytes, which results in the activation of the NF-κB pathway and a loss of myocardin, leading to increased apoptosis and inflammatory responses (Liu et al. 2020). Overexpression of NKILA, on the other hand, can improve myocardial ischemic injury by inhibiting NF-κB signaling and restoring myocardin levels. These findings demonstrate how a lncRNA can regulate different physiological outcomes depending on the respective cell and stress type via a single molecular mode of action, which is to function as a scaffold that binds NF-κB and IκB, thereby preventing IκB phosphorylation and NF-κB activation (Fig. 3b).

Another lncRNA that has been recently reported to function as a molecular scaffold is the small nucleolar RNA host gene 6 (SNHG6) (Liu et al. 2022). SNHG6 is significantly upregulated in hepatoma compared to normal liver tissue and has been shown to be a cholesterol effector that accelerates progression from non-alcoholic fatty liver disease (NAFLD) to hepatocellular carcinoma (HCC). Upon stimulation with cholesterol, SNHG6 expression is induced and associated with the endoplasmic reticulum (ER) and lysosomal compartments. SNHG6 forms a complex with the ER-associated protein Fas-associated factor family member 2 (FAF2) and mTOR at the ER-lysosome contact sites. This formation is regulated by cholesterol levels and requires SNHG6 for mTOR recruitment to the lysosomes and FAF2-mediated activation of mTOR signaling. Through targeted sequence deletion studies, two regions within SNHG6, loop 1 and loop 3, have been identified that specifically interact with FAF2 and mTOR, respectively. These interactions enhance FAF2-mTOR binding and the activation of mTOR signaling at the ER-lysosome contact sites and subsequently cholesterol biosynthesis (Fig. 3c). Of note, SNHG6 inactivation blocks mTOR signaling and inhibits tumor growth in a patient-derived hepatoma xenograft model. Hence, SNHG6 acts as a tether that is important for lysosomal recruitment and activation of mTOR by enhancing the ER-lysosome contacts. These findings suggest a new role for lncRNAs in organelle communication that facilitates and modulates organelle-specific crosstalk and signaling for example in the sensing of nutrients or other biomolecules, such as cholesterol, to maintain cellular and organismal homeostasis.

lncRNAs in stress-related liquid–liquid phase separation

The biophysical process of liquid–liquid phase separation (LLPS) is nucleated by multivalent interactions of RNAs and RNA binding proteins that often contain modular domains or intrinsically disordered regions (IDRs). LLPS leads to the formation of membrane-less compartments comprised of phase-separated RNA-protein condensates throughout the cell that have been described in many cell and tissue types and implicated in various biological functions (Banani et al. 2017). Phase-separated granules can for example control the specificity and the kinetics of biochemical reactions, sequester biomolecules to limit their available concentrations, or dynamically modulate complex biological processes through the active regulation of phase-separated condensates. lncRNAs have been increasingly shown to participate in LLPS and in stress-related responses involving phase separation and phase-separated compartments (Onoguchi-Mizutani and Akimitsu 2022).

Different types of cellular stress, including ER stress, oxidative stress, heat shock and hyperosmotic stress, result in the formation of cytoplasmic ribonucleoprotein granules, called stress granules. While the protein composition, including proteins like the G3BP stress granule assembly factor, TIAR, TIA1, or the eukaryotic translation initiation factor 4E (eIF4E), has been well-documented, relatively little has been known about the RNA composition of these phase-separated stress condensates (Campos-Melo et al. 2021). Studies on the stress granule transcriptome have revealed that mainly transcripts with poor translation and relatively long sequence length, including mRNAs as well as noncoding RNAs, are recruited and enriched in these ribonucleoprotein condensates (Khong et al. 2017; Namkoong et al. 2018). While stress granule formation does not per se appear to be a stress-specific response, it has been shown that certain stress types such as ER stress, heat shock or arsenite treatment can recruit similar, yet distinct subsets of RNAs to stress granules dependent on their translational activity, their length and the presence of AU-rich elements. However, how specific RNAs or mRNAs may be recruited to stress granules and translationally inhibited in response to a certain stress has remained an open question. In a recent study (Wang et al. 2021), a novel stress-related lncRNA was shown to recruit a specific mRNA within a ribonucleoprotein complex to stress granules. This less than one kb long noncoding RNA was identified in a screen for glutamine starvation-responsive lncRNAs and was hence termed glutamine insufficiency regulator of glutaminase lncRNA (GIRGL). Glutamine is an energy source as well as a source of nitrogen for the synthesis of many biomolecules. As a first step of glutamine utilization, glutamine is converted into glutamate by an enzyme called glutaminase, which is encoded by the genes GLS1 and GLS2. Glutamine deprivation results in downregulation of glutaminase (GLS1), which has been otherwise shown to be overexpressed in human cancers and to promote tumor growth. Glutamine starvation leads to GIRGL upregulation and inhibition of glutamine metabolism via translational repression of GLS1. GIRGL interacts with the stress granule-associated RNA binding protein CAPRIN1 (cell cycle associated protein 1) to promote the formation of a GIRGL-CAPRIN1-GLS1 ribonucleoprotein complex that is targeted to stress granules, preventing GLS1 translation (Fig. 4a). Under normal glutamine supply, GIRGL suppresses tumor cell growth in colon cancer cells, whereas under glutamine starvation GIRGL facilitates cell survival under prolonged glutamine deprivation, which may help tumors to adapt to a glutamine-low microenvironment. This suggests a role for a lncRNA in the recruitment of specific stress-related mRNAs to stress granules, where they are translationally repressed.

Fig. 4
figure 4

lncRNAs in stress-related liquid–liquid phase separation. a Induction of the lncRNA GIRGL upon glutamine deprivation and subsequent recruitment of GLS1 mRNA to stress granules in a lncRNA containing ribonucleoprotein complex. b NORAD binding to PUMILIO proteins (PUM) and the formation of NORAD-PUMILIO (NP) bodies as well as potential, yet to be shown, crosstalk with other condensates, such as stress granules. c Increased NEAT1 expression and paraspeckle formation upon various types of stress and the role of NEAT1-dependent TDP-43 nuclear bodies. (ALS = amyotrophic lateral sclerosis). d Induction of MALAT1 upon various types of stress and the role of MALAT1 in the heat-induced noncoding RNA containing nuclear bodies (HiNoCo bodies)

Full size image

Another noncoding RNA that is associated with stress granules is the noncoding RNA activated by DNA damage (NORAD) (Khong et al. 2017; Namkoong et al. 2018). NORAD and its interacting RNA binding proteins PUMILIO1 (PUM1) and PUM2 are strongly enriched in stress granules upon arsenite treatment and ER stress, although neither NORAD nor PUMILIO are required for each other’s recruitment to stress granules under ER stress. This suggests that NORAD is targeted to stress granules via other RNA binding proteins, such as TIAR or TIA1, presumably through AU-rich elements within the NORAD sequence, which is in line with the general notion that RNA targeting to stress granules requires multiple RNA-protein, protein-protein, and RNA-RNA interactions (Matheny et al. 2021). Interestingly, NORAD has been demonstrated to nucleate another phase-separated cytoplasmic condensate that is distinct from stress granules and that has been termed NORAD-PUMILIO (NP) body, which is present at steady state, induced upon DNA damage and critical for maintaining genome stability (Elguindy and Mendell 2021). NORAD has been initially described as a 5 kb long, predominantly cytoplasmic noncoding RNA that is induced by DNA damage in a p53-dependent manner and that regulates the activity of PUMILIO RNA binding proteins (PUM1 and PUM2) (Lee et al. 2016; Tichon et al. 2016). PUMILIO proteins bind to a very specific sequence, also referred to as PUMILIO response elements or PREs, in predominantly the 3’ untranslated regions (UTRs) of mRNAs, negatively affecting their stability and translation (Goldstrohm et al. 2018). NORAD harbors at least 15 PREs and represents the preferred RNA binding partner of PUM2 in human cells. Loss of NORAD results in PUMILIO hyperactivity and the repression of PUMILIO target mRNAs, which leads to chromosomal instability and mitochondrial dysfunction in human and mouse cells as well as to degenerative phenotypes in mice, resembling premature aging (Elguindy et al. 2019; Kopp et al. 2019; Lee et al. 2016; Tichon et al. 2016). Although NORAD is abundantly expressed and represents a multivalent binding platform with multiple PREs per molecule, it has not been understood how NORAD can compete with the transcriptome-wide pool of PRE-containing RNA targets through a simple titration model of PUMILIO proteins. Elguindy and Mendell recently showed that NORAD facilitates LLPS which enables the sequestration of a super-stoichiometric number of PUMILIO proteins into NP bodies, thereby out-competing the transcriptome-wide pool of PRE-containing RNA targets (Elguindy and Mendell 2021). They further demonstrated that this phase separation event is critical for NORAD’s molecular function, which is to regulate PUMILIO activity and maintain chromosomal stability (Fig. 4b). Interestingly, it has been shown that these NORAD condensates appear to increase in size when cells undergo DNA damaging stress (Elguindy et al. 2019; Elguindy and Mendell 2021), suggesting that NP bodies may be actively regulated upon different types of cellular stress in order to fine-tune PUMILIO activity depending on the cellular demands. Future investigations will be required to fully characterize and understand the role of the NORAD-PUMILIO axis and the formation of NP bodies and potential cross-talk with other condensates in the cellular stress response.

While stress granules and NP bodies are two examples of phase-separated condensates in the cytoplasm, there are also several ribonucleoprotein granules in the nucleus, such as nuclear stress bodies, nuclear speckles and paraspeckles, which have been also implicated in cellular stress response (Onoguchi-Mizutani and Akimitsu 2022). A well-characterized example of these nuclear condensates are the NEAT1-induced paraspeckles, which are thought to be involved in the regulation of transcription, RNA editing and nuclear retention of selected mRNAs (Fox and Lamond 2010). NEAT1 is a nucleus-retained lncRNA that associates with paraspeckles and that exists in two isoforms, a shorter 3.7 kb long isoform called NEAT1_1 (or MENε) and a longer ~ 23 kb long isoform called NEAT1_2 (or MENβ) (Clemson et al. 2009; Sasaki et al. 2009; Sunwoo et al. 2009). The longer isoform NEAT1_2 has been shown to be critical for paraspeckle formation (Naganuma et al. 2012; Yamazaki et al. 2018). NEAT1_2 is characterized by a modular structure, in which the middle module binds to the RNA binding protein non-POU domain containing octamer binding (NONO), which is necessary and sufficient for the initiation of LLPS and paraspeckle formation (Yamazaki et al. 2018). Interestingly, paraspeckle formation is highly dynamic and can be regulated by various types of stress. It has been shown that NEAT1 and paraspeckles are involved in the response to viral and bacterial infection (Imamura et al. 2014, 2018). Furthermore, NEAT1 and paraspeckles have been reported to be induced by stress conditions like heat shock and hypoxia (Choudhry et al. 2015; Godet et al. 2022; Lellahi et al. 2018). NEAT1 is also induced upon activation of the integrated stress response following ER stress in multiple myeloma cells, where it interacts with the apoptosis antagonizing transcription factor (AATF) in paraspeckles to prevent R-loop accumulation and activation of an inflammatory response (Bruno et al. 2022). In addition, mitochondrial stress or depletion of mitochondrial genes can also result in an induction of NEAT1 and increased numbers of paraspeckles, which leads to nuclear retention of mitochondrial and apoptotic mRNAs and subsequently to improved cell survival (Wang et al. 2018). Recently, it has been shown that NEAT1 promotes LLPS and the formation of TAR DNA binding protein 43 (TDP-43) nuclear bodies (Wang et al. 2020). Upon arsenite stress, TDP-43 is targeted partly to stress granules in the cytoplasm and mostly to NEAT1-mediated nuclear bodies that overlap with paraspeckles. TDP-43 has been associated with the development of amyotrophic lateral sclerosis (ALS) and an ALS-related mutation (D169G) of TDP-43 has been shown to impair NEAT1-mediated LLPS and nuclear body formation, resulting in an accumulation of TDP-43 in stress granules instead (Fig. 4c). Importantly, nuclear TDP-43 appears to be cytoprotective, whereas cytoplasmic TDP-43 is thought to be involved in the development of ALS. These findings suggest a critical function of NEAT1 and paraspeckles in response to a broad range of cellular stress types. The dynamic regulation of LLPS by varying the concentrations of a lncRNA may provide a powerful tool to the cell to adapt to different stress conditions, to titrate the necessary stress response, and to prevent pathophysiological conditions, such as ALS.

A stress-related function has been also ascribed to the metastasis-associated lung adenocarcinoma transcript 1 (MALAT1). Initially, MALAT1 has been associated with lung tumor progression and metastasis (Ji et al. 2003). Later studies showed that MALAT1 localizes to nuclear speckles (Hutchinson et al. 2007; Wilusz et al. 2008), a ribonucleoprotein granule that is known to contain RNA splicing factors and transcriptional regulators (Spector and Lamond 2011). Interestingly, MALAT1 is dispensable for the formation of these nuclear compartments, and neither do Malat1-deficient mice display abnormalities in pre-mRNA splicing nor do they have obvious developmental defects or a measurable reduction in viability (Eissmann et al. 2012; Nakagawa et al. 2012; Zhang et al. 2012). These findings raised the question regarding the actual biological function and the physiological role of MALAT1. Since then, MALAT1 has been shown to be induced upon different types of cellular stress, such as hypoxia in endothelial cells (Michalik et al. 2014), treatment with chemotherapeutic drugs in multiple myeloma (Handa et al. 2017), or lipopolysaccharide treatment in macrophages (Zhao et al. 2016). Recently, Malat1 has been also shown to be involved in CD8+ T cell differentiation upon viral infection (Kanbar et al. 2022), suggesting an important role in the immune response. Interestingly, upon heat shock MALAT1 appears to redistribute from nuclear speckles to a distinct nuclear compartment named heat-inducible noncoding RNA containing nuclear bodies (HiNoCo bodies) (Onoguchi-Mizutani et al. 2021). Although the precise composition of the HiNoCo bodies and the mechanism of how these granules are formed are currently unknown, these lncRNA containing nuclear granules may function as critical biosensors in the response to sudden temperature changes (Fig. 4d). While more research is needed to fully understand the role of noncoding RNAs like MALAT1 in the mammalian stress response, these examples demonstrate how lncRNAs can potently modulate the output of different stress responses through the efficient and timely regulation of cellular phase-separated condensates.

Conclusion

We have categorized the molecular functions of stress-related lncRNAs into three general classes: (I) gene regulatory functions in cis or trans, (II) roles as molecular scaffolds or tethers, and (III) functions in the regulation and formation of phase-separated condensates (Table 1). These three general modes of action and the underlying molecular mechanisms presented in this review should be however regarded as an initial framework that will need continuous revision. As new biochemical, molecular and genetic technologies evolve, an increasing number of stress-related lncRNAs and new molecular modes of action will be identified. On the other hand, individual functions of the currently existing and known lncRNAs may need to be revised or complemented by novel models based on future findings.

Table 1 Summary of stress-related lncRNAs and their modes of action
Full size table

Regardless, lncRNAs are perfectly suited molecules to exert important functions in the response to various types of cellular stress. Since there is no need for an additional step to translate the RNA, as it is the case for protein effectors and their encoding mRNAs, lncRNAs can be almost instantly induced upon the emergence of a stress condition, once the stress signal has been recognized and processed (e.g. in the cases of PANDAR and DINO). The functionality of a stress-induced lncRNA can be further adjusted through its abundance and/or stability, which can induce effects that are rather locally and timely restricted (e.g. in the case of Pvt1b) or extended to distant genomic or cellular sites (e.g. in the case of PINCR or NORAD). And lastly, the often-modular structure of lncRNAs make them pleiotropic effectors that can initiate a multifaceted stress response depending on the respective cellular demands and experienced stress conditions (e.g. in the case of GAS5). The modular structures of lncRNAs and how they can confer functionality to the noncoding transcript will be an interesting area of future research as well as the remaining issue of the frequently seen imbalance between the abundance of the lncRNA and the abundance of the interacting RNA or protein binding partners. While a few lncRNAs may simply display a sufficient abundance to support their ascribed molecular mode of action, others may fail to do so. However, a lncRNA’s molecular characteristics and underlying mode of action, such as the formation of phase-separated condensates or the recycling of low abundant lncRNA molecules in a certain cellular reaction (e.g. in the case of SPARCLE), may provide plausible evidence for its mode of action, even if the transcript is less abundant and expressed at a sub-stoichiometric ratio (Unfried and Ulitsky 2022). Precise biochemical testing of RNA-protein and RNA-RNA interactions, deciphering the RNA structure and associated structure-function relationships, as well as rigorous genetic analysis of the underlying modes of action will help to identify novel lncRNAs with valid molecular functions not only in the mammalian stress response, but also in normal cellular homeostasis, physiology and disease.