The Landscape of Subcellular Long Non-coding RNAs Links Organelle Metabolic Homeostasis

Organelles entail specialized molecules to regulate their essential cellular processes. However, systematically elucidating the subcellular distribution of functional molecules such as long non-coding RNAs (lncRNAs) in tissue homeostasis and diseases has not been fully achieved. Here, we characterized the organelle-associated lncRNAs from mitochondria, lysosome, and endoplasmic reticulum (ER), respectively, and revealed the diverse and abundant distribution of lncRNAs. Among them, we identied mitochondrial lncRNA Growth-Arrest-Specic 5 (GAS5) as a tumor suppressor in maintaining cellular energy homeostasis. Mechanistically, energy stress-induced GAS5 modulated mitochondria TCA ux by declining metabolic tandem association of FH-MDH2-CS, the canonical members of the TCA cycle. Remarkably, the expression of GAS5 negatively related with levels of its associated mitochondrial metabolic enzymes and breast cancer development. Together with the detailed functional annotations, this subcellular lncRNA identication revealed the human cell’s inquisitively complex architecture, aiding in the development of new strategies for the clinical application of organelle-associated lncRNAs. 3-5 samples/repeats per experiment/group/condition to detect a 2-fold difference with power of 80% and at the signicance level of 0.05 by a two-sided test for signicant studies. For immunohistochemical staining and immuno-blot, the representative images were shown. Each of these experiments was independently repeated for over 3 times. Relative quantities of gene expression level were normalized to B2M or GAPDH. Results were reported as mean ± Standard Deviation (S.D.) of at least three independent experiments. Comparisons were performed using two tailed paired Student’s t-test (*P < 0.05, **P < 0.01, ***P < 0.001), as indicated in individual gures. For survival analysis, the expression of indicated genes was tested as a binary variant and divided into ‘high’ and ‘low’ groups. Kaplan-Meier survival curves were compared using the Gehan-Breslow test with Prism Software (GraphPad, La Jolla, CA). The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.


Introduction
Organelles are microscopic semi-organs that underlie many cellular processes, including several important metabolic reactions, energy production, cellular signaling, and cell growth 1-6 . Each organelle, such as mitochondrion, lysosome, and endoplasmic reticulum (ER) carries out its faithfully characteristic functions as it possesses a unique set of proteins, lipids, and other molecular factors 7-10 . Multiple unique roles of organelles are revealed through searching the hierarchical dynamics of compartmentalized pools of molecules. For example, analyzing lysosome proteomics and metabolomics revealed its critical roles in regulating metabolic resource sensing and allocation 11,12 . Quantitative proteomics analysis of mitochondria Acylomes identi ed it as the holder of Acylomes 13 . It showed that it regulated the TCA cycle highly through the acylation of the critical targets (e.g.,MDH2) 13 . Therefore, comprehensively exploring and de ning such organelle-possessed unique molecular sets aid in unveiling the de novo functions of organelles, which provides novel insights into their associated cellular functions in human diseases like cancer.
Accumulating evidences have indicated that extensive lncRNAs are located in the cytosol and involved in multiple signaling pathways in homeostasis and human diseases [14][15][16][17] . Our recent study also showed that lncRNAs were naturally associated with the lipid components of the cellular membrane and played critical roles in signaling pathways 14 . This nding suggested that the localization of lncRNAs played a vital role in implementing their functions in various cellular processes. Decoding the subcellular distribution of organelle-associated lncRNAs will provide an important resource to interpret the complex subcellular architecture, cell dysfunction, and pathophysiology of human diseases.
Although the cytoplasmic RNAs are presumably thought to function in a wide range of organelleassociated biological processes, the understanding of subcellular RNAs is still limited. Most of the current techniques for detecting subcellular localization of RNAs follow the image-observing and fraction-localization 26,27 . Thus, we chose the lysosomal Lamp2 protein's coding mRNA as the marker for the lysosome RNA set. We validated the mitochondria, lysosome, ER sets by the other organelles-speci c markers (including protein and RNA levels), which helped exclude the organelle cross-contaminants. The relative enrichment of each RNA was calculated by 2^-(Ct Mito -Ct Total ) followed by normalizing all ratio value to the GAPDH in the control group (Value of the rst GAPDH column was normalized as 1), and the value of the cut-off line was determined as 1. The relative RNA enrichment of marker genes in each indicated organelle further con rmed the purity of isolation fractions (Fig. 1c). Notably, clear isolation between lysosome and mitochondria was di cult when using classical centrifugation-based organelles fraction methods because of their very similar sedimentation effect. However, we found little crosscontamination in isolated mitochondria and lysosome fraction (Fig. 1b, c), which attested the ability of this isolation method to distinguish the different subcellular RNAs among organelles.

Overview of the organelle-associated RNAs
To explore the subcellular organelle-associated lncRNAs, we sequenced ten samples from ve groups (i.e., total RNAs, LMF, mitochondria, lysosome, and endoplasmic reticulum) in HEK293T cells. Quality control analyses, including individual expression distribution, technical replicates' assessment, and principal component assays, were applied to attest these sequence results (R > 0.95, P < 0.05) ( Fig. 1d- Table 1), suggesting the consistency of the replicates within each group. Our sequence results utilizing the DESeq2 method 28 suggested the enrichment of organelleassociated RNA markers (highlighted in red) was enriched better than LMF fractionation (Fig. 1e, f).
Through it, we revealed 2292 organelle-associated lncRNAs (Organelle/Total, P < 0.05, fold-change > 1.5 for enrichment threshold) (Supplementary Table 1), which covered 5.2% of all the detected total lncRNAs in HEK293T cells (Fig. 1g). Moreover, we characterized both the unique and overlapping distribution patterns of the organelle-associated lncRNAs and mRNAs, respectively (Fig. 1d,h,i and Extended Data Fig. 1g,h). Through it, 370 lncRNAs were identi ed in all three organelles (i.e., mitochondria, lysosome, and endoplasmic reticulum) (Fig. 1h). Further lncRNA expression enrichment and cluster heatmap assays con rmed the speci city and diversity of the organelle-associated lncRNAs (Fig. 1j). As expected, the known nuclear-located lncRNA genes (NEAT1, XIST) were not enriched in the indicated organelles, underscoring the purity of our cytoplasmic fraction (Fig. 1j). Furthermore, using the organelle-associated speci c markers con rmed the constant validation of individual organelle components in our analysis (Extended Data Fig. 1i). In summary, our puri cation strategy showed its merit in speci c enriching organelle-associated lncRNAs.
Next, we analyzed the potential function of each organelle-associated lncRNA set using GO and KEGG pathway analyses. We found that each organelle lncRNA set was involved in different GO terms or pathways consistent with their associated organelles . For example, the lncRNAs in the mitochondrial set were found involved in cellular metabolic processes through various metabolism-related signaling events, such as AMPK signaling, alcoholism, and TNF signaling pathways (Extended Data Fig. 1j-l).

Functional validation of subcellular LncRNAs in cellular homeostasis
We validated the subcellular lncRNAs distribution and potential functions accompanied by their associated organelles. A high abundance of organelle lncRNA component enrichment was observed by picking up the lncRNA candidates of each set for the RT-qPCR validation. Among them, we respectively con rmed 21 out of 23 mitochondrial lncRNAs (Extended Data Fig. 2a), 10 out of 15 lysosome lncRNAs (Extended Data Fig. 2b), and 3 out of 14 ER lncRNAs (Extended Data Fig. 2c) for each isolated organelle. Through siRNA screening (Extended Data Fig. 2d, e), the subcellular LncRNAs were strikingly characterized functions in many important cellular processes (e.g., Glucose sensitivity and ATP production), which were consistent with their organelle-associated functions (Fig. 1l, m and Extended Data Fig. 2d-g). Interestingly, 2 out of 9 lysosome-associated lncRNAs candidates were involved in pivotal cellular energy sensor pathways such as the AMPK pathway ( Fig. 1l and Extended Data Fig. 2d, f), which was explicitly activated on the surface of the lysosome 6,10 . Meanwhile, 4 out of 12 mitochondriaassociated lncRNAs candidates were involved in mitochondrial ATP generation (Fig. 1m), and 9 out of 12 candidates were involved in whole-cell ATP production (Extended Data Fig. 2e, g), which was known as an essential function of mitochondria in energy homeostasis. Intriguingly, abundant subcellular lncRNAs could respond to the cellular energy stress condition (Fig. 1l,m,and Extended Data Fig. 2f,g), highlighting their potential functions in cellular energy homeostasis. Thus, mapping the subcellular lncRNAs distribution and their pivotal functions further sparked the novel non-canonical molecular mechanisms of organelles through their accompanying components.
Mitochondrial lncRNA GAS5 is identi ed as a glucose responder.
Organelles are involved in pivotal metabolic processes, where mitochondria acted as a nexus engine in cellular energy sensing and homeostasis 1,29 . Among the characterized mitochondrial lncRNAs candidates, GAS5 was identi ed as a mitochondria-located and functionally-related lncRNA (Fig. 1m). The partial colocalization between GAS5 and mitochondria was observed either by RNA uorescence in situ hybridization (FISH) or by , suggesting its potential role in mitochondria-related cellular processes. We also excluded the lysosome and ER distribution of GAS5, indicating its unique association with mitochondria (Extended Data Fig. 3a, b).
Next, we assessed the function of GAS5 in responding to cellular energy availability and regulating mitochondrial metabolism. Interestingly, the GAS5 expression and its association with mitochondria were remarkably increased under glucose-deprivation ). The glucose starvation-caused upregulation of GAS5 was reversed by glucose restoration (Fig. 2f). However, GAS5 showed less sensitivity to serum starvation, suggesting its speci c response to central carbon metabolism (Extended Data Fig. 3d). Our data further showed that GAS5 knockdown promoted oxygen consumption rate (OCR) (Extended Data Fig. 3e, f), ATP production ( Fig. 2g, h), and NADH generation (Fig. 2i,j), especially under glucose starvation condition (Extended Data Fig. 3g, h). It indicated a crucial role of GAS5 in mitochondria metabolic regulation.
The GAS5-Loop2 region dictates its mitochondrial translocation and function.
To identify GAS5-associated proteins potentially involved in the GAS5-related mitochondrial regulation, we performed an RNA pulldown assay followed by mass spectrometry (MS) analysis 15,30 (Fig. 2k,Extended Data Fig. 3i and Supplementary Table 2). Interestingly, the sense GAS5, rather than the antisense or beads control, bound to MDH2, a canonical member of the mitochondrial TCA cycle as well as a mitochondrial NADH/NAD + circulator ). The RNA-protein binding assay using cell lysates or recombinant MDH2 veri ed the direct interaction between GAS5 and MDH2 both in vivo and in vitro (Fig. 2m, n). The speci c interaction between GAS5 and MDH2 was also controlled by glucose, as indicated by the RNA immunoprecipitation (RIP) assay ( Fig. 2l and Extended Data Fig. 3j). As shown in Extended Data Fig. 3k, l, there were around 1828 copies of GAS5 per HEK293T cell, which was of relatively high abundance compared with several known functional lncRNAs: LINK-A as roughly 150 copy per MDA-MB-231 cell 14 and CamK-A as roughly 937 per MDA-MB-231 cell 16 . These pieces of evidence suggested that mitochondrial GAS5 might regulate MDH2-associated metabolism.
GAS5 regulates the FH-MDH2-CS tandem association to modulate mitochondrial metabolism.
Next, we examined the role of MDH2 in mitochondrial TCA cycling. As for the standard Gibbs free energy (ΔG 0' ) of the TCA cycle steps (Fig. 3a), we found that the ΔG 0' of MDH2 node was a highly positive value, suggesting that the forward reaction was theoretically unspontaneous in vitro. However, the in vivo ΔG of MDH2 node was almost 0, where the rapid oxaloacetate (OAA) consumption by citrate synthase (CS) and su cient malate supply by fumarate hydratase (FH) might count 31 . As expected, these three canonical members (FH-MDH2-CS) of mitochondrial TCA cycling were colocalized with mitochondria, and they were found to have a considerably high colocalization statistical index (Extended Data Fig. 4a-i). We then performed the co-immunoprecipitation (co-IP) assay in the isolated mitochondria fraction and found FH-MDH2-CS robustly associated with each other in mitochondria ( Fig. 3b and Extended Data Fig. 4j). Their physical interaction was further con rmed by in vitro protein pulldown assay (Fig. 3c, d). Besides, through mapping the CS interaction associated domain of MDH2 by co-IP assay, the enzyme activity core region (containing substrates binding sites, referring to PDB: 4WLE, 4WLF, 4WLU, 4WLV, 4WLN, and 4WLO MDH2 structure models) along with the MLS (mitochondria location signal peptides, 1-24 amino acid) of MDH2 was available for its interaction with CS (Extended Data Fig. 4k, l), suggesting their tight metabolic association. This complex formation was remarkably sensitive to glucose supply conditions (Extended Data Fig. 4m, n), suggesting that this exible FH-MDH2-CS complex formation facilitated an e cient metabolite stream modulation under various energy situations (Fig. 3e).
Interestingly, overexpression of GAS5-FL but not its Loop2-deletion mutant (GAS5-D2) could mimic the energy stress stimulation to block the FH-MDH2-CS complex formation (Fig. 3f,g and Extended Data Fig. 4p,q). However, GAS5 showed a mild effect on MDH2 enzyme activity in vitro (Extended Data Fig. 4o). Consistently, GAS5 knockdown signi cantly attenuated the disintegration of the FH-MDH2-CS complex under glucose starvation (Fig. 3h, i) and caused the ectopic malate and citrate level (Fig. 3j, k).
Next, we further con rmed the speci c metabolic function of mitochondria-distributed GAS5 in the GAS5knockout HEK293T cell. Reconstituting GAS5 could rescue the FH-MDH2-CS complex formation and the cellular level of ectopic malate/citrate, but this was not the case for the GAS5-D2 mutant (Fig. 3l-q and Extended Data Fig. 3p). The cell growth assay further highlighted that the Loop2-dependent mitochondrial distribution of GAS5 was vital for its mitochondrial metabolic function (Fig. 3r). Given the similar phenotype observed between the GAS5-D2and EV-transduced cells, the mitochondrial metabolic function of GAS5 mainly contributed to its inhibitory role in growth control (Fig. 3r). Thus, we indicated that mitochondrial GAS5 could transmit the energy stress signal, disrupt the FH-MDH2-CS complex formation, and eventually inhibit the mitochondria metabolism and cell growth (Fig. 4s).
GAS5 regulates the FH-MDH2-CS complex formation by targeting the MDH2 acetylation.
We next investigated the molecular mechanism underlying the GAS5-impaired FH-MDH2-CS complex formation. Accumulating studies suggested that the metabolic process was associated with the reversible acetylation of metabolic enzymes in cells controlled by cellular energy status 32-34 . It was known that the acetylation of MDH2 at K185, K301, K307, and K314 were essential for its enzymatic activity 34 , which was highlighted in red in the 3D molecule model of the acetylated MDH2 (Extended Data Fig. 5a). Considering the acetylation of MDH2 was remarkably impaired under glucose starvation and robustly boosted under glucose treatment ( Fig. 4a and Extended Data Fig. 5b), we hypothesized that MDH2 acetylation could be regulated by mitochondrial GAS5. Indeed, loss of GAS5 dramatically promoted MDH2 acetylation as controlled by glucose ( Fig. 4b and Extended Data Fig. 5c), while overexpression of GAS5, but not GAS5-D2, resembled the effect of energy stress in decreasing the MDH2 acetylation (Extended Data Fig. 4d). Similarly, re-expressing GAS5, but not GAS5-D2, rescued the MDH2 acetylation in the GAS5-KO HEK293T cell (Fig. 4c). Thus, GAS5 regulated TCA ux metabolic process, probably by negatively regulating MDH2 acetylation in response to energy stress.
Interestingly, we found that the treatment of nicotinamide (NAM, an inhibitor of SIRTs) but not Trichostatin A (TSA, an inhibitor of HDACs) not only arti cially increased the acetylation of MDH2 ( Fig.  4d) but also strikingly strengthen the FH-MDH2-CS complex formation (Fig. 4e). To further investigate MDH2 acetylation's effect on the FH-MDH2-CS complex formation, we generated a 4KR mutant to disrupt the MDH2 acetylation sites (K185, K301, K307, and K314) (Fig. 4f). In contrast to wild-type MDH2 (MDH2 WT), the 4KR mutant signi cantly lost the ability to maintain the FH-MDH2-CS complex formation under NAM treatment (Fig. 4g, h). As the acetylation of CS and FH was not sensitive or abundant enough (Extended Data Fig. 5e, f), these ndings further indicated GAS5 impaired the FH-MDH2-CS complex formation by declining the acetylation of MDH2.
The role of GAS5 in modulating mitochondrial metabolism depends on its mitochondria localization.
Mitochondria-associated GAS5 inhibits breast tumorigenesis by restricting TCA ux.
Mitochondria metabolism was critical for tumor progression 35 , and GAS5 was associated with human  Fig. 6g-m). Furthermore, overexpression of GAS5 inhibited tumor proliferation and anchorage-independent growth of breast cancer cells (Extended Data Fig. 6n, o). Collectively, these results indicated that GAS5 functioned as a key regulator of breast cancer metabolism to control breast tumorigenesis.
Given mitochondrial MDH2 was the critical effector of GAS5 ( Fig. 2-4), we examined whether MDH2 was required for the GAS5-mediated breast cancer suppression. The citrate/malate detection assay and the colony formation assay showed that expressing MDH2, but not its 4KR mutant, signi cantly attenuated the inhibitory role of GAS5 in regulating malate/citrate level and tumor cell growth ( Fig. 5j-l). Furthermore, although GAS5 could partly mimic the metabolic impact caused by energy stress, citrate restoration could signi cantly rescue the cell growth in the context of GAS5 overexpression (Extended Data Fig. 6p), suggesting that the citric acid cycle functions downstream of GAS5. Consistent with tumor cell growth assay ( Collectively, these results demonstrated a tumor suppressor role of mitochondrial GAS5 by modulating mitochondria TCA cycling. High expression of GAS5 and low TCA ux bene t clinical outcomes in breast cancer patients Next, we examined GAS5 and TCA metabolism enzymes' expression levels in a cohort of breast cancer tissues obtained from Sun Yat-sen University Cancer Center (SYSUCC) (Supplementary Table 3). Downregulation of GAS5 was found in breast cancer tumors compared with the corresponding adjacent normal tissues ( Fig. 6a; n = 48). We divided the patients into two groups (GAS5-low and GAS5-high), based on the expression of GAS5 compared to the median value of all patients. The GAS5-high group showed a better survival rate than the GAS5-low group ( Fig. 6b; n = 200). The immunohistochemistry (IHC) analysis further con rmed that tumors with low GAS5 expression harbored a high level of MDH2 and progression markers (i.e., Ki67and Cyclin D1) (Fig. 6c,d;n = 200). Therefore, these results showed an inverse relationship between GAS5 and tumor progress.
We also examined the MDH2 expression in breast tumors and normal tissues by RT-qPCR ( Fig. 6e; n = 48) and IHC (Extended Data Fig. 6w; n = 100) in a cohort of breast cancer tissues, nding that MDH2 was upregulated in the advanced breast tissues. Moreover, we categorized the patients into two groups (MDH2-low and MDH2-high) based on the expression of MDH2 and found that low MDH2 level bene ted the overall survival rate of breast cancer patients ( Fig. 6f; n = 200). The IHC staining assay also revealed the high MDH2 expression in a subset (MDH2-high subset) of breast cancer tissues (New Fig. 6g; n = 200). Moreover, tumors with high MDH2 expression also showed a high expression of cell proliferation markers (i.e., Ki67 and Cyclin D1) (Fig. 6g, h). Notably, FH and CS were also found highly expressed in the advanced breast cancer tissues ( Fig. 6i, j, and Extended Data Fig. 6v, w), while SIRT3 was found downregulated there (Extended Data Fig. 6s, t). These results suggested that high levels of MDH2, CS, and FH favored tumor progress, and SIRT3 negatively regulated the FH-MDH2-CS complex in both cells and tumors.
A further subgroup of breast cancer patients was conducted to investigate the relationship between GAS5-MDH2 and the patient survival rate. Notably, patients with the low level of MDH2 and high level of GAS5 were signi cantly associated with a better survival rate ( Fig. 6k; n = 200, P < 0.001) as compared with that of individual GAS5 (Fig. 6b, n = 200, P = 0.0039) or MDH2 (Fig. 6f, n = 200, P = 0.0198).
Collectively, these data suggested that the mitochondria-associated GAS5-MDH2 axis was involved in breast cancer metabolism and tumorigenesis (Fig. 6l), highlighting its potential diagnostic marker and therapeutic target roles for breast cancer.

Discussion
Membrane-enclosed organelles, such as the mitochondria, endoplasmic reticulum (ER), lysosome, and Golgi, de ne various critical cellular processes 7 . Comprehensively exploring the components of organelles will unveil substantial clues underlying cellular function and human diseases. Traditional approaches, despite the developed density medium from sucrose to Percoll, often result in contamination across factionations 37,38 . However, the approximation of the sedimentation coe cient between organelles (e.g., mitochondria and lysosome), especially the membrane-less granules, frequently compromises the sensitivity in the RNA detection. The previous study of mitochondria RNAs treated the density-centrifuged pellet with RNase to clear the outer RNAs 21 . However, the membrane-anchored RNAs were ablated, and the membrane-less granules with resistance to RNase would remain. In this study, we developed an effective organelle isolation method to obtain the non-arti cial treatment subcellular components by combining centrifugation with endogenous immunoprecipitation using speci c organelle protein markers. Recent developments applied proximity labeling techniques coupled proteomics to a subcellular RNAs study 27,39 . However, it required exogenous transduction and chemical treatment.
Compared with them 27,40 , our method does not rely on additional biochemistry treatment or cell transfection; therefore, it is more suitable and convenient for investigating the bona de cellular events under physiological conditions, including energy sources, drugs, and cytokines.
Using this approach, we identi ed thousands of organelle-associated RNAs, especially the subcellular lncRNAs that might function in numerous essential cellular processes. Notably, most of these newly identi ed lncRNAs were missed in the studies using either the APEX-RIP-Seq or the newly APEX-Seq 27,39 .
As the emerging pivotal roles of lysosome and mitochondria in regulating cellular energy homeostasis 2,10,29,41 , the identi ed mitochondria-and lysosome-associated lncRNA sets could function as messengers and regulators in organelle communication and cellular metabolism processes.
Among them, we identi ed a mitochondria-associated lncRNA named GAS5. GAS5 was identi ed in a subtraction cDNA library that hosts differentially expressed genes in growth-arrested cells 42 and was then characterized as a universal tumor suppressor 43 . Mechanistically, GAS5 was found to act as a DNA decoy for glucocorticoid receptor (GR) in response to growth factors 22 , regulate insulin receptor (INSR) gene transcription in adipocytes 23 , and act as an RNA sponger to buffer miRNAs 24 . However, these limited results were still hard to explain its roles in the unique cellular localization and glucose metabolism. Excitingly, we found mitochondria-associated lncRNA GAS5 could respond to the glucose condition and regulate TCA ux by suppressing the FH-MDH2-CS complex formation, which largely contributed to its tumor suppressor function (Fig. 3r, s).
Glycolysis, with actually low absolute energy generation e ciency, provides abundant intermediate metabolites and NADPH for anabolism and it is usually highly activated in cancers 35 . However, previous studies had found that members of the mitochondrial TCA cycle could bene t tumors in non-canonical ways 44,45 . It was also shown that GAS5 overexpression took no signi cant effect on glycolysis 46 . Strikingly, our ndings showed that mitochondrial TCA cycle enzymes FH/CS/MDH2 were all upregulated in tumor tissues (Fig. 6e, i, j, and Extended Data Fig. 6u-w) and GAS5 could suppress cell growth by directly blocking the TCA cycle by controlling these enzymes. Besides, mitochondria metabolismregulated ATP, NAD(P) + /NAD(P)H, and intermediate metabolites were also critical for macromolecular synthesis and redox control during cell growth and proliferation. Together with the previous studies 41,44,45,47,48 , our ndings proposed that the mitochondrial TCA cycle was both su cient and necessary for tumor development in both canonical metabolism manner and non-canonical manner.
Given the crucial role of this uncovered mitochondrial-associated lncRNA in regulating breast cancer mechanism and progression, our study paradigm-shifts our current understanding of the mitochondrial TCA cycle in physiology and cancer.

ACKNOWLEDGMENTS
We would like to thank Tian-hua Zhou (Zhejiang University), Jun Huang (Zhejiang University) and Pinglong Xu (Zhejiang University) for their support and suggestion on this study. We thank Prof. Jia-huai Han (Xiamen University) for gifting CS and FH template vectors. We thank Prof. Xu Li (Westlake University) for gifting SIRTs vectors and assistance for protein MS analysis. We thank Hai-long Piao (Chinese Academy of Science) for metabolites analysis. We also thank Cipher Gene, LLC, for the support in generating and processing the RNA-seq data, and for the help in data interpretation. This work was supported in part by

CONTACT FOR REAGENT AND RESOURCE SHARING
Further information and requests for resources and reagents should be directed to and will be ful lled by the Lead Contact, Aifu Lin (linaifu@zju.edu.cn). CVCL_0063) were purchased from American Type Culture Collection (ATCC) and characterized by Cell Line Core Facility (MD Anderson Cancer Center). These cell lines were maintained in Dulbecco modi ed essential medium (DMEM) supplemented with 10% fetal bovine serum (FBS) at 37 °C in 5% CO2 (v/v). All cells were negatively tested for mycoplasma contamination and authenticated based on STR ngerprinting before use.

Tissue Samples
Fresh frozen breast cancer tissues (Sun Yat-sen Cohorts) were obtained from Sun Yat-sen University Cancer Center (SYSUCC) as previously described 16 . The study protocol was approved by the Institutional Review Board of Sun Yat-sen University Cancer Center. All tissue samples were collected in compliance with informed consent policy. Detailed clinical information is listed in Supplementary Table 3.

Mice
All animal experiments were performed in accordance with a protocol approved by the Institutional Animal Care. Care of experimental animals was in accordance with guidelines and approved by the Laboratory Animal Committee of Zhejiang University. Female nude mice (4-5 weeks old) were purchased from the Shanghai SLAC Laboratory Animal Co., Ltd.

Cloning Procedures
The full-length MDH2 and GAS5 were cloned from HEK293T cDNA by PCR. CS, FH full-length template was gifted from Jia-huai Han lab. SIRT1-7 vectors were gifted from Xu Li lab. All these eukaryotic overexpression genes (wild type and mutants) were cloned into SFB-lv (S-tag, FLAG-tag and SBP-tag fused) vector using the Gateway system (Invitrogen) and pcDNA3.1-Flag/Myc/HA/His empty vectors using T4 ligase (Promega) or ClonExpress II One Step Cloning Kit (Vazyme). GAS5 and its deletion mutants were cloned into pGEM-T easy (Promega) for in vitro transcription. All single-point and deletion mutations were generated by PCR overlapping. Bacterial expression vectors for MBP-His-tagged MDH2 and SIRT3 (wild type and mutants) were constructed by cloning into pMBP28a vector.

Protein Recombination and Puri cation
Recombinant proteins His-MBP-MDH2 and SIRT3 (wild type and mutants) were expressed in E. coli strain BL21-CodonPlus ® (DE3)-RIPL (Agilent Technologies) and puri ed using Ni-NTA Se nose Resin (Sangon Biotech), respectively. Human bioactive SFB or FLAG tagged MDH2, CS and FH were puri ed from overexpression vectors transduced HEK293T cells. Puri cation was conducted by using FLAG (M2) magnetic beads (Sigma) to enrich and using 3×FLAG peptide (Sigma) to elute. All the concentration and purity of recombinant proteins were measured by SDS-PAGE and Coomassie staining with the standard BSA control.

LMF, Mitochondria and Lysosome Puri cation
Cells (at least 10 7 ) were washed once by KPBS (136 mM KCl, 10 mM KH 2 PO 4 , pH 7.25) and harvested by gentle scraping. Cells were pelleted down by centrifugation at 4 °C at 1000 g for 1 min. The cell pellet was resuspended with 500 μl KPBS containing the inhibitor cocktail and subjected to 2 ml Dounce homogenization. 40-50 strokes were su cient and the e ciency could be monitored by bright eld microscopy with trypan blue staining. After homogenization, the cell extraction was transferred to new 1.5 ml EP tubes and centrifuged at 4 °C at 1,000 g for 10 min. The resulting supernatant was reserved and further centrifuged at 4 °C at 1,000 g for 10 min. The resulting supernatant was the indicated LMF suspension. The real LMF pellet (regarded mainly as mitochondria and lysosome mixture) could be obtain by centrifugation at 4 °C at 13,000 g for 5 min.
500 μl above indicated LMF suspension should be incubated with 2 μl Tom20 antibody at 4 °C for 2 hr with gentle rotation, following by adding 20 μl KPBS-precleared protein A/G magnetic beads for further incubation at 4 °C for additional 1 hr rotation. After the incubation, the mitochondria were supposed to be captured and immobilized by Tom20 antibody coated beads. The enriched mitochondria were collected in the pellet of beads while the supernatant were discarded by physically magnetic enrichment. The beads enriched mitochondria should be further washed by pipetting 10 times in 500 μl KPBS. The washing step could be optionally replicated by 3 times and the resulting beads complex was the puri ed mitochondria fraction.
As the indicated Tom20 antibody was shift into the LAMP1 one, the procedure of the lysosome fraction was largely similar to the mitochondria fraction. However, considering the limiting content of the lysosome in cells, more cells would be additionally in need for better protein and RNA detection. buffer. Rotate the mixture solution for 20-30 min at 4 °C and occulent precipitate would gradually appear which was the ER microsome fraction. Centrifuge at 8,000 g for 10 min at 4 °C and the pellet was the ER microsomes. Remove the supernatant and wash the pellet twice with isotonic extraction buffer to clear the cytosol contaminant. The isolated ER could be used for RNA analysis and Western blot detection.

RNA Isolation and Sequencing
RNA degradation and contamination were monitored on 1% agarose gels; concentration was measured using Qubit ® RNA Assay Kit in Qubit ® 2.0 Flurometer (Life Technologies, CA, USA); RNA integrity was assessed using the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent Technologies, CA, USA).
A total amount of 1 μg RNA per sample was used as input material for the RNA sample preparations by rRNA depleting method. Libraries were sequenced on an Illumina NovaSeq platform and 150 bp pairedend reads were generated.

Data Analysis
Reference genome and gene model annotation les were downloaded from genome website Transcripts predicted to have coding potential by all of the two prediction tools (CNCI (Coding-Non-Coding-Index) (v2) 50 /CPC (Coding Potential Calculator) (0.9-r2) 51 ) were ltered out, and transcripts with no coding potential were selected as the novel candidate set of lncRNAs.
Ballgown was used to calculate FPKMs of lncRNAs in each sample. Technical replicates were evaluated by scatter plot with FPKM data. Differential gene were analyzed by DESeq2 28 , fold-change over 1.5 was selected as threshold as different expressed. By this criterion, heatmap of expression pattern was generated with R package (pheatmap); overlapping and speci c lncRNAs between organelles was showed by Venn and Circos maps.
In order to evaluate our isolation e ciency, LMF was utilized as the traditional fractionation separating method, all 13 mitochondrial genes and 9 Lysosome genes (C12orf66/LAMP2/LAMP3/RAB7A/RHEBL1/RPS6KC1/SNX6/STX7/VPS26A) were selected as marker to assess gene enrichment in our method.
With Gene Ontology and KEGG annotation results, we classify DGEs according to o cial classi cation, and we also perform GO 52 and KEGG 53 functional enrichment using clusterPro ler (a package of R program) for both lncRNAs and mRNAs' function.
siRNA, shRNA and RNAi All siRNAs sequences were designed according to http://sirna.wi.mit.edu/home.php and all shRNAs sequences were designed according to https://portals.broadinstitute.org/gpp/public/. siRNAs were commercially generated (GenePharma). All shRNAs sequences were cloned into pLKO.1-Puro vector, two shRNA producing the best knockdown e ciency were used in the following functional studies. Detailed sequences were listed in the Supplementary Table 4 Methylstat. Lysates were cleared by centrifugation at 13,000 g for 15 min at 4 °C. Supernatants could be applied for immunoblotting (IB) or immunoprecipitation (IP) with the indicated antibodies. As for IP, add the required primary antibody and the control IgG separately to the prepared lysates. After incubation at 4°C for 3 hr with gentle rotation, add 20 μl protein A/G magnetic beads (Pierce) each to the lysates and incubate another 2 hr at 4 °C with rotation. Wash the protein captured beads with NETN buffer 3 times for 5 min each at 4 °C with rotation. Then eluted beads with 50 μl 2×SDS loading buffer and the eluted protein or protein complexes could be detected by IB. The blotting signals were detected using Clarity Western ECL Substrate (Bio-Rad). As for peptide label tagged protein IP, the primary antibody and the protein A/G beads could be replaced with FLAG M2 magnetic beads (Sigma), S-protein agarose beads (Millipore), M-280 Streptavidin Dynabeads™ (Invitrogen) or HA magnetic beads (Pierce).

RNA Pulldown and Mass Spectrometry Analysis
RNA pull-down assay was performed as previously described with minor modi cations 16,30,54 . Biotin labeled RNA was generated using biotin-RNA labeling mix (Roche) and MEGAscript T7 or SP6 Transcription Kit (Thermo Fisher Scienti c) and puri ed by RNA Clean & Concentrator-5 kit (Zymo Research). Cell lysate was prepared using polysome buffer (25 mM Tris-HCl pH7.5, 150 mM KCl, 0.5 mM DTT, 0.5% NP-40) with complete protease inhibitor cocktail (Roche) and Ribolock RNase Inhibitor (Invetrogen). M-280 Streptavidin Dynabeads™ (Invitrogen) were prepared according to manufacturer's instructions and then incubated with 10 μg biotin labeled RNA (Sense and antisense separately) in RNA capture buffer (20 mM Tris-HCl pH 7.5, 1 M NaCl, and 1 mM EDTA) with RNase Inhibitor for 30 min at RT. Wash the RNA-captured beads once with NT2 buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM MgCl 2 , 0.05% NP-40). Incubate the RNA captured beads and the non-RNA-captured beads control separately with prepared 30 mg cell lysates for 2 hr at 4 °C with gentle rotation. Then wash the beads with NT2 buffer three times, NT2 high salt buffer (NT2 buffer with 500 mM NaCl) twice, and PBS once for 5 min at 4 °C and nally add 50 μl 2×SDS-loading buffer 95 °C heat for 10 to 15 min. The product could be subject to MS analysis. As for western blot detection, 0.5-1mg cell lysate and 1-3 μg biotin-RNA would be su cient. As for puri ed protein-RNA pulldown assay, 1-2 μg puri ed protein and 1-3 μg biotin-RNA would be su cient.

RNA Immunoprecipitation, RNA Extraction and RT-qPCR Detection
The enrichment of the interested protein process was mostly similar to the protein IP indicated in Cell Lysis, Immunoprecipitation and Immunoblotting section with the modi cation that all the processes should be in RNase-free way, additional Ribolock RNase Inhibitor (Invetrogen) was required, the lysis buffer was transferred to polysome buffer and the wash buffer was transferred to NT2 buffer. Then, use the TRIzol reagent (Invitrogen) to extract the associated RNAs according to the manufacturer's instructions. Reverse transcription was performed using the iscript cDNA synthesis kit (Bio-Rad) and the abundance of target RNAs was detected by iTaqTM Universal SYBR Green Supermix qPCR kit (Bio-RAD) according to the manufacturer's instructions.

RNA FISH and IF
RNA Fluorescence in situ hybridization (FISH) was performed with a FISH kit (Ribobio Co.) according to the manufacturer's instruction with minor modi cations. Brie y, cells with indicated treatment were xed in 4% formaldehyde for 10 min followed by washing with PBS. The xed cells were further dehydration through 70%, 90% and 100% ethanol. The air-dried cells were subjected to incubation with 40 nM FISH probe (Ribobio Co.) in hybridization buffer (100 mg/ml dextran sulfate, 10% formamide in 2×SSC) at 75°C for 3 min. The hybridization was then performed at 37 °C for 8 hr to overnight. Then, the slide was washed twice with 2×SSC (0.3 M NaCl, 0.03 M Na 3 Citrate, pH 7.0) at RT. The air-dried slide was mounted with Prolong Gold Antifade Reagent with DAPI for detection.
As for immuno uorescence (IF), cells were cultured in chamber slides overnight and xed with 3.7% formaldehyde in PBS for 10 min at RT, followed by permeabilization with 0.5% Triton X-100 in PBS for 10 min. Cells were then blocked with 5% FBS in PBS for 30 min at RT, and incubated with the indicated primary antibody for 1 hr at RT, followed by incubation with Anti-rabbit (or Mouse) IgG (H+L), F(ab')2 Fragment (Alexa Fluor ® 594 or 488 Conjugate) from Abcam for 30 min at RT. Coverslips were mounted on slides using anti-fade mounting medium with DAPI. IF images were acquired on a FV3000 confocal microscope (Olympus). For each channel, all images were acquired with the same settings.

NP
In vitro MDH2 Enzyme Activity Assay Immunoprecipitated Flag-MDH2 protein was eluted using 3×Flag peptide. Then, 1 μg FLAG-MDH2 was added to 200 μl reaction buffer (0.2 mM oxaloacetic acid, 0.1 mM NADH in 1×PBS). The reaction speed was measured by detecting the decreasing absorbance at OD 340 nm in microplate system.

In vitro Deacetylation Assay
Highly acetylated SFB-MDH2 protein was puri ed from SFB-MDH2 transduced HEK293T cells with TSA/NAM treatment using S-protein agarose beads (  NADH/NAD + Ratio Detection NADH/NAD + was detected using NAD/NADH-Glo™ Assay kit (Promega) in accordance to the manufacturer's instructions with miner modi cation. In brief, the cells were equally seeded in the 96-wells plate. When detection, cell culture medium should be removed and replaced with 50 μl PBS. Cells should be lysed by adding additional 50 μl base solution (0.2 M NaOH) with 1% DTAB to each well.
As for NAD + detection, transfer 25 μl the indicated cell lysate to a new tube and add 25 μl 0.4M HCl and heat at 60 °C for 15 min. Incubate the sample at RT for 10 min and add 25 μl 0.5 M Trizma ® base solution (Sigma). Add 50 μl freshly prepared NAD/NADH-Glo™ Detection Reagent to the sample to determine the NAD + abundance by luciferase report in the microplate reader.
As for NADH detection, transfer 25 μl the indicated cell lysate to a new tube and heat at 60 °C for 15 min.
Incubate the sample at RT for 10 min and add 50 μl HCl-Trizma ® solution (0.4 M HCl, 0.5 M Trizma ® base). Add 50 μl freshly prepared NAD/NADH-Glo™ Detection Reagent to the sample to determine the NADH abundance by luciferase report in the microplate reader.
Finally, the relative NADH/NAD + was determined as the NADH abundance/ the NAD + abundance ratio.
As for mitochondria NADH/NAD + measurement, the mitochondria were quickly isolated according to LMF, Mitochondria and Lysosome Puri cation section and the following process was the same.

Malate and Citrate Assay
Intracellular citrate or malate was detected by Citrate or Malate Assay Kit (MAK067 and MAK057,Sigma) in accordance to the manufacturer's instructions with miner modi cation. Brie y, cells (~10 6 ) were washed with ice-cold PBS and rapidly homogenized in 100 μl citrate or malate assay buffer. Samples were centrifuged at 15,000 g for 10 min at 4 °C to remove insoluble materials. Supernatants were ltered by 10 kDa MWCO spin lter (Millipore) and then assayed using the detection buffers in the kit. The measurement was performed using microplate reader at the speci c wavelength according to the manufacturer's instructions.

Measurement of Cell Respiration
The oxygen consumption rate (OCR) was determined in cell extracts using Seahorse Bioscience XF-24 Extracellular Flux Analyzer. 10 4 cells were seeded in XF24-well cell culture microplates (Seahorse Bioscience) for 24 hr. During respirometry, wells were sequentially injected at the indicated time points with 1.0 μM oligomycin (Oligo) to assess ATP turnover required respiration; 0.5 μM carbonyl cyanide p-[tri uoromethoxy]-phenyl-hydrazone (FCCP) to induce maximal respiration. Rotenone/Antimycin A was then added at a nal concentration of 1.0 μM to inhibit electron transport and non-mitochondrial basal respiration level would be detected.

Colony Formation Assay
Equal numbers of cells were seeded in 6-well plate at a density of 500 cells each well. The cells were cultured for 10-15 days with 5% FBS added DMEM in cell incubation. Then x cells using formalin and stain in crystal violet. The colonies were counted and the relative colony number was collected.

Xenograft Mouse Model
All animal experiments were performed according to the protocol approved by the Institutional Animal Care. Prepared tumor cells in 30 µl sterile PBS were injected separately into the ank of ve to six weeks old female nude mice, using the 100 μl sterile syringe. The tumor size was measured every two days using a caliper, and tumor volume was calculated using the standard formula: 0.54×L×W 2 , where the L referred to the longest diameter and the W referred to the shortest diameter. Mice were euthanized when they met the institutional euthanasia criteria for the tumor size and overall health condition. The solid tumors were removed, photographed and weighed.

Immunohistochemistry (IHC) Staining
The para n embedded tissues were depara nized in xylene followed by rehydration in a standard alcohol series, followed by antigen retrieval by 100 °C heating for 15 min in citrate buffer. The indicated primary antibody were diluted in 3% BSA and dropped to the tissue slides and incubated at 4 °C overnight. Simply wash the slides using PBS and incubate with 3% BSA diluted anti-rabbit or mouse HRPsecondary antibody for 60 min at RT. The slides were dehydrated in 50%, 70%, 80%, 95% and 100% ethanol, and stabilized with mounting medium. The images were acquired using Olympus BX43 microscope with Olympus cellSens Dimension software. The quanti cation of IHC staining density was measured using ImageJ software and calculated on the basis of the average staining intensity and the percentage of positively stained cells. A total score of protein abundance was calculated from both the percentage of positive cells and the intensity. High and low protein abundance was determined basing on the mean score of all samples as a cutoff line. Survival curves were plotted using the Kaplan-Meier method and compared by log-rank test.

QUANTIFICATION AND STATISTICAL ANALYSIS
The experiment was set up to use 3-5 samples/repeats per experiment/group/condition to detect a 2-fold difference with power of 80% and at the signi cance level of 0.05 by a two-sided test for signi cant studies. For immunohistochemical staining and immuno-blot, the representative images were shown. Each of these experiments was independently repeated for over 3 times. Relative quantities of gene expression level were normalized to B2M or GAPDH. Results were reported as mean ± Standard Deviation (S.D.) of at least three independent experiments. Comparisons were performed using two tailed paired Student's t-test (*P < 0.05, **P < 0.01, ***P < 0.001), as indicated in individual gures. For survival analysis, the expression of indicated genes was tested as a binary variant and divided into 'high' and 'low' groups. Kaplan-Meier survival curves were compared using the Gehan-Breslow test with Prism Software (GraphPad, La Jolla, CA). The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

DATA AND SOFTWARE AVAILABILITY
All the sequencing data in this study have been deposited in the NCBI with accession number (BioProject: PRJNA594757). No data with mandated deposition. Source data for supporting the ndings of this study are provided in the paper, and/or available from the corresponding author on reasonable request. Figure 1 The landscape of subcellular lncRNAs is established and quali ed. (a) Experimental scheme for the establishment of the speci c organelles' isolation process. The production could be used for RNA or protein detection. (b) Immunoblot (IB) detection of the puri ed organelles from HEK293T by the indicated protein markers to con rm the e ciency of the established organelles isolation method. Lamp2 for lysosome, CS and Tom20 for mitochondria, Calnexin for ER, GAPDH for cytosol and Lamin B1 for nucleus. (c) RT-qPCR detection of the puri ed organelles (mitochondria in the left panel, lysosome in middle panel and ER in the right panel) from HEK293T by the indicated RNA markers to con rm the e ciency of the established organelles isolation method. U6 was used as nucleus marker, GAPDH was used as cytosol marker, LAMP2 was used as lysosome marker, COX2 and ATP8 were used as mitochondria marker, secreted protein genes FGF2 and TJP1 were used as ER marker. The relative enrichment of each RNAs was calculated by 2^-(CtMito-CtTotal) followed by normalizing all ratio value to the GAPDH in control group (the rst GAPDH column). Cut-off line was thereby determined as 1. (d) Heatmap of enriched genes (the ratio of Total, fold change > 1.5, P < 0.05) in four groups. Each group contained two replicates (n = 2) and their mean value was used for heatmap.  show the peak overlapping (Right panel). GAS5 probe was labeled with Cy3 and mitochondria marker Tom20 was stained by Alex uo488. Scale bar, 20 μm. (c) The percentage of mitochondria-colocalized GAS5 per HEK293T cell under glucose su ciency and de ciency was revealed by Manders' Colocalization Coe cients (MCC) between GAS5 (Cy3 red) and Tom20 (Alex uo488 green), corresponding to GAS5-Tom20 uorescence images (a-b). MCC was calculated as the light intensity of Tom20-colocalized GAS5 divided by the light intensity of total GAS5. (d) RT-qPCR detection of GAS5, U6
(k) Biotin-GAS5 was in vitro transcribed and used for HEK293T cell lysate RNA-pulldown assay, followed by mass spectrum analysis to explore the GAS5-binding proteins. The representative candidates were listed and the mitochondrial protein MDH2 was identi ed. (l) Endogenous RNA immunoprecipitation (RIP) assay was performed using IgG and MDH2 antibody in HEK293T cells, under 8-hour glucose starvation or not (mean ± S.D., n = 3 biological replicates, Student's t-test, **P < 0.01). (m) In vitro transcribed biotinylated GAS5 sense (Sen.) or antisense (A.S.) transcripts were incubated with HEK293T lysate for RNA pulldown assay, followed by IB detection using the indicated antibodies. The input of biotin-RNAs was detected by dot blot using streptavidin-HRP. (n) IB con rmation of the GAS5-binding protein by in followed by normalizing all ratio value to the GAPDH in control group (the rst GAPDH column of each rescued group was normalized as 1, separately). Cut-off line was thereby determined as 1. (q) RNA FISH detection of GAS5 in GAS5 and its mutants (GAS5-D1, D2, D3) rescued HEK293T GAS5-KO cells.