Long noncoding RNA–mediated activation of PROTOR1/PRR5-AKT signaling shunt downstream of PI3K in triple-negative breast cancer

Significance

The phosphoinositide 3-kinase (PI3K)–AKT oncogenic pathway is hyperactivated in triple-negative breast cancers (TNBCs) and represents a prime therapeutic target in this highly aggressive breast tumor subtype. Unfortunately, PI3K pathway inhibitors have not shown overt clinical utility in TNBC patients due in part to incompletely understood compensatory cascades that sustain downstream PI3K activities. Here, we describe a previously unappreciated long noncoding RNA (lncRNA)–led promalignant pathway that is itself a downstream target of PI3K-AKT and that additionally functions as a positive feedback shunt that fosters AKT activation independent of PI3K. We demonstrate this lncRNA’s critical role as a determinant of TNBC growth and highlight its potential as an attractive therapeutic target in TNBC management.

Abstract

The phosphoinositide 3-kinase (PI3K) pathway represents the most hyperactivated oncogenic pathway in triple-negative breast cancer (TNBC), a highly aggressive tumor subtype encompassing ∼15% of breast cancers and which possesses no targeted therapeutics. Despite critical contributions of its signaling arms to disease pathogenesis, PI3K pathway inhibitors have not achieved expected clinical responses in TNBC, owing largely to a still-incomplete understanding of the compensatory cascades that operate downstream of PI3K. Here, we investigated the contributions of long noncoding RNAs (lncRNAs) to PI3K activities in clinical and experimental TNBC and discovered a prominent role for LINC01133 as a PI3K-AKT signaling effector. We found that LINC01133 exerted protumorigenic roles in TNBC and that it governed a previously undescribed mTOR Complex 2 (mTORC2)–dependent pathway that activated AKT in a PI3K-independent manner. Mechanistically, LINC01133 induced the expression of the mTORC2 component PROTOR1/PRR5 by competitively coupling away its negative messenger RNA (mRNA) regulator, the heterogeneous nuclear ribonucleoprotein A2/B1 (hnRNPA2B1). PROTOR1/PRR5 in turn was sufficient and necessary for LINC01133-triggered functions, casting previously unappreciated roles for this Rictor-binding protein in cellular signaling and growth. Notably, LINC01133 antagonism undermined cellular growth, and we show that the LINC01133-PROTOR1/PRR5 pathway was tightly associated with TNBC poor patient survival. Altogether, our findings uncovered a lncRNA-driven signaling shunt that acts as a critical determinant of malignancy downstream of the PI3K pathway and as a potential RNA therapeutic target in clinical TNBC management.

Hyperactivation of the phosphoinositide 3-kinase (PI3K) pathway is one of the most prevalent signaling features underlying human breast cancer pathogenesis (1), assuming critical roles in regulating several aspects of tumor cell biology, including metabolism, growth, motility, and survival (2). Its aberrant activation occurs via multiple different mechanisms, which consist predominantly of mutational activation of the p110α catalytic subunit; inactivation of PI3K negative regulators, such as phosphatase and tensin homolog (PTEN) or inositol polyphosphate-4-phosphatase type II (INPP4B); and/or as a result of overstimulation of upstream receptor tyrosine kinases (3). Among the different subtypes of breast cancer, PI3K signaling is most pronounced in the estrogen-receptor (ER), progesterone-receptor (PR), and epidermal growth factor receptor 2 (HER2) triple-negative breast cancers (TNBCs) (1), rendering it a particularly attractive therapeutic target in these tumor subtypes. However, PI3K inhibitors have not shown overt utility in clinical TNBCs (4) due in significant part to incompletely defined compensatory pathways that sustain signaling downstream of PI3K (5). While much research has focused on protein-led signaling networks triggered by PI3K (6), little is known as to whether regulatory RNAs, such as those belonging to the long noncoding RNA (lncRNA) family, participate in such cascades. Such findings would broaden our understanding of PI3K activities and provide avenues for actionable therapeutic interventions.

lncRNAs represent a family of non-protein-coding transcripts with a primary sequence of >200 nucleotides (7). Current GENCODE builds (version 41) describe ∼19,000 lncRNA members that occupy about 30% of the human genome. Initially dismissed as the offspring of junk DNA, lncRNAs have emerged as credible and biologically relevant regulators of a variety of cellular processes that include gene expression, protein modification, chromatin dynamics, and nuclear body assembly (7). As such, they exert important influences over biological and physiological processes in health and disease (8). Proximal roles for lncRNAs in PI3K signaling in TNBCs, however, remain incompletely characterized.

In the present work, we conducted multipronged analyses of clinical, patient-derived xenograft (PDX), and experimental TNBC tissues, and identified the lncRNA LINC01133 as a prime downstream effector of the PI3K-AKT signaling pathway. We found that LINC01133 potently promoted cancer cell growth in multiple in vitro and in vivo TNBC models and that it governed a molecular pathway that in turn fostered AKT activation in a PI3K-independent manner. Mechanistically, we determined that LINC01133 induced the expression of PROTOR1—a member of the mTORC2 complex also known as proline-rich 5 (PRR5)—by competitively binding to its negative regulator, the heterogeneous nuclear ribonucleoprotein A2/B1 (hnRNPA2B1). We show that PRR5 was both sufficient and necessary to activate AKT downstream of LINC01133 and that it mediated a signaling shunt that sustained proliferative signals in TNBC. Furthermore, we report that LINC01133 antagonism inhibited TNBC cell growth and that the LINC01133-led pathway was associated with poor TNBC patient outcome. Together, our results highlight critical contributions of a newly identified RNA-led molecular pathway downstream of PI3K and underscore its role as a potential therapeutic target in TNBC management.

Results

LINC01133 Is a Signaling Effector of the PI3K-AKT Pathway.

To identify lncRNAs that are tightly associated with PI3K-AKT activation in the context of TNBCs, we adopted three concerted approaches. First, we selected clinical TNBC specimens profiled in TCGA (n = 103) and extracted their reverse phase protein array (RPPA)–determined phospho-activated AKT^{Ser473/Thr308} levels. These specimens were then split into two groups, as follows: a “high AKT activation” group (pAKT^high), which occupied the top 20% of samples, and a “low AKT activation” group (pAKT^low) that encompassed the bottom 80% of specimens. We then conducted Xena-based differential lncRNA expression analyses between the pAKT^high and pAKT^low specimens, identifying 225 lncRNAs that were enriched (fold change [FC] of ≥1.5 and P < 0.05) in the pAKT^high tumors (Fig. 1A). Second, as PTEN loss is a frequent lesion in TNBC (1), we conducted gene expression analyses on PTEN^low versus PTEN^high TNBC PDXs profiled in GSE142767 (9) (n = 20) and identified 308 lncRNAs that were expressed at a FC of ≥1.5 and P < 0.05 in the PTEN^low when compared to PTEN^high tissues (Fig. 1A). Third, we analyzed microarray-derived (GSE93422) (10) lncRNA deregulations instigated by oncogenic mutant PIK3CA^H1047R when compared with wild-type (WT) PIK3CA in MCF10A cells, a model widely adopted in TNBC research, identifying 138 lncRNAs that were triggered by PI3K activation with a score of ≥1.5 (Fig. 1A).

Fig. 1.

LINC01133 is a PI3K-regulated TNBC lncRNA. (A) Screening flowchart for PI3K pathway–regulated lncRNAs conducted on pAKT^high versus pAKT^low TNBC specimens in TCGA, PTEN^low versus PTEN^high PDX specimens in GSE142767, and on MCF10A cells expressing PIK3CA^H1047R versus controls expressing PIK3CA^WT. (B) qRT-PCR measurements of LINC01133 in SUM159 cells treated with buparlisib (1 μM) or ipatasertib (2 μM) for 48 h (n = 3). (C) qRT-PCR measurements of LINC01133 in SUM159 cells (n = 3) treated with alpelisib or TGX221 (1 μM each) for 48 h. (D and E) qRT-PCR measurements of LINC01133 in MCF10A cells overexpressing PIK3CA^H1047R versus PIK3CA^WT (CTRL) (n = 2) (D) or MCF10A cells expressing one knock-in allele of PIK3CA^H1047R (^+/−) (n = 2) (E). All qRT-PCRs were normalized to 18S. Data are mean ± SEM (B–E). P values (**P < 0.01, ****P < 0.0001) are by one-way ANOVA for multiple comparisons (B and C), and by two-tailed unpaired Student's t test (D and E).

While several lncRNAs were found to be commonly induced in any two of the three approaches, only one lncRNA, LINC01133, was up-regulated in all three (Fig. 1A), firmly suggesting it laid downstream of PI3K signaling. In support of this notion, we found that inhibition of PI3K by the pan-PI3K inhibitor buparlisib (BKM120) or by the pan-AKT inhibitor ipatasertib (GDC0068) indeed suppressed LINC01133 expression in human SUM159 TNBC cells (Fig. 1B). We then probed if PI3K regulation of LINC01133 expression was isoform specific, focusing particularly on the PI3K p110α and p110β isoforms since PI3K isoforms p110δ and p110γ are restricted to lymphocytic cells (11). Here, we found that only the PI3K p110α-specific inhibitor alpelisib (BYL719), but not the PI3K p110β-specific inhibitor TGX221, significantly down-regulated LINC01133 expression in SUM159 cells (by ∼75%; Fig. 1C), indicating a likely preferential association between PIK3CA and LINC01133. Of interest, qRT-PCR revealed 3-fold to 10-fold inductions of LINC01133 in MCF10A cells expressing exogenous (Fig. 1D) or knock-in (Fig. 1E) oncogenic PIK3CA^H1074R, collectively indicating that LINC01133 falls under the upstream stimulatory control of the PI3K pathway.

LINC01133 Promotes TNBC Growth In Vitro and In Vivo.

We proceeded to characterize the functional contributions of LINC01133 to TNBC pathogenesis, particularly focusing on cellular growth, a key trait of PI3K signaling. First, we found that inhibition of endogenous LINC01133 expression using antisense oligonucleotides (ASOs) (SI Appendix, Fig. S1 A and B) caused reproducible reduction in the proliferation index (as measured by WST-1 staining 72 h after transient transfection with ASOs) of MDA-MB-468 and Hs578T TNBC cells in two-dimensional (2D)-growth conditions (Fig. 2A). Similarly, Hs578T cells transiently transfected with LINC01133-ASOs and plated in 3-wk spheroid assays (where cells are allowed to form one sphere/well in 96-well plates) formed ∼40% smaller growths when compared to their controls (Fig. 2B). Conversely, the stable expression of exogenous LINC01133 (SI Appendix, Fig. S1C) promoted the growth of MDA-MB-231 human TNBC cells in 2D (WST-1 at 72 h; Fig. 2C) as well as in anchorage-independent soft-agar three-dimensional (3D) conditions (Fig. 2D). Consistent effects were also observed in SUM159 cells (Fig. 2E) and in murine D2A1 (Fig. 2F) and 4T1 (Fig. 2G) TNBC cells. We also observed that LINC01133 promoted oncogenic anchorage-independent growth in immortalized human (MCF10A; Fig. 2H) and murine (NIH 3T3; SI Appendix, Fig. S2 A and B) cells, which indicated transforming activities for LINC01133 in certain models. Importantly, LINC01133 promoted the growth of orthotopic SUM159 tumors in xenografted NOD-Prkdc^em26Cd52IL2rg^em26Cd22/NjuCrl coisogenic immunodeficient NCG mice (Fig. 2I) concomitant with significant threefold increases in Ki67 positivity and fourfold suppression in apoptosis-associated cleaved-caspase-3 (CC3) in the ensuing tumors (Fig. 2 J and K). LINC01133 also accelerated 4T1 tumorigenic growths in immune competent BALB/c mice as early as day 13 after tumor implantation (SI Appendix, Fig. S2 C and D). These results highlighted the generalized growth-promoting activities for LINC01133 in TNBC development in multiple models and across species.

Fig. 2.

LINC01133 promotes promalignant phenotypes. (A) Proliferation of MDA-MB-468 (Left; n = 4) and Hs578T (Right; n = 3) cells transiently transfected with 100 to 200 nM LINC01133 ASOs or controls (NC5) measured by WST-1 (optical density [OD] at 450 nm) at 72 h. (B) Left: diagram of 3D spheroid formation assay in vitro; Middle: representative pictures of 3D spheres (1 sphere per well of a 96-well plate) of Hs578T cells after transient transfection of ASO-2MOE-NC1 and LINC01133#681 (200 nM each) for 3 wk; Right: the ratio of sphere volume in Left by ImageJ quantitation. (Scale bars, 50 µm.) (C) Proliferation of MDA-MB-231 cells (n = 3) stably expressing control (CTRL) or LINC01133 measured by WST-1 (OD at 450 nm) at the indicated time points. (D–H) Left panels: representative images of anchorage-independent growth patterns of controls and LINC01133-overexpressing MDA-MB-231 (D, n = 6), SUM159 (E, n = 6), D2A1 (F, n = 3), 4T1 (G, n = 6), and MCF10A (H, n = 6) cells. Right: ImageJ quantitation of colonies in Left. (I) Growth volume of orthotopic SUM159-CTRL (n = 5) and SUM159-LINC01133 (n = 4) tumors in NCG mice. (J and K) Left panels: Immunohistochemical determinations of Ki-67 (n = 4) (J) and Cleaved Caspase 3 (n = 4) (K) positivity in tumors in I. (Scale bars, 20 µm.) Right: ratios of positively stained cells over total cells in the counted fields estimated using ImageJ. Data are mean ± SEM. P values (*P < 0.05, **P < 0.01, ***P < 0.001) were performed by two-tailed unpaired Student's t test.

LINC01133 Activates AKT in a PI3K-Independent Manner.

We next investigated how LINC01133 fostered TNBC growth. For this purpose, we conducted RNA sequencing (RNA-seq) of LINC01133-overexpressing 4T1 cells as compared to controls, which revealed 1,630 transcripts that were up-regulated by LINC01133 with a FC of ≥2 and P < 0.001 (GSE212004). Database for Annotation, Visualization and Integrated Discovery (DAVID)-based functional annotation analysis on these induced differentially expressed genes (DEGs) highlighted several pathways that were enriched by LINC01133, with the top being “pathways in cancer” and “PI3K-Akt signaling pathway” (Fig. 3A). Prime among them was the Akt signaling node, which particularly attracted our attention considering 1) that it represented one of the most prominent threads common to these pathways and 2) that LINC01133, while serving as a downstream signaling effector of the PI3K-AKT pathway (Fig. 1), may also function as an upstream regulator of PI3K-AKT too since its expression alone was capable of revving up gene expression signatures associated with PI3K-AKT activation. In fitting with these analyses, we found that LINC01133 indeed caused AKT activation in MCF10A cells, as evidenced by robust increases in phosphorylated AKT^Ser473 levels (and to a lower extent, AKT^Thr308; SI Appendix, Fig. S3A) and of the AKT downstream targets PRAS40^Thr246 and GSK-3β^Ser9 (Fig. 3B), highlighting the functional activation of the pathway. LINC01133 also induced phospho-AKT^Ser473 in 4T1 and MDA-MB-231 cells (Fig. 3C). Conversely, AKT^Ser473 phosphorylation was significantly decreased in MDA-MB-468 cells transfected with LINC01133-ASOs (Fig. 3D), together underscoring critical regulatory roles for LINC01133 upstream of AKT.

Fig. 3.

LINC01133 causes mTORC2-dependent AKT activation. (A) KEGG enrichment analysis of DEGs (FC of ≥2, P < 0.001) in 4T1-LINC01133 compared with 4T1 control (CTRL) cells based on DAVID analysis. (B) Representative Western blotting (n > 3) of readouts of AKT activation in MCF10A cells stably harboring control vector (−) or exogenous LINC01133 (+). β-actin was used as a loading control. (C) Representative Western blotting (n = 3) of pAKT^Ser473 in MDA-MB-231 and 4T1 cells stably expressing LINC01133 (+) or controls (−). (D) Representative Western blotting (n = 3) of pAKT^Ser473 in MDA-MB-468 cells transfected with LINC01133 ASO#681 (100 nM) or negative control (NC5; 100 nM). (E) Representative Western blotting (n > 3) of pAKT^Ser473 in MCF10A cells stably expressing LINC01133 (+) or vector control (−) treated with alpelisib (1 µM) for 1 h. (F) Representative Western blotting (n = 3) of pAKT^Ser473 in stable LINC01133-expressing MCF10A cells or their controls treated with Torin 1 (1 µM) or DMSO vehicle for 1 h. (G) Representative Western blotting (n = 3) of pAKT^Ser473 in MEF-WT and MEF-Sin1^−/− cells after transient transfection with LINC01133 or control (CTRL) expression plasmids.

We asked how LINC01133 activated AKT, focusing specifically on the mechanisms leading to phospho-AKT^Ser473 induction. Here, we found that the culture of control cells in conditioned media derived from LINC01133-overexpressing counterparts did not induce phopho-AKT^Ser473 in recipient cells (SI Appendix, Fig. S3B), excluding the possibility that secreted factors induced by LINC01133 caused AKT activation in an autocrine/paracrine fashion. Moreover, activation of AKT did not appear to be caused by LINC01133 inhibiting the steady state levels of negative regulators of the pathway, such as PTEN or INPP4B (SI Appendix, Fig. S3 C and D), nor by enhancing the expression of positive pathway regulators, such as p110α, p110β, or p85 (SI Appendix, Fig. S3E). LINC01133-induced activation of AKT still occurred in the presence of BYL719 (Fig. 3E) and did not appear to involve the up-regulation of SKP2 (SI Appendix, Fig. S3F), for example, a protein previously shown to represent a major mechanism for PI3K-independent AKT activation (12). We did observe, however, that activation of AKT by LINC01133 was sensitive to the selective mTOR inhibitor Torin1 (Fig. 3F) and that it could not occur in cells in which the mTORC2 components Sin1 (Fig. 3G) or Rictor (SI Appendix, Fig. S3G) were down-regulated. These results indicated that LINC01133 activation of AKT occurred in a PI3K-independent and mTORC2-dependent manner. Of importance, Torin1 also inhibited LINC01133-promoted 3D-spehroid growth in vitro (SI Appendix, Fig. S3H), which was consistent with a prominent contribution of mTORC2 to LINC01133 oncogenic functions.

LINC01133 Activates AKT via PRR5/PROTOR1 Induction.

Considering its nuclear localization, determined by qRT-PCR (Fig. 4A), fluorescence in situ hybridization (FISH) (Fig. 4B), and RNAscope (Fig. 4C), we reasoned that LINC01133 primarily regulated gene expression—a common feature of nuclear lncRNAs (7)—and hypothesized that it activated AKT by regulating the gene expression of certain components of the mTORC2 complex. For this reason, we probed the messenger RNA (mRNA) levels of mLST8, RICTOR, DEPTOR, mTOR, SIN1, PRR5 (PROTOR1), and PRR5L (PROTOR2) by qRT-PCR. Interestingly, only the levels of PRR5 mRNA were increased (∼doubled) in LINC01133-overexpressing MDA-MB-231 cells (Fig. 4D), an observation that reproduced in MCF10A and SUM159 cells (Fig. 4E), and in sequenced 4T1-LINC01133 cells as well, which showed induction of PRR5 (Arhgap8) by >20-fold (P = 1.18E-20; see accompanying GSE212004). In addition, LINC01133 ASOs inhibited endogenous basal PRR5 expression in MDA-MB-468 and Hs578T cells (Fig. 4 F and G), further corroborating the positive genetic link between LINC01133 and PRR5 across multiple TNBC models.

Fig. 4.

LINC01133 is nuclear and promotes the expression of PROTOR1/PRR5. (A) qRT-PCR measurements of LINC01133 in cytoplasmic, insoluble nuclear, and soluble nuclear fractions in MDA-MB-231 cells. GAPDH was used as the cytoplasmic control, and XIST represented the control for the insoluble nuclear fraction (n = 2). (B) Localization of LINC01133 in MDA-MB-231-LINC01133 and MCF10A-LINC01133 cells (n = 3) detected by RNA-FISH. Nuclei were stained with DAPI, and GAPDH FISH served as a positive control for cytoplasmic staining. Scale bars, 10 µm. (C) Localization of LINC01133 in MDA-MB-231-LINC01133 and MCF10A-LINC01133 cells (n = 3) detected by RNAscope. Hs-PPIB and dapB represented positive and negative controls, respectively. (Scale bars, 20 µm.) (D) qRT-PCR measurements of mTORC2 components in MDA-MB-231 cells stably expressing control vector (CTRL) or LINC01133 (n = 4). (E) qRT-PCR measurements of PRR5 in MCF10A and SUM159 cells (n = 3 each) stably expressing LINC01133 or control vector (CTRL). (F and G) qRT-PCR measurements of PRR5 levels in MDA-MB-468 (F) and Hs578T (G) cells (n = 3) transfected with 100–200 nM LINC01133-ASOs or controls (ASO-NC1). (H) Representative Western blotting (n = 3) of AKT activation in MDA-MB-231 cells stably expressing control vector (CTRL) or LINC01133, transfected with esiRNA against PRR5 or Renilla luciferase (RLUC) (300 ng/mL each). GAPDH was used as a loading control. (I) Representative Western blotting (n = 3) of AKT activation in MCF10A and SUM159 cells stably expressing vector (−) or PRR5 (+). (J) Representative Western blotting (n = 3) of pAKT^Ser473 in MEF-WT and MEF-Sin1^−/− cells after transient transfection with PRR5 or control (CTRL) expression plasmids. (K) Growth of MDA-MB-231 cells stably harboring vector control or LINC01133 expression vector following transient transfection of esiRNA PRR5 or control esiRLUC (both at 300 ng/mL) measured at indicated time points (n = 4). (L) Cells in K were cultured in 3D for 7 d, and the growth area (as fold) was calculated by ImageJ (n = 4). (M and N) Representative images and quantitation ratio of anchorage-independent growth patterns of MCF10A (M, n = 4) and SUM159 (N, n = 4) cells stably expressing vector control (CTRL) or PRR5. (O and P) Growth volume (O) and weight (P) of orthotopic SUM159-CTRL (n = 5) and SUM159-PRR5 (n = 7) tumors in Nu/Nu mice. Data are mean ± SEM. P values (*P < 0.05, **P < 0.01, ***P < 0.001) determined by two-tailed unpaired Student’s t test.

We next assessed the role of PRR5 downstream of LINC01133. To this end, we inhibited its expression using pooled small interfering RNAs (esiRNAs) and found that esiPRR5 on its own inhibited basal phospho-AKT^Ser473 and that LINC01133 was effectively incapable of phospho-activating AKT^Ser473 when PRR5 was down-regulated (Fig. 4H and SI Appendix, Fig. S4). In addition, we found that exogenous human PRR5 was sufficient, on its own, in promoting AKT Ser473-phosphorylation in MCF10A and SUM159 cells (Fig. 4I), as well as in mouse embryonic fibroblasts (MEFs) (Fig. 4J), an activity that we propose depended on mTORC2 since it did not occur in Sin1^−/− cells (Fig. 4J). Of note, neither LINC0133 nor PRR5 promoted mTORC2-phosphorylation of SGK1 or PKCα at Ser422 and Ser657, respectively (SI Appendix, Fig. S5 A and B), suggesting the partiality of the presently described LINC01133-PRR5 pathway to AKT versus other mTORC2 substrates. Furthermore, esiPRR5 abrogated LINC01133-induced cell growth in both 2D and 3D conditions (Fig. 4 K and L), indicating the functional essentiality of PRR5 downstream of LINC01133 in TNBC cells. Finally, we observed that human PRR5 significantly promoted the growth of MCF10A and SUM159 cells in anchorage independence (Fig. 4 M and N) and that orthotopic tumors derived from PRR5-overexpressing SUM159 cells grew with faster kinetics in Nu/Nu female mice than their controls (Fig. 4 O and P). Combined, these results indicated that PRR5 was both necessary and sufficient for AKT activation and cellular growth downstream of LINC01133.

LINC01133 Binds to hnRNPA2B1.

To identify proximal mechanisms underlying LINC01133-regulated PRR5 expression, we conducted unbiased RNA pull-down assays of LINC01133 followed by mass spectrometry (MS) and did so using two approaches. First, we fused LINC01133 with 4 tandem repeats of the S1m adaptor sequence (4×S1m), a motif that is recognized by streptavidin and that can be used to pull down exogenous S1m-tagged LINC01133 using streptavidin-conjugated beads (13, 14). We then incubated total lysates of MDA-MB-231 cells expressing 4×S1m-LINC01133 with streptavidin beads and conducted MS on the proteins precipitated in the pull-downs (SI Appendix, Fig. S6A). These analyses identified 35 proteins potentially associated with LINC01133 (SI Appendix, Fig. S6A and Table S1). To complement this approach, we adopted a second, parallel scheme, in which we synthesized LINC01133 (and its antisense as a control) in vitro in the presence of biotin-labeled uracil. These synthetic RNAs were then incubated with nuclear lysates of MDA-MB-231, which was followed by RNA pull down using streptavidin-conjugated beads and by MS on the protein eluates. Eight proteins differentially precipitated with sense, but not antisense, LINC01133 (SI Appendix, Fig. S6B and Table S1). When cross-compared to one another, these two methods led to the identification of the hnRNPA2B1 as the only commonality between these two screens (Fig. 5A and SI Appendix, Table S1). Here, RNA pull down of overexpressed LINC01133 followed by Western blotting analysis for hnRNPA2B1 indeed demonstrated an association between the two (Fig. 5B). Similarly, RNA immunoprecipitates of endogenous hnRNPA2B1 from MDA-MB-231 cells revealed ∼threefold enrichment in endogenous LINC01133 when compared to IgG controls (Fig. 5C). Interestingly, sequence analyses of LINC01133 revealed two recognition pentamer motifs for hnRNPA2B1 (G/AGGGG and GGGUA; Fig. 5D), and truncation mutants lacking these motifs were in fact unable to bind hnRNPA2B1 in RNA pull-down experiments both in MCF10A and in MDA-MB-231 cells (Fig. 5E). Interestingly, such mutants were equally deficient in promoting AKT activation (Fig. 5F), consistent with the notion that LINC01133 coupling to hnRNPA2B1 was essential for LINC01133 activation of AKT. Structural modeling of the interaction between LINC01133 and hnRNPA2 (found structurally to be highly homologous to hnRNPB1; SI Appendix, Fig. S7) revealed that the predicted folding of LINC01133 (segment 751 to 1,154 nucleotides [nt] in particular) can dock into the RNA:protein binding cleft of hnRNPA2 (Fig. 5G and Movie S1), indirectly confirming our biochemical observations by showing tight docking of the expected protein–lncRNA domains for hnRNPA2 and LINC01133, respectively.

Fig. 5.

LINC01133 binds hnRNPA2B1. (A) Schematic diagram of LINC01133 pull-down strategy. Left: LINC01133 was pulled down from whole-cell lysates of MDA-MB-231 cells overexpressing 4xS1m-LINC01133 using Streptavidin beads. Right: In vitro–synthesized Biotin-U–labeled LINC01133 was incubated with nuclear lysates of MDA-MB-231 cells and then pulled down by Streptavidin beads. Pull-downs were then processed for MS analyses. (B) Representative Western blotting (n = 2) of hnRNPA2B1 in RNA pull downs of sense versus antisense LINC01133 in MDA-MB-231 cells. (C) Top: qRT-PCR measurements of LINC01133 (n = 5) after RIP of endogenous hnRNPA2B1 from MDA-MB-231 cells; Bottom: representative Western blotting of hnRNPA2B1 immunoprecipitation in MDA-MB-231 cells. Data are mean ± SEM. (D) Top: Logo plot highlighting the percentage enrichment of aligned hnRNPA2B1 recognition motifs (5mers) (https://www.encodeproject.org/experiments/ENCSR890PDQ/); Bottom: Schema of hnRNPA2B1 recognition motifs in the LINC01133 RNA sequence in ENCODE. (E) Representative Western blotting (n = 2 each) of hnRNPA2B1 in RNA pull-downs (immunoprecipitation [IP]) of indicated biotin-labeled LINC01133 variants incubated with MCF10A and MDA-MB-231 cell lysates. (F) Representative Western blotting (n = 2) of pAKT^Ser473 in MCF10A cells stably expressing the indicated truncation mutants of LINC01133. (G) Docking of the predicted structure of LINC01133 into the RNA binding pocket of hnRNPA2 using PyMol. hnRNPA2 is shown in surface mode, while LINC01133 is shown in wire-frame. The different regions of LINC01133 are colored based on truncated fragments in F.

LINC01133 Coupling to hnRNPA2B1 Relieves the Latter’s Inhibition of PRR5 mRNA Stability.

To test the functional significance of LINC01133 association with hnRNPA2B1 in regards to PRR5, we proceeded to investigate if/how hnRNPA2B1 itself exerted any influence over PRR5 expression. We found that the inhibition of hnRNPA2B1 expression using small hairpin RNA (shRNA) (#5; SI Appendix, Fig. S8A) in fact promoted twofold to threefold increases in the basal levels of PRR5 in MCF10A cells (Fig. 6A). Along similar lines, the expression of exogenous hnRNPA2 (SI Appendix, Fig. S8B) significantly down-regulated PRR5 protein expression in the same cells (Fig. 6B). These results demonstrated that hnRNPA2B1 is a negative regulator of PRR5 expression.

Fig. 6.

LINC01133 competitively binds hnRNPA2B1 and relieves its inhibition of PROTOR1/PRR5 mRNA stability. (A) qRT-PCR measurements of PRR5 in MCF10A cells (n = 3) stably expressing control shRNA plasmid (sh-Scramble) or sh-hnRNPA2B1#5, Data are mean ± SEM. (B) Representative Western blotting (n = 3) of PRR5 in MCF10A cells stably expressing control plasmid (−) or exogenous hnRNPA2. (C) qRT-PCR measurements of PRR5 in RIP of hnRNPA2B1 from MDA-MB-231 cells (n = 5). (D) qRT-PCR measurements of PRR5 in MDA-MB-231 cells (n = 3) stably expressing control plasmid (CTRL) or hnRNPA2 following treatment with actinomycin D (5 μM) for 8 h. (E and F) qRT-PCR measurements of PRR5 levels in MCF10A (E, n = 3) and MDA-MB-231 (F, n = 4) cells stably expressing control vector (CTRL) or LINC01133 treated with actinomycin D (5 μM) for 8 h. Data are mean ± SEM (C–F). (G–I) Correlation analyses between hnRNPA2B1 (probe: 205292_at) and PRR5 (probe: 47069_at) in 2,628 breast cancer patient samples from R2 (G), and between hnRNPA2B1 and PRR5 in 1,100 breast cancers (H) and 191 basal breast cancers (I) from TIMER 2.0. (J) Representative Western blotting (n = 2) of pAKT^Ser473 in MCF10A cells stably expressing control plasmid (−) or exogenous sh-hnRNPA2B1#5. (K) pAKT^Ser473 levels in hnRNPA2B1^lowPRR5^high (median for each gene) and the hnRNPA2B1^highPRR5^low group in breast cancer RPPA data. Box-and-whisker plots represent the median (centerline) and interquartile range (IQR; box). The whiskers extend up to 1.5 times the IQR from the box to the smallest and largest points. (L) KM plot of high levels of LINC01133 and PRR5 compared to the low levels of LINC01133 and PRR5 in breast cancer and basal breast cancers derived from KM plotter. (M) Schematic diagram of LINC01133-mediated activation of AKT. P values (*P < 0.05, **P < 0.01, ***P < 0.001) were determined by two-tailed unpaired Student's t test.

How hnRNPA2B1 down-regulated PRR5, however, was not clear. Considering their functions as RNA binding proteins with established and direct regulatory influences over RNA turnover, we next probed if hnRNPA2B1 affected PRR5 mRNA stability per se. To assess this notion, we were able to reproducibly amplify detectable levels of PRR5 mRNA from hnRNPA2B1 immunoprecipitates (Fig. 6C), indicative of a possible direct association between the two. Next, we found that hnRNPA2-overexpressing MDA-MB-231 cells treated for 8 h with the transcriptional inhibitor actinomycin D exhibited 50% reduction in their PRR5 steady-state mRNA levels when compared to their controls (Fig. 6D and SI Appendix, Fig. S8C). Additionally, LINC01133 was found to conversely enhance the stability of PRR5 mRNA both in MCF10A (Fig. 6E) and in MDA-MB-231 (Fig. 6F) following similar treatments, strongly suggesting that LINC01133 binding to hnRNPA2B1 relieves the latter’s destabilization of PRR5 mRNA.

In congruence with our experimental results, we found that hnRNPA2B1 exhibited strong negative correlation with PRR5 in 2,628 breast cancer patient samples from R2 (r = −0.446; P < 0.001; Fig. 6G), as well as in 1,100 breast cancer patients (r = −0.282; P < 10⁻²⁰; Fig. 6H) and 191 basal breast cancer patients from TIMER 2.0 (r = −0.36; P < 10⁻⁶; Fig. 6I). Furthermore, shRNA against hnRNPA2B1 induced phospho-activation of AKT^Ser473 (Fig. 6J) and TCGA breast cancer specimens that possessed low levels of hnRNPA2B1 and high levels of PRR5 (hnRNPA2B1^lowPRR5^high) displayed higher levels of phospho-AKT^Ser473 (determined by RPPA) when compared to their hnRNPA2B1^highPRR5^low counterparts (Fig. 6K). Finally, we found that elevated LINC01133/PRR5 levels were associated with poor patient outcome in breast cancer patients in general and in basal TNBC in specific (Fig. 6L), indicating the functional relevance of the presently described LINC01133-PRR5 pathway in supporting advanced disease in the clinic.

Discussion

In the work presented here, we conducted integrative analyses that leveraged TCGA data, PDX profiling, and cell line models to uncover a previously undescribed role of LINC01133 as a protumorigenic signaling effector of the PI3K pathway in TNBC. Mechanistically, we found that LINC01133 competitively couples to the RNA binding protein hnRNPA2B1, relieving the latter’s suppression of PRR5/PROTOR1 mRNA stability, thereby revving up AKT activation (Fig. 6M). Unexpectedly, although LINC01133 is itself induced by PI3K, its ability to activate AKT occurs in a PI3K-independent fashion, in contrast to previous reports describing similar activities for certain lncRNAs, namely, AK023948 or LINK-A, for example, in stabilizing p85β (15) or in facilitating PIP3-mediated AKT activation (16). Thus, our study delineates previously undescribed mechanisms for signal amplification downstream of PI3K, adds a new layer of complexity to the classical protein-to-protein signal transduction cascades that promote downstream regulation of the PI3K pathway (5), and suggests a cell-autonomous genetic mechanism that sustains mitogenic signaling longevity when upstream PI3K activity is discontinued.

PRR5 (or PROTOR1) has been simultaneously identified as a RICTOR-binding protein by the groups of Alessi et al. (17) and Kim et al. (18) in 2007 (17, 18). Consequent of this association, PRR5 was postulated to play regulatory functions over/within mTORC2; however, its precise functional contributions in these contexts have remained elusive (19). Indeed, Kim and colleagues (18) did not observe any direct effects of PRR5 on mTORC2 kinase activity per se. Nevertheless, they noted that shRNAs against PRR5, while not affecting EGF or Insulin signaling, did dampen PDGF signaling through PI3K to AKT in HeLa cells, an effect they attributed to undefined shPRR5-mediated down-regulation of PDGFRβ mRNA expression. Furthermore, they found that PRR5 knockdown caused preferential down-regulation in the phosphorylation of the mTORC1 effector SGK1 at Thr389 (and 4E-BP1 at Thr37/46) over that of AKT at Ser473, raising the possibility that PRR5 may actually regulate mTORC1 rather than mTORC2 (18). Additional perplexing results came from the Alessi group (17) in 2011, who found that a PRR5 (or PRR5L) whole-body mouse knockout model exhibited no discernable effects on mTORC2 activation in several tested tissues, with the only signaling impact restricted to down-regulation in mTORC2-dependent SGK1 phosphorylation in the kidney, suggesting that PRR5 was largely dispensable, especially for AKT activation (20). In contrast to these observations, we found that the LINC01133-PRR5 pathway severely down-regulated the expression of PDGFRB (see DEG in GSE212004) and had no impact on mTORC2-mediated SGK1 (nor PKC) phosphorylation (SI Appendix, Fig. S5). Furthermore, we provide evidence that PRR5 was indeed sufficient and essential for LINC01133-mediated and mTORC2-dependent AKT^S473 phosphorylation in TNBC cells. These results collectively underscore functions for PRR5 that are likely substrate and/or tissue specific. How PRR5 promotes AKT activation in TNBC, however, is presently unclear and is the subject of ongoing in-depth biochemical and structural determinations aimed at characterizing its specific contributions to mTORC2 integrity, assembly, localization, and functions, as well as full characterization of its cellular binding partners in our models.

Our work also indicated a negative PRR5-regulating role for hnRNPA2B1, a critical RNA binding protein that has two isoforms, namely, A2 and B1, with A2 possessing 12 amino acids less than B1 at the NH2 terminus and representing >95% of the gene in cells (21). It plays multiple physiological roles, particularly through its regulation of mRNA splicing, RNA trafficking, pre-mRNA 3′ end processing, and mRNA stability and translation, to name a few (22). Here, we found that hnRNPA2B1, on its own, bound to and inhibited PRR5 expression by decreasing its mRNA stability, in contrast to LINC01133 (Fig. 6). Interestingly, previous work indicated that hnRNPA2B1 associates with target mRNAs using specific motifs, such as GGGGG or AGGGG (23), AU-rich motif (24), (U)16 motif (25), UAGGG motif (26), and GA-rich region (27). In addition to identifying examples of these motifs in the LINC01133 RNA sequence (Fig. 5), we found multiple hnRNPA2B1 binding sites in the PRR5 isoform coding DNA sequence (CDS) we cloned in this study, such as 555 to 559, 720 to 724, and 1,144 to 1,148 nt for GGGGG motif or 170 to 174, 554 to 558, 571 to 575, and 662 to 666 nt for the AGGGG motif. These findings, together with results that LINC01133 truncations that cannot bind to hnRNPA2B1 cannot stimulate AKT activation (Fig. 5) and that hnRNPA2B1 binds PRR5 mRNA (Fig. 6) allow us to propose a model in which competitive binding of LINC01133 to hnRNPA2B1 takes it away from PRR5 mRNA, hence annulling its inhibitory activities on PRR5 expression.

About 9% of TNBC patients harbor oncogenic mutations in PIK3CA with an additional ∼25% exhibiting PI3K-AKT signaling pathway–related mutations (4). However, inhibitors of PI3K have not proven to be clinically useful in this class of tumors, neither when used alone [e.g., NCT01629615 (28)] nor when used in combination with other agents, such as paclitaxel [see NCT01572727 (29)]. These outcomes are attributed, in a significant part, to the emergence of mitigating pathways that sustain downstream PI3K signaling even in the presence of its inhibitors. In these respects, our data suggest we have identified one such, lncRNA-mediated, mechanism. Indeed, the ability of LINC01133 to still activate AKT in the presence of PI3K inhibitors (Fig. 3E), that LINC01133-induced genetic up-regulation of PRR5 was sufficient to promote oncogenic growth in vitro and in vivo (Fig. 4), and that the LINC01133-PRR5 pathway tightly associated with poor TNBC patient outcome (Fig. 6) are all consistent with this postulate. These data, when put together with findings in which LINC01133 (and PRR5; Fig. 4) inhibition decelerated cellular growth (Fig. 2) further emphasize the translational relevance of LINC01133 as a potential therapeutic target—a pressing need in the clinical management of TNBC.

Materials and Methods

Constructs.

LINC01133 (human sequence NR_038849.1, 1,154 bp) was synthesized by GenScript and cloned into the pLVX-puromycin plasmid (provided by P. Pandolfi, Beth Israel Deaconess Medical Center, Boston, MA) using EcoRI and SmaI restriction sites. Human PRR5 CDS (NM_181333.4, 1,167 bp) was synthesized by GENEWIZ and cloned into the pLVX plasmid using EcoRI and BamHI restriction sites. The MBP-hnRNPA2_FL_WT plasmid (#98662) was purchased from Addgene, and hnRNPA2 was subcloned into pLVX using EcoRI and BamHI restriction sites. The puroMXb-4×S1m (p2824) plasmid was kindly provided by G. Stoecklin (German Cancer Research Center, Germany) (13). The hnRNPA2B1 shRNAs (TRCN0000001058 [#1], TRCN0000001059 [#2], TRCN0000001061 [#3], TRCN0000010582 [#4] and TRCN0000377202 [#5]) were purchased from Millipore Sigma.

Cell Culture.

MDA-MB-231, MDA-MB-468, and Hs578T cells were purchased from American Type Culture Collection. SUM159, NIH 3T3, control and MCF10A-PIK3CA^H1047R and MCF10A-PIK3CA^{H1047R(+/−)} cells were provided by A. Toker (Beth Israel Deaconess Medical Center, Boston, MA). MEFs and MEFs-Sin1^−/− cells were provided by W. Wei (Beth Israel Deaconess Medical Center, Boston, MA). HEK293T, 4T1, and D2A1 cells were obtained from R. Weinberg (Whitehead Institute for Biomedical Research, Cambridge, MA), and MCF10A cells were a gift from J. Brugge (Harvard Medical School, Boston, MA). MCF10A cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM)/Ham’s F-12 medium supplemented with 5% horse serum (Gibco), insulin (10 μg/mL; Wisent Bioproducts), hydrocortisone (500 ng/mL; Sigma-Aldrich), EGF (20 ng/mL; R&D Systems), and cholera toxin (100 ng/mL; List Biological Laboratories). SUM159 cells were cultured in Ham’s F-12 medium with 5% fetal bovine serum (FBS), 5 μg/mL insulin, and 1 μg/mL hydrocortisone. Hs578T cells were cultured in DMEM supplemented with 10% FBS and 40 μg/mL insulin. Other cells were cultured in DMEM supplemented with 10% FBS. For viral transduction, lentiviral supernatants were collected after 48 h from HEK293T cells cotransfected with the plasmids pMD2.G (#12259, Addgene) and psPAX2 (#12260, Addgene) using polyethylenimine (PEI 25K) (Polysciences). Filtered particles were used to infect target cells in the presence of polybrene (10 μg/mL; Santa Cruz Biotechnology), and stable cell lines were selected in puromycin (2 µg/mL; Wisent Bioproducts).

qRT-PCR Determinations.

Gene expression was determined by qRT-PCR on total cellular RNA purified using the RNeasy kit (Qiagen) and quantified using NanoDrop ND-1000 (Thermo Fisher Scientific). First-strand synthesis was performed on equilibrated RNA using the Qiagen RT kit, and complementary DNA (cDNA) was amplified by SYBR green PCR master mix (Qiagen) in a CFX384 cycler (Bio-Rad). mRNA abundance was determined using the 2^−ΔΔCt method with 18S, GAPDH, or U1 used as normalization controls. Primers (for human target genes) used were as follows:

• 18S forward, 5′-GTAACCCGTTGAACCCCATT-3′;

• 18S reverse, 5′-CCATCCAATCGGTAGTAGCG-3′;

• GAPDH forward, 5′-ACAACTTTGGTATCGTGGAAGG-3′;

• GAPDH reverse, 5′-GCCATCACGCCACAGTTTC-3′;

• U1 forward, 5′-CAGGGCGAGGCTTATCCA-3′;

• U1 reverse, 5′-GCAGGGGTCAGCACATCC-3′;

• LINC01133 forward, 5′-GGCAAGGTGAACCTCAAAAA-3′;

• LINC01133 reverse, 5′-TTCCTGCAAGAGGAGAAAGC-3′;

• XIST forward, 5′-TGCTGATCATTTGGTGGTGT-3′;

• XIST reverse, 5′-TGACTTCCTCTGCCTGACCT-3′;

• PRR5 forward, 5′-CCTTCACCCATTCCTGCATCC-3′;

• PRR5 reverse, 5′-AGAGGCGTGTTGTAGCTCTTG-3′;

• mLST8 forward, 5′-TGTGGGCTTCCACGAAGAC-3′;

• mLST8 reverse, 5′-AGTTAATGGGTGCGTTCACCT-3′;

• RICTOR forward, 5′-TCCAAAGACTCGACAGTATGTGC-3′;

• RICTOR reverse, 5′-GGCTAGAAATCGTGCTTCTCT-3′;

• DEPTOR forward, 5′-GCGGAGGCGAAGACTGATG-3′;

• DEPTOR reverse, 5′-GGCTCACTGACATAAAGCTGGTA-3′;

• mTOR forward, 5′-TCCGAGAGATGAGTCAAGAGG-3′;

• mTOR reverse, 5′-CACCTTCCACTCCTATGAGGC-3′;

• MAPKAP1 (SIN1) forward, 5′-GGTGGACACCGATTTCCCC-3′;

• MAPKAP1 (SIN1) reverse, 5′-CGCTTCACTGCCTTCAGTAAGA-3′;

• PRR5L forward, 5′-CGGCTGTTGAAGAGTGAACTT-3′;

• PRR5L reverse, 5′-GCAGGGTAGGGAGAGTCTCAG-3′;

• HNRNPA2B1 forward, 5′-ATTGATGGGAGAGTAGTTGAGCC-3′;

• HNRNPA2B1 reverse, 5′-AATTCCGCCAACAAACAGCTT-3′;

• HNRNPA2 forward, 5′-AGCAGTGGGGTAAACTGACG-3′;

• HNRNPA2 reverse, 5′-TGGAGAAGGTGACGAAACCG-3′.

Cellular Growth Assays.

For proliferation assays, a total of 5 × 10³ cells were seeded per well in a 96-well plate, and growth was measured using the WST-1 proliferation kit (Millipore Sigma) according to the manufacturer’s instructions. Alternatively, a total of 3 × 10⁴ cells per well were plated per well in 12-well plate, and growth was measured by counting the cell numbers at indicated time points using the Trypan blue exclusion assay. For anchorage-independent growth, suspended cells were mixed with equal volume of 0.35% agar and seeded into 6-well plates precoated with 0.625% agar at a density of 5 × 10³ cells per well. Colonies were visualized ∼2 to 6 wk later using microscopy, stained with 0.002% crystal violet, imaged, and colony numbers were counted with Image J software (NIH Image).

ASO and esiRNA-Mediated Target Inhibition.

LINC01133 ASOs consisting of the following sequences were synthesized by Integrated DNA Technologies (standard desalting) : DNA-681, 5′-/C*T*T*C*C*A*C*T*T*C*A*G*C*A*C*A*C*C*T*T/-3′;

• NC5 PS G*C*G*A*C*T*A*T*A*C*G*C*G*C*A*A*T*A*T*G/-3′; DNA-681-2MOE: 5′-/52MOErC/*/i2MOErT/*/i2MOErT/*/i2MOErC/*/i2MOErC/*A*C*T*T*C*A*G*C*A*C

• /i2MOErA/*/i2MOErC/*/i2MOErC/*/i2MOErT/*/i2MOErT/-3′; DNA-NC1-2MOE: 5′-/52MOErG/*/i2MOErT/*/i2MOErT/*/i2MOErA/*/i2MOErA/*T*C*G*C*G*T*A*T*A*A

• /i2MOErT/*/i2MOErA/*/i2MOErC/*/i2MOErG/*/i2MOErC/-3′. Human RICTOR and PRR5 esiRNAs were purchased from Millipore Sigma. ASOs and esiRNAs were transfected into target cells using Lipofectamine 2000 or Lipofectamine RNAiMAX (Thermo Fisher Scientific) according to the manufacturer’s protocol.

Tumorigenesis Assays.

Animal experiments were performed according to Institutional Animal Care and Use Committee–approved procedures in Beth Israel Deaconess Medical Center (BIDMC) (protocol number: 061-2016). For orthotopic injections, 0.5 × 10⁵ 4T1 cells or 2.5 × 10⁵ SUM159 cells were injected into the mammalian fat pad of female BALB/c mice or female NCG mice (NOD-Prkdc^em26Cd52IL2rg^em26Cd22/NjuCrl Coisogenic Immunodeficient) (6 wk old), or 5 × 10⁵ SUM159 cells were injected into the mammalian fat pad of female Nu/Nu mice (7 wk old) (Charles River). For subcutaneous injections, 2.5 × 10⁵ 4T1 cells were injected into female BALB/c mice (6 wk old) (Charles River). Tumor volume was estimated using digital calipers and calculated according to the equation Volume = (W² × L)/2. Mice were euthanized at 2 to 7 wk postinjections, and tumors were excised and processed for weight measurements and/or for immunohistochemical determinations.

Immunohistochemistry (IHC).

IHC was performed using standard techniques. Estimation of Ki-67 (ab92742; Abcam) or Cleaved Caspase-3 (#9664, CST) positivity was conducted in a blinded fashion under microscopy in at least five random fields per slide at 200× magnification.

Computational Analyses.

To identify the pAKT^high and pAKT^low specimens in TCGA, TCGA-BRCA HTSeq-FPKM (n = 1,217) data were downloaded from XENA (https://xenabrowser.net/datapages/), and the RPPA-TCGA-BRCA-L4 (n = 901) data were downloaded from TCPA (https://tcpaportal.org/tcpa/download.html). There were 103 TNBC samples from RPPA-TCGA-BRCA that matched between the TCGA-BRCA-HTSeq-FPKM and RPPA-TCGA-BRCA-TNBC data. lncRNA identities (n = 5,079) were download from HUGO Gene Nomenclature Committee (https://www.genenames.org/download/statistics-and-files). For PTEN analyses in PDXs, data were downloaded from GSE142767, and samples were divided into PTEN^high and PTEN^low groups based on the median of PTEN value, with genes excluded if not detected in at least half the samples. For MCF10A analyses, differential gene expression was determined using RNA-seq among cells overexpressing exogenous PIK3CA^H1047R versus those expressing PIK3CA-WT from GSE93422. A score of ≥1.5 was used as a cutoff.

The hnRNPA2 and hnRNPB1 protein structures were predicted using AlphaFold v2.1.1 (30). The hnRNPA2 (NP_002128.1), hnRNPB1 (NP_112533.1), and LINC01133 (NR_038849.1) sequences were obtained from the National Center for Biotechnology Information. The predicted folding of LINC01133 was performed using the 3dRNA v2.0 web server (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6747482/). Using validated hnRNPA2 representations updated to Protein Data Bank, we were able to determine the structural location of the RNA binding pocket of hnRNPA2. Using this information, we manually docked LINC01133 to hnRNPA2 using PyMol 2.5.0.

Gene correlation analysis of PRR5 and hnRNPA2B1 was performed using the R2 Genomics Analysis and Visualization Platform (https://hgserver1.amc.nl/cgi-bin/r2/main.cgi?open_page=login) and TIMER2.0 (http://timer.cistrome.org/). For a Kaplan–Meier (KM) plot, we utilized Kaplan–Meier Plotter analyses (https://kmplot.com/analysis/) and used the mean expression for LINC01133 (probe: 239370_at) and PRR5 (probe: 47069_at).

Western Blot Analyses.

Western blot analysis was performed using standard techniques as previously described (31) with antibodies recognizing β-Actin (#4970, Cell Signaling Technology [CST]), GAPDH (#2118, CST), Vinculin (#V9131, Millipore Sigma), AKT (#2920, CST), phospho-AKT^Ser473 (#4060, CST), phospho-AKT^Thr308 (#4056, CST), phospho-PRAS40^Thr246 (#2997, CST), PRAS40 (#2691, CST), phospho-GSK-3β^Ser9 (#5558, CST), GSK-3β (#9315, CST), PI3K p110α (#4249, CST), PI3K p110β (#3011, CST), PI3K p85 (#4257, CST), phospho-PTEN^{Ser380/Thr382/383} (#9549, CST), PTEN (#9559, CST), INPP4B (#4039, CST), Skp2 (#2652, CST), hnRNPA2B1 (#9304, CST), PKCα (#2056, CST), and SGK1 (#12103, CST). Antibodies for phospho-PKCα^Ser657 (sc-377565) and phospho-SGK1^Ser422 (sc-16745) were purchased from Santa Cruz Biotechnology. The antibody for RICTOR (A300-459A) was purchased from Bethyl Laboratories. Antibodies for Sin1 (#05–1044) and for PRR5 (#PA5-32179) were purchased from Millipore Sigma and ThermoFisher, respectively.

RNA-Seq.

Total cellular RNA was extracted using RNeasy (Qiagen, Hilden, Germany) and assessed using NanoDrop ND-1000 (Thermo Fisher Scientific). A total of 100 ng RNA was further processed for ribosomal RNA (rRNA) removal using the Epicenter rRNA depletion kit. rRNA-depleted RNA was subsequently used to generate paired-end sequencing libraries using the Illumina RNA TruSeq Library Kit. The quantity and quality of RNA-seq libraries were analyzed by Qubit and Agilent Bioanalyzer, respectively, and the libraries were pooled at a final concentration of 12 pM and sequenced by Nextseq500. All processes were performed according to the manufacturer’s instructions.

mRNA Decay Analysis.

Cancer cells were treated with 5 μM actinomycin D (Thermo Fisher Scientific) and harvested at the indicated time points. The total RNA was isolated and validated by qRT-PCR.

Subcellular Fractionation.

Insoluble and soluble nuclear, as well as cytoplasmic, fraction preparations were performed on ice as follows. Pellets from 4.0 × 10⁶ cells were totally resuspended in 200 µL RLN1 buffer (50 mM Tris-HCl [pH 8.0], 100 nM NaCl, 1.5 nM MgCl₂, 0.1% Nonidet P-40, 2 nM ribonucleoside vanadyl complex) and then centrifuged at 2,000 rpm for 2 mins. Supernatants represented the cytoplasmic fractions. Remaining pellets were then totally resuspended in 200 µL RLN2 buffer (50 mM Tris-HCl [pH 8.0], 500 nM NaCl, 1.5 nM MgCl₂, 0.1% Nonidet P-40, 2 nM ribonucleoside vanadyl complex) and then centrifuged at 13,000 rpm for 2 mins. Supernatants from this fractionation represented the soluble nuclear fractions, while the pellets remaining, resuspended in 200 µL of RLN2 buffer, represented the nuclear insoluble fraction. All fractions were later subjected to RNA extraction by RNeasy kit (Qiagen, Hilden, Germany) as previously described and then subjected to qRT-PCR determinations on 1:10 diluted cDNA using GAPDH and XIST as controls for cytoplasmic and insoluble nuclear fractions, respectively.

RNA FISH.

RNA FISH for LINC01133 and GAPDH (control) detection was performed using a pool of 30 fluorescent probes (Quasar 570 Dye) purchased from Stellaris Biosearch Technologies, following the manufacturer’s instructions.

RNAscope.

In situ hybridization for LINC01133 was performed using RNAscope 2.5 HD detection reagent-Brown according to the manufacturer’s protocols (Advanced Cell Diagnostics). Fifteen ZZ probe pairs were used and were designed against sequences 179 to 1,139 bp of human-LINC01133 (NR_038849.1). Probes for human peptidylprolyl isomerase B (PPIB) and for bacterial dapB were used as positive and negative controls, respectively.

RNA Pull-Down Assay.

The 4×S1m sequence was cloned from the puroMXb-4×S1m (p2824) plasmid and subcloned into the pLVX or pLVX-LINC01133 plasmid by using an AfeI site (forward primer: 5′-TCTACTAGAGGATCGCTAGCAGGTGACACTATAGAACCAG-3′; reverse primer: 5′-CGAGATCTGAGTCCGGTAGCTACCGAGCTCGAATTCATCG-3′). After the transfection of these plasmids into MDA-MB-231 cells, cells were lysed in lysis buffer (150 mM KCl, 25 mM Tris-HCl [pH 7.4], 5 mM EDTA, 5 mM MgCl₂, 1% Nonidet P-40, 0.5 mM DTT, protease inhibitor, 100 U/mL RNAseOUT) 36 h later. Lysates were then cleared by adding 50 µL egg-white Avidin (A887, Thermo Fisher Scientific) and subsequently incubated with streptavidin beads (Dynabeads Streptavidin C1, Thermo Fisher Scientific) for 4 h at 4 °C under rotation. The beads were recovered and washed, bound proteins were released and then resolved by sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) electrophoresis. Gels were silver stained following manufacturer instructions (#24612, Thermo Fisher Scientific). RNA-bound proteins were sent for identification using MS (Harvard Medical School Taplin Mass Spectrometry Facility).

For biotin labeling of RNA (32), transcripts were amplified from pLVX-LINC01133 (sense-forward: 5′-TAATACGACTCACTATAGGGAGACTGATGTAACAGCCTTGGGAAAG-3′; sense-reverse: 5′-TTTTTTTTTTTTTTTTTTAAAATGAAAC-3′; antisense-forward: 5′-TAATACGACTCACTATAGGGAGATTTTTTTTTTTTTTTTTTAAAATGAAAC-3′; antisense-reverse: 5′-CTGATGTAACAGCCTTGGGAAAG-3′) and generated in vitro using the MEGAscript T7 high-yield transcription kit (AM1334, Thermo Fisher Scientific) with biotin-16-UTP. Biotin-labeled products were purified by MEGAclear transcription clean-up kit (AM1908, Thermo Fisher Scientific) according to the protocol provided by the manufacturer. Biotinylated products were subsequently heated to 65 °C for 5 mins and then allowed to slowly cool to room temperature in buffer (10 mM Tris [pH 7.5], 10 mM MgCI₂, 100 mM NH₄CI). Biotin-labeled RNAs were then incubated with precleared nuclear MDA-MB-231 lysates for 30 mins in binding buffer (100 mM Hepes [pH 7.0], 50 mM KCL, 10% glycerol, 1 mM EDTA, 1 mM DTT, 0.5% TritonX-100, 0.5 mg/mL heparin), followed by addition of streptavidin beads (Dynabeads Streptavidin C1, Thermo Fisher Scientific). After 15 min, beads were washed and boiled, and bound proteins were resolved by SDS-PAGE electrophoresis. Gels were silver stained following manufacturer instructions (#24612, Thermo Fisher Scientific). RNA-bound proteins were sent for identification using MS (BIDMC Mass Spectrometry Facility).

RNA-Binding Protein Immunoprecipitation (RIP).

RIP was performed using the EZ-Magna RIP kit (17-701, Millipore Sigma) according to the protocol provided by the manufacturer. Briefly, a total of 50 µL of protein A/G magnetic beads for each sample were first incubated with 5 μg antibody (normal mouse IgG, anti-SNRNP70, or anti-hnRNPA2B1; ab31645, Abcam) under rotation at room temperature for 30 mins. Then, 100-μL lysates from 2.0 × 10⁷ MDA-MB-231 cells were added to each mixture under rotation for 6 h at 4 °C. Samples were then treated with proteinase K for 30 mins at 55 °C, and bound RNAs were purified by phenol: chloroform: isoamyl alcohol extraction and ethanol precipitation. RNAs were quantified by qRT-PCR.

3D Spheroid Assay.

Autoclaved 50 µL of 1.5%/v agarose (A9539, Millipore Sigma) in appropriate cell culture medium (i.e., corresponding to the media in which tested cells are propagated in) were plated per well in 96-well plates and allowed to solidify at room temperature for 1 h. Two thousand cells were then seeded into each agarose-coated well of a 96-well plate in 100 µL of complete medium overnight. For drug treatment, cells were administered 5 µM of Torin1 or control DMSO at 24 h after cell plating. The dimensions of colonies were measured under microscopy, and volumes were calculated using the formula V = (W² × L)/2 using ImageJ.

Signaling Inhibitors.

Cells were starved overnight and treated for indicated periods with buparlisib (BKM120, S2247), alpelisib (BYL719, S2814), TGX221 (S1169), ipatasertib (GDC0068, S2808), or Torin 1 (S2827) obtained from Selleckchem.

Statistical Analysis.

Results were presented as means ± SEM, and significance was determined using unpaired Student’s t test after confirming normal distribution or by ANOVA with Dunnett’s test for multiple comparisons in GraphPad Prism 9. Correlation analyses were performed using bivariate Pearson correlation (SPSS: version 23). For all analyses, *, **, ***, and **** indicated P < 0.05, P < 0.01, P < 0.001, and P < 0.0001, respectively.

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix. RNA-seq data of LINC01133 overexpressing cells have been deposited in the Gene Expression Omnibus (GEO) repository under accession number GSE212004 (33).