Long non‐coding RNAs are significantly associated with prognosis and response to therapies in gastric cancer

the heterogeneity of in gastric cancer (GC) by a systems-level characterization of the long noncoding RNA (lncRNA). We identified molecular subtypes predictive of response to standard treatments and demonstrated therapeutic potential of tar-geting lncRNA for the refractory subtype.

We examined the association of LNC6 subtypes with chemo-response in a merged cohort where about half of the patients had received standard adjuvant chemotherapy (Figures 2A and 2B). 2,3 Among the major three subtypes comprising 90% of this cohort, L6B patients exhibited significant benefit from chemotherapy, while L6C and L6F patients did not. We also examined the association of LNC6 subtypes with response to pembrolizumab by directly analyzing lncRNA expression from the raw RNA sequencing data of a phase 2 clinical trial (n = 45). 4 The prediction model was constructed based on the expression data of subtype-specific lncRNAs in the TCGA cohort ( Figure S1B and Table S4). L6C probability was a positive predictor of clinical response to immune checkpoint blockade, whereas L6F probability was a negative predictor ( Figures S4A and S4B). Although the predicted probability of the L6E subtype could not stratify the responders and non-responders, gene set enrichment analysis (GSEA) of lncRNAs specifically upregulated in the L6E subtype successfully stratified them ( Figure S4C). Furthermore, we explored the relationship of LNC6 subtypes with other molecular subtypes reported in previous studies ( Figures 2C-2F). 2,3,5,6 The poor prognostic L6F subtype largely overlapped with the genomically stable, epithelialmesenchymal transition (EMT), mesenchymal phenotype, and invasive subtypes, all of which have been associated with poor clinical outcome. Enrichment of L6C with the microsatellite instability (MSI) subtype is well matched with its negative predictive value for chemotherapy. 7 In line with this, several L6F and L6C-specifc lncRNAs have previously been associated with poor clinical outcome and chemoresistance (Supplementary Discussion). Meanwhile, L6B was enriched with the chromosomal instability and proliferative subtypes, supporting association between chemosensitivity and proliferative signature. L6E subtype was 100% Epstein-Barr virus (EBV) subtype, and this is also in good agreement with favorable prognoses and response to pembrolizumab in MSI-high and EBV-positive tumors. Taken together, consideration of the lncRNA expression pattern may add predictive value for response F I G U R E 1 Prognostic association of LNC6 subtypes. (A) mRNA expression signature specific to LNC6 subtypes in the TCGA cohort (n = 258). Subtype-specific mRNA expression signatures were identified using multiple two-class t-tests (p < 0.001), yielding few hundred genes for each subtype. The top 200 mRNAs were selected for each subtype according to the log ratio. (B) Schematic diagram for prediction model. Samples in the test cohorts were assigned to one of the six subtypes according to the Bayesian probability scores. (C) Kaplan-Meier plots with global p-value from log-rank test of overall survival and recurrence-free survival of patients in the test cohorts (n = 1,933). In pairwise comparisons, the L6C demonstrated significantly shorter overall survival than all the other groups except the L6E, and the L6F showed significantly shorter recurrence-free survival than the L6B and L6C (adjusted p-value can be found in Table S2 to adjuvant chemotherapy and immune checkpoint blockade. Moreover, we carried out bioinformatics analyses and took advantage of transcriptomic data from other studies to establish the biological implications of the LNC6 subtypes. GSEA and ingenuity pathway analysis revealed biological pathways and regulators characterizing each subtype ( Figure S5; Table S5; Supplementary Discussion). In agreement with the activation of pathways associated with stemness, transcription factors overexpressed during the early stomach development were upregulated in the L6F subtype ( Figure S6). 8 Conversely, S-phase-enriched lncR-NAs were downregulated in the L6F subtype, supporting correlation between quiescence and stem-like characteristics ( Figure S7). 9 Such expression pattern of biological signatures might account for the bad clinical outcome consistently seen in the L6F subtype.
We investigated therapeutic approaches for the L6F subtype based on gene silencing experiment and pharmacogenomic analysis. First, we examined expression level of five L6F subtype-specific lncRNAs in six GC cell lines using RT-qPCR (primers listed in Table S6). Based on EMT phenotype characterization in a previous study, 10 the average expression level was higher in the EMT subtype cell lines for all five lncRNAs, among which ZNF667-AS1 achieved significance despite small sample number (Table S7). Then, we tested the functional association between ZNF667-AS1 and mesenchymal/stemlike characteristics with knockdown and overexpression studies ( Figures 3A and 3B). Knockdown of ZNF667-AS1 (ENSG00000166770.6) by siRNA in the EMT subtype cell lines decreased their migratory and invasive activity, as well as sphere formation under anchorage-independent growth ( Figure 3C). In agreement with this, silencing  (Figure 4A). Moreover, this increased sensitivity to drugs commonly used for GC treatment -oxaliplatin and 5-FU (Figure 4B). On the other hand, overexpression of ZNF667-AS1 in the non-EMT subtype cell lines induced mesenchymal marker expression and chemoresistance, while reducing their epithelial marker expression (Figures 4A and 4B). In addition, we looked into two independent pharmacogenomic datasets of GC cell lines to identify drugs that could specifically target the L6F subtype. By performing correlation analysis between drug sensitivity AUC values and L6F probability from lncRNA expression, we found three drugs that had negative correlation with significance in both datasets -YM155, PI-103, and Obatoclax ( Figure 4C). Previous studies reported the selective efficacy of these drugs against the mesenchymal and stem-like GC (Supplementary Discussion). Our study has some limitations -the retrospective nature of the clinical data and the lack of animal studies.
In conclusion, we demonstrated clinical implication of the lncRNA-based stratification and biological association of a specific lncRNA in GC. Significant association of lncRNA expression with prognosis and therapeutic responses indicate that it could be used to elaborate precision medicine for GC. Moreover, functional association of a lncRNA with clinically relevant phenotypes supports the notion of lncRNA-targeting therapeutics.

C O N F L I C T O F I N T E R E S T
The authors declare that they have no competing interests.

D ATA AVA I L A B I L I T Y S TAT E M E N T
All microarray or RNA-seq data used in this study are acquired from previous studies as described in text.