NKX2-1-AS1 negatively regulates CD274/PD-L1, cell-cell interaction genes, and limits human lung carcinoma cell migration

The function of most long noncoding RNAs (lncRNAs) is unknown. However, recent studies reveal important roles of lncRNAs in regulating cancer-related pathways. Human antisense lncRNA-NKX2-1-AS1 partially overlaps the NKX2-1/TTF1 gene within chromosomal region 14q13.3. Amplification of this region and/or differential expression of genes therein are associated with cancer progression. Herein we show higher levels of NKX2-AS1 and NKX2-1 in lung adenocarcinomas relative to non-tumor controls but no correlation between NKX2-1-AS1 and NKX2-1 levels across specimens, or with amplification of the 14q13.3 region, suggesting that NKX2-1-AS1 and NKX2-1 are independently regulated. Loss-and-gain of function experiments showed that NKX2-1-AS1 does not regulate NKX2-1 expression, or nearby genes, but controls genes in trans. Genes up-regulated by NKX2-1-AS1-knockdown belong to cell adhesion and PD-L1/PD-1 checkpoint pathways. NKX2-1-AS1 negatively regulates endogenous CD274/PD-L1, a known target of NKX2-1, and the transcriptional activity of -1kb-CD274 promoter-reporter construct. Furthermore, NKX2-1-AS1 interferes with NKX2-1 protein binding to the CD274-promoter, likely by NKX2-1 protein-NKX2-1-AS1 interactions. Finally, NKX2-1-AS1 negatively regulates cell migration and wound healing, but not proliferation or apoptosis. These findings support potential roles of NKX2-1-AS1 in limiting motility and immune system evasion of lung carcinoma cells, highlighting a novel mechanism that may influence tumorigenic capabilities of lung epithelial cells.


Supplementary
: Individual siRNAs A, B and D were able to down-regulate NKX2-1-AS1 expression by >80% in H441 cells, similar to the combination of all three siRNAs used in the experiments in this study, which was used to reduce non-specific effects of individual siRNAs. NKX2-1 mRNA expression was not significantly changed by NKX2-1-AS1 knockdown with any of the individual siRNAs. Treatment with individual siRNAs (A, B, or D) significantly increased expression of CD274, and treatment with B or D significantly increased expression of PTPN1, both genes identified as downstream of NKX2-1-AS1 in this study. n=3; ** p< 0.01; ***p< 0.001.

Summary of microarray samples and experimental design
This study is comprised of two independent experiments, each containing 6 Human Gene 2.0 ST arrays profiling H441 lung adenocarcinoma cells treated with control siRNA or a pool of three siRNAs (A, B and D) targeting the lncRNA NKX2-1-AS1 (n=3 per group in each experiment). Trizol RNA extraction was used to prepare samples for the first experiment (Experiment 1), while spin-column RNA extraction was used to prepare samples for the repeat of the first experiment (Experiment 2).

Normalization and quality assessment
All 12 arrays were normalized together using the Robust Multiarray Average (RMA) algorithm and a CDF (Chip Definition File) that maps the probes on the array to unique Entrez Gene identifiers. The result is a matrix in which each row corresponds to an Entrez Gene ID and each column corresponds to a sample.

Positive control gene expression
The expression of several constitutively expressed Y-linked genes (DDX3Y, KDM5D, RPS4Y1, USP9Y, and UTY) was assessed to estimate the dynamic range of the array, as these genes will serve as strong positive or negative expression controls in males and females, respectively. In both experiments the expression of XIST was very low (~1-2 log2 units) in all samples, and although the expression of most Y-linked genes was also relatively low (~2-4 log2 units), the DDX3Y had moderate expression (~5 log2 units) in all samples across both experiments. These findings are in agreement with the fact that the H441 cell line was derived from a male subject and contains a Y chromosome (http://www.atcc.org/products/all/HTB-174.aspx). As other housekeeping genes had high expression, there appears to be good dynamic range to discriminate true positive and negative controls in this experiment.
The expression of NKX2-1-AS1 was decreased ~1.2-fold in the knockdown group versus the control group in both experiments (reduction to 83% of original levels in Experiment 1 and 81% of original levels in Experiment 2).

Principal Component Analysis (PCA)
Following the initial QC analysis, Principal Component Analysis (PCA) was performed. PCA is a mathematical transform that collapses the variance between samples across a set of large set of variables (here, all ~25,000 genes on the array) into a much smaller set of variables called Principal Components (PCs). These "meta-variables" are arranged such that PC1 explains the most variance in the data, followed by PC2, etc.
PCA was performed using all genes across all samples, and a plot was made of PC2 vs. PC1 for both experiments independently and then combined. The plots can also be downloaded at: 2014-08-08_Ramirez_combined_PCA.pdf Samples shown in Figure 1 separate on the PCA plot by experiment along the PC1 axis, which explains 28% of variance in the experiment. This suggests that either the RNA extraction protocol or batch effect is the major cause of variance between the two experiments. However, in both experimental groups, control samples separate well from the NKX2-1-AS1 knockdown samples along the PC2 axis, which explains 11% of the variance in the experiment, suggesting that the siNKX2-1-AS1 treatment caused the most variance within each experiment.  Figure 2 shows the adjustment for experimental protocol leading to a distinct separation between all of the control and all of the NKX2-1-AS1 knockdown samples from both experiments along the PC1 axis, which is responsible for 16% of variance in the experiment. However, in both the control samples and the ABD clusters, the distinct experimental groups separate from each other. This suggests that after adjusting for experimental protocols, the treatment becomes the most important cause in variance across all samples, and the separation between the experimental groups may potentially be explained by the two different RNA extraction methods.

Moderated t tests: NKX2-1-AS1 knockdown versus control
To identify genes whose expression changed with respect to treatment after adjusting for experimental protocol, a linear model was created of the form expression ~ protocol + treatment where '~' means 'is a function of', and protocol and treatment are treated as categorical variables. The treatment effect therefore measures whether a given gene changes expression between siRNA groups after correcting for any technical effects introduced by the two experiments.
A t test was then performed for the treatment term in the model to determine its significance. A "moderated" t test was used, which is a Bayesian analysis that does not test each gene independently, but rather, leverages information from all of the genes on the array to increase statistical power over a standard two-sample Student t test. It is especially helpful when sample sizes are small.
Benjamini-Hochberg False Discovery Rate (FDR) correction was then applied to obtain FDR-corrected p values ('q' values), which represent the probability that a given result is a false positive based on the distribution of all p values on the array. Corrected/adjusted p values such as the FDR q are the best measure of significance for a given test when many hypotheses (e.g., ~25,000 genes) are tested at once. In addition, the FDR q value was also recomputed after removing probesets that were not expressed above the array-wise median value of at least one array. Probesets with low overall expression are more strongly affected by random technical variation and more likely to produce false positive results. Table 1 below shows the number of genes with a p value below various thresholds, as well as the number of genes expected by chance at each threshold. Similarly, the table below also shows the number of genes with a q value below various thresholds, either with or without expression filtering. As was expected, when combining all samples from the two experiments and adjusting for experimental protocols, there are considerable differential gene expressions changes between the cells treated with NKX2-1-AS1 siRNA and control siRNA. This is further reflected in the large number of genes that pass FDR q correction.