MicroRNA 4423 is a primate-specific regulator of airway epithelial cell differentiation and lung carcinogenesis

Small RNA Airway Transcriptome Sequencing.

Pools of small RNA (<40 nt) from the bronchial airway epithelium of healthy never smokers, healthy current smokers, current or former smokers with lung cancer, and current or former smokers with benign lung disease (three individuals per pool; see SI Appendix, Table S1 for subject demographics) were sequenced using ABI SOLiD platform (Applied Biosystems). Each read was trimmed and aligned to human genome build 19 (hg19) using Bowtie (18) allowing up to two mismatches per read. On average, 67 million reads were obtained per sample, of which 33.2 million reads aligned to the genome, and 9.8 million reads mapped to a known microRNA precursor from release 16 of miRBase (19) (see SI Appendix, Table S1 for subject demographics and SI Appendix, Table S2 for sequence alignment statistics). A total of 488 microRNAs had expression levels of one read per million reads (RPM) or greater in at least one sample.

Computational Prediction of Unannotated MicroRNAs.

The miRDeep algorithm (20) was used to identify transcribed regions that were predicted to fold into a canonical microRNA structure. Fifty-four microRNA hairpins were identified, five of which were not previously annotated as microRNAs (miRDeep score > 50 and RPM > 5; see SI Appendix, Table S3 for a list of predicted unannotated microRNAs). A predicted microRNA precursor mapping to chromosome 1 (chr1): 85,599,489–85,599,545 (Fig. 1) was the top scoring miRDeep prediction that did not overlap a known microRNA in release 16 of miRBase, and the ninth highest scoring miRDeep prediction overall. Within the predicted precursor, miRDeep predicted two mature microRNAs, 3p and 5p (Fig. 1). Both forms are highly expressed in airway epithelium, with the 3p and 5p being the 85th (91st percentile) and 107th (89th percentile) most highly expressed mature microRNAs, respectively. Quantitative real-time PCR (qRT-PCR) was used to validate the expression of both forms of the putative microRNA in human airway epithelial cells from bronchial brushings (SI Appendix, Fig. S1). During the preparation of this manuscript, the sequence of this putative microRNA was deposited in miRBase release 17 as hsa-miR-4423, based on a small number of sequencing reads in studies that characterized the small RNA transcriptome of melanoma (two reads) and cervical tumors (one read) (21, 22). Given the low level of expression in these studies and the high level of expression in airway epithelium, we investigated the regulation, expression, and function of miR-4423 in the respiratory tract.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Expression and evolutionary conservation of miR-4423. (Top) A coverage plot of sequencing reads aligned to miR-4423, scaled to reads per million (RPM). (Middle) The boundaries of primary transcript as predicted by miRDeep (chr1:85599425-85599600, human genome build 19), with 5p and 3p forms shown as black boxes and predicted seed regions in red. (Bottom) Alignment of human sequence with corresponding genomic regions of nine nonhuman primates and rat (R. norvegicus) (as determined from UCSC Genome Browser MULTIZ track). The seed regions of the 5p and 3p forms of miR-4423 are highlighted in red. Matches to human reference sequence are represented as dots over light gray, mismatches in dark gray, and gaps relative to the human reference as dashes. A 4-bp insertion in the mouse lemur (M. murinus) sequence (relative to the human reference) is indicated with an asterisk. Yellow, great apes; light green, old world monkey; dark green, new world monkey; blue, prosimian; purple, nonprimate.

Conservation of miR-4423.

A multiple species alignment revealed that the genomic locations of the primary transcript and precursor of miR-4423 are highly conserved in simians, i.e., the great apes (Homo sapiens, Pan troglodytes, Gorilla gorilla, Pongo abelii), Old World monkeys (Macaca mulatta, Papio hamadryas), and New World monkeys (Saimiri boliviensis, Callithrix jacchus). The regions corresponding to the miR-4423-5p and -3p seeds are perfectly conserved in simians, with the exception of single base substitutions in the miR-4423-5p seed region of the rhesus (M. mulatta) and squirrel monkey (S. boliviensis) genomes. In contrast, the regions of the primary transcript and precursor are less well conserved in two prosimians (Microcebus murinus and Otolemur garnettii) and are not conserved among nonprimate mammals (Fig. 1). These results suggest that miR-4423 is a recently evolved microRNA.

miR-4423 Is Processed by Dicer and Argonaute in Vitro.

We did not detect expression of either form of miR-4423 or the primary transcript in the 13 lung cancer cell lines or 8 undifferentiated bronchial epithelial cell lines that we examined (SI Appendix, Tables S4 and S5). To experimentally validate whether the identified sequence of miR-4423 encodes a functional mature microRNA, H1299 cells were transfected with a plasmid expressing the miR-4423 predicted precursor sequence (including ∼200 bp of flanking region) under the control of the CMV promoter, which resulted in the expression of mature miR-4423-3p and -5p (SI Appendix, Fig. S2). Expression of both mature forms was significantly reduced when Dicer was knocked down via siRNA in H1299 cells at a level comparable to what we observed for three known microRNAs (i.e., miR-10b, miR-21, and miR-26; SI Appendix, Fig. S3 A and B), suggesting that the transcript of miR-4423 is processed through the Dicer-dependent microRNA biogenesis pathway. In addition, we examined the data of Hafner et al. (23), who sequenced RNA associated with various components of the argonaute (AGO) complex in human embryonic kidney cells (GSE21918) and found three reads that mapped uniquely to miR-4423-5p and one read that mapped uniquely to miR-4423-3p, indicating that these microRNAs can be incorporated into the AGO complex (SI Appendix, Fig. S4). The low number of reads mapping to miR-4423 is likely due to the fact that the experiment was not performed in airway epithelial cells where its expression is high. Taken together, our results validate the computational prediction that the expressed sequence is a functional microRNA.

miR-4423 Expression Is Restricted to Mucociliary Epithelium.

The extent of miR-4423 expression in 24 human tissues was assayed using qRT-PCR. Both forms of miR-4423 were expressed in the respiratory tract, with moderate expression observed in the trachea and lung and high expression observed in the nasal and bronchial epithelium (Fig. 2A). Using in situ hybridization on sections from trachea, mainstem bronchi, and second-generation bronchi from nonsmoking donors, we found that the expression of miR-4423 is primarily restricted to the airway epithelium in these tissues (Fig. 2B and SI Appendix, Fig. S5). In addition, we observed expression of both forms of miR-4423 at low levels in the ovary. We hypothesized that this was due to contamination of the ovarian sample with mucociliary epithelium from the fallopian tube. Consistent with this hypothesis, we detected expression of both miR-4423-3p and -5p in an independent sample of total RNA from fallopian tube epithelium (Fig. 2A).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

The expression of miR-4423 is primarily restricted to mucociliary epithelium. (A) Expression of both forms of miR-4423 across 24 human tissues is detected in the respiratory tract (lung, trachea, and nasal and bronchial epithelia), ovary, and fallopian tube epithelium. (B) By in situ hybridization, miR-4423 is expressed in the epithelium of the trachea, mainstem bronchus, and second-generation bronchus. Arrowheads point to the regions with positive staining.

The miR-4423 primary transcript (pri-miR-4423) is located ∼600 bp downstream of WD repeat domain 63 (WDR63), which encodes a subunit of the inner dynein arm complex of motile eukaryotic cilia (24). Due to the close proximity of the two loci, and because the expression of WDR63 has also been reported in bronchial epithelium (25), we hypothesized that pri-miR-4423 and WDR63 are coexpressed. To test this hypothesis, qRT-PCR was used to assay the expression of pri-miR-4423, WDR63, and both mature forms of miR-4423 in the 24 human tissues. We found that the expression of WDR63 had a strong positive correlation with the expression of pri-miR-4423, suggesting that the two loci are coexpressed and that miR-4423, like WDR63, is expressed in the ciliated cells of the airway. Interestingly, the expression of pri-miR-4423 and both mature forms of miR-4423 were also highly correlated across most tissue types with a few exceptions. Most notably, we did not detect expression of miR-4423-3p or -5p in the testis, kidney, placenta, and brain, despite detecting high levels of WDR63 and pri-miR-4423 expression. These data suggest that although pri-miR-4423 is expressed in these tissues, it is not efficiently processed into the mature form (SI Appendix, Fig. S6A).

Additionally, the expression of miR-4423 together with pri-miR-4423 and WDR63 were highly induced when normal human bronchial epithelial cells (NHBEs) were differentiated into mucociliary epithelia at an air−liquid interface (ALI), beginning 6 d after the cells were raised to the ALI and continuing through day 13 (SI Appendix, Fig. S6B). The expression of miR-4423 was also strongly correlated with that of the ciliogenic regulator forkhead box J1 (FoxJ1) (SI Appendix, Fig. S7A), and this correlation was stronger than the correlation between miR-4423 expression and that of markers of goblet cells (MUC5B and MUC5AC), Clara cells (CC10), and neuroendocrine cells (ASCL1) (SI Appendix, Fig. S7B). Collectively, these data suggest that miR-4423 is expressed in the ciliated cell population of the airway epithelium.

miR-4423 Overexpression Results in an Increase in the Number of Ciliated Cells at an ALI.

To determine whether miR-4423 is functionally involved in the development of the airway epithelium, we stably overexpressed and knocked down both forms of miR-4423 in NHBE cells and differentiated them into mucociliary epithelium at an ALI. We found that overexpression of miR-4423 results in a significant increase in the number of cells expressing the ciliated cell markers FOXJ1 and beta-tubulin (β-tubulin) (Fig. 3 and SI Appendix, Fig. S8), suggesting that the ectopic expression of miR-4423 is sufficient to promote ciliated cell differentiation in the airway epithelium. However, knockdown of miR-4423 results in only a modest decrease in FOXJ1- and β-tubulin−expressing cells, indicating that either miR-4423 may not be required for ciliated cell differentiation or that the level of inhibition we were able to achieve is not sufficient to induce a miR-4423-deficient phenotype (SI Appendix, Fig. S9 A and B).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

MiR-4423 overexpression results in an increase in the number of cells expressing ciliated cell markers. NHBE cells overexpressing miR-4423 or control were differentiated into mucociliary epithelium at an ALI. NHBE cells overexpressing miR-4423 (Upper) show substantial FOXJ1 and β-tubulin staining compared with control (Lower). Representative images shown were taken at days 9 (FOXJ1, Left) and 11 (β-tubulin, Right). Arrows are pointing to regions of positive staining.

Loss of Expression of miR-4423 Is Associated with Lung Cancer.

As the expression of miR-4423 is increased during the differentiation of mucociliated airway epithelium, we examined its expression in cytologically normal epithelial cells from mainstem bronchial brushings of smokers with (n = 5) and without (n = 4) lung cancer (see SI Appendix, Table S6 for subject demographics) and found that both the 3p and 5p forms are significantly reduced in smokers with cancer (Fig. 4A). Moreover, the expression of miR-4423-3p and -5p, along with the highly correlated expression of pri-miR-4423 and WDR63, were also significantly reduced in a large fraction of squamous carcinomas (SCC) and adenocarcinomas (ADC) relative to matched adjacent normal tissues regardless of smoking status (Fig. 4B and SI Appendix, Fig. S10 A and B). We extended these observations to the lung cancer RNA-seq dataset from The Cancer Genome Atlas (TCGA) (26), using WDR63 expression levels as a proxy for the levels of miR-4423 expression, and found that WDR63 was down-regulated in a similar proportion of SCC and ADC (SI Appendix, Fig. S10C). Interestingly, miR-4423 expression is reduced in squamous metaplasia and is reduced further in SCC compared with normal airway epithelium (SI Appendix, Fig. S11), indicating that the decrease in miR-4423 expression might be an aspect of a process that occurs early in carcinogenesis. In TCGA data, we did not observe evidence of copy number variation that could account for the loss of WDR63 expression in lung tumors (SI Appendix, Fig. S12 A and B). We did, however, observe a negative correlation between methylation levels and WDR63 expression in ADC (but not in SCC) (SI Appendix, Figs. S13 and S14). Collectively, these results suggest that miR-4423 is down-regulated in a wide range of lung tumors and in premalignant lesions.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

MiR-4423 expression is associated with lung cancer. (A) Expression of miR-4423 was assayed in histologically normal epithelium from the mainstem bronchus of smokers with lung cancer (C) (n = 5) and smokers with benign disease of the chest (NC) (n = 4). The 3p and 5p forms of miR-4423 are significantly down-regulated in the bronchial epithelium of smokers with lung cancer compared with smokers without lung cancer (3p, P = 0.037; 5p, P = 0.045). (B) Expression of miR-4423-3p and -5p is significantly down-regulated in tumor tissue (T) compared with matched adjacent normal tissue (Adj.N) for: SCC (n = 15; 3p, P = 0.031; 5p, P = 0.01), ADC from current and former smokers (n = 10; 3p, P = 0.029; 5p, P = 0.028), and ADC from nonsmokers (n = 10; 3p, P = 0.01; 5p, P = 0.04). Error bars indicate SE, and P values were determined using Student t test in A and a paired t test in B.

miR-4423 Inhibits Anchorage-Independent Growth in Lung Cancer Cell Lines.

To investigate whether miR-4423 is an inhibitor of a cancer-associated process, miR-4423 was stably expressed in seven SCC cell lines (SW900, H1703, RH2, H2170, Skmes-1, H520, and Calu-1), and in three ADC cell lines (Caul-6, H1299, and A549). Expressing miR-4423 in Calu-6, SW900, H1703, and RH2 cells significantly decreased their ability to form colonies in soft agar (Fig. 5A), but it did not affect the anchorage-independent growth capacity of the remaining six cell lines. To characterize the biological basis of miR-4423 sensitivity with regard to anchorage-independent growth, we profiled baseline gene expression of two cell lines that were sensitive to miR-4423 in soft agar (Calu-6 and SW900) and one that was resistant (H2170). We found that genes that were expressed at higher levels in both miR-4423−sensitive cell lines relative to the miR-4423−resistant line H2170 were significantly enriched in genes important for cell differentiation, cell-to-cell contact/migration, and the cell cycle. In contrast, genes expressed at lower levels were significantly enriched for those involved in apoptosis (SI Appendix, Fig. S15).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

MiR-4423 inhibits lung cancer anchorage-independent growth in vitro and tumor growth in vivo. (A) Soft agar assays were performed in the indicated cell lines stably transfected with either a vector that overexpresses the miR-4423 precursor or the empty parent vector as a negative control (n = 10). Overexpression of miR-4423 in four of the cell lines tested decreases the number of colonies formed in soft agar (Calu-6, P = 2.8 × 10⁻⁷; SW900, P = 2.4 × 10⁻⁵; H1703, P = 7.4 × 10⁻⁹; RH2, P = 0.001). Error bars indicate SE, and P values were determined using Student t test. (B) H1703 (SCC) cells stably overexpressing miR-4423 or a control (1 × 10⁶) were injected s.c. into the backs of NSG mice (seven mice per group). Tumors derived from miR-4423 overexpressing cells were growth-suppressed relative to the control-derived tumors as shown in representative photographs of the tumors (Left), tumor volume over time (P = 1.55 × 10⁻¹¹) (Upper Right) and tumor weight (P = 0.01) (Lower Right). (C) Phosphorylated E-cadherin staining was performed in miR-4423-overexpressing tumors and controls (H1703). Mir-4423−overexpressing tumors exhibited focal areas of positive membrane phospho-E-cadherin staining consistent with the presence of tight junctions (Lower). Phospho-E-cadherin staining was not observed in control tumors (Upper). Arrows point to regions of positive staining.

miR-4423 Suppresses Xenograft Tumor Growth.

Based on these results, we sought to determine whether miR-4423 expression can inhibit lung tumor growth in mouse xenografts. To test this hypothesis, H1703, Calu-6, and H1299 cells stably expressing miR-4423 or a control vector were injected s.c. into the backs of immunodeficient mice. We found that although miR-4423 overexpression caused a modest decrease in tumor growth in Calu-6 and H1299 cells (SI Appendix, Fig. S16), it significantly reduced the size of the tumors formed by the SCC cell line H1703 (Fig. 5B). Upon histological examination of the H1703 miR-4423−overexpressing tumors we observed foci of altered morphology. These foci, which were not observed in the control tumors, had a more structured cellular organization (SI Appendix, Fig. S17). Because E-cadherin plays an important role in cell−cell adhesion and loss of its expression can promote increased tumorigenesis by inducing anchorage-independent growth and colonization of tumor cells (27, 28), we analyzed its pattern of expression in the H1703-derived tissues. We found that the miR-4423−overexpressing tumors contained foci of increased phosphorylated E-cadherin, which were not observed in the control tumors (Fig. 5C). This result suggests that miR-4423 overexpression promotes increased formation of cell−cell adhesions and provides a possible mechanism by which it suppresses tumor growth.

Transcriptomic Consequences of miR-4423 Modulation.

To explore the mechanisms by which miR-4423 modulates airway differentiation and lung cancer associated phenotypes, we predicted potential mRNA targets of miR-4423 using the TargetscanS v5.0 (29) and Miranda v3.3a (30) algorithms on 3′ UTR sequences obtained from Ensembl and Refseq (Dataset S1). A total of 809 and 1,578 genes were predicted to be targets of miR-4423-3p and -5p, respectively, by both algorithms in both sets of 3′ UTR sequences; 181 of these genes were predicted targets of both forms. Although the seed regions of the 3p and 5p forms of miR-4423 are not shared with any other known mammalian microRNAs, we observed a 4- to 5-nucleotide overlap between the seed region of miR-4423-3p and some members of the miR-449 and miR-34 families (SI Appendix, Fig. S18A) and a significant number of shared predicted targets (SI Appendix, Fig. S18B). Interestingly, these microRNA families are conserved across vertebrates, are up-regulated during mucociliary epithelium differentiation, and are key regulators of multiciliogenesis (4, 31).

To characterize the transcriptomic effects of miR-4423 modulation, we profiled gene expression in H1299 cells transiently overexpressing miR-4423 (n = 3) or empty vector controls (n = 3) and identified 1,231 genes significantly changed [false discovery rate (FDR) q < 0.25; SI Appendix, Fig. S19A and Dataset S2]. The set of genes whose expression was reduced upon miR-4423 overexpression was enriched in genes with putative 3p and 5p binding sites (P = 0.001; Fisher’s exact test; SI Appendix, Table S7). Likewise, genes with putative binding sites for miR-4423 were enriched among genes down-regulated with miR-4423 overexpression (P < 0.001; Kolmogorov−Smirnov test; SI Appendix, Fig. S19B).

Genes whose expression was reduced after overexpression of miR-4423 were enriched in chaperone proteins [FDR q < 0.05; Database for Annotation, Visualization, and Integrated Discovery (DAVID)], including Hsp70 and Hsp40 family members, which are important in protecting cells from apoptosis and promoting anchorage-independent growth (32). Among the down-regulated predicted targets, we found members of signaling pathways critical for anchorage-independent survival and growth of cancer cells, including a catalytic subunit of phosphatidylinositol 3-kinase (PIK3CA) (33) and the SH2-containg protein, SHC1, which couples activated growth factor receptors to the Ras pathway and promotes cellular transformation (34).

In comparison with previous studies, we found that genes whose expression was altered upon miR-4423 overexpression were concordantly differentially expressed during differentiation of NHBEs at an ALI [FDR q < 0.05; gene set enrichment analysis (GSEA); SI Appendix, Fig. S20] (35). Consistent with our observation that miR-4423 expression begins at day 6 of ALI differentiation (SI Appendix, Fig S7A), we found that many of the genes altered by miR-4423 overexpression were those whose expression levels began to change specifically between days 4–8 of ALI differentiation (SI Appendix, Fig. S20). With regards to lung cancer, using a previously published gene expression study from our group (17), we found that genes that decrease upon miR-4423 overexpression are enriched among genes whose expression is increased in airway epithelium from patients with lung cancer (FDR q < 0.001; GSEA). We also found in three datasets (36⇓–38) that the genes altered upon miR-4423 overexpression were significantly enriched among genes whose expression changed in the opposite direction in lung ADC and SCC relative to adjacent normal tissue (FDR q < 0.05; GSEA; SI Appendix, Fig. S21 A and B). Together, these results suggest that miR-4423 can regulate gene expression changes that occur during both the differentiation of airway epithelium as well as lung carcinogenesis.