|Home | About | Journals | Submit | Contact Us | Français|
The PicTar program predicted that microRNA-126 (miR-126), miR-145, and let-7s target highly conserved sites within the Hoxa9 homeobox. There are increased nucleotide constraints in the three microRNA seed sites among Hoxa9 genes beyond that required to maintain protein identity, suggesting additional functional conservation. In preliminary experiments, forced expression of these microRNAs in Hoxa9-immortalized bone marrow cells downregulated the HOXA9 protein and caused loss of biological activity. The microRNAs were shown to target their predicted sites within the homeobox. miR-126 and Hoxa9 mRNA are coexpressed in hematopoietic stem cells and downregulated in parallel during progenitor cell differentiation; however, miR-145 is barely detectable in hematopoietic cells, and let-7s are highly expressed in bone marrow progenitors, suggesting that miR-126 may function in normal hematopoietic cells to modulate HOXA9 protein. In support of this hypothesis, expression of miR-126 alone in MLL-ENL-immortalized bone marrow cells decreased endogenous HOXA9 protein, while inhibition of endogenous miR-126 increased expression of HOXA9 in F9 cells.
Hoxa9 is one of the 39 mammalian Hox genes, which are defined based on high sequence homology within their homeoboxes to Drosophila Antp (1). The 180-nucleotide (nt) homeobox encodes the DNA binding homeodomain motif that confers transcriptional regulatory activity to these proteins (17). Hoxa9 is important for skeletal (7), mammary gland (6), limb bud (14), urogenital tract (46), and kidney development (11). Hoxa9 plays a role in normal myeloid (33) and T-cell differentiation (21) as well as in hematopoietic stem cell (HSC) function (32). Hoxa9 is an oncogene (29) and is often upregulated in myeloid leukemias (34). In particular, aberrant Hoxa9 expression appears to be a common downstream event in several forms of leukemia marked by chromosomal rearrangements (reviewed in reference 10). In a study of 6,817 genes, Hoxa9 expression was most highly correlated with poor clinical outcomes in patients treated for acute myelogenous leukemia (18).
Regulation of HOXA9 protein or mRNA levels by microRNAs (miRNAs) has not been reported. The programs available at the inception of this project (23, 24, 28, 35) gave various predictions for putative miRNA target sites within the ~1,160-nt 3′ untranslated region (UTR) of the major Hoxa9 transcript (GenBank accession number BC006537). However, we were particularly intrigued by the fortuitous observation that one program utilized an alternatively spliced variant that designates the homeobox as the 3′ UTR to predict that miRNAs would target the Hoxa9 homeobox region. Since previous studies had shown that several mammalian Hox genes could partially functionally replace their fly homologues in situations where essentially only the homeobox and an EXD-PBX interaction motif are conserved (36, 37), much of the biological activity of HOX proteins appears to reside within the homeodomain, with little biological activity other than the PBX interaction motif (4) within the flanking regions. Thus, the prediction that miRNAs would target the homeobox region appeared to be an especially appealing area for investigation. Since the defining feature of the Hox genes is their highly conserved homeoboxes, part of the unexplained function of these genes is their differential activity when they are coexpressed in highly overlapping patterns (30). The elucidation of possible regulation of these genes by miRNAs, which are thought to exquisitely sense single nucleotide differences, might provide a new level of control of these master regulatory genes.
In the current study, Hoxa9-immortalized bone marrow (BM) cells were used in conjunction with retroviral vector-mediated expression of hairpin miRNA precursors in proof-of-principle studies which showed that miR-126, miR-145, and let-7a target their predicted sites within the Hoxa9 homeobox. The miRNAs downregulated HOXA9 protein and its associated biological activity. Sequence homology analysis demonstrated that the putative miRNA binding sites were constrained beyond the level of constraint required for maintenance of protein coding, suggesting additional functionality. Quantitative reverse transcription-PCR (QRT-PCR) of hematopoietic progenitor cells showed that miR-126 and Hoxa9 are expressed in parallel in HSCs but are inversely expressed in myeloid leukemic lines and primary leukemia BM samples, suggesting that variation in levels of miR-126 might regulate HOXA9. This prediction was supported by the findings that the forced expression of miR-126 downregulated endogenous HOXA9 in MLL-ENL cells while the inhibition of endogenous miR-126 upregulated HOXA9 protein in F9 cells. We hypothesize that miR-126, in conjunction with the ubiquitous let-7s, normally targets Hoxa9 during early HSC differentiation but that in myeloid leukemias, the loss of this regulatory miRNA might elevate the HOXA9 oncoprotein.
Site-directed mutagenesis was used to generate mutHoxa9, in which nucleotides within the three putative miRNA target sites were changed while the wild-type amino acid sequence was maintained. We discovered that, although a 1,975-nt full-length Hoxa9 clone (numbered from the protein start site; I.M.A.G.E. clone BC055059) would produce protein when transiently expressed from the murine stem cell virus (MSCV) long terminal repeat, following retroviral transduction, the HOXA9 protein could not be detected. Since this cDNA contained two putative polyadenylation signals in the 3′ UTR, we removed these sites to determine if they were interfering with retrovirally mediated protein expression. The site at nt 1798 was replaced using a primer containing a KpnI sequence in place of the AATAAA sequence together with a primer containing the BglII site at nt 749 to amplify a 1,052-nt BglII-KpnI fragment by PCR. A second primer containing the KpnI site in place of the polyadenylation signal at nt 1798 was used in conjunction with a primer containing an AAT-to-CGG change in the polyadenylation signal at nt 1975 and an adjacent XhoI site for cloning to generate a 150-nt KpnI-XhoI fragment. Standard methods were used to combine these fragments to create a 1,975-nt Hoxa9 cDNA encoding an N-terminal FLAG-tagged version of the mature protein but lacking the two putative polyadenylation signals. The resulting MSCV clone produced high levels of FLAG-HOXA9 protein and immortalized BM cells.
Murine genomic regions containing the miRNA hairpins and ~150 nt of 5′ and 3′ flanking regions were cloned into the MDH vector (5). High-titer retroviruses were produced by transient transfection of 293T cells and viral supernatants collected at 48 and 72 h posttransfection.
Immortalized lines were produced by transducing 5′-fluorouracil (5′-FU)-treated BM progenitors with a truncated Hoxa9 cDNA lacking the 3′ UTR, the same cDNA with mutated binding sites for the three miRNAs in the homeobox (mutHoxa9), or a modified full-length Hoxa9 cDNA, each cloned into an MSCV vector containing a neomycin selection cassette (12). F9 cells seeded at 0.4 × 106 in 1 ml of Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (FCS) were cultured at 37°C for 4 h, transfected with 4 μl of DharmaFECT-1 and 50, 100, or 200 nM of miRNA-126-3p inhibitor (I-310017-02) or miRIDIAN miRNA inhibitor negative control 1 (IN-001000-01) (Dharmacon Inc., http://www.dharmacon.com), and assayed after 72 h. Proteins were assayed by Western blotting with antisera against HOXA9 (Upstate catalog no. 07-178), followed by antiserum to tubulin (Santa Cruz Biotechnology catalog no. 5286) or actin (Sigma catalog no. A5316) (12).
Green fluorescence protein-positive (GFP+) transduced cells were sorted prior to Western blotting, colony assays, and growth (12). BM progenitors were sorted into long-term HSCs, short-term HSCs, and early progenitor pools in lysis buffer (Ambion catalog no. 1931) for RNA isolation through multichannel sorting (13). After approval by the UCSF Committee on Human Research, unidentified leukemia BM cells were placed in cell lysis buffer for RNA isolation. QRT-PCR was used to measure miR-126 and miR-145 or U6 RNA, by using Applied Biosystems kits (Applied Biosystems), and Hoxa9 or actin mRNAs (Table (Table1).1). Human leukemic cell lines were grown (34) and assayed for miRNA expression by Northern analysis, using antisense probes (Table (Table1).1). The Hoxa9 values were normalized to actin results and miR-126 values were normalized to U6 RNA results. To compare miR-126 expression levels between human and murine cells, miR-126 and U6 were measured in DAMI cells by QRT-PCR. miR-126 was measured by Northern analysis in all four human lines, and Hoxa9 values normalized to DAMI results are reported in Table Table44.
For myeloid colony assays, 2 × 104/ml of GFP+ FACS-sorted, Hoxa9-immortalized BM cells plated in triplicate in cytokine-rich methyl cellulose medium (Stem Cell Technologies catalog no. M3434) in the presence or absence of 1.3 mg/ml of G418 were scored on days 7 and 8 according to standard criteria. The contents of culture dishes were collected on days 7 and 8 and washed three times with RPMI medium for replating to establish Hoxa9-immortalized lines.
Statistical significance was calculated using the Student t test. All experiments were performed in duplicate, except where additional experiments were performed as noted.
Two alternatively spliced Hoxa9 transcripts have been described previously (15). One transcript encodes a full-length, homeodomain-containing protein. An alternatively spliced transcript contains a stop codon, is predicted to encode a truncated protein lacking the homeodomain, and contains the homeobox nucleotide sequence as part of the 3′ UTR. During a search for the miRNAs predicted to target Hoxa9, we serendipitously noted that although the prediction programs available at the inception of this project examined only 3′ UTRs (23, 24, 28, 35), the PicTar program utilized the alternative Hoxa9 transcript to predict that miR-126, miR-145, and several let-7 family members (let-7a, -b, -c, -e, -f, -g, and -i) might target Hoxa9 by binding the homeobox (Fig. 1A to C). These three sites were predicted based on seed sequence matches. The let-7s matched at seed nt 1 to 7, miR-126 matched at seed nt 1 to 7, and miR-145 matched at seed nt 2 to 8, with an extension to positions 9 and 10 by G-U pairing. All three seed targets are highly conserved across Hoxa9 genes (Fig. (Fig.1).1). Since 10 of the 11 members of the let-7 miRNA family share the same 8-nt seed sequence, they were all predicted by the PicTar program to bind the Hoxa9 homeobox. The three miRNAs exhibit substantial but variable binding to the Hoxa9 homeobox sequences throughout the remaining 12 to 14 nt of the respective miRNA (referred to below as extended miRNA binding sites), and most of these extended sites are conserved among Hoxa9 genes (Fig. (Fig.1).1). Given that the defining feature of the Antp-like Hox genes is their highly conserved homeoboxes, it was of interest to examine whether the three predicted miRNA seed sequences were unique to Hoxa9. Comparisons of the 39 human Hox homeobox sequences showed that no other gene contained a putative miR-126 target site, while only four were possible targets for miR-145 and a different four Hox genes exhibited seed matches for the let-7 miRNAs (Fig. (Fig.1D).1D). Additional analysis showed that the ~2.0-kb full-length Hoxa9 cDNA does not contain target seed sites for miR-126 or miR-145, while a single let-7 site is located near one of the polyadenylation signals in the 3′ UTR, suggesting that the three miRNAs would target the homeobox region.
There are few, if any, published reports on miRNA target sites within coding regions, primarily due to the fact that conservation of protein functionality prevents robust computerized prediction within these regions. We considered that the nucleotide conservation observed for the miRNAs might reflect merely the preservation of critical HOX coding regions. For instance, the amino acid sequences of the four human HOX9 proteins (A9, B9, C9, and D9) are 96.5% conserved across the homeodomain, while eight vertebrate HOXA9 proteins whose mRNAs are shown in Fig. Fig.1C1C are 99.6% identical within the homeodomain (data not shown). Thus, we assumed that nucleotide changes that would alter the coding information are disallowed by protein function constraints. Using these criteria, there are 75 possible first- and third-base wobble sites, which would maintain amino acid identity, among the eight Hoxa9 homeoboxes. These permissible variable nucleotide sites outside the extended miRNA target sites, the permissible variable nucleotides predicted to be miRNA binding sites, and those within the extended miRNA target sites but predicted to not bind miRNAs are distinguished in Fig. 1A to C. The amount of nucleotide variation outside the extended miRNA target sites (non-miRNA) was compared with the variation among base positions predicted to make direct contacts within each miRNA for the eight Hoxa9 homeobox sequences (Fig. (Fig.11 and Table Table2).2). As expected, there is little constraint and, thus, substantial variation across the non-miRNA regions, including the critical helix 3 DNA binding motif. In contrast, there is extremely low nucleotide variation in the miRNA sites. All three miRNA targets show a statistically significant decrease in nucleotide variability for bases predicted to contact the miRNAs compared to the rest of the homeobox (Table (Table2).2). Potentially variable codon bases within the extended miRNA binding sites, which are not predicted to directly bind the miRNAs, also exhibit higher conservation, suggesting that secondary-structure constraints may act to maintain base identity throughout the extended miRNA targets. We interpreted these data to reflect a biological function beyond protein coding for the proposed miRNA binding sites. Based on these observations, we initiated a project to discover whether the three identified miRNAs could regulate HOXA9 protein levels.
Since earlier studies suggested that miRNAs function more effectively when clustered (reviewed in reference 19), and since the three predicted miRNA target sites are closely clustered, we first tested the hypothesis that the three miRNAs might function to synergistically knock down the HOXA9 protein. In a separate experiment, analysis of miRNA expression in lineage-depleted, murine BM progenitor cells revealed that all eight of the let-7 miRNA species assayed (a, b, c, d, e, f, g, and i; Ambion catalog no. 1567) were expressed at levels 5- to 50-fold higher than those of miR-126 and miR-145 (Table (Table3).3). let-7a was arbitrarily chosen for coexpression in various combinations with miR-126 and miR-145 in the MDH retroviral vector, designed to express miRNA hairpins under the control of the PH1 promoter together with a GFP cassette (Fig. (Fig.2A).2A). The vector expressing all three miRNAs (triple vector) produced robust amounts of the mature 22-nt products following transient transfection of 293T cells (Fig. (Fig.2B).2B). In stably transduced BM cells, the expression of the mature miRNAs from the single or double vectors was ~8 or 4 times higher, respectively, than that observed for the triple vector (data not shown).
In the first biochemical proof-of-principle experiment to test whether the miRNAs could knock down HOXA9 protein in vivo, we used a previously described BM cell line (12), which was immortalized with a form of the Hoxa9 cDNA that expresses the full-length HOXA9 protein but lacks the extended Hoxa9 3′ UTR (43). This truncated Hoxa9 cDNA was chosen for the initial experiments because the 3′ UTR contains a putative let-7 target sequence that might confound results. The Hoxa9-immortalized murine BM cell line expresses high levels of the HOXA9 protein (Fig. (Fig.2C,2C, lanes 1 and 8). These cells were transduced with retroviral vectors expressing miR-126, miR-145, let-7a alone, let-7a with either miR-126 or miR-145, or the triple-expression plasmid. Although the transduction efficiencies of the single and double vectors were typically ~90% or greater, the triple-vector efficiency was ~40 to 50%. In order to compare knockdown efficiencies, cells were normally sorted for GFP+ expression prior to protein analysis. In replicate transduction experiments, the single miRNAs showed activity against HOXA9, with the triple vector exhibiting somewhat variable but normally the greatest amount of knockdown, with a mean decrease of 0.57% ± 0.23% (n = 10) (Fig. (Fig.2C).2C). A portion of the observed variability may have been due to differences in miRNA expression levels, as reflected by significant differences in GFP expression between experiments. In all experiments, a nonspecific miRNA hairpin was included to control for possible effects on Dicer and Drosha caused by the exogenous hairpin RNAs.
A mutant form of the Hoxa9 cDNA, in which the putative miRNA targets were altered at the nucleotide level while maintaining the wild-type amino acid sequence (mHoxa9) (Fig. (Fig.2D),2D), was produced. This cDNA was used to produce an mHoxa9-immortalized BM cell line, in which the HOXA9 protein was expressed at levels similar to that expressed by the wild-type cDNA. Transduction of these cells with the triple-miRNA vector produced an insignificant decrease in HOXA9 protein (0.17% ± 0.17%, n = 4) (Fig. (Fig.2E).2E). The effect of the triple vector on the wild-type HOXA9 protein (Fig. (Fig.2C)2C) was significantly different from the decrease observed with the mutant protein (P = 0.01). These data indicate that the miRNAs are targeting the predicted sequences within the Hoxa9 homeobox.
For historical reasons, all previous studies on HOXA9 immortalizing and leukemogenic activity (2, 12, 29) have used the truncated cDNA lacking ~1,160 nt of the 3′ UTR that we used for the initial studies of sensitivity to miRNAs directed against Hoxa9 described above. It was formally possible that the full-length endogenous Hoxa9 cDNA might be folded in such a manner as to be resistant to the miRNAs. Since initial experiments using a wild-type 3′ UTR revealed that the putative polyadenylation signals were incompatible with efficient HOXA9 protein expression, we immortalized BM cells with a modified version of the full-length cDNA in which two polyadenylation signals had been removed. Transduction of this line with the triple miRNA knocked down the HOXA9 protein (Fig. (Fig.2F),2F), showing that the targeting miRNAs could bind to a full-length Hoxa9 transcript (see also experiments described below using MLL-ENL cells).
HOXA9-immortalized cells possess the capacity to form colonies following serial replating in semisolid media. Small-interfering-RNA-mediated loss of HOXA9 protein in these cells results in the loss of colony-forming activity (D. Garcia and C. Largman, unpublished data). Following FACS for GFP+, BM cells immortalized with the Hoxa9 lacking the 3′ UTR and transduced with single, double, or triple miRNAs directed against Hoxa9 showed greatly diminished colony-forming capacity (Fig. (Fig.3A,3A, left). In contrast, cells immortalized with the mutant HOXA9 protein formed colonies in the presence of the triple-miRNA vector (Fig. (Fig.3A,3A, right). Similar experiments were performed to confirm that the combination of the three miRNAs reduces the biological activity of the full-length Hoxa9 cDNA (Fig. (Fig.3B).3B). To confirm the capacity of the miRNAs to block HOXA9 biological activity, the growth of Hoxa9-immortalized cells in liquid culture was assessed. FACS-sorted cells transduced with either the single, double, or triple vectors expressing the miRNAs targeting HOXA9 showed substantially decreased growth rates compared to the control miRNA (Fig. (Fig.3C3C).
Following the biochemical demonstration that shows that miRNAs against the Hoxa9 homeobox could regulate function, we embarked on an analysis of the expression patterns of the three miRNAs in various hematopoietic cell compartments and tissues. Since the preliminary microarray data showed that at least seven let-7s are highly expressed in the progenitor-enriched BM fraction, all of which are predicted to bind to the Hoxa9 homeobox, we presumed that there is sufficient endogenous let-7 to potentiate the action of endogenous miR-126 or miR-145. For this reason, we focused on miR-126 and miR-145 in assessing the possible hematological role(s) of miRNAs targeting the Hoxa9 homeobox region. To assess the possible regulation of Hoxa9 by miR-126 and miR-145, expression was measured in BM progenitor cells. FACS of lineage-depleted BM cells was used to fractionate the CD34+ population into stem and early progenitor pools. QRT-PCR analysis showed that miR-126 expression was strongest in the HSC pools and was downregulated as the cells progressed to more-committed progenitors (Fig. (Fig.4).4). The pattern of Hoxa9 expression closely parallelled that of miR-126, showing strong expression in HSCs and a gradual decrease in more-committed myeloid and erythroid progenitors. Since only extremely low levels of miR-145 were detected in a random pattern in the various pools (data not shown), we decided to focus on the role of miR-126 in the possible regulation of HOXA9.
Northern blotting was used to further explore the expression of miR-126 and Hoxa9 transcripts in human leukemic cell lines. We previously reported that the Hoxa9 gene is expressed in almost all leukemic cells with myelomonocytic features and is absent in almost all cell lines with erythroid/megakaryocytic or lymphoid phenotypes (34). miR-126 expression was not detected in the myeloid cell lines (Fig. (Fig.4C),4C), which was a pattern inverse to the previously observed pattern of Hoxa9 mRNA expression. However, miR-126 is strongly expressed in human erythroid and megakaryocytic cell lines, which express little or no Hoxa9 (Fig. (Fig.4C4C and Table Table4)4) (34).
To further explore the possible relationship between miR-126 and Hoxa9 expression, total RNA was isolated from BM samples from patients with acute promyelocytic leukemias. We noted that the samples with the highest Hoxa9 mRNA expression levels were low in miR-126, while samples with low Hoxa9 levels had some of the highest miR-126 levels (Table (Table4).4). Since, as discussed below, there are multiple pathways known to regulate Hoxa9 expression in leukemias, it is not surprising that there was not a statistically significant inverse correlation between Hoxa9 and miR-126 RNA levels. However, these data reenforce the cell line data that suggest a possible inverse relationship between miR-126 and Hoxa9 expression in leukemias. To assess whether the levels of miR-126 achieved by the viral vector were of similar magnitude to endogenous miR-126 expression, a series of human and murine cell lines were studied. As shown in Table Table4,4, miR-126 expressed from the triple vector was of the same order of magnitude as that observed in human nonmyeloid cell lines and some murine lines. The viral-vector-driven expression was ~50-fold higher than that measured in normal murine marrow or HSC progenitors but of similar magnitude to that detected in human erythro/megakaryocytic cells and several human leukemia samples.
To test the capability of miR-126 to regulate endogenous HOXA9 protein, we utilized a BM-derived cell line transformed with the MLL-ENL oncogene that had previously been shown to express high levels of endogenous HOXA9 (Fig. (Fig.5A,5A, lane 1) (49). Introduction of miR-126 alone into MLL-ENL immortalized cells knocked down the endogenous HOXA9 protein, confirming that miR-126, probably together with endogenous let-7s, can target the endogenous full-length Hoxa9 transcript. It was not possible to replicate this finding in the well-studied myeloid leukemic lines that express high levels of Hoxa9 mRNA (34), because surprisingly, none of these cells express sufficient HOXA9 protein for experimental manipulation (for an example, see reference 48).
To demonstrate that endogenous miR-126 regulates endogenous HOXA9, we chose to block miR-126 by using an RNA inhibitor (Dharmacon catalog no. I-310017-02). Since this approach requires the capacity to transfect at a high efficiency, primary BM cells and hematopoietic cell lines were unsuitable, due to their well-known poor transfectability. F9 cells were chosen because they express low but detectable levels of the HOXA9 protein as well as moderate levels of miR-126. Preliminary experiments were performed with a fluorescently labeled small RNA (siGLO Green; Dharmacon) to establish conditions in which ~95% of the F9 cells were transfected. Transient transfection with the miR-126 inhibitor resulted in increased HOXA9 protein relative to a control antisense RNA in a dose response fashion for the two higher doses used (Fig. (Fig.5B).5B). Separate replicate experiments using the middle (100 nM) dose, recommended by the manufacturer, produced 1.9-fold and 1.6-fold increases in HOXA9 protein (data available on request). These data confirm that endogenous miR-126 can regulate endogenous HOXA9 protein levels.
Relatively little is known as to how the Hox transcripts and/or their protein products, HOXA9 in particular, are regulated at the posttranscriptional level. In the current study, we present data suggesting a regulatory role for miR-126 in modulating the levels of the HOXA9 protein through binding to the homeobox region of the mRNA transcript. Although several previous examples of miRNA targeting of Hox transcripts have been reported (reviewed in reference 8), all describe targeting the 3′ UTRs. Although we have focused on miR-126 because of its expression relative to Hoxa9 in normal and leukemic hematopoietic cells, the broad expression of Hoxa9 in other embryonic and adult tissues presents the possibility that miR-145 and/or let-7s together with or separately from miR-126 may regulate Hoxa9 in other biologically important tissues.
We believe the current report to be one of the first to show examples of miRNAs targeting the coding region of a mammalian transcript. The earlier literature on miRNAs suggested that most miRNAs target the 3′ UTRs (39), a tendency that was reenforced by the fact that all of the earlier programs focused on the 3′ UTRs for target prediction (23, 28, 35). This project, which focuses on the Hoxa9 homeobox region, was initiated following a serendipitous observation that one of these earlier programs treated a portion of the Hoxa9 coding region as part of the 3′ UTR and predicted miRNA targets within the homeobox. Although there are substantial experimental data demonstrating that the 3′ UTR can mediate translational repression (reviewed in reference 47), there is no a priori reason to exclude possible targeting to the coding regions. Indeed, manipulation of let-7 binding sites from the 3′ UTR to the coding region did not prevent translational silencing in a model system (25). Although several programs that predict miRNA targets within the coding regions of mammalian genes have recently been described (20, 44), none of the programs predicted the three targeting miRNAs that we have discovered, illustrating the difficulty in making predictions given the limited seed requirements for miRNAs.
Much of the emphasis on the 3′ UTRs apparently derived from the use of sequence conservation across species as a major predictor of miRNA targets, coupled with the presumption that conservation of functional coding regions would confound target identification. Indeed, the sequence conservation of the proposed miRNAs we identified might merely reflect the canonical conservation of HOX homeobox coding regions. However, given that first- and third-base codon changes can permit amino acid maintenance while allowing substantial nucleotide drift, the observation of very high nucleotide conservation of the miRNA target site nucleotides, accompanied by the expected variation at non-miRNA contact bases, across eight Hoxa9 homeoboxes (Fig. (Fig.11 and Table Table2)2) implies functionality beyond conservation of amino acid identity. Thus, we propose that the miRNA targeting of the Hoxa9 homeobox is an ancient biological regulatory mechanism.
Grimson et al. (19) recently described an improved algorithm to predict miRNA targets based on five factors: (i) AU-rich composition near the target site, (ii) proximity to other miRNA sites; (iii) pairing to nt 13 to 16 within the miRNA, (iv) positioning within the 3′ UTR relative to the stop codon, and (v) positioning away from the center of the 3′ UTR. While parameters iv and v do not apply to the current miRNA targets, the other three conditions are at least partially met. The bias for AU-rich regions is consistent with a report of correlation between low ΔG values for 70-nt flanking sequences and miRNA targets (50). Given that homeoboxes are GC rich (1), analysis of the ΔG for 70-nt regions across the Hoxa9 cDNA did not identify the three sites within the homeobox compared to the overall sequence (C. Largman, unpublished data). However, the regions immediately around the three sites show that the miR-145 site is flanked by 7 of 10 and 6 of 9 A/Ts, while the 5′ flanking regions of the let-7a and miR-126 sites contain 6 of 9 and 9 of 11 A/Ts, respectively. Thus, there are localized low-energy mRNA structures adjacent to the three Hoxa9 miRNA targets that may enhance complex formation. The let-7 site is 9 nt from the miR-145 binding site, while the binding site of miR-126 overlaps that of let-7 and is only 17 nt away from the miR-145 site. With regard to binding to sites 13 to 16 within the miRNAs, miR-126 forms a perfect match with the Hoxa9 sequence at nt 13 to 16; let-7a can form matches at nt 12, 13, 14, 17, 18, and 19; and in miR-126, nt 14 and 15 can bind to the Hoxa9 target.
We were able to readily demonstrate downregulation of either the retrovirally expressed or endogenous HOXA9 protein in blood cells by exogenous miR-126. However, the well-known inability of miR-126 to efficiently transfect hematopoietic cells precluded showing regulation by endogenous miR-126 in blood cells. We therefore used the nonhematopoietic F9 line as a suitable test system. Despite incomplete transfection and possible inefficient inhibition of the endogenous miR-126, F9 cells transfected with the inhibitor showed a mean 2.3-fold increase in HOXA9 protein, demonstrating that the endogenous protein is regulated by endogenous miR-126. Since there are no other putative miR-126 sites within the full-length Hoxa9 transcript, these data strongly suggest that miR-126 regulates endogenous HOXA9 protein by binding to the site we have identified within the homeobox.
While this work was in progress, two studies described analysis of miRNAs in hematopoietic progenitor cells (31, 40). Although Landgraf et al. reported qualitative detection of miR-126 in the CD34+ pool that contains HSCs (31), they did not examine expression in purified common-myeloid, megakaryocyte-erythrocyte, or granulocyte-monocyte early-myeloid-progenitor compartments. Monticelli et al. (40) did not report detection of either miR-126 or miR-145 in their samples. miR-126 has also been reported to be downregulated during terminal megakaryocytopoiesis but upregulated in megakaryocytic cell lines (16). Our data show that miR-126 expression parallels that of Hoxa9 mRNA in normal murine BM; both are relatively elevated in the HSC compartment and decrease during early-myeloid-progenitor cell differentiation. Low levels of miR-145 are present in all pools. These data suggest that miR-126 may keep HOXA9 levels in check during normal hematopoietic development. We reason that if miR-126 functions to block translation of Hoxa9 mRNA, then both RNAs should be present in the same cells. Unfortunately, the very low levels of HOXA9 protein expression and the low levels of HSC progenitors preclude analysis of whether the HOXA9 protein is altered by miR-126 in normal HSCs. In contrast to the parallel expressions of miR-126 and Hoxa9 in normal HSCs, there was a partial inverse expression pattern in leukemic cell lines and primary leukemia samples. Although this inverse relationship did not achieve a significant regression, this was not anticipated perhaps because there are numerous mechanisms reported to cause high-level Hoxa9 expression (3, 10, 41, 42, 49).
There is increasing evidence for the misregulation of miRNAs in cancer (38). It is of interest that human miR-126 is located on chromosome 9.34.3. Chromosome 9q34 appears to be a relative hot spot for chromosomal breakpoints associated with both myeloid and lymphoid leukemias (http://AtlasGeneticsOncology.org). The majority of leukemias associated with this region are chronic myelogenous leukemias arising from the t(9;22)(q34;q11) translocation which results in the Bcr-Abl fusion (reviewed in reference 9). In addition, there are a range of other leukemias that have been associated with the loss or rearrangement of 9q34. While 9q34 remains a relatively large region, it is unclear if breakpoints distal to the miR-126 locus can influence its expression. In this regard, a recent study showed that Hoxa9 was upregulated in BM samples of patients with chronic myelogenous leukemia arising from the Bcr-Abl fusion (9). There is now substantial data supporting a role for Hoxa9 in myeloid leukemias (reviewed in reference 22). Our data are consistent with a mechanism in which the loss of miR-126 expression would lead to increased levels of the HOXA9 protein through changes in mRNA and/or protein levels.
miRNA modulation of HOXA9 protein levels might confer an additional level of regulation to tissues in which the paralogous Hoxa9 genes are simultaneously expressed. Indeed, one of the long-recognized but perplexing aspects of Hox gene expression is that multiple members of a single paralogue, as well as those of closely related paralogues, are coexpressed in the same tissues during development. For instance, all of the Hox-10 and Hox-11 members are coexpressed in the developing skeleton, while three Hox-9 genes, including Hoxa9, are expressed in parallel in the developing mammary gland (6). Because all members of a paralogue would be expected to bind to the same DNA elements (45), such coexpression would potentially lead to redundant functions for these transcription factors. Almost all of the data defining Hox gene spatial patterning has been derived from in situ hybridization of mRNAs (30), with only a few examples of HOX protein expression analysis (26, 27). The finding that the miRNAs target the HOXA9 protein has implications for whether the protein products of other Hox genes actually parallel the previously defined mRNA expression patterns. One of the conclusions of the current study is that a reexamination of Hox gene expression at the protein level in various developmental programs may be warranted.
This work was supported by the Research Service of the Department of Veterans Affairs (C.L.) and NCI grant CA80029 (C.L.).
We thank R. Slany for the MLL-ENL cell line, C.-Z. Chen for the MDH vector, M. Loh for providing samples from the UCSF leukemic bone marrow cell bank, R. K. Humphries, M. McManus, and H. J. Lawrence for helpful discussions, and S. Fong for excellent technical support.
Published ahead of print on 12 May 2008.