Isoprenoid biosynthesis is required for miRNA function and affects membrane association of ARGONAUTE 1 in Arabidopsis

Abstract

Plant and metazoan microRNAs (miRNAs) guide ARGONAUTE (AGO) protein complexes to regulate expression of complementary RNAs via base pairing. In the plant Arabidopsis thaliana, the main miRNA effector is AGO1, but few other factors required for miRNA activity are known. Here, we isolate the genes defined by the previously described miRNA action deficient (mad) mutants, mad3 and mad4. Both genes encode enzymes involved in isoprenoid biosynthesis. MAD3 encodes 3-hydroxy-3-methylglutaryl CoA reductase (HMG1), which functions in the initial C₅ building block biogenesis that precedes isoprenoid metabolism. HMG1 is a key regulatory enzyme that controls the amounts of isoprenoid end products. MAD4 encodes sterol C-8 isomerase (HYDRA1) that acts downstream in dedicated sterol biosynthesis. Using yeast complementation assays and in planta application of lovastatin, a competitive inhibitor of HMG1, we show that defects in HMG1 catalytic activity are sufficient to inhibit miRNA activity. Many isoprenoid derivatives are indispensable structural and signaling components, and especially sterols are essential membrane constituents. Accordingly, we provide evidence that AGO1 is a peripheral membrane protein. Moreover, specific hypomorphic mutant alleles of AGO1 display compromised membrane association and AGO1-membrane interaction is reduced upon knockdown of HMG1/MAD3. These results suggest a possible basis for the requirement of isoprenoid biosynthesis for the activity of plant miRNAs, and unravel mechanistic features shared with their metazoan counterparts.

MicroRNAs (miRNAs) are 20–25 nt small noncoding RNAs that regulate gene expression posttranscriptionally and control many biological functions, including development and stress adaptation both in plants and animals (1, 2). To effect gene regulation, miRNAs associate with ARGONAUTE (AGO) proteins, and guide an RNA-induced silencing complex (RISC) to complementary mRNAs via base-pairing. Recruitment of miRNA-loaded AGO as part of RISC promotes mRNA repression via enhanced turnover, translational repression, or a combination of both. In Arabidopsis thaliana, miRNAs preferentially associate with AGO1, 1 of 10 AGO paralogues. mRNA target degradation is believed to occur by AGO1-catalyzed endonucleolysis (“slicing”) potentiated by the high degree of complementarity between miRNAs and their targets in plants. Unsliced target mRNAs also undergo translational repression through undetermined mechanisms (1). In contrast, slicing is generally excluded in animals because of poor target mRNA:miRNA complementarity, and degradation results from increased bulk mRNA decay in parallel, or subsequent to, translational repression (2). Mammalian AGO2 was originally identified as a 95-kDa protein present in both soluble and membrane fractions; AGO2 was found peripherally attached to the golgi and endoplasmic reticulum, and was accordingly named GERp95 (3). More recent studies showed that AGO proteins in Drosophila and mammalian cells associate with late endosomes/multivesicular bodies (MVBs), and that protein sorting into MVBs is required for miRNA activity, perhaps by facilitating AGO loading with small RNAs (sRNAs) or RISC assembly (4, 5). Whether plant AGOs and miRNAs similarly associate with membranes remains unknown and, more generally, insights into the requirements for membrane association of AGOs are lacking in all organisms.

Isoprenoids are central metabolites in plant biology; they derive from the isomeric C₅ precursors isopentenyldiphosphate (IPP) and dimethylallyldiphosphate, and include key membrane constituents such as sterols, as well as C₁₅ and C₂₀ prenyl anchors that influence membrane association of many proteins by posttranslational modification. In plants, IPP can be synthesized via the cytoplasmic mevalonate (MVA) pathway, conserved in some prokaryotes and in all eukaryotes. The MVA pathway is tightly controlled at the step of MVA synthesis, catalyzed by the central enzyme 3-hydroxy-3-methylglutaryl CoA reductase (HMGR). Plants also use the plastidial methylerythritol 4-phosphate (MEP) pathway for IPP synthesis, common to eubacteria and algae (6). Cytoplasmic and plastidial IPP pools can interchange, but it is not entirely clear to what extent, and under which conditions they do so (7, 8).

We report here the cloning and characterization of two genes defined by miRNA-action deficient (mad) mutations previously isolated in a forward genetic screen in A. thaliana (9). MAD3 encodes the MVA pathway enzyme HMGR, and MAD4 encodes sterol C-8 isomerase in dedicated sterol biosynthesis. We show that at least two functions of AGO1, miRNA-directed mRNA repression and antiviral defense, are compromised in these genetic backgrounds. We further provide evidence that AGO1 is a peripheral membrane protein and identify AGO1 mutant alleles defective in membrane association. Moreover, AGO1 membrane association is decreased in a knockdown mutant of HMGR, suggesting a molecular basis for compromised miRNA function in mad3 and mad4 mutants. These results unravel links between isoprenoid and miRNA pathways, and suggest that membrane association of AGOs constitutes an important component of both plant and metazoan miRNA function.

Results

mad3 Is Defective in miRNA- and siRNA-Mediated Silencing.

To isolate mad mutants in Arabidopsis, we used a constitutively expressed GFP transgene silenced by endogenous miR171 (GFP171.1) and selected mutants showing GFP reactivation (9). The mutant mad3 showed robust GFP reactivation at both mRNA and protein levels, with little or no reduction in the levels of several miRNAs including miR171 (9). The high GFP levels in mad3 result from defective miR171-guided repression, because accumulation of a control GFP mRNA without the miR171 target site (GFPno miR) was equally high in mad3 mutant and WT seedlings (Fig. 1A). Accordingly, mad3 displayed increased protein and mRNA accumulation of endogenous miRNA targets, including Cu/Zn superoxide dismutase 2 (CSD2) (targeted by miR398), scarecrow-like transcription factor 6 (SCL6)-IV (miR171), and APS (miR395) (Fig.1 B–D). Contrary to other miRNAs analyzed previously, we observed a ∼40% reduction in miR398 levels in mad3 (Fig. 1B and Fig. S1), consistent with the unusual sensitivity of miR398 accumulation to perturbations at the miRNA effector level (10). We also detected a modest reduction (∼15%) in miR171 accumulation (Fig. 1C and Fig. S1) that was not seen in our previous analyses, perhaps because of differences between the second (9) and fifth backcrosses of mad3 analyzed here. MAD3 is also required for AGO1-mediated antiviral silencing, because recombinant Tobacco rattle virus (TRV) containing a fragment of phytoene desaturase (PDS) failed to induce PDS silencing, assessed by photobleaching, in mad3 mutant leaves compared with WT leaves (Fig. 1E). Moreover, although similar levels of viral siRNAs were produced in mad3 and WT leaves, the accumulation of TRV-PDS RNA was significantly higher in mad3 tissues (Fig. 1F), suggesting an alteration of AGO1 antiviral activity downstream of viral siRNA production, consistent with small RNA activity rather than biogenesis being compromised in mad3.

Fig. 1.

RNA silencing related defects in mad3. (A) RNA blots showing accumulation of GFP mRNA in MAD3 WT and mad3 mutant seedlings expressing either miR171-targeted (GFP171.1) or nontargeted (GFP no miR) GFP transcripts. Ethidium bromide stained ribosomal RNAs provides a control for equal loading of total RNA. (B) (Top) Western blots of total protein extracts from seedlings of GFP171.1, mad3, and dcl1-12 probed with CSD2 antisera. Coomassie-stained RbcL provides a control for equal loading. (Middle) RNA blots showing accumulation of CSD2 mRNA. (Bottom) RNA blots of miR398 accumulation. U6: control for equal loading. (C and D) Protein and RNA analyses analogous to those shown in B for the miRNA:mRNA target pairs miR171:SCL6-IV and miR395:APS1, respectively. mRNAs were quantified by real-time PCR, normalized to 18S rRNA levels. Error bars indicate SDs calculated from triplicate samples. The same Northern blot was used for consecutive hybridisations with miR398, miR171 and U6. miR395 levels were below the detection limit of Northern blots and are not shown. The APS antibody recognizes all four isoforms APS1 to -4. Total RNA and total protein was extracted from the same tissue for B–D; the intensity of all signals probed with the same antibody or radiolabeled nucleotides are directly comparable in all cases, as they come from the same gels containing larger series of mad mutants. (E) Virus-induced gene silencing (VIGS) assays on GFP171.1 and mad3. TRV containing a PDS segment was inoculated on three to four leaves, and VIGS, manifested as photobleaching, was scored 8 d later (SI Materials and Methods). (Scale bars, 2 cm.) (F) Northern analysis of viral RNA2 and siRNA accumulation in systemic leaves with U6 as a loading control. All lanes are from the same larger gel, and RNA2 and U6 signal intensities are directly comparable.

Accumulation and Loading of AGO1 Are Unaltered in mad3.

The mutant mad3 exhibits at least as strong silencing defects as the hypomorphic dcl1-12 allele, yet does not display as drastic changes in small RNA levels (Fig. 1 B–D and F) (9), suggesting that accumulation of the small RNA effector, AGO1, or its binding to miRNAs, is altered in mad3. Inflorescences are better suited than seedlings to analyze these questions because of their AGO1 abundance, so we first used the GFP171.1 sensor to verify that the mad3 phenotype manifests itself in inflorescences (Fig. S2). AGO1 accumulation and steady-state amount of bound miRNA were unchanged between inflorescences from mad3 and the parental line (Fig. 2A), suggesting a defect in either miRNA RISC assembly or in miRNA activity as part of assembled RISC. Because the AGO1-RISC protein composition is unknown, we only tested the latter hypothesis, exploiting the fact that plant miRNA action is in part mediated by AGO1-catalyzed slicing (11). The accumulation of TAS1 and TAS2 trans-acting (ta) siRNAs depends on slicing by AGO1-loaded miR173 (12). tasiRNA biosynthesis also requires RNA-dependent RNA polymerase 6 (RDR6) (13) that is mutated in the GFP171.1 parental line to avoid production of secondary GFP siRNAs from the miR171 sensor (14). We thus constructed mad3/RDR6 genetic backgrounds to test accumulation of the TAS1-derived siR255, which was unchanged in mad3 compared with WT plants (Fig. 2B). Collectively, these analyses position the mad3 defects downstream of AGO1 loading, and suggest that MAD3 is either not required for miRNA-guided slicing, or that slicing by miR173, in contrast to other miRNAs, might occur independently of MAD3.

Fig. 2.

MAD3 encodes HMG1. (A) (Upper) Accumulation of AGO1 protein in total protein extracts before immunoprecipitation (IP). (Lower) Analysis of miRNA and AGO1 levels in Protein A-agarose fractions following incubation with AGO1 antibodies (AGO1 IP) or buffer (mock IP). Spike refers to an oligonucleotide (50 pmol) that was added to each washed immunoprecipitate before RNA extraction to control for equal RNA yield. (B) RNA blot analysis of the accumulation of the TAS1-derived tasiRNA siR255 and miR173 in mad3 and dcl1-12 from which the rdr6 mutation had been outcrossed. U6: control for total RNA loading. (C) The schematic shows the positions of hmg1 mutant alleles used in this study. Gray boxes: untranslated exon regions; black boxes: coding exon regions; lines: introns. T-DNA insertions are shown by red triangles, the mad3 point mutation (R458H) by a red line. The T-DNA insertion in hmg1-3 is 35-bp downstream of the stop codon. Complementation of mad3 by a genomic DNA fragment containing HMG1. Results of a T2 family segregating WT and mutant (mut) individuals are shown. (Upper) Northern blots showing GFP mRNA accumulation in WT and mutant. Parental lines GFP171.1.1 and mad3 are included for direct comparison. (Lower) Genotyping of WT and mutant individuals using markers detecting two mutations specific to mad3 in HMG1 (At1g76490) and in At1g76410. WT individuals are mad3 homozygous (only mutant alleles at At1g76410), and contain the HMG1 transgene (mutant and WT alleles at At1g76490). Mutant individuals are mad3 homozygous, but do not contain the HMG1 transgene (only mutant alleles at both At1g76410 and At1g76490). (D) Western analysis of the miR398 target, CSD2, in mad3, and mad3 expressing the HMG1 transgene. (E) Western analysis of CSD2 accumulation in hmg1-1 and mad3. hmg1-1 seedlings with a WT morphology are denoted “normal”, hmg1-1 individuals with a mad3-like morphology are denoted “phenotype.” (F) GFP accumulation in F1 progeny of crosses between mad3 homozygous and either Col-0 (Upper) or hmg1-3 (Lower). Red fluorescence is because of chlorophyll; green fluorescence is because of GFP. Numbers in lower right corners indicate the number of F1 seedlings with the phenotype shown in the picture out of total number of F1 seedlings sampled. (G) Western analysis of HMG1 accumulation in seedlings of the GFP171.1 parental line and in mad3.

MAD3 Encodes HMG1.

We isolated MAD3 by positional cloning: two genes (At1g76410 and At1g76490) in an 80-kb mapping interval had nonsynonymous G-A transitions consistent with ethyl methanesulfonate-induced lesions. No complementation was observed for At1g76410, whereas several transgenic lines expressing native At1g76490 complemented the mad3 morphological and GFP silencing phenotypes (Fig. 2C). Low accumulation of the miR398 target CSD2 was also restored in homozygous mad3 mutants transformed with At1g76490 (Fig. 2D), indicating that At1g76490 is MAD3. At1g76490 encodes HMG1, and mad3 contains a missense mutation in the third exon, causing the amino acid substitution R458H (Fig. 2C).

A knockout allele, hmg1-1 previously isolated in ecotype Ws-0 (15) was further analyzed for possible miRNA-related defects. Ninety-percent of homozygous hmg1-1 seedlings developed normally, but 10% showed typical mad3 morphological defects, including stunted growth and narrow, undeveloped true leaves (see Fig. 3C for the mad3 phenotype). Molecular analyses of these two subpopulations showed that seedlings with a WT phenotype had normal levels of the miR398 target CSD2, although abnormal individuals showed distinctly higher CSD2 levels, as in mad3 (Fig. 2E). Similarly, an hmg1 knockdown allele from the SALK collection (16) [ecotype Col-0, SALK_001128, hereafter referred to as hmg1-3 (Fig. S3A)] showed distinct subpopulations of morphologically WT and mad3-like seedlings; hmg1-3 seedlings with mad3-like phenotypes had higher CSD2 levels than WT (Fig. S3B). When homozygous hmg1-3 was crossed to homozygous mad3, 6 of 11 F1 progenies showed noncomplementation with GFP reactivation and morphological defects similar to mad3 parents (Fig. 2F), and 5 of 11 F1 seedlings were WT. All F1 progenies of mad3 crossed to Col-0 were WT, ruling out that mad3 is semidominant in Col-0 (Fig. 2F). Therefore, hmg1 insertion mutants produce mad-related phenotypes with incomplete penetrance. Because mad3 mutants are sterile, we could not carry out similar analyses on homozygous mad3 populations. Although we observed, in some cases, segregation ratios of WT:mutant substantially greater than 3:1 in progenies from a heterozygous mad3 parent, systematic genotyping of 192 WT progenies failed to identify any homozygous mad3 mutants, suggesting that the mad3 R458H point mutant is fully penetrant. The underrepresentation of homozygous mutants may be a result of reduced gametophyte viability, because hmg1 hmg2 double-mutants are male gametophytic-lethal (17). Molecularly, the mad3 missense allele also behaved differently from insertion mutants, because the levels of mutant hmg1^R458H protein were substantially elevated in mad3 compared with WT (Fig. 2G), but HMG1 protein levels were below detection limit in hmg1-1 (Fig. S4).

Fig. 3.

Catalytic defects in HMG1 underlie the mad3 phenotype. (A) HMG1 encoded by mad3 (hmg1 R458H) is catalytically defective. (Upper) Spotting assays of S. cerevisiae Δhmg1Δhmg2 strains transformed with empty vector (EV), A. thaliana HMG1 (HMG1 WT), or A. thaliana HMG1 containing the R458H mutation (hmg1 R458H). Tenfold serial dilutions were spotted on medium supplemented with 5 mg/mL MVA (+MVA), or medium without supplement (−MVA). (Lower) Western analysis of HMG1 in total protein extracts from yeast cells grown in the presence of MVA. (B) Quantification of total sterols in leaves and in flowers of GFP171.1 and mad3, expressed as milligram per gram dry weight (DW). The histogram shows the results of one of three independent biological replicates that all showed the same trend. A detailed list of the sterol species quantified by gas chromatography (GC)-flame ionization detector and identified by GC-mass spectrometry can be found in Table S1. (C) Visual (Upper) and GFP fluorescence (Lower) phenotypes of mad3 grown on MS medium (Left), and GFP171.1.1 grown on MS supplemented with DMSO (Right, mock) or with 200 nM lovastatin (Center). Thirty-percent of the lovastatin-treated population had the phenotype shown, the remaining 70% had shoots similar to mock-treated seedlings. (D) Northern analysis of GFP mRNA and miR171 in seedlings grown for 18 d on MS supplemented with 9 ppm DMSO (mock) or with 200 nM lovastatin in DMSO. Lovastatin-grown seedlings showing a mad3-like phenotype (lova 1) were analyzed separately from those displaying WT morphology (lova 2). Transgenic lines expressing GFP targeted by miR171 (GFP171.1), and nontargeted GFP (GFP no miR) were analyzed.

HMG1 Catalytic Activity Is Required for miR171 Function.

Although in the primary structure, the conserved R458 residue mutated in mad3 is situated within 20 aa of key catalytic and substrate binding residues, it does not have a characterized function in either substrate/cofactor binding or catalysis (18) (Fig. S5). It was, therefore, unclear if the R458H mutation affects HMG1 catalytic activity, or if it might reveal a new HMG1 function in the miRNA pathway, unrelated to its biochemical function in isoprenoid biosynthesis. To address this issue, we exploited the fact that lethality of Saccharomyces cerevisiae Δhmg1Δhmg2 double-mutants is rescued by WT Arabidopsis HMG1 (19). The Δhmg1Δhmg2 double-mutant is also partially rescued by exogenous application of MVA, the direct product of the HMG1-catalyzed reaction. As previously shown, WT HMG1 fully complemented Δhmg1Δhmg2, regardless of MVA availability in the medium. In contrast, HMG1^R458H displayed clear growth defects on MVA-free medium (Fig. 3A). Because accumulation of both mutant and WT HMG1 proteins was comparable (Fig. 3A), we conclude that mad3-encoded HMG1^R458H has strongly reduced, but not fully abolished, catalytic activity. Accordingly, sterol quantification in mad3 plants showed a reduction by 40% in leaves and 60% in flowers (Fig. 3B), confirming that isoprenoid metabolism is compromised in mad3. To test if inhibition of HMG1 enzymatic activity is sufficient to reduce miRNA activity, we germinated seeds of GFP171.1 and GFPno miR on plates containing lovastatin, a specific inhibitor of HMGs, including Arabidopsis HMG1. Lovastatin had effects similar to hmg1 knockout mutations: ∼30% of the treated population exhibited morphological defects resembling those of mad3 and displayed markedly higher GFP accumulation in GFP171.1 (Fig. 3C), unlike treated plants with a WT phenotype. The high GFP levels were a result of defective miR171 action, because miR171 levels were not overtly affected. Moreover, GFP levels in abnormal individuals of the GFPno miR line were unchanged compared with those of treated plants with a WT phenotype, or untreated individuals (Fig. 3D). Stochasticity is unlikely to be caused by unequal drug uptake among individuals, because all seedlings showed strong root growth defects typical of HMG inhibition by lovastatin (20). We conclude that the MVA pathway is required for miRNA function.

Sterol Biosynthesis Is Required for a Functional miRNA Pathway.

Several classes of MVA-dependent isoprenoid metabolites could contribute to the miRNA-action defects of mad3 mutants, including sterols, essential components of all eukaryotic membranes. Further study of the previously isolated mad4 mutant provided evidence that sterols are indeed required for miRNA function: mad4 is pleiotropic, with a strong morphological and seedling lethal phenotype (9); mad4 seedlings are small, dark-green, and show misshapen structures in place of cotyledons and leaves in addition to strongly inhibited root development; mad4 plants exhibit robust reactivation of GFP171.1 at mRNA and protein levels, with unchanged miR171 levels (9). GFP171.1 reactivation in mad4 was a result of defective miR171 activity because, in contrast to the miR171-targeted GFP171.1 transcript, the GFPno miR transcript showed only marginally higher accumulation in mad4 compared with WT (Fig. 4A). Positional cloning of mad4 showed that it contains a G-A transition producing a nonsense mutation (W95STOP) in HYDRA1 (HYD1, At1g20050) (Fig. 4A), encoding sterol C-8 isomerase (21). Previously described hyd1 mutants show morphological phenotypes identical to mad4 (21). Moreover, HYD1 transgenic expression completely rescued GFP overaccumulation in mad4 homozygous mutants (Fig. 4B), indicating that HYD1 is MAD4.

Fig. 4.

MAD4 encodes sterol C8-isomerase. (A) Northern analysis of GFP mRNA targeted by miR171 (GFP171.1) or not (GFPno miR) in mad4. The schematic below shows the position of the nonsense mutation in mad4; gene nomenclature is the same as in Fig. 2C. (B) Complementation of mad4 by HYD1. GFP mRNA and GFP protein analyses of a T2 family segregating WT and mutant individuals. WT individuals were mad4 homozygous and contained the HYD1 transgene, because genotyping of 24 individual T2 seedlings with WT phenotype showed that they all contained mutant and WT MAD4 alleles, excluding that the T2 family was segregating for mad4 (Fig. S7).

AGO1 Is a Peripheral Membrane Protein.

Why are genes required for isoprenoid biosynthesis necessary for plant miRNA function? Isoprenoids are essential for correct functioning of membrane compartments and for membrane association/trafficking of several proteins to their destinations. Although disruption of isoprenoid biosynthesis may cause multiple cellular defects, the fact that AGO-containing silencing complexes associate with endomembranes in animal cells (4, 5) suggests that the miRNA activity defects in mad3 and mad4 could be linked to compromised membrane functions. Therefore, we tested if Arabidopsis AGO1 is membrane-associated.

Analysis of soluble and insoluble fractions upon hypertonic lysis that fragments all membrane compartments into microsomes showed that a sizeable fraction of AGO1 is in the insoluble pellet (Fig. 5A). This pellet contains the transmembrane proteins HMG1 (22) (Fig. 5A) and plasmodesma located protein (23) (Fig. S6A), but is free of the cytoplasmic, soluble protein phosphoenolpyruvat carboxylase (PEPC) (Fig. 5A) (24). A large proportion of AGO1 could be extracted from the insoluble pellet by low concentrations of mild detergent (0.5% Triton X-100), consistent with AGO1 being membrane associated rather than bound to other insoluble or aggregated material (Fig. 5B). In contrast to HMG1, which has two transmembrane segments (25), AGO1 was also extracted from the insoluble pellet upon washes in high-salt or alkaline solutions (Fig. 5B), indicating that it is a peripheral membrane protein that does not use hydrophobic interactions to associate with membranes.

Fig. 5.

AGO1 is a peripheral membrane protein. (A) Western analysis of AGO1, HMG1, and PEPC proteins in 100,000 × g supernatant (sup) and pellet (pel) fractions of cleared Col-0 inflorescence extracts prepared by hypertonic lysis. Five-percent of the supernatant fraction is loaded, 20% of pellet fraction is loaded, precluding any clear estimates of relative abundance in soluble versus insoluble fractions. HMG1 is used here solely as a positive control for a transmembrane protein. (B) Western analysis of AGO1 and HMG1 proteins in pellet fractions prepared as in A. Pellets were resuspended in either microsome buffer, or microsome buffer supplemented with 1 M KCl, 0.1 M Na₂CO₃, or 0.5% Triton X-100, and separated into supernatant (sup) and pellet (pel) fractions again at 100,000 × g. At longer exposures, an insoluble AGO1 fraction upon Triton X-100 treatment is visible. (C) Microsome fractionation of inflorescence lysates from Col-0, ago1-25, and ago1-38. (Top) Western analysis of AGO1 in total extracts. (Middle) Western analysis of AGO1 in equally loaded microsome fractions, developed by enhanced chemiluminescence. (Bottom) Same analysis as in the Middle, developed by less sensitive alkaline phosphatase staining that allows a clearer visualization of the difference in microsomal AGO1 abundance between Col-0 (WT) and ago1-38. (D) (Upper) Schematic depicting the position of several AGO1 hypomorphic mutations used in this study (black triangle) and the ago1-38 allele (red triangle) causing the G186R missense mutation. (Lower) The glycine residue mutated in Arabidopsis ago1-38 is highly conserved among various metazoan and fungal AGO proteins (highlighted in yellow). At, Arabidopsis thaliana; Ce, Caenorhabditis elegans; Hs, Homo sapiens; Sp, Schizosaccharomyces pombe. (E) Microsome fractionation of inflorescence lysates from Col-0 and hmg1-3. (Upper) Western analysis of AGO1 in total extracts. (Lower) Western analysis of AGO1 and HMG1 in microsome (pel) fraction. (F) Same analysis as in E performed with GFP171.1 and mad3 inflorescence lysates.

Hypomorphic ago1 Alleles and hmg1 Mutants Exhibit Reduced AGO1 Membrane Association.

We next investigated if AGO1 membrane association is important for its function in the miRNA pathway by examining a series of AGO1 missense alleles available from the literature, assessing AGO1 protein accumulation in total extracts and in membrane fractions. Most alleles (ago1-4, ago1-26, ago1-27, ago1-33, eco1-3, all with mutations within the MID and Piwi domains) showed protein accumulation similar to WT in both fractions or proportional reductions in both total and membrane fractions compared with WT (Fig. S6B). Two alleles, ago1-38 and ago1-25, showed disproportionate reductions in membrane-bound AGO1 levels (Fig. 5C). ago1-25 is mutated in the Piwi domain (G758S) (26) and displays strongly reduced total AGO1 protein levels, precluding its use to address the functional relevance of AGO1 membrane association. In contrast, ago1-38 protein abundance was similar to WT AGO1 in total extracts, but decreased markedly in the membrane fractions (Fig. 5C). ago1-38 harbors a missense mutation (G186R) (27) in the uncharacterized N-terminal part of AGO1, involved in neither miRNA binding nor AGO-catalyzed slicing (Fig. 5D). These data support the notion that membrane association is important for AGO1 activity, although we cannot exclude that ago1-38 is defective in other aspects of AGO1 function. We note, however, that the sequence segment corresponding to amino acids 175–328 of Arabidopsis AGO1 is well conserved in many eukaryotic AGOs, including human Ago1-4 and Drosophila Ago1-2 that are known to associate with membranes (4, 5). In all of these proteins, except Drosophila Ago2, where the first part of the sequence segment is missing, the Gly residue corresponding to G186 in Arabidopsis AGO1 is invariable (Fig. 5D). We finally asked whether knockdown of HMG1 affects AGO1 membrane association. AGO1 accumulation was unaffected in total extracts of hmg1-3 flowers, but membrane-associated AGO1 levels were lower in hmg1-3 compared with WT (Fig. 5E). Similarly, lower levels of membrane-associated AGO1 were observed in mad3, although the difference to WT was less striking than for hmg1-3 (Fig. 5F). These data show that HMG1 is required for full membrane association of AGO1, and suggest that this association is necessary for normal AGO1 function in the miRNA pathway.

Discussion

This study shows that miRNA and membrane function are linked in plants, and suggests that the MVA pathway is particularly important for miRNA activity and membrane association of the miRNA effector, AGO1. Our identification of MAD4 as an enzyme in dedicated sterol biosynthesis identifies sterols as one class of isoprenoid metabolites required for normal miRNA functions. Sterols influence many aspects of membrane biology, including organization of sterol/sphingolipid-rich microdomains, and membrane protein trafficking/localization (28). It is not clear, at present, which of these specific activities is (are) required for miRNA function. The mad3 mutants are defective in the MVA pathway common to all isoprenoids, but exhibit only about twofold reduced levels of total membrane sterols compared with WT. In contrast, previously characterized hyd1 mutants allelic to mad4 are devoid of the major membrane sterols, sitosterol and campesterol (21). The mad3 miRNA-related phenotype is, nonetheless, at least as pronounced as in mad4, suggesting that isoprenoids other than sterols could also influence miRNA function. We note, in this regard, that 3-hydroxy-3-methylglutaryl CoA synthase (HMGS), the first enzyme in the MVA pathway, was identified in a genome-wide RNAi screen for enhancers of a weak allele of the let-7 miRNA in Caenorhabditis elegans. HMGS RNAi led to defective function of both let-7 and lin-4 miRNAs without affecting their cellular levels, suggesting a defect at the effector stage of the worm miRNA pathway (29). C. elegans does not have a de novo sterol biosynthesis pathway and uses the MVA pathway to supply lipids for protein prenylation (30).

We cannot rule out, however, that cellular changes other than membrane perturbations could contribute to the miRNA-related phenotypes of isoprenoid biosynthesis mutants. Hence, several plant hormones are IPP-derived, and brassinosteroid biosynthesis also requires sterol biosynthetic steps catalyzed by HYD1/MAD4. Nonetheless, AGO1 membrane association and its partial dependence on HMG1 suggest that perturbed membrane function underlies at least part of the mad3 and mad4 miRNA-related phenotypes.

The precise identity of the membrane compartments with which AGO1 associates remains unresolved. Indeed, our attempts to localize AGO1 in planta using GFP fusions, immunofluorescence, or electron microscopy were so far unsuccessful. It is also not clear exactly what steps in the silencing mechanism require AGO1 membrane association. In Drosophila, the BLOC-3 complex on late endosomes/lysosomes modulates AGO-loading (5), and human miRISC assembly/turnover on late MVBs coincides with selective sorting of the AGO-binding protein TNRC6 into intralumenal vesicles formed by MVB inward budding (4). Whether Arabidopsis AGO1-RISC loading and assembly similarly require its association with endosome-like membranes is a key question for further studies. Finally, we note that statins are used extensively in humans to prevent cardiovascular disease, because HMGR inhibition reduces serum cholesterol levels (31). Intriguingly, several studies have shown effects of statins on cancers and neurodegenerative diseases that may not be solely explained by lowered serum cholesterol levels (32–34). Given that the plant- and animal-silencing mechanisms both involve membrane recruitment of AGOs, the possibility emerges that some of these unexplained effects of statins might entail altered miRNA activity.

Materials and Methods

Arabidopsis Genetic Analysis.

GFP171.1, dcl1-12, mad3, and mad4 are in ecotype C24 rdr6/sde1. Genes were mapped using polymorphisms between ecotypes C24, Ler and Col-0. Strains homozygous for the wild type allele of RDR6 were constructed for analysis of tasiRNA accumulation. Details for all procedures are given in SI Materials and Methods.

Biochemical and Molecular Analyses.

Standard protocols were followed for northern blot and quantitative PCR analyses of miRNA and mRNA detection. Immunoprecipitation and western analysis of AGO1 were carried out with a monospecific polyclonal AGO1 antibody raised against an 11-amino acid peptide at the extreme N-terminus of AGO1. For yeast complementation assays, an hmg1 hmg2 double mutant was constructed by crossing single mutants obtained from EUROSCARF, and transformed with the yeast expression vector pVT102U alone, or pVT102U containing WT or R458H mutant inserts of Arabidopsis HMG1. For quantification of sterols, non-saponifiable lipids were extracted by n-hexane, acetate derivatized, and steryl acetates were quantified by gas chromatography using lupenyl-3,28-diacetate as internal standard. Microsomes were prepared by centrifugation at 100,000 × g for 30 minutes of cleared lysates prepared by hypertonic lysis. Detailed protocols for all experiments are provided in SI Materials and Methods.