Alternative use of miRNA-biogenesis co-factors in plants at low temperatures

In nature, animals can migrate or modify their behavior in response to changes in the environmental temperature. Plants, as sessile organisms, do not have this possibility. With a few exceptions, such as the eastern skunk cabbage (Ito et al., 2004), plants are not thermogenic organisms, and thus lack a homogeneous and stable internal temperature to which biochemical reactions can be optimized during evolution. Instead, most molecular and enzymatic reactions in plants have to be adjusted to the fluctuating environmental temperatures surrounding the plant. Molecular mechanisms that sense temperature, allowing quick adaptive responses, are actively studied. PHYTOCHROME B, for example, was recently described as a main sensor of temperature and light (Jung et al., 2016; Legris et al., 2016). Morphological changes, coordinated by molecular responses, help plants to cope with environmental changes. Arabidopsis thaliana, for example, is well known to display differential growth patterns when cultivated at different temperatures (Quint et al., 2016). Key regulators of such temperature-sensitive changes in plant development are microRNAs (miRNAs), 21 to 23 nucleotide long, single-stranded RNAs that regulate mRNA transcript abundance (and translation) via sequence complementarity (Liu et al., 2018). Abundance and activity of miRNAs themselves, in turn, is regulated at multiple levels and by fine-tuning elements within the miRNA biogenesis machinery.

Under regular growth chamber conditions (22-23°C, long-day photoperiod) miRNAs are produced by the catalytic action of the type III ribonuclease DICER LIKE 1 (DCL1). During this reaction, HYPONASTIC LEAVES 1 (HYL1) and SERRATE (SE) are necessary for accurate excision of the mature miRNAs from their long and complex primary transcripts (pri-miRNAs). Mutations in the genes that encode these proteins have severe consequences for miRNA production, stability and homeostasis, and, ultimately, for plant phenotypic traits (Achkar et al., 2016). Here, we report that, upon a decrease in ambient temperature, the plant miRNA biogenesis machinery seems to become more robust and produces miRNAs even in the absence of the key DCL1 co-factors HYL1 and SERRATE. This restored miRNA production reduces the morphological and reproductive defects of mutations in these co-factors, leading to restored seed production. RNA-seq analysis, followed by a novel character-extraction protocol applied to pri-miRNA sequences points towards the secondary structure of pri-miRNAs as one of the main features enabling miRNA processing at low temperatures, despite a lack of (at normal temperatures crucial) co-factor activity.

A. thaliana plants with mutations in miRNA biogenesis co-factor-coding genes such as HYL1 or SE normally display strong developmental defects and low seed production when grown under standard chamber conditions [long-day conditions (LD) 23°C; Rogers and Chen, 2013]. However, when cultivated at 16°C, the same mutant lines perform better and show an overall attenuated phenotype (Fig. 1). Rosette size, leave shape and flower development in hyl1-2 and se-3 mutants resemble the wild-type phenotype more when grown at 16°C than at 23°C (Fig. 1A and Fig. S1A,B). This is paralleled by restored seed production, a reduced embryonic abortion and increased pollen viability (Fig. 1B-E and Fig. S1C). This effect of lower temperature, however, is not evidenced in mutants that affect proteins downstream of the core miRNA-biogenesis machinery, and hence downstream of DCL1, HYL1 and SE, such as ago1-25 and hen1-5 (Fig. 1 and Fig. S1A), suggesting that the effects caused by lowering the temperature occur during the initial steps of miRNA biogenesis.

To assess whether the increase in plant fitness was related to higher miRNA production at low temperature, we used small RNA blots to quantify miRNAs. We observed that the production of several miRNAs in hyl1-2 and se-3 mutants grown at 16°C was restored to almost wild-type levels (Fig. 2A). In contrast, in ago1-25 mutants, temperature did not affect miRNA levels (Fig. 2B), which is consistent with the hypothesis that temperature affects mainly the initial steps of miRNA biogenesis and processing. Notoriously, hen1-5 mutants present an increment in miRNA accumulation at 30°C (Fig. 2B). Most miRNAs, with a few exceptions, did not show an increased accumulation at 16°C in Col-0. This could perfectly reflect the hypothesis that, in wild-type plants, where the miRNA machinery is fully active, pri-miRNAs are processed equally well at both temperatures, reaching a potential maximum of miRNA production set by the amount of pri-miRNA transcribed. A few miRNAs, miR173, miR161 and miR172, showed decreased levels at 30°C in Col-0 plants, which may reflect differential regulation (Fig. 2A,B). The last mentioned experiments were performed using plants grown at the indicated temperatures for 5 days. To discard effects caused by the delayed development expected for plants cultivated at lower temperatures, we grew Col-0 and hyl1-2 plants for 5 days at 23°C and then transferred them to 16°C for variable periods of time, and vice versa. We then quantified several miRNAs with small RNA blots and observed that lowering the environmental temperature (i.e. transfer from 23°C to 16°C) led to an increase of miRNA accumulation in hyl1-2 and se-3, even after short exposure to a lower temperature (6 to 24 h, Fig. 2C,D). These kinetics were not mirrored by a similarly fast decrease in miRNA levels in plants grown at 16°C and transferred to 23°C. Here, it took almost 4 days to see the decrease in miRNAs, possibly owing to a slower turnover of already produced miRNAs (Fig. 2C). Interestingly, miR157 showed a reduction in Col-0 after 24 h of being transferred to 16°C, prior to increasing above initial concentrations, which may reflect a complex transcriptional regulation of this miRNA under this condition (Fig. 2D). Moreover, RNA-sequencing analysis showed that the observed increase in miRNA levels at low temperature could not be explained by increased miRNA transcription, as the expression of pri-miRNAs between samples cultivated at different temperatures was stable and did not recapitulate the observed miRNA dynamics (Table S1). This analysis was confirmed using MIRNA promoters fused to the reporter gene GUS. Plants transformed with these constructs were grown at 16°C or 23°C, and the promoter activity was assayed by histochemical staining, showing unchanged promoter activity and, hence, that the transcription of the tested MIRNA promoters was not increased at 16°C (Fig. 3A). This supported the notion that it is the enhanced processing of pri-miRNAs at 16°C, even in the absence of the co-factors, rather than increased transcription of miRNA genes, that leads to the observed mutant phenotype rescue. Interestingly, the time required to observe a clear temperature-mediated effect on miRNA accumulation varies depending on the miRNA; e.g. the increment of miR164 and miR167 is observed after 6 h at 16°C (Fig. 2D), whereas for miR161 and miR156, 24 h are necessary (Fig. 2C). These differences might be related to the transcription rate of each miRNA. However, this observation may also suggest that the miRNA production recovery depends on the nature of the complex and highly variable miRNA precursor and its secondary structure.

The fact that the observed temperature-dependent miRNA recovery was similar both in hyl1-2 and se-3 mutants suggests that the lower temperature may enhance miRNA processing itself, rather than inducing the expression of potential redundant homologous factors. Although SE does not have close homologs in Arabidopsis, the HYL1 protein family has five members that can potentially act redundantly. We evaluated this possibility by quantifying the transcript levels of the closest homologs of HYL1, DRB2, DRB3, DRB4 and DRB5, in Col-0 and hyl1-2 plants grown at 16°C and 23°C, using RNA-seq. The expression of none of these genes was induced by cold in either the WT or the hyl1-2 background, which would have been an expectation if any of these homologs were helping to balance the lack of functional HYL1 in the hyl1-2 background at low temperature (Table S1). To further confirm this, we obtained hyl1-2/drb2, hyl1-2/drb3, hyl1-2/drb4 and hyl1-2/drb5 double mutants and measured miRNA levels in plants grown at both 16°C and 23°C (Fig. 3B). The miRNA accumulation in all tested double mutants mirrored that of the hyl1-2 single mutant (Fig. 3B), dismissing a potential redundancy effect triggered by temperature as causal for the observed attenuated phenotypes.

Additionally, RNA-seq analysis comparing the transcriptome of Col-0 and hyl1-2 plants grown at 16°C or 23°C also did not identify any plausible non-homolog candidate to compensate HYL1 deficiency at low temperature (Table S1). We found 492 significantly differentially expressed genes (q-value<0.05) in hyl1-2 mutants grown at 23°C compared with 16°C, 365 of them exclusive to the hyl1-2 background and not affected in Col-0 plants treated in the same way (Fig. 3C, Table S1). If any factor would be taking over HYL1 functions at 16°C in the hyl1-2 mutant background, it could be expected to be found among the 302 genes expressed at higher levels at 16°C in hyl1-2 plants (i.e. appearing downregulated in hyl1-2 at 23°C, see Fig. 3C orange balloon). A gene ontology analysis showed that among those genes, only 37 have DNA- or RNA-binding properties, but the majority belongs to transcription factor families, proteins related to DNA damage repair or translation initiation factors (Table S1), which are unlikely to display HYL1-like properties or activity. Furthermore, when specifically tested, no miRNA-biogenesis-related factors appeared differentially regulated among these temperatures (Table S1).

In addition, we observed that the production of some miRNAs was not restored by a temperature reduction (i.e. miR157) or not affected in hyl1-2 mutants regardless of the temperature (i.e. miR168) (Fig. 2A,C). Aiming to sort miRNAs into categories, we performed small RNA-seq of Col-0 and hyl1-2 plants grown at 23°C and 16°C. We then categorized miRNAs according to their responsiveness to temperature changes into two groups: (1) those with increased expression in hyl1-2 when temperature changed from 23 to 16°C (TC); and (2) those that did not increase their expression with the temperature variation (TnC) (Fig. 3D). It is worth mentioning that, owing to the identical sequence of some mature miRNAs that belong to the same family, some miRNAs may be wrongly categorized if a single member of the family ‘hides’ the others (e.g. a temperature responsive member of a family among non-responsive ones). This can potentially reduce the power of our analysis.

Based on the miRNA-group sorting, we aimed to evaluate whether structural or sequence characteristics of the corresponding pri-miRNAs can help define how a miRNA responds to temperature, and to what extent its processing is HYL1 dependent. For this, 1590 pri-miRNA features were extracted from each group using miRNAfe (Yones et al., 2015), a feature selection method based on LASSO (Table S2, Zou and Hastie, 2005). In our analysis, a positive LASSO co-efficient for a feature indicates that the feature is important for the temperature responsiveness (TC), whereas a negative coefficient indicates that the feature is important for the temperature non-responsiveness (TnC). We found that the most important features of miRNAs that increase expression in hyl1-2 background at 16°C are the presence and amount of GCA and UGCA (Fig. 3E). Interestingly, the content of paired G or C nucleotide triplets in the pri-miRNA structure appeared at the top of the rank. This suggests that such high energy paired regions of the pri-miRNA are a defining feature for HYL1 dependency during pri-miRNA processing. As temperature change will affect the folding and secondary structure of pri-miRNAs, features highly dependent on the GC content of the paired region are particularly relevant. In our analysis, other characteristics, such as the number of loops or loop size, were not found to be relevant for the classification (Table S2). However, a comparison between our pri-miRNA classification and the DCL1-slicing mechanism (Bologna et al., 2013; Chorostecki et al., 2017) showed that HYL1-independent miRNAs tend to be processed via a short base-to-loop mechanism by DCL1 (Table S3). This correlation may indicate that proper pri-miRNA slicing by DCL1 can be independent of HYL1 for short base-to-loop processed precursors, whereas HYL1 becomes necessary to guide DCL1 when multiple slicing steps are required in long complex pri-miRNAs or when slicing starts from a less organized loop region.

The association of DCL1 with SE and HYL1 during miRNA processing has been shown to be essential to ensure the ribonuclease precision at the mature miRNA excision step (Dong et al., 2008; Liu et al., 2012). Given our results, it could be expected that a change in pre-miRNA stability/folding triggered by a drop in temperature could improve miRNA processing precision. We therefore analyzed processing precision by scoring the ratio of total miRNA-matched small RNA to the pool of imprecisely processed small RNAs, defined as those only partially matching the mature miRNA sequence. As reported, hyl1-2 mutants grown at 23°C showed a drastic reduction in processing precision when compared with Col-0 at the same temperature (Fig. 4A). However, reduction in temperature drastically improved miRNA-processing precision. This is particularly true in hyl1-2 mutants, as could be anticipated from our previous results, but also for wild-type plants, which display increased precision upon temperature decrease (Fig. 4A). This suggests overall increased robustness of the processing machinery at 16°C, coinciding with increased thermodynamic stability of pri-miRNA secondary structure.

It has been recently shown that the pool of active HYL1 is actively degraded when plants are exposed to insufficient light for a prolonged period (Achkar et al., 2018). Light reduction in the shade or complete darkness, is often accompanied by a drop in temperature. We wondered how such temperature changes could affect HYL1-dependent miRNA processing when HYL1 is degraded simultaneously. To test a combined scenario of low temperature and insufficient light, we sequenced small RNAs of plants grown at 16°C or 23°C in LD photoperiod for 14 days, or grown for 10 days in LD and then transferred to the dark for 4 days, and compared the miRNA accumulation (Fig. 4B). As previously reported, we observed an overall reduced miRNA accumulation when the plants were transferred to the dark, as consequence of HYL1 degradation, but also a partial recovery of the miRNA populations when the plants were kept at 16°C (Fig. 4B). Additionally, a processing precision analysis performed in these samples showed increased processing precision in Col-0 plants when the temperature drops, even in the absence of light (Fig. 4C), a result that is harmonious with our hypothesis.

Our results demonstrate that some, but not all, miRNAs could go through alternative processing that is less dependent on DCL1 co-factors when the environmental temperature is below standard growth chambers conditions. Given the extensive distribution of Arabidopsis worldwide and the fact that 16°C is the average temperature during spring/summer in central Europe, the center of the geographic distribution of A. thaliana, our finding gains ecological relevance. Variation in HYL1 itself seems unlikely to trigger a necessity for alternative processing pathways: In a polymorphism analysis using 309 Arabidopsis thaliana accessions from across the plant's natural geographic range (1001 Genomes Consortium 2016), we found very little variation in HYL1. We could not detect any polymorphism causing nonsense mutations, and all variants causing non-synonymous mutations fall in the C-terminal end of the protein, which has been reported to be dispensable for HYL1 activity (Fig. 4D; Wu et al., 2007). This lack of natural genetic variation affecting HYL1 functionality suggests that HYL1 may be necessary for plant homeostasis even in colder climates.

However, we have recently shown that HYL1 protein levels themselves change under certain environmental conditions – prolonged periods of insufficient light induce HYL1 degradation (Achkar et al., 2018). As darkness or shading, common and highly relevant ecological conditions, often coincide with a drop in temperature, the results described here prompt us to envision a complex regulatory mechanism where dark will induce HYL1 degradation, while at the same time the associated temperature drop will make HYL1 dispensable for the processing of some miRNAs. This implies a versatile and environmentally dependent miRNA biogenesis pathway that allows the production of subsets of miRNAs under given conditions. This gives the miRNA biogenesis pathway, which was once believed to be ubiquitous and constant, large flexibility to produce subsets of miRNAs. Finally, the finding that such a temperature recovery of miRNA production is ruled by the secondary structure of the pri-miRNAs may have significant evolutionary implications as a potential adaptation-driving mechanism.

Arabidopsis thaliana Columbia (Col-0, wild type, WT) seeds were surface sterilized with 10% (v/v) bleach and 0.5% SDS, and stratified for 2–3 days at 4°C. Plants were grown at indicated temperatures, for the indicated time periods, on soil or MS-Agar plates in LD (16 h light: 8 h dark), in ARALAB D1200 PL growing chambers. Temperature fluctuations greater than 1°C could buffer the observed effect on miRNA production in the plants. Thus, in all cases, the incubator temperature was set up to oscillate over a range of less than ±1°C and temperature was monitored during the whole duration of the experiments. Mutant lines were obtained from ABRC or NASC stock centers: drb2 (N349415), drb3 (SALK_022644C), drb4 (SALK_000736), drb5 (SALK_126609.29.25.N), hen1-5 (SALK_049197), hyl1-2 (SALK_064863), se-3 (Grigg et al., 2005) and ago1-25 (Morel et al., 2002).

Total RNA was extracted using TRIZOL reagent (Life Technologies). RNA blots were made as previously described (Tomassi et al., 2017). Briefly, 1-5 µg of total RNA were resolved in 17% (v/v) polyacrylamide gels under denaturing conditions (7 M urea) and then transferred to HyBond-N+ charged nylon membranes (Amersham) by semi-dry electroblotting. RNA was covalently fixed to membranes using a UV Crosslinker. Membranes were hybridized overnight with DNA oligonucleotide probes labeled with the second generation DIG Oligonucleotide 3′-End Labeling Kit (Roche); the signal was detected using CSPD ready-to-use solution (Roche), by exposure to Amersham hyperfilm ECL (GE Healthcare Life Sciences). See Table S4 for probe details.

Small RNA libraries were prepared as indicated by the TruSeq small RNA library prep kit (Illumina). Small RNAs (50 ng) purified with the ZR small-RNA PAGE Recovery Kit (Zymo Research) were used as input for library preparation. RNA libraries were prepared using 1 μg of total RNA as input, as described by the TruSeq RNA sample prep V2 guide (Illumina). Small RNA and mRNA library size selections were performed using the BluePippin System (SAGE Science) with a 3% and 1.5% gel, respectively. Single-end Illumina sequencing was performed using a HiSeq3000. Obtained sequence data have been deposited in the European Nucleotide Archive (ENA) under the project number PRJEB28610.

Small RNA reads were first processed to remove 3′ adapters using cutadapt (version 1.9.1; Martin, 2011) and then mapped using bowtie (version 1.1.2; Langmead et al., 2009)). The references used for mapping were the databases for hairpin and mature miRNAs for A. thaliana from miRBase (release 21);, in the latter, mature miRNAs with identical sequences were collapsed into single miRNAs. Ratios for each treatment were averaged and then divided by the average ratio in Col-0 in light conditions at 23°C. Only miRNAs for which more than half the samples had more than ten counts were included in the analysis. To assess absolute miRNA expression levels, read counts on the miRNA were normalized as counts per million mapped reads, the latter being the total amount of sRNA reads mapped to the A. thaliana genome in each library. For all the reads from Col-0 and hyl1-2 plants, the log2 fold-change (LFC) for each genotype at 16°C compared to 23°C was calculated using DESeq (Anders and Huber, 2010). Reads with LFC<0 and false discovery rate adjusted P-value<0.1 at 23°C were discarded. The remaining reads were divided into two groups: (1) temperature changing (TC) sequences, with increased expression in hyl1-2 when the temperature changed from 23 to 16°C; and (2) temperature non-changing (TnC) sequences that did not increase their expression with temperature variation. The median LFC (θ) was used as a threshold for separation into these two groups. This criterion left equal numbers of reads in each group, leaving in TC those with expression levels increased by more than 80% LFC upon a temperature change from 23 to 16°C. Feature selection was based on LASSO (Zou and Hastie, 2005). The optimization criteria were L=|LFC-AXβ|2+λ2∑jβj2+λ1∑j|βj|, where X is the matrix of pre-miRNA features and A has the confidence levels for each sequence. This is a diagonal matrix with elements αi, defined according to the difference between the distance of each sequence i to the threshold θ that separates the two groups (TC and TnC), and the standard error (SE) in the detection of the miRNA within the sequence, i.e. αi=2/(1+exp(-5(|LFCi-θ|-SEi)), for |LFCi-θ|>SEi; and αi=0 otherwise (the feature is ignored in the optimization). To solve the optimization problem, a grid search was performed over λ1 and λ2, calculating the optimal β for each combination in a cross-validation scheme. Finally, the combination of λ1 and λ2 that leads to the lower mean loss is used to estimate the final β (Yones et al., 2015).

The mRNA-seq reads were first filtered by mapping them to the A. thaliana 18 s ribosomal RNA with Bowtie2 (version 2.2.6; Langmead and Salzberg, 2012). Unmapped reads were then aligned to the A. thaliana genome using Tophat2 (version 2.1.0; Kim et al., 2013) with the TAIR10 gene annotation and the following parameters: no-novel-juncs, read-edit-dist 1, read-gap-length 0, max-multihits 1. Finally, differentially expressed genes were found with cuffdiff (version 2.2.1; Trapnell et al., 2012) using the default parameters, discarding genes with fewer than 10 counts and considering differentially regulated genes as those with an FDR adjusted P-value<0.05.

Flowers were fixed (10% acetic acid, 30% chloroform, 60% ethanol; 2 h) ∼16-24 h after pollination, incubated in NaOH (4 M; 4-6 h) and washed three times [50 mM potassium phosphate (pH 7.5)]. After staining with DABS [0.1% (w/v) Aniline Blue, 108 mM K3PO4 (pH 11), decolorized overnight at 4°C in the dark before use] for 12 h, at room temperature, pistils were imaged at 40× with a stereoscope (Leica MZ10F), using 420 nm excitation and a CFP filter. To evaluate fertilization and embryo production efficiency, fully elongated and filled, but not fully ripened, siliques were opened longitudinally and imaged using a stereoscope (Nikon SMZ800).

Plants containing MIRNA Promoter:GUS constructs were grown on MS-Agar plates for 8 days at 23°C in LD conditions, and then transferred to 16°C or kept at 23°C for 6 more days until a histochemistry assay was performed. GUS activity was detected by immersing whole plants in a 1 mM 5-bromo-4-chloro-3-indolyl-glucuronic acid solution in 100 mM sodium phosphate (pH 7.0) and 0.1% Triton X-100, applying vacuum for 5 min, and subsequent overnight-incubation at 37°C. Chlorophyll was cleared from the plant tissues by immersing them in 70% ethanol. Images were taken using a Leica MZ10F. All constructs used as miRNA promoter reporters are listed in Table S5.

D.A.R., A.L.A., G.S., D.M. and P.A.M. are members of CONICET; C.Y. is a fellow of CONICET.