Genome-Wide Detection of Condition-Sensitive Alternative Splicing in Arabidopsis Roots^,,

Transcriptional profiling of roots subjected to iron and phosphate deficiency revealed stress-specific changes in splicing patterns that are largely independent of differential gene expression, providing a mechanism adapting gene activity to environmental conditions.

Abstract

Iron (Fe) deficiency is a world-wide nutritional disorder in both plants and humans, resulting from its restricted bioavailability for plants and, subsequently, low Fe concentration in edible plant parts. Plants have evolved sophisticated mechanisms to alleviate Fe deficiency, with the aim of recalibrating metabolic fluxes and maintaining cellular Fe homeostasis. To analyze condition-sensitive changes in precursor mRNA (pre-mRNA) splicing pattern, we mapped the transcriptome of Fe-deficient and Fe-sufficient Arabidopsis (Arabidopsis thaliana) roots using the RNA sequencing technology and a newly developed software toolbox, the Read Analysis & Comparison Kit in Java (RACKJ). In alternatively spliced genes, stress-related Gene Ontology categories were overrepresented, while housekeeping cellular functions were mainly transcriptionally controlled. Fe deficiency increased the complexity of the splicing pattern and triggered the differential alternative splicing of 313 genes, the majority of which had differentially retained introns. Several genes with important functions in Fe acquisition and homeostasis were both differentially expressed and differentially alternatively spliced upon Fe deficiency, indicating a complex regulation of gene activity in Fe-deficient conditions. A comparison with a data set for phosphate-deficient plants suggests that changes in splicing patterns are nutrient specific and not or not chiefly caused by stochastic fluctuations. In sum, our analysis identified extensive posttranscriptional control, biasing the abundance and activity of proteins in a condition-dependent manner. The production of a mixture of functional and nonfunctional transcripts may provide a means to fine-tune the abundance of transcripts with critical importance in cellular Fe homeostasis. It is assumed that differential gene expression and nutrient deficiency-induced changes in pre-mRNA splicing represent parallel, but potentially interacting, regulatory mechanisms.

The removal of introns from the immature mRNA by a process called “pre-mRNA splicing” occurs in the vast majority of eukaryotic protein-coding genes. Pre-mRNA splicing is catalyzed by elaborate ribonucleoprotein megadalton complexes referred to as spliceosomes (Wahl et al., 2009). Multiple mRNA isoforms can be generated from a single gene locus by alternative splicing, potentially producing functionally distinct protein isoforms (Hsu and Hertel, 2009). In mammals, alternative splicing dramatically increases protein diversity, yielding proteins that differ in function, activity, binding properties, or subcellular localization, and is believed to account for the multiexonic gene expression diversity, generating an estimated 100,000 proteins encoded by 25,000 genes in humans (Modrek and Lee, 2002; Kelemen et al., 2013). In humans, more than 95% of the intron-containing genes are alternatively spliced (Pan et al., 2008). Alternative splicing is less prominent in plants, and its importance has been debated. Recent estimates based on RNA sequencing (RNA-seq) data suggest that in Arabidopsis (Arabidopsis thaliana), up to 61% of the intron-containing genes are alternatively spliced (Filichkin et al., 2010; Marquez et al., 2012).

About 80% of the nuclear genes in plant genomes contain noncoding introns (Alexandrov et al., 2006), many of which can be spliced in various modes. The dominant mechanisms in alternative splicing among species are intron retention, skipping of alternative (cassette) exons (exon skipping), and alternative 5′ and 3′ splice sites (alternative donor/acceptor). In contrast to mammals, in which exon skipping accounts for the vast majority of alternative splicing events, the predominant form of alternative splicing in plants is intron retention (Filichkin et al., 2010; Marquez et al., 2012; Walters et al., 2013). However, a recent study showed that many intron retention transcripts are low in abundance, suggesting that the importance of intron retention in plants might be overestimated (Marquez et al., 2012). In animals, exon skipping is believed to contribute most to phenotypic diversity by increasing protein diversity (Keren et al., 2010). Recent studies showed that changes in alternative splicing evolve faster than changes in gene expression and contribute largely to species-specific differences and evolutionary adaptation (Irimia et al., 2009; Barbosa-Morais et al., 2012; Merkin et al., 2012). Surprisingly, considering the high plasticity of plant postembryonic development, in plants the number of genes that undergo exon skipping was estimated to account for only a small percentage of the alternatively spliced genes (Wang and Brendel, 2006; Marquez et al., 2012; Walters et al., 2013).

Alternative splicing frequently produces isoforms that harbor a premature termination codon (PTC) and are degraded by the “nonsense-mediated mRNA decay” (NMD) surveillance pathway or are targeted by microRNAs. It has been assumed that the inclusion of a PTC and subsequent degradation via the NMD pathway represent a mode of controlling functional transcript levels, a mechanism referred to as “regulated unproductive splicing and translation” (Lareau et al., 2007). Recent estimates suggest that, in Arabidopsis, 13% to 18% of the intron-containing genes are regulated by alternative splicing and NMD (Kalyna et al., 2012), a percentage that is comparable to those that have been reported for other eukaryotes (Hansen et al., 2009).

Iron (Fe) deficiency, caused by the generally low bioavailability of Fe, is a major problem for both plants and humans. Approximately 1.6 billion people are affected by Fe deficiency, causing increased maternal and child mortality, retarded physical and cognitive development, and a general reduction in fitness. In particular, in places where plants are the major dietary source, anemia is often widespread. Furthermore, suboptimal availability of Fe limits plant growth in many agricultural soils, causing yield losses and decreased food and forage quality. Plants have evolved elaborate mechanisms to improve Fe acquisition when Fe is limited. In Arabidopsis, Fe is taken up after reduction of the trivalent Fe generally prevailing in Fe chelates by the FERRIC REDUCTION OXIDASE2 (FRO2) via the ferrous IRON TRANSPORTER1 (IRT1; Eide et al., 1996; Robinson and Lemire, 1996). This reduction-based Fe uptake is aided by proton extrusion mediated by the P-type PLASMA MEMBRANE PROTON ATPASE2 AHA2, increasing the solubility of Fe, which strictly depends on the pH (Santi and Schmidt, 2009). The control of cellular Fe homeostasis is thought to be mainly under transcriptional control. Two bHLH transcription factors, FER-LIKE REGULATOR OF IRON UPTAKE and POPEYE (PYE), control nonoverlapping subsets of genes that contribute to the acquisition, distribution, and storage of Fe (Bauer et al., 2007; Long et al., 2010). Proteomic studies revealed several Fe-responsive proteins that were not associated with changes of their corresponding transcripts, implying other, unexplored levels of homeostatic control (Lan et al., 2011).

Pre-mRNA splicing has been attributed to mechanisms that regulate stress responses in both animals and plants. In humans, yeast, and plants, abiotic stress can increase nonproductive isoforms, counteracting translation (Pleiss et al., 2007; Filichkin et al., 2010; Dutertre et al., 2011). In plants, genes encoding regulatory proteins are overrepresented in populations of genes that undergo alternative splicing, supporting a regulatory role of alternative splicing (Kazan, 2003; Duque, 2011). In principle, stress-induced changes in alternative splicing pattern could provide an alternative route of adjusting the active transcriptome and thus the protein profile to the prevailing conditions, acting in concert with the differential expression of genes. Microarray-based transcriptional profiling does not allow for monitoring changes in splicing patterns. Research into alternative splicing has been much advanced by next-generation sequencing of mRNAs (RNA-seq), which provides an integrative picture of gene expression changes in response to environmental signals. In this study, we set out to explore changes in splicing patterns that occur in response to Fe deficiency, analyzed with a novel toolbox that provides an easy-to-use package for the analysis of RNA-seq data. We chose Fe deficiency as a case study, since both the transcriptional responses and changes in the protein profiles have been well explored in previous studies. We report that differential alternative splicing is independently regulated from gene expression but that several genes that play critical roles in cellular Fe homeostasis were controlled both transcriptionally and posttranscriptionally by alternative splicing, indicating an orchestrated regulation of gene activity. A comparison of the changes in alternative splicing patterns induced by Fe deficiency with those induced by phosphate (Pi) deficiency revealed that changes in splicing patterns are nutrient specific and not, or at least not chiefly, caused by “weak” splicing sites. We further report that intron retention is largely induced in Fe- and Pi-deficient plants, potentially due to decreased splicing fidelity or a less efficient removal of nonsense transcripts by NMD under nutrient-deficient conditions. These data indicate that splicing patterns are adjusted to environmental conditions, providing a fast and efficient means of gene regulation.

RESULTS

Detection of Alternative Splicing Events

Gene expression and splicing patterns of pre-mRNA in Arabidopsis roots were explored using the RNA-seq technology on the Illumina GAIIx platform. Samples were analyzed in triplicate, resulting in a total of 49.6 and 59.7 million uniquely mapped reads for Fe-sufficient and Fe-deficient plants, respectively (Supplemental Data Set S1). Under control conditions, a total of 13,824 splice junction (SJ)-containing genes (out of a total of 20,633 detected genes) were found to be expressed in roots. Chromosome coverage profiles revealed an equal distribution of transcriptional activity over the five chromosomes (Supplemental Fig. S1). In samples from Fe-deficient plants, a comparable (not statistically different when applying the χ² test; P = 0.16) but slightly higher number (14,229) of SJ-containing genes was identified from a total of 20,897 detected genes. Splicing patterns were analyzed with the Read Analysis & Comparison Kit in Java (RACKJ) software toolbox, which separately computes the number of reads matching to exons and introns, mapping SJs according to the alignment of reads to the genome. Source codes for the RACKJ software are available online at http://rackj.sourceforge.net/. All four major types of alternative splicing (intron retention, exon skipping, alternative 5′ splice site, and alternative 3′ splice site), presumably covering the vast majority of the alternative splicing events, were investigated in control and Fe-deficient plants (Fig. 1). We first identified all SJs using conservative criteria (five or more reads per SJ in all three runs, at least two of which aligned to different starting positions). With these criteria, we detected 82,231 SJs under control conditions, corresponding to 14,177 intron-containing genes (Fig. 2A; Supplemental Data Set S2). Under Fe-deficient conditions, the number of detected SJs was increased by 5.1% to a total of 86,429 in 14,600 intron-containing genes (Fig. 2A; Supplemental Data Set S2). Using a threshold of five reads per alternative splicing feature in all three experiments, 6,855 genes were defined as being alternatively spliced (Fig. 2B). As anticipated from previous studies in Arabidopsis and Brachypodium spp. (Filichkin et al., 2010; Marquez et al., 2012; Walters et al., 2013), under control conditions, mRNAs with retained introns constituted the largest fraction of alternatively spliced transcripts (5,690 genes), followed by transcripts with alternative 5′ (donor) and 3′ (acceptor) splice sites (ADA; 2,266 genes; Fig. 2C; Supplemental Data Set S3). For ADA detection, a minimum of two out of five reads per alternative splicing feature had to align to different starting positions for consideration, to avoid the detection of false positives. Out of the 2,810 ADA events, alternative 5′ splice sites were slightly more abundant than alternative 3′ splice sites (1,406 versus 1,101 events); a small subset had variations in both 5′ and 3′ sites (303 events). Transcripts with skipped exons (523 genes) constituted only a small portion of the alternatively spliced genes (Fig. 2C; Supplemental Data Set S3). A substantial overlap was noted between the different forms of alternative splicing either in the same intron or a different intron, indicating sophisticated patterns of pre-mRNA processing.

Figure 1.

Major types of alternative splicing. The different forms of alternative splicing occurring in plants, intron retention, ADA, and exon skipping, are schematically illustrated. Exons are represented by boxes and introns by lines. [See online article for color version of this figure.]

Figure 2.

Alternative splicing in roots of Arabidopsis under control and Fe-deficient conditions. A, Genes with (orange) and without (yellow) introns in control and Fe-deficient plants. B, Number and percentage of alternatively spliced (AS) genes in control and Fe-deficient plants. C and D, Types of alternative splicing in plants grown in control (C) and Fe-deficient (D) conditions. Numbers represent alternatively spliced genes. ES, exon skipping; IR, intron retention. [See online article for color version of this figure.]

Differential Alternative Splicing upon Fe Deficiency

For plants grown under Fe-deficient conditions, a similar pattern in the distribution of alternative splicing events was observed (Fig. 2; Supplemental Data Set S3). Also, the total percentage of intron-containing genes that were alternatively spliced was substantially higher (χ² test; P = 2.6e-13) when plants were starved of Fe (Fig. 2B). For all alternative splicing features identified in the three samples of each growth type (control and Fe-deficient), a P value based on Student’s t test was calculated, and alternative splicing events with P < 0.05 and a fold change > 2 between control and Fe-deficient samples were defined as differentially alternatively spliced (DAS). Only features that were detected in all three parallel runs with at least five reads per run were considered. Using these criteria, a set of 313 DAS genes was identified, the majority of which had differential intron retention (DIR; Fig. 3; Supplemental Data Set S4). A subset of 59 ADA events was differentially distributed between samples from control and Fe-deficient plants (DADA), and five genes produced transcripts with exons that were differentially skipped upon Fe deficiency. The majority of the DIR events were induced upon Fe deficiency (i.e. the preference of an intron being retained in samples from Fe-deficient plants was higher than in those from control plants; Fig. 3; Supplemental Data Set S4).

Figure 3.

A and B, Distribution of DAS events in Fe-deficient (A) and Pi-deficient (B) plants. C, Numbers and overlap of DAS genes in roots from Fe-deficient and Pi-deficient plants. Data of Pi-deficient plants derived from a published RNA-seq data set (Lan et al., 2012a) were reanalyzed for alternative splicing patterns. Up, induced (i.e. the preference of an alternative splicing event in samples from Fe-deficient plants was higher than in those from control plants); down, repressed (i.e. the preference of an alternative splicing event in samples from Fe-deficient plants was lower than in those from control plants). [See online article for color version of this figure.]

Changes in Splicing Patterns Are Nutrient Specific

To investigate whether the observed changes in splicing pattern were specific to Fe deficiency, we analyzed a data set derived from plants that have been deprived of Pi but were grown under otherwise similar conditions (Lan et al., 2012a). The number of genes with differential exon skipping events was comparably low in both samples from Fe-deficient and Pi-deficient plants (Fig. 3B). Also, the distribution of the significantly altered alternative splicing events differed between the two nutrient regimes: while DADA was clearly lower in Pi-deficient plants, the number of DIR genes was about 2-fold higher in the latter condition. Similar to samples from Fe-deficient plants, the majority of intron retention events were induced upon Pi starvation, with this trend being more pronounced in Pi-deficient plants when compared with Fe-deficient plants (Fig. 3; Supplemental Data Set S4). Comparing the Gene Ontology (GO) category “biological process” in genes that were affected in their splicing pattern by Fe or Pi starvation revealed overrepresentation of different GO categories in both growth types (Fig. 4). With the exception of the categories “response to cadmium ion” and “response to salt stress,” which were strongly enriched for genes from plants grown under both Fe and Pi deficiency, the changes in splicing patterns appeared to be largely nutrient specific. In Fe-deficient plants, stress-specific categories such as “cellular response to iron ion” and “cellular response to nitric oxide” were found to be most strongly and exclusively overrepresented in Fe-deficient plants (Fig. 4). In Pi-deficient plants, the category “cellular response to phosphate starvation” and other processes important for Pi homeostasis, such as “phosphate transport” and “protein dephosphorylation,” were strongly overrepresented. Stress-specific regulation of DAS is also evident from the surprisingly small overlap of genes with alternatively spliced features in Fe-deficient and Pi-deficient conditions (Fig. 3C).

Figure 4.

Functional categorization (biological process) of DAS genes in Fe-deficient and Pi-deficient plants. Data of Pi-deficient plants derived from a published RNA-seq data set (Lan et al., 2012a) were reanalyzed for alternative splicing patterns. [See online article for color version of this figure.]

To validate the bioinformatic analysis, we confirmed predicted splicing events by quantitative reverse transcription-PCR following the delta-delta threshold cycle method. Samples from control, Fe-deficient, and Pi-deficient plants were analyzed, and the Fe/Pi deficiency-induced fold changes for the reference isoform (constitutively expressed) and the DIR isoform were calculated. All of the tested transcripts clearly validated the predicted event in three independent RNA preparations from Fe-sufficient and Fe-deficient plants with a clear trend (Fig. 5). These intron retention events remained unchanged when plants were subjected to Pi-deficient conditions. The 4-coumarate:CoA ligases CL1 and CL2 have been implicated in the production of phenylpropanoids that, after excretion into the rhizosphere, may chelate Fe from a pool of low availability (Yang et al., 2010). For EMBRYO SAC DEVELOPMENT ARREST7 (EDA7) and glutathion transferase L3 (GSTL3), no function in Fe acquisition or homeostasis has been assigned yet.

Figure 5.

Validation of alternative splicing events by quantitative reverse transcription-PCR. A, DIR features were determined on samples from control, Fe-deficient, and Pi-deficient plants, and the Fe/Pi deficiency-induced fold changes for the reference isoform (Ref; constitutively expressed) and the DIR isoform were calculated by normalizing the data to the reference isoform under control conditions. Four genes were tested: two (4CL1 and 4CL2) showed repressed DIR and two (EDA7 and GSTL3) showed induced DIR in response to Fe-deficient conditions. Splicing of all tested genes was predicted to remain unchanged in response to Pi deficiency. Note that different controls were used for Fe- and Pi-deficient plants. B, Schematic diagram of the gene model and primer design strategy. Primers 1 and 2 were used to measure the abundance of the reference isoform, and primers 3 and 4 were intron retention specific. Regions with retained introns are magnified. [See online article for color version of this figure.]

Alternatively Spliced and Differentially Expressed Genes Are Separately Regulated

Using the same criteria as for DAS, a subset of 429 genes was defined as differentially expressed (DE) between control and Fe-deficient plants (Supplemental Table S1). Marker genes for Fe-deficient plants, such as IRT1, FRO2, and bHLH100, were highly induced (57-, 58-, and 357-fold, respectively), indicating that the plants were truly Fe deficient. A subset of 26 DE genes also showed significantly altered intron retention, exon skipping, or ADA upon Fe deficiency. Categorization of the biological functions of DAS and DE genes revealed that overrepresentation of GO categories differed largely between both groups, indicating separate regulation of gene expression and alternative splicing in response to Fe starvation (Fig. 6). For example, some GO categories, such as “Suc biosynthetic process” and “Met biosynthetic process,” were only overrepresented in DAS genes, while many were only overrepresented in or largely dominated by DE genes (Fig. 6). Several biological functions that are important for cellular Fe homeostasis, such as “cellular response to nitric oxide,” “iron ion transport,” “cellular response to ethylene stimulus,” and “response to zinc ion,” were overrepresented in both DE and DAS genes. This indicates that the control of gene expression and alternative splicing is not completely independent and that genes with critical functions in Fe acquisition are regulated both transcriptionally and posttranscriptionally. Notably, most genes that were highly (more than 10-fold) up-regulated under Fe-deficient conditions also showed a highly significant tendency for at least one intron being retained under Fe deficiency (Table I). For example, the oxidoreductase FRO2, important for the obligatory reduction of ferric chelate, and the Fe transporter gene IRT2 produced significantly more transcripts with retained introns under Fe-deficient conditions (Table I). This is contrary to expectations and may point to either a decrease in splicing fidelity or to a fine-tuning mechanism that adjusts the amount of functional transcripts to the cellular Fe level. In both cases, more nonproductive transcripts might be produced, a scenario that may contribute to the generally low concordance of transcript levels and protein abundance (Baginsky et al., 2005; Rogers et al., 2008). However, this scheme does not hold true for all highly up-regulated genes, suggesting different modes of regulation. Also, only a small subset of the DE genes are also differentially alternatively spliced (Fig. 7), arguing against this assumption. It should also be noted that the basal (Fe sufficient) transcript level of many highly induced genes is very low, and alternative splicing features may be either very infrequent or not yet classified as such in transcripts from Fe-sufficient plants using our stringent criteria, which would overestimate the potential significance of DAS in this group of genes. A similarly small overlap of DE and DAS genes was observed for Pi-deficient plants (Fig. 7B). Notably, this overlap was substantially smaller when only those transcripts were considered for which a cognate protein was identified (Fig. 7). It can thus be assumed that alternative splicing and differential gene expression represent parallel, but potentially interacting, regulatory levels.

Figure 6.

Functional categorization (biological process) of DE and DAS genes. [See online article for color version of this figure.]

Table I. Transcriptional and posttranscriptional regulation of genes in GO categories that are relevant for cellular Fe homeostasis.

Genes were selected that were DE, DAS, or both. –, Not regulated; ∞, not expressed under control conditions. Genes that occur in different categories are only shown once.

Gene Identifier	Gene Description	Fe Deficient/Fe Sufficient	DAS Type-Regulation
			RPKM
GO categories response to iron ion and iron transport and homeostasis
At1g56160	MYB72, Myb domain protein72		∞	–
At5g55620	Unknown protein		∞	–
At3g12900	2-Oxoglutarate (2OG) and Fe(II)-dependent oxygenase superfamily protein		546.9	DIR-up
At2g41240	BHLH100, basic helix-loop-helix protein100		357.4	DIR-up
At3g53280	CYP71B5, cytochrome p450 71b5		128.0	–
At1g01580	FRO2, ferric reduction oxidase2		58.2	DIR-up
At4g19690	IRT1, iron-regulated transporter1		57.1	DADA-up
At5g38820	Transmembrane amino acid transporter family protein		49.6	–
At4g19680	IRT2, iron-regulated transporter2		21.8	DIR-up
At5g04150	BHLH101, basic helix-loop-helix protein101		18.0	DIR-up
At5g36890	BGLU42, β-glucosidase42		9.9	DIR-up
At5g03570	IREG2, FPN2, iron regulated2		7.7	–
At5g04950	NAS1, nicotianamine synthase1		4.4	–
At2g28160	FIT, FER-like regulator of iron uptake		4.2	–
At1g02500	SAM1, S-adenosylmethionine synthetase1		3.8	–
At3g21240	4CL2, AT4CL2, 4-coumarate:CoA ligase2		3.6	DIR-down
At1g51680	4CL1, 4-coumarate:CoA ligase1		3.2	–
At5g67330	NRAMP4, natural resistance-associated macrophage protein4		2.4	–
At2g05830	NagB/RpiA/CoA transferase-like superfamily protein		2.4	–
At5g26820	IREG3, MAR1, iron-regulated protein3		2.3	–
At3g47640	PYE, bHLH DNA-binding superfamily protein		2.2	–
At3g25190	Vacuolar iron transporter (VIT) family protein		0.5	–
At3g56090	FER3, ferritin3		0.5	–
At5g24380	YSL2, yellow stripe-like2		0.5	–
At4g04770	ABC1, ATP-binding cassette protein1		0.4	–
At1g21140	Vacuolar iron transporter (VIT) family protein		0.2	–
At1g76800	Vacuolar iron transporter (VIT) family protein		0.2	–
At2g40300	FER4, ferritin4		0.2	–
At3g11050	FER2, ferritin2		0.2	–
At5g01600	FER1, ferritin1		0.2	–
At2g28190	CSD2, copper/zinc superoxide dismutase2		–	DIR-down
GO categories Met biosynthetic process and phenylpropanoid biosynthetic process
At3g13610	2-Oxoglutarate (2OG) and Fe(II)-dependent oxygenase superfamily protein	10.3		–
At1g49820	MTK, S-methyl-5-thioribose kinase	2.7		–
At5g48930	HCT, hydroxycinnamoyl-CoA shikimate/quinate hydroxycinnamoyltransferase	2.1		–
At2g37040	PAL1, Phe ammonia lyase1	2.6		–
At3g53260	PAL2, Phe ammonia-lyase2	4.0		–
At2g45400	BEN1, NAD(P)-binding Rossmann-fold superfamily protein	2.2		–
At4g14710	ATARD2, RmlC-like cupins superfamily protein	2.4		DIR-up
At5g53850	Haloacid dehalogenase-like hydrolase family protein	2.1		DIR-up
At5g46845	MIR160/MIR160C, microRNA	0.5		–
At4g36220	CYP84A1, FAH1, ferulic acid 5-hydroxylase1	0.4		–
At1g71030	MYBL2, MYB-like2	0.2		–
At5g04410	NAC2, NAC domain-containing protein2	–		DIR-down
At5g20980	MS3, Met synthase3	–		DIR-up
At1g14810	Semialdehyde dehydrogenase family protein	–		DIR-up
At1g32100	PRR1, pinoresinol reductase1	–		DIR-up
GO categories response to zinc ion and zinc transport
At3g07720	Gal oxidase/Kelch repeat superfamily protein	23.0		–
At3g58810	MTPA2, metal tolerance protein A2	17.1		–
At4g33020	ZIP9, ZIP metal ion transporter family	7.2		–
At5g13740	ZIF1, zinc-induced facilitator1	2.8		–
At1g16150	WAKL4, wall-associated kinase-like4	2.7		–
At5g59520	ZIP2, ZRT/IRT-like protein2	0.2		–
At5g13750	ZIFL1, zinc-induced facilitator-like1	–		DIR-up
At1g04410	Lactate/malate dehydrogenase family protein	–		DIR-down
At3g09200	Ribosomal protein L10 family protein	–		DADA-down

Figure 7.

Venn diagram of the overlap of genes that are DE, DAS, or differentially regulated at the protein level (DP). Protein expression data were taken from Lan et al. (2011; Fe deficient) and Lan et al. (2012a; Pi deficient). [See online article for color version of this figure.]

Noise or Regulated Changes in Splicing Patterns?

To investigate whether the splicing patterns are affected by the gene architecture, we analyzed the number and length of introns in genes that produced isoforms with retained introns and in those in which no such events or any other type of alternative splicing under investigation were detected. We first analyzed the dependence of the probability to detect an intron retention event on the number of introns in a given gene. As anticipated, this ratio increased with the increasing number of introns; no further increase was observed for genes with 10 or more introns (Fig. 8A). A similar trend was observed for samples from Fe-deficient plants. Also with respect to intron length, there was a clear tendency for a higher probability for an intron being retained up to an intron length of approximately 1,000 nucleotides (Fig. 8B). Again, no differences were observed between samples from both growth types. It can thus be stated that the detection of intron retention is not affected by the nutrient regime. We next analyzed if the detection of DIR is affected by the gene architecture. For Fe-deficient plants, there was no clear dependence of detecting DIR events on the intron length (Fig. 8C). The DIR-intron retention ratio was generally higher for samples from Pi-deficient plants, indicating that the probability of identifying DIR events was somewhat higher under these conditions. The difference between the two growth types was less pronounced for very long (greater than 700 nucleotides) introns (Fig. 8C). With regard to intron number, there was a slight trend of increasing the likelihood to detect DIR events for genes with more introns, which was more pronounced in transcripts from Pi-deficient plants when compared with Fe-deficient plants (Fig. 8D). Generally, these trends were not very pronounced, and it can be stated that differential alternative splicing is not or not strongly biased by the gene architecture, except for genes with a high (more than 15) number of introns, which might be slightly more prone to intron retention under the tested nutrient conditions.

Figure 8.

Dependence of the probability of detecting intron retention events on the gene structure. A and B, Dependence of intron number (A) and intron length (B) on the ratio between intron retention events and detected intron-containing genes of samples from control and Fe-deficient plants. C and D, Ratio of DIR versus intron retention (IR) events with respect to the intron length (C) and the intron number of DIR versus intron retention genes (D). Data of Pi-deficient plants derived from a published RNA-seq data set (Lan et al., 2012a) were reanalyzed for alternative splicing patterns. [See online article for color version of this figure.]

We further compared the representation of the GO category “biological processes” in genes that underwent alternative splicing with those that were not alternatively spliced. This comparison revealed clearly that in both populations, different GO categories were overrepresented (Fig. 9). Generally, in genes that produce alternative transcripts, several GO categories related to stress, such as “response to salt stress,” “response to cold,” “defense response to bacterium,” and “response to wounding,” among others, were overrepresented, while housekeeping functions such as “protein folding,” “DNA repair,” and “chloroplast organization” were overrepresented in genes that were not alternatively spliced. For Fe-deficient plants, a generally similar picture emerged. In the stress-related categories, the possibility of genes that undergo alternative splicing was more pronounced than genes that were not alternatively spliced, supporting the assumption that, generally, stress-related genes are more predisposed to pre-mRNA processing than genes encoding proteins that are involved in basic cellular functions (Fig. 9). A similar pattern was observed for Pi-deficient plants, suggesting that these observations were general and not affected by the growth regime (Fig. 9).

Figure 9.

Functional categorization (biological process) of alternatively spliced (AS) and nonalternatively spliced (nonAS) genes in control and Fe-deficient conditions. Data of Pi-deficient plants derived from a published RNA-seq data set (Lan et al., 2012) were reanalyzed for alternative splicing patterns. [See online article for color version of this figure.]

DISCUSSION

While underestimated and almost neglected a decade ago, recent studies using massively parallel RNA sequencing show that a large percentage of genes in Arabidopsis undergo alternative splicing (Filichkin et al., 2010; Marquez et al., 2012), suggesting a massive effect on gene activity. This underestimation derived mainly from the widely used microarray platforms to catalog gene expression, which rather reports a mixture of different splicing isoforms matching a precast probe set and does not distinguish between different mRNAs that derive from the same gene. Our analysis here suggests that, under control conditions, 48% of the intron-containing genes detected in Arabidopsis roots are alternatively spliced. This number is slightly higher than that reported by Filichkin et al. (2010; 42%) and somewhat lower than a recent study that revealed alternative splicing for 61% of multiexonic genes for a normalized complementary DNA (cDNA) library, which facilitates the detection of alternative splicing events in transcripts with low abundance (Marquez et al., 2012). A comparable percentage (56%) was estimated for maize (Zea mays) leaves (Li et al., 2010). It can be assumed that, due to the distinguished phenotypic plasticity associated with the sessile lifestyle of plants, the extent of alternative splicing is elaborated by adaptive processes. This assumption has been supported by a recent analysis of the splicing patterns of various circadian clock genes under different stresses. For example, heat stress was shown to support the expression of a CCA1 isoform that harbors a PTC, while under cold stress the canonically spliced isoform was predominant (Filichkin and Mockler, 2012), suggesting stress-induced changes of alternative splicing to regulate gene activity.

Nutrient Deficiency Induces More Complex Transcript Isoform Patterns

Due to the differences in the architecture of mammalian and plant genomes, different types of alternative splicing are prevalent in the two groups. In contrast to animals, intron retention and alternative donor/acceptor are much more prominent than exon skipping, which makes the detection of alternative splicing features and predictions regarding the biological relevance more difficult. Supporting previous studies, the prevalent form of alternative splicing in Arabidopsis roots was intron retention, which seems to be conserved in plants. It is noteworthy to mention that a large subset of genes was affected by multiple forms of alternative splicing, suggesting complex patterns of mRNA isoforms. The estimate for exon skipping and alternative 5′ and 3′ splice sites appears to differ among species and/or tissues. Li et al. (2010) observed a substantially higher (32%) percentage of exon skipping in maize leaves, while the number of exon-skipping events appears to be much lower in Arabidopsis (Filichkin et al., 2010; Marquez et al., 2012; Walters et al., 2013; this study). Notably, in maize leaves, the estimate for 5′ and 3′ alternative splice sites was substantially lower than what we observed for Arabidopsis, indicating a highly variable amount of low-abundance splice isoforms. Fe and Pi deficiency enhanced the complexity of alternative splicing in Arabidopsis roots, increasing the percentage of alternatively spliced genes. This finding suggests that the total number of genes may increase when more genome-wide studies under different stress conditions become available. All types of alternative splicing were affected by the nutrient regime, which might be indicative of a decrease in splicing fidelity. However, the changes in splicing patterns were highly specific to the growth type, with a very small, random overlap between samples from Fe-deficient and Pi-deficient plants, arguing against such an assumption and in favor of the physiological significance of the observed changes. Furthermore, GO categories of growth type-relevant functions were significantly overrepresented in the respective conditions, further supporting regulated changes in splicing patterns.

Stress-Induced Changes in Splicing Patterns May Have Functional Implications

While it is clear that a large number of transcripts undergo alternative splicing, the biological significance of this process in plants remains enigmatic. One possible cause of alternative splicing is stochastic noise, imprecise operation of the splicing machinery that causes the production of a certain percentage of unproductive mRNAs that are not translated into proteins and degraded via the NMD pathway. Such stochastic noise may have functional advantage by stabilizing the system via a fast increase in functional heterogeneity (Blake et al., 2006; Simpson et al., 2009; Eldar and Elowitz, 2010). However, the fact that unrelated stresses induce unique molecular profiles argues against this assumption, without excluding that stochastic fluctuations may cause the production of a subset of alternatively spliced transcripts. Furthermore, the analysis of GO categories of genes that undergo alternative splicing and of those that are not alternatively spliced clearly shows differences between the two populations, suggesting the biological function of alternative splicing.

Alternatively spliced transcripts may be translated into novel proteins with altered functions, increasing the diversity of the proteome. However, the fact that most of the alternative splicing events in plants are intron retention, which may contain PTCs, does not support this scenario. A prerequisite for biologically meaningful alternative splicing is a regulated selection of splice sites. While our analysis does not allow for a clear-cut distinction between noisy and regulated alternative splicing, we favor the hypothesis that at least a subset of alternatively spliced genes is part of an additional, in plants largely unexplored, layer of gene regulation. Differential gene expression and differential alternative splicing may represent separate but cooperative regulatory levels for adjusting the abundance of a given protein to a given environmental condition. Such regulated unproductive splicing and translation has been described as a regulatory mechanism for several genes in yeast and mammals (Lareau et al., 2007). An argument in favor of this hypothesis is that the costs for transcription are relatively low compared with the production of a protein (and its potential degradation). Among the mRNAs with Fe deficiency-induced intron retention, some transcripts with crucial functions in Fe uptake and homeostasis were found, including FRO2, IRT1, IRT2, bHLH100, and bHLH101. All of the intron-retained isoforms contain PTCs. An attractive scenario that would explain our findings involves stress-specific changes in histone modification signatures that affect alternative splicing, either by the rate of Pol II elongation or by the recruitment of a specific set of splicing factors. In mammalian cells, methylation or acetylation of core histones can bias splicing patterns (Alló et al., 2009; Luco et al., 2010). In the case of plants, such a mechanism awaits experimental validation. Alternatively, or in addition to such a mechanism, the production of a mixture of functional and nonfunctional transcripts may provide a means to fine-tune the abundance of transcripts with critical importance for cellular Fe homeostasis without going to a much slower and more costly cycle of protein production and degradation to adjust the level of protein to fluctuating conditions. Such a dual regulation is common in biological systems and has also been observed for the control of Fe homeostasis. Two examples are the bHLH transcription factor PYE and the putative E3 ligase BRUTUS, which were suggested to have opposite effects on the expression of a subset of genes with important roles in Fe homeostasis by acting with common binding partners (Long et al., 2010). Both genes are up-regulated upon Fe starvation, possibly to allow for a fast adaptation of the activity of target genes. Alternatively (or additionally), retaining introns may serve as an adequate means to avoid the translation of transcripts that do not contribute to stress adaptation as an alternative to the repression of transcription. Intron retention and alternative 5′ and/or 3′ splice sites in conjunction with NMD may work in this way and may represent a mechanism to adapt the proteome profile to the prevailing conditions. In plants, intron retention transcripts are often insensitive to NMD, which may also provide an explanation for the high frequency of intron retention events in plants (Kalyna et al., 2012).

Nonfunctional Alternative Splicing May Represent a Fast Fine-Tuning Mechanism of Gene Activity

It is interesting that intron retention was repressed in a subset of transcripts, indicative of an increased splicing fidelity of the encoded proteins that is crucial for the recalibration of nutrient homeostasis. Alternatively, Fe and Pi deficiency may induce a more efficient decay of aberrant transcripts. It is further possible that the subset of genes with retained introns is enriched in more ambiguous splice signals that are, under control conditions, more frequently missed by the splicing machinery. Stress, such as Fe or Pi deficiency, may trigger a more efficient removal of the introns. Such a mechanism has been demonstrated for yeast (Saccharomyces cerevisiae), in which different stresses led to an increased or decreased splicing efficiency for stress-specific subsets of genes, suggesting a regulated splicing mechanism as a means to control gene expression in response to stress (Pleiss et al., 2007). A similar scenario may apply for plants. Since, in contrast to yeast, the majority of genes contain multiple introns, condition-specific changes in splicing patterns may be much more complex. In yeast, the response of the splicing machinery to environmental conditions is rapid and independent of protein synthesis. Thus, alternative splicing might represent a rapid component of a multilayered regulation of gene expression. What we describe in this study is a steady-state level of alternatively spliced transcripts that might be rapidly turned over. It can be speculated that condition-sensitive alternative splicing provides a means to regulate gene activity by fine-tuning the production of a certain subset of proteins, which may be specifically required under the respective circumstances, and shutting down others that are not required in a particular stress situation. A further possibility is that differential gene expression and differential alternative splicing represent separate but cooperative regulatory levels for adjusting the abundance of proteins to a given environmental condition. The expression level of several DAS genes was not changed significantly in response to Fe deficiency, indicating a regulation exclusively at the posttranscriptional level. Most of these genes have not been associated with responses to Fe deficiency, but a function in Fe metabolism might be inferred for some of these genes upon closer inspection. For example, ZINC-INDUCED FACILITATOR-LIKE1 (ZIFL1) was shown to possess a dual function in auxin transport and drought resistance (Remy et al., 2013). Interestingly, the function of ZIFL1 was regulated by alternative splicing (Remy et al., 2013). Auxin responsiveness is altered in Fe-deficient plants (Lan et al., 2012b), making it tempting to suggest that DAS could modulate auxin-related processes in Fe-deficient roots. Another gene for which a function of DAS can be assumed is the copper/zinc superoxide dismutase CDS2. DIR of CSD2 was repressed in Fe-deficient roots, possibly to compensate for the compromised activity of Fe superoxide dismutase. At3g09200 encodes one out of three paralogous ribosomal L10 family proteins, structural components of the large subunit of cytosolic ribosomes. The composition of ribosomes was shown to be responsive to environmental conditions (Hummel et al., 2012), possibly altering translation. The repression of DADA of At3g09200 transcripts may thus contribute to the selective translation of a subset of mRNAs coding for proteins with important functions in Fe homeostasis. While such a cooperative gene regulation by DE and DAS needs to be experimentally verified, the analysis presented here provides a straightforward way for directing experimental strategies.

MATERIALS AND METHODS

Plant Growth

Arabidopsis (Arabidopsis thaliana) plants were grown in a growth chamber on an agar-based medium as described previously (Estelle and Somerville, 1987). Seeds of the Columbia accession were obtained from the Arabidopsis Biological Resource Center. Seeds were surface sterilized by immersing them in 5% (v/v) NaOCl for 5 min and 96% ethanol for 7 min, followed by four rinses in sterile water. Seeds were placed onto Petri dishes and kept for 1 d at 4°C in the dark; the plates were then transferred to a growth chamber and grown at 21°C under continuous illumination (50 µmol m⁻² s⁻¹; Philips TL lamps). The medium was composed of (mm): KNO₃ (5), MgSO₄ (2), Ca(NO₃)₂ (2), KH₂PO₄ (2.5); (µm): H₃BO₃ (70), MnCl₂ (14), ZnSO₄ (1), CuSO₄ (0.5), NaCl (10), Na₂MoO₄ (0.2); and 40 µm Fe-EDTA solidified with 0.3% Phytagel (Sigma-Aldrich). Suc (43 mm) and 4.7 mm MES were included, and the pH was adjusted to 5.5. After 10 d of precultivation, plants were transferred either to fresh agar medium without Fe and with 100 µm 3-(2-pyridyl)-5,6-diphenyl-1,2,4-triazine sulfonate, medium without phosphate, or fresh control medium and grown for 3 d. The lower concentration of potassium because of the absence of KH₂PO₄ in the phosphate-free medium was corrected by the addition of KCl.

RNA-seq

Total RNA was extracted from roots grown under control and Fe- or Pi-deficient conditions using the RNeasy Plant Mini Kit (Qiagen). Equal amounts of RNA were collected from three independent experiments and used for sequencing. cDNA libraries for sequencing were prepared from 5 μg of total RNA following protocols provided by the instrument manufacturer (Illumina). The cDNA libraries were enriched by 15 cycles of PCR amplification. The resulting cDNA libraries were sequenced on a single lane per sample of the Illumina Genome Analyzer II. The RNA-seq and data collection followed published protocols (Mortazavi et al., 2008). The length of the cDNA library ranged from 250 to 300 bp with a 5′ adapter of 58 bp and a 3′ adapter of 63 bp at both ends. The fragment length of the cDNA ranged from 129 to 179 bp. For high-throughput sequencing, the library was prepared according to the manufacturer's instructions and submitted to 80-nucleotide sequencing using the Illumina GAIIx system.

Read Alignment to the Arabidopsis Genome

A total of 56,698,709 and 68,360,588 reads were obtained from Illumina sequencing for Fe-sufficient and Fe-deficient conditions. Reads were aligned to The Arabidopsis Information Resource 10 (TAIR10) reference genome using the BLAT program (Kent, 2002) with minimum 95% identity, but only the alignment with the highest identity for each read was considered for mapping. Multireads were distributed in proportion to the number of unique and splice reads recorded at similar loci using the Enhanced Read Analysis of Gene Expression strategy (Mortazavi et al., 2008). Subsets of 51,786,440 and 62,485,423 reads were mapped to the TAIR10 genome for each sample from treated and untreated roots.

Computing DE Genes

Adapters were trimmed from reads, with approximately 20.0 m 80-mer reads for each sample in three replicates. RPKM values (reads per kilobase of exon model per million mapped reads) were computed as described (Mortazavi et al., 2008). A transcript was defined as present when it was detected with at least five reads in all three experiments within one growth type (i.e. either treated or untreated). A gene was denoted as DE between control and Fe-deficient plants at P < 0.05 calculated by Student’s t test and a Fe-deficient/control fold change > 2.

SJ Detection

Splicing patterns were explored using the RACKJ software toolbox, which separately computes the number of reads matching to exons, introns, and SJs. Source codes for the RACKJ software are available online at http://rackj.sourceforge.net/. To diminish the number of potential false positives that are predicted by erroneous alignment, we performed a prediction of SJs with a minimum of eight-nucleotide identical alignment blocks to the genome and supported by at least five reads with at least two of them having different starting positions.

Detection of Condition-Sensitive Alternative Splicing Events

We investigated the four main types of alternative splicing in plants: intron retention, exon skipping (cassette exon), alternative 5′ splice site (alternative donor), and alternative 3′ splice site (alternative acceptor). Alternative splicing events were computed using the RACKJ program. We defined DAS events using the “splicing index” method (Griffith et al., 2010). Essentially, the expression of each event was normalized to the expression level of the gene to which it belongs (splicing index value) before two samples (control and treatment) were compared. Student’s t test was then applied to validate differences in the splicing index values for a particular alternative splicing event between samples from control and treated plants. The fold change was calculated as (Ti/Tj)/(Ci/Cj) for the i-th exon (intron) of the j-th gene in control (C) and treated (T) samples. Alternative splicing events with significant P values (P < 0.05) and fold change > 2 were denoted as DAS events.

Primer Design for the Validation of Alternative Splicing Events

For the validation of DIR, primers were designed to cover the predicted retained intron with either of the two primers corresponding to the putatively retained intronic sequence. Another pair of reference primers that amplify the exon next to the retained intron was designed to monitor the expression level of its corresponding gene. The melting temperature of primers was in the range of 58°C to 62°C. Primer pairs were selected using Primer3 (http://primer3.sourceforge.net/). Elongation factor1-β2 (At5g19510) was used as an internal control for transcript normalization. Primers used for the validation are listed in Supplemental Table S2.

Quantitative Reverse Transcription-PCR

Total RNA was isolated using the RNeasy Plant Mini Kit (Qiagen) and treated with DNase using the TURBO DNA-free Kit (Ambion) as suggested by the manufacturer. cDNA was synthesized using 4 μL of DNA-free RNA with oligo(dT)₂₀ primer and SuperScript II reverse transcriptase (Invitrogen). After incubation at 50°C for 1 h followed by 70°C for 15 min, 1 μL of RNase H was added and incubated for 20 min at 37°C. The cDNA was used as a PCR template in a 20 μL reaction system using the SYBR Green PCR Master Mix (Applied Biosystems) with programs recommended by the manufacturer in the ABI Prism 7500 Sequence Detection System (Applied Biosystems). Three independent replicates were performed for each sample. The delta-delta threshold cycle method was used to determine the relative amount of gene expression (Livak and Schmittgen, 2001).

GO Analysis

Enrichment analysis of GO categories was done in the Gene Ontology Browsing Utility (Lin et al., 2006) using the “elim” method from the aspect “biological process.” The elim method is a Java implementation of the TopGO elim algorithm (Alexa et al., 2006). The GO categories shown in the figures were chosen based on the following criteria: more than 10 genes must be listed in TAIR10, and P < 0.005 (P < 1e-6 in Fig. 9) calculated by the elim method. The selected categories were sorted from the smallest to the biggest P value.

Sequence data from this article can be found in the GenBank/EMBL data libraries under accession number SRA045009.

Supplemental Data

The following materials are available in the online version of this article.

• Supplemental Figure S1. Transcriptional profile of genes in roots of Fe-sufficient and Fe-deficient plants.

• Supplemental Table S1. DE genes between control and Fe-deficient plants.

• Supplemental Table S2. Primers used for the validation of DAS features.

• Supplemental Data Set S1. Read numbers and RPKM values from the transcriptome analysis of samples from control and Fe-deficient plants.

• Supplemental Data Set S2. SJs in samples from control and Fe-deficient plants.

• Supplemental Data Set S3. Alternative splicing features in samples from control and Fe-deficient plants.

• Supplemental Data Set S4. DAS features.

Glossary

RNA-seq

RNA sequencing

PTC

premature termination codon

NMD

nonsense-mediated mRNA decay

Fe

iron

Pi

phosphate

SJ

splice junction

RACKJ

Read Analysis & Comparison Kit in Java

ADA

alternative donor and acceptor

DAS

differentially alternatively spliced

DIR

differential intron retention

DADA

differential donor/acceptor

DES

differential exon skipping

GO

Gene Ontology

DE

differentially expressed

cDNA

complementary DNA

TAIR10

The Arabidopsis Information Resource 10

RPKM

reads per kilobase of exon model per million mapped reads