Multiple and Specific mRNA Processing Targets for the Major Human hnRNP Proteins

Alternative splicing is a key mechanism regulating gene expression,and it is often used to produce antagonistic activities particularlyin apoptotic genes. Heterogeneous nuclear ribonucleoparticle(hnRNP) proteins form a family of RNA-binding proteins thatcoat nascent pre-mRNAs. Many but not all major hnRNP proteinshave been shown to participate in splicing control. The rangeand specificity of hnRNP protein action remain poorly documented,even for those affecting splice site selection. We used RNAinterference and a reverse transcription-PCR screening platformto examine the implications of 14 of the major hnRNP proteinsin the splicing of 56 alternative splicing events in apoptoticgenes. Out of this total of 784 alternative splicing reactionstested in three human cell lines, 31 responded similarly toa knockdown in at least two different cell lines. On the otherhand, the impact of other hnRNP knockdowns was cell line specific.The broadest effects were obtained with hnRNP K and C, two proteinswhose role in alternative splicing had not previously been firmlyestablished. Different hnRNP proteins affected distinct setsof targets with little overlap even between closely relatedhnRNP proteins. Overall, our study highlights the potentialcontribution of all of these major hnRNP proteins in alternativesplicing control and shows that the targets for individual hnRNPproteins can vary in different cellular contexts.

INTRODUCTION

Top

INTRODUCTION

Alternative splicing is a critical process that ensures theproduction of a multitude of proteins from a limited set ofmammalian genes (7, 41). Alternative splicing decisions areregulated by a large collection of RNA-binding proteins (RBPs)that bind to pre-mRNAs in the nucleus (5). The heterogeneousnuclear ribonucleoparticle (hnRNP) proteins are among the mostabundant of such proteins, and more than 20 of them have beencharacterized and given alphabetical names based on size fromhnRNP A1 to hnRNP U (17). These proteins have been implicatedin a variety of biological processes including telomere biogenesis,translation, and RNA stability, and several (e.g., hnRNP A1,A2, F, H, I [PTB], G, and L) have documented roles in splicing(34, 38). hnRNP A1 has been implicated in the splicing controlof many genes, including the A1 gene itself, the caspase-2 gene,c-src, and the SMN2 gene (12), and several exons of human immunodeficiencyvirus type 1; the very similar hnRNP A2 protein (68% identity)appears to display comparable activity (4, 10, 29, 42). Whilethe related hnRNP F and H proteins play a role in splicing controlof many genes, including c-src, Bcl-x, cystathionine β-synthase,and several HIV alternative exons (38), it is unclear whetherF and H have completely redundant activities. hnRNP I (PTB)is another well-known splicing regulator that has been mostlyassociated with splicing repression (54, 59). A recent globalanalysis of PTB and its neural paralogue nPTB has revealed theirrole in the control of murine neuron-specific splicing (8).hnRNP G (RBM-X) has been implicated in the splicing controlof several pre-mRNAs including SMN and β_s-tropomyosin (38,45). The human hnRNP L protein promotes exon inclusion in theeNOS pre-mRNA (27) and exclusion of the variable exon in CD45(24). A recent microarray screen has identified several novelpre-mRNA targets of hnRNP L (28).

A role in splicing for the other major hnRNP proteins is lessconclusive, mostly because the number of examples remains limited.A function for hnRNP C in splicing is controversial since theinitial reports were never confirmed (13, 52). hnRNP K has beenassociated with splicing efficiency (18) and also with one caseof alternative splicing control in vivo using a minigene (56).Overexpression of hnRNP M can promote both exon inclusion andskipping events in a few alternative exons in 293T cells (25).hnRNP Q (Syncrip) is required for efficient pre-mRNA splicingin vitro (44), but the function of the very similar hnRNP Rprotein is unknown. It is unclear whether the mostly cytoplasmichnRNP D protein (AUF1) and the related DL protein can modulatesplicing.

To begin to assess the range of action of the major hnRNP proteinsthat function in mRNA processing and to explore the potentialfunction of the others, we have analyzed the individual capacitiesof several of the major hnRNP proteins to affect the productionof splice isoforms. We have selected alternative events thatbelong to a subset of apoptotic genes because the functionalconsequences of alternative splicing on apoptosis have beenamply demonstrated (51). Indeed, the complex and delicate balancebetween the activities of pro- and antiapoptotic variants producedby APAF-1, Bcl-x, Fas, and caspases is often controlled throughalternative splicing, and a number of studies have documentedthe contribution of hnRNP and hnRNP-like proteins in the controlof splice site selection in apoptotic genes. For example, PTBantagonizes U2AF binding to the upstream 3' splice site of Fasexon 6 by repressing exon definition (31). A role for PTB andhnRNP A1 has also been observed in the control of caspase-2pre-mRNA splicing (14). Likewise, hnRNP F/H, along with theRBPs ASF/SF2, and Sam68 have been implicated in the controlof Bcl-x splicing to produce pro- and antiapoptotic isoforms(21, 40, 48).

The first global study linking RBPs to alternative splicingwas carried out by examining the impact of downregulating hundredsof RBPs on three genes in the Drosophila S2 cell line (47).A subsequent study reported the effect of depleting four RBPson the alternative splicing of thousands of genes in the samecell line (6). In mammals, RNA interference assays were alsoused to study global effects of the hnRNP-like protein Nova2in the mouse neocortex, of PTB in a mouse neuronal cell line,and of hnRNP L in HeLa cells (8, 28, 57).

We have combined an RNA interference approach with our high-throughputreverse transcription-PCR (RT-PCR) screening platform (32) toanalyze the impact of downregulating 14 hnRNP proteins on 56alternative splicing events (ASEs) in apoptotic genes expressedin the human cell lines HeLa, PC-3, and BJT. We find that allhnRNP proteins tested can affect the ratio of some splice isoformsin human cells. The impact of the individual knockdowns variesconsiderably, ranging from <2% of ASEs being affected byhnRNP M to >40% of all events affected by hnRNP K. Althoughmany hnRNP proteins have the capacity to control the same splicingevents in two or three cell lines, the contribution of severalof these abundant proteins varies considerably in differentcellular environments.

MATERIALS AND METHODS

Top

MATERIALS AND METHODS

Cell lines and siRNAs. The cervical cancer cell line HeLa, the prostate cancer cellline PC-3, and an immortalized version of the normal foreskinBJ-TIELF (BJT) fibroblast cell line used in the present studyhave been described previously (49). HeLa, PC-3, and BJT cellswere, respectively, grown in Dulbecco modified Eagle medium,Ham F-12 medium, and alpha-minimal essential medium supplementedwith 10% fetal bovine serum. The siRNAs used to knockdown hnRNPprotein expression were purchased from IDT (Coralville, IA),and their sequences are listed in Table S1 in the supplementalmaterial. siRNAs were transfected into cells at a concentrationof 100 nM using Lipofectamine 2000 (Invitrogen). Proteins andRNA were extracted from mock-transfected and small interferingRNA (siRNA)-transfected cells 96 h posttransfection.

Western analysis and quantitative real-time RT-PCR. We used antibodies to assess the success of knockdown experiments.The same blot was decorated with an anti-actin antibody (SigmaA2066) to correct for total protein content in different lanes.Anti-hnRNP A1 and A2 antibodies and anti-hnRNP F and H antibodieswere described previously (49, 50). Anti-hnRNP C1/C2 and G (C-17)were from Santa Cruz Biotechnologies, Inc. (sc-10037 and sc-4583,respectively); anti-hnRNP D/AUF1 was purchased from UpstateBiotechnology, Inc. (catalog no. 07-260); anti-JKTBP (hnRNPDL) was kindly provided by Michiyuki Yamada (Yokohama City University,Yokohama, Japan); anti-hnRNP I (PTB) was kindly provided byDoug Black (University of California at Los Angeles); anti-hnRNPK was kindly provided by Karol Bomsztyk (University of Washington);anti-hnRNP L was purchased from Abcam (ab6106); anti-hnRNP Mwas purchased from Abnova (H00004670-A01); and anti-Syncrip(hnRNP Q) was kindly provided by Mike Mueckler (Washington University).

When antibodies were not available or the quality of the Westernanalysis was unsatisfactory, we conducted quantitative real-timeRT-PCR assays using Sybr green (Power Sybr green master mix,2x; ABI catalog no. 4367660) to assess transcript levels. AldolaseA (RTPrimerDB ID: 915) was used as the housekeeping gene onthe same samples. A total of 200 ng of RNA measured for integrity(using the Agilent Lab-on-Chip station) and quantification (usingthe Thermo Scientific NanoDrop) was reverse transcribed usingrandom hexamers (Roche catalog no. 11034731001) with Transcriptorreverse transcriptase (Roche catalog no. 03531317) in a finalvolume of 10 µl. Then, 10 ng of cDNA was used for quantificationin the presence of the specific primers at 0.2 µM in a10-µl reaction performed in triplicate. For the list ofgenes, oligonucleotides, and expected sizes of qPCR products,see Table S2 in the supplemental material. Reactions were carriedout in an ABI 7500 qPCR (Applied Biosystems, Foster City, CA)or Eppendorf Realplex. A first cycle of 10 min at 95°C wasfollowed by 40 cycles of 15 s at 94°C, 20 s at 55°C,and 20 s at 68°C.

RT-PCR assays. A set of 56 alternative splicing units from human apoptoticgenes was selected from the AceView database. Sets of primersmapping in the exons flanking the ASEs were designed by usingPrimer3 with default parameters. For the global analysis ofsplice isoforms from apoptotic genes, total RNA was extractedusing TRIzol and quantitated by using the Lab-on-Chip station(Agilent, Inc., Santa Clara, CA). A total of 2 µg of RNAwas reverse transcribed using a mix of random hexamers and oligo(dT)and the Omniscript reverse transcriptase (Qiagen, Germantown,MD) in a final volume of 20 µl. Then, 20 ng of cDNA wasamplified with 0.2 U/10 µl of HotStarTaq DNA polymerase(Qiagen) in the buffer provided by the manufacturer, withoutextra MgCl₂ and in the presence of the specific primers (IDT)for each splicing unit (at concentrations ranging from 0.3 to0.6 µM) and deoxynucleoside triphosphates. The list ofunits, oligonucleotides, and expected sizes of the RT-PCR productsare shown in Table S3 in the supplemental material. Reactionswere carried out in the GeneAmp PCR system 9700 (Applied Biosystems).A first cycle of 15 min at 95°C was followed by 35 cyclesof 30 s at 94°C, 30 s at 55°C, and 1 min at 72°C.The reaction was ended with the extension step of 10 min at72°C. Visualization and analysis of amplified products wasdone by using the LabChip HT DNA assay on an automated microfluidicstation (Caliper, Hopkinton, MA) (32).

Bioinformatic analysis. The search for hnRNP binding motifs in cassette exons or singlealternative 5' or 3' splice sites also included 150 nucleotidesof upstream and downstream sequences but did not include sequenceswhere the flanking introns were smaller than 150 nucleotides.The clustering was done by R using the R Project for StatisticalComputing (http://www.r-project.org/) with the Euclidean distancemetric.

RESULTS

Top

RESULTS

We carried out a loss-of-function study in three cell lines:the cervical cancer HeLa cell line, the prostate cancer PC-3cell line, and the BJT cell line (hTERT-immortalized foreskinfibroblasts derived from the BJ cell line) (49). Fourteen hnRNPproteins (A1, A2, C, D, DL, F, G [RBM-X], H, I [PTB], K, L,M, Q [Syncrip], and R) were each targeted with two distinctand nonoverlapping siRNAs. The combined knockdown of A1 andA2 was also performed. For the related hnRNP F and H proteins,we used one siRNA for each and a different siRNA targeting bothF and H. Depletion efficiencies were evaluated 96 h posttransfectionby Western blotting and/or quantitative RT-PCR assays (Table1 and Fig. S1 in the supplemental material). All assays indicatedthat depletion had been achieved, except for hnRNP H and hnRNPQ in BJT cells, which remained refractory to RNA interference-mediatedsilencing even after repeated attempts.

View this table:
[in this window]
[in a new window]

TABLE 1. Knockdown efficiencies

A set of 56 ASEs covering a variety of apoptotic genes was evaluatedto assess the impact of individual hnRNP protein knockdowns.Primers were designed to cover alternatively spliced regionssuch that two distinct splice products of sizes ranging between100 and 700 bp could be amplified. The relative abundance andsize of each amplification product was estimated by capillarymicrofluidic fractionation using Caliper reading stations. Foreach reaction, the relative abundance of the products was convertedto a percent splicing index ${Psi}$ , defined as the molarity of thelong product divided by the combined molarities of the longand short products. A ${Delta}$ ${Psi}$ value, defined as ${Psi}$ _knockdown – ${Psi}$ _control,was then computed. ${Delta}$ ${Psi}$ is therefore a measure of the relative shiftin splice site selection compared to the control transfectionperformed on the same day. ${Delta}$ ${Psi}$ values were calculated for eachASE in each cell line that was treated with a given siRNA. A ${Delta}$ ${Psi}$ with a |Z|-score of 1.5 (1.5 standard deviations above or belowthe average variation observed for all controls performed withall splicing units in this cell line) was considered significant(a "hit") only if it was detected with both siRNAs. Using twosiRNAs for each hnRNP protein is important because the correlationscores for pairs of siRNAs with similar knockdown efficienciesvaried greatly, indicating that some siRNAs can generate manyfalse positives.

From the nearly 5,000 ${Delta}$ ${Psi}$ values collected, 107 splicing eventshad ${Delta}$ ${Psi}$ values with |Z|-scores above 1.5 with both siRNAs (Fig.1). More than one-third of these hits resulted from knockingdown hnRNP K and C. To test the reproducibility of our hit assignment,we repeated the entire knockdown sets for hnRNP K and C. Asseen in Fig. S2 and S3 in the supplemental material, good agreementwas observed between the initial and the corroborative sets(sensitivity, 55%; specificity, 95%). Eighty-two percent ofall shifts and nonhits were confirmed.

View larger version (55K):
[in this window]
[in a new window]

FIG. 1. Alternative splicing of 56 ASEs in response to knockdown of 14 hnRNPs with two siRNAs. Individual hnRNP knockdowns and ASEs are shown to indicate which knockdowns caused a shift in alternative splicing in various cell lines (B = BJT, H = HeLa, and P = PC-3). The type of splicing events being followed is indicated on the right. Each column represents a distinct knockdown performed with the siRNAs listed in Table 1. Shifts with Z-scores of $≥$ 1.5 or $≤$ –1.5 are shown as bright red and blue bars, respectively, with shifts with Z-scores of $≥$ 2 or $≤$ –2 indicated with a darker color and Z-scores of $≥$ 1 or $≤$ 1 indicated with a lighter color. White areas indicate no shifts, and gray areas indicate no expression or failed PCRs or knockdown. ASEs that were significantly shifted by both siRNAs in all three cell lines are boxed in yellow. Also boxed in dashed yellow are examples where lower thresholds suggest an effect in all cell lines. (To access individual experiments, go to http://palace.lgfus.ca/to obtain an interactive version of the same figure. Then click on any bar to consult the electropherograms of RT-PCR products fractionated on a Caliper microfluidic station.)

Impact of the individual knockdowns. The profile for each individually targeted hnRNP protein isshown in Fig. 1. For the purposes of the discussion below, wehave combined the presentation of the hits for hnRNP A1 andA2, as well as for hnRNP F and H, because these proteins arestructurally very similar and are considered functional homologues.HnRNP D and DL, as well as hnRNP Q and R, are also combinedfor structural reasons.

hnRNP A1 and A2. The number of alternative splicing units controlled by A1 andA2 averaged 5% in HeLa and PC-3 cells for each protein. No hitswere obtained in BJT cells, presumably because the knockdownswere less efficient (Table 1). Five of the six A1 hits wereobtained when we knocked down both A1 and A2, but no new hitswere observed. A1 and A2 did not share common hits, indicatingdistinct target specificity. This is a surprising observationbecause A1 and A2 are considered functional homologues, althoughthis conclusion is based on in vitro studies (4, 10, 29, 42).The fact that only one of the five A2 targets shifted with adual knockdown is similarly intriguing. It may be explainedif downregulating A2 upregulates the expression of A1 and viceversa (49; data not shown). If A1 is the dominant splicing regulator,knocking down A1 would increase the expression of A2, but themajority of the hits would be caused by the A1 decrease. Incontrast, knocking down A2 would increase A1 expression to revealdistinctly sensitive units. Consistent with this view, we haveobserved in the data set with a lower |Z|-score two cases wherethe knockdowns of A1 and A2 shift splicing in opposite directions(data not shown).

Knocking down hnRNP A1 and A2 promoted a mixture of exon exclusionand exon inclusion events, a finding consistent with the recognizedbehavior of these proteins in splice site selection (39).

hnRNP C. Given the controversial role of hnRNP C in splicing, it wassurprising to note that a drop in hnRNP C expression affected30% of the ASEs, nearly half of them occurring in at least twodifferent cell lines. Moreover, knocking down hnRNP C only promotedexon skipping, suggesting that hnRNP C acts as a splicing enhancerand enforces exon inclusion. However, knocking down hnRNP Calso promotes the use of a more internal 3' splice site on LRDDwhen the Z-score is lowered. Such a result can be explainedif hnRNP C enhances the downstream 3' splice site or, alternatively,if it represses the upstream site. More work will be requiredto ascertain the mechanism(s) by which hnRNP C affects splicesite selection. Given the association of hnRNP C with the SWI/SNFchromatin remodeling complex (37) and the role of other SWI/SNFcomponents in alternative splicing (2, 3), hnRNP C may mediatethese effects cotranscriptionally. Among the hits seen in threecell lines, the depletion of hnRNP C promoted the skipping ofan alternative exon of 129 nucleotides in the poly(rC) bindingprotein 4 (PCBP4; Fig. 2). The corresponding protein lacks 43amino acids just downstream of its first KH domain. Exon exclusionin BCL2L12 is also observed upon hnRNP C depletion (Fig. 2),a situation that causes a frameshift and consequent deletionof the downstream BH2 domain.

View larger version (26K):
[in this window]
[in a new window]

FIG. 2. Examples of hnRNP-regulated ASEs (hits). Selected units that were affected by hnRNP C (PCBP4 and BCL2L12) and hnRNP K (APAF1 and PTK2B) knockdowns in all cell lines are presented. at the top is a diagram of the alternative splicing unit being monitored. Below each alternative splicing unit are gel-like representations of electropherograms corresponding to control and knockdown (+) samples. The position of the products (s = skipped, i = included) corresponds to the expected sizes.

hnRNP D and DL. Knocking down hnRNP D and DL had an impact on the ratio of afew isoforms, and there was no overlap in hits. Given that hnRNPD and DL are mostly known as cytoplasmic proteins involved incontrolling mRNA stability, it is possible that the changesthat we observed are caused by differences in the stabilityof the specific mRNA splice isoforms. Nonetheless, nuclear functionsfor hnRNP D have also been proposed (16, 30) and a direct impacton alternative splicing remains possible but will need to beconfirmed in future experiments.

hnRNP F and H. Three of the eleven hnRNP F and ten hnRNP H hits were common,indicating that these proteins are regulating different butoverlapping sets of ASEs. Interestingly, 80% of the events causedby downregulating F and H involved exon skipping, indicatinga prevalent role for these proteins in enforcing the inclusionof alternative exons, a finding consistent with their proposedrole in intron definition (39). Intriguingly, knocking downhnRNP F in PC-3 cells promoted the inclusion of a cassette exonin PCBP4, while the same treatment promoted its exclusion inBJT cells.

hnRNP G (RBM-X). Knocking down hnRNP G affected the splicing of seven ASEs, andmost hits were cell line specific. Both exon inclusion and exonskipping events were associated with the downregulation of hnRNPG. Notably, we observed that the depletion of hnRNP G promotedexon 8 skipping in MAPT (also known as Tau) in PC-3 cells. Arole for hnRNP G in Tau splicing was previously noted when overexpressingit in COS cells promoted Tau exon 10 skipping (60).

hnRNP I (PTB). Six ASEs were sensitive to the knockdown of PTB. Half of themchanged in at least two cell lines, making PTB a broad effectorlike hnRNP C and hnRNP K (see below). Downregulating PTB enhancedexon inclusion in four ASEs, which is consistent with the prototypicalinhibitory function of PTB (54). In one case, the depletionpromoted the inclusion of the penultimate exon of protein phosphatase3 (PPP3CB) in all three cell lines. In contrast, two cases ofimproved exon skipping were observed, suggesting that PTB doesnot always act as a repressor (46). A decrease in exon inclusionpromoted by PTB depletion was also recently observed in mousecell lines (8). The same study examined the impact of knockingdown PTB on the alternative splicing of approximately 1,300exons and showed that 25% of neuronal splicing in mouse is regulatedby PTB and nPTB. In N2A cells, PTB affected 5% of the globalASEs monitored. We had a slightly superior hit frequency forhuman PTB (10%), and half of these hits were seen in all threecell lines.

hnRNP K. Few natural ASEs have been associated with hnRNP K function.We show here that nearly 50% of ASEs are affected when hnRNPK is depleted. Moreover, most hits were detected in at leasttwo cell lines, and 25% of the hits occurred in all three celllines, making hnRNP K the most general effector in our set ofhnRNP proteins. The knockdown of Nova2, a member of the samefamily of KH domain-containing proteins affected 7% of the neuronalASEs (57). As was observed for Nova2, downregulating hnRNP Kpromoted a mixture of exon skipping and exon inclusion events,establishing hnRNP K as a bidirectional alternative splicingeffector.

The splicing changes elicited by the depletion of hnRNP K affectedimportant domains of apoptotic proteins. For example, downregulatinghnRNP K caused the skipping of a 129-nucleotide exon in apoptoticpeptidase activating factor 1 (APAF1) (Fig. 2). The corresponding43 amino acid residues are located in the first C-terminal WD40domain of this protein. Depletion of hnRNP K also promoted theinclusion of 44 amino acids just upstream of the C-terminalcaspase recruitment domain of the NLR family, pyrin domain-containing1 (NALP1). hnRNP K knockdown caused the exclusion of a 126-nucleotideexon (Fig. 2) encoding 42 amino acids between the tyrosine kinasedomain and the focal adhesion targeting region of protein tyrosinekinase 2 (PTK2B). Secreted phosphoprotein 1 (SPP1; also knownas osteopontin) had 14 residues excluded with both hnRNP K siRNAsin all cell lines. This splice variant is known as osteopontin-band has no known function. However, an adjacent splice variant(osteopontin-c) is implicated in breast and liver cancer (43).The breast cancer-associated protein C11orf17 (BCA3) is a proteinkinase A-interacting protein, and the splice variant lackingthe phosphorylation sites for various kinases is increased whenhnRNP K is depleted.

hnRNP L. Downregulating hnRNP L affected seven ASEs. Its depletion promotedsix cases of exon skipping and one case of exon inclusion, confirmingthe dual ability for hnRNP L to promote both exon skipping andexon inclusion events (24, 26-28). We found that BCL2L12 wasa target in all three cell lines, whereas PPP3CB was affectedin both HeLa and PC-3 cells. In a recent genomewide array analysis,Bindereif and coworkers (28) used the Affymetrix GeneChip Exon1.0 Array to find and validate 11 targets of hnRNP L in HeLacells; none of those ASEs was in our inquiry set.

hnRNP M. hnRNP M has been implicated in the splicing control of a fewgenes (25). Here we observed that downregulation of hnRNP Mwith two different siRNAs affected the alternative splicingof only one gene, APLP2. Induced inclusion was observed in allthree cell lines. Interestingly, this alternative exon of APLP2is also included in the MCF-7 breast cancer cell line and excludedin the MDA231 breast cancer cells relative to normal HMEC mammarycells (36).

hnRNP Q and R. Although hnRNP Q and R are considered structurally very similar,they did not target common ASEs. All hits (four for Q and threefor R) were cell line specific and while knocking down hnRNPQ only promoted the skipping of cassette exons, downregulatinghnRNP R increased exon inclusion in one case.

Generic and cell line-specific effects of hnRNP protein knockdowns. Although the depletion of hnRNP C, H, and Q only promoted exonskipping events, all other knockdowns produced a mixture ofexon skipping and inclusion events. Overall, downregulatingthe expression of hnRNP proteins promoted slightly more exonskipping events than exon inclusion events, further confirmingthat hnRNP proteins are not exclusively involved in exon repression.In only one case did the knockdown of an hnRNP protein havea different outcome in different cell lines; depleting hnRNPF in PC-3 cells promoted the inclusion of a cassette exon inPCBP4, while the same procedure promoted its exclusion in BJTcells. Except for the fact that all four exon skipping eventspromoted by the depletion of hnRNP G occurred in PC-3 cells,there was no apparent cell line bias toward exon inclusion orskipping.

The total numbers of hits were comparable in HeLa and PC-3 cells,but the total number was approximately half in BJT cells (Fig.3, total), most likely because of the smaller magnitude of someof the depletions achieved in the BJT cell line (Table 1), andthe fact that the initial level of many hnRNP proteins (A2,C, DL, F/H, G, K, L, and R) was generally lower in BJT relativeto the other cell lines (see Fig. S1 in the supplemental material).Out of nearly 5,000 reactions (56 ASEs in three cell lines knockeddown for 14 hnRNP proteins with two siRNAs), our analysis revealed10 ASEs that were controlled consistently by an hnRNP proteinin all three cell lines (solid yellow boxes in Fig. 1), with21 additional cases occurring in two cell lines. We noted 74cases where an hnRNP knockdown produced a hit in only one cellline. For five of them, the assessment in other cell lines wascompromised by the fact that the ASE was already fully splicedin the direction of the shift before treating with siRNAs. Twenty-onecases could be considered significant hits in all three celllines but with a |Z|-score inferior to 1.5 (indicated as dashedyellow boxes in Fig. 1). We identified 22 cases of clear cellline-specificity of hnRNP protein action (see Table S4 in thesupplemental material). The vast majority of these examplesof cell line specificity could not be explained simply by asuperior hnRNP protein knockdown efficiency. Further investigationswill be required to establish whether cell line specificityis due to a required threshold of hnRNP protein expression,to the existence of variant forms of hnRNP proteins, or to variationsin the combinatorial nature of splicing regulation.

View larger version (26K):
[in this window]
[in a new window]

FIG. 3. Representation of hits for each hnRNP protein in the three cell lines. Venn diagrams are used to indicate the number of significant alternative splicing units for each hnRNP (ASEs with shift |Z|-scores >1.5 with both siRNAs) that were found in BJT, HeLa, and PC-3 cells, as well as the overlap between hits.

Although PC-3 and HeLa cells shared ca. 40% of their hits, commonhits were more frequent with hnRNP K, C, and PTB (55% on average)than for other hnRNP proteins (22%). Thus, the range of actionof different hnRNP proteins varied considerably.

Networks of regulation. To assess the existence of regulatory networks involving ASEsand hnRNP proteins, we produced a cluster map (Fig. 4). Theclustering analysis confirms that hnRNP F and H had similaraction profiles. Overall, however, there was little target overlapbetween various hnRNP proteins. Even for the hnRNP proteinsassociated with most ASEs (K, C, and F+H), the hit overlap wasnot significantly above the frequency expected to occur by chance,suggesting that specific sets of hnRNP proteins do not systematicallyact in concert to control alternative splicing but rather havespecific and nonredundant roles.

View larger version (35K):
[in this window]
[in a new window]

FIG. 4. Clustered binary map of significantly shifting alternative splicing units. Binary ${Delta}$ ${Psi}$ data for the hnRNP knockdown affecting different ASEs in various cell lines. The cell line in which the shift was observed for a specific ASE is indicated (B = BJT, H = HeLa, and P = PC-3). Red indicates an increase in ${Psi}$ (e.g., exon inclusion or use of the more internal splice site in alternative 5' or 3' splice sites), and blue indicates a decrease, upon hnRNP knockdown. Only shifts with Z-score of ±1.5 or better obtained with two siRNAs are indicated. Gray bars indicate no change. Incomplete data representing depletions that failed (H and Q in BJT cells) or the absence of any RT-PCR products are shown as white bars.

A few genes in our selected set contained two distinct alternativesplicing units (C11orf17, CAPN3, NALP1, OPA1, and TNFRSF10Bb).In PC-3 cells, hnRNP K affected the alternative splicing ofboth ASEs of NALP1, while hnRNP F did the same for TNFRSF10B.In addition, hnRNP L influenced the splicing of both units ofCAPN3 in HeLa cells. Thus, a specific hnRNP protein can coordinatethe splicing of distinct exons in one gene. These results extenda previous observation indicating that the alternative splicingof mouse ASEs belonging to the same genes are often coordinatedin a tissue-specific manner (19).

Figure S4 in the supplemental material combines the data inFig. 1 to show which splicing events were sensitive to whichhnRNP proteins, regardless of cell line. Fourteen ASEs did notreact to any of the knockdowns, and nineteen were affected byonly one hnRNP protein. At the other end of the spectrum, afew units were sensitive to the downregulation of several hnRNPproteins. Nearly one-third of all units reacted to the downregulationof at least three different hnRNP proteins. In PC-3 cells, PPP3CBsplicing was affected by hnRNP A1, K, L, and PTB; CAPN3b byhnRNP K, DL, G, and C; and BCL2L12 by hnRNP H, Q, L, G, andC. These results indicate that an intricate and often cell line-specificnetwork of hnRNP proteins is dedicated to the control of individualASEs.

A few ASEs were controlled by two or more common hnRNP proteins.FGFR10P in PC-3 cells and CAPN3b in HeLa cells were regulatedby hnRNP K, F/H, and A2. In PC-3 cells, hnRNP Q, H, L, and Gall affected both BCL2L12 and MAPT. This last example may pointto the existence of networks of coregulation where the combinatorialaction of several different hnRNP proteins is converging tocoordinate the processing of a group of genes. However, in examiningthe positions of genes in the apoptotic pathway we could notdiscern biases in favor of specific portions of the pathwaycontrolled by individual or groups of hnRNP proteins (data notshown but see Fig. S5 in the supplemental material for the distributionof hnRNP K and C hits in the apoptotic pathway). Thus, hnRNPproteins act on a variety of genes involved at various stepsof the apoptotic pathway. Likewise, there was no indicationthat hnRNP proteins individually or collectively were dedicatedto the control of only pro- or only antiapoptotic events orproteins.

Our RT-PCR approach validated the control of 19 ASEs that werenot included as RefSeq events (see Table S5 in the supplementalmaterial). Many of the splicing events controlled by hnRNP proteinsare presumed to play a critical role in protein function becausethe events implicate residues directly located in functionalprotein domains (14 cases), change the reading frame (8 cases),or create a premature stop (2 cases) (see Fig. S6 and TableS5 in the supplemental material).

Binding motifs for hnRNP proteins in target genes. We have searched ASEs for known binding motifs of hnRNP proteins(Table 2), restricting our search to the 49 cases of alternativesplicing involving cassette exons or single alternative 5' or3' splice sites. The A1 binding motif UAGG (1, 15) was foundin all four A2 hits and one-third of the A1 hits. The A1 bindingmotif UAGAGU resembling the SELEX winner A1 binding site (9,29) was found in two A2 hits and more rarely in nonhits (5 of45), with a P value for motif enrichment of 0.086. Althoughno perfect match to the known six-nucleotide consensus bindingsites (e.g., UAGGGU and UAGAGU) was found in A1 hits, the submotifGGGU occurred in five of the six A1 hits and all four A2 hits.

View this table:
[in this window]
[in a new window]

TABLE 2. Prevalence of binding motifs for hnRNP proteins in ASEs

Only 5 of the 15 hnRNP C hits contained the motif UUUUU (22).Both hnRNP D motifs AUUUA and UUAG (35, 62) were found in oneof the two hits. For hnRNP F and H, nearly all 29 hits had atleast one GGGG binding motif (11). Examples of the distributionof these motifs in hits are shown in Fig. S7a in the supplementalmaterial. GGGG and GGG motifs were significantly more representedin the F hits than in the nonhits (P = 0.02 and 0.004, respectively),while this was the case only for the GGG motif in the H hits(P = 0.03).

The hnRNP G binding motif AAGU (45) was found in five of thesix hits. At least one of the PTB-binding motifs UCUU and CUCUCUwas found in every PTB hit, with a slight enrichment for theCUCUCU motif (P = 0.06). Half of the units reactive to a depletionof hnRNP K contained the optimal binding motifs AUC_3/4(A/U)₂(55). Although the enrichment for this motif in hits relativeto nonhits was modest (P = 0.06), it became more significant(P = 0.02) when only cases where K depletion increased exoninclusion were considered. This enrichment of AUC_3/4(A/U)₂ sitesupstream of included exons implicates hnRNP K in the suppressionof 3' splice site recognition and consequent inhibition of alternativeexons (see Fig. S7b in the supplemental material). Finally,the depletion in hnRNP L revealed diversely positioned bindingmotifs (ACAC, CACA, TACA, ACAT, and CACC) but no overrepresentationrelative to nonhits (see Fig. S7c in the supplemental material).

The presence of binding motifs in the vicinity of alternativesplice sites supports a direct role for the cognate hnRNP proteinin splice site selection. Although this was the case for themajority of the hits, there may be several reasons why not allhits contained the expected binding sites. First, the sequencesof all possible binding sites may differ from the binding motifsthat we have selected. Indeed, hnRNP proteins are notoriousfor their relaxed binding specificity (53). Second, bindingsites may act from a greater distance than the 150-nucleotidelimit that we imposed on our search. Elements controlling specificsplicing decisions can be located at a considerable distancefrom the splicing event that they control (20, 33). Third, hnRNPproteins may be components of a larger complex whose overallbinding selectivity may be dictated primarily by other RBPsin the complex. Fourth, changing the expression of one hnRNPprotein may affect the expression of other RBPs that in turncontrol splicing decisions. To assess this possibility, we carriedout quantitative RT-PCR assays to measure the transcript levelsof 14 hnRNP genes when hnRNP C or K were knocked down in PC-3cells (see Fig. S8 in the supplemental material). DepletinghnRNP C affected the expression of hnRNP D, DL, G, H, M, andR, but levels did not drop below 50% with both siRNAs. Onlytwo of the seven hnRNP C hits were detected in the other knockdowns(OPA1b and BCL2L12 with hnRNP H). Depleting hnRNP K affectedthe expression of hnRNP A2, D, DL, G, H, L, M, and R, with theexpression of D and DL falling below the 50% level. Of the 15hnRNP K hits in PC-3 cells, NALP1b, PTK2B, CASP1, and FGFR10Pwere also hits with hnRNP H, and CASP1 was a hit with hnRNPA2. On the other hand, one hit in common with hnRNP H (OPA1b)and two hits in common with hnRNP DL (BARD1 and CAPN3b) occurredin the opposite direction. Thus, reducing the expression ofspecific hnRNP proteins can affect the expression of other hnRNPgenes. Whether all of these changes in mRNA expression wereaccompanied by similar changes in hnRNP protein expression remainsto be examined. We could not detect significant changes in thelevels of hnRNP A1/A2 and F/H proteins in hnRNP K knockdowns,probably because more time is required before a difference inmRNA levels is converted into a difference in the abundanceof the cognate proteins. Nevertheless, these results raise thepossibility that some of the hits may be due to secondary effects,a situation that may explain the larger number of hits obtainedfor hnRNP C and K. A systematic analysis will be required todetermine the full extent of cross talks between different RBPs.

Despite a slight bias in hits for motifs representing bindingsites for some hnRNP proteins, additional features beyond themere presence of binding sites are clearly required for an hnRNPprotein to impact alternative splicing. A more extensive examinationof the structural environment, sequence, and position of bindingmotifs relative to splice sites may reveal additional insightsinto the requirements for splicing control. Moreover, the factthat a majority of splicing units is controlled in a cell line-specificmanner indicates that binding sites for a specific hnRNP proteinare not sufficient to impose regulation in all cellular contexts.Hit differences between cell lines may reflect differences inthe abundance or nuclear availability of given hnRNP proteinsor the dominance of other factors implicated in the combinatorialnetwork regulating specific ASEs.

DISCUSSION

Top

DISCUSSION

Our study is the first to examine the impact of downregulatingseveral mammalian RBPs in multiple cell lines (Fig. 1). Thisstudy was conducted with a unique RT-PCR screening technologythat provides high sensitivity in the identification of thevalidated targets, thereby offering a robust experimental systemfor elucidating the mechanism of hnRNP function in alternativesplicing.

Because hnRNP proteins bind RNA, it is likely that the effectson splice site selection result from a direct interaction withthe target pre-mRNAs. However, it is possible that a changein the relative abundance of splice isoforms may be achievedby differential mRNA stability, and many hnRNP proteins shuttleto the cytoplasm where they might affect mRNA turnover. In addition,because many RBPs are themselves alternatively spliced, changingthe expression of an hnRNP protein may affect the alternativesplicing and hence the activity of other RBPs in splicing control.Nevertheless, by examining the impact of reducing the abundanceof individual hnRNP proteins we have provided a first glimpseof the complex contribution of hnRNP proteins to the productionof splice isoforms in apoptotic genes, and our work sets thestage for a more detailed characterization of the mechanismsof action of the individual hnRNP proteins that target and controlalternative splicing and for a more extensive characterizationof the network of splicing factors that impact apoptosis.

Downregulating the expression of individual hnRNP proteins affectedalternative splicing in a very specific manner, with littleoverlap in the identity of the targeted events. An hnRNP proteinthat controls a specific ASE in one cell line does not necessarilyaffect the same splicing event in other cell lines. Althoughwe have identified 10 splicing events that are controlled bythe same hnRNP proteins in all three cell lines, many ASEs areregulated by hnRNP proteins in a cell line-specific manner.Therefore, comparable expression and knockdown efficienciescannot predict whether a specific splicing event will be regulatedby the same hnRNP proteins in different cell lines. This maybe because the activity of an hnRNP protein is modulated posttranslationallyto affect its cellular distribution and activity (23, 58, 61).Alternatively, other proteins may antagonize its binding and/oractivity. Given the combinatorial nature of alternative splicingcontrol, changes in the relative contribution of different RBPsmay change considerably in different cellular environments.This variation may offer more flexibility in the way alternativesplicing can be regulated, ultimately increasing the compositionof splice isoforms from different genes to expand phenotypicdiversity.

ACKNOWLEDGMENTS

We thank Johanne Toutant for preliminary RNA interference assaysand Western analysis. We thank Marco Blanchette for helpfulcomments on the manuscript. We thank Michiyuki Yamada, DougBlack, Karol Bomsztyk, and Mike Mueckler for antibodies.

This study was supported through funding from Genome Canada,Genome Quebec, and the Canadian Institutes in Health Research(B.C.). S.A.E. is a Chercheur-Boursier Senior of the FRSQ. B.C.is the Canada Research Chair in Functional Genomics.

FOOTNOTES

* Corresponding author. Mailing address: Département de Microbiologie et d'Infectiologie, Faculté de Médecine et des Sciences de la Santé, 3001, 12th Avenue Nord, Université de Sherbrooke, Sherbrooke, Québec J1H 5N4, Canada. Phone: (819) 564-5321. Fax: (819) 564-5392. E-mail: Benoit.Chabot{at}USherbrooke.ca

Published ahead of print on 21 July 2008.

Supplemental material for this article may be found at http://mcb.asm.org/.

REFERENCES

Top