A tour through the transcriptional landscape of platelets

Despite their small size and anucleate status, platelets have a rich repertoire of RNAs, including messenger RNAs (precursor-mRNA [pre-mRNA] and mature mRNA), structural and catalytic RNAs (ribosomal RNA [rRNA], transfer RNA, and small nucleolar RNA), and regulatory RNAs (microRNA [miRNA], long intergenic noncoding RNA, pseudogenes, and antisense RNA). Examination of RNA expression patterns in platelets has been used to identify biomarkers of disease, explain genetically or environmentally induced alterations in platelet function, and determine if genes are conserved between humans and mice.^1-13  The past decade has also revealed that platelets can translate mRNA into protein or transfer RNA to recipient cells where it regulates functional processes.^14-17  Because platelets are anucleate, transcriptional production of RNA (with the notable exception of mitochondria-derived transcripts) presumably occurs exclusively in the megakaryocyte. Therefore, the RNA profile in platelets also provides an accessible window into the transcriptional status of megakaryocytes, and the bone marrow signals involved, at the time of platelet production.

Several recent reviews have elegantly discussed the predictive value and functions of platelet RNAs.^2,18,19  This review explores the landscape of platelet RNA as mapped by next-generation RNA sequencing (RNA-seq). Specifically, we examine what a platelet mRNA looks like at the molecular level and how specific features of the mRNA impact biological responses. Along the way, we provide a series of Web links and access instructions (see Table 1)^1,20-22  for readers to follow as they explore key landmarks of platelet mRNA. Each stage of the tour (Figure 1) is discussed with reference to platelet function and in a broader biological context. Although screen shots are shown in figure format, this review is meant to be interactive and dynamic, with the goal of providing the tools necessary for researchers to explore the innumerable transcript features in platelets in the context of their own research interests.

To appreciate key features of platelet mRNA, one first has to understand the fundamentals of next-generation RNA-seq. RNA-seq has revolutionized medical research and, as shown in Table 2, several groups have recently used this powerful technique to profile platelet mRNA.^20,21,23 

RNA-seq libraries are prepared from either total RNA or a subfraction of RNA (ie, polyadenylated, rRNA-depleted, 5′ cap–containing small RNA, etc.) by random shearing of RNA into 100- to 400-bp fragments. Barcodes and sequencing primer binding sites are then added, and the RNA is briefly amplified, followed by size selection of 100- to 400-bp fragments (also called inserts). This size selection captures most protein-coding RNAs and long noncoding RNAs. For small RNA-seq, which captures mature mRNAs, piwi-interacting RNAs, and other small RNAs, fragmentation is skipped, and inserts <40 bp are selected. Following insert selection, ∼30 bp up to >150 bp of 1 end (single-end read) or both ends (paired-end reads) of each fragment are sequenced. Sequences (reads) are then computationally mapped by finding unique (usually) matches between the annotated genome/transcriptome and the sequenced end of the RNA fragment. If paired ends are sequenced, then each end is matched with the expected distance constrained to the approximate expected fragment size. If libraries are stranded, then the read can be mapped directionally to either the positive or negative strand of DNA. RNA-seq methodology is reviewed in more detail elsewhere.^24-27 

Four long RNA-seq data sets (ie, protein coding and long noncoding RNA) from human platelets and 1 from mouse platelets have been published.^20,21,23  Other small RNA-seq data sets (ie, mature miRNAs) have also been published but are not discussed here.^20,28  As shown in Table 2, each RNA-seq data set offers some complementary features not provided by the other. For example, in our experience, shorter (36 bp) reads map better to poorly annotated transcripts or novel small exons, whereas longer reads (101 bp) aid in isoform discrimination and map better to lower-complexity regions. Paired reads are useful for determining splicing and exon exclusion patterns. Stranded reads discriminate between genes overlapping in the genome, but transcribed from opposite strands. Oligo-dT enriches for protein-coding transcripts, whereas ribosomal depletion of total RNA will capture more noncoding transcripts. In 1 of their samples, Kissoupoulou et al²³  use a 5′ end–tagging strategy that marks the extreme 5′ terminus of transcripts. The aligned reads from 3 of these data sets are available for rapid navigation via Web links. Instructions to access and visualize the aligned data in the UCSC Genome Browser²⁹  are given in Table 1.

RNA-seq creates snapshots of mRNAs that provide a wealth of information, from gene expression to regional features and detailed sequence.²⁴  Before exploring subdomains of platelet mRNA, it is helpful to understand the topography of snapshots that are used to represent RNA-seq results. Figure 2A represents a visualization of RNA-seq alignments to glycoprotein IX (GP9). In this example, GP9 is annotated as it appears in platelets, a mature message that does not contain intronic sequences. Note that individual paired reads, which in this case are ∼36 bp in length, map across the entire GP9 annotated transcript. The reads that cover each nucleotide along the transcript are summed, and the count (y-axis) is displayed in the histogram-like graph (coverage or read depth graph) above the reads. Various methods, ranging in robustness, for calculating expression exist and are reviewed by Dillies et al and Garber et al.^30,31  One simple and commonly used index of abundance is the read (or fragment) per kilobase normalized to a million.²⁷  This is calculated from the sum of all the reads across a transcript, normalized to the transcript length and to the total number of reads mapping to the entire transcriptome. Variability in fragmentation, sequencing, and mapping efficiency for subsequences within the transcript affects the read depth at each loci, hence the “hills” and “valleys” of expression across the transcript.³¹  Thus, RNA-seq can quantify per-base, per-exon, per–annotated region, or per-transcript expression on a genome-wide scale without prior knowledge of the transcript.²⁴ 

Unlike the depiction in Figure 2A, RNA-seq data are typically visualized in genomic space using a genome browser such as the UCSC Genome Browser.²⁹ Figure 2B depicts the same transcript, GP9, in a genome browser. Note now that the exons are separated by the introns encoded in the genome, and like canyons between peaks of expression, the introns are devoid of aligned reads. This is because GP9 is completely spliced. Reads connected by thin lines represent either connected paired or split reads. Although the intron spanned by paired reads is much longer than the expected 200-bp fragment size, when the transcript is spliced the paired reads map at the expected distance.

Although anucleate, platelet mRNAs mirror mRNAs found in nucleated cells. Eukaryotic RNAs are prepared for nuclear export and translation by the addition of a 5′ 7-methylguanosine (7mG) cap, the addition of a 3′ poly-A tail of variable length, and the splicing out of introns.³²  The 5′ cap facilitates ribosome loading for protein translation via recognition and recruitment by eukaryotic initiation factor 4E (eIF4E).^33,34  The poly-A tail helps circularize mRNAs through interaction with the scaffold protein eIF4G, which initiates cap recognition by eIF4E.³³  Both ends protect mRNA from degradation.³³  The poly-A tail can be extended or shortened over the life of the RNA, but gradual poly-A shortening followed by decapping is the eventual fate of most RNAs.^35,36  In other mammalian cells, typically only 5′-capped, 3′-polyadenylated, spliced mRNA will be exported from the nucleus and used for translation.³²  The majority of the unspliced cytoplasmic message is rapidly degraded by nonsense mediated decay (NMD).³⁷ 

The customary use of oligo-dT to prime reverse transcription leaves little doubt that the majority of platelet transcripts are polyadenylated, consistent with descriptions of poly-A⁺ RNA in platelets reported by Roth and colleagues in 1989.³⁸  Platelets contain poly-A binding proteins such as activated-platelet protein-1³⁹  that may support poly-A–stimulated protein translation. Evidence that a subset of transcripts is 5′ capped comes from reports that protein translation can be initiated in platelets.⁴⁰  The poly-A tail is not genomically encoded, and stretches of As cannot be effectively aligned from RNA-seq reads. Therefore, poly-A reads are hidden in typical RNA-seq work flows. The 5′ cap is also not detected by normal RNA-seq reactions.

To characterize capping and polyadenylation in platelets, we have performed RNA-seq on transcripts captured with oligo-dT–conjugated beads or by a hyperaffinity mutant eIF4E protein that enriches for transcripts with a 7mG cap.⁴¹ Figure 3A offers a pictorial of this strategy. When comparing RNA-seq of poly-A–isolated mRNA with cap-captured RNA, poly-A isolation captures the majority of annotated protein-coding transcripts, whereas some transcripts with a very short or no poly-A tail are missed. Such is the case for many histone-coding transcripts with characteristically short or no poly-A tails,⁴²  which are missed by poly-A capture, yet readily observed when isolated by the 5′ cap–based strategy (Figure 3B). Conversely, a handful of mRNAs detected by poly-A–based methods are not captured by 5′-based methods (not shown). Aside from these exceptions, the bulk of annotated platelet transcripts that are captured by oligo-dT are also captured by cap-binding strategies (J.W. Rowley and A.S. Weyrich, unpublished data). A prime example is PAR1, whose mRNA was captured by both 3′- and 5′-based methods (Figure 3B). Other transcripts, in particular noncoding RNA, are undoubtedly present in platelets that contain neither a 5′ cap nor a 3′ poly-A tail.²⁰ 

The beginning of the poly-A tail can be inferred from the reads by a sharp decline in read depth at the end of the 3′ UTR accompanied by 3′-end mapped reads containing strings of As.^23,43  This type of analysis was used by Osman and coworkers to identify 2 polyadenylation start sites for transcript pro-platelet basic protein.²³  On the other hand, the length of poly-A tails in platelets and whether tail shortening occurs over time are not known. This is in part because new and old populations of platelets are difficult to separate in vivo.

In vitro studies suggest that some platelet mRNAs are not stable over time. Reticulated platelet counts potentially decrease after platelet storage.^44,45  mRNA for P-selectin and glyceraldehyde-3-phosphate dehydrogenase disappear in stored platelets at 22°C but decline only 40% over 5 days when stored at 4°C.⁴⁶  Sulfotransferase family cytosolic 2B member 1b (SULT2B1B) mRNA, which affects platelet functional responses via catalysis of the sulfonation of cholesterol,⁴⁷  also diminishes rapidly at 37°C.⁴⁸  SULT2B1B mRNA levels decrease by 40% within just 30 minutes at 37°C but are stable at 4°C. Interestingly, addition of high-density lipoprotein to the platelets slowed the decay. Of note, SULT2B1B protein levels and its product level of cholesterol sulfate tracked with mRNA levels suggest continued protein translation and functional implications of mRNA stability in platelets. Whether the poly-A tail and cap are involved in the stability and translation of these and other transcripts in platelets merits further investigation.

Surprisingly, platelets contain a subset of pre-mRNAs that have escaped the nucleus and NMD without splicing.⁴⁹  Platelet activation induces splicing and enables protein production from the newly formed mRNA. Unspliced pre-mRNA can be observed in RNA-seq data visually or bioinformatically by examining the ratio of intronic to exonic reads. Interleukin-1β, a prototypical unspliced mRNA in platelets,⁴⁹  contains intronic reads in relatively high proportion to exonic reads.²  According to our own analysis of platelet RNA-seq data sets, transcripts like interleukin-1β, with a large fraction of completely unspliced pre-mRNA, are infrequent. Despite this, additional examples of pre-mRNA splicing followed by protein production in platelets have surfaced in the literature.^50,51  In addition to these, using RNA-seq we have bioinformatically identified a handful of putatively unspliced transcripts in platelets that retain 1 or more introns. One example shown in Figure 3C is FOSB. FOSB is a transcription factor that regulates a wide array of cellular functions.⁵²  Whether FOSB can be spliced within platelets remains to be determined.

Serial analysis of gene expression in platelets previously demonstrated that the set of transcripts in platelets have longer UTRs than the set of transcripts found in other cells.⁵³  UTRs can also be assessed at the individual level. Figure 4A-B shows examples of transcripts in platelets where the actual UTR does not match the predicted UTR. The 5′ and 3′ UTRs regulate RNA stability and protein translation.^54-56  Within the 5′ UTR, secondary structure and primary sequence motifs direct ribosome scanning and initiation of translation.⁵⁷  A kozak sequence within the 5′ UTR that contains the AUG start codon flags the ribosome to start translation of the open reading frame (ORF). Premature upstream ORFs regulate the efficiency of initiation at the coding ORF, and secondary structures in 5′ UTRs further affect ribosome loading and scanning efficiency.⁵⁸  Proteins and small RNAs (ie, miRNAs) bind to sequence motifs within the 5′ UTR and 3′ UTR to regulate translation or degradation.^54,55,57,59  In platelets, for example, the 5′ UTR of B-cell lymphoma 3 controls translation of the mRNA in an mammalian target of rapamycin–dependent manner.⁶⁰  Given the many sequence-dependent effects of the UTR on RNA stability and translation, the predicted biological relevance of an annotated UTR sequence vs the actual UTR expressed in platelets could be profoundly different.

Growth factor independent 1B transcription repressor (GFI1B) controls megakaryocyte development,⁶¹  and GFI1B mutants can cause diseases of abnormal platelet formation and function like gray platelet syndrome (GPS).⁶²  In Figure 4A, the annotated 5′ UTR of GFI1B covers ∼350 bp of exons 1 and 2. RNA-seq reads in platelets suggest that transcription begins an additional 300 bp upstream from the annotated transcripts (Figure 4A, upper panel). The preponderance of 5′-terminally tagged reads (Figure 4A, lower panel) that map nearly exclusively to the start of the extended region suggests a dedicated transcription start site rather than aberrant run-on transcription. The additional 5′ sequence contains 3 short upstream ORFs.

miRNAs regulate RNA stability and protein translation by binding nearly complementary target sequences in the 3′ UTR of transcripts.⁶³  In platelets, miRNAs are abundant and stable.^64,65  They have been reported as biomarkers and as functional modifiers of platelets.^18,19,66  Interestingly, platelets can transfer miRNAs to regulate targets in other cells.^15-17  miRNA studies in platelets are increasingly popular and are the subject of several reviews.^18,19,67,68 

miRNA studies incorporate target prediction algorithms based on 3′-UTR annotations to identify potential miRNA-mRNA pairs.⁶⁹  Differences in 3′-UTR annotations may affect the outcome of such studies. P2RY1 accounts for ∼1/3 of adenosine diphosphate receptors on platelets.⁷⁰  Stimulation of P2RY1 by adenosine diphosphate induces intracellular calcium fluxes and platelet aggregation.^71,72  P2RY1 is a potential target of antithrombotics,^70,72  and differences in P2RY1 expression alter bleeding times and thrombosis in mice.⁷³  In Figure 4B, the annotated 3′ UTR of P2RY1 is nearly 1 kb, whereas the actual 3′ UTR in platelets is considerably longer. According to miRNA target prediction algorithms (miRanda⁷⁴ ), this additional 2.5 kb harbors additional miRNA binding sites (Figure 4B). This indicates that actual transcript data, not predicted annotation data, should be used for miRNA target site predictions in platelets. Sequences in view can be extracted directly in the UCSC Genome Browser by selecting “DNA” in the “view” menu.

Single nucleotide polymorphism (SNP) association studies⁷⁵  and mechanistic studies of platelet endothelial aggregation receptor 1 (PEAR1) link it strongly to an aggregatory role in platelets.⁷⁶  As illustrated in Figure 5A, RNA-seq distinguishes between multiple annotations for PEAR1. Alternative splicing affects nearly every multiexon eukaryotic transcript.⁷⁷  Exon skipping, intron retention, alternate exon usage, and addition or truncation of exons are examples of alternative splicing.^77,78 

Many alternative splice variants lead to NMD.³⁷  Other splice variants regulate protein synthesis or code for alternate protein products. Splice variants of the same gene can effect opposing functions, and some variants may alter physiological responses to drugs.⁷⁹  Proteomics detects many novel protein isoforms in platelets. In one proteomics study,⁸⁰  a number of proteins with novel exon–skipping events were identified. Three of these, integrin (ITG) A2, fumarate hydratase, and aminopeptidase puromycin sensitive, were further validated by polymerase chain reaction at the RNA level.⁸⁰  Interestingly, activation altered the number of exon-skipping events.⁸⁰ 

We have seen how alternate initiation and termination can affect regulatory elements within UTRs. Alternative splicing is another way to alter the 5′ and 3′ UTR. As shown in Figure 5B, RNA-seq can distinguish TFPI splice variants in platelets. TFPI inhibits tissue factor activity and thereby limits excessive coagulation.⁸¹  Multiple splice variants have been characterized for TFPI. Megakaryocytes and platelets express TFPIα, 1 of 2 major well-characterized isoforms⁸¹  produced by alternative splicing of the 5′ and 3′ end of the mRNA. Part of the 5′ UTR in exon 2, which can be removed by alternative splicing,^81,82  represses TFPI translation. On the other hand, the 3′ UTR of the TFPIα isoform in platelets relieves exon 2–mediated repression.⁸²  The proteins produced by TFPI splice variants are also different. The C termini of the TFPI isoforms differ in tissue factor inhibitory activity.⁸³  Furthermore, TPFIα is soluble and not linked to the cell surface like TFPIβ.⁸³ 

The production of a soluble vs membrane-bound form of proteins is a recurring theme of alternative splicing in platelets. Soluble P-selectin is shed from platelets after activation, mediated in part by extracellular cleavage.⁸⁴  Alternative production of soluble P-selectin may also be involved. Complementary DNA sequencing predicted the presence of a soluble splice variant of P-selectin.^85-87  Correspondingly, soluble P-selectin is found in the nonmembrane fraction of platelets, totaling up to 10% of total P-selectin levels.⁸⁸  Splice variants of ITGB3⁸⁹  code for a truncated protein, with the transmembrane and cytoplasmic domains missing. Protein assays have confirmed the expression of truncated ITGB3 protein in platelets.⁸⁹  Platelets contain multiple forms of the low-affinity immunoglobulin Fc region receptor IIa mRNA. Variants with and without the transmembrane exon are detected at approximately equal proportions in platelets.⁹⁰ 

Several other notable splice variants with relevance to platelet function have been reported. A variant of cyclooxygenase 2a is increased 200-fold after coronary artery bypass grafting.⁹¹  Variants of phospholipase C, β2 code for 2 different proteins in platelets.⁹²  An angiopoietin-1 splice variant that opposes its parent gene is found in platelet α-granules.⁹³  These examples only scratch the surface of the extensive collection of splice variants found in platelets. Some platelet-specific splice variants, including novel exons, have yet to be annotated within major databases. Such is the case with GFI1B (Figure 5C).

Noncoding RNAs regulate the production of target mRNAs or the translation of target mRNAs into protein, or they directly modulate the target proteins themselves. Platelets contain all major classes of noncoding RNAs including miRNAs (see “Regulatory regions: 5′ and 3′ UTRs”), long intergenic noncoding RNAs, pseudogenes, and antisense RNA. Of these, platelets are particularly enriched in miRNAs, pseudogenes, and antisense RNAs.²⁰ 

In eukaryotic cells, a major fraction of expressed genes are accompanied by an antisense.⁹⁴  Antisense transcripts regulate their complementary counterpart in numerous ways. Antisense RNAs mask miRNA binding sites, destabilize transcripts, repress translation, and promote or repress transcription.^94,95  They mediate both cytoplasmic and nuclear RNA-RNA and RNA-DNA interactions.⁹⁴ Figure 6 depicts an example antisense RNA found at the 5′ end of CD109. CD109 is a highly expressed human platelet alloantigen.⁹⁶  In this example, the antisense to CD109 begins at the 5′ end of CD109.

By zooming to the sequence level, individual SNPs, insertions, and deletions that are within the expressed transcript can be identified by individual RNA-seq reads. Searching for “platelet” within the “clinical synopsis” field within Online Mendelian Inheritance in Man (http://omim.org/, April 4, 2014 update) retrieves 79 different inherited platelet disorders, corresponding to 59 different defective genes. The inherited platelet disorders include well-characterized platelet gene/phenotypes such us ITGA2B/Glanzmann thrombasthenia⁹⁷  or neurobeachin-like 2/GPS.^98-100  Clicking on the genome coordinates (GRCh37) listed within any of these disorders while the RNA-seq platelet data UCSC Genome Browser window is open will bring you directly to the location listed overlaid with read sequence information. Of note, visualization of mutations in RNA-seq read sequences from a patient with GPS helped characterize a splicing defect in their neurobeachin-like 2 transcripts.⁹⁸ 

Genome-wide association studies (GWASs) have identified SNPs near >40 different genes associated with platelet size, count, or function.^101-105  Meta-analyses have extended this number to ∼80 (for a convenient list, see Bunimov et al¹⁰² ). In GWASs, a trait-associated SNP serves as a regional marker of allelic association without necessarily being the cause of the trait.^106,107 Figure 7 depicts a reported GWAS common variant associated with platelet counts, rs6065,^101,108  which is found within the RNA-seq read sequences of GP1BA compared with another individual without the variant. As we understand more about the meaning of GWAS SNPs, the ability to link RNA transcripts with marker SNPs and the causative mutations of trait variation may become important.

After completing this tour, it should be obvious that platelets have a complex mRNA signature that reflects and affects their function. Indeed, mRNA expression profiling was recently used to identify the molecular basis for differences in platelet reactivity between blacks and whites.⁷  The study found by microarray that phosphatidylcholine transfer protein (PCTP) mRNA is fourfold higher in blacks than in whites. This correlated with PCTP protein levels and platelet reactivity to PAR4. Another RNA, miR-376c, inversely correlated with platelet reactivity and regulated PCTP expression. This study is mentioned to underscore how platelet mRNA can be leveraged to identify and understand the molecular mechanisms that control human disease. Of note, the RNA-seq reads from 5 black and 5 white individuals’ platelets were subsequently published.²¹  In this data set (which is site 4 of Table 1 in this review), consistent with their microarray findings, PCTP mRNA is sevenfold higher in blacks than in whites.²¹ 

The intrinsic functional capacity and variability of platelets are determined by environmental signals and genetic factors seen by the megakaryocyte. As has been done for other cells,¹⁰⁹  integrating platelet and megakarycoyte RNA-seq with whole-genome sequencing or genome-wide microarray genotyping will provide valuable information regarding the relationship between the environment and genetics of altered gene expression within platelets. In addition to expression, RNA-seq analysis provides a tool for examining transcripts, and their plausible roles in platelet function and disease, at several other layers within the platelet terrain including the 5′ cap and poly-A tail, the 5′ and 3′ UTRs, intron retention, alternative splicing, strand, and single nucleotide sequences (summarized in Figure 8). In conjunction with published reports, we predict that forthcoming RNA-seq data sets will provide unprecedented opportunities to assess both “what’s in a platelet” and “what’s different about platelets in disease.”

We are grateful for the contributions of Diana Lim in preparation of the figures.

This work was supported by National Institutes of Health, National Institute of General Medical Sciences awards 1U54 HL112311-01, 1K01 GM103806-01, and 2R01 HL066277-11, as well as by the German Research Foundation (SCHU 2561/1-1).

Contribution: S.S. and J.W.R. drafted and prepared the manuscript; and A.S.W. reviewed and edited the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Jesse W. Rowley, Department of Internal Medicine, University of Utah School of Medicine, Eccles Institute of Human Genetics, Building 533, Room 4260, 15 North 2030 East, Salt Lake City, UT 84112; e-mail: jesse.rowley@u2m2.utah.edu.