Characterization of all small RNAs in and comparisons across cultured megakaryocytes and platelets of healthy individuals and COVID-19 patients

Abstract

Background:

The small noncoding RNAs (sncRNAs) in megakaryocytes (MKs) and platelets are not well characterized. Neither is the impact of SARS-CoV-2 infection on the sncRNAs of platelets.

Objectives:

To investigate the sorting of MK sncRNAs into platelets, and the differences in the platelet sncRNAomes of healthy donors (HDs) and COVID-19 patients.

Methods:

We comprehensively profiled sncRNAs from MKs cultured from cord blood–derived CD34⁺ cells, platelets from HDs, and platelets from patients with moderate and severe SARS-CoV-2 infection. We also comprehensively profiled Argonaute (AGO)–bound sncRNAs from the cultured MKs.

Results:

We characterized the sncRNAs in MKs and platelets and can account for ~95% of all sequenced reads. We found that MKs primarily comprise microRNA isoforms (isomiRs), tRNA–derived fragments (tRFs), rRNA–derived fragments (rRFs), and Y RNA–derived fragments (yRFs) in comparable abundances. The platelets of HDs showed a skewed distribution by comparison: 56.7% of all sncRNAs are yRFs, 34.4% are isomiRs, and <2.0% are tRFs and rRFs. Most isomiRs in MKs and platelets are either noncanonical, nontemplated, or both. When comparing MKs and platelets from HDs, we found numerous isomiRs, tRFs, rRFs, and yRFs showing opposite enrichments or depletions, including molecules from the same parental miRNA arm, tRNA, rRNA, or Y RNA. The sncRNAome of platelets from patients with COVID-19 is skewed compared to that of HDs with only 19.8% of all sncRNAs now being yRFs, isomiRs increasing to 63.6%, and tRFs and rRFs more than tripling their presence to 6.1%.

Conclusion:

The sncRNAomes of MKs and platelets are very rich and more complex than it has been believed. The evidence suggests complex mechanisms that sort MK sncRNAs into platelets. SARS-CoV-2 infection acutely alters the contents of platelets by changing the relative proportions of their sncRNAs.

1 ∣. INTRODUCTION

Hemostasis refers to the physiologic processes that maintain both blood fluidity under physiologic conditions and the integrity of a closed circulatory system after vessel injury. Blood platelets are small anucleate blood cells produced by megakaryocytes (MKs) [1] that serve as the cellular component of hemostasis. Platelets play a crucial role in the prevention of excessive blood loss by forming a plug at the site of a damaged blood vessel and by providing the phospholipid surface required by several enzymatic reactions leading to the formation of fibrin.

Inherited and acquired platelet disorders are common causes of pathologic hemorrhage in adults and children, most often due to low platelet numbers (thrombocytopenia) [2,3]. Qualitative platelet disorders also cause bleeding through acquired or inherited mechanisms [4,5]. At the other end of the hemostasis spectrum, platelet hyperre-activity is associated with occlusive platelet thrombi in coronary and cerebral arteries [6]. Also, abnormally high platelet counts in myeloproliferative neoplasms are a major cause of death and morbidity due to thrombosis and bleeding [7]. Aside from the well-studied role in hemostasis, viral infections are known to alter platelet function [8]. For example, a major complication of severe COVID-19 is a diffuse microvascular thrombosis: as we recently showed, platelets from these patients have an altered protein-coding transcriptome [9] and a prothrombotic platelet phenotype [10-12].

Human bone marrow is difficult to obtain and mature MKs represent only ~0.05% of nucleated marrow cells in adults. Consequently, the most common approach to studying human MK biology is to isolate hematopoietic stem cells from human umbilical vein cord blood (CBMKs) or adult venous blood (VBMKs)—often after “mobilization” with growth factors—and differentiate the hematopoietic stem cells into platelet-producing MKs in culture using thrombopoietin. CBMKs and adult VBMKs are very useful surrogates for primary mature MKs, having similar functions, agonist-induced signaling properties [13], a robust cytoplasmic invaginated membrane system, alpha and dense granules, and surface membrane proteins, and can form proplatelets [14]. CBMKs are fetal in nature and differ from adult MKs: CBMKs are smaller, have lower ploidy, and form ~10× more MKs in liquid culture than adult MKs. Mechanistic studies designed to understand the differences between CBMKs and adult MKs identified cell-intrinsic factors and local microenvironment–linked ones [14]. Heritability strongly influences platelet reactivity [15] and presumably the MK progenitor; however, there are gaps in understanding the responsible genetic mechanisms.

Most platelet transcriptome studies to date focused on interrogating messenger RNAs (mRNAs) [16-19]. Studies of small noncoding RNAs (sncRNAs) remain comparatively limited, and most focus on miRNAs [20]. The small number of studies, differences in the profiling platforms, preparation protocols, storage [21-23], and the known dependencies of miRNA production on sex, ancestry, and age [24,25] have led to differences in the reported miRNAomes [20]. It is important to understand the repertoire and function of platelet sncRNAs for several reasons [26-29]: platelets inherit their sncRNAome from MKs, and so, they serve as a window to MK gene expression; platelets translate their mRNAs to protein while delivering miRNAs to other vascular cells to regulate gene expression; and platelet sncRNAs can serve as biomarkers for platelet and nonplatelet diseases [27,30].

In 2013, we published one of the earliest studies of platelet sncRNAs [19] and reported the unexpected finding that, in addition to miRNAs, statistically significant platelet sncRNAs arose from regions encoding small cytoplasmic RNAs (scRNAs, also known as “Y RNAs”), transfer RNAs (tRNAs), and ribosomal RNAs (rRNAs). The first study of platelet miRNA isoforms (isomiRs) had just appeared [31], and there was limited knowledge of isomiRs, scRNA/Y-RNA–derived fragments, tRNA–derived fragments (tRFs), or rRNA–derived fragments (rRFs), so we did not pursue those early findings.

IsomiRs are coexpressed molecules that arise from the same miRNA arm and differ from each other by one or more nucleotides at either end [32]. IsomiRs can be detected and quantified easily by high-throughput (deep sequencing) methods [24,33]. Through loading on AGO proteins, isomiRs regulate mRNA and protein levels [32]. Sibling isomiRs from the same miRNA arm have been shown to affect different mRNAs and pathways [34-36]. tRNAs and rRNAs also produce fragments that are regulatory in nature—tRFs and rRFs, respectively [37-43]. Intriguingly, the levels of isomiRs, tRFs, and rRFs depend on “personal attributes” (eg, sex and ancestry) and “context” (eg, tissue type and disease) [32]. We now know that the scRNAs/Y RNAs (= RNY1, RNY3, RNY4, and RNY5) also produce fragments, the yRFs, which have found uses as cancer biomarkers [44-46]. IsomiRs, tRFs, rRFs, and yRFs are omnipresent: they are found in solid tissues [40,46-48], plasma [49,50], serum [49-51], exosomes [49], and platelets [52].

There have been some attempts to compare the transcriptomes of platelets and MKs. One of the earliest studies demonstrated an association between miRNAs and platelet reactivity [53]. A more recent study showed a strong correlation between (microarray-derived) miRNA profiles of day 13 cultures of CBMKs and primary bone marrow MKs isolated by laser capture microdissection [54]. Similarly, very strong correlations between mRNAs in neonatal cord blood–isolated and adult platelets [55] have also been reported, which might be extrapolated to the levels of parental cells (MKs).

The richness of the sncRNAome and the hindsight of a decade’s worth of sncRNA research prompted us to revisit and expand our original study of platelet sncRNAs [19]. This time, we focused on the exhaustive characterization of sncRNAs from and comparisons across 3 sources: CBMKs, platelets from HDs, and platelets from COVID-19 patients. In the case of the CBMKs, we additionally examined AGO-bound sncRNAs. For simplicity of notation, in what follows, we will use the abbreviation “MKs” to refer to CBMKs cultured from cord blood–derived CD34⁺ cells.

2 ∣. METHODS

2.1 ∣. Culture of CD34⁺-derived primary MKs and Argonaute immunoprecipitation

Cord blood was obtained from 3 independent umbilical vein cords from the New York Blood Bank and Cleveland Cord Blood Center and used with approval by the Institutional Review Board (IRB) of the University of Utah (IRB# 00108527 and 00149913). The study was conducted in accordance with the Declaration of Helsinki. CD34⁺ hematopoietic progenitors were isolated via positive selection with CD34 magnetic microbeads (Miltenyi, #130-046-702) and differentiated into MKs as previously described [13]. AGO immunoprecipitation (IP) was conducted with a Pan-AGO antibody (#MABE56, Sigma-Aldrich). For more information and the detailed protocol, see the Supplementary Material.

2.2 ∣. Collection of donor samples and platelet isolation

We studied a cohort of COVID-19 patients and healthy donors (HDs) whose platelet mRNAs we profiled previously [9]. Hospitalized COVID-19 patients (n = 10) were recruited from the University of Utah Health Sciences Center in Salt Lake City from March 17, 2020, to April 15, 2020. Four of the 10 patients were admitted to the intensive care unit (ICU). All patients were recruited under study protocols approved by the IRB of the University of Utah (IRB# 00102638, 00093575). Each study participant, or their legal authorized representative, gave written informed consent for study enrollment in accordance with the Declaration of Helsinki. Healthy, age- and sex-matched donors (n = 5) were enrolled under a separate IRB protocol (IRB# 0051506). Clinical and demographic information can be found in Supplementary Table S1. For more information on enrollment criteria and platelet isolation protocols, see the Supplementary Material.

2.3 ∣. Generation and analysis of sncRNA-seq datasets

For details about the experimental protocols for RNA isolation, sncRNA library preparation, and deep sequencing, see Supplementary Material. Raw sequencing data are available in the Sequence Read Archive (PRJNA946749). We used publicly available tools for the identification of isomiRs [56], tRFs [57], yRFs, rRFs, and other RNAs [58]. Reads were also mapped to the SARS-CoV-2 genome. We labeled all sncRNAs using the “license plate” approach [40,48,56]. We also used publicly available tools to find molecules that are differentially abundant between MKs and platelets from HDs, and between platelets from HDs and platelets from COVID-19 patients. For detailed information about these steps, including parameter settings and thresholds, see Supplementary Material.

2.4 ∣. Integrative isomiR-mRNA sequencing data analysis

We downloaded the previously published mRNA-seq datasets for the same platelet samples [9] from the Sequence Read Archive (PRJNA634489). Supplementary Table S1 shows the correspondence of labels between the sncRNA-seq data generated in this study and mRNA-seq data we generated previously [9]. For more information about isomiR target prediction and isomiR-mRNA data integration, see Supplementary Material.

3 ∣. RESULTS

We generated deep sequencing sncRNA data from cultured MKs and platelets, and studied their profiles. In MKs, we profiled the sncRNAs in total RNA and AGO IP RNA. In platelets, we profiled the sncRNAs in total RNA isolated from highly purified platelets of 4 HDs and 10 COVID-19 patients. Four of the patients were admitted to the ICU (see Supplementary Table S1 for clinical and demographic information).

3.1 ∣. Platelets and MKs contain sncRNAs from multiple RNA types

Using the pipeline described in Methods and the Supplementary Material, we profiled the sequenced reads from the platelet and MK datasets. We enforced exact matching when profiling “isomiRs,” “yRFs,” “tRFs,” “rRFs,” and “other repeats.” For reads not assigned to any of these 5 types, we sought matches across the genome by allowing ≤1 replacement but no indels (we denote it as “≤1 MM” type, where MM stands for mismatch). We accounted for ~95% of all starting reads, on average, across all datasets (Supplementary Table S2). For the COVID-19 patient samples, we also investigated the possibility that those platelet reads that did not map on the human genome originated in the SARS-CoV-2 genome but found no evidence of that. Figure 1 shows how each of the 6 types of sncRNAs contributes to the composition of all MK and platelet samples.

FIGURE 1.

Small noncoding RNA-seq read mapping summary. (A) Total RNA from 3 cord blood–derived cultured MK samples. (B) RNA from AGO IP from the 3 MK samples. (C) Platelets from 4 HDs. (D) Platelets from 10 COVID-19 patients. The heights of the bars and numbers on top of them stand for mean values; the error bars stand for SDs. The “≤1 MM” label refers to reads that either match isomiRs, yRFs, tRFs, rRFs, or repetitive sequences by allowing exactly 1 replacement or other parts of the genome with up to 1 replacement (no indels are allowed). Donut charts show compositions of the “≤1 MM” read sets. No expression level thresholding was used in this stage of the analysis. HD, healthy donor; isomiR, microRNA isoform; MK, megakaryocyte; MM, mismatch; rRF, rRNA-derived fragment; tRF, tRNA-derived fragment; yRF, Y RNA–derived fragment.

MKs contain a diverse and balanced sncRNAome (Figure 1A). In total RNA, isomiRs, yRFs, tRFs, and rRFs have relatively comparable abundances (14.1%-26.1%). In the AGO IP data, most of the reads (87.9%) are isomiRs, as expected (Figure 1B). In the platelets of HDs, yRFs are the most abundant species (56.7%), outpacing even isomiRs/miRNAs (34.4%)—figures correspond to the sum of exact matches and matches with 1 nucleotide replacement (Figure 1C). The situation reverses itself in the platelets of COVID-19 patients where miRNAs account for most (63.6%) sncRNAs and yRFs account for a mere 19.8% (Figure 1D), a statistically significant change (Mann–Whitney U-test P = 2.0 × 10⁻³).

Note that >10% of the total RNA in platelets and MKs belongs to the “≤1 MM” type (replacements, no indels). In platelets, this category comprises primarily reads mapping to Y RNAs and miRNAs other than the reference ones (donuts of Figures 1C, D). In MKs, the “≤1 MM” category comprises reads mapping to long noncoding RNAs and protein-coding sequences (Supplementary Table S2). Unless stated otherwise, in what follows, we will not consider molecules from the “≤1 MM” type.

3.2 ∣. The correlation between the sncRNAomes of MKs and platelets depends on RNA type

We used rank correlation to compare all datasets in an all-against-all fashion. We only analyzed molecules with an abundance of ≥5 RPM. As we see in Supplementary Figure S1A, there is high correlation among the MK samples (ρ = 0.76 ± 0.06 for total RNA and ρ = 0.91 ± 0.01 for AGO IP RNA) and the platelet samples (ρ = 0.81 ± 0.06).

MKs and platelets show a moderate correlation (ρ = 0.38 ± 0.04) when using all 4 types of sncRNAs (isomiRs, tRFs, rRFs, and yRFs). When we restrict ourselves to only miRNAs and combine same-arm isomiR abundances into a single number (to emulate expression arrays), the average correlation increases to 0.63 (Supplementary Figure S1B). This value is concordant to our report of the correlation (ρ = 0.64, P = 9.23×10⁻¹⁵) between miRNAs in platelets and day 13 cultured MKs, quantified using an Exiqon array [54]. Treating isomiRs as distinct molecules reduces the average correlation to 0.46 (Supplementary Figure S1C). Below, we will explain what causes this difference in correlation.

3.3 ∣. The various sncRNA types are characteristically diverse

Figure 1 captures the relative contribution of the various molecular types to the sncRNAome of each biological source. On the other hand, Figure 2 provides information about the cardinality and relative contribution of each sncRNA type to total RNA. Molecules belonging to the same molecular type are grouped together and listed left to right in order of decreasing abundance. Each molecular type’s span on the X-axis shows how many of its members exceed threshold. Each molecular type’s span on the Y-axis measures the relative contribution of the molecular type to the sncRNAome. Finally, the slope at any point of these curves captures the abundance of the respective molecule: points with steep gradient correspond to highly abundant molecules. Supplementary Table S3 lists the sequence and abundance of all sncRNAs used in Figure 2. The AGO IP data contain approximately twice the number of isomiRs detected in total RNA (despite the same threshold value being used): this is because the IP enriches for isomiRs while the sequencing depth remains unchanged, making now “visible” isomiRs with an abundance of <5 RPM in total RNA.

FIGURE 2.

Abundance distributions within and across the various sncRNA types. (A) Total RNA from 3 MK samples. (B) RNA from AGO IP from 3 MK samples. (C) Platelets from 4 HDs. (D) Platelets from 10 COVID-19 patients. The segment labels represent width (number of distinct molecules) and height (percentage of abundance). The distinct molecules are sorted within each type according to their abundance (higher abundance molecules are listed before lower abundance ones). The “≤1 MM” label has the same semantics as in Figure 1. For each of the 4 panels, only molecules with an abundance of ≥5 RPM in all corresponding samples were considered. HD, healthy donor; isomiR, microRNA isoform; MM, mismatch; rRF, rRNA-derived fragment; tRF, tRNA-derived fragment; yRF, Y RNA–derived fragment.

Note how platelets and MKs comprise a few but highly abundant yRFs, unlike the other molecular types. In the platelets of both groups, yRFs from RNY4 contribute >95% to the total population of yRFs. yRFs from RNY4 are the most abundant contributors to the yRFs in MKs (70.1%), followed by yRFs from RNY1 (13.5%), RNY5 (6.8%), and RNY3 (4.8%).

3.4 ∣. Noncanonical and nontemplated isomiRs abound in MKs and platelets

Previously, we highlighted the diversity of the isomiRs found in different tissues [34,56]. We extended those analyses to MKs and platelets further distinguishing among isomiRs produced from the 5p (Figure 3, top) and 3p (Figure 3, bottom) miRNA arm, respectively. Only isomiRs with an abundance of ≥5 RPM were considered in this analysis.

FIGURE 3.

The abundance of templated and nontemplated isomiRs in platelets and MKs. The heights of the bars stand for mean values; the error bars stand for SDs. Labels on top of the bars contain the numbers of distinct isomiRs, percentage of distinct isomiRs, and mean RPM value. Canonical isomiRs refer to isomiRs whose 5′ termini and 3′ termini sequences match the corresponding miRBase entries. Nontemplated isomiRs refer to isomiRs with 3′-end additions of nucleotide stretches that do not match the corresponding genomic sequence. Donut charts on top of the green bars represent the percentages of distinct isomiRs with A, C, G, or U 3′-end additions. For both MKs and platelets, only isomiRs with an abundance of ≥5 RPM in all corresponding samples were considered. The hairpins at the top and bottom of the figure were created at BioRender.com. IsomiR, microRNA isoform; MK, megakaryocyte.

We found that >50% of all isomiRs are noncanonical, ie, each differs from its reference sequence found in miRBase at either the 5′-end, 3′-end, or both. More than 80% of the noncanonical isomiRs are 3′-end isomiRs. These findings mirror our previous findings in lymphoblastoid cell lines [24] and cancer tissues [47,56].

Noncanonical 5′-end isomiRs are a minority. However, several of them are very abundant in MKs and platelets (Supplementary Table S4). These isomiRs are important since 5′-end sequence variations affect the miRNA “seed” and thus the isomiRs’ mRNA targets [34,35]. For example, a miRNA that has been linked to platelet reactivity, miR-126-3p [20], has 2 highly expressed (≥500 RPM) 5′-end isomiRs, suggesting the existence of many currently unexplored mRNA targets for this miRNA. The same 2 isomiRs are also highly abundant in the MK samples.

We also found many “nontemplated” 3′-end isomiRs—isomiRs containing 1 or more nucleotides that are not in the genome. Among the nontemplated 3′-end isomiRs from the 5p arm of miRNAs, adenine and uridine account for most additions. However, in isomiRs from the 3p arm, uridine is the most frequent addition. These observations hold for both the platelet and the MK samples.

We found many abundant miRNAs in MKs and platelets whose canonical (miRBase) isoform represents <1% of the parental miRNA arm’s abundance (Supplementary Table S5). For example, miR-224-5p has an abundance of 1542 RPM in MKs and 502 RPM in the platelets of HDs: in both cell types, its canonical isoform corresponds to <0.1% of miR-224-5p reads.

3.5. ∣. SncRNAs from day 13 MKs are selectively enriched/depleted in platelets

Supplementary Figure S1A shows a modest positive correlation between the MK and platelet samples. These values together with the differences in platelet and MK molecular composition (Figures 1A, C) suggest that only some of the sncRNAs found in MKs end up in platelets. To investigate this, we calculated for each molecule the following ratio: (mean RPM abundance in platelets)/(mean RPM abundance in MKs). Values of the ratio >1 indicate preferential sorting into platelets, whereas values of <1 indicate the opposite.

Figure 4 shows these ratios for isomiRs, yRFs, tRFs, and rRFs. Molecules whose ratios are not statistically significant are indicated in gray. Molecules that are enriched in platelets are to the right of the X = 0 line (red region), whereas molecules that are enriched in MKs are to the left of it (blue region). For all 4 RNA types, we find numerous molecules that are expressed in the MKs but are absent from the platelets (this is indicated by the Y = −X line in the left part of each plot).

FIGURE 4.

Comparison of the platelet and MK sncRNAomes. (A–D) The results of enrichment analysis between the platelets of HDs and MKs. Enrichment of a molecule is defined as the ratio of the molecule’s platelet abundance (in RPM) over its MK abundance (in RPM). The points were colored according to the enrichment significance criteria: (1) adjusted P ≤ .05, (2) enrichment ≥2 or ≤1/2, and (3) abundance of ≥5 RPM in HD platelets or MK samples. (E) Examples of platelet to MK enrichment of isomiRs (nonempty cells) originating from the same miRNA arms (Y-axis labels). The labels in the cells correspond to different isomiRs: the first number denotes an offset from the canonical 5′ terminus and the second number denotes an offset from the canonical 3′ terminus; parentheses indicate nontemplated 3′ isomiRs and contain information about the length and type of the nontemplated nucleotide additions. The mean RPM values of miRNA arms (sum of corresponding isomiR RPMs) in MKs are also shown. Only isomiRs with an abundance of ≥5 RPM in all MK samples are considered. We limited the number of displayed isomiRs for some miRNA arms (eg, miR-21-5p) by additionally requiring that an isomiR’s RPM be at least 1% of the parental miRNA arm’s abundance (in RPM). HD, healthy donor; isomiR, microRNA isoform; MK, megakaryocyte; rRF, rRNA-derived fragment; tRF, tRNA-derived fragment; yRF, Y RNA–derived fragment.

Most of the MK tRFs and rRFs are depleted in platelets. Only 7.6% of all tRFs and 5.3% of rRFs with mean abundance of ≥5 RPM in MKs are enriched in platelets. The platelet-enriched tRFs include many 3′-tRFs from the MT-encoded tRNA^ValTAC and tRNA^HisGTG (Fisher’s exact test P = 2.77 × 10⁻⁹). The platelet-enriched rRFs arise primarily from hotspots along the nuclear genome–encoded 18S and 28S rRNAs.

For completeness, we performed an enrichment/depletion analysis to identify molecules that are enriched in the MK AGO IP compared to MK total RNA. As expected, we found mostly isomiRs to be enriched in the AGO IP and a few tRFs and rRFs. Most of the abundant yRFs, tRFs, and rRFs in the total RNA of MKs are depleted in the AGO IP (Supplementary Figure S2A-D).

3.6 ∣. IsomiRs from the same miRNA arm, miRNA cluster, or miRNA family are enriched differently in platelets

Supplementary Table S6 lists all isomiRs that are statistically significantly enriched or depleted in platelets compared to MKs. We examine a few of these isomiRs in detail.

First, we looked at isomiRs from 3 well-known miRNA families: let-7 [59], miR-30 [60], and miR-103/107 [61]. The canonical miRNAs belonging to these families have very similar sequences. Figure 4E shows multiple isomiRs of the 8 members of the let-7 family: all are very abundant in MKs and consistently enriched in platelets. Note how each arm gives rise to both noncanonical (eg, 0∣−4, 0∣−1, and 0∣+1 [24]) and nontemplated (eg, 0∣−1(+1A) and 0∣−1(+1C) [56]) isomiRs. In this notation, which we introduced and used previously [24,56], numbers before and after a vertical bar represent offsets from the canonical isomiR’s 5′- and 3′-ends, respectively. On the other hand, miR-30 and miR-103/107 are examples of differential enrichment. MiR-30c-5p and miR-107 isomiRs are depleted in platelets, but isomiRs of miR-30d-5p and miR-103a-3p are enriched (Figure 4E). Notably, miR-103a-3p (AGCAGCAUUGUACAGGGCUAUGA) and miR-107 (AGCAGCAUUGUACAGGGCUAUCA) differ at the under-lined nucleotide but are enriched differently in platelets compared to MKs.

Next, we examined isomiRs from the miR-17/92 cluster (Figure 4E). IsomiRs from 5 of the cluster’s 6 miRNAs are characteristically depleted in platelets. For the sixth miRNA, miR-92a-3p, all but one of its isomiRs are enriched in platelets. This is unexpected given that the cluster comprises 6 miRNA precursors that are transcribed together [62].

We also encountered cases of sibling isomiRs, ie, isomiRs from the same miRNA arm, with opposite enrichments (Figure 4E). For example, the canonical isoform of miR-21 (miR-21-5p∣0∣0) is depleted in platelets (fold change (FC) = 1.95 log₂ units), whereas the nontemplated miR-21-5p∣0∣0(+1U) is enriched (log₂(FC) = 1.25). Other isomiR pairs with opposite behavior include miR-93-5p∣0∣ 0 (log₂(FC) = −2.22) and miR-93-5p∣0∣0(+1A) (log₂(FC) = 0.98), and miR-127-3p∣0∣0 (log₂(FC) = −1.85) and miR-127-3p∣0∣0(+1U) (log₂(FC) = 1.75).

Lastly, we highlight an instance of differential arm enrichment (Figure 4E). While the 5p and 3p arms of miR-106b-5p have comparable abundance in MKs, they behave differently: isomiRs from the 5p arm are depleted in platelets (5p arm log₂(FC) = −5.44), whereas isomiRs from the 3p arm are enriched (3p arm log₂(FC) = 2.76).

To help appreciate these striking differences, we note that they are not due to AGO loading differences (Supplementary Figure S2E). Also, while the cumulative abundance of isomiRs from the same miRNA arm is highly correlated between MKs and platelets (ρ = 0.64 ± 0.04), the correlation decreases considerably (ρ = 0.46 ± 0.05) when computed using isomiRs (Supplementary Figure S1B vs Supplementary Figure S1C) because of the evident selective sorting of isomiRs into platelets. Supplementary Table S7 shows the complete list of significantly enriched/depleted sncRNAs in AGO IP.

3.7 ∣. Platelet sncRNAomes differ greatly between HDs and COVID-19 patients

Next, we compared the platelets of HDs and COVID-19 patients. Principal component analysis showed that >85% of data variance is concentrated in 5 components (Supplementary Figure S3). Figure 5A shows the first and second principal components. Clustering of the platelet samples (Figure 5B) grouped samples in concordance with the principal component analysis findings. In both approaches, 2 samples from the non-ICU group stood apart from the group’s other samples.

FIGURE 5.

Abundance of sncRNAs in the platelets of COVID-19 patients and HDs. (A) Principal component analysis. (B) Hierarchical clustering of platelet samples (rows) and molecules (columns). (C–F) The detailed view on hierarchical clustering from panel (B) for different molecule types (row ordering preserved). Black dashed lines in panels (B–F) separate major clusters. (G–J) The results of differential abundance analysis between platelets of COVID-19 patients (ICU and non-ICU) and HDs. The points are colored according to significance: (1) adjusted P ≤ .05, (2) fold change of ≥2 or ≤1/2, and (3) abundance of ≥5 RPM in all COVID-19 or all healthy samples. HD, healthy donor; isomiR, microRNA isoform; rRF, rRNA-derived fragment; tRF, tRNA-derived fragment; yRF, Y RNA–derived fragment.

To evaluate each RNA type’s contributions, we created separate heatmaps while preserving the order of rows to facilitate comparisons (Figure 5C-F). We found pronounced changes in the yRF profiles of COVID-19 patients: almost all highly abundant yRFs in the platelets of HDs are depleted in the platelets of the COVID-19 patients (Figure 5D). Several large groups of tRFs and rRFs also varied markedly between patients and HDs (Figures 5E, F). The most substantial difference between ICU and non-ICU samples is due to the rightmost rRF cluster seen in Figure 5F. Surprisingly, isomiRs did not exhibit noteworthy differences among the 3 platelet groups (Figure 5C).

We also used differential abundance analysis to identify molecules that change significantly between HDs and patients (Figure 5G-J and Supplementary Table S8). Similar to the unsupervised analysis, we find that yRFs have lower levels in the patients (Figure 5H), whereas tRFs and rRFs have higher levels (Figure 5I-J).

3.8 ∣. The abundances of isomiRs and their putative target genes are perturbed in COVID-19

Among the isomiRs with an abundance of ≥ 5 RPM, nearly equal numbers increase or decrease in the COVID-19 patients. IsomiRs from the same miRNA arm generally change in the same direction. More-over, multiple nontemplated isomiRs are differentially abundant and exhibit notable biases. For example, of the isomiRs whose abundance decreases in the platelets of patients, 50% are 3′-uridylated, whereas only 4.5% are 3′-adenylated (down from an average of ~29%).

Supplementary Table S9 lists 15 isomiRs with abundance ≥100 RPM that are differentially abundant between HDs and patients. To assess the potential significance of those differences, we predicted targets [63] for these isomiRs among all mRNAs that we found previously to be expressed in the same samples [9]. For each isomiR, we subselected among its putative targets those whose abundance changes in the opposite direction from the isomiR.

Several platelet-relevant pathways are enriched among the genes whose mRNAs are anticorrelated with the isomiRs, including adhesion and proteasome (Supplementary Table S10). For 4 differentially abundant isomiRs (miR-7-5p∣0∣0, miR-7-5p∣0∣-1(+1C), miR-23b-3p∣0∣0(+1U), and hsa-miR-484∣0∣0(+1U)), their predicted targets are significantly over-represented among genes that changed in the opposite direction from the isomiRs (Supplementary Table S11). This finding links isomiR abundance changes to opposite changes in mRNA abundance in the platelets of patients and suggests direct molecular interactions.

3.9 ∣. Sibling yRFs, tRFs, and rRFs (from the same parental RNA) can have opposite enrichments

Supplementary Tables S6 and S8 include multiple sncRNAs from the same parental RNA with distinctly different enrichments. We discuss here 3 examples.

The first example pertains to yRFs. For clarity, we only considered yRFs with an abundance of ≥100 RPM. As Figure 6 shows, most MK yRFs are depleted in platelets except for yRFs from the 5′ region of RNY4. The figure shows each yRF’s abundance in MKs. The color coding captures the log₂(FC) of the ratio (platelet abundance)/(MK abundance). In complete analogy to isomiRs (Figure 4E), yRFs from the same parental Y RNA can exhibit opposite enrichments, even when their sequences overlap highly. Note the 3 yRFs from the 5′ region of RNY4 that are highlighted blue: while they are very abundant in MKs they are depleted in platelets.

FIGURE 6.

Enrichment and depletion of yRFs in HD platelets compared to MKs. The top 4 rows show a multiple sequence alignment for the 4 human Y RNA sequences. The left and right columns represent the mean MK RPM values of yRFs arising from the 5′ and 3′ regions of the Y RNAs, respectively. Only yRFs with an abundance of ≥100 RPM in all MK samples are shown. The changes for all shown yRFs are statistically significant. Boldface letters in the multiple sequence alignment indicate nucleotides common to at least 3 of the Y RNAs. HD, healthy donor; MK, megakaryocyte; yRF, Y RNA–derived fragment.

The second example pertains to tRFs. Among the tRFs that are enriched in the platelets of patients, the nuclear genome–encoded tRNA^GluCTC is most prevalent. While most tRFs are strongly depleted in the platelets of HDs compared to MKs, their abundance increases sharply in the platelets of patients compared to HDs. The tRNA^GluCTC tRFs account for 23.6% of all tRFs found in the patients. Figure 7 captures this change. On the left, the color coding captures the log₂(FC) of the ratio (platelet abundance)/(MK abundance), whereas on the right, it captures the log₂(FC) of the ratio (abundance in patients)/(abundance in HDs). tRFs whose nucleotide sequence is rendered in black change statistically significantly between HDs and patients; light gray font marks tRFs whose changes are nonsignificant.

FIGURE 7.

Differential enrichment (HD platelets vs MKs) and differential abundance (COVID-19 patient vs HD platelets) of tRNA^GluCTC fragments. Only fragments with an abundance of ≥5 RPM in all platelet samples of COVID-19 patients or HDs were considered. The yellow region on the tRNA^GluCTC sequence highlights the portion of the tRNA from which the shown tRFs arise. The nucleotides of differentially depleted (blue) and differentially enriched (red) tRFs are colored with black font if they satisfy the following criteria: (1) adjusted P ≤ .05, (2) enrichment (left part) or fold change (right part) of ≥2 or ≤1/2; (3) abundance of ≥5 RPM in all MK or HD platelet samples (left part), or in all platelet samples of COVID-19 patients or HDs (right part). The nucleotides of tRFs not satisfying one or more of the above three conditions are colored with light gray font. Also shown are the tRFs’ mean RPM values in MKs and HD platelets. HD, healthy donor; MK, megakaryocyte; tRF, tRNA-derived fragment.

The third example pertains to rRFs from the 5S rRNA. In MKs, we find multiple abundant rRFs from this rRNA. All but 4 of these rRFs are depleted in platelets compared to MKs. The situation reverses itself in COVID-19 patients: rRFs that were depleted in the platelets of HDs are now enriched, and vice versa. Figure 8 summarizes this finding in a similar manner to Figure 7.

FIGURE 8.

Differential enrichment (HD plateletes vs MKs) and differential abundance (COVID-19 patient vs HD platelets) of rRFs from the 5S rRNA. Only fragments with an abundance of ≥5 RPM in all platelet samples of COVID-19 patients or HDs were considered. The yellow regions of the 5S rRNA sequence highlight the portions of the rRNA from which the shown rRFs arise. The 2 red nucleotides (“TT”) indicate that the rRFs include nucleotides that are beyond the nominal boundaries of the 5S rRNA. The nucleotides of differentially depleted (blue) and differentially enriched (red) rRFs are colored with black font if they satisfy the following criteria: (1) adjusted P ≤ .05, (2) enrichment (left part) or fold change (right part) of ≥2 or ≤1/2, and (3) abundance of ≥5 RPM in all MK or HD platelet samples (left part), or in all platelet samples of COVID-19 or HDs (right part). The nucleotides of rRFs not satisfying one or more of the above 3 conditions are colored with light gray font. Also shown are the rRFs’ mean RPM values in MKs and platelets from HDs. HD, healthy donor; MK, megakaryocyte; rRF, rRNA-derived fragment.

4 ∣. DISCUSSION

We analyzed the sncRNAomes of cultured CBMKs, platelets from HD, and platelets from COVID-19 patients. We accounted for ~95% of the sncRNAs in these samples and can now provide the first comprehensive picture of the sncRNAome of MKs and platelets.

We uncovered a rich universe of sncRNAs whose relative abundance changes drastically with context (Figure 1). MKs contain primarily 4 types of sncRNAs with comparable abundances: isomiRs (26.1%), yRFs (14.1%), tRFs (16.7%), and rRFs (16.9%). Unlike MKs, the platelets of HDs comprise primarily yRFs (56.7%) from the 5′-region of RNY4, isomiRs (34.4%), and only a few tRFs and rRFs (<2.0%). SARS-CoV-2 infection profoundly affects the sncRNAome of platelets.

We also identified many 5′-end isomiRs not previously reported in either MKs or platelets (Supplementary Table S4). For example, we found several abundant 5′-end isomiRs of miR-126-3p, an important miRNA for the MK and platelet context [20]: miR-126-3p∣+1∣-1, miR-126-3p∣−1∣0, and miR-126-3p∣+1∣0. 5′-end variations compared to the canonical isomiR found in miRBase are significant because they translate into different “seeds” and, thus, different mRNA targets.

We also found many abundant isomiRs in MKs and platelets with nontemplated 3′-termini (Figure 3). The relative A/C/G/U contribution to the nontemplated portion of the isomiR differs between MKs and platelets, and for isomiRs from the 5p and 3p miRNA arms. On average, 21.3% of the nontemplated additions are cytosines, an unusual nucleotide choice: we are aware of nontemplated cytosine 3′ additions only in T cells and NK cells [64]. On a related note, recent work identified the proteins responsible for adenylation, uridylation, and guanylation of isomiRs, but not cytosylation [33]. Previous reports suggest that modification of the 3′-end of isomiRs can determine the identity of targeted genes [65,66]. It is known that 3′ isomiRs skew the measurements obtained through commercially available RT-qPCR assays due to crosstalk among the various siblings [67], posing an obstacle to quantifying these molecules with conventional approaches.

When we compared the sncRNAs of MKs with those of platelets, we observed prominent patterns of selective enrichment and depletion for different sncRNAs (Figure 4). The enrichment patterns are complex and depend on the sncRNA sequence. For example, isomiRs from miRNAs of the same family, or the same polycistronic cluster, can have opposite enrichment/depletion behavior in platelets compared to MKs (Figure 4). Even isomiRs from the same miRNA arm, or different arms of the same miRNA precursor, can exhibit acutely different enrichment patterns. Our findings on the selective enrichment of sncRNAs add to earlier findings of differential enrichments of MK-expressed mRNAs into platelets in humans [68] and mice [69].

Interestingly, isomiRs from miRNA loci that are linked to the maturation of cultured MKs do not exhibit concordant enrichments/depletions in platelets. See, for example, let-7c-5p, let-7e-5p, miR-125a-5p, miR-127-3p, miR-103a-3p, and miR-107 in Figure 4E. As we previously showed, all these miRNAs are upregulated in day 13 compared to day 6 MKs [54].

By comparing platelets from COVID-19 patients and HDs, we found that yRFs decreased drastically in the platelets of patients, whereas tRFs and rRFs increased (Figure 5). However, the behavior of isomiRs was more diffuse: nearly half of them decreased in abundance in the patients, whereas the rest increased. From a functional perspective, the differentially expressed isomiRs are linked to concomitant changes in the abundance of their predicted mRNA targets (Supplementary Tables S10 and S11). With the exception of some tRFs and rRFs that act like miRNAs [41,43,45], the mechanistic roles of the many sncRNAs found in the MKs and platelets are currently unknown. Indeed, we are unaware of any reports that studied isomiRs, tRFs, rRFs, or yRFs in platelets or MKs. In addition to operating through RNA interference, they could potentially play regulatory roles by presenting mRNA decoys to RNA-binding proteins. Given the dramatic changes of the platelet sncRNAome following a COVID-19 infection, we conjecture that isomiRs, tRFs, rRFs, and yRFs are essential to both homeostasis and systemic response to infections (eg, by the SARS-CoV-2 virus [9-12]).

The enrichment/depletion biases are particularly evident for specific yRFs, tRFs, and rRFs. Several yRFs from RNY4 are enriched in platelets compared to MKs, whereas yRFs from RNY1, RNY3, and RNY5 are depleted (Figure 6). Among tRFs, those from the nuclear tRNA^GluCTC are more abundant in the platelets of patients compared to HDs, but not enriched in HD platelets compared to MKs (Figure 7). Among rRFs, those from 5S rRNA show a very prominent inversion (enrichment to depletion, and vice versa) between HDs and patients (Figure 8).

Our findings suggest the existence of mechanisms in MKs that selectively sort sncRNAs into platelets. Conceivably, these mechanisms are sequence-based and analogous to those sorting cellular sncRNAs into exosomes [70,71]. We note that earlier studies showed platelets uptaking mRNAs from non-MKs, including vascular and cancer cells [72-75]. While this uptake has not yet been investigated for sncRNAs, it is reasonable to posit that it does exist. This possibility is supported by our findings: each panel of Figure 4 (especially evident in 4C and 4D) shows that a small collection of platelet-enriched molecules have very low abundances in MKs, suggesting the possibility of uptake from non-MK cells.

Lastly, we note that many sncRNAs change in abundance following a SARS-CoV-2 infection. We surmise that these changes are not unique to COVID-19 but occur in other diseases and conditions. This possibility is supported by the previous demonstration that changes in the platelet abundance of mRNAs and long noncoding RNAs can serve as powerful diagnostics in various settings including cancers [27,30]. Searching for alterations in the platelet sncRNAomes of patients with inherited platelet disorders is another area of future research.

Our study has several limitations. First, treatment of severe COVID-19 patients with heparin/heparinoids/other medications and the possible existence of comorbidities could alter platelet sncRNAomes and act as confounding variables. Second, we used “standard” RNA-seq that profiles only molecules with 5′-P/3′-OH termini, potentially missing sncRNAs with modified termini [32]. Because we did not use spike-ins during library preparation, we do not know whether the sncRNA composition changes we reported for the platelets of COVID-19 patients are accompanied by changes in the absolute amounts of platelet sncRNAs. Lastly, it is unclear whether the changes we observe in COVID-19 patients compared to HDs reflect modulation of the putative sorting mechanism in MKs, changes in the MK transcriptional program, or changes triggered in platelets after their release from MKs.

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1016/j.jtha.2023.07.028

Essentials.

• The megakaryocyte (MK) small RNAome primarily consists of isomiRs, tRFs, rRFs, and yRFs in comparable abundances.

• In healthy donors, yRFs (56.7%) and isomiRs (34.4%) account for most of the platelet small RNAs.

• In patients with COVID-19, isomiRs become the most abundant (63.6%) small RNA species in platelets.

• MK small RNAs are selectively depleted or enriched in platelets in a sequence-dependent manner.

DATA AVAILABILITY

The raw sncRNA sequencing FASTQ files are available in the Sequence Read Archive under accession number PRJNA946749.