Toward the human cellular microRNAome

In the paper

“While providing valuable insights into the relationship of microRNA expression and disease (Lu et al. 2005), these and subsequent studies (Cheng et al. 2015; Ludwig et al. 2016) have not unraveled cellular microRNA expression.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Cheng et al. 2015
• Supported

  The Introduction (page unavailable) says Ludwig et al. profiled miRNAs across 61 human tissues, and the Results report tissue-specific expression. The study therefore provides tissue-level patterns rather than resolving expression in individual cells. This verdict concerns Ludwig et al.; it does not assess the other cited studies. (LLM, conf=0.84)

  cites Ludwig et al. 2016 source

In the paper

“Additionally, the anonymity of microRNA nomenclature, with sequential numerical naming, has not allowed an intrinsic understanding of which microRNAs are ubiquitous and which have cell-restricted patterns of expression (Witwer and Halushka 2016).”

What the sources say

• Supported

  The paper says its numerical nomenclature does not intrinsically convey a microRNA’s localization. In “Areas of microRNA confusion and pitfalls to avoid” (page not supplied), it links the anonymity of microRNA names to difficulty identifying cellular expression, and its Figure 1 distinguishes ubiquitous from cell-type-restricted microRNAs. (LLM, conf=0.92)

  cites Witwer and Halushka 2016 source

In the paper

“We used surrogate variable analysis (SVA) and 26 loosely described “biologic clusters” of tissues and cells to clearly demonstrate that, after adjusting for surrogate variables, similar tissue and cell types clustered together across experiments (Supplemental Fig. S3; Leek and Storey 2007).”

What the sources say

• Supported

  In Results > SVA Applied to Human Expression Studies (page unavailable), Leek and Storey describe applying surrogate variable analysis and adjusting for a surrogate variable, supporting the citation’s identification of the method. The clustering across experiments is presented as the citing authors’ own result, illustrated by their Supplemental Fig. S3, rather than as a finding attributed to Leek and Storey. (LLM, conf=0.86)

  cites Leek and Storey 2007 source

In the paper

“In comparison to the 3707 novel TJU microRNAs (Londin et al. 2015), only 257 (52%) were shared between both data sets.”

What the sources say

• Supported

  Table 1 lists 3,707 novel miRNAs, and the Results describe the same total; the paper’s examples use TJU identifiers. The citation identifies the comparison dataset, not the citing authors’ finding that 257 (52%) were shared. (LLM, conf=0.92)

  cites Londin et al. 2015 source

In the paper

“Although technical factors may be responsible for some variation between cell types, the data clearly demonstrate a need to revise our understanding of the appropriate canonical microRNA sequences for better reproducibility and computational target prediction (Mestdagh et al. 2014; Agarwal et al. 2015).”

What the sources say

• ?Not checked — no full text of the sourcecites Mestdagh et al. 2014 source
• Partly supported

  In Materials and methods (‘MicroRNA sets for TargetScan7’; page not specified), the paper identifies abundant microRNA isoforms with different seed sequences and annotates them for target prediction. This supports considering alternative sequences in computational predictions, but the paper does not establish a need to revise canonical microRNA sequences for reproducibility. (LLM, conf=0.82)

  cites Agarwal et al. 2015 sourceExperimental — accuracy not yet measured

In the paper

“We determined that the main drivers of this difference were the changing methods of the Illumina sequencing library kits acting upon a small subset of microRNAs (Supplemental Fig. S10A; Baran-Gale et al. 2015).”

What the sources say

• Supported

  The Introduction reports that 50 highly expressed miRNAs were detected at levels more than tenfold different between Illumina v1.5 and TruSeq. The Discussion attributes the differential detection bias to the two library-preparation methods, supporting the cited explanation that a subset of miRNAs is particularly affected by the change in kits. (LLM, conf=0.88)

  cites Baran-Gale et al. 2015 source

In the paper

“These data further support the cell-type restricted microRNAs seen above and indicate that analyses of global tissue microRNA expression require an understanding of the source of each microRNA of interest, lest misinterpretations of the data result in spurious disease associations (Kent et al. 2014).”

What the sources say

• Supported

  The Introduction explains that tissue-level miRNA signals depend on cell-type composition and that understanding cellular expression is critical to interpreting them. The Conclusion warns that changes in cellular composition between diseased and non-diseased tissues can be mistaken for changes in miRNA expression within a cell type. Page numbers are not provided in our copy. (LLM, conf=0.93)

  cites Kent et al. 2014 source

In the paper

“yet, because blood is found in all tissues, they have been inappropriately assigned a variety of functions in epithelial and mesenchymal cells based on misinterpreted tissue-level data (Kent et al. 2014; Halushka 2016).”

What the sources say

• Partly supported

  The Introduction section on miR-451a (page not provided) says erythrocytes are present in every human tissue and that tissue-level miR-451a findings have been erroneously interpreted as expression in epithelial cells; it also discusses a liposarcoma example. It mentions miR-144 as part of the miR-144/451a cluster, but does not establish that miR-144 itself was inappropriately assigned those cellular functions. (LLM, conf=0.87)

  cites Kent et al. 2014 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Halushka 2016 source

In the paper

“Namely, miR-451a and miR-144 are exclusively expressed in red blood cells, yet, because blood is found in all tissues, they have been inappropriately assigned a variety of functions in epithelial and mesenchymal cells based on misinterpreted tissue-level data (Kent et al. 2014; Halushka 2016).”

What the sources say

• Partly supported

  The Introduction section “miR-451a—an erythrocyte specific microRNA” says the authors’ primary-cell comparisons found miR-451a exclusively in red blood cells. It identifies miR-144 as part of the same cluster, but does not say miR-144 is exclusively expressed in red blood cells. That part of the claim is unsupported by our copy. (LLM, conf=0.88)

  cites Kent et al. 2014 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Halushka 2016 source

In the paper

“This will improve the interpretation of tissue microRNA expression levels, which, to date, has significantly muddied our understanding of microRNA localization and biologically relevant function (Kent et al. 2014).”

What the sources say

• Supported

  The Introduction states that understanding cellular patterns of miRNA expression is critical to interpreting tissue-level data. The Conclusion says that failure to account for tissue composition complicates knowledge of miRNA function and that identifying the cellular source is needed to make biologically appropriate connections. (LLM, conf=0.93)

  cites Kent et al. 2014 source

In the paper

“Akin to the DEXUS results, among primary cell types, microRNA expression patterns generated four major groups: hematologic, mesenchymal, neural/ESC, and epithelial.”

In the paper

“Seven of these microRNAs (let-7a-5p, let-7c-5p, let-7b-5p, let-7f-5p, let-7g-5p, miR-26a-5p, and miR-30d-5p) were found in all tissues.”

In the paper

“Likewise, miR-126-3p, abundant only in endothelial cells and platelets, was present in 21 tissues, and miR-150-5p, abundant in lymphocytes, was present in 11 tissues.”

In the paper

“This method clustered cell types into hematologic, neural/embryonic stem cell (ESC), epithelial, and mesenchymal groups, identifying general patterns of microRNA expression.”

In the paper

“There were 320 microRNAs that had an RPM ≥1000 in any of the 46 normal cell types.”

In the paper

“Six microRNAs were present in all 46 cell types at this RPM threshold (miR-107, miR-103a-3p, miR-103b, miR-191-5p, miR-21-5p, and miR-92a-3p), and an additional nine microRNAs were present in all cells at a lower threshold of 100 RPM: miR-16-5p, miR-25-3p, miR-26a-5p, miR-26b-5p, miR-30d-5p, miR-101-3p, miR-128-3p, miR-140-3p, and miR-181a-5p (Supplemental Table S6).”

In the paper

“We then determined microRNA abundance from the 5p or 3p arm and found the guide/“driver” (more abundant, thermodynamically stable) microRNA to be equally from either arm of the hairpin microRNA, suggesting no strand bias in microRNA selection (Fig. 2D).”

In the paper

“Overall, 2319 of 2546 known microRNAs (miRBase v21) had a minimum expression of 1 read per million microRNA reads (RPM) in at least one sample (Supplemental Table S1).”

In the paper

“Of these, 94 (29%) were present in only a single class of cells (Fig. 2C).”

In the paper

“and an additional nine microRNAs were present in all cells at a lower threshold of 100 RPM: miR-16-5p, miR-25-3p, miR-26a-5p, miR-26b-5p, miR-30d-5p, miR-101-3p, miR-128-3p, miR-140-3p, and miR-181a-5p (Supplemental Table S6).”

In the paper

“Among tissues containing a mixture of cell types, 377 microRNAs were present at an RPM ≥1000.”

In the paper

“There were 204 (37%) microRNAs in which the miRBase v21 canonical sequence was never the most abundant sequence (Fig. 4A).”

In the paper

“while fibroblasts from three separate batches but obtained from different organ systems had a correlation between 0.75 and 0.9 (Supplemental Fig. S7).”

In the paper

“We also uncovered isomiR differences at the +1 position related to different library preparation kits.”

In the paper

“Additionally, 207 were the unassigned “passenger” 5p or 3p microRNAs from a known microRNA locus, and 15 were orthologous to a different species’ microRNA (primarily primate) (Supplemental Tables S8, S9).”

In the paper

“Importantly, cell-type restricted microRNAs showed active superenhancer activity matching those specific cells (Fig. 2E).”

In the paper

“Unsurprisingly, due to the specific high abundance of miR-126 in endothelial cells, JHU_ID_23828-3p was also significantly more abundant in four endothelial cell lines (avg. 3673 reads) than 30 other cell types (avg. eight reads; P = 0.001, Mann-Whitney U test) (Fig. 3F).”

In the paper

“We compared these to the 652 microRNA loci to ultimately produce 495 highest confidence putative novel microRNAs (Fig. 3A; Supplemental Table S10).”

In the paper

“For 205 common microRNAs, the most abundant sequence did not match the reported sequence in mirBase.org.”

In the paper

“Only 22% of microRNAs (versus 33% of primary cells) had the most abundant isomiR as the miRBase v21 canonical sequence in 100% of cells.”

In the paper

“Between cell types, these values were highly variable, with additional nontemplated adenines ranging from 28% (iPSC neurons) to 82% (H9 differentiated cells) of all additional nucleotides.”

In the paper

“As we showed, using the popular tool miRDeep2 to find novel microRNAs, we could obtain 21,338 loci, which is ∼10× more than are present in miRBase v21!”

In the paper

“Inflammation was also most prominent in the ulcerative colitis samples (Supplemental Fig. S5).”

In the paper

“A comparison of reads, all with the same 5′ starting location, revealed the most abundant isomiR to always be the canonical sequence for only 182 (33%) microRNAs.”

In the paper

“and miR-140-3p, in which the dominant species was 2 bp longer in 91 of 113 samples (Supplemental Table S11).”

In the paper

“About half, 264, were located within a gene locus.”

In the paper

“The median number of samples sharing a given microRNA loci was 33 (range 3–293) (Supplemental Fig. S8C).”

In the paper

“This experiment showed a reduction in epithelium as a percent of all cell types among the ulcerative colitis samples (avg. 20%) compared to Crohn's disease (avg. 44%) and diverticulosis (avg. 61%).”

In the paper

“Taking advantage of cell-specific microRNAs, we determined the feasibility of using cellular microRNA expression data to de-convolute an overall tissue signal to discern the individual components.”

In the paper

“The diversity of most abundant microRNA sequences between samples (as characterized by Shannon entropy) among cancer/immortalized cells was a ∼20% increase in disorder over primary cells (P < 0.005, Wilcoxon rank-sum test), suggesting increased DICER1 and DROSHA miscleavages (Fig. 4B; Supplemental Fig. S12; Supplemental Table S13).”

In the paper

“Across the primary cells, 84 microRNAs also had more abundant reads for template sequences that started proximal or distal to the canonical 5′ starting position (Supplemental Table S12), which is distinct from the 3′ changes reported in Figure 4A.”

In the paper

“A global comparison of microRNA differences identified miR-1304-3p to be the most extreme outlier among fibroblasts (Supplemental Fig. S6B).”

In the paper

“while endothelial cells were generally more consistent (4%–9%).”

In the paper

“On average, only 45% of all microRNA reads for each isomiR group were assigned to the most abundant isomiR sequence (Supplemental Fig. S11).”

In the paper

“This includes miR-10a-5p, in which a 1-bp-shorter sequence was the dominant species in 111 of 112 samples, and miR-140-3p, in which the dominant species was 2 bp longer in 91 of 113 samples (Supplemental Table S11).”

In the paper

“There were 411 unique seed sequences among the driver novel microRNAs.”

In the paper

“Surprisingly, the red blood cell component (blood) was a major and variable (18%–56%) part of each tissue, suggesting inconsistent and inadequate sample washing prior to RNA isolation (Fig. 2G).”

In the paper

“A small percent (394, 0.7%) were identified in more than 50 samples.”

In the paper

“As this may be related to library preparation, as described, we investigated just the 32 different primary cells that we generated in a shared batch and observed a twofold range, from 31% (dermal neonatal fibroblast) to 69% (bronchial epithelium), of all +1 additions being adenines, suggesting a real biological phenomenon was present (Supplemental Fig. S10B).”

In the paper

“Further, we found these differences were significant in only a small subset of microRNAs.”

In the paper

“MicroRNA expression in the presence of a superenhancer was significantly increased compared to microRNA expression at sites not adjacent to a superenhancer (Wilcoxon rank-sum test, P < 2.2−16).”

In the paper

“We found that nontemplated adenine additions were more common in data obtained from runs performed on older Illumina sequencers (Genome Analyzer I, II, and IIx) compared to the HiSeq (1000, 2000, 2500) systems (P < 2.4−9, paired t-test).”

In the paper

“We observed that only these miRGeneDB microRNAs have a separate distribution of high conservation.”

In the paper

“Total reads per microRNA did not correlate with the number of samples within which the microRNA was detected (R2 < 0.01) (Supplemental Fig. S9).”

In the paper

“Of these, 226 (55%) overlapped with a known microRNA seed sequence.”

In the paper

“Nontemplated adenines (62%) and uracils (30%) were again the dominant 3′ modifications in immortalized/cancer cell samples but could vary widely between both primary and cancer/immortalized cells as a reflection of either biological differences or technical factors (Fig. 4C,D).”

In the paper

“We provided two examples, using colon and adipose tissue, of how cell-specific patterns can de-convolute complex tissue expression patterns.”

In the paper

“The median number of reads per “driver” microRNA in these 495 putative novel microRNAs was 282 (range 51–257,582).”

In the paper

“miR-9-3p, 64-fold higher in the cancer-derived cells, has been previously reported as elevated in Hodgkin's lymphoma (Leucci et al. 2012) but not in these three cell types.”

What the sources say

• ?Not checked — no full text of the sourcecites Leucci et al. 2012 source

In the paper

“Lymphocytes were also variable (0%–5%) between samples (Fig. 2H), while endothelial cells were generally more consistent (4%–9%).”

In the paper

“Across the 126 cell types, 56% of nontemplate additions were adenines, followed by 41% uracils, 1% guanines, and 3% cytosines (Neilsen et al. 2012).”

What the sources say

• ?Not checked — no full text of the sourcecites Neilsen et al. 2012 source

In the paper

“A comparison of HeLa cells obtained from five sources had a range of expression correlation between 0.35 and 0.75, while fibroblasts from three separate batches but obtained from different organ systems had a correlation between 0.75 and 0.9 (Supplemental Fig. S7).”

In the paper

“By using cell, not tissue, data, we could observe cell-specific differences in the isomiR composition of microRNAs.”

In the paper

“We then validated nine novel microRNAs from six different cell types or tissue by PCR (Fig. 3C).”

In the paper

“As a result, there is a significant amount of data that can be collectively analyzed.”

In the paper

“suggesting no strand bias in microRNA selection (Fig. 2D).”

In the paper

“The use of RUV improved clustering of similar cell types from different experiments (Supplemental Fig. S2).”

In the paper

“Some well-known cell-specific microRNAs appear to be ubiquitous among tissues but merely reflect the presence of a certain cell type across tissues.”

In the paper

“The resulting cell type-specific patterns of discretized microRNA expression across the 161 primary cell types (Supplemental Fig. S1) are inherently robust to batch effects.”

In the paper

“We also likely lost some true positive microRNAs by retaining only those microRNAs with multiple sources of evidence.”

In the paper

“These ubiquitous and cell-specific patterns of microRNA expression, identified by RNA-seq data, are further supported by matching superenhancer data and highlight that there are many fewer ubiquitous microRNAs than currently believed.”

In the paper

“Although we report them as “highest confidence” putative novel microRNAs, they are still not definitive microRNAs.”

In the paper

“and highlight that there are many fewer ubiquitous microRNAs than currently believed.”

In the paper

“However, the risk of overinterpreting short RNAs as microRNAs is high.”

In the paper

“Here, we provide the first comprehensive delineation of cell-specific expression patterns of human microRNAs.”

In the paper

“This confirms that DICER1 is acting on some of these novel microRNAs.”

In the paper

“Despite this, technical factors certainly drive some of the variation and clustering in these samples.”

In the paper

“This indicates the potential for more shared regulatory control of genes often in a more cell-specific manner.”

In the paper

“Because much of the data is taken from public sources, in which the RNA-seq has been performed across different platforms and with different sequencing methods, significant technical variation in microRNA expression is present.”

In the paper

“This could explain some of the variability of microRNA expression across methods demonstrated by the miRQC project (particularly between RNA-seq and hybridization approaches) (Mestdagh et al. 2014) and impact the biologic activity of mimics and inhibitors relative to the true microRNA 5p end (Guo et al. 2014).”

What the sources say

• ?Not checked — no full text of the sourcecites Mestdagh et al. 2014 source
• ?Not checked — no full text of the sourcecites Guo et al. 2014 source

In the paper

“Human-specific microRNAs may represent a class of underreported microRNAs.”

In the paper

“Our approach to batch-effect adjustment and between-sample normalization resulted in clustering by cell or tissue type even when those samples came from multiple different experiments performed in different laboratories.”

In the paper

“suggesting increased DICER1 and DROSHA miscleavages (Fig. 4B; Supplemental Fig. S12; Supplemental Table S13).”

In the paper

“suggesting a real biological phenomenon was present (Supplemental Fig. S10B).”

In the paper

“suggesting methods that fail to acquire isomiRs significantly underestimate the presence of microRNAs in a variable fashion (Mestdagh et al. 2014).”

What the sources say

• ?Not checked — no full text of the sourcecites Mestdagh et al. 2014 source

In the paper

“Many are likely false positives and more proof of these RNA species acting as microRNAs must come from in vivo studies, which were beyond the scope of this project.”

In the paper

“suggesting inconsistent and inadequate sample washing prior to RNA isolation (Fig. 2G).”

In the paper

“To counter this, we performed extensive normalization.”

In the paper

“This difference, which has previously been reported on a smaller scale (Morin et al. 2008), can have an important effect on PCR- and hybridization-based strategies that may target a secondarily abundant microRNA in the isomiR family, altering the reported expression level of a microRNA.”

What the sources say

• ?Not checked — no full text of the sourcecites Morin et al. 2008 source

In the paper

“This association was observed with and without batch effect adjustment and was generally consistent across samples.”

In the paper

“In addition, some of these cell types have few to no replicates, and it will be important to continue adding data for these cells.”

In the paper

“These analyses suggest some key differences between immortalized/cancer cell lines and primary cells and highlight NIH concerns about the rigor and reproducibility of widely used cancer cell lines (http://grants.nih.gov/reproducibility/).”

In the paper

“In summary, these analyses suggest that the 495 highest confidence putative microRNAs reported here are similar to other known microRNAs.”

In the paper

“For many novel putative microRNAs, the sequence reads overlapped, with several bases of extension/difference between samples. Thus, the exact chromosome location of each novel microRNA is from a single sample of the collection and may not reflect the best absolute location on GRCh38/hg38.”

In the paper

“Finally, we are only moving toward a complete cellular microRNAome, as many cell types (hepatocytes, neutrophils, pneumocytes) were not available for this study.”