Knowing is not enough; we must apply: the case for rigorous microRNA annotation standards

Abstract

MicroRNAs are by far the most extensively studied and best-characterized class of small RNAs. This depth of knowledge may partly explain the persistent eagerness in the field to classify new sequences as microRNAs—since labeling a molecule as such often implies immediate functional insight and biological relevance. However, this enthusiasm has led to the publication and deposition of thousands of spurious microRNA entries in public repositories, increasing the risk of misinterpretation—particularly when these sequences are linked to human disease. To address this, we present a concise, four-part checklist for evaluating putative microRNAs, based on structural, genomic, functional, and expression-related criteria. Applying it to “Mir-690” and nine other examples, which are not listed in the curated microRNA database MirGeneDB, but highly cited, we find no support for their classification as microRNAs. They fail to meet core microRNA annotation criteria, lack structural features, conservation of targets, and reproducible or microRNA-typical expression evidence. Our findings underscore the need for heightened scrutiny and standardized validation in microRNA research. The checklist we provide offers a practical tool for authors, editors, reviewers, and readers to critically assess microRNA claims—particularly when such sequences are proposed as biomarkers or therapeutic targets in translational studies.

Graphical Abstract

Graphical Abstract.

Introduction

In 1993, the laboratories of Victor Ambros and Gary Ruvkun identified the first microRNA and one of its targets in nematodes [1, 2]. These seminal studies facilitated the identification of thousands of microRNAs in many eukaryotic organisms, the principles of their biogenesis and function and their importance for development and disease [3]. This led to an explosion of the noncoding RNA field [4] and was eventually awarded the Nobel Prize for Physiology or Medicine in 2024.

However, with annotation criteria being initially defined before the advent of next generation sequencing (NGS) [5], to know what is and what is not a microRNA has not been straightforward and led to confusion in the field. This resulted in the annotation of thousands of spurious microRNA candidates in public data repositories such as miRBase [6] and elsewhere (see [7, 8]) with up to two-thirds of the contained microRNA annotations being likely wrong (see Fromm et al. 2015) [9]. While those, in principle, are not directly harmful as eventually science will move on to correct such results, immediate care must be taken when human patients are involved, and clinical trials might be planned. To this end, uniform microRNA annotation criteria were established by us and others (see [9] and see below) and we also highlighted questionable annotation calls with possible consequences for human health [7, 10]. Nevertheless, these commentaries stand isolated in a field with >140 000 publications (microRNAs in title or abstract according to PubMed), of which many use “microRNAs” that do not fulfil annotation criteria. Importantly, these commentaries and other more general publications on the (failing) promise of microRNAs as biomarkers (e.g. [11, 12]) or therapeutics [13, 14] do not address the general issue of how to critically approach publications linking “microRNAs” to human diseases. The recent reports that the microRNA field is a field with one of the highest retraction rates (up to 4%), pointing toward concerted attempts at scientific fraud, are alarming and require immediate action [15]. To equip investigators, authors, reviewers, editors, and readers with the means to tell apart false from true positives, additional features, beyond annotation criteria, can further help to qualify or disqualify a microRNA candidate. However, a general checklist is currently missing.

We here present such a checklist for evaluating microRNA annotations and apply it, among nine other purported microRNAs, to “Mmu-mir-690”, a murine microRNA with uncertain annotation status that has been implicated in diabetes. We also examine its proposed, yet previously unrecognized, human orthologue, which was recently suggested as a potential drug target. We find that none of the 10 selected “microRNAs” adhere to widely recognized annotation criteria of bona fide microRNAs, have likely false-positive targets and insufficient expression levels or precision to be incorporated in a microRNA-like mechanism relevant to metabolism or disease. Concerningly, by investigating the impact of these purported microRNAs in the literature, we found that just those 10 previously rejected “microRNAs” are studied or mentioned by >8800 scientific publications that are cited by >260 000 papers. Given the fact that there are >1200 previously rejected microRNAs that are still studied, this is a truly gigantic problem of the field. We hope that providing a checklist for authors, reviewers, and editors can be a first meaningful step to manage this and help the field to focus resources on promising and real microRNAs.

The checklist

The four steps can be evaluated in any order.

Annotation

A crucial part of the process to evaluate a microRNA candidate presented in a manuscript is to check whether it is in MirGeneDB [16], or if any of the microRNA annotation criteria [5, 9, 17] are violated indicating origin outside the microRNA pathway with consequences for potential function.

Targets

When targets are presented in a manuscript those should be checked for their strength or nature of interaction and, if a more wide-reaching biological function is suggested, their conservation beyond human should be checked (we recommend TargetScan) [18, 19].

Genome

In case a microRNA candidate from e.g. mouse or rat is identified in human, it is important to check that the full microRNA sequence, i.e. the genomic locus, is present in the human genome. Ideally, syntenic information of neighboring protein coding genes should be supporting orthology-assignment as microRNAs a rarely change location during evolution.

Expression

Especially if the genomic locus has not convincingly been confirmed, but expression of a microRNA candidate is reported, the origin—and the biological meaningfulness [20]—of the expression signal must be clarified and should ideally be confirmed by orthogonal approaches.

The test case of “Mir-690”, a suggested drug target in human

A set of recent publications [21–24] has connected a microRNA candidate described in mouse (“Mmu-mir-690”) with obesity and fibrosis. Specifically, Rohm et al. [23] proclaims that treatment with small extracellular vesicles (EVs) exocytosed from adipose tissue macrophages from obese mice treated with the drug rosiglitazone, directly increases insulin sensitivity, suggesting a new mechanism for treating insulin resistance, a key feature of type 2 diabetes and obesity-related metabolic disorders. Moreover, the authors claim that mature microRNAs within those EVs, primarily “mir-690”, are solely responsible for these beneficial metabolic effects and, hence, that it could be a candidate for drug development for humans.

Annotation: "Mmu-mir-690" is not a microRNA

As the first laboratory that directly addressed the problem of spurious annotations within the field, the Bartel lab evaluated hundreds of sequences annotated as microRNAs in 2010 [17]. They found that >150 previously annotated as microRNAs “did not pass the filters for microRNA candidates”, including “Mmu-mir-690” [17]. Ectopic expression of bona fide microRNA sequences, such as the endothelial cell-specific Mir-126, are processed in mature and star sequences and robustly detected in cultured cells. In contrast, very few sequencing reads of “mir-690” were detected and the “mir-690” sequence data yielded no clear consensus 22-nucleotide mature “mir-690” sequence [17].

After publication of Chiang et al. [17], a broad systematic effort for the reannotation of vertebrate sequences, including the human microRNA complements led to the establishment of the manually-curated microRNA gene database MirGeneDB in 2015 [9, 16, 25, 26]. Annotation of microRNAs in MirGeneDB is based on updated annotation and naming criteria for microRNA families and genes [5], which include: (i) that the microRNA is derived from an imperfect base pairing microRNA precursor with at least 16 base pairing nucleotides in the stem and this precursor gives rise to (ii) transcripts from both arms, that (iii) show 5′ homogeneity, and (iv) 2-nt offsets (based on processing by Drosha and Dicer biogenesis proteins). “Mmu-mir-690” does not fulfill any of these criteria and hence does not appear in MirGeneDB and is a low confidence entry in miRBase (Fig. 1). Specifically, the structure of “Mmu-mir-690” rather resembles a cloverleaf/transfer RNA (tRNA) secondary structure and the sequence data does not show evidence of microRNA processing by Drosha or Dicer.

Figure 1.

“Mmu-mir-690” does not fulfill annotation criteria for microRNA sequences. Direct comparison of stem loop structures of the bona fide and endothelial cell-specific Mir-126 and “Mmu-mir-690” in addition to established annotation criteria for microRNA sequences.

Targets: “Mmu-mir-690” target site in Nadk does not exist in human

Despite the unclear nature of “Mmu-mir-690”, and hence limited proof of microRNA-like action by AGO-mediated RNA-translational inhibition or through decreasing mRNA stability, Olefsky and colleagues suggested the gene encoding NAD + kinase, Nadk, is a direct target of “Mmu-mir-690” [21]. The authors cite the TargetScan Mouse algorithm [27] and Rohm et al. [23, 24] continue to claim the direct targeting of Nadk by “Mmu-mir-690”. However, the results of the TargetScan Mouse algorithm (Fig. 2) clearly show the putative “mir-690” binding site in the 3′UTR of Nadk does not demonstrate cross-species conservation, a primary criteria for validating microRNA targets with a conserved biological function [18]. Despite the results showing treatment with EVs enhanced insulin signaling in primary human hepatocytes, since there is no conserved “mir-690” binding site in the human NADK 3′ UTR, it seems unlikely that this interaction can be the direct cause of the observation questioning thus the relevance of these findings to human physiology.

Figure 2.

Putative “Mmu-mir-690” binding site in the 3′UTR of Nadk is a 7mer-m8 binding site with no predicted relative knockdown (KD) score, low context score of −0.05, and a probability of conserved targeting of <0.1% as it is not conserved to either rat or human.

Genome: “Mir-690” does not exist in the human genome

From a genomic perspective, “Mmu-mir-690” is also unusual: in the mouse genome, “Mmu-mir-690” is located in an intron of the fibroblast growth factor 12 (FGF12) gene (GRCM39: Chromosome 16: 27 981 106–28 571 820). However, while the orthologue of FGF12 in the human genome exists (GRCh38.p14 Chromosome 3: 192 139 390–192 767, 76), BLAST matches to the purported microRNA sequence is confined to mice and the precursor microRNA sequence identified as “Mmu-mir-690” is not present in the human genome.

To rule out the unlikely possibility of genomic relocation of the microRNA in the human genome, potentially combined with sequence changes outside the mature, we next looked for the purported functional part of “Mmu-mir-690”, the 22 nucleotide long mature microRNA. This search returns 197 perfect, full-length hits in the mouse genome (Mm10), while no exact hits are found in the human genome, ruling out a genomic origin of potentially detected RNA sequences. As no bona fide microRNA in the literature displays such a high copy number as observed in the mouse genome, this again indicates “Mmu-mir-690” has been incorrectly identified as a microRNA. Interestingly, repetitive transposable elements such as SINEs (short interspersed repetitive DNA sequences) display high copy number throughout mammalian genomes and human and mouse SINEs have distinct evolutionary origins [28, 29]. Indeed, of the 197 hits in the mouse genome for “mir-690”, 74 hits overlap with repeatmasker [30] where 42 overlap with rodent specific B4 elements (SINE family). A separate search for Mmu-mir-690 in the curated telomere-to-telomere (T2T) human genome assembly also did not yield any matches.

Expression: “Mmu-mir-690” is not expressed in human tissue

Because the absence of a genomic locus in the human genome clearly contradicts claims that “mir-690” can be expressed in human cells [21], we next systematically investigated if expression of “mir-690” could be detected in human samples. For this, we used the miSRA database [31] and screened 57 400 human small RNA sequencing datasets totaling >525 billion small RNA sequencing reads of very heterogeneous nature for “mir-690”. Since it is well-established that sequencing data can be contaminated by molecules from other species [32], we first removed all datasets that displayed the unambiguous presence of mouse-specific microRNA sequences. We found “Mmu-mir-690” sequences in 20 (0.0003%) out of the remaining 56 878 datasets, comprising altogether 240 sequence reads out of the total of >500 billion sequences (Supplementary File 1). In contrast, we find the bona fide microRNA Mir-126 (Fig. 1) in 53 516 (93%) of the 57 400 samples, comprising >1.3 billion sequence detections. Given that computational methods to detect contamination in sequence data are imperfect, we conclude that the exceedingly few sequence detections of “Mmu-mir-690” most likely originate from mislabeling or contamination through e.g. sample handling or index hopping and that this molecule is not expressed in human tissues.

The study of “non-microRNAs” is widespread and has big impact

We next asked how widespread such cases of “microRNA” orthologues of spurious and previously rejected mouse microRNAs are. Two other studies on other spurious mouse entries in miRBase with a purported orthologues in the human genome, “mmu-mir-721” [33] (https://mirbase.org/hairpin/MI0004708) and “mmu-mir-6236” [34] (https://mirbase.org/hairpin/MI0021583) follow a similar discovery route as “mmu-mir-690”. Both are supposed human orthologues of rodent “microRNAs” that are not listed in MirGeneDB (but are found in miRBase), are not present in the human genome. While “mir-721”, a purported biomarker for myocarditis, is also not detected in NGS data at all [10], “mir-6236”, with proclaimed importance in insulin signaling, is of ribosomal (28S) origin matching LSU_rRNA_eukarya (RF02543) in Rfam [35] and clearly not a microRNA. All other checklist criteria are also clearly violated (see Table 1).

Table 1.

Summary of the checklist and three examples of purported human orthologues of mouse “microRNAs” with no genomic match in the human genome

Step	Key question	Red flags	“mir-721″	“mir-690″	“mir-6236″
Annotation	Does the candidate fulfill annotation criteria (e.g. hairpin, Drosha/Dicer processing, 5′ homogeneity, 2-nt offset)?	• tRNA-like structure • Missing mature/star reads • “Known microRNA” not in MirGeneDB	• “Hsa-mir-721” lacks all annotation features • Not Drosha/Dicer processed • Not in MirGeneDB	• “Mmu-mir-690” lacks all annotation features • Not Drosha/Dicer processed • Not in MirGeneDB	• Initial structure looks ok • Not Drosha/Dicer processed • Not in MirGeneDB
Targets	Are predicted targets conserved and supported by robust interaction data?	• Nonconserved sites • Weak context scores • Lack of cross-species evidence	• Claimed target PPARgamma not conserved in humans	• Claimed target Nadk not conserved in humans	• Claimed target PTEN has no bona fide binding site • Used sequence unclear
Genome	Is the full precursor present in the respective genome (human), ideally in conserved syntenic context?	• Absent from genome • High copy number • Repetitive element overlap	• Sequence present in rat only • Human locus only partially overlapping with purported mature.	• “Mmu-mir-690” absent in human genome • Overlaps with SINEs	• Not present as unique microRNA locus • Partial match of ribosomal RNA repeat regions
Expression	Is expression consistent, abundant, and confirmed acros orthogonal datasets?	• Rare or no reads • Likely contamination • Not replicated	• No NGS-reads reported • Just qPCR • qPCR artefact?	• Detected in only 0.0003% of 57 400 human datasets • Likely contamination as co-detected with mouse	• Locus is expressed, but not the purported mature

To understand how big the effect of such “non-microRNA” studies is in the field, we selected seven other microRNA candidates previously rejected by MirGeneDB, applied the checklist and counted how many publications mention them directly and how many publications cite those (Table 2).

Table 2.

The impact of false microRNA annotations on biomedical research

	Annotation	Targets	Genome	expression	Comment	Paper	Citations
“mir-690”	- lack most or all annotation features- not processed by Drosha or Dicer- not in MirGeneDB	Not listed in TargetScan	No	No (contamination)	Rejected from MirGeneDB. Rejected by Chiang et al. Low confidence miRBase.	906	28 376
“mir-721”			No	No NGS (PCR artefact?)	Rejected from MirGeneDB. Not sequenced by Chiang et al. No annotation confidence miRBase.	414	14 187
“mir-6236”			No	Yes, but not the “mature”	Rejected from MirGeneDB. Low confidence miRBase. Mature read shorter than 20 nt.	127	2885
“mir-720”			Yes	n.a.*	tRNA fragment and retracted from miRBase	1822	74 518
“mir-3607”		All predicted targets likely false-positives according to TargetScan	Yes	>1000 RPMM, smear	small nucleolar RNA (snoRNA) fragment and retracted miRBase	691	18 162
“mir-1182”			Yes	Below 50 RPMM, smear	Rejected from MirGeneDB. No annotation confidence miRBase	977	25 410
“mir-1202”			Yes	Below 5 RPMM, smear	Rejected from MirGeneDB. No annotation confidence miRBase	1080	32 818
“mir-1275”			Yes	>100 K RPMM, only one arm	Rejected from MirGeneDB. No annotation confidence miRBase. Mature product ∼16 nt	2463	74 200
“mir-4454”			Yes	>100 K RPMM, only one arm	Rejected from MirGeneDB. No annotation confidence miRBase. Mature product ∼16 nt	854	21 628
“mir-8485”			Yes	>50 K RPMM, smear	Rejected from MirGeneDB. No annotation confidence miRBase	324	6274

Ten chosen “microRNAs” all fail at least three of the four checklist criteria and are mentioned by hundreds of publications that are cited thousands of times. RPMM (reads per million microRNA reads) values from miRCarta are sums across all datasets that showed expression. * ”mir-720” is detectable in NGS data but not listed in miRCarta as it is a retracted entry.

Annotation

All candidates lacked most or all annotation criteria, were not processed by Drosha or Dicer and were not listed in MirGeneDB (Table 2). Two of them (“mir-720” and “mir-3607”) were previously retracted from miRBase as it was discovered they were a tRNA and snoRNA fragment, respectively. The other five “microRNAs” (“mir-1182”, “mir-1202”, “mir-1275”, “mir-4454”, “mir-8485”), despite clearly violating basic annotation rules, are listed in miRBase. “mir-1202” and “mir-8485” were previously used as example of very poorly performing microRNA candidates to explain the uniform set of annotation criteria in Fromm et al. 2015 (Fig. 1) [9]. These findings clearly indicate that all 10 candidates are not microRNAs.

Targeting

According to the TargetScan database [18]—with the exception of “mir-720”, which was not at listed all—all predicted targets for the selected “microRNA” should be considered false-positives, as none of them was in the list of confidently identified microRNAs [18]. This indicates that neither target was conserved either, which is a very common feature of biologically important processes found in many organisms (see the positive example of Mir-126 with 28 deeply conserved targets in TargetScan; Table 2) even though species specific targeting might also occur in highly species-specific gene-regulatory mechanisms.

Genome

In contrast to the previous three examples of “mir-690”, “mir-721”, and “mir 6236”, which could not be confirmed by us as having a genomic representation in the human genome, we found clear evidence for a genomic origin of the seven other microRNA candidates. This neither rejects nor confirms the validity of these candidates being microRNAs.

Expression

To test relative expression levels and distribution of expression of the microRNA candidates, we used miRCarta [36] were >18 000 sequencing datasets (altogether 169 billion smallRNA reads) were mapped to miRBase and other microRNA candidate precursors. We found that all microRNA candidates violated at least one of the criteria for expression as they were either (i) not detected in NGS data (of “mir-690”, “mir-721”, and “mir 6236”, “mir-720”), (ii) detected way below physiologically meaningful levels (“mir-1182”, “mir-1202”), or (iii) detected at meaningful levels, but as a “smear” with no clear mature product (“mir-3607”, “mir-8485”), or with just one 16 nt long arm (“mir-1275”, “mir-4454”). These results further support findings from the annotation and targeting part of the checklist (which are of course interconnected) and clearly show that none of the candidates is a bona fide microRNA.

These findings in 10 unconnected purported microRNAs show that any project, study, or trial assuming those were in fact functional microRNAs will be misled and set up to fail.

To assess the potential impact of incorrectly annotated sequences in biomedical research and using ‘publish or perish’ tool [37], we counted how many publications studied or mentioned each purported microRNA and how often those in turn were cited by other studies (between 2000 and 2025). Strikingly, each misannotated sequence appeared, on average, in nearly 1000 publications, which in turn were cited almost 30 000 times on average (Table 2 and Supplementary File 2). Altogether 8812 unique papers mention one or several of the ten “microRNAs” and those were cited by 267 753 unique papers. While the direct consequences remain unknown, the fact that just 10 erroneous microRNA annotations have influenced such a vast body of research is deeply concerning.

Conclusions

Already with the advent of NGS >20 years ago, many spurious microRNA candidates, including “Mmu-mir-690”, have long been known to fail the annotation criteria originally established in 2003 [5] and updated in 2015 [9]. Employing only bona fide microRNAs of the manually curated microRNA gene database MirGeneDB.org ensures the virtual absence of false-positives with no potential of interacting in a microRNA-like manner with any computationally predicted target give wrong mechanistic concepts and, when forwarded into biomedical applications and patients, false hopes. And while MirGeneDB today serves as repository for bona fide microRNA genes and products, it does not claim completeness. Even though the likelihood for discovery of hitherto undetected microRNAs in human is very low [7], novel microRNAs or what has been called “transitional microRNAs” might yet be discovered [38, 39]. Such case could potentially have been missed from rare developmental stages (fetal) or cell-types. Their broad biological significance in fundamental biological processes, however, is unlikely given the existing deep sequencing of closely related species. However, to know what is and what is not a microRNA has been difficult and remains challenging and until now a comprehensive checklist for assessing putative microRNAs has been missing.

We here presented such a list and applied it first, exemplary, to “Mmu-mir-690”. This revealed multiple, independent lines of evidence that strongly argue against its classification as a bona fide microRNA (Table 1). “Mir-690” fails to meet essential annotation criteria, like a hairpin-like secondary structure or clearly defined mature sequences and appears to be derived from a rodent-specific SINE element rather than the conserved microRNA biogenesis pathway. Its high copy number in the mouse genome and absence from the human genome, coupled with the lack of a conserved target site in NADK, suggest that any reported biological function is likely spurious or misattributed. The negligible detection of “mir-690” sequences in human samples points to contamination rather than endogenous expression. The conclusions drawn in the publications by Rohm et al. and others, asserting a role for “mir-690” in regulating insulin sensitivity via NADK targeting, are therefore most likely incorrect and should be revisited before any further translational steps are pursued. Regrettably, our cautionary submissions to the original publishing venues Cell Metabolism [21, 22] and Nature Metabolism [23, 24] were both declined, highlighting a broader issue of insufficient scrutiny in high-impact biomedical publishing.

While the urgency to discover novel therapeutic avenues for insulin resistance and any disease is both understandable and commendable, such efforts must be grounded in rigorous molecular validation. The presented case of “mir-690” is emblematic of a much wider problem, wherein spurious microRNA annotations—often carried forward from outdated or unverified databases, or through incorrect use of microRNA prediction algorithms—are inadvertently used as the foundation for mechanistic and therapeutic claims. We showed how wide-ranging the problem is by selecting 10 previously rejected “microRNAs” applying the checklist to them and counting the number of scientific publications directly mentioning them and those that cite them. With almost 9000 and 280 000 publications, respectively, the field should be deeply concerned and act, especially because there are at least 1200 other, previously claimed human “microRNAs”, that we have not investigated herein.

Previous studies have emphasized the need for rigor in microRNA research [40], as well as the challenges and opportunities within the field [8, 41] particularly in cancer research [15, 42]. We emphasize that the use of resources like MirGeneDB, in conjunction with the checklist presented here, can prevent the highlighted missteps by providing a standardized and evidence-based framework for microRNA evaluation. Our approach will not only safeguard the integrity of basic research but also ensure that downstream clinical translation efforts rest on a solid molecular foundation. We are currently developing user-friendly web tools to operationalize this checklist and assist researchers in vetting putative microRNAs systematically and automatically. In the spirit of Johann Wolfgang von Goethe’s words, “Knowing is not enough; we must apply. Willing is not enough; we must do”. it is critical that rigorous validation—not enthusiasm alone—guides the journey from molecular discovery to medical application. This responsibility extends to investigators, authors, reviewers, editors, and readers alike.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

TromsøForskningsstiftelse (grant number: 20_SG_BF ‘MIRevolution to B.F.).