Introduction

MicroRNAs (miRNAs) are arguably the most extensively studied noncoding RNA family, thought to fine-tune virtually all signaling pathways and homeostatic processes1,2. Dysregulation of miRNA expression is extensively documented in every major disease, including cancer, cardiovascular and neurological disorders3. The mechanistic paradigm of miRNA action states that they regulate gene expression predominantly at posttranslational level by targeting messenger RNAs (mRNAs) for degradation, translational repression or a combination thereof4,5,6.

Despite over two decades of studies and more than a million indexed publications focusing on, or containing miRNAs, major dilemmas persist in the field. For example, while each miRNA is suspected to regulate multiple genes, the exact number of biologically relevant targets remains unclear. To further compound the controversy, the term “confirmed targets” is used too liberally, without solid evidence that the corresponding interactions actually occur in vivo7.

While most miRNA experimental studies have relied on reagents that either mimic or inhibit the major single-stranded RNA product8,9,10, genetic interventions at these loci are thought to provide powerful biological insights11. A major potential flaw in interpreting such experiments is that the outcomes are typically assessed solely through the lens of targets, with rare or no consideration given to the genomic context. The case of miR-210 exemplifies this challenge. Whereas we have a relatively good understanding of its regulatory inputs, such as oxygen deprivation, immune and inflammatory signals9,10,12,13, the mechanisms responsible for its actions remain controversial. An in-depth examination of the available literature reveals a heavy bias for small sets or even unique targets, including ISCU, NDUFA4, SDHD, and HIF1A8,14,15. However, an exhaustive list of miR-210 candidate targets using the available bioinformatic resources can make the case for hundreds, if not thousands, potential genes16. Furthermore, it is unclear to what extent genes regulated by mature miR-210 remain relevant in knockout models, as published studies often overlook this issue or approach it with a bias towards a limited number of candidates. Of note, these dilemmas are highly relevant for the miRNA field as a whole.

We present a multicenter study integrating genetic interventions in mouse and human cells with pertinent insights from large OMICS datasets. Our investigation provides new evidence of MIR210 locus roles in cellular homeostasis and stress responses and emphasizes the importance of considering overlapping regulatory elements for a rigorous understanding of miRNA loci.

Results

Genetic inactivation of MIR210 locus in human and mouse cells

The human MIR210 gene, first identified as hypoxia-inducible through miRNA microarray analysis17, is located within an intron of a lncRNA gene known as the MIR210 host gene (MIR210HG) in chromosome 11p15.5 region. MIR210 knockout (KO) HEK cells were generated by deleting ~ 30 nucleotides via a bidirectional CRISPR-Cas9 approach. Using a similar strategy, we also engineered an ~ 2.5 kb deletion that eliminates the entire MIR210HG sequence, including the intronic MIR210 locus (Fig. 1a, Suppl. Fig. 1). Homozygous knockout clones (MIR210−/− and MIR210HG−/−) were identified by genomic PCR (Fig. 1b) and expanded for subsequent studies. The loss of RNA products was confirmed by RT-qPCR analysis of miR-210 and MIR210HG (exon 2 transcript) under both high and low oxygenation (Fig. 1c–d).

Mouse immortalized fibroblasts were derived from male mir210lox/lox mice11,13,18. Briefly, dermal fibroblasts were isolated from adult mouse ears and immortalized by shRNA- mediated p53 inactivation. Acute excision of mir210 was achieved by infection with Ad5Cre-eGFP versus Ad5eGFP-only control (Fig. 1e) and the results were confirmed by RT-qPCR for miR-210-3p (Fig. 1f). Reassuringly, in small RNA-seq profiling miR-210-3p and −5p were among the top mature miRNAs differentially expressed between WT versus KO cells (Fig. 1g-h).

Phenotypic repercussions of MIR210 locus loss

The relative fitness of locus-deleted and wild-type cells was investigated in competition assays (Fig. 2a). The results revealed that WT cells consistently outperformed KOs (Fig. 2a), with similar advantage over both small and large deletions. This finding is consistent with the presence of MIR210HG among the 3% of lncRNA loci important for cell growth19. In cellular respirometry and metabolic stress assays using the Seahorse XFp system, MIR210-deficient populations exhibited an abnormal dynamics of OCR shifts (Fig. 2b, d). Deletion also led to mild, but consistent alterations in cell cycle dynamics following oxygen deprivation (Fig. 2c, f). Compared with wild-type controls, MIR210-KO HEK cells are less responsive to hypoxia from the standpoint of S and G2/M phase accumulation (Fig. 2c). A similar effect was observed following the more extensive deletion of the entire MIR210HG (Suppl. Fig. 2). Furthermore, at least in murine fibroblasts, locus deletion is followed by excessive mitochondrial ROS production (Fig. 2e) and altered mitochondrial volume particularly under oxygen deprivation stress (Fig. 2g).

Collectively, these findings indicate that cellular homeostasis and at least some adaptive responses to stress are subtly but consistently altered as a result of eliminating MIR210 locus. Persistent cellular dysfunction of this magnitude may be particularly relevant for specific chronic diseases, which often take decades to manifest.

Consequences of miR-210 disruption on molecular programs

We performed transcriptomic profiling of mouse and human cells under low- and high-oxygen conditions, and pathway enrichment analysis was performed using publicly available tools: Enrichr (https://maayanlab.cloud/Enrichr/) and MSigDB hallmark gene sets to capture specific biological states. The results are summarized in Fig. 3 and Suppl. Fig. 3. Among the top hallmark differences between the KO and WT strains were hypoxia and mTORC1 signaling (Fig. 3a–b), as well as cycle-related signatures (E2F targets, and the G2-M checkpoint) (Fig. 3c–f). Additionally, glycolysis and mitochondrial OXPHOS were identified among the top differential metabolic signatures.

While bioinformatic analyses list tentative directional changes for specific signatures, we caution against a simplistic interpretation that the respective pathways are overactive or conversely inhibited after the loss of the miR210 locus. In our view, a more prudent interpretation is that KO cell responses to stress, including hypoxia, are suboptimally coordinated. For example, while the canonical hypoxic response is still present in KO cells, some HIF targets are excessively induced (SLC2A1, VEGFA, PGK1, NOS2) whereas others respond less or not at all compared to WT cells (CA9, GAPDH, EGLN3), at least at the time of investigation. Overall, our analysis is consistent with a suboptimal response to changes in ambient oxygen following MIR210 locus disruption.

We next focused our attention on miR-210 targets, which remains a controversial topic in studies based on miRNA deletion models. It is noteworthy that the number of miR-210 target candidates amounts to a significant fraction of all coding genes when multiple prediction programs are queried. As such, claims of mechanistic targets can be readily made based on convenient choices from differentially expressed genes (DEGs) in KO versus WT. Published studies rarely present the outcome of miRNA KOs tested in unbiased target enrichment analyses.

Specifically for our case, if mature miR-210 was the most important element eliminated by deletion, its targets would stand out in enrichment analyses, particularly in hypoxia. Contrary to this expectation, in Enrichr analysis using the miRTarBase computational prediction of miRNA-target gene interactions, miR-210 targets are collectively „lost” in a cloud of miRNAs, without significant enrichment in mouse or human KOs (Fig. 3g–i). Furthermore, widely validated targets such as ISCU, EFNA3, and HOXA3 fail to exhibit the expected pattern: higher expression in KOs, particularly in hypoxia. Arguably, the most relevant is the ISCU, which is a bona fide target of miR-210, as its interaction with mir210 is listed in high-throughput RNA-RNA interaction datasets in both human and mouse cells (http://bigdata.ibp.ac.cn/npinter5/).

Consistently, a transcriptomic comparison between MCF7 cells transduced with miR-210 mimic at “hypoxia-similar” level and scrambled controls revealed a highly significant difference in the expected target signature (Suppl. Fig. 3g). An obvious limitation of this result is that it was not derived from the cells subjected to genetic interventions. However, this outcome is consistent with previously published mimic-based interventions, experimentally validated targets and predictions by multiple algorithms.

Taken together, our results support the concept that the transcriptomic, biochemical and phenotypic consequences of MIR210 locus disruption cannot be solely attributed to mature miR-210 loss. Indeed, in specific cases such as ours, the anticipated transcriptomic “footprint” of the mature miRNA cannot be evidenced using unbiased approaches. On the other hand, deregulation of neighboring gene expression is clearly detectable, pointing to overlapping regulatory DNA elements disrupted during these types of interventions.

These observations suggest that our interventions, or indeed similar strategies leading to miRNA KO models, are prone to affect a multitude of genomic features and as such exhibit phenotypes not completely attributable to the elimination of mature miRNA products. Unbiased information on genetic elements and chromatin architecture contained in large public datasets provides important insights into the transcriptional landscape of genetically modified cells. To this end, we examined in detail the genomic neighborhood around MIR210, a region characterized by high synteny and evolutionary conservation (Suppl. Fig. 4a-b).

Indeed, the MIR210/MIR210HG locus exhibits extensive long-range chromatin interactions with nearby loci in a variety of human cells, including HEK293T cells (Fig. 4c; Hi-C data from the 3DIV database, http://3div.kr).

In locus-deleted cells, transcriptomic analysis revealed disrupted expression of multiple genes within the surrounding genomic region. We also noted that the magnitude, timing and direction of differences depended on deletion size and oxygen availability (Fig. 5a, c; Suppl. Figure 5a). We further tested whether DEGs around the MIR210 locus are significantly more affected by our interventions in comparison to randomly selected genomic regions, using complementary randomization strategies: (i) random genome-wide regions of identical size (1 Mb or 0.5 Mb) or (ii) sampling restricted to Chr11, with matched size and coordinates structure of MIR210 locus. Both analyses revealed significant enrichment in the target region. Enrichment was observed under both oxygen concentrations or knockout background (Fig. 5b and Suppl. Fig. 6).

Furthermore, time-course RT-qPCR assays in synchronized populations validated the downregulation of multiple neighboring genes (RNH1, HRAS, LRCC56, LMNTD2 ,RASSF7, PHRF1, IRF7, DEAF1, TMEM80, GATD1, and CD151), particularly in hypoxia (Fig. 5c). A similar study performed on murine fibroblasts further confirmed the broad impact of mir210 elimination on neighboring gene expression (Suppl. Fig. 5b-c).

We should also consider another potentially significant element affected by our genetic intervention, MIR210HG, a relatively abundant lncRNA in some cell types. This poorly characterized transcript is completely absent in the cells harboring the large deletion and also significantly downregulated in the clones with the 30-nt cut. Furthermore, although neither the mm10 or mm39 genomes list a murine MIR210HG ortholog, its existence is plausible based on previous research20, and information from public datasets: cis-regulatory elements (cCREs), RNAseq signals and sequence homology (Fig. 4a–b). Owing to the very nature of how the mouse KO was generated, the putative murine lncRNA would also have been eliminated following AdenoCre-mediated excision. Overall, inadvertent or unavoidable lncRNA elimination (partial or total) needs to be considered as a potential contributor to the biochemical or phenotypical outcomes.

Messages consistent with our results are conveyed by naturally occurring variations in human populations and their correlations with gene expression extracted from the Genotype-Tissue Expression (GTEx) database21. As summarized in Suppl. Fig. 7a-b local single nucleotide polymorphisms are almost exclusively associated with variations in the expression of neighboring genes (rather than MIR210HG itself), at least in some tissues. From this perspective, it only makes sense that comparatively dramatic genomic alterations, such as our engineered deletions, should also be consequential for a multitude of neighboring genes. Such effects, however, should not be considered as an automatic feature of interference (whether natural or experimental) with miRNA loci. As illustrated in Suppl. Fig. 8, naturally occurring variants at another hypoxia-responsive intergenic noncoding locus, miR193BHG/lincNORS22, appear particularly significant for the lncRNA itself, but less so for the abundance of neighboring transcripts.

Taken together, the above results indicate that the impact of MIR210 locus cannot be simplistically reduced to a handful of (conveniently chosen) targets of mature miR-210-3p. Indeed, its relevance for cellular homeostasis and stress responses needs to be understood in a cell-specific context as the combined activity of RNA products (miRNA and lncRNA) and regulatory DNA elements.

Discussion

To address long-standing dilemmas in miRNA biology, we conducted a comprehensive investigation using genetically engineered human and murine cells. Despite a vast and still expanding body of literature on this subject, the depth of our understanding of miRNA biology has been increasingly questioned. Persistent misconceptions and misunderstandings have impacted the overall quality of many publications on this topic that appear in PubMed each year23. Setting aside the unknown negative impact of papermills24,25, most miRNA studies rely on exogenous delivery of reagents (such as “mimic” or “anti-miR”). The specificity of these reagents is often assumed without rigorous testing and there is frequently minor consideration of the natural abundance of miRNAs. Comprehensive analysis of miR-210 target candidates predicted by various large-scale biological databases for miRNA-target interactions often identifies a significant proportion of genes in mammalian genomes, which is biologically improbable. Furthermore, claims of previously validated targets should be interpreted with caution.

Additional complexities arise when comparing the above experimental strategies with targeting miRNAs at the genomic level via Cre-Lox- or CRISPR-based interventions26,27,28. These studies typically report subtle phenotypes, in contrast to the more pronounced effects observed with mimic and anti-miRNAs, raising concerns that the latter systems may be tested outside of the physiologic range. Mechanistically, the relative contributions of individual targets to the observed phenotypes or their impact on gene expression in KO models are often overestimated. There is a notable lack of publications demonstrating that the targets of a given miRNA stand out collectively in unbiased analyses.

Our study demonstrates that the loss of miR-210 locus leads to reduced cellular fitness, a finding consistent with previous experiments involving miR-210 antagonists8,9,10. However, it seems unlikely that the phenotype results solely from the absence of mature miR-210, as our data did not reveal a specific target signature. More critically, we noted deregulation in the expression of neighboring coding genes, some of which are highly expressed. Our findings further suggest that, at least in some cases, experiments using miRNA mimics should not be expected to rescue transcriptional alterations or, indeed, phenotypes.

Importantly, miRNA KO studies often pay limited attention to the regulatory elements disrupted by the intervention, as well as the effects on neighboring genes and potential phenotypic or biochemical ramifications13,18,29,30,31. In contrast, our study revealed that altered expression of neighboring genes represents a major theme, and the specific miR-210 target signature was weak or absent. The common message from enrichment analyses is the collective disruption of MIR210 neighboring genes, suggesting that previously unrecognized regulatory elements are affected by these types of interventions. This theme is recapitulated in both MIR210-KO and MIR210HG-KO cells under varying oxygen conditions. It is noteworthy that MIR210 genomic locus engages in numerous chromatin interactions with neighboring genes, including RASSF7, HRAS1, PHRF1, TMEM80, and RNH1. It stands to reason that disruption of the locus may impact some of these genes, but the experiments provide valuable insights into the specific genes involved, the magnitude of the impact, and the dynamic nature of these effects. Databases such as GTEx and Dice provide potential clues that neighboring genes may be particularly sensitive to genetic changes in the MIR210HG/miR-210 locus. Therefore, it is reasonable to assume that a more substantial intervention at this location (i.e., Cre-Lox or CRISPR cuts) could significantly deregulate neighboring genes, potentially blurring or even erasing the anticipated miRNA-specific signature.

Last but not least, many miRNAs are located within introns of lncRNA genes, and targeting the miRNAs may inadvertently affect the expression of lncRNAs. Notably, in the GTEx database, MIR210HG shows similar or greater expression than lncRNAs with established biological activity (Suppl Fig. 9) and its disruption may therefore have significant biochemical or phenotypical consequences.

Integrated experimental models, such as those described above, are likely to become increasingly relevant in the future. Numerous copy number variations, including relatively small deletions affecting the miR-210 locus, have been cataloged in human populations and are often listed as having “unknown significance”. The region containing miR-210 on Chr11p15.5 is generally overshadowed by the more effectively studied centromeric segment, which includes the IGF2/H19 imprinted locus32,33. Currently, single nucleotide variations at the MIR210 genomic locus, including those in adjacent genes, are linked to chronic conditions, notably systemic lupus erythematosus34,35 aligning with the immune and inflammatory abnormalities observed in mir210-KO mice13,30. Our study underscores the need for further exploration of this genomic region, which harbors a wealth of insufficiently understood coding and regulatory elements. While the data should, at a minimum, prompt the need to revisit miR-210 knockout studies, our investigation is likely more broadly informative for the biology of miRNA loci.

Experimental procedures

Generation of mir210 KO mouse fibroblasts

C57BL/6 mir210-floxed mice were obtained by M. Ivan (Indiana University School of Medicine, Indianapolis, IN) in collaboration with N. Bardeesy (Massachusetts General Hospital, Boston, MA), as previously described18. Housing and all experimental animal procedures followed the NIH research guidelines and were approved by the Institutional Animal Care and Use Committee of the Indiana University School of Medicine. All methods were performed in accordance with the relevant institutional guidelines and regulations, and in compliance with the ARRIVE guidelines (https://arriveguidelines.org).

Primary dermal fibroblasts were isolated from the ears of mir210-floxed male mice (8–12-week-old), as previously described18. Two passages later, the cells where subjected to inactivation of p53 using an shRNA-expressing lentivirus followed by selection with puromycin (2ug/ml), a well-established immortalization method. Locus deletion was subsequently achieved via Ad5CMVCre-eGFP adenovirus infection, as previously described13. Ad5CMV-eGFP infection was performed on control cells. Deletion of miR210 was confirmed by genomic PCR, and the absence of mature mmu-miR-210-3p was confirmed via RT-qPCR using a TaqMan MicroRNA assay (A25576, Thermo Fisher Scientific) and TaqMan MicroRNA Reverse Transcription Kit (4366596, Thermo Fisher Scientific). U6 expression was quantified for data normalization.

CRISPR-Cas9 mediated deletions

HEK293T cell line purchased from the American Type Culture Collection (ATCC) was used to generate human KO cells for either MIR210 or MIR210HG. Plasmid construction and the design of the sgRNAs used to delete the selected fragments were conducted in collaboration with Transposagen Biopharmaceuticals (Lexington, United States). WT HEK cells were seeded in 6-well plates at a density of 50,000 cells per cm2. The following day, the cells were transfected using Lipofectamine 3000 (L3000015, Thermo Fisher Scientific) in OptiMEM reduced serum medium (11058021, Thermo Fisher Scientific), with two vectors containing nucleases targeting the 5′ and 3′ ends of the projected fragment for each KO type (MIR210 or MIR210HG). Specific sequences of the gRNAs are as follows: MIR210 5′ region (5’-GCGCAGTGTGCGGTGGGCAG-3′), MIR210 3′ region (5′-GCCGCTGTCACACGCACAGT-3′), MIR210HG 5′ region (5′- GCTGGCACCCTCTCGCCCCC-3′) and MIR210HG 3′ region (5′-GGGCAAGGAAGCCATCCACA − 3′. Fifteen hours later, the transfection medium was replaced with growth medium, and upon reaching confluence, the cells were passaged and sampled for genomic DNA isolation to confirm deletions. Two rounds of transfection were performed, followed by the derivation of single-cell clones, which were then expanded and screened for deletion status using specific genomic PCR primers to distinguish between heterozygous and homozygous deletions.

Cell culture and reagents

Mouse fibroblasts were cultured in DMEM/high glucose medium (31966047, Gibco) supplemented with 10% (v/v) heat-inactivated FBS (10500064, Gibco) and 8 µM 2-Mercaptoethanol (21985023, Gibco), under 5% CO2 at 37 °C. HEK293T cell line was maintained in high glucose DMEM (31966047, Gibco) supplemented with 10% FBS and 1% antibiotic-antimycotic mixture (15240062, Gibco). The cells were maintained in culture by routine splitting when they reached 70–90% confluence.

Time-course analysis of gene expression

Cells were seeded at a density of 60.000 cells per cm2 in 35 mm diameter cell culture dishes (0003700112, Eppendorf), and allowed to attach for 5 h. Overnight synchronization was achieved by switching to DMEM with 0.5% FBS. The next day, the medium was replaced with complete medium, and cells were exposed to either high (21%) or low (1%) oxygen concertations, with the culture medium in 1% O2 pre-equilibrated overnight in a H35 Hypoxystation (Don Whitley Scientific). Cells were harvested at 6, 12, 24, 36, and 48 h, lysed in TRIzol Reagent (15596018, Thermo Fisher Scientific) and kept at -800C until used for RNA extraction.

RT-qPCR

RNA extraction and qPCR analysis were performed using standard protocols according to the manufacturer’s instructions. Briefly, total RNA was isolated using TRIzol and reverse-transcribed with the High-Capacity RNA-to-cDNA™ Kit (4387406, Applied Biosystems). Quantitative PCR was performed using gene-specific primers and SYBR™ Select Master Mix (4472918, Applied Biosystems) on ViiA™ 7 Real-Time PCR System (Thermo Fisher Scientific). Relative RNA expression was assessed via the ΔΔCT method, normalizing to 18 S rRNA and Rpl32 mRNA levels. All primer sequences will be shared upon specific request.

Cell cycle analysis

First the cells were trypsinized and then resuspended in PBS for cell cycle analysis. Then, approximately 10^5 cells (in 300 µl PBS) were fixed with ice-cold pure ethanol added dropwise to the cell suspension, followed by 3-hour incubation at − 20 °C. After fixation, cells were stained with 20 µg/ml propidium iodide and 100 µg/ml RNAse A for 30 min at 37 °C. Samples were subsequently analyzed on a CytoFlex flow cytometer (Beckman Coulter), capturing data from 20,000 events per sample.

Competitive cell growth assay

Wild-type, miR210-KO, and MIR210HG-KO HEK cells were fluorescently labeled with either dTomato or EGFP through a 48-hour lentiviral infection using the plasmids pUltra-Chili (Plasmid #48687) and pUltra (Plasmid #24129) from Addgene (https://www.addgene.org). To ensure uniform fluorescent labeling, single-cell-derived clones were generated form the GFP/dTomato-expressing cells, which were then further propagated and used for proliferation assays. To assess competitive growth, equal ratios of EGFP-labeled wild-type and dTomato-labeled knockout cells, or the opposite, were cocultured. After 72 h, the growth rate of each genotype was determined by flow cytometric analysis using the CytoFLEX system (Beckman Coulter), which was used to measure the percentage of each cell type in the mixed culture.

RNA-sequencing of HEK cells

RNA-seq was performed by Novogene (UK) Company Limited. The quality of total RNA was assessed using an Agilent 2100 Bioanalyzer. Subsequent library preparation was performed, followed by high-throughput sequencing on the Illumina HiSeq 2500 platform using PE150 mode, generating approximately 20 million reads per sample. Paired-end clean reads were aligned to the reference genome using STAR software (version 2.6.1d). Quantification of reads mapped to each gene was achieved using FeatureCounts (version v1.5.0-p3). Differential gene expression analysis was performed using the DESeq2 package, with significance defined as an adjusted p-value < 0.05 and an absolute log₂ fold-change ≥ 0.3. This relatively permissive fold-change cutoff was selected based on prior experience indicating that miRNA-mediated gene modulation, particularly at the transcript level, often results in subtle yet biologically meaningful expression changes.

RNA sequencing of mouse fibroblasts

Two independent cultures were performed in either 21 or 1% hypoxia during 24 h. RNA was extracted with the miRNeasy mini kit (Qiagen) and libraries were constructed for both small RNA-seq and mRNA-seq as previously described36. Briefly, mRNA-seq was performed from 2 µg of RNA that was first subjected to mRNA selection with a Dynabeads® mRNA Purification Kit (Invitrogen). mRNA was fragmented for 10 min at 95 °C in RNAseIII buffer (Invitrogen) then adapter-ligated, reverse transcribed and amplified with the reagents from the NEBNext mRNA Library Prep Set for SOLiD. Small RNA-seq was performed from 500 ng RNA with the NEBNext Small RNA Library Prep Set for SOLiD according to manufacturer’s instructions. Both types of amplified libraries were purified with a Purelink PCR micro kit (Invitrogen) and then subjected to additional PCR rounds with primers from the 5500 W Conversion Primers Kit (Life Technologies). After Agencourt® AMPure® XP bead purification (Beckman Coulter), libraries were size-selected from 150 nt to 250 nt (for RNA-seq) and 105 nt to 130 nt (for small RNA-seq) with the LabChip XT DNA 300 Assay Kit (Caliper Life Sciences), and finally quantified with the Bioanalyzer High Sensitivity DNA Kit (Agilent). Libraries were sequenced on SOLiD 5500XL (Life Technologies) with single-end 50b reads. SOLiD data were analyzed with Lifescope v2.5.1, using the small RNA pipeline for miRNA libraries and whole-transcriptome pipeline for RNA-seq libraries with default parameters. Annotation files used for production of raw count tables correspond to Refseq Gene model for mRNAs and miRBase for small RNAs. Differential expression analysis was performed using the DESeq2, defining significantly differentially expressed genes as those with an adjusted p-value < 0.05 and an absolute log₂ fold-change ≥ 0.3.

Transmission electron microscopy

Cells were fixed in 2,5% buffered glutaraldehyde followed by post-fixation in 1% osmium tetroxide with 1.5% potassium ferrocyanide in 0.1 M cacodylate buffer. Sequential dehydration in 70%, 95% and 100% ethanol was followed by sample embedding in Agar 100 epoxy resin at 60 °C for 48 h. Ultrathin 80-mm-thick sections were cut via a diamond knife and double-stained with uranyl acetate and lead citrate. Ultrathin sections were examined using a Morgagni 268 transmission electron microscope (FEI Company, The Netherlands) at 80 kV. Digital electron micrographs were acquired using a MegaView III CCD using iTEM-SIS software (Olympus, Soft Imaging System GmbH, Germany). The quantification of mitochondria and endoplasmic reticulum relative volumes (V/v) were estimated using a grid superimposed onto electron microscopy images. The images were acquired from 2 sections/sample and 10 different images/section. The relative volumes were estimated by a point-counting method using stereological grids with regularly distributed test-points (at 0,2 μm) generated in iTEM-SIS software (Olympus, Soft Imaging System GmbH, Germany).

Cellular bioenergetics stress

Cells were subjected to Seahorse XFp Cell Energy Phenotype Test (103325-100, Agilent) and Seahorse XF Cell Mito Stress Test (103010-100, Agilent), using the Seahorse XFp Analyzer (Agilent, CA, USA) per manufacturer’s instructions. Briefly, the cells were seeded at 2 × 10^5 cells per well in miniplates for 24 h, and the growth medium was switched to the specially formulated assay medium before testing. The oxygen consumption rate (OCR) was measured under both basal and stress conditions induced by FCCP and oligomycin. Mitochondrial functionality was further investigated by challenging the cells with sequential additions of oligomycin, FCCP, rotenone, and antimycin A. Data normalization was based on DNA content (Hoechst 33258 staining), and data analysis was performed using Wave 2.6.3 software’s Report Generators.

Reactive oxygen species (ROS) production

ROS production in mouse fibroblasts was quantified using MitoTracker™ Red CM-H2XRos (M7513, Thermo Fisher Scientific) and flow cytometry, according to the manufacturer’s guidelines. Briefly, cells were cultured at a density of 2 × 10^4 cells/cm^2 for 24 h and then exposed to standard (21% O2) or low-oxygen (1% O2) conditions for another 24 h, as described above. After 48 h, without changing the oxygen exposure, cells were detached, resuspended in 300 µl DMEM w/o phenol red, and stained with MitoTraker dye at 400 nM for 30 min at 37 °C. The stained cells were then washed and immediately analyzed on a flow cytometer after the addition of propidium iodide, with data collected from 50,000 events per sample to ensure robust measurements.

Bioinformatic analysis

Pathway enrichment analysis of DEGs was performed using the Enrichr software tool37,38. Briefly, using the Appyter for bulk RNA-seq, DEGs were analyzed utilizing the MSigDB_Hallmark_2020 gene set to identify and cluster significant pathways. Additionally, miRTarBase_2017 gene set was utilized to estimate the odds ratios for microRNAs potentially regulating the input DEG set. To assess whether DEGs were statistically enriched in the extended MIR210 genomic locus compared to random genomic regions, we used the GenomicRanges package39, to compute overlaps between this region and genome-wide DEG coordinates. To establish an empirical null distribution, we implemented two strategies for generating randomized genomic intervals. First, we used a genome-wide approach, sampling 1000 random intervals of 500k or 1 Mb across chromosomes 1–22, X, and Y, proportionate to their sizes, while excluding centromeric and telomeric gaps based on BSgenome.Hsapiens.UCSC.hg38. Second, we used a chromosome-localized approach, where for each gene in the MIR210 locus, we sampled a coordinate-matched region of identical length randomly along chromosome 11, excluding out-of-bound coordinates. Using counbtOverlaps() we quantified overlaps and evaluated statistical significance with Fisher’s exact test, to determine if the DEG density in the MIR210 region was significantly higher than expected by chance. A Manhattan-style plot generated using ggplot240 visualized the distribution of DEG overlaps across all regions. The UCSC Genome Browser was used to display the genomic positioning and structure of the MIR210HG locus and the neighboring genes. Long-range chromatin interactions were examined using publicly available Hi-C data available through the 3DIV database (http://3div.kr)41.

Statistical analysis

Statistical analysis was performed with GraphPad Prism (version 7.05). Unless stated otherwise in the figure legend, the statistical significance of differences was determined by two-way ANOVA, and Tukey’s test was applied for multiple comparisons. Difference were considered statistically significant when the p-value was < 0.05. Data are presented as mean ± SD, unless otherwise stated.

Fig. 1
figure 1

Generation and validation of experimental models. (a) Schematic overview illustrating the generation of HEK knockout (KO) cell clones via CRISPR-Cas9 genome editing, created using BioRender.com (https://biorender.com/); (b) Genomic validation by PCR analysis confirms single-cell-derived KO clones for MIR210 or MIR210HG, identifying both heterozygous (+/−) and homozygous (−/−) genomic deletions. (c–d) Relative levels of miR-210-3p, and MIR210HG transcripts in wild-type (WT), MIR210–KO, and MIR210HG–KO cells under 21% or 1% O2 for 24 h. Data shown as mean ± SD and the statistical significance determined by two-way ANOVA (****p < 0.0001). (e) Experimental workflow of mir210 locus excision in immortalized primary fibroblasts from mir210 lox/lox male mice, created with BioRender.com (https://biorender.com/); (f) Data are shown as mean ± SD, with statistical significance determined by two-way ANOVA (**p < 0.01) (g–h) Volcano plots showing differential expression of miRNAs in mir210-KO mouse fibroblasts under 21% or 1% O2, with significantly downregulation of miR-210-5p and miR-210-3p.

Full size image
Fig. 2
figure 2

Phenotypic and metabolic impact of MIR210 locus deletion. (a) Competitive growth assays of WT and MIR210-KO or MIR210HG-KO clones labeled with GFP or dTomato, analyzed after 3 days of coculture. Flow cytometry results show significantly fewer KO cells compared to WT (mean ± SD from 3 experiments, with four different clones analyzed per experiment; two-way ANOVA, **p < 0.01, ***p < 0.001). (b) Mitochondrial respiration (OCR) in HEK WT and KO clones under basal or stress conditions. Data are mean ± SD from 2 experiments in duplicate (two-way ANOVA, n.s. not significant, ***p < 0.001, ****p < 0.0001). (c) Significant changes in cell cycle distribution in HEK cells (WT and MIR210-KO) under different oxygen levels (two-way ANOVA, *p < 0.05, ****p < 0.0001). (d) OCR profiles from a mito stress test in WT and mir210-KO mouse fibroblasts reveal reduced maximal respiration and spare capacity in KO cells (mean ± SD from a single representative experiment, three replicates). (e) Mitochondrial ROS quantification in WT and mir210-KO mouse fibroblasts after 24-hour exposure using MitoTracker™ Red CM-H2XRos and flow cytometry (mean ± SD from 3 experiments, two-way ANOVA, **p < 0.01). (f) Cell cycle alterations in mir210-KO mouse fibroblasts under different oxygen levels (two-way ANOVA, **p < 0.01). (g) Cellular ultrastructure by transmission electron microscopy shows structural variations in mitochondria and endoplasmic reticulum in KO fibroblasts under hypoxic conditions.

Full size image
Fig. 3
figure 3

Transcriptomic consequences of locus disruption in mouse and human cells. Pathway enrichment analysis of gene expression in miR-210-KO mouse fibroblasts and HEK cells under low O2 (1%) was performed using Enrichr and visualized with an Appyter for bulk RNA-seq (https://appyters.maayanlab.cloud/#/Bulk_RNA_seq). (a) Scatter plot with larger, black-outlined points highlighting significantly enriched pathways, grouped by gene set similarity. (b) Bar charts displaying the most significant gene sets from MSigDB_Hallmark_2020, with levels of significance indicated. (c) Volcano plots showing microRNAs potentially affecting downregulated DEGs in mir210-KO cells, with each microRNA represented by its odds ratio and statistical significance. (d–f) Similar analysis for MIR210-KO HEK cells under 1% O2, applying a stricter significance threshold (p-adjusted < 0.01) to limit gene number for more precise enrichment analysis. (g–i) Parallel analysis in MIR210HG-KO HEK cells under similar experimental setups, detailing transcriptomic shifts and pathway dynamics. For all analyses, DEGs were defined based on a log₂ fold-change cutoff of 0.3.

Full size image
Fig. 4
figure 4

Genomic features of human (a, c) and mouse (b) miR-210 locus. ENCODE database summary of transcript abundance, transcriptional regulator binding (ReMap density, ChIP-seq), chromatin states (red: active promoters; orange: enhancers) and mapped transcription start sites (TSS, FANTOM5). (c) Hi-C interactions in HEK293T cells anchored on MIR210HG locus (3DIVdatabase). Interaction frequency is represented by magenta dots and blue arcs correspond to interactions above the cut-off value (green line).

Full size image
Fig. 5
figure 5

Effect of MIR210 locus deletion on local transcriptional gene regulation in human cells. (a) Genomic localization of DEGs on Chr11 (p15.5) in HEK cells under hypoxia (1% O2, 24 h), comparing miR210-KO vs. WT. Genes are plotted based on their chromosomal position on the x axis and their differential expression (log2FoldChange) on the y-axis. Genes significantly differentially expressed (p adj < 0.05) were name-labelled and marked with red dots. MIR210 genomic locus is highlighted by a vertical dashed line. (b) Enrichment analysis of DEGs within the MIR210 region (± 250 kb), compared to 1,000 randomly selected genomic regions of similar size. The MIR210 region (red dot) shows a statistically significant enrichment (Fisher’s exact test p = 2.1 × 10−7) (c) Heatmap showing temporal expression profiles (log2 fold change relative to normoxic WT at 6 h) of selected hypoxia-responsive genes in HEK WT, MIR210-KO, and MIR210HG-KO cell lines under normoxia and hypoxia (1% O2) at indicated timepoints (6–48 h).

Full size image