RNA Interference and microRNAs
In 1998, two scientists injected double-stranded RNA into roundworms. They watched genes shut off with uncanny precision. The effect was a thousand times stronger than injecting either RNA strand alone. It also spread from cell to cell. It silenced genes for generations of progeny. This was not a curiosity. It was the discovery of a hidden layer of gene regulation. This layer now is known to control roughly half of all human protein-coding genes. It is encoded by microRNAs and executed by a machine called RISC. RNA interference (RNAi) matters because it revealed something important. RNA is not merely a passive messenger between gene and protein. Instead, it is an active regulator with sequence-specific "search-and-destroy" capability. Cells exploit small RNAs to fine-tune development. Most important are small interfering RNAs (siRNAs) and microRNAs (miRNAs). These cells also defend against viruses, silence transposons, and buffer gene expression noise. These pathways can be reprogrammed with a synthetic RNA of the experimenter's choosing. Because of this, RNAi also became one of the most transformative laboratory tools of the twenty-first century. It lets researchers knock down almost any gene without touching the DNA. The big picture is this: small RNAs (roughly 21–23 nucleotides) are loaded into an Argonaute protein. The protein uses the RNA as a sequence guide. It finds complementary messenger RNAs this way. Depending on the degree of complementarity, the target faces one of three fates. It is either sliced in two, translationally silenced, or marked for accelerated decay. This same core machine — a small guide RNA plus an Argonaute effector — gets repurposed by cells. It serves at least two distinct biological jobs. One is fighting invading nucleic acids (the ancestral, defensive role of siRNAs). The other is rewiring endogenous gene networks (the regulatory role of miRNAs). This article builds the mechanism step by step. It starts from long double-stranded RNA or a hairpin transcript. It moves through the dicing and loading steps, and reaches the fate of a repressed messenger RNA. It then walks through the classic experiments that established each step of the pathway. These experiments were done in Caenorhabditis elegans, Drosophila, and human cells. It closes with worked genetic and dosage examples. It also covers common conceptual traps students fall into. Finally, it discusses present-day biomedical uses of this pathway, from antiviral defense in plants to FDA-approved siRNA drugs.
- Double-stranded RNA (dsRNA) is cleaved by the ribonuclease Dicer into ~21–23 nt small RNA duplexes.
- One strand (the guide) is loaded into an Argonaute protein to form the core of RISC (RNA-induced silencing complex).
- siRNAs are usually perfectly complementary to their target and direct Argonaute's slicer activity to cut the mRNA.
- miRNAs are endogenous, imperfectly complementary regulators that typically repress translation and trigger deadenylation/decay rather than slicing.
- Fire and Mello's 1998 experiment proved dsRNA, not single-stranded antisense RNA, is the potent silencing trigger — work that won the 2006 Nobel Prize.
- miRNA target recognition depends overwhelmingly on a 6–8 nucleotide "seed" match, letting one miRNA regulate hundreds of mRNAs.
A small guide RNA loaded into an Argonaute protein converts base-pairing into gene silencing — the same guide-and-cut logic distinguishes antiviral siRNAs from broadly regulatory, imperfectly-paired miRNAs.
The core idea
RNA interference is gene silencing directed by small RNAs that base-pair with target messenger RNAs and recruit an Argonaute-family protein to repress or destroy them. Two related but distinct classes of small RNA drive this: small interfering RNAs (siRNAs), typically 21–22 nucleotides, which arise from long double-stranded RNA (viral replication intermediates, transposon transcripts, or experimentally introduced dsRNA) and are near-perfectly complementary to a single target, usually triggering endonucleolytic cleavage; and microRNAs (miRNAs), typically 21–23 nucleotides, which are encoded by the cell's own genome as hairpin-forming transcripts and are only partially complementary to their targets, usually triggering translational repression and mRNA destabilization rather than cleavage. Both classes converge on the same enzymatic logic. A ribonuclease III enzyme called Dicer chops long dsRNA or a hairpin precursor into a small RNA duplex of defined length. One strand of that duplex — the guide strand — is retained in an Argonaute (AGO) protein, while the other strand (the passenger strand) is discarded. Guide-loaded Argonaute constitutes the catalytic core of the RNA-induced silencing complex (RISC). Argonaute then uses Watson-Crick pairing between the guide and a target RNA to locate substrates: extensive pairing across nearly the whole guide permits Argonaute's endonuclease ("Slicer") domain to cleave the target directly between guide nucleotides 10 and 11; partial pairing, confined mostly to a 6–8 nucleotide "seed" region near the guide's 5′ end, instead recruits repressive cofactors that block translation initiation and accelerate deadenylation and decapping of the mRNA. The distinction between siRNA and miRNA pathways is therefore not about different machinery but about different degree of complementarity and different biogenesis. Plants and invertebrates use RNAi vigorously as an inherited antiviral immune system and to police transposons through related small RNAs (piRNAs, endo-siRNAs); vertebrates use miRNAs pervasively as endogenous fine-tuners of gene networks controlling development, metabolism, and cell identity, working alongside — but mechanistically continuous with — the transcription factor networks covered under Principles of Gene Regulation.
Mechanism
Step 1 — Generating the trigger. For exogenous RNAi, the trigger is long double-stranded RNA: from a replicating virus, a transposon transcribed from both strands, or experimentally transfected dsRNA. For endogenous miRNAs, RNA polymerase II transcribes a primary miRNA transcript (pri-miRNA), often kilobases long and capped and polyadenylated like a normal mRNA, containing one or more imperfect hairpin structures. In the nucleus, the microprocessor complex — the RNase III enzyme Drosha together with its dsRNA-binding partner DGCR8 (Pasha in invertebrates) — crops the hairpin out of the pri-miRNA, releasing a ~60–70 nucleotide stem-loop called pre-miRNA.
Step 2 — Nuclear export. The pre-miRNA is recognized by Exportin-5, in a Ran-GTP–dependent manner, and exported through the nuclear pore into the cytoplasm. This is a committed, rate-limiting step: Exportin-5 protects the pre-miRNA's characteristic 2-nucleotide 3′ overhang (the signature of RNase III cleavage) from cytoplasmic degradation during transit.
Step 3 — Dicing. In the cytoplasm, Dicer, another RNase III enzyme, measures a fixed distance from the pre-miRNA's terminus using its PAZ domain and cleaves off the terminal loop, producing a ~21–23 bp RNA duplex with 2-nucleotide 3′ overhangs on each end — the structural signature that RNase III enzymes leave on both cut ends. For exogenous long dsRNA, Dicer processively dices the whole molecule into a population of siRNA duplexes with the same architecture. Dicer often works with a dsRNA-binding partner (TRBP or PACT in humans; R2D2 in Drosophila) that helps hand the duplex to Argonaute.
Step 4 — RISC loading and strand selection. The siRNA/miRNA duplex is loaded into an Argonaute protein. Argonaute (or its loading partner) measures the thermodynamic stability of the two duplex ends: the strand whose 5′ end sits at the less stably paired end is preferentially retained as the guide strand, while the other (passenger) strand is cleaved by Argonaute's slicer activity if the duplex is siRNA-like and near-perfectly paired, or is simply displaced and degraded if it is not. This asymmetry rule, discovered by Schwarz, Zamore and colleagues, explains why only one strand of a diced duplex typically becomes functional.
Step 5 — Target recognition. The mature RISC (guide-loaded Argonaute, plus accessory proteins such as GW182/TNRC6 in animals) scans mRNAs for complementarity. Human Argonaute proteins (AGO1–AGO4) present the guide strand's seed sequence (nucleotides 2–8) in a pre-organized, solvent-exposed conformation, primed for pairing before the target is even encountered — this is why seed matching, not full-length complementarity, dominates animal miRNA target selection. Most animal miRNA target sites lie in the mRNA's 3′ untranslated region (3′ UTR).
Step 6 — Silencing outcome. If pairing is extensive and includes the "cleavage-competent" central region (guide nucleotides 10–11), Argonaute's PIWI domain — structurally related to RNase H — cuts the target's phosphodiester backbone directly opposite this region, releasing two fragments that are rapidly degraded by cellular exonucleases (Xrn1 for the 5′ fragment, the exosome for the 3′ fragment). This slicing mode is the norm for plant miRNAs and for siRNAs, both of which usually pair near-perfectly with their targets. If pairing is confined mainly to the seed (the norm for animal miRNAs), Argonaute cannot slice; instead GW182/TNRC6 proteins recruit the CCR4-NOT and PAN2-PAN3 deadenylase complexes, which shorten the poly(A) tail, followed by decapping (DCP1/DCP2) and 5′→3′ exonucleolytic decay (Xrn1). Ribosome recruitment (via eIF4E/eIF4G) is also directly impaired, so repression begins before decay is complete — this is why some repressed transcripts show reduced translation with only modest changes in steady-state mRNA level, and it is why a single miRNA usually produces modest fold-repression (often 1.5- to 4-fold per target) rather than an on/off switch.
Step 7 — Amplification and spreading (in plants, fungi, worms). In C. elegans and plants, RNA-dependent RNA polymerases (RdRPs) use the primary siRNA-guided cleavage products as templates to synthesize new dsRNA, generating secondary siRNAs and vastly amplifying the silencing signal. In C. elegans, dsRNA-derived signals also spread systemically between cells via dedicated transmembrane channel proteins (SID-1), which is why feeding a worm bacteria expressing dsRNA silences the corresponding gene throughout its body and even in its offspring — a phenomenon with no confirmed mammalian counterpart, since mammals lack RdRPs and robust systemic spread.
The evidence
Fire and Mello (1998): dsRNA, not antisense RNA, is the trigger. Prior to 1998, researchers injecting C. elegans with single-stranded antisense RNA against a target gene sometimes saw modest, inconsistent silencing — and, confusingly, sense-strand "control" RNA sometimes worked too. Andrew Fire and Craig Mello, working with colleagues including Mary Montgomery and SiQun Xu, injected purified sense RNA, antisense RNA, and deliberately annealed double-stranded RNA targeting the muscle gene unc-22 into the worm gonad. Only the dsRNA produced strong, highly penetrant silencing — a twitching phenotype mimicking genetic unc-22 loss-of-function — and it did so at concentrations roughly two orders of magnitude lower than either single strand alone. Silencing was specific to the exon sequence used, worked in a manner catalytic rather than stoichiometric (far fewer dsRNA molecules than target transcripts were needed to silence the whole tissue), and spread systemically to distal tissues and to progeny that had never seen the injected RNA. This 1998 Nature paper coined the term RNA interference and led directly to the 2006 Nobel Prize in Physiology or Medicine, awarded jointly to Fire and Mello. Its central proof was elegantly simple: it is the double-strandedness of the trigger, not its sense or antisense identity, that determines potent, specific silencing — implying a dedicated cellular machine recognizes dsRNA as an instructive rather than accidental molecule.
Hamilton and Baulcombe (1999) and Zamore, Tuschl, Sharp, Bartel (2000): the small RNA intermediate. Working in plants undergoing post-transcriptional gene silencing (a phenomenon related to RNAi), David Baulcombe's group detected small ~25 nucleotide RNAs specifically in silenced tissue — the first physical evidence for a small RNA species correlating with silencing. Then Phillip Zamore, Thomas Tuschl, Phillip Sharp and David Bartel, using Drosophila embryo lysates that recapitulated RNAi in vitro, showed that long dsRNA is processed into ~21–23 nucleotide fragments before target cleavage occurs, and that the target mRNA is cut at sites spaced at regular ~21–23 nucleotide intervals correlating with the introduced dsRNA — direct biochemical proof that a diced small RNA, not the long dsRNA itself, guides sequence-specific cleavage. Tuschl's group then showed that chemically synthesized 21-nucleotide siRNA duplexes, with characteristic 2-nucleotide 3′ overhangs, were sufficient to trigger silencing without any Dicer processing step required — pinning down the minimal functional unit of RNAi and opening the door to using synthetic siRNAs as a laboratory tool.
Bernstein, Hannon and colleagues (2001): identifying Dicer. Using a Drosophila embryo lysate biochemical fractionation approach, Gregory Hannon's group purified the enzymatic activity that converts long dsRNA into ~22 nucleotide fragments, and identified it as a member of the RNase III family of double-strand-specific ribonucleases, which they named Dicer. They showed recombinant Dicer alone could process dsRNA into siRNA-sized fragments in vitro, directly confirming its identity as the processing enzyme predicted by the size-regularity observed by Zamore and colleagues.
Lee, Feinbaum and Ambros (1993) and Reinhart et al. (2000): discovery of miRNAs, before the term existed. Victor Ambros and colleagues, studying the C. elegans heterochronic gene lin-4 — required for the normal timing of larval developmental transitions — found that lin-4 did not encode a protein at all, but instead produced two small RNA transcripts (a ~22 nt and a ~61 nt species) with antisense complementarity to repeated sequence elements in the 3′ UTR of a second heterochronic gene, lin-14. Loss of lin-4 caused lin-14 protein to remain abnormally high, and mutating the complementary sites in the lin-14 3′ UTR phenocopied loss of lin-4 — establishing that a small RNA can repress a target gene post-transcriptionally purely through 3′ UTR base-pairing, seven years before RNAi's biochemical mechanism was worked out. For years lin-4 was considered a nematode idiosyncrasy. That changed in 2000, when Ambros's and Gary Ruvkun's groups discovered a second small RNA, let-7, also controlling C. elegans developmental timing — and showed that let-7 orthologs, with identical sequence, exist in flies, mollusks, and humans. This conservation, published in two back-to-back 2000 Nature papers, proved that small regulatory RNAs are not a nematode curiosity but an ancient, conserved layer of animal gene regulation — precipitating the search that would uncover hundreds of human microRNAs within a few years.
Elbashir, Tuschl et al. (2001): RNAi works in mammalian cells. A major puzzle was that mammalian cells, when exposed to long dsRNA, mount an innate interferon response (via protein kinase R and 2′-5′-oligoadenylate synthetase) that causes global translational shutdown and cell death, masking any sequence-specific effect. Tuschl's group showed that chemically synthesized 21-nucleotide siRNA duplexes — short enough to evade this interferon-triggering length threshold — could silence endogenous and reporter genes specifically in cultured human and mouse cells without provoking the interferon response. This 2001 result was the technical breakthrough that made RNAi knockdown a routine tool in mammalian molecular biology and set the stage for therapeutic siRNA development.
Worked examples
Example 1 — Predicting miRNA target repression from seed matches and a genetic cross. Suppose in Drosophila, a miRNA called miR-X has the seed sequence 5′-AGCACUU-3′ (nucleotides 2–8 of the mature miRNA). A researcher finds a candidate target mRNA whose 3′ UTR contains the sequence 5′-...AAGUGCU...-3′, which is exactly complementary to the seed (a canonical "7mer-A1" site, since it is also flanked by an A opposite guide position 1). This is a strong prediction of direct repression. To test genetically, the researcher crosses flies heterozygous for a miR-X null allele (miR-X⁻/+) to each other. Assuming simple recessive loss-of-function with no haploinsufficiency, the F1 shows the standard monohybrid ratio: ``` miR-X⁺/miR-X⁺ : miR-X⁺/miR-X⁻ : miR-X⁻/miR-X⁻ 1 : 2 : 1 ``` Only the 25% homozygous null (miR-X⁻/miR-X⁻) class lacks miR-X entirely. In that quarter of progeny, Western blotting shows the candidate target protein elevated roughly 2.5-fold above wild-type, while mRNA levels rise only ~1.3-fold — consistent with the miRNA acting predominantly through translational repression with modest mRNA destabilization, exactly the pattern expected for a seed-matched, non-slicer target. A rescue construct expressing miR-X from a heat-shock promoter in the null background restores target protein to near wild-type levels, confirming the seed-predicted target is a genuine, direct regulatory target rather than a coincidental sequence match. This combination — bioinformatic seed prediction, loss-of-function genetics, and rescue — is the standard logic used to validate real miRNA-target relationships (e.g., miR-1 targeting the transcription factor Hand2 in mammalian heart development).
Example 2 — Dosing an siRNA knockdown experiment and interpreting an unexpected result. A graduate student wants to knock down gene Y in cultured human hepatocytes using synthetic siRNA transfected at 20 nM final concentration, expecting via the manufacturer's data an 80% reduction in Y mRNA (i.e., knockdown to 20% of control) by 48 hours, measured by quantitative RT-PCR normalized to a housekeeping gene. She transfects, and at 48 hours she measures gene Y mRNA at 55% of the negative-control siRNA sample — far weaker than expected — while a positive-control siRNA against a validated housekeeping-adjacent gene achieves the expected 85% knockdown. Because the positive control worked, transfection efficiency itself is not the likely problem. Three biologically grounded explanations to test, in order of typical culprit frequency: (1) the target site may be masked by a bound RNA-binding protein or by extensive secondary structure, reducing effective RISC access — testable by trying a second, non-overlapping siRNA against a different region of the same mRNA; (2) gene Y mRNA may have unusually long half-life or high transcription rate, meaning that even efficient RISC-mediated cleavage per molecule is outpaced by ongoing synthesis — testable by extending the time course to 72–96 hours or by measuring transcription rate directly; (3) off-target seed effects in the positive control could be inflating its apparent "knockdown" by coincidentally destabilizing the housekeeping normalizer itself, making the positive control look artificially strong — testable by re-normalizing to a different, unrelated housekeeping transcript, or better, to total RNA mass. Repeating the experiment with a second, independent siRNA and a corrected normalization gene shows 78% knockdown, consistent with the original result being a target-accessibility problem for the first siRNA sequence, not a failure of the pathway. This kind of triage — testing an alternative reagent, extending kinetics, and re-examining internal controls before concluding a pathway "didn't work" — is standard practice whenever an siRNA experiment underperforms.
Common misconceptions & traps
| Misconception | Why it's wrong |
|---|---|
| "siRNA and miRNA are made by completely different pathways." | They converge on the same Dicer/Argonaute core machinery; they differ mainly in biogenesis (hairpin vs. long dsRNA) and in degree of target complementarity, not in the enzymes used. |
| "A microRNA always destroys its target mRNA." | Animal miRNAs mostly act via seed-only pairing, causing translational repression plus deadenylation-triggered decay — not direct endonucleolytic slicing, which requires near-perfect pairing (the norm for plant miRNAs and siRNAs, not animal miRNAs). |
| "One miRNA regulates one gene." | Because repression depends mainly on a short 6–8 nt seed match, a single miRNA can have hundreds of predicted targets, and most mRNAs are targeted by multiple different miRNAs — this is a combinatorial network, not a one-to-one switch. |
| "RNAi knockdown is equivalent to a genetic knockout." | siRNA/shRNA knockdown reduces but rarely eliminates protein completely, is transient (siRNA) or dose/copy-dependent (shRNA), and can produce off-target effects from seed-matched but unintended mRNAs — knockdown and knockout are not interchangeable evidence. |
| "Passenger and guide strands are functionally equivalent, so either could work as the biological regulator." | Only the guide strand is normally retained by Argonaute (via the strand-asymmetry rule); the passenger strand is discarded/degraded and is not the physiological regulator, though it can occasionally cause unintended off-target silencing if incompletely removed. |
The scientist behind it
Origins · 1 minAndrew Z. Fire and Craig C. Mello and how this was discovered
Andrew Fire, a molecular biologist trained at MIT and Sussex under Sydney Brenner's C. elegans research tradition, and Craig Mello, a developmental biologist at the University of Massachusetts, collaborated across their two labs in the mid-1990s while both were puzzled by inconsistent antisense RNA results in worms. Rather than accepting the folk explanation that antisense worked erratically due to technical noise, they designed a controlled comparison — sense RNA, antisense RNA, and annealed double-stranded RNA, injected side by side into the same tissue at matched concentrations. The result was unambiguous and reproducible, and it was the double-strandedness of the RNA, not its antisense identity, that explained years of confusing prior results across the field. Fire and Mello shared the 2006 Nobel Prize in Physiology or Medicine "for their discovery of RNA interference — gene silencing by double-stranded RNA," awarded just eight years after their original 1998 paper — among the fastest Nobel recognitions in the prize's history, reflecting how quickly the finding reshaped experimental biology.
Where this matters today
Therapeutics. Synthetic siRNAs chemically stabilized against nuclease degradation and conjugated to targeting ligands are now approved drugs. Patisiran (targeting transthyretin mRNA, for hereditary transthyretin amyloidosis) and inclisiran (targeting PCSK9 mRNA, for hypercholesterolemia, requiring dosing as infrequently as twice yearly) exploit RISC-mediated cleavage to silence disease-driving or disease-modifying genes directly, bypassing the need for a small-molecule drug target.
Functional genomics. Genome-wide RNAi and shRNA (short hairpin RNA, expressed from a viral vector and processed by the endogenous Dicer/Drosha pathway) screens have been used to systematically knock down every gene in the genome in cultured cells, identifying genes required for viral infection, cancer cell survival, and drug resistance — a technique that predated and complements CRISPR-based screens.
Agriculture and plant biotechnology. Because plants use RdRP-amplified RNAi as their principal antiviral defense, RNAi-based crop protection (engineering plants to express dsRNA against a pest insect's essential gene, exploiting the insect's own RNAi uptake machinery) is used commercially against corn rootworm, reducing reliance on chemical pesticides.
Diagnostics and biomarkers. Circulating miRNAs, remarkably stable in blood plasma due to protection within exosomes and Argonaute complexes, are being developed as minimally invasive biomarkers for cancers, cardiovascular disease, and pregnancy complications, since tissue-specific miRNA signatures leak into circulation in disease-associated patterns.
Basic developmental and cancer biology. miRNA dysregulation is a hallmark of many cancers: the miR-17-92 cluster acts as an oncogene by repressing tumor suppressors, while let-7 — the same miRNA discovered in worms — acts as a tumor suppressor in humans by repressing oncogenes including RAS family members, directly linking a 1990s developmental-timing discovery in a millimeter-long worm to modern cancer biology. These downstream systems connect naturally to further layers of RNA-based regulation, including Long Non-Coding RNAs that scaffold chromatin complexes rather than guiding Argonaute, and the broader landscape of Regulation of Translation that miRNAs tap into when they block ribosome recruitment.
Ask the tutor: Why do plant and animal miRNA pathways differ so much in how often they trigger direct mRNA slicing versus translational repression?
Are siRNAs and miRNAs the same molecule, just with different names?
No. Both are ~21–23 nucleotide small RNAs processed by Dicer and loaded into Argonaute, so they share the same core machinery, but they differ in origin and behavior. siRNAs typically arise from long, near-perfectly double-stranded RNA — viral, transposon-derived, or experimentally introduced — and pair almost perfectly with one specific target, usually triggering direct endonucleolytic cleavage. miRNAs are encoded by the organism's own genome as imperfect hairpin transcripts processed through Drosha then Dicer, and they pair only partially (mainly via a short seed sequence) with typically hundreds of different target mRNAs, usually causing translational repression and mRNA decay rather than cleavage. Think of siRNAs as targeted antiviral/antitransposon strikes and miRNAs as broad regulatory rheostats.
What exactly is the "seed sequence" and why does it matter so much?
The seed sequence is nucleotides 2–8 (counting from the 5′ end) of a mature miRNA guide strand. Structural studies show Argonaute holds this short stretch in a pre-organized, exposed conformation, ready to pair with a target before the rest of the guide even engages — making seed complementarity the dominant, rate-limiting determinant of whether a given mRNA is recognized as a target. Because a 7-nucleotide sequence is short enough to occur by chance many times across a transcriptome, a single miRNA can have hundreds of predicted seed-matched targets, most repressed only modestly. This is also why off-target effects are a serious concern in siRNA experiments: any siRNA's seed can behave like a miRNA against unintended genes.
Why don't long double-stranded RNAs simply trigger RNAi in mammalian cells the way they do in worms and plants?
Mammalian cells evolved an additional, more ancient antiviral response to long dsRNA (roughly 30+ base pairs): activation of protein kinase R and the 2′-5′-oligoadenylate synthetase/RNase L pathway, which shuts down translation globally and can trigger interferon responses and cell death, masking any gene-specific silencing. This is why Elbashir and Tuschl's demonstration that short, ~21-nucleotide synthetic siRNA duplexes evade this length-dependent innate immune trigger while still loading efficiently into RISC was such a pivotal technical advance — it made sequence-specific RNAi usable in mammalian, including human, cells.
How does the strand-asymmetry rule decide which strand of a diced duplex becomes the functional guide?
After Dicer produces a small RNA duplex, Argonaute (often assisted by loading chaperones) senses the relative thermodynamic stability of the two duplex ends. The strand whose 5′ end sits at the less stably base-paired end is preferentially retained as the guide strand and threaded through Argonaute's active site; the other strand, the passenger, is either cleaved by Argonaute's slicer domain (if pairing is near-perfect) or unwound and degraded (if it is not). This is not strict or absolute — a "passenger" strand can occasionally be retained instead and can then cause unintended off-target silencing, which is one reason chemically modified siRNAs are engineered to bias selection strongly toward the intended guide strand.
Can RNAi actually spread between cells and across generations, and does this happen in humans?
Yes, but this is best documented in C. elegans and plants, not confirmed in mammals. In worms, dsRNA-derived silencing signals move systemically between cells through a dedicated dsRNA-import channel protein (SID-1) and can be inherited transgenerationally, amplified along the way by RNA-dependent RNA polymerases that synthesize secondary siRNAs from the primary cleavage products. Plants similarly propagate silencing signals through the vasculature as part of systemic antiviral defense. Mammals lack RNA-dependent RNA polymerases and a confirmed SID-1-like systemic spread mechanism, so mammalian RNAi is generally understood to be cell-autonomous, which is one reason therapeutic siRNAs must be delivered directly to target tissues (e.g., via lipid nanoparticles or GalNAc conjugates to liver cells) rather than relying on natural systemic spread.
How is a microRNA distinguished experimentally from a random hairpin-forming genomic sequence?
Establishing a genuine miRNA requires several converging lines of evidence: detection of a mature ~21–23 nt RNA by Northern blot or small RNA sequencing at the position predicted from a hairpin precursor; dependence of that mature RNA's production on Drosha/DGCR8 and Dicer (loss of either enzyme should eliminate or drastically reduce the mature RNA); evolutionary conservation of the hairpin structure and especially the seed sequence across related species, which argues against a coincidental fold; and functional repression of a predicted seed-matched reporter construct that is lost when the seed-complementary site is mutated. Databases such as miRBase apply these criteria (with substantial revision and pruning over the years) to distinguish confidently annotated miRNAs from spurious small RNA reads.
Why is RNA interference described as connecting to gene regulation more broadly, rather than being a separate, isolated pathway?
RNAi and miRNA silencing act as a post-transcriptional layer of control that operates downstream of, and in combination with, transcriptional regulation covered under Principles of Gene Regulation. A gene's expression level is jointly set by how much mRNA is transcribed (transcription factors, chromatin state) and by how efficiently that mRNA is translated and how long it survives (miRNAs, RNA-binding proteins, and general Regulation of Translation machinery). Many developmental switches use both layers redundantly — a transcription factor network establishes a expression boundary, and a miRNA sharpens or buffers it post-transcriptionally, filtering out noisy expression in cells where the gene should be fully off. This layering is also conceptually continuous with Long Non-Coding RNAs, which use RNA sequence and structure for regulation but typically act on chromatin or transcription rather than through Argonaute-mediated slicing.