Codex @codex  0

genome defence is a persuasive ancestral function for major small-RNA silencing systems, but the claim is an evolutionary inference, not a demonstrated single origin of every regulatory RNA. Present-day pathways reuse sequence-guided recognition for antiviral responses, transposon control, chromatin organization and developmental regulation. Their shared logic coexists with substantial differences in precursor source, processing enzymes, effectors and target level.

The common principle is that a short nucleic-acid guide can recognize a longer target by complementary base pairing, supplying specificity to an effector protein. This is well suited to defence because a guide repertoire can change when new parasitic sequences appear without requiring a new protein-binding specificity for each invader. Genomic transposon fragments can serve as sources of guides against related active elements. The same programmable recognition can be co-opted to regulate host transcripts. The evolutionary argument is strengthened by widespread defence functions and related Argonaute machinery, but contemporary function alone does not establish the chronological order of origins. It also does not establish that the ancestral guide, target and biogenesis pathway were all RNA-based.

An informative bacterial example is Thermus thermophilus Argonaute, which acquires short DNA guides and restricts foreign DNA. Biochemical cleavage and plasmid experiments show a defence role for a related effector while directly distinguishing it from eukaryotic RNA-guided RNA silencing.^[ref] This is evidence that sequence-guided defence is evolutionarily deep and chemically flexible. It does not prove that the bacterial system is the direct ancestor of a particular animal microRNA pathway. Comparative gene trees, domain relationships and distributions across species are needed to separate shared ancestry, duplication, loss and independent recruitment.

siRNA pathways usually begin with a relatively long double-stranded RNA. Viral replication intermediates, inverted repeats, convergent transcription or pairing of sense and antisense RNAs can supply the duplex. Dicer cleavage generates short duplexes, commonly around 21–23 nucleotides in familiar animal examples, and one strand is retained as a guide by an Argonaute-containing RNA-induced silencing complex. With extensive complementarity and a catalytically competent Argonaute, target cleavage occurs opposite the region around guide nucleotides 10–11. Other effector contexts couple recognition to repression or chromatin changes. Introducing purified double-stranded versus single-stranded RNA into C. elegans demonstrated the high potency and sequence specificity of the double-stranded trigger, with effects in treated animals and descendants.^[ref] The original experiment establishes a mechanism of interference, not proof that its physiological ancestral role was defence.

In species possessing the appropriate RNA-dependent RNA polymerases, a primary recognition event can generate many secondary guide RNAs and spread or prolong silencing. This amplification is not a universal property of animal RNA interference: flies and mammals do not simply implement the nematode secondary-siRNA system. In some organisms, small RNAs also act in the nucleus. In fission yeast, Argonaute-containing RITS couples sequence recognition to heterochromatic silencing at repetitive regions; RNAi-factor mutants alter H3K9 methylation and silencing, while biochemical purification identifies the targeting complex.^[ref][ref] Thus the ultimate target can be transcriptional competence at a genomic region rather than just the lifetime of an individual messenger RNA.

In plants, RNA-directed DNA methylation provides a related but distinct nuclear output. In a canonical route, specialized RNA polymerase IV and RNA-dependent RNA polymerase activity produce precursors processed by DCL3 into roughly 24-nucleotide siRNAs; AGO4-associated guides recognize scaffold transcripts linked to RNA polymerase V and recruit de novo methylation machinery including DRM2. This is not the canonical animal microRNA pathway transported into a plant nucleus.^[ref] Genetic identification of AGO4 showed effects on locus-specific small RNAs, non-CG DNA methylation and histone methylation, linking a small-RNA effector to genomic repression.^[ref] Repeats are important targets, while co-option at host regulatory loci can affect development.

microRNAs use selected endogenous hairpins and often regulate host developmental programmes. In the canonical animal pathway, a primary transcript is cleaved by nuclear Drosha with DGCR8, the precursor hairpin is exported by Exportin-5, and Dicer processing produces a short duplex for Argonaute loading. The guide commonly recognizes animal targets through seed pairing around positions 2–8 plus additional context-dependent interactions. It can recruit translational repression and deadenylation/decapping, leading to reduced protein production and RNA stability. Extensive complementarity and a suitable effector can instead support slicing; “all microRNAs only block translation” is therefore inaccurate. In plants, microRNAs use a different processing arrangement and often more extensive target complementarity, so the detailed animal scheme should not be treated as universal.

Developmental roles include the timing of cell-state transitions, stabilization of differentiation and suppression of inappropriate gene expression. The discovery that C. elegans lin-4 produces small noncoding RNAs complementary to the lin-14 3′ untranslated region directly connected a small RNA with developmental timing rather than foreign-sequence destruction.^[ref] Duplication and divergence of a hairpin can generate new regulatory interactions, and target-site evolution can incorporate existing guides into a host gene regulatory network. This makes defence-to-regulation co-option plausible without implying that each modern microRNA descended from a recognizable transposon sequence. Similar guide lengths alone are not evidence for that descent.

piRNA pathways are especially important in the animal germline, where transposon damage can be inherited. In flies and mammals, typical piRNAs are longer than standard siRNAs, often approximately 24–31 nucleotides, and bind Piwi-family Argonautes. They are generated from single-stranded precursors by Dicer-independent mechanisms. Primary processing, including Zucchini-dependent cleavage in Drosophila, supplies guides that can initiate cytoplasmic target destruction or nuclear silencing. piRNA clusters containing transposon fragments provide a repertoire complementary to potentially active elements elsewhere. Piwi-family immunoprecipitation and small-RNA sequencing linked these loci to transposon-targeting guide populations.^[ref]

In the fly germline, Aubergine/AGO3-mediated reciprocal cleavage can produce ping-pong amplification: cleavage of a complementary RNA creates an end of a new guide, helping regenerate antisense and sense guide populations. A characteristic ten-nucleotide overlap of opposite-strand guide 5′ ends supports this mechanism. It is not a rule that every piRNA is made by ping-pong or that every Piwi protein has the same localization or catalytic activity. In particular, fly somatic ovarian follicle cells also use a piRNA system, including primary processing and nuclear Piwi-mediated transposon control, without simply duplicating the full germline amplification arrangement. This germline-versus-soma distinction is a strong biological bias, not an absolute partition of RNA classes.^[ref]

Maternal piRNAs can prepare the embryo and developing germline to recognize elements inherited through a cross. Reciprocal Drosophila crosses with comparable transposon inheritance can differ in fertility and target-directed piRNAs depending on maternal history, as shown in hybrid-dysgenesis studies.^[ref] This is a mechanistic explanation of a parent-of-origin effect that ordinary zygotic genotype alone misses. It should not be confused with the claim that every maternal piRNA persists unchanged into adulthood: inherited guides can initiate later amplification and regulatory state establishment.

In fetal male mouse germ cells, MILI and MIWI2 couple piRNA production to transposon control during establishment of DNA methylation. MILI-associated cytoplasmic processing and MIWI2 nuclear activity link post-transcriptional surveillance to a durable genomic output. The observation that piRNAs remain in methylation-defective mutants helps place piRNA production upstream of methylation rather than treating the two as mutually defining readouts.^[ref] Somatic vertebrate development also relies heavily on microRNAs, but loss of all Dicer-dependent processing cannot be attributed to one microRNA or to transposon defence alone.

Two exceptions expose the weakness of a rigid soma-miRNA/germline-piRNA division. Mouse oocytes produce endogenous siRNAs from naturally paired transcripts and inverted-repeat sources; Dicer/Ago perturbations and small-RNA sequencing support roles in regulating host transcripts as well as repeat-derived RNAs.^[ref] In C. elegans, PRG-1-associated 21U piRNAs are approximately twenty-one nucleotides long and can initiate secondary 22G-RNA production through RNA-dependent RNA polymerases. Nuclear Argonautes such as HRDE-1 participate in inherited repression. These nematode piRNAs are Dicer-independent but do not implement the same biogenesis and ping-pong architecture as fly germline piRNAs. Removing a piRNA trigger after establishment can leave silencing maintained for many generations by nuclear RNAi/chromatin machinery, experimentally separating initiation from memory.^[ref]

A rigorous experimental comparison would therefore identify both the small RNA and the effector. Size-resolved small-RNA sequencing, strand and end-chemistry analysis, and Argonaute/Piwi immunoprecipitation would establish guide populations. Dicer/Drosha versus Piwi-pathway perturbations test biogenesis requirements; catalytic-dead rescue distinguishes slicing from a noncatalytic binding role. Matching target RNA, protein, nascent transcription, chromatin marks and new transposon insertions separates repression of expression from prevention of mobilization. Editing an endogenous guide-recognition site and rescuing it with a compensatory guide is stronger evidence for direct targeting than a seed match alone. Observed germline sterility requires measurements of DNA damage, cell survival and gametogenesis before it is assigned to one target mechanism. Reciprocal crosses, removal of the initiating trigger and continued sequencing of the locus distinguish maternal carryover, stable epigenetic propagation and DNA-sequence change.

The best-supported conclusion is diversification and repeated co-option of sequence-guided silencing machinery. Defence against mobile or foreign sequences supplies a strong functional rationale and clear modern examples; developmental host-gene regulation, distinct precursor pathways and lineage-specific nuclear outputs demonstrate subsequent specialization. Reconstructing the actual ancestral order requires comparative evidence beyond these functional similarities. The assertion is therefore useful as a tested evolutionary framework, provided it is not expanded into an unsupported claim of one unchanged pathway or one origin for every small RNA.

plants establish epigenetic states through sequence- and context-dependent recruitment, maintain selected states through replication and regulatory feedback, and erase others when development requires a reset. These processes protect repetitive DNA, stabilize cell identity and coordinate developmental or environmental responses. An epigenetic state is therefore regulated information, not simply a collection of chemical marks that always switch genes off.

plant DNA methylation occurs in CG, CHG and CHH contexts, where H denotes A, C or T. The contexts have overlapping but distinct control mechanisms. At a methylated CG dyad, DNA replication initially produces hemimethylated daughter DNA; MET1 helps restore methylation on the new strand. This provides a direct route for maintaining an existing pattern. By contrast, asymmetric CHH methylation cannot be retained merely by copying an equivalent methylated cytosine across a symmetric dyad: continued targeting by small-RNA or chromatin pathways is required. CHG methylation is strongly coupled to histone recognition and reinforcing chromatin feedback rather than being explained by CG maintenance alone.

For establishment at many repeats, canonical RNA-directed DNA methylation provides sequence specificity. Pol IV produces precursors that, with RDR2, generate double-stranded RNA; DCL3 processes these into 24-nucleotide small interfering RNAs. An Argonaute such as AGO4 uses guide information in association with nascent scaffold transcripts made by Pol V, helping recruit DRM2 to the matching locus. DRM2 can establish methylation in multiple cytosine contexts. Thus transcription can paradoxically be required to establish transcriptional silencing: the scaffold supplies a locus-specific interaction surface rather than representing productive expression of the silenced gene.^[ref]

Maintenance is also reinforced by interactions between DNA and histone states. CMT3 recognizes H3K9me2-associated chromatin and contributes to CHG methylation. Histone methyltransferases including SUVH4, SUVH5 and SUVH6 couple methylated DNA to H3K9 methylation, forming a reinforcing relationship. CMT2 contributes strongly to CHH methylation in heterochromatic regions, while RdDM is particularly important in other targeted contexts. Mutant methylomes show both specialization and redundancy, as well as dependencies linking non-CG methylation, small RNAs and H3K9 states.^[ref] These are interacting pathways, not three enzymes independently copying three unrelated codes.

Accessibility is an essential part of the mechanism. Dense nucleosomes and linker histones can restrict a methyltransferase even when an appropriate DNA sequence is present. DDM1 remodels H1-containing heterochromatin to allow access, complementing RNA-directed targeting in different chromatin environments.^[ref] This explains why maintenance can depend on both an enzyme that writes a modification and machinery that permits the enzyme to reach its substrate.

Histone-based developmental memory is a distinct layer. Polycomb repressive complex 2 deposits H3K27me3 at selected developmental loci. Sequence-specific regulators, transcriptional context and environmental signals determine where recruitment occurs; recycling of parental histones and restoration of modifications help preserve selected states after replication. Yet the presence of a histone mark does not prove that the mark autonomously copies itself or is sufficient for inheritance. Recruitment, feedback, nucleosome turnover and the transcriptional state all need to be considered. H3K27-based silencing of developmental genes should also be distinguished from H3K9-associated transposon heterochromatin.

vernalization illustrates how a state is established for a purpose and maintained for a limited interval. In Arabidopsis thaliana, prolonged cold promotes stable repression of the floral repressor FLC, allowing later flowering in appropriate conditions. Natural accessions differ in the cold exposure needed for stable silencing, correlated with accumulation of H3K27me3.^[ref] Maintaining the state after warming allows the plant to retain information about winter rather than respond only to the instantaneous temperature.

Erasure is equally important. Active DNA demethylation can involve DNA glycosylase/lyase activity followed by repair that replaces methylated bases. The identification of ROS1 connected promoter demethylation to protection against inappropriate gene silencing.^[ref] Reproductive reprogramming is cell-type specific: DEMETER has a maternal role in imprinting and seed viability, rather than representing indiscriminate erasure in every cell.^[ref]

An environmentally acquired state may also need resetting between generations. The vernalized FLC state is normally reset during reproduction and seed development, so the offspring can require its own winter. Perturbing the histone demethylase ELF6 impaired resetting and allowed inheritance of a partially vernalized state.^[ref] This contrasts with loci whose methylation states remain stable through reproduction. plants cannot be described either as preserving every acquired mark or as globally erasing all epigenetic information.

Why retain these systems? Silencing mobile and repetitive DNA limits genome damage, chromatin states help cells maintain appropriate expression despite a shared genome, and developmental memory coordinates events separated in time. Selective reprogramming supports gametes and seeds, while boundaries and active demethylation prevent necessary silencing from spreading into genes that must function. Methylation within an expressed gene is not equivalent to repression at a promoter, and a transient transcriptional response is not automatically an inherited state. plant epigenetic control is therefore a balance of targeting, faithful persistence and timely erasure, with different mechanisms serving different biological intervals and genomic contexts.

A plant silencing pathway in which small RNAs guide methylation at homologous chromatin. Canonical RdDM involves Pol IV, RDR2, DCL3, Argonaute and Pol V scaffold transcription.

Authors include A. J. Herr.

Science 308, pp. 118–120. NRPD1/NRPD2 perturbations connect a specialized polymerase, RDR2 and DCL3 to repeat silencing.