Gene Regulation and Epigenetics

Every cell in your body carries the same genome, yet a neuron and a liver cell behave like different organisms. The difference lies in which genes they use, when, and how much. This chapter explains the switches, dials and memory systems that decide gene expression, from bacterial operons to the chromatin marks and small RNAs that shape development and disease.

Section 1Principles of gene regulation

Why regulate genes at all?

Making RNA and protein is expensive. A bacterium that made all of its roughly 4,300 proteins at full speed all the time would waste energy on enzymes it does not need. Multicellular organisms face a second problem: they must build many cell types from one genome and keep each type stable through many divisions. Gene regulation solves both problems.

Genes expressed at a fairly steady level in nearly every cell, such as those for ribosomal proteins and glycolytic enzymes, are called housekeeping genes, and their expression is constitutive. Other genes are regulated: switched on or off, or turned up or down, in response to signals.

The levels of control

Gene expression is a pipeline, and a cell can intervene at any stage. In eukaryotes the main control points are:

1. Chromatin accessibility: tightly packed nucleosomes and heterochromatin hide DNA.
2. Transcription initiation: the most common and most economical control point.
3. RNA processing: alternative splicing and polyadenylation change which protein is made.
4. RNA export and localization.
5. mRNA stability: half lives range from minutes to days, set by 3′ UTR sequences, RNA binding proteins and microRNAs.
6. Translation: initiation factors, regulatory proteins and small RNAs control ribosome loading.
7. Protein modification, localization and degradation, for example phosphorylation or ubiquitin tagging for the proteasome.

In bacteria, where transcription and translation are coupled, most regulation acts on transcription initiation, transcription termination and translation initiation. Bacterial mRNAs often last only 1 to 3 minutes, so protein supply follows transcription closely.

Cis-acting elements and trans-acting factors

A cis-acting element is a DNA (or RNA) sequence that controls only the gene on the same molecule: promoters, operators, enhancers, silencers and 3′ UTR sequences are binding sites, not products. A trans-acting factor is a diffusible product, usually a protein or RNA, that can act on targets anywhere in the cell, including on another chromosome. Repressors, activators, sigma factors and microRNAs are trans-acting.

The distinction is testable in a partial diploid (merodiploid) carrying two copies of a regulatory region, one on the chromosome and one on a plasmid. A defective trans-acting factor is usually rescued by a wild type copy elsewhere, because the wild type product diffuses to both targets. A defective cis-acting site is not rescued; it affects only the genes physically linked to it.

Positive and negative control; induction and repression

Regulatory systems are classified along two independent axes.

• Negative versus positive control asks what the regulatory protein does when it binds DNA. In negative control a repressor binds and blocks transcription; the default without the protein is “on.” In positive control an activator binds and stimulates transcription; the default without the protein is “off.”
• Inducible versus repressible asks how a small signal molecule changes expression. An inducible system is normally off and is turned on by an inducer, typically a nutrient the pathway breaks down. A repressible system is normally on and is turned off by a corepressor, typically the end product of a biosynthetic pathway.

The logic fits physiology. Catabolic enzymes are worth making only when their food is present, so catabolic operons are usually inducible; biosynthetic enzymes are worth making only when their product is scarce, so biosynthetic operons are usually repressible. The signal usually works allosterically, binding the regulatory protein away from its DNA binding domain and changing its DNA affinity.

Operons, regulons and global control

Bacteria often cluster genes for one pathway into an operon: a set of genes transcribed from a single promoter into one polycistronic mRNA and controlled by shared regulatory sites. The lac operon carries three genes for lactose use, and the trp operon carries five genes for tryptophan synthesis. Operons guarantee that enzymes needed together are made together.

A single regulator can control many operons scattered around the chromosome, a set called a regulon. Alternative sigma factors act globally: in E. coli, σ70 directs RNA polymerase to most housekeeping promoters, while heat shock raises σ32, which switches on chaperones and proteases. Some mRNAs regulate themselves: a riboswitch is a structured 5′ UTR segment that binds a metabolite directly and changes shape to end transcription early or hide the ribosome binding site, so regulation does not always require a protein.

How fast can expression change? Synthesis, decay and steady state

The amount of an mRNA or protein reflects a balance between synthesis and destruction. If a molecule is made at a constant rate s and decays with first order kinetics, the decay constant is k = ln 2 ÷ t_1/2. At steady state synthesis equals decay, so N = s ÷ k. Doubling the half life therefore raises the level as much as doubling transcription does. The time to reach a new steady state depends only on the half life: halfway in one half life, 75 percent in two, and about 97 percent in five. Short lived mRNAs let cells respond quickly.

Section 2The lac and trp operons

The lac operon: genes and sites

The lactose operon of E. coli, worked out by François Jacob and Jacques Monod around 1960, was the first gene regulatory system understood in molecular detail. Lactose is a disaccharide of galactose and glucose. The operon carries three genes in the order lacZ, lacY, lacA:

• lacZ encodes β-galactosidase, which cleaves lactose into glucose and galactose. As a side reaction it also rearranges a small fraction of lactose into allolactose (a β-1,6 isomer), the natural inducer.
• lacY encodes lactose permease, a membrane symporter that brings lactose into the cell with a proton.
• lacA encodes thiogalactoside transacetylase, which is not required for lactose use.

Upstream of the structural genes lie the cis-acting sites. The promoter (P) binds RNA polymerase. The main operator (O1) overlaps the transcription start, so a bound repressor blocks polymerase; auxiliary operators O2 (in lacZ) and O3 (upstream) let one repressor tetramer bind two operators and loop the DNA, tightening repression. Just upstream of the promoter is the site for the catabolite activator protein (CAP). The nearby lacI gene, with its own constitutive promoter, encodes the tetrameric lac repressor.

Negative control by the repressor

Without lactose, the repressor binds the operator and blocks transcription. Repression is not absolute: a few molecules of permease and β-galactosidase escape, which the cell needs to import the first lactose and make the first allolactose. Allolactose binds the repressor and lowers its affinity for the operator, so it falls off and expression rises roughly a thousandfold. When lactose runs out, the repressor rebinds and the short lived mRNA is degraded within minutes.

Researchers often induce the operon with IPTG, a gratuitous inducer that binds the repressor like allolactose but is not broken down, so its concentration stays constant. X-gal, a colorless galactoside that β-galactosidase cleaves to a blue product, shows whether lacZ is active.

Positive control by CAP and cAMP

With both glucose and lactose present, E. coli uses glucose first and expresses lac only weakly, called catabolite repression. Two mechanisms produce it.

• Activation by CAP. The lac promoter is weak and is strongly transcribed only when CAP bound to cyclic AMP (cAMP) sits upstream, bends the DNA and contacts RNA polymerase. Adenylate cyclase makes cAMP when glucose is scarce. With plenty of glucose, cAMP is low, CAP does not bind, and lac transcription stays low even with lactose present.
• Inducer exclusion. While glucose enters through the phosphotransferase system, its component EIIA^Glc is mostly unphosphorylated and in that form inhibits lactose permease, so less inducer forms. This is thought to be the larger contributor, with low cAMP adding to it.

The result is a logic gate. High transcription requires two conditions at once: lactose present (repressor off) and glucose absent (CAP-cAMP on, permease active).

Regulatory mutations and the partial diploid test

Jacob and Monod deduced the logic of the operon from mutants, which remain the basis of many problems.

To solve a partial diploid problem, treat the cell as a pool of diffusible products and a set of fixed DNA molecules. Repressor made from either copy of lacI goes everywhere. Each operator and promoter controls only the structural genes on its own DNA. Work one structural gene at a time: find every functional copy of that gene, then ask whether the operator and promoter next to it will allow expression with and without inducer.

The trp operon: a repressible system

The tryptophan operon contains five genes, trpE, trpD, trpC, trpB, trpA, encoding the enzymes that make tryptophan. The trp repressor, encoded by the separate trpR gene, binds DNA poorly on its own. Tryptophan acts as a corepressor: it binds the repressor dimer and repositions its helix-turn-helix motifs to fit the operator, blocking transcription. When tryptophan is scarce, the repressor drops off. Repression alone changes expression about 70 fold.

Attenuation: fine tuning by coupled transcription and translation

Charles Yanofsky found that even trpR mutants lowered operon expression about eightfold when tryptophan was plentiful. The explanation is attenuation, premature termination controlled by translation of a 162 nucleotide leader (trpL) between the operator and trpE. Its RNA has four regions that can pair in alternative ways:

• Region 1 can pair with region 2.
• Region 2 can pair with region 3, forming the antiterminator.
• Region 3 can pair with region 4, forming the terminator (the attenuator): a GC rich hairpin followed by a run of U's, a classic intrinsic terminator.

Because region 2 cannot pair with both 1 and 3 at once, and region 3 cannot pair with both 2 and 4, the structures are mutually exclusive. Region 1 also encodes a 14 amino acid leader peptide that contains two tryptophan codons in a row. Because bacteria translate mRNA while it is still being made, the position of the ribosome on the leader decides which hairpin forms.

• Tryptophan plentiful. Charged tRNA^Trp is abundant, so the ribosome reads through the two Trp codons without pause and continues to the leader stop codon, where it covers region 2. Region 2 is unavailable, so when regions 3 and 4 are made they pair into the terminator. Polymerase stops before reaching trpE.
• Tryptophan scarce. Charged tRNA^Trp is rare, so the ribosome stalls at the Trp codons, sitting on region 1. Region 2 is free to pair with region 3 as soon as region 3 is made. The 2:3 antiterminator prevents the 3:4 terminator, and polymerase transcribes the structural genes.
• No translation at all (for example, if the leader start codon is mutated). Region 1 pairs with region 2, leaving 3 free to pair with 4. Termination occurs.

Repression (about 70 fold) and attenuation (about 8 to 10 fold) multiply to give a total range of roughly 600 fold. Attenuation senses charged tRNA, while repression senses free tryptophan. Leaders rich in the relevant codons also control the histidine, leucine and phenylalanine operons. Attenuation cannot work in eukaryotes, where transcription and translation are separated by the nuclear envelope.

Section 3Eukaryotic transcription factors and enhancers

The parts of a eukaryotic gene's control region

A typical protein coding gene in a eukaryote is transcribed by RNA polymerase II, which cannot recognize a promoter by itself. Its control region has three layers.

• The core promoter, within about 40 base pairs of the start site, is where the transcription machinery assembles. Common elements include the TATA box (consensus TATAAA, about 25 to 30 bp upstream), the initiator (Inr) spanning the start site, and the downstream promoter element (DPE) about 30 bp downstream. Many human promoters lack a TATA box and instead lie in CpG islands.
• Promoter proximal elements, usually within about 200 bp upstream, such as the CAAT box (bound by NF-Y) and GC boxes (bound by Sp1), raise the efficiency of initiation.
• Enhancers and silencers are distal elements that can lie upstream, downstream, within introns, or tens to hundreds of kilobases away.

The general transcription factors assemble at the core promoter to form the preinitiation complex. TFIID, containing the TATA binding protein (TBP), binds first; TFIIA and TFIIB stabilize it, TFIIF brings RNA polymerase II, and TFIIE and TFIIH join last. TFIIH opens the DNA and phosphorylates the polymerase C-terminal domain (CTD), releasing polymerase into elongation. These factors give only low, basal transcription; high levels need activators.

Activators have separable domains

A eukaryotic activator usually has at least two functional parts that can be separated and swapped like modules:

• A DNA binding domain recognizes a specific sequence of roughly 6 to 12 bp, usually by inserting an α helix into the major groove.
• An activation domain recruits the machinery that turns on transcription. Many activation domains are acidic or rich in glutamine or proline and are intrinsically disordered until they bind a partner.
• Many activators also have a dimerization domain and a domain that binds a regulatory ligand.

Modularity was shown by domain swaps with the yeast activator GAL4, which binds the UAS_G near the galactose genes. The GAL4 DNA binding domain fused to another protein's activation domain still activates UAS_G genes; the DNA binding domain alone binds but activates nothing. This modularity underlies the yeast two hybrid screen and the GAL4/UAS system used in fruit flies.

Without galactose, GAL4 is already bound to the UAS, but the protein Gal80 masks its activation domain. Galactose binds the sensor Gal3, which pulls Gal80 away. The signal acts on the activator, not on DNA binding.

DNA binding motifs

Enhancers and how they work at a distance

An enhancer is a cluster of binding sites for activators, usually 100 to 1,000 bp long, that raises transcription of a target gene. Enhancers have three defining properties that distinguish them from promoter elements:

1. They work at a distance, from a few hundred bases to more than a megabase away.
2. They work in either orientation; flipping an enhancer usually leaves its activity intact.
3. They work upstream, downstream or within introns of the gene they control.

Enhancers act by DNA looping. Activators at the enhancer contact the promoter through Mediator, a coactivator complex of about 25 to 30 subunits that bridges activators and RNA polymerase II. Activators also recruit histone acetyltransferases such as CBP and p300 and ATP dependent remodeling complexes such as SWI/SNF that slide or evict nucleosomes. Many genes have several enhancers, each active in a different tissue, so a mutation in one enhancer can alter one tissue alone. Point mutations in the ZRS enhancer, about a megabase from SHH, cause extra digits without other effects of SHH loss.

Silencers are the mirror image: binding sites for repressors. Eukaryotic repressors work in several ways: they compete with activators for overlapping sites, they mask activation domains (as Gal80 does), they block contact with Mediator, or they recruit corepressors that bring in histone deacetylases and methyltransferases.

Insulators and topological domains

What stops an enhancer from activating the wrong gene? Insulators (boundary elements) block enhancer-promoter communication when placed between them and stop the spread of heterochromatin. In vertebrates most are bound by the zinc finger protein CTCF, which with the ring shaped cohesin complex organizes the genome into topologically associating domains (TADs) of roughly 100 kb to 1 Mb. Deleting a TAD boundary can let an enhancer reach a gene it normally never touches, a cause of some human limb malformations.

Combinatorial control and signals

Humans have roughly 1,600 sequence specific transcription factors, far fewer than the number of expression patterns cells display. The answer is combinatorial control: each enhancer is active only when the right combination of factors is present. A small set of master regulators can therefore specify many cell types; four factors (Oct4, Sox2, Klf4 and c-Myc) can even reprogram skin fibroblasts into induced pluripotent stem cells.

Signals reach transcription factors in several ways. Lipid soluble steroid hormones enter the cell: cortisol releases the glucocorticoid receptor from the chaperone Hsp90, and the receptor moves to the nucleus, dimerizes and binds hormone response elements. Other signals act through kinases; protein kinase A phosphorylates CREB, letting it recruit CBP/p300. The signal changes the activity, location or partners of a factor, not the DNA.

Reporter genes and the dissection of enhancers

Enhancers are mapped with reporter genes. A candidate regulatory region is cloned upstream of a gene whose product is easy to measure, such as luciferase (light) or green fluorescent protein (fluorescence), and the construct is introduced into cells. Deleting or mutating pieces of the region and measuring the reporter reveals which pieces matter. Moving a fragment to new positions or inverting it tests whether it behaves as an enhancer.

Section 4Chromatin, DNA methylation and epigenetics

Nucleosomes as gatekeepers

In the nucleus, about 147 bp of DNA wraps around each nucleosome, an octamer with two copies each of histones H2A, H2B, H3 and H4, and histone H1 binds the linker. Nucleosomes block many factors from reaching DNA. Euchromatin is relatively open and contains most active genes; heterochromatin is condensed and largely silent. Constitutive heterochromatin at centromeres stays condensed in all cells, while facultative heterochromatin, such as the inactive X, is condensed only in some.

Cells change chromatin in two ways. ATP dependent chromatin remodelers (SWI/SNF and related families) use ATP to slide nucleosomes along DNA, eject them or exchange histone variants. Histone modifying enzymes add or remove chemical groups on the flexible N-terminal tails of histones, which protrude from the nucleosome.

Histone modifications

The best studied modifications are acetylation and methylation of lysines. Histone acetyltransferases (HATs) add acetyl groups to lysines, which neutralizes their positive charge, loosens the tails' grip on negatively charged DNA and creates docking sites for proteins with bromodomains. Histone deacetylases (HDACs) remove them. Acetylation is therefore a mark of active chromatin. Methylation does not change charge; instead, the effect of a methyl mark depends on which residue carries it and how many methyl groups (one, two or three) are present, because different reader proteins recognize each one. Proteins with chromodomains bind methylated lysines. The idea that combinations of marks are read by specific proteins is called the histone code hypothesis.

Repressive marks can spread: HP1 binds H3K9me3 and recruits more SUV39H1, which methylates the next nucleosome. This explains position effect variegation in fruit flies. When an inversion moves the white gene next to centromeric heterochromatin, silencing spreads into it in some cells but not others, and each cell's state is inherited, giving a mottled eye. Polycomb and trithorax complexes form a similar memory system that keeps Hox genes off or on in the segments where they were first set.

DNA methylation

In mammals the major DNA modification is 5-methylcytosine, made by adding a methyl group to carbon 5 of cytosine, almost always where the C is followed by a G (a CpG dinucleotide; the p denotes the phosphate between them). Because CpG is palindromic, a methylated site has methyl groups on both strands. About 70 to 80 percent of CpGs in the human genome are methylated. The exceptions are CpG islands, stretches of about 0.5 to 2 kb rich in G and C and in CpGs, found at roughly 60 to 70 percent of gene promoters and usually unmethylated. Methylation of a promoter CpG island silences the gene, both by blocking binding of some activators and by recruiting methyl-CpG binding proteins such as MeCP2, which bring in HDACs and H3K9 methyltransferases.

Three enzymes establish and maintain the pattern:

• DNMT3A and DNMT3B are de novo methyltransferases that place new methyl groups on unmethylated DNA, especially during early embryonic development and gamete formation.
• DNMT1 is the maintenance methyltransferase. After replication, each CpG is hemimethylated: the parental strand is methylated and the new strand is not. The partner protein UHRF1 recognizes hemimethylated sites and recruits DNMT1, which methylates the new strand. This copying mechanism is why methylation patterns are inherited through cell division.
• TET enzymes oxidize 5-methylcytosine to 5-hydroxymethylcytosine and further products, which are removed by repair or diluted by replication, giving active demethylation. Failure of DNMT1 maintenance during replication gives passive demethylation.

Methylated cytosines are mutation hot spots. Deamination of cytosine gives uracil, which repair recognizes as foreign, but deamination of 5-methylcytosine gives thymine, a normal base, leaving a G-T mismatch that is often repaired the wrong way. This is why CpGs are depleted to about one quarter of their expected frequency and why C-to-T changes at CpGs are common in human disease.

What makes a mark epigenetic?

In its strict sense, epigenetics means changes in gene function that are heritable through cell division without a change in DNA sequence. DNA methylation qualifies because of DNMT1 copying, and some histone marks qualify through reader-writer loops (HP1 with SUV39H1, PRC2 binding its own H3K27me3). Most acetylation turns over in minutes and reflects current activity rather than memory.

Genomic imprinting

For most genes, the maternal and paternal copies are expressed equally. For about 100 to 200 mammalian genes, one copy is silenced depending on which parent it came from. This genomic imprinting is set by methylation at imprinting control regions (ICRs) during gamete formation, maintained through all somatic divisions of the offspring, then erased and reset in the offspring's own germ cells according to that individual's sex.

At the Igf2-H19 locus, Igf2 (a fetal growth factor) is expressed from the paternal allele and H19 (a noncoding RNA) from the maternal allele; both share enhancers downstream of H19. On the maternal chromosome the imprinting control region between them is unmethylated and binds CTCF, an insulator that blocks the enhancers from Igf2, so they activate H19. On the paternal chromosome the region is methylated, CTCF cannot bind, the enhancers reach Igf2, and H19 is silenced.

Imprinting explains two disorders of chromosome 15q11 to q13. Prader-Willi syndrome (neonatal hypotonia, later overeating and obesity, intellectual disability) results from loss of paternally expressed genes, by a paternal deletion or maternal uniparental disomy (both chromosome 15s from the mother). Angelman syndrome (severe intellectual disability, absent speech, seizures, happy demeanor) results from loss of the maternally expressed UBE3A gene in neurons, by a maternal deletion, paternal uniparental disomy or a UBE3A mutation. The same deletion gives different diseases depending on the parent of origin.

X inactivation

Mammals achieve dosage compensation by silencing all but one X in each somatic cell, as Mary Lyon proposed in 1961. Early in the embryo each cell chooses one X at random. That X expresses the long noncoding RNA Xist, which coats it in cis and recruits Polycomb complexes; the chromosome gains H3K27me3 and promoter methylation and condenses into a Barr body. The choice is inherited by the cell's descendants, making every female a mosaic. The number of Barr bodies is the number of X chromosomes minus one.

The orange and black patches of a calico cat show this mosaic: the two alleles of an X linked coat gene are expressed in different patches. Carriers of X linked disorders may show patchy signs, or symptoms if inactivation is skewed toward the normal X. About 15 percent of human X linked genes escape inactivation, which helps explain why Turner (45,X) and Klinefelter (47,XXY) syndromes have phenotypes.

Measuring methylation: bisulfite sequencing

Ordinary sequencing cannot distinguish C from 5-methylcytosine. In bisulfite sequencing, sodium bisulfite deaminates unmethylated cytosine to uracil, read as T after PCR, while 5-methylcytosine stays C. Non-CpG cytosines are rarely methylated in somatic cells and should all convert, a built in check on the treatment.

Epigenetics in development, environment and disease

Most methylation is erased and rebuilt twice in mammalian life, in the early embryo (sparing imprints) and in primordial germ cells (including imprints), which is why most environmentally induced marks are not passed on. A well studied exception is the agouti viable yellow (A^vy) mouse: a retrotransposon upstream of the agouti gene drives yellow coat and obesity when unmethylated, and feeding pregnant mothers methyl donors (folate, choline, vitamin B12) shifts offspring toward the methylated, brown, lean state.

Cancer cells show genome wide hypomethylation together with hypermethylation of CpG island promoters of tumor suppressor and repair genes; methylation silencing of MLH1 causes many sporadic colorectal cancers. Because these changes are reversible, DNMT inhibitors (azacitidine, decitabine) and HDAC inhibitors (vorinostat) are used as drugs. Mutations in the machinery itself cause disease: Rett syndrome results from mutations in MECP2, which reads methylated DNA.

Section 5Regulation by RNA: RNAi and microRNA

The discovery of RNA interference

In 1998 Andrew Fire and Craig Mello found that injecting C. elegans with double stranded RNA (dsRNA) matching a gene silenced it far more strongly than sense or antisense RNA, at only a few molecules per cell. They called this RNA interference (RNAi). It is an ancient defense against viruses and transposons, and its machinery is shared with endogenous regulators, the microRNAs. The first microRNA, lin-4, had been found in worms by Victor Ambros and Gary Ruvkun in 1993.

Shared machinery: Dicer, Argonaute and RISC

Small RNAs of 20 to 30 nucleotides guide proteins to complementary targets. Two enzymes are central:

• Dicer, an RNase III family enzyme, cuts dsRNA or hairpin RNA into duplexes about 21 to 23 bp long with 2 nucleotide overhangs at each 3′ end.
• Argonaute (AGO) proteins bind one strand of the duplex, the guide strand, and discard the other, the passenger strand. The Argonaute with its guide is the core of the RNA induced silencing complex (RISC). In humans, AGO2 has slicer activity that cuts a target paired along the full length of the guide.

siRNAs and microRNAs compared

Small interfering RNAs (siRNAs) come from long dsRNA, such as a viral replication intermediate, a transposon transcript read in both directions, or an experimenter's synthetic duplex. They usually pair perfectly with their target, and AGO slicing cuts the target in the middle of the paired region, after which the fragments are degraded.

MicroRNAs (miRNAs) come from the cell's own genes. Their biogenesis has several steps:

1. RNA polymerase II transcribes a long primary miRNA (pri-miRNA), capped and polyadenylated, containing one or more hairpins. Many miRNAs sit in introns of protein coding genes.
2. In the nucleus, the Microprocessor, made of the RNase III enzyme Drosha and its partner DGCR8, cuts out a hairpin of about 60 to 70 nucleotides, the precursor miRNA (pre-miRNA).
3. Exportin-5, powered by Ran-GTP, carries the pre-miRNA to the cytoplasm.
4. Dicer removes the loop, leaving a duplex of about 22 bp.
5. One strand is loaded into Argonaute to form the active miRISC.

In animals, a miRNA usually pairs only partly with its target, in the 3′ UTR. The key region is the seed, nucleotides 2 to 8 from the miRNA 5′ end. Without full pairing, AGO does not slice; instead miRISC recruits GW182, which brings in deadenylases, represses translation and leads to decay. The effect on one target is modest, but each miRNA has hundreds of targets, and more than 60 percent of human protein coding genes carry conserved sites. Plant miRNAs usually pair nearly perfectly and direct slicing.

Finding miRNA targets

Target prediction begins by finding 3′ UTR sequences complementary to seed nucleotides 2 to 8, written antiparallel as the reverse complement of the seed. Such a site is a 7mer; one that also has an A across from miRNA nucleotide 1 is an 8mer, the most effective. Predictions must be tested with reporters.

Biological roles and other regulatory RNAs

MicroRNAs often control developmental timing. lin-4 and let-7 switch worm larvae between stages, and human let-7 represses oncogenes such as RAS. Some miRNAs act as oncogenes (the miR-17-92 cluster) and others as tumor suppressors (miR-15a and miR-16-1, deleted in many chronic lymphocytic leukemias).

PIWI-interacting RNAs (piRNAs) silence transposons in the germ line by cleaving their transcripts and guiding DNA and H3K9 methylation to their copies. Long noncoding RNAs (over 200 nucleotides), such as Xist, act as scaffolds that bring chromatin modifiers to particular loci.

RNAi as a tool and a therapy

RNAi made reverse genetics easy: feeding worms bacteria that express dsRNA silences the matching gene. In mammalian cells long dsRNA triggers an interferon response, so researchers use synthetic 21 nucleotide siRNAs or short hairpin RNAs (shRNAs). RNAi gives a knockdown rather than a knockout, and partial seed matches cause off target effects.

siRNA drugs, chemically stabilized and delivered to the liver in lipid nanoparticles or attached to GalNAc (which binds a hepatocyte receptor), are now approved: patisiran silences transthyretin in hereditary transthyretin amyloidosis, and inclisiran silences PCSK9 to lower LDL cholesterol with dosing about twice a year.

Words worth knowing