Eukaryotic Gene Regulation and Epigenetics

Nearly every cell in your body carries the same genome. Yet a liver cell and a kidney cell read different genes from it, and when either one divides, both daughters go on reading their parent’s set.

How does a cell choose, and how does it remember? In the lac operon a bacterial promoter is open unless a repressor sits on it. Eukaryotic genes start from the other side: their DNA is packed into chromatin, and packed DNA is hard to read.

Chromatin: the default state is closed

The unit of packing is the nucleosome. About 147 base pairs of DNA wrap nearly twice around a core of eight histone proteins, two copies each of H2A, H2B, H3 and H4. (The 1997 crystal structure from Karolin Luger, Timothy Richmond and colleagues held 146 base pairs.) The linker histone H1 binds the short linker DNA between nucleosomes and helps fold the string into compact fibres; it is not part of the octamer.

Think of thread wound on spools: a needle catches thread between spools, not thread pressed against one. A nucleosome on a promoter generally blocks the proteins that start transcription (transcription). ATP-driven remodelling complexes move nucleosomes; the yeast SWI/SNF complex slid them along DNA in a reconstituted test-tube system.

Each histone has a tail, rich in lysine and arginine, that sticks out of the core. Enzymes add and remove chemical marks on these tails and on the DNA itself.

Marks on histones and on DNA

Acetylation. Histone acetyltransferases (HATs) add an acetyl group to lysine, removing its positive charge; histone deacetylases (HDACs) take it off. Acetylated chromatin goes with active genes. One reason is structural: in nucleosome arrays built in the test tube, histone H4 acetylated at lysine 16 stopped the arrays folding into compact fibres. In 1996 C. David Allis’s laboratory cloned a histone acetyltransferase from the ciliate Tetrahymena that resembled yeast Gcn5, a protein known to be needed to activate certain genes, and recombinant Gcn5 acetylated histones. A known activator was a chromatin-modifying enzyme.

Histone methylation means different things on different residues. Marks are named by histone, residue and modification: H3K4me3 is histone H3 carrying three methyl groups on lysine 4.

• H3K4me3 sits at the promoters of active genes.
• H3K27me3, placed by Polycomb complexes, marks repressed genes, many of them developmental regulators.
• H3K9me3 marks heterochromatin, the densely packed, mostly silent chromatin. In 2001 two laboratories (Bannister and colleagues; Lachner and colleagues) reported that the protein HP1 binds H3 methylated at lysine 9 but not at lysine 4.

So “methylation silences” is wrong for histones in general. Brian Strahl and Allis proposed in 2000 that combinations of marks are read like a code; the “histone code” is still a hypothesis. A mark can also go with activity without causing it. In mouse embryonic stem cells, catalytically dead Mll3 and Mll4 enzymes lost the H3K4 monomethylation they place at enhancers, yet transcription changed only slightly.

DNA methylation. In mammals, DNA methyltransferases add a methyl group to carbon 5 of cytosine, mostly where C is followed by G (a CpG site). DNMT3A and DNMT3B lay down new methylation. After replication, DNMT1 copies the pattern from the old strand onto the new one. Dense methylation of a promoter’s CpG island goes with a silent gene. One reader is MeCP2, which binds methylated DNA. In cultured-cell assays it associated with a complex containing histone deacetylases, and the deacetylase inhibitor trichostatin A relieved its repression: the two marking systems work together. Mutations in MECP2 cause most cases of Rett syndrome (synapse disorders).

Two limits. Methylation can regulate one promoter rather than a whole gene, as at alternative promoters inside SHANK3. And it is not universal: budding yeast and C. elegans have little or no 5-methylcytosine, and Drosophila very little.

Enhancers: control from a distance, by combination

In 1981 Julian Banerji, Sandro Rusconi and Walter Schaffner put a rabbit β-globin gene into human HeLa cells with and without a piece of DNA from the virus SV40. With the SV40 piece, β-globin transcripts rose about 200-fold. The piece contained a 72 base-pair repeat, and it worked 1,400 base pairs upstream or 3,300 downstream of the start of transcription, in either orientation. Pierre Chambon’s laboratory reported the same year that the repeat was needed for SV40 T-antigen expression in monkey cells. Such an element is an enhancer: DNA that raises transcription from a promoter and can act at a distance and in either orientation. “Can” matters. Not every enhancer does both, and in Drosophila cells thousands of enhancers preferred one of two kinds of core promoter.

Enhancers are bound by sequence-specific transcription factors (some switched on by hormones, as in nuclear receptors, or by signaling to the nucleus). These recruit coactivators, including HATs, and the Mediator complex, which bridges the enhancer to RNA polymerase II at the promoter while the DNA between them loops out.

Think of an enhancer as a committee that acts only when the right members are present. Combined in different ways, a modest set of factors can specify far more patterns than there are factors, as in the motor neuron code.

Worked problem: mapping a liver enhancer

A 600 bp region upstream of a gene contains elements A, B and C. Each version is cloned beside the gene’s promoter driving luciferase and put into liver and kidney cells. Assume equal uptake has been corrected with a second reporter, each value is the mean of independent repeats, and activity is scaled so that promoter alone = 1.

Construct	Liver	Kidney
A + B + C	100	2
Delete A	95	2
Delete B	4	1
Delete C	30	2
B alone	5	–
C alone	2	–
B + C	90	–
B + C, reversed	88	–

1. A. Removing it leaves 95 of 100: dispensable in liver.
2. B. Removing it drops 100 to 4, a 25-fold fall (100 ÷ 4 = 25), so B is necessary. B alone gives only 5, so it is not sufficient.
3. C. Removing it drops 100 to 30, about 3.3-fold; alone it gives 2.
4. B and C together give 90, 90% of the full region. If their effects simply added, you would expect about 1 + 4 + 1 = 6; if they multiplied, about 5 × 2 = 10. The observed 90 is about nine times even the multiplied guess: B and C work together, as expected if factors on both must be present at once.
5. Orientation. Reversed B + C gives 88, 98% of 90: orientation does not matter here.
6. Kidney. Nothing exceeds 2. The simplest reading is that a factor B or C needs is missing in kidney cells; a kidney repressor would also fit.

Answer: the minimal liver enhancer is B + C, acting combinatorially. Its role at the native gene needs a test there, such as deleting B in the genome.

Factors bound at the enhancer

factor on element A factor on element B factor on element C

Chromatin at the promoter

open (no nucleosome on the promoter) closed (a nucleosome on the promoter)

Rule to keep: the enhancer works as a committee, so B and C together give far more than either alone, and packed DNA is hard to read. This is a sketch of the logic using the lesson’s liver numbers, not a prediction for any real gene.

Small RNAs: lin-4 and RNA interference

In 1993 Rosalind Lee, Rhonda Feinbaum and Victor Ambros reported that lin-4, a gene that times larval development in C. elegans, does not appear to encode a protein. It made RNAs of about 22 and 61 nucleotides, partly complementary to repeated sites in the 3’ untranslated region (3’ UTR) of lin-14 mRNA, and Gary Ruvkun’s laboratory showed that the lin-14 3’ UTR is needed for the repression. Nobody called lin-4 a microRNA then; the term came with the let-7 work of 2000 and 2001.

A microRNA gene is transcribed into an RNA that folds into hairpins. In the nucleus Drosha cuts the hairpin out; in the cytoplasm Dicer trims it to a duplex of about 22 nucleotides. One strand guides an Argonaute protein to target mRNAs, mostly to their 3’ UTRs. Pairing centres on a seed of about seven nucleotides near the guide’s 5’ end; outside it, animal microRNAs mostly pair imperfectly. They repress translation and destabilize the mRNA. In HeLa cells and mouse neutrophils, lower mRNA levels accounted for at least 84% of the reduced protein output from targets (Guo and colleagues, 2010). Plant microRNAs often pair fully and direct cleavage.

The second route, RNA interference, came from worms: in 1998 Andrew Fire, Craig Mello and colleagues found that injected double-stranded RNA silenced the matching gene far more effectively than either single strand. Dicer cuts long double-stranded RNA into short interfering RNAs (siRNAs) of 21 to 22 nucleotides. An siRNA pairs fully with its target, and Argonaute (Argonaute2 in mammals) cuts the mRNA at the paired site. In many mammalian cells long double-stranded RNA sets off broad antiviral responses, so RNAi there needed short synthetic siRNAs (Elbashir, Tuschl and colleagues, 2001). RNAi as a tool is in forward and reverse genetics.

What “epigenetic” does and does not mean

An epigenetic change is a heritable change in gene expression across cell divisions without a change in DNA sequence. Inheritance through mitosis is well established, through DNMT1 and more. X inactivation and imprinting keep one copy of a gene silent through many divisions (sex linkage and beyond). In mouse fibroblasts, heterochromatin induced by tethering HP1α to an engineered Oct4 allele persisted after HP1α was removed, when no activator opposed it.

Inheritance through the germline, across generations of whole animals, is another claim. It is well documented in plants and in C. elegans. In mammals it is limited and contested, partly because most DNA methylation is erased and rewritten in developing germ cells and again in the early embryo. Count generations carefully. Exposing a pregnant mother (F0) also exposes her F1 embryos and the germ cells inside them that will form F2, so F3 is the first unexposed generation; after exposure of a father alone, it is F2. Studies of the Dutch Hunger Winter found methylation differences, decades later, in people exposed to famine before birth; they were exposed themselves, so that is not transgenerational inheritance. In human data, shared genes, environment and culture confound the inference. None of this is health or diet advice.

Design problem

Pregnant mice fed a methyl-donor diet bear offspring with altered coat colour at a locus whose expression is controlled by a promoter derived from a transposon. Design an experiment to decide whether the effect is transmitted epigenetically to later generations.

One way to do it

1. Count generations. The F0 mother, her F1 embryos and their germ cells (future F2) are all exposed. Only F3 and later can show transgenerational inheritance; F1 and F2 effects can be direct exposure.
2. Remove the mother’s environment. Transfer embryos into unexposed females, or at least cross-foster pups to unexposed mothers, so neither uterus nor nursing carries the effect.
3. Separate the parents. Breed affected males and affected females separately to unexposed mates, and score coat colour in each line through F3 and F4. Sperm carry less besides DNA than eggs do, so the male line is the cleaner test.
4. Rule out a DNA change. Sequence the locus, transposon included, in each generation.
5. Measure the mark. Bisulfite-sequence the transposon promoter in tissue and in sperm or eggs of each generation, and ask whether methylation tracks coat colour.

Run unexposed control lines alongside and score blind. A shift that reaches F3 through either parent, with no sequence change and with methylation that follows it, would support epigenetic transmission at this locus.