The Epigenetics Revolution

Chapter 11. Fighting the Enemy Within

There are multiple instances in science of a relatively chance event leading to a wonderful breakthrough. Probably the most famous example is Alexander Fleming’s observation that a particular mould, that had drifted by chance onto an experimental Petri dish, was able to kill the bacteria growing there. It was this random event that led to the discovery of penicillin and the development of the whole field of antibiotics. Millions of lives have been saved as a result of this apparently chance discovery.

Alexander Fleming won the Nobel Prize for Physiology or Medicine in 1945, along with Ernst Chain and Howard Florey who worked out how to make penicillin in large quantities so that it could be used to treat patients. Isaac Asimov’s famous statement at the top of this page flags up to us that Alexander Fleming wasn’t simply some fortunate man who struck lucky. His insight wasn’t a fluke. It’s very unlikely that Fleming was the first scientist whose bacterial cultures had become infected with mould. His achievement came in recognising that something unusual had happened, and appreciating its significance. Knowledge and training had prepared Fleming’s mind to make the most of the chance event. He saw what probably many others had seen before him but he thought what nobody else had thought.

Even if we accept the role that odd events have played in research, it would still be very comforting to think that science generally proceeds in a logical and ordered fashion. Here’s one way we could imagine such progress in epigenetics …

Epigenetic modifications control cell fate – it’s these processes by which liver cells, for example, stay as liver cells and don’t turn into other cell types. Cancer represents a breakdown in normal control of cell fate, because liver cells stop being liver cells and become cancer cells, suggesting that epigenetic regulation has become abnormal in cancer. We should therefore aim to develop drugs that influence this epigenetic mis-regulation. Such drugs may be useful for treating or controlling cancer.

That’s a neat and tidy process, and makes a lot of sense. In fact, hundreds of millions of dollars are being spent in the global pharmaceutical industry to develop epigenetic drugs for exactly this purpose. But the clear-cut thought process outlined above is not how this process of cancer drug discovery started.

There are already licensed drugs which treat cancer and which work by inhibiting epigenetic enzymes. These compounds were shown to be active against cancer cells before they were shown to work on epigenetic enzymes. In fact, it’s the success of these compounds that has really stirred up interest in epigenetic therapies, and in the whole field of epigenetics itself – so much for a neat narrative arc.

Back in the early 1970s, a young South African scientist called Peter Jones was working with a compound called 5-azacytidine. This compound was already known to have anti-cancer effects because it could stop leukaemia cells from dividing, and had some beneficial effects when tested in childhood leukaemia patients[167].

Peter Jones is now recognised as the founding father of epigenetic treatments for cancer. Tall, thin, tanned and with thick close-cropped white hair, he is an instantly recognisable presence at any conference. Like so many of the terrific scientists mentioned in this book, he has researched for decades in an ever-evolving field. He remains at the forefront of efforts to understand the impact of the epigenome on health. He is currently spearheading efforts to characterise all the epigenetic modifications present in a vast number of different cell types and diseases. These days he is able to call on technologies that allow his team to analyse millions of read-outs from highly specific and specialised equipment. Back in the early 1970s, he made his first breakthrough by being incredibly observant and thorough – a classic case of a prepared mind.

Forty years ago, nobody was quite sure how 5-azacytidine worked. It’s very similar in chemical structure to base C (cytidine) from DNA and RNA. It was assumed that 5-azacytidine got added into DNA and RNA chains. Once there, it somehow disrupted normal copying of DNA, and transcription or activity of RNA. Cancer cells such as the ones found in leukaemia are extremely active. They need to synthesise lots of proteins, which means they need to transcribe a lot of mRNA. Because they divide quickly they also need to replicate their DNA very efficiently. If 5-azacytidine was interfering with one or both of these processes, it would probably hamper the growth and division of the cancer cells.

Peter Jones and his colleagues were testing the effects of 5-azacytidine on a range of cells from mammals. It’s remarkably fiddly to get many types of cells to grow in the laboratory if you just take them straight out of a human or another animal. Even when you can get them to grow, they often stop dividing after a few cell divisions and die off. To get around this, Peter Jones worked with cell lines. Cell lines are derived originally from animals, including humans, but as a result of chance or experimental manipulation, they are able to grow indefinitely in culture, if given the right nutrients, temperature and environmental conditions. Cell lines are not exactly the same as cells in the body, but they are a useful experimental system.

The type of cells that Peter Jones and his colleagues were testing are usually grown in a flat plastic flask. This looks a little like a see-through version of a hip flask for whisky or brandy, lying on its side. The mammalian cells grow on the flat inside surface of the flask. They form a single layer of cells, tightly packed side by side, but never growing on top of one another.

One morning, after the cells had been cultured with 5-azacytidine for several weeks, the researchers found that there was a strange lumpy bit in one of the culture flasks. To the naked eye, this initially looked like a mould infection. Most people would just discard the flask and make a silent promise to be a bit more careful when culturing their cells in future, to stop this happening again. But Peter Jones did something else. He looked at the lump more closely and discovered it wasn’t a stray bit of mould at all. It was a big mass of cells, which had fused to form giant cells containing lots of nuclei. These were little muscle fibres, the syncytial tissue we met in the discussion of X inactivation. Sometimes the little muscle fibres would even twitch[168].

This was very odd indeed. Although the cell line had originally been derived from a mouse embryo, it never usually formed anything like a muscle cell. It tended instead to form epithelial cells – the cell type that lines the surfaces of most of our organs. Peter Jones’ work showed that 5-azacytidine could change the potential of these embryonic cells, and force them to become muscle cells, instead of epithelial cells. But why would a compound that killed cancer cells, presumably by disrupting production of DNA and mRNA, have an effect like this?

Peter Jones carried on working on this when he moved from South Africa to the University of Southern California. Two years later, he and his PhD student Shirley Taylor showed that cell lines treated with 5-azacytidine didn’t only form muscle. They could also form other cell types. These included fat cells (adipocytes) and cells called chondrocytes. These produce cartilage proteins, such as those that line the surfaces of joints so that the two planes can glide smoothly over each other.

These data showed that 5-azacytidine wasn’t a special muscle-specifying factor. Very presciently, Professor Jones made the suggestion in his paper reporting this work that, ‘5-azacytidine … causes a reversion to a more pluripotent state’[169]. In other words, this compound was pushing the ball a little way back up Waddington’s epigenetic landscape. The ball was then rolling back down the valleys between the hills, into a different final resting place.

But there was still no theory as to why 5-azacytidine had this unusual effect. Peter Jones himself tells a lovely self-deprecating story about the turning point in our understanding. His original appointment at the University of Southern California was in the Department of Paediatrics, but he wanted a joint appointment with the Department of Biochemistry. Part of the procedure for obtaining this joint appointment included an extra interview, which he considered quite pointless. Peter Jones described his work with 5-azacytidine in this interview and explained that no-one knew why the compound affected cell pluripotency. Robert Stellwagen, another scientist at the same university who was taking part in the interview asked, ‘Have you thought of DNA methylation?’. Our candidate admitted he not only hadn’t thought of it, he hadn’t even heard of it[170].

Peter Jones and Shirley Taylor immediately began to focus on DNA methylation and in a very short time showed that this was indeed key to the effects of 5-azacytidine. 5-azacytidine inhibited DNA methylation. Peter Jones and Shirley Taylor created a number of related compounds and tested them for their effects in cell culture. The ones that inhibited DNA methylation also caused the changes in phenotype originally observed for 5-azacytidine. Compounds that didn’t inhibit DNA methylation had no effect on phenotype[171].

Cytidine (base C) and 5-azacytidine are very similar in chemical structure. They are shown in Figure 11.1, which for simplicity only shows the most relevant parts of the structure (called cytosine and 5-azacytosine, respectively).

^{Figure 11.1 5-azacytosine can be incorporated into DNA during the DNA replication which takes place prior to cell division. 5-azacytosine takes the place of a C base, but because it contains a nitrogen atom where there is usually a carbon atom, the foreign base cannot be methylated by DNMT1 in the way that was described in Figure 4.2.}

The top half of the diagram is very similar to Figure 4.1, showing that cytosine can be methylated by a DNA methyltransferase (DNMT1, DNMT3A or DNMT3B) to create 5-methylcytosine. In 5-azacytosine, a nitrogen atom (N) replaces the key carbon atom (C) that normally gets methylated. The DNA methyltransferases can’t add a methyl group to this nitrogen atom.

Thinking back to Chapter 4, imagine a methylated region of DNA. When a cell divides, it separates the two strands of the DNA double helix and copies each one. But the enzymes that copy the DNA can’t themselves copy DNA methylation. As a consequence each new double helix had one methylated strand and one unmethylated one. The DNA methyltransferase called DNMT1 can recognise DNA which has only got DNA methylation on one strand and can replace it on the other strand. This restores the original DNA methylation pattern.

But if dividing cells are treated with 5-azacytidine, this abnormal cytidine base is added into the new strand of DNA as the genome gets copied. Because the abnormal base contains a nitrogen atom instead of a carbon atom, the DNMT1 enzyme can’t replace the missing methyl group. If this continues as the cells keep dividing, the DNA methylation begins to get diluted out.

Something else also happens when dividing cells are treated with 5-azacytidine. We now know that when DNMT1 binds at a region where the DNA contains 5-azacytidine instead of the normal cytidine, the DNMT1 becomes stuck there[172]. This marooned enzyme is then sent to a different part of the cell and is broken down. Because of this, the total levels of DNMT1 enzyme in the cell fall[173][174]. The combination of this decrease in the amount of DNMT1, and the fact that 5-azacytidine can’t be methylated, means that the amount of DNA methylation in the cell keeps dropping. We’ll come back in a little while to why this drop in DNA methylation has an anti-cancer effect.

So, 5-azacytidine is an example of where an anti-cancer agent was unexpectedly shown to work epigenetically. Bizarrely, a rather similar thing happened with our second example of a compound which is now licensed to treat cancer[175].

In 1971 the scientist Charlotte Friend showed that a very simple compound called DMSO (its full name is dimethyl sulfoxide) had an odd effect on the cancer cells from a mouse model of leukaemia. When these cells were treated with DMSO, they turned red. This was because they had switched on the gene for haemoglobin, the pigment that gives red blood cells their colour[176]. Leukaemia cells normally never switch on this gene and the mechanism behind this effect of DMSO was completely unknown.

Ronald Breslow at Columbia University and Paul Marks and Richard Rifkind at Memorial Sloan-Kettering Cancer Center were intrigued by Charlotte Friend’s research. Ronald Breslow began to design and create a new set of chemicals, using the structure of DMSO as his starting point, and then adding or changing bits, a little like making new combinations of Lego bricks. Paul Marks and Richard Rifkind began to test these chemicals in various cell models. Some of the compounds had a different effect from DMSO. They stopped cells from growing.

After many iterations, learning from each new and more complicated set of structures, the scientists created a molecule called SAHA (suberoylanilide hydroxamic acid). This compound was really effective at stopping growth and/or causing cell death in cancer cell lines[177]. However, it was another two years before the team were able to identify what SAHA was doing in cells. The key moment happened more than 25 years after Charlotte Friend’s breakthrough publication, when Victoria Richon in Paul Marks’ team, read a 1990 paper from a group at the University of Tokyo.

The Japanese group had been working on a compound called Trichostatin A or TSA. TSA was known to be able to stop cells proliferating. The Japanese group showed that treatment with TSA altered the extent to which histone proteins are decorated with the acetyl chemical group in cancer cell lines. Histone acetylation is another epigenetic modification that we first met in Chapter 4. When cells were treated with TSA, the levels of histone acetylation went up. This wasn’t because the compound was activating the enzymes that put the acetyl groups on histones. It was because TSA was inhibiting the enzymes that remove acetyl groups from these chromatin proteins. These proteins are called histone deacetylases, or HDACs for short[178].

Victoria Richon compared the structure of TSA with the structure of SAHA, and the two are shown in Figure 11.2.

^{Figure 11.2 The structures of TSA and SAHA, with the areas of greatest similarity circled. C: carbon; H: hydrogen; N: nitrogen; O: oxygen. For simplicity, some carbon atoms have not been explicitly shown, but are present where there is a junction of two lines.}

You don’t need a chemistry degree to see that TSA and SAHA look fairly similar, especially at the right hand side of each molecule. Victoria Richon hypothesised that, just like TSA, SAHA was also an HDAC inhibitor. In 1998, she and her colleagues published a paper that showed this was indeed the case[179]. SAHA prevents HDAC enzymes from removing acetyl groups from histone proteins, and as a result, the histones carry lots of acetyl groups.

So, 5-azacytidine and SAHA both decrease cancer cell proliferation, and both inhibit the activity of epigenetic enzymes. Although we could take this as promising support for the theory that epigenetic proteins are important in cancer, perhaps we could just be leaping to conclusions? It might just be a coincidence that both drugs affect epigenetic proteins. After all, the enzymes targeted by the two compounds are very different. 5-azacytidine inhibits the DNMT enzymes, which add methyl groups to DNA. SAHA, on the other hand, inhibits the HDAC family of enzymes, which remove acetyl groups from histone proteins. Superficially, these seem like very different processes. Maybe it’s just coincidence that both 5-azacytidine and SAHA inhibit epigenetic enzymes?

Epigeneticists believe that it is far from being a coincidence. DNA methyltransferase enzymes add a methyl group to the cytidine base. High concentrations of this base are found in the long CG-rich stretches of DNA known as CpG islands. These islands are found upstream of genes, in the promoter regions that control gene expression. When the DNA of a CpG island is heavily methylated, the gene controlled by that promoter is switched off. In other words, DNA methylation is a repressive modification. DNMT activity increases DNA methylation and therefore represses gene expression. By inhibiting these enzymes with 5-azacytidine, we can drive gene expression up.

Histone proteins are also found at the promoters of genes. Histone modifications can be very complex, as we saw in Chapter 4. But histone acetylation is the most straightforward in terms of its effects on gene expression. If the histones upstream of a gene are heavily acetylated, the gene is likely to be highly expressed. If the histones are lacking acetylation, the gene is likely to be switched off. Histone deacetylation is a repressive change. Histone deacetylases (HDACs) remove the acetyl groups from histone proteins and will therefore repress gene expression. By inhibiting these enzymes with SAHA, we can drive gene expression up.

So there is a consistent finding. Our two unrelated compounds, which control growth of cancer cells in culture and which have now been licensed for use in human treatment, inhibit epigenetic enzymes. In doing so, they both drive up gene expression which raises the obvious question of why this is useful for treating cancer. To understand this, we need to get to grips with some cancer biology.

Cancer is the result of abnormal and uncontrolled proliferation of cells. Normally, the cells of our body divide and proliferate at exactly the right rate. This is controlled by a complex balancing act between networks of genes in our cells. Certain genes promote cell proliferation. These are sometimes referred to as proto-oncogenes. They were represented by a plus sign in the see-saw diagram in the previous chapter. Other genes hold the cell back, preventing too much proliferation. These genes are called tumour suppressors. They were represented by a negative sign on the same diagram.

Proto-oncogenes and tumour suppressors are not intrinsically good or bad. In healthy cells, the activities of these two classes of genes balance each other. But when regulation of these networks goes wrong, cell proliferation may become mis-regulated. If a proto-oncogene becomes over-active, it may push a cell towards a cancerous state. Conversely, if a tumour suppressor gets inactivated, it will no longer act as a brake on cell division. The outcome is the same in both cases – the cell may begin to proliferate too rapidly.

But cancer isn’t just a result of too much cell proliferation. If cells divide too quickly but are otherwise normal, they form structures called benign tumours. These may be unsightly and uncomfortable but unless they press on a vital organ and affect its activity, they are unlikely in themselves to be fatal. In full-blown cancer the cells don’t just divide too often, they are also abnormal and can start to invade other tissues.

A mole is a benign tumour. So is a little outgrowth in the inside of the large intestine, called a polyp. Neither a mole nor a polyp is dangerous in itself. The problem is that the more of these moles or polyps you have, the greater the likelihood that one of them will go the next step, and develop an abnormality that will take it further along the path towards full-blown cancer.

This implies something rather important, that has been demonstrated in a large number of studies. Cancer is not a one-off event. Cancer is a multi-step process, where each additional step takes a cell further along the road to becoming malignant. This is true even in cases where patients inherit a very strong pre-disposition to cancer. One example is pre-menopausal breast cancer, which runs in some families. Women who inherit a mutated copy of a gene called BRCA1 are at very high risk of early and aggressive breast cancer, which is difficult to treat effectively. But even these women aren’t born with active breast cancer. It takes many years before the cancer develops, because other defects have to accumulate as well.

So, cells accumulate defects as they move increasingly close to becoming cancerous. These defects must be transmitted from mother cell to daughter cell, because otherwise they would be lost each time a cell divided. These defects must be heritable as the cancer develops. Understandably, for a very long time, the attention of the scientific community focused on identifying mutations in the genes involved in the development of cancer. They were looking for alterations in the genetic code, the fundamental blueprint. They were particularly interested in the tumour suppressor genes as these are the genes that are usually mutated in the inherited cancer syndromes.

Humans tend to have two copies of each tumour suppressor gene, as most are carried on the autosomes. As a cell becomes increasingly cancerous, both copies of key tumour suppressor genes usually get switched off (inactivated). In many cases this may be because the gene has mutated in the cancer cells. This is known as somatic mutation – it has happened in body cells at some point during normal life. These are called somatic mutations to distinguish them from genetic mutations, the ones that are transmitted from parent to child. The mutations that inactivate the two copies of a tumour suppressor may be quite variable. In some cases there may be changes in the amino acid sequence, so that the gene can’t produce a functional protein any more. In other cases, there may be loss of the relevant part of the chromosome in the increasingly cancerous cells. In an individual patient, one copy of a specific tumour suppressor may carry a mutation that changes the amino acid sequence and the other may have suffered a micro-deletion.

It’s abundantly clear that these events do happen, and quite frequently, but often it’s been difficult to identify exactly how a tumour suppressor has mutated. In the last fifteen years, we’ve started to realise that there is another way that a tumour suppressor gene can become inactivated. The gene may be silenced epigenetically. If the DNA at the promoter becomes excessively methylated or the histones are covered in repressive modifications, the tumour suppressor will be switched off. The gene has been inactivated without changing the underlying blueprint.

Various labs have identified cancers where this has clearly happened. One of the first reports was in a type of kidney cancer called clear-cell renal carcinoma. A key step in the development of this kind of cancer is the inactivation of a specific tumour suppressor gene called VHL. In 1994, a group headed by the hugely influential Stephen Baylin from Johns Hopkins Medical Institution in Baltimore analysed the CpG island in front of the VHL gene. In 19 per cent of the clear-cell renal carcinoma samples that they analysed, the DNA of the island was hypermethylated. This switched off expression of this key tumour suppressor gene, and was almost certainly a major event in cancer progression in these individuals[180].

Promoter methylation was not restricted to the VHL tumour suppressor and renal cancer. Professor Baylin and colleagues subsequently analysed the BRCA1 tumour suppressor gene in breast cancer. They analysed cases where there was no family history of this disease, and the cancer wasn’t caused by the mutations in BRCA1 that we discussed a few paragraphs ago. In 13 per cent of these sporadic cases of breast cancer, the BRCA1 CpG island was hypermethylated[181]. Broader abnormal patterns of DNA methylation in cancer were reported by Jean-Pierre Issa from the MD Anderson Cancer Center in Houston, in collaboration with Stephen Baylin. Their collaborative work showed that over 20 per cent of colon cancers had high levels of promoter DNA methylation, at many different genes simultaneously[182].

Follow-on work showed that it’s not just DNA methylation that changes in cancer. There is also direct evidence for histone modifications leading to repression of tumour suppressor genes. For example, the histones associated with a tumour suppressor gene called ARHI had low levels of acetylation in breast cancer[183]. A similar relationship exists for the PER1 tumour suppressor in a form of lung cancer called non-small cell[184]. In both cases, there was a relationship between the levels of histone acetylation and the expression of the tumour suppressor – the lower the levels of acetylation, the lower the expression of the gene. Because these genes are both tumour suppressors, their decreased expression would mean that the cell would find it harder to put the brakes on proliferation.

This realisation – that tumour suppressor genes are often silenced by epigenetic mechanisms – has led to considerable excitement in the field, because this potentially creates a new way of treating cancer. If you can turn one or more tumour suppressor genes back on in cancer cells, there is a fighting chance of reining in the crazy proliferation rate of those cells. The runaway train may not run away quite so fast down the track.

When scientists thought that tumour suppressors were inactivated by mutations or deletions, we didn’t have many options for turning these genes back on. There are trials in progress to test if gene therapy can be used to achieve this. There may be circumstances where gene therapy will prove effective, but this is by no means certain. Gene therapy has struggled to deliver on the initial hopes for this technology, in all sorts of diseases. It can be very difficult to get the genes delivered into the right cells, and to get them to switch on when they are there. Even when we’re able to do this, we often find that the body gets rid of these extra genes, so any initial benefit is lost. There have also been relatively rare cases where the gene therapy itself has led to cancer, because it has had unexpected effects which have led to increased cell proliferation. The scientific community hasn’t given up hope for gene therapy and for some conditions it may yet prove to be the right approach[185]. But for diseases like cancer, where we would need to treat a lot of people, it’s expensive and difficult.

That’s why there is so much excitement about the development of epigenetic drugs to treat cancer. By definition, epigenetic changes do not alter the underlying DNA code. As we have seen, there are patients where one copy of a tumour suppressor has been silenced by the action of epigenetic enzymes. In these patients the code for the normal tumour suppressor protein has not been corrupted by mutation. So, for them there is the possibility that treatment with appropriate epigenetic drugs can reverse the abnormal pattern of DNA methylation or histone acetylation. If we can achieve this, the normal tumour suppressor gene will be switched back on, and this will help bring the cancer cells back under control.

Two drugs that inhibit the DNMT1 enzyme have been licensed for clinical use in cancer patients by the Food and Drug Administration (FDA) in the USA. These are 5-azacytidine (tradename Vidaza) and the closely related 2-aza-5′-deoxycytidine (tradename Dacogen). Two HDAC inhibitors have also been licensed. These are SAHA (tradename Zolinza), which we met earlier, and a molecule called romidepsin (tradename Istodax), which has a very different chemical structure from SAHA, but which also inhibits HDAC enzymes.

Following on from his successes in unravelling the molecular roles of 5-azacytidine, Peter Jones, along with Stephen Baylin and Jean-Pierre Issa, has played a hugely influential role in the last 30 years in moving this compound from the laboratory, all the way through clinical trials and finally to the licensed product. Victoria Richon played a major role in championing SAHA all the way through the same process.

The successful licensing of these four compounds against two different types of enzymes has given a major boost to the whole field of epigenetic therapies. But they have not proved to be universal wonder drugs, the silver bullets to treat all cancers.

That hasn’t been a surprise to anyone working in the fields of cancer research and treatment. There sometimes seems to be an obsessive determination on the part of certain journalists in the popular press to write about the cure for cancer. Generally speaking, scientists try to avoid being too dogmatic, but if there’s one thing most of them are agreed on, it’s that there will never be one single cure for cancer.

That’s because there isn’t one form of cancer. There are probably over a hundred different diseases with this name. Even if we take just one example – say breast cancer – we find that there are different types of this particular strain of cancer. Some grow in response to the female hormone called oestrogen. Some respond most strongly to a protein called epidermal growth factor. The BRCA1 gene is inactivated or mutated in some breast cancer cases, but not in others. Some breast cancers don’t respond to any of the known cancer growth factors but to some other signals which we may not even be able to identify yet.

Because cancer is a multi-step process, two patients whose cancers appear very similar may be ill because of very different molecular processes. Their cancers may have rather different combinations of mutations, epigenetic modifications and other factors driving the growth and aggressiveness of the tumour. This means that different patients are likely to require different types and combinations of anti-cancer drugs.

Even allowing for this, however, the results from clinical trials with DNMT and HDAC inhibitors have been surprising. Neither of them has yet been shown to work well in solid tumours such as cancers of the breast, colon or prostate. Instead, they are most effective against cancers that have developed from cells that give rise to the circulating white blood cells that are part of our defences against pathogens. These are referred to as haematological tumours. It’s not clear why the current epigenetic drugs don’t seem to be effective against solid tumours. It might be that there are different molecular mechanisms at work in these, compared with haematological cancers. Alternatively, it could be that the drugs can’t get into solid tumours at high enough concentrations to affect most of the cancer cells.

Even within haematological tumours, there are differences between the DNMT and HDAC inhibitor drugs. Both DNMT inhibitors have been licensed for use in a condition called myelodysplastic syndrome[186][187]. This is a disorder of the bone marrow.

Both HDAC inhibitors have been licensed for a different kind of haematological tumour, called cutaneous T cell lymphoma[188]. In this disease, the skin becomes infiltrated with proliferating immunological cells called T cells, creating visible plaques and large lesions.

Not every patient with myelodysplastic syndrome or cutaneous T cell lymphoma gains a clinical benefit from taking these drugs. Even amongst the patients who do respond, none of these drugs really seem to cure the condition. If the patients stop taking the drugs, the cancer regains its hold. The DNMT1 inhibitors and the HDAC inhibitors seem to rein in the cancer cell growth, retarding and repressing it. They control rather than cure.

However, this often represents a significant improvement for the patients, bringing prolonged life expectancy and/or improved quality of life. For example, many patients with cutaneous T cell lymphoma suffer significant pain and distress because their lesions are constantly and excruciatingly itchy. The HDAC inhibitors are often very effective at calming this aspect of the cancer, even in patients whose survival times aren’t improved by these drugs.

Generally speaking, it’s often very difficult to know which patients will benefit from a specific new anti-cancer drug. This is one of the biggest problems facing the companies working on new epigenetic therapies for the treatment of cancer. Even now, several years after the first licences were granted by the FDA for 5-azacytidine and SAHA, we still don’t know why they work so much better in myelodysplastic syndrome and cutaneous T cell lymphoma than in other cancers. It just so happened that in the early clinical trials in humans, patients who had these conditions responded more strongly than patients with other types of cancers. Once the clinicians running the trials noticed this, later trials were designed that focused around these patient groups.

This may not sound like a major difficulty. It might seem straightforward for companies to develop drugs and then test them in all sorts of cancers and with all sorts of combinations of other cancer drugs, to work out how to use them best.

The problem with this is the expense. If we check out the website of the National Cancer Institute, we can look for the number of trials that are in progress for a specific drug. In February 2011, there were 88 trials to test SAHA[189]. It’s difficult to get definitive costs for how much clinical trials cost, but based on data from 2007, a value of $20,000 per patient is probably a conservative estimate[190]. Assuming each trial contains twenty patients, this would mean that the costs just for testing SAHA in the trials at the National Cancer Institute are over $35,000,000. And this is almost certainly an under-estimate of the overall cost.

The researchers at Columbia University and Memorial Sloan-Kettering who first developed SAHA patented it. They then set up a company called Aton Pharma to develop SAHA as a drug. In 2004, after promising early results in cutaneous T cell lymphoma, Aton Pharma was bought by the giant pharmaceutical company Merck for over $120 million dollars. Aton Pharma had almost certainly spent millions of dollars to get SAHA to this stage. Drug discovery and development is an expensive business. The two companies that marketed the DNMT1 inhibitors have been bought relatively recently by larger pharmaceutical companies, in deals that totalled about $3 billion each[191]. If a company has paid a huge amount of money to develop or buy in a new drug, it would much prefer not to carry on spending like a drunken sailor when it comes to clinical trials.

Naturally, it would be a big improvement if we could run clinical trials with a much better idea of which patients will benefit, rather than having to take pot luck. Unfortunately, most researchers agree that many of the animal models used to test cancer drugs are very limited in their capacity to predict the most susceptible human cancer. To be fair, this isn’t just true of cancer drugs targeted at epigenetic enzymes, it’s also true of pretty much all oncology drug discovery.

In an attempt to get around this problem, researchers in both academia and industry are now looking for the next generation of epigenetic targets in oncology. DNMT1 is a relatively broad-acting enzyme. DNA methylation is rather all or nothing – a CpG is methylated or it isn’t. HDACs tend to be pretty non-discriminating too. If they can get access to an acetylated lysine on a histone tail, they’ll take that acetyl group off. There are a lot of lysines on a histone tail – there are are seven on histone H3, just for starters. There are at least ten different HDAC enzymes that SAHA can inhibit. It’s quite likely that each of these ten can deacetylate any of the seven lysines on the H3 tail. This is hardly what we would call fine-tuning.

This is why the field is now moving in the direction of assessing different epigenetic enzymes, which are much more limited in their actions, to see which are important players in different cancers. The rationale is that it will be easier to understand the cellular biology of enzymes with quite limited actions, and this will make it easier to determine which patients are likely to respond best to which drugs.

The first problem in doing this is rather a daunting one. Which proteins should we investigate? There are probably at least a hundred enzymes that add or remove histone modifications (writers and erasers of the epigenetic code). There are probably as many proteins that read the epigenetic code. To make matters worse, many of these writers, erasers and readers interact with each other. How can we begin to identify the most promising candidates for new drug discovery programmes?

We don’t have any useful compounds like 5-azacytidine and SAHA to guide us, so we have to rely on our relatively incomplete knowledge in cancer and in epigenetics. One area that is proving useful is considering how histone and DNA modifications work in tandem.

The most heavily repressed areas of the genome have high levels of DNA methylation and are extremely compacted. The DNA has become very tightly wound up, and is exceptionally inaccessible to enzymes that transcribe genes. But it’s the question of how these regions become so heavily repressed that is really important. The model is shown in Figure 11.3.

^{Figure 11.3 Schematic to illustrate how different types of epigenetic modifications act together to create an increasingly repressed and tightly condensed chromosome region, making it very difficult for the cell to express genes from this region.}

In this model, there is a vicious cycle of events that results in the generation of a more and more repressed state. One of the predictions from this model is that repressive histone modifications attract DNA methyltransferases, which deposit DNA methylation near those histones. This methylation in turn attracts more repressive histone modifying enzymes, creating a self-perpetuating cycle that leads to an increasingly hostile region for gene expression.

Experimental data suggest that in many cases this model seems to be right. Repressive histone modifications can act as the bait to attract DNA methylation to the promoter of a tumour suppressor gene. A key example of this is an epigenetic enzyme we met in the previous chapter, called EZH2. The EZH2 protein adds methyl groups to the lysine amino acid at position 27 on histone H3. This amino acid is known as H3K27. K is the single letter code for lysine (L is the code for a different amino acid called leucine).

This H3K27 methylation itself tends to switch off gene expression. However, in at least some mammalian cell types, this histone methylation recruits DNA methyltransferases to the same region of chromatin[192][193]. The DNA methyltransferases include DNMT3A and DNMT3B. This is important because DNMT3A and DNMT3B can carry out the process known as de novo DNA methylation. That is, they can methylate virgin DNA, and create completely new regions of highly repressed chromatin. As a result, the cell can convert a relatively unstable repressive mark (H3K27 methylation) into the more stable DNA methylation.

Other enzymes are also important. An enzyme called LSD1 takes methyl groups off histones – it’s an eraser of epigenetic modifications[194]. It does this particularly strongly at position 4 on histone H3 (H3K4). H3K4 is the opposite of H3K27, because when H3K4 is clear of methyl groups, genes tend to be switched off.

Unmethylated H3K4 can bind proteins, and one of these is called DNMT3L. Perhaps not surprisingly, this is related to DNMT3A and DNMT3B. DNMT3L doesn’t methylate DNA itself, but it attracts DNMT3A and DNMT3B to the unmethylated H3K4. This provides another way to target stable DNA methylation to virgin territory[195].

In all likelihood, many histones positioned at the promoters of tumour suppressor genes carry both of these repressive histone marks – methylation of H3K27 and non-methylation of H3K4 – and these act together to target the DNA methyltransferases even more strongly.

Both EZH2 and LSD1 are up-regulated in certain cancer types, and their expression correlates with the aggressiveness of the cancer and with poor patient survival[196][197]. Basically, the more active these enzymes, the worse the prognosis for the patient.

So, histone modifications and DNA methylation pathways interact. This may explain, at least in part, one of the mysteries of existing epigenetic therapies. Why are compounds like 5-azacytidine and SAHA only controllers of cancer cells, rather than complete destroyers?

In our model, treatment with 5-azacytidine will drive down the DNA methylation for as long as the patients take the drug. Unfortunately, many cancer drugs have serious side-effects and the DNMT inhibitors are no exception. The side effects may eventually become such a problem that the patient has to stop taking the drug. However, the patient’s cancer cells probably still have histone modifications at the tumour suppressor genes. Once the patient stops taking 5-azacytidine, these histone modifications almost certainly start to attract the DNMT enzymes all over again, re-initiating stable repression of gene expression.

Some researchers are carrying out clinical trials using 5-azacytidine and SAHA together to try to interfere with this cycle, by disrupting both the DNA and histone components of epigenetic silencing. It’s not clear yet if these will be successful. If they aren’t, it might suggest that it’s not low levels of histone acetylation which are most important for re-establishing the DNA methylation. It might be that specific histone modifications, of the types just described, are more important. But we don’t yet have drugs to inhibit any of the other epigenetic enzymes, so we’re stuck with Hobson’s choice at the moment, that is, no choice at all.

In the future, we may not need to use DNMT inhibitors at all. The link between DNA methylation and histone modifications in cancer isn’t absolute. If a CpG island is methylated, the downstream gene is repressed. But there are tumour suppressor genes that are downstream of unmethylated CpG islands and tumour suppressor genes that don’t have a CpG island at all. These genes may still be repressed, but solely thanks to histone modifications[198]. This has been shown by Jean-Pierre Issa at the MD Anderson Cancer Center in Houston, who has been so instrumental in the implementation of epigenetic therapies in the clinic. In these instances, if we can find the right epigenetic enzymes to target with inhibitors, we may be able to drive re-expression of the tumour suppressors without needing to worry about DNA methylation.

Is there something special about the tumour suppressor genes that get silenced using epigenetic modifications? There are two contrasting theories about this. The first is that there’s nothing special about these genes and the process is completely random. In this model, every once in a while a random tumour suppressor gets abnormally modified epigenetically. If this changes the expression of the gene, it may mean that cells with that epigenetic modification grow a bit faster or a bit better than their neighbours. This gives the cells a growth advantage and they keep outgrowing the cells around them, gradually accumulating more epigenetic and genetic changes that make them ever more cancerous.

The other theory is that the tumour suppressors that become repressed epigenetically are somehow targeted in this process. It’s not just random bad luck, these genes are actually at a higher than average risk of epigenetic silencing.

In recent years, as we’ve had the technology to profile the epigenetic modifications in more and more cell types, and at higher and higher resolutions, the field has shifted in favour of the second model. There are a set of genes that seem to be rather prone to getting themselves switched off epigenetically.

At first this seems incredibly counter-intuitive. Why on earth would billions of years of evolution leave us with cellular systems that render us prone to cancerous changes? Well, we have to put this in context. Most evolutionary pressures are connected with the drive to leave behind as many offspring as possible. For a human to reach reproductive age, it’s incredibly important that early development occur as efficiently as possible. After all, you can’t reproduce if you never make it past the embryo stage. Once we’ve reached reproductive age and had the opportunity to reproduce, there is little to be gained in evolutionary terms in us staying alive for several decades afterwards.

Evolution has favoured cellular mechanisms that promote effective early growth and development, including the production of multiple different tissue types. Many of these tissue types contain reservoirs of stem cells which are specific to that tissue. Our bodies need these for tissue growth as we mature, and for tissue regeneration following injury. The fates and identities of these tissue-specific stem cells are controlled by the precise patterns of epigenetic modifications. By using epigenetic modifications to control gene expression, the cells keep some flexibility. They have the potential to change into more specialised cells, for example. Perhaps even more importantly when considering cancer, the epigenetic modifications also allow cells to divide to form more stem cells. This is why we don’t run out of skin cells, or bone marrow cells, even if we live to be a hundred years old.

This requirement for gene expression patterns that aren’t completely set in stone is probably why epigenetic repression of tumour suppressor genes is not a random process. We can’t have things two ways. Regulatory systems that allow cells to be flexible are inevitably also systems that allow cells to go wrong. In evolutionary terms, it’s the price we have to pay for our Goldilocks scenario. Our epigenetic pathways make sure some of our cells aren’t completely pluripotent or completely differentiated. Instead, they are just right, hovering somewhere near the top of Waddington’s epigenetic landscape, but ready to roll down at any time.

Peter Laird, who like Peter Jones is based at the University of Southern California, has shown the knock-on effects of this system in cancer cells. His team analysed patterns of DNA methylation in cancer cells, especially focusing on the promoters of tumour suppressor genes. Tumour suppressor genes whose histones are methylated by the EZH2 complex in ES cells were twelve times as likely to have abnormally high levels of DNA methylation as those genes that aren’t targeted by EZH2. Peter Laird described this effect very elegantly, saying, ‘reversible gene repression is replaced by permanent silencing, locking the cell into a perpetual state of self-renewal and thereby predisposing to subsequent malignant transformation [sic].’[199] This is consistent with the idea that there is a stem cell aspect to cancer. If cells are locked into a stem cell-like state, where they can’t differentiate into cells at the bottom of the epigenetic landscape, they will be very dangerous because they will always be able to keep on dividing to form yet more cells like themselves.

Jean-Pierre Issa has described the genes that are epigenetically silenced in colon cancer as the gatekeepers. They are frequently genes whose normal role is to move the cells away from self-renewal, and into fully differentiated cell types. Inactivation of these genes in cancer locks the cells in a self-renewing stem cell-like state. This creates a population of cells that are able to keep dividing, keep accumulating further epigenetic changes and mutations, and keep inching towards a full-blown cancer state[200].

When we visualise the cells in Waddington’s landscape, it’s quite difficult to visualise the ones that linger somewhere near the top. That’s because instinctively we know that that’s a really unstable place to be. A ball that has started rolling down a slope is always likely to keep going, unless something can hold it back. And even if such a ball does come to a halt, there’s always the chance it will start moving again, rolling on down that hill.

What holds cells in this teetering position? In 2006, a group headed by Eric Lander at the Broad Institute in Boston, found at least part of the answer. A key set of genes in ES cells, the pluripotent cells we have come to know so well, were found to have a really strange histone modification pattern. These were genes that were very important for controlling if an ES cell stayed pluripotent, or differentiated. Histone H3K4 was methylated at these genes, which normally is associated with switching on gene expression. H3K27 was also methylated. This is normally associated with switching off gene expression. So, which modification would turn out to be stronger? Would the genes be switched on or off?

The answer turned out to be both. Or neither, depending on which way we look at it. These genes were in a state called ‘poised’. Given the slightest encouragement – a change in culture conditions that pushed cells towards differentiation for example – one or other of these methylations was lost. The gene was fully switched on, or strongly repressed, depending on the epigenetic modification[201].

This is really important in cancer. Stephen Baylin is the third person, along with Peter Jones and Jean-Pierre Issa, who has done so much to make epigenetic therapies a reality. He has shown that these poised histone modifications are found in early cancer stem cells and are really significant for setting the DNA methylation patterns in cancer cells[202].

Of course, other events must also be taking place. Many people do not develop cancer, no matter what age they live to. Something must happen in the people who do develop cancer, which results in the normal stem cell pattern getting subverted and hardened so that the cells are locked into their aggressively and abnormally proliferative state. We know that environment can have a substantial impact on cancer risk (just think of the hugely increased risk of lung cancer in smokers) but we’re not clear on how or if the environment intersects with these epigenetic processes.

There may also be an aspect of pure bad luck in who develops cancer. We probably all have random fluctuations in the levels, activity or localisation of proteins that target, write, interpret and erase our epigenetic codes. And there are the non-coding RNAs too.

The 3′ UTRs of both DNMT3A and DNMT3B mRNA contain binding sites for a family of miRNAs called miR-29. Normally, these miRNAs will bind to the DNMT3A and DNMT3B mRNA molecules and down-regulate them. In lung cancer, the levels of these miRNAs drop and as a consequence DNMT3A and DNMT3B mRNA and subsequently protein expression is elevated. This is likely to increase the amount of de novo methylation of susceptible tumour suppressor promoters[203].

It is likely that there will also be feedback loops between miRNAs and the epigenetic enzymes they control, if one component of the pathway becomes mis-regulated. This will reinforce abnormal control mechanisms in the cell, leading to yet another vicious cycle, and is shown in Figure 11.4. In this example, a miRNA regulates a specific epigenetic enzyme, which itself modifies the promoter of the miRNA. In this case, the epigenetic enzyme creates a repressive modification.

^{Figure 11.4 A positive feedback loop which constantly drives down expression of a miRNA which would normally control expression of an epigenetic enzyme that creates a repressed chromatin state.}

There is still much that we need to understand if we are to develop the next generation of epigenetic drugs to treat cancer patients. We need to know which drugs will work best in which diseases and which patients will benefit the most. We want to be able to work this out in advance, so that we don’t have to hope for the best when running large numbers of clinical trials. At least 5-azacytidine and SAHA have given us the comfort of knowing that epigenetic therapy is possible in cancer, even if improvements are needed.

As we shall see in the next chapter, epigenetic problems are not restricted to cancer. But sadly we are even further away from knowing how to use epigenetic therapies in one of the greatest unmet clinical needs in the western world – psychiatric disease.