The Power of Flower


We know we don’t ‘understand cancer’ — for if we did we would at least be well on the way to preventing the ten million annual deaths from these diseases and perhaps even stymieing their appearance in the first place. But at least, after many years of toil by thousands of curious souls, we might feel brave enough to describe the key steps by which it comes about.

Here goes!

Our genetic material, DNA, carries a code of four different units (bases) that enables cells to make twenty-thousand or so different types of proteins. Collectively these make cells — and hence us — ‘work’. An indicator of protein power is that we grow from single, fertilized cells to adults with 50 trillion cells. That phenomenal expansion involves, of course, cells growing and dividing to make more of themselves — and, along the way, a bit of cell death too. The fact that there are nearly eight billion people on planet earth testifies to the staggering precision with which these proteins act.

Nobody’s perfect

As sports fans will know, the most successful captain in the history of Australian rugby, John Eales, was nicknamed ‘Nobody’ because ‘Nobody’s perfect’. Well, you might care to debate the infallibility of your sporting heroes but when it comes to their molecular machinery, wondrous though it is, perfect it is not.

Evidence: from the teeming eight billion there emerges every year 18 million new cancer cases (that’s about one in every 444). And cancers are, of course, abnormal cell growth: cells growing faster than they should or growing at the wrong time or in the wrong place — any of which means that some of the masterful proteins have suffered a bit of a malfunction, as the computer geeks might say.

How can that happen?

Abnormal protein activity arises from changes in DNA (mutations) that corrupt the normal code to produce proteins of greater or lesser activity or even completely novel proteins.

These mutations may be great or small: changes in just one base or seismic fragmentation of entire chromosomes. But the key upshot is that the cell grows abnormally because regulatory proteins within the cell have altered activity. Mutations can also affect how the cell ‘talks’ to the outside world, that is, the chemical signals it releases to, for example, block immune system killing of cancer cells.

Clear so far?

Mutations can change how cells proliferate, setting them free of normal controls and launching their career as tumour cells and, in addition, they can influence the cell’s environment in favour of unrestricted growth.

The latter tells us that cancer cells cooperate with other types of cell to advance their cause but now comes a remarkable discovery of a hitherto unsuspected type of cellular collaboration. It’s from Esha Madan, Eduardo Moreno and colleagues from Lisbon, Arkansas, St. Louis, Indianapolis, Omaha, Dartmouth College, Zurich and Sapporo who followed up a long-known effect in fruit flies (Drosophila) whereby the cells can self-select for fitness to survive.

Notwithstanding the fact that flies do it, the idea of a kind of ‘cell fitness test’ is novel as far as human cells go — but it shouldn’t really surprise us, not least because our immune system (the adaptive immune system) features much cooperation between different types of cell to bring about the detection and removal of foreign or damaged cells.

Blooming science

So it’s been known for over forty years that Drosophila carries out cell selection based on a ‘fitness fingerprint’ that enables it to prevent developmental errors and to replace old tissues with new. However, it took until 2009 before the critical protein was discovered and, because mutant forms of this protein gave rise to abnormally shaped nerve cells that looked like bunches of flowers, Chi-Kuang Yao and colleagues called the gene flower‘.

Cells can make different versions of flower proteins (by alternative splicing of the gene) the critical ones being termed ‘winner’ and ‘loser’ because when cells carrying winner come into contact with cells bearing loser the latter die and the winners, well, they win by dividing and filling up the space created by the death of losers.

The effect is so dramatic that Madan and colleagues were able to make some stunning movies of the switch in cell populations that occured when they grew human breast cancer cells engineered to express different version of flower tagged with red or green fluorescent labels.

Shift in cell populations caused by two types of flower proteins. 

Above are images at time zero and 24 h later of co-cultures of cells expressing  green and red proteins (losers and winners). From Madan et al. 2019.

Click here to see the movie on the Nature website.

Winner takes almost all

The video shows high-resolution live cell imaging over a 24 hour period compressed into a few seconds. Cells expressing the green protein (hFwe1 (GFP)) were co-cultured with red cells (hFwe2 (RFP)). Greens are losers, reds winners. As the movie progresses you can see the cell population shifting from mainly green to almost entirely red, as the first and last frames (above) show.

How does flower power work?

Flower proteins form channels across the outer membrane of the cell that permit calcium flow, and it was abnormal calcium signalling that caused flowers to bloom in Drosophila nerves. It would be reasonable to assume that changes in calcium levels are behind the effects of flower on cancer cells. Reasonable but wrong, for Madan & Co were able to rule out this explanation. At the moment we’re left with the rather vague idea that flower proteins mediate competitive interactions between cells and these determine whether cells thrive and proliferate or wither and die.

Does this really happen in human cancers?

Madan and colleagues looked at malignant and benign human tumours and found that there was more ‘winner’ flower protein in the former than the latter and that ‘loser’ levels were higher in normal cells next to a tumour than further away. Both consistent with the notion that tumour cells express winner and this induces loser in nearby normal cells leading to their death. What’s more they reproduced this effect in mice by transplanting human breast cancer cells expressing winner.

So there we are! After all this time a variant on how cancer cells can manipulate their surroundings to promote the development of tumours. Remarkable though this finding is, in a way that is familiar it’s just the beginning of this story. We don’t know how flower proteins work in giving cancers a helping hand and we don’t know how effective they are. Until we answer those questions it would be premature to try to come up with therapies to block their effect.

But it is a moment to sit back and reflect on the astonishing complexity of living organisms and their continuing capacity to surprise.


Madan, E. et al. (2019). Flower isoforms promote competitive growth in cancer. Nature 572, 260-264.

Yao, C-K., et al., (2009). A synaptic vesicle-associated Ca2+ channel promotes endocytosis and couples exocytosis to endocytosis. Cell 138, 947–960.

Breaking Up Is Hard To Do


Thus Neil Sedaka, the American pop songster back in the 60s. He was crooning about hearts of course but since then we’ve discovered that for our genetic hearts — our DNA — breaking up is not that tough and indeed it’s quite common.

A moving picture worth a thousand words

When I’m trying to explain cancer to non-scientists I often begin by showing a short movie of a cell in the final stages of dividing to form two identical daughter cells. This is the process called mitosis and the end-game is the exciting bit because the cell’s genetic material, its DNA, has been duplicated and the two identical sets of chromosomes are lined up in the middle of the cell. There ensues a mighty tug-of-war as cables (strands of proteins) are attached to the chromosomes to rip them apart, providing a separate genome for each new cell when, shortly after, the parent cell splits into two. When viewed as a speeded-up movie it’s incredibly dramatic and violent — which is why I show it because it’s easy to see how things could go wrong to create broken chromosomes or an unequal division of chromosomes (aneuploidy). It’s the flip side if you like to the single base changes (mutations) — the smallest damage DNA can suffer — that are a common feature of cancers.

In Heir of the Dog we showed pictures of normal and cancerous chromosomes that had been tagged with coloured markers to illustrate the quite staggering extent of “chromosome shuffling” that can occur.

Nothing new there

We’ve known about aneuploidy for a long time. Over 20 years ago Bert Vogelstein and his colleagues showed that the cells in most bowel cancers have different numbers of chromosomes and we know now that chromosomal instability is present in most solid tumours (90%).

Knowing it happens is one thing: being able to track it in real time to see how it happens is another. This difficulty has recently been overcome by Ana C. F. Bolhaqueiro and her colleagues from the Universities of Utrecht and Groningen who took human colorectal tumour cells and grew them in a cell culture system in the laboratory that permits 3D growth — giving rise to clumps of cells called organoids.

Scheme representing how cells grown as a 3D clump (organoid) can be sampled to follow chromosomal changes. Cells were taken from human colon tumours and from adjacent normal tissue and grown in dishes. The cells were labelled with a fluorescent tag to enable individual chromosomes to been seen by microscopy as the cells divided. At time intervals single cells were selected and sequenced to track changes in DNA. From Johnson and McClelland 2019.

Genetic evolution in real time

As the above scheme shows, the idea of organoids is that their cells grow and divide so that at any time you can select a sample and look at what’s happening to its DNA. Furthermore the DNA can be sequenced to pinpoint precisely the genetic changes that have occurred.

It turned out that cancer cells often make mistakes in apportioning DNA between daughter cells whereas such errors are rare in normal, healthy cells.

It should be said that whilst these errors are common in human colon cancers, a subset of these tumours do not show chromosomal instability but rather have a high frequency of small mutations (called microsatellite instability). Another example of how in cancer there’s usually more than one way of getting to the same end.

Building bridges …

The most common type of gross chromosomal abnormality occurs when chromosomes fuse via their sticky ends to give what are called chromatin bridges (chromatin just means DNA complete with all the proteins normally attached to it). Other errors can give rise to a chromosome that’s become isolated — called a lagging chromosome, it’s a bit like a sheep that has wandered off from the rest of the flock. As the cell finally divides and the daughter cells move apart, DNA bridges undergo random fragmentation.

… but where to …

Little is known about how cells deal with aneuploidy and the extent to which it drives tumour development. This study showed that variation in chromosome number depends on the rate at which chromosomal instability develops and the capacity of a cell to survive in the face of changes in chromosome number. More generally for the future, it shows that the organoid approach offers an intriguing opening for exploring this facet of cancer.


Bolhaqueiro, A.C.F. et al. (2019). Ongoing chromosomal instability and karyotype evolution in human colorectal cancer organoids. Nature Genetics 51, 824–834.

Lengauer, C. et al., (1997). Genetic instability in colorectal cancers. Nature 386, 623-627.

Johnson, S.C. and McClelland, S.E. (2019). Watching cancer cells evolve. Nature 570, 166-167.

What’s New in Breast Cancers?


One of the best-known things about cancer is that it’s good to catch it early. By that, of course, we don’t mean that you should make an effort to get cancer when you’re young but that, if it does arise it’s a good idea to find out before the initial growth has spread to other places in the body. That’s because surgery and drug treatments are very effective at dealing with ‘primary’ tumours — so much so that over 90% of cancer deaths are caused by cells wandering away from primaries to form secondary growths — a process called metastasis — that are very difficult to treat.

The importance of tumour spreading is shown by the figures for 5-year survival rates. Overall in the USA it’s 90% but this figure falls to below 30% for cancers that have metastasized (e.g., to the lungs, liver or bones). For breast cancer the 5-year survival rate is 99% if it is first detected only in the breast (most cases (62%) are diagnosed at this stage). If it’s spread to blood and lymph vessels in the breast the 5-year survival rate is 85%, dropping to 27% if it’s reached distant parts of the body.

What’s the cause of the problem?

The other thing most people know about cancers is that they’re caused by damage to our genetic material — DNA — that is, by mutations. This raises the obvious notion that secondary tumours might be difficult to deal with because they have accumulated extra mutations compared with those in primaries. And indeed, there have been several studies pointing to just that.

Very recently, however, François Bertucci, Fabrice André and their colleagues in various institutes in France, Switzerland and the USA have mapped in detail the critical alterations in DNA that accumulate as different types of breast cancers develop from early tumours to late, metastatic forms. As is the way these days, their paper contains masses of data but the easiest form of the message comes in the shape of ‘violin plots’. These show the spread of results  — in this case the number of mutations per length of DNA.

Metastatic tumours have a bigger mutational load than early tumours. These plots are for one type of breast tumour (HR+/HER2−) and show results for 381 metastases and 501 early tumours. Red dots = median values: these are the “middle” values rather than an average (or mean) and they show a clear upwards shift in burden as early tumours evolve into metastases. From Bertucci et al., 2019.

The violin plots above are for one subtype of breast cancer (HR+/HER2−). Recall that breast tumours are often defined by which of three types of protein can be detected on the surface of the cells: these are ‘receptors’ that have binding sites for the hormones estrogen and progesterone and for human epidermal growth factor. Hence they are denoted as hormone receptors (HRs) and (human) epidermal growth factor receptor-2 (HER2). Thus tumours may have HRs and HER2 (HR+, HER2+) or various receptors may be undetectable. Triple negative breast cancer (TNBC) is an absence of receptors for both estrogen and progesterone and for HER2.

The plots clearly show an increase in mutation load with progression from early to metastatic tumours (on average from 2.4 to 3.8 mutations per megabase of DNA). Looking at individual genes, nine ‘drivers’ emerged that were more frequently mutated in HR+/HER2− metastatic breast cancers (we described ‘driver’ and ‘passenger’ mutations in Taking Aim at Cancer’s Heart).

So what?

For now these findings give us just a little more insight into what goes on at the molecular level to turn a primary into a metastatic tumour. The fact that some of the acquired driver mutations are associated with poor patient survival offers some guidance as to treatment options.

Don’t get carried away

It’s a familiar story in this field: another small advance in piecing together the jigsaw that is cancer. It doesn’t offer any immediate advance in treatment — mainly because most of the nine ‘driver’ genes identified are tumour suppressors — i.e. they normally act as brakes on cell growth. Mutations knock out that activity and at the moment there is no therapeutic method for reversing such mutations. (The other main class of cancer promoters is ‘oncogenes‘ in which mutations cause hyper-activity).

But such steps are important. The young slave girl in Uncle Tom’s Cabin gave us the phrase “grew like Topsy” — meaning unplanned growth. Cancer growth is indeed unplanned and a bit like Topsy but it’s driven by molecular forces and only through untangling these can we begin to design therapies in a rational way.


Bertucci, F. et al. (2019). Genomic characterization of metastatic breast cancers. Nature 569, 560–564.

Food Fix For Pharma Failure


If you held a global quiz, Question: “Which biological molecules can you name?” I guess, setting aside ‘DNA‘, the top two would be insulin and glucose. Why might that be? Well, the World Health Organization reckons diabetes is the seventh leading cause of death in the world. The number of people with diabetes has quadrupled in the last 30 years to over 420 million and, together with high levels of blood glucose (sugar), it kills nearly four million a year.

There are two forms of diabetes: in both the level of glucose in the blood is too high. That’s normally regulated by the hormone insulin, made in the pancreas. In Type 1 diabetes insulin isn’t made at all. In Type 2 insulin is made but doesn’t work properly.

When insulin is released into the bloodstream it can ‘talk’ to cells by binding to protein receptors that span cell membranes. Insulin sticks to the outside, the receptor changes shape and that switches on signalling pathways inside the cell. One of these causes transporter molecules to move into the cell membrane so that they can carry glucose from the blood into the cell. When insulin doesn’t work it is this circuit that’s disrupted.

Insulin signalling. Insulin binds to its receptor which transmits a signal across the cell membrane, leading to the activation of the enzyme PIK3. This leads indirectly to the movement of glucose transporter proteins to the cell membrane and influx of glucose.

So the key thing is that, under normal conditions, when the level of blood glucose rises (after eating) insulin is released from the pancreas. Its action (via insulin receptors on target tissues e.g., liver, muscle and fat) promotes glucose uptake and restores normal blood glucose levels. In diabetes, one way or another, this control is compromised.

Global expansion

Across most of the world the incidence of diabetes, obesity and cancer are rising in parallel. In the developed world most people are aware of the link between diabetes and weight: about 90% of adults with diabetes are overweight or obese. Over 2 billion adults (about one third of the world population) are overweight and nearly one third of these (31%) are obese — more than the number who are underweight. The cause and effect here is that obesity promotes long-term inflammation and insulin resistance — leading to Type 2 diabetes.

Including cancer

The first person who seems to have spotted a possible connection between diabetes and cancer was the 19th-century French surgeon Theodore Tuffier. He was a pioneer of lung and heart surgery and of spinal anaesthesia and he’s also a footnote in the history of art by virtue of having once owned A Young Girl Reading, one of the more famous oil paintings produced by the prolific 18th-century artist Jean-Honoré Fragonard (if you want to see it head for the National Gallery of Art in Washington DC). Tuffier noticed that having type 2 diabetes increased the chances of patients getting some forms of cancer and pondered whether there was a relationship between diabetes and cancer.

It was a good question then but it’s an even better one now when this duo have become dominant causes of morbidity and mortality worldwide.

We now know that being overweight increases the risk of a wide range of cancers including two of the most common types — breast and bowel cancers. Unsurprisingly, the evidence is also clear that diabetes (primarily type 2) is associated with increased risk for some cancers (liver, pancreas, endometrium, colon and rectum, breast, bladder).

With all this inter-connecting it’s perhaps not surprising that the pathway by which insulin regulates glucose also talks to signalling cascades involved in cell survival, growth and proliferation — in other words, potential cancer initiators. The central player in all this is a protein called PIK3 (it’s an enzyme that adds phosphate groups (so it’s a ‘kinase’) to a lipid called phosphatidylinositol bisphosphate, an oily, water-soluble component of the plasma membrane). It’s turned out that PIK3 is one of the most commonly mutated genes in human cancers — e.g., PIK3 mutations occur in 25–40% of all human breast cancers.

Signalling pathways switched on by mutant PIK3. A critical upshot is the activation of cell survival and growth that leads to cancer.

Accordingly, much effort has gone into producing drugs to block the action of PIK3 (or other steps in this signal pathway). The problem is that these have worked as cancer treatments either very variably or not at all.

The difficulty arises from the inter-connectivity of signalling that we’ve just described: a drug blocking insulin signalling causes the liver to release glucose and prevents muscle and fats cells from taking up glucose. Result: blood sugar levels rise (hyperglycaemia). This effect is usually transient as the pancreas makes more insulin that restores normal glucose levels.

Blockade of mutant PIK3 by an inhibitor. This blocks the route to cancer but glucose levels rise in the circulation (hyperglycaemia) promoting the release of insulin (top). Insulin can now signal through the normal pathway (bottom), overcoming the effect of the anti-cancer drug. Note that the cell has two copies of the PIK3 gene/protein, one of which is mutated, the other remaining normal.

Is our journey really necessary?

By now you might be wondering whether there is anything that makes grappling with insulin signaling worth the bother. Well, there is — and here it is. It’s a recent piece of work by Benjamin Hopkins, Lewis Cantley and colleagues at Weill Cornell Medicine, New York who looked at ways of getting round the insulin feedback response so that the effect of PIK3 inhibitors could be boosted.

Sketch showing the effect of diet on the potency of an anti-cancer drug in mice. The red line represents normal tumour growth. The black line shows the effect of PIK3 blockade when the mice are on a ketogenic diet: tumour growth is suppressed. On a normal diet the drug alone has only a slight effect on tumour growth. Similar results were obtained in a variety of model tumours (Hopkins et al., 2018).

They first showed that, in a range of model tumours in mice, insulin feedback caused by blockade of PIK3 was sufficient to switch on signalling even in the continued presence of anti-PIK3 drugs. The really brilliant result was that changing the diet of the mice could offset this effect. Switching the mice to a high-fat, adequate-protein, low-carbohydrate (sugar) diet essentially stopped the growth of tumours driven by mutant PIK3 treated with PIK3 blockers. This is a ketogenic (or keto) diet, the idea being to deplete the store of glucose in the liver and hence limit the rise in blood glucose following PIK3 blockade.

Giving the mice insulin after the drug drastically reduces the effect of the PIK3 inhibitor, supporting the idea that that a keto diet improves responses to PIK3 inhibitors by reducing blood insulin and hence its capacity to switch on signalling in tumour cells.

A few weeks prior to the publication of the PIK3 results another piece of work showed that adding the amino acid histidine to the diet of mice can increase the effectiveness of the drug methotrexate against leukemia. Methotrexate was one of the first anti-cancer agents to be made and has been in use for 70 years.

These are really remarkable results — as far as I know the first time diet has been shown to influence the efficacy of anti-cancer drugs. It doesn’t mean that all tumours with mutations in PIK3 have suddenly become curable or that the long-serving methotrexate is going to turn out to be a panacea after all — but it does suggest a way of improving the treatment of many types of tumour.


Hopkins, B.D. et al. (2018). Suppression of insulin feedback enhances the efficacy of PI3K inhibitors. Nature 560, 499-503.

Kanarek, N. et al. (2018). Histidine catabolism is a major determinant of methotrexate sensitivity. Nature 559, 632–636.

Taking Aim at Cancer’s Heart


Cancer is a unique paradox. At one level it’s as easy as can be to describe: damage to DNA (aka mutations) drives cells to behave abnormally — to make more of themselves when they shouldn’t.

But we all know that cancer’s fiendishly complicated — at least at the level of fine detail. Over the last decade or so the avalanche of sequenced DNA has revealed that every cell in a tumour is different: compare one cell to its neighbour and you’ll find variations in the individual units (the bases A, C, G & T) that make up the chains of DNA.

It’s a nightmare: every cancer is different so we need an infinite number of treatments to control or cure each one. Time to give up and retire to the pub.

Drivers and passengers

Not quite. DNA sequencing has also revealed that, amongst all the genetic mayhem, some mutations are more important than others. The movers and shakers have been dubbed ‘drivers’: those that come along for the ride are ‘passengers’. The hangers-on are heavily in the majority but, even so, several hundred drivers (i.e. mutated genes that give rise to abnormal proteins) have been identified. As it needs a group of drivers to work together to make a cancer we still have the problem that the number of critical combinations that can arise is essentially infinite.

One way of reducing the scale of the problem has been to look at what ‘driver’ proteins do in cells and to target those acting at key points to push cell proliferation beyond the normal.

Playing games

Just recently Giulio CaravagnaAndrea Sottoriva and colleagues at the Institute of Cancer Research, London and the University of Edinburgh have come up with a different approach. The idea goes back to the 1950s when a clever chap from Kansas by the name of Arthur Samuel came up with a program for IBM’s first commercial computer so that it could play draughts (checkers as our American friends call it) in its spare time. The program defined the patterns that could be formed by the pieces on the chequerboard so that, given enough of these, IBM 701 could indicate the optimal moves. Samuel called this machine learning, a precursor of the idea of artificial intelligence.

Perhaps the most famous moment in this saga came in 1997 when a later IBM computer, Deep Blue, beat the then world chess champion Garry Kasparov. Unsurprisingly, Kasparov was a bit miffed and accused IBM of cheating — to wit, getting a human to tell the machine what to do. Let’s hope that in the end he came to terms with the fact that Deep Blue could crank through 200 million positions per second and, however many games Grandmasters have in their heads, they can’t compete with that.

The cancer team realized that the mutations driving the evolution of cancer cells emerge as patterns in the sequence of DNA as a cell moves towards becoming independent of normal controls. Think of each cancer as a family tree of mutations, the key question being which branch leads to the most potent combination.

To pick out these patterns they applied a machine-learning approach known as transfer learning to the DNA sequences from a large number of cancers. They called this ‘repeated evolution in cancer’ — REVOLVER — aimed at picking out mutation patterns at the heart of cancer that foreshadow future genetic changes and can be used to predict how they will evolve.

Identifying patterns of mutation common to different tumours.

Samples are taken from different regions of a patient’s tumour (represented by the coloured dots). Their DNA sequences will have multiple variations that can mask underlying patterns of driver mutations present in some subgroups. The five trees show mutations picked up in those patients. REVOLVER uses transfer learning to screen the sequence data from many patients and pull out evolutionary trajectories shared by subgroups. The dotted red lines highlight common patterns that are represented in the lower strip. From Caravagna et al. 2018.

REVOLVER was applied to sequences from lung, breast, kidney and bowel cancers but there’s no reason it shouldn’t work with other tumours. The big attraction is that if these mini-sequence mutation patterns can be associated generally with how a given tumour develops they should help to inform treatment options and predict survival.

We have in the past referred to the ways cancers evolve as ‘genetic roulette’ — so perhaps it’s appropriate if game-playing computer programs turn out to be useful in teasing out behavioural clues.


Caravagna, G. et al. (2018). Detecting repeated cancer evolution from multi-region tumor sequencing data. Nature Methods 15, 707–714.

Now wash your hands!


You must have spent the last 20 years on a distant planet if you’re unaware that we’re heading for Antibiotic Armaggedon — the rise of “Superbugs”, i.e., bacteria resistant to once-successful medication. Microbes resistant to multiple antimicrobials are called multidrug resistant. It’s a desperate matter because it means trivial infections may become fatal and currently safe surgical procedures may become dangerous.

Time-line of the discovery of different antibiotic classes in clinical use. The key point is that the last antibiotic class to become a successful treatment was discovered in 1987.

What’s the problem?
It’s 30 years since we came up a new class of antibiotics. The golden age launched by Fleming’s celebrated discovery of penicillin is long gone and while the discovery curve has drifted ever downwards since 1960 the bugs have been busy.

Just how busy a bug can be was shown by a large-scale experiment carried out by Roy Kishony and friends. They built a “Mega-Plate” — a Petri Dish 2 ft by 4 ft filled with a jelly for the bacteria to grow in. The bugs were seeded into channels at either end so they would grow towards the middle. The only thing stopping them was four channels dosed with antibiotic at increasing concentrations — 10 times more in each successive channel.

The bugs grow until they hit a wall of antibiotic. There they pause for a think — and, after a bit, an intrepid little group start to make their way into the higher dose of drug. Gradually the number of groups expand until a tidal wave sweeps over that barrier. This is repeated at each new ‘wall’ — four times until the whole tray is a bug fest.

When they pause at each new ‘wall’ they’re not ‘thinking’ of course. They’re just picking up random mutations in their DNA until they are able to advance into the high drug environment. So this experiment is a fantastic visual display of bugs becoming drug-resistant. And it’s terrifying because it takes about 11 days for them to overcome four levels of drug. It’s even more scary in the speeded-up movie as that lasts less than two minutes.

Sound familiar?
It should do as this is a cancer column and readers will know that cancers arise by picking up mutations. To highlight the similarities the picture below is the left-hand half of the bug tray with new colonies shown as linked dots. You could perfectly well think of these as early stage cancer cells acquiring mutations in ‘driver’ genes that push them towards tumour formation.

So that’s pretty scary too and the only good news is that animal cells reproduce much more slowly than bacteria. The fastest they can manage is about 48 hours to grow and divide into two new cells and for many it’s much slower than that. Bugs, on the other hand, can do it in 20 minutes if you feed them enough of the right stuff.

Which is why we don’t all get zonked by cancer at an early age.

The evolution of bacteria on a “Mega-Plate” Petri Dish. The vertical red lines mark the boundaries of increasing antibiotic concentrations. You could equally think of each dot that represents a new bacterial colony being early stage cancer cells acquiring mutations in ‘driver’ genes (white arrows) that push them towards tumour formation. From Roy Kishony’s Laboratory at Harvard Medical School.

Enough of that!
But for once I don’t want to talk about cancer but about a really fascinating piece of work that caught my eye in the journal Cell Reports. It’s by Gianni Panagiotou, Kang Kang and colleagues from The University of Hong Kong and The Hans Knöll Institute, Jena, Germany and it’s all about their travels on the Hong Kong MTR (Mass Transit Railway). This is the network of over 200 km of railway lines with 159 stations that serves the urbanised areas of Hong Kong IslandKowloon, and the New Territories and has a cross- border connection to the neighboring city of Shenzhen in mainland China.

An MTR train on the Tung Chung line that links Lantau Island with Hong Kong Island.

Being scientists of course they weren’t just having a day out. They wanted to know the contents of the microbiome that they and their fellow travellers picked up on the palms of their hands when riding the rails. ‘Microbiome’ means all of the collection of microorganisms — though in practice this is almost entirely bacteria. So they swabbed the palms of volunteers and then threw the full power of modern DNA sequencing and genetic analysis at what they’d scraped off. Or, as they put it: “We conducted a metagenomic study of the Hong Kong MTR system.”

And if you’re thinking it might be possible to take a trip on the Hong Kong Metro without grabbing a handrail or otherwise engaging in what on the London Underground used to be called ‘strap-hanging’ you clearly haven’t tried it!

Hong Kong MTR.


The MTR System and Sampling Procedure. Left: The eight urban lines studied: the Airport Express line and Disneyland Resort branch were excluded. The Central-Hong Kong station and the cross-border rail stations connecting with the MTR and the Shenzhen metro system are labeled. Right: The sampling procedure included handwashing, handrail touching for 30 min and swabbing. From Kang et al. 2018.

Hold very tight please! 

It’s going to become a seriously bumpy ride. The major findings were:

  1. Four groups (phyla) of bacteria dominated: Actinobacteria [51%], Proteobacteria [27%], Firmicutes [11%] and Bacteroidetes [2%]. Followers of this blog will be delighted to spot the last two (B & F) as we’ve met them several times before (in Hitchhiker Or Driver?, Fast Food Fix Focuses on Fibre, Our Inner Self, The Best Laid Plans In Mice and Men, and, of course, in it’s a small world) — that’s how important they are in the context of cancer.
  2. The dominant organism (29% of the community) was P. acnes (one of the Actinobacteria — it’s the bug linked to the skin condition of acne).
  3. Some non-human-associated species (e.g., soil organisms) also popped up that varied enormously in amount from day to day — perhaps because of weather conditions (e.g., humidity).
  4. Variation in the make-up of the microbial communities picked up depended, more than anything else, on the time of day. There was a marked decrease in diversity in afternoon samples compared with those taken in the morning.
  5. Specific species of bacteria associated with individual metro lines. That is, sets of bug types are relatively abundant on a given line compared with all other lines, giving a kind of line-specific signature — though the distinction declines from morning to afternoon. The most physically isolated line, MOS (Ma On Shan), had a greater number of signature species. The MOS runs entirely above ground alongside the Shing Mun Channel, a polluted brackish river, and its ‘signature’ includes bacteria found in sewage.
  6. All of which brings us to bugs with antibiotic resistance genes (ARGs). Across the network 136 ARG families were detected including 24 that are clinically important. Strikingly, lines closer to Shenzhen (ER (East Rail) and MOS) tend to have higher ARG input during the day. Critically, the ER line a.m. signatures become p.m.-enriched in all MTR lines far from Shenzhen — that is, these ARG families spread over the network during the day.

Simplified map of the Hong Kong MTR indicating how antibiotic resistance genes spread during the day from the ER and MOS lines to the entire network. Tetracycline resistance genes: tetA, tetO, tetRRPP and tetMWOS; vancomycin resistance genes: vanC, vanX. From Kang et al. 2018.

These results clearly suggest that the ER line, the only cross-border line linked to mainland China, may be a source of clinically important ARGs, especially against tetracycline, a commonly used antibiotic in China’s swine feedlots. Antibiotics, including tetracycline, can be detected in the soil in the Pearl River Delta area where the cities of Hong Kong and Shenzhen are located.

It should be said that this is by no means the first survey of bugs on rails. Notable ones have looked at the New York and Boston metro systems and they too revealed the potential health risks of the bug communities found on trains and in the stations, including the presence of pathogens and antibiotic resistance. The Boston survey highlighted that different types of materials have surfaces that are preferred by different microbes with high variation in functional capacity and pathogenic potential.

One obvious suggestion from these studies is that world-wide we could do a lot to improve sanitation, e.g., by having hand sanitizer dispensers in all sensible places (at the exits of metro, railway and bike-sharing stations and airports and of course in hospitals). The Hong Kong data are seriously frightening and most people seem blissfully unaware that the invisible world they reveal carries the potential for the destruction of us all.

But, as ever, there’s two sides to the matter. We’ve evolved over millions of years to live with bugs and they with us. However you wash your hands you won’t get rid of every bug and anyway, as what’s-his-name almost says, “They’ll be back!” We all carry around micro-organisms that can be fatal if they get to the wrong place. But, if you’re reasonably fit, there’s a lot to be said for simply following sensible, basic hygiene rules with a philosophy of ‘live and let live.’

Have a nice day commuters, wherever you are!


Kang K., et al. (2018). The Environmental Exposures and Inner- and Intercity Traffic Flows of the Metro System May Contribute to the Skin Microbiome and Resistome. Cell Reports 24, 1190–1202.

Wu, N., Qiao, M., Zhang, B., Cheng, W.D., and Zhu, Y.G. (2010). Abundance and diversity of tetracycline resistance genes in soils adjacent to representative swine feedlots in China. Environ. Sci. Technol. 44, 6933–6939.

Li, Y.W., Wu, X.L., Mo, C.H., Tai, Y.P., Huang, X.P., and Xiang, L. (2011). Investigation of sulfonamide, tetracycline, and quinolone antibiotics in vegetable farmland soil in the Pearl River Delta area, southern China. J. Agric. Food Chem. 59, 7268–7276.

Leung, M.H., Wilkins, D., Li, E.K., Kong, F.K., and Lee, P.K. (2014). Indoor-air microbiome in an urban subway network: diversity and dynamics. Appl. Environ. Microbiol. 80, 6760–6770.

Robertson, C.E., Baumgartner, L.K., Harris, J.K., Peterson, K.L., Stevens, M.J., Frank, D.N., and Pace, N.R. (2013). Culture-independent analysis of aerosol microbiology in a metropolitan subway system. Appl. Environ. Microbiol. 79, 3485–3493.

Afshinnekoo, E., Meydan, C., Chowdhury, S., Jaroudi, D., Boyer, C., Bernstein, N., Maritz, J.M., Reeves, D., Gandara, J., Chhangawala, S., et al. (2015). Geospatial Resolution of Human and Bacterial Diversity with City-Scale Metagenomics. Cell Syst 1, 72–87.

Hsu, T., Joice, R., Vallarino, J., Abu-Ali, G., Hartmann, E.M., Shafquat, A., Du- Long, C., Baranowski, C., Gevers, D., Green, J.L., et al. (2016). Urban Transit System Microbial Communities Differ by Surface Type and Interaction with Humans and the Environment. mSystems 1, e00018–e00016.

Caveat Emptor


It must be unprecedented for publication of a scientific research paper to make a big impact on a significant sector of the stock market. But, in these days of ‘spin-off’ companies and the promise of unimaginable riches from the application of molecular biology to every facet of medicine and biology, perhaps it was only a matter of time. Well, the time came with a bang this June when the journal Nature Medicine published two papers from different groups describing essentially the same findings. Result: three companies (CRISPR Therapeutics, Editas Medicine and Intellia) lost about 10% of their stock market value.

I should say that a former student of mine, Anthony Davies, who runs the Californian company Dark Horse Consulting Inc., mentioned these papers to me before I’d spotted them.

What on earth had they found that so scared the punters?

Well, they’d looked in some detail at CRISPR/Cas9, a method for specifically altering genes within organisms (that we described in Re-writing the Manual of Life).

Over the last five years it’s become the most widely used form of gene editing (see, e.g., Seeing a New World and Making Movies in DNA) and, as one of the hottest potatoes in science, the subject of fierce feuding over legal rights, who did what and who’s going to get a Nobel Prize. Yes, scientists do squabbling as well as anyone when the stakes are high.

Nifty though CRISPR/Cas9 is, it has not worked well in stem cells — these are the cells that can keep on making more of themselves and can turn themselves in other types of cell (i.e., differentiate — which is why they’re sometimes called pluripotent stem cells). And that’s a bit of a stumbling block because, if you want to correct a genetic disease by replacing a defective gene with one that’s OK, stem cells are a very attractive target.

Robert Ihry and colleagues at the Novartis Institutes for Biomedical Research got over this problem by modifying the Cas9 DNA construct so that it was incorporated into over 80% of stem cells and, moreover, they could switch it on by the addition of a drug. Turning on the enzyme Cas9 to make double-strand breaks in DNA in such a high proportion of cells revealed very clearly that this killed most of them.

When cells start dying the prime suspect is always P53, a so-called tumour suppressor gene, switched on in response to DNA damage. The p53 protein can activate a programme of cell suicide if the DNA cannot be adequately repaired, thereby preventing the propagation of mutations and the development of cancer. Sure enough, Ihry et al. showed that in stem cells a single cut is enough to turn on P53 — in other words, these cells are extremely sensitive to DNA damage.

Gene editing by Cas9 turns on P53 expression. Left: control cells with no activation of double strand DNA breaks; right: P53 expression (green fluorescence) several days after switching on expression of the Cas9 enzyme. Scale bar = 100 micrometers. From Ihry et al., 2018.

In a corresponding study Emma Haapaniemi and colleagues from the Karolinska Institute and the University of Cambridge, using a different type of cell (a mutated line that keeps on proliferating), showed that blocking P53 (hence preventing the damage response) improves the efficiency of genome editing. Good if you want precision genome editing by risky as it leaves the cell vulnerable to tumour-promoting mutations.

Time to buy?!

As ever, “Let the buyer beware” and this certainly isn’t a suggestion that you get on the line to your stockbroker. These results may have hit share prices but they really aren’t a surprise. What would you expect when you charge uninvited into a cell with a molecular bomb — albeit one as smart as CRISPR/Cas9. The cell responds to the DNA damage as it’s evolved to do — and we’ve known for a long time that P53 activation is exquisitely sensitive: one double-strand break in DNA is enough to turn it on. If the damage can’t be repaired P53’s job is to drive the cell to suicide — a perfect system to prevent mutations accumulating that might lead to cancer. The high sensitivity of stem cells may have evolved because they can develop into every type of cell — thus any fault could be very serious for the organism.

It’s nearly 40 years since P53 was discovered but for all the effort (over 45,000 research papers with P53 in the title) we’re still remarkably ignorant of how this “Guardian of the Genome” really works. By comparison gene editing, and CRISPR/Cas9 in particular, is in its infancy. It’s a wonderful technique and it may yet be possible to get round the problem of the DNA damage response. It may even turn out that DNA can be edited without making double strand breaks.

So maybe don’t rush to buy gene therapy shares — or to sell them. As the Harvard geneticist George Church put it “The stock market isn’t a reflection of the future.” Mind you, as a founder of Editas Medicine he’d certainly hope not.


Ihry, R.J. et al. (2018). p53 inhibits CRISPR–Cas9 engineering in human pluripotent stem cells. Nature Medicine, 1–8.

Haapaniemi, E. et al. (2018). CRISPR–Cas9 genome editing induces a p53-mediated DNA damage response. Nature Medicine (2018) 11 June 2018.

Fantastic Stuff


It certainly is for Judy Perkins, a lady from Florida, who is the subject of a research paper published last week in the journal Nature Medicine by Nikolaos Zacharakis, Steven Rosenberg and their colleagues at the National Cancer Institute in Bethesda, Maryland. Having reached a point where she was enduring pain and facing death from metastatic breast cancer, the paper notes that she has undergone “complete durable regression … now ongoing for over 22 months.”  Wow! Hard to even begin to imagine how she must feel — or, for that matter, the team that engineered this outcome.

How was it done?

Well, it’s a very good example of what I do tend to go on about in these pages — namely that science is almost never about ‘ground-breaking breakthroughs’ or ‘Eureka’ moments. It creeps along in tiny steps, sideways, backwards and sometimes even forwards.

You may recall that in Self Help – Part 2, talking about ‘personalized medicine’, we described how in one version of cancer immunotherapy a sample of a patient’s white blood cells (T lymphocytes) is grown in the lab. This is a way of either getting more immune cells that can target the patient’s tumour or of being able to modify the cells by genetic engineering. One approach is to engineer cells to make receptors on their surface that target them to the tumour cell surface. Put these cells back into the patient and, with luck, you get better tumour cell killing.

An extra step (Gosh! Wonderful GOSH) enabled novel genes to be engineered into the white cells.

The Shape of Things to Come? took a further small step when DNA sequencing was used to identify mutations that gave rise to new proteins in tumour cells (called tumour-associated antigens or ‘neoantigens’ — molecular flags on the cell surface that can provoke an immune response – i.e., the host makes antibody proteins that react with (stick to) the antigens). Charlie Swanton and his colleagues from University College London and Cancer Research UK used this method for two samples of lung cancer, growing them in the lab to expand the population and testing how good these tumour-infiltrating cells were at recognizing the abnormal proteins (neo-antigens) on cancer cells.

Now Zacharakis & Friends followed this lead: they sequenced DNA from the tumour tissue to pinpoint the main mutations and screened the immune cells they’d grown in the lab to find which sub-populations were best at attacking the tumour cells. Expand those cells, infuse into the patient and keep your fingers crossed.

Adoptive cell transfer. This is the scheme from Self Help – Part 2 with the extra step (A) of sequencing the breast tumour. Four mutant proteins were found and tumour-infiltrating lymphocytes reactive against these mutant versions were identified, expanded in culture and infused into the patient.


What’s next?

The last step with the fingers was important because there’s almost always an element of luck in these things. For example, a patient may not make enough T lymphocytes to obtain an effective inoculum. But, regardless of the limitations, it’s what scientists call ‘proof-of-principle’. If it works once it’ll work again. It’s just a matter of slogging away at the fine details.

For Judy Perkins, of course, it’s about getting on with a life she’d prepared to leave — and perhaps, once in while, glancing in awe at a Nature Medicine paper that does not mention her by name but secures her own little niche in the history of cancer therapy.


McGranahan et al. (2016). Clonal neoantigens elicit T cell immunoreactivity and sensitivity to immune checkpoint blockade. Science 10.1126/science.aaf490 (2016).

Zacharakis, N. et al. (2018). Immune recognition of somatic mutations leading to complete durable regression in metastatic breast cancer. Nature Medicine 04 June 2018.

Another Fine Mess


Did you guess from the title that this short piece is about the seeming inability of the British Government to run well, most things but especially IT programmes? Of course you did! Provoked by the latest National Health Service furore. In case you’ve been away with the fairies for a bit, a major cock-up in its computer system has just come to light whereby, between 2009 and 2018, it failed to invite 450,000 women between the ages of 68 and 71 for breast screening. Secretary of State for Health, Jeremy Hunt (our man usually on hand with a can of gasoline when there’s a fire), told Parliament that “there may be between 135 and 270 women who had their lives shortened”. Cue: uproar, headlines: HUNDREDS of British women have died of breast cancer (Daily Express), etc.

Logo credit: Breast Cancer Action

I’ve been reluctant to join in because I’ve said all I think is worth saying about breast cancer screening in two earlier pieces (Risk Assessment and Behind the Screen). Reading them again I thought they were a reasonable summary and I don’t think there’s anything new to add. However, this is  a cancer blog and it’s a story that’s made big headlines so I feel honour-bound to offer a brief comment — in addition to sympathizing with the women and families who have been caused much distress.

My reaction was that Hunt was misguided in mentioning specific numbers — not only because he was asking for trouble from the press but mainly because the evidence that screening itself saves lives is highly questionable. For an expert view on this my Cambridge colleague David Spiegelhalter, who is Professor for the Public Understanding of Risk, has analysed the facts behind breast screening with characteristic clarity in the New Scientist.

Anything to add?

I was relieved on re-reading Risk Assessment to see that I’d given considerable coverage to the report that had just come out (2014) from The Swiss Medical Board.  They’d reviewed the history of mammography screening, concluded that systematic screening might prevent about one breast cancer death for every 1000 women screened, noted that there was no evidence that overall mortality was affected and pointed out that false positive test results presented the risk of overdiagnosis.

In the USA, for example, over a 10-year course of annual screening beginning at 50 years of age, one breast-cancer death would have been prevented whilst between 490 and 670 women would have had a false positive mammogram calling for a repeat examination, 70 to 100 an unnecessary biopsy and between 3 and 14 would have been diagnosed with a cancer that would never have become a problem.

Needless to say, this landed the Swiss Big Cheeses in very hot water because there’s an awful lot of vested interests in screening and it’s sort of instinctive that it must be a good thing. But what’s great about science is that you can do experiments — here actually analysing the results of screening programmes — and quite often the results turn to be completely unexpected, as it did in this case where the bottom line was that mammography does more harm than good.

This has led to the recommendation that the current programmes in Switzerland should be phased out and not replaced.

So we’re all agreed then?

Of course not. In England the NHS recommendation remains that women aged 50 to 70 are offered mammography every three years — which is just as well or we’d have Hunt explaining the recent debacle as new initiative. The American Cancer Society “strongly” recommends regular screening mammography starting at age 45 and the National Cancer Institute refers to “experts” that recommend mammography every year starting at age 25 for women with mutations in their BRCA1 or BRCA2 genes.

The latter is really incredible because a study published in the British Medical Journal in 2012 found that these mutations made the carriers much more vulnerable to radiation-induced cancer. Specifically, women with BRCA 1/2 mutations who were exposed to diagnostic radiation (i.e. mammography) before the age of 30 were twice as likely to develop breast cancer, compared to those with normal BRCA genes.

They are susceptible to radiation that would not normally be considered dangerous because the two BRCA genes encode proteins involved in the repair of damaged DNA — and if that is defective you have a recipe for cancer.


So it’s probably true that the only undisputed fact is that we need much better ways for detecting cancers at an early stage of development. The best hope at the moment seems to be the liquid biopsy approach we described in Seeing the Invisible: A Cancer Early Warning System? but that’s still a long way from solving a general cancer problem, well illustrated by breast mammography.

No It Isn’t!


It’s great that newspapers carry the number of science items they do but, as regular readers will know, there’s nothing like the typical cancer headline to get me squawking ‘No it isn’t!” Step forward The Independent with the latest: “Major breakthrough in cancer care … groundbreaking international collaboration …”

Let’s be clear: the subject usually is interesting. In this case it certainly is and it deserves better headlines.

So what has happened?

A big flurry of research papers has just emerged from a joint project of the National Cancer Institute and the National Human Genome Research Institute to make something called The Cancer Genome Atlas (TCGA). This massive initiative is, of course, an offspring of the Human Genome Project, the first full sequencing of the 3,000 million base-pairs of human DNA, completed in 2003. The intervening 15 years have seen a technical revolution, perhaps unparalled in the history of science, such that now genomes can be sequenced in an hour or two for a few hundred dollars. TCGA began in 2006 with the aim of providing a genetic data-base for three cancer types: lung, ovarian, and glioblastoma. Such was its success that it soon expanded to a vast, comprehensive dataset of more than 11,000 cases across 33 tumor types, describing the variety of molecular changes that drive the cancers. The upshot is now being called the Pan-Cancer Atlas — PanCan Atlas, for short.

What do we need to know?

Fortunately not much of the humungous amounts of detail but the scheme below gives an inkling of the scale of this wonderful endeavour — it’s from a short, very readable summary by Carolyn Hutter and Jean Claude Zenklusen.

TCGA by numbers. The scale of the effort and output from The Cancer Genome Atlas. From Hutter and Zenklusen, 2018.

The first point is obvious: sequencing 11,000 paired tumour and normal tissue samples produced mind-boggling masses of data. 2.5 petabytes, in fact. If you have to think twice about your gigas and teras, 1 PB = 1,000,000,000,000,000 B, i.e. 1015 B or 1000 terabytes. A PB is sometimes called, apparently, a quadrillion — and, as the scheme helpfully notes, you’d need over 200,000 DVDs to store it.

The 33 different tumour types included all the common cancers (breast, bowel, lung, prostate, etc.) and 10 rare types.

The figure of seven data types refers to the variety of information accumulated in these studies (e.g., mutations that affect genes, epigenetic changes (DNA methylation), RNA and protein expression, duplication or deletion of stretches of DNA (copy number variation), etc.

After which it’s worth pausing for a moment to contemplate the effort and organization involved in collecting 11,000 paired samples, sequencing them and analyzing the output. It’s true that sequencing itself is now fairly routine, but that’s still an awful lot of experiments. But think for even longer about what’s gone into making some kind of sense of the monstrous amount of data generated.

And it’s important because?

The findings confirm a trend that has begun to emerge over the last few years, namely that the classification of cancers is being redefined. Traditionally they have been grouped on the basis of the tissue of origin (breast, bowel, etc.) but this will gradually be replaced by genetic grouping, reflecting the fact that seemingly unrelated cancers can be driven by common pathways.

The most encouraging thing to come out of the genetic changes driving these tumours is that for about half of them potential treatments are already available. That’s quite a surprise but it doesn’t mean that hitting those targets will actually work as anti-cancer strategies. Nevertheless, it’s a cheering point that the output of this phenomenal project may, as one of the papers noted, serve as a launching pad for real benefit in the not too distant future.

What should science journalists do to stop upsetting me?

Read the papers they comment on rather than simply relying on press releases, never use the words ‘breakthrough’ or ‘groundbreaking’ and grasp the point that science proceeds in very small steps, not always forward, governed by available methods. This work is quite staggering for it is on a scale that is close to unimaginable and, in the end, it will lead to treatments that will affect the lives of almost everyone — but it is just another example of science doing what science does.


Hutter, C. and Zenklusen, J.C. (2018). The Cancer Genome Atlas: Creating Lasting Value beyond Its Data. Cell 173, 283–285.

Hoadley, K.A. et al. (2018). Cell-of-Origin Patterns Dominate the Molecular Classification of 10,000 Tumors from 33 Types of Cancer. Cell 173, 291–304.

Hoadley, K.A. et al. (2014). Multiplatform Analysis of 12 Cancer Types Reveals Molecular Classification within and across Tissues of Origin. Cell 158, 929–944.