Tag Archives: Cancer Genome Atlas

So much work, so little progress

Two years ago, I found going to a memorial service for a friend and classmate who died of Alzheimer’s curiously uplifting  (see the link at the end). The disease is far from ignored. A monster review in Neuron vol. 97 pp. 32 – 58 ’18 — http://www.cell.com/neuron/fulltext/S0896-6273(17)31081-4  contained references to over 400 research articles half of them published since January 2013.

Still I found it quite depressing.  Tons of work and tons findings, and yet no coherent path to the cause (or causes); something absolutely necessary for a rational treatment, unless we somehow stumble into a therapy.

In a way it’s like cancer.  The cancer genome atlas intensively studied the genome of various cancers, looking for ‘the’ or ‘the set of’ causative mutations.  They found way too much.  The average colon and breast cancer had an average of 93 mutated genes, of which 11 were thought to be cancer promoting.  Not only that, but the same 11 were not consistent from tumor to tumor.

So it is with this epic review.  Which of the myriad findings described are causative of the disease and which are responses of the nervous system to the ’cause’ (or causes).

In the review the authors posit that Alzheimer’s disease is due to failure of ‘homeostatic systems’ that maintain a ‘set point’ of neuronal firing.  Unfortunately what is measured to determine the set point isn’t known. This seems to be an example of redefining a question into an answer.  Clearly if you juice up neuronal firing rates by stimulation they come back down, or if you inhibit them, they come back up.  So you can operationally define set point without defining it mechanistically.   It must be due to some sort of feedback on whatever it is that is sensing ‘the set point’ , but what is it that is being sensed?

The following is from an earlier post but is quite relevant to homeostasis and set points.

The whole notion of control in the presence of feedback is far from clear cut.  Here’s the story of the first inklings of feedback in endocrinology.  I watched it happen.

Endocrinology was pretty simple in med school back in the 60s. All the target endocrine glands (ovary, adrenal, thyroid, etc.) were controlled by the pituitary; a gland about the size of a marble sitting an inch or so directly behind the bridge of your nose. The pituitary released a variety of hormones into the blood (one or more for each target gland) telling the target glands to secrete, and secrete they did. That’s why the pituitary was called the master gland back then.  The master gland ruled.

Things became a bit more complicated when it was found that a small (4 grams or so out of 1500) part of the brain called the hypothalamus sitting just above the pituitary was really in control, telling the pituitary what and when to secrete. Subsequently it was found that the hormones secreted by the target glands (thyroid, ovary, etc.) were getting into the hypothalamus and altering its effects on the pituitary. Estrogen is one example. Any notion of simple control vanished into an ambiguous miasma of setpoints, influences and equilibria. Goodbye linearity and simple notions of causation.

As soon as feedback (or simultaneous influence) enters the picture it becomes like the three body problem in physics, where 3 objects influence each other’s motion at the same time by the gravitational force. As John Gribbin (former science writer at Natureand now prolific author) said in his book ‘Deep Simplicity’, “It’s important to appreciate, though, that the lack of solutions to the three-body problem is not caused by our human deficiencies as mathematicians; it is built into the laws of mathematics.” The physics problem is actually much easier than endocrinology, because we know the exact strength and form of the gravitational force.



The butterfly effect in cancer

Fans of Chaos know all about the butterfly, where a tiny change in air current produced by a butterfly’s wings in Brazil leads to a typhoon in Java. Could such a thing happen in cell biology? [ Proc. Natl. Acad. Sci. vol. 112 pp. 1131 – 1136 ’15 ] comes close.

The Cancer Genome Project has spent a ton of money looking at all the mutations of all our protein coding genes which occur in various types of cancers. It was criticized as we already knew that cancer is effectively a hypermutable state, and that it would just prove the obvious. Well it did, but it also showed us just what a formidable problem cancer actually is.

For instance [ Nature vol. 489 pp. 519 – 525 ’12 ] is report from the Cancer Genome Atlas of 178 cases of squamous cell cancer of the lung. There are a mean of 360 exonic mutations, 165 genomic rearrangements, and 323 copy number alterations per tumor. The technical details in the rest of the paragraph can be safely ignored but the point is that there no consistent pattern of mutation was found (except for p53 which is mutated in over 50% of all types of cancer, which we knew long before the Cancer Genome Atlas). Recurrent mutations were found in 11 genes. p53 was mutated in nearly all. Previously unreported loss of function mutations were seen in the class I major histocompatibility (HLA-A). Several pathways were altered relatively consistently (NFE2L2, KEAP1 in 34%, squamous differentiation genes in 44%, PI3K genes in 47% and CDKN2A and RB1 in 72%). EGFR and kRAS mutations are rare in squamous cell cancer of the lung (but quite common in adenocarcinoma). Alterations in FGFR are quite common in squamous cell carcinomas.

This sort of thing (which has been found in all the many types of tumors studied by the Cancer Genome Atlas) lead to a degree of hopelessness in looking for the holy grail of a single ‘driver mutation’ which leads to cancer with its attendant genomic instability.

All is not lost however.

MCF-10A is an immortalized epithelial cell line derived from human breast tissue. It is capable of continuous growth, but is far from normal: (1) an abnormal complement of chromosomes ) (2) threefold amplification of the MYC oncogene, and (3) deletion of a known tumor suppressor . It does lack some mutations found in breast cancer. For instance, the Epidermal Growth Factor Receptor 2 (ERRBB2) is not amplified. The cell line doesen’t express the estrogen and progesterone receptors — making it similar to triple negative breast cancer.

A single amino acid mutation (Arginine for Histidine at amino acid #1047 ) in the catalytic subunit of a very important protein kinase (p110alpha of the PIK3CA gene) was put into the MCF-10A cell line (which they call MCF-1A-H1047R). The mutation was chosen because it is one of the most frequently encountered cancer specific mutations known. Exome sequencing of the entire genome showed that this was the only change — but the control sequences outside the exons weren’t studied, a classic case of the protein centric style of molecular biology.

In the (admittedly not completely normal) cell line, the mutation produced a cellular reorganization that far exceeds the known signaling activities of PI3K. The proetins expressed were stimilar to the protein and RNA signatures of basal breast cancer. The changes far exceeded the known effects of PIK3CA signaling. The phosphoproteins of MCF-1A-H1047R are extremely different. Inhibitors of the kinase induce only a partial reversion to the normal phenotype.

They plan to study the epigenome. This is signifcant as breast cancers are said in the paper to have tons of mutations changing amino acids in proteins (4,000 per tumor). In my opinion they should do whole genome sequencing of MCF-A1-H1047R as well.

The mutant becomes fully transformed whan a second mutation (of KRAS, an oncogene) is put in. This allows them to form tumors in nude mice. Recall that nude mice (another rodent beloved of experimental biologists — see the previous post on the Naked Mole Rat) has a very limited immune system, allowing grafts of human cells to take root and proliferate.

How close the initial cell line is to normal is another matter. Work on a similar cell line the (3T3 fibroblast) has been criticized because that cell is so close to neoplastic. At least the mutant MCF-1A-H1047R cells aren’t truly neoplastic as they won’t produce tumors in nude mice. However, mutating just one more gene (KRAS) turns MCF-1A-H1047R malignant when transplanted.

The paper is also useful for showing how little we really understand about cause and effect in the cell. PI3K has been intensively studied for years because it is one of the major players telling cells to grow in size rather than divide. And yet “the mutation produced a cellular reorganization that far exceeds the known signaling activities of PI3K”

Can losing one gene do all that? Yes it can — there’s still hope

The Cancer Genome Atlas has dashed our hopes of finding ‘the’ cause of cancer. It has sequenced the genomes of a large number of cancers — the following paper looked at 21 tumor types sequencing the protein coding parts (exomes) of 4,742 specimens, along with that of normal tissues [ Nature vol. 505 pp. 495 – 501 ’14 ].

The problem is that lots of mutations have been found in every type of cancer studied this way.

The following is typical — 178 cases of lung cancer (squamous cell variety) were studied. Some 360 mutations in exons, 165 genomic rearrangements, and 323 copy number alterations were found — but this doesn’t represent the results for the 178 cases as a whole. This was the average amount of genomic mayhem seen in each individual tumor . How do you find ‘the’ cause of the cancer in this mess? One way might be to find a gene mutated in all 178 cases (e. g. recurrent mutations). This would be the holy grail — the mutation driving cancer formation, the rest being the chaff of the well known genomic instability due to the high mutation rate of cancer cells. They found 11 such genes, but they were far from mutated in all cases. Pretty depressing isn’t it?

A recent paper [ Proc. Natl. Acad. Sci. vol. 111 pp. 14009 – 14010, E4066 – E4075 ’14 ] gave an example of a huge number of changes in the clinical activity of a cancer cell line due to the functional loss of just one gene (called COSMC). Here’s what happened. In a pancreatic cancer cell line, COSMC knockout produced malignant xenografts (e.g. placing the cells in an immunodeficient animal and watching what happens), which could be reversed by reintroduction of COSMC. The changes include (1) increased proliferation, (2)loss of contact inhibition of growth, (3) loss of tissue architecture, (4) less basement membrane adhesion and (5) invasive growth — remarkable that knocking out just one gene could do so much. Perhaps not a driver mutation, but certainly a delicious drug target. Before getting too excited, remember that this occurred in a cell line which was cancerous to begin with.

The quick and dirty explanation of what is going on is that COSMC is a protein chaperone for an enzyme adding a sugar to proteins destined either for secretion or for insertion into the cell membrane. Lose COSMC and the whole pattern of sugar attachments to these proteins changes. There are a lot of proteins modified by adding sugars (glycosylated proteins), actually 446 of them, with 1,471 sites for this to happen.

The rest of the post is for the cognoscenti and concerns the gory details.

From the paper itself — “Neoplastic transformation of human cells is virtually always associated with aberrant glycosylation of proteins and lipids.” The most frequently seen glycophenotype are the Tn and STn carbohydrate epitopes of epithelial cell cancers. They arise when mucin-type O-linked glycans (normally more complex) are truncated so that only a single -N-acetylgalactosamine (Tn) or N-acetylgalactosamine modified with sialic acid (STn) remains attached to the protein by a serine or a threonine. There are ‘up to’ 20 GalNAc transferases adding GalNAc to serine or threonine. Overall there are some 200 glycosyltransferase found in the secretory pathway. In most cases the GalNAc is modified with beta 1 –> 3 galactose by a single enzyme (called C1GalT1). This reaction is dependent on COSMC, a protein chaperone.

Although there weren’t mutations in the glycosyltransferases studied in 46 cases of pancreatic cancer, 40% of them showed hypermethylation of the COSMC (e.g. methylated cytosines in the promoter region, which shut down transcription of COSMC). This correlated with expression of truncated O-Glycans (e.g. the Tn and STn antigens) and loss of C1GalT expression.