Are we closer to personalized genomic medicine yet?

A large study of 200,000 people with breast, ovarian or prostate cancer and controls was reported widely last week as one of the most comprehensive studies ever of these diseases.  The study was carried out by the  Collaborative Oncological Gene-environment Study, COGS, and the thirteen papers published in five different journals were collated by Nature and presented in an open access site, with commentary.

The aim of the study was to identify risk factors with which to stratify the population by risk status, and thus to be able to determine who to most intensively screen to prevent or detect cancers early. They identified 74 new genetic loci (chromosome regions, whether or not an actual causal site has been found, as usually it hasn't, to date) associated with risk; 49 with breast cancer, 8 for ovarian cancer and 29 for prostate cancer.  There's a lot of material, and we can't cover it all, but we'll give it a go.  Keep in mind that they aren't yet saying they've identified genes 'for' these cancers, but that their genomewide associate studies have located chromosomal regions that are statistically associated with disease, and the assumption is that they are involved in affecting the probability or risk of disease.  

Prostate
The genetic loci that increase risk of prostate cancer that have been reported to date have been common and 'low-penetrance' alleles, meaning that even if a man carries one or more of them his risk of getting prostate cancer is low.  The new study identifies 29 new loci associated with susceptibility to prostate cancer.  They are less common than the previously identified risk factors, presumably found by the COGS now because their study is larger than previous studies.

The consortium stratified their sample into aggressive and non-aggressive disease and found different though overlapping patterns of loci associated with each.  Of the new loci, 16 are associated with both forms of the disease.  The researchers say that each of the loci contains 'plausible' candidate genes.  That is, the causal genes in the chromosomal location the disease mapped to have not yet been identified.  It is a slight stretch to say 'plausible' rather than 'possible', but that temptation's not unusual, and identifying the true causal genes can be tricky (e.g., see our previous posts on mapping genes for traits, here and here.)  But ok.  The paper discusses these candidates, but the consortium can't (yet) confirm any of them.

More than 70 loci have now been reported for prostate cancer, which, according to the researchers, explains "~30% of the familial risk for this disease." That means that this is found in those who already have a family history of risk--that is, an enriched subsample most likely to be carrying genetic risk factors. This study reports that those at highest genetic risk of prostate cancer, the top 1% of the risk distribution in the population, are at 4.7-fold higher risk than average.  High risk was considered to be aggressive prostate cancer or prostate cancer in patients younger than 55.

Essentially this means that if we tote up your alleles in these 70 regions, and add up their separate independently estimated effects, the net risk is 4.7 times higher than that of a random person in the population.  These are subtle and often slippery issues when it comes to assessing how important the actual risk assessment is--even if the method of computing summed risk is appropriate.
 
Breast and ovarian
Twenty-seven loci associated with familial breast cancer risk have previously been identified, accounting for about 30% of familial risk.  This study reports 49 new susceptibility loci, and estimates that approximately 1000 genes will eventually be found to be associated with risk of breast cancer. 

With respect to genes already known to be major risk factors for disease, the study looked for additional genetic markers in women who carry BRCA1 or 2 mutations, which put them at higher risk of breast and ovarian cancers than the general population.  They replicated previous findings, and also identified new 'modifier genes' that increase risk of breast cancer in BRCA1 carriers, and 8 new genes that increase risk of ovarian cancer, bringing the total to 10 and 12 known genes associated with breast and ovarian cancer, respectively. They identified a new risk allele specific to BRCA2 carriers, as well.  They also found 21 genetic loci associated with risk in east Asian women. 

The consortium reports that these results will one day allow those BRCA1 and 2 carriers at highest and lowest risk of developing cancer to be identified.  Risk estimates for BRCA2 carriers now range from "21–47% risk of developing breast cancer by the age of 80 years for the 5% of the BRCA2 mutation carriers at lowest risk compared to 83–100% risk for the 5% at highest risk" (source).

It is a somewhat separate question what the same non-BRCA sites do, if anything, in those not carrying BRCA mutations.

Gene-environment interactions and breast cancer
The consortium made a stab at addressing the question of gene-environment interaction, acknowledging that risk alleles are rarely determinative (they do apparently believe they've found at least one set of genes that may increase risk to 100% for some breast cancer).  Even the risk associated with BRCA1 and 2, among the most clear cut genetic risk factors known, has been shown to vary considerably by year of birth, presumably because of the interaction between the gene and some environmental risk factor.

King et al., 2003; Science 302:643-646

The study looked at whether relative risk associated with other risk alleles was modified by 10 factors already thought to be associated with risk: age at menarche, parity, breastfeeding, body mass index, height, oral contraceptive use, use of hormone replacement therapy after menopause, alcohol consumption, smoking, and exercise.

They replicated previously reported interactions between specific loci and parity and alcohol consumption, with risk increasing with lower parity and, they assume, a genetic variant of the LSP1 gene, and with a variant of CASP8 and alcohol consumption.

Import
A commentary in Nature Genetics discusses the public health implications of the work.  Remember that the aim of the study was to facilitate the identification, on an individual basis, of those at highest risk of breast, ovarian and prostate cancer, in order to prevent and detect disease.  But how to translate this to population public health measures is another question. 

But we've got a ways to go on this. Tests for prostate cancer are very unreliable, and there is no screen for ovarian cancer.  Whether a risk prediction is accurate may depend on whether the cancer is found because it is symptomatic or by extensive testing.  That's an issue since it's known that many or perhaps even most prostate cancers, and at least some breast cancers, regress on their own.

The chromosome regions now identified to increase risk of ovarian cancer are estimated to double the average population risk to just 3%.  The cost-effectiveness of mammography is still the subject of intense debate, and genetic susceptibility to cancer, even for those alleles known to be major risk factors, is never 100% and usually much lower. And testing carries its own risk, as does being told you're at higher risk--so such information had better be accurate if this is about public health!

Further, risk is always estimated from past environmental exposures, and future environments cannot be predicted.

Family history and the simulation of mendelism
We already know that family history is a good--generally, clearly one of the best predictors of these kinds of diseases.  So far, genetic risk studies haven't done much to improve on that; whether the current batch does isn't clear to us, at least not yet.

The other major risk factors are sex and age.  Generally these are far more powerful than individual gene identification, but of course the reason would be at least in part because family history reflects the presence of those genetic variants.  We also know that sampling and studying families with multiple cases enriches for whatever factors, genetic or environmental or even tendency to have testing done by high-grade clinics, in those family members.  This can make the presence of the disease seem Mendelian, as if due to a single or very few genes.  Such 'simulation of Mendelism' can raise the apparent risk of an identified gene shared by the affected family members to values far above those the same would have in the population at large.

This kind of bias is but one of several that make actual risk estimation very difficult, and GWAS type case-control studies are known theoretically and empirically to generate inflated estimates of effects.

So there is a lot that must follow these studies if they are to be judged to satisfy the media noise that they have generated.