Will Regular Screening of Men with Genetic Mutations for PSA Lead to Better Prostate Cancer Diagnoses?

Will Regular Screening of Men with Genetic Mutations for PSA Lead to Better Prostate Cancer Diagnoses?

We all know enough about PSA. Basically, as we’ve heard in earlier talks, there are a number of issues with screening for PSA. What I’m interested in is trying to explore whether or not there are genetic markers for PSA we might be able to use to improve screening. This is going to be actually a presentation of new data that hasn’t been published yet.

If we think about PSA and prostate cancer, we know there are a lot of different factors. For example, Barack Obama is probably being screened in part because they just checked everything off, but he’s an African American male of 50-something years old.

Family history also puts you in a slightly different category for PSA screening. What we know about PSA is that it’s genetically determined. It actually has a fairly high inheritability. There’s been a study showing that the inheritability of PSA is about 45%. That means the variation between all the men in this room and our PSA levels, almost half of are due to our inherited genetics. There’s gene KLK3, which encodes the PSA protein, which has been shown to be very strongly associated with PSA levels.

The really complicating factor here though is that we’ve got this weird situation where we don’t know whether these genetic factors affect prostate cancer or independently affect PSA. If they’re working through prostate cancer, then they’re really prostate cancer genes, not PSA genes. What I mean by that is if they increased PSA levels through their effect on the disease, then those are things that even though there’s variation between everyone in the room in our PSA levels, it’s due to actually cancer.

This red line here, these are the things that are inherent differences in PSA levels that have nothing at all to do with prostate cancer. If we could figure out what these were, we could actually adjust men’s PSA levels based on their genetic background.

DeCODE worked on this about five years ago and published a paper presenting I think it was four genes, including KLK3, for PSA. They showed how you could actually improve biopsy outcomes by leveraging the information from these four genes. The problem with only looking at four genes is that something like PSA and prostate cancer, these are very complex traits which we know are due not to just one major gene in most individuals, but due to lots of genes, maybe 100 or even more than 100. To actually get at this question, we needed to look at a much larger population that deCODE looked at because a lot of these genes have very small effects on PSA levels.

What I did was to get data from this big Kaiser cohort that’s been genotyped. They’ve run a genome-wide SNP array to look at single nucleotide polymorphisms on over 100,000 people. These individuals are primarily Caucasian. I pulled out the 43,000 men. Most of them were controlled. We had about 35,000 controls, men without prostate cancer.

Kaiser had been regularly screening for PSA, so these guys had a lot of PSA measurements, almost 300,000 total PSA measurements in these 35,000 men. We were actually able to correlate the relationship between all of these genes, genome-wide, and their repeated total PSA levels.

We could also look at other measures of PSA, which we’ve talked about a little bit this morning, but we’re going to focus just on total PSA right now.

This is actually what’s called a Manhattan plot. People publish these with genome-wide association studies. Down here on the bottom are just the different chromosomes. On this axis are P-values. These are increasingly small P-values. It’s called a Manhattan plot because you’re looking for peaks. That shows you what’s associated with here in this case PSA. We just studied controls and found a lot of things. This line here is what’s considered genome-wide significant.

This huge peak here is KLK3. No surprise. That has a P-value of less than 10 times to the -200. That’s 200 zeroes before the first number. That’s how small the P-value is and that’s reflecting how strong this is for PSA and how big the sample size is.

If you look more closely, I’ve got these color coded. All of the lines here that are pink were associated previously with prostate cancer. The things that are purple were the ones that deCODE and others pulled out as being PSA-specific genes. Things that are actually black and gray are all new things that we found in these data. These are all novel hits for PSA in the controls.

You might say maybe all of those novel things are just prostate cancer hits. We were able to actually go look in our cases and do genome-wide association study for prostate cancer and compared those two different results. This line is just showing P-values from a prostate cancer association study versus a PSA study. These things here circled are the things that aren’t associated with prostate cancer, but are strongly associated with PSA. We can see that there are some things that are prostate specific, some things that are both, and then quite a few that are independently associated just with PSA levels.

What we found was there were 31 different genes associated with PSA levels. Seven had been previously associated and then 24 novel. These were genome-wide significant, independent novel hits for PSA levels.

Following kind of what deCODE did with their four or five genes, we then went back and saw what happened if we genetically adjusted PSA levels. Basically, we would take a man’s observed PSA level and then divide it through by this adjustment value. The adjustment is just the product of these 31 different variants and which PSA locus is carried by that individual at each variant divided by the population average.

Just as an example, 4 KLK3. Let’s say a man had a PSA of 4 and that he carried two copies of this KLK3 gene. Well, the KLK average in the population–this is a very common variant, so it’s 1.48. The adjustment here would be 1.54 divided by 1.48, 1.04. His PSA would actually be changed or adjusted down from 4.0 to 3.8. You can either think of it as adjusting their PSA level or shifting what cutoff they should have.

This is a line graph just showing. We discovered all these in the controls and then we applied them to the cases in this large population. What this is showing is just the log PSA levels–this is a line here for 4–unadjusted and then adjusted for these genetic PSA variants. You can see the lines kind of go all over the place, but there’s a bit of a trend downward.

Kind of hard to see much there, so what we did instead was look at kind of classifying people. Here are their observed original PSA levels for everyone and then this is what they were classified with once you did this genetic adjustment. We could’ve used any cut point here. I’m just using 4 as an example. Then 10% of the men with a PSA greater than 4 on the original measure total PSA were then reclassified as having a genetically-adjusted PSA as being less than 4. Then 4% were reclassified as having a higher PSA level.

This is in the entire cohort. What was more interesting here was what happened when we stratified by whether or not they were a case or control, whether or not they had been diagnosed with prostate cancer or not. When we focus on the controls only, 15% were reclassified. It was still 4% of the low being reclassified as high; 15% with high were reclassified as having a low PSA, where the case is it was only 2.4%. This is a huge difference. Then 15% of the controls were classified from having a high PSA to a low using 4 as a cut point, whereas only 2.4% of the cases.

The last question we asked was whether or not PSA or this PSA prime, adjusted PSA, better predicts Gleason among the cases. We did a regression model of Gleason on PSA or PSA prime. They’re both, no surprise, highly significant predictors of Gleason score. But the PSA prime was actually a better predictor. We can calculate this kind of metric of how well they predict from the regression model.

There are kind of three main take-home points here. First of all, this genome-wide association, a very large genome-wide association study, replicated and identified many of the PSA, SNPs. Adjusting PSA level shifts down a larger percentage of men without prostate cancer than men with prostate cancer and PSA prime is more predictive of Gleason.

We’ve been talking a lot this morning at the breakfast session about using genetic information to determine diagnostics and treatment. Here actually we can use genetic information to try determine PSA levels and the relevance of them to prostate cancer. Hopefully if we actually use this, this might help us improve PSA screening and decrease unnecessary biopsies for men who produce high PSA because that’s what they produce. It’s due to their genetic factors.

In contrast, an increase in the necessary biopsies in men who are low PSA producers. Ultimately, the hope here is to reduce the total number of biopsies and improve the sensitivity and specificity of PSA.

Finally, just acknowledging especially the Kaiser group that put together this big cohort, which was genotype at UCSF. That’s it. Thanks.