Hacker News new | past | comments | ask | show | jobs | submit login
Blood Test That Spots Tumor-Derived DNA in Early-Stage Cancers (hopkinsmedicine.org)
361 points by ncw96 on Aug 17, 2017 | hide | past | web | favorite | 63 comments



I work for another player Guardant Health. We are the Liquid Biopsy market leaders right now. We just raised $360M Series E from SoftBank.

If you find this type of thing interesting and want to be part of it, we are hiring lots of folks. My team is looking for bioinformaticians, Python hackers, and machine learning people. Please reach out to me if you want to know more jgourneau@guardanthealth.com


I'm not sure if anyone can claim to be the market leader in liquid biopsy right now. If raising money is your metric for success, then Grail has raised $1B and has the backing of the world's leading DNA sequencing company (Illumina).


Also, with currently available techniques, how difficult is this field really?


Different tumor types have different tumor DNA shedding rates -- this has impacts on test sensitivity. A false negative is not great when dealing with a cancer diagnostic. Additionally, and more challenging, is the variant classification problem. Not all mutations are created equally. Some are pathogenic, some of benign, and some have unknown significance. If you simply report the genetic alteration, and I believe this is the approach taken by Guardant, the onus is on the oncology care team to determine whether the alteration is clinically relevant and actionable. Companies like Myriad and Foundation Medicine provide guidance on alterations which makes this less of a problem. The alternative is for a caregiver to do homework on the alteration and determine whether it is meaningful (through a literature search or ClinVar database query).

Variant classification alone is an incredibly challenging problem. Oncologists are not geneticists, and most hospitals do not have in-house geneticists. Additionally, there is potential liability involved with mischaracterizing genetic alterations.


Guardant Health is a member of the Blood Profiling Atlas for Cancer (https://www.bloodpac.org/members/) too with a commitment to opening up data for faster discovery. So they're good people!

Now you have to wonder why John Hopkins isn't party to it...


Hate to be that guy, but as an alum I have to correct this. It's Johns* Hopkins.


Interesting. I always heard it as "John Hopkins", and I guess every time I read it I also auto-corrected it to "John" in my head. It's actually his great-grandmother's maiden name[1]

[1]http://www.hopkinsmedicine.org/about/history/history1.html


The company mentioned in the original article (and where a lot of the research was done) is a Hopkins spinout -- Personal Genome Diagnostics. They are indeed a member of BloodPAC.


This might be the first time in the 7 years I've my degree in it that someone has said they're actually "looking for bioinformaticians"...


So is Illumina basically the "selling pickaxes during a gold rush" of this space?


Yes. Other cool sequencing out there, Pac Bio being the most frequent contender, but illumina has their HiSeq and MiSeq systems in the vast majority of human molecular pathology labs. I may be biased as I trained under their chief geneticist, who's now at UCSD, but I'm fairly confident that's a safe claim. The big issue, as I understand it, is QC. QC for illumina is well understood. QC for Pac Bio can hard.


Additionally, circulating tumor DNA is fragmented to short lengths, so there is no point in doing long read sequencing such as PacBio or Nanopore.


All fair enough. The two big immediate challenges in the field are i) that the tumor-derived fraction of total cfDNA can be as low as 1:10000 (stage I) and ii) that it is difficult to make Illumina sequencing more accurate than 1 error in 1000 sequenced bases (in which case the 1:10000 signal is drowned out). This paper uses some clever statistical tricks to reduce Illimina sequencing error; one of these tricks is to leverage population information, i.e. the more samples you sequence the better your understanding of (non-cancer-associated) systematic errors. This follows a long tradition in statistical genetics of using multi-sample panels to improve analysis of individual samples. There are also biochemical approaches like SafeSeq or Duplex Sequencing to reduce sequencing error.

Not-so-obvious point #1 is that the presence of cancer-associated mutations in blood != cancer. You find cancer-associated mutations in the skin of older probands, and assumedly many of the sampling sites would never turn into melanomas. A more subtle point is that cfDNA is likely generated by dying cells, i.e. a weak cancer signature in blood might also be indicative of the immune system doing its job.

Point #2 is that it's not necessarily about individual mutations, which are, due to the signal-to-noise ratio alluded to above, difficult to pick up. One can also look at the total representation of certain genes in cfDNA (many cancers have gene amplifications or deletions, which are easier to pick up because they affect thousands of bases at the same time), and the positioning of individual sequenced molecules relative to the reference genome. It seems that these positions are correlated with gene activities (transcription) in the cells that the cfDNA comes from, and cancer cells have distinct patterns if gene activity.


There is also Freenome, which raised a $65m Series A to bring something similar to market:

> Last year, we raised $5.5 million to prove out the potential of this technology. Now, it’s time to make sure that it’s safe and ready for the broader population.

https://medium.com/freenome-stories/freenome-raises-65m-in-s...


There's a bunch of companies doing this.


I haven't read the referenced study, but I'm sure this is using the same (or very similar) cell free DNA (cfDNA) sequencing techniques currently used clinically for Non Invasive Prenatal Testing (NIPT) to screen for genetic defects such as trisomy 21 (Down Syndrome).

NIPT is a non-invasive blood screening test that is quickly becoming the clinical standard of care. Many insurance companies now cover the entire cost of NIPT screening for for at-risk pregnancies (e.g. women of "Advanced Maternal Age" (35yo+)). The debate is moving to whether it should be utilized/covered for average-risk pregnancies as well.

[1] http://capsprenatal.com/about-nipt/

[2] https://www.genomeweb.com/molecular-diagnostics/aetna-wont-c...


As the linked article doesn't name the paper, here's a link to it:

Title: Direct detection of early-stage cancers using circulating tumor DNA

Link (no paywall): http://stm.sciencemag.org/content/9/403/eaan2415


My super-quick skim of the paper shows that they are indeed using cfDNA. The novelty presented lies in the creation and application of a technique they dub "targeted error correction sequencing" (TEC-Seq).

Figure 2 shows the sequencing error rate reduced from around 1e-4 using conventional NGS to 1e-6 using TEC-Seq.

This increased sensitivity allows for discovery of early cancer at levels that are undetectable using existing cfDNA testing techniques.


It's "funny" how they take all their high quality enzymatic reagents from the company NEB and do this error-repair step (which something NEB has claimed and shown that people should be doing with their NGS sample), and whole they use a high fidelity polymerase (Phusion), they don't use the highest fidelity (e.g. NEB Q5 or Kapa HIFI). Also, NEB makes a specific kit to repair NGS samples (i.e. Called pre-CR). Combined this would have translated into a 10-fold lower error rate (so regular NGS 10^-4, this paper 10^-6, then 10^-7 with even better enzymes).

Targeted capture using Agilent's SureSelect is pretty common in the cancer field. Their custom implementation allows for oversampling (30,000x) which, along the enzymatic error correction, allows them a high confidence in the target they chose to evaluate.

Overall, I think this is neat.


Slowly but surely. This isn't even close to a real diagnostic, but it's a hopeful proof of concept.

I really do wish detection studies would publish a ROC curve, though, or at least d'.


Yeah false positive vs negative is really critical when you tell people they might have cancer... people will do crazy stuff.


Possibly a silly question, but is it possible for a 'healthy' person who doesn't have any cancer risk factors to get a test like this done?


It's not a silly question, it's an interesting one to ask.

The short answer is: probably not, but maybe!, as some of these tests and assays are sometimes run on "healthy" people as part of testing and validation. So you might be able to convince some of these labs or companies to "test" you. But my spouse, who is one of the authors of the OP paper, ran into this issue when one of the lab members tested "positive" for a cancer-associated mutation during early-phase development of some of the assays. (They were running the stuff on their own blood.) The team was kind of stumped: the tests at that point were intended for characterizing cancer, with only inklings of ideas that it could be used for screening for cancer in seemingly-healthy people. It highlights the challenge: then what?

The outstanding question in this space is indeed: then what? We're getting to the point where we can detect tumor DNA with extremely high sensitivity. But if you are found to have DNA in your blood that could be from cancer, then what?

Whole-body CT scan? PET scan? MRI? And you find a node, a nodule, a mass... then what? Biopsy it? And then what? Cut it out? Chemotherapy? Watch it?

Each step is fraught with cost and worry and risk and potential morbidity, eg CT scans irradiate you... which can cause cancer.

We have no idea what to do with the information at this point. In fact, in some cases (as mentioned in other threads) even when we KNOW you have a specific cancer, it turns out to be better to watch and wait rather than do anything, because it can be expensive and harmful and risky to go hunting, when the cancer turns out to be slow and predictable. But it took us decades to figure that out, and to figure out which this was the case for.

So I would advise not getting such a test anytime soon, even if you could. There's nothing to do with information at the moment. As for the lab member, they ended up not doing anything about that result.

But it is still a good question, because the only way we begin to find out what to do with this kind of information is to begin collecting it.


If I am diagnose as early cancer, it could be a good motivation to switch to a more healthy lifestyle: more sleep, less food, ...


But if you don't, it's not worth it ? Everyone could benefit from a healthier lifestyle. I have a hard time believing that would be a wake up call for a significant percentage of people (see smokers, obesity, ...).


I recently discussed this very question with a colleague, and we both agreed that it probably would be a "wake up call". Currently, we really lack quality heath data, and so the loudest pundit or charlatan is who people listen to. And what they mostly present are anecdotes. Having high-fidelity health data about me as an individual and as part of various cohort study groups would allow my doctor to say "This is you now. These are the various future you's if you pursue these different lifestyles".


If you have no other risk factors or symptoms, it's usually not advised to do random testing like that (basically because Bayes' Theorem).


Before you head to this, ask them for false positive information


you're probably better off if you don't; incidentalomas are a thing


According to Craig Venter, early detection is what we need to eliminate cancer:

https://youtu.be/iUqgTYbkHP8?t=15m37s

I guess most are treatable if caught early?


Yes and no. With some cancers like prostate cancers, we're finding and treating cancers which would never kill the patient. And the deadliest cancers are the ones that grow quickly, quickly enough that early detection is extremely difficult.

For other kinds of cancers, like cervical cancer, which grows extremely slowly, early detection is already proven to be a lifesaver.


Something tied to what you said that is often not appreciated by people is that early detection can actually cause harm. In the example you gave of slow growing prostate cancers, in many cases people will now undergo treatment due to the early detection and the treatment winds up being far worse than the cancer itself would ever have been.

As the technology improves it's getting better at making predictions on the severity of the cancer and that can be used to guide treatment, or lack thereof. It's still far from perfect however.


> early detection can actually cause harm. In the example you gave of slow growing prostate cancers, in many cases people will now undergo treatment due to the early detection and the treatment winds up being far worse than the cancer itself would ever have been.

I have never quite bought this argument. ISTM it's not the early detection that causes harm -- it's the inappropriate treatment of early detected cancer that causes harm.


Yes, that's exactly what it is. Thus the argument is that indirectly early detection causes harm. As I mentioned later on in my post a lot of current research is going into determining if and when treatment must be done instead of the more common behavior of just diving in wholeheartedly.


"Early detecting causes harm" sounds absurd. That's akin to saying that "eating food" (and not dying from hunger) leads to harm from cancer treatment.


We also have a case study of over diagnosis from Korea:

http://www.npr.org/sections/health-shots/2017/05/09/52756929...

"South Korea is the poster child for the problem of overdiagnosis" of thyroid cancer, Welch says. About 15 years ago, doctors there started a mass campaign to screen for thyroid cancer. That vastly increased the rate of thyroid cancer, to the point that it exceeded cases of breast cancer and other common malignancies.

Yet Welch notes that the mortality rate from this cancer didn't change at all. "So all these extra cases were cases of thyroid cancer that weren't destined to bother people," he said. Instead, that rash of overdiagnosis sent people into needless surgery.

Harmless thyroid tumors are surprisingly common. Decades ago, pathologists in Finland noticed that more than a third of the people they saw in autopsy had thyroid tumors, but the cancers had no negative consequences.


No, it's not absurd at all.

Cancer treatment is not benign and (believe it or not) not all cancers will always causes harm.

https://sciencebasedmedicine.org/the-uncertainty-surrounding...

For instance, in 2012 a study in the New England Journal of Medicine (NEJM) by Archie Bleyer and H. Gilbert Welch found that, while there had been a doubling in the number of cases of early stage breast cancer in the 30 years since mass mammographic screening programs had been instituted, this increase wasn’t associated with a comparable decrease in diagnoses of late stage cancers, as one would expect if early detection was taking early stage cancers out of the “cancer pool” by preventing their progression. That’s not to say that Bleyer and Welch didn’t find that late stage cancer diagnoses decreased, only that they didn’t decrease nearly as much as the diagnosis of early stage cancers increased, and they estimated the rate of overdiagnosis to be 31%. These results are in marked contrast to the promotion of mammography sometimes used by advocacy groups. Last year, the 25 year followup for the Canadian National Breast Screening Study (CNBSS) was published. The CNBSS is a large, randomized clinical trial started in the 1980s to examine the effect of mammographic screening on mortality. The conclusion thus far? That screening with mammography is not associated with a decrease in mortality from breast cancer. Naturally, there was pushback by radiology groups, but their arguments were, in general, not convincing. In any case, mammographic screening resulted in decreases in breast cancer mortality in randomized studies, but those studies were done decades ago, and treatments have improved markedly since, leaving open the question of whether it was the mammographic screening or better adjuvant treatments that caused the decrease in mortality from breast cancer that we have observed over the last 20 years.


To echo the other responses with a practical example, suppose you were to have a slow growing prostate cancer. Suppose that there was a 0% chance it'd actually cause you any real harm. This is a not at all uncommon scenario.

Now suppose an early screen detects this cancer. Since we still have difficulty discerning whether or not your cancer is really "bad" or not you'll likely undergo immediate treatment. This treatment leaves you feeling horribly ill for the duration with all sorts of side effects. After treatment you're left with urinary leakage and erectile dysfunction.

I don't know about you but if you were to tell me that I could live the rest of my days with a minor cancer growing in my body minding its own business I'd choose leaving it there vs the side effects of the treatment.


> And the deadliest cancers are the ones that grow quickly, quickly enough that early detection is extremely difficult.

If a blood test can detect it, couldn't it be economically feasible to have people do it monthly, with a mailed kit and return envelope? Doesn't seem like much more work than taking vitamins, certainly less than going to the gym.


How do you think your blood test can tell the difference between slow-growing and fast-growing prostate cancers? Most men die with small prostate tumors.

I'm sure there would be types of cancer that would benefit, but it would involve a lot of study of outcomes.


> How do you think your blood test can tell the difference between slow-growing and fast-growing prostate cancers?

Would it need to? If your test ever came back postive, you'd go into the doctor, presumably retest to confirm, and then start appropriate treatment, regardless of how fast-growing the discovered cancer was.

To be clear: I'm not arguing, just genuinely curious as this isn't a subject I know much about.


The appropriate treatment depends on the risk of death. Chemotherapy is among the worst - probably 20% of people who get chemo die because of the chemo. If there's a 20% chance that the cancer will kill you, there's zero utility to a treatment that has a 20% chance of killing you.

Since none of the treatments we have are 100% effective, they can be worse than useless, actually increasing the overall death rate, and ensuring that the end of life is long, drawn-out, and horribly painful.


Understood. But my original question was inspired by greg's claim that early diagnosis of fast-growing cancers typically won't help, because they're so fast. I was asking if it would feasible to do monthly at home blood tests as a way to detect even rapid growth cancers (and assuming that would be enough time).

The point you're making seems orthogonal to me. Unless you're claiming that treatment is never worthwhile for such cancers, no matter how early they're detected, but Greg's original statement seemed to imply otherwise.


I read Greg's claim (and it's similar to mine) as meaning that there's no test to distinguish between fatal cancers and benign cancers, and this test doesn't claim to be such a test. Even if you have a test that's 100% accurate at detecting cancer, it's only useful if it only detects cancers that are likely to be fatal, otherwise unnecessary treatment will be expensive, both in dollars and in deaths.


>Chemotherapy is among the worst - probably 20% of people who get chemo die because of the chemo.

That's likely not right. I would believe something more boxed in...like 20% of patients who die within 30 days of starting chemotherapy died from it versus the cancer.


That's a recipe for over-treatment, which is already commonly happening for prostate cancer. As I mentioned earlier, it's very important to study outcomes! And not just do what seems reasonable.


Meaning you'd be treating false positives? I assumed the blood tests were coarse-grained, but surely there are more reliable tests that you could do if you tested positive on a blood test. Or are there not?


my understanding is that a big part of determining the stage of a cancer is based on tumor size and spread.

https://www.cancer.gov/about-cancer/diagnosis-staging/stagin...

I think most treatment is focused on killing/removing tumor cells and reducing tumor, so the earlier stage a cancer is detected, the less tumor have to remove, and the better the chances of removing all the tumor so there's nothing left to grow, spread, or metastasize beyond the initial tumor site.


Apparently, 2/3 of cancers caught early are curable.

Couldn't find a direct source but I read that here https://www.technologyreview.com/s/607944/grails-1-billion-b...

"Grail Inc., a Silicon Valley startup that last year began laying out a bet that the “cure” for most cancers isn’t a new drug, but tests that catch tumors at an early stage when they can still be successfully treated."


Didn't Grail raise a billion dollars to do just this?


I guess what's new is that the scientists found a way to distinguish between cancer dna and other altered/mutated dna so they reduce the occurrence of false positives.

http://fortune.com/2017/06/03/liquid-biopsy-grail-blood-canc... https://www.technologyreview.com/s/607944/grails-1-billion-b...


This is a fairly crowded space.


Yes but grail is doing deep sequencing I believe.


That's exactly what they did in the OP: "the sequencing method is based on deep sequencing"


> They found none of the cancer-derived mutations among blood samples of 44 healthy individuals.

Is 98% specificity adequate for a cancer test?


Certainly not for a general screen, but then we just know that the specificity is in the range of (98%, 100%) now, not the exact specificity. And we don't know what other refinements might improve upon what they did here.

But their data is too valuable to hold back until they had the final result of a test that could be taken to the FDA and approved for general use. This is valuable science, that impacts the directions that others are taking with their research.

I'm not saying you're doing this, but it's a common trend on HN and for other non-scientist technical people to forget that scientific journals show the process of science, not completed science. A journal is not a textbook that one can pick up and read a complete, final story, where all the false starts and uninteresting tangents are skipped. Journals are a way for scientists to communicate about their current progress, and that's an essential part of the scientific process. There has to be room to say: "here's where we are." Especially when they are able to get to the point of where they are right now.

Also, this was published in a very high profile journal, on a highly competitive topic, so take it all with a huge grain of salt. The senior author of this paper does good work, and many people are talking about similar preliminary results, but it takes time to prove these things.


> I'm not saying you're doing this, but it's a common trend on HN and for other non-scientist technical people to forget that scientific journals show the process of science, not completed science

How much would it have cost them to increase the number of controls? How did they choose 44?


It costs a lot, unfortunately. Sequencing is still fairly expensive, and the methods in OP are fairly complex on top of that.

There's also the interesting question of: if the test came back positive in a seemingly-healthy control, how would you know if that was a true positive or a false positive? It's still early in the game as to how to work up a cancer detected only in blood. The gold standard is tissue biopsy, but how far are you willing to go to hunt that down? It's a tough one.

Source: My spouse is one of the authors of the study in question.


> if the test came back positive in a seemingly-healthy control, how would you know if that was a true positive or a false positive?

You can estimate the likelihood from cancer prevalance.


Where did the 98% figure come from?

95% confidence intervals from the binomial distribution are for the specificity to be between 92% and 100%.


When this test has a near 100% success rate, how does it help the patients? Can it really prevent cancer?


> Can it really prevent cancer?

No.

What this is, is a test to detect cancer. The cool thing is that it should detect various cancers early (even before symptoms are showing).

The earlier cancer is detected the sooner a patient can start treatment. And the earlier treatment starts the better the odds of surviving.

And as a blood test this could end up being a standard part of an annual checkup. Especially for patients at greater risk (eg. men over N years, smokers, etc).


Tangent: I'm invested in a small cap stock, Sophoris Bio, that's in a P2B study for prostrate cancer with a drug developed out of Johns Hopkins called PRX302 (Topsalysin).

That and the article about blood tests shows there's a lot they're working on for noninvasive or minimally invasive procedures to help prevent cancer early on.




Applications are open for YC Winter 2020

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: