The Peter Attia Drive - #213 ‒ Liquid biopsies and cancer detection ｜ Max Diehn, M.D. Ph.D.

00:00:00.000 Hey, everyone. Welcome to the drive podcast. I'm your host, Peter Atiyah. This podcast,

00:00:15.480 my website, and my weekly newsletter all focus on the goal of translating the science of longevity

00:00:19.800 into something accessible for everyone. Our goal is to provide the best content in health

00:00:24.600 and wellness, full stop. And we've assembled a great team of analysts to make this happen.

00:00:28.880 If you enjoy this podcast, we've created a membership program that brings you far more

00:00:33.280 in-depth content. If you want to take your knowledge of the space to the next level,

00:00:36.840 at the end of this episode, I'll explain what those benefits are. Or if you want to learn more now,

00:00:41.740 head over to peteratiyahmd.com forward slash subscribe. Now, without further delay,

00:00:47.740 here's today's episode. My guest this week is Max Dean. Max is a professor of radiation oncology,

00:00:55.960 vice chair of research, and division chief of radiation and cancer biology at Stanford

00:01:00.480 University. Max is a co-founder of Foresight Diagnostics, a precision medicine company

00:01:05.320 developing novel liquid biopsy tests for measurement of minimal residual disease. He's

00:01:09.620 also the co-founder of CyberMet, a company that applies data science for biomarker discovery.

00:01:14.540 Max also consults and advises a number of companies in similar spaces. Max's current research involves

00:01:20.560 the development of novel methods for detecting circulating tumor DNA in the blood of cancer

00:01:25.340 patients. He also works to understand cancer cells by identifying molecular pathways and genes

00:01:30.000 associated with disease. And he's interested in uncovering biomarkers that can predict response

00:01:34.240 to therapy or predict patient survival and return of disease as early as possible, which is something

00:01:41.100 we'll get into in the discussion so you can understand why it's so important to predict recurrence

00:01:44.960 as soon as it happens. Clinically, Max is a radiation oncologist, and he specializes in lung cancer.

00:01:50.860 He manages a broad clinical research portfolio, and he focuses on improving these personalized

00:01:56.340 therapies for patients with lung cancer. In this episode, we talk about a lot of things. First of

00:01:59.640 all, Max and I were also classmates in medical school, so we catch up a little bit on that,

00:02:04.020 and we talk about his background and how he became interested in liquid biopsies. We go into great

00:02:08.320 detail here on sensitivity, specificity, negative predictive value, positive predictive value. These are

00:02:12.600 things that everybody needs to understand if they want to be smart on diagnostics and if they want to

00:02:18.160 understand cancer screening. We talk about why these things are important and, in particular,

00:02:22.680 how they play into cancer screening, especially when it comes to understanding prevalence and

00:02:26.800 pretest probability. We spend some time talking about lung cancer, which is the number one killer

00:02:32.020 for both men and women. And it's not just a smoker's disease. Remember, 15% of people who die of lung cancer

00:02:38.620 have never smoked a cigarette in their life. So this is an important cancer, whether or not you are a

00:02:42.660 smoker or not. From there, we dive really deep into liquid biopsies, the landscape, the history,

00:02:47.540 the possible future of liquid biopsies. For me, this was the high point of this interview. In fact,

00:02:52.400 in preparing for this interview, I myself had to get a lot smarter in liquid biopsies. Certainly,

00:02:58.240 I know more about them than the average bird, and I've spent a lot of time looking at them over the

00:03:02.200 past two years. But I think in this episode, we get a lot more granular around the nuances of the

00:03:09.140 different ways in which not just we can look at circulating tumor cells versus cell-free DNA,

00:03:15.020 but when looking at cell-free DNA, what are the different methods that we can use

00:03:18.820 to predict if a cancer is present? In other words, how can we look at the actual genes from the actual

00:03:26.380 cancer that we know we're searching for versus in a screening situation when we don't know the gene

00:03:32.460 that we have to look for other clues? So we talk about these cell-free DNA RNA signatures.

00:03:37.860 We talk about methylation patterns. We talk about the importance of knowing

00:03:40.780 mutation information. We talk about the difference in some of the screenings being approved by the FDA

00:03:45.720 versus those that are being permitted to use for patients without FDA approval formally.

00:03:49.860 There's a lot packed into this episode, but it is truly one of the most important subjects

00:03:54.880 given the difficulty in treating cancer when it becomes advanced. So without further delay,

00:03:59.740 please enjoy my conversation with Max Dean.

00:04:07.600 Max, thanks so much for making time today. So wonderful to see you again. It's been a lot of

00:04:12.860 years, huh? It's been a long time. Actually, I can't remember the last time we saw each other, but

00:04:16.940 we started together in medical school at Stanford, but then finished a little bit different times.

00:04:21.600 I think it's been a while since we saw each other.

00:04:23.100 As I was sort of joking earlier, we're going to have like a whole subset of the Drive podcast,

00:04:27.180 which is based on the Stanford MSTP students between Carl, you, and Josh. It kind of speaks

00:04:33.520 to the quality of people that were a part of that program. Let's tell folks a little bit about kind

00:04:38.120 of work you did. So as you mentioned, obviously, we started in medical school together. We have some

00:04:41.220 really funny stories from the beginning of medical school, which I think we'll refrain from telling at

00:04:45.760 this point in time. I could have a whole podcast just on some of that stuff. And then after the first

00:04:51.100 two years of medical school, I went right off into the clinical stuff, and then you went off

00:04:56.540 into the lab. Tell folks about whose lab you went to and what it is that you started pursuing and

00:05:01.640 frankly, how you even made that decision. As you pointed out, as part of the MD-PhD program,

00:05:06.600 the way it is done in most programs, you split the curriculum with splitting medical school in half,

00:05:11.400 doing the first two more classroom-based years first, then doing the PhD work in the laboratory,

00:05:17.600 and then going back for the clinical work. And there's an important transition point there when

00:05:21.660 you're finishing the classroom part of the medical school and deciding what lab to work in.

00:05:26.200 I ultimately chose to do my dissertation with Pat Brown, who is a professor in biochemistry here

00:05:33.100 at Stanford. He is no longer. A lot of your listeners may know of him, though, because he's now a CEO of

00:05:40.020 Impossible Foods, or was until recently. I think he just transitioned to a slightly different role,

00:05:44.320 founder and CEO of initially of Impossible Foods. But he was here, a faculty member. And what really

00:05:50.520 attracted me to his lab was that he had, around that time, invented technology for measuring the

00:05:57.060 expression of basically all the genes in the genome with a technique called DNA microarrays.

00:06:03.640 And that at the time was revolutionary. Before that, we were always measuring everything in one or a

00:06:08.420 handful of genes at a time in experiments. And now we could measure tens of thousands in one

00:06:11.800 experiment. And it just seemed to me that this was opening up whole new fields that we would be able

00:06:16.680 to learn so much. And that's why I chose to pursue that lab.

00:06:21.880 So what did you do for your actual dissertation? What was the project that you worked on?

00:06:26.160 I had a little bit unusual dissertation, then I worked on many different projects. It was a very

00:06:30.200 unique time in the lab and Pat's lab, as well as in labs in general, where, and I've seen this happen

00:06:35.680 since several times, but when there's a new technology that's developed that's sort of transformative,

00:06:39.700 it opens up so many doors simultaneously that, you know, you have this new tool that no one's ever

00:06:44.620 had before, a new lens through which to view biology. You can just immediately think of

00:06:48.940 thousands of questions that would be interesting to ask. And so I worked on a variety of different

00:06:53.540 things, two main areas. One is immunology, the other oncology or cancer biology. And those are my two

00:06:59.900 interests coming into medical school. And so I felt fortunate that I was able to do projects in some

00:07:05.160 of both. One of the projects was focused on T cells, which are a type, of course, of white blood

00:07:11.300 cell lymphocyte that are a critical part of the adaptive immune system. So we had this new tool

00:07:16.880 to measure all the genes in the human genome at one time to see how they go up and down after you

00:07:21.880 perturb cells. And so we were very interested to see what genes are turned on and off in T cells when

00:07:27.400 you activate them, when you stimulate them through their receptor, the T cell receptor, either by itself

00:07:33.160 or with a co-stimulatory signal. And so you would look at the T cell, it's stimulated. Are you

00:07:39.660 actually measuring protein? RNA. You're milking at RNA. Yeah, this technology was a way of measuring

00:07:45.020 RNA, so transcripts, so the intermediate between DNA and proteins, of course. We could see hundreds

00:07:50.780 and thousands of genes changing as we manipulated the cells. And so building a catalog of all the genes

00:07:57.460 that turned on or down-regulated, turned off when you activate T cells using different signals.

00:08:02.040 And that catalog then, of course, has been very helpful for subsequent studies and better

00:08:06.740 understanding and teasing apart the mechanism of T cell activation, which has gotten increasingly

00:08:11.760 more interesting, of course, now with the advent of immunotherapy.

00:08:15.340 There's a lot of things there that are interesting. One of them is the instability of RNA. This is going

00:08:19.140 to become relevant as we start to talk about the differences between cell-free DNA and RNA. But at the

00:08:23.540 time, how difficult was it to keep all of this RNA intact as you cataloged the generation of mRNA from DNA

00:08:33.640 as this signal of gene expression? I mean, was that one of the big technical challenges of this technique?

00:08:40.580 That for sure is a hurdle in all work that's focused on RNA, which, as you mentioned, is chemically

00:08:46.540 relatively unstable, particularly when you compare it to DNA, which is much more stable.

00:08:50.500 One has to be very careful in any experiment, whether you measure a single RNA or where you

00:08:54.860 measure 20,000 RNAs, to really try to preserve the sample in a way such that you minimize the

00:09:02.040 potential chemical degradation of RNA. So there are, you know, laboratory methods to do that,

00:09:07.180 of course. So, for example, if you're in that example of the T cell stimulation experiment,

00:09:11.960 the cells are alive, the RNA, of course, is maintained. It's really once the cells die,

00:09:15.900 the degradation issue starts happening. So you design the experiment in such a way that you're

00:09:19.440 very careful to immediately add solutions that protect the RNA after you kill the cells at the

00:09:24.800 end of the experiment so that there's no time for chemical degradation processes.

00:09:29.580 My other main project in the PhD was actually, we really struggled with this because it was a

00:09:33.880 separate project, my rotation project, which is the first project you do in a lab when you're kind

00:09:37.100 of trying to figure out if you want to go there, where we developed a method to isolate self-RNA

00:09:42.160 that's stuck to the endoplasmic reticulum inside cells, because that is where RNAs go for genes

00:09:48.260 that are secreted from the cells or that are surface proteins in the cells. And we were

00:09:52.100 interested in cataloging those because those are interesting for diagnostic purposes and

00:09:55.100 therapeutic purposes. So we had to purify the subset of the RNA that was stuck to these organelles

00:10:00.780 called the endoplasmic reticulum. There's a very long procedure where you have to build special

00:10:05.600 gradients and float, you know, gently slice the cells and then float the membranes that they're

00:10:10.420 stuck to to various levels in a test tube. And that required a lot of work in a cold room where,

00:10:14.920 you know, you're busy, you test tubes in a four degree room, but so is the experimenter. You,

00:10:18.860 you know, you have to wear jackets and stuff because you're working basically in a fridge

00:10:21.100 all to maintain the integrity of the RNA, as well as you can add chemicals to try to stabilize the RNA.

00:10:27.160 So this is something that was critical at the time to really try to work out methods to stabilize the RNA.

00:10:32.500 And how much of an insight could you get into non-coding sequences of genes where they're not making

00:10:38.900 proteins, but we now know, of course, that these non-coding segments can be very important as well.

00:10:43.960 Could you gain any insights there?

00:10:46.720 With the technique we were using at that time, we could not directly because we were focused on

00:10:51.940 measuring the coding portions of the genes of the transcripts that code for proteins. And we had

00:10:57.700 to decide at the beginning of each experiment, which genes we measured, because basically we had

00:11:01.340 to create a probe for each. You had to have the primers.

00:11:04.180 You had to have the primers for it. And so you had to make a decision. So at the time we were not

00:11:07.460 focused on that. Subsequently, this approach was used in some of the early work on long non-coding RNAs,

00:11:13.900 and I was largely led by a former postdoc from Pat Brown's lab, who was there while I was a

00:11:17.940 grad student named Howard Chang, who is a professor here at Stanford, as you may know.

00:11:22.080 I think you finished your PhD in about three years, correct?

00:11:24.460 Three, three and a half years, something like that.

00:11:26.100 And then you went back now and you started your clinical rotations. Did you have a sense when

00:11:30.240 you went back for the last two years of medical school that you definitely wanted to be a clinician,

00:11:36.780 which would mean not just finishing medical school, but then doing a residency?

00:11:39.860 Some people in the PhD program just say, look, I just want to be a pure scientist,

00:11:44.220 not a physician scientist. I'm going to finish medical school and get my MD,

00:11:47.040 but I'm not going to do any clinical training. Where were you on that spectrum at the beginning

00:11:51.540 of that final two years?

00:11:53.820 I was set on doing a residency. I wasn't sure yet in what, but I was set on doing it.

00:11:59.260 My decision point for me was made prior to med school. Initially, I thought I was going to do

00:12:04.180 graduate school and go for a PhD and focus on research for my career.

00:12:07.220 But then my father was diagnosed with lymphoma while I was junior in college. And the journey

00:12:13.440 that he went through and interacting with the oncologist and the medical team as part of that

00:12:18.160 and really seeing how little we knew about many things and how suboptimal treatments were

00:12:23.260 really convinced me that I want to be on the doctor's side, not just on the patient's side

00:12:27.600 with my father and be able to help people, help patients at the time of their life, what might be

00:12:31.800 the worst time of their life, as well as to try to move the field forward to improve treatments that

00:12:37.080 while they worked somewhat, obviously were not good enough.

00:12:41.400 And so does that mean that even coming into medical school, not only did you know you wanted

00:12:44.520 to be a physician and a scientist, but did you already have kind of an inkling that oncology was

00:12:48.820 where you wanted to be?

00:12:50.500 That's exactly right. Yes. I knew I wanted to work in one of the cancer specialties. What I didn't

00:12:55.480 realize when I was the first year of med school is how many options there are in that.

00:12:58.080 You were probably looking at it mostly through the lens of medical oncology.

00:13:01.360 That's right.

00:13:01.900 Whereas of course, surgical oncology, radiation oncology, and other sorts of avenues.

00:13:06.780 Remind me, were you born in Germany? Were you born in the US?

00:13:09.840 No, yeah, I was born in Germany. I was born in Munich and I didn't come to the US until I was 11

00:13:13.500 years old. And you grew up on the East Coast?

00:13:16.060 Yes. We initially moved to South Florida. When I was a sophomore in high school, we moved to

00:13:20.120 Connecticut, the Northeast. And then I went to Harvard as an undergrad. So I was in the Northeast

00:13:24.460 until I came out here, but then never left anywhere. And now I've lived longer here than

00:13:27.660 anywhere else in the world. So you go into the clinic and now it's basically a decision of

00:13:33.160 medical oncology, radiation oncology, or God forbid, something like surgical oncology,

00:13:37.280 though I'm guessing that was probably third on the list of three options.

00:13:41.520 You know, I did honestly strongly consider it. I do like the procedural aspects of it. And that's

00:13:46.700 the one reason why I chose what I chose. So how did you ultimately decide on radiation oncology then?

00:13:52.320 I was sort of leaning towards medical oncology because that's what I'd seen through my dad's

00:13:55.720 journey. I did a rotation in radiation oncology, actually in large part because my wife, who was

00:14:01.480 also a medical student at Stanford, and who you of course know, did a research year in the Department

00:14:06.620 of Radiation Oncology. She at the time was also interested in oncology. And so that actually is

00:14:10.980 how I got exposed to the field. While we were dating, I would often visit her in the lab. And so

00:14:15.400 got exposed in that way to the department in the field. And then so it made me want to try it as a

00:14:20.580 rotation. And then I really enjoyed it. From a patient care standpoint, it seemed to me that

00:14:24.860 radiation oncologists had a little more time in clinic to spend with each patient. We generally

00:14:28.860 see less patients in a day than might be seen in medical oncology. That seemed very attractive to

00:14:33.780 me. The other thing I really liked was the technology aspects. I've always been interested

00:14:37.600 in technology aspects, both new technologies in the research space, but then also, you know,

00:14:42.200 technologies for treating patients. And radiation oncology is very procedurally heavy. It's,

00:14:45.820 of course, a field where we do a lot of imaging to see where tumors are in a patient's body. And

00:14:49.720 then we have fancy robots that deliver the radiation very precisely. So it just seemed

00:14:54.580 like a great field to combine my love of oncology and patient care with my interests in technology

00:15:01.020 development. And then lastly, I kind of saw it as a field where they really, compared to medical

00:15:05.240 oncology, there wasn't as much work being done in the laboratory at the molecular level. And it seemed

00:15:11.040 like an opportunity to make a difference in a field where there weren't so many people working.

00:15:15.060 You know, one of the things I talked about with Carl was the challenge of keeping his hand in the lab

00:15:23.200 during those last two years of medical school. And then during his psychiatry residency,

00:15:28.700 he really found himself in a great situation when he did his residency,

00:15:32.160 where he was effectively allowed to do kind of a postdoc. And he was freed from, you know,

00:15:38.420 certain things. He didn't go to, you know, lab meetings and journal clubs and things like that,

00:15:41.480 but he still got to do a little bit of work. Were you able to do anything like that during your

00:15:47.320 residency? Or were you really strictly focused on just the clinical training for, was it four years?

00:15:53.780 So radiation oncology is, did an internship in medicine here at Stanford, and then it's four

00:15:57.880 years of radiation oncology. So five total during internship, of course, as you, I'm sure remember,

00:16:03.140 there is no time. I do remember trying to write some papers in the call room to finishing up work

00:16:07.940 from the PhD. Radiation oncology residency programs have a research track for individuals who are

00:16:12.380 interested in laboratory-based research called the Holman pathway, which is in other fields kind

00:16:16.360 of called short tracking or fast tracking. It's a similar idea where basically one can trade in some

00:16:21.560 of the clinical training time for research time. And so that's what I did, which gave me a postdoc time

00:16:26.800 of about two years during my four years of radiation oncology residency. During that postdoc time,

00:16:32.400 I had clinic activities about half a day to a day a week, and the rest of the other days were in the lab.

00:16:37.120 So it actually was a really good preview of what my life would be like once I finished all the

00:16:42.160 training and was a faculty member myself. That allowed me to keep a foot in the clinic while

00:16:46.560 mainly focusing on the postdoctoral research. So when you finished your training, obviously the

00:16:52.060 first decision you had to make is, do you want to stay at Stanford? I'm guessing that between the

00:16:56.360 connections you already had there, Jenny was probably by this point already out of her training and in

00:17:00.020 practice. That was probably a very high activation energy to leave. Tell me about the Department of

00:17:05.480 Radiation Oncology at Stanford. Was it a natural fit for the types of problems you wanted to solve

00:17:10.780 on the research side? The department here at Stanford has a very long history. It was actually

00:17:16.180 one of the very first departments of radiation oncology in the US. Radiation oncology grew out of

00:17:23.020 radiology, which is, as listeners probably know, is a diagnostic arm of radiology where you do imaging

00:17:28.380 only to diagnose disease. Initially, back in the 50s, 40s, 50s, it was part of, radiation oncology was

00:17:34.800 part of those departments. But then here, our first chairman, Henry Kaplan, who is a very famous

00:17:39.660 radiation oncologist and physician scientist, became the first chair of radiation oncology and sort of

00:17:43.720 led the movement to develop it as its own specialty. He quickly put the department on the map because of

00:17:49.700 his work in curing Hodgkin's disease with radiotherapy, which around the 50s, where, you know, those were

00:17:56.000 some of the first successes of taking patients, especially young patients with a Hodgkin's

00:17:59.540 disease or type of lymphoma that was incurable previously, who now, you know, the majority could

00:18:03.920 be cured with radiation. At the time, there were articles about how, oh, now radiation will cure

00:18:08.720 all cancer and look, we're on the path. That, of course, didn't turn out to be that way. We still have

00:18:12.300 many, many patients we can't cure. But it sort of was one of the first examples of this, the hype that

00:18:17.220 you get at each time a new therapy comes where, you know, then think, oh, again, now we've really

00:18:20.420 solved it. But ultimately, you know, it gets more complicated. But that, of course, established

00:18:24.780 department is one of the very leading departments in the world. And it has continued that the

00:18:28.400 department here throughout its history has had a very strong interest in laboratory-based research.

00:18:33.800 This is all thanks to Henry Kaplan, who was doing laboratory research at the time also while seeing

00:18:37.460 patients. So there were already faculty that were laboratory-based, even just PhDs who were fully

00:18:42.640 laboratory-based, which wasn't the case in many radiation oncology departments. And so I really like

00:18:47.240 that here, that within radiation oncology, this was a place where I could see that the kind of research I

00:18:51.580 wanted to do was valued. There was already, you know, mentors that had done it and successfully.

00:18:55.720 And those things are important when you're a young physician scientist to have mentors that can

00:19:00.140 show you, if you run into trouble, how to overcome obstacles.

00:19:04.480 At what point in your evolution did the idea of liquid biopsies start to become of interest?

00:19:10.720 Obviously, that's something I really want to talk about today, because it's something that I think

00:19:15.460 people are just starting to hear about. I mean, it's something I've been following for about six

00:19:21.200 years. Obviously, you've been following it for a lot longer than that. But I think even in the sort

00:19:26.580 of broader public eye, the past year, I think a lot has happened to bring this to the fore. But I still

00:19:33.340 think for many people, it's still a bit of a black box. How many years ago did this sort of become

00:19:37.780 something that landed on your screen as an area where you wanted to put focus?

00:19:41.600 As these things often happen, I did not start my lab with the idea of focusing on liquid biopsies.

00:19:48.060 I really got there by following the results we got from some experiments. And, you know, I think one

00:19:54.480 of the keys to being a successful scientist, whether you're a physician scientist or a purely

00:19:58.820 laboratory scientist, is following the data, following the data that you generate to where it

00:20:03.180 leads you, as opposed to, you know, saying I have to work in one area. So I was actually going to work

00:20:07.380 on a different area initially. I started this line of work in my laboratory out of a clinical

00:20:11.560 need I saw. And that's, in general, how I run my laboratory. All the research projects we do,

00:20:15.900 we start with a clinical need or a clinical, it's suboptimal we're doing in the clinic,

00:20:21.080 whether it's for diagnostic purposes or treatment purpose or whatever. You know, that's, of course,

00:20:26.060 one of the things that having, being a physician and taking care of patients is one of the things

00:20:30.180 that I spent a lot of years learning about and also what I have insights into that others might not.

00:20:34.780 That's one way I can be a useful contributor to the field in general. And so we always start

00:20:39.080 with clinical areas of clinical need. And in this case, it was one of the main things I struggled

00:20:42.420 with as a junior faculty member, which is that I decided to specialize in treating lung cancer

00:20:46.840 clinically. So one day a week, once I was on faculty, I was doing clinic one day a week and

00:20:51.200 I saw lung cancer patients exclusively. And I was very frustrated by the fact that after I treated

00:20:57.600 patients, so I treat a lot of early stage lung cancer patients, these are patients that are stage one

00:21:01.380 or two lung cancer, which means as far as we can tell, the lung cancer hasn't spread yet. It's just in

00:21:05.440 the lung where it started and hasn't spread anywhere. With targeted high dose radiation,

00:21:09.180 I can cure the majority of those patients. But there's, let's say, about 20, 25% of those patients

00:21:14.120 who ultimately develop recurrence where the cancer comes back. After I finished my treatment and that

00:21:19.520 first follow-up visit, let's say three months after I do the radiation, I could not tell who was going to

00:21:24.520 ultimately have the cancer come back and who had been cured. So state of the art at the time,

00:21:29.920 and actually, you know, it's still state of the art today in many ways, is just to see the patient

00:21:36.120 every three, six, 12 months as you get further out and to do scans and see if the cancer has come back

00:21:41.680 without any way of predicting in whom it will. It's a very reactive approach and very unsatisfying. And

00:21:46.560 this is something we went through with my dad also. Give people a sense of the resolution we have

00:21:52.960 with traditional imaging. Where are we at on CT scanners? Are we at 256-bit? What's the resolution

00:21:59.400 and speed of a CT scanner today? You may ask somebody who works in that specific field of

00:22:04.460 building those scanners, but to rephrase your question a little bit is, you know, how small of

00:22:08.220 a tumor can we see? That's where I'm really going, which is if you can see a one by one by one centimeter

00:22:13.400 or a one centimeter diameter tumor, you're doing pretty well, right? You can probably see a little

00:22:18.340 smaller than that. That's exactly right. So it is hard to see things much smaller than

00:22:23.800 one centimeter in diameter, maybe eight millimeters. It depends a little bit on the

00:22:27.260 location in the body. Some areas, of course, it's easier to see smaller things than others,

00:22:30.280 but that's generally true. And so how many cells is that? The general sort of rule of thumb is that's

00:22:35.260 about a billion cells already, which is, of course, a vast number of cancer cells, again, at the time when

00:22:41.480 we can just barely detect it in a patient. And I think it's important for patients to understand the

00:22:46.480 non-linearity of this. So you have everything shy of a billion cells is, by conventional detection

00:22:55.120 methods, undetectable. And then outside of the most rare tumors, anything at a centimeter is not

00:23:02.240 posing a direct threat to the organism. You know, maybe a really badly placed brain tumor at one

00:23:07.660 centimeter is problematic. But for the most part, if you talk about a lung cancer, liver cancer,

00:23:11.860 pancreatic cancer, colon cancer, etc., breast cancer, one centimeter is irrelevant. At what point

00:23:18.120 does a tumor become, in and of itself, threatening to the host? I mean, you could argue by the time

00:23:23.900 it's 10 centimeters by 10 centimeters by 10 centimeters, the burden of tumor itself is fatal.

00:23:29.420 And that's how many cells there, roughly, when you're at the point where the burden of the tumor

00:23:34.060 is actually fatal? Of course, whether tumors fail or not, as you point out, greatly depends on the

00:23:39.780 location. In certain areas, one can have a ton of tumor without it being fatal. And in other areas,

00:23:44.780 you have very little room for that. It's a volumetric. It grows by the cube, by the radius

00:23:50.420 multiplied by itself three times. And that is, of course, non-linear. It grows much faster. The number

00:23:56.660 of cells in a mass are much more than just the linear increase in the diameter of the lesion.

00:24:02.080 Yeah. And I think that's the part that's just hard to understand. We think linearly pretty well.

00:24:06.060 We don't think exponentially very well. But again, I think I want people to just anchor to this idea

00:24:11.060 that from zero to a billion cells, we're pretty much flying blind.

00:24:16.820 And relevant to the work, some of the work that we're now doing, many cancers, once they have

00:24:20.800 metastasized or spread, which is really ultimately what ends up killing patients is once the, usually

00:24:25.420 when the cancer spreads, if the cancer is only localized, usually a surgeon or a radiation

00:24:28.760 oncologist can cure that patient. But once the cancer spread and is in many places in the body,

00:24:33.680 you can't remove it all or radiate it all, right? So if you think about a patient who has

00:24:38.020 what's called micrometastatic disease or microscopic disease that has spread

00:24:41.560 elsewhere in the body, you can have dozens or even hundreds of these micrometastatic deposits

00:24:47.340 in different organs.

00:24:48.680 Right. That are under half a centimeter. And so you technically have tens of billions of cells

00:24:53.760 and it's still undetectable.

00:24:55.300 Exactly right. That is a critical issue where that's a limitation of blind spot of our imaging methods.

00:25:00.040 Since they're all spread out, we can't see them all. But that is one thing that

00:25:03.420 one could potentially get a handle on if one had a test that could be measured, let's say in the

00:25:08.040 blood where one could measure contributions from all these dozens of micrometastatic deposits and

00:25:14.220 that that one might get a higher sensitivity. And that exactly kind of was the motivation for

00:25:18.560 the liquid biopsy work in my lab. I had no idea how I was going to do that when I started it.

00:25:23.400 Now let's hit pause for one second, Max. Let's also give people a little bit of the historical

00:25:27.980 context of where there are at least a couple of areas where this was sort of being done, although it wasn't

00:25:34.560 great. So maybe explain for people how PSA, CEA, and CA-19-9 have been used historically. Because even when we were

00:25:42.860 in medical school, long before you and I got to medical school, those three biomarkers were used. Now you were

00:25:49.840 always cautioned that at least the CEA and the CA-19-9 could never be used for screening, but they could be

00:25:56.100 appropriate for following. PSA has a very complicated history, which we could go down the rabbit hole of

00:26:02.520 about how it could be used for screening, but you know, you can be misled. But tell folks a little

00:26:06.940 bit about what those things are and how effective you think they were for what they were trying to do

00:26:13.800 and which elements of those you wanted to replicate for lung cancer and where you wanted to improve upon

00:26:19.480 it. When we started thinking about developing a blood-based test for initially lung cancer, we of course

00:26:24.960 thought about exactly the examples you pointed out, which are the markers you mentioned, like PSA,

00:26:29.120 CEA-19-9 are protein biomarkers. So these are proteins made by the cancer cells that can be shed into the

00:26:36.520 blood and that one can then measure them in a non-invasive fashion. A large issue with these

00:26:41.900 protein biomarkers is that they are actually not that specific for cancer, meaning that these are also

00:26:48.520 proteins that normal cells can make. Now, cancers often tend to make more of them, which is why

00:26:54.220 they are potentially of interest. But the problem becomes that having low levels of any of those

00:27:00.320 markers, when a patient has that, you can't know whether that means they have a little bit of cancer

00:27:04.940 in the body or whether it's just that's their baseline, what the normal cells are making in their

00:27:08.780 body. And that lack of specificity is what it's called, meaning you're not so sure when you see it that

00:27:13.900 it's cancer or not cancer, was the major Achilles heel of the protein-based biomarkers. And a lot

00:27:19.540 of work, of course, had been done in lung cancer previously looking at similar biomarkers, including

00:27:24.100 CEA actually, which in some subset of patients can be a good biomarker if the tumor just pumps out a

00:27:28.800 lot of it and the patient has a decent amount of disease. So the total level is significantly higher

00:27:33.060 than what a normal patient would have. But in most patients, that was not the case. And so that was

00:27:37.160 protein-based approaches were really state-of-the-art, but had this major weakness of a lack of

00:27:42.000 specificity. Let's maybe use this as a moment to explain to people sensitivity and specificity,

00:27:48.480 which are going to become very important as we then factor in prevalence when it comes to screening

00:27:55.980 to understand positive and negative predictive value. But I think for now, if we can just start

00:28:00.260 with helping people understand what specificity and sensitivity are in terms of probabilities of

00:28:06.100 false positives and false negatives, I find this stuff to be so important. And yet,

00:28:12.000 if you don't explain it over and over again, it's easy for people to kind of get lost in the

00:28:17.800 details. And yet, it's so important for what we're about to talk about today that we'll probably end up

00:28:22.520 doing this twice, three times.

00:28:24.440 Yes. These are very important topics, words that are thrown around a lot and can be confusing.

00:28:30.540 Sensitivity is something that's also known as, or a synonym for it, is true positive rate, 0.95

00:28:35.860 which is basically the likelihood of having a positive test when the patient has the actual

00:28:42.760 condition. So let's say in the example of cancer, sensitivity is you have 100 cancer patients,

00:28:47.660 you have a new blood test. Out of those 100 cancer patients, how many of them is the blood test

00:28:53.680 positive? That's sensitivity.

00:28:56.520 Right. So a perfect test would obviously have 100% sensitivity, which means 100 out of 100 would test

00:29:04.120 positive. In reality, to give people a sense of this, a mammography might have an 85% sensitivity,

00:29:10.980 which means if you do a mammogram on 100 women who are known to have breast cancer,

00:29:16.380 it might only tell you that 85 of those 100 do. So it's correctly identifying 85. It's erroneously,

00:29:24.120 it's giving you a false negative in 15 of the 100.

00:29:27.780 In general, it's important to realize that there are no perfect tests, really, particularly when you

00:29:34.400 start pushing the envelope to trying to see smaller and smaller tumors. So while it would,

00:29:38.940 of course, be great to have a test that's 100% sensitive, meaning it catches every cancer patient,

00:29:43.620 that is really unrealistic in the vast majority, if not all cases. And I think this is something

00:29:47.220 that's sometimes not understood by patients where you go get your mammogram or your CAT scan for lung

00:29:52.220 cancer screening, and patients sometimes still develop a cancer even though they did all those things.

00:29:56.520 And why is that? And that can lead to a lot of frustration, but that's because these tests,

00:30:00.300 they're not always going to pick up a cancer even when it's present.

00:30:03.620 One of my former analysts, Bob Kaplan, came up with a great thought experiment to explain this

00:30:07.540 to people, which was the reason you can't have perfection is if you drive something to a very,

00:30:14.480 very, very high sensitivity, the specificity must go down. And here's the silly example.

00:30:20.180 And this is his example, so I don't want to take any credit for it, but it's brilliant. 0.73

00:30:22.900 If you wrote a letter, Max, to 1,000 women randomly and told every one of them they had

00:30:31.160 breast cancer with no additional knowledge, you technically have a test with 100% sensitivity.

00:30:38.120 Because let's assume 50 of those women have breast cancer. You have correctly identified them all.

00:30:44.100 There are no false negatives in that group. So you have a 100% sensitive test. The problem is

00:30:51.120 your specificity is in the toilet. You have so many false positives that the test is utterly useless. 0.53

00:30:58.420 So you can have 100% sensitivity if you really want to push that envelope, but it has no clinical

00:31:04.200 utility. And of course, I can play the experiment the other way. You could send 1,000 women a letter

00:31:08.520 that says you absolutely do not have breast cancer. And technically, you have no false positives,

00:31:13.580 but you clearly are going to have false negatives.

00:31:15.960 So I love that experiment because it explains that if you pull full throttle on sensitivity or

00:31:21.880 specificity, that's fine, but it doesn't make for a good test because you've undoubtedly

00:31:28.160 compromised the other parameter.

00:31:30.380 It's a great example to explain how that works. Actually, it explains a lot of what happens in the

00:31:34.920 research, the agnostic research field in general. You know, you try to push sensitivity, but that never

00:31:40.940 is meaningful to report if you don't also report the specificity, to look at specificity.

00:31:46.400 Because it's kind of like a yin and yang. If you push on one, the other, you know, usually one goes up,

00:31:51.120 most of the time the other goes down. Unless you really have a dramatic new insight or a new advance

00:31:55.740 where you can just increase sensitivity without also hurting specificity.

00:31:59.880 I guess just to finish off, so again, specificity since we haven't defined it.

00:32:02.760 So that is sort of the inverse of sensitivity, which is that for a patient who does not have cancer,

00:32:07.600 really is a true negative, what we call it, and doesn't have cancer. The test reports that the

00:32:12.780 patient doesn't have cancer. So it's sort of the exact inverse of sensitivity, but for the patients

00:32:16.500 who don't have cancer. And as you mentioned, they are connected. You can game it. You can always push

00:32:22.480 the sensitivity higher if you kind of ignore the specificity.

00:32:25.660 So now let's go back to your clinical problem. By the way, Max, you know, we've talked a little bit

00:32:31.840 about lung cancer on this podcast, but I can never resist the opportunity given the prevalence of lung cancer

00:32:37.180 and the, what I refer to as the loss given default, meaning the severity of the disease.

00:32:43.020 Do you want to give people just a sense of where we are today in 2022 with lung cancer? Is it the

00:32:49.040 first or second leading cause of cancer deaths for men and women?

00:32:52.520 It's the number one cause of cancer death. I think a lot of people don't realize because it doesn't get

00:32:58.000 nearly as much attention, at least historically, as some of the other cancers that are maybe more

00:33:05.280 common, especially if you do just in men or just in women like breast or prostate, but where actually

00:33:10.760 the outcomes are much better. We can cure a very large fraction of patients with breast and prostate

00:33:15.500 cancer. Of course, not everyone, and we still need to make more advances. But the difference with lung

00:33:20.140 cancer is that the vast majority of patients historically can't be cured. And so that's why it is the number

00:33:27.240 one cause of death, cancer-related death.

00:33:30.200 Is that going down or going up or flatline? I mean, because smoking rates are coming down.

00:33:35.720 And about 85% of lung cancer by incidence, I believe, is smoking-related. So some combination

00:33:43.220 of smoking-related. 15% is in non-smokers. So what's the trend line on lung cancer mortality?

00:33:49.860 Both the incidence of lung cancer has been going down in both men and women. Initially lagged in

00:33:55.520 women, but it's also coming down in women now. And this is all related to, as you mentioned,

00:33:59.060 the uptake and then now decrease of smoking in the population. So that means because there's less

00:34:03.860 smoking and because smoking is such a strong risk factor for developing lung cancer, there's less

00:34:08.340 lung cancer overall. And on top of that, our treatments have gotten much better, as has now

00:34:13.660 slowly the advent of screening. We have a way of screening for lung cancer, which is relatively

00:34:17.640 recent development. So those things combined have also decreased the number of patients who die from

00:34:24.120 lung cancer once they get it. Even with all that said, it still is the number one cause of cancer

00:34:29.880 deaths. But the trajectory is good. It's starting to head this way. And of course, we're very eager to

00:34:33.900 accelerate that slope, make it even less of a killer of patients. I want to point out one other point,

00:34:39.520 which is often comes up in these kinds of discussions, which is that while smoking is,

00:34:43.600 by and far away, the largest risk factor, it is not the only risk factor. And really,

00:34:48.140 anybody who has lungs can get lung cancer. You know, that can happen for reasons that are related to

00:34:53.140 pollution. You know, and that has been well described, that pollution or other things in the

00:34:57.160 environment like radon gas that people are exposed to can cause lung cancer. And there's genetic

00:35:02.100 associations also. For example, Asian populations, there's a much larger incidence of non-smoking

00:35:08.180 individuals who get lung cancer, particularly females. And so we're never going to get rid of 0.59

00:35:12.960 lung cancer, even if we could magically make smoking go away today, you know, across the globe,

00:35:17.620 you know, make all cigarettes disappear and never make a new one, there will still be lung cancer.

00:35:22.160 Let's go over kind of the distribution of how these things work out. You have small cell and

00:35:26.880 non-small cell is the big division, correct? That's right. And then within non-small cell,

00:35:32.840 you have adenocarcinoma, you have squamous cell carcinoma, large cell and carcinoid. Are those

00:35:38.740 the big ones on the large cell or in the non-small cell? I'm sorry. Carcinoid is not considered a

00:35:43.500 non-small cell lung cancer. It's its own category, but it's not a small cell. So that's true still.

00:35:48.180 The most common lung cancer isn't the non-small cell subtype of which the majority have adenocarcinoma,

00:35:54.920 but also a significant fraction of squamous cell carcinoma.

00:35:58.460 When a non-smoker gets lung cancer, it's usually adenocarcinoma. Is that correct?

00:36:04.520 The vast majority of time, it can sometimes be squamous cell carcinoma or even occasionally

00:36:08.080 small cell, but that is extremely rare. And as you mentioned, being a woman, 0.87

00:36:12.460 being Asian seem to elevate risk. In the case of women, it seems to be potentially related to 0.98

00:36:18.000 differences in estrogen, testosterone. That's been proposed as one plausible explanation. Do we know

00:36:23.320 what genes might account for this in Asians versus non?

00:36:27.420 We do not. And that's sort of one of the big mysteries that we have not identified the main

00:36:32.600 genetic drivers. And that's not for lack of trying, of course. It's multifactorial,

00:36:38.040 right? That it's one of these, as are many, of course, of our chronic illnesses that have genetic

00:36:42.860 components, the cardiovascular diseases, for example, right? Where it's not about a single gene,

00:36:47.260 the vast majority of the time. It is contributions probably from multiple genetic variants that in

00:36:53.200 aggregate elevate the risk. And then there may also be, of course, on top of that, then some further

00:36:58.320 environmental factors that are in smoking that may then interact with that genetic background.

00:37:03.360 I think you mentioned radon. We talked about that as definitely a big environmental exposure.

00:37:08.040 What do we know about the PM2.5 exposure? I've certainly seen data that show that people who are in,

00:37:14.760 and just for folks listening, particulate matter less than 2.5 micron is small enough to make its

00:37:19.500 way into the most distal part of the lung. There's clearly an increase in all-cause mortality associated

00:37:25.420 with higher PM2.5 exposure. Do we know if that's also true specifically for lung cancer? Is that

00:37:31.380 something that we can peg to an increase in lung cancer?

00:37:35.640 There are associations with that. So people have looked at individuals living in cities versus

00:37:39.280 rural areas and correlating with the particulate contamination. And that is also associated

00:37:44.500 with more lung cancer in the cities that have more of the particulate in the ambient environment.

00:37:49.860 There's definitely an association in that regard, as with epidemiologic studies,

00:37:53.480 to really prove it in a human is difficult, of course, but it does seem quite strong. And there's

00:37:58.120 some biological rationale for that, where that if that is causing irritation and chronic inflammation,

00:38:02.940 those kinds of things, of course, we know that those things are associated with developing

00:38:06.060 cancer in many organs. So it all makes sense. I think it's likely true, but there also are,

00:38:11.760 of course, chemicals in smog that, you know, sort of along the lines of what's in tobacco smoke,

00:38:16.460 similar class of chemicals that can be direct carcinogens.

00:38:19.740 Do we have a sense of what the dose response curve is in pack years of tobacco to relative risk

00:38:27.960 increase in lung cancer? So does a person who smoked a pack a day for 10 years, so they have a 10-pack

00:38:34.980 year history relative to a non-smoker. Is that 2x the risk, 50% more risk? I mean, do we have a sense

00:38:41.220 of how non-linear that curve is as well versus, you know, the 40-pack year smoker, the 5-pack year

00:38:47.000 smoker, et cetera? There are associations between the number of pack years and the risk of developing

00:38:53.620 lung cancer. Now, a pack year is not an ideal clinical variable because you can't measure it in

00:38:59.320 a completely unbiased way, the way just so your patients understand how a doctor tries to figure out

00:39:03.640 how many pack years a patient has smoked, meaning we basically multiply the number of years of

00:39:08.140 smoking, how many years they smoke for, by the number of packs per day that they smoked. Well,

00:39:12.900 how do we get that information? Well, we have to ask the patient. And the reality is, number one,

00:39:17.340 patients smoke different amounts over their time. And number two, you know, they don't always remember

00:39:21.020 perfectly. So it's not a great metric in that regard. So it's difficult to really pin it down

00:39:26.260 perfectly. I have seen studies that have for sure suggested there isn't, you know, increase as you go from

00:39:31.120 zero to higher levels. But then around 20 or 30 pack years, some studies argue that then kind of

00:39:35.700 plateau is that you may be sort of saturating. I don't have the exact slopes of those lines. I don't

00:39:40.280 know off the top of my head, but it is not a perfect variable. What do the data say about secondhand

00:39:45.320 smoke, which I would imagine is even harder to quantify than pack years? You know, for example,

00:39:49.480 I have some patients who grew up with parents who smoked. They themselves have never smoked. And let's

00:39:55.060 say those parents continue to smoke all the way until those kids, now adults, went to college.

00:40:00.040 Do you have a sense of how to quantify what their smoke exposure is? And should they be treated as

00:40:04.220 former smokers in terms of cancer screening, for example, with low dose CT and things like that?

00:40:08.980 It's very difficult to measure. That's even harder to measure than the pack years we just talked

00:40:12.460 about. We know that certain professions that were exposed to a lot of smoking, like waitresses and

00:40:18.560 waiters or stewards and stewardesses, that there was increases, you know, epidemiologically. Of course,

00:40:22.880 there's always other things that go along with that issue also. So I think the link is quite clear

00:40:27.260 how to quantitate it and how to know has one had enough secondhand smoke exposure that one's risk

00:40:31.980 is significantly elevated. I think that's a major problem. Actually, an error would be very useful

00:40:35.620 to have a biomarker that could be quantitately measured to actually measure like a clock of

00:40:40.760 amount of exposure you had. We don't have such a thing. Currently, that is not one of the criteria

00:40:45.920 that would get a patient eligible for lung cancer screening by low dose CT. So secondhand smoke is

00:40:51.700 not considered. The way that we decide on, you know, those criteria for who should get this test

00:40:56.780 and cover it by insurance and who shouldn't is really about trying to enrich for us, you know,

00:41:00.640 the highest risk population. You know, you want at least, you know, a certain percent, half percent,

00:41:04.560 a percent of risk of the cancer that you're going to screen for because otherwise you will screen

00:41:08.080 lots and lots of patients who will never get it. And then the specificity becomes a problem we

00:41:12.180 talked about earlier. Even though there are things like exposure to pollutants, secondhand smoke,

00:41:17.120 that we know increase the risk, we don't. And it can be frustrating enough for patients who are in

00:41:21.620 those categories to know whether, why can't they get access to screening? But there are public health

00:41:26.480 reasons for that. Yeah, which we'll then get to something we'll talk about in a little bit, which

00:41:30.860 is you can know the sensitivity and specificity of a test, but you then need to know the prevalence

00:41:36.540 or the pretest probability to know how to interpret the result. Without the prevalence, you can't

00:41:42.480 impute the positive and negative predictive value, which is what tells you when you have

00:41:46.960 a positive. How confident are you? It's positive. And similarly, when you have a negative, how

00:41:50.400 confident are you there? Let's talk a little bit about low dose CT, because I would say this has

00:41:54.640 been certainly in the last 10 years, one of the major changes in how we manage lung cancer. It's

00:42:00.180 been about 10 years since the data have made the case for the use of that as a screening technique.

00:42:04.560 Yeah, there'd been a long effort at trying to develop a screening test for lung cancer,

00:42:07.920 given that it is the number one cause of cancer deaths. And there were a lot of failures along that

00:42:12.340 road. A lot of initial studies focused on doing chest x-rays when that was the main thing that was

00:42:16.420 available. So a much lower resolution way of imaging the lungs, where those studies were unable to show

00:42:21.220 a benefit. But then there was a landmark study called the National Lung Screening Trial. In that

00:42:25.720 trial, they took patients that randomized them either to get a low dose CT scan, which is now a more high

00:42:30.480 resolution way of looking at the lungs that can see smaller nodules, versus getting a chest x-ray,

00:42:35.200 which was this technique that previously already we had kind of known that didn't work. And if you do that,

00:42:40.320 the patients who get the low dose CT scan, the high resolution imaging, in that group,

00:42:44.580 there was significantly lower rates of lung cancer deaths, you know, relative risk reduction of about

00:42:48.500 20% that argued that that's a major win for a screening test to actually be able to decrease

00:42:54.260 the number of deaths from the disease you're screening for. Do you remember what the absolute

00:42:57.940 risk reduction was? That was small. It was in the single digit percents. I don't remember it off the

00:43:02.760 top of my head. But it was single digit percents. So let's say it was 5%. Your NNT would be 20.

00:43:09.160 You could save a life for every 20 people. That still seems pretty good.

00:43:12.220 It's actually lower than that. So maybe it was like 1% to 2%.

00:43:15.440 Maybe it's 0.5% to 1% based on that argument. Now, that gets complicated also, because you'd

00:43:21.000 think doing a CT scan, you either see a nodule, you don't. So it should be pretty easy. But of

00:43:25.400 course, it's not that easy. A CT scan is a complicated thing to read. And so how you interpret

00:43:30.200 the scan and what you consider a positive will, of course, affect your sensitivity, specificity,

00:43:36.600 and will affect the number needed to treat downstream. So the way that we now read scans,

00:43:40.840 the low-dose CT scans is different than what the initial study did because of a high risk

00:43:44.360 of false positive, the way that they was doing the original studies. Low-dose CT is not my area

00:43:48.240 of expertise. I don't have all the numbers at my fingertips, but those are the sort of salient

00:43:52.620 issues that are in that field. So even though we have this test, and it is great, and it's

00:43:57.120 this major home run for screening because we can save cancer deaths, it's not a perfect test.

00:44:03.260 Do you know by any chance how many millisieverts of radiation the low-dose CT provides? I'm guessing

00:44:08.440 it's like one millisievert or less maybe? I think that's right. Yeah. I don't know

00:44:12.980 that off the top of my head, but it's significantly less than the scans that we routinely do for

00:44:17.740 patients who already have cancer. And this is an important concern that patients have.

00:44:22.820 Radiation can cause cancer. So is the risk worth it, right? It's called low-dose CT because the

00:44:27.740 amount of radiation used is much, much, much less than was historically done for a CT scan.

00:44:32.080 And so therefore the risk of causing a cancer is much, much lower.

00:44:35.860 You know, the NRC says that we shouldn't be exposed to more than 50 millisieverts in a year.

00:44:41.680 Is that sort of like saying you shouldn't drink more than 10 drinks in a day? Like,

00:44:46.580 do you have a sense of what would be your personal threshold for how much radiation you would want to

00:44:52.200 receive a year from imaging? Because you're going to get radiation from being on airplanes,

00:44:56.440 although that's relatively small. You know, just being at sea level is probably one to two

00:45:01.000 millisieverts a year. If you live in Colorado, you probably double that. But when do you start

00:45:05.920 to think about how much radiation a patient is getting from a practical standpoint? You know,

00:45:11.220 do you start to worry at 20 millisieverts in a year, which is obviously pretty easy to do with

00:45:15.420 a PET CT scan, for example? You know, we don't routinely track that at the individual patient level.

00:45:22.380 It's tracked in a less, you know, I guess, systematic way in that we don't do imaging tests unless we feel

00:45:29.080 that the benefit of doing that test outweighs the known but really low and actually very difficult

00:45:36.260 to quantitate harms. It's actually very difficult to quantitate exactly how much risk of, let's say,

00:45:42.100 a second milligancy, a future milligancy there is from a given protocol of a CT scan or whatnot,

00:45:47.000 just because to get that data and you can never do it randomized, all those things, right?

00:45:50.120 The way that medical practitioners who order imaging studies, which of course most or many

00:45:55.060 fields, in many fields they do, not just radiology, the calculus is always, do I need this test, right?

00:46:00.320 Will this help me manage the patient in a way that's beneficial to the patient? We absolutely

00:46:03.880 should not be doing imaging if there's no point. Like if no matter what the scan shows, I'm not going

00:46:08.580 to change what I'm doing, then I should not be ordering that scan. The question comes up, should I

00:46:13.040 image, should I not? And the way I and all of us in these areas think about it is, is it worth doing?

00:46:18.480 And so some patients who have lung cancer, particularly when they get radiation therapy,

00:46:22.120 which is very high doses of radiation, they have astronomical amounts of radiation that if you were

00:46:26.820 not a cancer patient, you would never want those amounts of radiation. But the vast majority of those

00:46:31.080 patients will not get a cancer from their imaging or radiation treatment. And their cancer, if we leave

00:46:35.380 it untreated, will kill them, right? And so if it's going to come back, if we don't catch it early,

00:46:39.040 well, we still might have a second chance at cure. So that's how we do the calculus. It's not that

00:46:43.760 there's badges that patients wear that track this so that if you get to a certain point,

00:46:47.860 you couldn't do no more imaging. It's always a sort of a case-by-case basis.

00:46:51.580 So let's go back now to the clinical problem that is at the root of this whole situation,

00:46:55.580 which is you have a patient who has either had a resection and or radiation to follow.

00:47:01.560 They're clinically, as we would say, NED, no evidence of disease. But you know that the

00:47:06.300 recurrence rate is there. Let's say it's 25% actuarially. 25% of these 100 people who are

00:47:11.860 currently NED are going to have a recurrence. And we know that the sooner we catch the recurrence,

00:47:18.700 the better the odds are for treatment. The lower the tumor burden, the lower the burden of mutation

00:47:24.240 in the patient, the greater the odds of therapy. So if this was 200 years ago, we'd be hosed,

00:47:30.120 right? Because we'd have to wait until they had folating, fungating masses. They were coughing up

00:47:36.740 blood. I mean, obviously that would be crazy. And we've already now just put in great detail that

00:47:41.880 look, even giving high resolution CT scans, at best, we have to wait until a billion cells

00:47:48.940 are cancerous. That's the best case scenario. So now we want to talk about something else.

00:47:55.220 So walk us down the path of how you would, or how you did think about going after another type of

00:48:04.780 biopsy, a liquid biopsy. As I mentioned, you know, I was frustrated by not being able to

00:48:11.000 diagnose a recurrence earlier, and that was an unmet need. So I wanted to start part of my

00:48:16.240 laboratory working on that problem. But to do be able to do that, there's two ways you could think

00:48:21.540 about doing that. You could start with mouse models to build mouse models, let's say what's

00:48:25.540 called preclinical models, in which you try to, you know, grow tumors in the animals, and then see if

00:48:30.580 you could, you know, have an idea, you know, hypothesis for what might be a good biomarker, try to test that

00:48:34.300 and see if it works in a mouse, and then ultimately go back to the human. And the approach that we take

00:48:39.940 in my group is more very tied to the clinic, very what's called translational research. And so for us,

00:48:44.260 really, the human being is the model organism. It's the final model. Of course, it's the model we

00:48:48.260 want to get the answer for. I thought we should do this work directly on blood samples, because if we

00:48:52.940 find something, then it likely will be directly applicable. Otherwise, you'd always still have that

00:48:56.260 second step where it works in the mouse, let's say, but you don't know. So I started just by collecting

00:49:00.400 blood samples. I actually didn't know what I was going to do with them.

00:49:02.160 And by the way, who was funding this work at the time, Max? Was this something you went to the NIH

00:49:06.000 and said, I want an R01, and then this is why I want to study it? Or at the outset, who is bearing

00:49:10.660 the risk financially? When you're a new faculty member, when you start a new lab, when you get

00:49:14.740 that first job, you get startup funds, which is basically funds your department or university

00:49:19.200 gives you, provides you sort of as a, to kickstart you, because you don't have grants at that moment,

00:49:23.280 you're just starting, you haven't had a chance to write them. And so I was using that money,

00:49:26.880 basically. So I was footing the bill on that pot of money I got to start my lab. Because as you know,

00:49:31.860 I'm sure your listeners have heard many times that one of the things with academic research

00:49:36.000 and funding is that oftentimes you kind of have to show it already works before you can get funding

00:49:40.280 for it. And that's a very common story because grant funding is limited. And so organizations

00:49:44.900 that fund research are often quite conservative that if the idea is sounds great, but you have

00:49:49.260 no proof that it works at all, they are not going to give you money. So I use money that I had for my

00:49:53.480 startup funds. So yeah, so I just started collecting blood. I started thinking about what biomarkers

00:49:57.440 would be good. So I read the literature on the protein biomarkers as we talked earlier.

00:50:01.040 And sorry, did you collect blood in patients prior to resection and post resection or just post

00:50:05.900 resection? Both prior to resection or radiation, whatever the treatment was. And then in subsequent

00:50:11.600 follow-up visits, usually, you know, at the time of their follow-up scans, when they're coming back to

00:50:15.940 see us. Protein biomarkers didn't seem very useful. I then spent maybe about a year looking at something

00:50:21.460 called circulating tumor cells, which is one subtype of what this field called liquid biopsy,

00:50:27.580 which is actually looking for intact cancer cells that can circulate in the blood of patients.

00:50:32.380 I tried that with a couple of different techniques and quickly realized, you know, something that many

00:50:37.440 others in the field have also observed, which is that while if one could really do that, it would

00:50:41.500 be super powerful. It is very complicated to actually measure those cells for writer reasons.

00:50:46.640 Two biggest ones probably are, I mean, three biggest ones are the cells are not very abundant when they

00:50:50.340 are there. There's very few of them. So they're very difficult to purify. Like a good purification,

00:50:54.580 they might still only be 1% or less of the cells in your purified sample because it's just so hard

00:50:59.280 to purify. It's finding a needle in a haystack. And because it's intact cells, you have to process

00:51:04.320 the samples really quickly. You have to process them basically the same day or within an hour or two

00:51:10.240 of the patients having the blood sample drawn. That makes it very difficult to build up biobanks of

00:51:15.680 frozen cells, frozen samples that you could, you know, get a large cohort to actually study

00:51:20.640 because you have to process them immediately. So, and then the last thing was that we did some

00:51:25.120 control experiments. This is critical, obviously. You always want to ask, is the thing I'm working

00:51:29.580 on really working? And where we drew blood from healthy individuals who don't have cancer,

00:51:33.500 some of these tumor, circ and tumor cell methods we're using found circulating cells. Those patients

00:51:38.480 don't have cancer. So they were picking up something else. So there was a lack of specificity.

00:51:43.240 So they were technically complicated. It seemed like it's going to be hard to get this into the

00:51:47.000 clinic anytime soon. And there was some weakness. So that's them when I went back and tried to really

00:51:51.240 see what else could we do. So let me go back to the first one. Let's just talk about the protein

00:51:55.460 because that's makes sense as the obvious place to start because we already kind of do that with PSA,

00:52:01.560 notwithstanding the limitation that normal cells, normal prostate cells, non-cancerous cells make

00:52:06.980 prostate specific antigen. Same with CEA, CA99. Did you find any proteins that were made that were

00:52:14.220 expressed by lung cancer cells that were unique to them? We at the time did not. We did not do a

00:52:20.960 holistic screen ourselves, but other people had. There's techniques now called mass spectrometry where

00:52:26.340 one can take a sample of proteins, let's say purified from the blood, run them through this machine,

00:52:31.480 and it basically tells you what proteins were there. And those studies had not found convincing

00:52:36.860 markers that were unique to lung cancer cells. Now, they found some markers that in some patients

00:52:43.320 were elevated. Actually, CEA is one of them that can be elevated in lung cancer patients. And in ones

00:52:48.160 where it is, it can even be a marker of recurrence like PSA can be in prostate cancer. But we already

00:52:52.860 knew that CEA has problems with specificity, right? So that was sort of the main concern. But the main

00:52:58.240 protein work we did was to actually, I drew some CEA and a few of these other markers like C99,

00:53:02.040 some of the ones that have been previously reported to be markers in lung cancers in some

00:53:05.700 of my patients. Because there were some studies that claimed really great performance in the

00:53:10.320 literature. But quickly realized those studies were, you know, maybe it was an artifact of that

00:53:14.600 cohort or something else was going on where, because we could not reproduce those results.

00:53:18.780 Let's talk about that a little bit, Max, because that's, I think what people don't

00:53:22.120 understand is how difficult it can be to reproduce results. How people understand

00:53:27.320 what that process is like, you or one of your graduate students or one of your postdocs reads

00:53:32.080 a paper where somebody says, hey, we've identified this protein, it seems really specific,

00:53:37.800 it looks fantastic, etc, etc. Presumably, these people are collaborative, you can read the paper

00:53:42.780 and get the methods, you can call the lab, they can tell you how they did it. Now you do the

00:53:46.860 experiment. What do you find? In many cases, unfortunately, one finds that one can't reproduce the

00:53:52.980 result. That can be not that it doesn't work at all, but that just it doesn't work as well as it

00:53:59.240 did in the initial study. But it works, you know, so much less well that it's not going to be

00:54:04.880 clinically useful. That's a very common event. And so why does that happen?

00:54:09.260 Now, do you still publish that result, Max?

00:54:11.560 The lack of validation?

00:54:12.800 Yes.

00:54:13.540 Sometimes it depends. Yeah, sometimes we do. Often people don't. The reason is complex,

00:54:20.080 of course. But one issue is that, you know, to definitively prove something doesn't work quite

00:54:26.120 as well as was initially reported. One actually has to do very large studies so that the what are

00:54:30.680 called error bars or estimate of the sensitivity, that's what we care about, is very accurate. But

00:54:35.980 even though by doing a smaller cohort, you can already tell it's not going to work out. But to

00:54:40.700 really ultimately prove it to the level of rigor that one would need for a publication would require

00:54:46.200 another year of, you know, thousands of dollars, right? And so there's these decisions one has to

00:54:51.580 make. Is this really worth it? And what's the opportunity cost?

00:54:54.840 But it's interesting, right, when you think about it, because it means that we have a bias system

00:55:00.500 in the publication realm. We are biased towards positive findings that are often untrue, because

00:55:09.080 let's say one person publishes that they can make this thing work. Okay. And again, we're going to

00:55:13.380 throw out fraud or anything like that. But we're just going to assume that in good faith,

00:55:17.680 these are people who did something, but they had an artifact. They didn't catch it. The experiment

00:55:21.760 looked like it was positive. They publish it. Let's assume that you and four other labs independently

00:55:27.320 try to reproduce it, and none of you can reproduce it. But you all come to the same conclusion, which is,

00:55:33.080 well, we only did it this one way, and it didn't work. Technically, I would need to do it three or

00:55:37.180 four other ways to be really sure. I'm going to put those resources of time and money elsewhere,

00:55:41.800 and away I go. And everybody comes to the same logical conclusion. But of course,

00:55:46.460 that means that that one finding gets disproportionately propagated as a positive

00:55:51.480 finding, when in reality, four labs have now found it to be highly unlikely, and that doesn't get

00:55:58.820 published. Of course, I'm not being critical of you or anybody, Max, other than just the system,

00:56:03.440 which is, I wish there was a way that the incentive structure could be such that we prioritize

00:56:08.320 negative findings as much as positive findings. And I agree with you 100%. In general, we don't

00:56:14.220 have a system where one lab could know that three other labs had done the same thing. Maybe eventually

00:56:20.220 you find out over dinner at a conference or something, right? But because there is no place

00:56:25.120 to look up to say what other people have done, there is no way to know. It would be great if we

00:56:29.780 could fix it. It's a very difficult problem, of course, because of many factors, including the

00:56:33.540 economic parts of limited resources and time. Even, you know, to publish such a study would

00:56:38.380 take, let's say, a grad student could still take six months of their time to get that out. But now,

00:56:44.340 you know, they're not spending those six months on their main project, which is to develop the new

00:56:47.720 task. And I've spoken with graduate students who will say, look, it's really going to be harder for

00:56:52.200 me to get my PhD publishing negative studies. I mean, I'm going to quickly turn my attention

00:56:56.320 to where I think my committee is going to be looking at me more favorably. I think the problem is

00:57:01.560 pretty significant. The selection bias of the positive results is a major issue. And so this

00:57:05.560 is why reproducing positive results is so critical and something that also not everyone likes to do

00:57:11.640 it because reproducing something someone else already did is not making a brand new discovery,

00:57:16.080 right? But more in the medical fields, we do do quite a bit of, right? There is quite a bit of

00:57:20.580 studies that try to reproduce positive results and then publish it when the results are also positive.

00:57:26.160 And that's really to really prove something works. Convince yourself that something is really real.

00:57:30.140 It's really helpful to see multiple studies from multiple groups, ideally with multiple methods

00:57:33.920 that find the same thing. And that's sort of how you can read through the literature and really try

00:57:37.680 to see where is likely a robust finding. All right. So let's go to the CTCs now. So protein

00:57:42.740 out. Circulating tumor cells. Okay. So talk to me about a patient who shows up with a big juicy

00:57:49.540 stage two lung cancer prior to resection. How many milliliters of blood do you take out of that

00:57:56.020 patient preoperatively? So that depends greatly, but these kinds of tests, whether we're doing them

00:58:01.180 for research purposes or ultimately, you know, if you were to develop something for the clinic,

00:58:04.680 usually take somewhere between 10 to 20 mLs of blood. So a few tablespoons.

00:58:11.360 So right out of the gate, how many CTCs do you see preoperatively when you should be at your highest,

00:58:16.540 right? That should be your greatest circulating tumor burden.

00:58:18.680 So in stage one to two non-small cell, most often zero.

00:58:23.440 Oh my God.

00:58:24.240 That was what we found when we looked. That meant the sensitivity isn't very good.

00:58:28.060 Did you have positive controls of stage four patients?

00:58:30.440 We did. And we did see some patients where we saw quite a lot. So you can definitely see patients

00:58:34.000 that have massive amounts. Occasionally, you even see early stage patients that have quite a lot.

00:58:38.340 This is work we didn't do, but another group in the field led by Carolyn Dive from

00:58:42.020 Manchester, where they actually took blood from the draining veins at the time of the surgery,

00:58:48.580 the lobectomy for lung cancer, and did circling tumor cell assays and found evidence that they

00:58:54.000 could find more signal sensitivity was better in that context because the hypothesis being that

00:58:59.120 you're basically measuring the blood outflow of the tumor. You're going to catch more cells that

00:59:04.240 way because once they hit the systemic circulation, they're getting diluted all over the entire body.

00:59:08.880 So there is some evidence along those lines.

00:59:11.040 Any findings about the differences in those early stage? So if you took a person with stage two cancer,

00:59:16.440 who is your typical patient, who has no circulating tumors, even preoperatively,

00:59:21.740 then you said there might be some who had a reasonably high burden. Does that predict who's

00:59:26.380 going to recur, even though based on TNM staging, they're both considered identical postoperatively?

00:59:33.240 Absolutely. There is evidence of that. There are studies that have shown that when you do see

00:59:36.760 high levels, that that is a negative prognostic marker, a bad thing, and that those patients are at

00:59:41.020 higher risk of recurrence. Now, one complication in the CTC field, the circling tumor cell field in

00:59:46.120 general, is that, as I mentioned, even non-cancer patients can have circulating cells that look like

00:59:52.660 circulating tumor cells using the markers that are generally used, which are, you know, when it's

00:59:57.400 looking for circling tumor cells, most assays stain the cells, meaning they look for markers on the

01:00:02.900 cells. They want the cells to not express markers from lymphocytes, from white blood cells. So there's a

01:00:07.440 marker that's highly expressed on white blood cells that should not be expressed on the circling tumor

01:00:11.460 cell. Which one do you use? CD3 or? People usually use CD45 as the most common marker, which is sort

01:00:17.100 of a pan white blood cell marker. On the flip side, for a positive marker, what should be expressed in

01:00:22.080 the cancer cell, but not the white blood cells, they look for things like cytokeratins, which are

01:00:26.220 structural proteins that are specific to epithelial cells. But as we and others have found,

01:00:32.100 healthy patients can also have cells that have cytokeratin expression, but no white blood cell

01:00:36.580 marker expression in the circulation. And other people have gone further to do single cell sequencing

01:00:41.620 of those to show that their genomes are actually normal. Those cells have normal genomes. They're

01:00:46.220 not mutated like the cancer cells are. So it definitely appears that there can be epithelial

01:00:52.120 cells circulating in us that are not cancer cells. Those cells can't divide forever, so they will not

01:00:58.140 ever set up shop anywhere. There's not very many of them. But that makes it complicated to look for the

01:01:03.960 presence of certain tumor cells if one doesn't have a very specific way of getting, removing that

01:01:08.600 signal. Yeah, so that right away tells you this is never going to be, at least that exact approach,

01:01:13.280 can never be a cancer screening tool. And it makes you worry that even in patients who are without

01:01:21.620 disease, but who have been resected, it might not be a great predictor of disease recurrence.

01:01:27.920 So far, the only thing that sounds somewhat promising from this approach

01:01:32.140 is it might help you determine the course of adjuvant therapy in a patient, correct? So if you

01:01:38.680 took a patient who was resected, radiated stage two, had high circulating tumor burden versus one who

01:01:46.640 didn't, you might take that former patient and say, even though the textbook says you don't need

01:01:51.580 chemotherapy, we're actually going to give you chemotherapy because we're going to treat you like 0.62

01:01:55.260 your stage three or worse. That's absolutely where we wanted to get. That exact situation is one of

01:02:00.360 the main motivations for starting liquid biopsy work in my lab. And yes, with very high levels

01:02:05.640 of circulating tumor cells, one could have envisioned if one were to do very large studies

01:02:08.860 to definitively prove that these patients with a certain threshold, even though they're a small

01:02:12.960 minority, that is really very highly associated with recurrence, that could work. But it would miss

01:02:18.060 most patients who ultimately develop recurrence. And so that was the negative. It's not to say in the

01:02:22.540 future, we might not get there. There's lots of work still going on in the circulating tumor cell

01:02:25.700 field, so I don't want to be too negative about it. It's just not as far along as the thing that we

01:02:31.860 ultimately settle on we'll talk about shortly, but more development is required if we're going to be

01:02:36.060 able to use that clinically.

01:02:37.960 Where did you go next? Did you look at cell-free DNA?

01:02:40.480 After that, we went to what's called cell-free DNA, which refers to DNA molecules that are found in the

01:02:47.080 circulation, but outside of cells. So they're circulating in what's called the blood plasma, which is the

01:02:51.760 non-cellular liquid portion of the blood. It's called cell-free because it's outside of cells,

01:02:56.340 free of cells. Of course, it comes from cells originally, but it's now circulating by itself

01:02:59.980 without membranes around it. The reason I got interested in that was actually largely from

01:03:04.760 reading about a different field, which is prenatal diagnostics, where around this time,

01:03:09.960 there was a lot of work being done led by a number of individuals like Dennis Lowe in Hong Kong and

01:03:14.700 Steve Quake, who was actually here at Stanford in the engineering department, basically showing that

01:03:19.200 in pregnant mothers can detect DNA from the fetus in the woman's blood, cell-free DNA,

01:03:25.960 again, not inside cells, but outside of cells. So it's pretty obvious that we can do that in

01:03:31.320 pregnancy. Then, you know, couldn't we try to do the same thing in cancer patients?

01:03:37.740 Just to be clear, Max, when you spin this down, when you take a tube of blood and you spin it,

01:03:42.200 you get all of the cellular matter below the Buffy coat, and then the plasma is above.

01:03:47.780 Is the cell-free DNA in the clear plasma and not stuck within the cellular material? It manages to...

01:03:55.380 Correct. The cell-free DNA is in the plasma, and that's exactly where we isolated.

01:03:59.580 That's actually a critical technical aspect of...

01:04:02.340 Yeah, because if it were in the cellular matter, it would be a disaster to find it.

01:04:05.720 You can't get it, right? And actually, we've thought about this idea. Could there be more

01:04:10.080 cell-free DNA mixed in with the cellular compartment? And I think it's quite likely that there could be.

01:04:14.240 If you could put a gradient on that and try to somehow extract it...

01:04:18.060 Get it out. That you might be able to get it out.

01:04:20.060 And you could enrich your cell-free DNA.

01:04:21.520 You could enrich for it again. And that might actually be a future interesting research direction.

01:04:25.060 But the problem is, currently, if you were to just take some of that cellular compartment,

01:04:29.280 there's vastly more DNA in the cells than there is in the cell-free space.

01:04:32.900 You would basically lose the signal. You would just have all... the vast,

01:04:35.980 vast majority is from the cells in that compartment. So we take the aqueous part,

01:04:38.900 the plasma part, at the top of the tube after you spin it.

01:04:41.380 And there's very low levels. It's only about... on a healthy individual,

01:04:45.760 about one to five nanograms per mil of cell-free DNA in the plasma.

01:04:50.560 By the way, you know what's interesting? I have beta thalassemia minor. So my phenotype is normal.

01:04:56.420 Well, I shouldn't say that. My hemoglobin hematocrit are normal. But my phenotype is that

01:04:59.960 I have very small RBCs. So my MCV is in the 50s or 60s or something like that. And my RBC count is very

01:05:09.280 high. So that's how I have a normal hematocrit because I have a lot of small red blood cells.

01:05:14.240 So our mutual classmate from med school, Matt McCormick, who was my roommate, used to always 0.99

01:05:18.360 refer to me as having shite for blood. And I just learned something very recently. I've spun my blood 0.98

01:05:24.680 a billion times. I've been more times than I can count. And I've always thought, man, I must be

01:05:29.000 dehydrated. Man, I must be dehydrated. Because relative to everybody else's blood, you'll get cellular

01:05:34.080 matter this much, buffy coat, plasma. They're about the same volume. And you can even estimate

01:05:40.940 hematocrit by spinning in a lavender top and looking at the separation. Well, you can't do any

01:05:46.180 of that in my blood because my plasma volume is tiny and my cellular fraction is enormous.

01:05:53.960 And so finally, I spoke to a hematologist who said, oh, the reason is because you have all those

01:05:58.860 little crappy red blood cells, they're actually keeping and trapping a lot of the plasma within

01:06:04.940 the cellular body at the bottom. So it's not that you don't have a normal plasma volume. You do,

01:06:10.600 but you just can't get it out of the separation, which means I'd be a lousy patient for this because

01:06:17.320 I'm cutting my sample size down so much by reducing my plasma volume. So you might have to donate larger

01:06:22.960 amounts of blood to get the right amount of plasma. But I think that's a very interesting observation.

01:06:26.220 I think it kind of does suggest this phenomenon of there potentially being some cell-free DNA among

01:06:30.140 the cells. You're the exceptional outlier, but it's probably also happening in the average patient

01:06:33.940 to a lesser extent. Now, again, it's DNA. So just to make sure everybody remembers,

01:06:38.020 this is a double-stranded or single-stranded DNA? These are largely double-stranded DNAs,

01:06:43.160 we think. We know there's done some experiments along those lines. These are double-stranded DNA

01:06:47.560 molecules. They are rather short, the cell-free DNA molecules.

01:06:51.880 How many? Yeah, it's about 170 bases, 160, 570. Exactly the length of DNA you would expect if the

01:07:00.280 DNA were wrapped around the core histones. Histones, of course, are, for your listeners,

01:07:04.440 are proteins that package the DNA in our cells. They're sort of the scaffold that lets the DNA,

01:07:10.380 which if it were linearly stretched out, would be extremely long, fit into the compact nucleus of

01:07:14.660 a 10-micron cell. It has this special packaging scaffold, which consists of the histones.

01:07:19.580 And so the DNA in the circulation is wrapped still around this core histones, and we think that's

01:07:26.080 critical because there's actually high activity of enzymes that chew up DNA in our blood and our

01:07:31.980 extracellular fluid. And so the DNA is only there temporarily, and it's temporarily protected by

01:07:36.980 these histones. So the DNA that's bound to the histones is present at a higher frequency in the

01:07:42.180 blood than the DNA that's between the histones. The histones are like pearls on a string, and the DNA

01:07:46.620 regions that are between two pearls are relatively depleted from the blood because these enzymes

01:07:51.840 are chewing up the DNA constantly, which is what contributes to these very low levels of DNA because,

01:07:57.340 of course, we have lots of cell death. So you might imagine we should have higher levels of DNA in

01:08:00.340 our blood, but we have a system to clear it out in this DNA's enzyme. And probably important because

01:08:05.220 if you ever work with DNA, if you get high concentrations of it, it becomes very sticky,

01:08:09.260 kind of a snot-like substance, which would be very probably bad to have in your microvasculature

01:08:13.940 for stroke, etc. Does this cell-free DNA primarily come via apoptosis? Or what is the predominant

01:08:23.260 method by which, or mechanism by which, the cells are evicting their DNA?

01:08:29.560 That is very poorly understood. That's a great question. It's still one of the mysteries.

01:08:34.080 And it's a mystery because it's really difficult to study because in a human being or even in an animal,

01:08:38.560 what we know is that most, if not all, tissues contribute DNA to the cell-free DNA pool.

01:08:44.780 You know, it's coming from everywhere, so it's very difficult to study release of something from

01:08:48.680 everywhere. Because it is chopped up in these small histone-bound fragments, there was historically a

01:08:55.340 lot of, most of the reviews and textbooks would argue that it's largely apoptosis that does that,

01:08:59.840 because in apoptosis, of course, there is a laddering of the DNA or chopping up of the DNA as part of the

01:09:04.640 programmed cell. Tell folks what apoptosis is so they know what we're talking about.

01:09:09.200 Apoptosis is a mode of cell death, so how cells in our bodies can die. It is a controlled programmed

01:09:15.400 cell death, meaning that there is a mechanism in our cells that is basically a suicide program

01:09:20.020 that can get activated. This is critical for development, our development, because at certain

01:09:25.060 points as we're developing from a, you know, the fertilized egg up to a full infant, and even then

01:09:30.680 later in life, cells sometimes have to get killed to make room for other cells, and that's just part

01:09:34.940 of our normal development homeostasis. So it's something that we have evolutionarily built.

01:09:39.120 It also is critical for getting rid of sick cells, because if a cell gets sick, there has to be a

01:09:42.700 way of killing it. So apoptosis is a programmed way of cells dying, and as part of that, once that's

01:09:47.720 activated, the cells chop up their proteins as well as their DNA, basically. It's sort of a tide to kind

01:09:54.120 of clean up after themselves in a way. But because of this chopping up the DNA that happens during

01:09:58.640 apoptosis, it was long thought that that likely could be a major mechanism. I think it still could

01:10:03.020 be. The complicating factor is now we have a much better understanding that there's also DNAs just

01:10:06.940 floating in the plasma that are also chopping up DNA. So if a long chunk of DNA were released

01:10:10.780 through a non-apoptosis process, it would likely also become chopped up. And so just because it's

01:10:15.900 chopped up, assume it comes from apoptosis. We don't know for sure. I think that's an important

01:10:19.400 area of research. We think it's likely multifactorial, apoptosis, necrosis, you know, basically any way

01:10:24.400 that a cell can die and release some of its contents into the blood. Okay. So let's say

01:10:28.980 we're talking about a normal person. You've taken 10 cc of blood. You take the plasma from them.

01:10:35.980 How do you quantify how much cell-free DNA you would expect to get out of that? And then furthermore,

01:10:40.920 how are you distinguishing what is potentially cell-free DNA from a normal cell, which presumably

01:10:45.760 is the bulk of it, versus cell-free DNA that is from the cell of interest, the cancer cell?

01:10:50.700 So once we purify the DNA, once we have the plasma, so the non-cellular compartment, there is

01:10:55.340 laboratory procedures for isolating DNA that have been, you know, well worked out for over many

01:11:00.100 decades that we use in all different fields of biology. And so we apply some methods there that

01:11:05.220 have been optimized for these low concentrations of DNA. That's one of the unique things about the

01:11:10.600 cell-free DNA work. The concentration is really low, like single-digit nanograms per mil. And often when

01:11:15.280 we do laboratory procedures, we like to work with at least micrograms of DNA. So there's nanograms,

01:11:19.100 a small amount, and that was one of the challenges early on in the field to try to overcome that.

01:11:22.760 So we purify the DNA, and then we can quantify how much there is. There's a variety of laboratory

01:11:26.900 tools we have. The most commonly used one in our lab is a fluorescent-based method where you

01:11:32.180 basically add a fluorescent dye to the solution. It binds to DNA, and that gives a specific fluorescence.

01:11:38.900 And then you can read that out in a machine that can read that fluorescence. So you can then

01:11:42.020 have a standard curve and read off how much DNA you have and what would I expect for a healthy individual.

01:11:46.720 So an in-situ hybridization, or...?

01:11:49.460 This is just purely a dye-based method. It's much simpler than that. You could do more

01:11:52.920 complicated things. The more complicated thing that we would do next, when we do sometimes in

01:11:56.680 certain situations, is called quantitative PCR, where basically you use an enzymatic method

01:12:02.540 that can amplify DNA. You are going to have a standard curve where you know how much DNA is in

01:12:07.320 there, and you compare how the amplification curves grow, and that can let you read out how much DNA was

01:12:12.300 in your sample. But most of the time we can use this more simple method, which is just add a dye,

01:12:15.240 put it in the machine, get the result. It's very fast.

01:12:17.440 And just give me some quantity. So in 10 ml of plasma, how many nanograms per milliliter in

01:12:25.660 any sample would you expect to have of cell-free DNA?

01:12:28.600 10 ml of plasma would be 20 ml of blood we collect, because the plasma in most individuals is about

01:12:33.460 half. From 20 ml of blood, we get 10 ml of plasma. Those 10 ml in healthy individuals would have

01:12:38.500 somewhere between 1 to 5 nanograms per mil. So that would mean 10 to 50 nanograms of DNA at the

01:12:46.020 end. But in some conditions, in some advanced cancer patients, patients who have trauma, have an

01:12:51.260 infection, the levels can be much higher, much higher than 5 nanograms per mil, because there's

01:12:56.240 active cell death happening in the body. And so when that happens, we sometimes see levels,

01:13:00.380 you know, in the hundreds of nanograms per mil that's rare, but it can happen.

01:13:04.000 So now the big question is, how do you separate the good actor and the bad actor in that pile?

01:13:11.540 The way that we and others in the field do that is to focus on a unique molecular property of cancer

01:13:17.920 cells that normal cells generally don't have, and that is the mutations that cause the cancer.

01:13:23.660 Of course, as many of your listeners will know, cancer is a disease that is caused by mutation

01:13:28.620 mistakes in the DNA of a normal cell that accumulate and ultimately lead to that cell not

01:13:33.660 being responsive to cues to stop growing or to kill itself and just to grow uncontrollably.

01:13:40.060 And so then those mutations, both the ones that cause the cancer as well as many thousands that

01:13:44.900 are long for the ride in most tumors, serve as really exquisitely specific markers of the cancer

01:13:50.660 cell, because these mutations are really only present in the cancer cells, not in the patient's

01:13:55.360 normal cells. So this is a really attractive biomarker. And for a variety of reasons, one is you're

01:14:01.260 tracking the cause of the disease. You're really at the molecular cause of this is a mutation. That

01:14:04.900 is what you're tracking. What better biomarker can there be for an illness? You're actually tracking

01:14:09.080 that. And the second thing is the specificity, particularly when you compare it to the proteins

01:14:13.480 like PSA can be made by normal prostate cells and by prostate cancer cells, but the mutations in a

01:14:19.420 prostate cancer would not be present in the normal prostate cells. They would be present in the cancer

01:14:23.660 cells. How do we tell them if there's circling tumor DNA present is that we look for mutations in these

01:14:32.160 short DNA molecules. So we look for mutant molecules. Most of the molecules in the blood come from the

01:14:37.540 healthy cells of the patient and they don't have mutations, but a small subset, and that can range

01:14:43.060 from an advanced lung cancer patient might have 1% of the circling DNA is from the cancer. So still a

01:14:49.260 small amount and that's high levels. In early stage patients, it's often well less than 0.01 or 0.001%

01:14:57.340 of the DNA that's from the cancer. How do you even detect that without, do you have the resolution

01:15:02.740 to detect 0.1% of the cell-free fraction, which itself is already very small and know that you're

01:15:11.360 not seeing an artifact? The methods we use to try to detect mutations, and there's a variety of them,

01:15:15.960 but the one that we chose to use in which the vast majority of the field has moved to since as well is

01:15:20.560 next-generation sequencing, which is these high-throughput molecular methods for sequencing DNA,

01:15:25.580 where you can identify the actual sequence of the bases, the AGTC bases that make up DNA. You can get

01:15:32.220 the exact sequence for hundreds of millions of molecules or billions of molecules even in parallel

01:15:37.060 in one experiment. It's as if we had a microscope that's so good that we can just read off the bases.

01:15:40.760 It's not a microscope. It's a complex molecular biology procedure, but the end result is the same.

01:15:44.460 We get a data set of millions of DNA sequences that are all about 170 bases in length,

01:15:49.020 and we can then compare those sequences to the patient's germline DNA, let's say, that we purify

01:15:54.740 from a buccal swab or from the cellular compartment of the blood. And we can say the vast majority of

01:16:00.360 the molecules look identical to the patient's germline, meaning their normal healthy cell DNA,

01:16:04.940 but some small proportion of the molecules have mutations. And in many of our experiments,

01:16:10.340 the way that we can be very certain that's the case is because we start by actually sequencing

01:16:14.560 the tumor. So let's say, particularly for this initial problem we had of telling who is cured or

01:16:18.880 not, those patients all had treatment. They're all known to have cancer. So we have a piece of their

01:16:22.880 tumor. We can sequence that and identify what mutations that tumor has compared to the patient's

01:16:28.520 healthy cells. Let's say we find in a particular panel we might be using, let's say we find 10 just

01:16:33.060 for sake of example. So 10 mutations the patient's tumor has that the healthy cells don't. We can then

01:16:38.900 go look in the blood in the sequencing data set we have for those molecules that carry one of those

01:16:44.020 10 mutations. That turns out to be exquisitely specific and sensitive for the presence of the

01:16:49.360 cancer. If we see even one or a handful of molecules that carry the mutation, one of these

01:16:54.060 mutations has borne out that that basically means there's still cancer cells in the body.

01:16:59.240 Now, the other thing that I want to make sure I understand here is we're talking about relatively

01:17:03.420 short fragments of DNA. Call it 150 bases, correct? 170 maybe, closer to 170.

01:17:09.300 99.9% of all of that cell-free DNA lines up to some segment of their germline. But no two of those

01:17:19.740 are the same either. It's like if you could lay their entire germline genome out, it would span

01:17:25.740 miles and miles and miles. You probably have no two segments that are the exact same. So you have a

01:17:30.680 whole bunch of 170 base segments that line up somewhere on that problem. That's a pretty interesting

01:17:37.880 computational problem to me because 170 is not that big. Presumably, some of that 170 also is present

01:17:47.020 in the cancerous, in the mutated. A cancer is only going to have maybe 100 mutations.

01:17:53.380 In the coding regions, exactly. I'd say 100 mutations in the coding regions. In the total

01:17:56.940 genome, if you include the regions between genes, maybe 10 to 20,000 or something. It depends,

01:18:01.080 of course, as a range. But having more than tens of thousands of mutations in the whole genome would

01:18:04.520 be unusual. I guess what strikes me as almost improbable is given the few number of mutations that

01:18:13.100 really exist in the cancer, if only 0.1% of the entire cell-free DNA, which itself are very short

01:18:20.660 segments, it seems to me you could very easily just miss it, right? You wouldn't happen to find

01:18:26.300 a segment that lines up with a mutation. Presumably, there must be a way to do the... I mean, I feel like

01:18:31.420 if you gave me an hour, I could figure out the numbers. It's just a statistics and probability

01:18:36.760 problem. So how infrequent is that? That is exactly one of the major challenges in the field.

01:18:43.160 Our first paper in this field that started really this whole part of my laboratory back in 2014,

01:18:48.340 we addressed that problem. So what there, of course, had been some prior work in cell-free

01:18:52.940 DNA and cancer prior to us. One is always building on the shoulders of former researchers when one

01:18:57.760 does the next project. Maybe just for a brief history lesson, mid of last century was the first

01:19:02.520 observation that there is cell-free DNA present by a French group. They had very, very insensitive

01:19:07.860 methods. So they could only see it in some very, very advanced cancer patients.

01:19:11.780 But even then, they were already at the foresight to say, well, this could be potentially useful.

01:19:15.980 But then for the intervening, I don't know, 50, 60 years, there wasn't that much progress because

01:19:21.680 of this hurdle of there being so little of it and the methods just not being there to measure it.

01:19:26.040 What people had been starting to do around 2005, 2010 in that range was to actually look,

01:19:31.080 use very sensitive PCR methods to look for single mutations. Let's say you know the patient has a

01:19:35.320 P53 mutation. You go looking for that P53 mutation. There weren't ever at that time then methods that

01:19:39.600 could do that more sensitively. But that ran into the problem that you mentioned, which is that,

01:19:43.440 well, now you're looking for one mutation, but often there's less than a cancer genome

01:19:46.820 present in the blood tube. And so you have lots of negative samples just because that mutation isn't

01:19:53.200 there. It's not in the blood draw you took. The way that we overcame that was actually fairly,

01:19:57.640 very simple, which is just to say, well, let's use this next generation sequencing technology

01:20:01.240 where we can get sequence for millions of molecules. And instead of looking for a single

01:20:04.880 mutation, let's look for dozens of mutations from the patient's cancer in parallel. Compared

01:20:09.180 to looking for one mutation, looking for 10 actually increases your sensitivity tenfold

01:20:13.580 by a whole order of magnitude because I don't have to see all 10 mutations to know the cancer

01:20:18.080 is still there. If the mutations are specific, I just need to see one of the 10.

01:20:22.200 And you don't care if they're coding or non-coding?

01:20:24.480 We don't care if coding or non-coding, and we've successfully used both. What is beneficial

01:20:28.380 is to have what's called a truncal or clonal mutation, meaning a mutation that's present

01:20:33.100 in every cancer cell. So as cancer develops, as we talked about earlier, by acquisition of mutations,

01:20:39.280 now the cell starts healthy, it gets the first mutation, and then ultimately gets to be a cancer

01:20:43.220 which has tens of thousands of mutations. Those happen over time, right? Years generally. Many of

01:20:48.440 those mutations are present, many of the 10,000 are actually present in all cells of the tumor,

01:20:53.180 even though they're not driving the cancer just because it's the history, the evolutionary memory of that

01:20:57.460 cancer, right? That the cell that ultimately was able to continuously divide already had these 10,000

01:21:03.120 mutations from the precursor process. And so once it then gets the final mutation that lets it divide

01:21:07.940 forever, it keeps all those to other 10,000 mutations. So we don't care. We just want the

01:21:12.560 mutation to be present in all the cells. And it's interesting because driver mutations have

01:21:16.140 surprising heterogeneity, right? I was actually talking to Steve Rosenberg about this on the podcast

01:21:20.780 last year, and I was amazed at the non-overlap of driver mutations. And so yeah, a lot of people had

01:21:28.000 p53, but it wasn't necessarily the driver. So it was present. And also, I mean, this gets back into your

01:21:33.300 other thing that you're interested in, which is the immunology, is how often it was not a mutation that

01:21:37.220 was antigen-inducing. That was immune response-inducing, yeah.

01:21:40.080 So let's now talk about, is there such a thing as cell-free RNA, or is that just so unstable that it's not even

01:21:46.200 worth thinking about? And is cell-free DNA the only place to be thinking about this?

01:21:51.140 No, absolutely not. There is such a thing as cell-free RNA. That field is much more nascent than the DNA

01:21:57.480 field. What advantages would it offer, given the instability of the molecule? I think there's some

01:22:02.500 critical questions initially about what is the stability of the molecule, you know, how much of it

01:22:07.040 is there, and how stable is it? And then you miss the non-coding mutations as well, right? If you're looking

01:22:12.580 for mutations, I don't think it has advantages, really. My vision is that ultimately, we want to

01:22:17.100 get to the point where we can, from a blood draw, determine as much as possible about a patient's

01:22:22.260 cancer, ideally, eventually to the point where we don't even need to biopsy it. Now, that is science

01:22:27.580 fiction currently, but if you extend the line of work decades into the future, that is where we're

01:22:32.560 trying to get. If we want to get there, we have to be able to measure things other than mutations.

01:22:38.060 Mutations are important. They're critical, but they're only a small part of the puzzle.

01:22:42.640 Measuring RNA could tell you about which genes are on and off in the cancer, theoretically.

01:22:47.340 And so, that is critically important. For example, in the immunotherapy field that you mentioned,

01:22:52.220 there's some markers like PD-L1 that can be expressed in the tumor cells that can basically

01:22:58.360 hide the cancer from the immune system. That marker is, that's not a mutational process in the

01:23:03.680 vast majority of patients. It's expressed from the promoter, you know, just turned on for reasons

01:23:08.780 having to do with signaling, epigenetic reasons, et cetera. You can't tell by looking at the sequence

01:23:13.900 in DNA of PD-L1 whether the tumor has this marker. It's not a mutation marker. So, we'd like to be

01:23:19.100 able to measure something like that from the blood. We'd like to be able to tell the difference between

01:23:23.680 an adenocarcinoma and a squamous cell carcinoma, or between a lung cancer and a breast cancer from

01:23:27.580 the blood. Some of those things you can do with DNA potentially, but a lot of them you can't,

01:23:31.060 and they might be better done with RNA. It's a fascinating area. My group is working in this area.

01:23:35.480 We haven't published our work, but we will hopefully soon. We have some exciting primary

01:23:39.980 results. I think in the future, RNA will be a big part of the liquid biopsy puzzle.

01:23:44.820 That doesn't replace DNA. It offers complementary pieces of information that you can't get from DNA.

01:23:51.180 Now, is there a difference between cell-free DNA and circulating tumor DNA? Sometimes I've heard

01:23:56.360 people use those interchangeably. Are they interchangeable, or is there a distinction?

01:24:00.060 They are different. Generally, cell-free DNA refers to the total DNA in the circulation. That

01:24:04.280 includes both the healthy cell-derived DNA and the cancer cell-derived DNA.

01:24:08.280 And circulating tumor is just the subfraction of that that's the tumor?

01:24:12.220 That 1% that's from the, or whatever that's from the tumor, exactly. So that's the difference

01:24:16.940 in the terminology. Are there any other signatures that can be helpful here? For example, looking at

01:24:23.020 methylation patterns on the DNA, can that be informative at all?

01:24:27.560 Absolutely. DNA methylation refers to a chemical modification of the DNA molecules. So again,

01:24:32.660 we have these about 170 base pair molecules in the circulation. And at certain of the nucleotides,

01:24:38.240 there can be chemical modifications that are put on by enzymes, particularly a methylation,

01:24:42.900 a methyl group, a chemical group. And that is put on not the same in every cell. Different cells,

01:24:49.080 based on largely what tissue they're from, have different patterns of this methylation. And it has

01:24:53.660 to do because these methylation marks, these kind of modifications can influence which genes are

01:24:58.420 turned on and off. You could envision that in a lung cell that needs genes for surfactant production,

01:25:02.480 or making LVLI or whatever, needs those genes on, but it doesn't need the genes on the fat cell to

01:25:08.380 make fat because it's not doing that. Some genes are methylated and turned on or off in different

01:25:12.920 cells in different ways. So they can be exquisite marks of the origin of the DNA. And so there's been

01:25:17.860 a lot of work in the last couple of years looking at methylation marks to do that. And it has some

01:25:24.020 potential advantages because unlike mutations, it's actually much more stereotyped. Methylation profile

01:25:28.880 of a lung cancer of two lung cancers is more similar than their mutation profile.

01:25:33.580 So in other words, you can use the mutations to identify and distinguish between tumor and

01:25:37.620 non-tumor, and you use the methylation to hopefully zero in on tissue of origin.

01:25:42.840 You could, for example, yes.

01:25:44.340 If you were using it as a pan screen.

01:25:46.000 For screening purposes, exactly. Which is a little, it's a different scenario than

01:25:49.420 this initial scenario of saying, I have a patient who I've treated with surgery or radiation.

01:25:53.560 I want to know if they're cured. Now, I think an important thing to recognize is that the

01:25:57.740 sensitivity of the methylation-based approaches is significantly inferior to that of the mutation-based

01:26:03.440 approaches. But from a practical standpoint, the methylation approaches have some advantages for

01:26:08.020 the screening question. But the sensitivity and the practicality aren't aligned. You'd want the most

01:26:13.600 sensitive assay possible for screening because the tumors are tiny. But that is not the current state

01:26:19.760 in the field is that the mutation-based methods are much more sensitive. So we recently published a

01:26:25.560 paper, a new method that's part of our third-generation mutation-based method, that's about

01:26:31.080 100 times more sensitive than our prior methods as well as other methods in the field that can now get

01:26:38.120 down to one part in a million. So we can detect maybe where the prior methods were kind of bottoming

01:26:44.480 out at about 0.01%. The prior mutation-based method around 0.01%, so one in 10,000. This new method can

01:26:50.760 get to one in a million.

01:26:53.040 So a two-log improvement.

01:26:54.420 Two-log improvements. A log improvement is a dramatic improvement in a diagnostic assay. And so

01:26:58.820 with that assay, we have lots of patients where we had sample leftovers so we could use our first-gen

01:27:03.720 assay and now come to this new assay. And we had false negatives with the first-gen assay that now we can

01:27:07.800 pick up with this newer method. It wasn't that there was no ctDNA. There was circling tumor DNA. It just was not

01:27:13.580 at the level that the prior generation could detect. With the methylation-based assays, it's actually,

01:27:19.020 the data is still not very mature to know exactly what their sensitivity is, but it's probably closer

01:27:24.280 to one in a thousand. It's a 0.1% the sensitivity. Of course, to be clear, what makes the mutation-based

01:27:31.340 method so much more sensitive is you know what you're looking for. You have the mutation a priori.

01:27:37.100 In the methylation method, you don't. If you're doing a pan screen for someone who doesn't

01:27:43.560 have cancer, by definition, you can't know what mutations to look for. So how does that gap get

01:27:49.820 closed? Because if you look at a test like GRAIL, for example, so GRAIL is, I believe, looking at

01:27:56.580 exactly this method, right? They're using, I think they're looking at methylation patterns of cell-free

01:28:02.440 DNA, but they're doing it as a pan screen. So it's give me 10, 20 ml of blood, and we're just going to

01:28:09.360 look for all the cell-free DNA. We're going to identify if there's cancer or non-cancer in there,

01:28:13.260 if there is cancer in there, we're going to try to predict the tissue of origin.

01:28:16.920 How good can that get? Because right now, the sensitivity for all stages is probably 50%

01:28:23.180 with a specificity of about 99%. It's above 99%. But if you look at stage one, if you're really

01:28:30.440 trying to catch an early cancer, we're talking about a sensitivity of maybe 20%, right?

01:28:34.940 This gets into a very interesting part of this diagnostic field in general, particularly this

01:28:38.920 liquid biopsy field with regards to early detection, which is how results are presented.

01:28:43.060 And you put your finger on it. We care about sensitivity and specificity. We want specificity

01:28:47.060 very high, 95%, 99%, something in that ballpark. And we want high sensitivity. When studies report

01:28:53.500 sensitivity across stages, I have an issue when that is done because you can easily bias that by just

01:29:00.340 including more stage four patients, right? You could include one stage one patient and 999 stage

01:29:05.040 four patients and say, I have sensitivity of blah from stage one to four. It's important to break it down

01:29:10.100 by stage. And actually, even within stage one, it should be broken down because for like a lung cancer,

01:29:14.760 let's say, we break stage one into stage 1A, stage 1B, and even stage 1A1, stage 1A2, stage 1. So it gets

01:29:20.760 very granular. And that all relates to small and smaller tumors that have better and better outcome

01:29:25.300 that are the ones you want to catch. In lung cancer, actually, the sensitivity for stage one lung cancer

01:29:31.040 in the grail day that's been presented is actually 5% or less for stage one lung cancer.

01:29:36.240 Let's make sure people understand the math on that. If it's bigger than stage one, you don't need a liquid

01:29:41.820 biopsy. You can do a low-dose CT and you'll see it. But in theory, the only reason you'd care about a liquid

01:29:47.860 biopsy is if we're talking about fewer than a billion cells. If you had a patient, you could do either test.

01:29:53.000 That's true. There are some potential advantages of the blood-based test as a way of getting more patients

01:29:57.960 screened. I think there are some practical reasons why you could argue. If people have less access to care.

01:30:03.340 Access or they're hesitant to get the scan. Yeah, fair enough. But let's just say I want to make

01:30:08.300 sure people understand the following. If a test has, and this is going to be, I think, amazing for

01:30:13.320 people, 50% sensitivity, 99.5% specificity. If you said that that was your stage one performance

01:30:23.280 in an unbiased sample, I think people would be like, wow, that's pretty good. For people who don't

01:30:29.040 have cancer, for whom we believe don't have cancer, in whom we don't have any mutational

01:30:34.000 information, I have a test that is 50% sensitive and 99.5% specific. Do the test. Okay. Here's how

01:30:43.140 I explain this to my patients. What is the pre-test probability that you have cancer? So if you're one

01:30:49.800 of my patients and you're 45 years old and you're not a smoker and your family history is relatively

01:30:54.960 normal and oh, by the way, your colonoscopy is negative. In other words, your pre-test

01:30:59.380 probability is 1%. And I encourage everybody to do this. And for the show notes, we're going to

01:31:05.240 include my calculators. I built a little calculator, but there are apps that do this. I just do it in

01:31:10.300 Excel, but you can download an app that will allow you to do this. You plug in sensitivity,

01:31:15.900 specificity, and prevalence, and it will spit out positive and negative predictive value. Well,

01:31:21.020 if you use the numbers I just gave you, 55%, 99.5% specificity, 1% on the prevalence,

01:31:28.460 if my memory serves me correctly, your negative predictive value goes to 99.7%. Let's just say

01:31:37.520 you put in 99% specificity. It goes to 99.5% negative predictive value. And they say, well,

01:31:44.660 that's really good, isn't it? And I say, yeah, but it's only a little bit better than what it was

01:31:49.140 at the outset. Your pre-test probability was 99% negative. I just took you from 99 negative to

01:31:54.920 99.5 negative. And by the way, your positive predictive value is 40%. If I get a positive

01:32:00.680 test in you, it's more likely than not a false positive. I don't want to be the guy that's

01:32:05.660 raining on the parade because we are using these tests in our patients, but never in isolation.

01:32:11.660 That's a lot I loaded into that. So feel free to comment on any and all of it. But where I want to go

01:32:16.460 with this, Max, is without the mutational information that makes the type of testing

01:32:22.180 you're doing in patients with known cancers so sensitive, how can we make these things more

01:32:29.260 than a parlor trick in patients who don't have cancer, at least at the surface?

01:32:34.960 Excellent point. You laid out some things that I often do in my talks about these things.

01:32:39.060 This is something that's even not well understood, not just by patients, but by a lot of

01:32:43.080 researchers and physicians, this issue that something that sounds great, 50% sensitivity,

01:32:47.520 99% specificity, even ignoring for the fact that the stage one sensitivity is actually less than 5%,

01:32:52.940 like there's still a state on the surface, the 50% where it's where we're at. That if you already

01:32:57.140 basically have no risk of having cancer, you're less than 1% chance of having cancer, the test isn't

01:33:01.720 really moving the needle for you significantly. And as a high risk of catching false positives, as you

01:33:07.080 mentioned, because the positive predictive value isn't so good, that leads to lots of further

01:33:10.760 testing anxiety. Often we can't find anything, but then patients are worried for years that is

01:33:17.100 there something brewing? Is it just the next scan will show it or the next time? So I have lots of

01:33:21.420 concerns about that. I'm one of the researchers that's really focusing on this field, and I very

01:33:25.080 much see these issues and am concerned about them. This is why I really strongly feel that we as the

01:33:30.980 field need to ultimately really do studies that prove cancer-specific survival benefits of these tests.

01:33:38.180 And that is currently not really on the roadmap. The commercial efforts, of which there are many,

01:33:44.780 they're not planning on doing this large randomized trials, like the low-dose CT study we showed,

01:33:49.120 we talked about earlier, that proved that you save lives from lung cancer. Why not? Well,

01:33:54.280 they're expensive and take years, and it's not attractive from an investment standpoint.

01:33:58.460 But if we really want to be sure that we are helping our patients and we're not just adding

01:34:02.680 costs to the healthcare system, that's what we have to do. And we don't know if the sensitivity

01:34:08.540 as of the best tests as they currently are, are good enough to save lives. And there's multiple

01:34:13.980 reasons for that. But an important reason is that it gets very complicated, but you got to look at

01:34:19.700 by stage, because you can game that. Even within stage one, you got to look by stage 1A1,

01:34:23.280 1A2, 1A3. But even once you're at that point, it gets even more complicated because there's actually

01:34:30.480 really, let's say we have stage 1 lung cancer patients. There's two types of stage 1 lung

01:34:33.660 cancer patients. There's a stage 1 lung cancer patients where the cancer cells have not left

01:34:37.820 the lung. That's where they are. That's the only place they are. And the surgeon removes the tumor

01:34:41.740 and the patient is cured. That's one type of stage 1 lung cancer patient. The second type of stage 1

01:34:47.000 lung cancer patient looks the same, has the exact same CT scan, has the same surgery,

01:34:50.280 but already has microscopic cells in the liver and the brain that we don't know about.

01:34:54.500 They're stage 1. We call them stage 1 because that's all we can see. So stage 1 is actually

01:34:59.160 heterogeneous. There's two subtypes of it. And in order for a screening test to be useful,

01:35:05.200 it has to catch a significant fraction of the first type of stage 1, the stage 1 that doesn't have

01:35:10.660 micrometastases, which the surgeon can cure. Because I think what's implicit in your comment,

01:35:16.200 Max, is that the surgeon's not going to cure the second one. Good point. Yes. So the surgeon

01:35:20.180 can cure what he can see or she can see, what they can cut out. If there's micrometastatic

01:35:24.740 metastases in the liver and the brain, the surgeon isn't cutting those areas because they don't know

01:35:28.300 about them. And so those will eventually grow back. That is why simply saying a test increases the

01:35:35.800 number of patients diagnosed at earlier stages doesn't automatically prove that that test will

01:35:40.980 save lives. When you look at the very granular level, you consider this issue about that there being

01:35:45.640 different kinds of stage 1 patients. You know, you can envision a test that mainly catches the

01:35:49.940 patients with micrometastatic disease. And actually, we have hints that the ctDNA assays

01:35:53.820 preferentially catch those patients because of the things we talked about at the beginning with

01:35:58.300 there being now, you know, lots of microscopic deposits spread throughout the body. If you were

01:36:03.180 to put those all into one place and add them up, you would have many billions of cells.

01:36:06.360 Therefore, overall, you have more circulating tumor DNA. And you can detect it, but you think they're

01:36:11.620 stage 1 because that's all you can see on the scan. How do we prove that the tests are not falling

01:36:16.620 into that pitfall is we have to do randomized studies. That's the only way we can prove it.

01:36:21.360 And to me, that's a major concern. And there's this push and pull about, well, these are great

01:36:25.540 tests. They're very promising. And obviously, I wouldn't work on it if I didn't think ultimately

01:36:28.560 these will save lives. But should we withhold them from patients?

01:36:32.240 From non-cancer patients especially.

01:36:33.940 Okay, we don't know yet if they're going to save lives, but isn't it unethical to not give them?

01:36:37.480 Because, you know, if we later find out, if we 10 years from now or five years from now find out

01:36:40.280 they save lives, now there's five years of patients who didn't have the benefit. That is

01:36:43.900 an argument that's often put forward. The problem with that argument, which is why ethically it

01:36:48.240 sounds, of course, very reasonable, is that if by that argument, we would never test anything new

01:36:52.560 that we develop in medicine, because there's always a possibility that that thing could improve things.

01:36:57.140 And that's, I think, why the FDA has taken the posture, because right now there are only four

01:37:01.580 approved liquid biopsies, Gardant and three others.

01:37:05.260 Correct. But those are for very different things, right? Those are for genotype. It's something we

01:37:08.620 haven't really talked about. Those are for identifying what mutations are present in

01:37:12.080 patients with advanced disease. That's what those tests are.

01:37:14.880 Let's actually talk about that. Tell folks what Gardant does. It's an FDA-approved diagnostic,

01:37:19.780 but explain to people what it's used for, because I get asked this question all the time.

01:37:23.420 So the Gardant assay, and they're one of the first companies in this field,

01:37:27.360 they use a very similar method to what we use, a next-generation sequencing to measure circling

01:37:31.760 tumor DNA. Their initial test, the one that's approved, was developed for a very specific

01:37:36.620 clinical problem. You have a patient with metastatic cancer, let's say lung cancer,

01:37:41.700 multiple parts of the body, and you want to identify what mutations that patient has,

01:37:46.580 because in lung cancer, we have drugs that work for patients with certain treatments based on

01:37:51.400 mutations. It's what we call actionable. If we know what the mutation is, and if we have a drug for

01:37:54.720 it, that's the first thing we should treat the patient with. Historically, we always had to do a

01:37:58.180 biopsy or surgery to get the piece of the tissue to identify the mutations. We, you know,

01:38:02.640 the field, Gardant, we, others have shown is that in many patients with advanced disease,

01:38:06.960 because like, say, 1% of the circling tumor DNA is from the cancer, that's enough that you can

01:38:11.160 identify the mutations without needing to do the biopsy. So that's potentially very useful,

01:38:16.360 particularly in patients who can't have a biopsy or where they had a biopsy, but all the tissue

01:38:19.600 was used up. There's lots of cases like this. Or where the biopsy missed the tumor, that happens

01:38:24.040 a lot, you know, a certain subset of the time. In those cases, it can be super useful to have a blood

01:38:28.660 test to say, oh, yet this patient does have an EGFR mutation, so they should get on the EGFR

01:38:32.660 tyrosine kinase inhibitor. That's what those tests were developed for. Now, that is the easiest problem

01:38:37.900 in the liquid biopsy space, because you're in stage four patients that are known to have cancer,

01:38:41.780 they have lots of disease in the... High tumor burden.

01:38:44.080 High tumor burden. And so those tests are optimized for that. They work well for that. They have good

01:38:51.400 agreement, a high 80% agreement with biopsy. They can sometimes find mutations that the biopsy

01:38:56.780 misses because of technical issues with the biopsy, as I mentioned. They're good tests for

01:39:00.540 that problem. They are not designed for, number one, early cancer screening or detection, where the

01:39:05.860 levels of circling tumor in a day are many logs lower. And they're not designed for this question

01:39:10.500 of, is the patient cured or not after treatment? Again, where the levels are much lower. They're

01:39:15.180 designed for this one thing, and that's very important to recognize. That test, if you were to

01:39:19.780 go to the doctor and say, you know, I want the test to see whether I'm cured or not, FDA-approved

01:39:23.420 tests, don't do that. Is that true of... Is it CellSearch and Foundation1 or a couple of the

01:39:28.800 others that are approved, right? Yeah. Foundation has a liquid biopsy test,

01:39:32.760 very similar to what we developed in Garden, and they do the same thing now. CellSearch is actually

01:39:36.660 a circling tumor cell assay. It was FDA-approved. It actually was approved before any circling tumor

01:39:42.080 DNA assay. You know, we had talked about the circling tumor cells at the beginning. At the very

01:39:45.760 beginning of the liquid biopsy field, like in the early 2000s, that was the focus. Everyone was

01:39:49.260 focusing on the cells, and so that's why that was the first test. It is FDA-approved. No one really

01:39:53.880 uses it because it doesn't provide you actionable information. Your doctor runs that test. Whatever

01:39:58.820 the result is doesn't impact how the patient is treated, and so therefore that has grown out of

01:40:04.040 favor. And there is a difference between FDA approval and actually clinical usefulness that

01:40:09.140 is important to recognize. Now, when you look at a test like GRAIL, for example, the FDA has not

01:40:13.820 approved it yet, but it is still permitted in use. Explain that distinction. Now, this is a complicated

01:40:21.060 area in diagnostic space in the U.S. where there's really two kinds of ways tests can be regulated by

01:40:30.480 the government such that they can be used on patients. One is the one that's well-known as FDA

01:40:35.720 approval, which is, of course, an agency that focuses on approving drugs and diagnostic tests

01:40:39.920 and evaluates them carefully and then allows companies to give them approval to market them

01:40:44.680 if they pass the bar of what you need to show. The second way is actually a much easier way to get

01:40:50.640 assays out to patients, and that is by setting up the assay in a lab that is compliant with a CLIA

01:40:57.380 act, an act of Congress that focuses on regulating laboratories that do diagnostic tests that is

01:41:03.400 independent of the FDA. The way that roughly works is that the lab itself gets this designation that,

01:41:08.480 yes, you guys are doing things in an appropriate way. They get inspected every year or two and

01:41:12.240 make sure they're doing things appropriately. And then those labs have the blessing to develop any

01:41:16.580 tests that they want using the procedures that they're supposed to follow with running control,

01:41:21.000 those kind of things, and then offer them to patients. So the FDA never reviews those. It's really

01:41:25.520 a much less highly regulated because basically the individual tests aren't regulated. It's the

01:41:30.380 laboratory that's regulated. And that is what GARDEN did before they got their FDA clearance. 0.80

01:41:35.320 That's what GRAIL is doing. That's what all the diagnostic companies, they start by building a

01:41:39.520 CLIA laboratory and then building their assays in that frame because they're much less regulated.

01:41:45.160 So you can start selling them to patients and providers long before you'd ever get FDA approval.

01:41:50.060 So from a technology standpoint, I mean, there are like 30 other companies that are out there doing

01:41:54.820 this right now. And not to use the word GRAIL again, but the holy grail is the blood test

01:42:02.020 that's going to take a patient who doesn't think they have cancer, screen them with a high enough

01:42:08.540 sensitivity and specificity that I do the blood test. And it says, Peter, we're afraid that you have,

01:42:15.420 you know, colon cancer. What? I had a colonoscopy two years ago. Well, it looks like you have it

01:42:21.000 again. Go get a colonoscopy next week. Lo and behold, there's a tiny little adenomatous polyp there

01:42:27.760 that is an early stage one. And gosh, I need just the smallest little partial colectomy and I'm

01:42:34.420 cured because that is a hundred percent curable cancer. Or, you know, a woman gets a blood test 0.99

01:42:40.460 and it says you have breast cancer. And she says, that can't be. I don't feel a lump. My mammography

01:42:45.760 was negative six months ago. Understood. Maybe you should go get a diffusion weighted image MRI and

01:42:51.340 look at that breast a little more closely. And sure enough, there it is. What would really be amazing

01:42:56.700 is if we had the confidence in these things that even if they were never showing up on imaging,

01:43:02.260 you could treat them. Now we're getting back into science fiction. What's science fiction?

01:43:06.420 Science fiction is in five years, you're going to have a one centimeter pancreatic adenocarcinoma.

01:43:14.000 You and I, Max, both know that even a one centimeter pancreatic adenocarcinoma has a 25%

01:43:21.520 five-year survival. That is a death sentence of a cancer, unfortunately, even at stage one.

01:43:28.500 So waiting until you can see a billion cells on the CT or MRI is not going to be the way to remedy

01:43:36.500 pancreatic cancer. It's either going to be much earlier detection or treatments that actually work.

01:43:44.080 So today we have neither of the above. Is there a path to this sci-fi world that I'm dreaming of,

01:43:50.040 where using the blood alone, we can actually start to develop a high enough sense of confidence that

01:43:57.720 that patient does indeed have cancer, but they might be three or four years away from it being

01:44:02.980 clinically detectable? I am hopeful that we could get there. It's not going to come in one step,

01:44:07.600 but we are already taking steps towards that. So building the, you know, the bricks on top of each

01:44:13.040 other. And the reason I say that is that the easier application, again, is in the patients who

01:44:19.040 are already known to have cancer where we have these super sensitive tests that we can then run

01:44:23.480 after and see who's cured or not. So that we now have, we have these tests that can detect the state

01:44:29.640 that's called minimal residual disease, meaning microscopic cells that are residual after your

01:44:34.140 treatment. And that have currently at least very high positive predictive value, meaning that if you

01:44:39.980 detect the CT DNA, the patient is very likely going to have the cancer grow back.

01:44:44.160 And is that guiding therapy? Is that immediately now sending you to adjuvant therapy right away?

01:44:49.640 That's exactly where it's going. There's now studies, clinical trials underway to do that. So

01:44:54.980 we've recently launched a couple here at Stanford and lung cancer where we're taking patients with

01:45:00.240 early stage lung cancer. They can have one of the trials that it's very broad. They can have radiation

01:45:04.740 or surgery, whatever it is appropriate. If they need chemo for their standard of care for their stage,

01:45:09.700 they get that too. When they're done with every standard thing, they get the blood test and we do

01:45:14.540 the, this minerals agility test. If it's positive, we give those patients immunotherapy. We can't see

01:45:21.340 any cancer anywhere in the body, but we know they have microscopic cells left behind. And that is the

01:45:28.020 moment in the patient's lifetime with their cancer that they will have the least number of cells they'll

01:45:32.760 ever have.

01:45:33.680 And the fewest mutations.

01:45:34.720 Exactly. The lowest heterogeneity, right? The lowest chance of having resistance.

01:45:39.380 So I'm very hopeful that by doing that, we will be able to cure more patients. And we already have

01:45:44.140 evidence of that because we know adjuvant treatment, even given the brute force way,

01:45:47.300 like we've done for decades, where we basically give it to everybody with a certain stage where

01:45:50.740 most are cured.

01:45:51.280 Yeah. The same drug in adjuvant is more effective than the same drug in metastatic.

01:45:56.320 Absolutely.

01:45:56.940 Yeah. I think this is a very important point that gets lost on the part of many people in the field,

01:46:02.800 which is early detection absolutely matters. And you have to look no further than that simple

01:46:09.980 point, which has, I've never seen it refuted. Less tumor burden, less heterogeneity equals better

01:46:16.440 outcomes. It's an axiom.

01:46:18.640 It's absolutely true. And that's because we now understand much better that why cancers are become,

01:46:22.720 initially respond and become resistance is because of this genetic heterogeneity that they already have

01:46:27.420 mutations. They're not all the same, the cancer cells actually all two, no two cells are probably the

01:46:31.380 same. They're all different. They all have slightly different, a handful of mutations that are

01:46:34.820 different in between each cell. And the more cancer you have, the higher chance that one of

01:46:38.520 those mutations will make the cancer resistant to whatever therapy you're using. So this idea of

01:46:42.820 guiding adjuvant therapy based on CTNAMRD, I think is the first step towards the future that

01:46:49.120 you're envisioning or dreaming of.

01:46:50.460 So let's think about which cancers that's going to be relevant in. It will not be relevant in pancreatic

01:46:55.660 because everybody's going to probably need adjuvant. I mean, if you're really being honest,

01:46:58.960 it certainly is relevant in your field of lung cancer. It's certainly going to be relevant in

01:47:04.800 breast cancer. It's certainly going to be relevant in colorectal cancer. You could potentially argue

01:47:11.020 prostate, although the treatment is pretty bad. That really takes care of kind of the big five right

01:47:16.660 there.

01:47:17.500 I throw in bladder, which is one of the six or seven most common cancer. Some of the more rare ones,

01:47:21.620 it'll be useful with melanoma. Of course, it's much more rare, of course, but where we also give

01:47:25.160 adjuvant. The way that I think of it actually is it's going to be useful in the cases where

01:47:29.560 a minority of patients develops recurrence. If you have a situation like pancreas cancer,

01:47:34.560 where the vast majority of stage one patients will develop recurrence, that's not the place to start

01:47:39.340 this kind of a developing this kind of a test because like you said, the tests aren't ever going

01:47:43.620 to be 100% perfect. No test gives you 75% chance of recurrence. That's not the place to do it. It's to do

01:47:49.140 it in places where there's a subset that recur. And in many cancers, actually, that may be at certain

01:47:53.200 stages you don't need it. Stage three colon cancer, well, maybe we don't need it there.

01:47:56.980 Yeah, we know the probability of recurrence is high enough. We should just treat as accordingly.

01:48:01.380 Maybe we should just treat it. But stage one, two, that's a different story, right? The minority 1.00

01:48:05.060 recur there. So I think that's how I think of it, where to first focus on.

01:48:09.520 We've talked a lot about where we are on call that stage one, program one, identifying the high risk

01:48:15.480 patient who needs adjuvant therapy, great progress on lung. What is the state of the art for that exact

01:48:21.900 problem on breast, prostate, and colorectal cancer?

01:48:25.900 So there are active studies underway. Colorectal cancer, actually, I would argue, might be the

01:48:30.860 furthest along. There actually are some Medicare-approved tests using ctDNA from a company called

01:48:38.140 Natera for colorectal cancer that I believe can be reimbursed in some patients. Actually, even though

01:48:44.100 there hasn't been proof of benefit of that at treating based on it matters, but that's something

01:48:49.260 that got approved recently. So I think in those patients, there's a subset of them that can already

01:48:52.780 be done in a doctor's office. But there are many trials going on in colorectal cancer, actually many

01:48:57.300 more than in lung cancer currently. And there are trials going on in breast cancer to also do similar

01:49:02.520 things. Now, breast and prostate are actually more challenging for a couple of reasons. But one is

01:49:08.620 they have very low levels of circulating tumor DNA. It seems that there's difference between tumor

01:49:12.340 types and how much circulating tumor DNA they shed for a certain amount of volume. And they seem to have

01:49:17.660 lower levels. So it's a little more challenging. So you need to use these very, very sensitive

01:49:22.320 mutation-based methods and probably even the subset of mutation-based methods that's most sensitive.

01:49:28.020 So I know there's studies going on in breast cancer. I'm actually not aware of prostate cancer studies

01:49:32.620 in the ctDNA space because of this issue of the very low concentrations. But I wouldn't be shocked if

01:49:38.060 there are now some places that are doing a few. Yeah, prostate seems to be a very privileged site.

01:49:42.800 Even the GRAIL test, I think they say, really, it doesn't screen effectively for prostate because

01:49:47.060 you're just not getting enough cell-free DNA in the circulation. Or breast, actually. The GRAIL's

01:49:52.160 development plan initially was largely around breast, but they then pivoted because

01:49:55.580 the early results were that breast also didn't do very well.

01:49:59.500 Would it be that the next stage of patients you're interested in treating are patients who are,

01:50:06.440 say, post-adjuvant therapy or who were deemed low enough risk to not need adjuvant therapy,

01:50:10.980 but who are still higher risk than the general population in whom you're screening for recurrence?

01:50:16.720 Would that be kind of the next group you would want to develop a liquid biopsy to assess?

01:50:21.120 If we can get proof of principle of that this approach works in these early studies,

01:50:24.720 I think the logical steps would be exactly what you say, which is you could envision a future where

01:50:29.280 maybe even you move into the more high-risk patients with higher risk of recurrence,

01:50:33.560 where there's still a substantial number that don't recur. But you do this test repeatedly,

01:50:38.640 let's say every three months, and you don't give adjuvant until you see a signal that the test is

01:50:43.400 positive. So maybe that way you minimize missing patients and you can still catch them months before

01:50:48.440 they have clinical recurrence. I think it's very likely that that kind of an approach where you

01:50:51.640 basically only give adjuvant if the test is positive, but you do the test longitudinally to

01:50:55.440 make sure that you don't minimize missing people, that that would very likely have equivalent outcomes

01:51:00.300 to giving everybody in that group treatment. Actually, maybe even better because you're not giving

01:51:03.520 toxicity to a big chunk of patients. So that's one. The second one, which I think is really

01:51:08.320 exciting, is analogous to that basically, which is to, it's the same set of trials,

01:51:13.340 but I think it's looking at it the other way around, which is to say that patients,

01:51:17.720 you would avoid overtreating patients. So one big problem we have right now in adjuvant therapy

01:51:21.640 in places like, many places, but breast cancer is a good example of it. The number you need to

01:51:25.620 treat is quite high because most patients don't already cure by the surgeon and don't have

01:51:29.980 micrometriotic disease. And then the treatments aren't perfect. So the ones even that have

01:51:33.480 micrometriotic disease, the significant number don't benefit. So actually, we have a huge problem

01:51:37.300 with overtreatment. And this approach that I outlined with the repeat testing could get us

01:51:41.060 away from that overtreatment situation where we only give the treatment to the patients where we

01:51:45.360 have substantial evidence that they still have cancer cells in the body. Those are three great

01:51:49.820 use cases for liquid biopsies that are all going to benefit from a very important insight, which is,

01:51:56.220 you know, the mutations of the cancer at the outset. So you get to pair a very important piece

01:52:02.380 of information, which is tumor identification genetically with cell-free DNA. We now need a

01:52:08.380 different approach if we're going to move to the sort of panacea of cancer screening, which is every

01:52:15.360 person once a year goes to their doctor, gives 10 or 20 ml of blood, and a test comes back that says,

01:52:22.660 one, do you have any cancer? Yes or no. Two, if you do, it's this organ. Let's go work it up,

01:52:28.520 which either means in the early stages, more diagnostics in the later stages means direct

01:52:33.020 to treatment. That would be, again, what we talked about earlier. It seems to me that

01:52:37.360 one way to think about that is if you were going to try to do that based on mutation analysis,

01:52:42.600 you would have to develop the world's most robust database of mutations for every given tumor.

01:52:49.060 Is that even possible? So we, about a year and a half ago, published a paper

01:52:55.120 where we developed actually a mutation-based lung cancer screening method using ctDNA. We call that

01:53:03.040 method lung clip for clip is cancer likelihood in plasma. And it's a method that is purely based

01:53:08.820 on mutations. The way it works actually is you sequence the plasma and the white blood cells of

01:53:14.860 the screening patient, the sampling on the patients you're trying to screen. You sequence both the CF

01:53:19.040 DNA, cell-free DNA, and the leukocyte DNA. The reason we do that is because we've learned that

01:53:23.540 in particularly older individuals, there are mutations present in the leukocytes often that

01:53:27.900 have been acquired through age, through a process called clonal hematopoiesis. So those mutations

01:53:32.480 that are in the white blood cells end up showing up in the plasma most of the time because the white

01:53:37.120 blood cells die in the circulation and release their DNA, and that's part of the cell-free DNA.

01:53:41.320 So those you want to get rid of, that's not coming from a lung cancer or breast cancer, right?

01:53:45.040 So we do both. We subtract the things that are in the white blood cells, and then we use a machine

01:53:49.980 learning algorithm that looks at the mutations that are left in the plasma and looks at things

01:53:55.220 like how long is the cell-free DNA molecule. We haven't talked about it, but we have, and others

01:53:59.700 have shown that the cancer-derived cell-free DNA molecules are a little bit shorter than that 170

01:54:04.620 basis. They're a few bases shorter on average. The reason we think is because most of the DNA in the

01:54:10.020 blood comes from white blood cells. We know that's true. About 80, 90% of the cell-free DNA is coming

01:54:14.700 from white blood cells, and that the other 10 or 20% comes from solid organs as well as cancer.

01:54:20.660 Probably the DNA has to traverse longer to get into the blood, right? If you could die in the

01:54:24.620 tissue and then to get into the blood takes a while, and that's probably a longer time.

01:54:28.920 To be exposed to the enzymes.

01:54:30.440 That's right. So we think that's part of the problem. It may also be that some of the way the

01:54:33.460 histones and the chromatin is configured may contribute to it. But for whatever reason, that is

01:54:37.580 true. We look at, is the mutation present? What gene is the mutation in? How long is the cell-free

01:54:42.460 DNA fragment? Is this mutation one of the mutations that is caused by smoking? All these

01:54:48.740 different things. Put that in the machine learning model. And the machine learning model isn't just

01:54:52.540 saying, have I seen this mutation before in lung cancer? But it's looking at these other properties

01:54:56.040 to say, do the properties of the mutations match what you'd expect in a lung cancer? And then the

01:55:01.140 model ultimately spits out a probability that that blood sample was from a patient with lung cancer

01:55:05.460 or not. So that is, I think, the way that I would envision doing it using mutation-based

01:55:12.060 approach where you're not actually doing this catalog of mutations because that won't work,

01:55:16.700 unfortunately, because every cancer is unique and you can have a mutation in any gene, in any position.

01:55:21.160 I assume you can use machine learning to basically predict what should be a cancer mutation versus

01:55:27.160 not. That's what we're doing. Exactly right. Yes. You try to use as much information as you can get

01:55:31.360 beyond just the mutation being there. In follow-up work, we are trying to extend that by adding more

01:55:35.980 such features, right? The more features you have that link with cancer, the better it is.

01:55:39.940 Again, the question is, can we combine that with methylation? Maybe the combination of the two,

01:55:44.100 you could kind of leverage the best of both worlds. Can we combine it with other analytes,

01:55:48.400 other approaches? So that's how I think we're going to move forward to try to develop ultimately

01:55:53.280 the best screening test. And it very likely will be a combination of not just methylation or just

01:55:58.220 mutation or just what's called fragmentomics, which is this size of the molecules or their

01:56:02.140 distribution. It likely will be a combo method. That's still very much in the early research phase.

01:56:07.160 One has to prove which of those things combine nicely and usefully. And then once you know that,

01:56:11.620 then you have to think about, okay, can I develop an assay that's affordable that one could imagine

01:56:15.580 doing on hundreds of millions of people? But we're still at the discovery phase.

01:56:20.440 Do you have a sense of what the theoretical limits look like? Assuming you can take information from each

01:56:28.680 of those data sets, right? So fragment size or features of the fragment length, methylation

01:56:34.860 patterns, predictive mutations, et cetera. And based on what you know about the frequency with which you

01:56:40.960 can identify cell-free DNA. And given that we're now exclusively talking about patients that do not

01:56:46.520 have disease on any other imaging modality, let's just limit it to pre-stage one or say at best stage

01:56:54.060 one, early stage one. Realistically, where do you think the sensitivity and specificity could be

01:56:59.920 pushed to? I think we don't really know yet. I think one critical experiment, which is what we're

01:57:05.580 currently doing, is to take the most sensitive method we have, this third gen method I mentioned

01:57:11.440 that has the one in a million sensitivity, apply that to a large cohort of stage one patients,

01:57:15.960 and see how many of those patients can we now detect circulating tumor DNA in when we know the

01:57:22.020 mutation. This is not the screening question, but it's sort of the preamble to it where we

01:57:25.800 can definitively say these patients have ctDNA and how much in stage one. So that experiment will tell

01:57:31.760 us what is actually the lay of the land for the amount of circunctumor DNA shedding in stage one

01:57:37.000 lung cancer. Once we have that, we will know what sensitivity do we have to get to with a naive

01:57:44.860 approach, meaning we don't know what the patient has cancer with this integrated approach. Before I can

01:57:49.880 say, I need to see that data, but I'm hopeful we can push it above the 5% we talked about earlier.

01:57:55.440 In our study from a year or two ago, within stage one, we could get to about 20% sensitivity,

01:58:01.700 including in stage 1a, non-small cell lung cancer in a validation cohort.

01:58:05.680 And what was the specificity?

01:58:07.240 The specificity was 99, 98%. I think it was tuned in that analysis, 98%.

01:58:11.520 Again, I know that's early stage. It's still just not going to make sense for people whose incoming

01:58:17.480 pre-test probability is 1% to 2%. It's just still too low.

01:58:21.440 It's too low.

01:58:22.140 My back of the envelope math, Max, is that if someone has a pre-test probability of 1%,

01:58:26.780 I mean, it sounds crazy. You're going to need 80% sensitivity and 99.5% specificity

01:58:32.660 to say, this is a test that matters. This is a test where, okay, now my positive predictive value

01:58:40.080 is high enough that I can do something with this. And just as importantly, my negative predictive

01:58:45.400 value means I can breathe easy. It's possible we can't get there with this approach. If you had

01:58:50.400 asked this question in the 50s with some of the protein biomarker work, which was exciting,

01:58:54.460 people would have said, well, okay, we're not there, but maybe in 10 years we will.

01:58:57.400 Do you think this is a decade away?

01:58:58.960 We don't even know what the real distribution of ctDNA is because most of the experiments have

01:59:03.700 used insensitive method where most of them are negative. Well, then we can't see what's negative,

01:59:07.700 right? So I'm hopeful we can increase it significantly. I'm hopeful we can get to at least 50%,

01:59:12.260 you know, 50 to 75%. Can we get to 80, 90%? 50 to 75, if it's truly early stage one,

01:59:19.000 that's a big step forward. Agreed. It would be a big step.

01:59:21.800 If you can keep your specificity north of 99%. Correct. So some liquid biopsy groups that have

01:59:27.800 kind of done that by saying, okay, let's go with 80% specificity because if you ever test 99%

01:59:32.780 specificity that has 20% sensitivity, now you drop the specificity to 80, your sensitivity magically

01:59:37.820 goes up by 20%, right? Because just the dumb luck part. And your negative predictive value goes out the

01:59:41.740 windows. So that's the problem, right? So for the foreseeable future, there's going to be sort of

01:59:46.380 two main areas, the way I think about applying these. One is in the cancers that have no screening

01:59:51.340 tests currently, the bar is, of course, much lower. We know that mammography, even low-dose ct,

01:59:57.940 we know colonoscopy is a little bit of an exception, but none of those are perfect, obviously. Yet we do

02:00:02.300 them, and at least for the case of lung cancer and the colon screening, we're very confident they save

02:00:06.360 lives. We think for breast cancer, although of course, as you know, there's controversy around that.

02:00:09.400 But in the cases where we don't have screening tests, if we can even get a test that's kind of

02:00:14.000 like that, not perfect, but can get us some, that could be helpful. It would be ideal in the case

02:00:18.080 like pancreatic cancer, but for reasons we talked about earlier, we don't know what to do with them,

02:00:21.440 and I don't think that's the right place to do it. But other cancers, you know, that we don't screen

02:00:24.200 for, there's that group, and there may be the bar is lower because we don't have a predicate.

02:00:29.280 Cancers that we already have screening tests for, I think it's really important how the tests compare to

02:00:33.200 the existing screening tests, obviously. If they're as good or better than what we're already doing,

02:00:37.940 well, then you could argue, well, then we should switch to those, right, even if they're not

02:00:40.640 perfect. But if they're worse, which is currently the situation, the liquid biopsies aren't as good

02:00:45.900 as those tests, then you really have to think hard about how you would use them. And then I think the

02:00:50.940 main use has to start to be something that's more around practical considerations or health system

02:00:55.140 considerations like low-dose CT, for example, we know tiny minority of patients who are eligible are

02:00:59.900 currently getting it in the US. Medicare pays for it, but the vast majority of patients who are eligible

02:01:03.580 don't get it. So why is that? Well, there's many reasons, but part of it could be, you know,

02:01:08.440 this concern of radiation exposure, the difficulty of having to book another appointment to go to an

02:01:12.900 imaging facility, access to the imaging facility. So if you're already a department care doctor,

02:01:16.920 you're getting your hemoglobin A1C tested or whatever, it's probably easier to draw another

02:01:20.720 couple of tubes of blood and send those off, right? So, but that's not the dream that's helpful,

02:01:25.680 but it doesn't solve the problem for us, I guess. So that's sort of how I would ration develop it,

02:01:30.300 but we still need to prove again, I think it's very important that these tests decrease cancer

02:01:35.440 specific death, even though they're exciting technologies and we're very hopeful that they

02:01:39.540 will, we shouldn't just give them a pass because they're high tech and not do the right clinical

02:01:45.420 science. Do you know of any companies out there that are actually in pursuit of what we're talking

02:01:50.540 about as the dream scenario where you have an actual high sensitivity, high specificity test for

02:01:57.740 low stage, early stage one? I think all the companies are working to try to improve their

02:02:03.600 tests constantly because they're being run in the CLIA environment. It's actually much easier to change

02:02:08.860 the test than it is under the FDA and approval. I think people are trying hard. I haven't seen it

02:02:14.580 solved. That includes my group. We're working on it hard. We have some ideas we think are promising,

02:02:18.960 but it's too early to know how much they're going to push the bar over what we showed in that prior

02:02:24.400 paper or what Grail showed in their recent papers. How much money is the NIH putting into this?

02:02:29.540 On a global scale, that's a great question. Actually, I've never seen a number of how much

02:02:33.300 funding are they doing for a liquid biopsy research. I can tell you that it's increased

02:02:38.480 dramatically. So just in general, when I started working in this area, like probably 2012-ish,

02:02:44.480 you know, with our first publication in 2014, when I was going to meetings around those times,

02:02:48.340 I would often be the only person talking about a circulating tumor DNA. There would be a bunch of

02:02:52.980 other diagnostics, imaging, and a lot of circulating tumor cell work at that time that was being

02:02:56.200 presented. Whereas now, the vast majority of presentations at meetings and stuff is focused

02:03:00.780 on liquid biopsy. The same thing I've seen at the NIH in study section, you know, reviewing grants,

02:03:05.800 many of us, including myself, serve on these NIH study section where one scores the grants that the

02:03:10.460 NIH, the scores that the NIH uses to fund research. And it's a lot of work because you've got to read

02:03:14.680 dozens of grants and, you know, you spend days of your week reading other people's grants and

02:03:18.460 commenting on them. But it's a service we do because, of course, otherwise the system doesn't work.

02:03:22.980 But now in the study section I sit on, which is focused on cancer biomarkers,

02:03:26.660 we see lots and lots of liquid biopsy work being submitted and often scoring high. So I think it's

02:03:32.160 substantial. It's increased dramatically from 2012. There's a lot of interest at NCI and NIH.

02:03:37.540 They see the value just like many of us do, that this is an area we absolutely have to push forward

02:03:43.160 and deserves more funding.

02:03:45.580 Max, this was super interesting. I mean, this is a topic I've been interested in for years,

02:03:49.220 and I learned a lot today. I really have a much better sense of the landscape and the history of

02:03:54.440 how it got where it got. So I'm positive that everybody listening to this will, even if they

02:03:58.260 came in with less information than me or more information, is going to have gleaned something.

02:04:01.580 Thanks so much for making the time and setting aside a few hours of your day to talk about this

02:04:05.700 and to catch up in general. It's just great.

02:04:07.400 Yeah, it was really a pleasure chatting with you. It's great to see you again.

02:04:10.300 Thanks for all the insightful questions. I really enjoyed the conversation. I hope

02:04:13.040 your listeners got something out of it.

02:04:14.180 I'm positive they will. Thanks, Max.

02:04:16.260 Thank you for listening to this week's episode of The Drive. If you're interested in diving deeper

02:04:20.400 into any topics we discuss, we've created a membership program that allows us to bring you

02:04:24.820 more in-depth, exclusive content without relying on paid ads. It's our goal to ensure members get

02:04:30.300 back much more than the price of the subscription. Now to that end, membership benefits include a bunch

02:04:35.820 of things. One, totally kick-ass comprehensive podcast show notes that detail every topic, paper,

02:04:41.520 person thing we discuss on each episode. The word on the street is nobody's show notes rival these

02:04:46.940 monthly AMA episodes or ask me anything episodes, hearing these episodes completely access to our

02:04:53.580 private podcast feed that allows you to hear everything without having to listen to spiels like 1.00

02:04:58.480 this. The Qualies, which are a super short podcast that we release every Tuesday through Friday,

02:05:04.360 highlighting the best questions, topics, and tactics discussed on previous episodes of The Drive.

02:05:08.620 This is a great way to catch up on previous episodes without having to go back and necessarily

02:05:13.460 listen to everyone. Steep discounts on products that I believe in, but for which I'm not getting

02:05:19.040 paid to endorse, and a whole bunch of other benefits that we continue to trickle in as time

02:05:23.940 goes on. If you want to learn more and access these member-only benefits, you can head over to

02:05:27.940 peteratiamd.com forward slash subscribe. You can find me on Twitter, Instagram, and Facebook,

02:05:34.700 all with the ID peteratiamd. You can also leave us a review on Apple Podcasts or whatever podcast

02:05:41.500 player you listen on. This podcast is for general informational purposes only and does not constitute

02:05:47.060 the practice of medicine, nursing, or other professional healthcare services, including

02:05:51.600 the giving of medical advice. No doctor-patient relationship is formed. The use of this information

02:05:57.460 and the materials linked to this podcast is at the user's own risk. The content on this podcast

02:06:03.360 is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Users should

02:06:10.240 not disregard or delay in obtaining medical advice from any medical condition they have, and they should

02:06:16.660 seek the assistance of their healthcare professionals for any such conditions. Finally, I take conflicts of

02:06:22.820 interest very seriously. For all of my disclosures and the companies I invest in or advise, please visit

02:06:29.120 peteratiamd.com forward slash about where I keep an up-to-date and active list of such companies.

02:06:52.820 Thank you.

The Peter Attia Drive - July 11, 2022

#213 ‒ Liquid biopsies and cancer detection ｜ Max Diehn, M.D. Ph.D.

Episode Stats

Summary

Transcript