An interlaboratory external quality assessment across 16 laboratories in the Dutch COIN consortium evaluated how diverse (pre)analytical workflows and analytical platforms affect detection and genotyping of ctDNA mutations in plasma.
0:10They night lab lights, a quiet hum in the air. A vile of plasma, hope hanging on prayer. Welcome to Base by Base, the paper cast that brings genomics to you wherever you are. Thanks for listening, and don't forget to follow and rate us in your podcast app.
0:28Hey, everyone. So I want you to imagine you are sitting in a doctor's office. It's a high stakes moment right. You've just had a standard blood draw. And that little vial of your blood is being boxed up and shipped off to a laboratory to hunt for circulating cancer DNA.
0:43Right, a liquid biopsy. Exactly. And this test, it could dictate your entire treatment plan. So here is the $1000000 question. Do you get the exact same result no matter which lab that vial goes to? I mean, it seems like a baseline assumption for modern health care, doesn't it?
1:00You would naturally expect uniform precision. You would assume it's a simple yes. But the reality is completely shocking. Depending on the specific preparation and testing method the laboratory happens to use, that exact same blood sample could yield completely different results.
1:15Which means entirely different treatment options for you. Yeah. Think about liquid biopsies. kind of like a baking challenge. If you give 16 different MasterChefs, the exact same ingredients, you know, flour, sugar, eggs, but they all use different mixing bowls, different oven temperatures and different methods for tasting the final product.
1:34You were gonna get 16 completely different cakes. That's a great way to put it. So how could this level of variation change the way we treat cancer today? Well, it is arguably one of the most pressing yet hidden vulnerabilities in modern oncology diagnostics.
1:49I mean, we have built an incredible medical infrastructure, but the plumbing connecting the pieces is surprisingly unstandardized. Today, we celebrate the work of the Dutch coin consortium and the research team, published in clinical chemistry in 2024, who have advanced our understanding of circulating tumor DNA testing.
2:07Yeah, their research really pulls back the curtain on the actual mechanics of diagnostic testing. It reveals just how much laboratory workflows influence the ultimate biological truth reported back to a clinician.
2:19Okay, let's unpack this. We're looking at liquid biopsies. Specifically, the detection of circulating tumor DNA, or CTDNA, from blood plasma. Right. And for an audience that closely follows genomics, We already know the incredible promise here, right?
2:35Dracking minimal residual disease, identifying resistance mechanisms. Or screening for early stage malignancies without needing a scalpel or a tissue biopsy. Exactly. The technology is phenomenal, but widespread clinical application seems to be hitting a bottleneck.
2:51Wait, isn't medicine standardized? Shouldn't every highly advanced medical laboratory be using the exact same protocol to find cancer DNA? You'd think so? I mean, if I get a standard metabolic panel or cholesterol test, the numbers mean the same thing everywhere.
3:06If we connect this to the bigger picture, we have to recognize that detecting microscopic fragments of mutant DNA in a C of normal DNA is vastly more complex than measuring a stable molecule like cholesterol.
3:17So it's an apples and oranges thing. Completely. Biology is messy, and more importantly, the technological evolution in genomics is moving much faster than our regulatory frameworks. Because the tech is constantly iterating, there is a severe lack of standardized plug and play protocols.
3:33Yeah, very few of these plasma-based CTDNA detection assays have formal FDA approval or European CE marking for broad clinical use. I'm trying to visualize what that means on the ground. If there isn't a universally approved machine that laboratories can just buy and turn on, how are they actually performing these tests for patients or are they essentially MacGyver in these critical diagnostic machines?
3:55In a sense, yes. The industry term is laboratory developed tests or LDTs. There are assays that are built in-house. Built in house. Wow. Right. Most of the chemical regions and sequencing machines are currently categorized by manufacturers as research use only.
4:09So a highly skilled molecular pathology laboratory uses their own expertise to build custom workflows. So they're putting the pieces together themselves. Exactly. They pieced together different DNA extraction kits.
4:21They optimize their own chemical washes, and they calibrate their own sequencing machines to get the job done. Which introduces an enormous amount of human choice into a process we assume is rigidly automated.
4:33really does. And this leads us directly to the core of this deep dive. The researchers recognize this variability. and essentially design the ultimate scientific sting operation to measure the fallout.
4:44They wanted to see what happens when you stress test this decentralized system. They engineered a comprehensive external quality assessment, or EQA. They needed to bypass theoretical debates and just look at hard data.
4:56So they sent identical blinded samples out to 16 different Dutch laboratories that routinely perform these liquid biopsies. But finding artificial DNA in a pristine test tube isn't the same as finding it in the chaotic environment of human plasma.
5:11How did they ensure the samples actually mimicked the real world challenge? That is where the study design shines. They sent two distinct categories of samples. First, they used three commercial reference plasma samples.
5:24Think of these as highly controlled laboratory products. The manufacturers take clean plasma and spike in precise amounts of mutated DNA. They engineered variant allele frequencies, or VFs, ranging from exactly 0% as a negative control up to just one percent.
5:41Let me stop and make sure we are clear on varying illegal frequency. If we are talking about a one% VAF. We mean that out of all the 1000s of DNA fragments floating in that specific sample, only one in a 100 carries the cancer mutation.
5:56Exactly. The other 99 are perfectly healthy normal DNA fragments from the patient's regular cells. That is the exact mathematical challenge. It requires an incredibly sensitive assay to spot that single meet and fragment without being blinded by the 99 healthy ones.
6:11But as you pointed out, commercial spike samples are a bit too clean, they don't have the complex proteins and fragmented structures found in a real patient. So the researchers introduced the 2nd category, which were real patient derived diagnostic leucophresis plasma samples.
6:26Diagnostic leucophoresis. Let's call them DLA samples to keep it simple. These came directly from patients diagnosed with advanced non-small cell lung cancer. Yeah. And the reason they use leucophresis is purely logistical, but brilliant.
6:40How so? Leucophoresis involves passing a patient's blood through a specialized centrifuge machine that skims off a large volume of plasma and mononuclear cells, and then returns the red blood cells to the body.
6:53Oh, wow. So they get a lot of material. Yes. This allowed the researchers to safely harvest an enormous volume of cancerous plasma from a single patient. They needed that massive volume so they could divide it into identical standardized aliquets to ship to all 16 labs.
7:08So every lab opens a package containing the exact same biological puzzle. Their mission was to find mutations in 3 specific gene regions heavily implicated in lung cancer and other malignancies. That's BRAF, Xon 15, EGFR, Xons, 18 through 21, and KRAS, Xons, 2 and 3.
7:27Just to translate the molecular geography for a second. When we say Exxon, we are basically talking about a specific chapter in the DNA instruction manual. That's perfect analogy. And the laboratories are looking for spelling errors within those specific chapters.
7:39Exactly. They are hunting for single letter typos, missing sentences, or duplicated paragraphs within those crucial chapters. Now, remember the baking analogy. Here is where we enter the Wild West of laboratory workflows.
7:54The researchers tracked exactly how each lab attempted to isolate and read those chapters. Out of 16 labs, they utilized eight completely different DNA extraction methods. And the extraction phase is critical.
8:08You are attempting to separate fragile microscopic DNA fragments from a thick soup of proteins, lipids, and cellular debris. It is harsh chemistry. And it gets more variable from there. They didn't even start with the same amount of plasma.
8:21No, they didn't. The input volume going into the machines vary between 2 and 4 milliliters. But the part that really caught my attention was the Aleutian volume. Ah, yeah. At the very end of the extraction process, you have to wash the purified DNA off a filter using a liquid buffer.
8:35The labs use dilution volumes ranging wildly from 25 microliters all the way to 200 microliters. A huge spread. I'm trying to process the chemistry here. If I have the same tiny amount of purified DNA, but I wash it off the filter with 200 microliters instead of 25, I'm radically diluting the final product.
8:54It's an eight fold difference in concentration. Why would a lab willingly dilute a sample when they are hunting for a microscopic one% signal? It is a calculated trade-off. If a laboratory washes the DNA into a larger volume, say 200 microliters.
9:11They have more physical liquid to work with. This allows them to run multiple different tests or keep them in the freezer as a backup. Oh, I see. But the penalty is severe dilution. If you wash it into just 25 microliters, the DNA is highly concentrated, maximizing the chance that your sequencing machine will actually suck up the mutated fragment, but you might only have enough liquid to run a single test.
9:33There is no universal standard for which trade-off is correct. And that leads directly to the analytical tools used to actually read the DNA. The labs deployed 10 totally different methodologies. 10 of them.
9:44Yeah. Some use single target droplet digital PCR or DDPCR. Some use small panel PCRs like the Coba system, and others use massive, broad next generation sequencing or NGS panels. You can visualize these 3 analytical methods, like different types of search parties, droplet digital PCR is like sending out a highly trained bloodhound given one exact cent.
10:07It is incredibly sensitive and fast, but it only finds the specific mutation you program it to look for. What about the small panel PCRs? That operates more like a standard police lineup? It can interrogate the sample for a few dozen common suspects very efficiently.
10:21Okay, and NGS. Finally, next generation sequencing is like casting a massive dragnet across the entire city. It sequences 1000000s of DNA fragment simultaneously to find anything out of the ordinary. It gives you a massive amount of data, but because it is looking at so much noise, the analysis becomes highly complex.
10:37What's fascinating here is how these varying workflows actually performed under pressure when looking at the exact same blood. On the surface, if you just glance at the top line data, the system appears functional.
10:4913 of the 16 labs, that is, 81%, hit a 100% overall detection rate for finding something abnormal in the samples. Here's where it gets really interesting. Finding something abnormal isn't the same as knowing exactly what you found.
11:04The real problem the study exposed wasn't finding the mutations. It was accurately identifying them. Exactly. The detailed genotyping, the exact spelling of the mutation failed across the board for several critical, actionable targets.
11:18I want to break down these statistics because we need to translate these numbers into human stakes. Let's look at a specific genetic typo called an EGFR deletion. Specifically, the S 752 to I 759 deletion.
11:31Going back to our book analogy, this is where a specific sentence has been ripped out of the chapter. That exact deletion was completely missed or misidentified by 69% of the participating laboratories.
11:43It's really concerning. Think about that for a second. Imagine 10 patients walking into a clinic with this specific lung cancer mutation. Seven of them will be sent home with a generic chemotherapy plan because their lab simply couldn't spot the deletion while the other 3 get a life-saving targeted pill.
12:00That is the reality of a 69% failure rate. The variation is staggering, and the pattern held true for other complex mutations. An EGFR duplication where a sentence is accidentally copied twice was missed by 50% of the labs.
12:15I'm looking at the KRAS mutation data, and I'm a bit lost. I know KRES is a massive driver in cancer, but the specific call for the KRES G12C mutation was only reported 33 out of a possible 64 times across the samples, meaning it was missed 48% of the time.
12:33Right. We were talking about a single letter variation this C at the end. Why does missing one specific letter caused such a catastrophic clinical failure? Biologically what changes? This raises an important question, and it is the crux of modern personalized medicine.
12:46Why does the exact spelling matter so much? The targeted therapies we use today are not broad spectrum poisons, like traditional chemotherapy. They are engineered molecules designed like highly specific keys for very specific molecular locks on the surface of the cancer cell.
13:01So how does that lock and key physics applying to that C in the KRES mutation? Well, if a lab's workflow isn't robust, and they just report a generic KRSG 12 mutation without specifying that it is exactly a G 12C mutation, the clinical interpretation halts.
13:17The patient likely receives nonspecific chemotherapy. But if the lab specifically identifies the G12C variant, it proves that the cancer cell has a unique physical pocket on its surface. That specific pocket unlocks access to incredibly effective targeted therapies, drugs like soda sib or ad aggressives.
13:35And those are specific to that pocket. Exactly. These pills are chemically engineered to bind exclusively to the unique shape created by that C mutation, shutting down the cancer cell's growth signal. If the lab misses the C, the doctor doesn't prescribe the drug.
13:50I want to apply that lock and key logic to the EGFR gene failures we discussed earlier. The study showed half the labs missed a specific EGFR Exxon 20 insertion. If an insertion physically changes the shape of the receptor, Is that like someone changing the locks on a door?
14:06Yes. The old key, a standard drug like Austin Martinium, just won't fit into the keyhole anymore. That is a brilliant way to conceptualize it. Common EGFR mutations respond beautifully to Asamertinub, but an Exon 20 insertion changes the lock.
14:20Asamertina becomes completely useless. What do they do? Those patients require a radically different approach, like a targeted antibody therapy called amivantomab, which binds to the outside of the cell differently.
14:32If your lab's workflow dilutes the DNA too much or the sequencing machine can't differentiate between a standard deletion and a complex insertion, you will be prescribed the wrong drug. I want to talk about how these machines are failing to read the spelling.
14:44We saw one lab in particular, Lab 4, which used a broad next generation sequencing approach performed terribly due to massive, false negative, and nonspecific calls. Right, Lab 4. I always assumed NGS was the gold standard.
14:58You know, the ultimate dragnet. How does a massive multimillion dollar NGS sequencer fail to find a mutation that a simpler PCR test picks up? It comes down to bioinformatics. An NGS machine doesn't spit out a clean, easily readable sentence.
15:14It generates a massive data dump of 1000000s of fragmented overlapping puzzle pieces. The laboratory has to use complex bioinformatic software pipelines to computationally stitch those tiny reeds together against a reference genome.
15:27If the software's algorithm is calibrated to be too strict, perhaps, to avoid false alarms, it might throw away real cancer puzzle pieces resulting in a false negative. And if it's too loose? If the algorithm is too loose, it forces the wrong pieces together, mistaking background sequencing errors for cancer, which is a false positive.
15:44Lab 4's bioinformatic pipeline was likely miscalibrated for the specific types of complex insertions and deletions present in the samples. This is the ultimate verdict of this deep dive. Divergent pre-analytical choices, like Aleutian volume combined with differing analytical bioinformatic algorithms, literally lead to discrepant clinical outcomes using the exact same blood.
16:08But looking closely at the paper. There's a massive real world caveat we have to address. The DLA samples used in this sting operation were from advanced stage patients. Because they came from advanced disease, those plasma samples had incredibly high tumor DNA fractions.
16:24The variant allele frequencies were ranging from one.6% all the way up to 9.3%. So they were playing on easy mode. Essentially, yes. If they failed this badly with a billboard size signal of 9%, what happens when the signal is just a whisper?
16:39That is the chilling implication? For early stage cancer screening or monitoring minimal residual disease after a surgical tumor removal, the fraction of cancer DNA might be 0.one% or lower. The signal to noise ratio is exponentially worse.
16:52And at those microscopic levels, you run headfirst into a phenomenon known as biological noise. Explain that biological noise. What is interfering with the signal? The primary culprit is something called clonal hematopoasis of indeterminate potential, or CHIP.
17:07As we age, the stem cells in our bone marrow naturally acquire random benign mutations. They aren't cancer, but they are genetically abnormal. These stem cells pump out 1000000s of white blood cells carrying these mutations, and as those white blood cells naturally die, they shed their mutated DNA into the blood plasma.
17:26So your own aging immune system is basically throwing off decoy DNA that looks identical to tumor DNA. Precisely. If your lab protocol isn't incredibly standardized with specialized bioinformatic filters, a highly sensitive NGS machine will detect a CHIP mutation in the plasma and flag it as a cancer recurrence.
17:43Which is terrifying. It is. That is a false positive that could lead to devastating unnecessary chemotherapy. Conversely, the sheer volume of CHIP background noise can drown out the real low level cancer signal, giving a false negative.
17:56The fact that the Koyan consortium showed labs struggling to accurately genotype these samples when the signal was loud and clear, strongly suggest that real-world clinical errors on low signal samples are likely significantly higher.
18:07It really paints a picture of a diagnostic landscape that is incredibly powerful, capable of mapping the molecular drivers of disease from a simple blood draw, but one that is currently resting on a very fragile, unstandardized foundation.
18:22We have the chemistry, but the harmonization is lagging severely behind. The fundamental science is sound. But until we treat the entire workflow from the volume of the blood draw to the extraction chemistry, to the exact parameters of the bioinformatic software as a single standardized medical device, we will continue to see these discrepancies.
18:42So what does this all mean? Liquid biopsies possess the power to fundamentally revolutionize cancer care. But this deep dive proves that the technology remains highly vulnerable to everyday workflow variations.
18:53Until the global medical community forces standardization across every single step, from the physical blood draw to the digital algorithmic analysis, patients risk receiving wildly different treatment plans based solely on which laboratory happens to process their sample.
19:08What does this mean for the future of global oncology standardization, and how can we trust our diagnostics in an era of personalized medicine? This episode was based on an open access article under the CCBY 4.0 license.
19:21You can find a direct link to the paper and the license in our episode description. If you enjoyed this, follow or subscribe in your podcast app and leave a five-star rating. If you'd like to support our work, use the donation link in the description.
19:34Now stay with us for an original track created especially for this episode and inspired by the article you've just heard about. Thanks for listening and join us next time as we explore more science, base by base.
19:57A quiet hum in the air. A vile plasma, hope hanging on prayer. We chase a signal in a river running thin. If the steps don't match, we can't read what's with it. Measure by measure. The smallest trace.
20:19One miss turn, and it fades from the page. We need one lage For what we see. So every result sets the right heart free. Same the same truth. Let it ring Let it rise Light it up, Lock it Make it clear for the eye.
20:43Don't let me answer slip through the cracks We can't see. When the word flow sings in harmony. Same love, same truth. Set the rhythm, set the proof. hey Yeah, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, Same truth, the rhythm, the proof.
21:03Different hands, different kids, different volumes in play. Heel goes up, or it vanishes drifting away. Some can't spot the shadow. I can't name the face. Action of a word lost in a blurry trace. So call and respond every bitch, every run from extraction to read out.
21:28We move as one broader pedal, deeper light in the code. So the right mutation. Let it ring. Standard step, steady hearts, under bright screen sky. No more misty. What matters when the states are high? Catch the signal, we'll verify.
21:58Same, but same truth. Hallelujah to the proof, hey Oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh
