So hi, hello. My name is Sarabeth Schommer, you just got a brief introduction. I’m going to be talking about automating cfDNA extraction for the aneuploidy screening. Brief overview of what we’re going to see today, I’m going to give an introduction on myself and on my company, BillionToOne, what we do.

And then we’re going to go into background, which is going to give motivation and goals for this project. I’m basically going to walk you through the development of, from the engineering perspective, the development. And then we’re going to look at performance and we’re going to conclude with future improvements.

Okay. I know you just got some of this, but I’m going to go over it really quickly.

As James mentioned, I’ve been in the automation industry for a while. I have worked with liquid handlers for almost eight years. But I got my degree in biomedical engineering and then I went to join Hamilton as an FAS where I got exposure to broad science applications and my first exposure to liquid handlers.

I then went to Mammoth where I got a chance to like apply that knowledge to product development as well as expose myself to the Agilent Bravo, which is a different liquid handler. That’s where I started my master’s in biotechnology and management with the intent to kind of supplement my industry experience as well as like deepen my understanding of biotech as an industry.

And now I work at BillionToOne and they have also a lot of liquid handlers, including the Lynx.

So quick overview of our company.

We are a screening company. We have two divisions. As James mentioned, I’m part of the prenatal one. I’ll touch on the oncology one briefly, but I don’t work in that department. We have two products though, both oncology-based liquid biopsies, both Northstar Select and Response are the product names. But then moving quickly back to prenatal UNITY Complete is our flagship product, which is a pregnant individual screening. So basically, we get maternal carrier status as well as fetal risk for the pregnancy. So as little, I think as early as nine weeks now, we can actually take this and it’s just a blood tube from a pregnant individual.

Great. So, the project we’re talking about today is under the aneuploidy, which is determining if there’s an abnormal number of chromosomes present. And so, this is a unique workflow in our CLIA-certified lab.

To quickly walk you through this workflow:

Basically, we received the blood tube from the clinics. We go through the accessioning process, which is ensuring that the blood tube is associated with the patient, identification information, as well as ensuring everything is lined up correctly. The sample looks okay. And then it gets accessioned and then it gets isolated into its components. So we take the plasma separately, 3.5 to 4 mils, and then the Buffy coat. For the cfDNA extraction, we do take the plasma. So it’s a high-volume input, 3.5 to 4 mils, as I mentioned. So coming in in a 24 format, and then we end in a 96 format with a low-volume elution, which then goes into our plastic kind of PCR and sequencing prep process. We’re not going to get into those today, but the sequencing prep is kind of classic bar coding, cooling, purification.

But yes, as we mentioned, cfDNA extraction is what we’re going to talk about. Most of these steps, I think all of them are actually already automated in a liquid handler, including cfDNA extraction. It’s currently done on a different platform in our lab. So as we’re looking to grow and expand, we get more tests all the time. We have record breaking days. We’re trying to consider other options. So basically as we scale, we’re looking for higher throughput, more integration friendly, more customizable, as well as always thinking about COGS, as well as more robust systems. So looking to reduce error rate, things that are easier to work on hardware wise. And then of course, it’s always nice to look at options that potentially could increase yield, because cfDNA extraction is notoriously low in concentration, because the particles are smaller or the DNA is smaller.

All right, jumping into the development process. So we’re first going to talk about configuration, jump into the specialty consumables and the work that went into that, and then end with the method development. Configuration is arguably the most important step in selecting a liquid handler and going forward. So we decided to go with the LM1800. So as I mentioned, this is a high-volume extraction. So input plates coming in require a lot of lab wear and specialized handling. So the main reason we selected this is based on the MCPA tool being one of the dual arms. Oh, let me back up a second. We chose the 1800. It has 66 positions in total. It also has the ability to have multiple arms, which is great, because we chose one to be the MCPA tool, which allows us to, on-the-fly, interchange heads.

So the 24 5 mL head is really valuable for handling those large-volume plasma samples, as you can see in the image below. Just makes for like a quick transfer really efficient, as well as like for the larger volume transfers and bead binding. Very helpful. The 24 1 mL head is great for when we’re doing that re-arraying. I know that was discussed as well. I think going the other direction, 96 to 24, we’re going 24 to 96. So having that stamp for the lower volume transfers is really valuable.

And then as the 24 magnetic rod head is also really appealing for that lower yield attempt, we’re going to talk more about that in a second. But yeah, the MCPA is really valuable. The other arm is a VVP head, which is kind of their iconic product. You saw really good examples of that in the previous presentation. And we handle all of our 96-plate handling with the VVP head. For all of our incubations, heating, cooling, and shaking, we use a Q1 BioShake. You can see in the back in that image as well.

It’s been working well so far. And the final part of the configuration won’t make sense right this moment, but in the next slide it well, it’s a tip comb change station. We have two options for that, which is a customized iMagZ. If you’re familiar with the Dynamic Devices product line, the iMagZ is their pneumatic elevator. But we’ll talk about that in a second. We’ll see a video and then a static raised platform.

So, jumping into specialty consumables. In order to use the magnetic rod head that was mentioned, we need to use a tip comb and plate combination that is unique. So essentially this combination. So if we just ignore the tip comb for a second, look at the plate. The plate has a very unique well geometry that is not off the shelf, especially for Q1 to have a specified adapter for it. So our great service engineering team, shout out Salma, went through and designed a specialty adapter for the Q1. We went through a couple versions to make sure that it was good for movement tests, fit, all of that. And then we got a part machined, went through and did thermal profile testing to ensure that we were reaching the correct temperatures and the correct amount of time. And there was even spread across the plate.

And once we got all that sorted out, we looked at that tip comb. So the tip comb is required for us to do magnetic rod extraction. So if you’ve seen a plate-based extraction for DNA, it’s the plate or there’s a magnet, there’s a plate, and then there’s reagents added to beads in the plate that are then holding your DNA. This is basically that flipped upside down. So you have the magnetic rods going in and out of the plate with a consumable, the magnetic tip comb placed on the outside of the magnet in which you then take the beads with bound to the DNA into the different reagents. So that’s your washes, your binding solution, yada, yada.

This is just a quick visualization to say that these are necessary per sample to have a unique tip comb because there was initially when we started this project, not a way to dynamically change tip combs. So essentially you would load one tip comb with one plate and it would process all the way through.

And that was your run. I think when it was initially designed, it was designed for the smaller platform. So we’re looking at LM1800. I need four tip combs that are clean and to go into four plates and for that to change on the fly.

Dynamic Devices was great, worked with us a lot on versions for this. So the elevator or slip load are mentioned here. We’re going to see some videos if it works correctly. So we have V1, which is just put four of them on the deck and load and unload them. That’s on the left there.

Yeah, not ideal. It takes up a lot of space. If you look at the instrumentation on the right and the left, you lose all the spaces. You lose twelve spaces to four positions. Not ideal. It does work. It is, oh, hold on. Wait, go back.

All right, all right. We’re okay. One, no. Go back again. Okay. Not ideal, not great. I don’t know if you can get that from a previous one. Can I show the one in the middle? Thank you. All right.

And this is live in our production lab right now. And this is working pretty well. So as I mentioned, it’s the iMagZ. You see that tip comb is right now elevated and being picked up and it’ll be tossed in the waste. So basically, the iMagZ has an elevator tray that pops up and down. So when it pops up, it is touching the dirty tip comb. And then it retracts down. You can then load a new tip comb and perform this assembly and then load it. So this was a lot better for loading to save spaces. So we only have to use the four spaces for the loaded tip combs, which is totally reasonable. It’s relatively robust. We’ve seen some issues where if we’re trying to do a recovery, it’s difficult because all of these are like manual movements, not like, oh, pick up tip comb, put it here, pick up this piece that holds down the top. The tip comb doesn’t come up.

Yeah, those steps don’t exist. So you’re just doing actual movements every single time. This is totally functional though. Great. Well, in our production lab right now. This, oh, we don’t need both. Oh my gosh. Can I do it? I don’t know if I can. Oh God, I’m so sorry everybody. My new videos were a bad idea. Okay. Okay. This one, can I just tell you in theory what it’s doing? Okay.

All right; we’re just going to watch this one. So looking at this one, we are picking up the tip comb, moving to a position in the back, just like the middle one, but it is a static position. Please don’t mind the video. This is relatively new. It is currently going through our validation process in the lab, but essentially you put it in the static position, the head comes down, it slides over, the magnet rods come down to hold the tip comb in place, and then it goes and it performs the process.

So much simpler, no assembly required, air recovery there is much better.

And so far it’s been very consistent. So, very promising. We’re all very excited, but it is not validated in our lab yet.

Final step with development, sort of, once you work out all the kinks, have your configuration all in place, you build the method. So we did this in-house, largely done by our automation engineering team, including myself and Beverly.

So, some highlights here. Really like how on the LM1800 you only lose two tracks on each side for each head. It’s a lot of accessibility, a lot of customization there. I also really appreciate that you can run any number of methods concurrently. Your BioShake is doing something, your left arm is doing something, your right arm is doing something, and you’re handling a file somewhere else. I think that is entirely powerful and great.

I also really appreciate the minimal traverse height, specifically for the beads. So when we’re transferring beads, we don’t have to fully retract up to a certain height. We can go right to the top of the plate and move over so we’re not carrying those beads all around the deck on that tip comb. Really valuable stuff.

Limitations here. We are a clinical lab, so from our perspective, we don’t really want anything to be dynamic as far as the user interface goes. We want a really, we want a big green button that says go.

And there’s not a big green button that says go. For all those customizable great engineering reasons that I like, our CLIA lab does not appreciate. They, however, came up with a solution that worked for them, which is coming up with these static images that they like and can use for their references, which is working very well in our lab at the moment.

These last two are us building sub methods to do things that the arms are not able to do. So basically, the MCPA with all those four tools initially wasn’t designed, I think, for on-the-fly dynamic exchange, so it does not know when it has the wrong tool on to perform the job it needs. So we basically, before any loop or any big sub method, we double check what head we have on, if it’s the wrong one, we unload it, reload it.

The arms also don’t know where each other are, so we have to make sure that in order for them not to run into each other, we do a quick check of where they are in space and then move forward.

I’m now going to touch on performance. So, we have the configuration placed. We’ve solved all the problems. We’ve built the method out, the method’s running smoothly, consistently.

We’re going to look at assay performance. So initial assay performance was not great. Shocker. First run throughs, not always good. Unfortunately, we are competing with an instrument that is validated in our lab and has had process improvements made to it over the last few years. So a little bit of an unfair, but it makes sense when you’re making a switch, you have to make sure you’re performing at least as well.

So basically, what we did was identified the four places we were, the four most crucial locations where we were trying to make improvements. Those are the ones in parentheses that are the differences from the validated process. So, there was a reagent addition order change, some bead binding, or the bead binding with the magnetic rods is a relatively significant change, should have low impact. Now to all temperature events were occurring on the Q1 versus a combination of pelts and incubators. And the elution is now occurring at an ambient temperature versus on a Peltier device.

So, we basically ran each of these steps on each of the instruments to narrow it down to find that the elution was the problem. Shocking.

But yeah, so once we identified that we were able to, without adding the Peltier, we explored options of just like longer incubation times, more mixing, let’s just be more aggressive. And it was able to work, only adding about five minutes of runtime actually, and we were able to see: exactly the same. So these are paired samples run through. This data was really nice to see when we saw it. It is, of course, repeatable as well. This was just the first experiment. So this is the instrument versus the Lynx. So making that shift was now entirely equivalent as far as genomic material on our end. So just equivalency isn’t great, but we have all these lab benefits. So as I mentioned, it passed the clinical, or as I mentioned, the performance was repeatable. So it passed validation.

And the CLIA lab operators highly preferred it. We had 100% preference. So we would pull them often. This process went on for a while. Often and the only thing they ever noted was the GUI, which is totally something we’re working on with Dynamic Devices, a fixable problem. But we get all these savings. So with the instrument runtime and operator time savings, we actually can fit another run in the day, which is really valuable for our operations. And then we save about 50 cents per sample, almost 10. 50 cents per sample.

So, with all these savings plus the equivalent performance and the opportunity to improve performance over time, this was the obvious choice.

There’s also a lot of customizability. A lot of our service engineering team is actually here. But when you’re looking at a CLIA lab, you also have to think about do the lab operators like it, and is it serviceable? And so far, they’ve taken to it really quickly, but there was a learning curve. It is very open, I guess, when you’re trying to recover. You can go back to any point in the method, which is really great. You just have to make sure that you’re tracking everything along the way.

But all good things from lab performance.

Concluding with some future improvements, we’d actually like this to process even more samples. But since we use the magnetic rod technology, it requires consumables to be on the deck per reagent. So they’re pre-filled.

Lots of consumables, lots of physical space that we need. So we’re looking to integrate robotic arm and external storage so that we can process more batches in parallel and overlapped.

I didn’t touch on it here, but there was some tip-static events. But they’ve already put mitigations in place, we just have to go through and do some testing to make sure that it’s completely eliminated the issue. I mentioned the CLIA perspective. User interface is not super intuitive to them. And then really exploring this idea of cleaner DNA with that magnetic rod technology moving out of the reagents versus into the reagents. So, more time spent on that. And with that, I would like to thank these people. Scott is great from Dynamic Devices. Victoria for all of her lovely data analysis. Beverly, Salma, and Henry.