Hi, everyone, I'm Dr. Brian Kruger. I'm an associate vice president and research and development at LabCorp and my team has been leading SARS-CoV-2 molecular testing development at LabCorp since the beginning of the pandemic. Today, I'm going to talk a little bit about the work that we have been doing to perform National pandemic surveillance using whole genome sequencing and I want to start out by highlighting LabCorp's reach and as I'm sure many of, you know, based on how often you see our courier vehicles on the streets, LabCorp can cast a very wide net to capture specimens from all across the country. From a diagnostic testing perspective, this means that we have a very broad, diverse reach into all areas of the country. And from a pandemic surveillance perspective, this sort of reach puts us in a really unique position where we are one of the few places that has immediate access to the samples required for tracking this virus and its variants throughout the country. For SARS-CoV-2, we have this access because of the very hard work that the team did at the beginning of the pandemic. We are one of the few companies that recognize the potential need for a high throughput SARS-CoV-2 assay in January of 2020 and the Molecular R&D Group began immediate development on a PCR assay to support detection and diagnosis. We began this development journey toward the end of January. We were the first commercial diagnostic lab to launch an assay on March 5th. We were the first commercial lab to get FDA authorization for testing. And those milestones were followed by many other firsts, including being the first to receive home collection approval and the first to receive direct to consumer testing approval. What this means is that because of our success with rolling out and scaling our own internally developed PCR based diagnostic, we had, and we still have, unparalleled access to samples that can be used in real time to track the spread of the virus and the emergence of new variants. The LabCorp COVID-19 RT-PCR test was developed as a very high throughput implementation of this original CDC assay, which consisted of 4 reactions with FAM labeled probes, 3 reactions targeted 3 different regions of the nucleocapsid sequence, and the 4th targeted human RNase P as the internal control. The original CDC implementation used a 16 sample per hour QIAGEN biorobot for extraction and they did detection in a 96 well plate on the QuantStudio 7500DX. So, it was a 16 sample extraction into a 24 sample assay, which isn't ideal on the best day, but it also wasn't a protocol that could be scaled to address a pandemic. So, to combat those obvious throughput issues with original CDC assay, we took their primer and probe sequences and optimize them for detection on the QuantStudio 7 in 384 well format. And we also validated 3 different high throughput extraction methods that could each produce 100 or more extractions per hour. In the original implementation of the high throughput 4 reaction assay, we could process 96 samples on a 384 well plate every hour. But it was immediately obvious to us that the assay needed to be multiplexed. So, while we ordered the 4 reaction FAM labeled probes, we also ordered oligos with floors that were compatible with multiplexing. For this, we kept the N1 target as FAM, but switched the N2 target to Yakima Yellow and the RP target to Cy5. And after a bit of optimization, we validated the multiplexed assay and submitted it for authorization, which we received from the FDA on April 14th, about a month after we received the initial authorization for the lower throughput 4 reaction version of the assay. All of the assays were automated on Hamilton Star liquid handlers and conversion to the multiplex quadrupled the hourly throughput of that instrument from 288 samples per hour to 1,152 samples per hour per instrument. More than half of LabCorp's PCR testing is done on the FDA authorized multiplex assay. And that's really important for this story, because without the bespoke PCR testing process, we would not have been able to quickly and efficiently scale the sequencing process. To date, more than 30 million PCR tests have been performed across all methods. So, even if just over half of those tests are on the multiplex PCR assay, we have access to really an unprecedented number of samples for performing surveillance activities. And this story wouldn't be complete without talking about the supply chain constraints that were needed to, that we needed to overcome. The molecular R&D group validated no less than 3 consumables or methods for every step of this process. This includes plastic plates, tips, enzymes, extraction reagents, collection devices, swabs. If you can think of any consumable that goes into any test that is performed, we have a backup validated for the backups backup. And this was essential for our success in PCR testing. Because if you run out of even 1 of these consumables, you lose the ability to test samples, you can't provide results, you lose customers, and ultimately you then don't have samples available for sequencing. And I'm really proud to know that we were 1 of the few national diagnostic testing companies to not exceed more than a 3 day average turnaround time for PCR testing during the pandemic. And that's because of supply chain preparedness and because we developed our own internal assay to perform PCR testing using materials that we can source from multiple other vendors. So I just said a bunch of words about PCR testing and supply chain shortage mitigation strategies for diagnostic testing and I told you that we were successful. But here's what I said visually. The 2 animations here track all the positive PCR samples that we have seen in our network since the pandemic began. The map on the left is a cumulative display of all the PCR positives and the map on the right is a visualization of the positives in 2 week intervals. This map on the right is a really good estimation of the dynamics of the virus. Since infection lasts about 2 weeks, this map allows us to see how the virus has spread throughout the country. Now, this view only shows us where the positives are and as we all know, it's not just important to know where the positives are, but we also need to know where all the different versions of the virus are because that determines how we respond to the emergence of newer, deadlier variants of the virus. But in order for us to know where those variants are, we need an assay that actually tests for those variants and the LabCorp PCR test detects 2 regions of the SARS-CoV-2 virus. This only provides us with a yes or a no answer about whether the virus is present in a sample. It does not tell us anything about the mutations in the virus or which version of SARS-CoV-2 has been detected. For that, we need to do genome sequencing. The PCR test detects 2 small regions within the viral genome. And these regions provide the instructions for making the end protein, which can be seen here inside of the viral particle, protecting the viral genome from degradation. The viral genome contains obviously more information than just the code for the end protein, and it's the sequence of the whole genome that we're interested in for tracking how the virus spreads. But as I said, it's the PCR testing that allows us to do the genome sequencing. We receive raw sample in the lab. We isolate the genetic material of the virus, put those samples on a plate, and then detect whether or not the virus is present in the sample by PCR. The genomic surveillance process leverages the PCR process because we're able to collect those leftover samples and pick out all of the positive samples for whole genome sequencing. We don't have to re-extract samples. We don't have to pry a primary tube out of the COBOS or the Hologic system, and everything is nicely organized in a barcoded plate. Those condensed positive plates then make their way into the whole genome sequencing process. I'll get into this in a little bit more detail, but whole genome process converts the entire viral genome into something that we can detect using a genome sequencer to then identify the specific versions of the SARS-CoV-2 virus present in each positive sample. And then we associate those viruses with the position that they fall in within the viral family tree. The sequencing protocol is one that we co-developed over many months in collaboration with Pacific Biosciences. It's loosely based off of the protocol Nikki Fried developed for the Oxford Nanopore System. You can see the citation for that paper here in the bottom right hand corner. But the enzymes and primer concentrations of that protocol were optimized for PacBio along with some modifications to the barcodes to make them amenable to the circular consensus sequencing that's used in the HiFi platform for Pacific Biosciences. The protocol itself is very straightforward. We start out by creating cDNA from residual PCR testing material, amplify that material using 2 pools of non-overlapping M13 tailed primer pairs. Those pools are then asymmetrically barcoded using M13 tags and pooled for smart bell preparation. Sequencing then proceeds on SQL 2 using a 20 hour movie for the data collection portion. And this process has been fully automated on Hamilton liquid handlers and we have released the exact protocol that we use for doing this on the PacBio website. So if you're interested in running this protocol on your own SQL or SQL 2, please stop by the Pacific Biosciences website and you'll be able to download the protocol there. HiFi yield on the SQL 2 has been really impressive. We're able to load to a P1 of 70% without failing runs with a mean read length of 30,000 or 37,000 bases and a mean sub read length of 3,600 bases. We're seeing loading very, coming across very consistently in the 60 to 70% range, which helps a lot when we're putting about 800 samples on each run. Using this method, we can get whole genome coverage from approximately 60% of the samples with a CT of less than 27, or we can get lineage coverage from about 80% of the samples, less than 30 CT. That's really the only major tradeoff between PacBio and short read approaches. To get these long reads, you need high quality starting material. I've also included an example here of the run metrics with a waterfall plot to show where we see performance drop off and highlight that for surveillance, we are sequencing samples up to a CT of 30, but performance appears to seriously degrade around 27 CT. There are a number of optimizations that are in the pipeline to try to recover data from these samples above 27 CT. However, the majority of this effect is due to the availability of target capable of producing long reads for sequencing on the SQL 2. So we do have a new protocol in development to increase yield from these higher CT samples utilizing 600 base pair reads and preliminary results with that protocol are very encouraging. But we can't talk about PacBio sequencing without highlighting the advantages of long reads in variant detection and phasing. And as you can see here, the 1200 base pair amplicons that we're using allow us to phase variants within samples where coverage is high enough to do so. This could be used in the future to identify interesting low frequency subtypes that occur within individuals and potentially discover early novel variant strains as they emerge. In addition to that minor variant phasing, we can also use long reads to capture large deletions that exceed the maximum insert size of short read approaches. And here you can see deletions in two samples that would have been mostly missed by the short read single tile amplicon approach. The pipeline that we've set up for processing is also pretty standardized. We take the PacBio generated CCS reads into alignment and variant calling. For this, we use CLC. But there are other tools like Minimap and Juliet that work just as well. One of the major problems that we ran into with generating consensus sequences was that there really weren't any great tools for doing this that didn't introduce frame shifts into the consensus. So once again, we co-developed a method with PacBio to solve this problem. And this is work that was spearheaded mostly by Liz Tseng at PacBio, along with Laxeyer and Chen Zhang at LabCorp who developed a program called the VCF cons to generate these consensus sequences from a VCF file based on read depth by replacing missing sequence in the reference with ends and inserting variants with greater than 50% frequency into the reference. And this completely eliminated the frameshift effect that we saw with the more traditional methods for performing consensus sequence generation. VCF cons has been published, and it's another joint preprint from PacBio and LabCorp, and the code for it is also available on GitHub. For the rest of the pipeline, clade calls are made with Nextclade, lineage calls are made with Pangolin, and all of the raw data and metadata and consensus information is assembled and packaged for delivery to the CDC. But that pipeline wasn't generated overnight, and the genome sequencing work originally started as an internal research and development project. We recognized early on that we were going to have a centralized location for collecting positive samples and began as soon as we possibly could to save these samples in the R&D freezers. And it was clear to us that no one had the same access to positive samples that we did, and that we could play a really critical role in understanding how the pandemic began. So every day we would go into the lab and steal all of the plates and primary tubes that were going to be thrown away, and at the end of the day, we picked out all of the positive samples for archive. We did all of this at the same time that we were developing the PCR testing infrastructure that is now used to support over half of the SARS-CoV-2 testing at LabCorp. And we spent most of the summer developing the sequencing protocol in our spare time, and then we sequenced 10,000 of these early positive samples between September and November of 2020. And on December 7th, Dr. Michael Lewandowski presented the results of that early sequencing work at the ASM Microbial NGS Conference, and then in that presentation, he highlighted this model that we had developed for doing national genomic surveillance. And on December 27th, we were awarded a CDC contract to continue doing this work on a national level in real time. We were awarded that contract because of the data we had generated. Here you can see the SARS-CoV-2 genome on the bottom of the graph, and the bars you see growing in the animation are the number of mutations we detected at those positions of the genome of the virus. Over the course of our surveillance study, these mutations matter a lot with respect to how the virus functions and its ability to infect the host, evade the immune system, or even thwart treatments that we've developed to combat it. And I want to draw your attention to one mutation in particular. We've all heard about spike protein mutations, and here I'm highlighting the D614G mutation. Everyone on this call probably knows that this is an important variant, but it became the most important variant in the U.S. much sooner than most people realize, and that's because it spread across the country during a time when there was very little surveillance going on at all. And it's this data that we use to support our engagement with CDC to expand this internal surveillance project to a national one. And it's these types of mutations, like D614G and the spike protein, that we're all interested in because they're cool scientifically, but more importantly, because it's mutations like this that drive our response and guide vaccine and therapeutic development. And that's because it's the spike protein that the virus uses to enter cells, and it's these spike mutants that make the new viruses a little bit scarier. We've heard about B.1.1.7 that originated in the U.K., P.1 that originated in Brazil, and B.1.351 that originated in South Africa. The mutations in these viruses make them more transmissible, but the mutations in the spike protein, in particular, the region of the spike protein known as the receptor-binding domain, or RBD, are the most worrisome because it's this region of the virus that our immune system recognizes to fight off the virus. And it is feared that mutations in the spike protein will make the vaccines we develop less effective because the virus might change so much that our immune system might not be able to recognize the virus anymore. And unfortunately, we've seen evidence of this already, and the main takeaway from the graphs here are that the immune response to either B.1.1.7 or B.1.351 are reduced significantly in people previously infected with SARS-CoV-2, and the immune response to both of these new viral strains is also reduced in people vaccinated with the new mRNA vaccines. But it's important to note that the vaccines still appear to be effective, but I think this is illustrative of why pandemic surveillance is so important. We need to be able to detect these new variants in order for us to plan for how to defeat them. And the work that aids in the preparation of our response to variants consists of us sequencing many thousands of samples a week. We at LabCorp have been tasked with sequencing a subset of those thousands as a part of CDC's baseline study. This allows us to help the CDC track the individual viruses as they spread across the U.S., and those results are presented on the weekly CDC SPHERES calls. To do this, we centrally collect thousands of sample plates a week in North Carolina, condense them into the positive plates, and sequence the genomes of those randomly selected positive samples. But what does this look like in practice? This map here, this animated map, shows where the positives we're picking come from, and the intensity of the states represent how many positives we pulled from those states relative to the others, and the animation ends on the cumulative total that we've sequenced so far within the CDC project. We can also generate clade donut charts to show how the prevalence of each clade has or hasn't changed in the entire data set at the state level. Obviously, there are some sample size effects here, but it's still interesting to visualize, and a few interesting things have appeared in the last few months, especially relating to trends of the variants of concern in some states, and a good example of this is the evolution of the clade distribution in Florida, which is having a significant outbreak of the B.117 variant, also known and listed here as 20I501Y.V1. You can see that variant approaching about 25% of the samples that we see every week out of Florida. But you might be asking yourself, well, the work we've done has actually made an impact, and I think the answer to that question would be yes. It was a LabCorp sequencing surveillance efforts that detected the first case of the South African variant in the U.S., proving that we're able to detect things when we look for them, and we can have an impact on our ability to combat the virus by providing information to state and federal teams as they respond to the emergence of these new variants. And this isn't the only set of first cases we've detected. After detecting the 351 variant in South Carolina, CDC asked us to focus our sequencing efforts geographically. And again, this proves out the power of the system we've developed and the enormous reach that we have nationally. CDC asked us to concentrate picking to the southeast, so we did, and increased sampling across the southeast from about 100 samples per state to an average of about 300. And that sampling paid off because we picked up five more 351 samples that week. And then that set we picked up two that were newly discovered in Virginia and in Illinois. Following that focused effort, we have since returned to unbiased sampling across the U.S. and have seen an increase in the cases of both B.1.1.7 and B.3.51. Obviously, the denominator matters here, but this is good early evidence that pandemic surveillance does actually allow us to capture variants as they spread across the U.S. More recently, there have been reports of a new variant of B.1.526 that was first identified in New York. This variant includes the E484 case spike mutation, which resides in the receptor binding domain. And we saw the first of these cases on February 6th in New York and Connecticut. And over the next two weeks, we saw the caseload increase to about 179 primarily in the northeast with a few emerging in the southeast in North Carolina and Georgia. B.1.526 now represents about 10 to 20 percent of the cases that we're seeing come out of New York and New Jersey. And watching variants like this through genomic surveillance is important. And hopefully with the renewed interest in science-based policy, we and others are further enabled to continue this work and sequencing thousands of samples a week. And B.526 itself represents an interesting case because New York and New Jersey had a pretty big outbreak of SARS-CoV-2 back around this time last year. So the idea would be that they have built up a reasonable amount of population immunity and it appears that this virus doesn't really care about that. But one of the most important aspects of the LabQuert dataset is that it's based on random sampling. Unlike other sequencing initiatives that target sequencing samples that they believe are variants of concern, we sequence samples randomly across the nation. This limits the level of sampling bias in our dataset, although there are still states that we end up sampling more than others. But generally, this dataset is much less targeted than other initiatives. Because of this, we can look at national trends and the distribution of clades across the country. The dominant clade throughout our sampling has been 20G, which is the major U.S. clade that was described early on in the pandemic in the United States. But as you can see here in the last few weeks, the prevalence of 20G has been reduced, while 20I501Y.V1, which is B.117, and clade 20B are seeing a bit of an uptick, indicating that the other two clades may be outcompeting 20G. It's potentially a little too early to say that, but that looks like that's where the trend is going. And this trend for B.117 is also in agreement with early predictions that this variant would become a major U.S. variant by the beginning to the end of March. So I hope that this presentation has helped convince you of the importance of genomic surveillance during a pandemic. Sequencing-based surveillance allows us to know the things that we don't know. We've deposited over 15,000 sequences in GISAID over the past nine weeks in collaboration with CDC, combined with the work of other state labs and universities. Approximately 10,000 new sequences are making it into the databases every week. With that, we're finally now getting the data and the tools we need to respond to this pandemic with a reasoned scientific approach. And here I'm showing a headline from a bastion of scientific integrity, the New York Post, and the headline might border on fear-mongering, but this is really what we're up against. We're doing surveillance to try to discover harmful variants before they reach community spread and allow us to prepare for how to respond to these viruses socially and therapeutically. But obviously, this work doesn't happen in a vacuum. An incredible number of people across LabCorp have supported this work with far too many to name here, but a huge thanks goes out to all the operations teams, corporate IT teams, and lab directors that have supported this work for the last year and who now continue to support the deployment and scaling of it. On the development side, I need to thank my team, our informatics group, and the PacBio team. The majority of the picking and archiving work, along with the high-level analysis, was done by Dr. Mike Levandosky. Scott Parker and John Pruitt optimized the wet lab process for the sequencing protocol. Kim Wagner developed all the automation, not only the sequencing automation, but the automation that has supported tens of millions of PCR results. Ayla Burns performed all of the wet lab RT-PCR work that led to the countless FDA approvals and authorizations for testing, all the way from the original submission through to pooling, heat extraction, asymptomatic testing, home collection, and validation of nearly every plastic consumable known to science. Stan Litofsky and his team at the Center for Bioinformatics developed the analysis pipelines in conjunction with Pacific Biosciences. And a special thanks really needs to go out to Lax Ayer and Chen Zhang, who have, for many weeks, been shepherding samples through a not-so-automated analysis pipeline. And finally, additional special thanks to Liz Zhang, George Yuen, Andy Larrea, and the rest of the PacBio team, and, of course, Duncan McConnell, Diwani Batra, and Scott Sammons at the CDC, who have all put up with our antics for the past year. With that, I'm happy to take any questions that you might have. Please contact me by email or search me out on social media. Thank you.

Dr. Brian Kruger

LabCorp

national pandemic surveillance

whole genome sequencing

SARS-CoV-2 virus

variants