Hi, everybody. I will be presenting the next topic, which is what's new in NGS. My name is Elaine Mardis, and I have the following disclosures for you to read through before I start my presentation. So the objective for this session of the educational topic on NGS is really to explore some of the new long-read technologies that are being utilized in terms of single-molecule sequencing and then to finish up with a focus on RNA sequencing. So we'll start out first looking at the comparison between short versus long reads. As we've already talked about in the prior topic focusing on short-read next generation or massively parallel sequencing, we know that, as seen on the left side of this slide, that NGS libraries are relatively easy to make. They don't require highly intact input DNA, and they don't really require large quantities of nucleic acids to make the libraries. And in many ways, as has already been discussed, the short-read NGS has really revolutionized resequencing, especially the human genome and increasingly in the clinical application of NGS. We know that in short-read sequencing, we can do de novo assembly, so not reference-guided alignment like we've talked about, but actually assembling the reads based on overlap in sequences. But this is really suboptimal, especially when you're talking about a genome the size and complexity of human or other model organisms. So even with this high-quality reference genome alignment approach, we also know that short-read alignment misses structural alterations, especially deletions. And we'll talk in specifics today about how longer-read sequencing can actually fill in some of the missing components of the genome in terms of these structural alterations, but also in terms of another thing that's notoriously missed by alignment to a fixed reference, which is any novel content that might be in the genome that's being sequenced but may not be available for alignment in the genome reference. So on the right-hand side of the slide, I just have listed out some of the advantages of long-reads, and I'll go through these in more detail in the subsequent part of the talk. So long-read sequencing really does have some significant requirements for entry. This includes having highly intact, high-molecular-weight DNA. There are some specialized procedures that are utilized for the DNA and RNA preparatory steps, and they still require relatively large input amounts of nucleic acids to make these preparatories, although those amounts have been going down over time with a variety of improvements, which we'll talk about. We also want to point out the fact that these long-read sequencing data really provide a substrate for de novo assembly, even of large and complex genomes with a high repetitive content. The resolution of structural differences, such as these deletions, as well as novel content can be identified using the de novo assembly as opposed to alignment to a fixed reference. And although, in particular, the long-read technologies that I'll talk about today tend to have a higher error rate on a single-read basis, we can see from good evidence that the combination of read length and a really semi-random or completely random error pattern permits consensus assemblies that overall have very low error rates, as well as high levels of contiguity, allowing for haplotype-level resolution along long stretches of the chromosome, for example. So more on all of this later. Let's talk first about instrumentation. So the first commercially available platform was from Pacific Biosciences, offering single molecule real-time sequencing. The way this technology works is highly unique. It is still a fluorescent-based technology, like some of the previous technologies that have been discussed in this session, but in particular, the Pacific Biosciences sequencing occurs entirely in this device that is shown here, which is a zero-mode waveguide. So the zero-mode waveguide is really just a nanomachined pinhole, if you will, that effectively focuses the laser excitation wavelength in a very, very small area, about 20 zeptoliters. And this very small area of focus for the excitation wavelength, in effect, gets focused on one thing, and that is an active DNA polymerase that is bound to a specific DNA molecule. Now, of course, in this technology, we don't just have a single zero-mode waveguide, but we have up to millions of these zero-mode waveguides that are on the sequencing substrate, the so-called smart cell, single molecule real-time cell on which sequencing takes place. So what we can see in this particular image now is just the way that the DNA polymerase complex is seated down in the zero-mode waveguide, and these are immobilized in place for the purposes of the sequencing reactions being observed by the instrument optics as they go on. This is how this works. So, of course, the natural substrate is required for the sequencing of the DNA molecule, this complex with the polymerase, and this occurs using differentially labeled, introduced all at the same time, fluorescent dNTPs. So you can see ACGNT here, each with a different fluorescent label. These get introduced at physiological conditions into the zero-mode waveguide by being added using the instrument fluidics onto the smart cell, and these nucleotides then, by diffusion, can essentially float in and out of the active site of the polymerase, and if it's the appropriate nucleotide for incorporation, the dwell time in the polymerase active site is significantly longer than if it samples in and is not a match for the complementary nucleotide that's in the active site. So what essentially happens in the PacBio sequencing, then, is that each zero-mode waveguide is observed by the instrument optics over a long period of time, of hours, the runtime, during which each time we have a nucleotide in the active site, we have fluorescence being excited by the excitation wavelength and the resulting fluorescent emission being captured by the instrument optics and assigned to a nucleotide identity based on the wavelength of fluorescence light that is being captured. And so this process, as I mentioned, already occurs in parallel across millions of zero-mode waveguides and is recorded in these movies over time to produce the sequence of that DNA molecule as the DNA polymerization process takes place. What do the libraries look like for the process of PacBio sequencing? So, as I've already mentioned, you need high molecular weight DNA, really of sufficient quality to allow for pieces of about a 30 kb on average to be produced, to produce these continuous long reads or the hi-fi reads that I'll talk about in a minute. We typically employ novel technologies to shear the genomic DNA to large fragments, and this is in the form of specialized instruments such as the Megarupter platform. We take these sheared DNAs then through extensive repair processes prior to adding on these very unique adapters that are shown in this figure on the right. The little blue loops are the specialized adapters for the smart bell production, so the templates for smart cells are called smart bells. As you can see, it looks like a little dumbbell in the linear form, and the way the adapter works is that right at the end where the ligation to each fragment occurs, you have complementary sequences that are double-stranded, but the majority of the adapter is a unique sequence that is single-stranded, and when denatured, essentially forms a circle, as you can see below. Just prior to the denaturation, however, the DNA polymerase is added in with a primer that corresponds to the sequences within the smart bell adapter, and you can see that this is then a polymerized, as we talked about already in the zero-mode waveguide, with the optics collecting the information from the DNA sequencing as it reads out around the Watson strand, then the Crick strand with the individual adapter sequences interspersed between Watson and Crick. So, as I mentioned earlier, the input DNA requirements have been decreasing over time. Currently, about three micrograms of genomic DNA is required to produce libraries suitable for coverage for sequencing a large mammalian genome, such as the human genome, and you can see down below that this movie that I talked about that takes place collecting data over a series of hours is processed once the run has completed, where the raw reads are filtered for quality, and then turned into high-quality consensus reads that essentially combine the sequences of Watson and Crick strands for each fragment in the library that's been sequenced in the sequencing process to produce final high-quality consensus reads. This is a picture of the platform now available commercially from Pacific Biosciences, the SQL or SQL2. You can see that the numbers of zero-mode waveguides differ in terms of the smart cell. Up to eight million are available, and you can just see a comparison below that around how much data can be collected in a 96-hour run across multiple smart cells, either using the consensus long-read sequencing or using this newer mode of sequencing, which is HiFi. This uses a new HiFi polymerase with an improved error rate for single reads and produces overall a consensus, sorry, a circular consensus sequence using this combined approach for the individual reads that improves error rates to about 0.1 percent overall. So again, the output of this instrument is a relatively high-accuracy long-read sequence that can be produced as follows. So this slide is really just showing the results of read lengths obtained either on the left from HiFi sequencing of DNA constructs made from a mammalian genome into smart bells and sequenced on the SQL2, and you can see that the peak of read length here is around 10,000 base pairs, hundreds of thousands of reads produced, and then on the right-hand side, this platform is also capable of sequencing RNA using a process called iso-seq, which can produce full-length RNA sequencing, again, with a very similar read length distribution profile from RNA sequencing. Here's just a close-up example of RNA-seq from the iso-seq method, just comparing in the center column data obtained from SQL1 and in the far right column from the SQL2 platform. These were identical samples from a human tumor that were studied using iso-seq. You can see the number of smart cells and the duration of movies, the huge increase in comparing the output reads and bases between SQL1 and SQL2, and importantly, the fact that many unique isoforms can be identified and mapped down to unique loci. So in this particular example, over 15,000 loci could be studied from the SQL2 data with a multitude of different isoforms. So as complex as it is to produce isoform predictions from short-read RNA-seq data, this platform actually can produce contiguous isoforms that can be studied in full detail to really understand the alternative splicing landscape of a given tumor, for example. What else can these data be used for? Here's just an example of long-read DNA data being sequenced through a particular gene, the OTS2 locus, that has been implicated in autism for complex structural rearrangements that may lead to autistic phenotypes. And in particular, the comparison here are packed bio reads that are showing these insertions within the OTS2 locus compared to, below them, Illumina reads that are showing a very low level of mapping, as indicated by the blue arrow, in this particularly challenging region, which, as you can see at the bottom, is highlighted by a very high tandem repeat content. So effectively, with the Pacific Biosciences sequencing, we are reading directly all the way through this locus on specific fragments in the library and not having to worry about alignment-based reconstruction of the locus, as one would have to do with Illumina sequencing reads. Here's yet another example. For those of you who are interested in pharmacogenomics, you'll recognize the difficulties of sequencing the cytochrome genes. One of the most difficult ones shown here, cytochrome CYP2D6, adjacent to its neighbor on the chromosome CYP2D7, and very nicely resolved to haplotype data from these very complex genes that are in the, you know, in the context of their segmental duplication regions on this chromosome and in comparison to the genome in a bottle of reference. So, getting very high quality and contiguity across what is well known as a very difficult region of the human genome. Then, just to finish this thought, we are looking at a study from last year, Nature Biotechnology, using 13 kb circular consensus reads in the human genome and actually showing that, in medically relevant genes, so-called NGS problem exons in 193 such genes, there are a very large number, 152 to be exact, where 100% of the problem exons are actually mappable using the circular consensus reads from the PacBio platform, and then some diminishing returns in the smaller number of genes listed down below, but still a significant resolution of these very difficult regions of the human genome. Another example from work that we've done is shown in the next couple of slides. This is just a patient from one of our cancer sequencing efforts with a diagnosis of a spindle cell neoplasm. This child was born with a right thigh mass that was really not clear whether it was cancer or not, and what we were able to do in our sequencing process was to show that there was a structural alteration in this child's somatic genome around this well-known cancer gene, platelet-derived growth factor beta, and we could tell from the single read alignment to the human genome that this was a complex alteration that really was affecting the area of the juxtamembrane segment. You can see that on the diagram on the far right. This is the region that's commonly mutated in cancers, and these mutations typically result in ligand-independent activation of this receptor tyrosine kinase. However, using the short read sequencing, we were really not able to get a full resolution on this complex alteration, which we could tell was spanning exon 12 through 15 of PDGFR beta, and so to address this challenge, we performed isoseq RNA sequencing after selectively amplifying this locus by PCR, and we were able to sequence through this region, assemble the reads to identify the unique messenger RNA structure that had resulted from the rearrangement, and this is shown here. So, on the top of this slide, you can see the normal arrangement of the wild-type locus between exon 11 and 12 of PDGFR beta. You can see in the intron between these two exons, there's a couple of landmarks that will become important as we understand the revised or structurally altered version of this gene, and then you can see the normal splicing below of the normal transcript, which juxtaposes exon 11 and 12 with the corresponding amino acid sequences listed underneath. Now, let's look at what we found from the isoseq sequencing. You can see in the DNA of the structural variant along the top of this lower part of the slide that intron 11 has been the subject of a complex rearrangement, inserting part of intron 15 as well as inverting the remainder of intron 11. You can see now the new location of that miR3 landmark. Following on there, that rearrangement, there's also exon 12 with an interspersion of insertion, rather, of intron 14, then exon 15, and then the remainder of exon 12, and so this is the very complex structural variant that we were able to reassemble in comparison to the splice variant transcript from isoseq along the bottom, which shows the rearrangement of exon 15 and 12 in the messenger RNA, the resulting protein sequence that resulted in in-frame. So you get the full-length PDGFR beta, but with this juxtamembrane rearrangement, it is turned continuously on and is signaling through to cause the cancer. Let's turn our attention now to a second instrument that is more recent in the commercial space, which is the Oxford Nanopore single molecule sequencing device. This is a very interesting approach that is without the use of any sort of fluorescent or other labeling, and really just involves the translocation of a DNA fragment through a protein nanopore that's set in an electrically resistant polymer membrane. So this nanopore-based approach is shown here diagrammatically, where you have a double-strand DNA that essentially is captured by this auxiliary protein on top of the nanopore that acts to do two things. One, to untangle it, so it unwinds the DNA double helix, and then helps to translocate the single-stranded part of that helix through the nanopore. How is sequencing done? Well, essentially because the protein nanopore sits in an electrically resistant polymer membrane, when the DNA molecule translocates through the nanopore, there are changes to the conductance of ions on either side of this electrically resistant polymer membrane. And that's shown stylistically in this gray box, a series of up and down squiggles, if you will, with the corresponding nucleotide sequence over top. So the obvious question then is how do you collect this data? Well, first of all, the data are collected by an ASIC, application-specific integrated circuit. Each nanopore is coupled with an ASIC, so the data from electrical conductance measurements can be recorded for that particular nanopore. And of course, the sequencing device is not, again, just a single nanopore, but multiple nanopores throughout the membrane with a corresponding ASIC to collect the data. The way that we convert the differential electrical current into sequence data then is a very interesting approach, which essentially starts with collecting the changes in electrical conductance across the membrane that occur when a polymer, a DNA polymer of known sequence is being pushed through the nanopore. This tells us the corresponding electrical current change for that particular sequence. And being done across all different combinations of DNA content results in a machine learning approach that can then go in and interpret these differential electrical conductance diagrams for each individual fraction. So, the pore current changes are compared to this model of all possible possible multimers and the resulting DNA sequence is produced for each fragment that's been sequenced in the nanopore. As you can imagine, this is subject to some level of error rate and we'll talk about that here. The error rate is around 10 to 15 percent. There have been steady improvements in this error rate over time with improved pores, improved software, and improved improvements in the machine learning model. In the past, this has really been utilized primarily for fairly straightforward genomes such as bacterial and viral genome sequencing. The technology can also be used for RNA sequencing. But in particular, the use of Oxford Nanopore has now expanded out to human genome sequencing and in some ways has really benefited from the assembly software that's been developed for the PacBio long read sequencing data that we just talked about because both of these long read technologies have similar error rates. The error models are a little bit different, but nonetheless, the same software can be utilized for de novo assembly of these long reads. The other thing that I didn't talk about for Pacific Biosciences as well as for Oxford Nanopore is there are ways to also interpret whether there are methylated residues on the DNA that's being sequenced. And so this is an add-on that doesn't require differential processing, but just comes along with the DNA data analysis for these two instruments. As I mentioned, the Nanopore sequencing more recently has been applied to human studies. This is a paper from Nature Biotechnology describing the assembly of human genomes using Oxford Nanopore data. And this is just a figure from that paper showing comparisons between different assemblies using the genome in a bottle benchmark where we're looking at, in particular, the improvements of concordance to that genome in a bottle benchmark along the x-axis measured by 100 kb assembled chunks of data. So these are cumulative percentages showing that there's a steady improvement from, at the time this paper was published, purely Oxford Nanopore data to Oxford Nanopore coupled with Illumina for error correction, then even better improvement with the PacBio consensus long reads, and then the best quality confirmation to the genome in a bottle benchmark concordance with the PacBio circular consensus sequence data. So just an indication that the use of long read technology for assembling and studying human genomes is really upon us. And while there are still some remaining challenges with respect to DNA quality and DNA amount that I talked about earlier, this is really approaching, in some ways, a clinical utility but may not quite be there just yet. So in conclusion, the long read sequencing data is really largely focused on this question of read assembly versus read alignment, the ability to identify novel content, to resolve haplotypes over long segments, to identify and accurately reassemble these and understand these complex genomic rearrangements. There are good evidence now that even short read technologies are really missing deletion events of 50 to 500 kb in size, and these can be very well mapped out, as I've tried to illustrate in some of the examples that I showed, using long read sequencing technologies. And these are actively being used, these technologies, to generate additional high quality reference genomes so that we have additional reference genomes for comparison, especially from individuals of different ancestries. And as I already alluded to, these medical applications of long read sequencing technologies really may benefit from having a higher diagnostic yield because of the identification of novel content and the ability to identify large deletions or complex structural rearrangements. So I want to turn my attention now, for the remainder of the talk, to some of the emergent trends in clinical genomics. And this is really, as I mentioned earlier, focused on RNA-seq. This has really been a focus for many reasons. There are some that are listed here. First of all, we are finding that just looking at DNA alone is insufficient. Because RNA is that intermediate molecule between DNA and proteins, we often find out what's encoded in the DNA at the level of RNA. And this includes epigenetic changes, such as chromatin remodeling or accessibility that may change in the context of health versus disease. Methylation as well can change, and both of these epigenetic events can influence DNA biology. They can be evaluated by studying DNA through specialized methods and accompanying analytics. But in many ways, if there is a source of affected tissue, or even sometimes just by studying blood cells, we can look at RNA-seq data and infer a lot of information that's coming out of epigenetic events. We can also, as I'll talk about, evaluate the RNA-seq data in other ways to yield some valuable biological insights that can be applied clinically. And there are some attendant challenges to studying RNA-seq, which we'll begin with. So first of all, what are the RNA gotchas? RNA is a much more labile molecule than DNA. It's sensitive to degradation from heat, from a variety of chemical treatments, and those include formalin fixation and paraffin embedding, which is commonly used in pathology to preserve protein in cellular structures, but ultimately, which damages RNA and DNA. We also can get genomic DNA co-isolating with RNA, and this can cause multiple problems. And so we're strong believers in DNA treating anything prior to using it in RNA-seq or other RNA-based assays. There are advantages to isolating messenger RNA in terms of the sample and studying it from a clinical standpoint. However, oftentimes clinical specimens do not yield a sufficient amount of total RNA that would be conducive to then isolating specifically messenger RNA, which only represents about 2% of the total RNA in a cell. And so there are other ways to get around this, which I'll talk about. We also, in many cases, want to pay attention to strandedness, which strand is actually being encoded, and there are now library construction methods that are appropriate for this. And in particular, as I've already talked about, short reads can make detecting certain RNA-specific attributes or aberrations difficult to sort out, and this includes exon skipping, isoforms, and others. So we'll talk more about that. First rule of business is, I mentioned RNA is labile, so QC of RNA is essential. These just show some adjuvant traces of different RNA samples. You can see, for example, the high-quality sample is sample 11 in the lower right-hand corner. The RIN is the RNA integrity number. It's very high here, 8, 10 is perfect. And you can see the nice ribosomal RNA peaks for this sample. Contrast this with the other three samples, which were obtained from FFPE tissues, and you can see, for example, that these have much lower RIN values. And also the other value that we pay a lot of attention to is this DV200 number, which tells you the proportion of molecules with longer than 200 nucleotides in the fragment. In particular, we can see that the two green-labeled samples have greater than 70%, which is the desirable DV200, but still a relatively low RNA integrity number. So what can we do about that? Well, one thing that can be done for poor-quality RNA is actually to turn it into an RNA-seq library and then take that cDNA library through a hybrid capture approach, such as with an exome capture reagent, that will really, as we showed in this paper from a few years ago in the Journal of Molecular Diagnostics, improve the yield of sequence reads that actually map to coding exons. So this is actually highly effective and can rescue RNAs with RIN values as low as 2 or 3. This is a generalized overview of RNA-seq and how it works. We'll talk a little bit about ribosomal RNA depletion. I already mentioned poly-A selection or mRNA selection, which is a very appropriate way to sequence RNA to remove ribosomal RNA sequences automatically, but does require a reasonably large input amount, which we don't always have the luxury of clinically. These approaches, once you've decided either to do ribosomal RNA depletion or poly-A selection, you can go through a cDNA synthesis to turn the RNA into DNA, a preparation procedure, and then concentration normalization followed by sequencing. So a fairly straightforward way to get some data from your RNA sample. Ribosomal RNA depletion is used by many labs. Again, in a clinical setting, if you have a relatively low amount of RNA available from total RNA, this straightforward hybrid capture step essentially can be used to deplete ribosomal RNA, which is a very large percentage of the population of RNA coming out of a cell. And these straightforward commercial reagents such as RiboZero or RiboZero Gold that are shown here can effectively knock down the ribosomal RNA so that you can sequence the coding protein genes that you're interested in for your study. This is just an example of a directional library construction approach that we have developed using the NEB Next Ultra 2. This is a stranded or directional library approach that can start from a fairly small amount, down to 100 nanograms of total RNA. It utilizes a hybrid capture-based RNA depletion effectively using DNA probes to capture ribosomal RNA that then gets degraded by RNase H. And you can go forward to the RNA fragmentation, cDNA synthesis, and other steps that I've already outlined in this simplistic approach. This also shows the use of hybrid capture, in this particular case using IDT probes, if needed, to enhance the yield of reads onto the coding exons prior to library QC, quantitation, and the sequencing of the RNA library. So what can we learn from RNA analyses? Here's one example of just using short-read sequencing data to look at the comparison between a primary AML, shown in the red diagram, and that same patient's relapse or secondary occurrence of AML. In the first or primary occurrence, we were able to identify a small subset of reads that actually correspond to this donor site mutation in FOXP1 that's shown in the blue gene diagram at the bottom of the slide. And the donor site mutation results essentially in the skipping of this second exon shown in the diagram. So some proportion of the reads in the primary RNA were skipping that exon. That mutation was in a subclone of the primary RNA that did not recur in the relapse, and so it's now gone and there's no evidence of that exon skipping event because this donor site mutation is no longer present. And the green diagram is just from a completely different leukemia patient showing, again, the normal splicing pattern that one would expect to see for comparison. In our current work in pediatric cancer, we need to have a very robust fusion detection pipeline starting from RNA sequencing, and that's shown here. In the orange boxes across the middle of this diagram, you see the names of different algorithms that are used to identify fusions from RNA-seq data. And the reason we are using so many in this diagram coming out of the paired-in read sequencing on our RNA library is because there's a high false positive rate of detecting fusions from RNA-seq data. And so this is essentially a way of getting different lenses to look at the possibility of fusions and then combining the results from all of these different collars using an overlap script where we know that if we have fusions that are detected by two or more of the fusion collars, we can usually then take those forward to cap CLIA validation if they look like they're clinically interesting and get a high return of around 100% of clinically confirmed fusions coming out of RNA-seq data. Here's just an example of some of the known and novel fusion events that have been identified through the pipeline that I just showed you where the purple bars are known fusions that have been described previously in pediatric cancers, and the height of the bars refer to the number of fusions that have been identified. And then in the green bars, some of the novel fusions that we've also been able to clinically confirm for our patients. Another way of looking at RNA-seq data is using principal components analysis to compare a single sample, often of unknown diagnosis or indeterminate diagnosis, to a variety of other samples which have the same RNA-seq data type available for them. We utilize this tool that's available through the Treehouse Initiative at UC Santa Cruz that also databases information on the top 500 expressed genes from a multitude of different cancer types, primarily focused on pediatric cancers. And so this tool has been very, very helpful to us in terms of sorting out information about the histological and often subtype level diagnosis of a given patient for which sometimes conventional pathology cannot determine an accurate diagnosis. Another way that we use the Treehouse dataset is shown here, which is really to evaluate our individual patients in the IgM patient cohort for gene expression outlier analysis. Here we use the data from the Treehouse and from our own patients through a Salmon analysis to obtain a transcript per million or TPM values for each RNA entity. And then we further refine those TPM values to just focus on known protein coding genes. These are then normalized and calculated in terms of their differential expression using the equation that's shown in the middle of this slide, and then compared using the approach described originally by Kotari et al. in a 2313 cancer discovery paper that's referenced at the bottom of this slide. Using this approach can either give us a variety of samples to put into ingenuity, a variety of genes, rather, to put into ingenuity pathway analysis to identify elevated pathways, and in some cases can present genes that are drugable, and we usually assess this drugability using the drug-gene interaction database that's illustrated on the left-hand side of the slide. So here's just one example from this patient with a sarcoma that I was mentioning earlier in terms of the complex rearrangement. In this particular patient, we can see that the MET pathway by ingenuity pathway analysis is being activated. You can see MET in the red triangle at the top, and there are a variety of other lighter pink and red genes that are also indicated by our outlier analysis through ingenuity pathway analysis showing this elevated pathway. In this particular diagram here, we're looking at the principal components analysis of this patient's top 500 expressed genes, and hopefully you can see that this patient in the black dot towards the bottom of the slide is clustering directly with a number of different other patients with a diagnosis of sarcoma of the bone, and this is consistent with the patient's diagnosis as well as with the MET activation being indicative of the potential for therapeutic vulnerability. One of the other newer ways that we're interpreting RNA-seq data is to understand the immune cell content in the tumor microenvironment. This is using an algorithm called Cybersort, now CybersortX, which has been trained on RNA-seq data, and what Cybersort does is it has already a pre-calculated expression matrix that has learned the gene expression profiles or signatures associated with each of 22 different immune cell types, and what happens in Cybersort is the RNA-seq data expression data are compared to this matrix and essentially put through a significance threshold for the quality of match to the LM22 matrix using permutation analysis. That output provides the relative fractions of all 22 immune cell types as well as a global p-value for confidence of immune cell signatures that have been identified in the tumor microenvironment. Here's just an example of how this comparison works. We're looking here at a patient with an NF1-associated glioma, a primary as well as a recurrent tumor, so the recurrent tumor has recurred since the removal of the primary tumor by surgery. What you can see on the left-hand side of this diagram are just the comparison of different mutations between the primary on the x-axis and the secondary tumor on the y-axis, where really there are very few shared mutations, including the second hit in NF1 shown in the middle of this diagram. On the right-hand side are the Cybersort calculations for immune cell types that are present in the recurrent glioma, shown in lanes A and B, and the primary glioma shown in lane C. What you can see in the comparison of primary to recurrent two samples is that this was a fairly—is that there's an enhancement of monocytes in purple and M2 macrophages in dark red in the recurrent tumor when compared to the primary tumor. This tells us that there has been an immune cell infiltration accompanying the recurrent NF1 tumor. This is really just one indication of how we're using genomic DNA to really tell us more about the immune microenvironment, as well as other aspects of what we commonly refer to as immunogenomics. This really is a teaser for the hot topics of a lecture that I'll be giving later in the meeting, talking about immunogenomics in clinical cancer. This is referencing a Nature Reviews clinical oncology paper that Mike Berger and I wrote and published a couple of years ago, which is really just showing that using the human genome reference and tumor versus normal NGS data alignment and variant analysis, we can come up with a variety of somatic and germline variants. Conventionally, what we think about is using the small molecule inhibitors that have been identified and developed and FDA-approved over the years to really forge this link between genomics and therapeutic decision-making, and this has been augmented by large-scale discovery projects like TCGA that have really reinforced and expanded the use of genomics to tell us about aspects of cancer onset, progression, and treatment resistance. In my hot topics lecture, I'm going to focus more now on how the success of checkpoint blockade therapies, for example, in high mutation load cancer types like melanomas, lung, and bladder cancer has also really utilized genomic evaluation to gauge the likelihood of response, mainly focusing on calculations of mutational load, such as tumor microenvironment. And in an even newer application of genomics, we were also able to predict neoantigens, these tumor-specific mutations that can be used to engage the immune system of the host. So more on that later in my hot topics lecture, but for right now, hopefully this has been a nice illustration for you of the different uses of DNA sequencing, RNA sequencing, long-read technology, and what's new in NGS. Thanks very much for your attention.

Next Generation Sequencing

long-read technologies

RNA sequencing

PacBio platforms

Oxford Nanopore

RNA extraction

gene expression profiles

immunotherapy