Scientist Stories: Jatinder Lamba, Pharmacogenomics saving lives with personalized treatment

Hello everybody, and welcome to the ESHG conference. My name is Jyothinda Lamba, and I'm a professor at the College of Pharmacy at the University of Florida, and my research area is basically looking into the multi-omics aspects in pediatric acute myeloid leukemia. For today's talk, I will present and share with you some of our data focused on current and future prospects of pharmacogenomics of CD33-targeted immunotherapy in AML. Before that, I will talk a little bit about AML. Acute myeloid leukemia is a common leukemia in adults and is the second most common childhood leukemia. It's a very heterogeneous disease with cytogenetic risk groups that affect the outcome. And the pie chart you see here actually shows, reflects some of the cytogenetic lesions or genomic lesions that occur commonly in AML. And these genomic lesions are used to define or classify patients into different cytogenetic risk groups as favorable risk group, intermediate risk group, and adverse risk group. And the survival curves over here, the Kaplan-Meier survival curves, show how these risk groups behave with respect to the treatment outcome. And as you can see, favorable risk group patients have better outcome, intermediates are in the middle, and adverse cytogenetic risk group patients have the worst outcome. And two graphs here reflect outcome in adults and pediatric AML. And you can very well see here that overall pediatric patients are doing better as compared to the adult AML patients. The overall 5-year survival rates for AML is 30 to 40% for adults and around 70% for the children, showing that this is pretty dismal and we really need to make efforts in order to design better treatment strategies for these patients. Talking a little bit about AML chemotherapy, the standard AML chemotherapy since 1970s has been combination of cytarabine, daunorubicin, and etoposide, commonly referred to as ADE. And even though this has been used for last 40 years, it fails to induce remission in around 10 to 50% of the patient— of the children and 20% of the adult AML patients. And despite the fact that significant proportion of patients achieve remission after first induction, among those who achieve remission, roughly 40% of children and around 70% of adults relapse. And the worst part is that over 90% of patients with resistant or relapsed AML, they die within 3 years. The patients within high-risk category are also candidates for the hematopoietic stem cell transplantation in AML. Since 2017, a couple of new agents have been approved for treatment of AML, one of them being CD33-targeted agents, and most commonly used, or the one that is so far approved by FDA is an antibody-drug conjugate known as gemtuzumab ozogemisin. And then there are other drugs that have been approved recently as BCL-2 inhibitor, or sonic hedgehog pathway, Vixios, and FLT3 inhibitor, Midostaurin. So we will focus for this talk on CD33 and CD33-targeted gemtuzumab ozogemisin. So CD33 is a myeloid differentiation antigen that is present on the AML blast in most patients. And gemtuzumab, or I will refer to it as GO for the rest of the presentation just for simplicity, it's a pretty big word. So gemtuzumab or GO is an anti-CD33 antibody which is conjugated with the cytotoxin caligemisin. So what happens in case of AML, so this is how gemtuzumab looks like, gemtuzumab ozogemisin or GO looks like. It is an anti-CD33 antibody which is linked with calychomyosin. So what happens when the patient is given GO, the CD33 antibody binds to CD33 antigen on the leukemic cells, gets internalized, and goes to lysosome. The acidic pH in the lysosome releases calychomyosin, and the calychomyosin release goes into the nuclei causing— it binds to the DNA minor groove causing double-stranded DNA breaks and the cell killing. A number of clinical trials have shown benefit of adding GO to the standard chemotherapy. In fact, it has a very unique story. Initially, GO was approved for treatment of CD33-positive AML in 2000, but a phase 3 study resulted in withdrawal from the market in 2010 due to increased number of deaths in the GO arm. But then subsequent studies like ALPHA study, AML1661 study, and the AML15 or AML16 MRC clinical trials showed that addition of GO to the standard chemotherapy improves outcome that resulted in FDA's approval of GO for CD33-positive AML in September 2017. So here is a snapshot from FDA showing from 2017 resulting in approval of Cemtuzumab for CD33-positive AML. Another one a little later in 2020, approval of GO for CD33-positive AML in pediatric AML patients. So what we— our lab was interested in looking at is what if there is genomic variation in CD33 because it's a direct drug target. So if the patients have variation in CD33, how does that impact treatment outcome? And over the years, we worked on a couple of clinical trials to identify CD33 SNPs, and I will share with you the results from one of the largest clinical trials in pediatric AML where patients were randomized to two arms. And I'm sorry, this— The number here is just moved somehow. But the patients were randomized to two arms with either the standard chemotherapy with 511 patients right here or the standard chemotherapy with addition of two dosages of GO, one in induction one and one in intensification here. So our goal was to look at the top CD33 SNPs that we have already established and see whether they impact the outcome in response to GO or not. And here is the list of SNPs that we have shown from this study and also from the previous study that these SNPs were associated with CD33 levels on the leukemic cells. And you can see this top SNP is the one that I'm going to focus a little bit more on. This is a missense polymorphism, which results in a C to T change. It also occurs in LD with the promoter SNP. And you can see it's very significantly associated with CD33 levels on the cell surface of leukemia, leukemic cells. So you can, you know, really see here that if the patients have low CD33 due to presence of the variant allele, this is going to result in low levels of the drug or very less amount of the drug binding and internalizing, thus resulting in poor response. And there are a bunch of other polymorphisms that I have listed here that also showed significant association with cell surface levels of CD33 in AML patients. We will talk a little bit more about the first SNP that I showed was a coding SNP. Interestingly, this SNP is also a splicing SNP. So, here is what you see, a schematic of CD33 gene with the SNP localized right at the beginning, 4 base pairs into exon 2. When there is a change from C to T, what it does is it impacts the binding score for a splicing protein, SRSF2. So, the binding site is here compromised. So, what this results in is loss of this exon 2. So basically exon 1 joins with exon 3. So, when C allele is present you have a full-length CD33 that codes for the full-length CD33 protein. And this figure over here reflects the CD33 protein structure. It has an IgV domain, IgC domain, and two ITIM which are the intracellular domains.. But due to this splicing defect or due to the SNP resulting in skipping of exon 2, we have named it as CD33D2 isoform. This IgV domain is gone. And you can see this reflected over here from either CD33 density or with the CD33D2 isoform that patients who had TT genotype had significantly high levels of this shorter isoform of CD33.. Now this is very, very important because GO targets IgV domain. So basically the patients, if they do not have IgV domain, the drug just does not have anything to bind and cause a therapeutic effect. Additionally, the antibody 67.6, which is used in clinics to quantitate the levels of CD33 in a patient's leukemic cells, also binds to IgV domain. And that is what you see here that presence of this T allele is associated with lower levels of CD33 when the quantification is done using this antibody that binds to IgV domain. So this also reflects not just with the levels of CD33 but with the clinical outcome. So I will show you here that what you would expect is that the patients who have CC genotype, they make wild-type CD33 with the domain that is recognized by COH. So these patients should have benefit from COH, and that is what you see here. So what is reflected in these curves is risk of relapse and comparing the patients in the standard arm, which is known as— which is reflected, shown here as no-GO or standard chemotherapy arm, or GO arm where patients were given standard chemotherapy with addition of 2 dosages of GO. So what you see here is that if patients have CC genotype, there is lower risk of relapse when you give them GO, so indicating that they benefit from addition of this GO to their standard chemotherapy. Unfortunately, if the patients have even one T allele present, this will result in skipping of exon 2 and this loss of IgV domain, which is recognized by GOW. You see the curves are right on top of each other, showing that whether the patients are given standard chemotherapy or if they were given standard chemotherapy with addition of GOW, they did not have any benefit with the addition of GOW. Similarly, you see the similar results with disease-free survival. The patients who received GO have a better disease-free survival if they had the CC genotype, but they do not benefit with addition of GO with respect to the survival outcome too when they're given standard therapy or standard therapy with GO if they had the CT or TT genotype. So this implicates— this implies that If you have to give a patient gemtuzumab or zytomycin to treat their AML, you should pay attention to their genotype for CD33 SNP. So this is overall the SNP map of CD33. These are the major SNPs that I have shown you in the table before. But here you could see that the SNPs occur in varying frequencies in different ethnic groups. So the SNP map over here reflects the SNP and the frequency in white patients and the frequency in the African American patients. So interestingly, this splicing SNP that I showed you in the previous slide, which has such a significant impact on the treatment outcome, occurs with 34% frequency in white patients, but the frequency is lower in the patients with African ancestry. However, if you pay attention to other SNPs, for example, 18301, this one has completely reverse frequency. And then there is a SNP here, this is very rare but very detrimental, that is CCG deletion, and this deletion results in premature termination codon. So, I will show you a little bit about this SNP in next couple of slides. So, here is what you see is that CCG deletion in CD33, it results in premature termination codon. So, you can see here the sequence when CCG deletion occurs, it results in a shorter CD33 isoform due to creation of a premature termination codon. And you can see here the flow data from a patient before transplant with the monoclonal CD33 is reflected here and CD14 here. So this patient had very low CD33, but post-transplant from the donor, the patient got the cells that were expressing high levels of CD33. So you can see here there is high levels of CD33, but when this patient relapsed again, there is loss of CD33. And when we genotype this patient, this patient was homozygous with CCG deletion for this indel in CD33. So looking into it, we wanted to see if we can create a score within the patients for the multiple CD33 SNPs that are either— that are associated with the CD33 cell surface levels and also with the clinical outcome. So, what we did, we compiled top 6 most significant SNPs and we gave them a genotype score. So, if the SNP was associated with low levels of CD33, it was given as 0, -1, or -2 based on number of variant alleles that are present. Or, if the SNP was associated with high expression, it was given 0, 1, and 2, which was only one SNP. RS2455069. So, and then we added the genotype score for all these 6 SNPs in order to come up with the CD33 6-SNP composite score for each patient. And then we looked at how this 6-SNP score is associated with CD33 intensity on leukemic cells. And you can see here beautifully, if the patients have negative scores, they have low levels of CD33, and as the score the CD33 levels increase. Based on this, then we dichotomized patients into two groups just to enhance the clinical utility of the score because if you go to a clinician and give them all these scores, it's very tough to figure out who is, which patients are likely to benefit or not benefit from CD33. So we dichotomized patients into two groups as CD33 six SNP score less than zero or more than zero.. And as you can see here, this score was very significantly associated with cell surface levels of CD33. And so was similar association was seen with treatment outcome. Higher CD33 SNP score predicted better response. You can see here a group of patients who had low CD33 score, and you can see that with respect to disease-free survival as well as risk of relapse, they did not benefit from addition of if their score was less than 0. The curves are right on top of each other whether they received a standard chemotherapy or standard chemotherapy with addition of GO. However, if their score was greater than 0, then adding GO really made impact in these patients with respect to improved survival and reduced risk of relapse. So in the end, it's like, what can we— where do we go from here is that we can take preemptively genotyped patients for these 6 SNPs, we can classify patients into low or high score groups and accordingly identify the candidates who are much likely going to benefit from addition of GO to their chemotherapy. So moving forward into, okay, what is next with respect to CD33 pharmacogenomics? So if you remember a couple of slides back, I showed you that presence of this SNP results in skipping of exon 2 and due to skipping of exon 2, the IgV domain is gone. GOW targets IgV domain. So do a lot of new agents that are currently under development also target IgV. But once they are in clinical trials, we will see the same thing that all the CD33-targeted therapeutics, either it is antibody-drug conjugate or BiTEs or CAR T cell therapy, they are not going to be beneficial in this group of patients. So one question that comes to your mind is, why not generate the new agents that target IgC domain instead of IgV domain? So, but one of the biggest challenges that we had in order to do that was there were no CD33 commercially available good antibodies that would target IgC domain. So we started another Going to the next step in this, we started generating IgC targeting antibodies, and we recently published this work where we have generated antibodies that targets IgC domain of CD33. So here you can see the structure of Ig— CD33, the 67.6, which is part of Goo, and the other chemotherapeutic agents that target CD33 are all targeting IgB.. And we generated these two antibodies that target IgC domain. And you can see here very clearly that blue is the BAF3 cell lines expressing D2 or shorter CD33 isoforms that express— that do not have IgV domain. And our antibodies can really— both these antibodies can detect the D2 isoform of CD33. And the similar results were seen by immunohistochemistry. Like by confocal microscopy here, you can see that only for D2 expressing BAF3 cells there was positivity with our HL2541 antibody, which targets the IgC domain of CD33. We again confirmed it with Western blotting. So, we used E6, which is the internal, which binds the IgM domain. So it should basically recognize both isoforms, and you can see that is reflected here, that full-length and D2 isoforms both are identified by E6, either with after glycosylated protein or unglycosylated protein over here. And our IgC-specific antibody also is able to recognize both full-length and D2 isoforms in westerns here, both glycosylated and unglycosylated phosphorylated proteins of CD33. So showing that the antibody that we generated is specific and it is able to recognize both full-length and D2 isoforms in the western blots. However, if when I showed you the flow data, it was on cell surface, especially in the cells that we tested, this was able to detect only CD33 D2 isoforms. We further tested a couple of AML cell lines with the inherently having different genotypes for the splicing steps, and we saw the same results. 67.6 is able to detect CD33 in cell lines that express the full-length CD33 isoforms, and our IgC-directed antibody is able to detect CD33 D2 isoform in the cell line that has TT genotype. Similar results were obtained in primary cells from AML patients, and you can see here our gating strategy. And here is 67.6, and our IgC— this is IgB targeting antibody, this is IgC targeting antibody, and you can see that our antibody is able to detect CD33 isoforms here. So while we were working on this, thinking about generating the IgC directing antibody as a next CD33 targeted agent, there was this result, the paper was published from Jannison where they had the similar idea. They went into looking into novel IgC2 binding domain of CD33 and generating a bispecific antibody. This is now in clinical trials for relapsed AML right here, their BiTE is. And so did another company called ViMaps. They also came up with a similar idea where they were targeting the IgC domain of CD33 in order to treat AML. The reason to target IgC antibody is that now you can see the benefit in patients independent of the splicing SNP genotype. And this is one of the graphical abstracts from the JENESON manuscript where you see that here on bottom panel is showing the structure of CD33 and their antibody targets IgC C2 domain just like ours. And then it's a BiTE. So you have CD33 IgC targeting, IgC domain targeting CD33 antibody as one part of the BiTE and CD3 antibody as a second part of the BiTE. And that results in cell killing of AML cells. So the goal here is that whether you have C genotype or T genotype for the splicing SNP, both group of patients should benefit if you use the CD33-targeted BiTE or CAR T cell or any other CD33-targeted agent that has antibody that targets this IgC domain but not the IgV domain. This is an example from the WyMaps where what they have done is they have generated this BC275 antibody that targets the IgC domain. And in order to generate their bispecific, specific antibody. So, in summary, what I have tried to show you is that CD33 pharmacogenomics could be used for personalisation of GO as well as for novel drug development. What our results have shown is that CD33 splicing SNP increases production of CD33 isoform that lacks exon 2 and hence the IgV domain. And the reason this is very important is because GEMPRISUMAB, ozogemisin, which is the only FDA-approved CD33 antibody-drug conjugate, as well as all the new CD33-targeted therapies that are in clinical trials, they target IgV domain. And the C to T change therefore compromises the clinical response to GO is what we have shown in our results by testing the patients patients who received GO. So in the clinical trial AML0531, our observation is that patients who have the splicing SNP T or have a score of less than 0 based on 6 CD33 SNPs, they derive no benefit from addition of GO to the standard chemotherapy. However, patients who have CC genotype for the CD33 splicing SNP, that means they will make full-length CD33 retaining the IgV domain, or if they have a score of more than 0, they show significant benefit from addition of GO to the standard chemotherapy. So the benefits— this also reflects that benefits of adding agents targeting IGP domain will be limited to patients who are homozygous for CC genotype for RS4519 or the splicing stem. So the efforts are now on in order to offer this testing to the patients before giving them CD33 targeted therapy. So we at University of Florida Path Labs are offering this test to guide the gemtuzumab treatment in AML, and so is another company, Hematologix Inc. in Seattle, is offering CD33 STIP analysis for gemtuzumab response in AML patients. So in a nutshell, what we can say here is that once the AML patients come in clinic, The patients can be categorized based on the presence or absence of the variant allele. If the variant allele is absent, that means patients have CC genotype, they could— they will be having— they will be expressing wild-type CD33 with exon 2, and they are good candidates for standard therapy with addition of GO. However, if patients have the variant allele, they will be expressing short form of CD33, and thus they will be lacking IgV domain, more these patients will be more likely not be benefiting from addition of GO. So for these patients, standard chemotherapy with other agents additions might be better options. And moving forward, targeting the IgC domain of CD33 will target both full-length and the d2 isoforms of CD33 C2 epitope, and thus may benefit patients independent of the splicing genotype. So I think this is very exciting because this is a classical example where pharmacogenomics can not only identify right patients to receive the right therapy, but can also move forward into the next step of novel drug development to benefit greater number of AML patients.