Systems level analysis of ovarian cancer

Video Transcript

• 00:00 - 00:15: Hello, everyone, and my name is Wendy Fantl, and I would like to thank you for joining
• 00:15 - 00:16: this webinar.
• 00:16 - 00:21: I’m a cancer biologist in the Department of Urology at Stanford University.
• 00:21 - 00:27: My lab applies multi-parametric single-cell proteomic technologies to study human malignancies,
• 00:27 - 00:31: and we focus on ovarian and kidney cancers.
• 00:31 - 00:38: The overarching goal of my lab is to capitalize on the single-cell attributes of these technologies
• 00:38 - 00:43: to uncover new mechanisms and cell types for clinical translation.
• 00:43 - 00:45: So now for my talk.
• 00:45 - 00:51: Systems-level analysis of ovarian cancer by CyTOF reveals natural killer cells with an
• 00:51 - 00:53: unexpected phenotype.
• 00:53 - 01:00: Outline of the talk, first, a brief background, the rationale for performing multiplex single-cell
• 01:00 - 01:05: proteomic studies, a brief background about ovarian cancer.
• 01:05 - 01:09: I will highlight one result from an earlier study.
• 01:09 - 01:15: It was the first application of CyTOF to studying ovarian tumors, and it was the identification
• 01:15 - 01:17: of tumor cell compartments.
• 01:17 - 01:23: And this is relevant for what is the main subject of this webinar, namely our study
• 01:23 - 01:30: of the tumor immune microenvironment in ovarian cancer, ovarian tumors, and the identification
• 01:30 - 01:37: of decidual-like natural killer cells that were positively correlated with tumor mass.
• 01:37 - 01:41: So multiplex single-cell technologies, why use them?
• 01:41 - 01:47: They reveal rare and subtly different cell populations that would be lost in bulk analyses.
• 01:47 - 01:55: And for cancer, these rare subpopulations are important for tumor initiation, relapse,
• 01:55 - 02:00: drug resistance, and epithelial and mesenchymal transition, all the things that make cancers
• 02:00 - 02:03: and tumors progress.
• 02:03 - 02:05: So depicted here is a tumor.
• 02:05 - 02:12: If we were to perform analysis on a bulk process sample, we would lose information about the
• 02:12 - 02:14: green cells and the pink cells.
• 02:14 - 02:18: And those are the cells we care about.
• 02:18 - 02:25: So the main proteomic technology, single-cell proteomic technology in my lab is mass cytometry,
• 02:25 - 02:32: which characterizes single intact cells that are in suspension based on their protein co-expression
• 02:32 - 02:33: patterns.
• 02:33 - 02:37: And currently, we are able to measure about 60 parameters per single cell.
• 02:37 - 02:44: It’s possible to include RNA, and it’s also possible to measure protein-protein interactions.
• 02:44 - 02:49: Although I’m not going to talk about it in this webinar, for the sake of completeness,
• 02:49 - 02:56: I’d like to mention the newly developed multiplex imaging technologies.
• 02:56 - 03:02: These maintain the deep phenotyping capabilities of CyTOF, but now they provide information
• 03:02 - 03:05: about tissue architecture.
• 03:05 - 03:10: And the idea will be that cellular neighborhoods, which are different cell types and their spatial
• 03:10 - 03:20: relationships will eventually be translated into the clinic for more reliable biomarkers.
• 03:20 - 03:25: So mass cytometry or cytometry by time of flight.
• 03:25 - 03:31: This single-cell multiparametric proteomic technology uses antibodies, which are tagged
• 03:31 - 03:34: with chelated metal polymers.
• 03:34 - 03:40: And these metals are from the lanthanide portion of the periodic table, and they are rare or
• 03:40 - 03:42: absent in biology.
• 03:42 - 03:47: And tagging an antibody with such a metal allows the readout to be the very sensitive
• 03:47 - 03:50: mass time of flight mass spectrometry.
• 03:50 - 03:55: And that’s in contrast to what you’re all familiar with, traditional fluorescence-based
• 03:55 - 03:57: flow cytometry.
• 03:57 - 04:04: So cells in suspension, single intact cells in suspension, are labeled with cocktails
• 04:04 - 04:07: of metal-tagged antibodies.
• 04:07 - 04:14: These antibodies recognize surface epitopes that delineate specific cell phenotypes, as
• 04:14 - 04:18: well as intracellular and nuclear proteins.
• 04:18 - 04:25: So after labeling single cells in suspension, they are introduced into the CyTOF machine.
• 04:25 - 04:30: That generates large CyTOF data sets, and as you can imagine, hundreds of thousands,
• 04:30 - 04:35: if not millions of cells in 60-parameter space.
• 04:35 - 04:39: And this spawned the development of many methods, many analytical methods.
• 04:39 - 04:44: And you’ll see some of them as I walk through this webinar.
• 04:44 - 04:49: So now a brief background to epithelial ovarian cancer.
• 04:49 - 04:56: It is the most lethal gynecologic malignancy, tubo-ovarian HGSC, which is high-grade serous
• 04:56 - 04:57: carcinoma.
• 04:57 - 05:03: And I’d just like to say through this webinar, I’ll be saying HGSC cells or ovarian tumor
• 05:03 - 05:08: cells, but I am referring to tubo-ovarian high-grade serous carcinoma.
• 05:08 - 05:10: It’s the most prevalent histotype.
• 05:10 - 05:15: It’s about 80% of epithelial ovarian cancers.
• 05:15 - 05:21: Most women arrive at the clinic with late-stage disease, and that’s because early-stage disease
• 05:21 - 05:23: is asymptomatic.
• 05:23 - 05:27: That is due to a lack of effective screening tests.
• 05:27 - 05:34: The disease is also very heterogeneous, so that does not bode well.
• 05:34 - 05:41: Women have been treated for the past few decades with exactly the same platinum-based chemotherapy.
• 05:41 - 05:48: At first, most patients will respond, but eventually most of them will relapse.
• 05:48 - 05:55: So the late-stage diagnosis combined with the extreme heterogeneity means that the five-year
• 05:55 - 06:00: survival rate is dismal, 25% to 40%.
• 06:00 - 06:04: This malignancy is in urgent need of new therapeutic approaches.
• 06:04 - 06:11: And indeed, the PARP inhibitors have changed the clinical management of ovarian cancer,
• 06:11 - 06:15: but only for a subgroup of patients.
• 06:15 - 06:20: And immunotherapy, while paradigm-changing for other malignancies, has had very little
• 06:20 - 06:27: clinical, if any, clinical benefit for women with ovarian cancer.
• 06:27 - 06:35: So the incidence of the disease in the U.S. is 21,000, and worldwide, 230,000, and there
• 06:35 - 06:38: are a disproportionate number of deaths.
• 06:38 - 06:45: The risk in the general population is less than 2%, but women with hereditary mutations
• 06:45 - 06:52: in BRCA1 or BRCA2, which are genes involved in repairing DNA damage, have an increased
• 06:52 - 06:58: risk of getting ovarian cancer, as you can see here.
• 06:58 - 06:59: Very different.
• 06:59 - 07:05: And one approach for these women who know they are at increased risk is to have a prophylactic
• 07:05 - 07:12: salpingo-oophorectomy, removal of the ovarian and fallopian tubes, but that, although that significantly
• 07:12 - 07:20: reduces the risk, it is accompanied by comorbidities, other comorbidities.
• 07:20 - 07:24: So now to the next part of the talk, where I’m going to highlight one result from an
• 07:24 - 07:31: earlier site of, from our first site of analysis of ovarian tumors, where we focused on the
• 07:31 - 07:33: tumor cells themselves.
• 07:33 - 07:38: So this is published in Cell Reports in 2018.
• 07:38 - 07:45: So we acquired 17 newly diagnosed chemo-naive tumors, all patients received platinum-based
• 07:45 - 07:46: chemotherapy.
• 07:46 - 07:53: The samples were processed into single-cell suspensions based on our established protocols.
• 07:53 - 08:00: We performed a CyTOF analysis, we designed and optimized three antibody panels, one against
• 08:01 - 08:05: the tumor cells, which I’m going to describe in a minute, and then two panels against the
• 08:05 - 08:12: immune, infiltrating immune cells, and that will be discussed later.
• 08:12 - 08:17: And then a variety of computational approaches were used to analyze the data and relate it to
• 08:17 - 08:22: patient outcomes and identification of new targets, and that actually information is
• 08:22 - 08:25: in that first paper.
• 08:25 - 08:32: So here are the antibodies that we validated, and on the right-hand side are the immune
• 08:32 - 08:38: antibody panels, one against B cells, macrophages, and dendritic cells that I won’t talk about,
• 08:38 - 08:44: but then the T cell and NK cell panel, which will be discussed later.
• 08:44 - 08:50: And now on the left-hand side is the panel against the ovarian tumor cells.
• 08:50 - 08:53: I’m going to give you a quick tour.
• 08:53 - 09:01: These were antibodies against ovarian tumor-associated antigens, namely HE4, mesothelin, and MUC16,
• 09:01 - 09:09: and then E-cadherin, a very important adhesion protein in epithelial cancers, and vimentin,
• 09:09 - 09:15: which is important in metastatic, when a cancer progresses to a metastatic phenotype.
• 09:15 - 09:22: And these two proteins are usually mutually exclusive, so the
• 09:22 - 09:28: loss of E-cadherin as a tumor progresses to a more metastatic state is accompanied by
• 09:28 - 09:30: an increase in vimentin.
• 09:30 - 09:36: Then CD151, a tetraspanin involved in cell adhesion, exopeptidases, stem cell markers,
• 09:36 - 09:42: then we can go inside the cell and we can look at proteins involved in the DNA damage
• 09:42 - 09:50: response, apoptosis, pleiotropic transcription factors, cell survival, proliferation, the
• 09:50 - 09:51: cell cycle, etc.
• 09:51 - 09:58: And importantly, the panel includes antibodies against fibroblast-activated protein for our
• 09:58 - 10:06: gating strategy, where we can gate out stromal cells, CD31 to delineate angiogenic cells,
• 10:06 - 10:10: and CD45 that will delineate immune cells.
• 10:10 - 10:18: And I’ll just give you an example to show you the strategy for us gating out these compartments
• 10:18 - 10:23: and allowing us to now fully focus on a tumor-enriched compartment.
• 10:23 - 10:30: So here we gate out the CD45 cells, we have this population here, and we gate out the
• 10:30 - 10:37: angiogenic cells and the stromal cells, and again, the tumor-enriched compartment.
• 10:38 - 10:44: And now, as I mentioned, we have a large CyTOF dataset, and obviously I don’t have time in
• 10:44 - 10:50: this webinar to explain and share with you all the different methods of analyses, but
• 10:50 - 10:53: overall, here are some general principles.
• 10:53 - 11:00: We combine the single-cell datasets from all the tumors, we select proteins, usually these
• 11:00 - 11:06: are surface marker proteins for clustering, and we find groups of similar proteins in
• 11:06 - 11:11: high-dimensional space based on their protein co-expression patterns.
• 11:12 - 11:18: Each cluster, or each phenotype, is then connected to its most similar cluster on a
• 11:18 - 11:20: minimum spanning tree.
• 11:21 - 11:26: And the size of the bubbles denotes the frequency of cells.
• 11:26 - 11:31: So if you’ve got a large bubble, you have a high number or high frequency of
• cells with
• 11:31 - 11:36: a particular co-expression, with a similar protein co-expression pattern.
• 11:37 - 11:46: In contrast, a small bubble denotes a lower frequency of cells with a specific, with a
• 11:46 - 11:49: particular protein co-expression pattern.
• 11:49 - 11:54: And color is usually biology, where blue is background, and any color that’s not blue
• 11:54 - 11:58: is above background, and red is high expression.
• 11:58 - 12:05: And as I mentioned earlier, ovarian tumors have a very complex molecular and genetic
• 12:05 - 12:06: heterogeneity.
• 12:06 - 12:14: Yet even with this complex genetics, there were, by CyTOF, a surprisingly limited
• 12:14 - 12:15: set of phenotypes.
• 12:15 - 12:19: And our hope is that this is a potential road toward precision medicine.
• 12:21 - 12:27: So the piece of data that I’m going to show you now that will be relevant for the tumor
• 12:27 - 12:35: immune part of this talk is that we generated a composite minimum spanning tree of all 17
• 12:35 - 12:37: tumor samples.
• 12:37 - 12:44: And this composite tree showed the cell clusters and the branches of the ovarian tumor cell
• 12:44 - 12:45: phenotypes.
• 12:45 - 12:52: And I’m showing you here the staining pattern for E-cadherin here, which is very common and is
• 12:52 - 12:57: a hallmark of epithelial tumors as well, including ovarian cancer.
• 12:58 - 13:01: And here we see the staining pattern for vimentin.
• 13:01 - 13:07: And remember I told you that E-cadherin here and vimentin, this is well established in epithelial
• 13:07 - 13:10: tumor biology, are mutually exclusive.
• 13:10 - 13:16: And it’s nice to see in this unsupervised analysis where we are making new discoveries
• 13:16 - 13:21: to see information that is already well established.
• 13:21 - 13:29: And what was interesting here was we could actually see by these cell clusters shown by
• 13:29 - 13:34: this boundary that they co-expressed E-cadherin and vimentin.
• 13:34 - 13:40: And presumably these cells are undergoing epithelial-mesenchymal transition.
• 13:40 - 13:49: So based on their stages of progression, we discovered different tumor cell compartments,
• 13:49 - 13:56: an E compartment, which is epithelial, a transitional epithelial-mesenchymal compartment,
• 13:56 - 14:03: EV, and a mesenchymal metastatic compartment based on the expression patterns of E-cadherin
• 14:03 - 14:04: and vimentin.
• 14:04 - 14:10: And these compartments, remember them, E, EV and E, will be relevant for the later part
• 14:10 - 14:10: of this talk.
• 14:11 - 14:18: And this is the later part of the talk, looking at the tumor immune microenvironment, where
• 14:18 - 14:23: we identified a decidual-like natural killer cell that was positively correlated with the
• 14:23 - 14:24: tumor cell mass.
• 14:25 - 14:30: So immunotherapy is a paradigm shift for treating cancer patients.
• 14:30 - 14:35: It harnesses the host immune system to attack and eradicate the tumor.
• 14:35 - 14:39: And it’s been very successful in a variety of malignancies such as kidney cancer,
• 14:40 - 14:45: bladder cancer, and lung cancer, and melanoma.
• 14:45 - 14:53: However, it has had very little clinical impact for women with ovarian cancer.
• 14:53 - 14:58: And that is despite the presence of exhausted T cells, in other words, T cells expressing
• 14:58 - 15:04: PD1 and CTLA4, immune checkpoint blockade has been disappointing for these patients.
• 15:05 - 15:13: We therefore ask the question whether alternate immune cell types could override any reversal
• 15:14 - 15:17: of the T cell-mediated immune suppression.
• 15:19 - 15:23: So in our earlier study, we characterized the tumor cells.
• 15:23 - 15:27: We identified these three tumor cell compartments.
• 15:27 - 15:34: And in the next study, we are now characterizing by CyTOF the tumor immune infiltrate.
• 15:34 - 15:38: And both these studies are now published.
• 15:39 - 15:48: So the immune panel that I’m going to focus on is the immune panel that interrogates T
• 15:48 - 15:50: cells and NK cells.
• 15:51 - 15:58: And the initial analysis for the immune infiltrate was to cluster the cells with the same algorithms
• 15:58 - 16:00: that we had used for the tumor cells.
• 16:00 - 16:03: So now we have two CyTOF data sets.
• 16:03 - 16:12: We have a CyTOF data set that interrogates the tumor cells and a CyTOF data set that
• 16:12 - 16:16: interrogates the T and NK cells from the same tumors.
• 16:17 - 16:24: And I think what we are really interested in is how do these tumor cells, how do these
• 16:24 - 16:27: cell types interact with one another?
• 16:28 - 16:33: And one way to analyze these data is with a network approach where we can measure correlations
• 16:33 - 16:36: to identify coordinated modules.
• 16:36 - 16:42: So we asked whether there were correlations between cell frequencies of the immune cell
• 16:42 - 16:44: phenotypes with the tumor cell phenotypes.
• 16:45 - 16:51: So we generated a heat map showing the pairwise correlations of the frequencies between all
• 16:51 - 16:57: the cell clusters, the 56 tumor and 52 T and NK cell clusters.
• 16:58 - 17:03: And thinking about how we were going to move forward, although this information would be
• 17:03 - 17:09: very useful, what we thought would be more useful would be to start by looking at
• 17:09 - 17:17: the frequencies of specific immune cell types that would correlate with total tumor and
• 17:17 - 17:18: EV cell frequencies.
• 17:18 - 17:25: Do these, is there a positive correlation with these immune cell phenotypes that correlate
• 17:25 - 17:27: with the tumor cell mass and EV?
• 17:27 - 17:30: So that is a growing and progressing tumor.
• 17:30 - 17:33: That seemed to be the first question to address.
• 17:34 - 17:35: So here is the heat map.
• 17:35 - 17:41: And you can see in the bottom left-hand corner, there’s this red block.
• 17:41 - 17:49: And these are Spearman correlation values of greater than 0.5 for correlates between
• 17:49 - 17:55: immune clusters and with the frequency of tumor cells and the frequency of the EV cells,
• 17:55 - 17:57: as I just mentioned.
• 17:57 - 18:04: And so now if we zoom in, we can indeed see what these clusters are.
• 18:04 - 18:10: And these clusters that are shown by these symbols turn out to be natural killer cells
• 18:10 - 18:16: that correlated positively, as shown by the red, with the tumor and EV cell frequencies.
• 18:17 - 18:21: And these NK cells have a decidual-like phenotype.
• 18:21 - 18:29: So in contrast to what is usually the function of NK cells, namely killing aberrant cells
• 18:29 - 18:35: such as virally infected cells and tumor cells that are floating around, because they have
• 18:35 - 18:40: a decidual-like phenotype, they were likely immune tolerant.
• 18:41 - 18:43: And why are we making this assignment?
• 18:43 - 18:47: Well, there are many phenotypes of NK cells.
• 18:47 - 18:53: But decidual NK cells, which were important in pregnancy, have the exclusive expression
• 18:53 - 18:55: of the tetraspanin CD9.
• 18:56 - 19:06: They also express high levels of the chemokine receptor CXCR3, CD56, and the KIR, the killer
• 19:06 - 19:11: immunoglobulin-like receptors that are inhibitory NK receptors.
• 19:12 - 19:19: And these phenotypes that we identified were negative for the T cell marker CD3, so they’re
• 19:19 - 19:25: not T cells, and also negative for CD16, which is another hallmark of decidual NK cells.
• 19:26 - 19:32: So decidual NK cells, well, as I mentioned, NK cells are lymphocytes of the innate immune
• 19:32 - 19:36: response, and they kill aberrant cells that are floating around.
• 19:36 - 19:42: But decidual NK cells provide maternal-fetal immune tolerance and facilitate placental
• 19:42 - 19:43: growth.
• 19:43 - 19:50: And in the first trimester of pregnancy, about 70% of the lymphocytes are, in fact, decidual
• 19:50 - 19:51: NK cells.
• 19:52 - 19:57: And if you think about it, there can be no stronger connection in biology than a mother
• 19:57 - 19:58: tolerating her fetus.
• 19:59 - 20:06: We therefore hypothesize that ovarian tumors co-opt this mechanism for their survival.
• 20:07 - 20:10: There are reports of decidual NK cells in other malignancies.
• 20:10 - 20:17: This is a very nice review earlier this year by Albini and Noonan, published in Cancer
• 20:17 - 20:17: Discovery.
• 20:18 - 20:25: So we hypothesize that, in fact, it was the ovarian tumor cells that endowed the natural
• 20:25 - 20:30: killer cells with their decidual-like immunosuppressive properties.
• 20:30 - 20:33: We went about trying to understand and prove this.
• 20:34 - 20:39: And one way this could happen would be by the tumor cells manifesting different expression
• 20:39 - 20:46: patterns of ligands for the NK-activating and inhibitory receptors.
• 20:46 - 20:54: So we designed and validated and actually modified our CyTOF antibody panel that was
• 20:54 - 20:57: generated to interrogate the tumor cells.
• 20:58 - 21:03: We kept the surface markers that define the tumor cells.
• 21:03 - 21:10: But now we added in antibodies against various NK receptor ligands, 12 in total.
• 21:10 - 21:16: Ligands, activating ligands for the NKG2D activating receptor.
• 21:16 - 21:23: So these activating and inhibitory receptors are present on NK cells, and the ligands are
• 21:23 - 21:24: present on the tumor cells.
• 21:24 - 21:30: And they can therefore orchestrate the behavior of the NK cells.
• 21:30 - 21:34: And the NKG2D ligands would be the ULBP family, MycA, MycB.
• 21:35 - 21:44: The ADAM10 and 17 proteases are important in regulating NK receptor and NK ligand function.
• 21:44 - 21:48: They have a protease activity that sheds extracellular domains.
• 21:49 - 21:56: We interrogated or measured levels of expression of the inhibitory ligands and ligands to DNAM1,
• 21:56 - 21:58: the activating NK receptor.
• 21:59 - 22:01: It’s also called CD226.
• 22:01 - 22:10: The inhibitory NK receptor CD96, and another important inhibitory receptor called TIGIT.
• 22:10 - 22:15: And these are the ligands, the nectin-like ligands that bind to these receptors.
• 22:17 - 22:20: So we analyzed 12 separate ovarian tumors.
• 22:21 - 22:27: We clustered the CyTOF data sets of the tumor cells with the tumor cell markers that we
• 22:27 - 22:28: had used in the first study.
• 22:29 - 22:35: And we examined the expression of the 12 NK receptor ligands and the two ADAM proteases.
• 22:35 - 22:42: And we visualized those expression patterns, those NK receptor ligands with force-directed layouts.
• 22:45 - 22:52: So when we look at the force-directed layouts, we recapitulate the E, EV, and V compartments by the
• 22:53 - 22:58: mutually exclusive and co-expression patterns of E-cadherin and vimentin.
• 22:59 - 23:04: And just giving you one example, looking at the expression patterns of the NKG2D ligands
• 23:04 - 23:11: and the proteases, you can see that the way they are expressed and the description I would use
• 23:11 - 23:12: would be very variable.
• 23:13 - 23:15: And we want to know more than that.
• 23:15 - 23:16: We need to quantify.
• 23:16 - 23:18: We need somehow to quantify that.
• 23:18 - 23:24: And so we turned to Boolean gating, which allowed us to look at different
• 23:24 - 23:27: combinatorial expression patterns.
• 23:27 - 23:32: And so this approach has been used for the NK receptor, for the NK receptors.
• 23:32 - 23:37: And we used it here to look at the NK receptor ligands.
• 23:37 - 23:40: So each row is a combination.
• 23:40 - 23:47: So this first row, it’s the combination of HLA-BC with HLA, with HLA-E.
• 23:47 - 23:54: And then we measured the frequency of cells with these combinations in the E, EV, and
• 23:54 - 23:57: vimentin compartments of the 12 tumors.
• 23:58 - 24:05: And by eye, you can see that there are actually fewer combinations in the vimentin mesenchymal
• 24:05 - 24:09: or metastatic portion of the tumors.
• 24:10 - 24:18: And also, it turns out that the combinations involve ligands, more ligands for the inhibitory
• 24:18 - 24:20: NK receptors.
• 24:20 - 24:25: So you can already see that we’re building a picture of more stringent immune tolerance
• 24:26 - 24:29: in the metastatic portion of the cells.
• 24:30 - 24:34: The numbers here on the Venn diagram show what I just showed you, what our eye sees,
• 24:34 - 24:41: which is more combinations, 103 and 101, in the E-cadherin and the E-cadherin-vimentin
• 24:41 - 24:46: compartments, and fewer combinations, 67, in the metastatic vimentin compartment.
• 24:47 - 24:54: And to summarize, we see different immune microenvironments within the E, EV, and V
• 24:54 - 24:59: compartments, increased ligand combinations in the E and EV compartments.
• 24:59 - 25:04: One possible explanation is to increase the likelihood of escaping immune detection.
• 25:05 - 25:12: And V cells have more stringent immune escape mechanisms, as most likely is expected.
• 25:13 - 25:17: So how do ovarian tumor cells modulate the function?
• 25:18 - 25:23: And can we have, are there, what are the functional readouts that we can use to see this?
• 25:24 - 25:31: For these experiments, we performed co-culture, we co-cultured ovarian tumor cells with NK
• 25:31 - 25:32: cells.
• 25:32 - 25:39: For the ovarian tumor cells, we found cells that modeled the E, EV, and V cells.
• 25:39 - 25:46: And these cells were selected from a seminal publication by Domcke et al, where they ranked
• 25:47 - 25:54: HGSC ovarian tumor cell lines based on their molecular and genetic similarity to tumors
• 25:54 - 25:57: that had been deposited in the TCGA database.
• 25:58 - 26:04: We then analyzed those cells, or characterized those cells by CyTOF, and we actually identified
• 26:05 - 26:12: cell lines that mirrored E, EV, and V, and not only in terms of those compartments, but
• 26:12 - 26:18: also in terms of their expression levels of the NK receptor ligands that I had shown you
• 26:19 - 26:22: from our characterization of the tumors.
• 26:22 - 26:29: And so we used for the E compartment, OVCAR4, the OVCAR4 cell line, for the EV compartment,
• 26:29 - 26:34: the Koromochi cell line, and for the V compartment, the TCN1 cell.
• 26:35 - 26:42: We set up the co-cultures with NK92 cells, and our readouts were decidual-like features,
• 26:42 - 26:45: cytokine production, and cytotoxicity.
• 26:46 - 26:51: And we chose the NK92 cell because of its clinical relevance.
• 26:51 - 26:55: It’s used in a variety of engineered forms.
• 26:56 - 27:03: And the first co-culture readout we looked at was CD9, the CD9 expression.
• 27:03 - 27:10: As you recall, CD9 is exclusively expressed in decidual NK cells.
• 27:10 - 27:15: And NK cells in monoculture have very background levels of CD9.
• 27:15 - 27:22: You see background levels of CD9, but upon co-incubation with the ovarian tumor cells,
• 27:22 - 27:27: there is a dramatic increase in the frequency of cells that express CD9,
• 27:27 - 27:30: a decidual-like NK cell feature.
• 27:30 - 27:35: And if we performed this experiment in the presence of a trans well,
• 27:36 - 27:41: you really abolish that CD9 uptake.
• 27:41 - 27:46: And this experiment indicates that you need cell-cell contact.
• 27:46 - 27:52: We ruled out exosomes because the pore size would have allowed exosomes to pass through.
• 27:53 - 27:56: So what is the mechanism for CD9 uptake?
• 27:57 - 28:04: We reasoned that perhaps the cell-cell contact somehow activated intracellular pathways,
• 28:04 - 28:07: resulted in some kind of epigenetic modification,
• 28:07 - 28:11: and maybe we could detect this by the use of small molecule inhibitors.
• 28:11 - 28:13: However, that was not the case.
• 28:13 - 28:21: What we did notice was the exceptionally high levels of CD9 in the ovarian tumor cell lines.
• 28:21 - 28:25: And this was not just for the OVCAR4 cell, I’m showing you an example,
• 28:25 - 28:29: but in pretty much all of the ovarian tumor cell lines.
• 28:29 - 28:34: And the NK92 cell was devoid of CD9 expression.
• 28:35 - 28:41: So our hypothesis was that CD9 is acquired from the HGSC tumor cells
• 28:41 - 28:44: by a process called trogocytosis.
• 28:44 - 28:47: What is trogocytosis?
• 28:47 - 28:52: Trogocytosis is one of many intercellular transport mechanisms.
• 28:52 - 28:56: You’re all familiar with endocytotic transport mechanisms,
• 28:56 - 29:00: phagocytosis bringing in solid particles into a cell,
• 29:01 - 29:05: pinocytosis bringing in liquid into a cell,
• 29:05 - 29:12: and other means of bringing substances into a cell through receptor-mediated endocytosis.
• 29:13 - 29:16: Trogocytosis, the word derives from Greek,
• 29:16 - 29:21: trogo meaning “to gnaw,” cytosis, cellular transport.
• 29:21 - 29:25: And it’s defined as the rapid, and it happens within minutes,
• 29:25 - 29:28: intracellular transfer of membrane fragments
• 29:28 - 29:32: and their associated molecules during intercellular contact.
• 29:32 - 29:35: And this acquisition of the associated molecules,
• 29:35 - 29:39: which those recipient cells didn’t have, now endows them with a new function.
• 29:40 - 29:43: And trogocytosis, there’s a vast literature,
• 29:43 - 29:47: and it happens very frequently and modulates immune responses.
• 29:47 - 29:51: And just a couple of examples, published examples,
• 29:51 - 29:54: for trogocytosis occurring in NK cells,
• 29:54 - 30:00: they acquire HLA-G and NKG2D ligands from tumor cells.
• 30:00 - 30:03: These are two nice studies published by EMBO a while ago.
• 30:05 - 30:09: And then just here, we see that the HGSC is the donor cell,
• 30:09 - 30:14: the NK92 is the accepting cell, it acquires CD9.
• 30:15 - 30:18: This was a hypothesis, but in order to definitively show
• 30:18 - 30:22: that the mechanism transfer was by trogocytosis,
• 30:22 - 30:29: it’s necessary to perform multiple experiments to really show this is indeed so.
• 30:29 - 30:31: And I’m just going to list them, but I’ll show you,
• 30:31 - 30:32: and I’ll show you in the interest of time,
• 30:32 - 30:35: I’ll just show you the results from one of these.
• 30:35 - 30:39: So quantitative real-time PCR showed that there was no,
• 30:39 - 30:43: no, there were no transcripts for CD9 before or after co-culture.
• 30:44 - 30:49: If we treated co-cultures with inhibitors of actin polymerization,
• 30:49 - 30:54: which are used to, which do, are known to inhibit trogocytosis,
• 30:54 - 31:00: with cytochalasins D, we observed a greater than 60% inhibition of CD9 transfer.
• 31:01 - 31:05: Using PKH67, a green fluorescent labeled
• 31:06 - 31:13: OVCAR4 cell, co-incubating these fluorescently labeled cells with NK92 cells,
• 31:13 - 31:19: we could see now the acquisition of the green onto the NK cells.
• 31:19 - 31:23: And finally, I’ll show you the results from the microscopy experiment.
• 31:24 - 31:31: If we have NK92 cells, we label them with the red fluorescent dye, PKH26,
• 31:32 - 31:35: CD45, because they’re immune cells, and here’s the overlay.
• 31:36 - 31:42: And for the ovarian tumor cells of OVCAR4 here, CD9, shown by blue,
• 31:43 - 31:49: they’re labeled with a green fluorescent dye, PKH67, and here’s the overlay.
• 31:49 - 31:53: And now with the merge, if you follow the white arrows,
• 31:53 - 31:57: these white arrows show that there are cells that are co-expressing
• 31:58 - 32:03: CD9, they’re green, they’re red, and they’re white.
• 32:03 - 32:09: And these are the cells, these are actually the NK cells that have undergone trogocytosis,
• 32:09 - 32:12: and they’ve acquired that membrane fragment from the ovarian tumor cells.
• 32:13 - 32:19: So I’ve told you that ovarian tumor cells are a rich source of CD9,
• 32:19 - 32:22: but what we really want to know, what about primary tumors?
• 32:22 - 32:28: So we actually measured the frequency of cells expressing CD9 and the median expression levels
• 32:28 - 32:34: of CD9 in primary, newly diagnosed ovarian tumors.
• 32:34 - 32:40: And as you can see, this protein is expressed by most of the tumor cells,
• 32:40 - 32:43: and there are variable but high levels of this protein.
• 32:43 - 32:51: So the tumor cells are a source of CD9 that could potentially, in vivo,
• 32:51 - 32:54: be a source for the infiltrating NK cells.
• 32:55 - 33:02: And trogocytosis not only occurs in NK92 cells, but in NKL cells.
• 33:02 - 33:10: And here I’m showing you trogocytosis by primary NK cells in peripheral blood mononuclear cells.
• 33:10 - 33:18: So let’s focus here on the NK CD56 bright, and you can see two replicates,
• 33:18 - 33:21: and you can see the frequency, low level of frequency.
• 33:21 - 33:27: And after co-incubation with the ovarian tumor cell line, this increases dramatically.
• 33:27 - 33:32: And the CD56 dim, which is the majority population in peripheral blood mononuclear cells,
• 33:32 - 33:36: also undergoes trogocytosis of CD9.
• 33:36 - 33:41: We showed this, but seemingly to a lesser extent.
• 33:42 - 33:48: So having shown that NK cells can acquire CD9 from tumor cells,
• 33:48 - 33:55: does this have any relevance for the function of these NK cells that have acquired CD9?
• 33:55 - 34:02: So we hypothesize that CD9 endows NK92 cells with immunosuppressive properties.
• 34:02 - 34:06: And to do this, we set up the co-culture experiment,
• 34:06 - 34:13: and we can gate out CD9 plus and CD9 minus NK92 cells.
• 34:13 - 34:16: And here we are measuring cytokine production,
• 34:16 - 34:21: actually specifically the production of anti-tumor cytokines
• 34:21 - 34:25: and cytotoxicity. NK cells kill aberrant cell types.
• 34:26 - 34:30: And so showing you here, the intracellular cytokine production
• 34:31 - 34:36: of interferon gamma, TNF alpha, and GM-CSF.
• 34:36 - 34:43: And what I’m showing you here, each of the columns shows you the frequency of cells expressing these,
• 34:44 - 34:48: producing these cytokines in CD9 plus with the light bars,
• 34:48 - 34:53: and the first in these couples, and CD9 minus.
• 34:53 - 34:57: And what you can see is that if we take GM-CSF, for example,
• 34:57 - 35:04: there is definitely a lower frequency of cells expressing GM-CSF in the CD9
• plus
• 35:04 - 35:09: versus the CD9 minus cells. And if we now look at expression levels,
• 35:10 - 35:14: the same holds true. So there are lower levels of expression,
• 35:14 - 35:22: again, just showing you the example, GM-CSF, lower level in CD9 plus than CD9 minus cells.
• 35:22 - 35:30: And these actually are the tumor cell lines that represent the different tumor cell compartments.
• 35:30 - 35:37: So it seems to hold this reduction in CD9 plus cells seems to be consistent,
• 35:38 - 35:46: dependent, regardless of which tumor cell line, in this case, we are using for the co-culture.
• 35:48 - 35:54: So as I keep saying, one function well known for NK cells is their killing activity.
• 35:54 - 35:59: What you can see here in this cytotoxicity assay, it’s actually a policy release assay,
• 35:59 - 36:08: that with varying increasing target effector ratio, the killing activity of NK92 cells toward
• 36:08 - 36:14: the different ovarian tumor cell lines is very suppressed when it’s compared to the positive
• 36:14 - 36:22: control, the gold standard of NK cells killing the erythroleukemia cell line, the K562 cell line.
• 36:23 - 36:32: If indeed, we wanted to determine whether CD9 was responsible for this attenuation of NK cell
• 36:32 - 36:39: mediated cytotoxicity. So we performed three independent experiments and I’m showing,
• 36:39 - 36:44: I’ll tell you what they are, but only show you the data from one. So after co-culture,
• 36:45 - 36:54: we fact-sorted CD9 plus and CD9 minus NK cells and showed that NK plus NK cells
• 36:55 - 37:02: had very attenuated killing activity compared to their CD9 minus counterpart.
• 37:03 - 37:10: We also performed a CRISPR, a CD9 CRISPR knockout. So when we performed a CRISPR9 CD9 knockout
• 37:10 - 37:18: in the ovarian tumor cells and then performed the co-culture, there was an increase in cytotoxicity
• 37:18 - 37:26: by the NK cells toward those tumor cells where the CD9 knockout, and I should tell you it was
• 37:26 - 37:34: about 97%. And that increase was about a 50% increase in cell killing. And the data I will
• 37:34 - 37:40: show you is with a CD9 blocking antibody in the co-culture that increased the cytotoxicity
• 37:40 - 37:47: of NK cells towards the ovarian tumor cells, like the CRISPR by greater than 50%.
• 37:48 - 37:56: And here are the data. And here again, you see the suppressed cytotoxicity in the co-culture
• 37:56 - 38:02: NK cells with, in this case of OVCAR4 cells. And that’s the case with a one-to-one
• 38:04 - 38:11: target effector ratio and a one-to-five. And now if we co-culture with one microgram per
• 38:11 - 38:20: mL of the CD9 blocking antibody, you can see this great enhancement in NK killing activity.
• 38:20 - 38:26: And it remains to be true. It’s true for two micrograms per mL, and it’s also true for this
• 38:27 - 38:34: one-to-five target effector ratio. To summarize the data from this study,
• 38:34 - 38:41: we identified an intratumoral decidual-like NK cell that was positively correlated with the tumor
• 38:41 - 38:49: and EV cell frequency abundance. And very lightly, these decidual-like NK cells provide
• 38:49 - 38:56: immune tolerance for the success of the tumor. And this is reminiscent of immune tolerance,
• 38:56 - 39:02: mother’s immune tolerance to her fetus. It is the tumor cells, the ovarian tumor cells,
• 39:02 - 39:08: that manipulate the NK cells, creating an immune suppressive, creating immune suppressive
• 39:08 - 39:14: microenvironments. This is lightly accomplished by different expression patterns of NK receptor
• 39:14 - 39:23: ligands, as well as ovarian tumor cells being a very rich source of CD9 for NK cell-mediated
• 39:23 - 39:33: trogocytosis. And the acquisition of CD9 by NK cells endows NK92 cells with immunosuppressive
• 39:33 - 39:39: functions. I want to emphasize, of course, these are in vitro experiments. And the CD9 blocking
• 39:39 - 39:47: antibody and CD9 CRISPR knockout reverse this immunosuppression. And what is the theme of my lab,
• 39:47 - 39:57: which is translational potential, get the results into the clinic ASAP. CD9 blockade as a therapeutic
• 39:57 - 40:05: target, and again, many possible avenues for translation. Such a blockade could sustain
• 40:05 - 40:14: the efficacy for NK immunotherapy and prevent CD9 uptake. A CD9 antibody could reactivate
• 40:14 - 40:23: hyporesponsive intratumoral decidual-like NK cells. And a CD9 antibody could actually attenuate
• 40:24 - 40:32: ovarian tumor cell growth. This needs all to be investigated. And in the diagnostic space,
• 40:32 - 40:39: the NK receptor ligand expression could be a companion diagnostic for NK cell immunotherapy.
• 40:39 - 40:47: For example, tumors with more NK, expressing more NK inhibitory ligands may not respond well
• 40:47 - 40:54: to NK cells that have been engineered to express an NK activating receptor such as NKG2D.
• 40:55 - 41:02: And also in data that I did not show you, both in vitro and in a small cohort of patient samples,
• 41:02 - 41:10: we showed that ovarian tumors after carboplatin therapy had increased expression of the
• 41:10 - 41:18: inhibitory ligands. So usually pretreated patients, heavily pretreated patients would be the subject
• 41:18 - 41:26: for NK immunotherapy. But post-carboplatin, one may need to rethink that strategy.
• 41:26 - 41:34: The frequency of decidual-like NK cells intratumorally may be a biomarker for response to
• 41:34 - 41:42: T and NK cell immunotherapy. For example, if a tumor has low frequencies of these decidual NK
• 41:42 - 41:48: cells, that may predict a response to T cell immune checkpoint blockade.
• 41:49 - 41:53: The final and most important slide are the acknowledgements of the people in my lab,
• 41:54 - 42:03: Ying, Kenyi, Antonio, Jacob, Veronica, and Alexis. Great support from my department of urology,
• 42:04 - 42:11: fantastic Stanford collaborators, particularly Ermelinda Porpiglia in the Blau lab, Angelica,
• 42:11 - 42:18: Shih-Yu, and Nikolay from Gary Nolan’s lab, at the Stanford Cancer Institute Tissue Bank,
• 42:18 - 42:26: Brooke Howitt, an ovarian tumor pathologist, and many other collaborators, and of course,
• 42:26 - 42:37: my funding. And I’d like to thank you for listening to this webinar.