Hello, my name is Maliha Ghazi and I'm a third year medical student at the University of Toronto and recent PhD graduate from the laboratory of Dr. Sheila Singh. Today I will be sharing with you our work in developing a model for glioblastoma recurrence and using it to characterize the minimal residual disease state. I want to begin by thanking members of the Singh lab and Bok lab at McMaster University, members of Moffitt lab, Kisslinger lab, Pew lab and Goyal lab at the University of Toronto and Dr. Ali Reza Mansouri at Penn State for their contributions to this work. I also want to acknowledge all the funding sources for this support. And most importantly, I want to thank our patients and their families. Glioblastoma is the most malignant primary brain tumor and despite aggressive therapy, all patients experience disease relapse. Whether glioblastoma recurs due to pre-existing therapy resistant populations or therapy driven events is not clear. And this is due to lack of recurrent GPM samples for comprehensive analysis, as well as the lack of clinically valid models of glioblastoma recurrence. We developed a therapy-adapted, patient-derived xenograft model to track tumor cells from treatment-naive human GBMs. Once human GBM cells were engrafted in immune-compromised mouse brains, we monitored the tumor growth using MRI and began in vivo chemotherapy with temozolomide and radiation upon visible engraftment. Approximately two weeks post-therapy, we observed improved clinical symptoms in treated mice, coupled with minimal radiographic evidence of disease, as is frequently observed in patients. We termed this state the Minimal Residual Disease, MRD, a clinically and biologically relevant stage of disease comprising the treatment-resistant pool of cells that comprise the imminent recurrence leading to disease endpoint. While treatment did lead to improved overall survival, phenotypic assessment of tumor cells collected at endpoint showed increased stem cell frequency, which we have previously shown is associated with poor clinical outcome and prognosis in glioblastoma. To investigate the clonal changes of GBM through therapy, we adapted a DNA barcoding strategy to track individual tumor cells at multiple time points during the course of disease progression and treatment. We transduced human GBM cells with our high-complexity barcode library and low multiplicity of infection to ensure single barcode representation per cell. This allowed for clonal tracking of hundreds and thousands of GBM cells through our chemoradiotherapy adapted xenograft model and identify the different modes used by glioblastoma to escape therapy and give rise to recurrence. When looking at the dynamics of these barcoded clonal populations, we observed that changes in clonal diversity from engraftment to endpoint in untreated control mice remains relatively constant. However, clonal diversity of different samples responded variably to treatment, as exemplified by BT.428 and BT.799, two GBM samples. While both samples showed an overall decrease in diversity, the effect was greater in BT.428 when treated with chemoradiotherapy. Treatment of BT.428 tumors induced a shift towards clonality, where fewer clones comprised the bulk of the recurrent tumor. A mean of 11 clones, in fact, comprised 98% of the treated tumors compared to an average of 3,000 clones, making up 98% of the untreated control tumors. On the other hand, despite a slight decrease in barcode diversity in BT.799, we did not observe a similar concomitant decrease in total barcodes after treatment. BT.799 control tumors, which are untreated, had an average of 100 clones, making up 98% of the tumor, while treated tumors had an average of 90 clones, making up 98% of the tumor. Closer inspection of the barcode distributions before and after treatment suggests that BT.428 has a more heterogeneous makeup, with a larger number of treatment-sensitive clones that are lost due to treatment and a smaller fraction of treatment-refractory barcoded cells. In contrast, there is little change observed in the distribution of BT.799, both untreated and treated populations, supporting a more homogeneous treatment-refractory population. When looking at clonal diversity at minimal residual disease, or MRD, BT.428, even after treatment, shows high clonal diversity, while untreated samples show a slight enrichment of low-abundance barcodes. Regardless, in BT.428, treatment itself is causing a change in clonal dynamics and population representation. Barcode diversity, on the other hand, in BT.799 shows little change with treatment at MRD. Untreated and treated samples show comparable diversity to that observed at engraftment, which suggests there is similar clonal kinetics in BT.799, irrespective of treatment. The quantitative observations of clonal complexity and expansion from MRD to recurrence suggest two possible mechanisms of recurrence. One, in which a non-reproducible, low-abundance clone rapidly expands post-treatment. We would refer to that as a priori equipotency. Or two, one in which a pre-existing clone of high abundance remains abundant through disease progression, survives treatment, and continues to expand thereafter. We would refer to that as a priori fitness. To assess these hypotheses, we qualitatively tracked independent clones, where each barcode is represented by a single-colored bubble, at minimum residual disease, and at recurrence. And what we find is, in BT.428, as you can see in these images, one color does not show up multiple times through the disease progression or in response to therapy, which suggests that there is very little similarity between replicates across all time points of BT.428, and therefore there is equipotency in all the different clones that are present within BT.428. In contrast, BT.799 reproducibly shows a selection of a few subset of clones at both MRD and recurrence, which are shown here by the same colored clones showing up at multiple time points and through the course of therapy. These results suggest that BT.428 may have clones of equipotent expansion potential that are sporadically selected after treatment and acquire treatment resistance, while BT.799 consists of pre-existing, highly fit, and treatment-resistant oligoclonal populations that dominate through disease progression. Next, we sought to define whether profiling the biology of MRD could provide further clinical and therapeutic insights for glioblastoma. To characterize the MRD stage, we performed single-cell RNA sequencing on two different GBM lines, MBT06, which represented the best outcome in our cohort with 43 months survival, and BT.799, which represented the worst disease outcome in our cohort with 3 months survival. We sought to explore differences with therapeutic and prognostic relevance. UMAP-based embedding revealed three major cellular populations. The first two distinct populations represented control in green and treated in orange MBT06 samples and showed pronounced treatment-induced differences, whereas the third population represented control in pink and treated in blue BT.799 cells co-clustered, suggesting a less pronounced response to treatment. These differential expression profiles were also in agreement with our barcode analysis. To assess the injury to moral heterogeneity, we classified cells into mesenchymal, astrocyte, oligodendrocyte progenitor, and neuroprogenitor-like subtypes. In untreated BT.799 samples, astrocyte-like cells were the most abundant, and the proportion of each state was not significantly affected following treatment. In contrast, the untreated MBT06 sample was predominantly composed of astrocyte and neuroprogenitor-like cells, and treatment induced a significant shift towards the mesenchymal-like state. Next, to identify innate tumor-specific differences in MRD, we performed differential expression analysis of control untreated MBT06 and BT.799 cells. MBT06 was enriched for tumor-specific immune signaling, including antigen presentation, whereas BT.799 was enriched for translation initiation and ribosomal RNA processing. To compare the regulation of these pathways and evaluate their clinical utility, we developed an immunomodulatory IM and translation-initiating TI signature composed of 13 and 12 genes respectively that were differentially expressed between the two samples and coherent across independent GBM datasets. On the right, we show that while these signatures were derived from untreated samples, the treated MBT06 and BT.799 samples consistently responded with an increased expression of the immunomodulatory signature and suppression of the translation-promoting signatures. We next sought to determine whether our immunomodulatory and translation-promoting signatures had prognostic value, as they were derived from the differential pathway activities of tumors associated with the best and worst disease outcomes in our cohort. We evaluated the prognostic value of our signatures using three independent GBM transcriptomic datasets, which included the public TCGA GBM dataset, the immunotherapy-responder and non-responder GBM patient dataset by Zhao et al., the adjuvant or neoadjuvant immunotherapy-treated GBM patient dataset by Golgesi, 2019 et al., and our in-house GBM proteomic dataset obtained by profiling the cerebral spinal fluid from GBM patients. Intriguingly, despite the immunomodulatory signature being derived from a sample associated with the best survival outcome, the immunomodulatory high status, shown here in red, was consistently associated with poor prognosis in the TCGA samples, the immunotherapy non-responders, the immunotherapy adjuvant-treated patients, and in our CSF samples. Taken together, the immunomodulatory-dependent activity offers a reliable biomarker in glioblastoma across all stages of disease at the bulk and single-cell transcriptomic as well as at CSF proteomic level. In conclusion, we have shown a therapy-adapted, patient-derived xenograft model of GBM recurrence which captures the MRD state. By investigating the MRD state further, we can use the MRD time point as a window to target and treat GBM patients before the imminent and heterogeneous clinical relapse. We also show that GBM recurrence post-therapy exists on a spectrum between a priori fitness, whereby distinct pre-existing clones remain abundant from engraftin through to recurrence, and a priori equipotency of clones, whereby treatment-derived subclonal events dominate clonal composition at each time point. We also show a prognostic immunomodulatory signature validated across five independent GBM datasets, including four transcriptomics and one unique CSF proteomics. By performing CSF proteomics, we can use it as a surveillance method to implement iterative detection, profiling, and targeting of MRD in GBM. In the end, I would like to thank the AANS committee for giving me the opportunity to present our work, and I thank you for listening to my presentation.

glioblastoma recurrence

minimal residual disease state

clonal tracking

single-cell RNA sequencing

targeted treatment