I hope this message finds you all well. My name is Rahul Kumar. I'm a rising fourth-year medical student at the University of Tennessee, and I've just completed my PhD at St. Jude in the lab of Paul Northcott. And it's truly an honor and a privilege to have the opportunity to share our work on exploring cerebrospinal fluid as a medium for liquid biopsy in medulloblastoma patients. Since Cushing's early description of medulloblastoma in the 1920s, tremendous advances have been made in the molecular understanding of the small round blue cell tumor of the posterior fossa, in particular, four core molecular subgroups termed Wnt, Sonic Hedgehog, Group 3 and Group 4 have been codified, each with distinctive clinical behaviors and genetic lesions. While many of our advances have centered on the multidimensional molecular profiling of tumors from initial neurosurgical resections, our ability to longitudinally characterize the molecular changes that occur over the course of subsequent chemo and radiation therapy has been limited. Furthermore, the current methodologies for disease staging, assessment, and surveillance have largely remained unchanged in the modern era. MRI of the neuroaxis along with cytologic examination of the CSF remain the staple workhorses in the assessment of medulloblastoma patients. However, both modalities are limited by issues related to sensitivity, specificity, and inner observer variability. Additionally, these modalities fail to divulge critical molecular features that might not be represented in the initially resected tumor mass or those that might change over time. As such, liquid biopsy using various biofluids has emerged as a robust platform for detecting and characterizing a wide variety of cancers, including CNS tumors. A short non-exhaustive list of seminal studies exploring the utility of CSF-based liquid biopsy in brain tumors is depicted here. In essence, CSF liquid biopsy leverages the fact that malignant cells shed genetic material into the CSF, which can then be isolated and profiled to detect tumor-specific molecular alterations, including somatic copy number alterations and mutations. Furthermore, liquid biopsy can provide critical insights into molecular targets in this era of precision medicine. In our current study, we utilize CSF collected by lumbar puncture to isolate small fragmented pieces of double-stranded DNA known as cell-free DNA for copy number profiling by low-coverage whole genome sequencing. Our decision to focus on copy number profiling as a tumor-specific signature was based on the fact that broad copy number variations are observed in approximately 90% of all medulloblastomas. In the current study, CSF was collected and banked longitudinally over the course of therapy and into follow-up for children with newly diagnosed medulloblastoma. Cell-free DNA from cryopreserved CSF was extracted using commercially available kits. While some of our samples exhibited the characteristic size distributions expected from cell-free DNA, with peaks corresponding to the lengths of nucleosome-associated DNA, yields from other samples were in the picogram range, further motivating us to focus on detection of broad copy number changes instead of mutations, which can be difficult to discern from background noise with extremely low DNA inputs. Isolated cell-free DNA was then prepared for whole genome sequencing with a target of 1-3x genomic coverage. Bioinformatic pipelines were then used to generate copy number profiles from CSF samples. As an example, here we see the copy number profile generated from CSF obtained approximately two weeks after initial resection, prior to the beginning of systemic therapy. In this plot, chromosomes are arranged sequentially along the x-axis, while the copy number states, either gains or losses, are plotted along the y-axis. In this example, we can clearly discern the loss of chromosome 6, inferred from the profile generated from CSF-derived cell-free DNA. We can also compare the CSF-derived copy number profiles to those obtained from the genomic profiling of resected tumor tissue itself, which in this case displays clear concordance. Thus, we established proof of concept for the detection of tumor-specific copy number variation using CSF of patients with previously resected medulloblastomas. In order to benchmark our approach and explore its utility, we assembled a longitudinal cohort of 369 samples from a total of 81 medulloblastoma patients. CSF was also collected at defined clinical time points over the course of therapy and into follow-up. Baseline samples corresponded to CSF that was collected approximately two weeks after initial neurosurgical resection and prior to the onset of systemic therapy. While we were able to generate sequencing libraries from over 99 percent of CSF samples from all time points, we were interested in determinants of tumor detectability using copy number profiling, so we initially focused on CSF samples obtained at baseline. The detectability of tumor-specific copy number alterations in baseline CSF samples was first assessed in univariate analyses. No statistical difference was noted between the exact timing of baseline CSF collection post-resection or the extent of initial resection, though the number of patients with near or subtotal resections was low in our cohort. However, both extended metastasis and molecular subgroup were associated with differences in detectability at baseline. In particular, patients with non-metastatic disease were less likely to have CNVs detected in their baseline samples. On multivariate analyses, detectability in sonic hedgehog tumors was lower than other subgroups, even when controlling for the extent of metastasis. Such a phenomenon may thus reflect the intrinsic differences in the anatomic localization of the various subgroups, as group 3 and 4 tumors often occur within or in close proximity of the fourth ventricle, while sonic hedgehog tumors have a propensity to involve the lateral cerebellar hemispheres. These results also align with previous studies, suggesting proximity to CSF reservoirs as key determinants of detectability of tumor-derived cell-free DNA. Next, we assess the agreement of CNV detection using CSF liquid biopsy with results from CSF cytology and MRI of the brain and spine at baseline. Overall, the agreement amongst all three modalities was only 35%. However, when compared to a combined disease assessment using both MRI and cytology versus CNV detection, the agreement was 65%. Strikingly, however, there were a notable number of samples in which disease was only identified by CNV detection, suggesting that liquid biopsy may enable resolution of microscopic disease not readily appreciable by MRI or cytology. We next turned our attention to the longitudinal assessment of patients over their disease course. Given a median clinical follow-up of over 10 years for our cohort, we were able to identify patients who did not or who did experience progressive disease as determined by conventional clinical radiographic disease assessments. The dashed vertical lines indicate the approximate end of therapy, and each solid circle represents an available CSF sample, colored by whether tumor-specific copy number alterations were detected. Last known follow-up is designated by open circles on the left plot of non-progressors, while clinical radiographic disease progression is shown as black Xs on the right plot of progressors. Amongst non-progressors on the left, only a single false positive was identified after the end of therapy. On the other hand, we can appreciate multiple cases amongst progressors on the right, where tumor-specific copy number variations were detected at or even months before progression was called clinical radiographically. In order to exemplify the utility of our approach in longitudinal disease assessment, we present here a case of a five-year-old male with gross totally resected group 3 medulloblastoma. The reference copy number profile generated from tumor tissue is depicted in the top left plot, and genomic profiling of CSF revealed persistence of tumor specific copy number alterations at the end of radiotherapy. These changes appear to resolve during subsequent chemotherapy. However, three months after the completion of therapy, the patient presented with an isolated spinal progression with reemergence of tumor-specific copy number variations in his CSF. However, reemergence of tumor specific copy number changes was identified at the end of therapy, three months prior to clinical radiographic disease relapse, even in spite of a clean MRI at the eventual site of relapse in the spine and no identifiable tumor cells in the CSF. Of note, similar longitudinal comparisons can also be leveraged to decipher temporal evolution of copy number landscapes. Finally, we explore the potential prognostic utility of CNV detection at the end of therapy. The Kaplan-Meier plot here depicts progression-free survival from the time at the end of therapy by whether or not abnormalities on MRI or cytology were present at that time. Conventional disease assessments by MRI and cytology at the end of therapy largely fails to discriminate patients who will subsequently relapse. On the other hand, detection of copy number variations by CSF liquid biopsy at the end of therapy was strongly associated with lower subsequent progression-free survival compared to patients for whom no such alterations were detected. While additional prospective benchmarking is needed, we believe that CSF liquid biopsy represents a promising approach for early disease detection in medulloblastoma patients. Taken together, I hope I've been able to demonstrate the feasibility and applicability of our approach using copy number profiling of CSF derived cell free DNA to detect and characterize medulloblastoma. While baseline detectability varies according to the subgroup and extent of metastasis, we are hopeful that longitudinal profiling may serve as a useful molecular adjunct to conventional clinical radiographic disease assessment and surveillance. Additionally, as we expand our approach to encompass additional molecular features, we anticipate an even broader applicability to both pediatric and adult CNS malignancies. Please feel free to reach out with any questions or comments. Finally, I've had the outstanding privilege to co-lead this study with Anthony Liu and Kyle Smith from our lab. Additionally, I'd like to thank all of our internal and external collaborators, as well as my research and clinical mentors. I look forward to seeing you all on the interview trail this season, and thanks again for this tremendous opportunity.

Rahul Kumar

University of Tennessee

cerebrospinal fluid

liquid biopsy

medulloblastoma