Targeting oncogenic pathways holds promise for brain tumor treatment, but inhibition of Sonic Hedgehog (SHH) signaling has failed in SHH-driven medulloblastoma. Cellular diversity within tumors and reduced lineage commitment can undermine targeted therapy by increasing the probability of treatment-resistant populations. Using single-cell RNA-seq and lineage tracing, we analyzed cellular diversity in medulloblastomas in transgenic, medulloblastoma-prone mice, and responses to the SHH-pathway inhibitor vismodegib. In untreated tumors, we find expected stromal cells and tumor-derived cells showing either a spectrum of neural progenitor-differentiation states or glial and stem cell markers. Vismodegib reduces the proliferative population and increases differentiation. However, specific cell types in vismodegib-treated tumors remain proliferative, showing either persistent SHH-pathway activation or stem cell characteristics. Our data show that even in tumors with a single pathway-activating mutation, diverse mechanisms drive tumor growth. This diversity confers early resistance to targeted inhibitor therapy, demonstrating the need to target multiple pathways simultaneously.

Background Medulloblastoma (MB) is a heterogeneous disease in which neoplastic cells and associated immune cells contribute to disease progression. We aimed to determine the influence of neoplastic and immune cell diversity on MB biology in patient samples and animal models. Methods To better characterize cellular heterogeneity in MB we used single-cell RNA sequencing, immunohistochemistry and deconvolution of transcriptomic data to profile neoplastic and immune populations in patient samples and animal models across childhood MB subgroups. Results Neoplastic cells cluster primarily according to individual sample of origin which is influenced by chromosomal copy number variance. Harmony alignment reveals novel MB subgroup/subtype-associated subpopulations that recapitulate neurodevelopmental processes, including photoreceptor and glutamatergic neuron-like cells in molecular subgroups GP3 and GP4, and a specific nodule-associated neuronally-differentiated subpopulation in subgroup molecular SHH. We definitively chart the spectrum of MB immune cell infiltrates, which include subpopulations that recapitulate developmentally-related neuron-pruning and antigen presenting myeloid cells. MB cellular diversity matching human samples is mirrored in subgroup-specific mouse models of MB. Conclusions These findings provide a clearer understanding of the diverse neoplastic and immune cell subpopulations that constitute the MB microenvironment.

While treatments that induce DNA damage are commonly used as anti-cancer therapies, the mechanisms through which DNA damage produces a therapeutic response are incompletely understood. Here we have tested whether medulloblastomas must be competent for apoptosis to be sensitive to radiation therapy. Whether apoptosis is required for radiation sensitivity has been controversial. Medulloblastoma, the most common malignant brain tumor in children, is a biologically heterogeneous set of tumors typically sensitive to radiation and chemotherapy; 80% of medulloblastoma patients survive long-term after treatment. We used functional genetic studies to determine if the intrinsic apoptotic pathway is required for radiation to produce a therapeutic response in mice with primary, Shh-driven medulloblastoma. We found that cranial radiation extended the survival of medulloblastoma-bearing mice and induced widespread apoptosis. Expression analysis and conditional deletion studies showed that p53 was the predominant transcriptional regulator activated by radiation and was strictly required for treatment response. Deletion of Bax, which blocked apoptosis downstream of p53, was sufficient to render tumors radiation resistant. In apoptosis-incompetent, Bax-deleted tumors, radiation activated p53-dependent transcription without provoking cell death and caused two discrete populations to emerge. Most radiated tumor cells underwent terminal differentiation. Perivascular cells, however, quickly resumed proliferation despite p53 activation, behaved as stem cells, and rapidly drove recurrence. These data show that radiation must induce apoptosis in tumor stem cells to be effective. Mutations that disable the intrinsic apoptotic pathways are sufficient to impart radiation resistance. We suggest that medulloblastomas are typically sensitive to DNA-damaging therapies because they retain apoptosis competence.

Medulloblastoma (MB) is a heterogeneous disease in which neoplastic cells and associated immune cells contribute to disease progression. To better understand cellular heterogeneity in MB we profile neoplastic and immune populations in childhood MB samples using single-cell RNA sequencing, immunohistochemistry and deconvolution of transcriptomic data. Neoplastic cells cluster primarily according to individual sample of origin which is in part due to the effect of chromosomal copy number gains and losses. Harmony alignment reveals novel MB subgroup/subtype-associated subpopulations that recapitulate neurodevelopmental processes and are associated with clinical outcomes. We identify discrete photoreceptor-like cells in MB subgroups GP3 and GP4 and nodule-associated neuronally-differentiated cells in subgroup SHH. MB immune infiltrates consist of both developmentally-related neuron-pruning and antigen presenting myeloid cells. We show that this MB cellular diversity is recapitulated in genetically engineered mouse subgroup-specific models of MB. These findings advance our understanding of both the neoplastic and immune landscape of MB.

It is unclear why medulloblastoma patients receiving similar treatments experience different outcomes. Transcriptomic profiling identified subgroups with different prognoses, but in each subgroup, individuals remain at risk of incurable recurrence. To investigate why similar-appearing tumors produce variable outcomes, we analyzed medulloblastomas triggered in transgenic mice by a common driver mutation expressed at different points in brain development. We genetically engineered mice to express oncogenic SmoM2, starting in multipotent glio-neuronal stem cells, or committed neural progenitors. Both groups developed medulloblastomas with similar transcriptomic profiles. We compared medulloblastoma progression, radiosensitivity, and cellular heterogeneity, determined by single-cell transcriptomic analysis (scRNA-seq). Stem cell-triggered medulloblastomas progressed faster, contained more OLIG2-expressing stem-like cells, and consistently showed radioresistance. In contrast, progenitor-triggered MBs progressed slower, down-regulated stem-like cells and were curable with radiation. Progenitor-triggered medulloblastomas also contained more diverse stromal populations, with more Ccr2+ macrophages and fewer Igf1+ microglia, indicating that developmental events affected the subsequent tumor microenvironment. Reduced mTORC1 activity in M-Smo tumors suggests that differential Igf1 contributed to differences in phenotype. Developmental events in tumorigenesis that were obscure in transcriptomic profiles thus remained cryptic determinants of tumor composition and outcome. Precise understanding of medulloblastoma pathogenesis and prognosis requires supplementing transcriptomic/methylomic studies with analyses that resolve cellular heterogeneity.

Patients with medulloblastoma are typically treated with a narrow range of therapies, but may experience widely divergent outcomes; 80-90% become long-term survivors while 20% develop incurable recurrence. Transcriptomic profiling has identified four subgroups with different recurrence risks, but outcomes remain variable for individual patients within each subgroup. To gain new insight into why patients with similar-appearing tumors have variable outcomes, we examined how the timing of tumor initiation effects medulloblastomas triggered by a single, common driver mutation. We genetically-engineered mice to express an oncogenic Smo allele starting early in development in the broad lineage of neural stem cells, or later, in the more committed lineage of cerebellar granule neuron progenitors. Both groups developed medulloblastomas and no other tumors. We compared medulloblastoma progression, response to therapy, gene expression profile and cellular heterogeneity, determined by single cell transcriptomic analysis (scRNA-seq). The average transcriptomic profiles of the tumors were similar. However, stem cell-triggered medulloblastomas progressed faster, contained more OLIG2-expressing tumor stem cells, and consistently showed radioresistance. In contrast, progenitor-triggered MBs progressed slower, lost stem cell character over time and were radiosensitive. Progenitor-triggered medulloblastomas also contained more diverse stromal populations, including tumor-associated macrophages, indicating that the timing of oncogenesis affected the subsequent interactions between the tumor and microenvironment. Our findings show that developmental events in tumorigenesis may be impossible to infer from transcriptomic profile, but while remaining cryptic can nevertheless influence tumor composition and the outcome of therapy. Precise understanding of medulloblastoma pathogenesis and prognosis requires supplementing transcriptomic data with biomarkers of cellular heterogeneity.

Cerebellar development requires regulated proliferation of cerebellar granule neuron progenitors (CGNPs). Inadequate CGNP proliferation causes cerebellar hypoplasia whereas excessive CGNP proliferation can cause medulloblastoma, the most common malignant pediatric brain tumor. Although sonic hedgehog (SHH) signaling is known to activate CGNP proliferation, the mechanisms downregulating proliferation are less defined. We investigated CGNP regulation by GSK-3, which downregulates proliferation in the forebrain, gut and breast by suppressing mitogenic WNT signaling in mouse. In striking contrast to these systems, we found that co-deleting Gsk3a and Gsk3b blocked CGNP proliferation, causing severe cerebellar hypoplasia. The GSK-3 inhibitor CHIR-98014 similarly downregulated SHH-driven proliferation. Transcriptomic analysis showed activated WNT signaling and upregulated Cdkn1a in Gsk3a/b-deleted CGNPs. Ctnnb co-deletion increased CGNP proliferation and rescued cerebellar hypoproliferation in Gsk3a/b mutants, demonstrating physiological control of CGNPs by GSK-3, mediated through WNT. SHH-driven medulloblastomas similarly required GSK-3, as co-deleting Gsk3a/b blocked tumor growth in medulloblastoma-prone SmoM2 mice. These data show that a GSK-3/WNT axis modulates the developmental proliferation of CGNPs and the pathological growth of SHH-driven medulloblastoma. The requirement for GSK-3 in SHH-driven proliferation suggests that GSK-3 may be targeted for SHH-driven medulloblastoma therapy.

Motivation: Low-dimensional representations of high-dimensional data are routinely employed in biomedical research to visualize, interpret and communicate results from different pipelines. In this article, we propose a novel procedure to directly estimate t-SNE embeddings that are not driven by batch effects. Without correction, interesting structure in the data can be obscured by batch effects.The proposed algorithm can therefore significantly aid visualization of high-dimensional data.Results: The proposed methods are based on linear algebra and constrained optimization, leading to efficient algorithms and fast computation in many high-dimensional settings. Results on artificial single-cell transcription profiling data show that the proposed procedure successfully removes multiple batch effects from t-SNE embeddings, while retaining fundamental information on cell types. When applied to single-cell gene expression data to investigate mouse medulloblastoma, the proposed method successfully removes batches related with mice identifiers and the date of the experiment, while preserving clusters of oligodendrocytes, astrocytes, and endothelial cells and microglia, which are expected to lie in the stroma within or adjacent to the tumors.Availability: Source code implementing the proposed approach in R and Julia is available at the link https://github.com/emanuelealiverti/BC tSNE, including a tutorial to reproduce the simulation studies.