Timing is everything: A connection between medulloblastoma prognosis and foetal cerebellar development

The childhood brain tumour medulloblastoma is typically classified into multiple discrete molecular subgroups with characteristic DNA methylation and expression patterns. Several of these subgroups are used as, or proposed to be, an effective basis for treatment stratification. Here, we highlight the close connection between the findings described in a recent series of studies which, together, strongly imply a continuous association between survival outcome, the transcriptional profile of a Group3/Group4 (i.e. non‐WNT/non‐SHH) medulloblastoma and the specific point during early foetal cerebellar development at which initial pathogenic disruption took place. This has important implications for future efforts to model the disease by incorporating driving molecular features into their specific developmental context. This further suggests that instead of relying upon discrete DNA methylation subgroups, using expression biomarkers as the basis of a continuous risk predictor may produce a more effective risk stratification of patients with Group3/Group4 medulloblastoma.

association between survival outcome, the transcriptional profile of a Group3/Group4 (i.e. non-WNT/non-SHH) medulloblastoma and the specific point during early foetal cerebellar development at which initial pathogenic disruption took place. This has important implications for future efforts to model the disease by incorporating driving molecular features into their specific developmental context. This further suggests that instead of relying upon discrete DNA methylation subgroups, using expression biomarkers as the basis of a continuous risk predictor may produce a more effective risk stratification of patients with Group3/Group4 medulloblastoma.
K E Y W O R D S cerebellar, development, medulloblastoma, scRNA-seq, transcriptomics Molecular-biological prognostication in medulloblastoma has been driven, over the last decade, by the discovery of increasing numbers of molecular subgroups and their associated mutations. Initially, medulloblastoma was divided by transcriptional profiling into SHH, WNT, Group3 (MB Grp3 ) and Group4 (MB Grp4 ) [1]; each has now been further subdivided into subgroups by DNA methylation patterns [2][3][4]. For example, the combined Group3/Group4 medulloblastoma (MB Grp3/Grp4 ) was divided into eight further subgroups (I-VIII) [5].
While of great biological interest, this atomisation of an already rare disease presents a practical problem to clinical trialists aiming to preserve statistical power. Here, we highlight the close connections between the findings described in a series of papers published in close succession by ourselves and others [6][7][8][9], which together support the prognostic potential of gene expression across MB Grp3/Grp4 medulloblastoma as a whole and its relationship to normal cerebellar development.
Our recent study [9] analysed transcriptional profiles-in MB Grp3/Grp4 combined-to describe a single transcriptional continuum, a pattern of continuous expression changes that links all MB Grp3 and MB Grp4 patients. We devised a 'G3/G4' score to represent this expression pattern and used this score to place each patient at a unique position along the continuum between two extremes (i.e., the archetypal MB Grp3 and MB Grp4 transcriptional states). Position on the continuum was reflective of an individual's clinicopathology and significantly related to 5-year survival.
Korshunov et al. [7] also recently performed a transcriptomic analysis of MB Grp3 , describing differences in gene expression between patients who died vs those who survived 5 years post diagnosis. They identified six differentially expressed genes which were associated with high-risk disease-MYC, KIRREL2, ITPRIPL1, DCAF4, NPW and CDT1-highlighting KIRREL2 expression in particular, as prognostic, independent of other clinico-molecular features, most notably MB Grp3/Grp4 DNA methylation subgroups (I-VIII) [5]. High KIRREL2 expression was present in all MB Grp3 subgroups and, in combination with other risk factors, stratified disease risk with good accuracy.
While Korshunov et al. [7] divided their patients into high and low expressors, the survival association we described was linear and continuous; the higher an individual's G3/G4 score (i.e., the more 'MB Grp3 -like'), the worse the prognosis. Notably, within our MB Grp3/ Grp4 cohort, the expression of KIRREL2 is log-linearly correlated to the G3/G4 score and a major contributor (22nd out of 56,546 transcripts) to the MB Grp3/Grp4 continuum signature ( Figure 1A); indeed, each of the top 6 genes described are significantly correlated with the G3/G4 score and with survival ( Figure 1B-D). This strongly suggests that the results described by Korshunov et al. [7] are substantially part of the phenomenon described by the G3/G4 continuum.
Our G3/G4 expression continuum-of which KIRREL2 expression is a key constituent-was also projected onto an scRNA-seq expression atlas of early human foetal cerebellar development [10], showing that the continuum is mirrored by a specific developmental trajectory-beginning with early rhombic lip (RL) precursors (most Smith et al. [8] and Hendrikse et al. [6] also recently published similar findings, that is, a single developmental RL/UBC lineage as the origin of the MB Grp3/Grp4 subgroups based on analysis of the same scRNA-seq atlas [12]. Both studies aligned similar G3/G4 transcriptional patterns to two spatially distinct ventricular and subventricular compartments within early human RL development. Hendrikse et al. to occur [4]. One interpretation is that 'earlier' disruption in a more progenitor-like/undifferentiated cell allows more rounds of division before postmitotic stalling, producing greater or longer lasting dysplasia. Consequently, a larger pool of premalignant cells is available to suffer a second hit and develop into cancer; therefore occurring statistically sooner. For the present, this remains speculation, although we note that methods to estimate the number of symmetric/asymmetric cell divisions prior to tumour initiation exist. For instance, by monitoring SNVs accumulating naturally through cell division coupled with multiregional tissue sampling and/or single-cell sequencing [15] or by cell tracing experiments within mouse models [16]. Regardless, the answer surely lies in observing and/or modelling these putative PeRLs whose predetermined character may dictate an individual's tumour biology and outcome. Critically, we note that two independent studies now confirm significant prognostic information, not readily apparent from DNA methylation subtyping, is encapsulated in a continuous manner within transcriptional patterns. This suggests a different approach to risk prognostication among MB Grp3/Grp4 -as an alternative or adjunct to subcategorisation by subgroup-whereby a continuous base level of risk can be assigned to an individual patient based on their transcriptional profile, after which further independent risk modifiers (e.g., MYC amplification, presence of metastases, MB Grp3/Grp4 subgroup VII) may then be applied ( Figure 1F).
The opportunities afforded by transcriptional analysis are perhaps best illustrated by those individuals who, by DNA methylation profiling, belong to MB Grp3/Grp4 subgroups II and III. These subgroups contain exclusively MB Grp3 individuals and, taken as a whole, demonstrate the classic poor prognosis historically associated with

Key points
• Recent studies describe a continuous association between survival outcome and expression profile in Group3/Group4 medulloblastoma.
• Tumour expression profile is related to disruption at specific points in early foetal cerebellar development.
• Expression biomarkers may be used as the basis of a continuous risk predictor for effective risk stratification.
F I G U R E 1 Legend on next page.  [5]. Nevertheless, according to our transcriptional study [9], only 40% of those patients could be considered to have a 5-year survival <50% by virtue of their biology alone, that is, with no other risk modifiers taken into account. Similarly, Korshunov et al. [7] reported that the addition of KIRREL2 expression to their biomarker scheme altered the risk category of up to 50% of MB Grp3 patients, with 25% of such patients assigned to a low-risk category with 5-year OS 95%. Put simply, expression profiling can be deployed to reassign substantial proportions of patients more accurately within each MB Grp3/Grp4 subtype into 'better' or 'worse' treatment stratification groups. The job of upcoming clinical trials/biological studies will be to determine how best to deploy such information in the future.
We note that a surrogate G3/G4 continuum score can be derived from DNA methylation profiles-widely used for diagnosis-if no expression profile exists [9]. Conversely, a focussed diagnostic transcriptomic assay (qRT-PCR, NanoString, etc.) could technically be used to achieve prognostication on its own, where no DNA methylation profile exists; however, we stress that we do not believe this to be desirable. We would expect progressive efforts in clinical trials and standard-of-care to continue, for the time being, to rely, in part, upon information more obtainable by DNA methylation analysis, and therefore, we imagine both expression and methylation assays operating side by side for the foreseeable future.
In short, there is now good evidence, backed up by a developmental/biological rationale, that expression is an effective prognostic determinant-particularly for MB Grp3 . Further prospective clinical trials data will be required to definitively settle whether continuous or subgroup-based prognostication, or a combination of both approaches, is most efficacious.