Advanced molecular pathology for rare tumours: A national feasibility study and model for centralised medulloblastoma diagnostics

Abstract Aims Application of advanced molecular pathology in rare tumours is hindered by low sample numbers, access to specialised expertise/technologies and tissue/assay QC and rapid reporting requirements. We assessed the feasibility of co‐ordinated real‐time centralised pathology review (CPR), encompassing molecular diagnostics and contemporary genomics (RNA‐seq/DNA methylation‐array). Methods This nationwide trial in medulloblastoma (<80 UK diagnoses/year) introduced a national reference centre (NRC) and assessed its performance and reporting to World Health Organisation standards. Paired frozen/formalin‐fixed, paraffin‐embedded tumour material were co‐submitted from 135 patients (16 referral centres). Results Complete CPR diagnostics were successful for 88% (120/135). Inadequate sampling was the most common cause of failure; biomaterials were typically suitable for methylation‐array (129/135, 94%), but frozen tissues commonly fell below RNA‐seq QC requirements (53/135, 39%). Late reporting was most often due to delayed submission. CPR assigned or altered histological variant (vs local diagnosis) for 40/135 tumours (30%). Benchmarking/QC of specific biomarker assays impacted test results; fluorescent in‐situ hybridisation most accurately identified high‐risk MYC/MYCN amplification (20/135, 15%), while combined methods (CTNNB1/chr6 status, methylation‐array subgrouping) best defined favourable‐risk WNT tumours (14/135; 10%). Engagement of a specialist pathologist panel was essential for consensus assessment of histological variants and immunohistochemistry. Overall, CPR altered clinical risk‐status for 29% of patients. Conclusion National real‐time CPR is feasible, delivering robust diagnostics to WHO criteria and assignment of clinical risk‐status, significantly altering clinical management. Recommendations and experience from our study are applicable to advanced molecular diagnostics systems, both local and centralised, across rare tumour types, enabling their application in biomarker‐driven routine diagnostics and clinical/research studies.


INTRODUC TI ON
Advanced upfront molecular diagnostics and pathology review are an increasing requirement in contemporary cancer care and clinical trials. As next-generation technologies such as high-throughput sequencing and DNA methylation-array profiling unearth novel and clinically significant insights in the research setting, their adoption into routine diagnostics and/or standard clinical care is imperative.
Diagnostics delivery in rare tumour types (typically <5 patients per 100,000 population per year 1 ) presents specific and significant challenges. Care pathways for such diseases vary with clinical setting but are commonly characterised by systems involving low patient numbers spread across multiple treatment centres. In such settings, co-ordinated multicentre approaches offer opportunities to address issues including access to specialist analytical technologies and expertise, analysis of tumour and associated biomaterials across multiple centres, tissue/assay quality control and standardisation, requirements for integrated diagnoses encompassing molecular genetic analysis and histopathological review and prompt reporting to accredited standards. For many tumour types, these processes must be completed prior to the commencement of adjuvant therapy, typically within an approximately 30-day window following primary surgical resection or biopsy.
Centralised diagnostics pathways offer significant potential to address these issues. However, the implementation and performance of centralised molecular pathology review (CPR) systems for rare tumours, and their clinical impact, have not been widely or systematically assessed. Such trials will be essential to support the clinical adoption of such schema and to generate experience which can be applied more widely in routine diagnostics.
Medulloblastoma is an important example of a rare tumour type where molecular advances are rapidly informing its diagnostic requirements and treatment. Medulloblastoma is a clinically and molecularly heterogeneous embryonal tumour arising in the cerebellum, which accounts for ~10% of all childhood cancer deaths. Until recently, medulloblastoma was classified solely by its clinically rele- improving outcomes and reducing late-effects. Early studies of the targeted inhibition of the SHH pathway 9 are in progress. The introduction and rapid upfront assessment of contemporary biomarkers is, thus, essential for entry of medulloblastoma patients into international biomarker-driven trials, and their adoption as 'standard of care'. Finally, exemplifying requirements in many rare tumour types, the assessment, cross-validation and definition of optimal methodologies for detection of specific biomarkers (e.g. MYC/N amplification, WNT subgroup status), including the relative performance of conventional and next-generation technologies, is a critical requirement in their standardisation and clinical adoption.
We assessed whether the introduction of national real-time CPR for medulloblastoma in the United Kingdom would lead to the effective and robust delivery of essential contemporary diagnostics and consequent improved clinical stratification. The annual incidence of medulloblastoma in the United Kingdom is ~80; patients are diagnosed and treated across over 16 specialised treatment centres associated with the UK Children's Cancer and Leukaemia Group (CCLG; https://www.cclg.org.uk/). In the United Kingdom, virtually all CNS tumour diagnosis is undertaken by specialist neuropathologists (with evidence of specialist training in brain tumour diagnosis and on-going expertise demonstrated by external quality assurance), according to ISO15189 standards within the UKAS (United Kingdom Accreditation Service) framework; this national network, thus, offers a unique opportunity to assess the experience of CPR implementation in practice.
We, therefore, undertook a national multicentre feasibility study of real-time medulloblastoma CPR, to establish a routine diagnostics infrastructure on behalf of all UK CCLG centres and to assess the experience and impact of CPR, incorporating next-generation technologies for biomarker assessment. We systematically assessed CPR in 135 patients, demonstrating that it is feasible, provides robust diagnostics to WHO standards and significantly alters clinical management in the setting of initial local diagnosis followed by referral to a national reference centre (NRC). Based on our findings and experience, we proffer recommendations generalisable to all rare tumours, and different diagnostic systems. These should help enable optimisation of pathology review and molecular diagnostics, recruitment to clinical trials and delivery of high-quality surplus tumour material suitable for comprehensive molecular research studies in these diseases.

Study design
This link-anonymised national trial was set up to investigate the feasibility of real-time medulloblastoma diagnostics within a clinically defined timeframe (i.e. before the start of adjuvant therapy) ( Figure 1A). Diagnostic assessments (fluorescent in-situ

Mutational analysis
TP53 (exons 4-9) and CTNNB1 (exon 3) mutation status was assessed by direct polymerase chain reaction (PCR)-based DNA sequence analysis according to the protocols described in Hill et al. 15 and Ellison et al. 16 Molecular subgroup assessment Intact DNA of >500 ng, in a total volume of <45 μl, was suitable for methylation-array (Illumina). RNA quantity and quality was assessed by Agilent 2100 Bioanalyser and those RNAs >800 ng and with an RNA integrity number (RIN) >5 were suitable for RNAseq and subsequent molecular subgrouping. Subgrouping according to methylation and expression profiles was achieved using established methods. 3 SHH and Group3/Group4 secondgeneration subtypes were assigned according to the 'Grp3 and Grp4 Classifier' found at https://www.molec ularn europ athol ogy. org/mnp/class ifier/ 7. 5

Copy number estimation
As well as iFISH (see above), copy-number for MYC(N) was estimated by 1) MLPA and 2) (additionally for chromosome 6) 450 k array probe intensity values using the R package conumee, 17 as previously described. 18 Copy-number assessment by MLPA (using SALSA reagents; MRC-Holland) for MYC(N) were as described 15

Turnaround times
A total of 135 samples were received by the NRC from 16 treatment/referral centres. The mean and median interval between date of surgery and arrival at the NRC was 24 and 14 days, respectively.
The range of 1-301 days reflected the study protocol; specifically, guidance was issued that allowed samples to be registered and submitted without specific time limits, to maximise cohort size and assessment of NRC performance.
A total of 127/135 (94%) samples were processed for iFISH within 8 days of receipt by the NRC (a calculated threshold, defined as the mean processing time plus 1 standard deviation (SD); dashed line Figure 1B). iFISH analysis and reporting was completed within 11 days of dispatch from the NRC (mean + 1 SD) for 130/135 (96%; Figure 1C). A total of 121/135 (90%) of tissues were processed for histology/IHC within 9 days (mean + 1 SD; Figure 1D), and subsequent central pathology review was completed within 15 days for 94% (127/135). Overall, results for 120/135 (88%) of cases were reported to the submitting local centre within 23 days of receipt by the NRC ( Figure 1F). For those for which the mean + 1 SD time threshold was exceeded (15/135; 12%), reasons included personnel unavailability (53%; 8/15), additional CPR (second opinion, additional requested histology and/or IHC; 5/15 [33%]) and technical IHC repeats required within the NRC (2/15; 14%; Figure 1G). The mean and median intervals between date of surgery and final reporting was larger, at 40 and 28 days, respectively (range 14-332 days), reflecting the permissive recruitment approach with regard to patient registration and sample submission to the NRC.

Sufficiency of biological material to support contemporary diagnostics and nextgeneration analyses
We examined whether current tissue handling practices across UK treatment centres were sufficient to provide the abundant F I G U R E 2 Next-generation molecular diagnostics: quality control and suitability of extracted nucleic acids for molecular subgrouping. Bar plots of DNA (A) and RNA (B) extraction yields from tumour material. Light purple, no remaining tissue postnucleic acid extraction; dark purple, remaining tissue. Red lines represent typical quantity thresholds that must be exceeded to allow for downstream research assessments including molecular subgrouping. (C) RIN assessment of RNA quality. Red line represents a quality threshold of 5, below which samples are ineligible for RNAseq. Non-negative matrix factorisation (NMF) clustering of (D) DNA methylation profiles and (E) expression profiles. Pie charts summarise molecular subgroup assignments derived from (F) DNA methylation-array or (G) RNAseq. Failed samples are shown in shades of grey; reasons for failure are poor nucleic yield, low quality RNA (RIN < 5) and insufficient tissue. H, Molecular subgroup assignment by DNA methylation-array (consensus 4-subgroup assignment and second-generation subtype) and RNAseq (consensus 4-subgroup assignment) [Correction added on 31 May 2021, after first online publication: Panel (C) was mislabelled previously and has been added in this version.]

Multi-assay WNT assessment improves subgroup assignment
WNT subgroup status was used as an exemplar to assess interobserver variability in IHC interpretation. WNT subgroup medulloblastomas are defined by activation of the Wnt/Wg signalling pathway, associated with nuclear accumulation of the β-catenin protein, CTNNB1 mutation and monosomy of chromosome 6. 6,16 βcatenin IHC methods have historically been used to evaluate WNT subgroup status, but inter-observer variability and sample heterogeneity may confound this assessment. 19 All cases showing any evidence of β-catenin nuclear accumulation in real-time CPR (1%-100%; n = 14), selected negative controls (n=10) and all CTNNB1 mut cases (n = 8) were subjected to a further round of independent, blind, pathology review by all three neuropathologists (total n = 32; Figure 4A,B). IHC consensus was defined as agreement between two or more pathologists and concurred with initial CPR in 25/32 (78%) of cases.
We then tested how diagnostic accuracy compared with molecular evaluation; methylation-array-based subgrouping, Chr6 monosomy and RNAseq-based subgrouping. The defining WNT consensus call was determined by either CTNNB1 mut or positivity for two or more of the four WNT status measures. Methylation-array and RNAseq subgroup calls had 100% sensitivity and specificity; Chr6 monosomy had a specificity of 100% and 88% sensitivity. Of the IHC evaluations, consensus 3-pathologist review status was more sensitive than initial CPR (82% vs 64%), with indistinguishable specificities (89%). Two cases were false positives by primary CPR IHC, compared with the WNT consensus call, and were negative for all other WNT measures.
Due to this failure rate and the increasingly common adoption of F I G U R E 4 Biomarker assay assessment and validation I: WNT molecular subgroup-comparison of IHC and molecular methods and blinded assessment of interpathologist concordance. A, Illustrative immunohistochemistry (IHC) micrographs showing variation in nuclear β-catenin accumulation in WNT subgroup cases (>20% and 1%-20%). A non-WNT subgroup case (0% nuclear β-catenin accumulation, membranous staining only) is also shown. Magnifications of ×20 are shown, scale bar = 200 μm. B, A comparative analysis of WNT subgroup assignment methods. Primary assessment of WNT status by 1) % nuclear β-catenin accumulation determined by CPR (white text in black boxes shows % nuclear accumulation >10%, those cases with evidence of nuclear positivity <10% shown in black text) and 2) CTNNB1 mutation status (amino acid changes given). % nuclear β-catenin accumulation >10% is also shown for a blinded, three-pathologist IHC evaluation exercise with consensus between 2 or more pathologists from CPR and the blinded evaluation exercise shown by black shading. Assessment of WNT status by molecular analyses are shown. Cases with monosomy 6 are identified by black shading, DNA methylationarray (consensus 4-subgroup, and second-generation subtype) and RNAseq subgroup calls are shown. An overarching WNT consensus call (defined by either CTNNB1 mutation status or IHC positivity plus at least one other molecular method) is shown, and each WNT status assignment methods is compared to this consensus (Sens., sensitivity; Spec., specificity). Grey boxes, negative results; white, not done  Figure 5C; illustrative examples). Of the nine cases shown to be MYC amplified by iFISH, seven also had elevated copy number by MLPA along with 11 false-positive cases. A total of 88 cases showed no evidence of MYC amplification, in agreement with iFISH data (not shown) (overall MLPA sensitivity, 78%; specificity, 88%). DNA methylation-array analysis showed complete specificity but poor sensitivity (62%; Figure 5D). These observations were recapitulated for MYCN amplification analysis, with 9 of the 12 iFISH positive cases also showed elevated copy number by MLPA, but with a number of false positives (n = 10; sensitivity, 75%; specificity, 92%; Figure 5E). DNA methylation-array analysis showed the same sensitivity (75%) but had a no false positives (specificity, 100%). MYCand MYCN-amplified cases showed the expected subgroup-specific distributions; the vast majority (8/9) of MYC-amplified cases were Group3 ( Figure 5D), whereas 11/12 MYCN amplified cases were SHH (n = 6) or Group4 (n = 5; Figure 5D,E).

Application and impact of CPR: diagnosis to SIOP-PNET5-MB and WHO (2016) criteria
The pan-European SIOP-PNET5-MB clinical trial of standard-risk medulloblastoma (NCT02066220) stratifies medulloblastoma patients (3) SHH-activated TP53 wild-type, (4) non-SHH/non-WNT (encompassing Group3 and Group4 as provisional variants). 2 In our cohort, non-SHH/non-WNT predominated as expected; 36% (n = 49) were Group4 and 24% (n = 33) were Group3. Fourteen (10%) WNT patients were identified ( Figure 6B). TP53 mutation status subdivided the SHH group into the respective TP53 mutant (n = 7, 5%) and wildtype (n = 26, 19%) groups. Overall, 96% of our cohort were successfully classified to WHO (2016) standards; this was not possible for the remainder due to a lack of molecular subgroup data.  (Table 1). These findings apply not only in medulloblastoma but also for other rare tumours, where collection and rapid assessment of biological material is key.
Histopathologic and associated review practice also varies significantly internationally. The UK neuropathology community provides a key framework to enable assessment of the value and impact of CPR within a highly specialised group of professionals. In the United Kingdom, neuropathologists qualify to practice following specialist training (assessed by examination) and participate in regular ongoing mandatory national EQA activities (https://www.bns.org.uk/eqaschem e/). Therefore, all tumours, prior to CPR, have been diagnosed by a specialist. We observed that, where local neuropathologists had assigned a medulloblastoma histological variant, there was a good degree of concordance with CPR calls, although significant discordances did occur (10/105; 10%); the distinction between certain DN variants and CLA tumours represented the most common discrepancy. Furthermore, in many cases, local neuropathologists did not assign a histological variant and CPR did; this may in part reflect local practice to use CPR for variant calling. Overall, variant was assigned or altered by CPR in 30% (40/135) of patients, strongly supporting the added value of the CPR process. In any central review system, it is essential to have mechanisms in place to identify and examine discrepancies between local and central opinion, and to reach consensus, and the involvement of multiple reference pathologists in the CPR team was critical to this. We further anticipate discordance Importantly, we observed a lack of prompt patient registration and shipment of biomaterials to the NRC. Of note, the study protocol did not permit intervention; guidelines were issued, and patient registration/sample submission was allowed to proceed without influence by the NRC. Marked improvements are anticipated in settings where real-time molecular diagnostics are mandated clinically to defined timescales. Indeed, this has been borne out by the improved submission rates and turnaround times on the SIOP-PNET5-MB medulloblastoma trial (NCT02066220), now underway; we estimate that >90% of UK medulloblastomas were submitted to the NRC in 2019. We anticipate the obligation to meet such eligibility criteria will be a strong driver of timely completion of the CPR pathway.
Once samples reached the NRC, outcomes were more consistent. The vast majority (88%) had completed diagnostic assessment within 23 days of receipt at the NRC; delays were typically due to personnel availability, the requirement for repeat analyses and further CPR. Insufficiency of submitted biomaterials was apparent, and we concur with previous recommendations 21 that a minimum of 1.5 cm 3 of FFPE tissue and 0.5 cm 3 of frozen tumour tissue should be collected where possible. However, in this study we found that the quality of frozen tumour material was a greater risk to successful delivery of molecular diagnostics. Although DNA is less sensitive to poor tissue-handling practices (reflected by the near-total ability to perform successful DNA methylation array), frozen material submitted was often not suitable for RNA-seq-based analysis. Although it is true that RNAseq sets a high bar for quality, and other molecular tests are suitable for use on RNA of lesser quality and/or quantity  Prompt liaison between local centres and NRCs is essential to ensure the timely and secure transit of biomaterials, with an infrastructure governance provided by the use of appropriate material transfer agreements. Once received by the NRC, robust tests performed in accredited laboratories, which first assess material quality and suitability for assessment, must be applied. 22 Although not collected in our study, workflows for the adequate collection and storage of other biomaterials such as blood (for the assessment of tumour predisposition syndromes and germline genetic variation), CSF (for liquid biopsy approaches) and additional frozen tumour material for proteomics should be built into practice.
The introduction of MB molecular biomarkers into the CPR process both (i) enabled determination of associated risk-status and (ii) highlighted requirements for the assessment and validation of any specific biomarker prior to its diagnostic use. Although specific biomarkers for any given disease will change over time -for instance, Interpretation of β-catenin IHC was highly variable in our blinded assessment of concordance between pathologists, supporting the requirement for objective molecular methods in this instance. The gold standard molecular analysis of MYC and MYCN status is iFISH, but failures were observed in our study. Improved success rates may have been observed if FFPE material had been used for iFISH, but nevertheless our experience suggests a requirement for collection of high-quality frozen tissues and a role for additional methodologies. DNA methylation-array approaches were highly specific but were prone to false negatives, 25 possibly reflecting tumour heterogeneity, whereas MLPA methods commonly yielded false positives and were unsuitable for primary detection of gene amplifications.
Benchmarking and validation vs the current gold standard is, thus, required in the development/adoption of any new biomarker method. For example, array comparative genomic hybridization (aCGH) has been shown to be perform equivalently to iFISH for the detection of MYC/MYCN gene amplification. 26 As novel (including NGS) technologies become available clinically for mutation, copy number and methylation assessment, robust validation against the current 'gold standard' becomes a key requirement.
Finally, reporting of molecular diagnostics to the local and/or treating centre must be in accordance with disease guidelines or the clinical study protocol. We (i) assessed diagnosis of our cohort to

ACK N OWLED G EM ENTS
We are grateful for the pathology support and expertise provided by Keith Robson. We are greatly indebted to CCLG (Study No: 2008 BS 12), the SIOP-Europe Brain Tumour Group and to NHS staff and patients for their participation in the study.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author on reasonable request.