see Molecular biomarker.
Bächli et al., report a single-institutional collection of pediatric brain tumor cases that underwent a refinement or a change of diagnosis after completion of molecular diagnostics that affected clinical decision-making including the application of molecularly informed targeted therapies. 13 pediatric CNS tumors were analyzed by conventional histology, immunohistochemistry, and molecular diagnostics including DNA methylation profiling in 12 cases, DNA sequencing in 8 cases and RNA sequencing in 3 cases. 3 tumors had a refinement of diagnosis upon molecular testing, and 6 tumors underwent a change of diagnosis. Targeted therapy was initiated in 5 cases. An underlying cancer predisposition syndrome was detected in 5 cases. Although this case series, retrospective and not population based, has its limitations, insight can be gained regarding precision of diagnosis and clinical management of the patients in selected cases. Accuracy of diagnosis was improved in the cases presented here by the addition of molecular diagnostics, impacting clinical management of affected patients, both in the first-line as well as in the follow-up setting. This additional information may support the clinical decision making in the treatment of challenging pediatric CNS tumors. Prospective testing of the clinical value of molecular diagnostics is currently underway 1).
It has been reiterated many times that molecular typing of (brain) tumors is more reliable and precise than histological classification, but data confirming this belief are largely missing. While it appears intuitive that searching for absence versus presence of a mutation is more straightforward and afflicted with less inter-rater variability than a diagnosis based on the bewildering variety of histological pictures, it still remains a hypothesis that needs to be tested in systematic inter-rater reliability studies. Preliminary endeavors have revealed surprisingly high inter-rater variability of molecular neuropathology. In an unpublished German study involving 22 neuropathology institutions, 20 gliomas were examined for MGMT promoter methylation. Uniform results of methylation versus non-methylation among all institutions were obtained in only four of 20 cases (20%), which is most probably lower than reliability expected for microscopical diagnosis. MGMT analysis may predispose to relatively high variability due to heterogeneous techniques and molecular targets, while assays for hotspot point mutations (IDH) or deletion (1p/19q) are expected to be more reliable, but this remains to be demonstrated and urgently calls for inter-laboratory studies and consensus protocols to guarantee reliable molecular and integrated diagnoses.
Are there valid and convenient surrogate markers? In general, molecular classification is performed using appropriate molecular methods, such as sequencing or methylome analysis. These techniques tend to be expensive and need to be well controlled. Existence of reliable and valid surrogate markers using more convenient and standard methods such as immunohistochemistry would be advantageous. While current agreement indicates that surrogate markers for 1p/19q codeletion do not exist, a few neuropathologists (including Banan and Hartmann, who are authors of the corresponding review article in this issue) believe that molecular classification of medulloblastoma (WNT, SHH, non-WNT/SHH) and ependymoma, RELA fusion-positive, can be performed using appropriate immunohistochemical markers. The problem is that the spectrum of markers as suggested by different experts is variable, and sensitivity and specificity of these markers is less than ideal. Furthermore, in any institution, immunohistochemical markers need to be validated against molecular methods in a large series of tumors before diagnostic application, because immunohistochemical methods and their evaluation may vary widely among institutions.
For example, immunohistochemistry for L1CAM has been suggested as a potential surrogate marker for the diagnosis of ependymoma, RELA fusion-positive. Unfortunately, L1CAM may also be expressed by other ependymoma subtypes and other brain tumors, and only 82% of RELA fusion-positive ependymomas have been shown to be positive for L1CAM in a systematic, well-controlled study 2).
It appears reasonable to assume that sensitivity and specificity will be even lower in a routine setting when immunostaining is performed in a single case every few weeks or months. In a similar vein, a variety of immunohistochemical markers have been recommended for the molecular classification of medulloblastoma, but their sensitivity and specificity are currently less clear than neuropathologists occasionally believe, who otherwise would be unable to classify medulloblastoma according to WHO 2016 or make only a NOS (“not otherwise specified”) diagnosis. Since molecularly defined entities of the WHO Classification are clinically, prognostically and potentially therapeutically relevant, the exclusive use of surrogate markers with 50, 80, or even 98% sensitivity does not appear diagnostically, scientifically, and ethically appropriate, if exact molecular methods are available elsewhere and neuropathology should still be considered the gold standard of diagnosis. Much more work remains to be done.
Economics or ethics?
Undoubtedly exact neuropathological diagnosis of brain tumors has become more expensive with WHO 2016. In an ideal world without financial constraints, every brain tumor would be comprehensively genetically characterized, including whole genomic sequencing and methylome analysis. In general, a classification system does not include statements about what is affordable, in part because there are huge differences between and within nations. However, the Blue Book makes an interesting point with respect to the molecular diagnosis of glioblastoma. The 2016 Classification includes glioblastoma, IDH wild type (also referred to as primary glioblastoma, about 90%) and glioblastoma, IDH mutant (also referred to as secondary glioblastoma, about 10%). The two tumor types differ with respect to age, length of clinical history, and prognosis, making a correct diagnosis clinically relevant. About 90% of IDH mutations are represented by IDH1 R132H, which can be reliably detected using an antibody specific for the mutant protein, whereas the other 10% mutations (IDH1 non-R132H, IDH2) can only be revealed by sequencing IDH1 and IDH2 genes. The proportion of glioblastomas with IDH mutation substantially decreases with age. Accordingly, the Blue Book states that it may be sufficient or “safe” in older patients to rely solely on negative immunohistochemistry for making the diagnosis of glioblastoma, IDH wild type, because in an immunohistochemically negative glioblastoma from a patient without prior lower-grade glioma, the probability of an alternative IDH mutation is <6% in a 50-year-old patient and decreases to <1% in patients aged >54 years. It is debatable whether saving cost and workload by refraining from sequencing IDH1/IDH2 genes in all glioblastomas justifies molecular mis-classification in <5% of patients.
How long does it take to make the final diagnosis? As molecular diagnostics is performed following histological and immunohistochemical analysis, time to final diagnosis inevitably increases for brain tumors with integrated diagnosis. For example, since oligodendroglioma requires molecular pathology, and criteria of anaplasia differ for astrocytic versus oligodendroglial neoplasms, it is not unusual that in a glioma with ambiguous histology and a few mitoses a preliminary diagnosis of diffuse glioma (without type and grade) has to be made for a week or so. Diagnostic turnaround times mainly depend on types of methods and frequency of assays in individual labs.
How can we move forward between WHO classification updates? Most probably, the number of molecularly defined brain tumor types will soon increase. Examples may include meningioma, atypical teratoid/rhabdoid tumor, diffuse astrocytoma IDH wild type, and pilocytic astrocytoma. Other molecular tumor types have not yet been introduced into the WHO Classification system, although they have been already included in consensus suggestions on clinical management, such as ependymoma with YAP fusion or infratentorial ependymoma types A and B 3).
Furthermore, new molecular or surrogate markers that are important for classification and diagnosis will be developed. The current intervals of 7–9 years between WHO Classification updates are certainly too long in this era of rapid progress. In order to provide prompt suggestions for the neuro-oncology community, members of the WHO Working Group and an associated Clinical Advisory Panel have recently constituted cIMPACT-NOW (Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy) 4).
As the suffix NOW indicates, this is Not Official WHO. Suggestions will be solicited from the neuro-oncology community, evaluated in working groups, and guidelines for diagnostics and suggestions for possible WHO updates will be regularly published 5).