Ultra-fast deep-learned CNS tumour classification during surgery

Central nervous system tumours represent one of the most lethal cancer types, particularly among children1. Primary treatment includes neurosurgical resection of the tumour, in which a delicate balance must be struck between maximizing the extent of resection and minimizing risk of neurological damage and comorbidity2,3. However, surgeons have limited knowledge of the precise tumour type prior to surgery. Current standard practice relies on preoperative imaging and intraoperative histological analysis, but these are not always conclusive and occasionally wrong. Using rapid nanopore sequencing, a sparse methylation profile can be obtained during surgery4. Here we developed Sturgeon, a patient-agnostic transfer-learned neural network, to enable molecular subclassification of central nervous system tumours based on such sparse profiles. Sturgeon delivered an accurate diagnosis within 40 minutes after starting sequencing in 45 out of 50 retrospectively sequenced samples (abstaining from diagnosis of the other 5 samples). Furthermore, we demonstrated its applicability in real time during 25 surgeries, achieving a diagnostic turnaround time of less than 90 min. Of these, 18 (72%) diagnoses were correct and 7 did not reach the required confidence threshold. We conclude that machine-learned diagnosis based on low-cost intraoperative sequencing can assist neurosurgical decision-making, potentially preventing neurological comorbidity and avoiding additional surgeries.

Verbal and General IQ Associate with Supragranular Layer Thickness and Cell Properties of the Left Temporal Cortex

The left temporal lobe is an integral part of the language system and its cortical structure and function associate with general intelligence. However, whether cortical laminar architecture and cellular properties of this brain area relate to verbal intelligence is unknown. Here, we addressed this using histological analysis and cellular recordings of neurosurgically resected temporal cortex in combination with presurgical IQ scores. We find that subjects with higher general and verbal IQ scores have thicker left (but not right) temporal cortex (Brodmann area 21, BA21). The increased thickness is due to the selective increase in layers 2 and 3 thickness, accompanied by lower neuron densities, and larger dendrites and cell body size of pyramidal neurons in these layers. Furthermore, these neurons sustain faster action potential kinetics, which improves information processing. Our results indicate that verbal mental ability associates with selective adaptations of supragranular layers and their cellular micro-architecture and function in left, but not right temporal cortex.

P12.02 Symptoms of patients with diffuse glioma during stable disease as a potential target for treatment

BACKGROUND

Patients with diffuse glioma experience many tumor- and treatment-related symptoms, including fatigue, cognitive deficits, and anxiety. It remains unclear how the frequency and burden of these symptoms change over time and specifically which symptoms patients experience during the stable phase of the disease. We assessed the frequency and burden of common symptoms in glioma patients during stable disease and whether patients were willing to undergo treatment for those symptoms. We suggest a top five of potentially treatable symptoms.

METHODS

Patients with a diffuse glioma during stable disease completed a questionnaire on symptoms they experienced during the past two weeks, such as irritability, fatigue, stress, pain, depressive thoughts and relationship problems. They rated the frequency and burden of seventeen symptoms on a seven-point numeric scale ranging from ‘never’ to ‘continuously’ and from ‘no burden’ to ‘unbearable’, respectively. Also, they stated whether they were willing to undergo treatment for each symptom. Data was collected retrospectively. The median values of the frequency and burden dimension were calculated for each symptom. We composed a top five of most frequent and burdensome symptoms. We used Kendall’s Tau to correlate frequency, burden and willingness to undergo treatment per symptom.

RESULTS

Fifty-two patients were included, 77% had a KPS score of 90–100, 23% a KPS score of 70–80 and 40% had a glioblastoma. The top five most frequent and burdensome symptoms were fatigue, memory problems, reduced physical fitness, concentration problems and drowsiness. Fatigue had the highest median value on the frequency (median 4.5, IQR 2.5) and burden (median 4.0, IQR 3.00) dimension and 21% of the patients were willing to undergo treatment for fatigue. Overall 35% of the patients were willing to undergo treatment for at least one symptom. Frequency and burden of symptoms correlated well (median Kendall’s Tau of 0.84 with a range of correlations of 0.73–0.95). Frequency and willingness to undergo treatment correlated less (median Kendall’s Tau of 0.35, range 0.24–0.57), as well as burden and willingness to undergo treatment (median Kendall’s Tau, range 0.28–0.61).

CONCLUSION

The top five symptoms most often reported by patients with diffuse glioma during stable disease were: fatigue, memory problems, reduced physical fitness, concentration problems and drowsiness. These symptoms may serve as a roadmap for future research on treatment strategies to alleviate these symptoms.

SUPPORT/DISCLOSURE

This project is funded by The Anita Veldman Foundation (CCA-2019-2-21).

P11.01 Symptom networks in glioma: a novel approach to study multidimensional symptomatology in glioma patients

BACKGROUND

Glioma patients experience a high symptom burden contributing to poorer quality of life. Symptoms include depression, cognitive impairment, and fatigue and vary throughout the disease. These symptoms are rarely studied from a comprehensive perspective, while their interdependence may be relevant for their development, perpetuation, and ultimately successful treatment. The emerging field of symptom network analysis uncovers the multidimensional symptom space. Nodes are the symptoms, and edges are operationalized as the full conditional association, or partial correlation, between two symptom severity scores across patients. Highly connected nodes are considered central and may be particularly relevant targets for treatment as disruption of these central nodes impact the entire network. We visualized the overall glioma symptom network, compared multidimensional results to known literature, and statistically compared networks between relevant patient subgroups.

MATERIAL AND METHODS

A dataset comprised of 355 observations of 180 glioma patients at different disease phases was analysed. Cognitive testing and questionnaires regarding health-related and glioma-specific quality of life, fatigue, depression, and cognition resulted in the definition of 30 symptom nodes. Symptom clusters were visually explored in the resulting networks, as were node strength, betweenness, and closeness centrality measures for each node. Networks were statistically compared between preoperative patients and patients during stable disease, as well as patients with low versus high-grade gliomas. Networks of patients with normal versus severe levels of fatigue were also compared as cancer-/glioma-related fatigue has a strong impact on quality of life and can correlate with other common symptoms such as pain, depression, and/or sleep disturbance.

RESULTS

Symptom clusters existed between: 1) bodily pain, headache and physical functioning; 2) concentration and motivation; and 3) fatigue and drowsiness. Fatigue and mental health were the most central nodes in the networks. Furthermore, the overall connectivity between symptoms was significantly higher in the severely fatigued patients than in patients with normal fatigue. No network differences were found between low versus high-grade, and preoperative versus stable disease networks.

CONCLUSION

Fatigue is a central node in glioma patients’ burden of disease, and symptoms are more tightly intercorrelated in patients experiencing severe fatigue. From our data, we hypothesize that fatigue co-exists with or perpetuates other symptoms. Thus, although these results are preliminary, the network approach may innovate hypothesis generation in symptom management.

Digital PCR quantification of MGMT methylation refines prediction of clinical benefit from alkylating agents in glioblastoma and metastatic colorectal cancer

BACKGROUND

O(6)-methyl-guanine-methyl-transferase (MGMT) silencing by promoter methylation may identify cancer patients responding to the alkylating agents dacarbazine or temozolomide.

PATIENTS AND METHODS

We evaluated the prognostic and predictive value of MGMT methylation testing both in tumor and cell-free circulating DNA (cfDNA) from plasma samples using an ultra-sensitive two-step digital PCR technique (methyl-BEAMing). Results were compared with two established techniques, methylation-specific PCR (MSP) and Bs-pyrosequencing.

RESULTS

Thresholds for MGMT methylated status for each technique were established in a training set of 98 glioblastoma (GBM) patients. The prognostic and the predictive value of MGMT methylated status was validated in a second cohort of 66 GBM patients treated with temozolomide in which methyl-BEAMing displayed a better specificity than the other techniques. Cutoff values of MGMT methylation specific for metastatic colorectal cancer (mCRC) tissue samples were established in a cohort of 60 patients treated with dacarbazine. In mCRC, both quantitative assays methyl-BEAMing and Bs-pyrosequencing outperformed MSP, providing better prediction of treatment response and improvement in progression-free survival (PFS) (P < 0.001). Ability of methyl-BEAMing to identify responding patients was validated in a cohort of 23 mCRC patients treated with temozolomide and preselected for MGMT methylated status according to MSP. In mCRC patients treated with dacarbazine, exploratory analysis of cfDNA by methyl-BEAMing showed that MGMT methylation was associated with better response and improved median PFS (P = 0.008).

CONCLUSIONS

Methyl-BEAMing showed high reproducibility, specificity and sensitivity and was applicable to formalin-fixed paraffin-embedded tissues and cfDNA. This study supports the quantitative assessment of MGMT methylation for clinical purposes since it could refine prediction of response to alkylating agents.