Direct image to subtype prediction for brain tumors using deep learning.

Background: Deep Learning (DL) can predict molecular alterations of solid tumors directly from routine histopathology slides. Since the 2021 update of the World Health Organization (WHO) diagnostic criteria, the classification of brain tumors integrates both histopathological and molecular information. We hypothesize that DL can predict molecular alterations as well as WHO subtyping of brain tumors from hematoxylin and eosin-stained histopathology slides. Methods: We used weakly supervised DL and applied it to three large cohorts of brain tumor samples, comprising N = 2845 patients. Results: We found that the key molecular alterations for subtyping, IDH and ATRX, as well as 1p19q codeletion, were predictable from histology with an area under the receiver operating characteristic curve (AUROC) of 0.95, 0.90, and 0.80 in the training cohort, respectively. These findings were upheld in external validation cohorts with AUROCs of 0.90, 0.79, and 0.87 for prediction of IDH, ATRX, and 1p19q codeletion, respectively. Conclusions: In the future, such DL-based implementations could ease diagnostic workflows, particularly for situations in which advanced molecular testing is not readily available.

Autonomous rhythmic activity in glioma networks drives brain tumour growth.

Diffuse gliomas, particularly glioblastomas, are incurable brain tumours1. They are characterized by networks of interconnected brain tumour cells that communicate via Ca2+ transients2-6. However, the networks' architecture and communication strategy and how these influence tumour biology remain unknown. Here we describe how glioblastoma cell networks include a small, plastic population of highly active glioblastoma cells that display rhythmic Ca2+ oscillations and are particularly connected to others. Their autonomous periodic Ca2+ transients preceded Ca2+ transients of other network-connected cells, activating the frequency-dependent MAPK and NF-κB pathways. Mathematical network analysis revealed that glioblastoma network topology follows scale-free and small-world properties, with periodic tumour cells frequently located in network hubs. This network design enabled resistance against random damage but was vulnerable to losing its key hubs. Targeting of autonomous rhythmic activity by selective physical ablation of periodic tumour cells or by genetic or pharmacological interference with the potassium channel KCa3.1 (also known as IK1, SK4 or KCNN4) strongly compromised global network communication. This led to a marked reduction of tumour cell viability within the entire network, reduced tumour growth in mice and extended animal survival. The dependency of glioblastoma networks on periodic Ca2+ activity generates a vulnerability7 that can be exploited for the development of novel therapies, such as with KCa3.1-inhibiting drugs.

Tumor cell plasticity, heterogeneity, and resistance in crucial microenvironmental niches in glioma.

Both the perivascular niche (PVN) and the integration into multicellular networks by tumor microtubes (TMs) have been associated with progression and resistance to therapies in glioblastoma, but their specific contribution remained unknown. By long-term tracking of tumor cell fate and dynamics in the live mouse brain, differential therapeutic responses in both niches are determined. Both the PVN, a preferential location of long-term quiescent glioma cells, and network integration facilitate resistance against cytotoxic effects of radiotherapy and chemotherapy-independently of each other, but with additive effects. Perivascular glioblastoma cells are particularly able to actively repair damage to tumor regions. Population of the PVN and resistance in it depend on proficient NOTCH1 expression. In turn, NOTCH1 downregulation induces resistant multicellular networks by TM extension. Our findings identify NOTCH1 as a central switch between the PVN and network niche in glioma, and demonstrate robust cross-compensation when only one niche is targeted.

Tweety-Homolog 1 Drives Brain Colonization of Gliomas.

Early and progressive colonization of the healthy brain is one hallmark of diffuse gliomas, including glioblastomas. We recently discovered ultralong (>10 to hundreds of microns) membrane protrusions [tumor microtubes (TMs)] extended by glioma cells. TMs have been associated with the capacity of glioma cells to effectively invade the brain and proliferate. Moreover, TMs are also used by some tumor cells to interconnect to one large, resistant multicellular network. Here, we performed a correlative gene-expression microarray and in vivo imaging analysis, and identified novel molecular candidates for TM formation and function. Interestingly, these genes were previously linked to normal CNS development. One of the genes scoring highest in tests related to the outgrowth of TMs was tweety-homolog 1 (TTYH1), which was highly expressed in a fraction of TMs in mice and patients. Ttyh1 was confirmed to be a potent regulator of normal TM morphology and of TM-mediated tumor-cell invasion and proliferation. Glioma cells with one or two TMs were mainly responsible for effective brain colonization, and Ttyh1 downregulation particularly affected this cellular subtype, resulting in reduced tumor progression and prolonged survival of mice. The remaining Ttyh1-deficient tumor cells, however, had more interconnecting TMs, which were associated with increased radioresistance in those small tumors. These findings imply a cellular and molecular heterogeneity in gliomas regarding formation and function of distinct TM subtypes, with multiple parallels to neuronal development, and suggest that Ttyh1 might be a promising target to specifically reduce TM-associated brain colonization by glioma cells in patients.SIGNIFICANCE STATEMENT In this report, we identify tweety-homolog 1 (Ttyh1), a membrane protein linked to neuronal development, as a potent driver of tumor microtube (TM)-mediated brain colonization by glioma cells. Targeting of Ttyh1 effectively inhibited the formation of invasive TMs and glioma growth, but increased network formation by intercellular TMs, suggesting a functional and molecular heterogeneity of the recently discovered TMs with potential implications for future TM-targeting strategies.

Tumor microtubes convey resistance to surgical lesions and chemotherapy in gliomas.

BACKGROUND: Primary and adaptive resistance against chemo- and radiotherapy and local recurrence after surgery limit the benefits from these standard treatments in glioma patients. Recently we found that glioma cells can extend ultra-long membrane protrusions, "tumor microtubes" (TMs), for brain invasion, proliferation, and interconnection of single cells to a syncytium that is resistant to radiotherapy. We wondered whether TMs also convey resistance to the other 2 standard treatment modalities. METHODS: Patient-derived glioblastoma stemlike cell (GBMSC) lines were implanted under a cranial window in mice. Longitudinal in vivo two-photon laser scanning microscopy was used to follow tumor growth, including the fate of single glioma cells over months. RESULTS: After a cylindrical surgical lesion, GBMSCs increasingly extended TMs toward the lesion area, which contributed to the repopulation of this area over many weeks. In fact, an excessive "healing response" was observed in which tumor cell densities significantly exceeded those of unlesioned brain regions over time. Inhibition of TM formation and function by genetic targeting of growth associated protein-43 robustly suppressed this surgery-induced tumor growth reaction, in contrast to standard postsurgical anti-inflammatory treatment with dexamethasone. After one cycle of temozolomide chemotherapy, intra- and intertumoral heterogeneity of TM formation and interconnection was strongly associated with therapy response: when tumor cells were integrated in TM networks, they were more likely to resist chemotherapy. CONCLUSION: TMs can contribute to the resistance against standard treatment modalities in gliomas. Specific inhibition of TMs is a promising approach to reduce local recurrence after surgery and lower resistance to chemotherapy.

Brain tumour cells interconnect to a functional and resistant network.

Astrocytic brain tumours, including glioblastomas, are incurable neoplasms characterized by diffusely infiltrative growth. Here we show that many tumour cells in astrocytomas extend ultra-long membrane protrusions, and use these distinct tumour microtubes as routes for brain invasion, proliferation, and to interconnect over long distances. The resulting network allows multicellular communication through microtube-associated gap junctions. When damage to the network occurred, tumour microtubes were used for repair. Moreover, the microtube-connected astrocytoma cells, but not those remaining unconnected throughout tumour progression, were protected from cell death inflicted by radiotherapy. The neuronal growth-associated protein 43 was important for microtube formation and function, and drove microtube-dependent tumour cell invasion, proliferation, interconnection, and radioresistance. Oligodendroglial brain tumours were deficient in this mechanism. In summary, astrocytomas can develop functional multicellular network structures. Disconnection of astrocytoma cells by targeting their tumour microtubes emerges as a new principle to reduce the treatment resistance of this disease.