Dysregulated Myc signaling is a key oncogenic pathway in glioblastoma multiforme (GBM). Yet, effective therapeutic targeting of Myc continues to be challenging. Here, we demonstrate that exosomes generated from human bone marrow mesenchymal stem cells (MSCs) engineered to encapsulate siRNAs targeting Myc (iExo-Myc) localize to orthotopic GBM tumors in mice. Treatment of late stage GBM tumors with iExo-Myc inhibits proliferation and angiogenesis, suppresses tumor growth, and extends survival. Transcriptional profiling of tumors reveals that the mesenchymal transition and estrogen receptor signaling pathways are impacted by Myc inhibition. Single nuclei RNA sequencing (snRNA-seq) shows that iExo-Myc treatment induces transcriptional repression of multiple growth factor and interleukin signaling pathways, triggering a mesenchymal to proneural transition and shifting the cellular landscape of the tumor. These data confirm that Myc is an effective anti-glioma target and that iExo-Myc offers a feasible, readily translational strategy to inhibit challenging oncogene targets for the treatment of brain tumors.

The subpopulations of tumor pericytes undergo pathological phenotype switching, affecting their normal function in upholding structural stability and cross-communication with other cells. In the case of pancreatic ductal adenocarcinoma (PDAC), a significant portion of blood vessels are covered by an α-smooth muscle actin (αSMA)-expressing pericyte, which is normally absent from capillary pericytes. The Desmin^lowαSMA^high phenotype was significantly correlated with intratumoral hypoxia and vascular leakiness. Using an in vitro co-culture system, we demonstrated that cancer cell-derived exosomes could induce ectopic αSMA expression in pericytes. Exosome-treated αSMA⁺ pericytes presented altered pericyte markers and an acquired immune-modulatory feature. αSMA⁺ pericytes were also linked to morphological and biomechanical changes in the pericyte. The PDAC exosome was sufficient to induce αSMA expression by normal pericytes of the healthy pancreas in vivo, and the vessels with αSMA⁺ pericytes were leaky. This study demonstrated that tumor pericyte heterogeneity could be dictated by cancer cells, and a subpopulation of these pericytes confers a pathological feature.

Tumor-derived extracellular vesicles (EVs) show great potential as biomarkers for several diseases, including pancreatic cancer, due to their roles in cancer development and progression. However, the challenge of utilizing EVs as biomarkers lies in their inherent heterogeneity in terms of size and concentration, making accurate quantification difficult, which is highly dependent on the isolation and quantification methods used. In our study, we compared three EV isolation techniques and two EV quantification methods. We observed variations in EV concentration, with approximately 1.5-fold differences depending on the quantification method used. Interestingly, all EV isolation techniques consistently yielded similar EV quantities, overall size distribution, and modal sizes. In contrast, we found a notable increase in total EV amounts in samples from pancreatic cancer cell lines, mouse models, and patient plasma, compared to non-cancerous conditions. Moreover, individual tumor-derived EVs exhibited at least a 3-fold increase in several EV biomarkers. Our data, obtained from EVs isolated using various techniques and quantified through different methods, as well as originating from various pancreatic cancer models, suggests that EV profiling holds promise for the identification of unique and cancer-specific biomarkers in pancreatic cancer.

This paper proposes a reinforcement learning approach to improve the performance of the random access procedure in the massive machine type communication (mMTC) environment, as defined by 3GPP, to accommodate an increasing number of user equipment (UEs). The existing random access process results in channel congestion when UEs that have failed to transmit due to a fixed backoff indicator choose a subsequent random access opportunity (RAO) slot. To address this, we introduce an agent that automatically adjusts the backoff indicator based on the state of the network environment. As a result of simulations using Q-learning and DDPG, DDPG achieves a higher success rate and reward, but the backoff indicator continually increases. In contrast, Q-learning exhibits relatively lower performance, yet the stability of the backoff indicator is maintained. From these findings, it is evident that the integration of reinforcement learning to optimize the random access procedure enhances performance. However, it necessitates careful algorithm selection and precise tuning of the reward mechanism.