Extracellular vesicles (EVs), including exosomes and microvesicles derived from different cell sources, are used as promising nanovesicles for delivering therapeutic microRNAs (miRNAs) and drugs in cancer therapy. However, their clinical translation is limited by the quantity, size heterogeneity, and drug or small RNA loading efficiency. Herein, we developed a scalable microfluidic platform that can load therapeutic miRNAs (antimiRNA-21 and miRNA-100) and drugs while controlling the size of microfluidically processed EVs (mpEVs) using a pressure-based disruption and reconstitution process. We prepared mpEVs of optimal size using microvesicles isolated from neural stem cells engineered to overexpress CXCR4 receptor and characterized them for charge and miRNA loading efficiency. Since the delivery of therapeutic miRNAs to brain cancer is limited by the blood-brain barrier (BBB), we adopted intranasal administration of miRNA-loaded CXCR4-engineered mpEVs in orthotopic GBM mouse models and observed a consistent pattern of mpEVs trafficking across the nasal epithelia, bypassing the BBB into the intracranial compartment. In addition, the CXCR4-engineered mpEVs manifested selective tropism toward GBMs by stromal-derived factor-1 chemotaxis to deliver their miRNA cargo. The delivered miRNAs sensitized GBM cells to temozolomide, resulting in prominent tumor regression, and improved the overall survival of mice. A simple and efficient approach of packaging miRNAs in mpEVs using microfluidics, combined with a noninvasive nose-to-brain delivery route presents far-reaching potential opportunities to improve GBM therapy in clinical practice.

Background: The accurate classification between malignant and benign breast lesions detected on mammograms is a crucial but difficult challenge for reducing false-positive recall rates and improving the efficacy of breast cancer screening. Objective: This study aims to optimize a new deep transfer learning model by implementing a novel attention mechanism in order to improve the accuracy of breast lesion classification. Methods: ResNet50 is selected as the base model to develop a new deep transfer learning model. To enhance the accuracy of breast lesion classification, we propose adding a convolutional block attention module (CBAM) to the standard ResNet50 model and optimizing a new model for this task. We assembled a large dataset with 4280 mammograms depicting suspicious soft-tissue mass-type lesions. A region of interest (ROI) is extracted from each image based on lesion center. Among them, 2480 and 1800 ROIs depict verified benign and malignant lesions, respectively. The image dataset is randomly split into two subsets with a ratio of 9:1 five times to train and test two ResNet50 models with and without using CBAM. Results: Using the area under ROC curve (AUC) as an evaluation index, the new CBAM-based ResNet50 model yields AUC = 0.866 ± 0.015, which is significantly higher than that obtained by the standard ResNet50 model (AUC = 0.772 ± 0.008) (p < 0.01). Conclusion: This study demonstrates that although deep transfer learning technology attracted broad research interest in medical-imaging informatic fields, adding a new attention mechanism to optimize deep transfer learning models for specific application tasks can play an important role in further improving model performances.

New precision medicine drugs oftentimes act through binding to specific cell-surface cancer receptors, and thus their efficacy is highly dependent on the availability of those receptors and the receptor concentration per cell. Paired-agent molecular imaging can provide quantitative information on receptor status in vivo, especially in tumor tissue; however, to date, published approaches to paired-agent quantitative imaging require that only “trace” levels of imaging agent exist compared to receptor concentration. This strict requirement may limit applicability, particularly in drug binding studies, which seek to report on a biological effect in response to saturating receptors with a drug moiety. To extend the regime over which paired-agent imaging may be used, this work presents a generalized simplified reference tissue model (GSRTM) for paired-agent imaging developed to approximate receptor concentration in both non-receptor-saturated and receptor-saturated conditions. Extensive simulation studies show that tumor receptor concentration estimates recovered using the GSRTM are more accurate in receptor-saturation conditions than the standard simple reference tissue model (SRTM) (% error (mean ± sd): GSRTM 0 ± 1 and SRTM 50 ± 1) and match the SRTM accuracy in non-saturated conditions (% error (mean ± sd): GSRTM 5 ± 5 and SRTM 0 ± 5). To further test the approach, GSRTM-estimated receptor concentration was compared to SRTM-estimated values extracted from tumor xenograft in vivo mouse model data. The GSRTM estimates were observed to deviate from the SRTM in tumors with low receptor saturation (which are likely in a saturated regime). Finally, a general “rule-of-thumb” algorithm is presented to estimate the expected level of receptor saturation that would be achieved in a given tissue provided dose and pharmacokinetic information about the drug or imaging agent being used, and physiological information about the tissue. These studies suggest that the GSRTM is necessary when receptor saturation exceeds 20% and highlight the potential for GSRTM to accurately measure receptor concentrations under saturation conditions, such as might be required during high dose drug studies, or for imaging applications where high concentrations of imaging agent are required to optimize signal-to-noise conditions. This model can also be applied to PET and SPECT imaging studies that tend to suffer from noisier data, but require one less parameter to fit if images are converted to imaging agent concentration (quantitative PET/SPECT).

Immuno-oncological treatment strategies that target abnormal receptor profiles of tumors are an increasingly important feature of cancer therapy. Yet, assessing receptor availability (RA) and drug-target engagement, important determinants of therapeutic efficacy, is challenging with current imaging strategies, largely due to the complex nonspecific uptake behavior of imaging agents in tumors. Herein, we evaluate whether a quantitative noninvasive imaging approach designed to compensate for nonspecific uptake, MRI-coupled paired-agent fluorescence tomography (MRI-PAFT), is capable of rapidly assessing the availability of epidermal growth factor receptor (EGFR) in response to one dose of anti-EGFR antibody therapy in orthotopic brain tumor models. Methods: Mice bearing orthotopic brain tumor xenografts with relatively high EGFR expression (U251) (N=10) or undetectable human EGFR (9L) (N=9) were considered in this study. For each tumor type, mice were either treated with one dose of cetuximab, or remained untreated. All animals were scanned using MRI-PAFT, which commenced immediately after paired-agent administration, and values of RA were recovered using a model-based approach, which uses the entire dynamic sequence of agent uptake, as well as a simplified “snapshot” approach which requires uptake measurements at only two time points. Recovered values of RA were evaluated between groups and techniques. Hematoxylin & eosin (H&E) and immunohistochemical (IHC) staining was performed on tumor specimens from every animal to confirm tumor presence and EGFR status. Results: In animals bearing EGFR(+) tumors, a significant difference in RA values between treated and untreated animals was observed (RA = 0.24 ± 0.15 and 0.61 ± 0.18, respectively, p=0.027), with an area under the curve - receiver operating characteristic (AUC-ROC) value of 0.92. We did not observe a statistically significant difference in RA values between treated and untreated animals bearing EGFR(-) tumors (RA = 0.18 ± 0.19 and 0.27 ± 0.21, respectively; p = 0.89; AUC-ROC = 0.55), nor did we observe a difference between treated EGFR(+) tumors compared to treated and untreated EGFR(-) tumors. Notably, the snapshot paired-agent strategy quantified drug-receptor engagement within just 30 minutes of agent administration. Examination of the targeted agent alone showed no capacity to distinguish tumors either by treatment or receptor status, even 24h after agent administration. Conclusions: This study demonstrated that a noninvasive imaging strategy enables rapid quantification of receptor availability in response to therapy, a capability that could be leveraged in preclinical drug development, patient stratification, and treatment monitoring.

No abstract