Advances in nanoparticle synthesis and engineering have produced nanoscale agents affording both therapeutic and diagnostic functions that are often referred to by the portmanteau ‘nanotheranostics’. The field is associated with many applications in the clinic, especially in cancer management. These include patient stratification, drug-release monitoring, imaging-guided focal therapy and post-treatment response monitoring. Recent advances in nanotheranostics have expanded this notion and enabled the characterization of individual tumours, the prediction of nanoparticle–tumour interactions, and the creation of tailor-designed nanomedicines for individualized treatment. Some of these applications require breaking the dogma that a nanotheranostic must combine both therapeutic and diagnostic agents within a single, physical entity; instead, it can be a general approach in which diagnosis and therapy are interwoven to solve clinical issues and improve treatment outcomes. In this Review, we describe the evolution and state of the art of cancer nanotheranostics, with an emphasis on clinical impact and translation.

Magnetic resonance imaging (MRI) is one of the most widely used diagnostic tools in the clinic. To improve imaging quality, MRI contrast agents, which can modulate local T1 and T2 relaxation times, are often injected prior to or during MRI scans. However, clinically used contrast agents, including Gd3+-based chelates and iron oxide nanoparticles (IONPs), afford mediocre contrast abilities. To address this issue, there has been extensive research on developing alternative MRI contrast agents with superior r1 and r2 relaxivities. These efforts are facilitated by the fast progress in nanotechnology, which allows for preparation of magnetic nanoparticles (NPs) with varied size, shape, crystallinity, and composition. Studies suggest that surface coatings can also largely affect T1 and T2 relaxations and can be tailored in favor of a high r1 or r2. However, the surface impact of NPs has been less emphasized. Herein, we review recent progress on developing NP-based T1 and T2 contrast agents, with a focus on the surface impact.

Delivery of nanoparticle drugs to tumors relies heavily on the enhanced permeability and retention (EPR) effect. While many consider the effect to be equally effective on all tumors, it varies drastically among the tumors’ origins, stages, and organs, owing much to differences in vessel leakiness. Suboptimal EPR effect represents a major problem in the translation of nanomedicine to the clinic. In the present study, we introduce a photodynamic therapy (PDT)-based EPR enhancement technology. The method uses RGD-modified ferritin (RFRT) as “smart” carriers that site-specifically deliver 1O2 to the tumor endothelium. The photodynamic stimulus can cause permeabilized tumor vessels that facilitate extravasation of nanoparticles at the sites. The method has proven to be safe, selective, and effective. Increased tumor uptake was observed with a wide range of nanoparticles by as much as 20.08-fold. It is expected that the methodology can find wide applications in the area of nanomedicine.

Macrophages hold great potential in cancer drug delivery because they can sense chemotactic cues and home to tumors with high efficiency. However, it remains a challenge to load large amounts of therapeutics into macrophages without compromising cell functions. Here we report a silica-based drug nanocapsule approach to solve this issue. Our nanocapsule consists of a drug-silica complex filling and a solid silica sheath, and it is designed to minimally release drug molecules in the early hours of cell entry. While taken up by macrophages at high rates, the nanocapsules minimally affect cell migration in the first 6–12 h, buying time for macrophages to home to tumors and release drugs in situ. In particular, we show that doxorubicin (Dox) as a representative drug can be loaded into macrophages up to 16.6 pg/cell using this approach. When tested in a U87MG xenograft model, intravenously (i.v.) injected Dox-laden macrophages show comparable tumor accumulation as untreated macrophages. Therapy leads to efficient tumor growth suppression, while causing little systematic toxicity. Our study suggests a new cell platform for selective drug delivery, which can be readily extended to the treatment of other types of diseases.

Nanoprobes for MRI and optical imaging are demonstrated. Gd@C‐dots possess strong fluorescence and can effectively enhance signals on T1‐weighted MR images. The nanoprobes have low toxicity, and, despite a relatively large size, can be efficiently excreted by renal clearance from the host after systemic injection.

No abstract

Adam and RMSProp are two of the most influential adaptive stochastic algorithms for training deep neural networks, which have been pointed out to be divergent even in the convex setting via a few simple counterexamples. Many attempts, such as decreasing an adaptive learning rate, adopting a big batch size, incorporating a temporal decorrelation technique, seeking an analogous surrogate, etc., have been tried to promote Adam/RMSProp-type algorithms to converge. In contrast with existing approaches, we introduce an alternative easy-to-check sufficient condition, which merely depends on the parameters of the base learning rate and combinations of historical second-order moments, to guarantee the global convergence of generic Adam/RMSProp for solving large-scale non-convex stochastic optimization. Moreover, we show that the convergences of several variants of Adam, such as AdamNC, AdaEMA, etc., can be directly implied via the proposed sufficient condition in the non-convex setting. In addition, we illustrate that Adam is essentially a specifically weighted AdaGrad with exponential moving average momentum, which provides a novel perspective for understanding Adam and RMSProp. This observation coupled with this sufficient condition gives much deeper interpretations on their divergences. At last, we validate the sufficient condition by applying Adam and RM-SProp to tackle a certain counterexample and train deep neural networks. Numerical results are exactly in accord with our theoretical analysis.

No abstract