Jing-Yu Lai

PURPOSE: Angiopoietin-1 (Ang1) plays a key role in maintaining stable vasculature, whereas in a tumor Ang2 antagonizes Ang1's function and promotes the initiation of the angiogenic switch. Specifically targeting Ang2 is a promising anticancer strategy. Here we describe the development and characterization of a new class of biotherapeutics referred to as CovX-Bodies, which are created by chemical fusion of a peptide and a carrier antibody scaffold. EXPERIMENTAL DESIGN: Various linker tethering sites on peptides were examined for their effect on CovX-Body in vitro potency and pharmacokinetics. Ang2 CovX-Bodies with low nmol/L IC(50)s and significantly improved pharmacokinetics were tested in tumor xenograft studies alone or in combination with standard of care agents. Tumor samples were analyzed for target engagement, via Ang2 protein level, CD31-positive tumor vasculature, and Tie2 expressing monocyte penetration. RESULTS: Bivalent Ang2 CovX-Bodies selectively block the Ang2-Tie2 interaction (IC(50) < 1 nmol/L) with dramatically improved pharmacokinetics (T(½) > 100 hours). Using a staged Colo-205 xenograft model, significant tumor growth inhibition (TGI) was observed (40%-63%, P < 0.01). Ang2 protein levels were reduced by approximately 50% inside tumors (P < 0.01), whereas tumor microvessel density (P < 0.01) and intratumor proangiogenic Tie2(+)CD11b(+) cells (P < 0.05) were significantly reduced. When combined with sunitinib, sorafenib, bevacizumab, irinotecan, or docetaxel, Ang2 CovX-Bodies produced even greater efficacy (∼80% TGI, P < 0.01). CONCLUSION: CovX-Bodies provide an elegant solution to overcome the pharmacokinetic-pharmacodynamic problems of peptides. Long-acting Ang2 specific CovX-Bodies will be useful as single agents and in combination with standard-of-care agents.

Bispecific antibodies (BsAbs) are regarded as promising therapeutic agents due to their ability to simultaneously bind two different antigens. Several bispecific modalities have been developed, but their utility is limited due to problems with stability and manufacturing complexity. Here we report a versatile technology, based on a scaffold antibody and pharmacophore peptide heterodimers, that enables rapid generation and chemical optimization of bispecific antibodies, which are termed bispecific CovX-Bodies. Two different peptides are joined together using a branched azetidinone linker and fused to the scaffold antibody under mild conditions in a site-specific manner. Whereas the pharmacophores are responsible for functional activities, the antibody scaffold imparts long half-life and Ig-like distribution. The pharmacophores can be chemically optimized or replaced with other pharmacophores to generate optimized or unique bispecific antibodies. As a prototype, we developed a bispecific antibody that binds both vascular endothelial growth factor (VEGF) and angiopoietin-2 (Ang2) simultaneously, inhibits their function, shows efficacy in tumor xenograft studies, and greatly augments the antitumor effects of standard chemotherapy. This unique antiangiogenic bispecific antibody is in phase-1 clinical trials.

CVX-045 is produced by covalently attaching a thrombospondin 1 (TSP-1) mimetic comprising a peptidic sequence and a linker to the Fab binding site of a proprietary scaffold antibody. CVX-045 possesses the potency of the TSP-1-derived peptide, along with the advantageous pharmacokinetics of an antibody. Antitumor activity of CVX-045 was evaluated in human xenograft models alone and in combination with standard chemotherapies and targeted molecules. In A549 and A431 xenograft models, CVX-045 demonstrated significant (P < .05) antiangiogenic activity, reducing tumor microvessel density and increasing the levels of necrosis within treated tumors. In an HT-29 xenograft model, CVX-045 in combination with 5-fluorouracil significantly (P < .01) decreased tumor growth rate compared with vehicle, CVX-045, or 5-fluorouracil alone. Cotreatment of CVX-045 plus CPT-11 delayed progression of tumor growth from day 28 to 60. In contrast CVX-045 alone treatment did not delay the progression of tumor growth, and CPT-11 alone delayed progression of tumor growth to day 39. Cotreatment of CVX-045 with sunitinib extended the time to reach tumor load from day 26 to 40. In summary, CVX-045 exhibits significant antiangiogenic activity in several tumor models and enhances antitumor activity in combination with chemotherapy or targeted therapies. These data suggest future avenues for effective combination therapy in treating solid tumors. CVX-045 has recently completed a phase 1 trial in solid tumors where it has been well tolerated.

Nowadays, the grinding process is mostly automatic, yet post-grinding quality inspection is mostly carried out manually. Although the conventional inspection technique may have cumbersome setup and tuning processes, the data-driven model, with its vision-based dataset, provides an opportunity to automate the inspection process. In this study, a convolutional neural network technique with transfer learning is proposed for three kinds of inspections based on 750-1000 surface raw images of the ground workpieces in each task: classifying the grit number of the abrasive belt that grinds the workpiece, estimating the surface roughness of the ground workpiece, and classifying the degree of wear of the abrasive belts. The results show that a deep convolutional neural network can recognize the texture on the abrasive surface images and that the classification model can achieve an accuracy of 0.9 or higher. In addition, the external coaxial white light was the most suitable light source among the three tested light sources: the external coaxial white light, the high-angle ring light, and the external coaxial red light. Finally, the model that classifies the degree of wear of the abrasive belts can also be utilized as the abrasive belt life estimator.

Surface roughness plays an important role in grinding; it can represent the grinding quality of machined parts. In previous research, analytical models and empirical models have been used to predict surface roughness. This research presented surface roughness prediction models based on linear regression and artificial neural networks of several types of model structures, then applied different features as model inputs, including force data directly get from the force sensor and those collected force data after statistical processing to reduce dimension. After conducting the prediction model, a self-developed grinding machine was used to collect the force data for model training and testing, and the mean absolute percentage error was used to evaluate the prediction performance. In the end, a neural network of three hidden layers was marked as the best model, which was useful for surface roughness prediction during grinding.