Crowdsourcing the design of swarming nanoparticles for cancer applications

The treatment of cancer is undergoing what could be called a revolution. The field has attracted the attention of bioengineers trying to design nanoparticles that can deliver drugs and therapeutics directly to tumors. Their size, typically 10 nm to 500 nm, is ideal to leak out of porous vessels deep in tumors while remaining in the blood stream throughout the rest of the body. This allows nanoparticles to passively accumulate in tumors while reducing side effects on healthy tissue. This article describes the ongoing project of HFSP Cross-Disciplinary Fellow Sabine Hauert in the laboratory of Sangeeta Bhatia at MIT.

Sabine Hauert is a Human Frontier Science Program Cross-Disciplinary Fellow at the Koch Institute for Integrative Cancer Research at MIT. As a swarm engineer she aims to design large collective systems that self-organize. Swarm strategies are either inspired from nature (ant colonies and bird flocks) or are automatically designed in simulation using machine learning and crowdsourcing. Demonstrated applications include designing swarming nanoparticles for cancer treatment and deploying large aerial swarms for communication relay. Underlying her work is a strong involvement in science communication and online learning as demonstrated by her widely recognized blogs, tweets and podcasts.
Link to Sabine Hauert's website
Sangeeta Bhatia is the John J. and Dorothy Wilson Professor of Health Sciences and Technology and Electrical Engineering and Computer Science at MIT and a Howard Hughes Medical Investigator. Her lab combines engineering and biology to develop micro- and nanoscale platforms for understanding and treating human disease. Dr. Bhatia received her B.S. from Brown University, M.S. and Ph.D. from MIT, M.D. from Harvard and completed graduate and post-doctoral training at MGH.  Prior to MIT, Dr. Bhatia held a faculty position at UCSD, and worked at Pfizer, Genetics Institute, ICI Pharmaceuticals, and Organogenesis.  Dr. Bhatia has published >100 manuscripts and over 40 issued or pending patents, and has co-founded two biotechnology start-ups, consults for industry, government and academic organizations, and advocates for diversity in science and engineering.
Link to the Bhatia Lab

Nanoparticles come in different sizes, shapes and materials. They can be loaded with drugs that are released in a controlled fashion, or coated with molecules that allow them to interact with their environment. Certain molecules can serve as a signature to uniquely identify and internalize in cancer cells. Nanoparticles can also be made of energy-receptive materials that heat up upon magnetic or laser excitation.

There are many ways to design a nanoparticle. Depending on the design, the nanoparticle will move, sense and act in different ways in the tumor environment (Fig. 1). Control is embedded in the design of the nanoparticles and their interactions with the environment rather than their computational capabilities. In other words, changing the body of the nanoparticle will change its behavior: we call this embodied intelligence. The challenge is to understand which nanoparticle designs will improve treatment outcome. This is a difficult problem because trillions of nanoparticles typically interact in a tumor with millions of cells. Predicting and optimizing the emergent behavior of all these nanoparticles is guess work at best.

Fig. 1: Changing nanoparticle size, charge, coating, materials or cargos changes their behavior in the tumor environment.

To address this challenge, and following a systems approach, we designed a simulator that models how nanoparticles interact with each other and the tumor environment. The simulator generates reaction diffusion networks for the different nanoparticle designs based on realistic parameters from the literature and uses the stochastic simulation compiler designed at MIT. We focus on a representative area of the tumor instead of modeling the entire system. The hope is that if we chose the tumor scenario wisely, we will be able to generalize the results to the rest of the tumor. We've successfully used this computational framework in the past to discover guidelines to improve the penetration of nanoparticles deep in tumor tissue in a generalizable manner (Hauert et al. 2013, under review).

Like our nanoparticles, flocks of birds, ant colonies, cells and robot collectives can exhibit seemingly complex swarm behaviors when large numbers of simple agents react to local information. By design, swarms are efficient, robust and scalable. Emergent swarm behaviors useful for real-world applications include amplification, optimization, mapping, structure assembly, collective motion, synchronization and decision making. Our goal is to explore how nanoparticles can cooperate, or swarm, to synergistically improve their therapeutic effect.

Recent work by our laboratory, which uses nanoparticles that communicate in vivo to amplify tumor-homing, shows promise in this direction (von Maltzahn et al., Nature Materials, 2011). In this work, a near-infrared laser was used to heat gold nanoparticles that had passively accumulated in a tumor, thereby causing damage to the tissue. A second wave of nanoparticles were engineered to bind to the damaged tissue, and would therefore accumulate at higher numbers there. Similar to ants forming trails to a picnic table by depositing pheromones in the environment, these nanoparticles work by depositing and interacting with information in the tumor.

There are many such tumor scenarios and swarm strategies. Each one takes time to explore and requires large amounts of trial and error and human intuition. Furthermore, each problem is different, making it difficult to program a computer that can automatically design the nanoparticles. Instead, we've decided to make our simulator available through Nanodoc: an online game we developed to crowdsource the design of nanomedicine. Crowdsourcing has been shown in the past to bring unthought-of solutions to complex scientific problems such as protein folding ( As shown in Fig. 2, bioengineers can design their own tumor scenarios (right side) and submit them to the crowd. Players can then design different nanoparticle strategies (left side), and test them using our scientific simulator. The first levels of the game are used to train new NanoDocs; licensed NanoDocs are then given challenges to solve.

Fig. 2: Screenshot of the online game NanoDoc to crowdsource the design of nanomedicine.

During the first two months of being online, NanoDoc attracted nearly 15,000 visitors and was featured on mainstream media including Scientific American, New Scientist and The Guardian. Over 2600 players have now performed over 50,000 simulations with many earning a NanoDoc license that can be posted to social media. This demonstrates the desire of the crowd to learn about nanomedicine and help in the fight against cancer. The crowd was able to solve a first challenge aimed at detecting a rare event in a tumor environment (e.g. mutation, stem cell) and amplifying the signal using a cooperative two-nanoparticle strategy. Interestingly, this challenge required players to design treatments that were beyond what had been taught during training, thereby showing the ability of the crowd to think outside the box and adapt to new challenges.

In parallel to NanoDoc, we've been working on ways to validate nanoparticle designs in reality. This will require expert bioengineers to implement the actual nanoparticles. Selected nano-treatments discovered using NanoDoc will be validated using 1) in vitro tumor-on-a-chip constructs that we have designed to emulate the extravasation of functionalized nanoparticles from artificial vessels into a compartment containing tumor cells and 2) robotic swarm systems (kilobots) in collaboration with Radhika Nagpal's lab at the Wyss Institute at Harvard University.

:::::Desktop:TissueDevice.png :::::Desktop:Kilobots_black51.jpg

Fig. 3: Left: Tissue-on-a-chip construct designed to emulate the extravasation of nanoparticles (red) from artificial vessels into a compartment containing tumor cells (green). Right: Diffusing nanoparticle robots (green) binding to cell robots (red) within communication range and internalizing within the cell (purple). Experiments were performed using kilobots designed by Mike Rubenstein at the Wyss Institute, Harvard.

By HFSP Cross-Disciplinary Fellow Sabine Hauert


Hauert, S., Berman, S., Nagpal, R., Bhatia, SN. (2013) A computational framework for identifying design guidelines to increase the penetration of targeted nanoparticles into tumors. Submitted.

von Maltzahn, G, Park, J-H, Lin, KY, Singh, N, Schwöppe, C, Mesters, R, Berdel, WE, Ruoslahti, E, Sailor, MJ, Bhatia, SN (2011) Nanoparticles that communicate in vivo to amplify tumour targeting. Nature Materials, 10: 545-552.