Biology and AWS: A Breakthrough in Cell Culture Monitoring
An AWS-based solution to count the number of cells in cell cultures in a laboratory
Published Nov 6, 2024
In collaboration with Dr. Igor Stefanini and Eng. Jacopo Pagnoncelli from the Institute of Digital Technologies for Personalized Healthcare – MeDiTech at DTI-SUPSI (Switzerland), I'm thrilled to share the progress of our innovative research project that marries biology with cutting-edge technology.
The main goal of cell cultures is to multiply the number of cells to harvest a large quantity for various purposes. For example, with a biopsy it’s possible to extract skin cells, then cultivate them for 3 to 4 weeks, to multiply them to 10,000 times their original amount. With this amount, it’s possible to perform a skin graft to heal scars, superficial wounds, or burns.
However, multiplying cells is not straightforward. Cell growth is not linear, meaning that the cell count increase is not constant every day. Numerous factors can alter the speed and disrupt the process, requiring the biologists to perform frequent inspections to check progress and make corrections.
In the simplest type of culture, the monolayer culture, cells lie on a flat plane in a Petri dish filled with a nourishing fluid (medium). This medium is a nourishing fluid that will support the growth of a small number of cells, initially placed in its center. As cells multiply from the center to the border, the cells in the center will be surrounded by other cells, and have less access to the media fluid, which contains nourishment.
This dish is placed in an incubator at a controlled temperature and humidity - mimicking a biological environment, to facilitate growth. There are some guidelines on how often to inspect, but no accepted standard and much variation. There are just too many factors. Therefore, the process heavily relies on human inspection.
The culture begins with a certain number of cells, which multiply exponentially but are adjusted by the rate of nutrient depletion, until they occupy all space available, filling the dish. This is the correct time to subculture, which means to remove the dish and split the harvested cells into 2 or 3 parts, with new Petri dishes and media, and repeat the growth process.
Another reason to perform a maintenance event is to change media during the multiplication process because the nourishment substances have been consumed. In addition to nourishment, the media may contain growth factors, which are molecules that signal the cells to grow. These growth factors can also deplete. As a result, first, the cells slow down their growth, then they may reach quiescence, meaning they stay idle for a time, without growing or multiplying. After too long spent in the quiescent state, eventually, they begin to die.
From this, you may think that a frequent inspection is a good idea. However, inspections are disruptive. Not only do they consume time for the biologist, but they consist of removing the cell from the incubator, the environment with the ideal temperature and humidity. The outside laboratory has a different temperature. Too frequent useless maintenance slows down the growth curve and even results in the death of a certain amount of cells.
However, waiting too long before an inspection is also wrong. Frequently, the culture will reach its maximum capacity, with a full dish, and just stop growing. A delayed inspection delays subculturing and results in a dead time. Additionally, as explained, the growth will slow down as the media is consumed. In any case, after entering quiescence, cells may begin to die. The goal is to hasten the growth as much as possible because we need the harvested cells for clinical application or research. As an example, a patient may be waiting for a skin graft, or the lab may have to postpone or halt some environmental research on bacteria that produce electricity or purify water.
Recognizing the inefficiency of the inspection process, we aimed to optimize it, by inspecting more efficiently, allowing the process to be properly controlled while minimizing losses due to inspection itself and the delays. Research is a long process and we can’t wait for these additional, avoidable, dead times. Our breakthrough involves leveraging bioimpedance as an indicator of cell growth.
When cells are subjected to an electrical field, they act as resistors, and their bioimpedance increases with cell count. This means the electrical resistance increases when the cell count increases. The exact bioimpedance curve depends on the type of cell culture, the type of cells, the media, and the temperature and humidity in the incubator. Although the relationship is non-linear, precise calibration allows for accurate measurements and results. The calibration is a one-off operation, compared to an inspection that must be repeated indefinitely. In other words, once a particular culture is calibrated, a measurement of resistance across the cells can provide a highly accurate model of cell growth, and this resistance measurement can be achieved without disrupting the growth process.
Given the observed variation, with proper optimization, we aim to make the process at least twice as efficient. Imagine a laboratory performing twice the research, with the same infrastructure, staff, and funds. Very interesting, isn’t it?
We place all Petri dishes with electrodes and check, at intervals of time, the bioimpedance, adjusted by the calibration parameters. At this point, it will be possible, without opening the incubator, to know when the growth has stopped, because:
- The media has depleted nourishment and needs changing
- The culture has reached full capacity and is ready for harvesting or subculturing
- Some anomaly or mutation
So the biologist can intervene only when necessary, but, also, as soon as it’s necessary.
It is our goal to ease the life of lab biologists, not to add extra work. Therefore, the simplest solution wins. An application with complex operations is definitely not our intent.
The computational needs of this project are fairly straightforward. For each cell culture, we need to:
- Take a resistance reading at regular intervals
- Authenticate with Amazon
- Do post-processing of the data
- Insert into a database
The AWS cloud comes as a great help for this project. It is in fact possible, with a serverless architecture, to serve requests from a web client that preprocesses observations to remove noise using a gliding window and sends them in batches. To accomplish this goal with AWS, we will be using 4 main services: API Gateway to route the request from the web client, AWS Cognito for authentication, AWS Lambda for post-processing computation, and DynamoDB for storing the data.
Every time we measure the bioimpedance, there is a random noise due to the electronics on the board, which average is zero. For example, we take 5 samples using the hardware components, in a short time, and return a single one with their average. This decreases the error and cleans the signal.
After a minute or so, we repeat the operation and collect another sample. Once we have enough of these averaged samples, we then batch them together and send them to AWS API Gateway, backed by a Lambda. The Lambda stores this data on DynamoDB, because it scales well vertically (with large numbers of records), being crowded with observations. Secondary indexes - global and local - help with all query patterns needed.
Additionally, the flexible schema means we can add all flags and fields necessary, as we figure them thanks to the feedback from biologists. Part of the success of a technical solution is the foresight in making decisions and reducing rework, but another part is successfully integrating feedback from actual users, which is impossible to foresee. DynamoDB affords the flexibility to evolve the schema on the fly, without any migration.
Using the web client, the biologist can add the first-class data entities (monitoring platforms, electrodes, and the cell cultures themselves) for which the observations will be collected. This data is relational; however, we opted to fit it in DynamoDB, because, even if we need extra code to guarantee the consistency of every operation, the global overall architecture is still simpler than running different databases.
Today the project is in a testing phase. We are calibrating the curves and figuring out the exact average, considering that an excessive number of impulses may interfere with the cell growth and consume the battery faster. Eventually, we plan to replace batteries with a more sustainable power source. The idea is for this measuring platform to be compatible with all incubators, so we can’t, for example, drill a hole, or assume that we can always fit a power cord inside the incubator. Hence, we started with a battery, so that we may place the entire device inside the incubator and close it. We wrap the electronics in resin to isolate them from the high humidity. We still have many details to figure out and have been wrapping our heads around a multitude of problems.
We expect the project to go live this year and we are excited to share further progress and some photos. There are many applications on our agenda. For example, a Swiss institute for oncology research will likely be an early adopter of this system, for cancer research.
My journey into the biomedical field began thanks to the AWS Community Builders program. Last year, they organized an AI hackathon and I participated with fellow community builder Govind Kumar. Although we didn’t win, this experience ignited a new passion in me, leading me to pursue further studies and dedicate myself to this exciting field.
I am deeply grateful to AWS for providing the ideal platform to support our project and advance laboratory biotechnology. The flexibility and efficiency of AWS's cloud services have been instrumental in our success. This project would have faced numerous challenges and delays with an on-premises solution, especially due to compliance and other logistical issues. Thanks also for the credits, which further removed constraints and sped up work. Instead of having to justify and have a budget approved, we could begin immediately.
On the technical front, AWS’s implementation of serverless architecture is unparalleled in the current cloud landscape. It has proven to be the perfect fit for our needs. Additionally, DynamoDB has become my preferred database due to its flexibility and suitability for various applications, as the perfect fit for this case further shows.
Thank you, AWS, for enabling us to push the boundaries of biomedical research and innovation.