At a classified meeting in Paris earlier this month, dozens of international delegates were handed plastic pipettes and asked to imagine doing one of the most basic tasks in genetic engineering: moving small, but exact volumes of liquid around. Collectively known as the Australia Group, the delegates from 43 mainly Western countries were meeting to discuss updates to their lists of restricted exports of materials and equipment that could potentially be misused to develop chemical and biological weapons. The plastic pipettes are not on these lists, and neither will they be. The question the group members were grappling with, rather, was how to deal with the modern, robotic equivalent to the old fashioned pipette—the new group of experimental biology platforms known as cloud labs.
While genetic engineering may sound exciting, the actual lab work involved to manipulate an organism’s genes is repetitive, painstaking, and tedious. Biological wet work, as it is called, is time-consuming, too. Pipetting, running gels, transfecting cells, growing bacteria, amplifying DNA, sequencing—it all takes time. Help, however, is at hand. Robotic machines have been developed to do many of these tasks, giving scientists more time to actually do science. And they no longer have to worry about manual mistakes or inaccurate measurements either.
But there’s a catch.
Robots are expensive. So much so, in fact, that 90 percent of the 4 million biologists around the world still do their pipetting by hand. But for large volumes of research work, the kinds pharma and biotech companies do, hand-pipetting is just not an option. Some companies, instead, invest in expensive robotic machines to support their work. And rather than buying their own lab hardware, an increasing number of companies are now outsourcing the work to so-called contract research organizations—third-party firms that are centralising and scaling the wet work of genetic engineering. One of them, the London DNA Foundry, at Imperial College London, for instance, can build and test about 15,000 different genetic designs in a day.
But even with various machines, like pipetting robots, to support the work, contract research organizations are still providing, essentially, a labor-intensive manual service. But as sophisticated AI and robotics begin to change how biology is done, these companies are facing competition from, of course, Silicon Valley tech firms.
Robotic cloud lab platforms. The self-proclaimed first robotic cloud lab for on-demand life science research was developed by a start-up company called Transcriptic. Founded in 2012, and backed by Google Ventures and the founders behind Pay Pal, the company now numbers 40 people and occupies a 22,000 square-foot facility in the heart of Silicon Valley. Here, the company builds and manages Plexiglas-enclosed robotic biology labs, or “workcells.” They’re containers about the same width and length as a parking space and that house about 20 devices each, including pipetting systems. The workcells are operated by computers, which receive experiment work orders and run the assemblage of machines. A robot on a gantry runs the length of the workcell, transferring plates from machine to machine to carry out the experiments.
To hire a robotic workcell, scientists sign up for an account with their name and email, provide some organizational information, and add payment details. A disclaimer on the company website notes that only commercial addresses are permitted; Transcriptic does not send samples to, or receive samples from, residential addresses.
Once scientists sign up, they access cloud lab services through a simple web interface. Transcriptic says its core mission is to turn “biology into information technology” and it’s at this stage where that happens. Each step of an experimental protocol is translated into machine-readable code. Transcriptic supplies scripted protocols for standard experiments, or offers to work with scientists to tailor-make experimental protocols. As one commentator noted: To carry out an experiment “you literally write [software] code. It really feels more like computing.” From each step of the experiment, scientists receive the data in real time on their computers, wherever they are.
There are multiple advantages to cloud labs, remote, real-time access being just one of them. The labs allow researchers to more easily reproduce results. They make documentation and standardization of the experimental process easier. Researchers can more efficiently analyze knowledge from all experiments stored on the cloud. All of this can translate to massively increased productivity.
These advantages seem reflected in demand for Transcriptic’s services. In 2017, for instance, 79 million microliters of fluid were pipetted by Transcriptic robots, 21 million microdroplets were transferred, 24,000 thermocycles were completed, and nearly 150,000 scientific operations were executed. User testimonials paint a similar picture. Barry Canton, co-founder of Ginkgo Bioworks (a $1-billion Boston-based organism foundry), said, “Transcriptic’s ability to translate an organism designers’ vision into reality via lab automation is unparalleled, and brings an unprecedented scale to our organism foundry.” Bret Huff, vice president of small molecule design and development at drug giant Eli Lilly, said, “We believe this capability will transform how new drugs are discovered internally and enhance Lilly’s partnership with external innovators.”
In the last couple of years, Transcriptic has entered into a five-year agreement with Ginkgo Bioworks valued at more than $10 million. Transcriptic and Ginkgo Bioworks have additionally been awarded a $9.5 million grant from the Department of Defense. In 2018, Transcriptic entered into a multi-year collaboration with Eli Lilly. This is all to say that money is certainly flowing into remote controlled wet labs, and the market is growing with companies like Emerald Cloud Laboratory, Riffyn, and Synthego offering similar services.
While these cloud labs can be seen as an evolution of outsourcing to contract research organizations, there is a clear difference. One CEO highlights, for instance, that cloud lab customers retain the ability to specify all the exact, low-level instructions on how to carry out experiments, while customers outsourcing to the older third-party labs are only able to give general directions to, say, synthesize a particular molecule or determine the toxicity of a given compound. There are at least two other more significant differences. The first is the lower cost of experimenting. Transcriptic, for instance, offers protocols such as cloning or mutagenesis, which involves creating mutations, at about the same cost or even less than what they cost at an academic lab. Their prices are about half of what a conventional contract research organization would charge. The second is privacy. When scientists outsource work to a contract research organization, they have to tell them what experiments they want done. Once they’ve secured access to a remote cloud lab, the service provider does not necessarily know what experiments are being run.
Both of these aspects, cost and privacy, have security implications.
Worries over cloud labs. Cloud labs give more people access to experimentation. Currently, most clients of Transcriptic and another platform, Synthace, are large pharmaceutical firms, but both claim they are working their way down to smaller users. The average experiment at Emerald Cloud Laboratory already costs only $25. In the investor technology forum Nanalyze, cloud labs are portrayed as making “an entire lab available to aspiring life scientists.”
Transcriptic’s laptop-and-credit-card approach to genetic engineering means that, as one prominent venture capitalist wrote in Forbes, scientists using cloud labs might not need PhDs or scientific degrees. The labs represent “citizen and crowd-performed science… where a large multiple of today’s existing number of researchers have an entire army of connected life science robots at their disposal to carry out experiments. Almost endless possibilities…are within reach,” Josh Wolfe wrote.
Cloud labs may offer up endless possibilities for experimentation, but not all research will necessarily be for good purposes. It is conceivable, for instance, to carry out experiments for one, or a few, point mutations that would confer resistance on a pathogen against therapeutically useful antibiotics; or that would alter the host range of a pathogen from one affecting, say, birds to one affecting, say, humans; or that would enhance the virulence of a pathogen or even render a non-pathogen virulent. While there may well be legitimate reasons to carry out these sorts of experiments, that legitimacy hinges on intent, and to evaluate intent, service providers would need to know what is going on in their workcells.
And that’s not all. Cloud labs lower barriers to experimentation. An article in the technology magazine IEEE Spectrum explained how cloud labs are “black-boxing” biology: “Just as most of us don’t care how a printer works, a scientist using these platforms doesn’t have to track (or care about) every detail of the experiment. The software does it for her, and saves it in a way that can be stored and shared with others.” Synthace’s CEO, Tim Fell, put it this way: “Say I want to assemble 40 genes into an organism. Then all of the complex details involved, be it liquid handling or using the equipment, just disappear from view.”
This doesn’t mean, however, that the need to understand the biology the robots are doing goes away. Biology is not just about programming the right moves and sequence of steps in a smart way. If an experiment doesn’t work as expected, trouble-shooting is often the trickiest and most time-consuming part. But what cloud labs do is obscure how to make equipment work properly or correctly perform techniques like pipetting.
Nature profiled one Transcriptic client, Justin Siegel. He heads a lab at the University of California, Davis that designs, builds, and tests new enzymes. His students used the cloud lab to build a biosensor that detects the chemical profile of olive oil. Using the lab, Siegel’s students became more efficient, but also “a little bolder,” he said. “Instead of making just 10 designs, they want to try a couple of extra. They’ll go a little bit farther out on a limb because all of a sudden they don’t have to physically build the stuff.”
Transcriptic and other cloud labs may be great for creativity and innovation, but there are certainly security questions that arise as scientists go that little bit farther out on a limb.
A third security concern is that the scale enabled by sophisticated AI and robotics provides a new avenue for misuse. Cloud lab investors draw parallels between genetic engineering and computer programming, saying, for instance: “With the advances being made in synthetic biology, it’s not hard to imagine a future where you can sit down and code your own synthetic organisms in the same way you can create a software program today.”
It’s sensible to retain some skepticism about how transformative cloud labs might actually be–particularly when many of the perspectives and portrayals of them come from the start-ups themselves as well as their investors and partners. But it’s equally prudent to take them seriously, and to think through possible security implications. Whether or not the technology lives up to all its promise, the bottom line remains: Cloud labs will make it easier for a larger range of people to use biology for malevolent purposes.
Reducing the risk from cloud labs. Discussions on cloud labs are at an early stage within the Australia Group, which helps member countries strengthen their compliance with the international chemical and biological weapons treaties. Member countries are still weighing potential policy options. One of the least intrusive of which (other than doing nothing) is to actively encourage cloud labs to self-regulate, much as the gene synthesis industry did a few years ago when it voluntarily implemented standardized customer and sequence screening practices. In terms of cloud labs, self-governance should encompass, at a minimum, customer-screening, controlled access to substances, experiment-screening for contextual understanding, as well as secured networks and firewalls.
A more direct policy option would be to introduce some form of export control. Yet, this would pose significant challenges to export regulators as a large portion of the control measures would have to cover ill-defined “intangibles.” While current export controls already capture electronic technology transfer directly associated with listed goods (such as equipment, organisms, and DNA sequences) through, for instance, email, phone, or video conferencing, cloud labs deal primarily in data generation and this would lift intangible technology transfer to a whole new level of complexity.
A third, more intermediate policy option would be to ensure adequate safety and security measures in all aspects of scientific research, that is, in chemical safety, biosafety, biosecurity, cyber security, etc. This would mean the privacy approach to experiments, where service providers do not necessarily know the content of the experiments carried out, would be unacceptable. This might be a hard sell to cloud labs, for whom it would involve more resources and higher costs, but they might be concerned enough about their reputations—which lab wants to be known as the place where some terrorist group developed a pandemic, for instance—to see the importance of doing so.
Without much oversight, cloud labs offer up the tools of genetic engineering to a broad array of people. Yes, they create a space potentially for more innovation, but also the opportunity for someone or group to exploit them for malevolent reasons. Although they could prove an incredibly useful bioengineering innovation, cloud labs also pose a perplexing risk—one that biosecurity experts are only beginning to tackle.