How Do We Know If a Virus Is Bioengineered?
Almost as soon as the coronavirus appeared in the news, so too did speculation that it was purposefully engineered, the result of experimentation at one of several Wuhan laboratories. The idea that the virus, whether natural or engineered, came from a scientific facility was pushed by some politicians. The White House reportedly pressured spy agencies to look into lab links.
Most scientists agree, based on the virus’s genetics, that it probably hopped from animals to humans. On April 30, the U.S. Office of the Director of National Intelligence declared, on behalf of the 17 different organizations that make up the U.S. intelligence community, that “the Covid-19 virus was not manmade or genetically modified.” The organizations decided to continue investigating two alternatives: the more likely explanation that the virus jumped from an animal to a human, and the more remote possibility that it was a natural virus released in a lab accident, which still hasn’t been ruled out.
So, the U.S.’s spy sector “concurs with the wide scientific consensus,” as the statement put it, that the virus wasn’t created by people. But how did its people come to that conclusion? While the full scope of its investigation isn’t known, one program within the intelligence community, FELIX, did specifically investigate the hypothesis. FELIX’s analysis revealed that the virus hadn’t been engineered using “foreign” genetic sequences, indicating that SARS-CoV-2, the virus that causes Covid-19, was not man-made or engineered using pieces of other organisms.
But detecting “bioengineering” is a fraught task for any organism. Just as there are many ways to determine whether a virus was engineered, there are many ways to engineer a virus, leading to a constant tug of war — and a lot of uncertainty.
FELIX stands for Finding Engineering-Linked Indicators, and it’s run by IARPA, the Intelligence Advanced Research Projects Activity. IARPA does high-risk research and develops next-next-gen technology under the Office of the Director of National Intelligence. In 2018, FELIX began funding six external teams to develop tools that can detect the fingerprints of bioengineering. These genetic signs are clear indications that someone messed around with an organism’s genome.
A genome is the full list of genetic bases that make up an organism. In DNA, those bases are A, G, C, and T; in RNA, they’re A, G, C, and U. Strung together, they make up “sequences,” which can either refer to all the letters, in order, that describe an organism, or a smaller subset.
Within a genome, the fingerprints of engineering can take a few forms, according to FELIX program manager David Markowitz, PhD. They can appear as foreign genetic material in a given sequence, or the off-kilter duplication, insertion, or deletion of bases. Other flags, says Isaac Plant, PhD, who worked on FELIX as a graduate student at Harvard, include sequences known to encode antibiotic resistance and short sequences called “scars,” which “show a change was made to a DNA sequence.” There’s no totally comprehensive list of “engineering sequences” like the ones Plant describes, but services like Addgene, which provide the molecular tools to manipulate DNA in the lab, host big databases.
FELIX’s tools could determine whether someone has stolen biological IP from someone else — like if a custom yeast strain from one company appeared in a competitor’s lab — and investigate the naturalness of new germs. In SARS-CoV-2, an RNA virus, FELIX got its first big real-world test.
“FELIX [teams] spent 18 months developing the first working prototypes of their engineering detection platforms,” says Markowitz. “They were poised to work on SARS-CoV-2 when this biosecurity threat first emerged.”
And so in January, a team from the MIT-Broad Foundry deployed its FELIX tools to “test the veracity of online stories claiming that SARS-CoV-2 was engineered in a laboratory,” according to IARPA’s website. Though the results of that test aren’t fully public, a pop-up on the page says that the system compared the virus’s genome to 58 million known genetic sequences — including “genomes from closely and distantly related viruses.” In 10 minutes, the tool determined that the virus’s makeup matched those of naturally occurring coronaviruses better than any other organisms: “This analysis indicates that no sequences from foreign species have been engineered into SARS-CoV-2,” IARPA wrote.
It sounds definitive. “But that actually doesn’t rule out engineering,” says Filippa Lentzos, PhD, a senior research fellow at King’s College London focused on biosecurity. It just means that the virus wasn’t engineered in specific ways.
The MIT-Broad team that worked on SARS-CoV-2 declined an interview. Not much is known about what’s behind their curtain, but other FELIX-funded teams were more willing to share.
Eric Young, PhD, studies yeast engineering at Worcester Polytechnic Institute, but he started thinking more about biosecurity when he noticed policy folks and government ethicists at synthetic biology conferences. They were concerned about the implications of being able to build custom organisms so easily.
“What we’re creating here are things that are going to be great for civilization,” like developing new pharmaceuticals, Young says. “But, as the history of every technology has shown, a lot of times, over the course of human history, those technological advances have been repurposed for weapons.” To date, he adds, we don’t have any examples of someone using synthetic biology to create a bioengineered weapon.
The tool Young worked on for FELIX first sequences an organism’s whole genome. Then, it delivers those letters back to the scientist with annotations, calling out which genetic parts, if any, seem to have been tinkered with. It checks for engineering by comparing the sequence against a standard list of yeast engineering sequences or a custom sequence list that the user inputs. “Future versions will use machine learning to flag DNA sequences as engineered without needing a list of parts,” says Young. For now, it works with yeast sequences.
Ginkgo Bioworks, another FELIX awardee, is coming at the problem using computational biology and its own database of known engineered sequences. The company often sits on the other side of the equation: designing and engineering organisms itself. “For this program alone, we created over 6 million simulated, engineered genomes that reflect a variety of genetic engineering techniques and design styles, across a panel of organisms chosen by IARPA and the national labs,” says Joshua Dunn, PhD, the company’s head of design.
“We really wanted to be the white-hat hackers at this table,” Dunn continues. To hack thusly, Ginkgo uses several different methods. First, they compare a sequence to known, natural reference sequences, like the MIT-Broad Foundry seems to have done for SARS-CoV-2. Then, they look for known engineering signatures in that same input sequence. A third tack analyzes the distribution of its genetic alphabet. Finally, all that information flows into a fusion engine the team is working on that assesses at all three results in tandem to come up with a final conclusion.
At the Draper Laboratory, a FELIX group led by Kirsty McFarland, PhD, has focused on a different aspect of detection: how to pick out engineered organisms from environmental samples that are chock-full of life, like soil or water. They’re developing separate methods aiming to pinpoint two sorts of bioengineering: unknown change to a known organism, and known or suspected change to a potentially unknown organism. SARS-CoV-2, being a new pathogen, falls into the latter category.
The first method works by comparing the genomes it finds to a database of gene sequences it expects to find. The second searches for sequences of interest — bits of code that might be a signature of engineering — within the genomes of previously unknown organisms. Currently, the second method can detect a single engineered being from a sample that also contains a million, or maybe more, totally natural beasts.
At Harvard, a team led by Elizabeth Libby, PhD, is working on a “biosensor”: an engineered cell that senses other engineered organisms. Naturally, the Boston group began by sketching the sensor’s basics on a Dunkin’ Donuts napkin. “It can basically vacuum up any DNA that comes nearby,” says Libby. Then, it picks up on whatever DNA signatures they’ve preprogrammed it to recognize, boosts the signal, and then lights up to say, “Got it!”
For FELIX, they’ve programmed the biosensor to react to common signatures of engineering. But the technology is broadly applicable. “What we envision, down the road, is if you wanted to detect a pathogen in, say, an air-handling system or a hospital surface or a water supply, you could have a cartridge that’s disposable and passively senses all the time,” says Libby. If it detects what she calls “the thing of interest” — whether that’s an engineering tip-off, a coronavirus, or some other genetic sequence — the cartridge lights up and sends a signal. Illumination achieved. In a pandemic like the one we’re living through now, such a technology could be useful for figuring out whether the enemy is lurking inside your office building.
In their current form, these methods largely share the same limitation: They rely on records of known organisms, or known signatures of engineering. In other words, they need a reference catalog to compare against. The statement on IARPA’s website implies that the MIT-Broad Foundry’s does, too.
“You’re never going to have completely the ability to detect something that’s engineered.”
And therein lies the rub: At the moment, most analysis relies on existing data and on assumptions about engineering’s multiple personalities. “Everyone who’s trying to engineer knows just as much as the people who’re trying to detect engineering,” says Plant. “So you have, functionally, an arms race. You’re never going to have completely the ability to detect something that’s engineered.”
The MIT-Broad Foundry’s SARS-CoV-2 analysis weeded out the idea that the virus was put together with other organisms’ parts — a common method among the FELIX teams. But a lot of pathogens have genomes that don’t populate databases. “You can’t control for all the viruses that have been discovered but not shared,” says Alina Chan, PhD, a postdoctoral researcher at the Broad Institute. Many scientists, she says, “take years to publish their work. I could hoard some data for 10 years if I wanted to. There’s no law.”
Though such delays are normal, they nevertheless hamper efforts like FELIX because some engineering-alert tools work like a plagiarism detector, checking to see whether an organism contains stolen text. A lazy plagiarist will just CTRL-C, CTRL-V whole paragraphs: easy to spot. A better detector can determine that someone thesaurus.commed a few words. A super detector could perhaps tell whether the term paper is modeled on a Slate essay. All plagiarism detectors, like engineering detectors, require a robust catalog of published text. If someone manipulated an unpublished virus they’d found in the wild, it would be much harder to spot.
Plant uses a different analogy for tinker-detection more broadly. “Detecting an engineered organism is like trying to detect whether a word has been purposely invented,” he says. “To do that with perfect confidence, you have to know all the words that have ever existed, as well as all the words that are currently being created by accident. That, like detecting an engineered organism with perfect fidelity, is impossible.” Efforts like FELIX may never be fully equipped to determine whether an organism has been engineered or not.
FELIX’s existing analysis of SARS-CoV-2 would detect the work of scientists who put publicly known sequences into pathogens. A less hapless bioengineer would — whatever organism they were trying to gin up — use a sneakier method. “If you wanted to infiltrate an event incognito, you’d disguise yourself as someone unknown,” Chan says, “as opposed to putting on multiple facets of the most recognizable celebrities in the world and waltzing into the room.”
What FELIX has provided is evidence that SARS-CoV-2 isn’t made of one celebrity’s eyes and another’s ears. Which is a useful hypothesis to have ruled out. “This was a limited application of a toolkit,” says Gregory Koblentz, PhD, associate professor and director of the graduate program in biodefense at George Mason University.
Even if FELIX were to perfect genetic-engineering detection, its discoveries would raise even more questions than answers. In the case of a disease outbreak, intelligence and health officials would still want to know where the germ had been engineered, by whom, and why. “That requires much more information,” says Koblentz — not just of the technical sort but also of the intelligence and law-enforcement variety.
FELIX doesn’t address those questions. But it does bring up different ones. While FELIX’s stated aim is to increase biosecurity, its technology is necessarily dual-use: offense, too, not just defense. If you know how to detect bioengineering, you theoretically understand how to hide your own. In that way, FELIX could be seen as a passive flex on the international community, broadcasting offensive ability without violating bioweapons-related conventions. IARPA did not respond to questions about dual use in time for publication.
Programs like FELIX beam a second message to the globe, too. “[This research] is driven by this perception that the diffusion of increasingly sophisticated biotechnology is creating new potential threats that we are not prepared to detect,” says Koblentz. “This is trying to prevent a Pearl Harbor, trying to prevent another surprise attack. FELIX is just one example of that.
In other words, with these programs, the United States is telling the world that it thinks that biothreats could present a clear and present danger. That implication could, Lentzos believes, lead other countries down more pathogenic research paths, to avoid being left behind. “While you’re just trying to protect yourself,” she says, “you’re actually creating the threat.”