There Are More Viruses in a Liter of Seawater Than People on Earth

Lex Fridman opens with a statistic Vincent Racaniello likes to use on the first day of virology class: there are more viruses in a liter of coastal seawater than people on Earth. A Nature piece titled *Microbiology by Numbers* puts the count at 10^31 viruses on the planet, with 10^23 infections per second in the ocean and a daily removal of 20 to 40 percent of bacterial cells. Racaniello, a professor of microbiology and immunology at Columbia and co-host of the podcast This Week in Virology, says students walk out stunned, which is the point. The 10^31 figure, he clarifies, is just the bacterial viruses in seawater, tallied from particle counts around the world.

The mass of those ocean particles, he says, exceeds that of all the elephants on Earth by a thousandfold. Lined up end to end, they would stretch 200 million light years into space. Each second they punch holes in bacteria and dump organic matter back into the water, feeding what oceanographers call the biogeochemical pump: cells die, sink, get metabolized, and a large share is released as carbon dioxide. Lex calls it a mechanism for mass murder. Racaniello shrugs. It is simply what happened, and it is not even uniformly fast. Some viral cycles take twenty minutes, some take weeks; the per-second rate is a census of how many particles are working at once.

To explain why there are so many, he pulls the clock back billions of years. His guess is that viruses, or something very like them, were among the first organic entities after the Earth cooled: short self-replicating molecules that today would look like RNA. They were naked and fragile, so they probably never got very big. Proteins arrived, then cells, and DNA replaced RNA as the way to store a long genome. One of the earliest enzymes, he thinks, was reverse transcriptase, the protein that copies RNA into DNA, later co-discovered by his former boss David Baltimore and by Howard Temin. Once cells existed, those self-replicators wandered in, found a hospitable pool of building blocks, stole proteins for a shell, and learned to leave. His colleague Dickson Despommier calls viruses safe crackers: building a safe is expensive, cracking one is cheap. "There's no planning, no thinking about it," Racaniello says. "It just happens."

Lex wants the "magic moment" on the way from bacteria to something with tissues. Earth spent a long stretch with only single cells, and he wonders what was hard if you were trying to engineer a planet. Racaniello's answer is energy, not imagination. Bacteria form colonies and biofilms, but every cell in those mats is still doing the same job. Skin, eye, and brain cells are a different trick, and bacteria never pull it off. "Bacteria cannot make enough energy to do that," he says.

The turn, in this telling, is a bacterium that entered an archaeon and stayed, becoming the mitochondrion, the cell's power plant. With a surplus of energy the host could get more complicated, and the lineage we now call eukaryotes took off. Racaniello is careful to credit the idea: it is not his. It comes from Nick Lane, an evolutionary biologist in the United Kingdom who has done experimental work on the energy argument. After that, the pre-cellular replicons already in the story had a richer interior to invade, which is why, he thinks, everything on the planet is infected. Most of those viruses cannot infect us. The ones to watch live in the mammals and birds we live beside: bats, about 20 percent of mammal species; rodents, about 40 percent; and birds, which keep giving influenza back.

Asked for a definition, Racaniello stalls on purpose. Lead with the definition, he says, and people turn off; lead with the numbers, and they stay. The textbook line is *obligate intracellular parasite*: a virus takes energy and building blocks the host would otherwise use, and it cannot reproduce outside a cell. That sits awkwardly next to the earlier claim that many viruses are beneficial. Definitions, he reminds Lex, are constructs that make study easier, not revelations. Pluto was a planet until it was not.

His own view is two-phased. Fill a mug with virus particles and they will sit there for years, slowly losing infectivity, doing nothing that looks like life. Get one inside a cell and the cell becomes a virus factory; that infected cell is alive. A bacterial spore is a useful contrast. It also lies dormant until you add water and nutrients, but it does not need to enter another cell in order to divide. People picture the particle, the virion, when they say "virus," which is why he refuses to call that object alive. Once it is in, it is taken apart; the new particles that leave are inert again.

Entry looks like trickery and is not. Cells display receptors for nutrients and signals; when something binds, they often swallow it. The virus happens to fit, and it slips in. Lex reaches for a Trojan horse. Racaniello allows the metaphor and then takes it back: the virus is not deceiving anyone, it is hitchhiking. He tells students to be careful with anthropomorphism. It helps understanding until you start assigning motives. Variants are emerging, people say, because antibodies are selecting them. That is a good hypothesis, he says, and it may not be the only thing going on.

Almost everything else on Earth stores genetic information in DNA. Some viruses still use RNA, relics of an RNA world that Racaniello does not treat as fossils. In his view they are the most successful viruses we have: HIV, influenza, SARS-CoV-2. DNA viruses can be extremely successful too. Everyone, he notes, carries a herpesvirus. They just do not make the news, because they are conservative and slow.

The mechanical reason is copying quality. RNA viruses, he says, exist at their error threshold: they cannot make any more mutations when they reproduce without going extinct. DNA viruses sit hundreds of times below that threshold, a gap you can measure in the lab. RNA polymerases lack proofreading, so every round of copying imports mistakes. That is a terrible way to keep a genome intact and a brilliant way to adapt. A bat in a cave is not hosting one coronavirus genome. It is hosting millions of slightly different ones, and somewhere in that cloud may be a genome that happens to fit a person.

Lex supplies the business analogy and Racaniello takes it. RNA viruses are startups. Most of the changes fail; one survivor is enough. DNA viruses are IBM and Google: large, slow, expensive to reproduce, loaded with bureaucracy. The nucleic acid is never left naked for long. It is wrapped in protein, or in a stolen fatty membrane studded with spikes that bind cells. Plant and fungal viruses are the exceptions. Yeast cell walls are hard to cross, so those viruses often just ride cell division. Plant viruses hitchhike on sap-sucking pests or on contaminated farm equipment.

Asked for a virus that is fascinating, terrifying, and beautiful at once, Racaniello splits the adjectives. He is not terrified as a virologist; the human side is always messier. The terrifying example is rabies. Unvaccinated, it is essentially 100 percent fatal, still killing about 70,000 people a year, mostly via feral dogs. It is also the rare vaccine that works after exposure, because disease takes so long to climb the nerves. Ebola's case fatality can run 50 to 90 percent in parts of Africa with thin hospital care; cases that reached the United States were manageable, and a vaccine now exists, tested in a few thousand people during the 2013–2015 West African outbreak of about 25,000 cases. Neither virus is a good traveler. Rabies needs a bite. Ebola patients are too sick to walk around shedding at parties.

That is his answer to the philosophical question: why is there no virus that is both extremely deadly and extremely transmissible? Compare SARS-1 and SARS-2. With SARS-1, more than half of infected people ended up in hospital, and peak shedding came after they were already in isolation. Containment was almost a side effect of severity. SARS-2 reproduces hard in the nasopharynx before anyone feels anything, and many people never feel anything. "If the infection is too lethal you're simply not a good transmitter," Racaniello says. Transmission, in his view, is among the strongest selection pressures a virus faces. Fail to find a new host and you are a startup that dies.

The cleanest experiment is not a thought experiment. In the 1800s, Australian hunters imported European rabbits because the native ones were too fast to shoot. The imports exploded into the millions. Authorities released myxoma virus, a poxvirus of another rabbit, spread by mosquitoes. The first year, 99.2 percent of the rabbits died. The remaining 0.8 percent carried resistance. In later years the virus itself became less lethal, because a rabbit that lived longer gave mosquitoes more time to move the infection on. Hosts and virus both survived. It is, Racaniello says, almost the only clean field experiment we have of lethality and transmissibility pulling against each other.

Coronaviruses got their name from an electron-microscope silhouette that looked like the solar corona, nothing more mystical than that. They are enveloped, spiked, and packed with RNA, the longest RNA genomes known: SARS-CoV-2 at about 30,000 bases, some relatives at 40,000. In the 1960s, two human coronaviruses turned up as mild colds. In 2003 a severe respiratory disease started in China in November and was not announced until February, by which time it was moving. It reached 29 countries, infected about 8,000 people, and stopped. Live markets in Guangzhou, palm civets, and then bats in countryside caves supplied the trail. In one cave, researchers later found all the pieces that could assemble SARS-1. That was the first clear proof that bat coronaviruses could land in people.

Racaniello thinks the world should have been ready. Some groups started vaccines and got as far as mice. Some started antivirals. Industry had no disease left to sell against, and NIH reviewers called the grants too risky because the virus was gone. "People were really short-sighted," he says. Molnupiravir, the oral antiviral later in phase three, had been developed five years earlier and never taken into humans. MERS in 2012, spilling from camels on the Arabian Peninsula with almost no sustained human-to-human spread, was another ignored alarm. Then 2019. SARS-2's genome is about 20 percent different from SARS-1. It still binds ACE2, and the spikes look similar, but enough amino acids changed to rewrite the biology, and animal models still do not get the human split: many mild cases, about 20 percent severe and prolonged, some fatal.

The practical barrier in RNA virology used to be that you cannot edit RNA the way you edit DNA. In 1979, as a postdoc at MIT, Racaniello got a job from David Baltimore: the recombinant-DNA moratorium on viruses had just lifted, so make a DNA copy of poliovirus and see if it would launch an infection. Poliovirus is only 7,500 bases. He cloned it into a plasmid, spent a year sequencing it with methods that now take fifteen minutes, put the DNA into cells, and polio came out. That became the template for influenza, coronaviruses, everything. In January 2020, as soon as China released the SARS-2 sequence, labs synthesized the 30,000-base DNA and recovered virus. Cut a gene, put the DNA back, and you learn whether that gene is optional or lethal. The limit is not technique. SARS-2 is a BSL-3 agent, so not every bench can touch the whole virus. Vaccine work does not need to. The spike alone is enough.

Lex's prior was that influenza, not coronaviruses, was the family that should keep civilization up at night. Racaniello grants the history and then draws the mechanical contrast. Both families have membranes and spikes. Influenza's RNA is negative-sense and chopped into eight segments, so two viruses in one cell can shuffle pieces and throw out a pandemic cocktail the immune system has never seen. Coronaviruses carry one long positive-sense RNA that ribosomes can translate immediately. Some flu vintages are killers, 1918 among them; the 2009 swine-origin pandemic transmitted well and killed relatively few. Shorebirds will keep the reservoir stocked. A perfect human vaccine would still not erase influenza from gulls.

The first influenza vaccines still sound like folk magic. Jonas Salk, working in the 1940s for the U.S. Army, grew virus in ten-to-twelve-day chick embryos, harvested about ten milliliters of fluid, treated it with formaldehyde, and injected the wreckage. Cells in culture did not yet exist as a tool, so eggs were the factory, and eggs are still in use. Chemical inactivation leaves an antigen that is no longer very close to a real particle, which is one reason today's inactivated flu shots top out around 60 percent efficacy against illness, not infection. Efficacy in trials is scored on symptoms, which only the patient feels, plus a PCR confirmation. Culture, Lex notes, changes who reports a sore throat. Racaniello agrees the number is fuzzier than it looks.

The 1950s brought replication-competent, live-attenuated vaccines. Max Theiler passed yellow fever through hundreds of chick embryos, checking every ten passages in mice or monkeys, until after more than 200 passes he had a virus that did not cause disease and still taught the immune system. Measles, mumps, and rubella followed that empirical path. The failure case is Albert Sabin's oral polio vaccine: elegant delivery, strong gut immunity, and shedding of virus that had genetically reverted in the intestine. In the 1960s and 1970s, some parents of vaccinated children caught polio from that shed virus. The rate was about one in 1.5 million doses, eight to ten paralyzed children a year in the United States. Public health called it an acceptable cost of eradication. Racaniello, looking back, does not.

Later generations get more precise. Vectored vaccines take a virus that infects people without making them sick, cut out the dangerous genes on purpose, and splice in an antigen, as with adenovirus or vesicular stomatitis virus carrying an Ebola or coronavirus spike. Racaniello thinks most cancers will be treatable with targeted viral vectors in ten to fifteen years, and the same toolkit can return a missing gene in a monogenic disease. DNA vaccines, injecting a gene on a plasmid, have no licensed human product; there is a West Nile DNA vaccine for horses. Nucleic-acid thinking then moved to RNA, which brings the fifth generation.

Racaniello did not believe it. He has worked with RNA his whole career. Look at it the wrong way, he jokes, and it breaks; your hands are covered in enzymes that chew it up. Companies had been tinkering with experimental mRNA vaccines for years before 2020. When they said they would try it on SARS-2, he thought the molecule was too labile. The fix was a lipid nanoparticle, a greasy capsule whose formulation Moderna and Pfizer spent years optimizing in mice. Fat sticks to fat. Cell membranes are greasy, the particles bump, they get swallowed, and there is no receptor in the story. Lex calls it cracking the safe with a hammer made of fat. Racaniello agrees it is amazing that it works.

Once inside, the mRNA is translated into spike protein, and the immune system files a memory. The spike in the vaccine is not the fusion machine on the virus. Two amino-acid changes lock it in a pre-fusion shape so it cannot zip cells together, a tweak that came from structural biology. mRNA itself is gone in days; the protein lasts at most a few weeks. That is an advantage over DNA, which can reach the nucleus and linger. The shot goes deep into the deltoid, not into a vein. In mice given a thousand times the human dose, nanoparticles showed up in every tissue. At the doses used in people, Racaniello says, most of the material stays in the muscle, with small amounts elsewhere.

Near-term harms, he notes, tend to declare themselves in the first months: myocarditis with some shots, clotting with AstraZeneca, Guillain-Barré in small numbers. Millions of doses later, those signals are known and rare. The harder question is the long tail. Lex compares it to putting Splenda in coffee, a thing that tastes too good and therefore feels like it must collect a bill later. Racaniello will not cash that check with a slogan. If someone tells you there are no long-term effects of the COVID vaccines, "they can't say that because they haven't done the long experiment." Fifty or sixty years would be a long experiment. What you can do is weigh. COVID, he says, damages more than the lungs; the cognitive effects are what scare him. Long COVID was already real at the time of the taping, observed out six to eight months, duration unknown.

Racaniello does not pretend medicine has a clean record. The first Salk polio vaccine, released in 1955, paralyzed a few hundred children within months because inactivation was botched. Parents took it anyway; summers were bringing 30,000 paralyzed children a year. The later live vaccine kept causing a handful of cases after the evidence was in. He has sat in lawsuits brought by those families. The first failure, he says, was preventable. The second should have ended the product. "That destroys public confidence." He does not think those stories still govern 2021, because most people have never heard them. He tells them because the mechanism can run again. Lex's opener had already named the other half of the fear: some people are scared of a virus, some of a government, and dogmatic certainty plus division, he said, is more destructive in the long term than any virus.

When the talk turns to Anthony Fauci, Lex piles up praise and then the indictment. Leadership in a new pandemic is hard. The communication still failed on three counts: a tone the internet reads as inauthentic, speaking from authority when nobody was an authority on a brand-new virus, and too little openness about uncomfortable origin hypotheses, including a laboratory leak. Masks, tests, lockdowns, and durability of vaccine protection were all unknown in early 2020. Stating them as settled, Lex argues, is how you spend trust. His deepest objection is the result: a visible drop in trust in science, which is the tool that actually gets societies out of holes. The person at the top carries that bill even when the failure is collective.

Racaniello answers more narrowly. As far as he knows the data are public; FDA rules require clinical-trial releases, and if something were hidden it would be hidden from him too. Large-scale fabrication, he thinks, tends to show up as inconsistencies across trials, a bit like faking a moon landing being harder than flying there. He has long disliked making health depend on for-profit firms and daydreams about well-paid nonprofit vaccine institutes. Lex, who grew up with the Soviet alternative, says capitalism's ugly virtue is that it fires people who cannot deliver. American healthcare, they agree, is a muddle of regulation, subsidy, and politics, and politics should never have been in the examining room. Racaniello adds an observation he finds strange: vaccines and antivirals take the conspiracy heat, while everyday drugs with worse side-effect sheets barely get a question.

Ivermectin, a very safe antiparasitic that can prevent river blindness with one dose a year, became a culture war. Racaniello discloses that his wife worked on the drug at Merck for twenty years. An Australian cell-culture study early in the pandemic showed clear inhibition of SARS-2, at concentrations you cannot reach with approved human dosing. The drug binds receptors in the brain; taking ten times more would need a safety study first. He wanted a proper randomized trial and notes a 4,000-person study then enrolling. Off-label use if a physician offers it does not offend him. He will not pretend he knows it works. Manufacturing is another mess: ivermectin is hard to synthesize, so it is fermented from a bacterium Merck owns. Two employees once stole the strain, were arrested, and went to jail. You cannot scale it like a molecule you cook in a tank.

Hydroxychloroquine is the cleaner scientific miss, and it poisoned the well for other repurposed drugs. It blocks the endocytic pathway whose falling pH helps many viruses escape a vesicle, so of course it inhibited SARS-2 in kidney cells, which only have that door. Human lung cells offer a second door at the cell surface. The drug does not block that route, so the virus walks around it. A wrong cell-culture conclusion, Racaniello says, drove a lot of subsequent politics. HIV antivirals and other repurposed compounds had similar unhappy tours. The deeper timing problem is shared. People ask for an antiviral when they cannot breathe. By then the problem is inflammation, not virus. Viral load crashes within about three days of the first symptom; hospital antivirals, remdesivir included, are often late. Monoclonal antibodies have the same clock. What you want is a pill on the day of a fifteen-minute positive test, scratchy throat still in the room.

Even a good oral drug is not a strategy by itself. Molnupiravir looked promising in phase three, a pill that had sat on the shelf. If it becomes the only oral antiviral, Racaniello warns, resistance will arrive in months. HIV and hepatitis C taught the field to use two or three drugs at once. The money, meanwhile, went to vaccines: he cites a high-level decision, involving Fauci, to put on the order of $24 billion there because shots can protect for years, while pills need repeated dosing. He thinks that bet left testing, masks, and early antivirals underfunded in the first year.

Joe Rogan had caught heat for saying vaccines create selective pressure for mutants. Racaniello wants the verbs fixed first. Viruses are always mutating. No vaccine and no drug makes them mutate. Immune responses, from shots or from infection, apply selection: if one particle can slip past antibody, it emerges. Influenza does this every year. Early 2020 sequences already carried, at very low frequency, the changes later seen in variants of concern. They had nothing to select them until millions of people were immune, mostly from infection, some from vaccination. Alpha then rose in England around November. Stop vaccinating and infection will do the same selecting, with a much higher chance of severe disease. People who recovered from COVID and then took one vaccine dose, he notes, suddenly made a response that covered every variant then in circulation, broader than two doses alone. That, for him, is proof a rationally designed broad vaccine is possible, the same bet influenza researchers are making on the unchanging stem of the spike.

Testing is the other cheap lever he thinks was dropped. PCR detects fragments of viral RNA, not infectious particles, after reverse transcription and cycling. Cycle threshold is the gotcha even physicians mix up: CT 35 means very little RNA, CT 10 means a load. Below about a million copies in a sample, he says, you are not infectious. Early policy quarantined every positive, which was wrong, and fourteen-day isolation was longer than shedding lasts; ten days is enough. The typical curve peaks four to five days after infection, often before symptoms, then crashes in three or four days. Antigen tests look for protein instead and can sit in a drugstore. Harvard's Michael Mina argued in 2020 for a daily paper strip costing less than a dollar: positive twice, stay home. It was never mass-manufactured. Racaniello thinks cheap antigen tests should start in week one of the next outbreak, even if raising antibodies in animals takes a few weeks that PCR does not need.

Spread, he says, is mostly airborne. Talking throws droplets; large ones fall, small ones can travel a hundred feet and usually carry less virus. Infectious dose is unknown, probably at least several thousand particles. About 80 percent of transmissions come from 20 percent of infected people. A University of Colorado study of student swabs over six months found most of the RNA copies in 15 to 20 percent of people. That is why one room can be safe for two hours and another pass-through is not. Mask data are thinner than vaccine data because you cannot run a clean human challenge, so the field leans on household epidemiology and on droplet-only material tests, then on models. Daniel Griffin's working numbers, 50 to 60 percent from a mask and another 30 from distance, are estimates. The political wound, in Racaniello's telling, was January 2020: CDC and WHO said masks do not work, then reversed without saying they had been wrong. "That was the kiss of death for masks."

Lex asks what to make of the public argument between Sam Harris, who treats skipping vaccination as irresponsible, and Bret Weinstein, who dwells on long-term unknowns and on ivermectin. Racaniello's rule is that conversation is healthy and definitive statements are not, because they imply knowledge you do not have. There is evidence for long COVID and, at the time, no equivalent evidence for long vaccine injury. Scaring people with a hypothetical while downplaying a documented sequela is not, in his view, a fair argument. Telling everyone they are stupid if they hesitate is not the other answer; people dig in. The middle is a calm ledger: here is the safety profile, here is the benefit, here is what I did, you decide. Columbia required vaccination to work or study on campus, which he treats as an employer's choice, not a priestly one. If someone still wants ivermectin after the talk, he says that is fine.

He had cancelled a This Week in Virology recording to sit for the interview. Before COVID the show rotated among many viruses; he is trying to sneak those back in, because they are informative, not because the pandemic is over. On Wednesday nights he runs a live stream under the handle profvrr where, thirty minutes in, hundreds of questions stack up. Lex quotes Camus's *The Plague*, a closed town that reveals the best and worst of people, and the doctor who is freed from his own mortality by serving others. Racaniello's shorter line is that a tiny, invisible virus can bring Earth to its knees, not only by killing but by tearing the social fabric animals need. Airports in August 2021 looked almost normal. New York the year before had been eerie and quiet. Most people, he still thinks, wanted to help, including the lab workers who built the shots.

Advice for the young is not a five-year plan. His father wanted a physician; he was not interested in taking care of people, so he took a lab job after college because he liked science, and each step produced the next. Be curious, or you will be bored. Be passionate about the thing that stops feeling like a job. He tells classes he loves viruses, which sounds morbid until you hear that he means their emergence and their mechanics, not their body count. Asked for the meaning of life, he declines the assignment. There is no meaning. It is an accident, a rare storm of chemistry, and he doubts there is life elsewhere. Sentience is the exception: we can take medicines and shove some selection aside. He does not contemplate his own death. Telomeres shorten, immortality is a vampire story, and bacteria, little bags of chemicals that split, have no stake in the matter at all. Isaac Asimov, in Lex's sign-off, gets the last word: science gathers knowledge faster than society gathers wisdom.