A History of Antibiotics

Stories about antibiotics usually begin with mold. But there is a way to look much further back. In 2011, researchers examined 30,000-year-old DNA extracted from Beringian permafrost sediments. In it, they found resistance genes against beta-lactams, tetracyclines, and glycopeptides. When they compared the 3D structure of an enzyme produced by one of those genes side by side with its modern counterpart, the difference was minimal, and its function was no different.

There is evidence closer to our time, too. In 1940, two researchers working with penicillin at Oxford published a paper on "an enzyme from bacteria able to destroy penicillin." Back then, penicillin was not yet a medicine. It had not even been used on its first patient. The mechanism that disarms the drug was recorded before the drug itself.

So saying "overusing antibiotics created resistant bacteria" is not an accurate description. Using a drug does not create that trait; it simply creates an environment favorable to the ones that already possess it. When a drug is introduced, those that can endure survive and reproduce. An individual microbe does not alter its body—the composition of the population changes.

Let us make two clear distinctions. First, antibiotics act on bacteria. They do not work against viruses. Both cause disease and often get lumped together, but their structures are completely different. Second, this story looks nothing like the story of smallpox or polio. In those cases, the question was, "Can we erase this from Earth?" Bacteria live in soil, in water, and inside other animals, and resistance genes have always existed in the natural environment. Eradication is not the goal here.

That is why the question this series follows is not "Why did resistance emerge?" That resistance is an inevitable, recurring condition is our starting point, and only one conclusion follows from it: We constantly need new drugs. The real story begins from what comes next.

So, when did these medicines first appear? There is no single answer to this question. Depending on what kind of "first" you are asking about, both the person and the year change completely. Let's break it down along five different axes.

First, there is the first drug chemically synthesized to fight bacteria. That is Salvarsan. It is considered the first effective treatment for syphilis, and its nickname, "Compound 606," came from its order among about 2,000 synthetic molecules tested by Ehrlich's research team. In 1920, 2 million doses of this drug and its modified versions were manufactured each year in the United States alone.

Second, there is the first class of antibacterials widely used in clinical practice. Chemists at a dye company synthesized and handed over more than 300 azo compounds, and one red dye among them proved effective against bacterial infections. It was commercially released in 1935, and that very year, the Pasteur Institute in Paris discovered that the active ingredient was not the dye itself, but a much simpler molecule called sulfanilamide. Over the following decade, more than 5,000 variations of this substance were synthesized.

Third, there is the first widely used medicine produced by a microorganism. Penicillin belongs here. Fourth, there is the method of systematically screening soil microbes to find new drugs. Waksman's laboratory established that approach.

The fifth is fascinating: the word "antibiotic" itself. Waksman was the first to use this word as a noun, and that was in 1941. What he meant was "a small molecule made by a microbe that inhibits the growth of other microbes." If you apply that definition strictly, neither Salvarsan nor sulfa drugs are antibiotics, because both were chemically synthesized by humans. The meaning we use today has expanded since then to include synthetic antimicrobials as well.

There is a widely told story: Fleming accidentally discovered a mold, saved the world, and warned people not to overuse it. Since the original transcript of his Nobel lecture still survives, let's take a look. It is a lecture he delivered on December 11, 1945.

First, he deliberately narrows the scope of his own contribution. He notes that "my 1929 publication was the starting point of the work of others" and openly states that he "failed to concentrate and stabilize" the substance. He also writes for himself that it was a pure accident. He admits he could have claimed that he searched for molds after extensive reading and deep reflection, but that simply was not true. And toward the end of his lecture, he says he leaves the rest of the story to Florey.

There is another fascinating part. He writes that the phenomenon of microbes inhibiting one another's growth was commonplace among bacteriologists of his generation. Anyone who looked into culture plates, he explains, saw it about once a week. He then adds that the fact that it was so common and well known actually prevented rather than helped research in this field from taking off. All he did, in his own words, was simply not let one of those ordinary sights slip by.

The section with his warning points in a different direction. What he saw as dangerous was using too little. He explains that exposing microbes to concentrations too weak to kill them can easily educate them to resist the drug, something that is not difficult to do in a laboratory. Then he gives a hypothetical example: a person buys penicillin for a sore throat, self-administers too small a dose, passes the resistant bacteria to a family member, and subsequent treatment fails. That section of the lecture concludes with: "If you use penicillin, use enough."

The tone is quite different from how it is commonly summarized today. The danger he envisioned was one where anyone could buy the drug over the counter in shops, and he viewed that danger as a matter of dosage. How today's prescription guidelines are determined and on what grounds is outside the scope of this story. What we should take away here is just one thing: the more widely a quote is cited, the more its original context may point in a completely different direction.

For nearly ten years after the 1929 paper, this substance failed to become a medicine. Two things shifted the board: the arrival of sulfonamides convinced the medical community that chemical drugs could treat bacterial infections, and Dubos demonstrated a potent antibacterial substance produced by bacteria. Fleming wrote that only then did Chain and Florey take up the problem.

The Oxford team divided up the work. Chain handled purification, Heatley took charge of mass-culturing the mold, and Florey oversaw animal trials. Their results were published in The Lancet in August 1940, and the drug was given to its first patient the following February. He was a police officer in Oxford. His condition improved visibly within a single day, but the supply ran out, and he died a few weeks later. What filled the dozen-odd years between discovery and medicine were entirely different tasks: purification, cultivation, and clinical trials.

Next came the United States. On July 14, 1941, the researchers arrived at the Department of Agriculture laboratory in Peoria, Illinois, and work began the next day. What emerged here was the method of pumping air into deep vats with corn steep liquor, a byproduct of cornstarch processing. It was faster and yielded far more than growing mold on surfaces. After scouring the world, the strain that produced the most came from a spoiled melon in a Peoria market. The price shifted accordingly. An item virtually impossible to price in 1940 cost $20 per dose in July 1943, and 55 cents three years later. By September of that year, enough had been secured to meet the demand of the Allied forces.

Yet how Fleming himself recorded this chapter is what matters. He wrote that the Oxford team's findings were published "in the middle of a great war when normal economics were in abeyance and production could be continued regardless of cost." He also noted that American factories were built "at great cost." That this drug became cheap and plentiful was the result of a few years when cost didn't matter. Once that exception ends, the arithmetic returns.

Today, newly developed antibiotics carry a label right from the start: "Use sparingly." The World Health Organization categorizes antibiotics into three groups: Access, Watch, and Reserve. The Reserve group is the last-resort option for severe infections caused by multidrug-resistant bacteria, to be used only when all other drugs fail. Newly approved antibiotics are almost always placed in this group.

Regulators follow the exact same direction. In December 2016, the United States established the "Limited Population Pathway." It allows antimicrobial drugs to be approved specifically for a limited patient population suffering from serious infections, but requires the drug's labeling and all advertising to prominently display "Limited Population." Regulators have carved out an official category for "drugs designed to be used less."

Returning to the company from our introduction, the picture looks entirely different now. In its annual report, the company stated that its drug was labeled for patients with "limited or no alternative treatment options." When prescribed strictly by the label, prescriptions will naturally be scarce. Its 2018 net sales of $800,000 were the direct outcome of that. That same year, its net loss was $186.5 million, and its accumulated deficit reached $559.4 million. Nor was it an isolated case. In December 2019, another company selling four approved antibiotics entered the same restructuring process, swapping 100% of its equity for $140 million in secured debt.

This is the very heart of the story. The better the rules work, the fewer drugs are sold. Conserving them is the right thing to do, but using them sparingly means no sales, and no sales means the next drug never gets developed. When an immense need fails to translate into demand, we call that a market failure. In the conclusion of its 2025 report, the World Health Organization likewise described this condition as a "broken market."

Major pharmaceutical companies pulled out of this field one by one. It is easy to say they were simply chasing profits, but looking at how their math actually worked gives a far more accurate picture.

First, there is the shape of the market. A 2016 review commissioned by the UK government estimated the entire global antibiotic market at around $40 billion a year. Yet the share of patented antibiotics was just $4.7 billion. The report noted that this was "roughly comparable to the annual sales of a single blockbuster cancer drug." Off-patent generic drugs accounted for the rest.

Second, the timing is misaligned. To borrow the words of the same report, the commercial viability of a new antibiotic only becomes attractive once resistance to older-generation drugs has spread widely, but by then the drug's patent has either already expired or is about to. The moment when the need is greatest is the exact moment when you cannot get paid.

Third, it is a question of where to put the same money. Between 2003 and 2013, out of $38 billion in venture capital poured into pharmaceutical R&D, only $1.8 billion went to antimicrobial research—and even that amount shrank by more than a quarter over that period. In 2014, the oncology pipeline had nearly 800 drug candidates, while antibiotics had fewer than 50. Public research funding told a similar story. From 2009 to 2014, resistance-related research received just 1.2% of the National Institutes of Health (NIH) budget, compared to 18.6% for cancer.

The numbers also reveal what the space left behind looks like. Among the 118 preclinical developers counted by the World Health Organization in its 2025 report, 90.7% had fewer than 50 employees, and 61.9% had fewer than 10. The developer turnover rate was about 33%, and eight programs were left stranded when their parent companies were liquidated or exited the field.

There is a widely circulated claim: that no new classes of antibiotics have emerged since 1987. But when you look into the source of this statement, it becomes clear what kind of "none" it really refers to.

An academic review from 2011 puts it this way. The author first separates the year of discovery from the year of introduction. Oxazolidinones were first reported in 1978, and a drug in that class came out in 2000. Acidic lipopeptides were discovered in 1987 and introduced in 2003, while pleuromutilins were discovered in 1952 and introduced in 2007. It takes twenty to fifty years from discovering a class to turning it into medicine. Then, the author writes that when looking at the year of discovery rather than the year of introduction, there has been no successful discovery of a new class since 1987.

What comes right after that sentence is crucial. The author calls this "not a gap, but an open-ended void," adding that while there are a few new compounds in early clinical trials, whether they will succeed remains uncertain. The author intentionally avoided making an absolute claim.

If you look at approvals instead, the picture looks different again. In December 2012, a diarylquinoline tuberculosis drug received accelerated approval in the United States. The manufacturer stated that it was the first tuberculosis drug with a new mechanism of action in 40 years. In March 2025, an oral antibiotic in a class called triazaacenaphthylene was approved, and the manufacturer introduced it as the first drug in a new class. Let us note that both are the manufacturers' own characterizations. Even so, it remains true that "no new class" is a statement along the axis of discovery, while things look quite different along the axis of approval.

Looking at one more yardstick brings balance. The World Health Organization assesses clinical-stage antibacterials by whether they show "no known cross-resistance to existing classes." As of February 2025, only 5 out of 27 traditional antibacterial agents—or 19%—met that standard. For tuberculosis, it was 6 out of 18. The overall pipeline stood at 90 candidates, down from 97 in the previous survey. The same report concluded that even when combining clinical-stage antibacterials with those approved in recent years, they "remain insufficient to tackle the growing threat of the emergence and spread of drug-resistant infections." It does not mean that nothing is coming out; it is a judgment that, when counted, there are simply not enough.

There is another number in this field that must be written in two lines: the number of people dying from drug resistance.

An international collaborative study published in The Lancet in 2022 calculated two figures using 2019 data. One was 4.95 million deaths associated with resistant infections. The other was 1.27 million deaths attributable to resistance. That is a difference of nearly four times.

The numbers differ because their underlying assumptions differ. The former asks what would have happened if those resistant infections had never occurred in the first place. The latter asks what would have happened if those infections had been caused by drug-susceptible strains that responded to treatment. That is why the first number is larger. Any figure that does not state which scenario it is measuring cannot be interpreted correctly. The same study also drew two separate lines by pathogen. The top six pathogens, including Escherichia coli and Staphylococcus aureus, accounted for 929,000 attributable deaths and 3.57 million associated deaths.

Measuring at the national level reveals yet another measuring stick. In its 2019 report, the US Centers for Disease Control and Prevention estimated that more than 2.8 million resistant infections occur each year in the United States, killing more than 35,000 people. When certain intestinal bacterial infections are included, the figures exceed 3 million cases and 48,000 deaths. Even within the same report, the numbers divide based on what is counted.

The next edition released by the same research team in 2024 used 2021 data to report 4.71 million associated deaths and 1.14 million attributable deaths. Recorded alongside these numbers were two trends moving in completely opposite directions. Since 1990, resistance deaths among children under 5 have fallen by more than half, while in those aged 70 and older, they have risen by more than 80%. Summarizing this into a single line erases one of those two realities.

The widely cited figure of "10 million people every year by 2050" requires the same caution. That value was a scenario developed for six pathogens by a 2016 UK review, and on the very same page, it carried the sentence: "predicting the path that resistance will take is impossible." The 2050 scenario presented by the 2024 study projected 1.91 million attributable deaths and 8.22 million associated deaths. Because the methods and assumptions differ, reading this as "the projection has changed" would be a mistake.

If the problem lies in a structure where "you make money only the more you sell," that is also where the fix has to happen. Models that sever the link between sales volume and payment amounts are actually being put to the test.

The United Kingdom went first. In July 2022, it signed contracts to pay two companies a fixed £10 million a year each for 3 to 10 years. It was the first model in the world that did not link payment to prescription volume. An evaluation body valued the two drugs at £11 million to £19 million a year. Later, as this approach was formalized, four tiers were established, allocating £5 million to £20 million a year for England. On August 12, 2024, a contract valued at about £1.9 billion over 16 years was announced, with an effective start date of April 1, 2026.

Why it has to look like this comes down to two reasons. One is that the benefits of conserving a new drug do not end where it was conserved. The payoff of an antibiotic staying effective over the long run ripples out to other hospitals and future generations. Conversely, an effective drug is a shared resource, so when everyone uses it at their own convenience, what everyone relies on wears out together. The other reason is that the payer and the beneficiary are split. One analysis found that 84% of sales across ten new antibiotics came from a single country: the United States.

Other countries made moves too. Japan included an antibacterial procurement support project worth 1.1 billion yen in its fiscal year 2023 budget, launching a revenue guarantee model for 3 years that covers the gap between actual sales and projected revenue. Sweden also ran a pilot payment model.

Still, what has not passed has still not passed. In the United States, bills allowing the government to enter into subscription-style contracts have been introduced multiple times over several years, but none have ever passed the full floor. On February 4, 2026, a bill with the same name was introduced again and referred to committee. And regarding the UK model, the World Health Organization noted in a 2023 report that "it is too early to evaluate whether this model has stimulated research, development, and innovation."

You don't need to predict the future. Instead, guided by the rules this story has shown us, let's wrap up by looking at where to watch so you can spot the game shifting before anyone else does.

First, post-approval revenue. The sheer number of approvals is not a healthy sign on its own. The company mentioned at the very beginning earned just $800,000 in revenue the year its drug was approved. It is far more accurate to look at how much a new drug brings in over the years following approval, and whether that sum comes anywhere close to its development costs.

Second, payment models. The things to keep an eye on are whether the UK contracts lead to renewals and new deals, whether other countries follow suit, and whether US legislation makes it through committee. We won't settle here which model is right. Third, the profile of the developers. Right now, most of the preclinical pipeline is carried on the shoulders of companies with around 10 employees. The total count and turnover rate of those firms offer the fastest read on this field's stamina.

Fourth, the discovery side. Rather than how many new drugs are approved, the more revealing story is how many among them share no cross-resistance with existing treatments. According to recent tallies, that was about one in five. It also helps to remember that going from discovery to a finished drug takes 20 to 50 years. What happens in laboratories today won't reach pharmacy shelves until long from now. Put the other way around, the empty spaces we see today began clearing out decades ago.

Fifth, the habit of reading numbers carefully. In this subject, figures almost always come in pairs: the discovery pipeline versus the approval pipeline, associated deaths versus attributable deaths, and the total market versus the patented share. Whenever you see a number without a label showing which side it belongs to, it pays to pause.

Lastly, just one more thought. Several players in this story chose to step back. The companies that abandoned antibiotic research saw an unrecoverable investment, while the guidelines locking new drugs away as reserve treatments were designed to keep them working as long as possible. From their own vantage points, each made the rational choice. Yet right where all those sound decisions overlap, we ended up with medicines that are hard to find anyone to make. Beside the history of what gets made, this is how the history of what rarely gets made is left behind.