The Drugs Won’t Work
The Rise and Fall of Rational Drug Design
I was in a diner in Santa Fe, New Mexico. It was 1997 and I was having an Italian soda with Anthony Nicholls, the founder and president of one of the hottest up-and-coming informatics companies in America.
By all rights Nicholls should have been out in one of the town’s expensive restaurants, being wined and dined by venture capitalists hoping to get in on the ground floor of this chemoinformatics thing, but he was tired of pretending to explain what it is he does to people who pretend to understand. So he explained it to me instead.
The problem with drug discovery — OK, it’s not just one problem, I’ll get to the rest soon — is the size of the challenge. The number of drug-like molecules that could potentially exist is huge. ‘Mind-bogglingly huge,’ according to Nicholls. How huge? ‘It would take several universes to hold all these molecules, if one of each existed in real life.’
The huge number he means is 10^200. A number that, when he gave a lecture to a roomful of chemists not long after we met, made everyone gasp. I would hear the same line trotted out about once a week for the next few years, though I didn’t know that yet.
Flash forward to now and the number of new drugs is dwindling. Costs continue to spiral. What looked in the 1990s like a quantum leap forward in medicine — developing drugs to eradicate rare and lethal diseases while making personalised regimens of treatment available to the masses — now looks more and more like a vanishing mirage.
There are still ways to make a buck in the industry. Companies are churning out tweaked variations of existing drugs to try to keep themselves in profit. Rich Western customers can expect to have new conditions spun out of thin air just to sell them a pill. And other companies, like Martin Shkreli’s Turing Pharmaceuticals, are jacking up prices on old drugs while expecting patients and insurance providers to pay.
Things shouldn’t be like this. With 10^200 potential molecules to choose from, finding a customised fit for all diseases should not be so hard.
That’s what everyone thought, anyway.
It probably helps to start with some basics. What are drugs? Usually they are small molecules of between 15 and 20 heavy (non-hydrogen) atoms. Almost every drug currently available is a small molecule of some kind. Aspirin, for example, contains 13 non-hydrogen atoms and has a typical profile of qualities drug researchers are looking for. Finding molecules like these is the bread-and-butter of the pharmaceutical business.
If you take that average size, and work in a few more back-of-the-envelope calculations, you come up with Nicholls’s number of 10^200 potential molecules. Finessing the rest of it is where the real work begins.
Nicholls was born in Plymouth, England and studied physics at St John’s College Oxford before going to FSU to study for a PhD in molecular biophysics. He is perhaps inevitably a fan of the Hitchhiker’s Guide to the Galaxy, and urges anyone overwhelmed by the hugeness of 10^200 molecules: Don’t Panic. His job is helping researchers navigate this vast ‘chemical space’ in the search for new drugs.
And then comes the ‘but.’
But, the process of sorting the wheat from the chaff is not simple. Many things look like they could be drugs — exhibit drug likeness, as Nicholls explains — but aren’t. They need to be the right size, soluble to be transported by the body, so on. And that’s only the beginning. Lipinski’s Rule of 5 is a loose set of rules describing common properties of many such orally active drugs, and is widely used as the first port of call when winnowing the herd. But everyone can name execptions to it: only about half of all approved drugs fit inside its narrow bands. Discard outliers too early and you could be missing valuable leads. That’s far from all. Understanding the interaction of small molecules inside the body, when they bind to proteins, is not straightforward.
Many on the outside of chemoinformatics believe that Big Data will revolutionise drug research. That all possibilities can be calculated in an instant, with likely candidates passed on to the clinical scientists who test the compounds in animals and humans. Given the advances in computing the public constantly hears about it is easy to imagine that the earliest steps in bringing a drug to market are the simplest ones — and that we are only one charity run, one supercomputer, one Nobel Prize away from a cure.
Analysing any items when there could be 10^200 of them is complex. Nicholls stirred the cream, soda, and syrup layers of his drink to demonstrate. ‘There is more going on in this glass than all the processing power on Earth can calculate,’ he assured me. ‘The scale of the problem with the early stages of drug discovery is similar to that.’ Brute force approaches simply wouldn’t work. Companies like his, OpenEye Scientific Software, specialise in programs that attempt to find the drugs without having to resort to brute force.
Nicholls is softly spoken (most of the time) and has a way of turning ideas into products, nudging scientists out of the lab and into coding. He cut his teeth developing the molecular visualisation program GRASP, a competitor of Roger Sayle’s RasMol and a Pretty Big Deal in the early days of computer molecular modelling. That is why he was being pursued by venture capitalists. And that’s the thing that got me to take a Greyhound bus back home to Tallahassee, Florida, pack my bags, and move to Santa Fe to work for him.
New Mexico isn’t the sort of place where you would have expected to have this conversation. With its adobe buildings and art galleries and cultural history, it is a world away from tech start-ups of the Left Coast. On the other side of the Plaza from where Nicholls and I were talking, Native Americans sat on blankets and sold silver jewellery to tourists keen to adopt the Santa Fe “look” of floaty skirts and handwoven chic. You would hardly guess that the city was an epicentre in the early years of what we now call Big Data — or as it was dubbed in a long-ago Wired feature, and a book of the same name, the Info Mesa.
The name Info Mesa may be more apt than those Wired editors realised, since the prospects of a paradigm shift in drug development rose rapidly only to flatten out just when everyone thought they were getting to the top.
One reason is the sheer size of chemical space. Even the most basic molecule screenings are still prohibitively expensive, and the research rarely goes to plan. But there’s also a lot we simply don’t know about why some molecules are good drugs and others aren’t.
Knowing what drugs have in common, such as their size and solubility, is not necessarily a predictor of success. Looking like a drug is not the same thing as being a drug. Take methyl salicylate, for example: a small molecule that’s incredibly similar to aspirin. Perhaps it could be a powerful painkiller too? But probably its only significant contribution to medicine is as the ingredient that gives us the characteristic smell of Ben-Gay.
For all the promise of rational drug design, results have been slow coming. Finding a target on a protein, where a drug molecule would join up, is like looking for a needle in a haystack. Finding the drug that fits that target in a virtual database? Like looking for a needle in ten thousand haystacks. Where every single piece of hay looks like a needle.
The term chemoinformatics is unfamiliar to most outside of the pharmaceutical industry. It refers to the uses of computer science and data technology in chemistry. And no use has aroused more excitement than the potential for rational drug discovery.
All pharmaceutical companies have libraries of chemical structures they keep in databases — some fraction of that potential 10^200. They contain chemicals that have actually existed and many more that are merely virtual. Searching these for drug candidates is a datamining problem of enormous proportions, and one the industry has poured a lot of resources into. By 2001 it was estimated that an average drug cost about $802 million before it hit the market and of that, $335 million was ‘preclinical research,’ i.e. before human testing.
There are two foci of research in this area: the small molecules that could be drugs, and the proteins that drug-like molecules bind to. Both are more complex than their formulae suggest, but the technological revolution was supposed to make such concerns irrelevant.
On paper, molecular formulae are simple. But molecules in reality look nothing like the stick-and-ball models from school. They are flexible around their bonds, and coding this is a considerable challenge. The surfaces, instead of looking like tennis balls with pencils in between, are lumpy and sometimes hard to access. That’s even before getting to properties like surface electrostatic charge. The metrics — criteria used to sift the candidates, preferably (as the name suggests, though not always) measuring something — are complex and widely debated. Visualising these things for one molecule can be computationally expensive… never mind for any significant fraction of 10^200.
I have observed and sometimes worked in this field, in Santa Fe and London, first as a webmaster and later as a software developer. At first it was an exciting community — at least in part because of the personalities around the tech scenes in Santa Fe and Boston, and the sense of being part of the next big thing. The field also benefitted from its conceptual kinship to another 90s big hitter: the Human Genome Project. Headlines from the time made many believe advances in genomic medicine were imminent. Celera Genomics founder Craig Venter and the media whirlwind surrounding him promoted the idea that we would soon be enjoying the benefits of personalised pharmaceutical treatments, both affordable and widespread.
We are only slowly waking up to the idea that genetic sequencing is not the quick fix everyone thought it was, and that the promised cures seldom pan out as expected, in part becasue single-gene-single-trait connections only explain a small percentage of illnesses. Common diseases more likely to come from MAGOTS, or ‘multiple assorted genes of tiny significance.’ What Venter was really selling was a myth at the heart of the industry’s flawed business model: the idea of the lone genius breakthrough in biochemical research.
For all practical purposes, this myth starts with Paul Ehrlich, who pioneered the treatment of syphilis with the discovery of the drug arsphenamine in 1910. Ehrlich’s work codified the enduring image of pharmaceutical scientists as we imagine them now, and also supplied the template for the process of modern drug discovery.
Ehrlich looked for compounds that targeted disease-causing organisms, hoping to find one toxic enough to kill the organism without killing the patient. He called this the magische Kugel — the magic bullet. He focused on derivatives of atoxyl, or arsanilic acid, which is both highly toxic and showed high binding affinity with its target, the treponemal bacterium that causes syphilis. He then tested these leads to determine which ones had suitably drug-like qualities.
Ehrlich’s magic bullet theory worked and created something new under the sun: a drug that had no prior relation to any traditional remedy (unlike, say, aspirin, which was a derivation from a known ancient cure found in willow bark). Arsphenamine was supplanted by penicillin as a treatment for syphilis by the 1940s, but the principle of the magic bullet was already enshrined as a founding precept of modern drug discovery.
Ehrlich succeeded not merely in making a drug, but also in birthing a legend. The 1940 Hollywood film Dr Ehrlich’s Magic Bullet popularised a fable of Ehrlich — and by extension, any heroic research scientist — as a rule-breaking, bureaucracy-defying maverick. In the Tinseltown account of drug discovery, Ehrlich’s other character traits were shunted aside. His struggles to establish his career as a German Jew, for example, received no extended treatment, even though it was obviously a timely and topical subject in 1940. Such would only clutter the film’s central narrative point: the canonisation of a new kind of scientific folk hero — free of religion, prejudice, or any of the normal collaborations and struggles that mark most scientists’ careers.
This view of Ehrlich’s singular genius became so widely accepted that the process he created became a template — which meant, ironically enough, the complete standardisation of research. The steps Ehrlich used, from target identification, to candidate production, to screening, are a lock-step method that still governs the course of drug discovery today. This is the pipeline that drugs take before they move into animal and human clinical trials and then, if all goes well, into production.
For a while it worked. As the century advanced, so did the science. Researchers acquired some understanding of what drugs do: interfering with certain cell products, or mimicking them, or stopping cells from reproducing. Knowing more about both the small molecules themselves and the larger proteins they interacted with was helpful. For several decades the number and availability of effective pharmaceuticals grew dramatically.
By the 1990s the pace of discovery had started to slow. The time it took for a drug to be developed increased, in part due to more stringent trial requirements, in part because medicine had moved on to more difficult problems like chronic conditions rather than acute infections. Researchers and pharma companies were untroubled by the slowdown — they assumed that it was a prelude to the industry’s next big breakthrough. Investors and researchers pinned their hopes on a new age of rationally designed drugs.
The first part of this approach was the needle-in-a-haystack quest for the detailed structure of a biological target on a protein. Crystallography allowed scientists to visualise proteins at the precise point where they join up with a small molecule. Results were stored in the Protein Data Bank, providing a freely available map of the target future researchers might be looking for.
Advances in combinatorial chemistry were expected to stretch the field of potential drugs to match these targets. Where Ehrlich had to synthesise his potential drug molecules by hand, a time-consuming process that made hardly any dent on the size of chemical space, combinatorial chemistry used newly available equipment and methods to produce orders of magnitude more compounds in the lab. Producing new compounds quickly allowed their basic properties to be recorded and stored in case of future use.
As the price of computer processing power and storage tumbled in the 1980s and 1990s, pharma companies began stockpiling all these data in their proprietary virtual libraries, which housed the profiles of billions of small molecules. This was still only a fraction of Anthony Nicholls’s estimate of 10^200 potential drugs, but growing all the time.
In order to be effective, big libraries require big screening. Pharma researchers pioneered a process known as high-throughput screening (HTS). The procedure employed robotic lab setups to test thousands of small molecules for drug-like activity. This was an encouraging start, the industry agreed; but the holy grail of library-screening was the idea that it might eventually be done in silico, also called virtual screening, without even having to make the molecules at all.
With luck, if candidate molecules are effective but not toxic, and don’t cost the earth to produce, then a compound might become a drug. But did the much-lauded advances prove to be only hype? Well, yes. Decades of combinatorial chemistry and high-throughput screening managed to produce exactly one drug from scratch: sorafenib, used in treating advanced renal cancer. It costs US patients about $100,000 a year and is thus out of reach for many. A very muted success considering the investment of billions of dollars across the entire industry.
You wouldn’t have guessed that, though, if in the mid-90s you’d happened to pick up a copy of The Billion Dollar Molecule, Barry Werth’s tome celebrating the early years of Boston’s Vertex Pharmaceuticals.
The book, a study of the head research team at Vertex, reads more like a technological thriller than straight journalism. We are introduced to the thrusting young boffins of Vertex holding court at the World Trade Center — a touch that now seems sadly prophetic — as they confidently bat away questions from misguided investors who have committed the cardinal sin of the Information Age: not getting it. The ruthless self-belief as Vertex’s founders subvert Wall Street suits is lauded like a second round of David vs Goliath.
What readers of the bestseller would have missed amid this personality-driven clash of cultures was that any ‘billion dollar molecule’ was at that stage only a theory. And unlike many tech fields that were gaining attention in the mid 1990s, drug discovery is not exclusively the domain of the young and ballsy. On the contrary — like all scientific research, it builds its base of knowledge slowly and accretively. Behind the swashbuckling boardroom intrigue, Vertex’s researchers were relying on the considerable scientific base built by well-established professors and their labs from the 1980s and earlier: Harold Scheraga, Johann Gasteiger, Peter Willett, among many others. Academics first and almost exclusively.
But, as with the film version of Paul Ehrlich’s life, that is anything but a bestselling narrative — and so The Billion Dollar Molecule hinged instead on the dot-com idea that emerging technologies and a fairytale dusting of investment cash would produce the necessary fuel to propel drug research into its new millennial frontiers. Vertex were, of course, bounteously rewarded for their audacity with heaps of investment.
The actual course of the research in question never came close to following the Silicon Valley script. Vertex had been in business six years when Werth’s book was published in 2004, and a publicly traded company for four. It would be another 16 years before the company would show a profit. Indeed, anyone reading Werth’s book today would have trouble matching up the next-generation-of-geniuses narrative with Vertex’s modest track record. Vertex researchers collaborated with GlaxoSmithKline on a pain treatment in 2005. They released a hepatitis C treatment in 2011, with marketing carefully targeted at rich Baby Boomers to justify the eye-watering cost.
While the hepatitis drug was indeed a success, Vertex managed to burn through $4 billion in R&D in its short history. Telaprevir, marketed under the names Incivek and Incivo, brought in $2.5 billion in revenue — not even balancing the books before it was forced off the market by a competing drug in 2014. Not exactly an investment win.
These turned out to be ‘billion dollar molecules’ only in the drearier sense that they involved a huge commitment of resources, trial-and-error research, and incremental progress resting on a vast corpus of prior discoveries.
Even so, boosters of pharma might still contend that Vertex is an outlier — and that, once the game-changing breakthroughs in molecular data happen, they will be outshone and largely forgotten. Werth’s update to the story, The Antidote, published in 2014, skims over past disappointments and focuses almost solely on now-unavailable telaprevir. The timing of the book must be a bitter pill for those Vertex had to let go along the way. In November 2013 the company announced plans to lay off 370 people, or 15% of their workforce, half of those from their Boston offices. Somehow, out of this, Werth conjures a company in rude health. There’s one problem with his prognosis: It’s the downsized Vertex of disappointment and layoffs that is emerging as the real outcome for research across the industry. It’s a story that Werth minimises in his quest for tautly written biotech business adventures starring solo adventurers.
Vertex’s job losses were not unique. In an unfortunate coincidence, the Swiss pharmaceutical company Novartis announced precisely the same number of highly-skilled job losses from its research unit in Horsham, West Sussex, in the same month. If these been isolated incidents, it would be worrying enough. But this was far from isolated.
In a single quarter of 2012, AstraZeneca cut two thousand R&D jobs in an effort to ‘streamline’ its drug discovery process. In 2013, Merck announced 20% of its 80,000 jobs would go. GSK predicted $250 million savings would be found from its research units, and Pfizer unceremoniously dumped $3 billion from its research budget after acquiring Wyeth in 2009.
Whether a company is pharmaceutical stalwart like Pfizer or a small biotech like Vertex, this much holds true: you are only as good as your next product. The Antidote is on shelves, influencing a new generation of excitable business types whose pulses race at the sight of financial success, not deep science. Treating Hep C might make them feel like gilded age philathropists but if the result is only another erectile dysfunction drug, well, that’ll do. They are desperate for successes that are now few and far between. The reality behind the swagger is that no one can predict when the next breakthrough pharmaceutical product is coming.
No one ever could.
When you head north out of Santa Fe on Highway 84, keep your eyes open for a black and yellow sculpture peeking out from the scrub on the side of a hill. This giant stick-and-ball model of a molecule was erected in a more confident time to launch the new headquarters of Daylight Chemical Information Systems in 1999. The sculpture, called Cognition Enhancer, depicts a once-promising drug lead that never made it into development.
The sculpture was commissioned from Steve Klein, a New Orleans sculptor. When unveiled it was a glossy and impressive thing, a bit like some giant child’s toy dropped into the desert. The years since 1999 have not been kind, either to the oversized geegaw or the company that it symbolises. Cognition Enhancer these days is peeling and starting to rust; graffiti taggers have left their marks on the lower atoms. The headquarters building, too, is unkempt and empty: testimony to a vision of a bustling workplace that never took firm root in the desert sands.
When it was built, though, things looked very different indeed. The tech scene in Santa Fe in the 1990s loosely coalesced around David Weininger and his company Daylight Computer Information Systems. Weininger, a self-taught polymath and school dropout with a past in motocross racing, was well known for being the inventor of SMILES, a language for expressing the structure of a chemical (not simply its formula) in a single line of computer-friendly ASCII.
For example, take the aspirin-alike molecule methyl salicylate again. Its chemical formula C8H8O3 only gives us the number of each of its consituent elements but tells us nothing about what the molecule looks like. The SMILES notation, on the other hand, O=C(OC)c1ccccc1O shows the benzene ring, the attached functional groups — all for not very many more characters (hydrogens in normal valence arrangements don’t need to be shown). From this, a 2D or 3D representation can be created.
Scientists originally came to northern New Mexico from the 1940s onwards to work in the Sandia and Los Alamos labs. Fifty years later, research establishments still thrived, though with perhaps less purpose than they originally had. A proportion of the state’s scientists in the 1990s — many influenced by Weininger himself — left their careers in lab research to start companies of their own. In spite of a lack of business-friendly infrastructure, Santa Fe was becoming a hotbed of tech activity. Everyone thought they were going to found the company that would write the software that would cure cancer or stop ageing.
Weininger’s reputation attracted dozens of people to the area, probably more. Colleagues paid tribute to his ‘reality focusing field’ – a play on the infamous ‘reality distortion field’ used by Apple’s Bud Tribble to describe Steve Jobs’ powers of persuasion. Weininger, too, had a way of making people believe anything was possible.
Anthony Nicholls was one of the scientists enchanted by Weininger’s vision; he left a position at Columbia University for the research frontier in Santa Fe. I joined the cross-country pilgrimage not long after and was similarly under the Daylight spell.
Perhaps inevitably, because it was the 90s, Nicholls’s company consisted of him, a Silicon Graphics box, and a corner in his front room piled high with O’Reilly paperbacks and back issues of Nature and Science. This was fine by me: I’d just come from hanging out in a laser lab with physical chemists for four years, as a kind of anthropologist/mathematician/mascot.
My first task was to write OpenEye’s first website in a vi editor at a computer lab, then buy Ant an accordion folder to organise the company’s paperwork. One of my brush calligraphy squiggles became the company’s logo. When Nicholls and his colleagues wrote papers, I translated their English into English (anyone who has worked extensively with computational chemists will know what I mean by this). In return, as well as pay, he taught me to program in C and introduced me to a group of people I started to think of as “my tribe.” People who had opinions about Blade Runner, compilers, and the novels of Ursula K LeGuin. Nicholls’s corner of the Info Mesa was, admittedly, not yet as glamorous as it would later become.
At the time Nicholls and Weininger were thick as thieves. And while Nicholls had a strict no-VC stance, that didn’t stop him letting them buy him dinner and drinks. Sometimes I went along and we’d laugh afterwards: a couple of scruffy dorks rabbiting on about obscure punk bands while some rich guys in suits pretended to be interested.
But others were interested in more than a free meal, and Santa Fe soon had no shortage of tech start-ups rolling in dough. Plenty of people went from conference t-shirts and bicycles to linen suits and Jags, seemingly overnight.
Weininger’s Wednesday high teas were the stuff of legend. Weininger and his then-partner Dawn Abriel held court in their eclecticly furnished adobe house on the hill that had once belonged to sci-fi writer Roger Zelazny. The assembled company, too, was eclectic, and smitten with the all-but-palpable hope of greater things to come — a mix of scientists, writers, artists, and anyone else who caught the couple’s roving interest.
Within an hour of my introduction to Weininger he had winkled out my idea for a business. In short order, he put me on the phone to a potential investor — who turned out to be Michael Nesmith, late of the Monkees — even though I had nothing in the way of a proper pitch, and the brooding guitarist, quite sensibly, never forked over any cash for me to play with. Abriel, an emergency room doctor and jewellery designer, had an elaborate workshop creating custom furnishings from exotic woods like wenge and purpleheart. Weininger fought a seemingly never-ending battle with the city council over the observatory he put up in the back yard. I once sat up all night looking through the telescope there, to learn at sunrise that the tall English guy I’d been chatting with was none other than Douglas Adams. Nobody who has ever met Weininger would find any of this unusual.
(If at this point you’re wondering if a particular couple in Breaking Bad was modelled on any of the people described here: know that I’ve wondered the exact same thing myself.)
One of Daylight’s heavily funded spin-outs, Metaphorics, seemingly consisted of Roger Sayle coding in a small office, while two other guys wind-surfed on the Pacific Coast in between presentations about products that were never released. I’m exaggerating, but only a little. Everyone in the company had the title ‘Vice President.’
Drug discovery was having its moment of irrational exuberance.
If you’ve made it this far in the story waiting for the moment of implosion, well, there wasn’t a moment as such. Unlike that other 90s phenomenon the dot-com crash, no bubble suddenly burst. It was more like a red giant star slowly sloughing off its layers. Things didn’t so much collapse as deflate, break up, and drift away. Big investment and big data did not lead to big breakthroughs. Money moved elsewhere and so, eventually, did the people.
On the day of the Daylight CIS building’s official opening, a photo was taken of staff and friends (me among them) standing under the huge hunk of metal. There were about thirty of us. Little did we know it was probably the biggest group of people who would ever visit the place. That photograph is now nowhere to be found — much like the company headquarters itself.
One tourist who spotted the sculpture a few years ago was intrigued by the unusual molecule and sought out Weininger’s business partner Yosi Taitz, who runs what’s left of Daylight CIS from an address in California. The blogger commented that no one seemed to be in the building the day he visited Daylight. Little did he know that was true every day.
Another website theorises that Daylight and its sculpture were part of a vast worldwide conspiracy of Illuminati, suggesting the software tool DAYCART was part of an ‘engineered pandemic’, and that an old Daylight press release from the chemoinformatics glory days constituted sufficient evidence for ‘criminal prosecution of suspected murderers.’
Such theories might make for lurid clickbait, but the truth is more banal. By the time the building opened Daylight was already losing key staff. Weininger hardly spent a single day in the office whose design and luxury finishings he had obsessed over for years.
From the mid-Noughties onward, investors in cheminformatics startups and small biotechs got antsy. Where were the new drugs? What happened to the era of computerised drug design we were promised? Countless people inside and outside the pharma scene who believed in some version of a billion dollar molecule assured everyone the breakthrough was coming. Alas however, none came to pass.
As the failures became more noticeable, factions entrenched themselves. Software companies that once had enjoyed genial relationships became suspicious of each other. Scientists circled the wagons around their favoured approaches, trying to show why this analysis or that metric really was the solution we all had been looking for. Shape, chemistry, surface properties. Support vector machines. Clustering. Fingerprints. Fragment based drug design, ligand efficiency. The more the belt tightened, the faster the cycle of fashionable buzzwords seemed to spin. This desperation, however, did not stop the cuts that now were inevitable.
Many have taken a stab at identifying the exact problem but here, too, a magic-bullet theory has proved elusive. It could be that the low-hanging fruits — the drugs that are easy to find and produce — have all been harvested. Or it might be that we ask too much by expecting every next drug to be better than the last. Maybe the diseases now crying out for treatment are in fact the ones most resistant to drug therapy. The science might be insufficient, the hardware might be limiting. So many mights and maybes where previously there were soons and nexts.
One thing seems certain: technology can’t plug the gaps in scientific knowledge.
The flow of big data has not proven effective in stopping drugs that are destined to fail from making it to later stage development. The numbers are sobering. Of the drugs that fail in clinical trials, 43 percent are withdrawn as not cost effective, while 52 percent fail for reasons of efficacy and safety. Critics like Ben Goldacre have excoriated the industry’s methods, noting self-serving ways that pharmaceutical companies report the results of trials; now, some are also starting to question the track record of animal testing.
The costs of a failure at clinical trial stage far exceed the cost of abandoning a drug in the earlier, informatics-heavy stages. As a result, the industry is becoming more risk-averse — another ironic, though entirely sensible, turn in a field founded on the lone-genius model of research. Public pressure for faster approvals and new classes of drugs are tough to reconcile with what happens if it all goes wrong: bad side effects, lawsuits, or worse.
US and EU legislation has offered incentives for companies to pick up unfunded “orphan” drugs, including extended patents before the drugs go generic. Drugs for rare diseases are now the fastest-growing sector in US pharmaceuticals. Martin Shkreli — the man who raised the price of a single pill of Daraprim from $13.50 to $750 per tablet — originally founded his company to focus on treatments for exactly this class of disease. Critics have accused companies of reaping excessive profit from rare diseases, including childhood cancers. Those critics are probably right.
At the same time, charity-based funding for basic research dried up in the wake of the 2008 recession, leaving profit-oriented businesses the only ones tending the flame. As a researcher into causes of paediatric carcinomas — the field I moved into after chemoinformatics — I watched sources of cancer research funding wither, then disappear, every year after 2006. My disillusionment with drug discovery pushed me back into basic science, but even there the fallout was unavoidable.
The numbers tell only part of the story. If all drugs were headline-grabbing breakthroughs, research units in big pharma would not be worried about whether their new tech was working or not. Not every drug earns back the amount of money put into its development. Not every drug will be a runaway success, a Viagra or a Lipitor, buoying a company’s many failures. Big spends on marketing the few successes are also, some claim, outstripping R&D investment to the industry’s detriment.
Meanwhile august organs like the Wall Street Journal claim the “answer” to scandals about high prices is “more drugs, faster” with no less than Hillary Clinton and Bernie Sanders bravely leading the call. Well turn me over and call me a biscuit… do they really think this hasn’t occurred to anyone before now?
The media’s sudden obsession with arguing the how and whys seems quaintly out of touch when you consider the situation on the ground: namely, the layoffs that are shuttering R&D divisions across the industry. The way things are going, pharma — especially UK pharma — risks losing both a generation of accumulated knowledge and the best young scientists working in the field.
Friends who still work in R&D, many of whom refused to be quoted on the record, feel the culture of drug companies is the problem. On a macro level, the business is about finding treatments and cures. On the micro level, it is about meeting internal goals that are not always compatible with discovery.
One anecdote involved a molecular modelling group in a UK-based pharmaceutical company. According to my source, staff were told not to criticise anything in a meeting unless they also had a solution to the problem. But what if something is obviously wrong, with no obvious solution? Too bad. It gets passed forward and becomes someone else’s problem. Others described being put into teams whose objectives were designed to fail from the outset, spending the last several years of their careers on work that would inevitably be shelved.
While I had long since moved on from chemoinformatics, I watched the fallout as friends and former colleagues were affected. As the prognosis for drug discovery grew worse many decamped to other countries, other careers. One close friend who moved from chemoinformatics research with a pharmaceutical company in the UK to software support in the US talks casually of having been ‘trained for a job that doesn’t exist.’ Will he ever go back to working in research again? Probably not.
The loss of talented researchers couldn’t come at a worse time. Bernie and Hillary are at least in the right ballpark. The need for new drugs is urgent: with medics warning of treatment-resistant diseases emerging worldwide, the success of the pharmaceutical industry and rehabilitating its public image requires more than churning out new ways to treat first-world problems. It simply seems unlikely that a top-down order from the giddy heights of political pointscoring would achieve this.
I heard about projects that were destined to fail being supported while others with potential were shelved, in the name of meeting internal benchmarks that had nothing to do with cures. Management with no lab experience were sold on technologies from some genius-come-lately without understanding how integrating them might affect other research. Your ongoing project uses software from a supplier that has fallen out of favour? Say bye-bye to your results. New tech was shoehorned into the development pipeline and expected to produce immediate results.
‘The research leadership put too much faith in technology,’ agrees Peter Kenny, a computational chemist who previously worked at AstraZeneca. ‘The individual technologies weren’t that bad; they just needed to be used together.’ Having accepted early retirement, Kenny is free to speak his mind (not that this was ever one of his problems). He admits that elements of R&D departments were bloated, but by managers, not scientists. Excellent scientists like his colleague Andy Grant were forced out of day-to-day science and up the corporate ladder while managers with no experience of running long-term research were parachuted in.
Kenny notes that for a long time researchers have been afraid to tell the truth to executives when things aren’t working. Saving embarrassment in the short term has stored up problems for the long term. To borrow the term from another 90s relic, the X Files, we wanted to believe.
The last two times I’ve seen Anthony Nicholls had nothing to do with work. He was a witness at my wedding in Tesuque, New Mexico in 2010, and me and my husband spent about an hour chatting with him over eggs and green chile just last year. I moved away from Santa Fe in 2000 to pursue a PhD in forensic pathology, inspired in part by an OpenEye surface model of a crown ether that reminded me of the human os coxae. This kind-of sort-of led to me becoming an internationally bestselling author. Ant is now the kind of guy with millions of frequent flier miles, who FedEx’s his luggage to Japan instead of checking it in. He bought, then sold, Bob Dylan’s ranch. I think it’s even possible he doesn’t get girlfriends to cut his hair anymore. Yeah, it’s fair to say things have changed.
OpenEye is rather a larger company than when I worked there, with sprawling main office and branches in five countries. It is one of the few vestiges of the late 90s Info Mesa still hanging on, in spite of — or perhaps because of — not taking VC funding and not offering public shares.
These days Nicholls describes the pharmaceutical industry as ‘so manifestly important — and so manifestly stupid’. As he sees it, for every step forward there have been two steps back. Relaxing advertising restrictions, says Nicholls, ‘changed forever how the industry is both viewed and run.’ When budgets were rerouted to adverts, the lab was relegated to second place.
Another problem according to Nicholls (and rather predating the computing hype) was the 1980 Bayh-Dole Act and similar moves in Europe that encouraged academia and nonprofits to leverage intellectual property. With the go-ahead to profit from patents that come out of government sponsored research, universities started to operate more like companies. Scientists now have to keep an eye on patents and profits as well as their research. Many must secure their own salaries through attracting grants, even if they are contracted university employees. Plenty of universities act as incubators hoping to get a slice of any profits. Ideally science should happen in public and be peer-reviewed for all to see; in reality, protecting IP prevents much work from being shared. The conflict between the idealism of science and the bottom line of profit has always dogged the pharmaceutical industry, now that conflict is endemic in universities.
As technology transfer — or ‘TechXFer’ as it is sometimes cringingly known — took off, the effects of Bayh-Dole really started to show. Venture capital in the Info Mesa certainly wasn’t restricted to the former academics and their start-ups, a goodly number of the sleekly-appointed offices and prestige conferences springing up also belonged to nonprofits. After all, it’s not profit if you spend it on salary and perks, is it?
The money part wouldn’t matter as much if the research worked, but it didn’t. ‘Streamlining’ early R&D fails to looks at the problem holistically. The difficulties in drug discovery may start in early stage research but they don’t end there. There is still concern about the rest of the process. The high dependence on specific strains of animals for testing is one of these. When some diseases such as, say, tuberculosis manifest differently in mice than they do in humans, are potential treatments being lost because the animal analogues we use aren’t appropriate?
Critics have also hit out at the failure of female animals to be used in testing, and of human trials to recruit enough women to understand sex-specific effects. Only in 2013 were guidelines updated to recommend half-doses of Ambien for women, because the drug takes longer to clear their systems. Considering it’s been on the market since 1991, and was in development for years prior, this inspires little confidence in the safety of drugs that do make it into production.
One problem with the pharmaceutical approach is its very mechanism of action. The fact that a drug can stop a symptom or infection is not always the same thing as a cure. The problem with the approach is brought home especially with drugs prescribed for mental health conditions. Depression is probably not caused by a deficiency of serotonin per se (as levels of serotonin in the living brain are impossible to measure, no one knows), but for the last several decades we have treated it as if that’s exactly what it is. With, it hardly needs saying, disappointing results.
It would be a great irony if Ehrlich’s magic bullet that opened the door to the modern drug design pipeline in the first place was also responsible for holding back discovery and producing substandard results. The steps he used worked so well for the discovery of arsphenamine, but does that mean the pipeline based on them is the only valid way to find cures? What might a new paradigm even look like?
There are animal studies that might provide a start of an answer. Researchers have been trying to give mole rats cancer for years, but these long-lived creatures have resisted it. Not through any treatment; they appear to be naturally resistant to carcinogens kill lab rats (and people) in countless numbers. Meanwhile, life extension scientists who have long noted the tendency of animals fed on low-calorie diets to have increased lifespans have started theorising on the mechanisms why this not only seems to work, but also to combat many chronic diseases. In both cases, the real answer is probably multifactorial and would require truly cross-disciplinary research to fully understand. And the end product to consumers, if we ever get one, is unlikely to come in pill form.
These and similar research hints at a new way of fighting disease. Rather than identifying a malady and searching endlessly for target and candidate drug, what can we do to advance the science of prevention? We already know to eat better and exercise more, but that’s like teaching the alphabet over and over when we need a whole new Shakespeare’s Folio of work. What more can be brought to this fight?
I wish I had a better answer. I parted ways with chemoinformatics after an unsatisfying spell with a London start-up in the mid-2000s, and moved on to epidemiology and cancer research. I keep an ear to the ground for any gossip. Partly out of professional concern, partly out of the irresistible urge we all have to stop and rubberneck at the scene of a disaster. But there is little to report as more friends and colleagues leave the field behind. Perusing the literature shows the same questions being asked, in the same ways, by a decreasing circle of people scared of being too critical. No one who still has a job wants to lose it.
Back in Santa Fe the names and offices of those who took venture capital’s golden ring have faded into obscurity. One friend went back to driving the 1998 Subaru he showed up in town with. Those who could cashed out, those who couldn’t, moved on. Nobody talks about IPOs anymore. I’m kind of glad that era has passed and try not to ask what they lost along the way.
A solution to pharma’s current crisis is unlikely to come from a single direction. Ben Goldacre in his 2013 book Bad Pharma highlights the need for transparency in late-stage clinical trials. Goldacre laid bare the dirty secret that many had whispered about for ages: not only are drug companies ruthless in promoting their products, but just as ruthless in making certain they make it to market at all. If that happens to mean stopping unsuccessful trials in humans and burying the results, so be it. Goldacre called for all clinical trials to be registered and reported, no matter what the results, and the British Medical Journal-supported AllTrials initiative is the result of that.
But even that necessary reform only addresses the end result of a process that is crying out for change at all stages, the early ones most of all. The tendency to pass bad results forward results from a corporate structure where mistakes can’t be criticised and scientists are afraid to speak up. The habit of buying in the newest, shiniest technology regardless of whether it has a track record of working — or worse, preferring companies that actually have no proven results simply because they’re novel — poisons the well for drug discovery long before any human trials even take place.
Make no mistake: by the time you have heard of a drug at all, so much has been invested in it that no one wants to lose the money or the face in admitting failure.
Any changes to the development process will adversely affect the bottom lines of big pharma companies even if they introduce efficiencies in time — which means, in turn, that venture capitalists and shareholders will continue to demand more and faster drug discoveries, despite the fast-multiplying obstacles to the process.
If politicians really do commit to changing the future for the pharma industry, who are they most likely to take their cues from: the market with its eye forever on the bottom line, the suits who don’t know what is going on, or the scientists who are struggling to keep their heads above water?
Almost everyone got into this business convinced it was about finding cures. Almost everyone had that belief ripped out of them a long time ago.
Still, for all the chastening developments that have assailed drug discovery over the past decade-plus, Nicholls is still a true believer. He wipes up the last of his green chile with a scrap of tortilla and argues, not for the first time, the science will out. ‘Surely by now we as a community have figured out that most spectacular advances do not come from expected directions, from directed research?’ Nicholls says. ‘They come from discovering the unexpected, from the blue sky, from just doing something because it’s cool.’
Dedicated to the memory of J Andrew Grant.