Happy Friday!

This should be my last newsletter sent through Beehiiv. I've been moving everything over to Substack this week, and apparently migrating a newsletter involves a lot more fiddling with settings, old posts, and email configurations than I expected.

I thought it would be a fairly straightforward process, but of course, these things never are.

Anyway, if all goes according to plan, next Friday's newsletter will arrive through Substack!

The Secret Life of Drugs After We Take Them

Hand-drawn editorial illustration of a northern lake ecosystem inside a transparent medicine dosing cup, featuring pine trees, a yellow canoe, swimming fish, and a setting sun in navy and mustard yellow on a textured cream background.

When we talk about drug discovery, one of the things we strive for is potency. We spend years designing drugs, wanting to produce some kind of a biological response at incredibly low concentrations, yet remain stable long enough in your body to work, and hopefully don’t interfere with too many other things along the way.

But there is one consequence that we often neglect and probably never even think of. Once those same drugs leave our bodies, we treat them as waste, flushing them down the toilet and into our environment. And when they’re detected in rivers at concentrations of just a few nanograms per litre, those numbers can seem almost insignificant. Except we just spent years designing some of these drugs to be so biologically active at extremely low concentrations.

Take ethinylestradiol, a synthetic estrogen used in birth-control pills. In the early 2000s, researchers in northwestern Ontario deliberately added small amounts of this hormone to an experimental lake, maintaining concentrations around five to six nanograms per litre.

Well, it turns out fish have hormonal systems too, and they are sensitive to some of the same chemicals that affect us. Over the following years, male fathead minnows began developing female reproductive characteristics, reproduction largely stopped, and eventually the population collapsed.

What’s particularly interesting is what happened to the rest of the lake. With fewer minnows around, insects and zooplankton became more abundant, while lake trout, which relied partly on those fish for food, declined in biomass by as much as 42%.

So a tiny concentration of a hormone drug, designed for a specific purpose in humans, ended up changing an aquatic food web.

Why are we talking about all of this? This week, a paper was published last Thursday in Nature Geoscience. Researchers combined machine learning, environmental measurements, and over 1,000 toxicity assays covering 53 pharmaceuticals to estimate the ecological risks of drug pollution in rivers worldwide.

Their model suggested that around 13% of river grid cells worldwide could exceed a defined toxicity threshold from pharmaceutical mixtures. Now, that does not mean 13% of rivers are being destroyed. Remember, these are model predictions, not direct measurements of ecological damage everywhere. Still, it's quite an impact for chemicals we usually think of as healthcare products.

But the damage doesn’t just end there, it affects everything around them.

In a 2018 Australian study, researchers detected 69 pharmaceutical compounds in aquatic invertebrates and 66 in spiders living along nearby riverbanks. They even estimated that platypuses feeding on contaminated insects could ingest certain antidepressants at a substantial fraction of a human therapeutic dose, when adjusted for body weight.

That is an estimate of exposure, not evidence of antidepressant effects in platypuses, but the idea is rather extraordinary.

Now, none of this is particularly new to environmental scientists. Pharmaceutical pollution has been studied for decades, and there's even a field exploring benign-by-design medicines, where environmental persistence and biodegradability become part of medicinal chemistry.

But I wonder whether AI drug discovery could eventually make this more practical.

We’re already using computational models to evaluate candidate drugs across different properties before deciding which molecules to develop. Imagine having two candidates with similar therapeutic performance, except one persists in waterways while the other breaks down into relatively harmless compounds. If we could predict those differences early enough, perhaps they might influence which candidate moves forward.

Of course, I suspect this would be considerably harder than it sounds. A molecule that resists degradation long enough to work inside the body might also resist degradation in the environment. And predicting what happens across different organisms and entire ecosystems is a very different problem from predicting binding to a particular protein.

Better wastewater treatment and pollution controls could probably accomplish more in the near term, something the new study also highlights. I’ll also add that there is probably very little financial incentive to save the environment.

Either way, since AI is allowing us to explore an increasingly large chemical space, I think there’s an opportunity to reconsider what makes a successful drug candidate. Perhaps someday we’ll design medicines that remain active long enough to help patients, but become relatively harmless once they’ve left our bodies.

After all, we already spend so much effort thinking about what happens to a drug before it reaches the patient. Maybe we should be thinking a little more about what happens afterward.

Chart Of The Week

need seo alt text Hand-drawn infographic showing that 59% of AI drug development programmes (458 of 773) involve small-molecule drugs, compared with 41% (315) involving biologics and other therapies. Yellow tablets float in a river, illustrating the potential persistence of small-molecule pharmaceuticals in waterways.

What’s interesting is that most pharmaceutical pollution comes from small-molecule drugs. Unlike antibodies and other protein-based medicines, which generally get broken down by the body, small molecules can remain intact and eventually make their way into rivers and streams.

And small molecules happen to make up the majority of AI drug discovery pipelines. In our database, 458 of the 773 programs are small molecules, or roughly six in ten.

Many of these are being designed to work at extremely low concentrations and remain stable long enough to do their job. Which is great when they’re treating patients, but those same properties could become a problem once the drugs leave our bodies.

What Caught My Eye

Isomorphic Labs just raised another $5 billion, only six months after raising $2.1 billion. That's quite a lot of money for Alphabet's AI drug discovery company, especially considering its own drug candidates haven't reached human trials yet. The new funding will help expand its IsoDDE platform and move those candidates closer to the clinic. [Link]

NVIDIA is getting further into the business of AI scientists. This week it introduced Nemotron for Science, a collection of open models and tools that researchers can use to build AI agents for scientific work. It also released LitmusBench, with nearly 50,000 chemistry questions to evaluate these systems. The idea is to move beyond AI that answers scientific questions toward AI that can actually help carry out research. [Link]

Roche received FDA approval for giredestrant (Wilserdly), a new oral treatment for certain patients with advanced breast cancer. The drug targets estrogen receptors, including those carrying ESR1 mutations that can make tumours resistant to hormone therapy. In its Phase 3 trial, combining giredestrant with everolimus reduced the risk of disease progression or death by 62% compared with standard endocrine therapy plus everolimus. A promising result for patients whose cancers have become harder to treat. [Link]

Have a Great Weekend!

❤️ Help us create something you'll love—tell us what matters!

💬 We read all of your replies, comments, and questions.

👉 See you all next week! - Bauris

Reply

Avatar

or to participate