Fifteen minutes into helping our younger son move back to college last Saturday, he gave me the mildly patronizing, empathetic "I think you're too old to lift heavy things" face while taking boxes out of my hands and holding the door open with his foot. I spent the day watching for signs of the parent-child inflection point and asking, "Does he need my help or is he just letting me feel helpful?"
From the dorm, we flew to Reno two days late for my wife's grandfather's 100th birthday. The day before we arrived (24 hours after his birthday party), he took a turn for the worse, and our trip ended up being at his bedside for his last days. It was sad, understandable, and remarkable. He was an MIT engineer, a pilot, a stockbroker, a man with strong views about the world and exactly how his corner of it should run.
Flying home from Reno through Las Vegas to Baltimore was illuminating. The Reno airport had Burning Man stragglers heading home, shoes and luggage powdered in playa dirt and rave wear. One man wore a floor-length snow leopard coat, which seemed thematically on point but impractical for a dirt-filled desert. The Vegas airport had more men wearing leopard print, this time tights, accompanied by a large gold word necklace declaring, "It's my Birthday, Bitch." I was pleased to see the comma, which clarified the statement's focus. Near the gate, a woman who looked like a retired showgirl held the leash of a Shih Tzu. The dog wore a denim skirt and a toddler-sized KISS T-shirt, Gene Simmons's tongue running down its back.
Life is long, life is short, celebrate every birthday, live on your own terms, and I don't own enough faux leopard.
Gemini Notebook AI-generated podcast version of this week's newsletter.
Latest episode at the top
https://wair.ajwein.com/podcast
Science and Technology Trends
Physicians at Massachusetts General Hospital and Harvard Medical School reported the first successful use of a transgenic pig kidney as bridge therapy in The Lancet last week. This means a gene-edited pig kidney kept a 66-year-old man off dialysis for 271 days. Then it failed, and he went back on dialysis for another 82 days before a human kidney arrived. (He is blood type O, which has a roughly seven-year wait on the deceased-donor list). This advances our knowledge of how patients tolerate these xenotransplants, and it is an interesting (albeit probably pretty expensive right now) way to keep a patient with dialysis-related complications alive while waiting for a human transplant.
The xenotransplant did not fail from rejection as you would expect. He did well for six months. Then his immunosuppression was dialed back to let him fight off an ordinary bacterial infection, and the porcine kidney developed thrombotic microangiopathy. This microvascular injury is difficult to treat and destroyed the xenotransplant.
The Article: https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(26)01295-X/fulltext
AI-Supported Analysis: https://wair.ajwein.com/porcine-kidney-xeno-2026-09-11-v1
There is a short list of things people like to talk to doctors about at parties. Cold plunging has been on that list for a few years now, which sounds exceedingly uncomfortable to me. This week I found an X post about cold showers rather than plunges, claiming that "30 seconds of cold showering at the end of a hot shower reduces sick days by 29%." I pulled this thread and found myself deep in the rabbit hole of clinical evidence on cold exposure. There is measurable physiology. Thirty seconds of cold is enough to spike hormones and set off an acute inflammatory response, i.e., some approximation of "activating the immune system." What I could not find was a data showing a reduction in objectively measured illness. All the studies use self-reported sick days, including the trial behind this post. That trial also counted illness days, and those did not differ between groups. In other words, the cold-shower group got sick as often as everyone else, and they went to work anyway. Since it is hard to create a placebo group, I imagine the cold-shower people felt motivated to report they felt less sick (taking cold showers would be less justifiable?) Fortunately, there is very little harm in short bursts of cold water that are "cold enough to make you gasp." And watching people get into ice baths on Instagram makes for entertaining video.
X post: https://x.com/anishmoonka/status/2098705681847079100?sB
Article: https://pubmed.ncbi.nlm.nih.gov/27631616/
AI Review of cold exposure clinical outcomes: https://wair.ajwein.com/cold-shower-sick-days-2026-09-13-v1
Anti-Anti-Science
Last week I found a paper describing a naturally occurring furin cleavage site in a Brazilian bat coronavirus genome. This caught my eye because much of the recent focus on Anthony Fauci runs through the fact that SARS-CoV-2 has one. This portion of the COVID virus's RNA is offered as evidence that the virus was built in a gain-of-function experiment and, from there, that it escaped from a lab. Whether COVID came out of a wet market or a research laboratory remains unsettled.
This paper does not settle it. The Brazilian coronavirus lacks the same furin cleavage site as SARS-CoV-2 and, is from a different limb of the coronavirus family tree. Nor is "furin sites occur in nature" new news, as other coronaviruses, such as MERS-CoV, have them. For instance, a 2020 paper is titled, "Furin cleavage sites naturally occur in coronaviruses."
However, the finding is a good example of how an argument from absence works. The claim that mattered was that nobody had found a furin cleavage sit in the RNA of the coronavirus family that SARS-CoV-2 belongs to and since there is one in SARS-CoV-2, it had to be artificial. That argument treats a gap in biological knowledge as evidence for an agenda. Impugning a man's reputation based on which viruses somebody has gotten around to sequencing is a claim about the sequencing backlog, not about the man or his expertise.
Ironically, this Brazilian coronavirus paper has also been mischaracterized. The Brazilian work was posted as a preprint on October 27, 2025, and published, peer-reviewed, on September 5, 2026. During the ten-month gap, a letter appeared in a Brazilian tropical medicine journal describing the bat virus as having the same cleavage site as SARS-CoV-2. It does not. The letter also renamed the virus, which the authors term "BRZ" for Brazil. The letter redefined BRZ to be "bat-related zoonotic," implying it can infect people, which no one has shown. The letter also does not cite the Brazilian bat study anywhere in its references. It cites a vaccine review that mentioned the preprint. So, a claim traveled from a preprint, through a summary, into a peer-reviewed letter that changed the wording and was published nearly three months before the final paper it relied on.
The letter closes by calling for transparency and data sharing, "to reduce speculation and fake news about the origin of viruses."
Background on the furin cleavage site argument: https://wair.ajwein.com/furin-absence-argument-2026-09-08-v1
The paper (Nature Communications): https://www.nature.com/articles/s41467-026-77599-6
AI analysis of the paper: https://wair.ajwein.com/bat-coronavirus-furin-2026-09-08-v1
AI Impact
Everyone expects AI to shrink the clinical workforce. Dr. Dhruv Khullar thinks the economics argue against this. His NEJM editorial combines three borrowed ideas: the Jevons paradox, the lump-of-labor fallacy, and the task-versus-skill distinction from labor economics.
Jevons Paradox: more efficient steam engines burned less coal per unit of work and, for exactly that reason, burned far more coal in total. Cataract surgery got cheaper per case, and there are now vastly more cases. Cheaper cognition may produce more care opportunities.
Lump of labor: There is no fixed quantity of clinical work to be divided up. Demand exceeds supply (roughly a quarter of Americans live in a primary care shortage area), so AI extends clinicians before it substitutes for them. (Combined with Jevons and there is no limit?)
Tasks are not skills: a surgeon decides, operates, and documents. Automate the documentation, and you have automated a task, not a job. Automating the cheap steps lets one expert supervise far more output. That makes the expert more valuable and, at the same time, means you need fewer of them. (Both can be true, and the editorial argues only the first.)
He also borrows the O-ring theory: in work where a single failure ruins the whole outcome, the value of the part that cannot fail goes up, not down. In other words, the remaining humans in their loops are going to be increasingly critical.
Given all the talk of "AI is going to kill us" over the last week, it was heartening to find an editorial that was a little more upbeat about offloading tedious, time-consuming tasks that feel lower value. Shifting eyeballs away from screens and giving more time to engaging with patients is the win I hope for.
Editorial: https://www.nejm.org/doi/10.1056/NEJMp2607831
Analysis and comparison to other arguments: https://wair.ajwein.com/khullar-ai-clinical-workforce-2026-09-13-v1
The J&E Random Kidney Facts of the Week (JERKFoW!)
Loyal readers have met the glomeruli already, but it is worth focusing on them this week. Each kidney has roughly one million glomeruli, which are the parts of the kidney (the filters) that do the actual blood cleaning. To be clear, human kidneys have a wide range - autopsy counts run from about two hundred thousand to over two and a half million, and by age seventy most adults are down roughly half of the glomeruli they were born with.
Each glomerulus is a tangled ball or tuft of somewhat leaky (think sieve or strainer) tiny blood vessels, fed by one small artery and drained by another, slightly narrower one (the afferent and efferent arterioles). The size mismatch is deliberate; the narrower exit keeps the pressure inside the tuft high, so blood flowing through is pressurized, and the fluid component of the blood is pressed out (the beginning of urine). The glomerular wall is only three layers thick, with the outer layer made of cells called podocytes whose interlocking feet leave slits so fine they sort molecules by both size and electrical charge. Water and waste molecules pass through, while blood cells and proteins stay behind.
This filtering goes on 24/7/365 for decades, without maintenance. Many kidney diseases either directly damage the glomerulus or the surrounding tissue and share a final common pathway: the loss of glomeruli over time. We track this by estimating the total filtering capacity of the kidneys (through labs) and watching for blood cells or protein in the urine, which signals that the filters are damaged. Some of the most effective medications we prescribe protect the glomeruli directly (ACE inhibitors, angiotensin receptor blockers, and SGLT2 medications like Farxiga), and more recently, GLP-1 medications like Ozempic (which have been shown to slow kidney disease progression in type 2 diabetes).
The problem is that our tools for measuring glomerular function are imprecise and lag behind the damage. You can lose much of your glomerular capacity, often approaching 1/3 to 1/2, before routine blood work shows changing function. So, we kidney doctors spend much of our time worrying about risk factors, treating diabetes and hypertension as early as we can to spare the glomeruli (and the rest of the body) from damage we cannot easily see.
My job comes down to protecting a million of these little tangled balls.
Click here to see naked glomeruli! (Kidneys gone wild?)
Things I learned this week
My Southwest flight to Reno was one of the eight planes in an 817-aircraft fleet that have Starlink, which felt like winning a very small lottery. It got me thinking about my travel over the last 15 years. I gave Claude access to my Gmail to review all of my flight receipts and flight numbers, and used an FAA database to find the airframes I have flown on. While not perfectly accurate, 28 minutes of compute built me a database of my travels. Since 2011, I've been on 256 flights on 110 different aircraft. I have flown on one Southwest airframe 6 times, one 5 times, and many 3 to 4 times. I should start carving my initials in the back of seats.
My flight log: https://awflights.ajwein.com/
This week I learned the word "connectome," a complete wiring diagram of which neurons connect to which, and how strongly. On September 3rd, Janelia's FlyEM team, the Cambridge Drosophila Connectomics Group, and Google Research published the first finished connectome of an adult male fruit fly's entire central nervous system, eyes through legs. It has 166,700 neurons and 125 million synapses; it is the largest brain map anyone has made, by neuron count.
The dataset (MaleCNS v1.0) went public in June. The internet noticed early this month and got to work connecting and tasking virtual fruit fly brains and nervous systems to various computational tasks:
For now, most of these demos are not showing great outcomes. The Doom fly has not learned to survive; the trading fly has shown no profit; and someone's Mario 64 fly spends its days flying into a wall. The engineering is amazing but imagine the existential trauma if the connectome could talk. Fruit fly brains, it turns out, are best suited to letting fruit flies be fruit flies.
male-cns.janelia.org: the project site. Data downloads, the Cell paper, and a Dimorphism Explorer for comparing male and female neurons side by side.
The paper: Berg et al., "Sexual dimorphism in the complete connectome of the Drosophila male central nervous system," Cell 189(18):5504-5526 (2026). doi:10.1016/j.cell.2026.08.015
Google Research - A connectomics milestone: the machine-learning side of the fly brain map.
Cambridge Neuroscience: the most readable write-up, and the one that credits the Cambridge half of the team. Good quotes.
Janelia news release: the Gerry Rubin and Greg Jefferis quotes.
awesome-fly: a running catalog of every demo.
AI art of the week
A visual mashup of topics from the newsletter and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.
https://wair.ajwein.com/wair-art-loteria-gallery-2026-09-14-v1
This week's style: lotería, the Mexican card game that hangs on the wall of many Mexican restaurants in America.
It works like bingo with pictures instead of numbers. The deck runs to fifty-four cards, each one an image with a name and a number. Every player gets a tabla, a board of sixteen of those images in a four-by-four grid. A caller works through a shuffled deck; you cover your matches with dried pinto beans, and the first person to complete the agreed pattern shouts ¡Lotería! (or, in the older habit, ¡Buenas!).
Clean hands and sharp minds, team
Adam