CT Scan #4: Fascinating 2023 breakthroughs
I decided to make this a yearly tradition. Similar to last year, there’s no particular curation criteria for this list other than developments that I find fascinating and have the potential to change and advance humanity. Unlike last year, I didn’t look at other breakthrough lists (i.e. from The Atlantic), so hopefully I include breakthroughs that might be overlooked by the media.
1. AI as engine for scientific discovery
The most remarkable breakthrough of this year has been the use of AI for scientific discovery. So far, we have only had one mechanism to do science, it involves human brains, a lot of work, luck, and insights. Basically, a lot of things need to go right to produce a scientific discovery like penicillin or CRISPR. What AI offers is a different mechanism, one where a great deal of the scientific process is done by algorithms and machines, and not humans.
There exist billions of inorganic materials (materials not based on chains of carbon atoms) that haven’t been discovered by humans that could vastly improve batteries, microchips, etc. The first story involves two AI systems developed by DeepMind and Berkeley University that create novel inorganic materials. One of the AI systems, graph networks for material exploration (GNoME), tweaks the compositions of known materials from the training data, computes their stability, and predicts their crystal structures. This system was able to come up with 2.2 million new inorganic compounds, adding an order of magnitude to the stable structures known by humanity. The other AI system, A-Lab, is a robotic system that can synthesize these compounds. The robotic system takes the recipes created by GNoME, weights and mixes ingredients, tweaks temperatures, adds gases to create the new materials, and finally, analyzes the materials to inform the iterative AI system. This system doesn’t only create the materials, but it comes up with synthesis recipes proposed by an LLM and learns to generate better procedures to make the synthesis more efficient.
In another paper, also by DeepMind (how do they do it?), an AI based on LLMs called FunSearch was able to create new mathematical knowledge. The LLM system is based on Codey, basically Google’s PaLM2 but fine tuned on computer code that outputs functions that solve specific problems. These code functions are assesed by a systematic evaluator that scores the functions and sends back the best ones to the LLM. This system was able to come up with code that correctly solves in a complete novel way the cap set problem (a central problem in extremal combinatorics).
2. AI can create novel proteins and discover antibiotics
Proteins are responsible for all biological functions that dictate life, from the glow of a jellyfish and the growth of a plant, to the the creation of antibodies that fight diseases in our bodies. However, it takes billions of years of evolution to create them. The breakthrough that kick-started the AI-driven scientific movement was AlphaFold (guess who developed it?), which is able to predict the structure of a protein based on its amino acid sequence. Although there exist a vast number of proteins in the natural world, there are many problems that evolution has not solved, hence the need for novel proteins. In a paper by David Baker’s lab, an AI system based on diffusion models called RFdiffusion, was able to achieve unprecedented performance in the design of novel proteins. In another system also based on diffusion models called Chroma, it was shown that the generated sequences result in proteins that can fold and have favorable biophysical properties.
In another study, researchers from the Broad Institute at MIT and Harvard, created an AI system that discovered a new class of antibiotic candidates. The system tested the antibiotic properties and cytotoxicity of more than 12 million compounds, from these, a subset of 283 compounds were empirically tested and some were found to be effective against a couple of the hardest-to-kill pathogens.
3. GLP-1 drugs
The main character of this story is a hormone called glucagon-like peptide-1 (GLP-1). This year, clinical trials based on GLP-1 agonists showed remarkable outcomes as a drug for obesity and beyond. One clinical trial of 529 people with obesity and heart failure showed that after one year, people taking the GLP-1 agonist drug doubled their heart improvement. Another clinical trial of 17,000 people with excess weight and cardiovascular disease, indicated that people taking the drug had 20% less chance of developing heart attacks and strokes. Also, more clinical trials are underway to evaluate this same drug in treating drug addictions, Alzheimer’s, and Parkinson’s.
How does this drug work? Everything starts with the Gila monster. This species is the only venomous lizard in America, and it turns out they have a hormone in their venom called exendin-4 that is very similar to the hormone produced in the small intestine of humans (GLP-1). The GLP-1 hormone triggers insulin in the pancreas, helping to regulate blood sugar. What’s interesting about exendin-4 is that it degrades way slower than the human form of GLP-1, making it attractive for a drug. Exendin-4 was used to create a synthetic hormone called extenatide, which was approved by the FDA in 2005 for type 2 diabetes. Extenatide was the first drug that mimics the function of GLP-1.
You might ask what’s the relation between type 2 diabetes and obesity? It turns out that when blood sugar (glucose) is high, food cravings are also high. This might sound counterintuitive, if there’s sugar in the body why is there a craving for more food? Well, this is where insulin comes in. Without insulin (the hormone that allows cells to absorb glucose), our brains cannot use sugar to function, so our brains send signals to eat more. This is one of the reasons why GLP-1 drugs help with obesity and weight loss. As they help with the production of insulin, blood sugar is regulated, and hence the control of food craving. Additionally, GLP-1 slows the rate of absorption of nutrients into the blood stream by reducing gastric emptying, making people feel full longer. GLP-1 also impacts the hypothalamus, which controls hunger signals (reducing hunger and cravings).
Novo Nordisk developed semaglutide, which is a peptide similar to GLP-1. It is sold as Ozempic for diabetes (similar to extentide), and as Wegovy for weight loss. They are self-administered once a week with an auto-injector. These are the drugs used in the clinical trials mentioned above. Because of these blockbuster drugs, Novo Nordisk just became the largest company in Europe.
4. Functional egg cells from male cells
Babies from 2 dads? Something like that. Researchers from Osaka University published a paper that showed the creation of female germ cells (oocytes) from male mice cells. This is kind of a big deal because so far, the only way to create oocytes is from female cells. The field of in vitro gametogenesis (reproduction through stem cell reprogramming) is getting very exciting. A breakthrough from last year included the creation of an embryo like structure using mice stem cells. This paper used chromosomal arrangement to show that oocytes can be generated from male cells. This is relevant as it opens the door for same-sex reproduction, fertility treatments, reducing sex-chromosome disorders, and protecting endangered species.
5. Brain-Spine prosthesis restores motion
In an impressive paper by Courtine’s group, researchers showed the restoration of standing, walking, and climbing stairs of a human after spinal cord injury. Spinal cord injury interrupts the communication of signals between the brain and the spinal cord, preventing signals from reaching muscles and resulting in permanent paralysis. This group has previously shown that by implanting electrodes in the epidural region of the spinal cord and providing patterns of electrical stimulation, locomotor circuits in the legs can be activated to restore walking. In this paper, the team also implanted a set of electrodes in the brain. The signals from these electrodes were decoded and transmitted to the spinal cord implant, allowing transmission of cortical signals to the spinal cord. What’s really exciting is that after several sessions of rehabilitation with the stimulation, the participant was able to walk with crutches even when the stimulation was switched off.
6. Realistic in vitro embryo models
The creation of a human being is a remarkable process. What’s even more remarkable is that everything starts with a single cell. We go from a single cell to a whole body. The mechanisms that occur during this development are hard to study, as there are ethical considerations and so far we haven’t had good tools. There has been tremendous progress in the past years. A breakthrough from last year included the creation of an embryo like structure using mice stem cells that could recapitulate embryos up to 8.5 days post-fertilization. A paper from this year showed in vitro embryo models from human stem cells that recapitulated embryonic states up to 14 days.
7. Gene therapies for deafness and blindness
There’s a gene that produces a protein called otoferlin, which helps transmit signals from hair cells in the inner ear to the brain. There’s a rare condition that affects this gene and causes deafness at birth. This year, a team from Fudan University in Shangai, reported a gene therapy in a group of children that restored some hearing functionality (60–65% of normal hearing). The therapy works by adding a working copy of the otoferlin gene that is packed in an adeno-assocated virus (AAV) and injected in the cochlea. Once inside the cochlea, the gene is integrated and guides the creation of otoferlin.
Similarly, the company GenSight reported safety and efficacy data for an analogous therapy for retinis pigmentosa, a condition that leads to photo-receptor loss in the eye and loss of vision. The therapy comprises an optogenetic molecule (a photo-sensitive protein) that is packed in an adeno-assocated virus and injected in the eye. Once the virus is injected, cells in the eye express the optogenetic molecule, making them sensitive to light. A pair of googles converts images in the real world to patterns of light stimulation to the retina. The clinical trial data indicated that this therapy is safe after 1 year, with patients being able to locate and count objects compared to barely perceiving light prior to treatment. There’s a similar promising method in pre-clinical stage that involves the implantation of a tiny display.
8. Whole-brain mouse atlas
Mammalian brains are some of the most complex systems; they have millions to billions of cells that form complex networks that produce emotions, movement, and everything mammals experience. In a true tour-de-force, a team of researchers reported a cellular atlas of the entire mouse brain. They used methods of single-cell sequencing to identify cell types and spatially resolved transcriptomics to know the diversity of cells in the brain while retaining spatial information. The team imaged 1,100 genes in 10 million cells, identifying more than 300 different cell types.
9. Discovery of the role of autoimmune diseases
A paper by scientists from Washington University described the role of the immune system in a mouse model of Alzheimer's. The study provides new evidence of the role of the innate immune system as well as the adaptive immune system (T and B cells) in tau-mediated neurodegenerative diseases. They show that T cells are recruited by microglia to areas of tau aggregation. The removal of T-cells prevented brain atrophy. What is more interesting is that T cells were responding to a pathogen presented by the innate immune system! This suggests that diseases like Alzheimer’s might be autoimmune.
10. Gene editing gets FDA approval
This year, a CRISPR therapy (Casgevy by Vertex Pharmaceutical and CRISPR Therapeutics) was approved by the FDA to treat sickle cell disease and beta thalassemia. Both conditions involve mutations in the gene that codes for hemoglobin, and in the case of sickle cell disease, blood cells tend to clog up. In this therapy, doctors extract blood stem cells from the patient, and using CRISPR (Cas-9), a gene in the patient’s blood stem cells is edited to resume production of fetal hemoglobin.
