Every featured story this week sits in the same spot: a treatment, or an urgent need for one, that has outrun the biology meant to explain it. Where last week was about precision arriving — a vaccine built per patient, a narcolepsy drug that treats a cause instead of a symptom — this week is about how often the causal story arrives last. The feature that says it plainly is the one on SSRIs: the most-prescribed drug class on Earth, taken by roughly one in six US adults, and “we fundamentally don’t know how they work.” The serotonin-imbalance story never fit the data; what’s left is a fragmented picture of neuroplasticity, stress hormones, gene expression and plain expectation. A new single-cell map of the depressed hippocampus lands squarely in that gap — it locates a specific stalled step in the brain’s neuron-making assembly line, which is exactly the “restore the brain’s capacity to remodel itself” mechanism the antidepressant field keeps invoking. A companion-in-spirit study does the same for cognitive aging, pinning it partly on the brain’s myelin-maintenance cells and a pathway called NRF2 rather than on neurons alone.
The other two features are mechanism-first in a different way. The first injection of a person’s own mitochondria into a human eye was explicitly a safety exercise — the authors say outright they are “not trying” to prove it restores vision, and it didn’t. And a glioblastoma study names one precise trick the tumor uses to disarm the immune system: a neuropeptide, galanin, better known from mood and pain biology, redeployed to stock the tumor with immune-suppressing cells. Below the five features, Also This Week covers everything else the scan turned up — two firsts against Bundibugyo ebolavirus in the DRC, rapid genome sequencing rolled out city-wide, prime editing scaling to whole-gene size, computational pathology’s benchmarking fights, and a busy week for the wellness-industrial complex.
In February, a brain bleed injured a woman’s optic nerve and retina, leaving her almost completely blind — her pupils no longer even contracted when a light was shone into her eyes. This month, a team led by David Putrino, a neuroscientist at the Icahn School of Medicine at Mount Sinai in New York City, reported what they did next: they took mitochondria — the structures that generate energy inside cells — from the woman’s own leg muscle and injected them into the vitreous fluid of both her eyes.
The idea is that damaged retinal cells might take up the healthy mitochondria and get enough of an energy boost to survive and work again. The specific target was the retinal ganglion cells, the neurons that carry visual signals to the brain; rodent studies have shown that mitochondria delivered into the vitreous can keep injured versions of these neurons alive. Mitochondrial transplants have been tried before in the human heart and brain, but never the eye, and this attempt happened under an emergency approval the FDA granted in May for this one patient.
On safety, the result was reassuring: no inflammation, no damage to either eye. On vision, it was not. Her pupils did regain some response to light, but the effect faded after about four weeks (staying slightly above where it started), and a low-vision assessment noted only that she could perceive shapes and shadows in her left eye. The findings are in a preprint posted on 10 August that has not been peer reviewed. The authors are unusually blunt about the limits: “We can’t prove efficacy at all, nor are we trying to,” Putrino says. An outside researcher, Temurkhan Ayupov of the Institute of Molecular and Clinical Ophthalmology Basel, agreed the procedure looked “relatively safe” but said it does nothing to confirm that mitochondrial transplants actually work.
Mitochondrial transplantation is inching organ by organ toward the clinic, and this is its first try in the eye. But it is a single compassionate-use case with a transient, hard-to-interpret signal — a hypothesis, not evidence of benefit. The authors’ own framing (“we’re not trying to prove efficacy”) is worth noting in a field that has a habit of overclaiming from exactly this kind of n = 1 result.
Selective serotonin reuptake inhibitors are among the most prescribed drugs on Earth — more than 100 million US prescriptions for sertraline, escitalopram and fluoxetine in 2024 alone, with roughly one in six American adults taking some antidepressant. And yet, as psychiatrist Maurizio Fava of Massachusetts General Hospital puts it: “We fundamentally don’t know how they work.” This Nature feature walks through why a drug class this entrenched still has no agreed mechanism.
The familiar story — depression is a serotonin shortage, and an SSRI tops it back up — never fit the evidence. Deliberately depleting serotonin’s raw material doesn’t make healthy people depressed, though it does trigger relapse in people who had responded to antidepressants. So serotonin is central to how the drugs act, but probably not the root cause of the illness. What SSRIs do at the synapse is clear and fast: they block the transporter that reabsorbs serotonin, leaving more of it around within hours. What takes weeks is the actual symptom relief, and what happens in that gap is where the picture fragments — into neuroplasticity, raised BDNF (a growth factor that’s low in the blood of people with depression), shifts in gene expression, and a partial reversal of chronic stress’s toll on regions like the hippocampus and amygdala.
Two threads stand out. Catherine Harmer at the University of Oxford has shown that SSRIs quickly and subtly soften the brain’s bias toward negative information — people start registering a stranger’s smile or a bit of praise before they consciously feel any better, which is often why a friend notices the improvement first. And Gerard Sanacora at Yale points to expectation: a 2022 meta-analysis of more than 200 randomized trials found a median improvement of about 10 points on a standard depression scale for people on an SSRI versus about 7 for placebo. The drug effect is real and shows up in every meta-analysis — but it is modest on average, and averages hide the fact that some people improve a lot and others barely at all. “They’re no panacea,” says Imperial College London’s Sameer Jauhar.
The most-prescribed psychiatric drug class rests on a mechanism nobody can fully specify, for a diagnosis with no objective test. The useful question isn’t “do SSRIs work” — they beat placebo consistently — but why the average effect is small, who the real responders are, and why a meaningful minority find the drugs genuinely hard to stop. That last point is what turned this into a political fight this year, and it is still under-researched.
For over two decades, some researchers have suspected that depression involves a breakdown in neurogenesis — the brain’s ongoing production of new neurons in the hippocampus. A new study in Nature Medicine, “Dysregulated adult hippocampal neurogenesis in major depressive disorder”, with Columbia University neuroscientist Maura Boldrini Dupont among its authors, is the first to pin down where in that process things go wrong, at the level of individual cells.
The team analysed post-mortem hippocampal tissue from 30 people — 11 who had major depressive disorder and 19 without any mental-health condition — and sequenced the RNA in the nuclei of nearly 500,000 cells. Because the genes a cell switches on tell you how mature it is, they could reconstruct the assembly line of neurogenesis: neural stem cells become intermediate progenitors, then neuroblasts, then immature neurons. In the brains of people with depression, that line was backed up — more stem cells sitting at the start, fewer neuroblasts further along. “It looked like the trajectory was stalled, lagging behind,” Boldrini says. As a bonus, the work is also the most detailed cell-type map of the adult human hippocampus so far, and it adds weight to the still-contested idea that adult human brains keep making new neurons at all.
The caveats are real. This is a correlation in a small, post-mortem sample; it doesn’t show that stalled neurogenesis helps cause depression, and depression clearly involves more than the hippocampus. Notably, the study did not find fewer fully immature neurons in the depressed brains — which is what you’d expect if the process were completely halted — so exactly when and how it falters is still open. But Boldrini’s group says it surfaced “several potentially druggable pathways.”
This connects directly to the SSRI question above. The leading mechanistic story for antidepressants is that they restore the brain’s capacity to remodel itself — and this is the sharpest human evidence yet that that capacity is specifically blunted in depression, at an identifiable step, with candidate molecular targets attached.
Glioblastoma, the most aggressive brain cancer, has been almost untouched by the immunotherapy revolution — the drugs that transformed melanoma and lung cancer barely dent it. This paper offers one reason why. The tumor cells secrete a small signalling molecule called galanin, better known from the biology of mood, pain and seizures, and use it as an immune-evasion tool.
Galanin binds a receptor, GALR3, on monocytic myeloid-derived suppressor cells — immune cells whose job, once corrupted by a tumor, is to shut down other immune cells. Through a downstream cascade, this signal both draws more of these suppressor cells into the tumor and makes them resistant to ferroptosis, an iron-dependent form of cell death, so they stick around. When the researchers blocked GALR3 and forced ferroptosis, tumors in mice grew more slowly and antitumor immunity switched back on. Adding an anti-PD1 checkpoint drug on top produced durable, complete tumor regression in roughly 60% of the mice. A second glioblastoma study this week goes after the same problem from a different angle — the tumor’s electrical synapses onto neurons.
This names a specific, potentially druggable secreted signal that glioblastoma uses to stock its surroundings with immune-suppressing cells — and a two-part combination that clears most tumors in mice. It is early, mouse-stage and mechanism-dense, with human relevance still unproven, but glioblastoma has so few leads that a concrete target and a striking combination result are worth tracking.
Why do some people’s thinking hold up well into old age while others decline steeply? Most research on that question looks at neurons and at amyloid plaques. This study looks instead at white matter — the brain’s wiring insulation — and at the cells that build it, the oligodendrocytes.
The team drew on a rare resource: a brain bank with more than 70 years of linked cognitive-test data for its donors, so each brain could be matched to how fast that person’s cognition had actually fallen. Worse cognitive trajectories were associated with a consistent white-matter signature — smaller myelinated nerve fibres, thicker myelin, more oligodendrocytes, and, inside those cells, a turned-down NRF2 pathway (a master switch for cellular stress defence). To test whether that was cause or bystander, they knocked out NRF2 specifically in the oligodendrocytes of aged mice; those mice showed blunted cognitive gains over time and reproduced the same white-matter pathology seen in the human brains.
It reframes cognitive aging as partly a wiring-maintenance problem, not just a neuron problem, and points at a specific pathway — NRF2 — that already attracts drug-development interest. It pairs naturally with the depression-neurogenesis study above: both use single-cell data from human brain tissue to isolate a non-obvious, potentially druggable cellular step in a brain condition. The human data are correlational; the mouse knockout supports causation for the pathology if not the whole cognitive effect.
Following last week’s featured phase III readout for Moderna and Merck’s personalized melanoma vaccine, a Nature news analysis argues the harder problem now is manufacturing: a positive trial signal is one thing, but building a genuinely new mRNA drug per patient, on a post-surgery clock, at the scale of routine oncology is a logistics problem the field has not solved.
Alongside this week’s featured galanin study, a second glioblastoma paper attacks the tumor’s connections to the brain itself: an implantable hydrogel that locally delivers siRNA to silence the TrkB receptor at the tumor–brain interface disrupted the malignant electrical synapses glioblastoma forms onto neurons, more than doubling median survival in mice and easing tumor-associated anxiety and memory deficits without hurting motor function.
Two papers in the same issue of Nature Genetics converge on the same picture of pancreatic ductal adenocarcinoma. One finds that essentially every tumor, whatever its mutations, is built from the same six molecularly distinct cell states, and that a tumor’s ability to slide between them is how it shrugs off single-drug treatment. The other identifies the developmental and MAPK-responsive transcription factors that decide which state a cell adopts — along with the genetic dependencies specific to each, which is where combination therapy would have to aim.
An outbreak of Bundibugyo ebolavirus in the Democratic Republic of the Congo produced two firsts for the virus. A healthcare worker was treated with the investigational monoclonal antibody cocktail MBP134 plus remdesivir under emergency authorization and recovered, clearing detectable virus by day 22. And five exposed family members — one vaccinated adult and four unvaccinated children — received MBP134 as post-exposure prophylaxis and all stayed disease-free through 21 days. The context is grim: overcrowding at treatment centers is itself fuelling transmission, and armed conflict and humanitarian funding cuts in the east are simultaneously undermining malaria, TB and HIV control.
“Little Falcon,” a citywide rapid whole-genome sequencing program across Dubai’s neonatal and pediatric intensive care units, ran trio sequencing on 100 critically ill children with a median turnaround of 3.4 days. Against a matched historical cohort on standard testing, it cut diagnostic time from 38 days, raised the diagnostic yield from 30% to 53% (80% in consanguineous families), and changed clinical management for 53% of patients versus 18%. An accompanying piece discusses what it takes to scale this into routine intensive-care practice.
On prediction: an APOE-independent, multiancestry polygenic risk score for Alzheimer’s predicts cognitive decline and neuropathology consistently across diverse populations, fixing a major weakness of earlier ancestry-skewed scores. On mechanism and clearance: a mouse study ties aberrant ERBB4 activity in excitatory neurons to microglia over-consuming synapses, a specific molecular switch upstream of the synapse loss that tracks with cognitive decline; and engineered autophagy receptors delivered in extracellular vesicles cleared pathological Tau and TDP-43 aggregates in models of frontotemporal dementia and ALS.
Two papers push prime editing past small edits toward inserting or replacing entire genes without double-strand breaks. Donor-complementary prime editing (DoPE) handles precise kilobase-scale, library-compatible insertions; “prime assembly” is a programmable gene-replacement tool that swaps in large DNA fragments in human cells — both aimed at the long-standing gap between editing a base and rewriting a locus.
Model-guided protein evolution with sparse experimental data produced smaller, more efficient RNA-guided nucleases — compact editors are easier to deliver. Compact deep-learning models trained on chromatin-accessibility data designed synthetic tissue-specific enhancers, 15 of which fired in the right tissue (heart, limb, CNS) in mouse embryos. And a review of automated prototyping of non-standard genetic codes surveys how new codon–amino-acid pairings could unlock new therapeutic chemistries and biocontainment — and how hard recoding a genome still is.
A cluster of mechanism papers: the endocannabinoid system acts as a “gain control” dial on reward signalling in mice, tuning how strongly inputs shape engagement with rewards; cerebellar granule cells help the cortex reorient its activity to tell similar contexts apart rather than over-generalizing; and zebrafish thalamocortical-like circuits use the same hierarchical sensory processing as mammals, in a far simpler brain. On the genetics side, systematic mapping shows autism-linked mutations converge by rewiring shared protein-interaction networks, and a pleiotropy analysis dissects the shared biology between major depression and its physical-disease comorbidities.
Human brain organoids grown for five years show epigenetic aging patterns that mirror real brains, making them a plausible long-term model of postnatal human brain development. Further afield, comparative genomics across Myotis bat species links their long lifespans and viral tolerance to specific genome-evolution patterns — a natural experiment in longevity and antiviral defence.
Two closely related papers put multiple-instance learning — the standard way to turn patch-level features into a whole-slide prediction — through large-scale testing: nnMIL, evaluated across 40,000 whole-slide images and 35 clinical tasks, and a companion generalizable-MIL framework focused on efficient large-batch training. Separately, HisToSpatialCNV infers spatial copy-number changes directly from routine H&E images, reconstructing tumor subclones without separate sequencing. The counterpoint: a comment argues that despite the data-and-compute boom, drug-development success rates have stayed flat, and calls for tighter feedback loops between clinical and lab data.
Eric Topol interviews investigative journalist and physician Deborah Cohen on how the supplement, wearable-testing, tech and pharma industries commercially amplify health anxiety in healthy people. The governance question is live: a proposed regulatory sandbox, ANTLiON, for consumer-sensor data entering clinical care; a feature on how sleep-enhancing tech risks turning rest into a surveilled, auditable obligation; and wearable exoskeletons moving out of rehab clinics and into consumer fitness.
A counterpoint to the featured eye study’s ambitions: Mobilio, a smartphone app combining camera sensing, machine learning and personalized audio guidance, helped people with blindness or severe visual impairment navigate outdoor routes 13% faster and with 41% fewer obstacle contacts than a white cane plus Google Maps — roughly matching the reliability of a human guide, without any biology to fix.
A prespecified secondary analysis of CARDIO-TTRansform (1,432 patients) revisits the TTR-silencer eplontersen — which missed the trial’s primary endpoint overall — splitting results by whether patients were also on a TTR stabilizer, to ask whether silencing only helps on top of stabilization. The open-label EPIDAURUS trial tested dual antithrombotic therapy with a potent P2Y12 inhibitor (rather than clopidogrel) in atrial fibrillation plus acute coronary syndrome, aiming to keep the bleeding advantage of dropping aspirin while closing the early ischemic-event gap. And on mechanism, SGLT2 inhibitors turn out to directly activate pantothenate kinase 1 in the human heart — a candidate explanation for why the diabetes drugs cut heart-failure mortality independent of glucose lowering. Adjacent: a large observational study links the recombinant shingles vaccine to lower later cardiovascular risk.
A surgically implanted, electrically pulsating muscle graft gave mice more muscle mass, strength and bone density than a static graft — essentially simulating the systemic benefits of exercise for animals that can’t exercise. A review marks two decades of induced pluripotent stem cells, from discovery to an actual clinical treatment platform. And “orbital alchemy” surveys efforts to manufacture drugs and engineered tissues in microgravity as launch costs fall.
A signalling axis that converts inflammatory necroptotic cell death into quiet apoptotic death in the gut turns out to underlie the chronic inflammation of inflammatory bowel disease. Androgen signalling through sympathetic neurons dampens allergic lung inflammation in mice — a mechanistic hint at why some boys’ asthma eases after puberty. A phase 1 trial of CD19 CAR T cell therapy for treatment-refractory rheumatoid arthritis reports early data from the COMPARE trial, extending the B-cell-reset idea from lupus into RA. And a review of systems vaccinology argues vaccine design should target the immune system’s full multilayered architecture, not just antibody titres.
A large mother–infant study finds the maternal gut is a major source of the bacterial strains that colonize a baby’s gut, with downstream effects on outcomes like eczema. Transferable genetic toolsets for hard-to-engineer gut Clostridia let researchers reversibly switch two microbiota-derived metabolites on and off in mice. And DartUniFrac, a GPU-accelerated version of a standard microbiome diversity metric, is built for datasets that have outgrown the old tools.
A nonviral lipid-based delivery system for chemically modified suppressor tRNAs restored functional protein production past disease-causing nonsense mutations, demonstrated for cystic fibrosis. A paired perspective, “rethink the cargo, rethink the carrier,” argues these tRNA drugs need lipid nanoparticles designed for tRNA rather than mRNA hand-me-downs. Related delivery chemistry: a binding-to-release strategy for antibody- and small-molecule-drug conjugates improves payload uptake and release, widening which surface targets are usable.
Deep-learning 3D structure prediction was used to design RNA pseudoknots from scratch — a notoriously hard-to-engineer fold — expanding the range of functional RNA molecules that can be rationally designed. A companion perspective, “the knotty problem of RNA structure prediction,” puts the result in context.
A fast, reference-free method discovers RNA sequences directly from single-cell and spatial data without aligning to a pre-built genome. A large cell-type-specific eQTL map shows many complex-trait genetic associations only make sense once you know which cell type the regulatory effect happens in. Long-read sequencing of human sperm finds non-crossover gene conversion is more frequent and earlier than thought. An improved ChIP-seq spike-in normalization (ChIP-wrangler) argues H3K27ac is largely independent of ongoing transcription — against recent claims. A method predicts missense-variant effects from experimental functional data rather than clinical correlations. And primate-specific regulation of the lipid enzyme UGCG shows how evolution fine-tuned a pathway implicated in neurological and metabolic disease.
From the Bioinformatics advance-access feed: SNPannotator automates post-GWAS functional annotation via Ensembl, GTEx, the eQTL Catalog and STRING; the Plotgardener App puts a graphical interface on a publication-figure R package; nf-core/pacsomatic is a scalable somatic-variant pipeline for PacBio HiFi long reads; SpaAlign integrates spatial multi-omics by contrastive learning; and st2traj infers cell trajectories from multi-timepoint spatial transcriptomics.
TargetPrior embeds miRNA-signature information into an evolutionary-learning framework to prioritize drug targets, demonstrated in acute myeloid leukemia. A study of finetuning strategies for foundation models on temporal clinical transcriptomics asks how to make general-purpose models capture disease-progression dynamics. And AdmixLD does fast genome-scale inference of ancestry disequilibrium in hybrid zones.
A science-integrity investigation found more than 18,000 potentially manipulated or duplicated images in commercial antibody-validation catalogues from 15 companies — a reliability problem for a reagent used everywhere in biomedical research. The Gates Foundation pledged $540 million to the Global Burden of Disease study, which has itself faced transparency criticism. The new chief of a US child-health research institute drew scrutiny over “Make America Healthy Again” political ties. A feature on China’s expanding role in global health asks whether cheaper diagnostics and medicines will come with local capacity-building or dependence. And two conflict-zone pleas: public-health support for El Obeid, Sudan, and a deadly tick-borne virus gaining ground in Europe.
Children’s lung function has grown faster since London introduced its ultra-low-emission zone — real-world evidence that urban air-quality policy measurably helps child respiratory health. A commentary frames GLP-1-era obesity drugs as a “disruptive innovation” that health systems’ delivery models and reimbursement are structurally unready for. And a review of sickle cell disease notes that gene therapies have finally arrived, but cost, access and durability keep it short of a declared victory.