LSD promotes neuroplasticity. In mouse prefrontal cortex, it increases dendritic spine density by roughly 40% and synapse density by about 20%. These are the effects that make psychedelics therapeutically interesting for depression, PTSD, and addiction — the ability to rewire damaged neural connections. But LSD also causes hallucinations, which makes it unsuitable for patients with psychotic disorders like schizophrenia, precisely the population that might benefit most from neuroplastic rewiring.
Olson and colleagues at UC Davis (PNAS, 2025) swapped the positions of two atoms in LSD's indole ring — one carbon and one nitrogen, exchanged. The resulting molecule, JRT, produced a 46% increase in dendritic spine density and an 18% increase in synapse density. It improved cognitive and negative symptoms in mouse models of schizophrenia. It did not cause hallucinations.
Two atoms. Same elements. Different positions.
The mechanism is specific. LSD makes a hydrogen bond to a serine residue in the active site of the 5-HT2A serotonin receptor. This bond stabilizes the receptor in a fully active conformation — full agonism — which triggers the downstream signaling cascade that produces hallucinations. The carbon-to-nitrogen swap repositions the hydrogen-bond donor. JRT still binds the receptor. It still activates neuroplasticity pathways. But it cannot make the one bond that produces full agonism. It is a partial agonist — enough activation for growth, not enough for hallucination.
The selectivity is geometric. The 5-HT2A binding pocket has a specific three-dimensional shape. Dozens of molecular contacts hold the drug in place and initiate signaling. The serine hydrogen bond is one of many, but it is the one that tips the receptor from partial to full activation. The C-N swap doesn't weaken the drug's overall binding. It doesn't reduce affinity or change the drug's ability to find the receptor. It disables one contact among many — the contact that happens to control the distinction between therapeutic and hallucinogenic.
This is skeletal editing: modifying the molecular backbone itself rather than decorating it with different functional groups. The conventional approach to drug design adds or removes chemical groups from a fixed scaffold. The scaffold is treated as identity — what the molecule is. The functional groups are treated as properties — what the molecule does. Skeletal editing treats identity itself as a variable. The scaffold changes. The elements are the same. The properties diverge.
The deeper principle: in a system with many interactions, a single positional change can silence one output while leaving the others intact, if the silenced output depends on a geometric relationship that the others don't share. The hallucination pathway requires a bond at a specific angle to a specific residue. The neuroplasticity pathway does not. The swap is selective because the two pathways have different geometric dependencies on the same binding event.
A drug that heals without distorting — same atoms, rearranged — is a proof that the therapeutic and the hallucinogenic were never the same thing. They shared a molecule. They did not share a mechanism.
Note (2026-11-16; checked against the open-access PNAS paper, 'Molecular design of a therapeutic LSD analogue with reduced hallucinogenic potential', full text as summarized by a page fetch, plus its abstract via Europe PMC): the 46 percent spine-density and 18 percent synapse-density increases after a single dose are in the paper, and the abstract says (+)-JRT has 'lower hallucinogenic potential' and produces therapeutic effects in behavioral assays relevant to depression and cognition without exacerbating behavioral and gene-expression signatures relevant to psychosis. The essay's hydrogen-bond account is presented there as a design rationale: JRT lacks an indole N-H that LSD may use to hydrogen bond with S242 or G238, JRT is a potent partial agonist at 5-HT2A (Emax 81 percent), its dissociation from the receptor is about ten-fold faster than LSD's, and the authors say the lack of hallucinogenic effect 'is likely due to' partial agonism, to be confirmed with 5-HT2A knockout mice, adding that it is unclear whether hallucinogenic potential can be predicted in vitro. The essay's 'it cannot make the one bond that produces full agonism' and 'it doesn't reduce affinity or change the drug's ability to find the receptor' are therefore firmer than the paper, which also reports the faster off-rate. The mouse result is the head-twitch response, where JRT did not induce a robust response at any dose, a proxy for hallucination, so 'did not cause hallucinations' is a statement about mice; the cognitive result is a rescue of deficits in a reversal-learning task after unpredictable mild stress, and in the paper text I could see, the schizophrenia link rests on that, on antidepressant-like effects and on the absence of exacerbated psychosis markers, and I did not find an improvement of negative symptoms in a schizophrenia model stated, which is how the essay (and the university release) put it.