The experiment that gives the paper its shape is quietly brutal. Take human stem cell derived microglia, the brain's resident immune cells, and put them next to dying neurons. The microglia respond the way immune cells do, by ramping up production of a protein called GPNMB, glycoprotein nonmetastatic melanoma B. Enzymes then snip the protein's outer portion loose from the cell surface, and it drifts away into the surrounding fluid. Now take that fluid, the conditioned medium, nothing else, and pour it onto healthy neurons that have been seeded with fibrils of alpha-synuclein, the protein that clumps into Lewy bodies. The neurons take up more fibrils and go on to form more pathology. Medium from microglia engineered to lack GPNMB does not do this. Damage, in other words, leaves a signal in the water, and the signal makes the next cell more willing to let the poison in.
Marc Carceles-Cordon, Alice Chen-Plotkin and colleagues at the Perelman School of Medicine published the work in Neuron on 12 May 2026. The team had form here: in Science in August 2022 they linked a chromosome 7 Parkinson's risk locus, tagged by the variant rs199347, to GPNMB itself, with risk alleles tracking higher GPNMB expression. What is new is the geography. Until now GPNMB mattered as something anchored in a cell's own membrane. The new paper shows the free-floating extracellular domain works on cells that never made it. Adding recombinant GPNMB extracellular domain to the medium of GPNMB knockout neurons restored their ability to internalise fibrils (p=0.007), and increased uptake even in ordinary neurons (p=0.017). That distinction is not academic. A protein stuck inside a neuron is a hard thing for a drug to reach. A protein floating in the space between cells is exactly the kind of thing an antibody can grab.
So they made antibodies: 42 monoclonals raised against the GPNMB extracellular domain, screened first in HEK293 kidney cells engineered to display GPNMB. Fifteen of the 42 blocked fibril uptake, and four blocked it at both low and high doses. Two of the candidates, mAb-1 and mAb-26, were then taken into human iPSC neurons, where both blocked the development of phosphorylated synuclein pathology at high doses and mAb-1 worked at low doses too; non-targeting control antibodies did nothing. One correction worth making plainly, because it has been circulating: this is not a mouse treatment study. The only mice in the paper are five BALB/c mice immunised with the human GPNMB extracellular domain so that hybridomas could be made from their spleens. No animal was given an anti-GPNMB antibody and tested for protection. Everything functional here happened in a dish.
What lifts the work above cell biology is the brain bank. The team genotyped 1,675 postmortem neurodegenerative disease cases at rs199347 and scored how far Lewy pathology had travelled, from none, to amygdala only, to brainstem, to limbic, to diffuse neocortical. Genotypes associated with higher GPNMB expression tracked with more widespread alpha-synuclein pathology (p=0.010 in a regression adjusted for age at death and sex). Tau and amyloid staging showed no such pattern. That specificity is the most human thing in the paper, and it is also the most easily overread: this is an association in donated tissue, not proof that GPNMB drove the spread in any of those brains. For families living with Lewy body dementia or the dementia that arrives later in Parkinson's, the honest translation is that a genetic dial already known to raise Parkinson's risk also appears to sit near how far the pathology goes, and that the dial's product is, unusually, out in the open where a drug could theoretically reach it.
The field's real state of play is visible if you set this beside a paper from Stephen Strittmatter's group at Yale, published in Nature Communications on 25 December 2025. Azucena Perez-Canamas and colleagues combed the neuronal membrane proteome for anything that binds alpha-synuclein fibrils and landed on two other candidates, mGluR4 and NPDC1, which form a complex on the surface of nigral neurons. Mice missing either Grm4 or Npdc1 were protected from dopamine neuron loss after fibril injection into the striatum, and mice carrying one faulty copy of each were protected too. So the field now has several candidate doorways into the neuron, one with mouse survival data and no antibody, one with an antibody and no animal treatment data, and no human trial for any of them. Penn has filed a provisional patent on the anti-GPNMB antibodies, with Chen-Plotkin, Carceles-Cordon and three colleagues as inventors, a disclosure the paper makes openly and readers should weigh alongside the authors' own caution.
The authors argue their target may be tolerable to block, noting that people born without functional GPNMB have only a mild skin condition, autosomal recessive amyloidosis cutis dyschromica. That is a reasonable argument about biology, not a safety finding, and it is a long way from what a person with a diagnosis wants to know. As Chen-Plotkin herself put it to Penn's news office, there is still a lot of work to do before this could reach people.