Deep divefrom Episode 1Brain & mood23 August 202618 min

TMEM106B structural haplotypes and neurodegeneration risk

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Fine-mapping of the TMEM106B locus (medRxiv preprint)

Chapters

Also linked

Companion: Neurodegeneration risk variants promote lysosomal TMEM106B fibril accumulation (bioRxiv)

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper. One paper today. A gene that was supposed to be simple turns out to be four different things, and one of them is breaking lysosomes from the inside.

Host 2Salazar, Tesi, Holstege and colleagues at Amsterdam UMC used long-read genome sequencing to fine-map the TMEM106B locus, one of the strongest genetic risk factors for neurodegeneration. What they found overturns fifteen years of assumptions about how this gene works.

What is TMEM106B

Host 1The paper is called Fine-mapping of the TMEM106B locus reveals four haplotypes differentially associated with risk for neurodegeneration. Let's start at the beginning. What is TMEM106B?

Host 2TMEM106B is a transmembrane protein that sits in the membrane of lysosomes. Lysosomes are the recycling centres of the cell — acidic compartments filled with enzymes that break down waste proteins, damaged organelles, and cellular debris. TMEM106B helps regulate lysosomal size, motility, and acidity.

Host 1And why do neuroscientists care about a lysosomal protein?

Host 2Because in 2010, a genome-wide association study by Van Deerlin and colleagues identified variants near TMEM106B as the strongest genetic risk factor for frontotemporal lobar degeneration with TDP-43 pathology. That is the most common molecular subtype of frontotemporal dementia.

Host 1For context, frontotemporal dementia is the second most common early-onset dementia after Alzheimer's. It typically hits people in their fifties and sixties and affects personality, behaviour, and language.

The original risk variant

Host 2Correct. And over the next fifteen years, TMEM106B variants were linked not just to frontotemporal dementia but also to Alzheimer's disease, ALS, Parkinson's, Lewy body dementia, and a condition called LATE, which is a TDP-43 proteinopathy of older adults. One gene, implicated across nearly the entire spectrum of neurodegeneration.

Host 1What was the specific variant?

Host 2The lead signal from the 2010 GWAS was a single-nucleotide polymorphism called rs1990622. It sits between genes, but it is in near-perfect linkage disequilibrium with a coding variant inside TMEM106B — a threonine-to-serine substitution at amino acid position 185. The threonine version was called the risk allele. The serine version was considered protective.

Host 1Considered. Past tense.

Host 2Past tense. Because this paper shows the picture is far more complicated than a single amino acid change. But before we get to the haplotypes, we need the 2022 plot twist.

The 2022 surprise

Host 1What happened in 2022?

Host 2Three independent groups — Eisenberg at UCLA, Goedert and Scheres at the MRC Laboratory of Molecular Biology in Cambridge, and Bharat and Bhatt at Columbia — were using cryo-electron microscopy to search for TDP-43 fibrils in the brains of people who had died with frontotemporal dementia. All three groups found abundant amyloid fibrils. But the fibrils were not made of TDP-43.

Host 1They were made of TMEM106B.

Host 2Specifically, the C-terminal fragment of the protein. TMEM106B gets cleaved by an unknown protease inside the lysosome, releasing a fragment spanning residues 120 to 254 into the lysosomal lumen. That fragment was forming tightly wound beta-sheet amyloid fibrils, architecturally similar to the tau tangles seen in Alzheimer's disease.

Host 1Were these only in diseased brains?

Host 2No. And that was the unsettling part. TMEM106B fibrils were found in virtually every neurodegenerative condition examined — frontotemporal dementia, Alzheimer's, Parkinson's, ALS — and also in neurologically normal elderly controls. They appeared to be a universal feature of brain aging, present in nearly everyone over seventy.

Host 1So the question became: are these fibrils driving disease, or are they inert debris, like lipofuscin?

Host 2Exactly. That debate ran from 2022 through to the papers we are discussing today. Now, back to the haplotypes.

Long-read sequencing

Host 1What did the Amsterdam group do differently from all the GWAS that came before?

Host 2They used PacBio long-read whole-genome sequencing with de novo assembly on 513 individuals. Previous genetic studies relied on genotyping arrays and short-read sequencing, which are effectively blind to structural variants — insertions, deletions, rearrangements, and mobile genetic elements longer than a few hundred base pairs. The TMEM106B locus, it turns out, is full of them.

Host 1How full?

Host 2Over sixty structural variants clustered near the gene. And when they combined this structural data with conditional association analysis from nearly six thousand Alzheimer's disease cases and controls, they found that the TMEM106B locus contains not one independently associated signal, but two.

Host 1Two separate signals, meaning two blocks of variation that independently influence risk?

Host 2Yes. One block covers the coding region of the gene, including the threonine-to-serine variant at position 185. The other covers an upstream regulatory region containing distal enhancers. Each block has two forms, giving four possible combinations.

The four haplotypes

Host 1Walk me through the four haplotypes.

Host 2The coding block comes in two versions. The risk version, which they call HA, carries threonine at position 185 and contains an AluYb8 retrotransposon insertion — a 317 base pair mobile genetic element embedded in the three-prime untranslated region of the gene. The alternative version, Ha, carries serine at position 185 and lacks the retrotransposon.

Host 1And the regulatory block?

Host 2Also two versions. The common version, HB, and a rarer version, Hb, which carries a 19 kilobase rearrangement in a distal enhancer region along with additional structural variants including transposable elements and tandem repeats.

Host 1So the four combinations are HAB, HaB, Hab, and HAb.

Host 2Correct. In European populations, the frequencies are roughly 57 percent for HAB, 34 percent for HaB, 9 percent for Hab, and less than one percent for HAb, which is too rare to assess.

Host 1And here is where fifteen years of interpretation gets overturned?

Host 2Here is where. For fifteen years, the field assumed that carrying the serine variant at position 185 was what made you protected. But the 34 percent of Europeans who carry the HaB haplotype — serine in the coding region with the common regulatory region — turn out to be at essentially neutral risk. Not protected. Neutral.

Host 1So where does the protection come from?

Host 2From Hab. The nine percent haplotype that combines the serine coding variant with the rearranged upstream regulatory region. That is the truly protective haplotype. Previous genome-wide association studies averaged the effects of HaB and Hab together and called the blend protective, when it was Hab doing the heavy lifting the entire time.

Host 1That is a substantial correction.

Host 2It is. And they confirmed it with two independent lines of evidence. First, brain proteomics from 182 post-mortem samples showed that Hab carriers had lower TMEM106B C-terminal fragment levels in the temporal cortex, while HAB carriers had higher levels. Second, cognitively healthy centenarians were significantly enriched for the Hab haplotype compared to middle-aged controls.

The retrotransposon

Host 1Centenarians with intact cognition. That is a strong natural experiment.

Host 2Now, one of the most intriguing structural features is the AluYb8 retrotransposon on the risk haplotype. Alu elements are primate-specific mobile genetic elements that propagate by a copy-and-paste mechanism: the DNA is transcribed to RNA, reverse-transcribed, and inserted at a new genomic location. This particular Alu is 317 base pairs, sits in the three-prime UTR, and is present exclusively on the risk haplotype HA.

Host 1And they propose it is not just a passive marker.

Host 2Holstege's group proposed an age-dependent activation model. The Alu element contains fourteen CpG sites that are heavily methylated in younger brains, keeping it silenced. As the brain ages, global DNA methylation declines. Simultaneously, TDP-43 function deteriorates with age, and TDP-43 normally helps suppress Alu element transcription. The loss of both silencing mechanisms could create a feed-forward loop that increases TMEM106B expression specifically in aging neurons.

Host 1Which would explain why age is the single largest risk factor for neurodegeneration.

Host 2That is the proposal. It is mechanistically coherent as a model. It has not been experimentally demonstrated in a causal manner.

A word from our sponsor

Host 1Attention. Does your nose contain hair?

Host 2The Hastrup Review of Science is brought to you by nose-hair trimmers. Precision-engineered for the passages you forgot you had.

Host 1At a certain age, your nose starts producing more hair than your scalp. Evolution offers no explanation. Neither do we.

Host 2A good nose-hair trimmer is the difference between a distinguished academic and a man whose students are too distracted to listen.

Host 1Nose-hair trimmers. Because your nasal cavity is not a nature reserve. Our referral link is in the show notes.

The mechanism

Host 1Back to the science. So we have fibrils accumulating in lysosomes, a risk variant at position 185, and a retrotransposon that may wake up with age. How do these connect mechanistically?

Host 2This is where a companion preprint from Ward, Petrucelli and colleagues becomes essential. They used induced pluripotent stem cells to show that cells homozygous for threonine at position 185 — the risk variant — accumulated roughly twice as much C-terminal fragment as cells homozygous for serine.

Host 1Twice as much. What drives that difference?

Host 2Glycosylation. Position 185 sits directly adjacent to an N-linked glycosylation site at asparagine 183. Threonine at position 185 is more efficiently glycosylated than serine in that sequence context. The sugar group appears to protect the C-terminal fragment from lysosomal protease degradation, either by stabilizing a protease-resistant conformation or by physically blocking enzyme access.

Host 1So the chain is: threonine means more glycosylation, more glycosylation means slower degradation, slower degradation means more fragment accumulates, and more fragment means more raw material for fibril formation.

Host 2Correct. And progranulin fits directly into this pathway. Progranulin is processed in the lysosome into smaller peptides called granulins, which function as cofactors for lysosomal enzymes. Mutations in the progranulin gene, GRN, cause familial frontotemporal dementia through haploinsufficiency. Ward's group showed that complete progranulin loss produced a nearly tenfold increase in TMEM106B C-terminal fragment levels.

Host 1And that was reversible?

Host 2Yes. Adding recombinant progranulin back to progranulin-knockout cells reduced fragment levels in a dose-dependent manner. The two genes are functionally coupled at the lysosome: TMEM106B produces the substrate, progranulin provides the machinery to clear it.

Lysosomal rupture

Host 1What happens when enough fragment accumulates? What do the fibrils actually do inside lysosomes?

Host 2Ward's group used cryo-electron tomography to look inside individual lysosomes from post-mortem brains. In a control brain, they found a single fibril across 246 lysosomes examined. In sporadic frontotemporal dementia, fibrils appeared in roughly twenty percent of lysosomes, with a median length of 321 nanometers. In a patient with a progranulin mutation, the fibrils were longer — median 436 nanometers — and they were physically rupturing lysosomal membranes from the inside.

Host 1The fibrils are mechanically breaking the lysosomes open.

Host 2From the inside. And when a lysosome ruptures, two things happen. Acidic hydrolases leak into the cytoplasm, which is directly toxic. And the cell loses degradative capacity. Proteins that should be cleared — TDP-43, tau, alpha-synuclein — accumulate because the machinery that was supposed to destroy them is broken.

Host 1Which would explain something that has puzzled the field for years. Why does one gene affect risk for so many different neurodegenerative diseases?

Host 2Exactly. If TMEM106B fibrils compromise lysosomal integrity generically, then the downstream proteinopathy depends on which protein a given neuron is most vulnerable to accumulating. In frontotemporal dementia, it is TDP-43. In Alzheimer's, it contributes to tau pathology. In Parkinson's and Lewy body dementia, it is alpha-synuclein. The upstream mechanism is the same: lysosomal failure. The downstream consequences vary by cell type and disease context.

Causal evidence

Host 1Is there direct causal evidence? Not just human association data, but an animal model?

Host 2Yes. Ward's group engineered mice to express the TMEM106B C-terminal fragment targeted specifically to lysosomes. Those mice developed increased TDP-43 inclusions in cortical neurons. That is the first in-vivo demonstration that TMEM106B fibril accumulation can drive TDP-43 proteinopathy, rather than merely correlating with it.

Caveats

Host 1What are the major caveats?

Host 2Several. First, the haplotype study was conducted primarily in European-ancestry populations. People of African ancestry carry at least two additional hybrid haplotypes with recombined risk and protective features that do not map onto the four-haplotype framework. The genetic architecture may differ across populations, and the findings need replication in diverse cohorts.

Host 1Second?

Host 2The proteomics effect sizes are modest. Hab is associated with 1.1-fold lower C-terminal fragment abundance compared to HaB. HAB shows 1.16-fold higher levels. These are typical for common genetic risk factors — they shift probability across populations, not destiny for individuals.

Host 1Third?

Host 2The cryo-electron tomography from the companion paper examined only three brains, all male, all homozygous for the risk variant. The images are striking, but the sample is far too small for population-level conclusions about fibril burden.

Host 1Fourth?

Host 2The mouse model uses overexpression of the C-terminal fragment, not a knock-in of the human risk variant at physiological levels. Overexpression models can produce phenotypes that would not occur naturally. And the retrotransposon age-dependent activation model, while coherent, remains a hypothesis.

Host 1And both the companion cryo-ET paper and the haplotype fine-mapping study originated as preprints.

Host 2Correct. The haplotype paper has since been published in Alzheimer's and Dementia, but the Ward and Petrucelli companion paper with the fibril mechanism data is still on bioRxiv as of this recording.

What comes next

Host 1What would it take to make this fully convincing?

Host 2Three things. First, knock-in mouse models carrying the actual human threonine and serine variants at position 185, showing differential fibril accumulation and neurodegeneration across a natural lifespan. Second, functional characterisation of the 19 kilobase rearrangement in the protective Hb regulatory region — does it reduce TMEM106B transcription, alter chromatin architecture, or act through some other mechanism? Third, replication of the four-haplotype framework in non-European populations.

Host 1Is there a therapeutic angle?

Host 2If the mechanism holds, several approaches become conceivable. Inhibiting the protease that cleaves TMEM106B to reduce C-terminal fragment production. Enhancing lysosomal clearance of the fragment, perhaps by supplementing progranulin or its granulin derivatives. Or modulating glycosylation at asparagine 183 to make the risk variant behave more like the protective one. All speculative at this stage, but the mechanism is specific enough that each step in the chain is a potential drug target.

Host 1Bottom line for someone following neurodegeneration research: how significant is this?

Host 2It is a genuine advance on two fronts. The haplotype work corrects a fifteen-year misattribution of genetic risk and reveals that most protective effect comes from the nine percent Hab haplotype, not from the serine coding variant alone. The companion work provides the first direct evidence that TMEM106B fibrils are not inert debris but active agents of lysosomal destruction. Together, they offer a coherent mechanism linking one gene to multiple diseases through a single upstream pathway: lysosomal fibril accumulation and membrane rupture.

Sign-off

Host 1That is the deep dive. Fine-mapping of the TMEM106B locus, from Salazar, Holstege and colleagues at Amsterdam UMC, with companion data from Ward, Petrucelli and colleagues. Links in the show notes.

Host 2This has been the Hastrup Review of Science. See you next week.