When we talk about the tiny bags that cells throw out — extracellular vesicles (EVs) — the star of the show is usually the “exosome,” roughly 50–150 nm across. On this site too, we have covered how those particles work many times over.
The star of today’s paper, however, is a bag more than 600 nm in diameter. The big ones reach 1–2 µm: a bag the size of a single bacterium, dropping out of a cell. The authors call it an extra-large extracellular vesicle (XLEV).
And that bag was loaded with sonic hedgehog (SHH), the instruction molecule that decides the shape of the body and turns up in every developmental-biology textbook. When EVs collected from culture supernatant rich in these bags were applied to vascular cells, the rudiments of blood vessels clearly increased — that is the substance of the paper (though, as I will set out in detail later, this was not an experiment in which XLEVs alone were isolated and applied).
The most interesting part was which way the switch that lets the bags out is turned. In EV secretion, the field has long emphasized that “a Rab protein works in its ON state, the form holding GTP.” In this paper, however, the OFF state — the form holding GDP — was the one that made the large bags come out. In the authors’ own words, this is a logic that contrasts with conventional models emphasizing the GTP-bound form.
One caveat up front. This is a study carried out in cultured cells. It has not reached the stage of confirming an effect by administering anything to animals, and it is not a story about a treatment that can be used in patients. With that said, as groundwork for the question of how a cell builds a bag and where it lets it out, I think this is a piece of work worth reading.
Journal information
- Paper title: Unconventional Secretion of Angiogenic Sonic Hedgehog–Containing Extra-Large Extracellular Vesicles is Driven by PI3K–Rab18-GDP Signalling
- Type of paper: Original research article (cell biology; mechanisms of extracellular vesicle biogenesis)
- Journal: Journal of Extracellular Biology, Volume 5, Issue 2, article number e70112 (received 10 July 2025; revised 3 January 2026; accepted 7 January; published online 28 January)
- DOI / link: 10.1002/jex2.70112
- License: Open access (CC BY). Anyone can read the full text
- About the journal: Founded in 2021 as the second journal of the International Society for Extracellular Vesicles (ISEV) and launched in January 2022, this is a relatively new specialist title (the first was the Journal of Extracellular Vesicles, launched in 2012). It is published by Wiley Periodicals, LLC. It is fully open access, and the article processing charge (APC) is published as up to US$3,340
- Impact Factor: The publisher’s page lists 4.8 (the year it refers to, however, is not stated. This appears to be the journal’s first IF, based on the Journal Citation Reports released in June 2026 — that is, on 2025 citation data — but I could not find an official page that specifies the year). The journal is indexed in PubMed/PMC, but MEDLINE indexing is not listed in the catalog of the US National Library of Medicine
- Authors: Four in all — Shuo Wang (first author), Rio Imai, Yuya Kaneko, and Yosuke Tanaka (corresponding author)
- Affiliation: Division of Cellular Architecture (Molecular Cell Dynamics), Department of Cell Biology and Anatomy, Graduate School of Medicine, the University of Tokyo
- Funding: The University of Tokyo GAP Fund Program (8th, 12th and 14th rounds); AMED (JP22ym0126805, JP25ym0126805j0004, via the Clinical Research Promotion Center of the University of Tokyo Hospital); JSPS KAKENHI (20K06634, 25K09626); AMED BINDS (JP25ama121002). The GAP Fund is an internal University of Tokyo grant scheme supporting “proof-of-concept validation with commercialization in view”
- Conflict of interest disclosure: The corresponding author, Tanaka, and the first author, Wang, have a patent application pending that relates to this work (Japanese application 2022-078750; US application 18/864610; European application 23803594.3). The other authors declare none. As I will describe below, the content of this patent bears on how the paper should be read, so I will come back to it later
- Data availability: The stated arrangement is provision “upon reasonable request” to the corresponding author
What kind of researcher is the corresponding author?
The corresponding author, Yosuke Tanaka, is a Lecturer (PI) at the Graduate School of Medicine, the University of Tokyo, and has led a laboratory in molecular cell dynamics and cellular architecture since September 2018. After graduating from the School of Medicine at the University of Tokyo he went on to its graduate school, earning a PhD in medicine in 1999. If his research axis had to be put in a single phrase, it would be the motor molecules that carry things around inside cells.
He worked for many years in the laboratory of Nobutaka Hirokawa, known worldwide for his work on kinesins — as a research associate from 1996 to 2017, and as a lecturer from 2017. KAKENHI records show him listed as a co-investigator on Hirokawa’s Specially Promoted Research and other grants from 1996 through 2015 (Hirokawa is not an author on the present paper). Line up the papers on which Tanaka is first author and the background to this study comes into view as a single continuous thread.
- 1998, Cell — In mice lacking the kinesin heavy chain KIF5B, mitochondria accumulated abnormally around the nucleus (Tanaka Y, et al. Cell 1998;93(7):1147-1158). “Lose the motor and the organelles crowd toward the center” is the prototype of exactly the same logic that sits at the core of the present paper.
- 2005, Nature — This is the single most important piece of prehistory for the present work. In the “node” of the mouse embryo, the authors discovered that FGF signaling triggers the secretion of membrane-enclosed objects 0.3–5 µm in diameter, which they named nodal vesicular parcels (NVPs). What those objects contained was sonic hedgehog and retinoic acid (Tanaka Y, Okada Y, Hirokawa N. Nature 2005;435:172-177). It was part of the machinery that decides left from right in the body.
- 2022, Developmental Cell — In the forelimbs of mice with weakened KIF3B function, the gradients of FGF/PI3K signaling and of Shh were disturbed, producing polydactyly, and the strength of PI3K signaling sorted where Shh ended up (Wang S, Tanaka Y, et al. Dev Cell 2022;57:2273-2289). Wang, the first author of the present paper, was first author here too.
- 2024, EMBO Reports — Kinesin-1 helps fold the calcium channel CaV1.2 in the endoplasmic reticulum and thereby maintains glucose homeostasis in mice (Tanaka Y, et al. EMBO Rep 2024;25:4777-4802). The chaperone Hsp90 is again one of the leading players here.
So the present paper amounts to work that pins down, at the molecular level and in cultured cells, how the “large bag that carries SHH” — found inside the embryo roughly 20 years ago — is built and where it comes out. Kinesin, Hsp90, SHH, “organelles that crowd toward the nucleus”: nearly all of the materials this researcher has worked with converge, almost unchanged, in this paper.
One more facet, apart from the research. Tanaka is also a poet. His collection “Pinku no Sunabukuro” (Pink Sandbag; Shichosha, 2021) won the 37th Shiika Bungakukan Prize, and the prizewinner profile at the Japan Museum of Modern Japanese Poetry, Tanka and Haiku lists him as “poet, cell biologist.”
What was it that we did not know until now?
Bags come in size classes
Extracellular vesicles are not one homogeneous population. MISEV2023, the guidance issued in 2024 by the International Society for Extracellular Vesicles (ISEV), recommends using the operational labels small EV (broadly under 200 nm) and large EV (broadly over 200 nm) when naming them by size, and advises against using terms that presuppose a route of biogenesis — “exosome” among them — unless that route has actually been verified (Welsh JA, et al. J Extracell Vesicles 2024;13:e12404).
Most research has concentrated on the smaller side. The reason is simple: they are easier to handle. They are easy to collect by centrifugation, and they fall inside the working range of the instruments most commonly used. Put the other way round, large bags exceeding 600 nm may have been discarded by the experimental procedures themselves. What this paper stepped into is precisely that discarded side.
Sonic hedgehog cannot travel far on its own
During development, SHH forms gradients of high and low concentration that tell cells “this is a finger, this is a palm.” But this protein is modified with lipids. Because cholesterol and palmitic acid are attached to it, it is poorly suited to drifting freely through water on its own.
👦 Student: Why is being greasy a problem?
🧬 Exotaro: Drop permanent-marker ink into water and it will not spread out — it beads up, right? SHH is the same: left as it is, it cannot reach far through water. So it seems the cell uses a trick — loading it onto a little membrane boat and sending it off. The NVP found in 2005 was exactly that boat.
In 2018, a different group also reported a specialized EV carrying SHH: a population bearing three proteins — Axl, Rab18 and Tmed10 — as markers, which those authors called ART-EVs (Coulter ME, et al. Cell Rep 2018;24:973-986). Rab18 is among those markers — and that is the entrance to today’s story.
Rab, the “ON/OFF tag”
Rabs are a family of small proteins that run the logistics inside a cell. When one is holding a molecule called GTP it is ON; once that has been broken down to GDP, it is OFF. ON and OFF swap out the partners (effectors) it binds to.
For EV secretion, the view that the ON side does the work has long been emphasized. The classic example is Rab27a, known for docking multivesicular bodies to the plasma membrane and thereby releasing exosomes (Ostrowski M, et al. Nat Cell Biol 2010;12:19-30).
To be fair, though, it is not as simple as “ON only.” This very paper notes in its discussion that small-EV secretion involves both the GDP form of Rab7 and the GTP forms of Arl13b and Rab27a. So the novelty of this study is not that it “overturned the established view of EV secretion wholesale,” but lies in one point: the OFF form of Rab18 was driving the release of the large bags.
👦 Student: So an OFF Rab isn’t doing anything?
🧬 Exotaro: That is the usual assumption — that OFF is a standby posture. Yet this paper reports that the OFF form was itself the trigger that let the large bags out. That was the single biggest surprise this time.
What did this paper find?
The clue was a drug that stops Rabs
The authors began by applying two drugs to human bone-marrow-derived mesenchymal stem cells (MSCs): CID1067700 (hereafter CID, 40 µM) and SF1670 (250 nM). The latter inhibits an enzyme called PTEN and, as a result, strengthens signaling downstream of PI3K.
Collecting and examining what came out into the culture medium gave a clear difference.
- TSG101, a marker of small EVs, was present even without drug treatment (the DMSO control)
- SHH, however, was detected only when SF1670 or CID was applied
When the supernatant of CID-treated hMSCs was then separated on a density gradient, the SHH-containing particles collected in fractions heavier than the CD81-positive small EVs (fractions 8–10), while CD81 appeared in the lighter fractions (2–7). The two are different things.
By nanoparticle tracking analysis (NTA), particles larger than 600 nm were detected in 3 of 3 runs in the SF1670 group and in 3 of 3 runs in the CID group, whereas in the DMSO control they were detected in none of the 3 (p = 0.01, chi-square test). Particles under 600 nm appeared under every condition. Electron microscopy — here on samples of immortalized hTERT-MSCs (UE7T-13) — also showed that CID treatment markedly increased the number of sparse-content bags exceeding 600 nm, and that most of them were enclosed by a lipid bilayer. The authors state that this is consistent with their being EVs rather than protein aggregates or lipoproteins.
Counting the glowing particles that come out
Next the authors had mouse NIH3T3 fibroblasts express SHH that glows green (SHHN-EGFP) and filmed the particles moving in the culture medium directly. Glowing particles roughly 1 µm in diameter drift through the liquid in Brownian motion. Applying CID significantly increased their number.
Here the authors tested whether two drugs could stop this.
- Inhibiting Hsp90 with TAS-116 (0.5 µM, 48 hours) made the CID-driven increase in release disappear
- Inhibiting neutral sphingomyelinase 2 (nSMase2) with GW4869 (1.25 µM) also made it disappear
Both are required, in other words. Incidentally, TAS-116 goes by the name pimitespib, and in Japan, since June 2022, it has been a real drug approved for the treatment of gastrointestinal stromal tumor (GIST) that has progressed after chemotherapy (trade name Jeselhy). Here it is used as a research tool.
And then came the clincher: the Rab18 mutants.
- In cells made to express Rab18S22N (the GDP-form mutant, which mimics OFF), intracellular SHH decreased and the SHH-XLEVs coming out increased
- With Rab18Q67L (the GTP-form mutant, which mimics ON), SHH stayed inside the cell and release fell markedly
The same direction of result came not only from the drugs but from the genetic side. Flow cytometry likewise showed that, under CID treatment, supernatant from cells expressing fluorescent SHH contained 2–3 times more of the large fluorescent-particle population than supernatant from cells that did not, confirming that SHH really is inside those large particles.
Where does it come out? From around the nucleus
From here on is, I think, the most beautiful part of this paper.
Filming CID-treated cells over time, within 30 minutes the SHH signal gathered around the nucleus (the perinuclear region), and within an hour structures could be seen moving from there vertically toward the plasma membrane.
Then, using lattice light-sheet microscopy — an instrument that can image in three dimensions at high speed without damaging the cell — they captured tubular and vesicular SHH-XLEVs of 1–2 µm being released only from the plasma membrane around the nucleus. The particles that emerged clung to the cell surface for a while and then abruptly let go.
👦 Student: Does it really matter where in the cell something comes out?
🧬 Exotaro: It matters a great deal. To use a factory analogy, it is the question of which door of the building the product ships from, and if the exit is fixed, that means there is a dedicated loading line there. What is more, in a developmental setting, which direction something is sent decides the shape of the body. Directional secretion is itself information.
Why does it gather around the nucleus? A tug-of-war between motors
So why does OFF-form Rab18 move toward the nucleus? Inside a cell, two kinds of motor run in opposite directions along rails called microtubules. Kinesin heads outward, toward the cell periphery; dynein heads inward, toward the nucleus. The results of in vitro binding experiments (GST pull-down) split cleanly.
- Rab18-GTP (ON) bound well to kinesin-1 (KIF5A)
- Rab18-GDP (OFF) bound well to Hsp90α and nSMase2
In other words, going OFF severs the tie to the outward-bound motor. Once one side of the tug-of-war lets go of the rope, the vesicle is pulled by inward-bound dynein and pools around the nucleus — this is the model the authors propose. They did not catch dynein in the act of pulling; this part is an inference.
Indeed, applying PI3K inhibitors (ZSTK474, LY294002) scattered Rab18-bearing vesicles throughout the cell, whereas strengthening PI3K’s downstream signaling with SF1670 made them assemble distinctly around the nucleus. The GDP form, S22N, sat significantly closer to the perinuclear region than the GTP form, Q67L. That assembly site was an endosome bearing the marker AP-1 (overlap was significantly stronger than with AP-3) — one of the cell’s sorting hubs.
Where Hsp90α sat was also set by Rab18’s state. In cells made to express the GDP form, Hsp90α gathered around the nucleus in 123 of 125 cells, whereas with the GTP form it did so in only 8 of 88. And CID treatment significantly increased intracellular ceramide (the lipid nSMase2 makes); the increase was cancelled by GW4869 and restored by adding SF1670. Ceramide is known for bending membranes and driving the formation of bags.
The ingredients are all there. PI3K signaling comes in → Rab18 pools around the nucleus in its OFF form → Hsp90α and nSMase2 are recruited there → the membrane is remodeled and a large bag forms → it goes out through the plasma membrane around the nucleus. That is the storyline the authors drew.
Prolonged stimulation builds clumps outside the cell
When CID was kept on for a long stretch, 24–30 hours, large Hsp90α-rich structures 20–40 µm in diameter appeared outside the cell, with SHH-XLEVs inside them. These were often connected to the cell surface by thin, cytoneme-like protrusions.
In cells made to express the GTP form of Rab18, by contrast, the picture differed: large SHH deposits roughly 50–100 µm across formed on the glass beneath and around the colonies. The authors interpret these as structures consistent with aggregates of SHH bound to the extracellular matrix. ON and OFF change the very manner in which SHH is put out, in other words.
Examining SHH-positive particles near the cell surface, they also found that the group exceeding 2 µm had significantly stronger surface SHH signal than the group of 2 µm or less (the figure legend gives n = 27 vs 90, with cells as the unit). This suggests the surface of the bag may be remodeled at the moment of release.
And blood vessels formed
Last comes the functional experiment. Human umbilical vein endothelial cells (HUVECs) were seeded on Matrigel and the number of blood-vessel rudiments (tubes) formed was counted.
When an EV fraction collected from MSC culture supernatant (roughly 10 µg/mL by protein) was added along with them —
- Compared with the DMSO control, roughly 4–5 times as many tubes formed in the SF1670 and CID groups (number of tubes per 0.72 mm². n = 5, one-way ANOVA)
Furthermore, when the HUVECs themselves were made to express SHH, significantly more tubes formed than in the control (per 1.62 mm². n = 3–4). The main text says “roughly fourfold.” But reading the means off the figure gives about 15 for the control and about 45 for the SHH group, which looks closer to threefold. Either way, it supports the idea that SHH itself has the power to make blood vessels form.
How will the future change? (The road to the clinic)
“How much can you get?” is the practical wall for MSC therapy
When you set out to turn MSC-derived EVs into a drug, the first thing you run into is yield. Culture cells in bulk and the amount of vesicle you collect is still modest. So a drug that can tell cells to “put out more” has great value. The CID and SF1670 of this paper are exactly such candidates.
Here I return to the patent I set aside earlier. The published application (international publication WO2023/219119; in Japan, JP-A 2023-167506) is titled “A method for producing extracellular vesicles containing a physiologically active substance.” The applicant is the University of Tokyo and the inventors are Tanaka and Wang. Claim 1 reads: “A method of producing extracellular vesicles, said method comprising obtaining extracellular vesicles secreted from a cell that has been subjected to a treatment that inactivates a Rab.” The claims that follow run through mesenchymal stem cells; both Rab inhibitors and Rab18-GDP-form mutants; the Shh-containing extracellular vesicles themselves, as “Shh-containing extracellular vesicles of 800 nm or larger and/or a volume of 10⁹ nm³ or greater”; and even a pharmaceutical composition containing them as an active ingredient.
So this paper is, at the same time as an academic discovery, also supporting evidence for a production method that has already been filed for patent. Since it is properly disclosed as a conflict of interest this is not misconduct, but it is background a reader is better off knowing. The author’s publicly posted research project records also list, as AMED translational research for 2022–2024, a project titled “Development of methods to promote the qualitative and quantitative secretion of human mesenchymal stem cell–derived extracellular vesicles contributing to early recovery after myocardial infarction” (I could not confirm how this corresponds to the grant numbers listed in the paper). The aim is consistent throughout.
There is also something one cannot be unreservedly pleased about. What CID increased was the larger bags, while at the same time small-EV secretion tended, if anything, to be suppressed. So it is not that “everything increases” but that things are swapped over. Depending on what you want to turn into a drug, this compound can be an ally or an enemy.
A drug that stops Hsp90 already exists in Japan
TAS-116 (pimitespib), used here to inhibit Hsp90, is approved in Japan as a treatment for GIST. In the context of this paper, the use runs in the opposite direction: what is happening to this kind of vesicle secretion inside patients whose Hsp90 is being suppressed as cancer therapy? As a question it is coherent, but this paper does not examine it. It remains strictly a task for the future.
What can a bag that carries SHH be used for?
SHH does not only make blood vessels form; it is a molecule broadly involved in development and in tissue repair. The very idea of “taking a molecule that is lipid-modified and cannot travel alone, and delivering it aboard a little membrane boat” strikes me as interesting from a drug-delivery point of view.
That said, I want to stress that the distance from here to the clinic is long. This paper contains no therapeutic experiment in animals. The only place a mouse appears is where brain extract is taken for the in vitro binding experiments. “Blood vessels formed” is a story inside a dish; vessels were not built inside a living body.
How to read this study critically (limitations, and ways to raise its quality)
First, what deserves credit
- The imaging is serious. They mobilized spinning-disk confocal, Airyscan, and even lattice light-sheet microscopy, and actually showed the phenomenon — “a large bag goes out from around the nucleus” — on film. They do not argue from indirect readouts alone.
- They push on two wheels, drugs and genes. With the drug (CID) alone it would end at “maybe it is a side effect of the compound,” but by producing opposite results with the GDP-form and GTP-form Rab18 mutants, the support under the claim became much thicker.
- They state the limits of their own design explicitly. The authors themselves write that their aim was “not to emphasize purity, but to demonstrate the existence, enrichment and regulated secretion of SHH-XLEVs.” That is an honest stance.
- They state explicitly that EVs were handled in accordance with MISEV2023, they published under CC BY, and they disclosed the patent application.
Points to read with care
First, the paper’s account of its central drug conflicts with that drug’s primary source. This paper repeatedly describes CID1067700 as “a pan-Rab inhibitor that increases the GDP form of Rab.” But the paper that first characterized this compound says something different. CID1067700 is a competitive inhibitor that blocks more than 97% of both BODIPY-GTP and BODIPY-GDP binding, and the GTP-mimicking Rab7Q67L and the GDP-mimicking Rab7T22N were both inhibited to the same extent — which makes it natural to think of the result as a nucleotide-free state (Agola JO, et al. ACS Chem Biol 2012;7:1095-1108). The target there was Rab7, six protein constructs including a Rac1 mutant were examined (five GTPases — Cdc42, Rac1, Rab2, Rab7 and H-Ras), and the paper also states “at least 40% inhibitory activity against all GTPases tested.” Nor is it Rab-specific.
This discrepancy is not a small one. The paper’s very title is “Rab18-GDP”, and half of the pharmacological grounding rests on this drug. The saving grace is that the genetic evidence — the opposite phenotypes of S22N and Q67L — exists independently. The conclusion itself is supported from that side. But the explanation “because CID increases the GDP form” has, as things stand, no backing.
Second, care is needed in how the figure showing that “only the large bags increased” is read. The vertical axis of the NTA figure is not the number of particles but their volume (nm³/mL). Multiply the diameter by 20 and the volume goes up 8000-fold, so a single large particle easily outweighs a crowd of small ones on a volume axis. On top of that, the statistic is a chi-square test counting “detected or not detected” across three runs each (3/3 vs 0/3) — not a test of how much there was. NTA is also a method that was never good at quantifying above 1 µm in the first place. Because there is electron-microscopy backup I think the increase itself can be believed, but one should keep in mind the possibility that the difference is not as dramatic as the graph makes it look. Incidentally, this “volume” metric overlaps with the fact that the patent mentioned above defines its vesicles by a volume criterion of 10⁹ nm³ or greater.
Third, there is a large logical leap in the interpretation of the blood vessels that formed. The title says “angiogenic SHH-XLEVs,” but what was applied to the endothelial cells was the whole EV fraction collected from MSC culture supernatant; there is no experiment in which XLEVs alone were isolated and administered. Which means that the contribution of secreted components other than vesicles, or of residual drug, cannot be excluded on the basis of the experiments presented (the authors themselves state that purity is not their subject). And there is no experiment that knocks SHH out — neutralizing it with an antibody, knocking SHH down in the donor cells, inhibiting Smoothened on the receiving side. Any one of these would have made the claim of SHH dependence much stronger. The experiment in which the HUVECs themselves expressed SHH does show that SHH has the power to make blood vessels form, but it is no proof that the bag carried it and that this is what worked.
Fourth, the time point at which tube formation is assessed is early. This paper images at 90 minutes. A widely used protocol notes that non-transformed endothelial cells — HUVECs are in this class — form tubes within 6 hours (Arnaoutova I, Kleinman HK. Nat Protoc 2010;5:628-635), and another published procedure states that cells begin to line up at around 1 hour, with lumen-bearing tubes appearing from about 2 hours, and that counting is best done at 3–12 hours, when tube formation peaks (DeCicco-Skinner KL, et al. J Vis Exp 2014;(91):51312). The 90-minute point certainly sits short of that range, which means they are looking at a single point while things are still half-formed. A time course including later points, together with measures such as total tube length and the number of branch points, would have made the assessment far more secure.
Fifth, the identity of what is being detected as “SHH” is not clear. On immunoblots, bands appear at roughly 50 kDa and 27 kDa. But SHH is a molecule that undergoes cleavage: the signaling entity proper is an N-terminal fragment of about 19 kDa, the 27 kDa species corresponds to the C-terminal fragment, and 45–50 kDa corresponds to the uncleaved precursor (Bumcrot DA, et al. Mol Cell Biol 1995;15:2294-2303). The observed molecular weight given by the manufacturer of the antibody used is likewise 45–51 kDa. Whether what is loaded into the bag is the active entity or the precursor — the point that ought to matter most of all for any discussion of function — cannot be read off.
Sixth, there is a small but definite inconsistency. The graph in Figure 6C has the cell numbers printed directly on it, and the totals are 112 and 169. Yet the legend of the same figure says “n = 88–125 cells.” Those 88 and 125 match the numbers in Figure 5D exactly. It is presumably a mix-up made while writing the legend. It does not shake the conclusion, but the numbers should be checked.
Seventh, there are two places where the citation does not support the claim.
- The introduction cites Mackie et al. 2012 for the statement that “loading SHH into human MSC-derived EVs has been shown to enhance therapeutic efficacy in a model of ischemic heart failure.” But what that paper modified was human CD34-positive cells, not MSCs; the model was acute myocardial infarction; and what was administered was cells, not EVs (the exosome analysis was in vitro). The story that SHH-modified cells hand SHH over via exosomes is itself genuine, but both the cell type and the name of the model are different (Mackie AR, et al. Circ Res 2012;111:312-321).
- Tanaka et al. 2024, cited as having “reported similar EV-associated fibrous structures in the embryo,” is, in the reference list, the paper on kinesin-1, a calcium channel and glucose homeostasis (EMBO Rep), and it deals with neither EVs nor fibrous structures nor embryos. The item in the authors’ own reference list whose content does match is rather the 2023 Developmental Cell paper (nodal flow carrying a fibrous network containing PKD1L1). Which of the two was intended cannot be stated for certain, but a mix-up of reference numbers seems likely.
Eighth, there are limits in the design itself. The Rab18 mutants were introduced by overexpression; there is no experiment in which endogenous Rab18 was knocked down or knocked out. What is more, the authors themselves write that cells expressing S22N “tended to be flatter and more spread out (not quantified).” If shape changes, secretion can change too, so this is a confounder that cannot be ignored. The material used in the binding experiments was also mouse brain extract, not material from the MSCs or fibroblasts that are the stage for the experiments. And the word “XLEV” is not standard MISEV terminology; the 600 nm cut-off is the authors’ own definition as well. When comparing with other studies, this has to be borne in mind.
What would have made this a higher-quality study
- Isolate XLEVs alone and administer them. If, by size exclusion, filtration or fluorescence sorting, the XLEV fraction and the small-EV fraction were applied separately to endothelial cells with particle numbers matched, the claim in the title could be stated from the experiment just as it stands.
- Knock SHH out. Vesicles from donor cells in which SHH has been knocked down; vesicles neutralized with an anti-SHH antibody; endothelial cells with Smoothened inhibited on the receiving side. These would greatly strengthen the verification of SHH dependence (reading downstream Gli transcriptional activity would be better still). To show that “the XLEV carried it,” though, they would need to be combined with the separated administration above.
- Assess tube formation at 4–6 hours, with multiple metrics. Taking not just tube counts but total length and the number of branch points would also make the result more robust to scatter.
- Take Rab18 itself down. Whether the phenotype disappears when endogenous Rab18 is reduced by siRNA or knockout is the decisive move. Measuring the nucleotide-binding state of the Rab directly in CID-treated cells would, alongside that, allow the account of the drug to be rewritten correctly.
- Settle the identity of the cargo. Determine which fragment of SHH is loaded, using fragment-specific antibodies or mass spectrometry.
- Confirm it in a living body. Looking at whether vessels increase in a Matrigel plug or a hindlimb ischemia model would give the phrase “angiogenic” its backing in vivo.
In fairness, though, let me note that many of these lie outside the scope of this paper. The authors state explicitly that their aim was to show existence and a regulatory mechanism, and against that aim they have done the work. The problem, I think, is rather that the title and the abstract step one pace ahead of the design of the main text.
Exotaro’s view
Let me set out a personal matter first. Of the four references cited in this paper’s introduction as grounds for the regenerative-medicine potential of MSC-derived EVs, two are my own papers (Nakazaki, M., et al., Journal of Extracellular Vesicles 2021;10(11):e12137, and Nakazaki, M., et al., International Journal of Molecular Sciences 2024;25(24):13460). And in the acknowledgements, my own name is listed as someone they discussed the work with. So this article is not an entirely third-party critique. With that said, I have tried to read it as fairly as I can.
What I myself have worked on is the smaller bags. In a rodent model of severe spinal cord injury, the MSCs administered intravenously do not themselves reach the injury site, and yet the sEVs that the MSCs released inside the body did migrate to the injury site and associated specifically with the M2 macrophages present there. And even at the same total dose, a single large bolus did not work: only when it was delivered split across three days did we obtain motor-function recovery equivalent to that from the MSCs themselves. In the course of this, TGF-β expression rose, the permeability of the blood–spinal cord barrier fell, and tight junction proteins increased.
So what sat at the center of my own work was also stabilizing blood vessels. The present paper says that the same stem cell, using bags of an entirely different size, makes blood vessels form. Defense and offense; stabilization and new growth. If the same cell is doing different jobs by choosing between bag sizes, that is rather interesting.
Here is something I want to put down honestly, addressed to my own field. We often throw the large bags away. In protocols for handling small EVs, it is common to spin cells and debris down at low speed, pass the sample through a 0.22 µm filter, and then collect the small particles. Take that filtration step and bags larger than 600 nm fall outside what is recovered. (The present paper, too, detected large EVs after clarifying at low speed, so whether they are removed depends on the steps taken.)
If that is so, then we should ask: of what we have been measuring as “the effect of MSC-sEVs,” how much consists of activity that properly belonged to the side we discarded? Or conversely: if part of the efficacy lay precisely in what was discarded, have we been crediting it to the small particles without ever looking? This paper does not pose that question directly, but it was the first thing that came to mind when I finished reading.
The other thing that stayed with me strongly is the perspective of where it comes out. We tend to measure secretion as “out or not out,” “a lot or a little,” but this paper used film to show a directional phenomenon: release only from the membrane around the nucleus. In a developmental context, direction is information itself. If this is to be applied therapeutically, the direction in which EVs are released at an injury site may in truth bear on efficacy too.
Finally, on the points I read critically. The two I weighted most heavily in this article are that the account of the central drug’s mechanism of action conflicts with its primary source and that there is no experiment in which the XLEVs themselves were isolated and administered. This does not in any way mean that the study has no value. “A large bag, loaded with a lipid-laden instruction molecule, going out through a fixed exit around the nucleus” — filming that is, first of all, an achievement in itself. The next turn is to confirm, with the bags themselves, whether they really are building blood vessels.
A small parcel found inside a mouse embryo about 20 years ago has come back into a dish of human stem cells and is trying to build blood vessels. So this is how time flows in basic research — the paper left me thinking that quite plainly.
This article was written with the assistance of AI.