Have you ever wondered what became of a wisdom tooth after the dentist pulled it for you? The extracted tooth goes into a medical waste container and is incinerated just as it is.
And yet inside it, in the dental pulp, there are stem cells. These cells branched off during development from the neural crest—the cell population that gives rise to the peripheral nerves (Chai Y, et al., Development 2000;127(8):1671–1679). Unlike bone marrow and adipose stem cells, which come from mesoderm, the dental pulp stem cell alone is a relative of nerve. What we read today is a paper asking whether the exosomes released by the dental pulp mesenchymal stem cells (DPMSCs) obtainable from this “material that gets thrown away”—sacs roughly 100 nm in diameter in which a cell wraps up its contents and sends them outside, one kind of the particles collectively called extracellular vesicles (EVs)—can be used against spinal cord injury.
Let me start with the conclusion. The authors lined up the exosomes of 3 kinds of mesenchymal stem cell (MSC)—umbilical cord, bone marrow, and dental pulp— and compared them side by side in the same experimental system; the dental pulp-derived ones were the best, so that is what they chose. Next, proteomic analysis identified 727 proteins, from which they narrowed down to proteasome 26S subunit non-ATPase 1 (PSMD1)—a structural part of the proteasome, the machine that breaks proteins down. Vesicles loaded with it tipped microglia in a culture dish toward the “repair side,” neural stem cells cultured in the room next door survived, and the motor scores of mice rose 1 week after injury—so the authors report. Looking for a treatment for spinal cord injury may be less a matter of “curing the patient” than of looking for a way to “change the air in the room the patient is in.”
Let me set the honest frame first. The only places this paper tried the treatment were cells in a culture dish and mice. The human data amount to nothing more than the observation that PSMD1 was lower in specimens from 6 patients than in those from 6 healthy controls; they are not “it worked in humans too.” And on top of that, it was a cross-species setup in which vesicles from commercially available human dental pulp MSCs were put into immunologically normal mice.
Before we read, let me block off two traps. First. PSMD1 is a part of the proteasome, but this paper never once measured proteasome activity. “The part worked as the machine” is no more than a hypothesis the reader is apt to supply on their own. Second. When this article says “tipped toward the M2 side,” it refers to this and nothing more: in flow cytometry (an instrument that runs cells past one at a time and counts markers), the proportion of cells falling inside the range where two markers—CD68 and CD206, or CD68 and CD163—light up at the same time, a demarcation called a “gate,” increased.
Note that among the references this article cites, the three the paper itself cites are Liu 2022, Tomko 2013, and Bencomo-Alvarez 2021. The other references, along with the journal metrics, the clinical trial registrations, and the authors’ publication records, are things this article looked up outside the paper.
Publication Details
- Paper title: Exosomal PSMD1 Derived From Dental Pulp Mesenchymal Stem Cells Promotes Microglial M2 Polarization to Alleviate Spinal Cord Injury
- Paper type: Original research article. A preclinical study using mouse cell lines and a mouse spinal cord injury model; there is no administration to humans (the human specimens are a 6-versus-6 observation)
- Authors: Xiaohui Xing (co-first author), Yuan Hao (co-first author), Guanghui Zhang, Xiaoyang Xing, Hongfei Jia, Chenlong Liu, Xin Wang, Tao Li (corresponding author), Xueyuan Li, and Yilei Xiao (corresponding author)—ten in all
- Affiliation: Centered on the Department of Neurosurgery at Liaocheng People’s Hospital, affiliated with Shandong First Medical University, together with that hospital’s key laboratory. Also the Second Hospital of Shandong University; the Department of Spine Surgery at Shandong Provincial Hospital, affiliated with Shandong First Medical University; and the Department of Neurosurgery at Brigham and Women’s Hospital / Harvard Medical School in the USA
- Journal: CNS Neuroscience & Therapeutics (published by Wiley), 2026, Volume 32, Issue 9, Article No. e71133
- DOI / Link: 10.1002/cns.71133
- Peer review and publication dates: Received April 2, 2026; revised manuscript received August 7; accepted August 24
- Open access: Yes (CC BY. Free to reuse with attribution)
- Impact Factor: We could not confirm a definitive value that a JCR year could be attached to (Journal Citation Reports was unreachable, and the figures shown by secondary sources are split, ranging from 4.3 to 5.7). As alternative primary data, the 2-year mean citedness calculated by OpenAlex is 5.99 (retrieved September 9, 2026)
- About the journal: Indexed in MEDLINE, and fully open access since 2019. Its publication volume grew roughly 4-fold, from 143 papers in 2020 to 560 in 2025, and the majority of what it published in 2025 includes authors with affiliations in China
- Funding: Six grants in all: Shandong Province Traditional Chinese Medicine Science and Technology Project M-2022123, Liaocheng Key Research and Development Program 2023YD42, Shandong Provincial Natural Science Foundation ZR2021MH303, Shandong Province Taishan Scholar Program tsqn202103200, Liaocheng People’s Hospital Youth Research Fund 201910915, and Affiliated Hospital of Shandong Second Medical University Research and Development Fund 2024FYM138
- Conflicts of interest: Declared none. The data are stated to be “available from the corresponding author upon reasonable request,” and no public repository where the raw data have been deposited is named (ethics approvals: human 2024258 / animal 2024–02-19A)
About corresponding authors Tao Li and Yilei Xiao: Tao Li (Department of Spine Surgery, Shandong Provincial Hospital affiliated with Shandong First Medical University, Jinan) returns only one hit when PubMed is searched with the contact email given in the paper—a 2011 study of the rat thoracic vertebral body growth plate in Nan Fang Yi Ke Da Xue Xue Bao (Journal of Southern Medical University). Since “Li Tao” is a common name, I will not list publications that do not match. That said, a 2021 paper in the European Journal of Pharmaceutical Sciences by a Tao Li listed under the same hospital and the same department was retracted in 2023 on the grounds of serious similarity to an earlier paper from a different group. The contact email differs, however, and I could not confirm whether this is the same person. The other corresponding author, Yilei Xiao (Department of Neurosurgery, Liaocheng People’s Hospital, ORCID 0000-0002-6651-4122), is identified with certainty because the email matches. In 2018 he reported that bone marrow MSC-derived exosomes prevent apoptosis of oligodendrocytes (Journal of Cellular Biochemistry 2018, PMID 30191592), and in 2024 that the lncRNA LRRC75A-AS1 carried by adipose-derived stem cell exosomes has an anti-inflammatory effect after spinal cord injury (Applied Biochemistry and Biotechnology 2024, PMID 38165592). The latter has almost the same design as the present paper, which makes this a team that has pursued the theme of protecting the central nervous system with MSC exosomes for 8 years. Note that of the co-first authors, only Xiaohui Xing also lists a hospital affiliated with Harvard Medical School as one of their affiliations, and there is no publication record other than this paper to show any tenure there. The most reliable route by which that hospital appears is co-author Xin Wang, an associate professor in its Department of Neurosurgery (ORCID 0000-0002-9233-2170), but his role was “manuscript revision” only.
What Wasn’t Understood Before?
The cruelty of spinal cord injury (SCI) lies in the fact that what happens in the instant of the accident and what happens afterward are two different things. On top of the initial trauma comes a cascade of secondary injury—in the paper’s words, “excessive inflammation, glial activation, neuronal apoptosis, edema, and oxidative stress” (apoptosis is programmed cell death, in which a cell throws its own switch and dies by folding itself up). The primary injury is over in zero seconds; by the time the patient reaches hospital, it can no longer be taken back.
Let us look at the ceiling in numbers. In a prospective study of 313 cases of cervical spinal cord injury at 6 North American centers, comparing an early group operated on at a mean of 14.2 hours against a late group at a mean of 48.3 hours, the proportion whose AIS grade (the ASIA Impairment Scale, a yardstick that expresses severity on 5 levels from A, complete paralysis, to E, normal) improved by 2 grades or more at 6 months was 19.8% in the early group and 8.8% in the late group (Fehlings MG, et al., PLoS One 2012;7(2):e32037). Because the timing of surgery was the attending physician’s judgment, this cannot be read as causal, but even with close to the best available treatment, only about 1 patient in 5 improves substantially.
The one who runs the scene is not the injured neuron but the immune cell living next door, the microglia. They are a separate lineage from blood monocytes, born in the yolk sac at the very earliest stage of embryonic development and taking up residence in the brain and spinal cord before birth (Ginhoux F, et al., Science 2010;330(6005):841–845). A caretaker who has been there since before we were born changes character at the moment of injury, and can turn either into the side that kills neurons or into the side that helps repair them.
👦 Student: If that caretaker runs wild and makes the inflammation worse, why not just silence it?
🧬 Dr. Exotaro: That is exactly the idea the field once believed and was then forced to revise. Both of these are mouse experiments, mind you, but depleting microglia causes scar formation to break down and even more neurons to die, and motor function recovery actually got worse (Bellver-Landete V, et al., Nature Communications 2019;10(1):518). The scar itself is the same: when mice are prevented from forming one, even axon regeneration falls significantly (Anderson MA, et al., Nature 2016;532(7598):195–200). The point is not to drive them out but to change their mood.
The words used for that “mood” were M1 / M2. A study that brought the scheme into mouse spinal cord injury reported that only M1 showed neurotoxicity, while M2 promoted axon outgrowth even in the presence of inhibitory substrate (Kigerl KA, et al., Journal of Neuroscience 2009;29(43):13435–13444). The present paper starts from here too. The yardstick itself has wobbled, however: in 2016 came an essay stating outright that the concept is obsolete (Ransohoff RM, Nature Neuroscience 2016;19(8):987–991), and in 2022 a consensus document calling for an end to the dichotomous labels (Paolicelli RC, et al., Neuron 2022;110(21):3458–3483). Neither of them appears among this paper’s 60 references.
Dental pulp is chosen not only for its neural crest origin but also because it secretes more neurotrophic factors than bone marrow or adipose tissue (Mead B, et al., PLoS One 2014;9(10):e109305). But “neural crest-derived, therefore suited to nerve repair” is an inference, not a proven therapeutic principle. And that exosomes from dental pulp stem cells (DPSCs) suppress M1 macrophage conversion and hold back the progression of spinal cord injury is something the present paper itself introduces as prior work (Liu C, et al., J Nanobiotechnol 2022;20:65). That the sacs work was already known. What was blank was which of their contents works.
Which brings us to the main subject: the cell’s garbage disposal plant. The “garbage disposal plant” in the title means this machine. A cell sticks onto defective proteins a sticky note saying “this may be discarded”—ubiquitin. That system is the ubiquitin-proteasome system (UPS). The machine itself comes in two parts. The 20S core particle is a hollow barrel, and the blades of its degradative enzymes face inward only. The lid is the 19S regulatory particle, which grabs the sticky note, unfolds the protein with ATP, and pushes it into the barrel. 20S + 19S = the 26S proteasome.
So where does PSMD1 sit? PSMD1 is a structural subunit of the 19S regulatory particle—neither a blade nor a motor, but the scaffold on which the other subunits are assembled: a pillar, in the language of buildings (Tomko RJ Jr, Hochstrasser M, Annual Review of Biochemistry 2013;82:415–445). To say that this backstage part rides a vesicle out of the cell and moves a pathway in a different cell is tantamount to saying that a structural part works as a messenger all on its own.
👦 Student: If somebody sends me just one pillar, I don’t think a house gets built.
🧬 Dr. Exotaro: That is exactly where it snags. In cancer research, PSMD1 has been a molecule treated as a “target to be struck down.” In chronic myeloid leukemia it acts oncogenically by stabilizing NF-κB (the transcription factor that serves as the command center of inflammation; once it enters the nucleus, the genes for TNF-α and IL-1β all start moving at once) (Bencomo-Alvarez AE, et al., Oncogene 2021;40(15):2697–2710). Villain in cancer, hero in the spinal cord. And this is a paper that the present paper itself cited, as reference 54.
What Did This Paper Find?
Putting Three Stem Cells in the Same Ring
Exosomes were recovered by a precipitation method from the culture supernatants of human umbilical cord MSCs, bone marrow MSCs, and dental pulp MSCs (purchased), and the authors confirm that they were structures of roughly 100 nm mean diameter that are wrapped in a lipid bilayer, and that the positive markers CD9, CD81, and TSG101 were detected while Calnexin, the endoplasmic reticulum protein whose presence would mean unwanted contamination, was negative. When the three were added to mouse BV2 cells exposed to 1 μg/mL lipopolysaccharide (LPS) for 48 hours to stand in for “microglia in which inflammation has been provoked,” all three partially restored the viability that had fallen (p < 0.001), but the dental pulp-derived ones were the most marked (p < 0.001), and dental pulp was the strongest at suppressing apoptosis as well (p < 0.001).
From 727 to 20, and Then to PSMD1
Only the dental pulp MSC exosomes were run through liquid chromatography–tandem mass spectrometry (LC–MS/MS)—an instrument that sends a mixture of proteins through in sequence and infers their identities from mass—and 727 proteins were identified. The narrowing-down leaves room for argument. The paper writes that “neurological disease-related pathways, including prion disease and Parkinson’s disease, were significantly represented,” but the top entries in the KEGG figure are “COVID-19,” “Ribosome,” “ECM-receptor interaction,” and “Proteasome”—not one neurological disease among them. Prion disease was 11th and Parkinson’s disease 15th. The “significant involvement of the NF-κB pathway in the Reactome analysis” likewise rests on a figure that lines up only the NF-κB-related entries, and of the seven entries, the bottom three do not reach 1 on −log10(q value).
Crossing the neurodegenerative disease-related set with the NF-κB pathway-related set (the Venn diagram breaks down as 51 on the neurodegenerative side only, 20 in both, and 2 on the NF-κB side only—the NF-κB side comes to only 22 in total) yields 20 overlapping proteins. They were 19 proteasome subunits—PSMA1–PSMA7, PSMB1/2/4/5, PSMC3/5/6, and PSMD1/2/11/12/14—plus UBA52. Nearly every candidate is a part of one and the same machine. Both the 19S side and the 20S side are present. But because this is an analysis that extracts and digests before measuring, there is no way to distinguish whether they were loaded as the 26S complex or separately.
Three reasons are given for why PSMD1 was chosen. First, there are reports that it is associated with the progression of neurodegenerative disease. Second, it was detected in this study. Third—and this is the only human data in the paper—comparing 6 patients with spinal cord injury (more than 24 hours from onset, 18 years or older) against 6 age-matched healthy controls, PSMD1 was significantly lower in the patient-derived exosomes (blood-derived, p < 0.01; cerebrospinal fluid (CSF), p < 0.05). But this is a cross-sectional observation, and neither the direction of causation nor confounding can be distinguished.
👦 Student: If 19 of the candidates are parts of the same machine, is there any point in picking one out of them?
🧬 Dr. Exotaro: That question goes straight to this paper’s weak spot. The only thing supporting the selection is the third item, the patient data. The figure legend says “of these, only PSMD1 was detected in exosomes,” but taken literally that does not square with the main text’s claim to have identified 727 proteins. It probably means “only PSMD1 was verified by Western blot,” but that one sentence is the sole justification for pursuing PSMD1 on its own.
The Key to Tipping Toward Repair Was JAK2/STAT3
In BV2 cells given LPS, the M1 index CD68⁺iNOS⁺ rose (p < 0.001) and the M2 index CD68⁺CD206⁺ fell (p < 0.01); adding dental pulp MSC exosomes moved both in the opposite direction (both p < 0.001), and CXCL12, IL-1β, and TNF-α—which the paper calls “downstream mediators of M1 activation”—were suppressed as well (p < 0.05).
As for the pathway, JAK2 and STAT3—relay runners that carry a signal from the cell surface to the nucleus, and that switch “on” when a phosphate group, the switch, is attached (when they are phosphorylated)—are the ones that move. What the paper measures is the ratio of the phosphorylated form to the total. LPS stimulation significantly “altered” the p-JAK2/JAK2 ratio and the p-STAT3/STAT3 ratio (both p < 0.001), and the exosomes raised them (p < 0.05 and p < 0.01, respectively). “Altered”—whether they went up or down is never once written in the main text. Meanwhile, there were no significant differences in p-STAT6/STAT6, p-ERK/ERK, or p-CEBPβ/CEBPβ. Adding 10 μM of the STAT3 inhibitor stattic cancelled out even the rise in the M2 markers CD163 and CD206 (both p < 0.01). The authors explain M2 conversion by way of JAK2/STAT3, but the center of classical M2 is STAT6. What it means that STAT6 did not move is not touched on in the discussion.
PSMD1 itself is also added and subtracted. Overexpressing it in BV2 raised the M2-side indices (CD68⁺CD163⁺ and CD68⁺CD206⁺) (p < 0.001); siRNA (small interfering RNA, a short RNA that temporarily lowers the activity of a targeted gene) moved them in the opposite direction (p < 0.01); IL-1β and TNF-α fell with overexpression (both p < 0.05) and rose with knockdown (both p < 0.01); and the pathway moved in both directions as well (all p < 0.001). Yet this figure contains no panel showing that the amount of PSMD1 protein in BV2 actually went up or down.
Knocking PSMD1 down on the producer side lowered PSMD1 in both the cells and the exosomes; exosomes from overexpressing cells raised the M2 side in the recipient, and knockdown cancelled that out. For this rescue experiment, however, the paper prints not a single p value, and the legend does not even define the significance symbols. Next, dental pulp MSCs expressing a PSMD1-GFP fusion protein were made, and their exosomes raised the M2 side more strongly than those derived from the negative control vector (p < 0.01).
The Neural Stem Cells Were Protected From the Room Next Door (So the Paper Says)
Moving the recipient to the mouse neural stem cell line NE-4C, LPS lowered viability (p < 0.05) and increased apoptosis (p < 0.001), the exosomes improved both (p < 0.05), and the paper writes that the PSMD1-loaded ones had the larger effect (though the main text carries no p value for this between-group comparison; the legend of Figure 8 merely defines ”& = comparison with the OE-NC-Exo group,” and the numbers cannot be read off the main text). Under electron microscopy, autophagosomes—the sacs of the system by which a cell wraps up its own contents and sends them for degradation (autophagy)—decreased, and Beclin-1 and LC3II, the markers of autophagy, and caspase-3, the executioner of cell death, went down, while Bcl-2, which puts a brake on cell death, went up (all p < 0.05).
👦 Student: Autophagy is the cell’s housecleaning, isn’t it? How does suppressing that come out as a good thing?
🧬 Dr. Exotaro: That is the most uncomfortable part of all. In neurons, basal autophagy is if anything protective, and shutting down autophagy in the nervous system of mice causes neurodegeneration (Hara T, et al., Nature 2006;441(7095):885–889). What’s more, a drop in LC3-II can be read as “autophagosomes can no longer be made” and equally as “they are being degraded fast as soon as they are made”—it happens in either of two opposite situations. Telling them apart requires the “flux measurement” that the consensus guidelines call for (Klionsky DJ, et al., Autophagy 2021;17:1–382), and that was not done. Besides, what is written as having gone down in culture is caspase-3, and it is not written separately from the cleaved caspase-3 of the animal experiments. Whether the activated cleaved form decreased cannot be read from that one line. And since they did not test whether blocking autophagy abolishes the improvement in survival, whether it is “they survived because it went down” or “it merely went down because the stress was reduced” cannot be settled either. The abstract’s “inhibited autophagy” is one step out in front of the data.
The most ambitious move is the Transwell co-culture. BV2 cells and PSMD1-loaded exosomes go in the upper chamber, LPS-treated NE-4C in the lower. In the BV2 above, the M2-side gate increased (p < 0.01, and more strongly still in the PSMD1 group, p < 0.001), and the NE-4C below showed higher viability (p < 0.05) and less apoptosis (p < 0.001). The authors conclude from this that the neuroprotection is indirect, mediated by M2 polarization of the BV2. The claim is that rather than handing a drug directly to the ailing resident, they changed the air in the next room and the resident got better—but what is happening here is an event inside a culture dish. On top of that, there is no control group in which the vesicles alone were placed without BV2, and the paper’s own data show that the vesicles raise NE-4C viability even without BV2. The correct conclusion is: there is a direct effect, and an indirect route may also exist—the ratio between them is unknown.
In Mice, They Looked for One Week Only
Adult female BALB/c mice (26–30 g) were randomly assigned to 4 groups of n=7 (Control + PBS, Model + PBS, Model + Exo from negative control vector-transduced MSCs, Model + PSMD1-Exo), anesthetized with 50 mg/kg sodium pentobarbital, and then the vertebral arch at the level of the tenth thoracic vertebra (T10) (the arch of bone that covers over the spinal cord) was removed to expose the spinal cord, and a contusion was delivered with a striking device (an impactor). The impact conditions are not in the main text. 2 hours after injury, 100 μL of exosomes labeled with the fluorescent dye DiR was administered—the original reads “intramuscular injections … at the injury site,” that is, an intramuscular injection at the injury site, which means it went into the surrounding muscle rather than the spinal cord, but there is no explanation of the precise location.
Motor function was measured with the BBB score (the Basso–Beattie–Bresnahan locomotor scale, a gait measure that rates hindlimb movement from 0 to 21 points; note that this BBB is a different abbreviation from the familiar “blood–brain barrier”) at a single time point only, 1 week after injury; it fell steeply in the injured group (p < 0.001) and improved significantly in the PSMD1 group (p < 0.05). The immunostaining, however, consists of representative images with neither quantification nor statistics—PSMD1 is said to have fallen in the injured group and “recovered” with administration, Ki67-positive cells (a marker of proliferation) were more numerous in the PSMD1 group, fluorescence for the M1-side marker iNOS decreased, and the M2-side marker CD206 was strongest in the PSMD1 group.
Quantification is limited to Western blots and serum. Cleaved caspase-3 and Bax rose in the injured group (p < 0.01) and fell in both exosome groups (p < 0.01). NGF and NT-3, if anything, rose in the injured group (p < 0.01), and rose further with exosome administration (p < 0.05). GFAP and vimentin—markers of activated astrocytes, whose rise is a sign that “the glial scar is growing”—rose in the injured group (p < 0.001) and fell in the PSMD1 group (p < 0.01). Serum IL-1β and TNF-α rose markedly in the injury + PBS group (p < 0.001) and fell significantly with exosome administration (p < 0.05). Note that the authors are more cautious than the abstract: “was associated with” in the discussion, and “may contribute” in the conclusion. That difference in temperature is a hint.
How Will the Future Change? (The Path to the Clinic)
This is a preclinical original research article. As far as a search of ClinicalTrials.gov on September 9, 2026 goes, there is not a single registered clinical trial in which dental pulp MSC exosomes were administered to humans (ChiCTR and jRCT were not checked, so I cannot say “none exists anywhere in the world”). Beyond that, there is not a single interventional trial of extracellular vesicles for traumatic spinal cord injury at all. Multiple Phase I/II trials are running for stroke, amyotrophic lateral sclerosis (ALS), and Alzheimer’s disease, yet spinal cord injury alone is a gaping blank. For the cells themselves, there are 38 registered trials targeting spinal cord injury.
The modality itself, though, has entered the human body. A Phase I trial in which allogeneic human umbilical cord MSC-derived exosomes were given intrathecally to 9 patients with subacute complete spinal cord injury was reported in 2024; over 12 months of follow-up there were no adverse events attributable to the intervention, and on the sensory assessment of the same ASIA (American Spinal Injury Association) scheme as the AIS above, the scores for pinprick (P = 0.039) and light touch (P = 0.038) improved significantly (Akhlaghpasand M, et al., Stem Cell Research & Therapy 2024;15(1):264). But this was single-arm, unblinded, 9 cases. It is a first step on safety, not on efficacy.
👦 Student: If you can get them from tissue that was going to be thrown away, shouldn’t I just save my own wisdom teeth?
🧬 Dr. Exotaro: The idea is perfectly sound. But what this paper actually did used cells from “somebody else, and from a different creature at that.” What they used was commercially available human dental pulp MSCs, and the recipients were immunologically normal BALB/c mice; the paper does not discuss this cross-species combination in a single word. In the acute phase, “we will now start culturing” will not arrive in time, so in practice it would become an off-the-shelf allogeneic product. And then the cargo varies from donor to donor. Autologous or allogeneic, deciduous teeth or permanent teeth, how many donors and how many passages—not one of these is written in this paper. Saying in the limitations section that “cargo composition may vary between donors” while writing down no donor information at all does not sit well.
Dental stem cells “themselves” have reached rigorously designed clinical trials. Teijin Pharma’s JTR-161, an allogeneic stem cell product derived from the dental pulp of healthy adults, completed a double-blind, randomized, placebo-controlled Phase I/II trial in 79 patients with acute cerebral infarction in 2021, designed around a single intravenous administration of either 1×10⁸ or 3×10⁸ cells per patient. The trial results, however, are not registered on ClinicalTrials.gov (as of September 9, 2026; domestic registries such as jRCT were not checked), so within what I could confirm, whether it was effective is unknown. Even so, the cells have reached humans. What has not is the exosomes those cells release.
How to Critically Read This Research (Limitations and Paths to Improve It)
Before the criticism, the virtues. That they measured three cell sources before choosing one, drew their candidate molecule out of proteomic analysis, and connected to humans with patient specimens. That they went from overexpression to knockdown to rescue and placed a pathway inhibitor on top of that. That they asked “direct or indirect?” of themselves with the Transwell. That while advertising regeneration, they did not hide the finding that NeuN (a marker that stains the nuclei of mature neurons) went down. The limitations of a good study and the limitations of a bad study are different things. This is the former.
The limitations the authors themselves acknowledge number five. (1) The effect is short-term. (2) The precipitation method may co-isolate contaminants, and the cargo may vary between donors. (3) Both the patient specimens and the in vivo work are small in scale, and the clinical information is limited as well. (4) Overexpression and knockdown alone cannot establish PSMD1 as the active ingredient. (5) The effect may also derive from other cargo, and the relationship between M2 polarization and neural stem cells is complex. That they wrote the last two themselves is honest. Now, the problems that go untouched.
The dose is printed nowhere. For the animal experiments there is only the volume, “100 μL”; for the culture experiments, not one line states how much was added. LPS and stattic have concentrations, but the protagonist alone does not.
👦 Student: Is 100 μL not good enough? There is a proper number written there.
🧬 Dr. Exotaro: Volume is not dose. It is like saying “we gave them a glass of sugar water” without writing how many grams of sugar were in it. Without a concentration, it cannot be converted into a particle count or a protein amount. And this paper does measure particle concentration—the vertical axis of the nanoparticle tracking analysis figure is Particles/mL. The means to measure it were there, and yet the numbers for the administered solution are not printed.
What this hits directly is the three-way comparison. Because it is not written whether particle numbers were matched, there is no way to decide whether dental pulp’s win means “the cargo is superior” or merely “dental pulp MSCs put out the most particles.” The effect size has the same hole: nothing tells us how many of the 0–21 points the BBB’s “significant improvement (p < 0.05)” amounts to, or what percentage IL-1β’s “significant decrease” was. For 4 groups of n=7 there is no prior power calculation, no pre-specification of a primary endpoint, and no reporting of surgical deaths or exclusions (all items required by the ARRIVE 2.0 reporting guidelines). The culture work says only “triplicate measurements, independently three times,” so there is no way to decide whether the n in each figure is 3 or 9. Taking technical replicates as n makes p values smaller, but which of the two was used as the unit of analysis is not written—multiple comparison correction is applied to the between-group comparisons themselves, but pseudoreplication is not fixed by that correction. The BBB is a scale validated for rats (Basso DM, et al., Journal of Neurotrauma 1995;12(1):1–21); for mice, the Basso Mouse Scale was created later (Basso DM, et al., Journal of Neurotrauma 2006;23(5):635–659). This study applied one-way ANOVA plus Tukey to an ordinal scale, and no baseline assessment immediately after injury is described.
The proteomics lacks the description needed for reproduction. The methods name the instruments EASY-nLC and Orbitrap Eclipse Tribrid (Thermo) while also listing parameters for PASEF MS/MS, an acquisition scheme normally described for instruments from a different manufacturer (Meier F, et al., Molecular & Cellular Proteomics 2018;17(12):2534–2545), so which instrument the measurement was made on cannot be read off. Neither the search engine, nor the FDR threshold, nor the number of replicates is reported. The “727” cannot be retested until one asks the authors and looks at the raw data. And although the filter for selection was NF-κB-related, the mechanistic verification was pushed onto JAK2/STAT3, and no actual measurement of NF-κB is reported.
The experiment that directly tested pathway dependence is a single inhibitor. That JAK2/STAT3 moves along with increases and decreases in PSMD1 is shown, but the only thing used to stop the pathway was 10 μM stattic, and the treatment time is not written, nor is there data showing cell survival in the stattic group. Whether the M2 markers disappeared because the pathway was stopped or merely because the cells were weakened cannot be separated out, and there is neither a JAK2 inhibitor nor a STAT3 siRNA. On top of that, although stattic is an inhibitor of STAT3, it cancelled out the exosome-driven rise in the p-JAK2/JAK2 ratio along with it (p < 0.01). The paper attempts no explanation, and there is no data ruling out non-specific action or cytotoxicity. The inhibitor I listed among the virtues was a virtue only as far as having placed it there.
It puts the proteasome in its name, yet never once measures what the proteasome does. There is no activity assay, no inhibition with MG132, and no confirmation of incorporation into the 26S complex. Uptake of the vesicles is shown with PKH26-labeled fluorescence images (Fig. 1d, Fig. 7c, Fig. 8a), and that PSMD1 increased in the recipient BV2 is in Fig. 6e. That, however, is not proof that the protein itself was handed over; it is equally explained by the recipient having made it on its own. Although both a PSMD1-GFP fusion line and an anti-GFP antibody are on hand, there is not one lane in which the lysate of the BV2 that took the vesicles up was run with the anti-GFP antibody. The in vivo evidence consists only of near-infrared fluorescence images of DiR, a dye that inserts into membranes, and there is no quantification, no statistics, no control group given free dye alone, and no organ-by-organ distribution. Since T10 after contusion is a place where the blood–spinal cord barrier is broken, injecting dye alone could make the injury site glow. On top of that, there is no description of having checked whether transfection reagent or expression plasmid remained in the fraction. If they were contaminating it, “PSMD1 increased in BV2” would equally be explained by PSMD1 that BV2 itself made—and yet there is no qPCR for the vector backbone.
The crudeness of the isolation method hits the one piece of human data directly. Polymer precipitation brings lipoproteins down along with the vesicles, and neither the orthogonal validation required by the international reporting standards (Welsh JA, et al., Journal of Extracellular Vesicles 2024;13:e12404) nor registration with EV-TRACK, the international database that lets third parties inspect experimental conditions, can be confirmed from the main text. Because recovery efficiency changes with a rise in acute-phase proteins, with dilution from intravenous fluids, and with hemolysis, the difference that “PSMD1 is low in patients” cannot be separated into a difference in the contents of the vesicles or a difference in the composition of the co-precipitated fraction.
And this culture system is not spinal cord injury to begin with. The methods text says “To mimic SCI in vitro,” and what is actually done is to apply 1 μg/mL LPS for 48 hours. LPS is an endotoxin, a stimulus that hits TLR4 down a single road. But what is happening in the acute phase of trauma is sterile inflammation with no bacteria present, and what lights the fire is the DAMPs (HMGB1, ATP, and so on) leaking from broken cells. That something moved with LPS is no guarantee that it moves with trauma. BV2 is an immortalized microglia-”like” line, and NE-4C is a neuroectodermal line immortalized by deleting p53. What it means to be running experiments whose main readout is apoptosis in cells that lack the p53 pathway is written nowhere.
The very act of counting proportions has its own trap. The main readout for M1/M2 is percentages (immunofluorescence and Western blotting are used alongside). The paper itself reports that LPS lowered BV2 viability at 48 hours (p < 0.001). If LPS is selectively killing one of the phenotypes and the exosomes are merely rescuing it, the proportions will move without any cell changing its character. The in vivo side is thinner still: the title asserts “Microglial,” yet there is not a single stain that distinguishes resident microglia from infiltrating macrophages; the Control group had no surgery, so the whole of the surgical insult is mixed into the difference; and on top of that, there is no knockdown group in vivo, so the necessity of PSMD1 has not been verified in the living animal.
So how could the quality have been raised? In order of cost. With zero additional experiments. (1) Deposit the raw data in PRIDE and state the instrument, the FDR threshold, and the number of replicates. (2) Rewrite every dose in particle number and μg of protein, and pin down the administration site and the impact conditions. (3) Reanalyze the BBB with Kruskal–Wallis plus Dunn. With material already in hand. (4) Run one lane of an anti-GFP Western on BV2 lysate. (5) Measure the remaining 18 subunits in the patient specimens as well. In a few weeks. (6) A three-source comparison matched by particle number; (7) rescue with an siRNA against a different region; (8) MG132 and an activity assay; (9) a proteinase K protection assay; (10) a “no BV2” arm in the co-culture; (11) qPCR for the vector backbone and a check for residual reagent; (12) orthogonal verification with a JAK2 inhibitor and STAT3 siRNA, plus actual measurement of NF-κB. In a few months. (13) Add sham, unmodified exosome, and knockdown groups in vivo, in both sexes, and take the BBB from day 1 once a week out to 8 weeks. (14) Reproduce the main results in primary microglia with DAMP stimulation, and see whether the effect disappears in mice whose microglia have been eliminated with a CSF1R inhibitor.
In the limitations section the authors correctly write that “additional validation is needed.” But the only concrete measure they name is a dual-luciferase reporter assay—that and nothing else. What the assay is designed to measure is not written, and as for “whether it was inside the vesicle,” “whether it reached the recipient,” or “whether it moved JAK2”—how any of these would be confirmed cannot be read off. The one additional experiment named by name was a single line whose content is unclear—that one point, I think, symbolizes the whole of this paper’s mechanistic analysis.
All in all, every good piece of design stops one step short of the end. They compared three sources but did not write the dose; the two “two independent” siRNAs matched on the sense strand over 20 bases. They cast a proteasome subunit as the protagonist and never once measured the proteasome.
Dr. Exotaro’s Perspective
This paper connects straight to my own research theme. I have continued to work on applying small extracellular vesicles (sEVs) derived from bone marrow mesenchymal stem cells (MSCs) to the treatment of spinal cord injury. In a study I reported earlier—experiments in a rodent model of severe spinal cord injury—the MSCs administered intravenously never reached the injury site themselves, yet the sEVs those MSCs released inside the body did migrate to the injury site and associated specifically with the M2 macrophages there (Nakazaki, M., et al., Journal of Extracellular Vesicles 2021;10(11):e12137). When the sEVs were administered in doses divided over 3 days, TGF-β expression within the M2 macrophages rose, permeability of the blood–spinal cord barrier fell, and motor function recovery equivalent to direct administration of MSCs was obtained.
Let me write three things honestly. First, I have written for a long time that myeloid cells after spinal cord injury do not sort into two boxes labeled M1 and M2 but move along a continuous spectrum, and that steering that movement is the key to treatment. This paper’s two-parameter gate is an operational definition that draws an artificial line across a continuum, and the moment the line is drawn, a story is born in which “the cells swapped sides, left and right.” What snagged me most was the title asserting “Microglial” outright. In my own paper, too, the breakdown is not perfectly separated out, so this criticism comes back to me first.
Second, something inconvenient for me as well. What we used was a natural MSC-sEV with nothing added to it. Given that unmodified vesicles stabilized the barrier and brought about functional recovery, this paper’s story of attributing the effect to a single cargo protein sits a little awkwardly. And the control in the animal experiments was not unmodified vesicles but vesicles derived from the negative control vector, and whether the two groups differ on the BBB locomotor score cannot be read off the main text.
Third, the point that stings me most. I have criticized papers on extracellular vesicles for not counting the dose. But I admit that I too have stood before the same wall. It is a demand on others and at the same time homework for myself.
Even so, this paper has left something behind. The weight of the word “indirectly.” What became visible by separating the upper and lower chambers is the possibility that the real substance of firefighting in the acute phase may not be throwing water on the flames but changing the judgment of the caretaker on the scene. Even with the proof only half done, the question was framed correctly. Which is exactly why, knowing it is impertinent of me, I have one request. I wish they had used just one proteasome inhibitor. It is the most straightforward single move for seeing whether function as a proteasome is involved in the effect. That said, one inhibitor’s result does not settle a mechanism. Whichever way the result came out, this paper would have become more interesting. This is, of course, a preclinical study, and one week in mice does not demonstrate efficacy in humans. Even so: cells harvested from teeth that were going to be thrown away load a drab part of the proteasome onto a vesicle and rewrite the mood of the spinal cord’s caretaker—the picture is honestly beautiful. For the day when what was going to be discarded moves somebody’s legs, I will keep at the research today as well.
Note: this article was written with the assistance of AI.