Cancer drugs have long carried a dilemma: make them stronger and they work better, but the patient is injured along with the tumor. Conventional cytotoxic agents do not distinguish cancer cells from normal ones. Over the past decade or so, one technology has begun to offer a practical answer to that dilemma: the antibody-drug conjugate (ADC). A cytotoxic drug is tied to an antibody — an address label — and delivered by following that address, which reduces exposure to everything other than the target. One of the papers this review cites calls it a “biological missile.”
Meanwhile, our bodies already possess a system for delivering cargo from cell to cell: the exosome — a small vesicle 30–150 nm in diameter, secreted by nearly every cell type.
The review we read today asks whether these two can be fused: load the artificial missile onto a transport ship the body built itself. The target is hepatocellular carcinoma (HCC). The authors call the hybrid an “immuno-exosome.”
Yet what remains after reading is less excitement than one quiet fact. This is a blueprint, not a report on a finished machine. And several boxes on that blueprint have been left blank. The name of the antigen the antibody should target. The names of the drugs it would be compared against. The number of molecules that can be loaded onto a single vesicle. And the integration with diagnostic imaging promised by the abstract appears nowhere in the body of the paper.
And here is the point I most want to make in this article: the authors themselves correctly acknowledge the center of that blank space. Their conclusion states that the claim that immuno-exosomes will decisively improve the clinical management of HCC patients remains speculative at this stage. It is unusual for authors to write this way about the technology they are advocating. In this article we will look at both the value of the blueprint and the boxes left blank.
Journal Information
- Paper title: Understanding the Potential of Antibody-Drug Conjugates Functionalized Engineered Exosomes in Hepatocellular Carcinoma Therapy: A Comprehensive Narrative Review
- Article type: A narrative review. It contains no experiments and no primary data of the authors’ own. The authors state explicitly in their methods section that it is not a systematic review either, and give their reasons. Two figures, one table, 98 references
- Authors: Subham Sarkar (first author) and five others, six in all. There are two corresponding authors: Ajoy Kumer and Bikram Dhara
- Affiliations: Department of Biotechnology and Department of Microbiology, St. Xavier’s University (Kolkata, India); Department of Chemistry, IUBAT–International University of Business Agriculture and Technology (Dhaka, Bangladesh); Department of Microbiology, Saveetha Medical College and Hospital (Chennai, India)
- Journal: Health Science Reports (Wiley), 2026, Volume 9, Issue 10, article number e73289
- DOI / link: 10.1002/hsr2.73289 (PMID 42799406)
- Peer review and publication dates: Received 3 October 2025 → revised manuscript received 2 September 2026 → accepted 16 September. That is 348 days from receipt to acceptance, but the breakdown is 334 days until the revision came back, and 14 days from receipt of the revision to acceptance
- Open access: Yes (CC BY 4.0). A fully open-access journal; the article processing charge is up to USD 3,440
- Impact Factor: 2.9. ⚠️ However, this figure comes from Wiley’s own product page, and neither the JCR edition, nor the category, nor the quartile is given alongside it (the Clarivate original has not been checked). The Q2 one often sees is the Scimago quartile derived from Scopus, not a JCR ranking
- ⚠️ Where the journal sits: This journal is not indexed in MEDLINE. It can be read on PubMed only because its contents are deposited in PMC (confirmed two ways: the NLM Catalog and a PubMed subset search). Its output went from 815 articles in 2024 to 1,344 in 2025 (+65%), and by September 2026 it had already published 1,564, exceeding the previous full year’s total. Retractions number 16 of the 5,330 articles indexed in PubMed, or roughly 0.30%; most were a batch of systematic reviews from a single author group retracted in the summer of 2023, but there is also one each in August and November 2025. This paper itself is not among them, and the journal does not appear on the so-called predatory lists
- Funding: “Nothing to report” / Conflict of interest: “The authors declare no conflict of interest”
- Use of generative AI: Disclosed. Grammarly was used for English-language polishing, and the paper states explicitly that no generative AI tool was used for drafting, data analysis, or content generation
About the two corresponding authors. On this site we profile the corresponding authors every time. This time, that profile turned out to bear on the content of the article. What follows is a mechanical tally of ORCID-registered output and of a bibliographic database (OpenAlex), information not found in the paper itself. ⚠️ What does not appear in these records is not proof that it was never done — registration depends on self-reporting, and completeness is not guaranteed. Please read on with that caveat in place.
Ajoy Kumer (ORCID 0000-0001-5136-6166) is an assistant professor in the Department of Chemistry at IUBAT (the paper’s transparency statement writes “Prof.”, but his title on the university’s departmental page is Assistant Professor). His field is computational chemistry and in silico drug design. Counting by title the 169 papers registered in ORCID gives 34 on density functional theory (DFT), 23 on molecular docking, and 16 on molecular dynamics; even the roughly 27 whose titles contain “cancer” are mostly computational studies of target proteins. Of those 169 papers, only two — including this one — have exosomes in the title, and the other one is also a review (counted by querying the ORCID API directly while writing this article).
Bikram Dhara (ORCID 0000-0002-5433-1994) has registered a professorship at Saveetha Institute of Medical and Technical Sciences in ORCID, but the institution’s own staff page could not be retrieved, so the title rests solely on self-reported registration data. The backbone of his experimental primary research is environmental microbiology, plant pathology, and mycology — the first report of a halophilic pathogen causing leaf blight in cucumber, an evaluation of silver nanoparticles synthesized from oyster mushroom extract, and the like. By contrast, all six of his papers with exosomes in the title take the form of reviews or opinion pieces (sweat-derived exosomes, plant-derived exosomes, a narrative review of the gut-brain axis, and so on).
For neither author could primary research on exosome isolation or characterization, on antibody-drug conjugates, or on hepatocellular carcinoma be found in these records. The first author, Subham Sarkar, likewise has 28 papers registered in ORCID, all of them from 2024–2026 and mostly reviews (all four of the self-citations in this review are reviews as well).
This information needs to be handled with care. I have laid out the character of their output not in order to pass judgment on the authors’ ability. A review that organizes concepts has a value all its own, and it can be written — and is worth writing — by researchers who do not run a laboratory. Still, the observation this article is about to make, that the review contains no report of a prototype, is consistent with the fact that no primary extracellular vesicle research can be found in the writing team’s published record. Whether the blueprint in front of us was drawn from experience in design or from experience on the construction site — as a reader, I think that is worth knowing.
What Was It That We Did Not Know?
Hepatocellular carcinoma (HCC) arises from the liver’s own cells and is the most common form of primary liver cancer. The risk factors the review lists are chronic hepatitis B and C virus infection, non-alcoholic fatty liver disease (NAFLD), and cirrhosis. What makes it troublesome is that its therapeutic difficulty comes in four layers.
(1) It is found late. Early symptoms are vague or absent, and because it usually arises in people who already have cirrhosis or similar disease, the pre-existing abnormality masks the shadow of the new tumor. The curative window of resection or liver transplantation closes before it opens. (2) It comes back even when it has been fully removed. The review cites a post-resection recurrence rate of “50–70% within five years.” The reasons are micrometastases invisible at the time of surgery, and a cirrhotic liver that keeps producing the next tumor. (3) Drugs work poorly. Multidrug-resistance proteins, efflux pumps, and altered apoptosis (programmed cell death) pathways compound one another, and conventional systemic chemotherapy is largely ineffective in advanced disease. (4) The tumor is not uniform. Cells differ not only between patients but within the same tumor. It is hard to imagine a single drug knocking all of them down.
👦 Student: If the drugs do not work, why not just make stronger ones?
🧬 Dr. Exotaro: Make them stronger and the patient goes down first. Conventional chemotherapy is brutal because it cannot tell cancer cells from normal ones. Raise the potency and you raise the toxicity. So the thinking changed: instead of raising the drug’s power, raise the precision of its delivery. A drug of the same potency, if it enters only cancer cells, is in practical terms a far stronger drug.
The flagship example of this precision of delivery is the antibody-drug conjugate (ADC). Its structure has three parts: an antibody that grips a particular antigen on the cancer cell surface, a linker designed to be cleaved inside the cell, and the cytotoxic drug (the payload) hanging off the end of that linker. When the antibody grips its target, the whole conjugate is swallowed into the cell, the linker is cleaved in the acidic, protease-rich late endosome and lysosome, and the drug is released into the cytoplasm. The aim is not to hit only the target but to reduce exposure to everything other than the target. One of the papers the review cites calls this a “biological missile.”
The other protagonist is the exosome. It is a small vesicle 30–150 nm in diameter, secreted by nearly every cell, and it arises from an intracellular compartment called the endosome. Inside are proteins, lipids, and nucleic acids such as microRNA (miRNA), and it serves as a messenger from cell to cell.
What matters here is the review’s point that the exosome is a double-edged sword. Engineered exosomes can become therapeutic carriers. But the exosomes that cancer cells release naturally do exactly the opposite. The review calls these cancer cell-derived exosomes (CDEs) and lists six roles for them in HCC. (1) They convert fibroblasts into cancer-associated fibroblasts (CAFs), which then support tumor growth and invasion. (2) They deliver VEGF and FGF to endothelial cells, drawing in new blood vessels (angiogenesis). (3) They hand miR-21 and similar molecules to neighboring tumor cells, inducing epithelial-mesenchymal transition (EMT, the process of shedding epithelial character and becoming more mobile) and raising metastatic capacity. (4) They travel ahead through the bloodstream to lung and bone, preparing soil in which cancer cells can later settle (formation of the pre-metastatic niche). (5) They hand the immunosuppressive molecule PD-L1 to T lymphocytes and NK cells, disabling them. (6) They hand over drug efflux pumps and miRNAs, transferring drug resistance itself horizontally.
👦 Student: The same exosome can be an ally or an enemy?
🧬 Dr. Exotaro: Because the vesicle is nothing more than a container. Picture a delivery truck. The truck itself is neither good nor evil. Carry medicine and it becomes treatment; carry contraband and it becomes a crime. The exosomes that cancer cells release are the cancer’s own logistics network, driving around loaded with cargo that suits the cancer. So researchers thought: what if that network could be hijacked? Empty the cargo bed, load our drug, and rewrite the delivery slip. That is the idea behind the engineered exosome.
And this field has a predecessor: the immunoliposome — an artificially made lipid vesicle decorated with antibodies on its surface and packed with drug inside. The review positions the success of immunoliposomes as having established the paradigm of functionalizing a biocompatible nanocarrier with an antibody in order to deliver a cytotoxic payload with precision. If so, the authors say, the next move is obvious: do the same thing with a natural vesicle instead of an artificial one.
That is the question of this review. Can exosomes functionalized with antibody-drug conjugates — what the authors call the “immuno-exosome” — serve as a therapeutic carrier for hepatocellular carcinoma?
What Did This Paper Find?
First, let us set the premise precisely. This is a review, and it contains not a single experiment performed by the authors themselves. They state this explicitly in their methods section and even give their reason for choosing a narrative rather than a systematic review: they wanted to integrate conceptually a topic that spans several fields, and narrowing to the single question a systematic review demands would have constrained them. The three databases searched (PubMed, Web of Science, Scopus), the search period (1991–2026), the actual search strings, and the inclusion and exclusion criteria are all disclosed. This deserves credit.
How Far Does the Natural Vesicle Fill In the Artificial One’s Weaknesses?
The core of the review is Table 1, which compares immunoliposomes and immuno-exosomes along seven axes and is also the most information-dense part of the paper.
The immunoliposome’s strengths are essentially those of a finished industrial product. Drug loading efficiency is high; PEGylation (coating the surface with polyethylene glycol) can extend the circulating half-life beyond 30 hours; manufacture under GMP (good manufacturing practice) has already scaled and is low-cost; and because the lipid ratios can be controlled precisely, batch-to-batch variability is small. And for liposomal formulations the regulatory path at FDA and EMA is established. The weaknesses: being synthetic, repeated dosing triggers accelerated clearance from the blood (accelerated blood clearance, ABC); they are taken up by phagocytes in the healthy liver and spleen; and their rigid structure keeps them from penetrating deep into a fibrotic tumor such as HCC. Reaching the tumor relies mainly on the EPR effect (macromolecules leaking out of and accumulating in the tumor’s coarse vasculature) — a passive mechanism.
The immuno-exosome is drawn as the mirror image. Being derived from human cells, it has high biocompatibility and low immunogenicity (Table 1 states that “homologous cell sources minimize immune rejection and complement activation”). It is less rapidly cleared by the reticuloendothelial system and has high penetration into the tumor — the mechanisms cited are fusion with biological membranes and cell-to-cell transport (though the table does not show how far this reduces dependence on the EPR effect). It also possesses natural tropism (a tendency to home to particular tissues). The review, however, lists the disadvantages honestly as well. Loading efficiency is low to moderate (endogenous cargo already fills the vesicle, leaving limited lumenal space). The natural half-life is short, minutes to hours (the table does note that surface engineering such as CD47 expression can extend it, and that it is not accompanied by the ABC phenomenon). Manufacture requires complex cell culture, isolation is difficult, stringent quality control is needed, and costs are high. Because output depends on the metabolic state of the producing cells, batch-to-batch variability is also large. And the regulatory path is not yet established.
👦 Student: They built a table of strengths and weaknesses, and wrote in the weaknesses of the side they are advocating too.
🧬 Dr. Exotaro: That deserves plain credit. A paper proposing a new technology tends to build its comparison table in its own favor. This table does the opposite: on five items — loading efficiency, half-life, manufacturability, batch reproducibility, and regulation — it says the immunoliposome is ahead. The exosome side wins on two: low immunogenicity and tumor penetration. The table’s own caption concedes that immuno-exosomes face regulatory and translational challenges and have inferior loading efficiency. It is a table that spells out the points against its own case. One caveat, though: this is a matter of honesty, not of quantitative comparison. There is not a single number in the table showing by what percentage either side is superior. And, as we will come to later, one of the table’s citations is in a state that needs checking.
The Blueprint — How Would It Be Assembled?
The manufacturing process the review describes runs as follows.
First, high-purity exosomes are isolated from a parent cell (the example given is genetically modified mesenchymal stem cells) by methods such as size-exclusion chromatography or immunoaffinity capture. In parallel, the ADC is synthesized: a cytotoxic drug or a PROTAC (proteolysis targeting chimera — a molecule that bridges a target protein to the cell’s own degradation machinery and has that protein destroyed) is attached to an antibody through a cleavable linker. The ideal attachment site is the constant region of the antibody. The linker is designed to be stable in the bloodstream and readily cleaved by intracellular enzymes (lysosomal proteases and the like) or stimuli (pH changes in the tumor microenvironment or the endosome).
That ADC is then conjugated to the exosome surface. Here the review singles out click chemistry as especially promising. The argument is that using bio-orthogonal reactions — that is, reactions that do not interfere with other reactions inside the body — allows the ADC to be attached precisely and efficiently “without disrupting its structural integrity or functionality.”
The behavior after administration is described too. The immuno-exosome circulates, and its antibody moiety recognizes and binds a cognate receptor that is overexpressed on the surface of HCC cells. The whole complex is then internalized by receptor-mediated endocytosis and carried to the late endosome and lysosome. The acidic environment and the high protease concentration cleave the linker, and the freed drug or PROTAC is released into the cytoplasm.
What happens beyond that is laid out concretely in the caption to Figure 2, which divides the action into three compartments — plasma membrane, cytoplasm, and nucleus — and names real drugs. Vinca alkaloids that depolymerize microtubules, HSP90 inhibitors that block chaperones, HDAC inhibitors that disturb chromatin, XPO1 inhibitors that halt nuclear export of mRNA — and, in the case of a PROTAC, recruiting the cell’s own E3 ubiquitin ligase to the target protein and having it degraded by the proteasome.
The Diagnostic Side — Reading the Tumor from Blood
The review’s other pillar is diagnosis. Exosomes released by HCC cells carry cargo that mirrors the genetic and phenotypic state of the parent cell. Capture them from the blood and the contents of the tumor can be read — the idea of the liquid biopsy.
The candidates the review lists are concrete. Among proteins: GPC3 (glypican-3) and AFP (alpha-fetoprotein), the known HCC tumor markers, plus ADAM10 and CD147. Among miRNAs: miR-21, miR-122, miR-155. Among lncRNAs: H19 and HOTAIR. And oncogenic mutations such as mutations in the TP53 gene may be detectable as well. The name GPC3 will come up once more later. That moment is the climax of this article.
👦 Student: If a blood draw alone reveals the tumor’s genes, surely biopsies are no longer needed?
🧬 Dr. Exotaro: That is more than can be said. What the review lists are candidates, and for none of them does the text give a single number of the form “this assay achieved this sensitivity and this specificity.” Between a list of candidate molecules and an assay that can actually be used lies a long process called validation. Having one’s name on the list and being usable are two different things.
How Will the Future Change? (The Road to the Clinic)
The Five Walls the Review Itself Lists
Here the review becomes quite candid about its own proposal. Section 5, “Technical and Manufacturing Challenges of Exosome-ADC Therapeutics,” is the most practical part of the paper.
The loading-efficiency wall. Unlike synthetic lipid nanoparticles, exosomes already contain endogenous cargo, leaving limited lumenal space. As a result, the review writes that the efficiency of loading ADCs into or onto exosomes is significantly lower than the high loading efficiency achieved with synthetic liposomes (the original English is “significantly lower” — plain English for well below, not a claim of statistical significance).
The post-conjugation stability wall. Surface engineering and functionalization techniques such as click chemistry carry the risk of altering the integrity of the exosome membrane. The consequences may be vesicle aggregation, premature drug leakage, or masking the natural homing ligands that are essential to tropism for the liver and other organs. This is an important observation: the very act of attaching something to the surface can destroy the properties that depended on that surface.
The manufacturing wall. There is no standardized high-yield isolation method, and scale-up and GMP-compliant manufacture lag behind. Ultracentrifugation is the preclinical standard but is unsuited to mass production, and moving to a combination of TFF (tangential flow filtration) and SEC (size-exclusion chromatography) requires rigorous optimization to avoid destroying the vesicles.
The quality-control wall. Because exosomes are cell-derived, their composition varies with the metabolic state of the producing cells. Quality-control metrics aligned with MISEV (the minimal reporting requirements for extracellular vesicle research) need to be established, the review says.
The regulatory wall. Because an exosome-ADC is a combination of three things — a biological carrier (the exosome), a targeting moiety (the antibody), and a cytotoxic payload (the anticancer drug) — it does not fit neatly into the existing FDA or EMA frameworks. Defining criteria for pharmacokinetics, toxicity, and mechanism of action remains a task to be completed before broad clinical trials can begin — these are the review’s own words.
The conclusion is quite cautious too. Quoting the original directly: “the claim that immuno-exosomes (ADC-functionalized) will decisively improve the clinical management of HCC patients remains speculative at this stage.” And: the clinical superiority of exosome-based ADC delivery over conventional drug delivery platforms must be treated as a working hypothesis rather than an established clinical reality.
👦 Student: That is a remarkably timid conclusion, for a technology they are advocating.
🧬 Dr. Exotaro: It is not timid; it is accurate. And that is this paper’s greatest virtue. Reviews all too often close on “promising” and “much anticipated.” Those words promise nothing, yet they land on the reader like a promise. This paper named it itself: speculative, a working hypothesis. It does not swap a preclinical story for a clinical one. Reading as a physician, I value that kind of honesty highly.
From Here On, This Is Outside the Review — How Far Has the Clinic Actually Come?
The review writes that clinical translation is the challenge, but it does not write what is on the other side of that bridge. Not a single clinical trial registration number (NCT number) appears in the text, and no specific trial is ever named. The phrase “clinical trials” appears exactly once, in the abstract’s general statement that the approach should be confirmed in future by sufficient experimentation and clinical trials. So, with everything below drawn from outside the review, let us establish where things actually stand.
First, the scale of the disease. Liver cancer (liver and intrahepatic bile duct, ICD-10 C22; hepatocellular carcinoma accounts for more than 90% of it) is diagnosed in 843,045 people worldwide each year and kills 732,489 — seventh in incidence and third in mortality (GLOBOCAN 2024 estimates, IARC). In Japan there were 32,673 incident cases (2023) and 22,465 deaths (2024), and the five-year relative survival after diagnosis is 35.2% (National Cancer Center cancer statistics, cases diagnosed in 2018).
Next, the baseline a new treatment has to clear. First-line therapy for unresectable advanced hepatocellular carcinoma has moved a great deal over the past five years. Atezolizumab plus bevacizumab gave a median overall survival of 19.2 months (versus 13.4 months for the sorafenib control, hazard ratio 0.66, 95% CI 0.52–0.85); these figures come from an updated analysis, however, and the p values are descriptive (Cheng AL, et al., Journal of Hepatology 2022;76(4):862–873). Tremelimumab plus durvalumab gave 16.43 months (sorafenib 13.77 months, hazard ratio 0.78, 96.02% CI 0.65–0.93, P=0.0035; Abou-Alfa GK, et al., NEJM Evidence 2022;1(8):EVIDoa2100070). And nivolumab plus ipilimumab gave 23.7 months (against investigator’s choice of lenvatinib or sorafenib at 20.6 months, hazard ratio 0.79, 95% CI 0.65–0.96, p=0.018; Yau T, et al., Lancet 2025;405(10492):1851–1864), and the FDA approved it as first-line therapy on 11 April 2025. ⚠️ Because the comparators, the eras, and the populations all differ, these numbers cannot be lined up side by side. Even so, this is the level into which a new treatment has to fit.
And what about antibody-drug conjugates themselves? As of 29 September 2026 there are 15 FDA-approved ADCs. Not one of them carries an indication for hepatocellular carcinoma (confirmed by mechanically searching the indications sections of all 15 package inserts).
👦 Student: So nobody is trying ADCs in liver cancer yet.
🧬 Dr. Exotaro: No — they are. And in precisely the place this review should most have mentioned.
The target the review never named — where GPC3 stands today. This review brings GPC3 (glypican-3) on stage only as a diagnostic biomarker. In reality, GPC3 is the most intensively pursued therapeutic target in hepatocellular carcinoma. The clinical trial registries hold 72 hepatocellular carcinoma trials involving GPC3 or glypican. There is history here as well: codrituzumab, a naked anti-GPC3 antibody with no drug attached, failed in a phase II trial. Median overall survival was 8.7 months (placebo 10 months, hazard ratio 0.96, p=0.82), with no difference in progression-free survival either (Abou-Alfa GK, et al., Journal of Hepatology 2016;65(2):289–295). That failure was published ten years before this review — and in the same journal.
Developments since have been rapid. GPC3-targeted CAR-T cell therapy was reported in Nature in 2026 (36 patients, response rate 44.4%, median progression-free survival 4.2 months, median overall survival 14.2 months; Zhang Q, et al., Nature 2026;657(8131):495–504). That paper identifies loss of the GPC3 antigen and an increase in TGFβ as mechanisms of resistance, a finding that bears on every design aimed at GPC3.
And here is the decisive part. Antibody-drug conjugates targeting GPC3 have already been given to patients. A phase I/II trial of an agent called MRG006A (registration number NCT07093970; the registry lists the population as advanced solid tumors, and it is still recruiting) reported, for a cohort of 26 patients with advanced hepatocellular carcinoma (25 evaluable), a response rate of 23.1% and a disease control rate of 68.0%. Restricted to the 12 patients with high GPC3 expression, it was a response rate of 33.3% and a median progression-free survival of 7.0 months (Zhao H, et al., Journal of Clinical Oncology 2026;44(16_suppl):3028). ⚠️ This is a conference abstract, however: median follow-up is still only 5.7 months, and grade 3 or higher treatment-related adverse events occurred in 42.3% (thrombocytopenia in 88.5%), so the safety questions are equally clear. It should be stressed that this is a stage at which one ought to wait for the peer-reviewed full paper. Beyond that, a second GPC3-targeted ADC (ZW251, NCT07164313) is also in phase I.
In other words: the very substance of what this review proposes — aim with an antibody and deliver the drug — is already being tested in humans without being loaded onto exosomes, and numbers are starting to come in. The review never once touches on this situation.
Finally, where the exosome side stands. This one is far harsher. There is still not a single FDA-approved exosome product anywhere in the world (the FDA’s official position, held since its safety notification of 6 December 2019; the page content is current as of 9 April 2024). Clinical trials administering EVs themselves for cancer therapy number 9 by my count. And —
There has been exactly one clinical trial of EV therapy in hepatocellular carcinoma, and that one was terminated. It was a phase I trial of exoASO-STAT6 (CDK-004, registration number NCT05375604), an antisense oligonucleotide targeting STAT6 loaded onto exosomes, in advanced hepatocellular carcinoma and liver metastases. It was run by Codiak BioSciences, started in June 2022, and ended in May 2023. A total of 9 patients were enrolled, the entry in the registry’s termination-reason field is the single phrase “Company Bankruptcy,” and no results have been published.
For reference, the EV therapy trial in oncology that got furthest and was actually completed — a phase II giving dendritic cell-derived exosomes as maintenance therapy in non-small cell lung cancer — failed to meet its primary endpoint (22 patients, median time to progression 2.2 months; Besse B, et al., Oncoimmunology 2015;5(4):e1071008).
👦 Student: …This is much earlier along than I had imagined.
🧬 Dr. Exotaro: It is. And I think this is the single most important correction to make when reading this review. Section 5 of the review honestly laid out the technical walls of manufacturing, quality control, and regulation. That was right. But the height of a wall cannot be judged without looking at the trials that are actually running. On the antibody side there are numbers in humans; on the exosome side, the one trial in this disease stopped because of a bankruptcy. The phrase “translation is the challenge” is the same for both halves, yet the two halves are standing in completely different places.
How to Read This Study Critically (Limitations, and Routes to Higher Quality)
Before the criticism, the good points first. (1) The authors declare it a narrative review and even give their reason for choosing that format. (2) They disclose the databases searched, the period, the search strings, and the inclusion and exclusion criteria (procedures often omitted in narrative reviews). (3) In Table 1 they honestly list five items on which the side they favor comes off worse. (4) In Section 5 they enumerate the manufacturing, quality-control, and regulatory walls concretely. (5) In the conclusion they say plainly “speculative” and “working hypothesis.” (6) They disclose their use of generative AI explicitly — Grammarly for English polishing, and no generative AI for drafting, data analysis, or content generation.
With that said, the limitations.
First, this is a blueprint, not a report on a prototype. Section 4 of the review describes the manufacturing process in detail, but the description is consistently an account of procedure — this is how it could be done — rather than a report of results, this is what happened when we did it. And among the 98 references this review cites, as far as the titles could be matched, no study that actually conjugated an antibody-drug conjugate to an exosome and tested it in hepatocellular carcinoma could be confirmed. The citations whose subject is ADCs are general reviews, and exosomes do not appear in their titles. Exosomes carrying antibodies do appear among the citations (a study of an exosome vaccine displaying a PD-L1 antibody), but that is display of an antibody, not the loading of an antibody with a drug attached to it. ⚠️ Note, however, that titles alone cannot establish what each study actually did inside — there is, for instance, a citation titled along the lines of anticancer drug delivery by tumor-targeting exosomes, and the details of its design cannot be read off the title. All that can be said here is that no title states it explicitly. And the study that the caption to Figure 2 names as the concrete example of delivering a PROTAC against hepatocellular carcinoma used a galactose-modified liposome, not an exosome. In other words, the worked example of the very paradigm this review proposes to extend with exosomes is still the unextended liposome.
👦 Student: Is it acceptable to publish a paper about something for which no validation results have been reported?
🧬 Dr. Exotaro: Yes. A paper that organizes concepts has a value all its own — the work of connecting scattered fields and making the next experiment visible. The problem is not what was written but the gap with the impression the reader takes away. Figure 2 depicts, with precision and with the names of real drugs, how the drug acts in three compartments: extracellular, cytoplasmic, and nuclear. That precision is the precision of a design, not the precision of a validation. Yet a figure reaches the viewer as something that has been confirmed.
Second, the integration with imaging that the abstract and conclusion hold up is never developed in the body of the paper. The abstract states that “Co-loading of imaging probes can enable integrated theragnostics,” helping to visualize biodistribution, target engagement, and therapeutic response (the original wording is “can enable” — a statement of possibility), and the conclusion likewise holds up the potential to enable real-time monitoring of therapeutic response. Yet reading the body of the paper through, neither imaging probes nor any imaging modality — MRI, PET, fluorescence, radiolabeling — is discussed even once. The word “imaging” appears in the body only inside the search strings and inclusion criteria of the methods section. Section 3 is titled “Exosomes as Theragnostic Agents,” but its contents are therapy (Section 3.1) and liquid biopsy, reading molecules out of blood (Section 3.2). In other words, what the body of the paper actually develops as the diagnostic half is blood testing, not imaging. A function held up in front and then not dealt with in the body — this is the kind of gap a reader is least likely to notice.
Third, the antigen that the ADC’s antibody targets is not specified. The linchpin of the immuno-exosome is the antibody, and an antibody’s value is determined by what it grips. It is not that the review names no targets at all — Section 3.1 names αvβ3 integrin as the target of exosomes modified with a peptide called iRGD. But for the ADC antibody proposed in Section 4, the text goes no further than the general expression “a cognate receptor that is overexpressed on the surface of HCC cells”; which antigen is to be targeted is not specified. GPC3 and AFP both appear, but in the context of diagnostic biomarkers, not as candidate ADC targets. Without a settled target no antibody can be chosen, and without an antibody this blueprint cannot break ground. And this blank, as the preceding section showed, looks larger still when held against how the clinic is actually progressing.
Fourth, no comparator is placed on the table. If a new treatment is proposed, the question is how much it adds over the best existing treatment. Yet not a single name of an approved drug for hepatocellular carcinoma appears in the body of this review. The word sorafenib occurs only inside the titles of cited references. The limitations of existing therapy are discussed qualitatively, but which drug, which trial, and which number those limitations are measured against is not written.
Fifth, there are almost no numbers. The figures printed anywhere in the body are the exosome diameter (30–150 nm), the recurrence rate within five years of resection (50–70%), and the half-life of PEGylated liposomes (over 30 hours) — essentially that is all. Not one p value, not one dose, not one n, not one effect size is printed. Loading efficiency is described as significantly lower, but what percentage against what percentage is never shown. How many ADC molecules can be loaded onto a single vesicle? — the most basic number determining whether this technology succeeds or fails is nowhere to be found.
Sixth, there is disorder in the management of the citations. Within the range I was able to cross-check, here are four confirmed instances. (1) Table 1 lists expression of CD47 as a means of extending the half-life of immuno-exosomes and attaches two references, one of which is a paper titled the role of CD147 in the development and diagnosis of hepatocellular carcinoma. CD47 and CD147 are different proteins. (2) A single author-year citation, “(Zhao et al., 2021),” is mixed into the text in one place. This journal uses numbered citations, and no entry that would resolve this notation could be found in the reference list. (3) Section 5 refers to the MISEV guidelines by name as the standard for quality control, yet the MISEV original is not among the 98 references. (4) Section 3.1 writes “immune checkpoint inhibitors such as anti-PD-1 and anti-CTLA-4 antibodies” and attaches two references, one of which is titled for an exosome vaccine displaying an anti-PD-L1 antibody. PD-1 and PD-L1 are different molecules in the same pathway. None of these is fabrication; they are management problems that could be fixed by cross-checking.
Seventh, the review’s procedure as a review has a hole that could be filled cheaply. The authors declare it a narrative review, and the fact that it is not a systematic review is not in itself a criticism. However, the number of records the search returned and the number of those that were adopted are not given. This requires no additional experiments. If the screening record still exists, the number of hits and the number retained could have been written.
Eighth, part of the text depends on a prior review by the same author. The routes of administration listed in Section 3.1 — microneedle patch, intravenous, intratumoral, and oral delivery via enteric-coated capsules — are all four attributed to one and the same reference, which is a prior review by the first author himself. Self-citation is in itself a legitimate practice, but the fact that the evidence for something as practically important as route of administration is concentrated in a single review rather than in primary research is information the reader should have.
Ninth, one sentence in the conclusion stands out for being insufficiently supported. It is the clause “While early translational data is highly encouraging.” This review never once presents translational data on the immuno-exosome it proposes. Precisely because the sentences around it correctly qualify things as remaining speculative at this stage and as something that must be treated as a working hypothesis, this clause floats free.
Look and you find other slack of the same kind. The abstract writes that the dual-targeting concept supports increased intratumoral delivery and reduced off-target exposure compared with conventional approaches, but no data actually measuring that comparison is presented. Section 4 writes that click chemistry enables conjugation “without disrupting its structural integrity or functionality,” while Section 5 writes that the same surface processing carries the risk of altering membrane integrity. The latter is the more cautious, and the correct, way to write it.
So how could the quality have been raised? Listed in order of least effort.
(1) Redo the cross-checking of every citation (no additional experiments needed). (2) State the number of search hits and the number adopted (if the screening record survives, this requires almost no extra work). (3) Name candidate target antigens for the ADC and build one comparison table — with a column for each candidate giving expression rate in HCC, expression in normal liver, whether an antibody already exists, and, most important, whether it is absent from normal hepatocytes. Since the organ being targeted is the liver, this technology does not stand up unless a margin against normal liver can be obtained. (4) Put in a comparison table against existing therapy and define a target product profile — write in numbers what has to be exceeded, and by how much, for the result to be clinically meaningful. (5) Estimate the stoichiometry on the back of an envelope — how many antibody molecules fit on the surface area of one vesicle, how many particles are needed per dose, and whether that quantity can be made with current manufacturing capacity. The ADC world has a specification item called the drug-to-antibody ratio (DAR), but what the corresponding specification would be once it is loaded onto an exosome is not written in the review. (6) Attach the reference type (primary research or review) to each row of Table 1. (7) Reconcile the abstract with the body — either devote a section of the body to integration with imaging, or, failing that, drop it from the abstract and conclusion. Either is fine, but what is held up in the abstract should be dealt with in the body. (8) Write the skeleton of a phase I trial in one paragraph — which stage of disease, which patients, and what is measured as the primary endpoint.
If the criticism has to be put in one sentence, it is this. This paper is an excellent blueprint, and an honest one. What is missing is the first pillar showing that the blueprint can actually be built.
Dr. Exotaro’s Perspective
To be honest, hepatocellular carcinoma is not my own field. What I have worked on for a long time is how to use the small extracellular vesicles (sEVs) released by mesenchymal stem cells (MSCs) in neurological disease such as spinal cord injury (SCI). I have never been involved in cancer pharmacotherapy, nor in the design of antibody-drug conjugates. Even so, there are three points I could not read as someone else’s problem.
First, routes of administration are enumerated, while dose, number of doses, and dosing interval are not examined. In our own work, intravenously injected MSCs themselves did not reach the injury site, yet the sEVs the MSCs released did migrate there and associated specifically with M2 macrophages. And — even at the same total dose, a single large bolus did not work; only when it was divided across three days was motor recovery equivalent to that of the MSC cells themselves obtained (Nakazaki, M., et al., Journal of Extracellular Vesicles 2021;10(11):e12137; a rodent model of severe spinal cord injury). What determined the outcome was not how much was given but in what temporal shape the exposure was delivered. This review does enumerate routes of administration, but it does not write a single line on dose, number of doses, or dosing interval. For a carrier whose half-life the authors themselves describe as minutes to hours, that blank is not a small one.
Second, the question of what and how many are in a single dose. This is a criticism of myself as well. Among papers on extracellular vesicles, very few — mine included — can answer both the molecular copy number per vesicle and the particle count per dose. Yet in the ADC world, the drug-to-antibody ratio (DAR) is a specification item that is controlled. How many drug molecules are attached on average to one antibody molecule is what determines both efficacy and toxicity. Loading an ADC onto an exosome means that a further ratio is added on top: how many antibody molecules per vesicle. The review never once asks for this number. Without a definable potency specification, no product can be made. I suspect that the first wall this technology has to clear is neither chemistry nor biology, but measurement.
Third, the irony of the liver as an organ. That intravenously injected nanoparticles accumulate in the liver and spleen has been treated as a problem in this field for many years: the intended destination is elsewhere, yet the liver takes delivery on its own. But in hepatocellular carcinoma that troublesome property may switch sides — because the destination is that very liver. The review writes about this from a somewhat different angle: exosomes are less prone to nonspecific uptake by the liver’s resident macrophages (Kupffer cells), so tumor-specific delivery improves. Both points matter, and they are different points. Going to the liver, entering the tumor within the liver, and not being eaten by the liver’s macrophages are three separate problems. I would like to read a review that discusses these three separately and puts a number on each.
Let me draw a line as well. What I have worked on is the injured central nervous system, exploiting a state in which the blood-brain and blood-spinal cord barriers are broken and the door is open. In liver tumors, where the barrier situation is entirely different, the meaning of the word “reaches” changes. And I have never once conjugated an ADC to an exosome, nor modified a vesicle surface by click chemistry. On the actual difficulty of this technology, I stand in the same place as the reader.
Even so, I will not end on disparagement. Table 1 was the first comparison table I have seen in this field in a long while that contains information unfavorable to its own side. And the self-limitation in the conclusion — that the claim remains speculative at this stage, and that it must be treated as a working hypothesis rather than an established clinical reality — shows that the authors consciously refused the temptation to swap a preclinical story for a clinical one. The value of the blueprint stands on that honesty. What is needed next is one unglamorous, decisive paper: name a single target antigen, measure the margin against normal hepatocytes, count how many molecules were loaded per vesicle, and line up the same number of unmodified particles for comparison. This review acknowledges, in its own words, that the clinical superiority of the immuno-exosome is at this stage no more than a working hypothesis. A blueprint written with that acknowledged is far more trustworthy than a success story written without it.
Note: this article was written with the assistance of AI.