Giving a drug-carrying lipid nanoparticle (LNP) an address label: don't stick it on the surface, insert it afterwards. Before or after dialysis, the particle's fate changed

Lipid nanoparticles (LNPs) became household objects with the COVID-19 mRNA vaccines. They are couriers: they wrap fragile RNA in a bead of lipid and carry it inside the cell. A vaccine only has to go into a muscle, and it does its job if it works there and in the nearby lymph nodes, so “where does it end up?” was never much of a problem.

The moment you want to hit one particular cell type, though, everything changes. Put a particle into the bloodstream and, left alone, it collects in the liver. If you want to name a recipient — leukemia cells only, neurons only — you have to put an address label on the particle’s surface: a ligand or an antibody that binds the target.

The paper we are looking at today takes on nothing but how you attach that label. It may sound like a modest question, but there is no point labeling a parcel if the labeling crushes the parcel. That is exactly what this paper measured.

Two caveats up front. First, this is a paper from 2022. I am covering it now because it is the foundation under the same group’s 2023 animal work (more on that below), and because the method it uses runs straight into the surface engineering of exosomes. Second, this is a study in test tubes and cultured cells. It is not a treatment that can be given to patients.

Publication details

• Paper title: A robust post-insertion method for the preparation of targeted siRNA LNPs
• Type of paper: original research article (a methodology study in formulation and drug delivery)
• Journal: International Journal of Pharmaceutics, volume 620, article number 121741 (received 1 February 2022, accepted 8 April, published online 11 April)
• DOI / link: 10.1016/j.ijpharm.2022.121741
• License: open access (CC BY 4.0). Anyone can read the full text
• Impact Factor: roughly 5 or thereabouts (it moves year to year; in formulation science and drug delivery this is a long-established specialist journal)
• Authors: seven in all — L.E. Swart (first author), C.A. Koekman, C.W. Seinen, H. Issa, M. Rasouli, R.M. Schiffelers and O. Heidenreich (corresponding author)
• Affiliations: Princess Maxima Center for Pediatric Oncology (Utrecht, the Netherlands); the Department of Clinical Chemistry and Haematology, University Medical Center Utrecht (UMC Utrecht); the Wolfson Childhood Cancer Research Centre, Newcastle University (UK); and the Department of Paediatrics, University Hospital Frankfurt (Germany)
• Funding: a programme grant from KiKa, the Dutch charity that supports childhood cancer research (grant 329)
• Conflict-of-interest disclosure: the corresponding author, Heidenreich, has a patent application pending on “Targeted lipid nanoparticle formulations” (in the paper’s own wording, filed as assigned to the Princess Maxima Center)

What kind of researcher is the corresponding author?

The corresponding author, Olaf Heidenreich, leads a research group at the Princess Maxima Center and is also professor of molecular haematology and theme leader for childhood cancer research at Newcastle University in the UK. His work has had one consistent axis: how do you hit the “fusion genes” that form when chromosomes swap pieces in leukemia. Above all he has spent more than 20 years on RUNX1/ETO, the product of the t(8;21) translocation (also written RUNX1/RUNX1T1, or AML1/ETO). In 2003 he published a paper in Blood showing that silencing AML1/MTG8 — that is, RUNX1/ETO — with siRNA pushes t(8;21)-positive leukemia cells toward myeloid differentiation.

A fusion gene is an ideal target, because it exists only in the cancer cell, and at the same time a hard one for conventional small molecules (a protein made of two fused transcription factors is generally held to have no deep pocket for a drug to wedge into). That is what makes siRNA, which aims by sequence, so attractive. And siRNA needs a courier to get there. This paper is the work of getting that courier right.

Co-author Raymond M. Schiffelers, who contributed conceptualization and supervision here, is professor of nanomedicine at UMC Utrecht. He is the key to today’s story, so let me introduce him at a little more length. His doctoral work (Erasmus University Rotterdam, 1996–2001) was on liposomal targeting of antibiotics: in other words, he was raised on liposomes. In 2010 he won an ERC grant for a microvesicle-inspired drug delivery system, and he is a founding member of the International Society for Extracellular Vesicles (ISEV) as well as an associate editor of the Journal of Extracellular Vesicles. So someone at the very centre of the extracellular vesicle (EV) world was in on the conception of a formulation paper about LNPs, which are synthetic particles. One disclaimer, though: the paper itself never mentions extracellular vesicles once. What it compares is liposomes and LNPs. Even so, the fact that the author list straddles all three of these particles leads into something I come back to in the second half.

First author Laura E. Swart established the methods, ran the experiments, did the analysis and wrote the manuscript; she is also on the 2023 follow-up described below.

What was it that nobody knew?

Liposomes and LNPs look alike, but their insides are not the same

The old hand among drug-carrying nanoparticles is the liposome. It is a bag made of the same lipid bilayer as a cell membrane, with water inside. It has been in clinical use since the 1990s.

An LNP that carries siRNA is built differently. According to the paper, the interior of an LNP is an essentially hydrophobic core packed with inverted lipid micelles, and it is those micelles that enclose the nucleic acid. Not a bag of water, but a solid, filled bead. And indeed, in this paper’s electron micrographs the LNPs show up as spheres whose insides are dark and dense.

👦 Student: If they are both “lipid particles”, surely you can treat them the same way?

🧬 Dr. Exotaro: That is exactly the trap. Picture a liposome as a balloon and an LNP as a dumpling stuffed with bean paste. Pushing an ornament through the skin of a balloon and pushing one into the surface of a dumpling are two completely different things, both in how it goes in and in how the thing falls apart afterwards.

Labeling a liposome is a settled craft. Give the ligand a lipid tail and the tail buries itself in the bilayer of its own accord. An LNP, however, has no bilayer for that tail to bury itself in. That is where the paper starts.

There is one more constraint to respect: the particle’s hydrodynamic diameter. Citing earlier work, the paper calls 40–150 nm the “Goldilocks zone”. Below that the kidneys filter the particle out; too far above it and the particle cannot get into tissue. Size is not a secondary readout — it decides whether the thing is usable at all.

Which of the two approaches is better?

The paper compares these two (Figure 1 of the paper).

1. Direct surface modification — make the particle first, then attach the ligand directly to the reactive groups displayed on its surface (Figure 1A)
2. post-insertion, that is, inserting afterwards — join the ligand and the lipid in advance to make a single component, then push that component into the finished particle (Figure 1B)

For the joining chemistry they chose a click reaction, and specifically a copper-free one. Its formal name is strain-promoted azide–alkyne cycloaddition (SPAAC).

👦 Student: Why avoid copper?

🧬 Dr. Exotaro: Copper(I) is an excellent catalyst, but it is poison to living things. So instead you use an alkyne deliberately bent into a strained ring — DBCO. The strain stores energy, and that energy lets it react with an azide all on its own, with no catalyst. Press the fastener home and it snaps shut — that snap is where the name “click reaction” comes from.

And the last variable is timing. After an LNP is made, there is a step in which the ethanol is removed by dialysis. Dialysis does more than strip the solvent: it neutralizes the surface charge and lets the lipids settle into a stable arrangement, “ripening” the particle. Do you attach the address label before this dialysis, or after? Nobody had looked at that systematically.

What did this paper find?

How it is made — materials and procedure

The base LNP is made by dissolving DLin-MC3-DMA (an ionizable lipid), DSPC, cholesterol, DMG-PEG(2000) and DSPE-PEG(2000)-DBCO in ethanol at a molar ratio of 50:10:38.5:1.2:0.3. For post-insertion runs that ratio was changed to 50:10:38.5:1.5:0.1, to leave room for the insert.

The siRNA is diluted in 25 mM acetate (pH 4) to give an N/P ratio 4 and mixed in a microfluidic device, the NanoAssemblr Benchtop, at a total flow rate of 4 ml/min and a lipid:siRNA flow-rate ratio of 1:3. The product is then dialyzed against PBS through a 10 kDa cut-off membrane overnight at 4 °C, which removes the ethanol and neutralizes the particle.

There are two cargo siRNAs. siRE is the real thing, aimed at the RUNX1/ETO fusion junction; siMM is a mismatch control with two bases swapped. Using a control that differs from the active sequence by no more than a nudge is the basic design that lets you show the sequence did the work, rather than merely that the particle got in. The address-label component itself is built at a molar ratio of DBCO:azide = 3:1, overnight at room temperature. In practice the finished components were held at 4 °C for up to 7 days before use — which is not the same as having validated a shelf life.

① Couple directly on the surface, and the particle fell apart

This is the clearest result in the paper. Four conditions were compared both by size, using dynamic light scattering (DLS), and by appearance, using cryo-electron microscopy (cryoEM).

(mean ± standard deviation, n = 3. PDI = polydispersity index; the closer to 0, the more uniform the particles.)

The cryoEM images explain what those numbers mean. In A the dense spheres line up neatly, and in B the core structure was still intact. But in C — the one coupled directly on the surface — the spherical shape has broken down, micelle-like fragments appear, and there are even large multilamellar particles that look like several LNPs fused together (the white circles in Figure 2C of the paper). D, by contrast, kept both its spherical shape and its dense core.

👦 Student: Already at B it is 110 nm. Nothing has been added but DBCO, so why does it swell?

🧬 Dr. Exotaro: The paper’s guess runs like this. When the DBCO has not been joined to an azide, that group ends up facing into the particle. Earlier work reported PEG burrowing into the interior of a particle during self-assembly, and this is presumably the same phenomenon. Which means you are trying to react, from the outside, with groups that point inward. That is why “assemble first, then insert” is the more natural order — although the paper did not actually measure the mechanism beyond that point.

The pattern did not depend on which ligand or which cargo was used. Supplementary data in which the ligand was swapped for a VHH (a nanobody) and the cargo for a DNA barcode behaved the same way, with post-insertion holding the size increase below 50%.

② Before or after dialysis changed the size

Next they varied the timing. Using LNPs labeled with the fluorescent dye Cy3, they prepared two series differing in PEG content (DMG-PEG 1.5% and 1.2%).

(mean ± standard deviation, n = 3. The p values are from a two-tailed Student’s t-test. The paper does not state which comparison each p value refers to, but since the “after dialysis” rows are blank, they read as comparisons against that reference.)

The trend is unambiguous. Add it before dialysis and you end up well over 100 nm; insert it after dialysis and the particle stays smallest of all. The paper sums this up by saying that the point at which the solvent is removed was decisive.

PEG content made a difference too. In the 1.2% series, even inserting after dialysis pushed the PDI from 0.136 up to 0.246, whereas in the 1.5% series it barely moved, from 0.147 to 0.161. The discussion reasons that LNPs with less PEG are intrinsically less stable and more prone to aggregation, and so clump together during post-insertion itself. The authors took the 1.5% formulation as their standard from then on. That said, the p value column deserves a careful look: not one of the comparisons in the 1.5% series reaches significance. I return to this in the second half.

👦 Student: What is the PDI number actually looking at?

🧬 Dr. Exotaro: How uniform the particles are. The closer to 0, the more they are all one size; the larger it gets, the more scattered they are. In formulation work it is weighted about as heavily as the mean size, because a population of mixed large and small particles behaves unpredictably in the body and is not stable as a product either.

③ It works anywhere in the 25–55 °C range, and is over in 30 minutes

In liposomes, post-insertion efficiency has been reported to depend strongly on temperature, governed by the phase transition temperature of the lipid. What about LNPs?

LNPs loaded with Cy5-labeled siRNA were reacted with Cy3-tagged components for 2 hours each at 25, 35, 45 and 55 °C. The clever part is that the dye on the siRNA and the dye on the inserted component were kept separate. Efficiency was then judged by co-elution on FPLC. On a Sepharose CL-4B column the large LNPs come off early and the small free micelles come off late. The trick is this: if the component has inserted, the Cy3 peak overlaps the Cy5 peak; if it is left over, a separate peak appears later.

The result: at every temperature the Cy3 peak overlapped the Cy5 peak completely. Sizes also stayed within 64–79 nm, which against the bare LNP (78 nm) cannot be called an increase (n = 2, reported as mean and range). From there on the standard was set at 45 °C.

Time was probed the same way: 0, 0.5, 1 and 2 hours at 45 °C. Only in the 0-hour sample did a small residual Cy3 peak remain; in all the others the peaks overlapped completely. And here the paper adds a satisfying footnote of its own. Even the “0 hour” sample spends 10 minutes being injected and passing through the column. So the reaction has in all likelihood already finished inside those 10 minutes. That is why the authors concluded it completes within 30 minutes.

For liposomes the optimum has been put at 2–4 hours, and in some reports as long as 48 hours. An LNP that is done in 30 minutes is therefore 4 to 8 times faster than that optimum, and faster by orders of magnitude than the longest reports. In practice that is a very large difference.

④ There is a ceiling on how much PEG you can add

So how much can you pack in? Cy3-tagged components were varied from 0.5% to 4% and reacted for 30 minutes at 45 °C.

(mean ± standard deviation, n = 3)

Everything stays tidy up to 2%, which is 3.5% total PEG. Go above the 2% mark, though, and components that failed to insert turn up as free micelles while the LNP itself starts to break. At 3% the size jumped to 215 ± 269 nm, and at 4% to 565 ± 164 nm. The paper puts the ceiling somewhere between 2% and 3% added PEG, and notes that 0.1–0.3% is plenty for practical use.

Here too the gap with liposomes shows. Liposomes routinely contain more than 10% total PEG, and their post-insertion ceiling has been put at around 10% total PEG. An LNP, by contrast, tops out at 2–3% of added PEG (3.5–4.5% in total). The two figures are measured on different bases, total versus added, so you cannot simply compare the ratios — but it does mean that an LNP tolerates considerably less PEG than a liposome does.

⑤ Labeling does not blunt the effect, and an address label speeds up uptake

Finally, the biological check. The cells used were Kasumi-1 and SKNO-1 — both human acute myeloid leukemia (AML) lines that carry t(8;21) and express RUNX1/ETO.

First, does adding a dye blunt the effect? siRE was delivered in LNPs with and without a Cy3 label, and RUNX1/ETO transcript was measured by qPCR 24 hours later. Both showed roughly a 2-fold knockdown relative to the mismatch control — the paper’s phrasing; roughly a halving against the control — with no significant difference between labeled and unlabeled (the “ns” in Figure 6A). The fluorescent label does not get in the way of the effect.

Now the main event. The ligand used here is LDV, a molecule that binds VLA-4 (integrin α4β1); VLA-4 is broadly expressed on cells of the haematopoietic system. Two different components were inserted into the same particle — Cy3 for tracking and LDV for targeting — and uptake was followed by flow cytometry. Measured at 1, 3, 6 and 24 hours, LDV-bearing LNPs were taken up significantly more than LNPs without it (n = 4, two-way ANOVA; the significance is reported as a between-group comparison of targeted versus untargeted, not timepoint by timepoint). That difference also showed up in how quickly RUNX1/ETO knockdown got going (Figure 6C; this experiment was run at 0.1 µg/ml).

👦 Student: What is so good about being able to attach two labels?

🧬 Dr. Exotaro: Two things. One is that it separates tracking from targeting: the dye that tells you where the particle went and the ligand that decides where it goes can be designed independently. The other is an application the paper itself hints at — the possibility of reaching awkward targets that are taken up only when two receptors are engaged at once. Think of a door that will not open unless you hold both keys.

How does the future change? (the road to the clinic)

What did this method make possible — looking at the follow-up

Formulation methodology on its own cannot tell you how far away the clinic is. Fortunately, this paper has a sequel. The following year, in 2023, the same group published Nanoparticle-mediated targeting of the fusion gene RUNX1/ETO in t(8;21)-positive acute myeloid leukaemia in Leukemia (Issa H, Swart LE, et al., Leukemia 2023, vol. 37, no. 4, pp. 820–834, DOI: 10.1038/s41375-023-01854-8). Both Issa, the first author, and Swart are authors of the paper discussed here.

According to the abstract, transiently knocking RUNX1/ETO down with siRNA-loaded LNPs reduced its binding to target genes and altered the binding of RUNX1 and its cofactor CBFβ as well. Median survival in a t(8;21) AML mouse model was substantially extended, and the report goes further: a transient knockdown produced long-term suppression of the leukemia cells’ proliferative and colony-forming capacity, drove myeloid differentiation and markedly reduced re-transplantation capacity. The authors sum this up as demonstrating the preclinical feasibility of targeting a fusion gene. Hit it once, briefly, and the effect lasts. Which makes this formulation paper the foundation that let that experiment happen at all.

👦 Student: So a treatment is close, then?

🧬 Dr. Exotaro: No — this is still preclinical. There is a wide gap between longer survival in mice and working in patients. In humans you still face the particles’ bias toward the liver, the immune response to PEG on repeat dosing, and the problem that bone marrow is a hard place to reach. Even so, it is fair to say that the idea of aiming at a fusion gene by sequence has started to acquire substance.

What it means on the manufacturing floor

The practical headline is that with post-insertion you can attach the address label at the end of the process. Antibodies and nanobodies are expensive and fragile toward both heat and organic solvent; mix them in from the start and you drag them through the mixing and the dialysis as well. Finish the particle first, then a 30-minute reaction at 45 °C puts the label on, and the cargo does not leak (although another overnight dialysis follows the reaction, so the whole process is not over in 30 minutes). Make a large batch of one base and change only the label according to the application — in other words, modularity. If you no longer have to rebuild a production line for every target, the structure of development costs changes.

How to read this study critically (its limits, and how it could have been better)

What deserves credit

To be fair first. The comparison is designed correctly — direct clicking and post-insertion are set side by side on the same base, the same ligand and the same cargo. It checks itself with both numbers and images — size is quantified by DLS, while shape and internal structure are seen directly by cryoEM. DLS is a measurement that gets dragged upward by large particles, so corroborating it with images is worth something (although the paper does not quantify a size distribution from the cryoEM). The efficiency readout is clean — tagging the particle and the component with separate dyes and asking whether the peaks overlap is a design with nothing forced about it. And the paper confirms that the cargo is not leaking. If the labeling step lets the contents out, the whole exercise is pointless.

Points to read with care

① The statistical support is weak. The comparisons in the 1.5% series of Table 2 come out at p = 0.13, 0.33 and 0.099, none of them reaching significance (and again, the paper does not state which comparison each p value refers to; this is inferred from the blank “after dialysis” rows). So one of the paper’s central claims — that adding before dialysis more than doubles the size — turns out not to be statistically supported in the very 1.5% series that was adopted as the standard condition. Where significance did appear was the 1.2% series, the one the authors themselves ruled out as too unstable to use.

② n is small, which is a weakness for negative conclusions. Tables 1, 2 and 4 use n = 3; Tables 3 and 5 use n = 2 and report a mean and range rather than a standard deviation. Asserting negative conclusions — “it does not depend on temperature”, “it does not depend on time” — from two measurements calls for caution. You cannot tell whether there is no difference or no power to find one. On top of that, the 3% condition in Table 4 is 215 ± 269 nm, a standard deviation larger than the mean; that is circumstantial evidence the particle is broken, but it cannot be read as “3% gives you a particle of this size”.

③ The same numbers appear in Tables 3 and 5. LNP 1.0 in Table 3 and LNP 2.0 in Table 5 are both 78 [65, 90] nm with a PDI of 0.205 [0.144, 0.266], and the 45 °C, 2-hour entry in Table 3 and the 2-hour entry in Table 5 are both 64 [60, 68] nm with a PDI of 0.197 [0.174, 0.219]. The natural reading is that one experiment is shown twice, but nothing says so, which leaves room to mistake them for independent experiments.

④ Text, tables and figures disagree with one another. Because this could do real damage to a reader trying to reproduce the work, let me be specific.

• The reaction time in Figure 5 differs between the figure and its caption. The arrow in Figure 5B is drawn with “30 min at 45 °C”, yet the caption on the same figure says the reaction is run for 2 hours at 45 °C. The conclusion in the text is 30 minutes at 45 °C. In the one figure meant to present the optimized procedure, the times do not agree.
• Several cross-references and numbers do not match the tables. The caption to Table 1 says the symbols correspond to the cryoEM panels in Figure 1, but the cryoEM is Figure 2. The text’s “diameter increased by 50% (Table 1, second and fourth entries)” does not work either: the second entry is 110 nm and the fourth is 93 nm, so in fact it decreased (the roughly 50% increase is against the first entry, 63 nm). “Increased the 1.5% PEG LNPs by 10–30%” is the same story: the relevant comparison in Table 2 is 66 nm → 93 nm, about a 41% increase. And the text gives the 3% condition as a mean of 218 nm, where Table 4 says 215 ± 269 nm.

None of this overturns the conclusions; these are the kind of errors a proofread catches. But the value of a protocol paper lies in being able to build exactly what it says, so for anyone reproducing it the problem is not a small one.

⑤ Do not misread what “targeting” means here. VLA-4, which LDV binds, is, as the paper itself writes, broadly expressed on cells of the haematopoietic system. It is not a leukemia-specific target. What this experiment demonstrated is not that leukemia cells alone could be picked out and hit, but the principle that attaching an address label speeds up uptake and brings the effect on faster. The paper claims no more than that, but it is an easy thing for a reader to conflate.

⑥ All of this happens in test tubes and cells. There are no animal experiments. Only two cell lines were used, and no patient-derived cells. Stability in serum, circulation time, immune response — the factors that decide clinical fate even more than size does — lie outside this paper’s scope. And although the method for measuring encapsulation efficiency is in the Methods section, the actual percentage is nowhere in the tables of the main text. The storage interval before measurement is also a range, “up to 7 days”, which could itself contribute to variability.

⑦ Background the paper does not mention — public information. According to his institution’s public profile, co-author Schiffelers serves as an editor of International Journal of Pharmaceutics, the journal in which this paper appears. There is no mention of it in the conflict-of-interest section. Journals have standard procedures for handling submissions that involve people on the editorial side, and this does not in itself imply anything improper; I note it because it strikes me as useful background for a reader to have. The corresponding author’s patent application, for its part, is properly disclosed.

What would have made it a stronger study

• Increase n, and fix the design of the tests in advance. Negative conclusions in particular — “independent of temperature”, “independent of time” — need a design that demonstrates equivalence, and a statement of power
• Reconcile the numbers across figures, tables and text mechanically. The inconsistencies listed above could be eliminated with a pre-submission checklist
• Report encapsulation efficiency, zeta potential and serum stability as numbers. Size and PDI alone are not enough to characterize a formulation
• Compare against a more specific ligand. VLA-4 suits a proof of principle but not an assessment of selectivity; setting it beside a more restricted target such as CD33 or CLEC12A would make the meaning of “targeting” clear
• Put patient-derived cells and normal haematopoietic progenitors side by side. How much uptake into normal cells increases is a number you want from a safety point of view
• Measure encapsulation efficiency and leakage in the direct-click condition too. That would give quantitative backing to the claim that the particle is broken

Dr. Exotaro’s perspective

Let me be honest with you. This paper is not my own field. What I work on is treating neurological disease, spinal cord injury (SCI) above all, using mesenchymal stem cells (MSCs) and the extracellular vesicles (EVs) they release. Neither childhood leukemia nor LNP formulation design is territory I have worked in with my own hands. So on the fine points of formulation chemistry I am reading this from the same side of the page as you are.

I chose it anyway, because what is written here bears directly on our own field.

First: post-insertion is used just as it is here in the surface engineering of exosomes. A researcher who wants to aim an EV somewhere builds a lipid–PEG component carrying a peptide or a ligand and pushes it into the finished EV. The thinking and the parts are all but identical to this paper’s. So the cautions this paper raises — the timing of insertion changes the particle’s properties, there is a ceiling on how much you can load, the structure can break during the reaction itself — become questions those of us working with EVs ought to take home. Honestly, if you ask me whether I have carefully re-measured size distributions after inserting components into EVs, I cannot say that I have.

Second — and this is the heaviest point. The real lesson of this paper is that a procedure does not transfer across particle types. In their conclusion the authors write that the kinetics and thermodynamics of post-insertion may differ between liposomes and LNPs, and that a procedure established on one particle cannot casually be carried over to the other. And indeed: a reaction said to take several hours to 48 hours on liposomes finishes in 30 minutes on an LNP, and against liposomes that routinely carry more than 10% total PEG, an LNP will accept only 2% added PEG. The before-or-after-dialysis distinction is not something that has been reported for liposomes either.

So what about EVs? Structurally an EV has a lipid bilayer, which puts it closer to a liposome. But that membrane is densely studded with membrane proteins, tetraspanins among them; the surface is covered in a layer of protein; and above all, because living cells make them, every single particle is different. An EV is neither a liposome nor an LNP — it is a third kind of particle. Extend this paper’s logic straightforwardly and the answer is that what happens with an EV will not be known until it is measured on EVs.

The claim carries weight because Schiffelers, who contributed the conception, took his doctorate on liposomes, won an ERC grant on extracellular vesicles and became a founding member of ISEV — a person who has genuinely moved back and forth among all three particles. The fact that the acknowledgements carry the name of the extracellular vesicle researcher Pieter Vader tells you something, too, about the soil this paper grew out of.

Third: pulled back toward my own subject, the distance still to be covered comes into focus as well. What this paper worked with is leukemia cells floating in culture medium. A cell sitting in the blood is about the most reachable customer a particle can have. What we want to reach, by contrast, is injured neural tissue on the far side of the blood-brain and blood-spinal cord barriers. Having the technology to write an address and being able to arrive at that address are separate problems.

Even so, I think the order of work is right. Arguing about destinations without first nailing down the technology of delivery produces nothing you can hold in your hand. It is not glamorous work, but it is work somebody has to do. And the quality of work like this shows up a few years later as a survival curve in another paper. Which is just what happened: the year after this formulation paper, the same group’s survival data appeared.

One last thing, about the writing. When their own results clashed with the earlier liposome literature, these authors did not write that the liposome reports were wrong. They wrote that the particles are different, so the procedure does not transfer. Not which one is correct, but a question of the domain of applicability. It is an unshowy stance, and it is the kind of writing I trust.

Note: this article was written with the assistance of AI.