“The Primary Efficacy Endpoint Was Not Met” — What PISCES-2, Which Implanted Neural Stem Cells into the Brain, Really Left Behind

Between 2014 and 2016, at eight hospitals in the UK, 23 patients were put under general anesthesia, had a hole drilled in the skull, and had 20 million neural stem cells (NSC) injected deep into the brain, into the putamen. What was used was a frozen product clonally derived from human fetal cortical neuroepithelial cells — a cell line, the identical preparation for every participant. The patients were people living with the aftermath of ischemic stroke, unable to grasp a 2.5 cm block with the paralyzed arm. Of them, 9 had no arm movement at all; 14 had a trace of movement remaining.

The report sits in the 2020 Journal of Neurology, Neurosurgery and Psychiatry, and the discussion says this: “the primary efficacy endpoint was not met.” Yet most of the summaries in circulation are causal statements — “stem cell implantation improved upper limb function.” What the original paper wrote was an observation in a single-arm study: “improvement was observed at 3, 6 and 12 months.”

Publication Details

• Paper title: Intracerebral implantation of human neural stem cells and motor recovery after stroke: multicentre prospective single-arm study (PISCES-2)
• Paper type: Original research article. A prospective, multicenter, single-arm, open-label exploratory clinical trial. There is no control group
• Authors and affiliations: 14 in all, including Keith W Muir (corresponding author, Institute of Neuroscience & Psychology, University of Glasgow). A collaboration across eight UK sites, with 4 co-authors from ReNeuron Ltd.
• Journal: Journal of Neurology, Neurosurgery and Psychiatry (JNNP, published by BMJ), 2020, Volume 91, Issue 4, pages 396–401. DOI 10.1136/jnnp-2019-322515, PMID 32041820
• Peer review and publication dates: Received November 25, 2019; accepted January 8, 2020; published online ahead of print February 10
• Impact Factor: 7.7 in JCR 2025 (the figure published by BMJ). 161 citations (OpenAlex, checked August 4, 2026)
• Open access and registration: Hybrid OA, CC BY-NC 4.0. Individual participant data and the revision history are public as well. EudraCT 2012-003482-18
• Funding and conflicts of interest: Funded by ReNeuron Ltd. and Innovate UK. The disclosure is refreshingly plain: “JH, PS, KP and JS are employees of ReNeuron.” KWM has previously served on ReNeuron’s advisory board and sits on the steering committee of PISCES-3. The protocol was designed by these 4 employees, with input from the clinical investigators

About corresponding author Professor Keith W Muir: Professor Muir holds the SINAPSE Chair of Clinical Imaging at the University of Glasgow in the UK and is a neurologist at the Queen Elizabeth University Hospital, specializing in acute stroke treatment, and brain imaging for treatment decisions. Professor Muir has been a central investigator on many stroke clinical trials, including ATTEST, PISCES, PISTE and WAKE-UP.

What Wasn’t Understood Before?

Stroke, as the paper says in its opening line, is “the second commonest cause of death and the leading cause of long-term disability in adults.” In Japan too, cerebrovascular disease is the second leading cause of needing long-term care (2022 Comprehensive Survey of Living Conditions). The aftermath comes with a “wall of time.” The body of prior work the paper cites is cold-eyed: if there is no recovery by day 10 after onset, severe disability remains at day 90 as well, and what determines hand dexterity at 6 months is only the initial severity and the recovery within the first month.

👦 Student: Does the brain’s recovery really “end” half a year after onset?

🧬 Dr. Exotaro: Strictly speaking it does not “end” — the closer image is that a steep downhill grade flattens into level ground. What the paper cites is a body of work saying outcome is largely settled within the first weeks after onset, and it does not say that recovery falls to zero. It says the momentum slows sharply over a few months. PISCES-2 was a trial asking whether you can put a gradient back under someone left standing on flat ground.

So what were the stem cells expected to do? According to the paper, cell therapy may promote recovery through “modulation of inflammation, neuroplasticity and angiogenesis via the secretion of cytokines and growth factors.” CTX0E03 in particular is described as “modifying the local inflammatory response and, in animal models, promoting host cell neurogenesis after stroke and host cell angiogenesis after limb ischemia” (note that the evidence for angiogenesis does not come from a stroke model).

👦 Student: So this isn’t a story about the transplanted stem cells themselves turning into neurons?

🧬 Dr. Exotaro: The action the paper mainly assumed is not neural replacement — that is as far as an accurate statement can go. The implanted cells are less a “pinch hitter” than a coach sent onto the field. That does not mean CTX0E03 is a cell that has lost the ability to differentiate, though. The paper states explicitly that switching off proliferation in culture “restores the cells’ capacity to differentiate.” And what actually became of the cells once they were inside the patients’ brains was never verified in this trial. What does not change is the framework: if an effect appears, it is decided by how much “room to be roused” is left in the host brain.

CTX0E03 is an allogeneic cell line clonally derived from human fetal cortical neuroepithelial cells. Thanks to a retrovirally introduced c-mycERTAM transgene, it is designed as a “conditionally immortalized” line that proliferates only in the presence of 4-hydroxytamoxifen (4-OHT); remove the 4-OHT and proliferation stops and the capacity to differentiate returns. The fact that the driver is the proto-oncogene c-myc is something we will come back to in the safety section.

What Did This Paper Find?

Enrollment and treatment ran from July 2014 to August 2016 at eight UK sites, with follow-up completed in August 2017. Eligible patients were aged 40 or over, had had a supratentorial ischemic stroke 2–13 months earlier, had an NIHSS (National Institutes of Health Stroke Scale) arm motor score of 2, 3 or 4, and an ARAT (Action Research Arm Test) item 2 score of 0 or 1. Of 41 patients, 18 were excluded (the commonest reason, in 10, was antibodies against the HLA (human leukocyte antigen) expressed by CTX0E03 testing positive), 23 went on to injection, and 20 completed 12 months of follow-up. Of the participants, 96% were white, and the median time from onset to enrollment was 7 months. The decisive sentence is this one: “Baseline ARAT item 2 scores were 0 in 22/23 and 1 in one participant.” (at baseline, 22 of the 23 scored 0 on the primary endpoint measure).

Surgeons injected into the putamen ipsilateral to the infarct, delivering 20 million cells (400 μL) split across 20 deposits — 4 tracts × 5 sites each, and the procedure succeeded in all 23 patients. Every patient also received at least 1.5 hours a week of rehabilitation of the affected arm for 6 weeks (any additional therapy was at each site’s discretion).

The Primary Endpoint — Not Met

The primary endpoint was “an improvement of 2 points or more on ARAT item 2 at 3 months after implantation.” Here it is in the original wording.

“At the 3-month evaluation, only one patient met criteria for ARAT test 2 response, and so the primary efficacy endpoint was not met.”

The number meeting the criterion was 1/23 (4%). The primary endpoint was not met. At 6 months it was 3/22 (14%), at 12 months 3/20 (15%). But the responder counts at each timepoint run 1 → 3 → 3: there is no increase after 6 months. The apparent rise from 14% to 15% is because the denominator fell from 22 to 20.

👦 Student: Is the primary endpoint really that special? If things improved on the other measures, isn’t that good enough?

🧬 Dr. Exotaro: A primary endpoint is the target you declared, before seeing any data, as the one point on which you would stake the trial. If you redraw the target after the arrow has flown, anyone can hit the bullseye every time. Missing is not, in itself, a defeat. The defeat is missing and then insisting you hit.

Secondary Endpoints

Let me unpack the scales first: ARAT (upper limb function, total score 0–57), mRS (modified Rankin Scale, which grades disability from 0 to 6; median 3 in this trial), Barthel Index (independence in activities of daily living, 0–100), FMA (Fugl-Meyer Assessment, motor function; upper limb 0–66). Response was defined, in that order, as an improvement of 6 points, 1 grade, 9 points, and 10 points or more.

Caveats are needed. NIHSS has vanished from the table — because no one improved by 10 points or more; but only 4 patients had a baseline score above 10, so for most patients the threshold was close to unreachable. The Barthel hit a ceiling effect, leaving 6 patients unassessable. The FMA’s “improvement of 10 points or more” was defined as “10 points or more in either the upper or the lower limb” (table footnote), so in a trial aimed at the arm, leg improvement counted too, and it was administered in only 10 patients. The table’s “at least one of ARAT, mRS or Barthel” (57–70%) does not appear in the Methods list of secondary endpoints, and it is a figure that rises mechanically through multiplicity.

And here is what I most want to convey. Every number in the table is “how many people were over the threshold at that timepoint,” not evidence that the same people held on to their improvement. Of the 2 patients whose mRS improved from 4 to 3, the paper writes “both of whom returned to grade 4 at subsequent visits,” and on the ARAT there is 1 patient reported who responded at 6 months but not at 12. This trial laid down no definition for judging whether a response was sustained.

The Paper’s Headline Finding — But It Is post hoc

What the paper pushes hardest is a finding about patient selection. Comparing the 9 patients with an NIHSS arm score of 4 (no movement at all) with the 14 scoring 2 or 3: “Responses on ARAT item 2 or total score were seen in none of those with baseline NIHSS arm score of 4” — not a single ARAT responder came from the group with no movement at all. For improvement of 1 grade or more on the mRS as well, at 12 months the arm 2–3 group had 6/12 (50%) against 1/8 in the arm 4 group.

But this is an analysis the Methods explicitly label a “Post hoc subgroup analysis,” and the two groups differ substantially from baseline: the arm 4 group was significantly older (70±8 vs 57±9 years, p=0.002), had a higher total NIHSS (p=0.019), and a lower Barthel (p=0.003). You cannot pull “presence or absence of residual movement” alone out of two groups that differ in age, severity and ADL (activities of daily living). On top of that, the arm 4 group was not in the eligibility criteria when the trial began; it was added by a mid-trial amendment, so the enrollment periods themselves differ — a triple caveat attaches.

👦 Student: If even three people got better, doesn’t that mean the cells worked?

🧬 Dr. Exotaro: I understand the feeling. But this trial has nobody to compare against. And on the primary endpoint, 22 of 23 started at 0, pinned to the floor. You cannot measure below 0, so when variability goes downward it never shows up in the numbers, and only when it goes upward does it get counted as “improvement.” That does not mean “it always goes up on its own.” Only the upward moves are visible; the downward ones are not — which is why, without a control group, you cannot tell genuine improvement apart from measurement variability and regression to the mean.

Safety

There were 17 serious adverse events (SAEs) in 11 patients. Two deaths (sepsis on day 241, and a suicide 7 days after the final visit) were judged unrelated to the trial procedure. Of these, 6 events in 4 patients were surgery-related, and all resolved — ischemic stroke, subdural hematoma, headache, vomiting, and a partial seizure and sepsis on postoperative day 22. That one case the original text describes as “possibly related to CTX0E03 cells.”

This is the widest gap in temperature in the whole paper. The abstract asserts that “no cell-related adverse events occurred up to 12 months of follow-up,” while the discussion writes of “few potentially cell-related adverse events.” Not a single adverse event was confirmed to have been caused by the cells. But there was an event that the investigators on the ground recorded as “possibly related,” and the abstract’s “zero” drops that record. Note also that although allogeneic cells were implanted into the brain, this paper says nothing at all about immunosuppressive drugs.

One more thing. CTX0E03 is an immortalized cell line whose proliferation is controlled by a transgene derived from the proto-oncogene c-myc, and whether those cells inside the brain will one day form a tumor is, in theory, the heaviest risk of all. Yet the paper does not discuss it, and following 23 patients for 12 months is nowhere near enough to rule out tumor formation (PISCES-1 is reported to have shown no problems out to 8 years, but only 2 of its 11 patients received this dose).

How Will the Future Change? (The Road to the Clinic)

The abstract concludes like this: intracerebral implantation is “feasible in a multicentre trial,” and “improvements in upper limb function were seen at 3, 6 and 12 months, but no improvement was seen in patients with no arm movement at all, suggesting a candidate target population for future controlled trials.” The main text likewise closes with “further evaluation in a randomised controlled trial is warranted.” So what happened to that trial?

PISCES-III (NCT03629275) was that trial. Multicenter, randomized, sham-surgery controlled, with a primary endpoint of “the proportion improving by 1 point or more on the mRS at 6 months.” But the planned figures in the public documents do not agree: a 2023 design paper gives 6–12 months after onset and 110 patients (Laskowitz DT, et al. Front Stroke 2023;2:1182537), whereas the final Version 3.0 protocol posted on ClinicalTrials.gov (September 26, 2019) says “approximately 130 patients,” “2:1 randomization” and “6–24 months after onset,” leaving an amendment trail that widened the enrollment window. The ending was TERMINATED, with 15 patients actually enrolled. The reason for stopping given in the registry is “a strategic decision to advance the stroke disability program through regional partnerships” — it does not say the trial stopped because it did not work. Follow-up is marked complete as of March 2, 2021, but hasResults is false (checked August 4, 2026; there is no record of results submission, and the registry was last updated August 12, 2021). Neither a results table nor a peer-reviewed efficacy paper can be found. The trial PISCES-2 called “essential” ended without an answer.

The developer, ReNeuron, has itself stepped back from stroke. On March 20, 2024 it entered administration (the UK insolvency procedure), on September 2 it was delisted from AIM, and on March 17, 2025 it restarted as a private company (ReNeuron release, checked August 4, 2026). The company’s official history records that it out-licensed the stroke program to Shanghai Fosun Pharma in 2019 and concentrated in-house research on exosomes in 2022. But the agreement reported in the April 2019 release was an exclusive license limited to China, and where the rights outside China went cannot be traced from published material. That the focus of in-house development shifted can be confirmed; whether stroke development was formally terminated cannot. The long-term safety follow-up study of PISCES-III participants (NCT05598775, planned n=9) is “enrolling by invitation”, with an estimated completion date of December 31, 2026 — but the registry was last updated in March 2023, and its actual progress cannot be confirmed. “The trial failed” and “the company did not survive” are two different things. A result that is a failure still becomes an asset for the next researcher, but a result that is never published becomes nobody’s asset.

Other companies reached the same ending. The sham-controlled phase 2b trial in chronic ischemic stroke of SB623 (SanBio), implanted by stereotactic neurosurgery — ACTIsSIMA (n=163) — missed its primary endpoint in 2019. The intravenous MASTERS (n=129) also missed its primary endpoint (day 90) (odds ratio 1.08, 95% CI 0.55–2.09, p=0.83; Hess DC, et al. Lancet Neurol 2017;16:360–8), and there was no between-group difference on the prespecified secondary endpoints either. Japan’s TREASURE (n=206) went the same way. Athersys filed for Chapter 11 in January 2024, and in April Healios acquired the assets. Healios in turn disclosed on December 9, 2025 that for acute ischemic stroke it “will not pursue a rolling submission and will reconsider its development strategy” (its priority is ARDS). This is not abandonment; it is a filing pathway left hanging in the air.

The exception is SB623 in traumatic brain injury (TBI). The sham-controlled trial STEMTRA (n=63) met its primary endpoint (between-group difference 6.0, 95% CI 0.3–11.8, p=0.040), and in Japan “Acuugo® Intracerebral Implant Injection” (vandefitemcel) received conditional, time-limited approval on July 31, 2024 for improving motor paralysis in the chronic phase of TBI, with an NHI price listed on May 20, 2026 (72,716,528 yen per dose). But the basis is an interim analysis with a confidence interval whose lower bound barely clears 0, and the secondary endpoints showed no significant difference. This is a system that sets a time limit while efficacy is still at the stage of being “presumed,” and what was approved is not stroke.

👦 Student: So right now in Japan, can’t people get stem cell therapy for the after-effects of stroke?

🧬 Dr. Exotaro: As far as I could verify, as of August 2026 there is no approved cell therapy for stroke in Japan. The only intracerebrally implanted cell medicine available under insurance is Acuugo, and its indication is traumatic brain injury. One more distinction. “Regenerative medicine products” under the Pharmaceuticals and Medical Devices Act, and “regenerative medicine and similar practices,” in which a medical institution processes cells itself, are two different frameworks. The latter is a system that requires notification and committee review; it is not a system that reviews efficacy and grants approval. What you can conclude from an uncontrolled case report in private-pay practice — please work that out backwards for yourself.

How to Read This Study Critically (Limitations, and Ways to Raise the Quality Further)

First, the honest points. The paper states plainly in the discussion that the primary endpoint was not met, and states in the Methods, of its own accord, that the central finding is post hoc. Nor is there any self-serving manipulation of the thresholds: the Barthel response threshold of 9 points sits far above the minimal clinically important difference (MCID) of 1.85 points reported in stroke patients (Hsieh YW, et al. Neurorehabil Neural Repair 2007;21:233–8), and the 10-point NIHSS threshold was set almost out of reach. Conflict-of-interest management is likewise equipped with safeguards: an independent DSMB (Data Safety Monitoring Board), an independent CRO (contract research organization), an independent statistician, and “The funder was not aware of outcomes until database lock.”

That said, the Contributors statement also records “KWM undertook analyses,” and who performed the headline post hoc analysis cannot be determined. The corresponding author also sits on the steering committee of the successor trial, so the closing line “further evaluation is warranted” has to be read as a sentence written by someone with a structural incentive to continue.

The way the limitations are written deserves fair credit. The authors state explicitly that “the open-label design may bias functional assessment” and that “further evaluation in a randomised controlled trial is essential,” list age, comorbidity, infarct volume and location, the corticospinal tract (CST), time from onset, and concomitant rehabilitation and its intensity as candidate confounders, and write that the possibility that other factors contributed cannot be entirely excluded.

Even so, things that a single-arm design cannot distinguish in principle remain. (a) Spontaneous recovery. The lower bound for enrollment was 2 months from onset, a period in which spontaneous recovery can still be under way. The foundation of the design is the assumption that “the probability of spontaneous improvement was predicted to be under 5%,” but that 5% is never measured within this paper, and no citation is attached to it. That the responders at 6 months were significantly younger (53±6 vs 64±11 years, p=0.025) is also consistent with the rival hypothesis, given that youth is a known predictor of good spontaneous recovery (though this difference disappears at 12 months, and it rests on 4 patients versus 18, unadjusted for multiplicity). (b) The effect of rehabilitation. The mandated rehabilitation is an inseparable part of a package intervention together with cell implantation. Additional rehabilitation was decided site by site and patient by patient, and neither an upper limit nor the total hours delivered is reported. The authors counter by citing the literature, but that is an indirect argument, not a measurement. (c) Regression to the mean (22 of 23 sat at the floor of 0). (d) Surgical and expectation effects. Patients knew they had received experimental cells, and both the ARAT and the mRS are scales into which effort and subjectivity enter. There is also no mention of independent blinded assessors.

The statistical design has problems too. The primary assessment timepoint was changed mid-trial (day 180 → day 90, amendment 8), and when day 90 came back as not met, the paper justifies continuing on the grounds of the 6- and 12-month results and the “arbitrariness of choosing 3 months.” The description of the confidence level is inconsistent as well: the Methods say “the lower one-sided 50% CI,” the discussion says “90% CI.” The lower bound of a one-sided 50% confidence interval is nearly identical to the point estimate, so it controls the error probability almost not at all. From the counts in the paper, the Clopper-Pearson 95% confidence interval I calculated for 1/23 at 3 months is 0.1–21.9%.

Biological verification is missing as well. There is no evidence of engraftment, no imaging finding at the implantation site, no biomarker — a design in which, whether it works or not, you cannot know why. Generalizability is limited too. The commonest reason for exclusion was antibody positivity against the HLA of CTX0E03, in 10/41 (about 24%), and 96% of the participants were white. Whether that roughly 24% carries over to other populations cannot be judged from this trial.

So how could the quality have been raised? The most practical option is a delayed-start design that randomizes the timing, while ultimately offering implantation to everyone. It clears a lower ethical bar than sham surgery, and even with 23 patients it makes a same-timepoint comparison possible. The primary endpoint should have been the ARAT total score or the FMA upper limb, which can pick up graded change (with the total score, changes across a range of 1–54 points were in fact detected). Beyond that: do not change the assessment timepoint mid-trial; blinded independent scoring from video recordings; a record of total rehabilitation hours. The greatest missed opportunity is DTI (diffusion tensor imaging) and TMS (transcranial magnetic stimulation) assessment of the CST — one of the principal predictors of upper limb motor recovery went unmeasured, even though the authors themselves list it as a candidate confounder. Had it been there, “arm 4 does not respond” could have been converted into a biological hypothesis: because the CST is severed.

👦 Student: With this many problems, does that mean the trial wasn’t worth running?

🧬 Dr. Exotaro: No, that’s not it. This trial was not designed as a trial that decides “whether it works.” Beyond feasibility, the paper goes as far as concluding that “functionally meaningful improvement in upper limb movement was observed in patients with residual arm movement.” But it did not fix that as an effect of the cells; it wrote that a randomized controlled trial is “essential.” Building a cryopreserved product and completing the procedure successfully in all 23 patients is real progress too. The problem is those numbers walking off on their own, detached from the limits of the design. When only “70% improved” is cut out and passed around, what gets lost is the triple caveat: “single-arm,” “any one of the 3 scales will do,” and “the denominator is 20.”

Dr. Exotaro’s Perspective

I have kept working on mesenchymal stem cell (MSC) and extracellular vesicle (EV) research in spinal cord injury (SCI), so I cannot read PISCES-2 as somebody else’s business.

The action mainly expected of CTX0E03 was not “the implanted cells become neurons” but “modulate inflammation on the host side and draw out angiogenesis and neurogenesis” — the same framework as MSC-derived EVs. In other words, PISCES-2 was a trial that set out to ask how far the strategy itself — “move the host,” rather than any particular cell type — carries in the human brain from the late subacute into the chronic phase. But a single-arm, open-label trial in 23 patients with no evidence of engraftment is not a design that can answer that question. What we learned goes only as far as “at the one point fixed in advance, the numbers were not enough to say anything moved” — the strategy was neither refuted nor supported.

The endpoint question stings more. Which scale you measure with determines what a trial is able to prove — and the same thing can happen in spinal cord injury trials. What PISCES-2 left behind — “ARAT responders came only from those with residual arm movement” — was a post hoc hypothesis carrying a triple caveat: cell therapy may not be a drug that works for everyone, but may act only in those who still have “room to be roused” in the host — and it disappeared without being tested. The line, though, moves with the scale: on the mRS, even the completely paralyzed group had 1 of 8 improve. Even so, it is the question I think about most seriously in my EV research.

One ironic coincidence. ReNeuron, the company that made CTX0E03, has moved its internal center of gravity away from the most invasive approach of injecting cells into the brain, and is now shifting its footing to an exosome drug discovery platform. I cannot bring myself to cheer without reservation. That PISCES-III stopped at 15 patients and its results were never published means that the time and the burden shouldered by the 15 people who enrolled never reached the next researcher.

To patients and their families, let me put it plainly. PISCES-2 is not a paper showing that stem cells work for stroke; it is a paper recording “what is still missing before we can find out whether they work.” I do not want to inflame hope, and I do not want to sneer. The fact is this: this road is not yet closed. But neither is it open. What decides whether it opens is not the story of one dramatically improved patient, but whether an unglamorous trial with a control group can be carried through to the end.