Why SELLAS Might Fail: The Hidden Risks Behind GPS, REGAL, and the AML CR2 Bull Case

Ticker: SLS
Company: SELLAS Life Sciences Group
Lead asset: Galinpepimut-S, also known as GPS
Key catalyst: Phase 3 REGAL trial in acute myeloid leukemia patients in second remission
Core question: Is the delayed REGAL readout evidence that GPS is working, or could the market be overreading blinded survival data?

Summary

SELLAS Life Sciences has become one of the more debated small-cap biotech catalyst stories. Its lead asset, Galinpepimut-S, or GPS, is a WT1-targeted peptide immunotherapy being tested in the pivotal Phase 3 REGAL trial for patients with acute myeloid leukemia, or AML, who are in second complete remission.

The bullish argument is straightforward: REGAL is event-driven, the final analysis is triggered by 80 deaths, and events have taken longer than many expected. Therefore, bulls argue, patients must be living longer because GPS is working.

That conclusion is possible.

But it is not the only explanation.

Because REGAL remains blinded, the public does not know how deaths are distributed between the GPS arm and the best available treatment arm. A delayed event curve can be caused by GPS benefit, but it can also be caused by a stronger-than-expected control arm, better modern AML care, trial selection effects, long recruitment timelines, or a survival tail in both arms.

The most important question is not whether pooled survival is longer than expected. The real question is whether GPS improves overall survival versus the control arm by enough to produce a statistically convincing and clinically meaningful result.

This article examines why SELLAS might fail, even if the surface-level trial timeline looks encouraging.

Key terms explained

Before getting into the SELLAS case, it is worth defining the terms.

AML: Acute myeloid leukemia, an aggressive blood cancer.

CR1: First complete remission. The patient has responded to treatment and no leukemia is detectable by standard criteria.

CR2: Second complete remission. The leukemia has relapsed once, the patient has received additional treatment, and the patient has again entered remission.

CR2p: Second complete remission with incomplete platelet recovery.

OS: Overall survival. This measures how long patients live.

mOS: Median overall survival. The point at which half the patients in a group have died and half remain alive.

PFS: Progression-free survival. The time before disease progression or death.

MRD: Measurable residual disease. Very small amounts of disease that can remain even when a patient is in remission.

BAT: Best available treatment. In REGAL, the control arm is investigator’s choice of best available treatment.

HR: Hazard ratio. A measure of the relative risk of an event, such as death, over time. An HR below 1 favors the experimental drug. For example, an HR of 0.70 means the risk of death is estimated to be 30% lower in the treatment arm than in the control arm.

Event-driven trial: A trial where the final analysis is triggered by a pre-specified number of events, not by a fixed calendar date. In REGAL, the key event is death.

Blinded or pooled data: Data where the company and public can see the overall event count but not which arm the events occurred in.

What is GPS?

Galinpepimut-S, or GPS, is a peptide immunotherapy targeting WT1, a protein that is overexpressed in several cancers. The idea is to train the immune system to recognize WT1-expressing cancer cells.

WT1 is a legitimate cancer target. It has long been studied in cancer immunotherapy and is especially relevant in AML. That makes GPS scientifically interesting.

But investors should separate two different claims:

WT1 is an attractive biological target.
GPS is clinically validated as a broad cancer platform.

The first claim is reasonable.

The second claim is much harder to prove.

A target can be compelling without a specific drug product being broadly effective. Cancer vaccines face several practical challenges: tumor burden, immune suppression, antigen presentation, HLA type, T-cell exhaustion, tumor microenvironment, prior therapies, and whether the patient is in remission or has active bulky disease.

GPS may work best in a low-disease-burden setting, such as remission or maintenance therapy. That is very different from saying it is a proven platform for many large solid tumor markets.

What is REGAL testing?

REGAL is the pivotal Phase 3 trial of GPS in AML patients who have achieved second complete remission or CR2p after second-line antileukemic therapy and who are ineligible for or unable to undergo allogeneic stem cell transplantation.

The study compares GPS monotherapy in the maintenance setting against investigator’s choice of best available treatment. The primary endpoint is overall survival from time of study entry.

This is a stronger design than the earlier small studies because it is randomized and uses overall survival as the primary endpoint. However, it is still open-label, and the control arm can be heterogeneous.

REGAL is the real test. Earlier GPS studies can support the hypothesis, but they cannot prove the Phase 3 outcome.

Why the delayed REGAL timeline is not proof that GPS works

The most common bullish argument is that REGAL has taken longer than expected to reach 80 deaths. Bulls interpret that as direct evidence that GPS is extending survival.

That is too simple.

REGAL is blinded. The public knows the pooled event count, not the arm-by-arm event distribution. If 78 or 79 deaths have occurred among roughly 126 patients, we still do not know whether those deaths are mostly in the control arm, mostly in the GPS arm, or relatively balanced.

A simple example helps.

If the trial has roughly 63 patients per arm, the median overall survival in one arm is roughly reached when about 31 or 32 patients in that arm have died.

At 78 total events, several scenarios are possible:

GPS deaths	BAT deaths	Interpretation
39	39	Both medians likely reached; no obvious arm separation from event count alone
34	44	GPS likely better, but GPS median may still be reached
30	48	GPS median may not yet be reached; strong potential signal
44	34	BAT may be doing better than GPS
48	30	GPS would look worse by event count

The same total number of events can support very different interpretations.

That is why pooled survival should not be treated as GPS survival.

A slow event curve can mean:

GPS is working.
BAT is better than expected.
Both arms are living longer than historical assumptions.
The trial enrolled a selected, better-prognosis CR2 population.
Recruitment was prolonged and staggered over time.
There is a long survival tail in both arms.
Modern AML care and supportive care improved outcomes.

Only unblinded data can distinguish these explanations.

The control arm may be stronger than the old AML CR2 assumptions

A central risk for REGAL is that the control arm may outperform the assumptions used in older models.

Historical AML CR2 survival assumptions often anchor around poor outcomes. Some investors use the old 5–8 month control survival range as if it were a hard baseline for REGAL.

That may be dangerous.

REGAL enrolled patients in a modern treatment era. AML care has changed. Venetoclax-based regimens, hypomethylating agents, improved supportive care, better patient monitoring, and clinical trial selection effects can all improve outcomes.

If BAT lives 15–20 months instead of 5–8 months, GPS has a much higher bar.

A trial can look bullish on a pooled basis while becoming harder to win statistically.

For example:

Scenario	GPS mOS	BAT mOS	Possible interpretation
Null case	18–20 months	18–20 months	Long pooled survival, but no meaningful treatment effect
Mild benefit	21–23 months	17–19 months	Could be clinically interesting but statistically uncertain
Strong benefit	26–30+ months	14–17 months	More convincing win
Bearish outcome	16–19 months	19–23 months	BAT outperforms GPS

The key point: a long trial does not automatically mean a positive trial.

The early AML CR2 data: interesting, but highly vulnerable to bias

The bull case often relies on earlier AML CR2 data showing GPS survival of 16.3 months, later updated to 21 months, versus 5.4 months in a matched historical control group.

Those numbers are striking.

But the study design matters.

The Brayer AML/MDS pilot enrolled 16 patients in total: 14 AML and 2 MDS. Among the AML patients, 10 were in CR2 and 4 were in CR1. Formal PFS and OS analysis was limited to 10 AML patients who had received more than two GPS doses.

That is important.

More than two doses means the patient had to survive long enough and remain clinically stable enough to receive at least three vaccine doses. Since early GPS dosing was given every two weeks, the third dose came around week 4.

This creates an exposure-selection problem. The analysis does not simply represent all AML patients who entered the study. It represents patients who were able to receive sufficient vaccine exposure.

In addition, survival was measured from the time of remission, not necessarily from the first GPS dose. That can materially favor the GPS group. A patient could achieve CR2, remain in remission, enter the trial, receive several vaccine doses, and then have survival counted from the earlier remission date.

The comparison group was not randomized. It was a contemporaneously treated historical matched cohort.

The reported results were:

PFS: 10.5 months with GPS versus 4.3 months in controls
P-value for PFS: 0.19
OS: 16.3 months with GPS versus 5.4 months in controls
P-value for OS: 0.0175

The PFS result was not statistically significant, while the OS result was. That mismatch does not disprove GPS benefit, but it should make investors cautious. If a therapy is preventing leukemia progression, one might expect a clearer PFS signal. A stronger OS than PFS signal in a very small, non-randomized study can reflect baseline differences, post-relapse therapy, control-arm weakness, or selection bias.

The old AML CR2 data justify running a Phase 3 trial. They do not make Phase 3 success inevitable.

The AML CR1 study: another source of overinterpretation

GPS was also tested in AML patients in first complete remission.

The headline number often cited is long overall survival, including a median OS figure of 67.6 months in company materials.

But the CR1 setting is heavily selected.

Patients had already:

received induction therapy
achieved first complete remission
received planned post-remission therapy
recovered sufficiently
remained eligible for a vaccine maintenance study

Fast relapsers and medically fragile patients are less likely to make it into such a trial.

Another important issue is how survival is measured. If OS is measured from diagnosis rather than from the first GPS dose, then the survival figure includes time before GPS treatment could have had any effect.

That does not make the CR1 data useless. It supports safety, immunogenicity, and the maintenance-setting hypothesis. But it should not be treated as randomized evidence of large survival benefit.

Why CR2 survival curves can produce a long tail

AML CR2 patients are heterogeneous. Some relapse quickly and die early. Others have better biology, longer first remissions, better performance status, lower residual disease, or more effective salvage options.

This creates a survival curve with phases:

Early drop-off from aggressive disease.
A middle phase where many median events occur.
A long tail of patients who remain alive much longer.

If REGAL is now in the long-tail portion of the curve, the final events can take a long time even if both arms have already reached median survival.

That matters because investors may see the slow pace from 60 events to 78 or 79 events and conclude that GPS must be producing a long survival tail. It may be. But a long tail can also exist in both arms.

At 78 or 79 events out of roughly 126 patients, both medians may already be reached unless the event split is meaningfully imbalanced.

GPS in other cancers: optionality, not yet platform proof

Another bullish argument is that GPS is not just an AML drug but a WT1 platform. There is some truth to this. GPS has been studied outside AML.

However, the non-AML data are not strong enough to support a fully de-risked platform valuation.

Mesothelioma

In malignant pleural mesothelioma, GPS was tested in a randomized, double-blind, placebo-controlled Phase 2 study involving 41 patients.

The reported median OS was 22.8 months in the GPS arm versus 18.3 months in the control arm, with an HR of 0.79.

This is a real signal, and it is more methodologically meaningful than many single-arm studies. But it is not a dramatic result. The absolute OS difference was around 4.5 months, and the sample size was small.

This supports biological activity. It does not prove a broad oncology platform.

Ovarian cancer with nivolumab

GPS was tested with nivolumab in recurrent ovarian, fallopian tube, or primary peritoneal cancer patients in second or later clinical remission.

The study was open-label and non-randomized. It enrolled 11 patients, 10 of whom were evaluable. The company reported a one-year PFS rate of 70% among patients who received more than two doses of GPS and nivolumab.

This is interesting, but it again uses a small sample, a remission setting, and a more-than-two-dose analysis group. That structure is vulnerable to the same kinds of selection issues seen in AML.

Ovarian cancer with pembrolizumab

GPS was also tested with pembrolizumab in WT1-positive relapsed or refractory platinum-resistant advanced ovarian cancer.

The reported overall response rate was 6.3%. Disease control and survival metrics were more favorable than historical checkpoint inhibitor comparisons, but the lack of a randomized control group makes interpretation difficult.

An ORR of 6.3% is not a strong platform signal in an active metastatic solid tumor setting.

Multiple myeloma

In high-risk multiple myeloma after autologous stem cell transplant, GPS showed a median PFS of 23.6 months in a small Phase 2 setting, with median OS not reached.

Again, this is a low-disease-burden or maintenance-like setting. It is scientifically interesting, but not enough to de-risk a large commercial platform.

The pattern is consistent: GPS may be most plausible in maintenance or minimal residual disease settings. The evidence does not yet support a broad, late-stage, multi-cancer platform thesis.

The NCI WT1 argument: true fact, stretched conclusion

A common bullish talking point is that WT1 was ranked as a high-priority cancer antigen by the National Cancer Institute, and therefore GPS should be viewed as a major platform asset.

This is an example of a true fact being stretched too far.

WT1 is an attractive target. That does not mean GPS has proven broad clinical utility.

A buyer would not pay only for the target. A buyer would ask:

Does REGAL show a clean OS benefit?
How large is the hazard ratio?
How clinically meaningful is the absolute survival difference?
How narrow is the initial label?
Can GPS be expanded into CR1, MRD-positive AML, MDS, myeloma, ovarian cancer, or mesothelioma?
Are those expansion paths supported by randomized data?
What is the patent life?
What are the MSK license economics?
How much dilution occurs before any transaction?

WT1 biology creates option value. It does not automatically create a $40 billion platform.

The addressable market may be smaller than the bull case implies

AML CR2 is a serious unmet need, but it is not a large mass-market indication.

The likely initial label would be narrow: AML patients in second remission or CR2p after salvage therapy who are not going to allogeneic stem cell transplant.

Even if GPS succeeds, the commercial opportunity depends on:

number of eligible patients
pricing
duration of therapy
physician adoption
reimbursement
competition from other maintenance strategies
label language
real-world tolerability and logistics
whether the drug moves into earlier lines

A successful REGAL result could create meaningful value. But it does not automatically justify extreme buyout assumptions.

A realistic buyout framework after a good REGAL result might be:

Outcome	Potential enterprise value range
Marginal positive result	$1.5B–$3B
Solid positive result	$3B–$5B
Very strong OS result with clean regulatory path	$5B–$8B
Broad multi-indication platform valuation	Requires much more evidence

A $40 billion buyout would require a very different level of proof: multiple large indications, de-risked platform data, strong patent/exclusivity protection, and a much larger expected peak sales base.

Dilution risk matters

A biotech’s enterprise value is only part of the story. Per-share value depends on the fully diluted share count.

SELLAS has already expanded its share count materially. The company also has an at-the-market offering program that allows it to raise up to $150 million over time.

If REGAL succeeds and the stock rises, raising capital may be rational. The company may need money for:

BLA preparation
manufacturing scale-up
regulatory work
commercial readiness
medical affairs
ex-US strategy
pipeline development

That does not mean dilution is bad for the company. It may be prudent. But it can reduce per-share upside for existing shareholders.

For example, a $5 billion transaction means very different per-share values depending on fully diluted shares:

Fully diluted shares	$5B value per share
180M	$27.78
220M	$22.73
260M	$19.23
300M	$16.67

Investors discussing buyout values should always model fully diluted share count, not just headline enterprise value.

MSK license and IP: important but not fully transparent

GPS is licensed from Memorial Sloan Kettering Cancer Center. That adds scientific credibility, but it also introduces economics that investors need to understand.

The key questions are:

What royalties are owed to MSK?
What milestone payments apply?
Does MSK receive a share of sublicense revenue?
What are the termination rights?
What patent families protect GPS?
What is the actual patent expiration timeline?
Are the strongest claims composition-of-matter, method-of-use, or combination claims?
How much value rests on orphan exclusivity rather than patent life?

These questions do not kill the bull case. But they matter for buyout valuation. A licensed asset with uncertain long-term patent breadth and undisclosed royalty economics deserves a different multiple than a wholly owned, broadly protected oncology platform.

Management, messaging, and change-of-control provisions

In June 2026, SELLAS filed changes to executive severance and change-of-control arrangements. These do not prove a buyout. They do not prove REGAL success.

But they do show that management’s own protection around a possible change-of-control event has been clarified before the pivotal readout.

This can be interpreted as transaction readiness. It can also be interpreted as ordinary governance. Either way, it should not be treated as clinical evidence.

A change-of-control package is not an OS hazard ratio.

What the bear article gets right

A recent bearish article argued that SELLAS had rerated sharply without a matching fundamental update, that REGAL remains a binary high-risk trial, and that delayed events may reflect a stronger control arm rather than GPS benefit.

The most useful part of that argument is not the rating itself. It is the structure:

The trial is blinded.
The control arm may be stronger than historical assumptions.
AML CR2 is a niche market.
The pipeline outside REGAL is mostly option value.
Dilution remains possible.
A positive readout may already be partly priced in if valuation has run far ahead of risk-adjusted fundamentals.

That is a reasonable critical framework.

However, a fair analysis should also note the risk to the bearish view: if REGAL shows a clean, statistically significant OS benefit with a strong hazard ratio and a favorable safety profile, SELLAS could be worth significantly more than conservative rNPV models imply.

This is a binary biotech setup. Both sides have risk.

What would make the bull case stronger?

The bull case becomes much stronger if REGAL reports:

a statistically significant OS benefit
a strong hazard ratio
a clinically meaningful absolute survival gain
clean Kaplan-Meier separation
manageable censoring
good safety
supportive immune response data
supportive MRD or biomarker data
a credible FDA and EMA path
evidence that GPS can move into CR1 or MRD-positive AML

The strongest version of the bull case is not “the trial took longer.” It is “the unblinded data show a clear survival advantage.”

What could make SELLAS fail?

SELLAS could fail in several ways.

1. REGAL misses statistical significance

This is the cleanest failure mode. If the OS endpoint is not statistically significant, the core GPS thesis is severely damaged.

2. REGAL is technically positive but clinically underwhelming

A small p-value win with modest absolute benefit, heavy censoring, or difficult-to-interpret curves may not support the valuation bulls expect.

3. BAT outperforms historical assumptions

If the control arm lives far longer than expected, GPS may not produce enough relative benefit even if patients overall live longer.

4. Earlier studies overestimated the treatment effect

The old CR2 pilot was small, non-randomized, exposed to selection bias, and measured survival from remission. REGAL may reveal that the historical 21 versus 5.4 month narrative overstated the true effect.

5. The initial market is too narrow

AML CR2 is important, but it is not a broad oncology market by itself.

6. Platform value remains speculative

Other GPS indications are not yet strong enough to support a large platform valuation.

7. Dilution reduces per-share upside

Even a good clinical result may be followed by equity financing.

8. Buyout expectations are too high

A $3B–$5B transaction after a strong readout may be plausible. A $40B transaction would require much more evidence than GPS currently has.

Bottom line

SELLAS is not an obvious fraud, and GPS is not an obviously worthless drug. The science is interesting. WT1 is a real target. GPS has shown immunogenicity and signals of activity in several settings. REGAL is a legitimate pivotal trial.

But the market may be treating uncertainty as confirmation.

The delayed REGAL event timeline is not proof of GPS efficacy. It is proof that the pooled trial population has lived longer than some expected. Without unblinded arm-level data, investors cannot know whether that is due to GPS, BAT, patient selection, improved modern care, trial effects, or survival tails in both arms.

The earlier AML CR2 data were hypothesis-generating, not definitive. The CR1 data were encouraging but selection-biased. The non-AML studies provide optionality, not platform proof. The addressable market may be narrower than some bulls assume. Dilution and license economics matter. Buyout expectations may be inflated.

REGAL can still succeed. If it does, SELLAS could be worth meaningfully more.

But “could succeed” is not the same as “will succeed.”

For investors, the central risk is simple:

SELLAS may fail not because GPS has no biological activity, but because the market has priced a high-risk, blinded, event-driven trial as if success were already known.