Retraction Watch

Tracking retractions as a window into the scientific process

“We do not want to create false hope”: Authors retract Cell paper they can’t replicate

with 24 comments

A few years ago, researchers in Sweden had something to celebrate: They thought they had discovered a chink in the armor of the most common type of malignant brain cancer.

In a 2014 Cell paper, the team — led by Patrik Ernfors at the Karolinska Institutet — reported that they had identified a small molecule that could target and kill glioblastoma cells — the cancer that U.S. Senator John McCain was just diagnosed with — and prolong survival in mice with the disease. 

Satish Srinivas Kitambi, the paper’s first author, who is also based at the Karolinska Institutet, said the results got the team “really excited:”

We believed we had found a molecule that raised a new therapeutic hope for patients with glioblastoma.

Given the promising results, Kitambi said, “Our [goal] was to use this molecule as a monotherapy.” Unfortunately, their initial excitement soon faded.

The study created buzz. The paper was featured in Research Highlights in Nature Reviews Drug Discovery and in commentaries in Cell and Cell Research, all of which explained why the findings could be important for treating glioblastoma; Cell called the 2014 paper “a true tour de force.”

To carry on the work, the paper’s second-to-last author, Lars Hammarström, told us that the authors performed follow-up studies “to validate the compound’s efficacy” and “to make sure we had a robust response in vivo and to probe minimum efficacious dosing, toxicity etc before proceeding with further preclinical development.”

But when the researchers attempted to test the molecule’s effectiveness in rats and other mouse models of glioblastoma, they could not replicate the original results. Kitambi explained:

We repeated our experiment multiple times and saw that the molecule only extended survival in the animals sometimes, not all the time. That was a cause for concern.

Hammarström, a senior scientist at the Karolinska Institutet, told us:

In the new studies, the control mice lived much longer than we had observed in the initial Cell study.

Hammarström added:  

We were puzzled by these results for a long time until we did careful additional histopathological analysis of the brains from the original study, which we fortunately had kept in formalin.

When the researchers re-examined the data from the 2014 paper, for “Vulnerability of Glioblastoma Cells to Catastrophic Vacuolization and Death Induced by a Small Molecule,” they identified another possible explanation for why the untreated mice didn’t live as long as the treated ones: The control mice had developed a tumor in another location in their brain, called the meningeal compartment, which the researchers believe may have caused the control mice to die more quickly. Hammarström added that the the mice in the treatment group may have also developed tumors in the meningeal compartment, but were “cured” by the compound.

Kitambi said the researchers initially missed this other tumor in the control mice because “We never looked at meningial regions … there was [no] real reason to do at that time.”

When performing subsequent experiments, the researchers modified their protocol to ensure the mice did not develop tumors in the meningeal compartment. As a result, Hammarström explained, “the life expectancy was much longer and overall survival unaffected by the compound.”

Hammarström said that the authors immediately contacted Cell to inform the editors:

We chose to retract the paper so that others would not do any additional study on the compound expecting to repeat the original results.

He added:

It’s of course a pity since the remaining 95% of data, including all in vitro data in the paper have been thoroughly validated by both us and others, but in the end a retraction was the right decision since the in vivo data was the most important piece of the puzzle.

Kitambi agreed that a retraction was “the best way to move forward:”

We feel a high degree of responsibility when it comes to treating patients. We do not want to create false hope.

Here’s the retraction notice:

This article has been retracted at the request of the authors.

Our study reported the discovery of a class of small molecules that induces the massive vacuolization and cell death of glioblastoma cells in vitro, attenuates disease progression, and prolongs survival in a glioblastoma animal model. In the process of generating additional pre-clinical data to support the transition of vacquinol-1 to the clinics, we found that we are unable to replicate the original results showing that vacquinol-1 treatment extends overall survival of mice implanted with glioblastoma cells (Figure 7U in the original paper). Retrospective histopathological analysis of the brains from the original Figure 7U in vivo study indicates that tumor growth in the meningeal compartment of mice in the control group may have contributed to the difference in survival observed between vehicle and vacquinol-treated animals. These results call into question the extent of in vivo efficacy of vacquinol-1 as a monotherapy, and we therefore believe that the responsible course of action is to retract the paper. We apologize for any inconvenience we may have caused.

The 2014 paper has been cited 56 times, according to Clarivate Analytics’ Web of Science.

Hammarström told us:

We are continuing to investigate the mechanism of action of the compound and what is causing the discrepancy between in vitro-in vivo results.

Kitambi hopes the decision to retract the paper will highlight the fact that “retractions are not necessarily a bad thing.” In this case, Kitambi felt that retracting the study and alerting readers to the replication issue would:

…improve what compounds actually reach the clinic and hopefully help researchers engineer more efficient patient therapies. Ultimately, our goal is to help patients.

Hat tip: Rolf Degen

Like Retraction Watch? Consider making a tax-deductible contribution to support our growth. You can also follow us on Twitter, like us on Facebook, add us to your RSS reader, sign up on our homepage for an email every time there’s a new post, or subscribe to our daily digest. Click here to review our Comments Policy. For a sneak peek at what we’re working on, click here.

 

Written by Victoria Stern

July 20th, 2017 at 11:05 am

Comments
  • Robert Eibl July 20, 2017 at 12:07 pm

    Interesting to see a Cell paper being retracted. Since I once had developed a children brain tumor model for medulloblastoma-like tumors in rats (Eibl, Kleihues, Jat, Wiestler 1994), it appears to me hard to beleive that anyone on the microscope could overlook any part of the brain with an unexpected growth of tumor. It appears to be difficult to position cells in the exact same part of a mouse or rat brain, and perhaps, some cells then grow in the injection canal, but one would check that. One way to achieve a good control group, when implanting from one source of cells is to use every other animal as control, i.e. when the prepared cells are, for example, stored on ice, operating and injecting one animal for one group then the next for the other, and NOT (I do not know if that might have been the case) operate fresher cells in the first group and older (longer stored) cells for the other group.

    • Hira kyiko July 20, 2017 at 2:37 pm

      I would also suggest for those using groups of 5 or 10 animals to stagger the injections. It is common knowledge cells undergo rapid apoptosis when centrifuged and stored in high numbers and small volumes ~ 30% apoptosis/hour.

      For example, preparing 1 million cells in 150 microlitres for injection requires a staggered injection protocol. Prepare enough cells for injecting say 5 mice (assuming 30 seconds/injection). THEN prepare fresh cells for the other animals.

      When harvesting, harvest in the same order the injections took place.

      • DocMartyn July 22, 2017 at 12:31 pm

        We use a flank tumor as source, remove and place in PBS, homogenize in a BeadBug then 30 sec spin. Directly into 50% ice-col MatiGel and then direct injection into the brain.
        We can do 75 mice in 150 minutes and get very tight survival curves.

        • Hira kyiko July 23, 2017 at 2:10 am

          thank you DocMartyn.

          May I ask, what are the apoptosis levels in the cells after 5 min, 30 min and an hour?

          I ask as if one injects 100,000 viable non apoptotic cells after 5 minutes and inject 100,000 cells after 1 hour which are say, 60% apoptotic, it may be reasonable to assume tumours would then grow at very different rates and the resulting datasets would not be “tight” at all.

  • Peter Muller July 20, 2017 at 1:23 pm

    Nice to see that the authors display a higher degree of responsibility (towards patient care). They could have just made corrections to their original manuscript and moved on (like most scientist). Hats off for their courage. More should live up to the standard set by them.

  • raj R July 20, 2017 at 1:33 pm

    I think these guys are excellent and responsible scientist. Their comments “We repeated our experiment multiple times and saw that the molecule only extended survival in the animals sometimes, not all the time. That was a cause for concern.”, most of them would have corrected their article and moved on.
    “improve what compounds actually reach the clinic and hopefully help researchers engineer more efficient patient therapies. Ultimately, our goal is to help patients.” Happy to see scientist retract out of concern towards patients.

  • Hira kyiko July 20, 2017 at 2:25 pm

    Why would the “control” mice develop tumours spontaneously and nobody was aware? Was it all the mice, 50%, 10%?

    I am puzzled, there should always be at least 5 groups in preclinical studies-

    1. untreated control (no administration of ANY compound). This is the true control group of the study. These animals should be healthy

    2. vehicle control (administration of same volume of formulation minus the active drug).This determines there is no effect administration route. This is the control for the drug excipients on a healthy animal.

    3. disease with no administration. This is the diseased model.

    4. Disease with vehicle(administration of same volume of formulation minus the active drug). This determines there is no effect administration. This is the control for the drug excipients on a diseased animal.

    5. disease with drug administration (even better still, high, medium and low dose treatment groups). This is the experiment to be studied.

    Without the proper controls, there is no experiment.

  • Robert Eibl July 20, 2017 at 2:55 pm

    Hira kyiko
    Why would the “control” mice develop tumours spontaneously and nobody was aware? Was it all the mice, 50%, 10%?

    Tumors did not develop spontaneously, the authors used a glioblastoma model in mice: they injected glioblastoma cells into the brain of mice. The control group did not receive the putative anti-tumor compound. With just overlooking the retracted article, I could not figure out how the authors tried to get the cells into the same region (I didn’t find the word “sterotactically”)- so I could only speculate that there is some variation in the procedure and some of the cells injected may have appeared (in addition to the region wanted, i.e. within the brain), perhaps, more outside the brain and grew at a different site than expected. For me it appears hard to understand, how tumors at different regions in the brain could have overlooked in the first place.

    • Hira kyiko July 20, 2017 at 7:05 pm

      So the “control” group was actually the disease group?

      • Robert Eibl July 21, 2017 at 3:53 am

        Yes.
        One group gets the putative treatment, the other is the control group.

        • Cancer doc UK July 21, 2017 at 11:58 am

          And what was the control for the disease group?

          Furthermore, what was the control for the excipients of the drug formulation?

          • Robert Eibl July 21, 2017 at 1:17 pm

            To compare a candidate drug one uses one group of animals which get the drug, the other not, but all animals got the tumor cells injected into the brain. So there are two groups, one with the putative treatment, one without. All animals may develop brain tumors, but the control group should either die faster (or develop faster symptoms of a growing brain tumor) than the treated group. Although brain tumors overall are rare tumors, the glioblastoma is considered the worst malignant tumor, due to its location in the brain and only limeted survival time. Therefore, many scientists try since a century to find anything helpful for these tumors. On the other hand, all other tumors are also important and – in my view – may have a much better ratio for much better results per 100 billion dollars, but that are the societies investing tax payers money.

  • Yoshihiro E July 20, 2017 at 4:48 pm

    IMPRESSIVE to see authors STANDUP for patient care, even though they could have added a correction to their paper. It was nice to read “We repeated our experiment multiple times and saw that the molecule only extended survival in the animals sometimes, not all the time. That was a cause for concern.” Usually this ends up as ERRATA or CORRIGENDUM as the molecule can prolong survival sometime. But the authors still retracted shows their respect and belief to what they want to achieve. IMPRESSIVE

  • Muzee July 21, 2017 at 4:56 am

    The authors did the best possible thing in this case, in particular since they don’t look for cheap excuses, offering a frank and perfectly believable explanation of what happened. This behaviour can be only praised. But seen in a bit larger perspective, I would like to point out an issue that is not directly raised here and that might not apply at all in this specific case, but that is often the culprit in related situations: the “hurry” to publish a seemingly exceptional, but not thoroughly investigated, discovery. Too many times scientists need to fend off the attempts of overexcited PIs to hurriedly publish some greatly promising result, which, as every experimentalist knows, always need some further, deeper study in order to be confirmed. The need to publish “before the competitors” is too often producing fake positive results, which could have been prevented just with a bit more careful screening (ie, submitting the paper a few months later). Again, probably this is not the case in this specific example, but works like this remind us of the priority that should be given to solid, reproducible results rather than just noisy headlines, especially in crucial biomedical publications.

  • Gupta K July 21, 2017 at 10:14 am

    “We believed we had found a molecule that raised a new therapeutic hope for patients with glioblastoma.” “Our [goal] was to use this molecule as a monotherapy.” “We repeated our experiment multiple times and saw that the molecule only extended survival in the animals sometimes, not all the time. That was a cause for concern.” “retractions are not necessarily a bad thing.””…improve what compounds actually reach the clinic and hopefully help researchers engineer more efficient patient therapies. Ultimately, our goal is to help patients.”
    so true, my utmost respect to you. With all that academic competition, Authors did it risking their funding, reputation and career. I wish you the best and hope more people live up to your standards. Excellent.

  • Robert Eibl July 21, 2017 at 1:00 pm

    I may not fully understand these explanations, although I understand that the (untreated) original control group died earlier due to tumor growth at an unwanted site (meningeal compartment), which escaped the original, macroscopic as well as microscopic analysis. Now, the authors claim that they could not reproduce these results, although in some cases the compound may have targeted such ectopic growth significantly, or in some cases may have worked as expected and in some cases not. Perhaps, the authors will investigate the reasons, why in some cases, there seemed to be the expected effect – and what additional treatment (radio-, chemo-, immune therapy) may help to get the compound, perhaps, working.

    • DocMartyn July 22, 2017 at 12:43 pm

      The simple explanation is that their drug worked nicely on tumor tissues outside the blood-brain barrier, but lousy for deeply embedded tumor. When you inject your cell into the brain the GBM can, and normally does, grow up along the wound track, and sometimes the tumor grows on the brains surface and kiss cancers attach to the skull. We have had GBM growing down the optic nerve or into the cheek.
      The mice that get infiltrates into the Medulla or Cerebellum reach ethical endpoints very quickly. I have inspected a mouse at 8:30 in the morning, and have it give all the appearance of phenotypically normal, and receive a sac-notice at 11:30 due to immobility; they have strokes and the cavity is filled with blood.
      Very difficult model and groups of n=7 are the minimum we use.

      • Robert Eibl July 22, 2017 at 3:12 pm

        DocMartyn
        The simple explanation is that their drug worked nicely on tumor tissues outside the blood-brain barrier, but lousy for deeply embedded tumor. When you inject your cell into the brain the GBM can, and normally does, grow up along the wound track, and sometimes the tumor grows on the brains surface and kiss cancers attach to the skull. We have had GBM growing down the optic nerve or into the cheek.

        Thank you for this explanation; so, it seems, that in the long run there is still great value of these results in order to develop a working modification of the molecule to pass the blood-brain barrier, or to get the original compound better to the tumor cells. I am not sure how solid such animal models (even with human cells in this case) are – a mouse brain is considerably smaller than even a glioblastoma in a human brain – and, perhaps, the human capillaries are slightly different or leaky in some areas of the glioblastoma in a human, perhaps also in some cases in areas difficult for the neurosurgeon to reach easily. So, it will be intersting to see the developing story again somewhere.

      • Hira kyiko July 23, 2017 at 2:16 am

        When you say “drug” – how do they know the active pharmaceutical ingredient effected the tumour at all, if they did not control for the excipients ?

        Presumably they tested this novel compounds stability?

      • Hira kyiko July 23, 2017 at 3:32 am

        Not according the the PK data the authors claim….

        ” The pharmacokinetics of Vacquinol-1 showed a maximal plasma exposure
        of 3,279 ng/ml and good penetrance into the brain with exposure
        of 1,860 ng/ml following a single oral dose of 20 mg/kg (Figures
        S7H, S7J, and S7K)”

        Reading the article we can see there were HIGHLY divergent numbers of animals in certain groups.

        “. During dissection, brains from
        vehicle treated mice displayed hemorrhage, areas of necrosis,
        and increased brain weight (n = 7), whereas brains from Vacquinol-1-treated
        mice (n = 13) showed a normal brain morphology
        and had a normalized brain weight (Figures 7S and S7I)”

  • Hira kyiko July 21, 2017 at 4:39 pm

    Robert Eibl
    To compare a candidate drug one uses one group of animals which get the drug, the other not, but all animals got the tumor cells injected into the brain. So there are two groups, one with the putative treatment, one without. All animals may develop brain tumors, but the control group should either die faster (or develop faster symptoms of a growing brain tumor) than the treated group. Although brain tumors overall are rare tumors, the glioblastoma is considered the worst malignant tumor, due to its location in the brain and only limeted survival time. Therefore, many scientists try since a century to find anything helpful for these tumors. On the other hand, all other tumors are also important and – in my view – may have a much better ratio for much better results per 100 billion dollars, but that are the societies investing tax payers money.

    This, in my view, is a VERY flawed experimental plan. Drugs are not single molecules. The active pharmaceutical ingredient is formulated with other compounds – these must be tested in the model. Drug excipients must be tested – not only is there no control for these, but the authors don’t appear to grasp the very concept of proper controls in preclinical studies.

    There is no CONTROL group using only 2 groups: 1. disease 2. disease with drug, you mention, despite calling it a control group.

    There is no experiment without proper controls. This study lacks any controls, let alone proper controls.

    • Robert Eibl July 21, 2017 at 5:18 pm

      Hira kyiko

      This, in my view, is a VERY flawed experimental plan. Drugs are not single molecules. The active pharmaceutical ingredient is formulated with other compounds – these must be tested in the model. Drug excipients must be tested – not only is there no control for these, but the authors don’t appear to grasp the very concept of proper controls in preclinical studies.
      There is no CONTROL group using only 2 groups: 1. disease 2. disease with drug, you mention, despite calling it a control group.
      There is no experiment without proper controls. This study lacks any controls, let alone proper controls.

      Please, do not judge the authors or their experiments on my rudimentary explanations. I tried to get you the point, which you appear to miss. Please read the paper instead, than referring to my simple explanations how such experiments are done. Obviously, you never worked with brain tumor models. It may help you for understanding to read their original, although retracted paper, which is still online.

  • Hira kyiko July 23, 2017 at 3:45 am

    The key part of the paper is summarised below:

    “Vacquinol-1 significantly enhanced survival as compared to
    vehicle-treated mice. Vehicle-treated animals (n = 8) showed a
    median survival of 31.5 days, whereas only two of the eight Vacquinol-1-treated
    mice died during the 80 days of the experiment
    (Figure 7U, p = 0.0004 Mantel-Cox test). Hence, we conclude
    that oral administration of Vacquinol-1 substantially impairs
    disease progression and prolongs survival”

    A p value of 0.0004.

    Clearly, this was not reproducible.

    The numbers of animals used in each group appears to alter as earlier in the article they wrote
    “During dissection, brains from
    vehicle treated mice displayed hemorrhage, areas of necrosis,
    and increased brain weight (n = 7), whereas brains from Vacquinol-1-treated
    mice (n = 13) showed a normal brain morphology
    and had a normalized brain weight (Figures 7S and S7I)”

    Were there 8 animals in the groups, or 7 or 13?

    To administer 20mg/kg is an extraordinarily high amount of drug in a preclinical study.

  • Abhi July 24, 2017 at 6:13 am

    “improve what compounds actually reach the clinic and hopefully help researchers engineer more efficient patient therapies. Ultimately, our goal is to help patients.”
    This is how all scientist should be

  • Post a comment

    Threaded commenting powered by interconnect/it code.