Adrian Hayday refutes challenge on influential autoantibody paper by Karolinska scientists
Back in 2016, Adrian Hayday and his colleagues at King’s College London — together with some collaborators in Germany and Estonia — made a splash in the immunology world with a paper, published in Cell, suggesting that patients with a particular genetic defect produce more autoantibodies than previously known. These self-generated antibodies, they proposed, offered therapeutic potential for autoimmune diseases, most prominently Type 1 diabetes.
The results formed the foundation of a German biotech dubbed ImmunoQure, which later inked a partnership with France’s Servier to pursue an autoantibody that found to neutralize interferon-α. Servier pledged up to €164 million to the deal.
It also piqued the interest of Nils Landegren, a professor at the Karolinska Institute who’s spent years decoding the autoimmune regulator gene (AIRE) and its deficiency. While he’s helped identify a number of bona fide autoantigens in autoimmune polyendocrine syndrome type 1, the vast number of autoantibodies identified in the Cell paper caught him by surprise. And he wanted to see if there was really a link between one of the known autoantigens to Type 1 diabetes. So he took a couple of existing datasets and then ran his own analysis with a team from Sweden and the US.
Long story short: They couldn’t replicate the results.
“We could show that using random data, we got basically the same result they did,” Landegren told MedScape. “We have a very strong case their results are false.”
In a response to the paper, Hayday stood by his results and the methods brought into question, countering that “flaws in design and implementation invalidate this challenge.”
“While we welcome open, constructive discourse about science, we are disappointed by this dispute because we believe it reflects simple but important differences between our approaches that could have been easily resolved, had Landegren and co-workers approached us directly,” his team wrote.
The disputes are two-pronged. The first has to do with the number of autoantibodies APS1 patients harbor. Hayday’s group — with Steffen Meyer and Martin Woodward — as co-first authors found that collectively hit over more than 3,700 human proteins out of the 9,000 screened. But when Landegren’s team ran their samples against the same assay, they found “only a very restricted set of proteins (<1% of the protein panel) are targeted by autoantibodies in multiple APS1 patients.”
The statistical method deployed for the patient group versus the control, Landegren wrote, skewed the results to show higher levels of autoantibody signals than in reality.
To better determine how the analytical bias influenced the overall results, we applied a reversed cutoff to our dataset based on the values obtained for the 51 APS1 patients instead of those for the 21 healthy controls. The healthy controls now turned out to show greater numbers of high-level (Z ≥ 5) autoantibody signals than the APS1 patients, revealing that the skewing effect of the data analysis was dominating the results.
To be sure, the 2016 paper had a stated interest in finding patients’ “private autoantibody repertoire” — which might not be shared with others.
“When you undertake a statistical treatment of data, there is no silver bullet, you have to use what makes sense,” he said to MedScape.
Even if they apply a more conservative statistical approach, he wrote in his response, they “identified reactivities overlapping 81% with our original study: again, these comprised broadly shared autoantigens and from ~30 to~100 private specificities that collectively composed a substantial fraction of the proteome.”
Perhaps more pointedly, Landegren’s team attempted to poke holes in the correlation between having neutralizing autoantibodies of interferon-α and being free of Type 1 diabetes. Hayday’s theory was that anti-IFNα antibodies appeared to protect APS1 patients from ultimately developing that disease, supporting previous claims that type I IFN contributes to Type 1 diabetes,
Writing in eLife — where they went after getting turned down by Cell — Landegren reported strong neutralization of type 1 IFN in all APS1 samples, not just the ones from diabetics (a finding that Hayday dismissed as a result of a “shoddy experiment” conducted with less sensitive methods.)
“[A]t high concentrations, the sera of patients with and without T1D showed comparable activities, but at lower, sub-saturation concentrations [50-fold dilutions], the cohort without T1D showed significantly greater capacity to limit IFNα activity,” Hayday et al wrote in their response.
They added that the other central observations from their 2016 paper remained unchallenged: that APS1 patients share naturally arising autoantibodies to a small subset of proteins that are of extremely high affinity.
ImmunoQure is still in the CMC stage for its drug candidate targeting IFN-α, but they won’t exactly be the first. AstraZeneca recently conceded a Phase III defeat for anifrolumab in lupus, an antibody that hits the receptors for IFN-α, IFN-β and IFN-ω. The hope is that by hitting the cytokines directly — a “very different modality” — their novel pan anti-Type I IFN neutralizing antibody will work “far more effectively” than the existing mAbs produced by AstraZeneca or Genentech, Hayday added.