Retraction Watch

Tracking retractions as a window into the scientific process

Paper linking cell phones during pregnancy to behavior problems in mice corrected

with 15 comments

scientificreportsThe authors of a study published last year looking at the effects of cell phone exposure on mice in utero have corrected a figure after it was questioned. New experiments, they write, confirm the original conclusions they drew from the figure.

Here’s the corrected figure from the paper in Scientific Reports, published by Nature Publishing Group:

The authors identified an error in the determination of the maternal corticosterone levels as presented in Figure 4. The ELISA was repeated using stored serum samples from the original experiment. The levels have been corrected and are shown below in Figure 1 (the revised Figure 4). The mean corticosterone level in the pregnant control females was 5.4 ng/ml and in the exposed female mice was 6.1 ng/ml. There was no significant difference between the corticosterone levels of the control and experimental groups. The correction does not change the conclusions.

phones mice

Corticosterone levels were measured on pregnant mice on day 12 of gestation. The mean corticosterone levels in pregnant control mice was 5.4 ng/ml and in exposed pregnant mice was 6.1 ng/ml. The differences were not statistically significant (P=0.786 by T test). Twelve representative mice housed separately were used, six from each exposure group.

Here’s the original Figure 4, which the authors used to rule out the effects of stress, rather than the cell phone radiation, on the mice:

fig 4

The mean corticosterone level in the pregnant control females was 69.94 ng/ml and in the exposed female mice was 69.91 ng/ml.

Alexander Lerchl, who has questioned other cell phone risk studies, noted in a comment on the study that

Figure 4 (corticosterone data) is critical for the whole manuscript, as can be read in the text: “In order to exclude the possibility that impaired memory and behavior in exposed mice was caused by stress resulting from experimental manipulation, we measured serum corticosterone levels on day twelve of gestation using an ELISA assay. The mean corticosterone level in the exposed female mice (69.91?ng/ml, n = 6) was not significantly different from that in the control females (69.94?ng/ml, n = 6) [Figure 4], eliminating stress as a source of the observed behavioral and electrophysiologic differences.

I have analyzed the data and the extremely small variations with the help of a ruler and found that the coefficients of variation in both the control as well as in the exposed animals is approximately 8%, thus in the same range as the unavoidable “noise” of this assay (so called intra-assay-variation), according to the manufacturer. In other words, if the data were correct, there was virtually zero biological variation within both groups and virtually zero difference between all animals’ values. And this for a stress marker which is known for being extremely variable within and between individual animals.

I have worked with assays like this one for many years, and I have never seen such small variations.

The original study garnered some media attention, and has been cited by one other paper, also in Scientific Reports, according to Thomson Scientific’s Web of Knowledge.

Written by Ivan Oransky

February 18th, 2013 at 11:01 am

  • Prof. Dr. Alexander Lerchl February 18, 2013 at 11:14 am

    Thanks for reporting. This is NOT the end of this story, promise.

  • jimgthornton February 18, 2013 at 1:54 pm

    The original is shockingly bad. The whole enormous figure with all that black ink provides only two data points – even the error bars aren’t defined – and since both are also in the text, the figure contains no new data. An infinitely small data:ink ratio. Is this another example of the low editorial standards, even of publishers like Nature, when they go Open Access?

    • Tim D. Smith (@biotimylated) February 18, 2013 at 5:32 pm

      This kind of plot (sans errors, one hopes) is about par for the course for the biological sciences; you see them everywhere. One of the world’s small tragedies.

  • StrongDreams February 18, 2013 at 2:16 pm

    Nature publishes based on “sex factor” (controversy, timeliness, hot button issue of the day), not science.

    • Marco February 19, 2013 at 1:19 am

      This is a NPG journal, not Nature itself. Scientific Reports even supposedly does not take potential impact into account, but only checks whether the paper is technically sound. It’s the PLoS One approach.

  • Sylvain Bernès February 18, 2013 at 4:33 pm

    I note that in the original Figure 4, corticosterone level scale is given in mg/ml, while the revised figure has the same axis in ng/ml… Well?
    Or maybe, there is something I’ve not understood.

    • littlegreyrabbit February 18, 2013 at 5:07 pm

      I am guessing freezing-thawing may have destroyed much of the steroid – something in the order of 60-70 ng/ml or mg/L might have been unsurprising

      There are some reference ranges in section 7 here – it also gives an indication of the diurnal variability that Alexander Lerchl talks of.
      Unless there is something I have not understood

      • Tim D. Smith (@biotimylated) February 18, 2013 at 5:30 pm

        The values given in the paper are in ng/ml; mg/ml must be an error in the figure legend (sloppy, but it happens). I’m concerned that the new values are still tenfold lower than the original values. I wonder if they made a mistake in their calibration curve (!) or if they actually lost most of their steriod during the freeze-thaw? I might not have expected them to be *that* fragile.

      • littlegreyrabbit February 18, 2013 at 5:46 pm

        D’oh ng/ml or ug/L – but Tim seems to have sorted that issue out.

        According to the ELISA spec sheet I linked to
        “8.1 Analytical sensitivity
        The lowest analytical detectable level of corticosterone that can be distinguished from the Zero Calibrator is
        4.1 ng/ml at the 2SD confidence limit.”
        About half of the values are below that.

    • Stewart February 18, 2013 at 5:09 pm

      If I may also add, in the dotplot – there are only five in the experimental group, not 6, as the authors wrote.

      In the histogram, of course, we have no way of knowing.

    • StrongDreams February 18, 2013 at 5:24 pm

      mg/ml is pretty impossible. It could be that it was supposed to be 60ng/ml (labeling error), or that they really calculated 60mg/ml because there was a huge mistake in how they did the assay. We don’t know what specific error was called to the author’s attention, so it’s hard to know anything for sure. (I suspect sample #6 overlaps one of the others and the plot should have been edited.)

  • Terence Pang February 18, 2013 at 7:21 pm

    I have to agree with Lerchl in that over the years of performing corticosterone assays, I have never come across standard errors of that minute a magnitude. This is not the first time I’ve come across this issue, in fact, I frequently query corticosterone data with small variance during reviewing of manuscripts and request for dot plots. The exception to this is for dexamethasone suppression test data.

    I find it impossible to comprehend how a re-sampling of the samples could yield such different values. Corticosterone does not degrade that quickly, and I have previously managed to obtain similar values after 3 freeze-thaw cycles.

  • Prof. Dr. Alexander Lerchl February 20, 2013 at 8:27 am

    In addition to the many strange features of this story the corrigendum adds one more: the “new” data (re-analysis of the serum samples for corticosterone values) were known since September 1, 2012. How come that the corrigendum has been published only now???

  • Post a comment

    Threaded commenting powered by interconnect/it code.