One issue that concerned me was that proteins were identified by mass spectroscopy from healthy and sick honey bees. However, only non-honey bee proteins were described. One would expect the vast majority of identified proteins to be honey bee proteins. If we took a skin sample from someone and did mass spectroscopy to see what proteins were there, we would find a ton (figuratively) of human proteins and a pound of microbial proteins from microbes that were living on the piece of skin. A response from Leonard J. Foster highlights how excluding the vast majority of proteins (honey bee), one greatly increases the detection of false positives. This goes a long way to explaining why the most abundant viral proteins were rarely detected. The more a specific kind of protein is in a sample, the stronger the signal you get from the specific protein (actually we are dealing with protein fragments not entire proteins, but the point is the same).
The reason for this is a signal-noise problem. In these types of approaches, you want to remove the strong signals, which overwhelm the data set, to identify the weak signals. The problem is that as we get closer to the limit of detection the difference between a bona fide weak signal and garbage (aka noise) becomes negligible. Think of an eye chart. Your eye is the detector and you want to identify letters. At the top, most people can identify the 'E', which has a strong signal because its so damn big. As you go down the chart it becomes more difficult to identify the letters. With my glasses on I can identify all the letters with 100% accuracy. Without my glasses and from a distance of about 18 inches, I got 1 wrong on line 6 (~83% accuracy), 3 wrong on line 7 (~57% accuracy), and all wrong on line 8 (0% accuracy). So as the signal got weaker, smaller sized letters for my eye detector, my accuracy was less. Importantly, I could see there was something there, but what I interpreted the letter to be was wrong. So, if we asked a population of people to identify these letters, there would be a strong detection of 'E' 'F' and 'P' in lines 1 and 2 and poor detection of 'D' 'O' and 'C' in line 8. The noise comes in because if we ask people, our detectors, to come up with answers on line 8, instead of 'D' 'O' and 'C', we may often get 'P' 'Q' and 'O' respectively. This is our noise. There's something there and we know it, but it is below our ability to accurately figure out what it is. All those Iridovirus sequences identified by Bromenshenk et al may be noise.
So why was IIV6 identified Bromenshenk in the first place? In the October 2010 paper, mass spectroscopy was used to identify proteins. Basically, portions of the amino acid sequence that make up the protein were identified. With these amino acid sequence, we can search protein databases for things that match these sequences. The only genome (DNA) sequence for an Iridovirus that is available is for IIV6. Thus, we can deduce all the protein sequences encoded by this virus. So the amino acid sequences identified by Bromenshenk appear to be most closely related to IIV6 proteins, as opposed to octopus proteins. (see Foster's short rebuttal for why this may likely be an artifact).
To test the idea that IIV6 or an IIV6-like virus is associated with CCD, Tokarz et al tried to detect viral DNA in sick honey bees (and healthy bees, which are predicted to lack the virus if the hypothesis is correct). They made primers that would detect IIV6 directly as well as related viruses based on what we understand to be the most conserved viral proteins. They also made primers to detect IIV24 directly. What did they get? Nothing, Zip, Nada, Zilch. They detected a positive control for honey bee genomic DNA, but obtained nothing for the viral sequences. They also conducted positive controls to show that they could detect viral sequences if it was added to honey bee DNA prior to the PCRs. This sounds like a conclusive result that Iridovirus are not associated with CCD.
However, even though this dataset supports my trepidation of the Bromenshenk conclusions, I am concerned that this study lacks a critical control. I would be much more convinced if the authors were able to detect at least some kind of virus in any of the honey bee samples, even some unrelated endogenous retrovirus. Just something to show they could detect a viral signal from actual samples, not mocked infected samples. The authors did note that it's possible the sick honey bees are infected with an Iridovirus that cannot be amplified with the primers used, which is true.
Finally, I want to hit on an issue you often hear about in the sciences. The myth that 'you cannot publish negative results.' This paper is in fact nothing but a negative result. Yet it is published. Now the authors use other studies to bolster their conclusions, so in fact this work is not simply a negative result. It's a negative result in context. There are many papers that are essentially negative results, although they are cast in a way that is positive. Now you probably won't see papers like this in Science, Nature or Cell, but you also won't find them retracted as often either.
Tokarz R, Firth C, Street C, Cox-Foster DL, & Lipkin WI (2011). Lack of Evidence for an Association between Iridovirus and Colony Collapse Disorder. PloS one, 6 (6) PMID: 21738798