Today I will be coming back from our little camping trip (hopefully!) Until I get back to my labtop, I'll entertain you with another post on microscopy.
Flipping through Nature, I stumbled onto this commentary: The good, the bad and the ugly. Here's a taste:
To correctly capture images using a modern microscope, researchers must have a good grasp of optics, an awareness of the microscope's complexity and an obsession for detail. Such skills can take months or even years to master, and yet, owing to inexperience or the rush to publish, are all too often squeezed into hours or days. Popular methods such as fluorescence microscopy are particularly fraught with dangers.
The problem? Here's more:
It is now a routine part of many studies to investigate, using microscopy, where in the cell a fluorescently labelled protein is concentrated and where it goes. This type of microscopy has hooked cell biologists because it allows them to gaze inside living tissues and monitor molecules in their native environment. But although most biologists graduate with some training in chopping and splicing DNA, few will have laid their hands on a pricey fluorescent microscope. "Your average molecular biologist can make all these fantastic fluorescent tools," says Kurt Anderson of the Beatson Institute for Cancer Research in Glasgow, "but then imaging is just a little bit tacked on the end."
Anyone who has hung out with me, knows that I have gone apoplectic when it comes to microscopy. Most of the microscopy data in papers I read is just poor AND IT DRIVES ME CRAZY. And it also drives my microscopist friends crazy too. I have resisted writing too much on this topic (here is a rare exception), but now that I've read this article I'll have to rant a bit ...
OK let's start:
1) As far as I'm concerned, most qualitative microscopy lies at the bottom of the hierarchy of data. Poorly controlled and poorly presented microscopy is far too common. And the data is often overinterpreted. It drives me nuts. 95% of the endoplasmic reticulum staining in papers look like amorphous blobs. How can you derive any useful information from that? And then people adjust the brightness/contrast levels so that the images are what we call overcontrasted. And then there are the journals ... why are your images so freakin' small????? Microscopy is only valuable if can derive information, if your images are uninterpretable and do not add anything to your argument, why bother?
From the article (that quotes Jennifer who is right downstairs):
... one expert contacted by Nature estimated that as many as half of all experiments that report two proteins in one spot have not been performed properly. Another estimated that 5-10% of images don't match what is reported in the text. "It's easy to pick up any journal -- even Nature -- and see poor microscopy data," says Jennifer Waters, who directs the Nikon Imaging Center at Harvard Medical School. "I don't know how often the results are blatantly wrong, but I do worry about the accuracy."
2) Colocalization. From the article:
One of the most common uses of fluorescence microscopy -- and therefore the source of many problems -- involves looking for two proteins labelled with different coloured tags in order to determine whether they sit in the same place in a tissue or cell. Each fluorescent protein is excited by a particular range of wavelengths and emits at different wavelengths that are collected through microscope filters. If a researcher uses GFP in combination with a tag that emits red light then, in places where the two proteins are close together, combining digital images of these two tags will create a yellowish signal.
I hate it. Just because you flash some "yellow" at me, it does not follow that the green and red proteins colocalize!!!!! Please perform controls. Make sure that each protein/immunostain does not fluoresce in multiple channels. Bleedthrough is a horrible artifact. And here is another request show the distribution of each protein separately and please show the distributions of each protein with black and white images. This brings me to #3 of things that bug me.
3) Color. From the article:
.. journals and reviewers are too often impressed by pretty images. North tells the story of one postdoc who slaved to capture images of her small cells only to have the paper turned down because the reviewer said the images were not good enough. "I've seen how long she spent getting the highest quality images she could possibly achieve," says North. "I think it's a big problem when the reviewers are more concerned with how aesthetically pleasing an image is than whether the scientific content is clear."
Yes colour pictures are pretty, but they are awful when it comes to presenting data. If you are presenting the distribution of a single protein there is no need for color. This practice is wrong. Your eyes can distinguish more shades of grey then shades of any color (especially dark blue on a black background). I hate color. I hate color. Data is more important than esthetics.
4) FRET. ARRGGHHHH! For those that know me, you'll laugh because I've been ranting about this forever. This passage made me happy:
Microscopists save most of their expletives for more sophisticated techniques such as FRET (fluorescence resonance energy transfer), one of several four-letter acronyms for methods that are both popular and treacherous.
The idea is that you activate fluorophore A it will donate its energy to a nearby fluorophore B. So if A and B are close you should be able to flash A and see B. Sounds nice right? The problem is that this FRET signal is usually very small. We're talking about fluorescent changes of 5-10% in the best case scenario. And to calculate the amount of FRET you need to know how much A and B you have at any one spot on the coverslip. Add on background and bleaching effects and you can see why many microscopists call FRET the "photoshop effect". The only FRET that I can truly believe is cuvet FRET where the entire spectrum of both fluorophores can be analyzed. That is the only way to ensure that the spectral properties of each fluoropore have been altered.
The answer to all these problems? From the article:
... many biological questions simply cannot be answered without a working knowledge of microscopy. "In competitive biology, you don't necessarily need to be a mechanic but you need to be able to operate the machine," Davidson says. "If you don't know how it works you'll get creamed in the race."
Translation: get educated, get trained, be careful, and don't overinterpret your data. And please follow some basic principles like performing controls.
You are now warned.
interesting that a journalist (disclaimer: i'm one too -- though once a microscopist) is raising this issue.
are journals enforcing quality control on fluorescent microscopy, or are they so worried about intentional fraud that they've ignored perhaps a larger problem?
Ewen Callaway, formerly of Complex Medium? How the heck are ya? Hope that J-school is treating you well.
Alex, when you are back I'd love to hear your comments on the virtues (or lack thereof) of quantifying a single fluorophore, say, in the nucleus that increases or decreases in response to drug treatment.
good -- i'm done with j-school and interning at nature in washington, writing a lot of online news (www.nature.com/news) and the occaisonal magazine story. I really should be doing the carl zimmer thing, expounding on published stories online. It's a good model for journalist bloggers, I think.
alex, scifoo looks fascinating. one of our editors is there.
You are completely correct, Alex. The problems become even more extensive when you consider electron microscopy where shoddy staining, selection bias on behalf of the researcher, and half-hearted attempts at quantification make many of the articles that rely on such techniques unreadable. When I was doing EM for some neuroanatomical work, we would search for months for a single image - perhaps a strange synapse, or some sort of stain colocalization - and then the PI would say, "OK, now we can publish."
Confocals with spectral detection systems (Zeiss and Leica at least make them) can be used for FRET in situ if done correctly, and with the proper controls. But yeah, you better know the relative amounts present... for genetically encoded FRET sensors this means they have be on the same protein, e.g. a few cAMP and Ca2+ detectors that are available.
I'm particularly fond of people who claim colocalization when they see yellow on confocal projections.
Such is the state of the review process... editors are more taken up with whether or not something is flashy and has "mechanistic insight" than finding reviewers competent (or willing) to assess the actual quality of the data. Who was it who said "Not everything published in Cell is wrong"?