Monday, December 03, 2007

What Cell Is This, Who Laid To Rest?

With the holiday season upon us, I break out my seasonal music. So it is only fitting while I ponder various versions of Greensleeves that the NIH is finally starting to lay down the lay about the identity of cell cultures.

Cell cultures are great, but the problem is most of the cells pretty much look the same -- especially when growing in a culture dish. Now, I'm sure folks expert in the field will say they can distinguish many of them even without a microscope, but the fact remains that errors have frequently occurred and more than a few published studies are wrong because their cell culture wasn't what they thought it was. At my previous posting we burned a lot of effort on one large dataset that turned out to be useless for just this reason.

The prod in this case was an open letter by a number of researchers, and the response will apparently be to encourage referees to downgrade grant proposals and such which do not authenticate their cultures. 'bout time.

At Millennium we had licensed the Ingenuity database, which is a spectacular collection of biomedical facts culled from the primary literature (spectacular, but neither perfectly correct nor comprehensive, but amazing nonetheless). When gearing up for a big experiment on cell line X, we might try to pull all the knowledge of the database derived from experiments on cell line X -- the level of detail which Ingenuity provides. Of course, some of these would be contradictory -- and I found other cases where two published experiments were claimed to be precisely the same, but with very different results. The letter cites estimates that 20% of cell cultures are the wrong thing, which might explain a few of these.

Particularly in cancer research, this lack of a critical control is downright stupid. Nowadays, a somewhat pricey but powerful cell typing method are SNP chips. These will clearly disambiguate mouse from human (you won't get much signal if from the wrong species!), but can also probe deletions & copy number variation (though not balanced translocations). Given that most tumor cell lines are pretty fouled up in the DNA operations & maintenance department, expecting these cell lines to be stable is pretty unreasonable. Particularly if the experimenter is deliberately selecting for something (such as adherent growth, or growth in the absence of specific factors, or such), checking for major changes just makes sense -- and will also catch sneaky invaders who might take over such a culture.

Science published a good news article on the topic last winter -- alas, it will require subscription access.

1 comment:

felciano said...

Thanks for the encouraging words about our knowledge base, Keith! We worked hard to get cell-specific context into the Ingenuity Knowledge Base for as many findings as possible, but at the end of the day, the identity verification is directly dependent on what is reported in the publication. If this initiative takes off it would enable us to add additional experimental metadata, for example as a confidence metric on the identity. We use similar evidence codes for gene annotations, and the same approach should apply to other concepts.

It would be interesting to know what percentage of publications that have reportedly mis-identified cell lines have retrospectively retracted them or at least flagged those articles for new readers. This should be relatively easy to do (at least for online versions), but I don't recall seeing it anywhere.