Skip to Content

Biological News

Cancer Cells Are Even Worse Than We Thought

There are a lot of cancer cell lines out there, and many of them get used a lot, too. It’s not surprising, in a way, because these are cells that have already (and unfortunately) proven themselves to be robust and fast-growing, so many of these lines tend to take to cell culture conditions pretty well. In the case of HeLa cells, too well – over the years, there have been more cases than you’d like to think about where HeLas have contaminated other cell lines and crowded them out without the the researchers involved noticing for a while. That doesn’t happen as often as it used to (cheaper sequencing and people in general being more aware of the problem), but it’s always out there.

As you’d imagine, these cells lines are very important in oncology drug development, being located way up near the top of the screening cascade. They’re used both to see if you’re affecting the sorts of tumors you hope to target, and to get some idea of selectivity across different types. Now, everyone knows that these lines don’t really reflect what’s going on in an actual tumor – it’s a different environment in general, and after cells have been in culture a while they tend to be different beasts than they were at the start – but it’s still a lot better than running no cell assay at all.

Here’s a new paper, though, that goes into detail about just how different those cells are, and not so much from the original tumor samples, as from other batches of what are supposed to be the exact same cells. That’s not so good. Cancer cell lines tend to be genomically unstable in general, which is a problem, but here are the numbers for just how much of a problem that is. This team looked at 106 tumor cells lines and found that there’s a very large amount of heterogeneity in them when examined closely (deep sequencing, RNA-seq, cell painting assay, and more).

Our results show that established cancer cell lines, generally thought to be clonal, are in fact highly genetically heterogeneous. This heterogeneity results both from clonal dynamics (that is, changes in the abundance of pre-existing subclones) and from continuous instability (that is, the appearance of new genetic variants). Moreover, genetic heterogeneity leads to varying patterns of gene expression, which in turn result in differential drug sensitivity. These findings have a number of important implications. . .

For example, they took 27 different samples of what are all labeled “MCF7” cells (a widely used breast cancer derived line) and compared 321 known oncology compounds across them. At least 75% of the compounds that showed strong inhibition of one MCF7 line were totally inactive against others. That’s going to confound experiments big-time, and this paper is a loud warning for people to be aware of this problem and to do something about it. On the flip side, this heterogeneity can be an opportunity to learn more about cancer cell biology and compound effects on it – but only if you’re aware that it exists!

38 comments on “Cancer Cells Are Even Worse Than We Thought”

  1. E. Briand says:

    Deficient DNA repair machinery is a hallmark of cancer, but is it necessary for ongoing survival once the cells have acquired the qualities that make them good for the lab ?

    Could we restore repair function, thus genomic stability, on existing cell lines ?

    1. Peter S. Shenkin says:

      And if we can do it in vitro, can we do it in vivo?

      That is, stabilize the genome of an actual tumor, in order to maintain the effectiveness of a drug known to work against its current genome.

      1. Hannah says:

        This is a bit of a double-edged sword, since many therapies used against cancers (intercalating agents, radiotherapy) involve *causing* DNA damage that will hopefully kill the unstable cells more quickly than the stable ones, hence the unpleasant side effects. You’d have to have a combination therapy which is entirely free of those DNA damaging treatments in order to take advantage of DNA stabilising agents.

  2. Mobio says:

    This is well known to any cell biologist. Cells in culture change and even ‘clonal’ cells are not necessarily clonal.

    1. a. nonymaus says:

      Certainly this is the case for tumour cell lines. Is the situation any better with allegedly clonal non-tumour and/or stem cell lines?

  3. al says:

    The cynic in me is wondering why this experiment wasn’t done long ago. Yes it’s expensive, but not as expensive as the money that was spent chasing results of dubious quality. Same holds true for iPS cells and organoids – deep characterization not being done because it’s much higher priority to the practitioners to get papers and funding (I guess it’s understandable, if in their shoes). And I cannot agree that we will learn more about cancer and compounds by studying these types of “cancer” cells. Time to think different.

    1. secret sauce says:

      yes – and it makes me wonder what the senior author (Golub) would have to say regarding the value of the LINCS transcriptomics data set, of which he is a PI. NIH has invested millions of dollars on LINCS (and still counting). The caveats about transcript copy number vs. protein expression levels is one thing. Genomic instability of the cell lines, underlying the transcriptomics, is yet another. When it comes to biology or systems biology, it feels like we keep spending money on “houses built on sand”, rather than getting the foundation(s) right first.

      Comments from the computational biologists/bioinformaticians in the audience?

      1. bks says:

        I worked with plants, so I have nothing to say specifically about cancer cell lines. But in my experience molecular biologists could always get rave reviews, funding and FTEs from C-level executives for the most frumious “leads” but exposing a show-stopping problem with a million-dollar experiment was received with frosty radio silence. This is somewhat at odds with the CS tribe where skill in finding bugs is treated with the utmost respect.

        1. Matthew K says:

          It’s not hard to see why flaky leads are treated like gold and exposing flakiness is shunned, if the reward system is geared to generate investment. Scepticism is not something people feel they are likely to turn a profit on. Rigor and reproducibility runs a very distant second to hope / saleability.

    2. biology *sighs* says:

      Bit late to the comments, but I would like to add the entire field of host-microbe interactions to this list. The area is ripe for the funding, but we’re skipping some really basic-science that is needed to characterize our models. Apparently folks don’t want to spend extra NRG studying microbial physiology or metabolic differences in gnotobiotic animals when NIH is happy to fund them to simply gavage mice with poop from cancer patients and sequence the 16s of what comes out.

      Garbage in; garbage out.

  4. Niek says:

    I guess there is value in screening against multiple “identical” cell lines in that case. If the cells can mutate to be resistant against a drug in the cell line form, the corresponding tumor will probably be able to do the same. If something kills all examples of a particular cancer cell line, my gut feeling us that will be a good sign.

    1. Nesprin says:

      Yep, my handle has a link to a paper by a collaborator that did exactly this looking at MCF10a subclones. In this paper, the difference between subclonal lines was not due to genomic differences- ATM levels were variable due to epigenic differences.

  5. Wavefunction says:

    In general we aren’t very good at taking heterogeneity into account.

    1. John Wayne says:

      On that note, we demand that you stop thinking and comply. Repeat after us, “the screening funnel will provide a drug.”

  6. Wavefunction says:

    Ok, “the screening funnel will provide a dud, I mean, drug.”

  7. Daen de Leon says:

    I wrote a review of literature on mechanisms of cisplatin resistance in 2002 where I noticed exactly this problem: A2780 (a human ovarian cancer cell line) seemed to be particularly popular — and especially variable.

  8. MrRogers says:

    This has long been known and even taken advantage of. A lab that I looked at for grad school 25 years ago was studying drug resistance by culturing MCF7 cells in various chemotherapeutics to induce resistance and then looking at what changes they observed in the cells. Unfortunately, this also occurs in patients. Careful analysis over the last few years has demonstrated that treatment failure in CML typically results from the outgrowth of preexisting resistant clones. I don’t imagine it’s much different for other cancers.

    1. E. Wold says:

      Interesting comment. More thought might be needed towards the difference between emergent (cultured) heterogeneity and intrinsic (pre-existing) heterogeneity. Especially given that many drug-resistant patient cancer samples are thought to have been pre-existing subpopulations.

  9. JB says:

    Wouldn’t you actually WANT a heterogeneous population of cells derived from the the same cell line that randomly occurs over time? After all, that’s what exists in cancer, so if your drug works against the worst case scenario of a genetically complex group of cells wouldn’t it stand to reason it’d have a higher probability of working better against a tumor that has a complex set of cells?

    Besides being genomically unstable, cells have all sorts of pathways – Hippo for example – that are mechanosensing and can alter transcript profiles simply in response to how dense cells are on a dish. Thus, even if you had a homogeneous population of perfect cells, you could still get variance in the results simply because it is impossible to culture and passage cells the exact same way 100% of the time to the same density.

  10. AI says:

    To an experienced biologist, it is nothing surprising and worth worrying too much. In the end, your drugs will be characterized against a panel of cell lines based on your hypothesis, not a single one. We do have pay extra attention to publications built around only a couple of cell lines.

  11. sgcox says:

    Interestingly, all HDAC inhibitors and Proteosome inhibitors worked across all MCF7 clones, in contrast to other drug classes. Still, none are approved for breast cancer yet…

    1. Old Pump Kicker says:

      My hunch is that, after HDAC / Proteosome inhibitors become widely, harvested cell lines will start showing varying sensitivity, even if the donating patient didn’t receive the inhibitor. When each tumor is already multiple genomes, what’s a few more?

    2. john adams says:

      (Good) HDAC inhibitors are too toxic against (many) non-transformed cells….

  12. Anon the 3rd says:

    Oh that’s why my PI’s compounds work sometimes!

  13. steve says:

    All tissue culture is artifactual. Cells are grown stuck on plastic in a 2-dimensional configuration that has nothing to do with how they grow in vivo. Serum, which is a pathological fluid only found in wounds, is used to help them grow. They are grown in ambient oxygen, instead of the hypoxic conditions usually found in tumors. Techniques also vary widely from lab to lab. So the fact that some lines are heterogenous and show genetic drift is hardly shocking.

    1. Inconvenient truther? says:

      Well said

    2. Anon says:

      You can’t get a nature paper with this attitude!

  14. Inconvenient truther? says:

    Well said by Steve

  15. Kling says:

    Decades ago I ran a CRO that offered cell line testing service. A Big VC firm asked us to reproduce the survival / kill curve of a Secret Peptide invented by a Big Name Institute. So we got the ATCC tumor line and showed Secret Peptide had No Effect. We got the “tumor” cell line from the Big Name Institute lab instead, and the Peptide worked marvelously, miracle cancer drug verified. The VCs spun out the the company got buckets of cash, and ditched before the company crashed into smithereens.

  16. Anonymous says:

    Paywall on the paper, but I get the gist from Derek’s description. I especially appreciate the report of non-uniform response to drugs across different samples of the “same” cell lines. I also agree that there is still info to be gleaned by these cell based assays, just maybe not the easy pickings which can be either misleading or completely wrong.

    I think that many here have understood the basic message for MANY years. Cancer cells are not stable. I read Gerald B Dermer’s “The Immortal Cell” (Avery Press, 1995) shortly after it came out 23+ years ago. Dermer is a pathologist who has been looking at cells from cultures and from real tumors his whole life. He said that cultured cancer cells are really screwed up and almost irrelevant to finding a cancer treatment in vitro.

    You only need to read the wikipedia description of HeLa cells to know that something is weird. Some consider that the descendent cells are not even a human cell species anymore. (“However, this proposal has not been taken seriously by other prominent evolutionary biologists …”)

    I will try to tie this in with “Med-Chem Labs: What’s Changed?” from 27-Aug. Many things have sped up in research, including the replacement of rodents with cell lines. “More data! We need more data! ASAP! ‘Kill Kill Faster Faster!’ ” But when you are churning out numbers from unreliable or irreproducible assays, it’s a waste of time and money. Some (e.g., the authors of the cited article) can get away with questioning it but many here have reported the demise of the messengers. Kill Kill Faster Faster, indeed.

  17. This is not news to experimental oncologists. Four levels of heterogeneity:
    – Some years ago the Cell Line Encyclopedia group published that even the presumptive ID of cancer lines based on a snp fingerprint was wrong most of the time
    – There is dynamic chromosomal heterogeneity even in a single dish of cancer cells. Not too surprising given the incidence of incorrect mitotic events, and well demonstrated by highly accurate counting techniques such as Sequenom and NGS giving non integer copy numbers for genes from cells in a single dish
    – Epigenetic changes alter growth pathways and hence susceptibilities
    – As recently understood bettereffects of things like PIK3CA inhibitors vastly dependent on the metabolic state of the cell.

  18. Cialisized says:

    @Peter Shenkin: Funny you ask: Apogen is a new biotech startup–targeting cancer cell resistance mechanisms (APOBEC pathway)
    See link in my handle.

  19. Barry says:

    For a price, you can test your drug candidates against (heterogeneous) primary tumor isolates (biopsies and excised tumors) in nude rates. But it’s hard for a reader/referee/the FDA to compare your data to anyone else’s

    1. Barry says:

      Aaargh. “nude rats” not “nude rates”

  20. Pranab says:

    In a process where failure rate is >99%; how much concern we should carry for one or two potholes here and there. How well a molecule can withstand all the potholes to survive till the end, will eventually be a “drug”. Isnt this one of those potholes. Not any of the single step in the full screening cascade, from designing a molecule to all the way when it gets approved, is perfect on its own term. So why bother for a cell line heterogeneity. Although its good to know, but it hardly helps anyway.

  21. PJI says:

    None of it is surprising – but it does mean as a field we need to be more rigorous in ensuring we understand the system we are using. Fortunately the tools for profiling the results of this inherent genetic instability and sensitivity to conditions are becoming ever more available/affordable.

    It just needs to be built into workflows, with the results analyzed in the context of the actual system we have been using rather than say assuming every MCF7 looks like the CCLE snapshot.

  22. swattie91 says:

    Categorical statements that cultured cells are irrelevant to tumors are as false as the belief that they are perfect models for cancer. As always, the specifics matter and the truth is somewhere in the murky middle.

  23. Vladimir Nikolayevich says:

    Cancer cells (and lines too) survived under immune selection pressure. Cancer cells are clever than you thought. Stop counting inevitable numerous mutations, care about cure.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.