You may have heard a big buzz around the recent theoretical cancer study, which is trying to explain why probability of malignancies from some tissue much greater from the others. And the reason is a number of stem cell divisions in particular tissue. Study made very provocative claims and triggered huge wave of discussions about its methodology and interpretation. Significant part of discussions on the blogs was about statistical methods and interpretation. You can read some good posts here, here, here, here and here. It also caused very good discussion on PubMed Commons. However, it seem to me, not much attention was paid to author’s strategy to calculate the stem cells number and their divisions in every analyzed tissue. I’d like to bring your attention to few posts, which touched potential problems with stem cell frequency calculations in different tissues. I’m in agreement with these comments:
The most difficult issue is how to consistently and accurately calculate the number of lifetime stem cell divisions in a series of diverse tissues. This is extremely difficult and it’s not difficult to imagine such calculations being off by one or more orders of magnitude.
Another factor is that not all cancers are going to originate with stem cells. Some tumors arise from progenitor cells or from differentiated cells that de-differentiate.
Further, many organs have more than one type of stem or precursor-like population. How does one handle that in modeling?
from the Signals blog:
It is also worth mentioning that the majority of the data used in the estimation of the total lifetime stem cell divisions are largely derived from mouse models. In the current paper, the authors identify that mouse and human intestinal tissues are not equivocal (supplementary data); so then, why are mouse data used in the other calculations? The data that are of human origin are typically from in vitro (in a dish) or mathematical models. While animal, cellular, and computer modeling are acceptable for basic sciences research, applying those conclusions to whole-body human beings is a little difficult (and not good science).
“Acute myeloid leukemia” section, they described the reports claiming that the cell division cycles of the stem cells are 15, 17, 30, 57 days and the author of the current manuscript approximated it as 30 days. However, the reference number (38) is from the experiment of mouse, not human. For (37) I am not sure but as carcinogen BrdU was used, it should not have been a human experiment. Since the cell cycle of mouse is ~ 2 times faster than that of human, it violated the approximation with growth and is also be able to explain an apparent power law of the model. It is also notable that more than 3-fold differences in the reported cell cycles violate the approximation in this model.
To illustrate how author’s calculations of stem cell number and divisions could be flawed, I’d like to cite “acute myeloid leukemia” part from supplementary materials:
The lifetime incidence of acute myeloid leukemia is 0.41% (www.seer.cancer.gov) (3). There are a total of ~3·1012 blood cells and 7.5·1011 nucleated cells in the bone marrow (32). 0.18% of these normal bone marrow cells are CD34+ (33). The CD34+ are a heterogeneous population of cells and only about 10% of the CD34+ are hematopoietic stem cells (HSC), phenotypically defined as (Lin−CD34+CD38−CD90+CD45RA−) (34). The number of hematopoietic stem cells therefore equals 7.5·1011· 0.0018 · 0.1 = 1.35·108. Hematopoietic stem cells have ben estimated to divide every ~15 (35), 17 (36), 30 (37,38), 57 (39) days. Thus, we assume they divide every 30 days.
First, hematopoietic stem cells within CD34+ is defined population, however there are other HSC populations, which are CD34-. Second, they picked Weissman’s group 2011 study, to define phenotype of human HSC. It’s not very clear why this study was picked, if more recently Dick’s group showed that human HSCs could be either CD90+ or CD90-. Why don’t use the most recent and accurate study? Third, all cited studies, which calculate HSC divisions rate, were done on mice. How authors can extrapolate it to human? They took 4 numbers from mouse studies and simply calculated average (15+17+30+57)/4 = 30. Assumption = mouse + mouse + mouse + mouse/4 = human. The authors did bad work with literature. They can find very nice 2011 simulation study by Sandra Catlin, which estimate a rate of human HSC replication as 1 time in 40 weeks. Now compare: every 30 days versus every 40 weeks.
This is just an example from AML. If you look at other tissues, you can find comments about the same issues. For example, a comment from PubPeer about testicular cancer:
The estimation for the number of stem cell divisions is, in my opinion, highly speculative.
Looking at how testicular stem cell divisions were calculated (my particular field of knowledge), there are so many assumptions being made, that I wouldn’t be surprised if the calculation was off by many orders of magnitude. I wouldn’t be surprised if the calculations for other tissues was subject to similar margins of error.
Some of the dubious assumptions:
1) A-single spermatogonia are the stem cells of the mouse, which is contested; it is even unclear whether there is a fixed, deterministic pool of stem cells present.
2) Human spermatogenesis is essentially a scaled-up version of murine spermatogenesis, for which there is no data;
3)The number of lifetime stem cell divisions can be deduced from the number of total sperm produced in a lifetime
Overall, the authors are making a number of assumptions, which are not reflecting reality:
- Data on tissue adult stem cell number in human are available and valid. Not true! There are very limited data on number of defined stem cell populations in adult human tissues. Most of these studies are hypothetical and their validity is unknown. That’s why, for example, the authors were not able to find good studies on such frequent cancers as breast and prostate.
- There is only one well defined stem cell population in human adult tissues. Not true! There are different stem cell populations within tissues. For example, for hematopoietic tissue I cited a few above. Markers for some of them are overlapping, for some are unique. The question – which one to pick for calculation?
- There is only one valid study, describing well defined adult stem cell population. Not true! There are many studies from different groups, most of them are conflicting! The question – which one to pick?
- Mouse = human. Not true! Stem cell calculations for mouse cannot be used for math models in human.
- Only divisions of stem cells responsible for carcinogenesis. Not true! It is well known that progenitor cells are more rapidly dividing then adult stem cells. Due to rapid division history, some progenitor populations in some tissue are more prone to accumulation of mutations and carcinogenesis. Both differentiated cells and progenitor cells could acquire self-renewal via mutations and become cancer stem cells. Most adult stem cells are quiescent. Stem cell quiescence could be dynamic and functions as a “safe harbor” from exhaustion and “acquisition of multiple mutations”.