Note to users. If you're seeing this message, it means that your browser cannot find this page's style/presentation instructions -- or possibly that you are using a browser that does not support current Web standards. Find out more about why this message is appearing, and what you can do to make your experience of our site the best it can be.
Comment on " 'Stemness': Transcriptional Profiling of Embryonic and Adult Stem Cells" and "A Stem Cell Molecular Signature" (I)
Ramalho-Santos et al. (1) and Ivanova et al. (2), comparingthe same three "stem cells"— embryonic stem cells (ESCs);neural stem cells (NSCs), referred to as neural progenitor/stemcells (NPCs) in the present study; and hematopoietic stem cells(HSCs)—with their differentiated counterparts, each identifieda list of commonly expressed "stemness" genes, proposed to beimportant for conferring the functional characteristics of stemcells. The ability to capture expression profiles of cells usingmicroarrays offers the possibility of defining a stem cell byits constellation of active genes. An intriguing question, however,is whether the functional commonalities (self-renewal and pluripotency)(3) among stem cells can be defined at the genetic level. Doall stem cells express a similar set of "stemness" genes necessaryfor their unique properties, or do different stem cells expressdifferent sets of genes that confer stemness?
We have independently carried out gene expression profilingof three types of stem or immature progenitor cells: ESCs, NPCs,and retinal progenitor/stem cells, or RPCs (Fig. 1) (4, 5).The intersection of ESC-, NPC- and RPC-enriched genes defineda list of 385 genes that are collectively expressed by all threestem cells (6). It can be inferred that these genes may representor include putative "stemness" genes. For the approach takenhere to be able to define and support the notion of "stemness"genes, however, would also require that very similar sets ofgenes can be identified regardless of the type of stem cellsused. To test the validity of this notion, we have collectivelyanalyzed our results along with those from the studies of Ramalho-Santoset al. (1) and Ivanova et al. (2) (Figs. 2 and 3). To our surprise,a comparison of the three independently derived lists of "stemness"genes showed only one gene (integrin alpha-6) commonly identifiedin the three studies (Figs. 2A and 3A) (6). This finding raisedserious concerns about the conclusions reported in (1) and (2),as was also critically highlighted by Burns and Zon (7).
Fig. 1. (A) Hierarchical clustering of stem cells and their differentiated progenies (6), as identified in this study. The clustering of NPCs and RPCs with their differentiated progenies [tissues adjacent to lateral ventricles of adult brain (LVBr) and mature retina (Ret)] suggest that somatic stem cells are already primed to become specific lineages. (B) Venn diagram showing the number of genes enriched in ESCs, NPCs and RPCs, as identified in this study. The overlap of these genes identified a common list of 385 stemness genes enriched in all three "stem cells" (6).
[View Larger Version of this Image (24K GIF file)]
Fig. 2. Venn diagrams showing overlap of "stemness" genes and stem cell–enriched genes among studies by Ramalho-Santos et al. (1), Ivanova et al. (2), and Fortunel et al. (this study). Ivanova et al. used three different Affymetrix chips (U74v2 A, B, and C); Fortunel et al. and Ramalho-Santos et al. used only the U74v2 A chip (Fig. 3 shows same comparison, limiting Ivanova et al. results to the A chip). (A) "Stemness" genes found by the three groups overlap by only one gene. (B) ESC-enriched genes identified by each study overlap by 332 genes; the probability that such overlap occurs by chance is extremely low (P < 10–8). (C) NPC-enriched genes overlapping by 236 genes between the three groups (P < 10–8). (D) Overlap of "stemness" genes—two types of stem cell (ESC/NPC)-enriched genes—is limited to 10 genes. The probability of this number of genes overlapping by chance is greatly increased. P > 10–4 is not significant because there are more than 104 genes studied (8).
[View Larger Version of this Image (45K GIF file)]
Fig. 3. (A to D) Same as Fig. 2, but with data limited to the Affymetrix U74v2 A chip for all three groups. Although the total number of genes identified by Ivanova et al. (2) is reduced, the number of genes found in the intersecting regions among the three studies are not changed significantly: The same single "stemness" gene common among the three studies remains in (A); the number of ESC- and NPC-enriched genes in (B) and (C), though it has declined, remains in the same order of magnitude; and the number of ESC/NPC-enriched genes in (D) has declined by only one, from 10 to 9. The decrease in the number of genes found in the intersecting regions of ESC- and NPC-enriched genes (panels B and C) is due to the fact that probes for some genes are present in more than one chip and that some genes that were detected in chip A by Ramalho-Santos et al. (1) and Fortunel et al. (this study) were detected by Ivanova et al. in either chip B or chip C but not in chip A. The P values remain low in both Fig. 2 and Fig. 3.
[View Larger Version of this Image (43K GIF file)]
We then examined whether the same extreme discrepancy was observedfor the lists of genes expressed in one specific stem cell type.In marked contrast, there was a very significant overlap inthe lists of stem cell–specific genes from the three studies.A total of 332 "ESC-enriched genes" (Fig. 2B) (6) and 236 "NPC-enrichedgenes" (Fig. 2C) (6) were identified by all three investigators.Statistical analysis (8) showed that these numbers are highlysignificant (P < 10–8). These results strongly go againstmajor differences in cells and analytical methodologies betweengroups as an explanation for the lack of overlap in the threelists of "stemness" genes. However, when we computed for "stemness"genes, based on genes commonly expressed in two types of stemcells, we found that the three studies overlap by only 10 genes,with P = 1.4 x 10–4 (Fig. 2D). Therefore, as the numberof stem cell types intersected to identify "stemness" genesis increased, the overlap between datasets from different investigatorsdrops dramatically.
To examine why that may be the case, we plotted a comparisonof significance scores (Fig. 4) for the different categoriesof genes derived by each group [fold change (FC) value or lowerconfidence bound (LCB) score] (9). It is quite apparent thatas one increases the number of stem cell types for comparison,the genes that occur in the intersection are genes that showprogressively lower differential expression between stem anddifferentiated cells (Fig. 4). For example, we found that thelog of the significance scores for individual ESC or NPC-specifictranscripts are typically in the range of 6 to 10 over differentiatedprogenies in all three groups. Transcripts of "stemness" genesfrom intersection of two or more stem cells, from all threegroups, commonly showed log significance scores in the 2 to4 level. In contrast to ESC and NPC, for which each study cameup with overlapping sets of genes that are uniquely or highlyexpressed exclusively in each stem cell type, none of the putative"stemness" genes in the three studies were highly expressedgenes compared with differentiated cells. These results indicatethat the expression of "stemness" genes common to all stem cells,if such genes exist at all, is only relatively elevated comparedwith differentiated cells. This would create a significant variationin the genes identified by differential expression, amplifiedby subtle differences in experimental conditions between differentstudies. We propose this as the most important basis for theextreme discrepancies in the lists of putative "stemness" genes.
Fig. 4. (A to C) Comparison of relative expression levels of"stemness"genes by fold changes. The graphs depict significance scores for log FC or log LCB changes in the expression levels of the top 80 genes from the list of ESC-enriched genes (blue), NPC-enriched genes (orange), ESC/NPC-enriched genes (red), and "stemness" genes (green). "Stemness" genes here are (A) the top 80 genes from the 385 genes identified in this study (Fig. 1), (B) the 230 genes identified in (1), and (C) the 283 genes identified in (2), as published; values in (B) and (C) were derived from published data. Fold changes are highest for genes enriched in a single stem cell type; with two stem cell types (NPC and ESC), the top 80 genes have much lower fold change compared to one stem cell type. With three stem cells, the fold changes compared with individual stem cells are markedly reduced.
[View Larger Version of this Image (20K GIF file)]
Our observation does not rule out the possibility that genesunique to all stem cells which are expressed at low levels mayexist. However, it is clear that no one single study can confidentlyidentify the bona fide genes that specify "stemness," and cross-validationof lists generated independently by different investigatorsis crucial. For instance, the total of 24 genes commonly identifiedin two of the studies (Figs. 2A and 3A) is statistically significantand warrants further investigation (6). It is possible alsothat there are "stemness" genes that have not yet been identifiedand are not represented in the chips used. Genes that may beimportant for stem cell functions such as self-renewal but thatare also expressed in non-stem cells (for example, Stat3, gp130)are unlikely to be identified by a comparative microarray approach.Another possibility is that different stem cell types may usedifferent gene networks to achieve self-renewal or multipotency(10). Finally, "stemness" genes may only be transiently expressed,so that they are easily missed by comparing two homeostaticstates. Further efforts to identify these genes will requiredifferent strategies.
In summary, speculations made from independent studies (1, 2)about identity of stemness genes do not hold up when the studiesare compared. Our explanation for this also demonstrates theinherent problem of testing the stemness hypothesis using aprofiling approach.
Nicolas O. Fortunel*
Department of Medicine Beth Israel Deaconess Medical Center Harvard Institutes of Medicine 77 Avenue Louis Pasteur Boston, MA 02115, USA and Genome Institute of Singapore 60 Biopolis Street Singapore 138672 and Laboratoire de Biologie des Cellules Souches Humaines CNRS UPR 9045 Villejuif, France
Hasan H. Otu*
Department of Medicine Beth Israel Deaconess Medical Center
Huck-Hui Ng*
Genome Institute of Singapore and Department of Biological Science National University of Singapore Singapore
Jinhui Chen
Genome Institute of Singapore and Departments of Physiology and Biochemistry National University of Singapore
Xiuqian Mu
Department of Biochemistry and Molecular Biology The University of Texas M. D. Anderson Cancer Center Houston, TX 77030, USA
Timothy Chevassut Xiaoyu Li Marie Joseph Charles Bailey
Department of Medicine Beth Israel Deaconess Medical Center
Jacques A. Hatzfeld Antoinette Hatzfeld
Laboratoire de Biologie des Cellules Souches Humaines
Fatih Usta
Department of Medicine Beth Israel Deaconess Medical Center
Vinsensius B. Vega Philip M. Long
Genome Institute of Singapore
Towia A. Libermann
Department of Medicine Beth Israel Deaconess Medical Center
Bing Lim
Department of Medicine Beth Israel Deaconess Medical Center and Genome Institute of Singapore E-mail: blim{at}caregroup.harvard.edu
* These authors contributed equally to this work.
References and Notes
1. M. Ramalho-Santos, S. Yoon, Y. Matsuzaki, R. C. Mulligan, D. A. Melton, Science298, 597 (2002).[Abstract/Free Full Text]
3. J. E. Till and E. A. Mc Culloch, Biochim. Biophys. Acta605, 431 (1980). [Medline]
4. E14 pluripotent ESCs were cultured following established methods on embryonic fibroblast feeder layer in the presence of leukemic inhibitory factor. NPCs were cultured as neurospheres derived from E12 embryos following an established protocol (11). Although the frequency of stem cells cannot be ascertained, this population consisted largely of cells that exhibit stem cell properties such as self-renewal and the capacity to generate the major cell types found in the central nervous system. RPCs were obtained from manually dissected from E14.5 mouse embryo retinas, a stage in which more than 90% of cells in the retina are undifferentiated neuroblasts (12). Mature retinas were dissected from day 10 postpartum (P10) mice. Lateral ventricular brain (LVbr) came from adult mice. Primary embryonic fibroblasts (PEF) were isolated from E13 embryo. All RNAs were extracted using Trizol.
5. Duplicate samples of total RNAs from stem/progenitor cells and differentiated cells were probed using the Affymetrix U74Av2 chip following established manufacturer's protocol. Samples that passed a priori quality control criteria were scaled using a 2% trimmed mean to perform comparative analyses. The high reproducibility of the samples for the replicate arrays was indicated by the high correlation coefficient values (between 0.95 and 0.98). An unsupervised-learning technique was applied by constructing an unweighted pair group method with arithmeticmean (UPGMA) tree using Pearson's correlation as the metric of similarity (13). This tree represented the inherent grouping in the sample set and the degree of closeness between samples. To find genes enriched in a given cell type, we followed the supervised-learning approach as described by Tusher et al. (14). The statistical significance of the results was assessed using permutation testing. Genes that showed at most a 5% false discovery rate were reported as differentially expressed. Programs used in the analyses can be obtained through www.biostage.org. We have also used dChip software, used by Ramalho-Santos et al. (1), to compute for transcripts enriched in stem cells (data not shown). The results for the list of genes obtained were highly similar.
8. The significance of the number of overlapping genes after intersection of gene lists was measured by P values, the probabilities that such overlaps could occur by chance alone. Monte Carlo simulations were employed to approximate the P values of the three-group comparisons (6, 15). Ivanova et al. (2) used Affymetrix U74v2 A, B, and C chips, each containing 12,000 probes. Since at most 12,000 genes [the number in this study and Ramalho-Santos et al. (1), using U74Av2 chip] from each study could be in any list of overlapping genes, our Monte Carlo analysis provides a conservative estimate of the significance of overlaps, including the study by Ivanova et al. (2). The P values arising from our Monte Carlo analysis are estimates of how unlikely it is that a degree of overlap in the gene lists of the different studies would arise by chance if the gene lists were chosen independently at random. The fact that P is small should be interpreted as suggesting that the processes through which these gene lists were generated have something in common (for example, if there was a small-to-moderate number of truly differentially expressed genes, these were very likely to be discovered in each of the studies, even if each of the studies also had a large number of false positives). In particular, the P values should not be interpreted as a measure of the confidence that all or even most of the overlapping genes are the correct ones."
9. ESC- and NPC-enriched genes were listed with their significance scores, as evaluated in the three studies by fold changes (FC) value or lower confidence bound (LCB) score, and tabulated for their log value. A graph is plotted as shown for the first 80 scores of ESC- and NPC-enriched genes (blue and orange curves in Fig. 4). To measure the extent to which the data implicate a gene as being differentially expressed in more than one stem cell type (for example, ESC and NPC), we took the average of the scores for differential expression between the two stem cells. A graph is plotted for ESC/NPC-enriched genes in each of the three studies (red curves in Fig. 4). A similar approach was taken to get the curves for the ESC/NPC/RPC (this study) or ESC/NPC/HSC (other two studies) "stemness" genes for the intersection of three stem cells (green curves in Fig. 4).
10. T. S. Burdon et al., Trends Cell Biol.9, 432 (2002).
15. A. C. Davison and D. Hinkley, Bootstrap Methods and Their Application (Cambridge Univ. Press, Cambridge, 1997).
Received for publication 2 May 2003. Accepted for publication 28 July 2003.
The editors suggest the following Related Resources on Science sites:
In Science Magazine
REPORTS
Miguel Ramalho-Santos, Soonsang Yoon, Yumi Matsuzaki, Richard C. Mulligan, and Douglas A. Melton (18 October 2002) Science298 (5593), 597.
[DOI: 10.1126/science.1072530] |Abstract »|Full Text »|PDF »|Supporting Online Material »
REPORTS
Natalia B. Ivanova, John T. Dimos, Christoph Schaniel, Jason A. Hackney, Kateri A. Moore, and Ihor R. Lemischka (18 October 2002) Science298 (5593), 601.
[DOI: 10.1126/science.1073823] |Abstract »|Full Text »|PDF »|Supporting Online Material »
12,000 probes. Since at most 12,000 genes [the number in this study and Ramalho-Santos et al. (
