From Parts List to Architecture

LAYING THE FOUNDATION

In the first Hot Paper, a team from Harvard Medical School compared gene expression profiles for ESCs, NSCs, and HSCs.¹ The second paper, from Princeton University, did much the same, with controls of fetal mouse and human HSCs.² Both groups used subtractive hybridization and cDNA microarrays to identify expressed genes shared among the stem-cell types as well as unique to each. "The main idea is to get at the core program that makes these cells tick, to see how they decide to self-renew and proliferate, or commit to lineages," said Ihor Lemischka, professor of molecular biology at Princeton, while at the International Society of Stem Cell Reseach meeting in Boston this past June.

Both papers identify genes involved in stress resistance, chromatin remodeling, translation regulation, and cell-cycle control. But the analysis is descriptive, rather than functional, says Miguel Ramalho-Santos, of the Harvard team and now a fellow at University of California, San Francisco. Beyond factors needed to maintain stemness, stem cells may sit in a dynamic cocktail of lineage-specific regulatory factors synthesized before they are needed. "They are all potentially there, but the stem cell doesn't really mind, because it is busy renewing itself. At the same time, it is poised to go ahead and follow a particular lineage," he says.

The Princeton researchers call their results "a foundation for a more detailed understanding of stem cell biology." Building on that base, investigators are now taking a more systems-oriented view that echoes engineering. "A network of transcription factors emerges that responds to inputs and can compute them into outputs, which are the decisions that stem cells make," says Bertal Gottkgens, leukemia research fund lecturer at the Cambridge Institute for Medical Research, UK.

Both research teams acknowledge limitations in global gene-expression profiling: Microarrays detect only what you tell them to. Stem-cell preparations contain a cacophony of expressed genes that have nothing to do with stemness. "Every group that has a new way to enrich for adult stem cells is developing gene microarray profiles. These analyses depend on the purity of the isolated cells, and are desperately confounded by the generally high level of noise in microarray data," says Sherley.

CRITICISM A YEAR LATER

Nearly a year after publication of the Hot Papers, two Technical Comments in Science compared the studies and noted "negligible overlap" and "extreme discrepancy" in the results. Alexei Evsikov of the Jackson Laboratory in Bar Harbor, Maine, and Davor Solter of the Max-Planck Institute of Immunobiology in Freiburg, Germany, determined that only six genes, or 1.2% of the genes expressed commonly among the three stem cell types, matched in the two studies.³ They fault a focus on poorly expressed genes. "In such a situation, it is very easy to obtain spurious two- to threefold change either way, without biological significance," says Solter.

The other multiauthor Technical Comment presented new Venn diagrams, contrasting expression profiles for ESCs, NSCs, and retinal progenitor cells with the results presented in the Hot Papers. They found only one stemness gene in common with all of those identified in the two studies.⁴ But when they compared their results to the other groups within stem-cell type, such as NSC to NSC, they saw more shared genes. If the type-to-type comparisons are robust, but the search for stemness genes among both ESCs and more restricted stem or progenitor cells is not, then the methodology is not at fault, they argue. A design flaw should limit all comparisons, not just some.

The Hot Papers groups responded, citing possible sources of error and reporting better correspondence after standardizing statistical methods and definitions of enrichment.⁵ Nuances of protocol in gene expression profiling or stem-cell derivation, plus low levels of gene expression, might have contributed to the differences in the profiles from the two studies. But assumptions may have oversimplified stem-cell function (see "Dedifferentiation: More than Reversing Fate" on pg. 20). That is, a "roll call" of genes expressed at a certain time might not reveal as much about stem-cell distinctions as protein function, abundance, and interactions.

While researchers try to decipher why results vary, many are turning to RNAi to make these descriptive studies functional. "You start with purified stem cells, and now have a snapshot of what genes are activated. ... Then, you take one away. Can the stem cell cope?" asks Ramalho-Santos. He foresees RNAi and microarrays teamed to probe cells after each gene is silenced, "to discover the genetic circuits in stem cells." At the Boston meeting, Natalia Ivanova, a Princeton postdoctoral researcher, reported on how their group uses RNAi: Their microarray of 44,000 genes found 901 that are downregulated as retinoic acid induces differentiation of murine ESCs. They have selected 70 for further analysis.

"We selected the transcription factors and expressed-sequence tags that are highly expressed in undifferentiated cells and rapidly downregulated during differentiation," she said, adding, "We have begun to define molecular networks that control cell-fate decisions."

OTHER PARTS LISTS

Eclectic studies have followed publication of the Hot Papers, most zeroing in on genes for which expression changes with ensuing differentiation. Studies range from the specific to the general, and probe the earliest to the most specialized cells.

© 2002 AAAS

(A)The Ramalho-Santos group identified 230 expressed genes shared among the three stem cell types (embryonic, hematopoietic, and neural), but pared this number down to 216 to account for genes recognized by more than one probe on the microarray. Embryonic stem cells and neural stem cells appear the most intimately related. (B) indicates several shared genes that are co-expressed in ESCs, and NSCs, considered by general functional category. (Reprinted with permission, Science, 298:597–600, 2002.)

Kevin D'Amour and Fred Gage, at the Salk Institute for Biological Studies in La Jolla, Calif., highlight differences between ESCs and NSCs, rather than seeking the similarities reported in the Hot Papers.⁶ Elaine Fuchs' group at Rockefeller University captures stem cells as they exit hair follicles to spread out over adjacent epidermis and differentiate, in response to signals generated from wounding.⁷ Gretchen Darlington at Baylor College of Medicine and postdoctoral researcher Scott Ochsner are scrutinizing expression profiles of three liver progenitor cells. And Ralph Brandenberg and colleagues at Geron in Menlo Park, Calif., followed genes upregulated or downregulated as human ESCs develop into embryoid bodies, prehepatocytes, or neuroectoderm, finding a "quite small" overlap with the Hot Papers' data. They conclude that it might be unrealistic to try to define a static set of stemness genes.⁸

Sherley points out a shortcoming of the papers' global transcriptional-profiling approach. "They looked for genes that, compared to non-stem cell populations, were expressed in both embryonic stem cells and adult stem cell-enriched populations. Thus, their stemness signatures exclude important genes that are preferentially expressed in one stem-cell type."

He and student Min-Soo Noh pursued a characteristic of stem cells, the asymmetric cell division of adult stem cells that self-renew as well as yield more differentiated daughters. They identified asymmetric self-renewal-associated (ASRA) genes, and compared them to the stemness genes found in the Hot Papers and Fortunel et al. "When we compared ASRA genes to their individual adult and embryonic stem-cell gene sets, we find highly significant overlap for upregulated ASRA genes and downregulated ASRA genes, respectively," Sherley says, supporting the hypothesis that self-renewal asymmetry is an important aspect of stemness for adult but not embryonic stem cells.

Given the newly-appreciated complexities of stem cells, the Hot Papers have laid a foundation for a very elaborate architecture. As Lemischka put it: "We have a molecular parts list for stem cells, but how do we build that into an architecture? There's no easy solution. We need a wiring diagram, a logic circuit, and that is still very far away."

Ricki Lewis rlewis@the-scientist.com