Science, Vol. 298, Issue 5593, 601-604, October 18, 2002 A Stem Cell Molecular Signature Natalia B. Ivanova, John T. Dimos, Christoph Schaniel, Jason A. Hackney, Kateri A. Moore, Ihor R. Lemischka* Mechanisms regulating self-renewal and cell fate decisions in mammalian stem cel ls are poorly understood. We determined global gene expression profiles for mous e and human hematopoietic stem cells and other stages of the hematopoietic hiera rchy. Murine and human hematopoietic stem cells share a number of expressed gene products, which define key conserved regulatory pathways in this developmental system. Moreover, in the mouse, a portion of the genetic program of hematopoieti c stem cells is shared with embryonic and neural stem cells. This overlapping se t of gene products represents a molecular signature of stem cells. Department of Molecular Biology, Princeton University, Princeton, NJ 08544, USA. * To whom correspondence should be addressed: E-mail: [email protected] ton.edu -------------------------------------------------------------------------------- Adult and embryonic stem cells (SCs) hold great promise for regenerative medicin e, tissue repair, and gene therapy (1). Hematopoietic stem cells (HSCs) have bee n the most extensively studied and serve as a prototype model to define the gene ral biological properties of mammalian SCs. Distinct developmental stages of the hematopoietic hierarchy can be identified and arranged in a hierarchical tree t hat begins with the long-term (LT) functional HSC. A single LT-HSC is both neces sary and sufficient for life-long sustenance of the entire hematopoietic system (2, 3). LT-HSCs produce less potent short-term (ST) functional HSCs, and these i n turn, give rise to lineage-committed progenitor (LCP) cells. The LCP cells are directly responsible for the generation of at least 10 mature blood cell (MBC) populations. Many nonhematopoietic tissues also depend on tissue-resident SCs fo r their maintenance and regeneration (4). Totipotent embryonic stem cells (ESCs) , derived from blastocysts, and neural stem cells (NSCs), derived from the germi nal zones of the nervous system, are two examples of SCs that can be propagated in vitro (5). Because all SCs share fundamental biological properties, they may share a core set of molecular regulatory pathways. It is likely that at least so me components of these regulatory pathways are preferentially expressed by SCs. We therefore attempted to define a general gene expression profile of the SC "st ate." We have adopted the approach outlined in Fig. 1 that first separately identifies gene expression profiles for murine fetal and adult HSCs. These profiles are th en compared to derive a shared HSC profile. This profile should include gene pro ducts that are necessary for LT hematopoietic function. We also generated gene e xpression profiles for human HSCs and for two murine nonhematopoietic SC populat ions, NSCs and ESCs. The comparison of murine with human HSCs defines evolutiona rily conserved components in HSCs, whereas the comparison of hematopoietic with nonhematopoietic SCs identifies the gene products expressed in multiple SC types . The samples were processed as shown in fig. S1. Tissue or cell replicates were isolated and functionally evaluated to measure the purity of SC-containing frac tions. In vitro amplified RNA probes were hybridized to Affymetrix oligonucleoti de arrays. We estimate that these arrays currently allow for the monitoring of a pproximately 80% of HSC-related gene products (fig. S2). Arrays were scanned and processed using Affymetrix MAS 4.0 software. Genes were assigned to distinct cl usters according to their expression patterns within the hematopoietic hierarchy . NSC and ESC enrichment scores were calculated to define the expression of the transcripts in these two SC populations. Bioinformatics analyses were performed for the SC-specific gene products. Details of the SC purification procedures, bi ological assays, and data analyses are available in supporting online material ( 6). -------------------------------------------------------------------------------- Fig. 1. Stem cell phenotypes profiled. Cells at key stages of the murine and hu man hematopoietic hierarchy were isolated as shown, and include LT-HSCs, ST-HSCs , LCPs, and MBCs. Nonhematopoietic SCs were cultured (ESCs) or purified (NSCs). This approach identifies three groups: genes specific for both fetal and adult m urine HSCs (blue boxes), genes specific for murine and human HSCs (red box), and genes enriched in diverse SCs (green box). [View Larger Version of this Image ( 26K GIF file)] -------------------------------------------------------------------------------- To translate the biological phenotypes of key hematopoietic populations into the language of gene expression, we used a series of hypothetical expression patter ns that correlate with distinct, quantitatively measured biological activities p resent in the hematopoietic hierarchy (Fig. 2, A to C). A total of 4289 informat ive genes were assigned to seven clusters (Fig. 2D), characteristic of key stage s of hematopoiesis, progressing from stem through progenitor to terminally diffe rentiated cells. HSC-related clusters i to iii include many known HSC markers su ch as c-Kit, Tie1, Ly-6E/Sca-1, Tek, Mpl, Meis1, Gata2, and Abcb1b/MDR1. At leas t 72% of the above-defined HSC-related genes are also up-regulated in CD45+c-Kit +Sca-1+ Hoechst 33342 side population cells (7). These cells have been shown to contain LT-HSCs (8). Furthermore, 54% of genes assigned to these clusters were p reviously identified through a global subtractive hybridization screen for HSC-s pecific gene products (7, 9) (fig. S2). This demonstrates a strong correlation b etween HSC-specific gene sets identified by different strategies. The expression specificity of 22 HSC-related genes was confirmed by quantitative reverse trans cription-polymerase chain reaction (RT-PCR) (fig. S5). Gene products were groupe d into categories according to their function as reported in the literature or a s predicted on the basis of the presence of diagnostic protein motifs. Regulator y molecules, such as transcription factors, proteins involved in intracellular s ignaling, cell-surface receptors, and ligands account for 45% of the HSC-related gene-products (Fig. 2E). -------------------------------------------------------------------------------- Fig. 2. Correlating biological function and gene expression. (A) Competitive re populating activity of the isolated hematopoietic populations was determined (23 ). Mice were transplanted with graded doses of purified Ly5.2 fetal liver (FL) o r bone marrow (BM) SCs, mixed with 2 × 105 Ly5.1 whole BM cells. Ly5.2 peripher al blood content at 6 months is shown. The repopulating stem cell frequency in t hese purified populations is 1 in 10 to 20 cells for both FL and BM SCs. (B) The number of colony-forming cells (CFCs) in the isolated stem and progenitor cell populations was determined. Colonies were scored as high proliferative potential -granulocyte macrophage (HPP-GM), GM, MIX (three or more lineages: GM, megakaryo cyte, erythrocyte), and HPP-MIX. (C) The hematopoietic hierarchy subgrouped into different stem and progenitor populations and (D) their corresponding expressio n clusters (i to vii). Individual genes were assigned to expression clusters as described (6). Relative expression levels are displayed by red (highest) to gree n (lowest) coloration. Predicted cellular roles of identified HSC-specific gene products: (E) distribution within the HSC profile for gene products with known o r putative functions, (F) distribution of the annotated gene-products between HS C subtypes, and (G) between fetal and adult HSCs. [View Larger Version of this I mage (43K GIF file)] -------------------------------------------------------------------------------- We have defined genomewide transcriptional changes during early stages of hemato poietic differentiation by comparing four distinct sets of genes that are up-reg ulated in LT-HSCs (i), in both LT and ST-HSCs but not in LCPs (ii), in both HSCs and LCPs (iii) and, in ST-HSCs and early progenitor population (iv), respective ly. The distribution of genes within these four sets across functional categorie s is shown in Fig. 2F. Molecules thought to be involved in cell-cell communicati on, such as signaling ligands, receptors, extracellular matrix, and adhesion mol ecules, tend to be overrepresented in the HSC-specific gene set. LT-HSC-specific ligands include Bmp8a, Wnt10A, EGF-family members Ereg and Hegfl, the angiogene sis-promoting factor Agpt, a ligand for the ROBO receptor family Slit2, and the ephrin receptor ligand EfnB2. These molecules may be involved in signaling betwe en HSCs and their microenvironment. It is interesting that HSCs coexpress severa l ligand-receptor pairs, such as Wnt10A/Frizzled and Agpt/Tek, which suggests th at HSC regulation may be partly autocrine. The complete set of HSC-related genes is presented in table S2. ST-HSCs and early progenitors express molecules associated with the initiation o f the cell cycle such as Wee1 kinase, Cdk4, replication licensing factor Mcmd, a nd the critical hematopoietic proliferation protein, Myb. Genes involved in DNA repair and protein synthesis are also up-regulated in these compartments. This i s consistent with the exit from G0 arrest at the onset of differentiation. ST-HS Cs also express a set of gene products with RNA-binding domains, which is sugges tive of posttranscriptional regulation. Hox genes are likely to play a role in HSC regulation. Four HoxA genes are expre ssed in different subsets of HSCs. Hoxa5 and Hoxa10 are specific for the LT-HSCs , Hoxa2 is expressed in both LT and ST-HSCs, and Hoxa9 is expressed both in HSCs and LCPs. It is noteworthy that overexpression of Hoxa9 in murine HSCs induced stem cell expansion (10), whereas Hoxa5 and Hoxa10 perturbed their differentiati on activity (11, 12). In addition, Hoxb4, which is detected both in HSCs and LCP s, has been shown to promote specification and expansion of definitive HSCs (13, 14). Fetal and adult HSCs share the key stem cell properties of self-renewal an d multilineage differentiation potential. In agreement with this, comparing the gene expression profiles of fetal and adult HSCs reveals broad molecular similar ities (Fig. 2G). More than 70% of all HSC-related gene-products are expressed in both fetal and adult HSCs. We next asked whether the HSC genetic program is conserved between mouse and hum an. Human fetal liver Lin-CD34+CD38- cells provide long-term engraftment of nono bese diabetic immunodeficient NOD-SCID mice and, therefore, are functionally sim ilar to murine LT-HSCs (15). Human gene products with an increase in expression of at least twofold in HSCs compared with MBCs were defined as HSC-enriched. Mou se-human homologous pairs were identified by direct sequence comparison of expre ssed sequence tag (EST) assemblies as described (6). We found 822 human homologs for murine HSC-related genes that are expressed in fetal liver (Database S3). O f these, 322 (39%) were enriched in human fetal HSCs. The probability of observi ng such an overlap by chance as estimated using hypergeometrical distribution (6 , 16) is extremely low (P = 1011 ). These genes likely represent the conserved m olecular components expressed in HSCs. Homologous gene products expressed in the LT-HSC subset are listed in Table 1. The remaining homologous pairs did not sho w coordinate expression. This may reflect technical difficulties in purifying ho mogeneous HSC fractions. Alternatively, related but not identical populations ma y function as HSCs in different organisms. Table 1. select known mouse-human homologs expressed in LT-HSC subset. The compl ete list of homologous pairs is presented in Database S3. FC, fetal cells, GPCR, G protein-coupled receptor; LDL, low density lipoprotein; MHC, major histocompa tibility complex; TF, transcription factor; UTR, untranslated region. -------------------------------------------------------------------------------- Gene Name Mouse GenBank ID Human GenBank ID Mouse FC Human FC Annotation -------------------------------------------------------------------------------- Ches1 AW046392 U68723 3.3 2.8 Checkpoint suppressor 1, DNA damage, cell-cycle ar rest SREC AA986099 D86864 10 3.6 Acetyl LDL scavenger receptor Blr1 AI608284 X68149 2.8 3.6 Burkitt lymphoma-associated chemokine GPCR Procr L39017 L35545 39.3 3.1 Endothelial cell protein C receptor Fzd4 U43317 AI927489 9.2 2.4 Frizzled-like GPCR (Wnt receptor) Igf1r AF056187 X04434 4.6 2.1 Insulin-like growth factor I receptor Mtap7 Y15197 X73882 3.3 6.3 Microtubule-associated protein MYO5C AW214321 AA195002 9 6.4 Myosin 5 motif Pclo Y19186 AB011131 7.8 2 Presynaptic cytomatrix protein Sparcl1 AV347505 X86693 4.6 3.1 SPARC-like protein Ocln AW209088 U49184 6.3 2 Tight junction component Jcam2 AI853724 AI199779 15.2 3 Tight junction component Jcam3 AI850297 AA149644 11.1 8.7 Tight junction component Mpdz AV244715 AF093419 18.4 5.2 Multiple PDZ domains, interacts with GPCRs Nbea AI154580 AI052524 29.3 3.2 Protein kinase A regulator SCOP AI836256 AB011178 3.7 2.7 Protein phosphatase 2C domain Ptpn21 D37801 X79510 32.4 3.8 Protein tyrosine phosphatase, nonreceptor type Ndr2 AV349686 AI201607 14.7 2.1 Regulated by N-myc Rras M21019 AI201108 9.8 2.7 R-ras oncogene Agpt U83509 U83508 9.8 20.9 Angiopoietin-1, binds TIE-2/Tek receptor Efnb2 U30244 AI765533 39.4 2.9 Ephrin B2, Eph receptor ligand Rbp1 X60367 M11433 49.8 3.8 Involved in metabolism of retinoids Aldh2 AV329607 X05409 21.9 4.1 Mitochondrial aldehyde dehydrogenase Fkbp7 AF040252 AI271550 3.3 2.9 Peptidyl-prolyl cis-trans isomerase Smpd1 AV347445 M81780 13.2 2.1 Lysosomal sphingomyelin phosphodiesterase Tapbp AV361189 AA767887 7.9 2.3 MHC-like antigen-processing transporter Nnp1 AV260279 AI860822 2.5 5 Nucleolar protein 52-like Htf9c AV325777 AW007779 6.5 2 RNA recognition motif-containing protein Elavl4 AV241912 AA102788 13.9 6.4 Uridylate-rich UTR binding Tcf3 AJ223069 AI916838 5.6 2 General immunoglobulin TF-3 Pphn AW123178 M95585 33.8 201.4 Hepatic leukemia factor implicated in apoptosis inhibition Hoxa5 Y00208 AC004080 3 4.1 Up-regulates p53 and progesterone receptor expressio n P2rx4 AF089751 U83993 9 2.9 Purinergic receptor ligand-gated ion channel Slc12a2 U13174 N56950 10.6 2.9 Sodium-potassium-chloride cotransporter To establish the gene expression profile common for diverse types of SCs, we per formed analyses of ESCs and NSCs. Gene products with an increase in expression o f at least twofold compared with both fetal and adult MBCs were defined as ESC/N SC-enriched. Correct detection of ESC- and NSC-enriched genes was verified by co mparison with published data sets (17, 18) and are presented in tables S3 and S4 . ESC/NSC-enriched gene sets were compared with each hematopoietic cluster. These results are summarized in Fig. 3. Gene products enriched in all three SC types b elong to a variety of functional categories. Several identified gene products ha ve been previously implicated in the regulation of different types of SCs. Trans cription factors Edr1 and Tcf3 have been shown to sustain the activity of HSCs ( 19) and epidermal SCs (20), respectively, whereas EfnB2 and Hes1 have been impli cated in control of NSC proliferation (21, 22). Analyses of EST collections indi cate that many of the HSC- ESC- and NSC-enriched genes are also expressed in oth er tissues (7). This may suggest more general functional roles in a broader arra y of SC populations. -------------------------------------------------------------------------------- Fig. 3. Overlapping gene expression in diverse murine SCs. (A) Venn diagram det ailing shared and distinct gene expression among NSCs, ESCs, and HSCs. (B) A sum mary of the number of different genes expressed in diverse stem cell compartment s in relation to each other and compared with the above defined hematopoietic cl usters. A complete list of HSC-related genes also enriched in NSCs and/or ESCs i s presented in Database S4. [View Larger Version of this Image (16K GIF file)] -------------------------------------------------------------------------------- In summary, we have determined the molecular similarities and differences among five distinct SC populations, specifically, human fetal HSCs, murine fetal and a dult HSCs, NSCs, and ESCs. The similarities define a common SC genetic program o r SC molecular signature. It is likely that hallmark properties shared by all SC s, such as the ability to balance self-renewal and differentiation, will be gove rned by shared molecular mechanisms. As such, numerous components of these molec ular mechanisms are likely to be contained within the SC molecular signature pre sented here. REFERENCES AND NOTES 1. I. L. Weissman, Science 287, 1442 (2000) [Abstract/Full Text]. 2. C. T. Jordan and I. R. Lemischka, Genes Dev. 4, 220 (1990) [Abstract]. 3. M. Osawa, K. Hanada, H. Hamada, H. Nakauchi, Science 273, 242 (1996) [Abstrac t]. 4. E. Fuchs and J. A. Segre, Cell 100, 143 (2000) [ISI][Medline]. 5. I. L. Weissman, D. J. Anderson, F. Gage, Annu. Rev. Cell Dev. Biol. 17, 387 ( 2001) [Abstract/Full Text]. 6. Material and Methods are available as supporting online material on Science O nline. 7. N. B. Ivanova, K. A. Moore, I. R. Lemischka, unpublished observations. 8. M. A. Goodell, K. Brose, G. Paradis, A. S. Conner, R. C. Mulligan, J. Exp. Me d. 183, 1797 (1996) [Abstract]. 9. R. L. Phillips, et al., Science 288, 1635 (2000) [Abstract/Full Text]. 10. U. Thorsteinsdottir, et al., Blood 99, 121 (2002) [Abstract/Full Text]. 11. G. M. Crooks, et al., Blood 94, 519 (1999) [Abstract/Full Text]. 12. C. Buske, et al., Blood 97, 2286 (2001) [Abstract/Full Text]. 13. J. Antonchuk, G. Sauvageau, R. K. Humphries, Cell 109, 39 (2002) [ISI][Medli ne]. 14. M. Kyba, R. C. Perlingeiro, G. Q. Daley, Cell 109, 29 (2002) [ISI][Medline]. 15. G. Guenechea, O. I. Gan, C. Dorrell, J. E. Dick, Nature Immunol. 2, 75 (2001 ) [CrossRef][ISI][Medline]. 16. S. Tavazoie, J. D. Hughes, M. J. Campbell, R. J. Cho, G. M. Church, Nature G enet. 22, 281 (1999) [CrossRef][ISI][Medline]. 17. D. L. Kelly and A. Rizzino, Mol. Reprod. Dev. 56, 113 (2000) [CrossRef][ISI] [Medline]. 18. D. H. Geschwind, et al., Neuron 29, 325 (2001) [ISI][Medline]. 19. H. Ohta, et al., J. Exp. Med. 195, 759 (2002) [Abstract/Full Text]. 20. B. J. Merrill, U. Gat, R. DasGupta, E. Fuchs, Genes Dev. 15, 1688 (2001) [Ab stract/Full Text]. 21. J. C. Conover, et al., Nature Neurosci. 3, 1091 (2000) [CrossRef][ISI][Medli ne]. 22. T. Ohtsuka, M. Sakamoto, F. Guillemot, R. Kageyama, J. Biol. Chem. 276, 3046 7 (2001) [Abstract/Full Text]. 23. D. E. Harrison, C. T. Jordan, R. K. Zhong, C. M. Astle, Exp. Hematol. 21, 20 6 (1993) [ISI][Medline]. 24. We thank C. Jordan for providing the human hematopoietic samples, A. Beavis for expert flow cytometry, and T. Doniger and M. Pritsker for assistance with bi oinformatics. We also thank N. Stahl and F. Santori for critically reviewing the manuscript. This work was supported by grants from the NIH DK54493 and DK42989 (to I.R.L.). Additional support was provided by ImClone Systems, Inc., New York. 10 May 2002; accepted 3 September 2002 Published online 12 September 2002; 10.1126/scienc