Leading Edge Review Mechanisms of Asymmetric Stem Cell Division Juergen A. Knoblich1,* 1Institute of Molecular Biotechnology of the Austrian Academy of Sciences, Dr. Bohr Gasse 3, 1030 Vienna, Austria *Correspondence: juergen.knoblich@imba.oeaw.ac.at DOI 10.1016/j.cell.2008.02.007 Stem cells self-renew but also give rise to daughter cells that are committed to lineage-specific differentiation. To achieve this remarkable task, they can undergo an intrinsically asymmetric cell division whereby they segregate cell fate determinants into only one of the two daughter cells. Alternatively, they can orient their division plane so that only one of the two daughter cells maintains contact with the niche and stem cell identity. These distinct pathways have been elucidated mostly in Drosophila. Although the molecules involved are highly conserved in vertebrates, the way they act is tissue specific and sometimes very different from invertebrates. A hallmark of all stem cells is the ability to simultaneously gener- ate identical copies of themselves but also to give rise to more differentiated progeny. Work mostly done in the fruitfly, Drosophila, has suggested two different mechanisms by which this can be achieved (Horvitz and Herskowitz, 1992) (Figure 1). When an intrinsic mechanism is used, regulators of self-renewal are localized asymmetrically during mitosis so that they are inherited by only one of the two daughter cells (Betschinger and Knoblich, 2004 Yu et al., 2006). Already in interphase, cells which undergo such intrinsically asymmetric divisions use apical-basal or planar polarity of the surrounding tissue to set up an axis of polarity. As they enter mitosis, this axis is used to polarize the distribution of protein determinants and to orient the mitotic spindle so that these determinants are inherited by only one of the two daughter cells. Alternatively, the stem cell is in close contact with the stem cell niche and depends on this contact for maintaining the potential to self-renew (Li and Xie, 2005). By orienting its mitotic spindle perpendicularly to the niche surface, it ensures that only one daughter cell can maintain contact with the stem cell niche and retain the ability to self-renew. In contrast to intrinsically asymmetric cell divi- sions, which usually follow a predefined developmental pro- gram, niche-controlled stem cell divisions offer a high degree of flexibility. Occasionally, the stem cell can divide parallel to the niche, thereby generating two stem cells to increase stem cell number or to compensate for occasional stem cell loss. For this reason, niche mechanisms are more common in adult stem cells, whereas intrinsically asymmetric divisions predomi- nate during development. Collectively, both types of cell division are referred to as asym- metric cell division. An asymmetric cell division is defined as any division that gives rise to two sister cells that have different fates���a feature that can be recognized by differences in size, morphology, gene expression pattern, or the number of subse- quent cell divisions undergone by the two daughter cells (Horvitz and Herskowitz, 1992). Although some stem cells���in particular hematopoietic and embryonic stem (ES) cells���do not quite fit this definition when kept in culture, it is safe to assume that most, if not all, stem cells undergo asymmetric cell divisions when they are in their natural environment. In Drosophila, neuroblasts and ovarian stem cells are well-studied examples for the intrinsic and extrinsic mode of asymmetric cell division, respectively. Although these simple categories may not apply as exclusively to mammalian stem cells and both pathways seem to be combined in some cell types, they provide a conceptual framework that will help us to understand the complexity of mammalian stem cell biology. Below, I describe the anatomy and molecular machineries of asymmetric cell division in Drosophila neuroblasts and ovarian germline stem cells and use neural, muscle, and hematopoietic stem cells as examples to illustrate the similarities and differ- ences in higher organisms (see Table 1 for a summary of the model systems described). Asymmetric Protein Segregation in Drosophila Drosophila Neural Precursor Cells Drosophila sensory organ precursor (SOP) cells and neuroblasts (the progenitors of the peripheral and central nervous system, respectively) are well-studied examples of intrinsically asymmet- ric cell divisions (Figure 2). SOP cells give rise to the four cells present in external sensory organs (Figure 2A). Although they are not stem cells, SOP cells have revealed many of the funda- mental principles for asymmetric cell division. This is mainly due to their simple and highly reproducible lineage: SOP cells delaminate from a polarized epithelium and then divide into an anterior pIIb and a posterior pIIa cell. After SOP division, pIIa and pIIb divide once more to generate the two outer and the two inner cells of the organ, respectively. Asymmetry in all of these divisions is generated by different levels of Notch activity in the two daughter cells (Schweisguth, 2004 Le Borgne et al., 2005). It is thought that SOP cells inherit epithelial planar polarity and use it to segregate regulators of the Notch signaling pathway into one of the two daughter cells. In contrast to SOP cells, Drosophila neuroblasts undergo multiple rounds of stem cell-like divisions (Figure 2B). During each division, they give rise to a large cell that retains neuroblast properties and a smaller cell that is called the ganglion mother cell (GMC) and divides only once more to generate two differen- tiating neurons. Neuroblasts come in two flavors embryonic neuroblasts give rise to the relatively simple nervous system of Cell 132, 583���597, February 22, 2008 ��2008 Elsevier Inc. 583

the larva. They are specified within a monolayered epithelium called the ventral neuroectoderm and delaminate from the epithelium to undergo repeated rounds of asymmetric division along the apical-basal axis. It is thought that certain aspects of epithelial polarity are inherited by the neuroblasts and used to polarize the first mitotic division. Although the reproducible position and the relatively simple lineages of embryonic neuro- blasts have made them a valuable system to discover basic principles of asymmetric division, their restricted self-renewal capacity limits their usefulness as a true stem cell model. Mainly for this reason, the field has recently begun to focus on larval neuroblasts. Larval neuroblasts generate the thousands of neurons found in the central nervous system of an adult fly. Unlike embryonic neuroblasts, which become smaller with each division, larval neuroblasts regrow back to their original size after each division and can divide hundreds of times (Ito and Hotta, 1992 White and Kankel, 1978) (Figure 2B). Several types of larval neuroblasts can be distinguished based on their position within the larval cen- tral nervous system (Figure 2C). In the ventral nerve chord, 30 ventral nerve chord neuroblasts per hemisegment divide repeatedly along the apical-basal axis to form the neurons of the thoracic and abdominal ganglia (Truman and Bate, 1988). In eachofthetwobrainlobes,approximately 85centralbrainneuro- blasts give rise to most of the neurons present in the adult brain (Ito and Hotta, 1992). Central brain neuroblasts are heteroge- neous in cell cycle length and regulation of self-renewal. In partic- ular, a group of less than 10 dorso-posterior (DP) neuroblasts seems to be particularly susceptible to mutations in tumor sup- pressor genes (Betschinger et al., 2006). Compared to other cen- tral brain neuroblasts, these precursors generate many more progeny and they might even have a different lineage in which GMCs divide more than once. It is worth noting that much of the earlier experiments on Drosophila larval neuroblasts did not distinguish between these subgroups and might have to be reinvestigated. In addition to the central brain neuroblasts, the fly brain contains the mushroom body and optic lobe neuroblasts. In each brain hemisphere, four mushroom body neuroblasts give rise to 2500 neurons called Kenyon cells that form the learning and memory centers (Ito and Hotta, 1992 Ito et al., 1997). To generate this large number of neurons, they start dividing much earlier than central brain neuroblasts and proliferate throughout most of the pupal stages of development. Whereas mushroom body and central brain neuroblasts are already spec- ified during embryogenesis and simply reactivate their prolifera- tion programs during larval stages, optic lobe neuroblasts follow a distinct program of neurogenesis (Egger et al., 2007). They arise from two multilayered neuroepithelia called the inner- and outer-proliferation centers (White and Kankel, 1978). Neuroepi- thelial cells divide symmetrically in parallel to the epithelial surface. Neuroblasts are generated on the rims of these epithe- lia. They lose their adherens junctions, turn on neuroblast markers, and start dividing asymmetrically and perpendicularly to the epithelial plane. Following a canonical neuroblast lineage, optic lobe neuroblasts give rise to the neurons in the visual processing centers of the fly brain. Segregating Determinants The different fate of the two neuroblast daughter cells is thought to be induced by the unequal segregation of several proteins into one of the two daughter cells (Figure 3). Due to their combined activity in specifying daughter cell fate, these proteins are referred to as segregating determinants. Because determinant segregation can even occur in individual cultured neuroblasts, it is thought to be governed by a cell-intrinsic machinery (Broadus and Doe, 1997) (note, however, that partially redundant extrinsic cues exist as well���see below and Siegrist and Doe, 2006). Before mitosis, the proteins Par-3, Par-6, atypical PKC (aPKC), Inscuteable, Pins, Gai, and Mud (see below for their individual functions) accumulate on the apical side of the cell cortex (Betschinger and Knoblich, 2004 Suzuki and Ohno, 2006 Goldstein and Macara, 2007). Although they are preferen- tially inherited by the apical daughter cell, which remains a neuro- blast, they are not thought to influence cell fate directly. Instead, they induce the asymmetric localization of cell fate determinants to the opposite, basal side of the cell and their segregation into the basal GMC. Below, I discuss those determinants for which functions in Drosophila neural stem cells have been shown. The first segregating determinant was called Numb and was actually identified in SOP cells (Rhyu et al., 1994). In numb mutants, both daughter cells of the SOP assume the fate of the cell that normally does not inherit the Numb protein. Conversely, Figure 1. Extrinsic and Intrinsic Regulation of Stem Cell Self- Renewal (A) Stem cells can set up an axis of polarity during interphase and use it to localize cell fate determinants asymmetrically in mitosis. Orientation of the mitotic spindle along the same polarity axis ensures the asymmetric segrega- tion of determinants into only one of the two daughter cells. (B) Stem cells may depend on a signal coming from the surrounding niche for self-renewal. By orienting their mitotic spindle perpendicularly to the niche surface, they ensure that only one of the two daughter cells continues to receive this signal and maintains the ability to self-renew. 584 Cell 132, 583���597, February 22, 2008 ��2008 Elsevier Inc.

numb overexpression results in the transformation into the oppo- site cell fate. Numb acts as a tissue-specific repressor of the Notch pathway (Le Borgne et al., 2005 Schweisguth, 2004). It binds to the endocytic protein a-Adaptin (Berdnik et al., 2002) and might control the intracellular trafficking of Notch intermedi- ates. When Numb is mutated in the larval brain, the mutant neuroblasts overproliferate and form a tumor-like phenotype (Lee et al., 2006a Wang et al., 2006). Lineage analysis shows that this is due to occasional divisions in which a neuroblast still divides into a larger and a smaller daughter cell but both daugh- ter cells eventually show the gene expression and proliferation pattern of a neuroblast. Similar (but not identical) brain pheno- types are observed upon mutation of other segregating determi- nants and have made Drosophila neuroblasts an ideal model system to investigate the biology of cancer stem cells (Figure 4, see below) (Gonzalez, 2007). Like Numb, the transcription factor Prospero (Pros) segre- gates asymmetrically in neuroblasts. Although Pros is already present in neuroblasts, it only enters the nucleus once asymmet- rically segregated into the GMC (Betschinger and Knoblich, 2004). When Pros is mutated in embryonic neuroblasts, the GMC continues to express neuroblast markers and undergoes multiple rounds of division (Choksi et al., 2006). Several cell- cycle regulators including Cyclins A and E and Cdc25 (string in Drosophila) are upregulated and may be responsible for this phenotype (Li and Vaessin, 2000). In larval neuroblasts, muta- tions in Pros cause stem cell-derived tumors (Betschinger et al., 2006 Lee et al., 2006c Bello et al., 2006). Pros contains a homeodomain and binds upstream of over 700 genes many of which are involved in neuroblast self-renewal or cell-cycle control. However, Pros can also induce the expression of neural differentiation genes indicating that it can act both as a transcrip- tional activator and inhibitor (Choksi et al., 2006). More recently, a third important regulator of neuroblast self- renewal has been identified (Lee et al., 2006c Bello et al., 2006 Betschinger et al., 2006). This protein is called Brat and was previously shown to act as an inhibitor of ribosome biogen- esis and cell growth (Frank et al., 2002). Brat is a member of a new conserved protein family that is characterized by the pres- ence of a C-terminal NHL domain, a coiled-coil region and an N-terminal Zinc binding B-box (Slack and Ruvkun, 1998). In Drosophila, Brat, Mei-P26, and Dappled are members of this family. Given that all three proteins act as tumor suppressors, growth control might be a common function of this protein family. During embryogenesis, Brat cooperates with Pros to specify GMC fate. Although only a small subset of GMCs is affected in pros mutants, pros/brat double mutants show an almost complete loss of all GMCs (Betschinger et al., 2006). In larval brains, brat causes the formation of stem cell-derived tumors consisting almost entirely of large cells expressing neuroblast markers. This has led to the hypothesis that Brat might inhibit cell growth in one of the two neuroblast daughter cells to prevent self-renewal and induce terminal differentiation. The molecular mechanism by which Brat regulates cell growth and cell fate is currently unknown. Brat has a second function in specifying the anterior-posterior body axis and for this function, it binds to Nanos and Pumilio to repress translation of the posterior iden- tity gene hunchback (Sonoda and Wharton, 2001). In neuro- blasts, however, neither the phenotypes nor the expression patterns of Nanos, Pumilio or Hunchback suggest that Brat acts in a similar manner. Instead, Brat was suggested to be a transcriptional activator of Pros (Lee et al., 2006c Bello et al., 2006) because brat mutant tumors are Pros negative and overexpression of Pros can rescue tumor formation in brat mutants. However, this hypothesis neither explains why brat enhances the pros null mutant phenotype in embryonic neuro- blasts nor why it regulates cell growth even in tissues that do not express Pros. Given that brat tumors arise specifically in DP neuroblasts (see above), lower expression levels of Pros in these cells would also explain why the brat tumors are Pros Table 1. Model Systems for Asymmetric Cell Division Mother Cell Daughter Cell Types Polarity Cue Mechanism of Unequal Fate Specification Drosophila sensory organ precursor cell Four cell types forming external sensory organs: socket, hair, sheath, neuron Planar polarity Asymmetric segregation of Numb results in differential Notch regulation Drosophila neuroblast Neuroblast, ganglion mother cell Epithelial polarity Asymmetric segregation of Numb, Prospero, and Brat results in self-renewal versus cell-cycle exit Drosophila ovarian germline stem cell Stem cell, cystoblast Niche architecture Diffusible signal (Dpp and Gbb) from stem cell niche Mouse brain progenitor cells Progenitor cell, neuron (occasionally: intermediate/basal progenitor) Apical-basal polarity of neuro- epithelium Unidentified segregating determinant or apical membrane compartment or basal fiber Mouse muscle satellite cells Stem cell (Myf5 ), committed progenitor (Myf5+) Unclear, maybe integrin contact with basal lamina Segregating determinant (Numb), signal from basal lamina or both Mouse hematopoietic stem cell Hematopoietic stem cell, committed progenitor Signal from stem cell niche (blood vessel or osteoblast) Different levels of Notch signaling (maybe induced by Numb segregation) Mouse T-lymphocytes Effector T cell, memory T cell Immunological synapse Unequal segregation of Numb, CD8 and Interferon g receptor Cell 132, 583���597, February 22, 2008 ��2008 Elsevier Inc. 585