Strategies for enrichment and selection of stem cell-derived tissue precursors

Human embryonic stem cells have the capacity for self-renewal and pluripotency and thus are a primary candidate for tissue engineering and regenerative therapies. These cells also provide an opportunity to study the development of human tissues ex vivo. To date, numerous human embryonic stem cell lines have been derived and characterized. In this review, we will detail the strategies used to direct tissue-specific differentiation of embryonic stem cells. We also will discuss how these strategies have produced new sources of tissue-specific progenitor cells. Finally, we will describe the next generation of methods being developed to identify and select stem cell-derived tissue precursors for experimental study and clinical use.


Introduction
Stem cells have the ability to maintain long-term proliferation and self-renewal. Under specifi c conditions, stem cells can diff erentiate into a diverse population of mature and functionally specialized cell types. Th ere are two main types of human stem cells classifi ed according to their source and developmental potential: embryonic and adult, or tissue-specifi c, stem cells. Human embryonic stem cells (hESCs) are pluripotent cells that can diff erentiate into all types of somatic -and, in some cases, extraembryonic -tissues. Human adult stem cells are derived from non-embryonic tissues and are capable of generating specifi c cells from the organ or tissue of origin. Because of the unrestricted potential of hESCs, these cells have become a highly desirable experi mental tool for understanding human development and are especially attractive for therapeutic applications. In addition, methods for inducing a pluripotent state in human somatic cells have created the opportunity to study and use patient-specifi c stem cells, or induced pluri potent stem cells (iPSCs), and their derivatives. For these reasons, methods for directing the diff erentiation of pluripotent stem cells and selecting these for analysis and clinical use have become an area of intense investigation.

Enrichment of tissue-specifi c precursors
Pluripotency is defi ned as the potential of a stem cell to diff erentiate into tissues originating from all of the three embryonic germ layers: endoderm (for example, gastrointestinal tract and lung), mesoderm (for example, heart, muscle, bone, blood, and urogenital), and ectoderm (for example, nervous system and epidermis). At the molecular level, highly regulated transcriptional circuitry as well as extrinsic factors control the pluripotent state [1,2]. In addition, chromatin remodeling and small noncoding RNAs have been implicated in the regulation of pluripotency [2,3]. Pluripotency can be tested by using in vivo and in vitro methods. A test of pluripotency in vitro involves determining the ability of hESCs and iPSCs to form human embryoid bodies (hEBs) when cultured in a non-adherent cell suspension in the absence of feeder cell layers. hEBs are spherical colonies of diff erentiating stem cells that contain cell types representative of all three embryonic germ layers [4]. Th e most commonly used in vivo method to test pluripotency involves the trans plantation of undiff erentiated stem cells into immunodefi cient mice to induce the formation of teratomas [4][5][6]. Teratomas are benign tumors composed of disorganized tissue structures characteristic of the three embryonic germ layers. Analysis of embryonic tissues found in teratomas from engrafted stem cells can be used to test their diff erentiation potential ( Figure 1).
Th e ability of hESCs and iPSCs to mimic in vitro and in vivo the events occurring during human development makes them not only valuable tools for understanding the mechanisms involved in developmental processes but also stepping stones toward the generation of desired cell types suitable for cell therapies. Recent studies have shown that it is possible to generate lineage-restricted pro genitors that are capable of diff erentiating into specialized post-mitotic cell types such as cardio myocytes, pancreatic islet cells, chondrocytes, hematopoietic cells, endothelial cells, and neurons. Furthermore, the ability of pluripotent stem cells to divide indefi nitely makes these a potential large-scale source of specifi c progenitors. In the following sections, we will provide examples of how stem cell diff erentiation can be directed toward specifi c cell/tissue types.

Developmental programming
Some of the most eff ective strategies for directing the diff erentiation of pluripotent stem cells into specifi c cell types have taken advantage of our understanding of human development (Table 1). Endodermal derivatives include cells that populate the lung, liver, and pancreas. Directing the diff erentiation of hESCs and iPSCs toward defi nitive endoderm would help generate specifi c cell types, such as islet cells or hepatocytes, which could be used in the treatment of diseases such as diabetes or liver disease, respectively. D' Amour and colleagues [7] showed that selective induction of endoderm could be achieved through the addition of high concentrations of Activin A, under low serum conditions, and in a stage-specifi c manner. Activin A mimics the action of Nodal, a ligand that activates transforming growth factor-beta (TGFβ) signaling, which in turn leads to the induction of endoderm diff erentiation. Th e eff ect of Activin A in inducing defi nitive endoderm is enhanced when additional factors such as Wnt3a [8] and Noggin [9] are present or when coupled with the suppression of the phosphoinositide 3-kinase pathway [10].
Induction of defi nitive endoderm can lead to the genera tion of specifi c progenitor populations after the addition of other factors. Among the most successful examples to date is the generation of pancreatic islet progenitors, devised by Kroon and colleagues [11] and accomplished through the sequential exposure of hESCs to Activin A and Wnt3A, followed by the addi tion of keratinocyte growth factor or fi broblast growth factor 7 to induce the formation of the primitive gut tube. Subsequently, retinoic acid, cyclopamine, and Noggin are added to inhibit hedgehog and TGFβ signaling and thus induce the diff erentiation of posterior foregut cells, the source of pancreatic cell progenitors. Th ese are cultured further to generate pan creatic endoderm cells. When engrafted in immuno defi cient mice, these cells display the histological and structural characteristics of pancreatic islet cells and are able to sustain insulin production for at least 100 days [11].
In a similar manner, hepatocytes can be obtained after diff erentiation of hESCs into defi nitive endoderm [12,13]. A robust population of functional hepatocytes was generated with the sequential addition of low serum medium, collagen I matrix, and hepatic diff erentiation factors that include fi broblast growth factor (FGF), bone morphogenetic protein 4 (BMP4), hepatocyte growth factor, oncostatin M, and dexamethasone [13]. Th ese cells expressed known markers of mature hepatic cells, exhibited appropriate function, and were able to integrate and diff erentiate into mature liver cells when injected into mice with liver injury [13].
Directing the diff erentiation of stem cells into mesoderm requires activation of the TGFβ signaling pathway and can be accomplished through the stepwise and dosage-dependent addition of Activin A, BMP4, vascular endothelial growth factor (VEGF), and basic fi broblast growth factor (bFGF) [14]. Meso dermal derivatives have also been successfully obtained by spontaneous diff erentiation of hESCs through hEB formation without fi rst directing them toward mesoderm. Robust diff erentiation of hESCs into hematopoietic lineage cells, which give rise to all blood cell types and components of the immune system, has been achieved under serum-free conditions through spin hEB formation [15]. Specifi c hematopoietic cells, such as functional dendritic cells, have been success fully diff erentiated from hESCs through spontaneous hEB formation under serum-free conditions with the addition of BMP4 at specifi c time points [16]. Hemato poietic progenitor cells that give rise to functional T and natural killer cells capable of targeting human tumor cells both in vitro and in vivo have also been derived from hESCs co-cultured with stromal cells [17]. Th us, the ability to diff erentiate hESCs into hematopoietic lineage cells promises to be useful in improving existing therapies that require blood cell transplantation and in immune therapies that require induction of the immune response in an antigen-specifi c manner [18].
Cardiomyocytes, which represent another thera peutically important derivative of mesoderm, have been successfully generated from pluripotent stem cells by using several methods [19]. Spontaneous diff eren tiation of stem cells under appropriate culture conditions can produce cardiomyocytes that exhibit morphological, mole cular, and electrophysiological properties similar to those of adult cardiomyocytes [20] and display quantifi able responses to physiological stimuli reminiscent of atrial, ventricular, and pacemaker/conduction tissue Table 1

Embryonic germ layers Diff erentiation factors or culture conditions or both Example of diff erentiated cells
Endoderm FGF, BMP4, hepatocyte growth factor, oncostatin M, dexamethasone Hepatocytes [12,13] [21 -24]. Cardiomyocytes have also been generated by directed diff erentiation with Activin A and BMP4 of a dense monolayer of stem cells; these cells successfully form specifi c cardiac lineages when transplanted in vivo [25]. Another study used additional medium supplements that included VEGF and the Wnt inhibitor, DKK1, followed by the addition of bFGF to promote cardiomyocyte diff erentiation in culture [26]. Th e success of these studies was measured by the expression of proteins specifi c for mature cardiac cells, such as cardiac troponin T, atrial myosin light chain 2, and the cardiac transcription factors Tbx5 and Tbx20. Th e dominant diff erentiation pathway in hESC cultures leads to the formation of ectoderm, which makes up cells of the nervous system and the epidermis. hESC-derived neural progenitor cells are characterized by rosette-like neural structures that form in the presence of growth factors FGF2 or EGF either through spontaneous diff erentiation from an overgrowth of hESCs or after hEBs are plated onto adherent substrates [27,28]. Th ese neural rosettes have become the signature of hESC-derived neural progenitors, capable of diff erentiation into a broad range of neural cells in response to appropriate developmen tal signals. Th us, many studies are exploring ways to enhance the formation of neural rosettes in order to generate an enriched population of specifi c neural cell types. One example is the use of specifi c stromal cell lines [29]. With this method, stromal cells provide ectodermal signaling factors required for neural induction, as determined in animal model studies, and therefore promote the formation of neural rosettes [30,31].
Th e withdrawal of FGF2 and EGF and the addition of specifi c compounds can lead to the diff erentiation of neural rosettes into specifi c neural subtypes. For example, hESC-derived neural progenitors treated with FGF8 and sonic hedgehog give rise to dopaminergic neurons [32], whereas treatment with sonic hedgehog and retinoic acid induces motor neuron diff erentiation [33]. Neural crest stem cells derived from neural rosettes can diff erentiate into peripheral sympathetic and sensory neurons by withdrawing FGF2/EGF and adding brain-derived neurotrophic factor (BDNF), glial cell-derived neurotrophic factor (GDNF), nerve growth factor (NGF), and dibutyryl cyclic adenosine monophosphate (dbcAMP) or into Schwann cells in the presence of ciliary neuro trophic factor (CNTF), neuregulin 1β, and dbcAMP [34]. Neuroglial cells, such as oligodendrocytes, are generated with B27, thyroid hormone, retinoic acid, FGF2, epidermal growth factor, and insulin [35].

Chemical enrichment
In some cases, the manipulation of key pathways during germ layer development has allowed signifi cant enrichment of specifi c cell types without the complete recapitu lation of embryonic development. Treatment with 5-aza cy ti dine at days 6 to 8 of hESC diff erentiation signifi cantly increases cardiac alpha-myosin heavy chain (αMHC) expression and enhances cardiomyocyte diff erentiation, suggesting that DNA demethylation is a key factor in directing tissue-specifi c diff erentiation [36]. Similarly, exposure to SB203580, a small-molecule inhibitor of p38 MAPK , has been shown to signifi cantly improve cardiomyocyte diff erentiation of hESCs grown in medium conditioned by mouse END2 cells, supporting a role for p38 MAPK signaling in regulating human cardiomyocyte diff erentiation [37]. SB203580-treated hEBs display an increase in expression of both early mesoderm markers (Brachyury T, Tbx6, and Mesp1) and cardiac αMHC as well as increased cardio myocyte numbers. Gaur and colleagues [38] subsequently showed that p38 MAPK inhibition occurs in a dose-and stage-dependent manner, that it also causes the accelerated diff erentiation of hESC-derived cardiomyocytes by using the standard hEB formation method, and that it appears to act at the ectoderm/mesoendoderm branch point during hESC diff erentiation.
In the original study with SB203580, cells were subjected to an adapted diff erentiation system in which hESCs were diff erentiated in a suspension culture by using serum-free medium conditioned by the mouse END2 cell line [39]. END2-conditioned medium alone exhibits cardiomyocyte-inducing activity during hESC diff erentiation [37], and biochemical as well as microarray analysis of END2-conditioned medium and END2 cells, respectively, identifi ed PGI2, a product of prostaglandin synthase enzymes, as an inducing factor in hESC cardiac diff erentiation [40]. Two key enzymes involved in PGI2 synthesis are upregulated in END2 cells compared with control MES1 cells [41], which lack cardiogenic activity. PGI2 levels are between 6-and 10-fold higher in END2-conditioned medium compared with control conditioned medium from MES1 cells. More over, insulin, a common supplement in media formu lations, was discovered to be an inhibitor of hESC cardiac diff erentiation. END2-conditioned medium supple mented with increasing concentrations of insulin results in a dramatic decrease in hESC cardiomyocyte diff erentiation. Th us, addition of PGI2 in combination with insulin-free, unconditioned medium yields eff ective cardiac induction similar to that produced by END2conditioned medium. Cardiac diff erentiation is further augmented in the presence of SB203580. Taken together, these three components provide a basic synthetic recipe for directing cardiomyocyte diff erentiation of hESCs.

Mechanical factors that infl uence diff erentiation
Since cardiac muscle is one of the few tissues that develop under the eff ects of dynamic force, it is not surprising that conditions generated by the force of fl uids in motion can enhance cardiomyocyte diff erentiation as well. Supply ing a constant rotary orbital motion for 7 days to suspension cultures of diff erentiating mouse EBs (mEBs) results in a signifi cantly increased number of beating mEBs compared with mEBs cultured in static suspension [42]. Analysis of gene expression shows higher levels of mesodermal and cardiac proteins (Brachyury, GATA4, Nkx2-5, MEF2c, αMHC, and MLC2v) in rotary mEBs than in static mEBs. In addition, a greater proportion of rotary mEBs stain positive for α-sarcomeric actin compared with static EBs. Th e enhanced cardiomyocyte diff erentiation is independent of rotary speed ranging from 25 to 55 revolutions per minute as determined by the expression of cardiomyogenic genes [43].
Domian and colleagues [44] have examined the eff ects of surface tension on cardiomyogenic diff erentiation of murine cardiac progenitors. Progenitors derived from embryos and mouse ESCs (mESCs) are cultured on either fi bronectin-coated slides or micropatterns of fi bronectin alternating with a surfactant that blocks cell adhesion. When grown on these micropatterned surfaces, a population of cells form longitudinally aligned myocardial fi bers. In addition, culturing this population on micropatterned surfaces results in a statistically signifi cant increase in the proportion of cardiomyocytes, supporting a role for microenvironmental forces in cardiac muscle diff erentiation.
Studies of substrate stiff ness and elasticity during stem cell diff erentiation have also demonstrated eff ects on skeletal muscle and bone development. Myosin/actin striations occurred only when myoblasts were cultured on gels with stiff ness typical of normal muscle [45]. In addition, muscle stem cells cultured on hydrogel substrates that mimic the elasticity of muscle self-renew and fuse to existing myofi bers with greater effi ciency than cells grown on rigid plastic [46]. Substrate stiff ness has also been shown to favor mesodermal formation from ESCs, specifi cally osteogenic diff erentiation [47].
Th ese observations, as well as an evolving under standing of the tissue environment in which stem cells diff erentiate in vivo, underlie the recent application of biomaterials to promote the diff erentiation and retention of stem cells in a number of tissues [48,49]. For example, optimal sizing of pancreatic β-cell clusters derived from hESCs is essential to preventing nutrient and oxygen deprivation in β-cell transplant therapy for diabetes. As shown by Van Hoof and colleagues [50], the size of β-cell aggregates can be controlled by using micropatterned laminin. Polymer scaff olds have also been studied to facilitate transplantation of retinal progenitor cells for the treatment of macular degeneration. Steedman and colleagues [51] demonstrated that biodegradeable thin fi lms of polycaprolactone could be topographically fabri cated to enhance the attachment and organization of progenitor cells and induce their diff erentiation to photoreceptor cells.

Epigenetic reprogramming
MicroRNAs (miRNAs) are small non-coding RNAs thought to regulate the expression of 30% of proteincoding genes [52]. Th e biological importance of these RNAs in stem cell biology is underscored by recent studies demonstrating that mESCs lacking the miRNAprocessing enzyme Dicer display diff erentiation and proliferation defects [53][54][55][56]. miR-1 and miR-133 specifi cally are expressed in the mouse heart [57,58]. Targeted deletion or knockdown of these miRNAs results in dysregulation of cardiac morphogenesis, electrical conduction, cell-cycle, and cardiac hypertrophy [57][58][59][60]. Recently, Ivey and colleagues [61] showed that miR-1 and miR-133 regulate the diff erentiation of mESCs and hESCs into the cardiac lineage. Both miRNAs are enriched in mESCderived cardiomyocytes. Lentiviral introduction of either miR-1 or miR-133 into mESCs enhances early mesoderm diff erentiation as evidenced by increased expression of Brachyury. miR-1 and miR-133 also reinforce mesoderm lineage decisions by repressing endoderm and neuroectoderm diff erentiation. When stimulated to diff erentiate into either endoderm or neuroectoderm lineages, mEBs expressing either miR-1 or miR-133 express lower levels of endodermal and neural markers compared with control mEBs. However, further diff erentiation revealed opposing roles of miR-1 and miR-133. miR-1 promotes diff erentiation of mesoderm into the cardiac and skeletal muscle lineages as determined by enhanced Nkx2-5 and myogenin expression, respectively, whereas miR-133 blocks induction of both markers. Importantly, the diff erentiation of hESCs in the presence of miR-1 behaves comparably to mESC diff erentiation. Overexpression of miR-1 in hESCs increases Nkx2-5 expression and yields more than a threefold higher number of beating hEBs compared with wild-type controls.
Whereas miRNAs direct cell lineage determination by controlling protein dosage, epigenetic regulation through chromatin remodeling has been shown to control cell fate as well. Takeuchi and Bruneau [62] identifi ed a minimal set of factors necessary to execute the cardiac transcriptional program. Baf60c, a cardiac-enriched subunit of the Swi/Snf-like BAF chromatin remodeling complex, in combination with cardiac transcription factors GATA4 and Tbx5, is able to induce cardiac diff erentiation in mouse embryos when ectopically expressed. With this combination, 90% of the transfected embryos display expression of the early cardiac marker Actc1 and 50% of the transfected embryos exhibit beating tissue. GATA4, together with Baf60c, is essential in initiating the cardiac gene program as assessed by expression of Actc1. Neither of the other transcription factors tested alone (Tbx5 and Nkx2-5) or in concert with Baf60c is able to induce Actc1 expression. GATA4/Baf60c, however, is not suffi cient for generating spontaneously contracting embryonic tissue: Tbx5 is required to achieve beating cardiomyocytes.

Selection of tissue-specifi c precursors
As discussed earlier in this review, enrichment of specifi c cell types can be achieved by using molecules introduced at specifi c time points during culture. However, many of these methods yield only moderate enrichment that is not yet scalable for clinical application. In addition, it may be desirable to enrich fi rst for partially diff erentiated, proliferative stem cell intermediates with specifi c fates. Th ese could then be expanded before further diff er entiation into cells for therapy. In the following sections, we will discuss the variety of methods used to select tissuespecifi c precursors and their derivatives.

Use of cell surface markers
Even before the tools of genetic engineering were employed to manipulate stem cells, proteins expressed on the cell surface were used to identify distinct populations of stem cells and their ontogeny. Perhaps the best-known application of this approach is to the intermediates of hematopoiesis [63]. Over the past 10 years, similar attempts to create lineage maps of other tissues have focused on the identifi cation of lineage-specifi c cell surface markers (Table 2). For example, pluripotent stem cells can be identifi ed by the expression of stage-specifi c embryonic antigen (SSEA) 3 and 4 and the embryonal carcinoma marker Tra-1-60 on their surface [4]. Th e expression of the cell surface antigen CD133 on proliferating hESCs identifi es cells predestined toward a neuroectodermal fate [4]. Hemangioblasts that ultimately give rise to hematopoietic stem cells, smooth muscle progenitors, and endothelial progenitors can be selected on the basis of the surface expression of CD143, Ecadherin, VE-cadherin, platelet endothelial cell adhesion molecule 1 (PECAM-1), and vascular endothelial growth factor receptor 2 (VEGFR2) [64]. Mesenchymal stem/ stromal cells that diff erentiate into muscle, fat, cartilage, and bone cells can be identifi ed by the expression of the cell surface proteins CD105/endoglin, CD73, and CD90/ Th y-1 in the absence of surface expression of CD45, CD34, CD14 or CD11b, CD79α or CD19, and HLA-DR [65]. Recently, cardiomyocytes that have, until now, eluded detection by specifi c surface protein expression were shown to express signal-regulatory protein alpha (SIRPA) [66].
Two main approaches have been used to identify lineage-specifi c surface markers. One employs labeled cell fate mapping to assess the tissue-specifi c fate of subpopulations of pluripotent stem cells expressing specifi c surface proteins [4], whereas the other uses a high-throughput fl ow cytometry screen of available antibodies against stem cell derivatives [66]. Although there are still large gaps in our ability to identify and select derivatives of all three germ layers on the basis of surface marker expression, progress continues to be made.

Biochemical purifi cation
Although detection of lineage-specifi c genes underlies most methods for selecting specifi c cell types from diff erentiating stem cell cultures, a few cell types, such as cardiac muscle cells, bear unique biochemical properties that aid in their isolation. For example, Percoll gradient centrifugation has been reported to purify hESC-derived cardiomyocytes [36,67], based on their buoyancy character istics. Diff erentiating hESCs are applied to a discontinuous Percoll gradient consisting of 40.5% Percoll layered over 58.5% Percoll. After centrifugation, the majority of cardiomyocytes reside within the 58.5% Percoll layer and express cardiac troponin I, sarcomeric MHC, αMHC, βMHC, and N-cadherin. hESC-derived cardiomyocytes of almost 70% purity have been obtained by using this approach.
A second purifi cation strategy is based on the observation that cardiomyocytes have high mitochondrial content compared with that of non-myocytes [68]. Using the fl uorescent dye tetramethylrhodamine methyl ester (TMRM), which freely diff uses into the mitochondrial matrix, Hattori and colleagues [68] found that TMRM fl uorescence in embryonic rat hearts increases with developmental stage, suggesting that mitochondrial biogenesis is linked to myocardiogenesis. In whole rat embryos, TMRM fl uorescence in the heart is more robust than in other tissues, and when analyzed by fl ow cytometry, fl ow-sorted populations with the highest TMRM fl uorescence are observed to express cardiac αactinin. TMRM-labeled cardiomyocytes derived from mESCs are positive for both Nkx2-5 and α-actinin. Th e cardiomyocyte content in cultured cells sorted from day-12 to day-25 mEBs is greater than 99% as determined by Nkx2-5 and α-actinin expression. Most notably, greater than 99% cardiomyocyte purity is also obtained in cultured cells sorted from diff erentiating hEBs.

Transgenic reporter cell lines
Th e use of fl uorescent reporters of gene expression off ers an approach to selecting tissue-specifi c cells by highthroughput fl uorescence-activated cell sorting [69]. Alternatively, the use of reporters that activate expression of a selectable marker can drive a population of cells to near-homogeneity by restricting survival to a specifi c cell type. Th e paucity of specifi c cell surface markers with which to select cardiac cells has led to the development of a variety of reporter lines with which to select stem cell-derived cardiomyocytes and cardiac progenitors. Huber and colleagues [70] used lentiviral vectors to produce stable hESC lines in which enhanced green fl uorescent protein (eGFP) is expressed under control of the cardiac-specifi c human ventricular myosin light chain 2 promoter. Xu and colleagues [71] generated stable hESC lines by using a reporter plasmid consisting of the cardiac-specifi c mouse αMHC promoter driving expression of the neo mycin resistance gene. Kita-Matsuo and colleagues [72] designed a set of lentiviral vectors to generate multiple stable hESC lines with eGFP and mCherry reporters or with puromycin resistance downstream of the mouse αMHC promoter. Ritner and colleagues [23] generated a cardiac-specifi c hESC reporter line by using a lentiviral construct consisting of Glial-restricted progenitor Oligodendrocyte FGFR [110][111][112] MOG, CD140a [113,114] Type 2 astrocyte Type 1 astrocyt FGFR3 [110,111] Motor neuron progenitor Motor neuron Surface ectoderm Epidermal stem cell Epithelial cell α6-integrin high , CD71 low [115] CEACAM-1, EpCAM [116,117] Mesoderm Skeletal muscle α7β1-integrin [122] Pre-adipocyte Adipocyte Glut4 [123] Chondrocyte precursor Chondrocyte CD44 [124] Osteoblast Osteocyte Endoderm Primitive endoderm Defi nitive endoderm Hepatic progenitor Hepatocyte c-Met, CD49f [125] Il-6 [126] Pancreatic progenitor α cell β cell Glut2 [127] PPγ cell a fragment of the mouse αMHC promoter upstream of eGFP. Th e specifi c promoter fragment allowed the identifi cation and analysis of early cardiac progenitors expressing Nkx2-5 but before the onset of cardiac troponin T or chamber-specifi c MLC expression [23]. Whereas isolation of hESC-derived cardiomyocytes from these reporter lines is based on positive selection, Anderson and colleagues [73] implemented a negative selection strategy to deplete undiff erentiated proliferating hESCs from cultures of hESC-derived cardiomyocytes. Th eir hESC line used a herpes thymidine kinase/ gancyclovir (HTK/GCV) suicide gene system under the control of a constitutive phosphoglycerate kinase promoter. After administration of the antiviral drug GCV, cells expressing HTK-phosphorylated GCV incorporated the guanosine analog into nascent DNA chains of proliferating cells and caused chain termination and cell death. Th e increased number of α-actinin-positive cells after GCV treatment led to an almost seven fold enrichment of cardiomyocytes. An important caveat of this approach is that other non-proliferating cell types would remain in the culture while proliferating cardiomyocytes would be depleted. Th e culture would still need to undergo a cardiac purifi cation step, and the excluded proliferating cardiomyocytes and cardiac progenitors may be of greater benefi t for transplantation than fully diff erentiated, non-proliferating cells [19].
Reporter lines have similarly been derived for the isolation of other tissue-specifi c cells of biological and clinical interest, including motor neurons [74], germ cells [75], hepatocytes [76], and pancreatic progenitors [77]. However, the eff ects of random genomic integration on disrupting endogenous gene expression, as well as unregulated transgene expression, are potential limitations to this approach. Reporter lines must be rigorously evaluated to confi rm authentic representation of gene expression without altering cell behavior [23].

Transgenesis through homologous recombination
To address the limitations of virally mediated intro duction of reporter transgenes (for example, constitutive transgene expression and eff ects of random integration), many investigators have approached reporter development by using homologous recombination to 'knockin' a fl uorescent protein into a specifi c genetic locus [78]. Initially, gene targeting of ESCs by homologous recombination was slowed by the diffi culties of single-cell cloning [79][80][81][82], transfection effi ciency using conventional methods [83,84], and karyotypic instability with dissociation of ESCs [85,86]. However, these obstacles have been overcome to some degree, and several successfully targeted ESC lines have resulted [78].
Th e hESC reporter lines currently in use represent a variety of targeting strategies. An H1.1-derived OCT4-eGFP line targeting exons 3 to 5 of the octamer-binding transcription factor 4 gene by positive selection alone has been used to sort pluripotent hESCs [83]. Similarly, an MIXL1-GFP line targeting exon 1 of the Mix1 homeobox-like 1 gene in HES3 cells by positive selection has been used to identify blood precursors [79]. Ruby and Zheng [84] used the FEZ family zinc fi nger 2 promoter to direct eYFP expression in HUES9 cells as a means of isolating neuronal precursors by using cre-mediated recombination with both positive and negative selection steps, and a similar approach using the oligodendrocyte lineage trans cription factor 2 promoter driving eGFP expression in BG01 hESCs identifi ed a subset of glial-restricted progenitors with oligodendrocyte fate [87]. Cre-mediated homologous recombination has also been used to generate two cardiac reporter lines, one using the second heart fi eld-specifi c Islet-1 promoter directing dsRed expression in H9 hESCs [81] and another recently described pair of hESC lines in which the promoter of the early cardiac transcription factor NKX2-5 was used to drive eGFP expression in MEL1 (male) and HES3 (female) cells [88]. While these accomplishments demon strate that the various unrelated ESC lines currently available are all amenable to homologous recombination, optimal targeting vector designs and selection strategies are still being explored.

Use of dual-FRET molecular beacons
Although homologous recombination technology avoids some of the concerns with transgenic reporter cell lines generated by random integration, the resulting cells have still undergone genomic modifi cation. Seeking an alternative to genetically modifi ed stem cell lines for tracking and isolating tissue-specifi c pro genitors, King and colleagues [6] adapted dual-fl uores cence resonance energy transfer (FRET) 'molecular beacon' technology for transient, real-time detection of gene expression during ESC diff erentiation. Molecular beacons are singlestranded oligonucleotide probes that have been employed to assay gene expression in vitro, as in real-time polymerase chain reaction assays, and in vivo by using microscopy [89]. Th ese consist of short sequences capable of forming stem-loop structures bearing a fl uorescent reporter group at one end and a fl uorescent quencher at the opposite end [89]. In the absence of a target sequence, the oligonucleotide self-anneals, forming a stem that brings the reporter and quencher in close proximity, thereby quenching fl uores cence (Figure 2). In the presence of a target sequence, the oligonucleotide anneals to the target, separating the reporter and quencher, thereby allowing fl uorescence. Th ese investigators showed that appropriately designed, dual-FRET molecular beacon pairs can identify the expression of specifi c mRNAs by microscopy and fl ow cytometry and facilitate the collection of specifi c stem cell populations by cell sorting while leaving the stem cell genome intact [6].

Spectral fl ow cytometry: next-generation analysis
Conventional fl ow cytometers and cell sorters use a series of dichroic mirrors and fi lters matched to the fl uorescence emission profi le of reporter fl uorochromes to identify and quantify specifi c biomarkers by using photomultiplier tubes (PMTs) to detect each parameter. Historically, this technology has relied on dichroic mirrors that refl ect light of specifi c wavelengths and bandpass fi lters that allow incident light within a specifi c wavelength range to get to the detector. Th e detector, or PMT, detects photons and turns them into electrons or current. Th e fl uorescent photon hits a photocathode and turns into an electron on a one-to-one basis; the electron then passes through a series of dynodes that amplify the electrons on the basis of the voltage generated across the dynode. One electron is multiplied and becomes two or more electrons, creating a cascade of electrons based on the single initial photon. At the end of the dynode chain, an anode collects the electron wave into an electronic signal; the more photons delivered to the PMT, the bigger the electronic signal.
A key criterion for distinguishing specifi c biomarkers over background is the optical signal-to-noise ratio of the detector [90]. For cytometric analysis of ESCs and iPSCs in altered diff erentiation states and treatment conditions, variable cellular autofl uorescence and fl uorescence crossover between multiple fl uorochromes contribute to the background or 'noise' [91][92][93]. Increased noise decreases the resolving power for dim signals. A new cytometer developed by Sony Corporation (Tokyo, Japan) replaces fi lter-based optics with a spectral detection system based on a multi-anode spectral PMT. In the spectral cytometer, the light is directed on the basis of wavelength into discrete channels corresponding to diff erent regions of the color spectrum by using a series of prisms. Over 30 channels are distributed across the 500-to 800-nm color spectrum, which allows almost all photons to be processed, whereas at least 20% of photons may be lost in the dichroic mirror/bandpass fi lter system. Th e spectral cytometer is able to measure and subtract varying autofl uorescence, permitting increased signal-tonoise ratios and improving the resolution of dim signals (Figure 3). Multiple fl uorochromes are mathematically unmixed by using component analysis. Fluorochromes with overlapping emissions like fl uorescein isothiocyante and Alexa 532 can now be used simultaneously where their spectral overlap would prevent their use in a conventional fi lter-based cytometer. Unique spectral 'fi ngerprints' can be created to distinguish tissue precursors transfected with combinations of molecular beacons described above but without the need for genome integration of reporter constructs or surface antibody staining that might cause unintended pleiotropic cellular responses. Th is technology holds great promise for its ability to exploit functional biosensors [94][95][96][97], such as dual-FRET molecular beacons [6], to iden tify specifi c stem cell diff erentiation states of therapeutic value.

Conclusions
Research on pluripotent stem cells has progressed signi ficantly since the fi rst derivation of hESCs from discarded blastocysts in 1998 [98] and the discovery that somatic cells can be reprogrammed to a pluripotent stem-like state in 2006 [99]. Th e international scientifi c community has discovered the enormous potential of pluripotent stem cells as newly derived lines continue to be developed and diff erentiation methods into various types of cells are optimized. Th e ability to select and isolate stem cellderived precursors and diff erentiated cells with tissuespecifi c properties holds the key to fully exploiting these cell lines and diff erentiation methods for scientifi c investigation and clinical use.
Abbreviations bFGF, basic fi broblast growth factor; BMP4, bone morphogenetic protein 4; dbcAMP, dibutyryl cyclic adenosine monophosphate; EGF, epidermal growth factor; eGFP, enhanced green fl uorescent protein; ESC, embryonic stem cell; FGF, fi broblast growth factor; FRET, fl uorescence resonance energy transfer; Figure 3. 'Next generation' spectral fl ow cytometry. (a) With a conventional fl ow cytometer, lasers excite cell-associated fl uorochromes, and emitted light is fi ltered by a combination of dichroic mirrors (DMs) and bandpass fi lters (BFs) that refl ect and fi lter light of specifi c wavelengths, respectively. Light within narrow selected wavelength ranges arrives at a photomultiplier tube (PMT), which converts light as photons to an electronic signal. (b) In spectral fl ow cytometry, laser diodes similarly provide initial excitation of reporter fl uorochromes; however, emitted fl uorescence passes through a prism array into a spectral PMT. Component fl uorescence, including autofl uorescence, is linearly unmixed by using spectral lookup tables. In spectral fl ow cytometry, unlike conventional cytometry, almost all light signals are analyzed, and signal-to-noise resolution is dramatically improved.

Competing interests
HSB and WCH acknowledge Sony Corporation for providing the funds and the spectral fl ow cytometer required for their evaluation of the spectral detection system.