PERSPECTIVES 38. Hewish, D. R. & Burgoyne, L. A. Chromatin substructure. The digestion of chromatin at regularly spaced sites by a nuclear deoxyribonuclease. Biochem. Biophys. Res. Comm. 52, 504–510 (1973). 39. Oudet, P., Gross-Bellard, M. & Chambon, P. Electron microscopic and biochemical evidence that chromatin structure is a repeating unit. Cell 4, 281–300 (1975). 40. Wolffe, A. Chromatin Structure and Function (Academic Press, San Diego, 1998). 41. Kornberg, R. & Lorch, Y. Twenty-five years of the nucleosome, fundamental particle of the eukaryotic chromosome. Cell 98, 285–294 (1999). 42. Thomas, G. J., Prescott, B. & Olins, D. E. Secondary structure of histones and DNA in chromatin. Science 197, 385–388 (1977). 43. Mardian, J. K., Paton, A. E., Bunick, G. J. & Olins, D. E. Nucleosome cores have two specific binding sites for nonhistone chromosomal proteins HMG 14 and HMG 17. Science 209, 1534–1536 (1980). 44. Sandeen, G., Wood, W. I. & Felsenfeld, G. The interaction of high mobility proteins HMG14 and 17 with nucleosomes. Nucl. Acids Res. 8, 3757–3778 (1980). 45. Olins, D. E. et al. Electron microscope tomography: transcription in three dimensions. Science 220, 498–500 (1983). 46. Olins, A. L., Olins, D. E. & Bazett-Jones, D. P. Balbiani ring hnRNP substructure visualized by selective staining and electron spectroscopic imaging. J. Cell Biol. 117, 483–491 (1992). 47. Olins, D. E. & Olins, A. L. The replication band of ciliated protozoa. Int. Rev. Cytol. 153, 137–170 (1994). 48. Richmond, T. J., Finch, J. T., Rushton, B., Rhodes, D. & Klug, A. Structure of the nucleosome core particle at 7 Å resolution. Nature 311, 532–537 (1984). 49. Luger, K., Mader, A. W., Richmond, R. K., Sargent, D. F. & Richmond, T. J. Crystal structure of the nucleosome core particle at 2.8 Å resolution. Nature 389, 251–260 (1997). 50. Uberbacher, E. C. & Bunick, G. J. X-ray structure of the nucleosome core particle. J. Biomol. Struct. Dyn. 2, 1033–1055 (1985). 51. Harp, J. M., Hanson, B. L., Tim, D. E. & Bunick, G. J. Asymmetries in the nucleosome core particle at 2.5 Å resolution. Acta Crystallogr. (Section D Biol. Crystallogr.) 56, 1513–1534 (2000). 52. Harp, J. M., Hanson, B. L. & Bunick, G. J. The Structure of the Nucleosome Core Particle (Elsevier Science B. V., Amsterdam) (in the press). 53. Arents, G., Burlingame, R. W., Wang, B. C., Love, W. E. & Moudrianakis, E. N. The nucleosomal core histone octamer at 3.1 Å resolution: a tripartite protein assembly and a left-handed superhelix. Proc. Natl Acad. Sci. USA 88, 10148–10152 (1991). 54. Richmond, T. J. & Davey, C. A. The structure of DNA in the nucleosome core. Nature 423, 145–150 (2003). 55. Woodcock, C. L. F. & Dimitrov, S. Higher-order structure of chromatin and chromosomes. Curr. Opin. Genet. Dev. 11, 130–135 (2001). 56. Horn, P. J. & Peterson, C. L. Chromatin higher order folding: wrapping up transcription. Science 297, 1824–1827 (2002). 57. de la Serna, I. L. & Imbalzano, A. N. Unfolding heterochromatin for replication. Nature Genet. 32, 560–562 (2002). 58. Turner, B. M. Decoding the nucleosome. Cell 75, 5–8 (1993). 59. Jenuwein, T. & Allis, C. D. Translating the histone code. Science 293, 1074–1080 (2001). 60. Fry, C. J. & Peterson, C. L. Unlocking the gates to gene expression. Science 295, 1847–1848 (2002). 61. Heun, P., Laroche, T., Shimada, K., Furrer, P. & Gasser, S. M. Chromosome dynamics in the yeast interphase nucleus. Science 294, 2181–2186 (2001). 62. Hutchinson, C. Lamins: building blocks or regulators of gene expression. Nature Rev. Mol. Cell Biol. 3, 848–858 (2002). 63. Burke, B. & Stewart, C. L. Life at the edge: the nuclear envelope and human disease. Nature Rev. Mol. Cell Biol. 3, 575–585 (2002). 64. Hoffmann, K. et al. Mutations in the gene encoding the lamin B receptor produce an altered nuclear morphology in granulocytes (Pelger–Huët anomaly). Nature Genet. 31, 410–414 (2002). 65. Shultz, L. D. et al. Mutations at the mouse ichthyosis locus are within the lamin B receptor gene: a single gene model for human Pelger–Huët anomaly. Hum. Mol. Genet. 12, 61–69 (2003). 814 66. Waterham, H. R. et al. Autosomal recessive HEM/Greenberg skeletal dysplasia is caused by 3β−hydroxysterol ∆14-reductase deficiency due to mutations in the lamin B receptor gene. Am. J. Hum. Genet. 72, 1013–1017 (2003). 67. Grandville, J. J. Un Autre Monde (H. Fournier, Paris, 1844). 68. DuPraw, E. J. Quantitative constraints in the arrangement of human DNA. Cold Spring Harb. Symp. Quant. Biol. 38, 87–98 (1974). Acknowledgements The authors express their gratitude to Bowdoin College and to the German Cancer Research Center (Heidelberg) for providing stimulating intellectual and scientific environments. H. Herrmann and P. Lichter, our generous hosts at the German Cancer Research Center, supplied helpful comments on the manuscript. Several anonymous referees made significant contributions towards the improvement of this essay. The authors dedicate this review to the memory of H. G. Davis (formerly at the Department of Biophysics, King’s College, London), an excellent microscopist and our good friend. Online links DATABASES The following terms in this article are linked online to: LocusLink: http://www.ncbi.nlm.nih.gov/LocusLink/ H2A | H2B | H3 | H4 | HMGN1 | HMGN2 | lamin A | lamin B1 | lamin B2 | LBR OMIM: http://www.ncbi.nlm.nih.gov/Omim/ Emery–Dreifuss muscular dystrophy | Pelger–Huët anomaly Access to this interactive links box is free online. OPINION Can transcription factors function as cell–cell signalling molecules? Alain Prochiantz and Alain Joliot Recent data support the view that transcription factors — in particular, homeoproteins — can be transferred from cell to cell and have direct non-cellautonomous (and therefore paracrine) activities. This intercellular transfer, based on atypical internalization and secretion, has important biotechnological consequences. But the real excitement stems from the physiological and developmental implications of this mode of signal transduction. Transcription factors are present in the nucleus, and sometimes in the cytoplasm, but on the whole they are not thought to travel between cells. This is because of their hydrophilic properties and the absence of a signal peptide. But there are exceptions and, in fact, some transcription factors travel between cells because they contain protein domains that allow them to do so. This is the case for the HIV transcription factor TAT1 and for several homeoproteins, such as Engrailed2,3, Hoxa5, Hoxb4, Hoxc8, Emx1, Emx2, Otx2 and Pax6 (G. Mainguy, A. Maizel, A.P. and A.J., unpublished observations). On the basis of the conservation of the internalization and secretion signals that have been identified in Engrailed (see below), it is anticipated that this property is shared by most homeoproteins. Homeoproteins are known to contribute to cellular positioning. They were actually dis- | OCTOBER 2003 | VOLUME 4 covered in the fly on the basis of mutations that affect the spatial identity of segments and appendages (for example, antennae can be transformed into legs). Within a single structure, such as the spinal cord, specific combinations and concentrations of homeoproteins define the anterior–posterior and dorso–ventral positions of cells. Furthermore, the homeoprotein Engrailed can define the midbrain and the position of cells within the anterior–posterior axis of the midbrain. It is widely thought that homeoprotein function involves the regulation of genes that encode signalling molecules such as surface receptors or growth factors. By contrast, direct paracrine homeoprotein activity is not generally envisaged, although in theory it represents a parsimonious way for neighbouring cells to coordinate positional information. So the ability of homeoproteins to transfer between cells is extremely exciting. There are more than 400 of these proteins in mice and humans, and they are involved in all the main developmental decisions. Many of them also function in the control of adult physiology. For example, Engrailed 1 and Engrailed 2 (EN1 and EN2; collectively known as Engrailed) are expressed in adult aminergic nuclei that control motor behaviour, mood and addiction4. Because the transfer of positional information is a general phenomenon that occurs during development and throughout adulthood, because homeoproteins contribute to www.nature.com/reviews/molcellbio © 2003 Nature Publishing Group PERSPECTIVES Homeodomain Secretion sequence AQELGLNESQ Internalization sequence (Penetratin) RQIKIWFQNRRMKWKK Nuclear export sequence QSLAQELGLNESQIKI Figure 1 | Functional domains for homeoprotein intercellular transfer. Within the homeodomain, three domains that are required for secretion, internalization and nuclear export have been characterized by loss-of-function (deletion) or gain-of-function (synthetic peptides) studies. The secretion sequence (in green) is part of the nuclear export sequence (in yellow) and its deletion blocks nuclear export. The internalization sequence (in red) has been used as a vector to introduce cargoes into live cells and is therefore also known as Penetratin. It is important to note that the secretion and internalization sequences are distinct and that their presence within the homeodomain makes it impossible to define mutations that would block intercellular transfer without modifying transcription properties. encoding such information, and because homeoproteins can transfer between cells in vitro, we shall focus on homeoproteins to outline our model that transcription factors can function as true signalling proteins. In support of this hypothesis is the well-established transfer of homeoproteins in higher plants5, which occurs through intercytoplasmic bridges known as plasmodesmata. It is, however, noteworthy that the intercellular transfer of Knotted-1 (KN1), a maize homeoprotein, through plasmodesmata is poorly affected by conditions that strongly affect the passage of other types of protein6,7. So there might be other mechanisms of transfer that work in parallel. Homeoprotein entry into cells The evidence. The finding that the 60-aminoacid DNA-binding domain (homeodomain) of the Antennapedia homeoprotein translocates across biological membranes was serendipitous8. But, once observed, the phenomenon had to be investigated further. It was then shown that Antennapedia homeodomain internalization properties are conferred by its third helix (now referred to as Penetratin9; (FIG. 1)) and that this region absolutely requires the tryptophan at position 48 (W48) in a basic environment. Exogenously added fluorescein isothiocyanate (FITC)-labelled homeodomain and Penetratin peptides can be detected in the nucleus of live cells. Internalization is inhibited by the replacement of W48 by a phenylalanine. This, and the visualization of internalized homeodomains and homeoproteins in nonfixed cells8, precludes internalization being a post-fixation artefact (as has recently been suggested for several basic peptides and proteins10). The property of internalization into live cells is shared by all homeodomains that have been tested, probably because the third helix is highly conserved. Potential entry mechanisms. Biophysical data indicate that Penetratin fully translocates into artificial lipid vesicles11, which confirms that chiral receptors are not required for internalization. This was also shown by the internalization of a Penetratin peptide composed of 12 D-enantiomers , which cannot, therefore, recognize a protein receptor of opposite chirality (composed of L-amino acids). Current models of internalization involve peptide binding to the cell surface through charge interactions and membrane destabilization after W48 insertion in the bilayer. Destabilization might assist in the formation of inverted micelles13, which would allow peptides to translocate across the membrane and be released into the cytoplasm. Full-length homeoproteins are also internalized by live cells14 and, when tested, the same mutations (for example, the deletion of W48 in the homeodomain) block the internalization of the third helix, of the homeodomain or of the entire protein3,15. So, it is assumed that the same mechanisms are responsible in all cases when homeoproteins are internalized. identify a short sequence (∆1 sequence; FIG. 1) that is necessary for secretion. This sequence, which spans part of helices 2 and 3 of the homeodomain, is also required for efficient cEN2 nuclear export16, which indicates that secreted cEN2 might originate from the nucleus (FIG. 1; see below). As already mentioned, the homeodomain has a highly conserved structure and this nuclear export/secretion domain is also highly conserved, indicating that intercellular transfer might be a conserved property of homeoproteins. Accordingly, intercellular passage has been confirmed for almost all homeoproteins that have so far been tested, including Hoxa5, Hoxb4, Hoxc8, cEN2, Emx1, Emx2, Otx2 and Pax6 (G. Mainguy, A. Maizel, A.J. and A.P., unpublished observations). Potential secretion mechanisms. Subcellular localization studies showed that, in cultured cells, a pool of cEN2 was associated with membrane fractions and, in part, was protected against proteolysis17. By contrast, the non-secretable variant (∆1 deletion) is accessible to proteases3, which indicates that there might be an exportable pool of cEN2 (~5% of the entire Engrailed Cytosol D NLS ? Nucleus NES B ? NLS Secretion of homeoproteins from cells The evidence. The concept of a ‘messenger protein’ (FIG. 2) requires that transcription factors — in this case, homeoproteins — be also exported into the extracellular medium. This is despite the absence of a classical signal peptide, which is also absent from other hydrophilic proteins — in particular, interleukin-1 (IL-1) and mammalian fibroblast growth factors (FGFs). In co-cultures of primary neurons and COS-7 cells that express chick Engrailed 2 (cEN2), cEN2 is found — intact — in the 100,000 g supernatant (not in the pellet) of the culture medium and within the neurons3. This indicates that the protein has been secreted by the COS cells into the surrounding medium, and has then been taken up by the neurons. Interestingly, the cellular export and import pathways seem to be distinct, as mutating W48, which is mandatory for internalization, does not affect cEN2 export. Systematic mutagenesis has been used to NATURE REVIEWS | MOLECUL AR CELL BIOLOGY C A Figure 2 | Intercellular transfer of Engrailed homeoprotein. At equilibrium, Engrailed is detected in the cytosol and nucleus. In principle, this dual distribution can result from an early nucleocytoplasmic partition of newly synthesized Engrailed (A) or, alternatively, from Engrailed continuous shuttling between the nucleus and cytosol owing to its nuclear import (NLS) and export (NES) sequences (B). Our data favour the latter hypothesis and indicate that nuclear export might comprise an important step to gain access to the secretory compartment. So, Engrailed transport from the nucleus to the cytoplasm (binding to transporters, dissociation from DNA) could be a site of regulation for its secretion. Only forms of Engrailed that can be secreted are directed into the lumen of vesicular compartments (C) by a mechanism that is still undefined but that is distinct from the internalization mechanism (D). Engrailed internalization (D) is endocytosis independent, and requires the third helix of the homeodomain as the driving domain. VOLUME 4 | OCTOBER 2003 | 8 1 5 © 2003 Nature Publishing Group PERSPECTIVES Growing axon hydrophilic proteins seems unlikely. In fact, in genetically well-characterized organisms (such as bacteria), six different unconventional secretion pathways have been identified, even though they usually do not co-exist in the same organism. Target Cytosol a Stabilization a′ b Collapse Nucleus Mature axon a a a′ mRNA a′ b Presynaptic axon Ribosome Nascent polypeptide b Postsynaptic element Figure 3 | Hypothetical signalling with homeoproteins. The example taken here is signalling between either a navigating axon (growing axon) and an intermediate target, or the pre- and postsynaptic elements of a functioning synapse (mature axon). The main advantage of this model is that, in addition to a classical signalling activity (for example, a neurotransmitter-induced change in ionic conductance), it also confers positional information. Indeed, because the combination of homeoproteins expressed within a cell is a signature of its position in the organism, the cells are reciprocally informed of their topological origins. The exchange of transcription factors between the axon (growing or mature; blue) and the target cell (green) can be interpreted at the level of translation (a, a′) or transcription (b). For example, homeoprotein transfer into the growth cone might regulate (up or down) the translation (a) of a messenger encoding a receptor or an adhesion molecule, leading to stabilization or turning away (or collapse) of the growth cone. Reciprocally, homeoprotein passage from the growth cone or nerve terminal into the target cell might regulate the translation (a′) of messengers (for example, sub-synaptic dendritic messengers in adult synapses) or the transcription of specific genes (b) after transfer into the target-cell nucleus. Indeed, it cannot be precluded that, following transfer, homeoproteins show unsuspected physiological activities (for example, activation of second messengers). Green circles represent proteins originating from the growth cone or presynaptic axon, and the red circles represent proteins travelling in the opposite direction. cellular pool) that exists within vesicles. These vesicles have ‘caveolae-like’ properties, such as high cholesterol and glycosphingolipid contents, which confer resistance to Triton and separation in sucrose density gradients17. The same proportion of endogenous rat EN1 and EN2 that is expressed in the embryonic mesencephalon is resistant to proteolysis and present in these caveolae-like vesicles in vivo. Although intercellular transfer has only been directly visualized in vitro, the presence of EN1 and EN2 in a secretion compartment in vivo gives strength to the idea that this phenomenon is of physiological relevance. So how might these homeoprotein-containing vesicles arise? An interesting possibility is that homeoproteins are incorporated into nascent vesicles as they form; exosomes, with their inside–out membrane orientation (their cytosolic face is inside), are ideal candidates. In fact, galectin-3 and annexin II — two 816 secreted hydrophilic proteins — have been detected in exosomes by mass spectroscopy18. In some cases, for example for IL-1β19 and FGF2 (REF. 20), protein translocation channels of the multidrug resistance (MDR) family seem to be used for unconventional secretion. However, the correlation between Engrailed secretion and its localization in the lumen of vesicles3 indicates the potential existence of an unidentified mechanism that directs Engrailed into secretory vesicles. Unconventional secretion does not necessarily require cell-type-specific machineries, and so it might reflect an ancestral mode of secretion. For instance, MDR-dependent secretion is found in organisms ranging from bacteria to humans. Another example is the unconventional secretion of thioredoxin that is observed in bacteria21 (Helicobacter pylori), plants22 and vertebrates23. Taken together, a single unconventional pathway for all secreted | OCTOBER 2003 | VOLUME 4 An active role for the nucleus? Intriguingly, a vast number of secreted hydrophilic proteins find their way to the cell nucleus (VP22, galectin-3, FGFs, thioredoxin, high-mobility group box (HMGB) proteins and homeoproteins) and some of them are capable of nucleocytoplasmic exchange24. This exchange might control the amount of protein that is available for secretion, as alluded to below. The nucleocytoplasmic distribution of FGF2 (including its 18-kDa secreted isoform) is highly regulated during development25. In culture, phosphorylation modulates (directly or indirectly) the nucleocytoplasmic distribution of both FGF2 (REF. 26) and thioredoxin27. FGF1 accumulates in the nucleus before reaching the cytoplasm28. Transfer of nuclear non-histone HMGB1 protein to the cytoplasm precedes its lysophosphatidylcholine-induced secretion29. Both annexin II 30 and Engrailed16 contain an active nuclear export signal. Furthermore, secretion often correlates with nuclear, rather than cytoplasmic, localization. The phorbol ester phorbol 12-myristate 13-acetate (PMA) promotes both thioredoxin secretion31 and its nuclear localization27. Similarly, amino-terminal deletion abolishes galectin-3 secretion and its nuclear accumulation32. In the case of cEN2, the fact that an active nuclear export sequence that takes the protein out of the nucleus is necessary for cEN2 secretion might be seen as a counter example. However, the addition of an extra nuclear localization signal (NLS) does not impair secretion (A. Maizel and A.J., unpublished observations) and many modifications that decrease cEN2 secretion correlate with a pronounced cytoplasmic localization. For example, Engrailed phosphorylation blocks its secretion33 and favours its accumulation in the cytoplasm (A. Maizel and A.J., unpublished observations). So, although exit from the nucleus is important to gain access to the secretory compartment, a passage through the nucleus seems necessary, therefore indicating an active role for the nucleus in the secretion of secreted hydrophilic proteins. Whether this role is related to the nuclear accumulation of some endocytic proteins34 or to the nucleus–membrane shuttling of some proteins (for example, the scaffold proteins Ste5 (REF. 35) or regulator of G-protein signalling 4 (REF. 36)) remains to be investigated. www.nature.com/reviews/molcellbio © 2003 Nature Publishing Group PERSPECTIVES Transfer in other phyla Behaviours that are indicative of transcription-factor transfer have been observed in several phyla. This generality might indicate that the phenomenon is important. Yeast and bacterial functions. In many lower eukaryotes, some homeoprotein functions are related to intercellular communication. Following cell fusion that accompanies mating, the homeoprotein content of each contributing cell determines the behaviour of the resultant diploid cell38,39. Direct transfer of Functions in higher plants. Movement proteins that undergo intercellular transfer are commonplace in plants and some of these proteins are transcription factors. This explains why paracrine activity of transcription factors, including homeoproteins, has been proposed in higher plants. In specific cases, homeoproteins might even be used as chaperones for protein transfer, as implied by the observation that a single mutation antagonizes the intercellular transfer of a plantvirus movement protein and its binding to a homeodomain protein42. Because movement proteins traffic through plasmodesmata, it is thought that homeoproteins use the same pathway. However, the homeodomain of the shuttling plant homeoprotein KN1 can be transferred between animal cells (which lack plasmodesmata), and the only mutation — located in the homeodomain — that is known to inhibit KN1 intercellular transfer in plants5 also inhibits the intercellular transfer of the KN1 homeodomain between animal cells (M. Tasseto, A. Maizel, A.P. and A.J., unpublished observations). This indicates a possible homology between metaphytes and metazoans with respect to homeoprotein transfer. In fact, it would be curious if, even though animal proteins have efficient import and export signals, animals are unique in not using this powerful signalling pathway. In search of paracrine functions Most secreted hydrophilic proteins have extracellular functions that involve interactions with plasma-membrane receptors (these could be on the cells that secrete them, as well as on different cells). However, for some of them, such as TAT43 and FGFs44,45, internalization is known to be required for them to elicit their full range of biological effects. Homeoproteins, however, seem to be atypical secreted hydrophilic proteins that so far have no characterized extracellular functions in metazoans. It is therefore anticipated that their paracrine targets are intracellular. Because homeoproteins regulate transcription and, in some instances, translation, these two modes of action are most likely to affect paracrine action. NATURE REVIEWS | MOLECUL AR CELL BIOLOGY Transcription and translation. It seems evident that homeoproteins that have been transferred from one cell to another should regulate transcription in the nucleus of the receiving cell. This is likely, but nonautonomous target genes might differ from autonomous ones — for example, the proteins might become modified during transport, or they might take up a different role in a VE Epi VE Otx2 Epi Otx2 ? Otx2 VE cell Epi cell ANE transcription factors from pathogen to host also happens during infection by pathogenic bacteria40. Also, although it is not considered to be a transcription factor, the T-DNA-binding protein VirE2 of Agrobacterium is directly involved in the penetration of the host plasma membrane by the T-DNA, and its subsequent nuclear import41. AVE The data described above show that cEN2 and many other homeoproteins can pass between cells by two original and distinct mechanisms of secretion and internalization. Secretion involves a category of vesicles that contain the protein within their luminal compartment, and EN1 and EN2 are found in such vesicles in vivo. Finally, the sequences that are necessary for unconventional export and import both reside within the homeodomain and are conserved among homeoproteins. How might this intercellular transfer be regulated? The absence of chiral receptors and a mode of entry that is primarily based on the interaction of the proteins with charged phospholipids almost certainly preclude any regulation of intercellular transfer at the entry site. By contrast, the secretion of homeoproteins is probably highly regulated. Engrailed is a phosphoprotein in vivo and is phosphorylated by casein kinase 2 (CK2). In CK2–cEN2 co-transfection experiments, CK2 blocks cEN2 secretion by phosphorylating a short serine-rich domain that is upstream of the homeodomain33. Secretion of proteins in which serine residues within the serine-rich domain were mutated to alanine residues was insensitive to CK2. By contrast, the replacement of the same serine residues by glutamate residues — mimicking phosphorylated serine residues — produced a protein that was unable to be secreted. Other putative modes to regulate secretion control are the regulation of the nucleocytoplasmic distribution or the fusion of caveolae-like vesicles with the plasma membrane. The latter process is known to be regulated, in particular, through protein kinase C activation37, and therefore any mechanism regulating the fusion of these vesicles with the plasma membrane might also regulate homeoprotein exocytosis. So, homeoprotein secretion is probably a highly regulated process that can only happen in specific physiological and developmental situations. AVE Regulation of intercellular transfer Otx2 b Neural tube (E 8,5) Otx2 MHB Anterior Posterior c MHB Wild-type d Otx2 e Gbx2 MHB Otx1–/– Otx2+/– f En1Otx2 MHB +/– Figure 4 | A model for homeoproteins as infectious proteins. a | Signalling and inductive activity between two layers is presented here in the case of Otx2. Otx2 that is first expressed in the anterior visceral endoderm (AVE; left) is transported into the anterior epiblast (Epi), where it activates Otx2 transcription (middle panels), which is necessary for the formation of the anterior neural ectoderm (ANE). This auto-induction could be dependent of, or amplified by, co-factors. Following its induction in the epiblast, Otx2 spreads by cell–cell transfer into the posterior part of the neuroepithelium (b) until it encounters a nuclear environment (such as the presence of Gbx2) that represses its replication. The latter model is hypothetical and supported by the following observations: first, homeoproteins can be secreted and internalized; second, in the wildtype (c), the mid–hindbrain border (MHB) forms where Otx2 and Gbx2 meet; third, Gbx2 and Otx2 are auto-activators and reciprocal inhibitors (d); fourth, downregulation of Otx expression shifts the expression of Gbx2 and the position of the MHB into a more anterior position (e); and, finally, conversely forcing the expression of Otx2 in the En1 domain represses Gbx2 and shifts the MHB position into a more posterior position (f). VOLUME 4 | OCTOBER 2003 | 8 1 7 © 2003 Nature Publishing Group PERSPECTIVES a different cellular context. But additional functions can also be envisaged. As mentioned, homeoproteins might regulate messenger RNA stability and/or translation (FIG. 3a). Direct evidence that a homeoprotein can regulate translation is the repression, in Drosophila, of caudal mRNA translation by Bicoid46,47. The Bicoid mRNA-binding domain has been identified48 and its conservation among homeoproteins raises the possibility that other homeoproteins bind RNA. Indirect evidence is that Bicoid-dependent repression of translation involves its binding to the eukaryotic translation initiation factor 4E (eIF4E)49, a property that is shared by proline-rich homeobox proteins and Hox11 (REF. 50). Most importantly, the domain of interaction with eIF4E is present in 200 other homeoproteins50. Homeoproteins are known to have a role in shaping neuronal arbors8,15,51,52. The development of the nervous system involves the formation of many pathways and synapses with an extremely precise topology. In addition, the functioning of adult neuronal networks requires that each synapse be informed of its position within the network. So, economical mechanisms for position coding are of primary importance in the developing and adult nervous systems. Because homeoproteins contribute to positional information, homeoprotein transfer would indeed be a very economical way to combine signal transduction and topological information (FIG. 3). We speculate that homeoproteins might affect axonal navigation during development and modify synaptic properties within adult neuronal networks by regulating transcription and/or translation in the receiving compartment (FIG. 3). For example, the levels of cell-surface receptors or cell-adhesion molecules might be changed, which, in turn, would affect axon guidance. However, we do not exclude other unsuspected modes of paracrine activity, such as, for example, activating second messengers. Could homeoproteins be morphogens? During development, the transfer of positional information is either vertical, between layers of cells, or tangential, within layers. A clear example of vertical induction is the expression of the homeoprotein Otx2 in the anterior visceral endoderm, from where it is transported into the anterior epiblast. Here, it induces its own transcription, which is necessary for the formation of the anterior neural ectoderm. The sequential expression of this anterior genetic marker in the two layers involves a non-cellautonomous (direct or indirect) function of Otx2 (FIG. 4a; reviewed in REF. 53). 818 Tangential induction within an epithelium has been proposed as a patterning mechanism54. We speculate that homeoprotein tangential transfer might be involved in border formation within the neuroepithelium, thereby participating in the formation of the compartments of the future brain. For example, the isthmus is an important embryonic brain structure that forms where the expression of Otx2 meets that of Gbx2 (REF. 55; FIG. 4b), and genetically decreasing Otx2 or Gbx2 levels anteriorizes or posteriorizes (respectively) the position of the isthmus. The same holds true for Emx2 and Pax6 in the compartmentalization of the cortex56, of Gsh2 and Pax6 for the dorso–ventral patterning of the telencephalon57, or of Engrailed and Pax6 for the mesencephalon–diencephalon border58. It is striking that each homeoprotein in a pair activates its own transcription and antagonizes that of its ‘partner’. In the hypothetical models outlined in FIG. 4, homeoproteins can be considered as ‘infectious entities’ — they invade the neuroepithelium until they encounter a nuclear environment that is unsuitable for their own induction, partly because of high expression of the antagonist homeoprotein (FIG. 4b). In fact, the protein does not have to travel far — just one cell — but propagates rapidly by inducing and amplifying its own transcription. All this happens as if the genes themselves were diffusible, a term that evokes the definition of morphogens by Alan Turing in 1953 (REF. 59): “The substances will be called morphogens, the word being intended to convey the idea of a form producer. It is not intended to have any very exact meaning, but is simply the kind of substance concerned in this theory…The genes themselves may also be considered to be morphogens. But they form rather a special class. They are quite indiffusible…” Testing hypotheses Homeoprotein intercellular transfer in vivo has been observed in plants5 but not in metazoans. Although it is possible that this is the true scenario, it might be that homeoprotein transfer in animals is below the threshold of detection or is highly regulated in time and place. In favour of homeoprotein transfer is the presence of endogenous Engrailed in the lumen of vesicles17, and that of Emx1 in the axonal terminals of olfactory receptors60. In addition, the existence of highly conserved but distinct sequences that allow secretion and capture is probably not purely coincidental. In others words, we think that the main question is not whether | OCTOBER 2003 | VOLUME 4 homeoproteins are transferred between cells but when, where and why they are. To test the proposed hypotheses, it will be necessary to block the intercellular transfer of homeoproteins in vivo without modifying their intracrine activity. A main objective would be to identify and mutate sequences that confer direct paracrine properties and to insert the mutated gene into the normal locus to produce an ‘intracrine-only’ mouse. As shown in FIG. 1, the two sequences that are necessary for secretion and internalization are in the homeodomain. This is very interesting with respect to homeodomain conservation, but makes it impossible to introduce mutations that would not modify the roles of these proteins as transcriptional regulators. So, the strategy that is being used at present is to design mini-genes that encode homeoprotein-binding polypeptides and to force their expression in the extracellular milieu at the appropriate developmental periods. Such anti-homeoprotein agents are being produced and it is our hope that their expression under the control of the appropriate regulatory sequences will allow us to prevent or retard homeoprotein transfer without interfering with their intracrine functions. Alain Prochiantz and Alain Joliot are both at Centre National de la Recherche Scientifique, Unité Mixte de Recherche 8542, Ecole Normale Supérieure, 46 rue d’Ulm, 75230 Paris Cedex 05, France. e-mails: prochian@wotan.ens.fr; joliot@wotan.ens.fr doi:10.1038/nrm1227 1. Ensoli, B. et al. Release, uptake, and effects of extracellular human immunodeficiency virus type 1 Tat protein on cell growth and viral transactivation. J. Virol. 67, 277–287 (1993). 2. Prochiantz, A. Messenger proteins: homeoproteins, TAT and others. Curr. Opin. Cell Biol. 12, 400–406 (2000). 3. Joliot, A. et al. Identification of a signal sequence necessary for the unconventional secretion of Engrailed homeoprotein. Curr. Biol. 8, 856–863 (1998). 4. Simon, H. H., Saueressig, H., Wurst, W., Goulding, M. D. & O’Leary, D. D. Fate of midbrain dopaminergic neurons controlled by the engrailed genes. J. Neurosci. 21, 3126–3134 (2001). 5. Lucas, W. J. et al. Selective trafficking of KNOTTED1 homeodomain protein and its mRNA through plasmodesmata. Science 270, 1980–1983 (1995). 6. Lee, J. Y. et al. Selective trafficking of non-cellautonomous proteins mediated by NtNCAPP1. Science 299, 392–396 (2003). 7. Kragler, F., Monzer, J., Xoconostle-Cazares, B. & Lucas, W. J. Peptide antagonists of the plasmodesmal macromolecular trafficking pathway. EMBO J. 19, 2856–2868 (2000). 8. Joliot, A., Pernelle, C., Deagostini-Bazin, H. & Prochiantz, A. Antennapedia homeobox peptide regulates neural morphogenesis. Proc. Natl Acad. Sci. USA 88, 1864–1868 (1991). 9. Derossi, D., Joliot, A. H., Chassaing, G. & Prochiantz, A. The third helix of Antennapedia homeodomain translocates through biological membranes. J. Biol. Chem. 269, 10444–10450 (1994). 10. Lundberg, M. & Johansson, M. Positively charged DNA-binding proteins cause apparent cell membrane translocation. Biochem. Biophys. Res. Commun. 291, 367–371 (2002). www.nature.com/reviews/molcellbio © 2003 Nature Publishing Group PERSPECTIVES 11. Thoren, P. E., Persson, D., Karlsson, M. & Norden, B. The antennapedia peptide penetratin translocates across lipid bilayers — the first direct observation. FEBS Lett. 482, 265–268 (2000). 12. Derossi, D. et al. Cell internalization of the third helix of the Antennapedia homeodomain is receptorindependent. J. Biol. Chem. 271, 18188–18193 (1996). 13. Berlose, J. P., Convert, O., Derossi, D., Brunissen, A. & Chassaing, G. Conformational and associative behaviours of the third helix of Antennapedia homeodomain in membrane-mimetic environments. Eur. J. Biochem. 242, 372–386 (1996). 14. Chatelin, L., Volovitch, M., Joliot, A. H., Perez, F. & Prochiantz, A. Transcription factor Hoxa-5 is taken up by cells in culture and conveyed to their nuclei. Mech. Dev. 55, 111–117 (1996). 15. Le Roux, I., Joliot, A. H., Bloch-Gallego, E., Prochiantz, A. & Volovitch, M. Neurotrophic activity of the Antennapedia homeodomain depends on its specific DNA-binding properties. Proc. Natl Acad. Sci. USA 90, 9120–9124 (1993). 16. Maizel, A., Bensaude, O., Prochiantz, A. & Joliot, A. A short region of its homeodomain is necessary for Engrailed nuclear export and secretion. Development 126, 3183–3190 (1999). 17. Joliot, A. et al. Association of engrailed homeoproteins with vesicles presenting caveolae-like properties. Development 124, 1865–1875 (1997). 18. Thery, C. et al. Proteomic analysis of dendritic cellderived exosomes: a secreted subcellular compartment distinct from apoptotic vesicles. J. Immunol. 166, 7309–7318 (2001). 19. Hamon, Y. et al. Interleukin-1β secretion is impaired by inhibitors of the Atp binding cassette transporter, ABC1. Blood 90, 2911–2915 (1997). 20. Gupta, S., Aggarwal, S. & Nakamura, S. A possible role of multidrug resistance-associated protein (MRP) in basic fibroblast growth factor secretion by AIDS-associated Kaposi’s sarcoma cells: a survival molecule? J. Clin. Immunol. 18, 256–263 (1998). 21. Kim, N. et al. Proteins released by Helicobacter pylori in vitro. J. Bacteriol. 184, 6155–6162 (2002). 22. Ishiwatari, Y. et al. Thioredoxin H is one of the major proteins in rice phloem sap. Planta 195, 456–463 (1995). 23. Rubartelli, A., Bajetto, A., Allavena, G., Wollman, E. & Sitia, R. Secretion of thioredoxin by normal and neoplastic cells through a leaderless secretory pathway. J. Biol. Chem. 267, 24161–24164 (1992). 24. Davidson, P. J. et al. Shuttling of galectin-3 between the nucleus and cytoplasm. Glycobiology 12, 329–337 (2002). 25. Dono, R. & Zeller, R. Cell-type-specific nuclear translocation of fibroblast growth factor-2 isoforms during chicken kidney and limb morphogenesis. Dev. Biol. 163, 316–330 (1994). 26. Stachowiak, M. K. et al. Regulation of bFGF gene expression and subcellular distribution of bFGF protein in adrenal medullary cells. J. Cell Biol. 127, 203–223 (1994). 27. Hirota, K. et al. AP-1 transcriptional activity is regulated by a direct association between thioredoxin and Ref-1. Proc. Natl Acad. Sci. USA 94, 3633–3638 (1997). 28. Cao, Y. & Pettersson, R. Release and subcellular localization of acidic fibroblast growth factor expressed to high levels in HeLa cells. Growth Factors 8, 277–290 (1993). 29. Gardella, S. et al. The nuclear protein HMGB1 is secreted by monocytes via a non-classical, vesicle-mediated secretory pathway. EMBO Rep. 3, 995–1001 (2002). 30. Eberhard, D. A., Karns, L. R., VandenBerg, S. R. & Creutz, C. E. Control of the nuclear–cytoplasmic partitioning of annexin II by a nuclear export signal and by p11 binding. J. Cell Sci. 114, 3155–3166 (2001). 31. Sahaf, B. & Rosen, A. Secretion of 10-kDa and 12-kDa thioredoxin species from blood monocytes and transformed leukocytes. Antioxid. Redox Signal. 2, 717–726 (2000). 32. Gong, H. C. et al. The NH2 terminus of galectin-3 governs cellular compartmentalization and functions in cancer cells. Cancer Res. 59, 6239–6245 (1999). 33. Maizel, A. et al. Engrailed homeoprotein secretion is a regulated process. Development 129, 3545–3553 (2002). 34. Vecchi, M. et al. Nucleocytoplasmic shuttling of endocytic proteins. J. Cell Biol. 153, 1511–1517 (2001). 35. Mahanty, S. K., Wang, Y., Farley, F. W. & Elion, E. A. Nuclear shuttling of yeast scaffold Ste5 is required for its recruitment to the plasma membrane and activation of the mating MAPK cascade. Cell 98, 501–512 (1999). 36. Chatterjee, T. K. & Fisher, R. A. Cytoplasmic, nuclear, and Golgi localization of RGS proteins. Evidence for N-terminal and RGS domain sequences as intracellular targeting motifs. J. Biol. Chem. 275, 24013–24021 (2000). 37. Smart, E. J., Ying, Y. S. & Anderson, R. G. Hormonal regulation of caveolae internalization. J. Cell Biol. 131, 929–938 (1995). 38. Goutte, C. & Johnson, A. D. Yeast a1 and α2 homeodomain proteins form a DNA-binding activity with properties distinct from those of either protein. J. Mol. Biol. 233, 359–371 (1993). 39. Gillissen, B. et al. A two-component regulatory system for self/non-self recognition in Ustilago maydis. Cell 68, 647–657 (1992). 40. Van den Ackerveken, G., Marois, E. & Bonas, U. Recognition of the bacterial avirulence protein AvrBs3 occurs inside the host plant cell. Cell 87, 1307–1316 (1996). 41. Dumas, F., Duckely, M., Pelczar, P., Van Gelder, P. & Hohn, B. An Agrobacterium VirE2 channel for transferred-DNA transport into plant cells. Proc. Natl Acad. Sci. USA 98, 485–490 (2001). 42. Desvoyes, B., Faure-Rabasse, S., Chen, M. H., Park, J. W. & Scholthof, H. B. A novel plant homeodomain protein interacts in a functionally relevant manner with a virus movement protein. Plant Physiol. 129, 1521–1532 (2002). 43. Thomas, C. A., Dobkin, J. & Weinberger, O. K. TAT-mediated transcellular activation of HIV-1 long terminal repeat directed gene expression by HIV-1infected peripheral blood mononuclear cells. J. Immunol. 153, 3831–3839 (1994). 44. Bailly, K., Soulet, F., Leroy, D., Amalric, F. & Bouche, G. Uncoupling of cell proliferation and differentiation activities of basic fibroblast growth factor. FASEB J. 14, 333–344 (2000). 45. Wiedlocha, A., Falnes, P., Madshus, H., Sandvig, K. & Olsnes, S. Dual mode of signal transduction by externally added acidic fibroblast growth factor. Cell 76, 1039–1051 (1994). 46. Rivera-Pomar, R., Niessing, D., Schmidt, O. U., Gehring, W. J. & Jäckle, H. RNA binding and translational suppression by bicoid. Nature 379, 746–749 (1996). 47. Dubnau, J. & Struhl, G. RNA recognition and translational regulation by a homeodomain protein Nature 379, 694–699 (1996). NATURE REVIEWS | MOLECUL AR CELL BIOLOGY 48. Niessing, D. et al. Homeodomain position 54 specifies transcriptional versus translational control by Bicoid. Mol. Cell 5, 395–401 (2000). 49. Niessing, D., Blanke, S. & Jäckle, H. Bicoid associates with the 5′-cap-bound complex of caudal mRNA and represses translation. Genes Dev. 16, 2576–2582 (2002). 50. Topisirovic, I. et al. The proline-rich homeodomain protein, PRH, is a tissue-specific inhibitor of eIF4Edependent cyclin D1 mRNA transport and growth. EMBO J. 22, 689–703 (2003). 51. Marie, B., Cruz-Orengo, L. & Blagburn, J. M. Persistent engrailed expression is required to determine sensory axon trajectory, branching, and target choice. J. Neurosci. 22, 832–841 (2002). 52. Bloch-Gallego, E. et al. Antennapedia homeobox peptide enhances growth and branching of embryonic chicken motoneurons in vitro. J. Cell Biol. 120, 485–492 (1993). 53. Simeone, A. & Acampora, D. The role of Otx2 in organizing the anterior patterning in mouse. Int. J. Dev. Biol. 45, 337–345 (2001). 54. Ruiz i Altaba, A. Planar and vertical signals in the induction and patterning of the Xenopus nervous system. Development 116, 67–80 (1992). 55. Li, J. Y. & Joyner, A. L. Otx2 and Gbx2 are required for refinement and not induction of mid-hindbrain gene expression. Development 128, 4979–4991 (2001). 56. Bishop, K. M., Rubenstein, J. L. & O’Leary, D. D. Distinct actions of Emx1, Emx2, and Pax6 in regulating the specification of areas in the developing neocortex. J. Neurosci. 22, 7627–7638 (2002). 57. Toresson, H., Potter, S. S. & Campbell, K. Genetic control of dorsal-ventral identity in the telencephalon: opposing roles for Pax6 and Gsh2. Development 127, 4361–4371 (2000). 58. Matsunaga, E., Araki, I. & Nakamura, H. Pax6 defines the di-mesencephalic boundary by repressing En1 and Pax2. Development 127, 2357–2365 (2000). 59. Turing, A. M. The chemical basis of morphogenesis. Bull. Math. Biol. 1990, 153–197 (1953). 60. Briata, P. et al. EMX1 homeoprotein is expressed in cell nuclei of the developing cerebal cortex and in the axons of the olfactory sensory neurons. Mech. Dev. 57, 169–180 (1996). Acknowledgements We would like to thank the members of our groups for many helpful discussions and C. Goridis for his critical reading of the manuscript. Many of the experiments discussed in this article were supported by the Centre National de la Recherche Scientifique and the Ecole Normale Supérieure, and by grants from the Association Française de lutte contre les Myopathies, the Human Frontier Research Program and the European Economic Community. Online links DATABASES The following terms in this article are linked online to: FlyBase: http://flybase.bio.indiana.edu/ Antennapedia | Engrailed LocusLink: http://www.ncbi.nlm.nih.gov/LocusLink/ eIF4E | FGF1 | FGF2 | Gbx2 | Gsh2 | IL-1 | IL-1β | HMGB1 | Otx2 | Pax6 Access to this interactive links box is free online. VOLUME 4 | OCTOBER 2003 | 8 1 9 © 2003 Nature Publishing Group